A Survey of Molecular Communication in Cell Biology: Establishing a New Hierarchy for Interdisciplinary Applications
Dadi Bi, Apostolos Almpanis, Adam Noel, Yansha Deng, Robert Schober
11 A Survey of Molecular Communication inCell Biology: Establishing a New Hierarchy forInterdisciplinary Applications
Dadi Bi,
Student Member, IEEE , Apostolos Almpanis,
Student Member, IEEE ,Adam Noel,
Member, IEEE , Yansha Deng,
Member, IEEE , and Robert Schober,
Fellow, IEEE
Abstract —Molecular communication (MC) engineering is in-spired by the use of chemical signals as information carriers incell biology. The biological nature of chemical signaling makesMC a promising methodology for interdisciplinary applicationsrequiring communication between cells and other microscaledevices. However, since the life sciences and communicationsengineering fields have distinct approaches to formulating andsolving research problems, the mismatch between them canhinder the translation of research results and impede the de-velopment and implementation of interdisciplinary solutions. Tobridge this gap, this survey proposes a novel communicationhierarchy for MC signaling in cell biology and maps phenomena,contributions, and problems to the hierarchy. The hierarchyincludes: 1) the physical propagation of cell signaling at thePhysical Signal Propagation level; 2) the generation, reception,and biochemical pathways of molecular signals at the Physicaland Chemical Signal Interaction level; 3) the quantification ofphysical signals, including macroscale observation and controlmethods, and conversion of signals to information at the Signal-Data Interface level; 4) the interpretation of information incell signals and the realization of synthetic systems to store,process, and communicate molecular signals at the Local DataAbstraction level; and 5) applications relying on communicationwith MC signals at the Application level. To further demonstratethe proposed hierarchy, it is applied to case studies on quorumsensing, neuronal signaling, and communication via DNA. Finally,several open problems are identified for each level and theintegration of multiple levels. The proposed hierarchy provideslanguage for communication engineers to study and interfacewith biological systems, and also helps biologists to understandhow communications engineering concepts can be exploited tointerpret, control, and manipulate signaling in cell biology.
Index Terms —Cell biology, chemical reactions, diffusion, hier-archy, interdisciplinary applications, level, microfluidics, molec-ular communication, signaling pathways, synthetic biology.
I. I
NTRODUCTION L IKE human beings, cells have their own “social ac-tivities” and are in constant communication with eachother. One way they achieve this is by continuously sensing,
Manuscript draft. Funding information here. (Dadi Bi and ApostolosAlmpanis are co-first authors.) (Corresponding author: Yansha Deng.)D. Bi and Y. Deng are with the Department of Engineering, King’sCollege London, London WC2R 2LS, U.K. (email: { dadi.bi, yan-sha.deng } @kcl.ac.uk).A. Almpanis and A. Noel are with the School of Engineering, Uni-versity of Warwick, Coventry CV4 7AL, U.K. (email: { tolis.almpanis,adam.noel } @warwick.ac.uk).R. Schober is with the Institute for Digital Communications, Friedrich-Alexander-Universit¨at Erlangen-N¨urnberg, 91058 Erlangen, Germany (e-mail:[email protected]). receiving, and interpreting extracellular signaling molecules,and then coordinating their behaviors in response. This formof information exchange, termed molecular communication(MC), is a biologically-inspired communication paradigm,where information is exchanged via chemical signals [1],[2]. The basic concepts and the architecture of MC wereinitially proposed and described to the research communityin 2005 [3], [4]. After empirical work aimed to validate thefeasibility of MC, this novel field has been primarily occupiedand developed by the theoretical communications researchcommunity [5].Significant progress has been made over the last decade witha flourish of activity to understand the biophysical characteris-tics of molecule propagation using tools and mechanisms fromcommunication engineering. The focus of channel modelingresearch has spanned from basic Brownian motion [6] tomolecular transport with fluid flow [7] and active propagationthat relies on energy sources, such as molecular motors [8]and bacterial chemotaxis [9]. The interactions between in-formation molecules and the receiver have been extensivelystudied for passive reception [10] and full absorption [11],and recent works have modeled receiver-side reaction kineticsmore precisely, e.g., reversible adsorption [12] and ligand-binding [13]. While many works have been based on trans-mission using simple on-off keying modulation [14], moresophisticated modulation and coding schemes have been de-veloped for molecular transmission with higher data rates andimproved communication reliability [15], [16]. AccompanyingMC system design has been information-theoretical researchto quantify the fundamental limits of molecular signaling,i.e., the communication capacity [2]. In addition to theoreticalresearch, experimental research on MC has sought to validatetheoretical models and provide pathways towards applications,both at microscale [17]–[21] and macroscale [22], [23]. Moredetails on channel modeling, modulation and coding, com-munication capacity, physical design, and biological buildingblocks can be found in recent surveys [24]–[28], respectively.With the ultimate goal of enabling practical and paradigm-shifting applications, such as disease diagnosis, drug deliv-ery, and health monitoring, the MC community has soughtexploitation in cross-disciplinary research. For example, fordisease diagnosis, evaluating the capacity of the brain toencode and retrieve memories could reveal the dysfunctionand loss of synaptic communication due to Alzheimer’s andother neurodegenerative diseases [29]. For drug delivery, MC a r X i v : . [ c s . ET ] A ug theory has been applied to characterize the transport of drugparticles in blood vessels with the aim to optimize the druginjection rate while reducing its side effects [30]. For healthmonitoring, MC could coordinate the movement of intra-body nanoscale sensors to collect health data, which couldbe further transmitted to external devices via micro-to-macrointerfaces for real-time monitoring [19], [31]. Additional MCapplications are identified in the surveys [32], [33]. To leadtowards successful implementation of MC for the aforemen-tioned applications, both synthetic biology and microfluidicshave been regarded as promising tools for the design, test,and manufacture of microscale MC systems. Synthetic biologyoffers tools to engineer MC transceivers with modulation andcoding functionalities via genetic circuits [34], [35]. Microflu-idics provides microscale experimental platforms to flexiblymanipulate and control molecular transport to realize MCfunctionalities with high performance and reagent economy[18], [36].Clearly, MC research incorporates elements from differentscientific communities and this survey seeks to bring themcloser together. MC theory can provide valuable insights forboth man-made and natural systems. However, life scientistsand engineers tend to have quite different ways of thinkingand employ different language. In biology, the typical way toconsider a natural (or synthetic) system is to describe its partsin appropriate detail and how these parts integrate to providea functional system from start to finish. A representativeexample of this approach is cell signaling with G-protein-coupled receptors (which we describe in further detail inSection IV-B); this system is conventionally described as achain of events from receptor activation to the subsequentcell response that is experimentally observed [37]. Anotherexample is the Wnt signaling pathway, which is a highly-conserved system present in all animals that regulates manyimportant processes like cell proliferation, differentiation, andcell survival [38]. Wnt signaling is typically depicted as aseries of molecule-release incidents that can trigger a responsein neighboring cells and determine their fate. The trend inbiology to detail functional components is consistent withthe discipline’s focus on understanding biological mechanisms(especially when a number of components are unknown) andcontrolling biological systems to maximize production yield(e.g., from a bioreactor).On the other hand, it is common in engineering to designsystems with a more modular approach; the different sub-systems and their theoretical limits are studied and tested inisolation. For example, in a synthetic communication system,the physical channel through which information propagatesis distinct from the encoding and decoding techniques thatare applied, and these can be described separately, althoughthey must be combined to implement a functioning system.This way of thinking is evident in [39] where different levelsfor MC systems are described as part of a hierarchy. Despitethe progress that has been made within the MC researchcommunity, translation of results to enable the desired inter-disciplinary applications has been limited, in part due to themismatch of different perspectives and the distinct methodsthat each community uses to formulate research problems. Biological systems tend to be difficult to study, both dueto their complexity and also because technology is not alwayssufficient, so parts of a natural system might remain unknownuntil technological developments enable observation. Thus,there is a tendency in life sciences to sequester complicatedsystems into small manageable parts, with the risk of losinghigher-level interactions. Quorum sensing is an informativeexample, where individual microbes were being studied fordecades, but only recently came the realization that commu-nication via quorum sensing between microorganisms is offundamental importance for the coordinated behaviors that weobserve (e.g., biofilms, virulence) [40]. Systems biology triesto enforce a more holistic view of natural systems and toexploit concepts in biology that originated in other disciplines,including engineering [41]. The tools commonly employedinclude big data and network motif analysis, i.e., the studyof individual biological systems in engineering terms (e.g.,biological circuits as logic gates [42]). However, the systemsbiology approach is still decidedly biologically-focused, andthere can be a benefit to studying systems biology problemsfrom a more structured engineering perspective. Inspired bysystems biology, we wish to highlight the evident relevanceof communication theory to signaling in microscale biologicalsystems. A holistic view of natural and synthetic biologicalsystems from a modular communication-centred perspectiveis a missing link that would help bridge contributions in MCto biological applications.
A. State-of-the-Art in Communication Hierarchies
The notion of a communication hierarchy is commonlyfound in the design of communication networks. By stan-dardizing the role of each layer, the layers can be designedin isolation without compromising the functionality of thesystem. We seek such a communication hierarchy for signal-ing in cell biology (whether natural or engineered), not tofacilitate communication network design or to map existinglayering approaches to cell biology, but rather to enable aninterdisciplinary understanding of natural and synthetic MCsystems. Nevertheless, existing approaches provide a usefulreference against which we can compare our approach.The formal communication standard within the MC commu-nity is IEEE 1906.1, “Recommended Practice for Nanoscaleand Molecular Communication Framework”; see [43]. It in-cludes a definition for a nanoscale communication system thatmaps to the basic communication elements (i.e., transmitter,receiver, medium, message carrier, and message). However,it does not specify a particular protocol for communication.Interestingly, it explicitly excludes purely natural systems byspecifying that at least one component must be synthetic. An-other standardization effort is the Molecular CommunicationsMarkup Language (MolComML), which specifies the essentialcomponents of MC systems for making different simulationplatforms more comparable and cross-compatible [44].For telecommunication systems, the Open Systems Inter-connection model (OSI) [45] and Transmission Control Pro-tocol/Internet Protocol (TCP/IP) [46] are popular frameworks,and both were designed for interoperability in heterogeneous digital communication networks. Entities that are at the samelayer in different communicating devices interact via a pro-tocol designed for that layer. Tasks managed by the layersinclude interaction with programs that need to communicate,establishing connections between devices, transmission errorcontrol, and the physical transmission of bits over the commu-nication channel. Incidentally, both of these frameworks havealready been considered for the design of MC systems.Works proposing protocols for MC systems include [39],[47]. In [39], the authors present a layered architecture that isinspired by both OSI and TCP/IP. Preliminary descriptionsof this architecture were presented in [2], [5]. The layerscomprise an application layer, a molecular transport layer, amolecular network layer, a molecular link layer, and a physicallayer (comprised of signaling and bio-nanomachine sublayers).These layers map to those in TCP/IP, and facilitate the designof synthetic (though bio-inspired) communication systems.Similarly, the authors of [47] also present an architecture basedon TCP/IP, with the goal of operating a synthetic communica-tion system over a range of tens of microns ( µ m). In particular,the protocol specifies how to establish connections betweendevices, how to reliably transfer data, and how the receivercan control the transmission rate. The protocols in [39], [47]are designed to establish digital communications and assumethat the system designer has full control over the specificationof the communicating devices.A roadmap for the development of synthetic biologicalMC systems is proposed in [28]. It involves five stages andillustrates the steps to facilitate MC-enabled commercial ap-plications. The main purpose of the roadmap is to help definethe scope of [28] (i.e., transmitter and receiver building blocksfor different signaling molecules) as opposed to establishinga communication hierarchy. B. Contribution Summary
The intended audience for this survey are those whoare interested in how communications engineering conceptsemerge and apply to understanding and controlling signalingin cell biology. This includes members of the communicationengineering community who may not be familiar with MCor cell biology, and researchers in synthetic biology and bio-engineering who may not be familiar with communicationsystems and networks. Ultimately, this survey is written tobuild and support a bridge between these distinct domainsby linking them together and identifying opportunities forinterdisciplinary collaboration.To facilitate our objectives, this survey introduces a com-munication level hierarchy for microscale biological systems.Our perspective in the design of a communication levelhierarchy for signaling in cell biology is primarily not tocreate a protocol and build synthetic communication systems.Instead, we seek to map our hierarchy and communicationsconcepts directly to biological behavior. Thus, our approachwill help engineers and biologists understand communicationand signal processing (including computation and control) incell biology, while providing language for synthetic biologyand new opportunities to interface with biological systems.
1. Physical Signal Propagation(Section III)2. Physical and ChemicalSignal Interaction (Section IV)3. Signal-Data Interface(Section V)4. Local Data Abstraction(Section VI)5. Application(Section VII)
Fig. 1: The proposed communication level hierarchy for MC signalingin cell biology that is also used to organize this paper.
Furthermore, to emphasize that we are not designing a formalcommunication technology protocol, we refer to the tiers inour proposed hierarchy as “ levels ” instead of “layers”. Thelevels, also summarized in Fig. 1, are: 1)
Physical SignalPropagation ; 2)
Physical and Chemical Signal Interaction ;3)
Signal-Data Interface ; 4)
Local Data Abstraction ; and 5)
Application . Within the context of this hierarchy, this surveymakes the following contributions:1) We map the communication processes of cell biologysignaling to the levels of the proposed hierarchy. Thisincludes macroscale interactions (i.e., experimental ob-servation and control) with the microscale biologicalsystems.2) We provide biological case studies on quorum sensing,neuronal signaling, and communication via DNA, thatmap to all levels of the proposed hierarchy.3) We link contributions in the MC engineering domainwith applications in biology and synthetic biology. Thisenables us to identify many opportunities for interdis-ciplinary contributions that advance understanding andcontrol of signaling in cell biology.There are several challenges faced by this survey to ef-fectively serve an audience with these diverse backgrounds,i.e., communications engineering and cell biology. First andforemost is the disparity in background knowledge. Althoughmuch of this survey describes cell biology and cellular sub-systems, we do not expect a reader with a communicationsbackground to be familiar with these topics. Thus, we havesought for the cell biology in this survey to be self-contained,and we frequently cite [48] as a model background reference.We have also sought for the communications theory in thissurvey to be self-contained. However, a reader with a biologybackground and no foundation in communication systems mayfind it helpful to refer to a fundamental text in wirelesscommunications (such as [49]). Additional background on thecorresponding mathematics and signal processing can be foundin [50]. Furthermore, we include glossaries of biological and communications terms in Tables VII and VIII, respectively, inthe Appendix.The second challenge for this survey is one of language.Different domains have distinct ways of articulating researchproblems and presenting results. This makes it difficult torecognize when research groups from different fields areseeking answers to the same question, or when an answer hasalready been obtained but from a different perspective. Wehope that this survey and its proposed hierarchy is a usefulguide for expanding a reader’s language for interpreting resultsfrom both communications engineering and bio-engineering.The third challenge for this survey is one of research focus.As we have already established, contributions in MC from thecommunication engineering community have focused on thedesign of new communication systems, whereas the subjectof biology is concerned with understanding existing systems.Nevertheless, we intend for the hierarchy in this survey toidentify ample common ground.
C. Survey Organization
The remainder of this survey is organized as follows. Weprovide an overview of the proposed communication levelhierarchy for cell biology in Section II. We also present ageneral definition of a communicating device so that ourdiscussions of communication between devices are in theappropriate context.Sections III to VII sequentially detail the communicationlevels in a bottom-up approach, such that we traverse the levelsin Fig. 1 from Level 1 to Level 5. A graphical summary ofthe content of these levels is provided in Fig. 2. We start inSection III with the underlying fundamental concepts for thephysical propagation of molecules (Level 1). We summarizethe mathematical modeling of diffusion-based propagation butwe also detail other mechanisms including cargo transport andcontact-based signaling.In Section IV, we discuss the biochemical and biophysicalprocesses for devices to generate and receive signals (Level2), including signaling pathways and physical responses. Thissection discusses initial and boundary conditions that areassociated with molecule propagation. It also considers thebiochemical signaling pathways associated with gene expres-sion and corresponding transcription networks.Section V addresses the mathematical quantification ofphysical signals and how they are observed and controlled(Level 3). We discuss microscale signal operations in terms ofgene regulation and metabolic control, then review experimen-tal methods for observing and controlling microscale molec-ular signals. We complete the section by mapping quantifiedphysical signals to information bits.Section VI focuses on the meaning of information in cellbiology signaling (Level 4). We discuss the information con-tained in cellular signals, the realization of signal processingunits with chemical reactions and synthetic biology, and thedevelopment of synthetic communication functionalities. Wealso describe DNA as a potential mechanism for microscalestorage. In Section VII, we reach the top of the hierarchy and discussapplications (Level 5). We focus on biosensing and therapeu-tics as exemplary applications relying on the integration ofnatural and engineered cell biology systems.Section VIII presents three end-to-end case studies that spanall of the levels of the proposed hierarchy. We discuss bacterialquorum sensing, neuronal signaling, and communication viaDNA. We map each case study to all of the communicationlevels, thus demonstrating the flexibility of our approach.Section IX provides a selection of open interdisciplinaryresearch problems that can be identified and formulated usingthe proposed hierarchy. This includes problems that map toparticular levels or the integration of multiple levels, as wellas questions that are formulated by applying the hierarchy tothe end-to-end case studies.We conclude our survey in Section X. We re-iterate theintent of the hierarchy as a bridge to understanding andcontrolling communication in cell biology. We also emphasizekey open research problems that are introduced throughoutthe work and for which we hope our framework will help todevelop solutions.
D. Comparison with Existing MC Surveys
In the past 12 years, there have been several surveys focusedon MC. To differentiate the scope of our survey with thoseof other surveys, in Table I we compare and summarize thedifferences of existing surveys according to the structure ofour survey. This format also emphasizes our perspective andnew contributions.II. C
OMMUNICATION H IERARCHY FOR S IGNALING IN C ELL B IOLOGY
In this section, we elaborate on our broad definition ofcommunicating devices as we understand them in the contextof cell biology. We then present our proposed communicationhierarchy for signaling in cell biology. We briefly discusseach of the levels and compare them with the layers inexisting communication protocols. The hierarchy is then usedthroughout the rest of this survey to articulate the differentstages of a communication process between devices.
A. Defining Communication Devices
A minimum requirement for any communication system isthat there must be communicating devices . A device can actas a transmitter if it is a source of information and has amechanism to translate that information into a physical signalfor other devices to detect. A device can act as a receiver ifit needs to detect a signal from a transmitter and recover theembedded information. Of course, a single device can behaveas both a transmitter and a receiver, i.e., as a transceiver .Throughout this survey, we use the term “device” instead of“transceiver” to emphasize that the communicating devices canhave diverse functions in addition to communication, and whenrelevant these functions are integrated within our proposedhierarchy. As we will see, one key difference between a fully-engineered communication system and a living cell system
Section IIILevel 1 - Physical Signal PropagationAdvection ReactionDiffusionCargo ContactMolecularMotors Chemotaxis PlasmodesmataGap Junctions Appendage-BasedCommunicationSection IVLevel 2 - Physical and Chemical Signal InteractionReceptionandResponsesMoleculeGenerationandReleaseManagement MathematicalModeling TranscriptionNetworkMoleculeReception ReceptionResponses InitialConditionsof Release ChannelBoundaries ReceptionBoundaries GeneExpression NetworkMotiffsSection VLevel 3 - Physical/Data InterfaceMicroscaleObservationsofMicroscalePhenomenaMicroscaleSignalOperations MicroscaleControlofMicroscaleChange Conversionfrom Signalsto BitsGeneRegulation MetabolicControlOpticalMicroscopy Optofluidics
In Vivo
Imaging MagneticNanoparticle THzCommunication pH-MeasuringInstrumentsChemical Electric Optical Temperature Mechanical MagneticMicroscaleMod-Demod MacroscaleMod-DemodSection VILevel 4 - Local Data AbstractionDigitaland AnalogCircuits InformationinCellular Signals RealizingCommunicationsFunctionalities MicroscaleStorageChemicalReactions SyntheticBiology Modulation-DemodulationFunctionalities Coding-DecodingFunctionalitiesSection VIILevel 5 - ApplicationsTherapeuticsBiosensing MC-AssistedApplicationsMicrofluidics SyntheticBiology Magnetic FieldSection VIIIEnd-to-End Case StudiesNeuronalCommunication Quorum Sensing Communicationvia DNASection Subsection Subsubsection
Fig. 2: Organization and content of Sections III-VIII of this survey. Sections are shown in blue, subsections in orange, and subsubsectionsin beige.
TABLE I: Comparison of MC Surveys.
Reference [1] [5] [51] [39] [25] [52] [29] [53] [32] [54] [24] [26] [27] [31] [55] [56] [57] [28] This paperYear 2008 2012 2013 2014 2016 2016 2016 2017 2017 2018 2019 2019 2019 2019 2019 2019 2019 2020 -Section III:Level 1PhysicalSignalPropagation Diffusion-BasedPropagation (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Advection-Diffusion-Based Propagation (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Advection-Diffusion-Reaction-BasedPropagation (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Cargo-BasedPropagation (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Contact-BasedPropagation (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Section IV:Level 2DeviceInterface Molecule Generationand Release Management (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Molecule Receptionand Responses (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Mathematical Modelingof Emission, Propagation,and Reception (cid:88) (cid:88) (cid:88) (cid:88)
BiochemicalSignaling Pathway:Transcription Network (cid:88)
Section V:Level 3Physical/DataInterface Microscale SignalOperations (cid:88) (cid:88) (cid:88) (cid:88)
Macroscale Observationsof Microscale Phenomena (cid:88) (cid:88) (cid:88) (cid:88)
Macroscale Controlof Microscale Change (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Conversion fromSignals to Bits (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Section VI:Level 4Local Data Information inCellular Signals (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88)
Digital andAnalog Circuits (cid:88) (cid:88)
RealizingCommunicationFunctionalities (cid:88)
MicroscaleStorage (cid:88) (cid:88)
Section VII:Level 5Application Biosensing (cid:88)
Therapeutics (cid:88) (cid:88) (cid:88) (cid:88)
MC-Assisted Application (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) is that the transfer of information is explicitly present in thedesign of the engineered communication system, whereas itmay only be implicitly represented in the biological system.For example, cells in a bacterial colony do not always directlycommunicate in the sense of information transfer betweenone specific cell and another specific intended target cell.Nevertheless, the release and detection of cellular signals,and the corresponding responses to those signals, indicate thatcommunication is in fact taking place.Due to the unstructured yet complex nature of biologicalcommunication, there is some flexibility in how we define acommunicating device. In this survey, we take a very generalapproach so that we can identify devices as appropriate in therespective context. Thus, examples of devices include:1) Organelles and large macromolecules that engage inintracellular signaling. Depending on their functionalityand mobility, macromolecules could also represent thesignaling molecules between other devices.2) Individual cells in intercellular signaling networks. Again,individual cells could also represent the signal or its trans-port mechanism, depending on the context. For example,synthetic bacteria that move via chemotaxis have beenproposed to carry information via plasmids in an artificialnanoscale network [58]. Another example can be foundin the animal immune system, where sub-populations ofT-cells can act as intermediates for information aboutpathogens [59].3) A colony of cells, including tissues and multi-cellularorganisms, as an aggregation of individual cells in inter-population or inter-species communication. 4) Experimental equipment or other synthetic means tointroduce a signal or observe signals in a cell biologyenvironment.An important detail is that any of these devices could benatural or synthetic (e.g., genetically-modified cells, syntheticmacromolecules, microscale robots, etc.). Even though somedevices listed above can be much larger than individualcells (e.g., multi-cellular organisms, cellular tissues), we stillinclude them within our framework if they have identifiablesignaling with microscale devices (including individual cellsthat are part of the larger device). Nevertheless, we oftenfind it most convenient throughout this survey to consider acommunicating device to be an individual cell.
B. Our Proposed Hierarchy
We summarize our proposed hierarchy in Fig. 1 and applyit to a sample cellular signaling scenario in Fig. 3. The levelsof the hierarchy are as follows:1)
Physical Signal Propagation – how molecules are trans-ported between communicating devices, e.g., via diffu-sion, fluid flow, or contact-based means. This level is not defined within devices themselves, but directly connectsdevices that are communicating; e.g., the two cells inFig. 3.2)
Physical and Chemical Signal Interaction – how thephysical signal is generated at a transmitting device andsampled at a receiving device, e.g., the stimulation ofgeneration, release, and binding of the molecules. This
Level 5: ApplicationLevel 3: Signal-DataInterface Level 1: SignalPropagation0100Level 4: Local Data Abstraction0100Action A vs B Level 2Level 2: SignalInteraction
Cell A Cell BExtracellularSpace
Fig. 3: The proposed communication level hierarchy applied to anexample cellular signaling environment. The communicating devicesare two cells separated by extracellular space. In this scenario, Level 1describes the signal propagation across the extracellular space. Levels2-4 describe the biochemical pathways, signal-data interface, and dataabstraction within the individual cells, respectively. Level 5 describesthe overall behavior requiring communication between the two cells. also includes the biochemical signaling pathways thatprocess molecular signals.3)
Signal-Data Interface – how physical signals are math-ematically quantified, observed, and controlled. This in-cludes the conversion of data between its mathematicalrepresentation and its physical form, i.e., modulation anddemodulation in communication networks.4)
Local Data Abstraction – the meaning of quantified dataat a local device, e.g., timing information, quantization ofconcentration, or the genetic information in a strand ofDNA. This includes the information theoretic limits ofmolecular signals. We classify the field of bioinformaticsto be primarily at this level [60]. The level also includesencoding and decoding in communication networks.5)
Application – the top-level behavior that is relying oncommunication. This could be entirely within a biologicalcontext, e.g., differentiation of cells in a multi-cellular or-ganism or symbiosis between different species, or withina mixed synthetic and biological context, e.g., diseasedetection by medical sensors.We can compare our hierarchy with the TCP/IP commu-nication protocol, which has similar concepts but severalkey differences. For clarity, we make a direct comparisonwith the TCP-based protocol proposed in [39]. In [39], theapplication layer is an interface for applications to access thecommunications functionality; in our hierarchy, the applicationlevel is the application itself. In [39], the transport, network,and link layers provide operations that are critical to theoperation of a synthetic communication network; our focusis on biological systems, which in general do not imple-ment these specific operations, and so they are not directlyrepresented in our approach. In particular, we focus on themathematical representation of the information (through thelocal data abstraction level) rather than the management ofthat information (as provided by the aforementioned layersand which is also the focus of [47]). In [39], the signaling sublayer includes modulation anddemodulation, transmission and sampling, and signal propaga-tion. We separate these three critical tasks into distinct levels,i.e., Levels 1–3, because of the diversity of implementationsat each of these levels and their interoperability for differentbiological and engineering applications. Finally, [39] has abio-nanomachine sublayer to define routine operations for theongoing maintenance of a device; we do not focus on suchbehavior so we do not define a corresponding level, thoughwe do integrate life-preserving tasks within the hierarchy(e.g., gene expression in Level 2; gene regulation and cellmetabolism in Level 3).Important advantages of our approach are its flexibilityand scalability, which are both useful for the study of andinteraction with propagating signals in cell biology. If wetried to map a natural communication system to a structuredcommunication protocol, then we would observe many in-stances where components are missing, e.g., the network layeris defined for packet management in [39] but such a functionis not evident in cell-based diffusive signaling. This is in partbecause the quantity of information that is communicated inbiological systems is often relatively small (on the order ofa few bits; e.g., see [61]) when compared with the objectivesof synthetic MC systems. Important exceptions are DNA (asthere are millions of base pairs – 2 bits each – in the genomesof most organisms; see [48, Fig. 1.32]) and the aggregateinformation stored in the brain (as the human brain has about neurons and synaptic connections between them [48,Ch. 11]). Our approach enables us to be more holistic andflexible in our mapping by identifying multiple and potentiallyoverlapping hierarchies within a single environment withoutconcern for how the hierarchies would all map and adhereto a single uniform communication protocol. Thus, we cansimultaneously characterize internal communication within acell, signaling between cells, and experimental observation ofindividual cells or an entire population, which is not practicalwith a formal communication protocol. There can also beasymmetry in the communicating devices, and some may noteven exhibit all of the levels, so long as they are joined bysome physical propagation channel. For example, chains ofmotor neurons propagate action potentials that convey thesignal for muscle contraction, but the neurons do not need tounderstand the information that is being relayed. Additionally,the newly-discovered telocytes can act as intermediates for theflow of information between different types of cells [62], [63].III. L EVEL
HYSICAL S IGNAL P ROPAGATION
A fundamental characteristic of any communication networkis how information propagates between the devices. In thissection, we survey the means by which MC signals physicallypropagate. These correspond to Level 1 of our proposedcommunication hierarchy (see Figs. 1 and 3). We focus ondiffusion-based phenomena (Sections III-A to III-C) becausethey are prevalent at the microscale and they have receivedsignificant attention within the MC engineering community[24]. We include mathematical descriptions for diffusion,which integrate with the mathematical characterizations of
TABLE II: Summary of Physical Signal Propagation . Propagation Mechanism Example Speed ReferenceDiffusion-Based Propagation Calcium ions Diffusion coefficient: . × − m /s [64]Pheromones Diffusion coefficient: . × − m /s [65] lac repressor protein Diffusion coefficient: − − × − m /s [66]Advection-Diffusion-Based Propagation Human skin capillary Mean velocity: . − − . × − m/s [67]Advection-Diffusion-Reaction-Based Propagation Antibody-antigen interaction Diffusion coefficient: ≈ − m /s [68]Cargo-Based Propagation Kinesin propagation Mean velocity: ≈ . × − m/s [69]Contact-Based Communication Gap junctions Mean velocity: ≈ × − m/s [70] initial and boundary conditions (for Level 2 in Section IV)to determine the corresponding channel response. In addition,we describe cargo-based transport with motors and chemotaxis(Section III-D), as well as contact-based propagation includinggap junctions and plasmodesmata (Section III-E). A summaryof the propagation mechanisms that we discuss, includingrepresentative molecules for each mechanism, is also providedin Table II.It is helpful to have a sense of the scope in diversity ofmolecules and mechanisms that are used in cell signaling.In biological systems, regardless of physical scale but espe-cially in plants and animals, there is a tremendous varietyof molecules with distinct characteristics (e.g., size, shape,electrical charge) that act as messengers between cells orwhole organisms, in addition to the biochemical machinery tosupport them. For example, extracellular vesicles have beenfound to carry more than 19 distinct proteins that are believedto be involved in signaling between animal cells [71]. Asearch through the in silico Human Surfaceome database [72] reveals 1201 surface receptor proteins and 137 ligands.Considering that not all molecules or receptors are known,that other communication modalities (e.g., ion channels, gases,contact signaling) are equally or more important depending onthe cell type, and that there is significant cross-talk betweenmodalities, these examples give a glimpse of the scope indiversity and complexity that characterizes communication innatural systems. Nevertheless, many of the molecules propa-gate according to one (or more) of the approaches surveyedin this section. Examples include the diverse cell signalingprocesses using ions (e.g., Ca + , K + ) that convey a differentmessage depending on the type of the recipient cell, microbialquorum sensing where organisms secrete their own variants ofmolecules as communication signals, or pheromones carryingmessages over long distances [73].Furthermore, in contrast to fully-engineered systems wherethere is usually an effort to standardize components andminimize signal variability, natural systems (and those thatare synthetically derived from natural systems) can expressvariation within populations of the same species or cells ofthe same type. This variation, which can include strength ofgene expression or response to external stimuli [74], [75],leads to variability in signal density, duration, and timing.Environmental factors also influence the propagation of any Publicly available at: http://wlab.ethz.ch/surfaceome/. molecular signal (e.g., temperature, pH, presence of flow).Last but not least is the fact that competition for resources orbetween predators and prey has led to an arms race betweendifferent organisms where survival depends on the successfuldetection of or interference with each other’s signals (e.g.,“eavesdropping” in bacteria [76]; discussed in Section VIII-A).With signal interference and also the inherent stochasticity ofmolecule release and propagation, cell signaling is inherentlyvery noisy and these features must be taken into considerationin order to understand a natural system’s reliability or to designa synthetic MC-based system.
A. Diffusion-Based Propagation
Diffusion refers to the random walk, namely Brownianmotion, of molecules in a medium arising from the molecules’thermal energy [77]. It is a simple and efficient movementparadigm without a need for infrastructure or external energysources. Therefore, there are many examples found in nature,including calcium signaling among cells [78], pheromonalcommunication among animals [65], and propagation of DNAbinding molecules over a DNA segment [79].The mathematics of Brownian motion are often modeledusing Fick’s laws of diffusion. As a conceptual example, wefind it useful to describe Fick’s first law of diffusion from firstprinciples using the macroscopic approach presented in [79,Ch. 2]. We consider the simplified case shown in Fig. 4, wheremolecules move one step at a time along only one axis witha displacement step ∆ x and a time step ∆ T . We assume thateach molecule walks independently and the probabilities ofmoving forward and backward are both / . Let N ( x ) denotethe number of molecules at position x and time t . During thetime interval [ t , t + ∆ T ] , we expect that half of the moleculesat x will move to x + ∆ x and traverse the normal face thatis orthogonal to the axis and located at ( x + ∆ x )/ . At thesame time, we expect that half of the molecules at x + ∆ x willcross the face in the opposite direction. Hence, the net expected number of molecules coming to x + ∆ x will be [ N ( x )− N ( x + ∆ x )] . Dividing by the face area S and time step ∆ T , the netflux J Diff crossing the face by diffusion is J Diff (cid:12)(cid:12) = − ∆ T [ N ( x + ∆ x ) − N ( x )] S . (1) x x x Face area S ( ) N x x ( )
N x
Fig. 4: A macroscopic one dimensional (1D) random walk modelwhere the small circles represent molecules and move along the x -axis. The dotted red square is in the plane that is orthogonal to the x -axis. If we further multiply the right-hand side of (1) by ∆ x / ∆ x ,then it becomes J Diff (cid:12)(cid:12) = − ∆ x ∆ x ∆ T [ N ( x + ∆ x ) − N ( x )] S ∆ x = − ∆ x ∆ x ∆ T [ C ( x + ∆ x ) − C ( x )] , (2)where C ( x + ∆ x ) = N ( x + ∆ x )/( S ∆ x ) and C ( x ) = N ( x )/( S ∆ x ) are the molecular concentrations at locations x + ∆ x and x ,respectively. By considering ∆ x → and defining the diffusioncoefficient D = ∆ x /( ∆ T ) , we arrive at Fick’s first law in 1Dsapce [79, eq. (2.1)], i.e., J Diff (cid:12)(cid:12) = − D ∂ C ( x , t ) ∂ x . (3)Correspondingly, Fick’s first law in three dimensional (3D)space is J Diff (cid:12)(cid:12) = − D ∂ C ( d , t ) ∂ x , (4)where vector d = [ x , y , z ] specifies the molecule position.Fick’s first law describes the relationship between thediffusion flux and the concentration gradient. The value ofthe diffusion coefficient D determines how fast a certaintype of molecule moves. In general, D is dependent uponthe environment (e.g., temperature, viscosity) as well as themolecule size and shape. For example, in a given environment,smaller molecules tend to diffuse faster. However, even when amolecule’s diffusion coefficient is on the order of µ m /s(a relatively large value), it is estimated that it would takenearly half a millisecond for such a molecule to travel over µ m (the width of a bacterium) [79], demonstrating thatdiffusion alone is quite a slow process.The impact of diffusion on concentration change withrespect to time can be described by Fick’s second law as [79,eq. (2.5)] ∂ C ( d , t ) ∂ t = D ∇ C ( d , t ) , (5)where ∇ is the Laplace operator. Solutions to (5) can beobtained under different initial and boundary conditions, de-pending on the diffusion environment. We discuss examplesof initial and boundary conditions in greater detail in Sec-tion IV-C. B. Advection-Diffusion-Based Propagation
The diffusion process can be accelerated by introducingadditional phenomena. In particular, molecule transport can beassisted by two physical mechanisms: 1) force-induced drift,and 2) advection, i.e., bulk flow. Force-induced drift is causedby applying an external force directly to the particles ratherthan the fluid containing them. Examples include applying amagnetic field to magnetic nanoparticles, an electrical field tocharged particles, and a gravitational force to particles withsufficient mass [24]. Advection refers to molecule transportassisted by bulk movement of the entire fluid, including themolecules of interest. Examples include endocrine signalingin blood vessels and the manipulation of fluids in microfluidicchannels (we elaborate on applications using microfluidics inSection VII). Here, we focus on advection and in the fol-lowing, we present a mathematical framework to approximatemolecular transport assisted by advection.Analogous to diffusion, the advection process also results ina flux of concentration crossing the surface of a given region. Ithas been shown that the concentration flux caused by advectionis simply a concentration shift over time; thus the flux J Adv with local velocity u can be described by [80] J Adv = u C . (6)The temporal change in concentration is jointly determinedby the diffusion flux and the advection flux, and can beexpressed as [81, eq. (4.3)] ∂∂ t ∫ V CdV = − ∫ S ( J Diff + J Adv ) · n dS , (7)where V is the volume of a given region with differentialelement dV , S is the surface of the volume with differentialelement dS , and n is a unit outward normal vector. Substituting(4) and (6) into (7), and applying the divergence theorem, weobtain the advection-diffusion equation in differential form as[82] ∂ C ( d , t ) ∂ t = D ∇ C ( d , t ) − u · ∇ C ( d , t ) , (8)where ∇ is the Nabla operator.It is clear from (8) that the flow properties, such as thevelocity u , have an impact on the distribution of the moleculeconcentration. Several dimensionless numbers have been de-fined to classify and characterize the transport behavior offluids. Two important examples are the Reynolds number andP´eclet number, which we describe in the following.
1) Reynolds number:
The Reynolds number (Re) is definedas [82, eq. (2.39)] Re = ρ u eff d µ , (9)where ρ is the flow density, u eff is the mean velocity, and µ is the fluid dynamic viscosity. d is the characteristic lengthscale, and for flows in a pipe or tube, it becomes the hydraulicdiameter of the pipe [7, eq. (3)]. The Reynolds numberdetermines whether the flow is in the laminar regime orthe turbulent regime. In the laminar regime, the Reynoldsnumber is normally less than , and regular streamlineflow patterns can be experimentally observed. In contrast, in the turbulent regime, the Reynolds number is larger than , and a single stable flow pattern cannot be observedin practice. For microscale channels (whether synthetic orblood vessels), the Reynolds number is frequently less than ; thus, laminar flows are often assumed [83]. For example,it has been demonstrated that most blood vessels (except theaorta with Re ∈ [ , ] ) are laminar [84], [85]. Based onthis, the authors of [30] derived a time-varying drug deliveryconcentration profile based on the advection-diffusion equationin (8). This work provided an initial understanding of drugpropagation and laid the foundation to establish advancedtherapeutic methods.A typical example for laminar flow is Poiseuille flow,where a pressure drop exists between the inlet and outlet ofa microfluidic channel. If the flow only moves along the x direction, and if the channel cross-section is circular, then thevelocity distribution u x ( r ) can be expressed as [82, eq. (3.42a)] u x ( r ) = ∆ P µ L ( R − r ) , (10)where ∆ P is the pressure drop, L is the channel length, R is theradius of the cross-section, and r is the radial location. Eq. (10)indicates that the velocity follows a parabolic distribution, suchthat the flow velocity increases from the boundary towardsthe center of the channel. The velocity distributions for othercross-section shapes can be found in [82, Ch. 3].
2) P´eclet number:
The P´eclet number (Pe) compares therelative dominance of advection versus diffusion and is definedas [82, eq. (5.53)] Pe = u eff LD . (11)For Pe = , the molecular movement is purely diffusive; forPe → ∞ , the movement becomes a pure bulk flow process.The P´eclet number is useful to predict the moleculardistribution under Taylor dispersion, which describes howaxial advection and radial diffusion jointly affect moleculartransport in pressure-driven bulk flow [82]. Specifically, asshown in Fig. 5, a homogeneous band of solute is injected at x = to travel through a cylindrical microchannel with radius R . A very short time after injection, the solute molecules arestretched into a parabolic plug by the flow having the velocityprofile in (10). Subsequently, two concentration gradients areestablished at the front and back ends of the solute plug.Due to these gradients, there is a net migration of solutemolecules at the front end from the high concentration area(i.e., the channel center) to the low concentration area (i.e., thechannel boundary). On the contrary, there is a net migrationof molecules at the back end from the channel boundaryto the area around the channel center. We use R /( D ) to characterize the expected diffusion time along the radialdirection, and use L / u eff to represent the time of moleculetransport at average fluid velocity u eff over distance L . If R /( D ) (cid:29) L / u eff (i.e., Pe (cid:29) L / R ), then the cross-sectional diffusion cannot be ignored and fully counteractsthe parabolic plug, which leads to a uniform distribution ofthe solute over the cross-section [86], [87]. Thus, in this case,the 1D advection-diffusion equation with a modified diffusion (a)(b1) (b2) (b3)(b) Fig. 5: The schematic of Taylor dispersion in Poiseuille flow. (a)In a microfluidic channel, the velocity increases from the boundariesinwards, following the parabolic distribution in (10) ( R : cross-sectionradius, r : radial location, x : direction of flow). (b) Taylor dispersionprogression inside a microfluidic channel: (b1) A homogeneoussolute band is injected into the channel. (b2) After injection, thesolute band is stretched into a parabolic plug due to the parabolicvelocity distribution. Then, the concentration gradients established atthe front and back ends, cause the net motion of solute moleculesto counteract the parabolic plug. (b3) Finally, the molecules areuniformly distributed over the cross-section. coefficient can be used to approximate the 3D Poiseuille flow[88]. C. Advection-Diffusion-Reaction-Based Propagation
In addition to the diffusion and advection processes, chem-ical reactions often occur simultaneously during molecularmovement. Examples include the polymerase chain reaction(for synthetically copying DNA [89]) and surface capture [68],[90]. To analyze molecular transport under chemical reactions,we consider the example of a second-order (bimolecular)reaction S i + S j k f → S l , where species S i reacts with species S j to generate product S l under the rate constant k f . Ifmolecular transport is subjected to diffusion and reaction, thenthe concentration changes of the reactant S i (analogously S j )and the product S l can be expressed as ∂ C S i ( d , t ) ∂ t = D ∇ C S i ( d , t ) − k f C S i ( d , t ) C S j ( d , t ) , (12a) ∂ C S l ( d , t ) ∂ t = D ∇ C S l ( d , t ) + k f C S i ( d , t ) C S j ( d , t ) . (12b)A general diffusion-reaction equation is given by ∂ C S i / l ( d , t ) ∂ t = D ∇ C S i / l + q f [ k f , C S i ( d , t )] , (13)where q = − holds for reactants, q = holds for products,and f [·] is the reaction term which in general can accountfor the presence of multiple reactions. Furthermore, if molec-ular propagation is simultaneously governed by advection, diffusion, and chemical reaction, then the spatial-temporalconcentration distribution can be expressed by the followingadvection-diffusion-reaction equation [24]: ∂ C S i / l ( d , t ) ∂ t = D ∇ C S i / l ( d , t ) − u · ∇ C S i ( d , t ) + q f [ k f , C S i ( d , t )] . (14)With certain initial and boundary conditions, the expectedtime-varying concentration of each type of molecule can bederived. Some analytical solutions of the 1D form of (14) canbe found in [91]. D. Cargo-Based Propagation
In contrast to the free transport of signaling molecules,there are also signal propagation mechanisms that rely on themolecules of interest being transported as cargo within somekind of biological container. These can be classified as active transport, as energy is expended for loading, unloading, andmoving the container. Such propagation mechanisms are usedby nature to overcome the slowness and limited directionalityof diffusion-based signaling, particularly over larger distances.In this subsection, we review cargo-based propagation usingmolecular motors and using bacteria participating in chemo-taxis.
1) Molecular Motors:
A common example in intracellularsignaling, particularly within eukaryotes (including plant andanimal cells) as these cells are generally larger than bacteriaand other prokaryotes, is transport via translational motorsalong the cytoskeleton. The cytoskeleton has different “track”-like components that help to control the shape, movement, anddivision of the cell. The motors are proteins and a given type ofmotor protein can move in a single direction along a particularcytoskeleton element. Motor proteins can be classified into oneof three families: myosin, kinesin, and dynein [92, Ch. 16].Myosin motors travel along actin filaments whereas kinesinand dynein motors move along microtubules (e.g., see Fig. 6).Each type of motor binds to the cytoskeleton element at oneend (i.e., the head) and binds to its cargo or cargo container atthe other end (i.e., the tail). A sequence of energy-consumingreactions hydrolyzes ATP to induce changes in the motorbinding conformation at the head so that the motor movesalong the cytoskeleton.The possibilities for cargo are quite diverse and can in-clude mitochondria (which produce ATP), messenger RNA(intermediate molecules needed to produce proteins), anddifferent types of vesicles (a more comprehensive discussionon transport with vesicles for Level 2 is given in Section IV).A common function is to transport the cargo from where it issynthesized in the cell to where it is used, e.g., a secretoryvesicle could bind to a motor to carry external signalingmolecules from the Golgi apparatus where they are producedto the cell surface where they are released. Even thoughthe cargo can be much larger than the motor, relatively fasttransport speeds are possible; propagation of secretory vesicleshas been measured at speeds of up to µ m / s [93].Despite the infrastructure in place for molecular motors tomove along cytoskeleton elements, the propagation still hasstatistical behavior over short time scales due to the precise CargoMotorMicrotubule
Fig. 6: Representative example of a motor protein carrying cargoalong a microtubule. Motor proteins are used for directed transportwithin a cell. Motor proteins typically have a head (shown in red)that “walks” along the cytoskeleton, a tail (shown in blue) that bindsto its cargo, and a long stalk (shown in green) that joins the head andthe tail. The cargo can be much larger than the motor, e.g., vesiclesand mitochondria. timing of the ATP hydrolysis reactions [92, Ch. 16]. It is evenpossible for a motor to detach, diffuse around, and rebind tothe cytoskeleton [94]. Thus, as with diffusion-based transport,random walk models apply, and the expected behavior can bedescribed analogously to an advection-diffusion channel (asdiscussed in Section III-B).From a communications engineering perspective, molecularmotor-based signaling was one of the first mechanisms pro-posed for synthetic nanoscale communication [95] and alsoconsidered in early experimental work [96]. However, therehas been limited work on characterizing communication ina system where devices are connected by cytoskeleton-likeelements that are traversed by motors [8], [9]. Some attentionhas been given to the design of synthetic systems that reversethe roles of cytoskeleton and motor, such that cytoskeletonelements become the information molecules. In these systems,a surface is covered with motor proteins and these proteinspush microtubules between transmitter and receiver devices[97], [98]. Such designs are envisioned to be more suitablefor implementation in lab-on-a-chip platforms.
2) Chemotaxis:
A cellular mobility mechanism that hasbeen proposed for adoption as a cargo-based transport methodis chemotaxis. The most common example associated withchemotaxis is that of bacteria moving along concentration gra-dients, but more generally chemotaxis refers to any organismmovement in response to a chemical stimulus. In the case ofbacteria, they engage in a series of runs and tumbles to performlocal concentration sensing and bias their motion towards foodsources or away from toxins; further details can be foundin [92, Ch. 19]. While moving, bacteria coming into contactwith each other are able to exchange genetic information. Asa result, there have been proposals to capitalize on this anduse bacteria as cargo-carrying organisms between networkingnodes that need to communicate [99], [100]. Under such aprocess, a receiver node releases molecules to attract cargo-carrying bacteria to run toward it from the correspondingtransmitter, and information (i.e., the cargo) is shared be-tween the bacteria and nodes using plasmid conjugation (wediscuss conjugation further in the context of contact-basedcommunication in Section III-E). The more general concept ofguiding nanomachine motion to target sites using the release of attractant or repellent molecules has been presented in [31],[101], where the nanomachines are envisioned to be carryinga drug payload or perform some other therapeutic tasks.Signal propagation under chemotaxis has multiple compo-nents, i.e., the diffusion of attractant and repellent molecules,the motion of cells (or more generally nanomachines) inresponse to molecule gradients, and the process of conjugationupon contact. The processes of bacteria runs and tumbles canbe mathematically modeled as a biased random walk [102],and such a model was also adopted to describe cargo-carryingbacteria in nanonetwork design [103], [104]. E. Contact-Based Communication
While the diffusion-based and cargo-based transport mech-anisms described thus far take place over the open spaceseparating communicating devices, there are mechanisms thatrely on direct or indirect contact between devices. Suchmechanisms tend to have better reliability as a result. Here, wediscuss direct contact-based signaling mechanisms includingthe use of gap junctions and plasmodesmata. We also discussappendage-based processes, where cells have mechanisms to“reach” toward or into each other to make contact, includingconjugation, tunneling nanotubes, and telocytes.In biological systems, the distinction between propagationmechanisms might not always be clear. For example, a signalpropagating through any number of gap junctions (GJs) orplasmodesmata (PD) (both described below) could be targetinga population of cells that are relatively far away from thesource. One could then argue that a series of GJs is in effectthe communication channel (in contrast to a series of localdiffusion events). As the biophysical constraints of channelregulation across many sites along the signal path can greatlyaffect signal propagation and reception, we describe thesebiological systems and their regulation in moderate detail tosupport future work for channel modeling. This also leads toseveral open research problems discussed in Section IX.
1) Gap Junctions:
Gap junctions are clusters of membranechannels that connect adjacent cells and they are present inevery tissue of any multicellular organism, a hint on theimportance of these communication channels. They are tightlyregulated and permit the diffusion of small molecules directlyfrom the cytoplasm of one cell to that of its neighbor, therebyavoiding the extracellular matrix. For general reviews on gapjunctions, see [105], [106].GJ structure was first visualized in the 1960s using electronmicroscopy [107]. Subsequent experiments revealed the three-dimensional structure of gap junctions in sufficient detail toenable the description of an individual GJ as the combinationof two hemichannels constructed by specific proteins ( con-nexins ), one from each cell, that are connected head-to-head[108], [109]; see Fig. 7. Clusters of gap junctions arranged ina hexagonal lattice connect the plasma membranes of adjacentcells. Each hemichannel (a connexon ) is comprised of sixsubunits that create a cylindrical pore connecting the cells.By creating direct pathways between cells, it is known thatgap junctions play an important role in intercellular communi-cation. They permit exchange of ions, miRNA, and other small
Fig. 7: Gap junction structure. Individual proteins (connexins) forma hemichannel, or connexon. Two connexons from adjacent cells arejoined to create one gap junction. A small gap remains between thecells, giving gap junctions their name. The channel opens selectivelyto allow ions (e.g., calcium ions) and other small molecules to passthrough. molecules such as metabolites and second messengers. Dueto this important function, GJs have to be tightly regulated.Opening and closing of GJs can be achieved via chemical,electrical, and mechanical means. Two different mechanismsare known to be involved in the regulation of GJs’ openingand closing [110], [111]:1) A fast-gating mechanism, where rectification of ioniccurrents passing through a fully-opened channel occursdue to selective permeability. Transitions are fast and atleast three intermediate states between open and closedare known.2) A slow voltage-sensitive mechanism, also termed loop-gating . With this mechanism, transitions between statesoccur in many small steps, resulting in a slower response.Regulation of GJ permeability is achieved by means ofan electrostatic barrier created by Ca + [112]. Calcium ionsbinding to specific side chains on each hemichannel createa positive gradient that inhibits any other positive ion suchas K + from entering the pore. The mechanism, though notyet fully understood, involves the interaction between parts ofthe intracellular domains of GJ and Ca + -bound calmodulin[113]. Connexin subunits in gap junctions can bind Ca + ionsand create a strong positive surface potential. The result isan effective electrostatic barrier that can block the entranceof other positive ions. This allows a rapid response of thegating mechanism, much more so than if large conformationalchanges were needed.Protein phosphorylation (inducing structural changes by theaddition of a phosphate group) is another major mechanism forGJ regulation. It acts at several levels affecting the traffickingof connexins from inside the cell to the plasma membrane, andalso the clustering, localization, and recycling of GJs [114].Individual GJ channels have a fast turnover rate, enabling theadjustment of the communication level between the two cells Fig. 8: Plasmodesmata structure (not to scale). Both simple andcomplex forms are depicted in the diagram (dashed lines A and B,respectively). Adjacent cells have connected cytoplasms and ERs viadesmotubules. Molecules can pass through the cytoplasmic sleeve. Abuildup of calose (in red) closes the channel. The walls of each cellare separated by the central lamela. by modulation of the number of connexons produced by eachcell and thus the surface area of the lattice [115].Despite the crucial role of GJs in cell communication,communications engineering-based analysis of systems usingGJs is still at an early stage. The authors of [116] consideredthe assembly of a synthetic communication network basedon GJs. They demonstrated successful transmission via GJsexpressed in genetically modified HeLa cells by propagatinga calcium wave at about µ m/s. In [117], an information-theoretic model was developed to derive a closed-form ex-pression for the GJ channel capacity where the GJs createand propagate action potentials. The model was applied tocorrelate increases in the incidence of cardiac diseases withdysfunction in communication. GJs were also included in thechannel modeling of calcium propagation in [118].
2) Plasmodesmata:
In plants, structures comparable to gapjunctions are called plasmodesmata (PDs, singular plasmod-esma ). Plant cells are surrounded by a rigid cell wall thatprovides structural rigidity but at the same time constrains thepassage of molecules and hence communication. To overcomethis barrier, PDs serve as channels in the cell wall that connectadjacent cells; see Fig. 8. By connecting virtually every cellwithin a plant, PDs create an avenue that permits the transferof metabolites, nutrients, and signals to the remotest tissue[119], [120].Plasmodesmata are nanoscale structures, so they are notclearly visible using an optical microscope. However, withadvances in electron microscopy it became possible to ob-tain detailed images of their structure [121]. The channel ismembrane-lined and connected tightly with the cell wall. Thismeans that there is a continuum of plasma membrane betweenthe different cells that enables the endoplasmic reticulum(ER) of each cell to connect with that of its neighbor. In the center and along the length of the channel there is astructure resembling a pole and called the desmotubule . This isconnected to the surrounding membrane by spikes. Moleculescan travel through the gap between the desmotubule and theplasma membrane, i.e., the cytoplasmic sleeve . Regulation oftraffic through PDs is currently believed to happen by theadjustment of the width of the cytoplasmic sleeve, and inturn this is due to deposition or removal of the protein calose around the mouth of the PD channel, thus restricting accessas needed [122].
3) Appendage-based communication:
We now briefly dis-cuss several contact-based signaling mechanisms (namelyconjugation, tunneling nanotubes, and telocytes) where cellsconnect with neighbors via appendages. Conjugation is awidespread mechanism of genetic material exchange betweenbacteria, and a key factor of microbial genomic plasticity[123], [124], as it facilitates the transfer of DNA betweencells in close range [125]. Different types of mobile geneticelements (MGE) are responsible for initiating and establishingconjugation [126]. Conjugative plasmids are the most widelyknown and studied MGE, as they are both easily identifiedand ubiquitous across bacterial species. Plasmids are alsoone of the most important factors of pathogenicity and ofthe development of antibiotic resistance in prokaryotes. MGEplasmids are small double-stranded DNA elements separatefrom the rest of the bacterial genome. They contain all thenecessary sequences for coding their own replication andtransfer to other cells [127]. Two distinct mechanisms areknown to be used by plasmids for their transfer. One isparticular to actinobacteria such as species of
Streptomyces andis mediated by a protein related to DNA re-positioning duringcell division [128]. The second mechanism is more complexand involves a single-stranded DNA transfer apparatus that iswidespread among diverse bacterial species [127].Conjugation is typically initiated by the donor cell carryingthe plasmid to be copied. A filamentous hollow structureknown as a pilus extends towards the acceptor cell; see Fig. 9.Upon contact the pilus fuses into the recipient cell’s membraneand is destroyed at both ends simultaneously to bring thetwo cells into close proximity. The conjugative plasmid isthen separated into single strands of DNA, one of which istransported through the open channel into the acceptor cell.At the same time, in both donor and acceptor cells, the singlestrands are converted back into double-stranded DNA. Withthe conclusion of DNA transport the pilus extends from bothends to separate the two cells and is then recycled after thefinal separation [125], [129].Recent evidence suggests that other types of MGE, suchas the integrative and conjugative elements (ICE) also knownas conjugative transposons, might in fact be more abundantand more important for bacterial communication than plasmids[123], [130], [131]. Similar to plasmids, ICE encode all thenecessary machinery for their excision from host DNA, trans-fer to, and integration into the host DNA. Unlike plasmids,however, ICE are incorporated into the host DNA (or anexisting plasmid) [130].There have been several works within the MC engineeringcommunity that proposed conjugation as a propagation mech- (a) (b)(c) (d) Fig. 9: Steps of bacterial conjugation. (a) The donor cell, carryingthe conjugative plasmid (red circle), extends a pilus that finds its wayto the acceptor (i.e., recipient) cell through chemotaxis. (b) After theconnection, the pilus contracts to bring the two cells in close contact.(c) As the plasmid DNA is being replicated (blue) in the donor cell,it is being simultaneously passed through the open channel to theacceptor as a single strand, where it is also immediately convertedback into double-stranded cyclic DNA (dotted red circle). (d) At theend of the process, the cells separate by extending the pilus whichthen breaks up and is recycled. anism for synthetic networks, including [58], [100], [132].These contributions have tended to focus on behavior thatoccurs higher in our proposed hierarchy, i.e., quantifying thetransmission of information using bacterial conjugation, andnot characterizing signal propagation.Tunneling nanotubes (TNTs) are long channels formedbetween cells that can be several micrometers apart. Theyare temporary structures that can dynamically form in a fewminutes and directly connect the cytoplasms of the cellsinvolved [133]; see Fig. 10. While ions and small moleculescan diffuse freely along a TNT, TNTs also facilitate the activetransport of a diverse range of molecules, organelles, andmicro-vesicles, e.g., mitochondria and membrane components[134], [135].Relatively recently, a specialized type of cell called a telocyte was discovered [62]. Telocytes form a network onthe extracellular matrix of all body tissues. This network iscomprised of very long thin channels ( telopodes ) betweentelocyte cells; see Fig. 11. They form connections with theother cells in the surrounding environment and permit inter-cellular communication by diffusive, contact, electrical, andmechanical signaling. Their confirmed and speculated rolesspan a diverse array of processes in animal physiology, in-cluding cell signaling, extracellular vesicle release, mechanicalsupport to surrounding tissues, muscle activity, guidance formigrating cells, tissue homeostasis, and even the transmissionof neuronal signals in cooperation with other specialized cells[62], [63], [136], [137]. Thus, telocytes could play an essentialrole in several signal propagation mechanisms and are hencea very interesting target for future MC research. (a)(b)
Fig. 10: Tunneling nanotube (TNT) structure (not to scale). (a)TNTs connect adjacent cells at relatively long distances. (b) Ionsand small molecules are able to freely diffuse through the channel,while larger molecules and organelles are actively transported via theactin filament.Fig. 11: Telocytes (TC) connect various types of cells via longappendages (telopodia). Telopodia can connect with each other,enabling long distance communication between cells.
IV. L
EVEL
HYSICAL AND C HEMICAL S IGNAL I NTERACTION
Communicating devices that use MC channels require in-terfaces to interact with the channels. A transmitter needs amechanism to insert molecules into the channel, and a receiverneeds a mechanism to observe (i.e., sample) the moleculesthat are in the channel. It is common in the MC literatureto assume that these processes are perfect by making ideal assumptions about the generation and sampling of molecules;see [24]. These typically include the instantaneous creationof a desired number of molecules at a fixed point, and thena perfect counting of the number of molecules that arriveat the receiver (whether or not they are removed from thephysical propagation channel). These assumptions shift thefocus of the analysis to the characterization of the physicalsignal propagation (as discussed for Level 1 in Section III), andcan be accurate if the physical and temporal scales of moleculerelease and molecule sampling are sufficiently small relativeto the physical channel. However, if these constraints are notsatisfied, then the interface to the physical signal has to be anintegral component of the end-to-end channel characterization.The biochemical machinery for generating and receivingcellular signals can be rather complex and serve importantroles in cellular function. For example, ions are commonlyused for signaling and also directly regulate behavior viabiochemical signaling pathways. Ca + ion signals control,among others, muscle contraction, cell division, exocytosis,fertilization, metabolism, neuronal synaptic transmission, cellmovement, and cell death [138]. The propagation of ions,usually via diffusion, is only part of the role that they play.Ions actively interact with the mechanisms that release andreceive them (e.g., ion-activated gap junctions and cell surfacemolecular pumps), so they are an essential part of the signalproduction and reception apparatuses.In this section, we review mechanisms for generating andsampling the physical signal, including intermediate biochem-ical and biophysical processing and pathways where the phys-ical signal is an input or output. These mechanisms corre-spond to Level 2 of our proposed communication hierarchy(see Figs. 1 and 3). We discuss the physical storage andrelease of molecules, in particular via vesicles (Section IV-A).We present the common methods for signaling moleculesto be detected and the diversity of biochemical responses(Section IV-B). We mathematically link Level 1 with Level2 by discussing commonly-considered initial and boundaryconditions for diffusion-based propagation, which describehow molecules are added to and how (if any) molecules areremoved from the environment (Section IV-C). The boundaryconditions are needed to derive the channel response andthus directly constrain analytical solutions if they exist. Wefinish this section by introducing the biochemical influenceof molecular signals on transcription networks, which controlprotein production, and discuss how we can study transcriptionnetworks by separating them into isolated network motifs(Section IV-D).Much of the functional communication complexity forcomputation and control in cell biology pertains to Level2. The behavior in natural organisms that we associate withthe higher levels of the hierarchy is generally achieved viamathematical abstraction. So, although our focus in thissection is on natural mechanisms, we briefly note that theexperimental (i.e., macroscale) addition and observation orcapture of molecules is also at Level 2, but we elaboratefurther on experimental methods in the context of Level 3in Section V. In synthetic devices, we are more likely to havedirect (e.g., digital) implementations of higher-level behavior and require less complexity at the channel interface, evenfor nanoscale designs of electrical-transistor-based biosensors[139]. Thus, even though Level 2 is the lowest level within anindividual device, we can already observe distinctions betweennatural and artificial behavior.It is helpful in this section for the reader to have someunderstanding of cellular composition and the importanceof lipid bilayers. Lipid bilayers are thin yet stable polarmembranes that are hydrophilic on the outside (i.e., watersoluble) and hydrophobic on the inside (i.e., they repel water)[48, Ch. 10]. Lipid bilayers are the key basic componentof biological membranes and they help to compartmentalizecells and maintain molecule gradients because many moleculescannot pass through them, particularly if they are chargedor strongly polar. The outermost boundary of a cell, i.e.,the plasma membrane, is comprised of a lipid bilayer andmany other types of molecules whose functions can includemaintaining the membrane’s structure or to facilitate thetransport of specific molecules across the membrane (e.g.,through gap junctions or plasmodesmata as shown in Figs. 7and 8, respectively). Thus, cellular mechanisms for generatingor sampling molecular signals need to account for the plasmamembrane. For example, a typical cytosolic Ca + concentra-tion is . µ M, while in the extracellular fluid it is more than10,000 times higher at about 1.2 mM. This creates a verypowerful ion gradient that results in a rapid influx of Ca + towards the interior of the cell when there is a chance to do so.This difference is tightly controlled using pumps that activelytransfer Ca + ions out of the cytosol and Ca + channels thatare normally closed and impermeable to the ions [48, Ch. 15]. A. Molecule Generation and Release Management
The transmitter in an MC system needs to be able togenerate and release a molecular signal. These molecules maybe harvested from within the transmitter or its surroundingenvironment, or synthesized from its constituent components.If the molecules do not need to be released as soon as they areready, then the transmitter also needs a mechanism for storingthe molecules until they are needed. For example, Ca + ionsstored in the ER are released via Ca + gates to restore thecytosolic ion concentration when it is depleted [140].A common technique for storing molecules within eu-karyotic cells, either for transportation or until the storedmolecules are needed, is within vesicles. Vesicles are usuallyspherical or near-spherical shapes that are composed of a lipidbilayer. Thus, they can securely hold many types of molecules,e.g., cholesterol, proteins, neurotransmitters, or even invadingbacteria. Vesicles can vary in size from about nm (synapticvesicles) to several microns in diameter [92], and even smallervesicles can contain many thousands of molecules. To emptytheir contents, vesicles merge with another bilayer (such as acell’s plasma membrane) and release their molecules onto theother side of the other bilayer (e.g., outside the cell as shown inFig. 3) via exocytosis. Thus, molecules can be directly releasedfrom an intracellular vesicle into the extracellular space, whichcan occur very quickly; synaptic vesicles released by neuronscan empty their contents within about a millisecond or less[141]. While many transport vesicles are produced at a cell’s Golgiapparatus, processes that rely on rapid and precise vesiclerelease can fabricate them locally [48, Ch. 13]. For example,synaptic vesicles are produced locally from budding at theplasma membrane to help ensure a steady supply. No matterwhere they are produced, vesicles are generally too large toefficiently move by diffusion alone. So, they are carried alongcytoskeletal fibers by motor proteins (as introduced for cargo-based transport for Level 1 in Section III-D). Proteins that“coat” the outside of a vesicle are used to identify its intendeddestination so that it can bind to a suitable molecular motor.For example, a vesicle could be intended for an endosomeinstead of the plasma membrane. Additional surface proteinsare used to control both vesicle docking and fusion once ithas reached its target.A key advantage for using vesicles is the precise regulationthat is provided for molecule release, since particular proteinsneed to be available and in the correct state for a vesicle to betransported, docked, and fused with the destination membrane.However, vesicles in the constitutive exocytosis pathway areused for immediate uncontrolled release of their contents whenfusing with the plasma membrane [48, Ch. 13]. These providematerials to grow a plasma membrane, but can also carryproteins for secretion to outside the cell. In this pathway,proteins can be secreted as fast as they are produced; the onlydelay is in transport. In other cases, released molecules canbypass vesicle pathways entirely if they are able to directlypass through the plasma membrane [48, Ch. 11]. This is truefor small uncharged or weakly polar molecules, e.g., nitric ox-ide, or molecules that have dedicated transmembrane channels,e.g., the common ions sodium, potassium, and calcium.As noted, MC models typically treat molecule generationand release as instantaneous processes, or at least as steps thattake negligible time relative to molecule propagation acrossthe channel of interest [24], [27]. Exceptions include [142],[143], which have modeled transmitter molecule release withchemical reaction kinetics. The authors of [144] modeled theimpact of vesicle preparation and release on the informationcapacity in a chemical neuronal synapse.
B. Molecule Reception and Responses
The receiver in an MC system needs to be able to detectand respond to a molecular signal. Depending on the typeof received molecule and the receiver’s sensitivity, somethreshold signal quantity may need to be observed in orderto stimulate a corresponding response.
1) Molecule Reception:
In cells, extracellular signalmolecules generally fall into one of two families: 1) moleculesthat are small or hydrophobic enough to easily cross thereceiver cell membrane, and 2) molecules that are too large ortoo hydrophilic to cross the receiver cell membrane, as summa-rized in Table III. The first family of molecules can directlypass the cell membrane to activate intracellular enzymes orbind to intracellular receptor proteins, while the second familyof molecules relies on receptors at the surface of the target cellto relay their messages across the cell membrane [48, Ch. 11].In the MC literature, these two reception paradigms are usuallyreferred to as passive and active reception, respectively [2]. Dissolved gases and steroid hormones are representatives ofthe first family [48, Ch. 15]. Most dissolved gases can cross theplasma membrane and enter the cell interior to directly activateintracellular enzymes. For example, smooth muscle relaxationin a blood vessel wall can be triggered by Nitric Oxide (NO).Unlike molecules that directly activate intracellular enzymes,the detection of steroid hormones (such as cortisol, estradiol,and thyroxine) relies on intracellular receptors. All of thesemolecules cross the plasma membrane of the target cell andbind to their protein receptors distributed either in the cytosol (i.e., the liquid inside the cell) or the nucleus to regulate geneexpression.The vast majority of extracellular signal molecules belongto the second family. They are either too large or hydrophilicto cross the plasma membrane, so their detection requiresthe use of surface receptor proteins; see Fig. 12. Accordingto their biochemical signaling pathways, the surface-bindingreceptors can be further classified into three classes: ion-channel-coupled receptors, G-protein-coupled receptors, andenzyme-coupled receptors [48, Ch. 15]. • Ion-channel-coupled receptors are prevalent in the ner-vous system and other electrically excitable cells. Thiskind of receptor binds with ion molecules and cantransduce changes in ion concentrations into changes inmembrane potential. • G-protein-coupled receptors associate with a G protein inthe cytosolic domain. Once extracellular signal moleculesare bound to G-protein-coupled receptors, these receptorsare able to activate membrane-bound, GTP-binding pro-teins (G proteins), which then turn on or off an enzymeor ion channel on the same membrane and finally alter acell’s behavior [37], [145]. Examples of this type of re-ception include the transduction of a heartbeat slowdownsignal for heart muscle cells, a glycogen breakdown signalfor liver, and a contraction signal for smooth muscle cells.A recent review of G-proteins can be found in [146]. • The cytoplasmic domain of enzyme-coupled receptorseither acts on an enzyme itself or associates with anotherprotein to form an enzyme once signaling molecules bindto the outer surface of the plasma membrane. Enzyme-coupled receptors play a significant role in the responseto the growth factor molecules that regulate cell growth,proliferation, differentiation, and survival.
2) Reception Responses:
There is a broad diversity in howbiochemical receptors respond to molecular signals, and evenreceptors sensitive to the same kind of signaling moleculecan behave differently in different cells [48, Ch. 15]. Forexample, responses to acetylcholine include decreasing thefiring of action potentials, stimulating muscle contraction, andstimulating saliva production. Another example is calciumsignaling. The same stimulus can trigger a Ca + wave acrossone cell, local calcium oscillations in another cell, or causeonly a localized increase in the concentration in yet anothercell [147]. The different responses are due to the ability ofCa + to bind to a large variety of different proteins. Thus, thesame signal activates different signaling pathways dependingon the cell type and the available proteins. TABLE III: Reception Mechanism Summary.
Reception Type ExampleReception Site Receptor ProteinIntracellular reception: Molecules cancross cell membrane Intracellular enzyme Dissolved gasesIntracellular receptor Cortisol, estradiol, and testosteroneSurface reception: Molecules cannotcross cell membrane Ion-channel-coupled receptor Acetylcholine, glycine, γ -aminobutyric acid, ionsG-protein-coupled receptor Neurotransmitters, local mediators, hormonesEnzyme-coupled receptor Insulin, nerve growth factor SignalingMolecule(Ligand) CytoplasmExtracellularSpaceReceptorPlasmaMembrane (a)
Ligand-ReceptorComplex (b)
SecondaryMolecule (c)
Fig. 12: Steps of a generic molecule reception process. (a) There isa receptor embedded in the plasma membrane that separates a cell’scytoplasm from the extracellular space. The receptor can bind to aligand, which in this case is the signaling molecule of interest. (b)The ligand binds to the receptor to form a ligand-receptor complex.This instigates a conformational change in the receptor. (c) Theconformational change leads to a response, e.g., the release of aninternal secondary signaling molecule as shown.
The diversity in biochemical responses means that a giventype of receptor (or a collection of coupled receptors alonga pathway) has several distinguishing properties [48, Ch. 15].The timing of responses can vary by many orders of magni-tude, from milliseconds for muscle control and other synapticresponses [148], [149], to seconds for bacteria using chemo-taxis to respond to chemical gradient changes [79], to hoursor even days for changes in the behavior or fate of a cell (e.g.,gene regulation, differentiation, or cell death). Correspond-ingly, the persistence of a response could be very brief (asis usually needed in synapses) or even permanent.
Sensitivity to a signal can be controlled by the number of receptorspresent or by the strength of a secondary signal created by anactivated receptor. Similarly, a biochemical system’s dynamicrange specifies its responsiveness over a range of molecularsignal strengths. More complex responses can be achievedusing biochemical signal processing , e.g., applying feedbackto implement switches and oscillators. Some responses arecontrolled by the integration of multiple molecular signals,which we can study with a mathematical understanding oflocal data abstraction (i.e., Level 4 in Section VI). Conversely,a single molecular signal can coordinate multiple responses simultaneously within the same cell, e.g., to stimulate bothgrowth and cell division.
C. Mathematical Modeling of Emission, Propagation, andReception
The release and reception processes can be mathematicallymodeled by defining initial conditions (ICs) and boundaryconditions (BCs) for the propagation equations, such as those discussed for diffusion in Level 1 (Section III). In thisway, the spatial-temporal concentration distribution can beobtained by solving the partial differential equations (PDEs)that describe propapgation channels with ICs and BCs. Inother words, the release strategy, propagation channel, andreception mechanism jointly determine the channel responseand the observed signal. The recent survey [24] summarizedchannel impulse responses (CIR) under different models forthe transmitter, physical channel, and receiver, where the CIRwas formally defined as the probability of observation ofone output molecule at the receiver when one molecule isimpulsively released at a transmitter. It is noted that althoughthe CIR definition implies impulsive release of signalingmolecules, the transmitter geometry and molecular generationmethod still affect the CIR. Unlike [24], here we focus on themathematical formulation of specific (mostly ideal) conditionsso that the ICs and BCs can be mapped to the discussionsin Sections IV-A and IV-B. With these conditions, we alsoprovide a brief summary of some known channel responses inTable IV.
1) ICs on Release Strategies:
As stated earlier, the simplestscenario is that N molecules are released from a point in animpulsive manner at time t , so the IC can be expressed asIC : C ( d , t ) = N δ ( d − d TX ) , (15)where δ (·) is the Kronecker delta function and d TX is thelocation of the release point.Although the point transmitter has been widely used in MCresearch, it is quite idealized. Another idealized transmitteris the volume transmitter, which occupies physical space andits surface does not impede molecular movement. Signalingmolecules are released from a releasing space ˜ V TX or areleasing surface ˜ S TX of the volume transmitter. Therefore,a volume transmitter can be regarded as a superposition ofmany point transmitters that are located at different positions,and the corresponding IC can be expressed by extending (15)as follows:IC : ∫ d TX ∈ ˜ V TX N δ ( d − d TX ) dV or ∫ d TX ∈ ˜ S TX N δ ( d − d TX ) dS , (16)where d TX is a location within the releasing volume ˜ V TX or onthe releasing surface ˜ S TX . We note that (16) can also describethe molecule release from an ion-channel-based transmitter ifit has many open ion channels [24].
2) BCs on Propagation Channels:
An unbounded envi-ronment is a common assumption to simplify the derivationof the channel response. However, in practice, the molecular TABLE IV: Comparison of Diffusion-based Propagation Mechanisms.
Release Strategy Propagation Environment Reception Mechanism Channel ResponseTX Type IC Boundary Equation BC RX Type BCPoint IC Unbounded Eq. (3) BC N/A N/A [79, Eq. (2.8)]Spherical fully absorbing BC [11, Eq. (22)]Reversible absorbing BC [12, Eq. (8)]Spherical bounded BC Spherical fully absorbing BC [150, Eq. (13)]Rectangular bounded Fully absorbing walls BC [151, Eq. (19)]Rectangular/Circular bounded BC , BC N/A N/A [152, Eq. (14.4.4), (14.13.7)]Unbounded Eq. (6) BC [153, Eq. (18)]Cylindrical bounded [87, Eq. (11)]Unbounded Eq. (12) BC Reversible absorbing BC [13, Eq. (23)]Passive receiver [10, Eq. (9)]Partially absorbing BC [154, Eq. (16), (17), (29), (30)]Eq. (13) N/A N/A [155, Eq. (8)]Volume IC Unbounded Eq. (3) BC Passive & active receiver BC [156, Eq. (12)] propagation medium is often much more complex. Molecularpropagation can be constrained by various boundaries, such asthe tunnel-like structure of a blood vessel, oval shape of livercells, and the rectangular geometry of plant cells. A boundedmedium can provide molecules with guided transmission,limits dispersion, and can have beneficial effects for long-rangecommunication. The boundaries of a constrained medium areoften assumed to be reflective, and the corresponding BC isgiven as BC : ∂ C ( d , t ) ∂ d i (cid:12)(cid:12)(cid:12) d i = d b = , (17)where d i ∈ [ x , y , z ] is an element of the position vector d and d b is the position of the propagation boundary along direction d i .In addition, for both unbounded and bounded environments,the concentration at locations sufficiently far away from thereleasing source is usually assumed to be zero, which can bemathematically described asBC : C ((cid:107) d (cid:107) → ∞ , t ) = . (18)
3) BCs on Reception Mechanisms:
As stated earlier, thetwo conventional paradigms for molecule reception in the MCliterature are active and passive, where molecules do and donot participate in chemical reactions at the receiver, respec-tively. If a receiver is passive, then molecules are transparentlyobserved by the receiver without disturbing their propagation.If the receiver is active, then the molecules are usually detectedby surface receptors via absorption. However, if moleculescan be adsorbed (i.e., “stick” to the surface) instead of justbe absorbed (i.e., removed from the surface), then it is alsopossible that the receiver is capable of desorbing the moleculesthat were previously adsorbed. This type of receiver can becalled a reversible adsorption receiver and examples includethe reception of hormones and neurotransmitters [157]. Thecorresponding BC is given as [158]BC : D ∂ C ( d , t ) ∂ d (cid:12)(cid:12)(cid:12) d ∈ ˜ S RX = k C ( d ∈ ˜ S RX , t ) − k − C a ( t ) . (19) where k is the adsorption rate, k − is the desorption rate, ˜ S RX is the adsorbing surface of the receiver, and C a ( t ) is theaverage adsorbed concentration on the receiver surface at time t . We note that BC in (19) is a general formulation and canbe reduced to relevant special cases as follows. When k →∞ and k − = , i.e., every collision leads to absorption andthere can be no desorption, then the receiver becomes a fullyabsorbing receiver, and BC in (19) reduces to [11]BC : C ( d ∈ ˜ S RX , t ) = . (20)When k is a non-zero finite constant and k − = , then thereceiver becomes a partially absorbing receiver [12].We note that the aforementioned ICs and BCs are verygeneral, and one type of IC or BC can be represented invarious forms. The reason for this is that the different modelscan be expressed in terms of different coordinate systems, e.g.,Cartesian coordinates, cylindrical coordinates, and sphericalcoordinates, as appropriate for a given MC environment. Forexample, cylindrical coordinates are preferred in scenarios thathave some rotational symmetry about the longitudinal axis,such as a circular duct channel. D. Biochemical Signaling Pathway: Transcription Network
The molecule release and reception functions within a cellare carried out by proteins, such as the bacteriorhodopsinprotein that functions as a light-activated proton pump andtransports H + ions out of the cell, and the aforementionedsurface receptors that control the passage of molecules intothe cell [48]. Thus, the careful production and timely deliveryof these proteins is of utmost importance for a cell’s survival.Tight control of protein production is achieved through theinteraction of a number of genes, forming what is knownas a transcription network [41]. As shown in Fig. 13, atranscription network can be represented by circles and edges,where circles represent genes and edges represent their inter-actions. The building blocks of a transcription network area small set of recurring interactions between genes. Theseinteractions are called network motifs . In a network motif, Fig. 13: Overview of a transcription network, network motif, and feed-forward loops (FFL). In the transcription network, circles indicategenes and edges indicate gene interactions. Network motifs (dotted blue oval) are small sets of recurring interactions and are the buildingblocks of a transcription network. Feed-forward loops, one of the fundamental network motifs, are comprised of three nodes connected inone of eight possible configurations, i.e., Coherent FFL (C1-C4) and Incoherent FFL (I1-I4). Arrows denote activation and ⊥ symbols denoterepression of the corresponding node (gene).Fig. 14: Overview of protein production from DNA. (i) RNA polymerase guided by a trasncription factor binds to the beginning of a geneand transcribes the DNA sequence into mRNA. (ii) The newly formed RNA molecule is being modified immediately after production togive rise to a mature mRNA which then, in eukaryotes, is transported outside the nucleus. (iii) In translation, a ribosome uses mRNA as atemplate for the assembly of peptides. Raw materials are brought in by tRNA. (iv) A number of peptides are combined into proteins. Afterfurther modifications and folding, a mature protein is produced. the interaction between two genes is realized through geneexpression and regulation, where the product of one geneacts as the transcription factor to regulate the expression ofthe other. In the following, we first provide mathematicaldescriptions of gene expression and regulation. Then, wedescribe the feed forward loop (FFL), the typical networkmotif.
1) Gene Expression and Regulation with Mathematical De-scriptions:
Gene expression initially starts with transcription,where DNA is used as a template to synthesize mRNA, andthen mRNA will be converted to proteins through translation, as shown in Fig. 14. DNA transcription begins when theenzyme RNA polymerase (RNAP) recognizes and binds tothe promoter region. The promoter region is unidirectionaland can be found at the beginning of a gene. In addition,it decides not only the starting point of mRNA synthesis, butalso the synthesis direction. After RNAP binds to the promotersequence, RNAP unwinds the DNA at the starting point andbegins to synthesize a strand of mRNA. Once the mRNA isproduced, it is translated by a ribosome into protein moleculeswith the help of transfer RNA (tRNA). The production ofmRNA is controlled by transcription factors that bind to operator sites near promoter regions. The transcription factorsact as activators (or repressors) to enhance (or obstruct) thebinding ability of RNAP to promoter sites, thus controlling thetargeted gene expression rate. It is important to note that bothtranscription and translation establish two major control pointsfor protein regulation, as their products are being commonlymodified or even degraded before reaching the next stage( post-transcriptional and post-translational modification ).The aforementioned gene expression and regulation can bemathematically modeled by the Hill function once the bindingof a transcription factor to its site on the promoter reaches asteady state, i.e., equilibrium [41]. Let E x denote the inputsignal that carries information from the external world, and X denote a transcription factor with active form X ∗ . For activators , the input-output relation is d [ Out ] dt = β X ∗ n K n + X ∗ n , (21)where [ Out ] is the concentration of the output protein, K is the activation coefficient, β is the maximal expressionlevel of the promoter, and n is the Hill coefficient. Theactivation coefficient K has units of concentration and dependson the chemical affinity between the transcription factor andits operator region. With an increase in concentration of thetranscription factor, it is more likely for the transcription factorto bind to the operator region. However, since the bindingprobability cannot be larger than , the output protein levelis unable to increase infinitely and approaches a saturatedmaximal expression level β . The Hill coefficient n determinesthe steepness of the Hill function [41, Fig. 2.4]. For repressors ,a similar relation exists with the same parameters and can beexpressed as d [ Out ] dt = β K n K n + X ∗ n . (22)The values of K , β , and n may change with cell evolution.For example, K will change if a DNA sequence suffers frommutations that alter the transcription factor binding site.
2) Network Motif:
As the building blocks of transcriptionnetworks, network motifs have resisted mutations to persistover cell evolution, and the study of their dynamics canfacilitate the understanding of complex transcription networks.A typical network motif is the FFL [41], which has beenstudied in hundreds of gene systems in some organisms, suchas
E. coli [159], [160] and yeast [161], [162]. In the structureof an FFL, transcription factor X regulates proteins Y and Z ,and Y is also a transcription factor for protein Z . Due to thepossibility of three edges with each being either activationor repression, there are eight variations of this motif; seeFig. 13. The eight signaling pathways can be classified intotwo categories: coherent FFL and incoherent FFL, accordingto whether the regulation (i.e., activation or repression) ofthe direct path from X to Z is the same as the overallregulation going through Y (i.e., the regulation from X to Y and the regulation from Y to Z ) [41]. In the most well-studiedtranscriptional networks in E. coli and yeast, the coherent type-1 FFL (C1-FFL) and incoherent type-1 FFL (I1-FFL) occurwith a high frequency and so we discuss them in detail below.
Delay
Time ZY AND
Fig. 15: C1-FFL with an AND input function at the Z promoter. E x is the input signal for X . Transcription factor X is an activator ( ↓ ) for Y and Z , and Y is also an activator for Z . The AND gate indicatesthat both X and Y are needed to regulate Z . Time ZY AND
Fig. 16: I1-FFL with an AND input function at the Z promoter. E x is the input signal for X . Transcription factor X is an activator ( ↓ ) for Y and Z , while Y is a repressor ( ⊥ ) for Z . The AND gate indicatesthat both X and Y are needed to regulate Z . • C1-FFL:
The signaling pathway of C1-FFL is depictedin Fig. 15. A dynamic feature for C1-FFL is the abilityto distinguish spurious input square signals. The accu-mulation time of Y depends on the duration of the inputsignals. If a transient spike signal arrives, the accumulatedconcentration of Y is too low to satisfy the thresholdcondition and Z will not be produced, i.e., the systemdoes not respond to this input signal. This feature preventsthe C1-FFL motif from responding to spurious inputsignals. • I1-FFL:
The signaling pathway of I1-FFL is depictedin Fig. 16. Compared with C1-FFL, Y regulates Z viarepression instead of activation in I1-FFL, such that Z shows a pulse-like profile in response to a sustainedinput signal. Once induced by an input signal E x , theexpressions of the genes encoding protein Y and protein Z are both activated, and Z is instantly produced. Here,the delay that appears in C1-FFL is eliminated becauseprotein Y needs some time to reach the repression thresh-old for the Z promoter, which gives a chance for protein Z to accumulate. Once the concentration of Y crossesthe repression threshold, it starts to repress the protein production rate of Z . As a result, the concentration of Z begins to decrease and either reaches a steady state ordrops to zero depending on the repression strength of Y .V. L EVEL
HYSICAL /D ATA I NTERFACE
Level 3 of the proposed hierarchy is the interface betweenphysical signals at communicating devices and quantifyingthese signals mathematically. From the perspective of commu-nication systems, this includes: 1) how signals are modulatedat a transmitter and demodulated at a receiver; 2) how mod-ulated signals control the propagation (i.e., communicationchannel), and 3) how the physical signals (i.e., channel re-sponses) are observed for demodulation. In other words, giventhat there is information to transmit, how does the transmittertranslate this into a molecular (or some other physical) signal?Then, how does a receiver translate the observed signal backinto information?Depending on whether the data interface is at microscaleor macroscale, with reference to Fig. 17 we can catego-rize the physical/data interface of a microscale communica-tion system into 1) microscale modulation and demodula-tion (Section V-D1), and microscale signal operation (Sec-tion V-A); and 2) macroscale modulation and demodulation(Section V-D2), macroscale control of microscale change(Section V-C), and macroscale observation of microscale phe-nomena (Section V-B).In this section, we start with a general overview of quantify-ing microscale signals from the perspective of gene regulationand metabolic control. We proceed to review methods forobserving and quantifying microscale phenomena from themacroscale, i.e., in a laboratory environment. This leads toa discussion of macroscale control of microscale change. Wefinish the section by discussing the quantification of cellularsignals as information, with both modulation and demodula-tion processes.
A. Microscale Signal Operations
The interface between physical signals and their mathemat-ical quantification can be perceived as being relatively simplefor many cell signaling processes. It is often a matter ofdetecting whether the signal is stronger than some threshold,e.g., detecting a sufficiently high autoinducer concentrationin quorum sensing (which we describe as a case study inSection VIII). The creation of a signal and then the detectionof its presence is a common communication methodology forcells, and is sufficient to link many processes at Level 2 (i.e.,biochemical pathways detecting signals) with activity at Level4 (i.e., the device-level state and the actions that the devicetakes). For example, there are biochemical signaling pathwaysthat modify protein function directly (i.e., without requiringchanges to gene expression) [48, Ch. 15]. However, receivinga signal can also require more precision than simple detection,as can often be observed in the context of gene regulation , i.e.,the activation and deactivation of different genes to control We emphasise that gene regulation is distinct from the genetic informationembedded within DNA or RNA itself; gene regulation controls which DNAsequences are made accessible for transcription into RNA. the proteins that are produced within a cell, as we introducedfor Level 2 in Section IV. Due the impact of gene regulationon cell behavior, in the following, we discuss the quantificationof signals from the perspective of gene regulation. We choosecontrol of the metabolism as a specific example of generegulation, given the metabolism’s importance for cell growthand reproduction. We further discuss conversion betweenquantified microscale signals and bits in Section V-D1.
1) Gene Regulation:
Regulation is often described usinggenetic circuits, which show how a combination of inputs (i.e.,signals) leads to activation of the gene in question (e.g., see thegeneric transcription network in Fig. 13). Depending on thesensitivity to the inputs and on the possible ranges of outputs,the quantification of these processes can be understood asbeing analog or digital. For example, if there is an appreciabledifference in response according to input signal concentrations,such that the output varies continuously with the input, then thequantified response is analog. This can occur in the fine tuningof metabolic processes by some hormones [48, Ch. 15]. Ifthere is a discrete (i.e., readily countable) number of responselevels, regardless of input concentrations, then the response isdigital. A threshold-based response (i.e., most existing workfrom the MC engineering community [24]) is digital, whetherthe response threshold is a single detected molecule or somelarger quantity. There are often only two response levels (e.g.,on and off), where the bistability of the circuit is achievedthrough positive feedback that pushes the response to one ofthe two levels [48, Ch. 15].
2) Metabolic Control:
We highlight the gene regulation ofcell metabolism as an example of microscale signal operations.The metabolism of a cell refers to all chemical reactionsthat take place inside the cell and that are necessary forreproduction and growth [48, Ch. 2]. These chemical reactionsare highly interdependent and chained into signaling pathways,where the product of one reaction is the on-demand substratefor the next reaction in the pathway. These reactions requirespecialized proteins (i.e., enzymes) in order to proceed, thusthey offer an effective means of regulation. By varying theamount of enzyme that controls each reaction, a cell isable to control its metabolism precisely. This control usingenzymes, in turn, relies on a cascade of reactions triggeredby extracellular cues that eventually stimulate the release oftranscription factors (TFs) inside the cell to activate or repressenzyme production.It is important to note that the release of TF is determinedby a combination of different chemical components (e.g.,hormones) with particular concentrations, which can resultin a digital ON/OFF enzyme activation mechanism. Suchmechanisms can be controlled by the cell’s environment,where variations in the surrounding chemical compositioncan prompt the up- or down-regulation of the enzymes tocontrol cellular growth or production of chemical compoundswithin the cell. Each TF can enter the nucleus and interferewith the expression of specific clusters of genes to alterthe type and amount of proteins produced, which ultimatelyestablishes the cell’s metabolism. For example, detection ofthe glucocorticoid hormone by a liver cell triggers an increasein energy production via the enzyme tyrosine aminotransferase A.MicroscaleSignalOperations
B.MacroscaleObservationsofMicroscalePhenomenaC.MacroscaleControl ofMicroscaleChange
Optical Microscopy OptofluidicsIn Vivo ImagingMagneticNanoparticle ImagingTHz CommunicationpH-measuringInstrumentsMagneticUltrasoundTemperatureOptical Electrical Chemical
D. 1) Microscale Modulation/Demodulation C o n v e r s i o nb e t w ee n s i g n a l s a nd B i t s D. 2) Macroscale Modulation/Demodulation
Conversion betweensignals and Bits
Fig. 17: Schematic diagram showing typical workflows when dealing with microscale signal operations in MC. The label indexing in thefigure corresponds to the subsections of Section V. Signal operations in MC systems (A) depend on microscale data (D.1) and can bedetected using a number of available methods such as microscopy or optofluidics (B) and then be decoded (D.2). The signal operations (A)can also be controlled (C) based on target data (D.2) to modify the microscale data (D.1). [48, Ch. 7].
B. Macroscale Observations of Microscale Phenomena
Observing signaling phenomena in laboratory experimen-tation is important to understand their behavior and inferinformation about the system . However, we are generallyconstrained by the level of detail that we can readily observe(especially microscale behavior). For example, a living cellhas a typical size of about µ m, interactions between cellscan occur at a scale from a few µ m to a few mm, and thereis also communication between different organs, which mightspan up to a few meters. Signaling molecules can vary insize from around pm for individual ions to nm forextracellular vesicles [164]. Furthermore, temporal scales varywidely. Chemical reactions occur typically in milliseconds[165], ionic diffusion in biological tissues occurs at a rate ofa few tens of µ m /s [166], and physiological tissue responsesto stimuli can occur in milliseconds or over many hours.Currently, there is no single technique that enables theinspection of biological signaling processes across all spatialand temporal scales simultaneously. Thus, observation and ver-ification often relies on a combination of established methods.In the following, we review state-of-the-art methods for ob- Although macroscale tools provide a way to observe microscale phenom-ena, imperfections in experimental tools can lead to a noisy and non-idealinterface. One approach to model this uncertainty and try to enhance theaccuracy of observations are learning-based models [163]. serving microscale phenomena, including optical techniques,magnetism, THz waves, and pH sensing.
1) Optical Microscopy:
Perhaps the most commonly knownobservation method is optical microscopy. Since its conceptionin the 16th century, the optical microscope is one of the mostvaluable instruments in any laboratory that investigates themicroscale world. The major limitation of optical microscopyis the diffraction barrier, namely the inability of the lens todistinguish between objects which are separated by a distanceless that half the wavelength of the light used. Confocal lasermicroscopy greatly improved image resolution using visiblelight [167], though it was only since the end of the last centurythat it was finally possible to overcome this limit and obtaininstruments such as the near-field scanning microscope, thescanning tunneling microscope, and the atomic force micro-scope. Subsequent refinements led to an increase in resolutionto the point that a single molecule can now be distinguished[168], [169]. Together with fluorescent microscopy, these toolsremain among the most accessible and valuable in cell biologyimaging.There is increasing demand in modern science to visu-alize dynamic spatial and temporal events at the micro-and nanoscale. It is now possible to obtain nanometer sizeimages of cells while simultaneously measuring subjectedmechanical forces [170], [171]. Protein motion has also beenobserved in great detail at the microsecond timescale usinginterferometric scattering microscopy [172]. Concerning therelease of molecules by the cell into extracellular space, various methods such as fluorescence microscopy [173], [174]and electrochemical techniques [175], [176], either alone orcombined, are particularly suited for capturing the traffickingof molecules [177], [178].
2) Optofluidics:
When analyzing biological samples, it isoften desirable to separate and sort out individual molecules.Conventional microscopy is cumbersome and already closeto its limit in terms of spatial resolution, therefore is notusually suitable for this purpose. Optofluidics technology wasdeveloped with this in mind, combining advanced opticalmicroscopy with microfluidics [179]–[181]. It was developedas a way to miniaturize analytical instruments and it laterlead to lab-on-a-chip technology [179], [182]. Optofluidicstechnology is particularly suitable for analysis of very smallworking volumes, in the range of nanoliters or femtoliters.This is because it combines the analytical mechanism withsample preparation. By taking advantage of low energy con-sumption, nanoscale sample handling, and being free fromrequirements for very specialized electronics, optofluidics hasbeen combined with other techniques such as flow cytometry[183], interferometry [184], and Raman spectroscopy [185]with good results in cell and molecular microscopy imaging.Its characteristics have also enabled integration into biosensors[186] and on-chip technologies [187].
3) In Vivo Imaging:
Despite the popularity of optical-basedimaging, there is strong scattering of light by biological tissueand so its application is limited beyond optically transparentsystems or cultured cells. A long-standing challenge is tomake non-invasive observations of in vivo activity. Conven-tional methods of in vivo imaging continue to be studied toimprove their sensitivity and spatial resolution, e.g., there issignificant research in the design of contrast nanoparticlesfor magnetic resonance imaging [188]. Recent advances inultrasound imaging with synthetic biology have overcomeultrasound’s lack of specificity to make it a strong candidatefor in vivo observation of cellular functions [189]. Air-filledprotein nanostructures called gas vesicles have already beenengineered for introduction in mammalian cells and havehelped produce high-resolution ultrasound imaging of geneexpression in living mice [190].
4) Magnetic Nanoparticle:
Driven by the wide applicationsof magnetic nanoparticles in drug delivery systems [191]–[193], the properties of magnetic nanoparticles have been usedto observe microscale processes [19], [194]. In [194], theauthors presented a magnetic-nanoparticle-based interface andproposed a wearable susceptometer design to detect magneticnanoparticles. In [195], an experimental platform that used asusceptometer to detect magnetic nanoparticles was proposed,where the susceptometer can generate an electric signal ifmagnetic nanoparticles pass through it.
5) THz Communication:
The integration of nanosensorsand terahertz (THz) communication modules can also supportmacroscale observations of microscale phenomena. This isrealized by the fact that chemical nanosensors are capable ofmeasuring the concentration of a given gas or the presenceof a specific type of molecule, which can then be communi-cated from intrabody to outside the body via THz signaling.For nanosensors that are made of novel nanomaterials, such as Graphene Nanoribbons (GNRs) and Carbon Nanotubes(CNTs), the sensed and absorbed molecules can change theelectronic properties of the nanomaterials by either increas-ing or decreasing the number of electrons moving throughthe carbon lattice. With nano antennas, the change in thenumber of electrons can enable the conversion of molecularinformation to THz waves [196]. One example of using THzcommunication to observe microscale phenomena is [197],where a nano antenna array operating in the THz band wasdesigned to detect different carbohydrate molecules and theirconcentrations.
6) pH-Measuring Instruments:
Hydrogen ions (i.e., pro-tons) are a popular signal molecule type with advantages ofsmall size and easy production. More importantly, hydrogenions have the physical property that their accumulation canlead to a reduction of the solution pH. Therefore, the concen-tration variations of protons can be observed at macroscaleusing a pH meter [21], [198]. The same approach is consideredfor communication in [199], where pH meter values are usedto determine whether acids or bases are being transmitted.
C. Macroscale Control of Microscale Change
Macroscale instruments do not only enable us to observemicroscale phenomena, but also make it possible to controlmicroscale systems, which establishes an interface from themacroworld to the microworld and would expand the ca-pability of MC. An example application that benefits frommacroscale control is drug delivery, where precise guidanceto the diseased cells and controllable release of drugs couldlargely improve their therapeutic effect. There are many ap-proaches that have been developed towards macroscale con-trol. Here, we briefly review the controlled release of signalmolecules via macroscale stimulation.
1) Macroscale Chemical Control:
It has been a commonchoice to use genetically modified
E. coli bacteria in ex-perimental MC testbeds [18], [21], [198], [200]. In [18], amicrofluidic chamber was used to trap
E. coli . These bacteriawere genetically modified by introducing a plasmid from
V.fischeri to produce fluorescence in response to the C6-HSLsignaling molecule.In [200], communication between two physically separatedpopulations of E. coli was controlled and observed. Thepopulations were grown on a microfluidic chip and separatedby a filter composed of cellulose nanofibrils between rowsof polydimethylsiloxane (PDMS) pillars. The filter preventedthe populations from mixing but enabled the passage ofsignaling molecules such as the quorum sensing moleculeacyl-homoserine lactones (AHL). Furthermore, the microflu-idic chip was designed to flush excess bacteria and thusconstrain the population sizes. The sender population couldproduce AHL and fluoresce cyan in response to the additionof arabinose, and the receiver population fluoresced green inresponse to AHL. Fluorescence patterns of the two populationswere observed with negligible delay, suggesting that rapidsignaling from sender to receiver enabled the populations tobehave synchronously.
2) Macroscale Electric Control:
External electric stimulusis a method to bridge the macroworld and the microworld.The electrically controllable release of DNA molecules im-mobilized in layer-by-layer (LbL) thin film was investigatedin [201]. Upon an electric signal on the LbL film, DNAmolecules are disassembled and released with an electrodis-solution of the layers. The DNA molecule release processcan be switched off when the electric stimuli is removed,and the released number of DNA molecules is proportionalto the amplitude of the electric signal, which allows for atunable release of signal molecules. It is noted that externalelectric stimulus can also trigger biological responses viaredox reactions, such as the patterning of biological structureand the induction of gene expression [56].
3) Macroscale Optical Control:
Light-sensitive cellular en-tities can be controlled by external light sources. One exampleis the release of biomolecules from photoremovable containersupon illumination [202], which achieves a conversion ofoptical signal to chemical signal. Similar signal transductionhas also been realized in the MC community [198] and [21],where
E. coli was modified with light-driven proton pumps(i.e., bacteriorhodopsin), which can be excited by externallight sources to induce proton release, with increased pH valuemeasured via a pH sensor.
4) Macroscale Temperature Control:
External temperaturecan be another macroscale stimulus to control microscaleprocesses. Some nanocapsules that are temperature-sensitive,such as the liposomes in [203], the dendrimers in [204], andthe polymersomes in [205], can undergo a conformation orpermeability change and release encapsulated signal moleculesas a response to a temperature increase. In this way, thermalsignals from the exterior of devices are converted into chemicalsignals. It is noted that the morphological changes in [205]are reversible, meaning that sustainable temperature controlcan be achieved. One method to achieve temperature controlis via focused ultrasound, which has been proposed to controlcellular signaling and the expression of specific genes [189].Candidate targets include temperature-sensitive ion channelsand transcription repressors.
5) Macroscale Mechanical Control:
In addition to tem-perature control, focused ultrasound can provide momentumand energy to interact with molecules, cells, and tissues viamechanical mechanisms [189]. For example, ultrasound wavescan be amplified by microbubbles and provide mechanicalforces on a millisecond timescale, i.e., with much greaterprecision than temperature changes. This approach has beenused in vitro to open mechanosensitive ion channels expressedin mammalian cells. A current constraint for use in vivo is the difficulty in delivering such microbubbles beyond thebloodstream.
6) Macroscale Magnetic Control:
Magnetic nanocarriersare important carriers for drug delivery. The magnetic behaviornot only allows magnetic nanocarriers to be manipulated inspace towards targeted locations by external magnetic fields,but also facilitates their visualization by increasing their imag-ing contrast in magnetic resonance imaging (MRI), which inturn provides a way of monitoring their movement through thebody. After nanocarriers arrive at desired sites, the magnetic energy can be converted into internal energy to induce localheating, thus triggering the release of loaded drugs [206].
D. Conversion from Signals to Bits
Throughout this section, we have been referring to thequantification of physical molecular signals, how these signalsare observed, and how such signals can be controlled. Wehave mentioned that these signals contain information, but wehave not yet directly linked the mathematical abstraction tothe quantification of information. Level 3 of the proposedhierarchy includes not only the mathematical abstraction ofphysical signals, but also how a quantified signal containsinformation. We now elaborate on this idea and re-visit someof our examples from this perspective.The MC community already has an understanding of infor-mation transmission that is directly inspired by conventionaltelecommunication systems [207]. A transmitter in a commu-nication system has information to send. Information that is ina quantifiable form is typically represented as a sequence ofdigital bits, i.e., 1s and 0s, and the sequence is packaged into aseries of symbols, each of 1 or more bits. The transmitter needsa scheme to represent each symbol as a different physicalsignal. Generating a physical signal that corresponds to thecurrent information symbol is called modulation . Demodula-tion at the receiver then uses the observed signal to attemptrecovery of the intended symbols and hence the original bitsequence. Thus, the observed physical signal is somehowquantified and then translated back to information. There havebeen significant research efforts to effectively and efficientlydemodulate diffusion-based signals to recover sequences ofdigital bits [27].The simplest modulation scheme and also the most popularone in the MC literature is binary concentration shift keying(BCSK). In BCSK, the transmitter releases a certain numberof molecules to send a 0 (i.e., bit-0), and a higher number ofmolecules to send a 1 (i.e., bit-1). When zero molecules arereleased to send a 0, then BCSK is also known as ON/OFFkeying (OOK). BCSK sends a single bit of information witheach symbol. Other modulation schemes use more variationsin the number of molecules to send more bits, or they varyfeatures, such as the type of molecule used or the precise timeinstant when molecules are released. Given the prevalence ofON/OFF signals in cell signaling, we can readily understandsignaling in many biological systems as BCSK.
1) Microscale Modulation and Demodulation:
Thetelecommunications engineering approach to modulation anddemodulation does not always precisely align very well withsignaling in cell biology, in particular when it comes to theability to represent information with a long sequence of bits.This is evident in some existing platforms developed by theMC community, including the tabletop MC system [22] andits iterations, which are actually macroscale systems, as wellas droplet microfluidic channels [208]; these testbeds focuson the physical or chemical properties of signal propagationand detection and not on integration with a biological system.While there are specific instances where biological data canreadily map to sequences of bits, such as strands of DNA or RNA (where each base pair is a 2-bit symbol), many MCschemes are not structured in this manner and often 1 or afew bits is sufficient to represent all the information beingmodulated, e.g., whether a target threshold concentrationhas been reached to stimulate an action. Nevertheless, adigital representation can still be useful. For example, genesare often represented as switches that are turned on oroff by transcription factors. Thus, there can be one bit ofinformation for each switch, and this bit can change with thedemodulation of the corresponding gene regulation signal.This signal could come from within the cell, e.g., via acoupled internal signaling pathway, or from outside the cell,e.g.,
E. coli demodulating a chemotactic signal from itssurrounding environment to decide whether to proceed alongits trajectory (i.e., run ) or change direction (i.e., tumble ) [79].
2) Macroscale Modulation and Demodulation:
Broadlyspeaking, making macroscale observations (as reviewed inSection V-B) and trying to recover information about a cellularsystem corresponds to demodulation at a receiver, whereas us-ing macroscale methods to control such a system (as reviewedin Section V-C) corresponds to modulation by a transmitter.At macroscale, we have the benefit of easy access to moderncomputing devices. Macroscale MC systems often includea connection with a microcontroller board (e.g., Arduino)or a computer to perform modulation to convert sequencesof symbols into physical signals or to demodulate signalsinto received symbols. Thus, any of the macroscale methodsdiscussed in this section could be abstracted and interpretedas a transmitter or receiver of quantified information. In thefollowing, we highlight works that did so explicitly to quantifythe transmission of bits.At a macroscale transmitter, electrical signals representingbit sequences can be directly modulated as chemical signals[201] or through an intermediate signal form, such as an optical signal in [21], [198] and a thermal signal in [203]–[205]. In [201], the authors realized OOK modulation bytranslating electrical signals into biological DNA signals.For transmission of bit-1, a rectangular electrical signal withan amplitude of 5 V and a duration of 10 s was applied tostimulate the release of DNA molecules from a multilayerfilm, while for transmission of bit-0, the electrical signalwas switched off. This setup could achieve a bit rate of1 bit/minute. In addition, the authors also found that thenumber of released DNA molecules was dependent on theamplitude of the electric stimulus, which could enable a higherorder concentration shift keying modulation by modulatingdifferent symbols with different amounts of DNA.In [21], [198], OOK modulation is achieved by an optical-to-chemical conversion. The intended symbol sequence doesnot directly induce the light-driven proton pump to emitprotons, but it is first modulated as an optical signal to switchan LED on or off. The LED is switched off during theentire symbol interval to represent bit-0 while it is turnedon to transmit bit-1, thus controlling the release of protonsby modified
E. coli . The proton releases were measuredwith a pH sensor , and channel estimation techniques andadaptive transceiver methods were implemented at macroscaleto demodulate the signal and recover the symbol sequences. A reliable throughput rate of about 1 bit/minute was achieved,which was much faster than the 6-7 hours to recover ON/OFF fluorescent patterns made by modified
E. coli bacteria inresponse to C6-HSL signal molecules in [18].VI. L
EVEL
OCAL D ATA A BSTRACTION
Level 4 of the proposed hierarchy is the interface betweenthe mathematical quantification of physical signals (i.e., theoutput of Level 3) and how the information in these signalsis manifested and manipulated within an individual communi-cating device. In other words, Level 4 is concerned with thecontext for information in cell biology signaling. By definition,this level is more mathematically abstract than the lowerlevels, but is also manifested as individual behavior. We expectthat synthetic devices, whether they are at a microscopicor macroscopic scale, will generally have more functionalcomplexity than natural microorganisms have at this level.For example, digital computing and memory devices canenable significant data processing capabilities. While naturedoes have means for storing and manipulating large quantitiesof information, e.g., DNA and memory in the brain, thefunctional complexity of communication is primarily in thebiochemical processes that physically manipulate the signal,i.e., at Level 2 of our proposed hierarchy. Nevertheless, Level 4describes data, where the data comes from, and how individualdevices use it.From a communication perspective, the transmitter is re-sponsible for encoding its information into a quantifiableform such as a bit sequence (that is then modulated, i.e.,in Level 3). Once the receiver has demodulated the receivedmolecular signal, it is then decoded to recover the embeddedinformation. The encoding and decoding processes are usuallyignored in contributions by the MC community, because it isoften assumed that a bit sequence of interest already exists(or one is randomly generated if needed). The fundamentalcommunication problem is for the receiver to recover thebit sequence, typically without consideration of how thisinformation is subsequently utilized (as this is beyond thescope of a conventional communication engineering problem).However, since behavior in cell biology is tightly coupled withthe information that cells receive, it is particularly relevant forour holistic approach to consider the significance of the data.The remainder of this section is organized as follows.We describe the meaning of information in cellular signals,including limits on how much information these signals cancarry (Section VI-A). Contexts for cellular information includegenetic information in DNA and RNA, collecting informationabout the external cellular environment, and controlling actionssuch as cell division, cell differentiation, and cooperation.We then transition to a discussion of the design of analogand digital circuits based on chemical reactions and syntheticbiology (Section VI-B), and how these can be used to realizecommunication functionalities in an engineered cell biologysystem (Section VI-C). Finally, we elaborate on the physicalstructure of DNA and its potential for synthetic storage (Sec-tion VI-D). A. Information in Cellular Signals
While we consider DNA holistically as a case study inSection VII, we summarized the translation and transcriptionprocesses in Section IV, and we will elaborate on microscalestorage using DNA in Section VI-D, it is worthwhile to brieflydiscuss it here in the context of local cellular data. BothDNA and RNA are linear polymers composed of nucleotidesubunits with 4 distinct bases (DNA and RNA both useadenine, guanine, and cytosine; DNA has thymine while RNAhas uracil) [48, Ch. 6]. Thus, each subunit carries 2 bits ofinformation, which get copied when DNA is transcribed toproduce RNA. While some RNAs have specific standaloneroles including reaction catalysis and regulation of other genes,mRNAs are RNAs that are created for translation to protein.In this latter case, triplets of bases called codons are usedto encode each of the 20 amino acids that are commonlyfound in proteins. Since 3 nucleotides, each having one of4 bases, can be combined to make = distinct codons,many amino acids are specified by multiple codons, and thereare also codons that indicate the end of a sequence. Whilethere are many biochemical steps to go from DNA to protein(some of which were described in Section IV-D), there is aclear mapping from nucleotide bases to amino acids.Besides genetic information, many cellular signaling pro-cesses are driven by a single bit of information [48, Ch. 21],e.g., the presence or absence of an event or a change of state.From a communications perspective this can seem incrediblysimplistic, but this is consistent with existing bounds onmutual information and capacity, including the estimation ofenvironmental signals using biochemical reaction networks[209], intracellular signaling in an individual cell [210], andan individual signal transduction channel [211]. However,natural options do exist to transmit information beyond suchconstraints. Generally, individual signaling pathways couldbe chained together to drive more complex functions andbehavior. It has been shown that noise filtering in E. coli enables it to detect antibiotic concentrations with up to 2 bitsof resolution, thereby distinguishing sublethal levels [61]. Theauthors of [212] showed that temporal signal modulation canreduce the information loss induced by noise and increase theaccuracy of biochemical signaling networks.Communication between cells is also used to augment theavailable information [213]. Noise at a single-cell level canbe exploited to increase information at a population level upto several bits by smoothing out individual cell responses thatwould otherwise lead to abrupt ON-OFF changes in populationbehavior [210]. There are limits to the gains available and thisis in part due to the constraints imposed by communicationreliability [214]. While it may be intuitive to think thatcommunicating cells should be as close together as possibleto maximize the precision in concentration estimation, it hasactually been shown that sparse packing of a large populationis optimal for concentration sensing [215].As we have noted, cellular information is tightly coupledwith behavior. Even DNA, which is stored analogously todigital information, leads to the RNA and proteins that drivemany cellular tasks. In the case of
E. coli measuring antibiotic concentrations, detection of sublethal levels can signal when toproduce costly resistance mechanisms to improve populationfitness [61]. Information shared across a cellular populationcan include the fraction of the population that is preparing fora major event such as cell division or cell death [210].Other examples of the significance of local cellular data canbe readily identified. For example, quorum sensing is usedin many communities of bacteria to coordinate decisions byreleasing signaling molecules [216]. A simplified understand-ing of quorum sensing is that the accumulation of signalingmolecules is treated as a proxy for the local estimation ofcurrent population density. While the density is not estimatedprecisely, bacteria can distinguish between “high” and “low”states and gene expression is switched to favor cooperativebehavior when the estimate becomes sufficiently high. Wediscuss quorum sensing in greater detail as a case study inSection VII.Cell differentiation is the specialization of cells into par-ticular roles and is a fundamental process for multi-cellularorganisms [48, Ch. 21]. One way in which cell differen-tiation is controlled is via the reception of signals fromneighboring cells. Diffusion creates concentration gradientsbased on proximity to the source signal, enabling cells tospecialize according to their location. Additional diversity canbe provided by controlling differentiation with multiple typesof signals, such that each molecule type corresponds to onebit of information.Theoretical and experimental studies in [217]–[220] estab-lished methods to characterize the limits and information flowrates for cell metabolism, and quantify the amount of controlthat the external environment can exert on a cell in terms ofmetabolic fluxes. By using different combinations of chemicalcompounds with varying concentrations, temperatures, andacidity, the chemical composition of a cell’s environment canbe manipulated in order to trigger a specific response, such asthe secretion of a useful metabolite.Additional works that have sought to describe the infor-mation in natural cellular signals include calcium signalingspecificity in [221], and the insulin-glucose system in [222].The authors of [223], [224] modeled the information carriedin the action potentials between plant cells.
B. Digital and Analog Circuits
An MC transmitter encodes information into a quantifiableform and then modulates it into a physical signal. An MCreceiver demodulates a received chemical signal to recoverthe transmitted information. To guarantee successful infor-mation delivery, signal processing units that process infor-mation flow over molecular concentrations are envisioned tobe indispensable components for synthetic MC transmittersand receivers with complex communication functionalities,including modulation-demodulation and encoding-decoding.In general, biochemical signal processing functions can berealized in two fashions: 1) chemical circuits [225] basedon “non-living” chemical reactions, and 2) genetic circuits[226] in engineered living cells. In chemical circuits, a set ofchemical reactions is designed for a target desired chemical response, whereas in genetic circuits, a gene regulatory net-work based on synthetic biology is designed to achieve desiredfunction. Considering the scalability of digital design and thediscreteness of molecules, it is logical to start by designingcircuits to process digital signals that switch rapidly from adistinct low state representing bit-0 to a high state representingbit-1. However, biological systems do not always operatewith reliable ‘1’ and ‘0’ signals; instead, many signals areprocessed probabilistically and show a graded analog responsefrom low to high level [227]. In addition, motivated by thefact that biological systems based on analog computationcan be more efficient compared with those based on digitalcomputation [227], [228], analog circuit design also receivesattention from biologists and engineers. In the following, wereview some synthetic digital and analog circuits which aredesigned based on chemical reactions and synthetic biology.These circuits can not only achieve certain computationaloperations by themselves, but can also be integrated to realizesome communication functionalities.
1) Digital & Analog Circuits via Chemical Reactions:
Many types of digital circuits have already been designed andrealized via chemical reactions, demonstrating their capabil-ities to process molecular concentrations. Designing digitallogic functions has also attracted increasing research attention.Combinational gates, including the AND, OR, NOR, and XORgate, were designed in [229] based on a bistable mechanism.For a single bit, the HIGH and LOW states are indicated bythe presence of two different molecular species. The designedgates were mapped into DNA strand-displacement reactionsand validated by generating their chemical kinetics. The au-thors of [230] also used the bistable mechanism, where fivegeneral and circuit-free methods were proposed to synthesizearbitrary combinational logic gates. The AND, OR, NOR, andXOR gates were also realized via joint chemical reactions andmicrofluidic design in [231] with a different bit interpretation,where bit-1 is represented by a non-zero concentration valueand bit-0 is represented by zero concentration. A mathematicalframework was proposed in [232] to theoretically characterizethe designed gates, and insights were also provided into designparameter selection (e.g., species concentrations) to ensure anexhibition of desirable behavior.An architecture of analog circuits to compute polynomialfunctions of inputs was proposed in [233], where the circuitswere built on the basis of analog addition, subtraction, andmultiplication gates via DNA strand displacement reactions.Relying on the help of Taylor Series and Newton Iterationapproximations, these analog circuits can also compute non-polynomial functions, such as the logarithm. However, anaccurate logarithm computation over a wide range of inputsrequires a large number of reactions, due to the high-orderpower series approximation. In [234], the authors presenteda method to accurately compute the logarithm with tunableparameters while maintaining low circuit complexity. In [235],a systematic approach to convert linear electric circuits intochemical reactions with the same functions was presented.The principle of the approach is that both electric circuitsand chemical circuits can be described by ordinary differen-tial equations (ODEs), no matter what quantities the ODEs represent (e.g., voltages or concentrations). Based on this, anelectric high pass filter circuit was realized by a set of chemicalreactions.
2) Digital & Analog Circuits via Synthetic Biology:
Afundamental objective of synthetic biology is to control andengineer biochemical signaling pathways to build biologicalentities that are capable of carrying out desired computingtasks. The single input logic gates were synthesized to carryout simple computations, and these include the BUFFER gate[236] and the NOT gate [237], which are directly inspiredby mechanisms of gene expression induced by activators andrepressors, respectively. To expand the information processingability, multi-input logic gates, including a / -input ANDgate [237], [238], / / -input NAND gate [239], [240], and / -input OR gate [240], were also designed. The authors in[237] further optimized their designed multi-input logic gateswith modularity (i.e., having exchangeable inputs and outputsto increase the reusability) and orthogonality (i.e., no crosstalkwithin the host cell to increase robustness and stability). Forinstance, the proposed -input AND gate in [237] can notonly be rewired to different input sensors to drive variouscellular responses, but can also show the same functionalityin different types of cells. It is noted that multiple logic gatescan be combined to realize much more complicated cellulartasks, such as multicellular biocomputing [236] and the edgedetection algorithm [241].Many synthetic analog circuits have also been proposed.One example is the wide-dynamic-range, positive-logarithmcircuit [242], which consists of a positive-feedback componentand a ‘shunt’ component, demonstrating an ln( + m ) input-output transfer characteristic for a scaled input concentration m . A comprehensive review of different analog circuitsis provided in [243]. An intuitive way to understand thedesign of analog circuits is to interpret the synthetic processas tuning the behavior or response curve of a biologicalcomponent. In particular, the Hill function (introduced forLevel 2 in Section IV-D1) provides a semi-empirical approachin capturing the desired response curves [244]. For example,in [35], the parameters of the Hill function were optimizedto tune the relationship between the temporal change of theoutput protein and the input transcription factor as close aspossible to a hyperbolic tangent and a logarithmic function.Integrating analog circuits with digital circuits is a strat-egy to achieve more complicated computations. A digitallycontrolled logarithm circuit was designed in [242], wherea positive or negative logarithm circuit is connected to adigital switch. This combined circuit achieves a positive ornegative logarithm function in the presence of the inputinducer IPTG/AraC, whereas it shuts OFF in the absence ofthe inducer. C. Realizing Communication Functionalities
The digital and analog circuits realized either by chemicalreactions or synthetic biology provide the communicationcommunity with novel tools for processing chemical signals. A buffer gate can maintain the input and output logic relationship, andcan be regarded as a delay gate. In the following, we review some theoretical circuit designsthat enable modulation-demodulation and coding-decodingfunctionalities.
1) Modulation & Demodulation Functionalities:
For con-centration shift keying (CSK) modulation and demodulation,binary CSK (BCSK) and quadruple CSK (QCSK) realizationswere presented in [88] and [232], respectively. The BCSKtransmitter designed in [88] was capable of modulating arectangular input signal representing bit-1 as a pulse-shapedoutput, where the involved chemical reactions were directlyinspired by the I1-FFL discussed for Level 2 in Section IV.The corresponding receiver used an amplifying reaction tooutput a rectangular signal if the received signal exceededa threshold. For the QCSK modulation and demodulation in[232], the transmitter design was inspired by the electric 2:4decoder that activates exactly one of four outputs according toa combination of two inputs. As an electric 2:4 decoder can beeasily implemented using logic gates, the QCSK transmitterused the chemical reactions-based AND and NOT gates tomodulate two inputs to four different concentration levels.At the receiver side, three detection modules proposed in[88] with different thresholds were connected with two ANDgates and an XNOR gate to achieve QCSK demodulation.In addition, the demodulation of rectangular signals havingidentical durations but different concentrations was analyzedin [245], where the demodulator was based on the maximuma-posteriori probability (MAP) framework and can be imple-mented by several chemical species and reactions found inyeast.In addition to the CSK modulation scheme, chemical cir-cuits have also been applied to implement other modulationschemes, such as frequency shift keying (FSK), molecular shiftkeying (MoSK), and reaction shift keying (RSK). The realiza-tion of binary FSK (BFSK) demodulation was investigated in[246]. With two symbols encoded with different frequencies,the BFSK receiver consisted of two branches of enzymaticreaction circuits, which is analogous to the design of anelectric BFSK decoder. The parameters of the two brancheswere carefully selected according to the transmitted symbolsso that each symbol could only trigger one branch. ForMoSK, the receiver architecture was presented in [247], wherechemical reactions were exploited to determine if the samplednumber of bounded signaling molecules exceeded a predefinedlevel. For RSK, different chemical reactions were exploited formodulating transmission information into different signalingmolecule emission patterns [248]. To demodulate RSK signals,the authors in [248] investigated two types of ligand-receptorbased chemical circuits, and demonstrated the positive impactof feedback regulation on symbol error rate reduction. Theamount of information transferred by chemical reactions-based transceivers was quantified in [249], where optimaltransmitter circuits that maximize the mutual information ofthe whole communication link were derived for four types of In modulation schemes for wireless communication, “Q” usually stands for“quadrature” and refers to phases, e.g., quadrature phase shift keying (QPSK)modulation. However, in MC, “Q” usually stands for “quadruple” and refersto four concentration levels. receiver circuits (i.e., ligand binding, degradation, catalytic,and regulation reactions).An engineered bacteria-based biotransceiver architecturewith modulation and demodulation functionalities was pro-posed in [34]. In this architecture, the transmitter employeda modulator to realize M-ary amplitude modulation, and wascapable of generating a transmitted signal via a transmissionfilter; the receiver first processed a received signal via thereceiver filter with low-pass filtering characteristic to reducenoise and then used the demodulator to decode transmitted bitsequences.
2) Coding & Decoding Functionalities:
Classic codingschemes have been studied for MC to improve the reliabilitybetween communication links. A uniform molecular low-density parity check (LDPC) decoder to retrieve transmittedinformation from received signals was designed in [250]with chemical reactions. To execute the belief-propagationalgorithm, a chemical oscillator was introduced to schedulethe iterative message passing and trigger corresponding com-putations in each phase. The proposed LDPC decoder designis flexible and can deal with arbitrary code lengths, code rates,and node degrees.A transceiver design with single parity-check (SPC) en-coding and decoding functionalities was developed in [35]using both chemical circuits and genetic circuits. The proposedtransmitter is able to generate a parity check bit and modulatethe corresponding codeword with CSK, and the proposedreceiver acts as a soft analog decoder that calculates the a-posteriori log-likelihood ratio of received noisy signals toretrieve transmitted bits. During the aforementioned processes,chemical reactions are used to realize degradation, subtraction,and storage, while engineered gene expression processes areemployed to implement some complicated operations, suchas amplification, the hyperbolic tangent function, and thelogarithm function.
D. Microscale Storage
To end this section, we elaborate on the physical structureof DNA and the potential for DNA as a storage mechanismfor synthetic systems. A DNA molecule is comprised of twoantiparallel chains (DNA chains or strands), each composedof nucleotide subunits [48, Ch. 4]; see Fig. 18. Each subunitcontains one of 4 bases: adenine (A), cytosine (C), guanine(G), and thymine (T), and it is common to use the nameof the base to label an entire nucleotide subunit. Knowingthe bases along one chain is sufficient to know the sequencealong both chains, because an adenine subunit is always pairedwith a thymine subunit in the other chain, and cytosine isalways paired with guanine. The chemical properties of thechains mean that they arrange in a “double helix” shape, whichperforms one complete turn for every 10 base pairs. The mainskeleton of the chemical structure of each base consists ofone or two carbon rings, with their carbon atoms denotedwith a primed number from 1’ to 5’. Depending on where theconnection with the next base occurs, it is typical to describethe 5’ and 3’ ends of a DNA strand.RNA is synthesized from DNA (in a process called tran-scription, proceeding from the 5’ end to the 3’ end; see Fig. 18: Antiparallel DNA strands consisting of combinations of thefour bases: adenine (A, green), cytosine (C, light blue), guanine (G,black), and thymine (T, red). The pairing up of the bases alongopposite strands is very specific (A with T and C with G) andthis facilitates reliable replication. The primed numbers indicate thedirection of each strand, as transcription always proceeds from the5’ to 3’ end. Source: [251].
Section IV-D) and produces a single strand [48, Ch. 6]. RNAis also composed of nucleotide subunits, but its bases areadenine (A), cytosine (C), guanine (G), and uracil (U) insteadof thymine.Utilization of DNA as an information storage medium isperhaps the most mature and promising of the applicationsof DNA in MC (for recent reviews see [252], [253]; we alsodiscuss DNA storage as a case study in Section VIII-C). Cur-rent magnetic drives are close to the limit of that technologywith a storage capacity of 1 TB per square inch [254]. On theother hand, DNA has a maximum storage density of 2 bits pernucleotide. This means that each gram of single-stranded DNAhas a theoretical maximum storage capacity of 455 exabytes.With such a technology, only 4 grams of DNA would berequired to store the man-made digital information producedglobally in 2016, including newspapers, books, and internetsites [255]–[257]. However, in practice, the actual storagecapacity of DNA is lower than the theoretical maximum.GC base pairs form one more hydrogen bond than AT. Thisresults in a different melting temperature because GC bondstake more energy to break. Thus, the replication efficiency ofDNA by polymerase chain reaction (PCR; discussed further inSection VIII-C) varies depending on the ratio of GC to AT inthat sequence which affects DNA synthesis and informationretrieval [258]. Similarly, repetitive occurrences of the samebase (e.g., sequence AAAAAAAA) introduce errors duringsequencing. These factors, together with natural DNA decay,mean that many copies of the same sequence are requiredfor effective storage of information and this constrains storagecapacity.It has been shown that systems of DNA data storage have anerror rate close to 1 % [259], a number comparable to currentmagnetic media. Errors are mainly produced during the writingand reading processes, so some redundancy with duplicateddata is needed for reliable information retrieval. In naturalsystems, DNA polymerase II proceeds at a speed of about70 bases per second [260], although actual DNA transcriptionhappens at less than half of that rate, around 30 bases persecond [261], mainly due to polymerase activity pauses. State- of-the-art engineered systems can achieve a writing-to-retrievalperiod of approximately 21 hours as demonstrated recentlyin a proposed end-to-end system [262]. However, one of thekey advantages of DNA storage lies with the large capacityfor parallelization, e.g., any DNA sequence can be easilyreplicated into millions of copies in a short run of PCR.VII. L EVEL
PPLICATIONS
When communication is used to send information, it is ameans to an end. Organisms would not have evolved to engagein the costly activities to send and receive signals if therewere no clear benefits of doing so. Thus, the top level ofour proposed hierarchy (see Figs. 1 and 3) is the applicationlevel, which defines and describes the behavioral interactionsbetween communicating devices. These interactions could becompetitive or collaborative (e.g., a predator-prey dynamicversus coordination within a cooperative population). They canalso apply over very different physical scales, e.g., within anindividual cell, between a pair of cells, across a populationof cells, between different species or kingdoms, or over amacroscopic-microscopic interface.Interactions between communicating devices can either benatural or artificial. Since we provide detailed examples ofapplications in natural systems in the subsequent section oncase studies (i.e., Section VIII), in this section we focus onsynthetic cell biology applications. Specifically, we select twopromising applications of MC to demonstrate Level 5: biosens-ing and therapeutics . To reveal how these two applicationsrely on all of the lower levels of the proposed hierarchy,we use boldface font to refer to previously-covered topicsin this survey and summarize these mappings in Table V.However, as indicated in this table, the potential for local datamanipulation at Level 4 has not been fully explored in thecurrent literature on these synthetic applications. Thus, at theend of this section, we present an envisioned automatic drugdelivery system to demonstrate how MC system design mightbe applied to improve the state of the art of these cell biologyapplications. In this way, this section not only demonstrateshow the components described at each level have already beenutilized in biosensing and therapeutics, but also reveals promis-ing collaboration opportunities for researchers from differentcommunities, especially for those in the communication andsynthetic biology fields. In particular, we describe how thesetwo representative applications can be realized or facilitatedwith the tools from the microfluidics community discussed forLevel 1 (in Section III) and the synthetic biology communitydiscussed for Level 4 (in Section VI). In addition, this sectionincludes applications of therapeutics based on magnetic fieldsas this macroscale control technique can reshape drug deliverysystems and other in vivo applications.A microfluidic device processes and manipulates smallamounts of fluids using channels in dimensions of tens to hun-dreds of micrometers (i.e., − ∼ − litres). Advantages ofmicrofluidic systems include rapid analysis, high performance,design flexibility, and reagent economy [267]. Synthetic biol-ogy lies at the intersection of engineering, biological sciences,and computational modeling. It borrows tools and concepts TABLE V: Application Examples.
Application(Reference) Section III: Level 1Signal Propagation Section IV: Level 2Device Interface Section V: Level 3Physical/Data Interface Section VI: Level 4Local DataBiosensing viaMicrofluidics([263]) Diffusion-BasedPropagation (III-A);Advection-Diffusion-BasedPropagation (III-B);Chemotaxis (III-D2) Gene Expression (IV-D1) Optical Microscopy (V-B1) MC-AssistedApplications (VII-C)Biosensing viaSynthetic Biology([264]) Diffusion-BasedPropagation (III-A) Gene Expression (IV-D1) Optical Microscopy (V-B1) Digital/Analog Circuits viaSynthetic Biology (VI-B2)Therapeutics viaMicrofluidics([265]) Advection-Diffusion-BasedPropagation (III-B) Molecule Reception andResponses (IV-B)
In Vivo
Imaging (V-B3);Macroscale MechanicalControl (V-C5) MC-AssistedApplications (VII-C)Therapeutics viaSynthetic Biology([42]) Diffusion-BasedPropagation (III-A) Molecule Reception andResponses (IV-B);Gene Expression (IV-D1) Macroscale Observations ofMicroscale Phenomena (V-B) Digital/Analog Circuits viaSynthetic Biology (VI-B2)Therapeutics viaMagnetic Field([266]) Advection-Diffusion-BasedPropagation (III-B) Molecule Reception andResponses (IV-B)
In Vivo
Imaging (V-B3);Macroscale MagneticControl (V-C6) MC-AssistedApplications (VII-C) from these disciplines to engineer non-existing biologicalsystems or to redesign existing systems to achieve user-definedproperties [268]. Over the past few decades, microfluidics andsynthetic biology have proven their potential as tools that offerunprecedented solutions for biosensing and therapeutics.
A. Biosensing
Biosensors are devices used to detect the presence ofchemical substances. A biosensor is normally composed ofa bioelement and a transducer [269]. The bioelement en-ables microscale detection by binding an analyte of interest,and the transducer modulates the variation of the analyte toan electrical or optical signal that can be observed at themacroscale. This application maps to Level 5 of our proposedhierarchy because communication occurs when we observeinformation encoded by the transducer, when the transducerreceives information from the bioelement, and possibly alsowhen biosensors communicate with each other. Biosensingplays a significant role in our daily life, and has been appliedto many fields from disease monitoring to pollutant detection.
1) Microfluidics:
Conventional biosensing methods areusually time-consuming and the corresponding equipmentis big and expensive. There is a need for biosensors withfaster analysis, higher cost-effectiveness, and smaller size[270], [271]. Microfluidic platforms have become realizable tomeet the above requirements. The authors of [263] proposeda chemical biosensing microfluidic chip based on bacterialchemotaxis. As shown in Fig. 19, the microfluidic chip consists of one middle channel and two side channels. Thebacteria are introduced in the middle channel, and the flowing (Level 1) buffer with and without attractant (‘source’ and‘sink’, respectively) are injected into the side channels. Theconnection between the middle channel and the side channelsonly enables the diffusion (Level 1) of attractant molecules, It is noted that the microfluidic architecture in Fig. 19 can also be usedfor biomedical research by including chemical reactions. One example is toemulate the scenario of oxygen-glucose deprivation to study stroke [272].
Microscope Glass Slide
PDMS
Attractant Bacteria Buffer (a) Schematic side view of the microfluidic sensing chip.
INOUTIN
OUT
INOUT
Attractant BufferBacteria
Bacteria Accumulation
Attractant Gradient (b) Schematic top view of the microfluidic sensing chip.
Fig. 19: Illustration of the microfluidic bacterial chemotaxis biosens-ing system [263]. and this can create a microscale concentration gradient. Asa response, the bacteria bias their motion towards the attrac-tant using chemotaxis (i.e., cargo-based transport guided bymolecule gradients; presented for Level 1). The signaling at-tractant can also activate the expression of a fluorescent gene (Level 2) embedded in bacterial cells so that the chemotacticintensity (i.e., the spatial distribution of bacterial cells) can bevisualized using fluorescent optical microscopy (Level 3). Ifa liquid sample that is taken from a natural environment (e.g.,a river) is injected into the attractant side channel, then thefluorescent intensity provides a tool for estimating the cells’ AND rbs30 rbs30 rbs31
ZraR
ZntR
Fig. 20: Zn + specific biosensor using an engineered AND logic gate,where P zraP and P zntA are cognate promoters for ZraR and ZntR,rbs and rbs are both ribosome binding sites, P hrpL is anotherpromoter that is activated when the genes hrpR and hrpS are bothexpressed, and gfp is the gene encoding a green fluorescent proteinthat works as a biosensor readout [264]. living conditions (i.e., information in cellular signals , pre-sented for Level 4). A higher attractant concentration leads toa stronger fluorescent intensity. The integration of chemotacticsensing in microfluidic chips enables rapid and quantitativesensing readouts, and the miniaturization of sensing devicesalso significantly reduces the power and reagent consumption.Microfluidic devices can also lower the cost and time ofDNA detection. The authors of [273] demonstrated a paper-based microfluidic device that combined DNA extraction,amplification, and antibody-based detection of highly specificDNA sequences associated with malaria infection. The devicewas able to produce results in the field with high sensitivity( > % ) in less than one hour. In addition, the manufacturingwas simple and cost effective, enabling production of thedevice in great numbers.
2) Synthetic Biology:
The selectivity of a biosensor de-scribes its ability to distinguish targeted molecules amongother similar chemicals. A biosensor with low selectivity canbe activated by targets with similar chemical properties. Forexample, this is a concern when detecting toxic heavy metalsfor water pollutant monitoring [264]. Examples of nonspecificmetal biosensors include the triggering of the regulator CadCin
S. aureus by cadmium, lead, and zinc, the regulator CmtRin
Mycobacterium tuberculosis by cadmium and lead, and theregulator ArsR in
E. coli by arsenic, antimony, and bismuth[274].From a communication engineering perspective, a generalsolution to increasing the selectivity is to endow biosensorswith more signal processing capabilities, and the engineered digital synthetic biological circuits reviewed for Level 4 canbe applied to build biosensors with increased selectivity [264].Fig. 20 shows the schematic design of a biosensor that isonly sensitive to zinc (Zn + ) but not to palladium (Pd + ) orcadmium (Cd + ). The sensor senses the targeted metal ionsthat diffuse (Level 1) in the extracellular environment and canphosphorylate their respective regulators (i.e., ZraR for Zn + and Pd + , and ZntR for Zn + and Cd + ). The transcriptionfactors ZraR and ZntR regulate the gene expression (Level2) of hrpR and hrpS , respectively, and the protein productsbecome the inputs of the engineered AND logic gate. Atthe AND gate, the expression of gene gfp is activated onlywhen both hrpR and hrpS are expressed. In this way, thereadout green fluorescent protein is driven by a single bit of Reservoir Cannula
Drug DeliveryDevice
CorneaRectus Muscles
Fig. 21: Schematic of a manually-actuated drug delivery device forchronic eye diseases [265]. information, i.e., the presence or absence of Zn + ( informationin cellular signals , presented for Level 4), and can be observedvia optical microscopy (Level 3). B. Therapeutics
Therapeutics is a discipline developed to treat and carefor a patient with the purpose of preventing and combatingdiseases or alleviating pain. Drug delivery systems play animportant role in therapeutics by controlling the release andadsorption of pharmaceutical compounds to achieve desiredtherapeutic effects. They map to Level 5 of our proposedhierarchy because communication occurs when we observechanges at the disease site (e.g., reduction in tumor size),when therapeutic agents bind with their receptors at diseasedcells, and when drug delivery devices receive informationfrom the extracellular environment or body-area stimuli. In thefollowing, we present some therapeutic drug delivery methodspowered by microfluidic platforms, synthetic biology, andmagnetic fields, which show improved therapeutic efficiencycompared with conventional systems, such as oral ingestionand intravascular injection.
1) Microfluidics:
Microfluidic systems are beneficial fornovel drug delivery applications by improving drug deliv-ery accuracy and reliability at reduced size [275]–[277]. Amanually-actuated drug delivery device for the treatment ofchronic eye diseases was developed in [265].As shown in Fig. 21, benefiting from microfluidic systems,the drug delivery device was miniaturized to allow its place-ment in the limited space within the eye. The device worksusing macroscale mechanical control (Level 3). More specif-ically, a pressure force, mechanically actuated by a patient’sfingers, can induce the advection-diffusion-based propagation (Level 1) of phenylephrine drugs from the reservoir to intraoc-ular tissues. The released phenylephrine molecules undergo ligand-receptor binding (Level 2) to adrenergic receptors andfinally lead to a temporal change in pupil size, which canbe measured via in vivo imaging (Level 3). The observationof pupil size changes not only demonstrates successful druginjection, but also indicates that the observed output of pupilsize can be manipulated by the phenylephrine concentration(i.e., information in cellular signals , presented for Level4). This device is refillable, such that only one surgicalintervention and the associated pain is needed. In contrast, dueto the presence of the blood-retina barrier, conventional oralmedications require large doses in order to reach therapeutic levels and can have serious negative side effects. Traditionalintraocular injections for chronic diseases require frequentinjections, which can induce trauma in ocular tissues [265].
2) Synthetic Biology:
Designing and engineering biologicalparts via synthetic biology has enabled novel therapeuticplatforms to target specific pathogenic agents and pathologicalpathways [278]. Cancer involves abnormal cell growth andproliferation with the potential to invade nearby healthy tissueand spread to other organs. A significant shortcoming ofcurrent cancer therapies is that cancerous cells are difficultto distinguish and remove from surrounding healthy cells.One potential solution is to synthetically link the invasin( inv ) gene (from
Yersinia pseudotuberculosis ) with the fdhF promoter. The reason for this synthesis is that tumor mi-croenvironments are low in oxygen and the fdhF promoter isstrongly expressed in such an environment. Thus, the invasinproteins become controlled to only effectively express in theoxygen-deprived environment ( gene expression described forLevel 2). The invasins diffuse (Level 1) within the cellularmedium and can bind with the β -integrins distributed onthe surface of cancer cells ( molecule reception describedfor Level 2), which triggers the internalization of bacteriainside cancer cells. The invasion ability can be quantifiedand observed by macroscale instruments (Level 3) after agentamicin protection assay. Furthermore, the authors of [42]also synthetically linked the invasion of cancer cells to bacteriadensity, which is achieved by placing the inv gene under thecontrol of a lux quorum sensing system (we describe quorumsensing in further detail in Section VIII-A). Hence, bacteriainvasion of cancer cells is driven by two bits of information,i.e., the oxygen level and the bacteria density ( information incellular signals , presented for Level 4). This can be interpretedas an application of the genetic AND gate (Level 4) that inte-grates multiple inputs to achieve more accurate environmentalsensing. This characteristic makes invading cells ideal carriersto release therapeutic agents to enhance tumor treatment.
3) Magnetic Field:
To prevent drug absorption or degra-dation before reaching the affected target sites, one efficientapproach is to place the drug as close as possible to the targetsites. It has been demonstrated that accurately manipulatingmagnetic microrobots via magnetic fields is feasible, and thehuman body is ‘transparent’ to magnetic fields (i.e., in termsof biocompatibility and safety). Motivated by this, the use ofmagnetic microrobots for drug delivery through macroscalemagnetic control (discussed for Level 3) has been widelystudied and applied [279]–[281]. In [282], a microrobot wasinjected into the posterior area of a rabbit eye. Once an externalmagnetic field was applied, the injected microrobot couldachieve rotational and translational mobility, thus presenting anopportunity for ocular drug delivery. In [266], mitoxantrone-loaded magnetic nanoparticles were injected into the femoralartery. By applying an external magnetic field above thetumors implanted in the limb of rabbits, nanoparticles movedtowards the tumor region via advection-diffusion-based prop-agation (Level 1). A higher accumulation of mitoxantrone wasfound near the tumor region, and a clear reduction in tumorsize could be observed through in vivo imaging (Level 3) asmitoxantrone is able to bind with the DNA of tumor cells, thus halting tumor growth and division ( molecule receptionand responses described for Level 2). Therefore, cancer celldifferentiation is controlled via the reception of mitoxantronesignals ( information in cellular signals , presented for Level4).
C. MC-Assisted Applications
In the following, we envision an MC-enabled automaticdrug delivery system, with the aim to illustrate how MCcould facilitate and enhance the aforementioned biosensingand therapeutics applications. An automatic drug deliverysystem largely reduces the dependency on manual operationsand should be composed of a biosensor and an actuator.The biosensor senses the extracellular environment (e.g., theconcentration of glucose), and could be connected with theactuator including drug reservoirs to cooperatively support thedrug regulating mechanism. Nevertheless, the biosensor andthe actuator, such as the microfluidic biosensor in [263] and themicrofluidic actuator in [265], are often designed separately byresearchers from different fields and are likely to be physicallyisolated. Thus, the communication between a biosensor and anactuator is of great importance because it is the only featurethat enables them to work in a synchronous and cooperativemanner to reach a common goal. To address this issue, MCcan be used to establish a point-to-point communication linkbetween the two of them. In this scenario, the biosensor wouldserve as a transmitter, and the actuator would function as thecorresponding receiver. Once the biosensor detects a relevantphenomenon, it modulates this information to a chemicalsignal that can be received and demodulated by the actuator.As a response, the actuator releases drug molecules to aspecific area of cells.On some occasions, the actuator may be controlled by morethan one bit of information (i.e., the presence or absenceof a phenomenon), implying the involvement of multiple biosensors. In this sense, the signal processing capability ofan actuator should be expanded accordingly to manipulatesignals received from multiple biosensors. One example is theintroduction of digital logic gates, as in [264] and [42], to con-trol the drug-regulating mechanism. Moreover, this envisioneddrug delivery system can be further optimized by integratingother MC-based concepts. For example, the implementationof coding functions at Level 4 could be added to mitigatethe effects of noise, thus providing a more reliable and robustcommunication link.VIII. E ND - TO -E ND C ASE S TUDIES
From Sections III to VII, we individually discussed andpresented examples for each of the five levels of the proposedcommunication hierarchy. While we drew connections be-tween the levels, we did not directly apply the entire hierarchyto any one example. In this section, we have selected severalprominent exemplary biological systems as case studies fora complete mapping to the proposed hierarchy, as summa-rized in Table VI. In particular, we present quorum sensingby bacteria (Section VIII-A), signaling within and betweenneurons (Section VIII-B), and information encoding in DNA (Section VIII-C). Quorum sensing is an example that alignsclosely with diffusion-based MC. Neuron signaling includes amix of diffusion-based and action potential wave propagation.While the propagation of DNA information has been less ofa focus of study in the MC community, its implementation ofthe higher levels is widely known and well understood and soit provides a useful supplement. While all three of these casestudies can map to the entire hierarchy as natural systems,they also demonstrate opportunities for synthetic interactionsincluding control. A. Quorum Sensing
The classical view of bacteria depicts them as individualorganisms that act independently as isolated entities. Whilethis is true to the extent that a bacterium is a distinctautonomous cell, we have known for a few decades thatbacteria can form groups comprised of many individuals thathave been shown to exhibit coordinated behavior [283]. Thisincludes bioluminescence (one of the first collective microbialbehaviors to be characterized) [284], biofilm formation [285],production of virulence factors and secondary metabolites[286], and induction of competence for foreign DNA uptake[287]. These processes are made possible by communicationbetween bacteria via a process termed quorum sensing (QS).A recent review of QS can be found in [216] and a visualsummary is provided in Fig. 22. QS relies on the exchangeof small extracellular signaling molecules called autoinducers .Exchanging signals in this way enables bacteria to assimilateinformation conveyed by different types of autoinducers tocontrol specific genes. This enables communication betweenthe same and distinct species and even between bacteria andanimal cells [288].QS is very energy efficient as a communication system.Signaling molecules are based on intermediates that have a keyrole in the central metabolism [289], [290]. Thus, investmentin a specialized production chain is not required and thehigh affinity and selectivity of the molecules means that avery small amount is sufficient for effective communication,resulting in a very small production cost. Cost effectivenessis further improved by the fact that many QS molecules canserve multiple purposes. Examples include photopyrones, asmall QS molecule in
Photorhabdus luminescens that in highconcentrations can act as an insect toxin [291], or dialkylre-sorcinols (DARs) that can act as an antibiotic [292].Understanding biological systems is inherently complicated,with many components often serving multiple functions inhighly-interconnected networks that make separation of func-tionality into layers a challenging task, and QS is no different.From the perspective of our proposed hierarchy, Level 1 andLevel 2 are the diffusion of molecules and the mechanisms forthe release and reception of autoinducers, respectively, as listedin Table VI. Subsequent levels are less intuitively defined.We propose concentration threshold detection as Level 3, andfunctions such as estimation of cell density and individualsswitching behavior as Level 4. Level 5 describes featuresthat emerge across the bacteria population using QS (e.g.,coordinated behavior, cooperation, eavesdropping). (a) (b)(c)
Fig. 22: Simplistic representation of quorum sensing in
Vibrio fis-cheri . Autoinducers (in green) produced by LuxI gene are excreted bythe cell and accumulate in the environment. (a) At low cell densities,autoinducer concentration is also low. (b) Higher cell densities causean accumulation of molecules. (c) Accumulated molecules can besensed by the cell and activate the Luciferase genes that inducebioluminescence.
1) Level 1:
In QS communication, Level 1 is the random diffusion of the autoinducers in the environment. Diffusion hasbeen summarized mathematically in Sections III-A to III-C ofthis survey. Bacteria that have formed a biofilm merit separatediscussion, as fluid flow is non-existent or very restrictedinside their extracellular matrices [293]. High cell densitiesinside a biofilm also have a significant effect on both thediffusion distance of a molecule, and the speed at whichdiffusion occurs. Experimental and theoretical studies havedetermined the reduction of the diffusion coefficient withinbiofilms to be between 0.2 and 0.8 when compared withdiffusion in water [294]. The distance that a molecule cancover while diffusing through a biofilm is effectively given bythe dimensions of the biofilm cluster [294].
2) Level 2:
Level 2 concerns the mechanisms of releaseand reception of autoinducers. QS signaling molecules differbetween bacterial types. Gram-negative bacteria typically useacyl-homoserine lactones (AHL) as autoinducers. AHLs aresmall molecules that can freely diffuse through membranes.The system depends on two proteins, LuxI and LuxR (seeFig. 22(c)). LuxI helps in the synthesis of the autoinducerN-3-(oxo-hexanoyl)-homoserine lactone (30CC6HSL) familyof proteins [295], [296]. To describe the system here, werefer to individual proteins of a specific QS system (
V. fis-cheri ), although we clarify that LuxI, LuxR, and 30CC6HSLare members of protein families found among all gram-negative bacteria, with each bacterial species carrying theirown version. After the AHL is synthesized, it diffuses freelythrough the cell membrane in both directions, and its concen- TABLE VI: Case Study Summary.
Case StudyName(Subsection) Section III: Level 1Signal Propagation Section IV: Level 2Device Interface Section V: Level 3Physical/Data Interface Section VI: Level 4Local Data Section VII: Level 5ApplicationQuorum Sens-ing (A) Diffusion of autoinducermolecules Release, capture, and detec-tion of autoinducers Measuring threshold con-centration(s) (e.g., high ver-sus low) Behavior based on estimateof local population Bacteria cooperation andcoordination; eavesdroppingand surveillanceNeuronalSignaling (B) Action potential along neu-ral axon; neurotransmittersacross chemical synapse Membrane potential changesdue to ion channels; releaseand capture of neurotrans-mitters Spike timing and frequency Messages to transmit nervestimuli and motor actions Functioning of nervous sys-temCommunicationvia DNA (C) Storage in genome Transcription and translationof genes; replication of DNA Controlling gene regulation;modulation with 4 bases (A,C, G, T for DNA; A, C, G,U for RNA) Encoding of amino acids forproteins Life (e.g., cell growth, divi-sion, differentiation) tration rises as the microbial population increases [297] (seeFigs. 22(a), 22(b)). LuxR is the receptor for 30CC6HSL in thecytosol, as well as the transcriptional activator of the luciferaseoperon [295], [298]. The role of 30CC6HSL is to stabilizeLuxR (otherwise LuxR naturally degrades rapidly) and enableit to persist long enough to recognize and bind to a consensussequence encoding for luciferase and accessory proteins [299],[300] (see also the general discussion on gene expression inSection IV-D1). Stabilized LuxR also activates LuxI in a feed-forward loop so, when the QS system is engaged, productionof autoinducers accelerates and the environment is filled withthe signal molecule.Gram-positive bacteria have a different membrane structurethat is impermeable to AHLs. They instead use oligopeptides(also referred to as autoinducing peptides) as autoinducers andthese are actively transported to and from the cell surface.Reception of these molecules is based on cell surface-boundreceptors collectively termed two-component signaling pro-teins that upon activation set up a series of reactions insidethe cell (i.e., a signal cascade ) [297].Recent findings suggest that for all types of bacteria, largermolecules such as hydrophobic AHLs that cannot pass throughthe membrane are instead being released in the environmentvia membrane vesicles [301], [302]. The size range of thesevesicles appears to be between 40 and 500 nm [303], [304]. Inaddition to QS, these vesicles have been shown to be involvedin horizontal gene transfer for the exchange of virulencefactors. A review of QS vesicles can be found in [305].
3) Level 3:
In QS circuits, the default mode is the continu-ous expression of the gene that encodes the required behavior(e.g., luminescence, virulence). In a typical arrangement forbiological systems, this gene expression is suppressed by therapid degradation of the mRNA molecules transferring theinformation for the corresponding protein production. Thus,in low cell densities the product cannot be synthesized asthe mRNA is not produced or is readily degraded. Whenautoinducers in the environment reach a critical (i.e., thresh-old) concentration, this suppression ceases and the mRNAare able to reach their target destination [306]. Thus, thedesired behavior of a QS system is digital, in the sense thatQS acts as a sensing mechanism that regulates the transitionfrom one behavior to another. By implementing a thresholdmechanism, a cell is able to quantify autoinducer concentrationand translate it into information to infer cell density and the presence of other species.Function in biological systems is usually tied to physicalstructure, so in this discussion of threshold measurements itis also appropriate to outline the mechanisms that implementthese measurements, as the way each system is implementedmirrors its behavior and suggests how it might be controlled.QS systems differ in the arrangement of their internal com-ponents, reflecting different needs in their implementationof signal quantification. For example, there are QS systemscontaining circuits that act in parallel or in series, others withdifferent system components acting in opposition, and alsosystems that upon activation confer a permanent change tothe organism [297], [306]. In the following, we summarizeparallel and series implementations to emphasize the diversityof microscale signal operations in QS.In a parallel QS architecture, it is typical to implement morethan one autoinducer in separate signal transduction cascades.Because they all have the same result (e.g., suppression ofthe mRNA suppressor), their signals reinforce one another. Inaddition, the need for the simultaneous presence of two (ormore) signals for the activation of the system ensures thatspecific requirements are met (e.g., availability of nutrients,presence of another species), not unlike an AND gate. Thisparallel sensing approach might be helpful for noise reductionand to filter foreign signal-mimicking molecules [306].Series QS circuits differ in the sense that activation ofone circuit is required for the activation of the subsequentcircuits. This is the mechanism
P. aeruginosa employs forvirulence. Experiments have shown that unlike what happensin a parallel system, some genes in these circuits may beexpressed in response to one autoinducer only, while othersrespond to any of the legitimate signals, and yet others requirethe simultaneous presence of all signals for their expression.In addition, their activation occurs at different times duringcell growth, an indication that timely ordered gene expressionis very important for these organisms [306], [307].A number of techniques are currently used to observe QSbehavior, depending on the nature of the expected microbialresponse to QS signals. Microbial antibiotic assays can beemployed for the detection of antimicrobial agents secreted bybacteria in response to the presence of other organisms.
Fluo-rescence microscopy coupled with microfluidics is a suitabletool for the observation of gene expression at the level of anindividual cell [75], [308]. Detection and real-time tracking of autoinducers has been achieved using bacterial reporterstrains [309], high performance liquid chromatography [310],and nanosensors [311].
4) Level 4:
Autoinducer release, random diffusion outsidethe cell, and subsequent detection provide bacteria with a toolfor the estimation of presence and density of microorganismpopulations around them, such that behavior can update oncethreshold conditions are observed. The QS processes enableprecise regulation of a large number of genes and the fine-tuning of responses, e.g., input-output range and dynamicbehavior, synchronization, and noise control [312]–[315]. Inlarge microbial populations (particularly in biofilms), thereare inevitable variations in each individual cell’s local envi-ronment due to differential access to resources, accumulationof metabolic byproducts in pockets, and oxygen penetration.Thus, the QS behavior of a bacterial population is not nec-essarily homogeneous. For example, in a biofilm, oxygenpenetration is slow, creating a gradient from the outside tothe center. The cells at the periphery then sense a completelydifferent environment than those further inside the matrix,leading to variable individual responses that are essential tomaintaining the biolfilm. Recent evidence also suggest thatsome microbial populations can exhibit a stochastic expressionof QS genes, resulting in segments of the population being indifferent QS modes [75], [308].
5) Level 5:
The highest level in the hierarchy correspondsto aggregate behavior, e.g., QS stimulating coordination be-tween bacteria. This is of great interest to the scientificcommunity as the results of coordinated microbial actionshave major economic importance (e.g., biofilms, virulence,biofouling). For example, model estimates in 2013 predictedthe economic burden of antibiotic resistance on global GDPto be between US$14 billion and US$3 trillion by 2050 [316].Naval biofouling by barnacles is initiated by microbial matsthat enable barnacles to attach to a ship hull. This adds aconsiderable annual cost that can reach a few million US$ per vessel due to the subsequent increase in drag [317]. Theseprocesses all rely on the coordination between microorganismsrealized through the exchange of molecular signals. As anexample of inter-species cooperation, consider the QS systemused by the gram-negative bacterium
Vibrio fischeri . It is thecanonical example of a QS system in gram-negative bacteriaand was also the first to be described during an investigationof bioluminescence in the Hawaian Bobtail squid
Euprymascolopes [284]. Favorable conditions inside a specific organof the squid allow
V. fischeri to reach high cell densities, andthrough the activation of a QS system to induce the expressionof the luciferase operon. The light produced benefits the squidhost by providing protection from predators [284].There are several interesting communication security ap-plications and problems that can be observed in QS sys-tems. Since the autoinducers released can be unique for eachspecies using QS, different bacteria can send signals thatonly individuals of the same species will detect. This canestablish a secure communication channel, although it canbe compromised when other species are able to detect thesame signals (i.e., “eavesdropping”) [76]. For example, thesoil bacterium
Myxococcus xanthus ( M. xanthus ) is a predatory species that actively seeks other bacteria as prey.
M. xanthus is able to detect a range of QS molecules used by differentgram-negative bacteria, which enable it to infer the presenceand direction of many species [318]. Some QS signals can bedeliberately detected by a number of different species to enableinter-species communication. This function in gram-negativebacteria relies on variations in the structure of the autoinducermolecules. Gram-positive bacteria exert more control on thefinal structure of the peptides used as signal molecules, asthese are DNA-encoded, resulting in a unique genetic sequencefor each organism [297].Microbes have the ability to attach to surfaces and formbiofilms [319]. Examples include plaque in teeth and rock-coating slime in water. It has become increasingly apparentover the last couple of decades that biofilm communities arethe predominant form of microbial life and that they are ofgreat importance to medicine, industry, and the environment[40], [320], [321]. Biolfilms help microbes to engage insymbiosis with other species, avoid predators, and be shieldedfrom antibiotic compounds. Biofilms are also very dynamic;they can have significant heterogeneity within a populationand also change behavior depending on the conditions (e.g.,nutrient availability) [75], [322]. In the canonical example of
V.fischeri , it is reported that although QS signals are flooding theenvironment, there can be a significant variation in the level ofbioluminescence between individual microbes [323]. QS playsa key role in biofilm formation, and bacterial species can havediverse biofilm-forming strategies. For example,
Pseudomonasaeruginosa creates biofilms when cell density is high, while
Vibrio cholerae and
Staphylococcus aureus form biofilms atlow cell density [286], [324]. In the latter cases, autoinduceraccumulation suppresses the excretion of biofilm molecules.Another example of physiological activity regulated by QSis the releasing of virulence factors to destroy tissues in targethost cells during the initiation of an infection [325]. Thesynthesis and secretion of virulence factors are expensive andthey are needed in a large quantity to successfully attack a host.Thus, since autoinducer molecules are less expensive, QS isused to regulate the expression of virulence factors in bacteriaso that they are produced only once the bacterial populationdensity is sufficiently high.
B. Neuronal Communication
Neurons are important cells for storing and processinginformation in most animals. In order to swiftly carry in-formation throughout the body, they require very rapid andreliable communication mechanisms. Neuron signaling is aninteresting example of microscale signal propagation becausethe physical dimensions and performance requirements ofneurons demand a diversity of propagation techniques bothwithin and between neurons and other cells. Since neuronscan be extremely elongated (see Fig. 23(a)), signaling within neurons is as important as signaling between neurons. Neuronshave branching dendrites around the cell body ( soma ) toreceive inputs from other neurons, and usually one long axon to signal outputs (at terminal branches) to distant targets(including other neurons). Axons in vertebrates are typically (a) (b) Fig. 23: Schematic drawing of a neuronal synapse and neuronalchemical signaling. (a) Main parts of a neuronal cell, showing theconnection of an axon to the dendrites of another neuron. (b) Acalcium wave propagating along the axon triggers the opening of cal-cium channels in the pre-synaptic neuron’s outer membrane. Increasein Ca + concentration causes the brief release of neurotransmittermolecules, before their rapid re-absorption by both the presynapticneuron and adjacent glial cells. Released ions bind to and activatereceptors in the postsynaptic neuron, leading to an influx of ions intothe second neuron, and to the propagation of the signal. from less than mm to more than m in length [48, Ch. 11].The contact sites between neurons are known as synapses . Themost common modality are chemical synapses [48, Ch. 11](see Fig. 23(b)), which are uni-directional, though there arealso bi-directional electrical synapses and there is evidencethat chemical and electrical synapses functionally interact witheach other [326].To apply the proposed communication hierarchy, we rec-ognize the dichotomy of signal propagation both within andbetween neurons to identify two distinct implementations ofLevel 1 and Level 2 behavior. As summarized in Table VI,the propagation of an action potential spike along the axoncarries information within a neuron, whereas neurotransmittermolecules diffuse across a chemical synapse to carry informa-tion between neurons (or other cells connected to neurons).However, these implementations merge at Level 3, since bothtypes of signals carry information in the timing of actionpotential spikes. Level 4 concerns the information in neuronalsignals and Level 5 describes where and why these signals areused.We have already discussed neuronal communication in sev-eral instances throughout this work. We have briefly mentionedneurons as examples in the context of calcium signaling inLevel 1 (Section III), storage and release of neurotransmittersin Level 2 (Section IV), and the speed of synaptic responsesin Level 3 (Section V). The reader may re-visit the detailsin those sections to supplement the holistic discussion of this case study.
1) Level 1:
The signals within neurons are changes in thelocal electrical potential across a neuron’s plasma membrane.Active mechanisms are needed to amplify signals in largerneurons so that the signals can propagate along the axonwithout attenuation. The active mechanisms create a travelingwave known as an action potential, which can propagate atspeeds of
100 m / s or more. The primary components arevoltage-gated ion channels (e.g., Na + and K + ), which openwith positive feedback (to trigger the opening of neighboringchannels) and then close with a refractory period (to preventrepetitive firing so that the wave travels along the axon). Thefirst contributions to quantitatively model the propagation ofaction potentials in neurons were by Hodgkin and Huxley[327], who treated the membrane as an electrical circuit withvariable conductances due to the transfer of Na + , K + , andother ions. Thus, the propagation of an action potential spikecan be modeled as an electrical transmission line using thecable equation [92, Ch. 17].The signals across chemical synapses, as shown inFig. 23(b), are more similar to the reaction-diffusion processesthat we described for Level 1 in Section III. The synapse hasthe transmitting neuron at the pre-synaptic side and the re-ceiving neuron (or other cell, e.g., muscle) at the post-synapticside. While there is significant diversity in the components andprecise function of chemical synapses, they generally signalby releasing neurotransmitter molecules that diffuse acrossthe synapse to the post-synaptic neuron’s outer membrane.There are many different types of neurotransmitters; commonones include acetylcholine, glutamate, serotonin, glycine, and γ -aminobutyric acid [48, Ch. 11]. Since chemical synapsesare quite narrow (only – nm wide [328, Ch. 12]), thisdiffusion process is fast. Nevertheless, signaling pathwayswithin the cleft provide mechanisms to destroy neurotransmit-ters, recycle them via re-uptake by the pre-synaptic neuron,or remove them via re-uptake by glial cells [48, Ch. 11].Cleansing the synapse of neurotransmitters helps to make thesynapse available for future transmissions.
2) Level 2:
For Level 1, we explained two distinct prop-agation mechanisms of neuronal signaling. For Level 2, wediscuss the transitions between these two mechanisms. Thepropagation of an action potential along the axon is controlledby the opening of voltage-gated ion channels that are at thedendrites. These ion channels are opened by external signalsthat can include both biological and synthetic sources. If thetransmitter is another neuron, and the environment betweenit and the receiver neuron is a chemical synapse, then theopening of the receiver’s ion channels are controlled by thebinding of neurotransmitters to the receptors [48, Ch. 11].The distribution of ion channels (e.g., Na + and K + ) in amembrane dictates how it reacts to the synaptic inputs; neuronscan be tuned to their individual roles based on where the ionchannels are expressed. For example, chemical synapses canbe either excitatory or inhibitory (i.e., generate or suppressaction potential firing in response to stimulus), depending onthe ion channels present and the current ionic conditions.Generally, the overall firing rate of an excitatory neuron isproportional to the strength of the stimulus. Action potentials travel along the axon until they reach theaxon terminal. For terminals that are connected to cells via chemical synapses, the arrival of an action potential triggersthe fusion of synaptic vesicles with the pre-synaptic neuron’souter membrane [328, Ch. 12], as we described for Level 2 inSection IV. This releases neurotransmitters into the synapse,triggering ion channel activity in the following cell and thecycle continues.Due to the diversity of neuron and synapse structures, therelative timing of Levels 1 and 2 can vary considerably. Theauthors of [329] showed that, depending on the size of achemical synapse and its associated reaction rates, commu-nication via the synapse could be either diffusion-limited orrate-limited.
3) Level 3:
The network of neurons in the body creates anenormous number of connections to relay and store informa-tion. Broadly speaking, an individual neuron does not directlymodulate and demodulate the information that it relays, butit is generally understood that information is contained in thenumber and timing of the action potential spikes. A sequenceof such spikes is referred to as a spike train .Synthetic means to interface with neurons try to controlthe membrane potential via the ion channels, e.g., using macroscale electric control and optical control as describedfor Level 3 in Section V. The authors of [330] report recentexperimental work that used electrodes to transmit digitalmessages across anesthetized earthworms, thereby creatingan artificial communication link across a biological channelwhere the neurons are used literally as relays. Optical stim-ulation has been popular in the biology community, wherelight-sensitive ion channel membrane proteins (called opsins)are expressed by the introduction of genes to make a cell arti-ficially sensitive to light. When an opsin protein is activated,it opens to enable a current pass through the membrane. Thisapproach, known as optogenetics , has been used to controlaction potentials in neurons and also other cells [331]. Becauseof the directionality of light, optogenetics is a promisingsolution for precise control of the behavior of individual cells.
4) Level 4:
As noted, neurons are primarily relaying infor-mation. We do not have a complete understanding of exactlyhow neurons modulate information; different metrics exist toquantify the information in spike trains, e.g., measuring thenumber of spikes over some period of time or the rest intervalbetween spikes [332]. There has also been recent research toquantify the information transmissible over the different stagesof neuron signaling. For example, the authors of [333], [334]analyzed axon memory and propagation and [144] consideredthe capacity of vesicles released into the synapse. The authorsof [335] measured the channel capacity of information in achemical synapse. The authors of [336] maximized the trans-missible bit rate across an axon and synapse by optimizing theaction potential spike rate and the receiver decision threshold.Moreover, the authors of [148] modeled neurons as filters andconsidered the effect of information filtering.
5) Level 5:
The connections between neurons can be quiteextensive; one neuron can receive inputs from thousands ofneurons and have synapses connecting to thousands of neuronsand other cells [48, Ch. 11]. There are about neurons in a human brain and synaptic connections. Collectively, theyenable the capacity to learn from and react to external stimuli.One particular example is at the neuromuscular junction,where acetylcholine is released by a motor neuron into achemical synapse with a skeletal muscle cell. This scenario isvery well-studied due to its accessibility, unlike most synapsesin the brain and spinal cord. Reception of acetylcholine leadsto a rapid influx of Na + and triggers muscle contraction.Due to their nature as relays of information, neurons arerecognized as key junctions for having an interface betweenthe external macroscale world and the in vivo microscaleworld for biomedical applications. Besides “high-jacking” anorganism’s nervous system to build an artificial communica-tion channel, as demonstrated with a worm in [330], there aremany opportunities to develop technologies for brain implantsand interfaces to detect and treat neurological diseases withartificial systems [337]. State-of-the-art implant technologiesinclude electromagnetic, chemical, and optical stimulation.Optogenetics has also been proposed as part of a bridge tointerface between biological systems and computer networks[57]. C. Communication via DNA
DNA is the foundation language of life. It is inherently astorage mechanism, as it contains the information required toencode proteins, but it also includes the supporting machineryto control when to produce each protein and how much. Thus,it is more useful to think of DNA communication as thesharing of information that propagates over time instead ofspace. Nevertheless, as we have mentioned in earlier sections,there are also characteristics of DNA communication thatinclude signaling over physical space.To apply the proposed communication hierarchy to DNA,we emphasize our perspective that DNA is primarily a storagemedium. As summarized in Table VI, Level 1 deals with thestorage of genes in DNA, though there are also aspects ofDNA communication that rely on spatial propagation. Level 2covers the biochemical processes of transcription from DNA toRNA and translation from RNA to protein. Level 3 covers howgenetic information is modulated and how it is controlled bothlocally and experimentally. Level 4 describes how proteins areencoded in DNA, and Level 5 considers what proteins areneeded and how they support life.DNA communication and processes supporting it havealready been mentioned in all of the sections discussingthe individual levels of the proposed hierarchy, even thoughgenetic information is often an exception to many of thegeneral trends of microscale MC (e.g., it supports far higherinformation rates than diffusive signaling). This includes therole of diffusion for DNA binding and the sharing of DNAvia bacterial conjugation in Level 1 (Section III), biochemicalpathways for gene expression in Level 2 (Section IV), generegulation and macroscale control of it in Level 3 (Section V),the information in DNA bases and using DNA synthetically torealize storage in Level 4 (Section VI), and as a component ina microfluidic biosensing application in Level 5 (Section VII).In this case study, we tie these ideas together with a focus onDNA’s role in storing genetic information.
1) Level 1:
The primary function of DNA is the preserva-tion and function of life. From a communication perspective, itstores information until it is needed, which could be on-goingor in response to particular life events or external signals.While there have been limited works within the MC commu-nity to model DNA as a “conventional” information molecule(such as [338] where DNA was proposed for a diffusion-basedcommunication system), there are aspects of DNA signalingthat can benefit from spatial propagation modeling. Theseinclude the diffusion of proteins that travel along DNA toregulate what genes are expressed [79], the propagation ofRNA out of a cell’s nucleus for translation into a protein byribosomes in the cytoplasm [48, Ch. 6], and the exchange ofDNA by conjugation when two bacteria come into contactwith each other [125]. We highlight a particular laboratorymethod because it has been modeled using the advection-diffusion-reaction equation that we discussed for Level 1 inSection III. Polymerase chain reaction (PCR) modeling isan important tool for making copies of a region of DNA[48, Ch. 8]. It includes a three-step process to 1) separateDNA into single strands; 2) bind primers to the ends ofa single strand; and 3) generate the compliments of thesingle strands. These steps take place in different regions withdifferent temperatures, which causes a flow that affects themovement of chemical molecules. Since all of the componentsare affected by circulatory flow, diffusion, and temperature-dependent chemical reactions, the advection-diffusion-reactionequation can be used to analyze PCR [89].
2) Level 2:
There are several distinct DNA processes thatwe associate with Level 2 behavior, i.e., at the interfacebetween device and the propagation medium and the bio-chemical signaling pathways therein. These include the stepsof gene expression , i.e., transcription from DNA to RNAand translation from RNA to protein using ribosomes, whichwe have already discussed in some detail for Level 2 inSection IV. Additional processes include replication of DNAand gene mutations. Our general understanding of Level 2 asdiscussed in Section IV includes the generation of informationmolecules. DNA is unique in this sense because the storedinformation is persistent ; nature obtains more DNA by copyingexisting DNA. We do not go into the biochemical details ofDNA replication here (the reader can learn more in [48, Chs. 4,5]), but we draw attention to one of the profound ideas thatdrive evolution. The processes that maintain and repair DNAare extremely precise but imperfect. The imperfections areactually quite important, because they lead to the mutationsthat enable evolution.Synthetic creation of gene mutations is referred to as geneediting. The basic idea behind gene editing is to modifya particular gene (i.e., create mutations) and then observethe effect on the organism [48, Ch. 8]. Mutations include“gene knockouts,” where the gene is simply removed from thegenome, and modifications where the experimenter controlsthe conditions under which the gene is expressed, e.g., tomake the gene sensitive to a signaling molecule that can turnthe gene on and off, and which we associate with Level 3behavior. The general goal is to understand the role of agene and the proteins that it produces. From the perspective of our proposed hierarchy, such gene editing is altering thedevices themselves. We are able to artificially introduce orremove physical interfaces to the environment and thereforecontrol communication within a cell, between cells, or withan experimenter. For example, a common modification is tofuse a gene with one that encodes a fluorescent protein. Wecan then monitor gene expression by measuring the level offluorescence.
3) Level 3:
We discussed gene regulation, i.e., the processescontrolling when to activate or deactivate the expression of aparticular gene, in some detail for Level 3 in Section V. Here,we emphasize how the bases in DNA (and RNA) correspond tosequences of information. Both DNA and RNA are composedof subunits labeled as one of four bases. These subunits areplaced sequentially in a chain, such that we can read theinformation in a chain as a sequence of bases. Thus, DNAis usually described as a sequence of As, Cs, Gs, and Ts, andRNA is usually described as a sequence of As, Cs, Gs, andUs. Each subunit stores log = bits of information.Not surprisingly, there is considerable interest in readingDNA and RNA sequences at macroscale. Technology fordoing so has been in development since the 1970s and theemergence of dideoxy (or Sanger) sequencing. Two commonsequencing methods today are Illumina sequencing and iontorrent sequencing, which both rely on PCR to amplify DNA[48, Ch. 8]. Newer methods are in development to avoidthe amplification stage and read a sequence directly fromindividual molecules. There are also possibilities for applyingoperations normally expected of a word processor, i.e., cut,copy, and paste, using restriction enzymes, PCR, and ligases,respectively [339]–[342]. Data storage in DNA is expected tobe orders of magnitude more energy efficient than currentlyavailable technologies [256], [257], [342].A macroscale control method for DNA that was not in-troduced for Level 3 in Section V is optogenomics [57].An optogenomic interface uses light to activate or deactivatespecific genes in eukaryotic cells with subcellular resolutionand high temporal accuracy. It has already been validated forthe activation of cellular responses and expressing individualgenes but could furthermore be applied to regulate and correctDNA structure.
4) Level 4:
Level 4 behavior for DNA is relatively wellunderstood, since we know how the nucleotide bases in DNAmap to amino acids in protein. As we discussed in Section VI,triplets of bases encode the twenty amino acids that are com-monly found in proteins. However, not all DNA maps directlyto amino acids [48, Ch. 6]. For many genes, transcription toRNA is the final step and there is no corresponding translationto protein. RNAs themselves can play important structural orcatalytic roles, such as being a base for ribosomes (whichconduct translation). RNAs can also be used to regulate geneexpression, e.g., by degrading other target RNAs.
5) Level 5:
The natural applications of DNA are somewhatself-evident since it is the foundation language of all life. DNAencodes proteins, which perform thousands of distinct cellularfunctions [48, Ch. 2]. The propagation of genetic informationover time, regulated to manifest at the right moment or inresponse to the right stimulus, facilitates the cell life cycle and correspondingly the function and behavior of multi-cellular or-ganisms. For example, cell signaling, mobility, growth, mitosis(i.e., division of eukaryotic cells), and differentiation (i.e., acell changing to a different type) are all behaviors that aredriven by the biochemical actions of proteins.As discussed in Section VI-D, DNA can be a promising so-lution for data storage due to its information density and long“shelf life”. Another application that shows potential is theuse of DNA molecules as building blocks for nanomachines.The forces between DNA bases that define its double helicalshape are well understood. By carefully selecting the sequenceof base-pairs, it is possible to create synthetic double-strandedDNA molecules that self-assemble into predetermined struc-tures (“DNA Origami”) with a specific size and shape [343],[344]. Additionally, because DNA assembly occurs due tohydrogen bonding between base-pairs, the conformation ofthese kinds of structures can be controlled by temperature.Heating or cooling affect the level of association-dissociationbetween complementary strands and change the shape ofthe molecule. Using this principle it is possible to constructmolecular motors for the movement of nanomachines [345].Such nanomachines can be controlled in a number of ways,such as temperature, light, pH, metal ions or other externalstimuli [346]. Using this technology, it is possible to constructnanoscale devices that perform complicated tasks such asmonitor physiological functions [347] or targeted drug release[348]. IX. O PEN P ROBLEMS
In this section, we summarize open challenges and op-portunities with the support of our proposed hierarchy forsignaling in cell biology. We intend for this discussion to guidefurther interest and research in this interdisciplinary field. Ourproposed hierarchy enables us to organize these problems andgain some insight on how to effectively tackle them.We emphasize that there are many opportunities for theapplication of MC theory and communications analysis inbiological systems that are already relatively well studied, inaddition to the design of synthetic communication networks.Natural scientists focus on describing systems end-to-end, inthat they provide as much detail as possible for individualcomponents and often omit interactions with other systems.However, given life’s reliance on communication, studyingthese same systems from a communications engineering pointof view can provide tools to inform understanding and developmethods for control. We facilitate this kind of exploration byenabling researchers to map system components and how theyintegrate in a communications networking sense.In the following, we discuss problems that align with eachof the five levels of the proposed hierarchy. We then describeproblems that derive from the integration of the differentlevels. Finally, we describe opportunities associated with ourend-to-end case studies of QS, neuronal communication, andcommunication via DNA.
A. Level 1
The existing literature on diffusion for MC focuses ontheoretical descriptions of particle propagation using the math- ematics of diffusion [24]. For tractability, these models usuallymake simplifying assumptions about the system (e.g., idealpropagation, homogeneous molecule characteristics, simplifiedchannel geometry). Despite the progress that has been made,both natural and synthetic systems can be much more complexthan what existing contributions can sufficiently model [27].For example, molecules can participate in intricate chemicalreaction networks while they are diffusing, and practical fluidflow patterns can be more spatially-varying than the models wehave summarized here. Such details can make the correspond-ing differential equations for propagation more complex andheterogeneous. There are opportunities to identify closed-formsolutions to such scenarios, or to develop robust numericalmethods where closed-form solutions are not achievable.An important related question is how realistic a modelneeds to be in order for it to be useful in practice, i.e.,to make informed predictions or to effectively guide systemdesign. There likely is scope to effectively apply existingMC channel modeling to biological systems. In particular,some biological communication modalities could be describedas the integration of multiple communication channels. Forexample, neuron signals propagate as both a traveling electricalpotential wave (along the axon) and as a reaction-diffusionsignal (across a synapse) [48, Ch. 11]. Other examples includegap junctions and plasmodesmata [105], [119], which createlocally-regulated parallel channels between cells with diffusionon either side. There is also a range of communicationmodalities that have so far received limited attention from theMC community but that could benefit from their engagement(e.g., contact-based communication, including cell conjugation[125]).
B. Level 2
As summarized in Table IV, channel responses have beenderived under ideal molecule generation and reception models.However, the physical and temporal scales of molecule gen-eration and reception may not be sufficiently small comparedwith the propagation channel, which requires us to take thecorresponding biophysical and biochemical processes intoaccount to understand the channel response, such as mRNApropagation from nucleus to ribosome and stochastic chemicalreactions in gene expression. Not surprisingly, the inclusion ofthese practical processes will complicate the initial and bound-ary conditions applied to propagation equations, and thusintroduce challenges when deriving channel responses [27].The shift of theoretical research to more realistic models alsoimposes a requirement on related biological software to notonly verify closed-form solutions but also provide numericalresults for intractable problems. Moreover, through analyticalcharacterization and verification via simulation, guidelinesfor choosing the optimal molecule generation method andreception strategy should also be developed to facilitate MCsystem design.An important signaling mechanism discussed in Section IVis the use of extensive and interconnected transcription net-works. Although the functions of some basic building blocks(e.g., the feed forward loop) have been identified, there is still a need to have a more comprehensive understandingof transcription networks, including the inputs, the outputs,and how a change of inputs influences the outputs [41]. Thiswould be helpful for controlling and synthesizing transcriptionnetworks to achieve target functions. Another question inspiredby transcription networks concerns signaling complexity inbiological systems. It would be helpful to develop a categoriza-tion or quantified “metric” of signaling complexity in termsof the number and variety of molecules used for the signalingarchitecture of a given cell type [349], [350]. Such a metricmight provide a rule of thumb that enables us, given data ofwell-understood cells, to predict what is unknown about othercells that are sufficiently similar (i.e., same kingdom, similarsize, etc.). C. Level 3
Although existing experimental tools make the control andobservation of microscale phenomena possible, the operationof some devices (e.g., optical microscopy [168]) requires skillsand interdisciplinary technical knowledge that may impedetheir adoption in communications-focused research. Moreover,as stated in Section V-B, no single existing technique caninspect biological signaling processes across all spatial andtemporal scales simultaneously. Thus, new experimental toolsthat span multiple spatial and temporal resolutions would beincredibly useful for microscale systems. These must satisfythe requirements of biocompatibility, non-invasiveness, andminiaturization. It would also be helpful for the new exper-imental tools to facilitate communication analysis, e.g., tocapture the probability distributions needed to determine com-munication capacity and bit error rate. In addition, it is worth-while to seek unique combinations of existing macroscalecontrol and observation tools (e.g., see Fig. 17) for targetedapplications. One example is the guidance of drug carriers totarget areas via in vivo imaging and the subsequent release ofdrug molecules via macroscale magnetic control [351].With the vision of the Internet of Bio-NanoThings (IoBNT)[352], more attention is needed to develop feasible interfacesto connect the microscale world with macroscale wirelessnetworks. In particular, we can draw from the expertise ofother related domains to support efforts to have an effectivemicro-macro interface, such as image processing, machinelearning, and optical physics.
D. Level 4
Quantifying the limits of molecular signaling usinginformation-theoretic approaches provides a way to study thepotential of cell signaling. Although specific cell types andsignaling pathways have been studied, such as the
E. coli bacterium strain K-12 MG1655 [217] and the JAK-STAT path-way in eukaryotic cells [219], it remains to be seen whetherthe obtained results and research methods can be generalizedto other pathways and cell types. Moreover, the relationshipbetween communication capacity and cell behavior needs anaccessible interpretation so that scientists from different fieldscan easily understand and apply each other’s research results. Various chemical circuits are introduced in Sections VI-Band VI-C with the aim to realize computation and com-munication functionalities [34], [225], [229], [232], but wehighlight that most contributions have been theoretical worksthat have not yet been validated with physical experiments,mainly due to the tedious, laborious, and expensive nature ofwet lab experimentation. However, it is essential to develop arobust testing framework for validation and to optimize designparameters.Communications engineers try to minimize the complexityof a system design in terms of different components orvariations in the types of signals. We might also assume thatthis is true for natural systems, since every extra componentor function has an associated cost, e.g., metabolic or fitness.However, in nature, we see potential over-engineering in theconstruction of signaling systems. For example, cells usemultiple pathways and a variety of different molecules forthe activation of the same gene [353]–[355]. It is not entirelyclear whether this fulfills a need for robustness via redundancyor whether designs constitute a locally optimal (but globallysub-optimal) solution obtained by evolutionary optimization.We believe that it is important to ask whether this redundancy(if any) can be identified by comparing predictions with ob-servations. Such knowledge might lead to better understandingof minimal sets of required system components for particularfunctions, or increased robustness in synthetic systems.
E. Level 5
While significant progress has been made over the last twodecades in applying communications ideas to biological andsynthetic systems, most of this work has taken the form oftheoretical models proposed for MC schemes or exploringtheir limits [24]–[27]. For MC to progress as a field, validationof these schemes is necessary through proof-of-concept appli-cations that demonstrate their feasibility and usefulness. Ad-ditionally, as a large proportion of MC work concerns low- tomid-level interactions (e.g., diffusion or modulation) [1], [24],[29], [31], there is scope for applying communications networktheory to larger systems such as interconnected populations ofcells or nanomachines. Such modeling could enable predictionand observation of the emerging behavioral dynamics of thesesystems, but requires the development of suitable algorithmicor computational tools to do this efficiently. Inspiration can betaken from agent-based modeling in synthetic biology, whichis an approach that can simulate large networks of cells [356].
F. Multi-Level Problems
The proposed hierarchy provides insight by separatingsystem components and tasks into levels, but one of thehierarchy’s salient features is to help articulate, understand,and solve problems that span multiple levels. For example,there is a direct mathematical link between Levels 1 and 2,since Level 2 provides the boundary conditions that are neededto solve the differential propagation equations at Level 1.Thus, Levels 1 and 2 are both necessary to determine a cell’ssignaling channel impulse response, which can then be usedto determine a receiving device’s observed signal given what was transmitted. Significant research efforts have been made todetermine impulse responses for diffusion-based MC channels[24], but this survey can assist to identify important scenariosthat have not received such analysis. For example, models forgene expression could be expanded to include the propagationof RNA out of the nucleus.An important problem is to understand how the constraintsand limitations of one level impact the design and performanceof other levels. In particular, the biophysical and biochemicalactivities at Levels 1 and 2 are inherently noisy; moleculepropagation and chemical kinetics are both modeled at mi-croscale as stochastic processes. These features impact thereliability of cell signaling channels, the rate and quantity ofinformation they carry, and how life evolved to accommodatethem. There are many questions that can be posed regardingthe impact of biophysical and biochemical noise on higherlevels, e.g., on gene regulation and other microscale signaloperations (Level 3), on our ability to experimentally observeand control cell signaling systems (Level 3), on how accuratelycells can infer information about their surrounding environ-ment (Level 4), on how robust synthetically designed MCdevices can be (Level 4), and on how heterogeneous behavioremerges in large cellular populations (Level 5). Levels 1 and 2also impose constraints on the overall communication speed.The approach to analyze the communication speed of chemicalsynapses in [329] might be generalized to other communi-cation systems to discern bottlenecks and their impact onhigher levels, e.g., device sensitivity and responsiveness toenvironmental changes.Given the scalability of our proposed hierarchy, interestingproblems can arise when deciding at what scale to definea communication channel. For example, gap junctions formchannels between adjacent cells. However, signals passingthrough gap junctions can be relayed and reach cells at largedistances from the initial transmitter. We propose to investigatewhether it is suitable to describe the communication link to adistant cell as a single aggregated channel. If so, then we candetermine the reliability of this channel as the receiving cellis placed further from the source. G. Case Studies
Finally, we highlight several open problems associated withour case studies. Concerning QS, the majority of work todayis concerned with the simulation and analysis of ideal systems.Studies that consider collective behavior tend to be theoretical,in part because of the inherent complexity of large cellpopulations [357], [358]. Nevertheless, simulations of a largenumber of cells and multiple autoinducers could span all levelsof our proposed hierarchy and provide numerical insightsregarding the underlying causes of natural behavior and auseful testbed for the design of synthetic systems. In particular,one underdeveloped area is security in cellular signaling (e.g.,secure communications using QS and eavesdropping on QSsources). Furthermore, inspired by the multi-level problemspresented in the previous subsection, rigorous analysis of theimpact of autoinducer propagation could extend our under-standing of how QS architectures in individual cells contributeto collective behavior. There is already a vast literature of theoretical, experimental,and computational studies of neuronal signaling [359]. How-ever, there is still scope to apply our proposed hierarchy tothis field and address consequential problems. For example,neurological diseases could be modeled as application-levelproblems that arise from deficiencies in the propagation ofneuronal signals, and could be treated via communication andcontrol of neurons using optogenetic tools and other brainimplants [337]. Another scenario is synaptic plasticity, whichrefers to biophysical processes that change synaptic strengthsover time [359, Ch. 25], and could be studied as a dynamiccommunication link that affects learning and memory.We have already mentioned several open problems forsignaling via DNA in our discussion of multi-level problems,e.g., the impact of RNA propagation on gene expression.Similarly, all of the biochemical pathways involved in thegene transcription and translation processes can be impairedby noise, even in a managed scenario such as PCR, which hasan impact on the resulting protein production levels. Modelsfollowing our proposed hierarchy to include this stochasticitycould be used to determine the distribution of protein produc-tivity and predict the reliability of DNA-based storage.X. C
ONCLUSION
Unleashing the potential of MC for interdisciplinary ap-plications requires substantial efforts from diverse scientificcommunities. However, the distinct approaches to articulateand study research problems gives rise to a mismatch betweenthe different disciplines. To bridge this mismatch, in thissurvey, we proposed a novel communication hierarchy todescribe signaling in cell biology. The proposed hierarchy iscomprised of five levels: 1)
Physical Signal Propagation ; 2)
Physical and Chemical Signal Interaction ; 3)
Signal-DataInterface ; 4)
Local Data Abstraction ; and 5)
Application .While the nominal communicating “device” is assumed tobe an individual cell, the hierarchy readily describes com-munication between any devices using or observing chemicalsignals in a biological system, including cellular organellesand macroscale experimental equipment.Our proposed hierarchy enabled us to map communicationconcepts to infrastructure and activities in biological signaling.Specifically, we started with the
Physical Signal Propagation level (Level 1) to discuss the fundamental mechanisms ofmolecular propagation. This level focused on mathematicalformulations of diffusion-based phenomena, and also detailedcargo-based propagation and contact-based transport. For the
Physical and Chemical Signal Interaction level (Level2), we reviewed physical signal generation and receptionmechanisms and the associated biochemical and biophysicalsignaling pathways. In addition, we provided a mathematicalcharacterization of different release and reception strategies,corresponding to different initial and boundary conditions (andhence channel responses), and mathematically described geneexpression pathways. For the
Signal-Data Interface level(Level 3), we described the mathematical quantification of thephysical signals that are released and received, including theconversion between quantification and data, i.e., modulation and demodulation in communication networks. This discussionalso included a survey of methods for macroscale observationand control of cell signaling behavior. For the Local DataAbstraction level (Level 4), we considered the significance ofinformation in individual cells, limits on how much informa-tion is carried in natural MC signals, and how synthetic devices(including chemical and genetic designs) might be realized torepresent, store, and process information. For the top of theproposed hierarchy, i.e., the
Application level (Level 5), weselected biosensing and therapeutics as exemplary applicationsto show how they might benefit from the integration of naturaland synthetic systems at lower levels to realize their potential.To further demonstrate the utility and flexibility of ourproposed hierarchy, we mapped all of the levels to casestudies of QS, neuronal signaling, and communication viaDNA. Finally, we identified a selection of open problemsassociated with each level and in the integration of multiplelevels. We anticipate that our proposed hierarchy providesresearchers from different fields with language to interpret andunderstand results on MC signaling from other disciplines,while simultaneously realizing the potential of opportunitiesfor interdisciplinary collaboration. Ultimately, we intend forthis survey to support the advancement of interdisciplinarycell signaling applications.A
PPENDIX
In Tables VII and VIII, we define common biologicaland communication terms that appear throughout the survey,respectively. R
EFERENCES[1] I. F. Akyildiz, F. Brunetti, and C. Bl´azquez, “Nanonetworks: A newcommunication paradigm,”
Comput. Netw. , vol. 52, no. 12, pp. 2260–2279, Aug. 2008.[2] T. Nakano, A. W. Eckford, and T. Haraguchi,
Molecular Communica-tion . Cambridge, U.K.: Cambridge University Press, 2013.[3] T. Suda, M. Moore, T. Nakano, R. Egashira, and A. Enomoto,“Exploratory research on molecular communication between nanoma-chines,” in
Proc. ACM GECCO , Jun. 2005, pp. 1–5.[4] S. Hiyama, Y. Moritani, T. Suda, R. Egashira, A. Enomoto, M. Moore,and T. Nakano, “Molecular communication,” in
Proc. NSTI Nanotech-nol. Conf. , May 2005, pp. 392–395.[5] T. Nakano, M. J. Moore, F. Wei, A. V. Vasilakos, and J. Shuai, “Molec-ular communication and networking: Opportunities and challenges,”
IEEE Trans. Nanobiosci. , vol. 11, no. 2, pp. 135–148, Jun. 2012.[6] A. W. Eckford, “Nanoscale communication with Brownian motion,” in
Proc. IEEE CISS , Mar. 2007, pp. 160–165.[7] A. O. Bicen and I. F. Akyildiz, “System-theoretic analysis and least-squares design of microfluidic channels for flow-induced molecularcommunication,”
IEEE Trans. Signal Process. , vol. 61, no. 20, pp.5000–5013, Oct. 2013.[8] Y. Chahibi, I. F. Akyildiz, and I. Balasingham, “Propagation modelingand analysis of molecular motors in molecular communication,”
IEEETrans. Nanobiosci. , vol. 15, no. 8, pp. 917–927, Dec. 2016.[9] T. Suda and T. Nakano, “Modeling the stochastic behavior of kinesin-coated beads,” in
Proc. IEEE ICC , May 2019, pp. 1–7.[10] A. Noel, K. C. Cheung, and R. Schober, “Improving receiver perfor-mance of diffusive molecular communication with enzymes,”
IEEETrans. Nanobiosci. , vol. 13, no. 1, pp. 31–43, Mar. 2014.[11] H. B. Yilmaz, A. C. Heren, T. Tugcu, and C.-B. Chae, “Three-dimensional channel characteristics for molecular communications withan absorbing receiver,”
IEEE Commun. Lett. , vol. 18, no. 6, pp. 929–932, Jun. 2014.[12] Y. Deng, A. Noel, M. Elkashlan, A. Nallanathan, and K. C. Cheung,“Modeling and simulation of molecular communication systems witha reversible adsorption receiver,”
IEEE Trans. Mol. Biol. Multi-ScaleCommun. , vol. 1, no. 4, pp. 347–362, Dec. 2015. [13] A. Ahmadzadeh, H. Arjmandi, A. Burkovski, and R. Schober, “Com-prehensive reactive receiver modeling for diffusive molecular commu-nication systems: Reversible binding, molecule degradation, and finitenumber of receptors,”
IEEE Trans. Nanobiosci. , vol. 15, no. 7, pp.713–727, Oct. 2016.[14] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “On the character-ization of binary concentration-encoded molecular communication innanonetworks,”
Nano Commun. Netw. , vol. 1, no. 4, pp. 289–300, Dec.2010.[15] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Interferenceeffects on modulation techniques in diffusion based nanonetworks,”
Nano Commun. Netw. , vol. 3, no. 1, pp. 65–73, Mar. 2012.[16] P.-J. Shih, C.-H. Lee, P.-C. Yeh, and K.-C. Chen, “Channel codesfor reliability enhancement in molecular communication,”
IEEE J. Sel.Areas Commun. , vol. 31, no. 12, pp. 857–867, Dec. 2013.[17] I. Akyildiz, F. Fekri, R. Sivakumar, C. Forest, and B. Hammer,“Monaco: fundamentals of molecular nano-communication networks,”
IEEE Wirel. Commun. , vol. 19, no. 5, pp. 12–18, Oct. 2012.[18] B. Krishnaswamy, C. M. Austin, J. P. Bardill, D. Russakow, G. L.Holst, B. K. Hammer, C. R. Forest, and R. Sivakumar, “Time-elapsecommunication: Bacterial communication on a microfluidic chip,”
IEEE Trans. Commun. , vol. 61, no. 12, pp. 5139–5151, Dec. 2013.[19] T. Nakano, S. Kobayashi, T. Suda, Y. Okaie, Y. Hiraoka, andT. Haraguchi, “Externally controllable molecular communication,”
IEEE J. Sel. Areas Commun. , vol. 32, no. 12, pp. 2417–2431, Dec.2014.[20] N. Tuccitto, G. Li-Destri, G. M. L. Messina, and G. Marletta, “Reactivemessengers for digital molecular communication with variable trans-mitterreceiver distance,”
Phys. Chem. Chem. Phys. , vol. 20, no. 48, pp.30 312–30 320, Dec. 2018.[21] L. Grebenstein, J. Kirchner, W. Wicke, A. Ahmadzadeh, V. Jamali,G. Fischer, R. Weigel, A. Burkovski, and R. Schober, “A molecularcommunication testbed based on proton pumping bacteria: Methodsand data,”
IEEE Trans. Mol. Biol. Multi-Scale Commun. , vol. 5, no. 1,pp. 56–62, Dec. 2019.[22] N. Farsad, W. Guo, and A. W. Eckford, “Tabletop molecular commu-nication: Text messages through chemical signals,”
PLoS One , vol. 8,no. 12, p. e82935, Dec. 2013.[23] S. Giannoukos, D. T. McGuiness, A. Marshall, J. Smith, and S. Taylor,“A chemical alphabet for macromolecular communications,”
Anal.Chem. , vol. 90, no. 12, pp. 7739–7746, Jun. 2018.[24] V. Jamali, A. Ahmadzadeh, W. Wicke, A. Noel, and R. Schober,“Channel modeling for diffusive molecular communication: A tutorialreview,”
Proc. IEEE , vol. 107, no. 7, pp. 1256–1301, Jul. 2019.[25] N. Farsad, H. B. Yilmaz, A. Eckford, C.-B. Chae, and W. Guo, “Acomprehensive survey of recent advancements in molecular commu-nication,”
IEEE Commun. Surv. Tuts. , vol. 18, no. 3, pp. 1887–1919,2016.[26] I. F. Akyildiz, M. Pierobon, and S. Balasubramaniam, “An informationtheoretic framework to analyze molecular communication systemsbased on statistical mechanics,”
Proc. IEEE , vol. 107, no. 7, pp. 1230–1255, Jul. 2019.[27] M. Kuscu, E. Dinc, B. A. Bilgin, H. Ramezani, and O. B. Akan,“Transmitter and receiver architectures for molecular communications:A survey on physical design with modulation, coding, and detectiontechniques,”
Proc. IEEE , vol. 107, no. 7, pp. 1302–1341, Jul. 2019.[28] C. A. Soldner, E. Socher, V. Jamali, W. Wicke, A. Ahmadzadeh, H.-G.Breitinger, A. Burkovski, K. Castiglione, R. Schober, and H. Sticht, “Asurvey of biological building blocks for synthetic molecular commu-nication systems,”
IEEE Commun. Surv. Tutorials , pp. 1–1, Jul. 2020.[29] O. B. Akan, H. Ramezani, T. Khan, N. A. Abbasi, and M. Kuscu,“Fundamentals of molecular information and communication science,”
Proc. IEEE , vol. 105, no. 2, pp. 306–318, Feb. 2017.[30] Y. Chahibi, M. Pierobon, S. O. Song, and I. F. Akyildiz, “A molecularcommunication system model for particulate drug delivery systems,”
IEEE Trans. Biomed. Eng. , vol. 60, no. 12, pp. 3468–3483, Dec. 2013.[31] T. Nakano, Y. Okaie, S. Kobayashi, T. Hara, Y. Hiraoka, andT. Haraguchi, “Methods and applications of mobile molecular com-munication,”
Proc. IEEE , vol. 107, no. 7, pp. 1442–1456, Jul. 2019.[32] U. A. K. Chude-Okonkwo, R. Malekian, B. T. Maharaj, and A. V.Vasilakos, “Molecular communication and nanonetwork for targeteddrug delivery: A survey,”
IEEE Commun. Surv. Tutorials , vol. 19, no. 4,pp. 3046–3096, 2017.[33] I. F. Akyildiz, M. Pierobon, and S. Balasubramaniam, “Moving forwardwith molecular communication: from theory to human health applica-tions [point of view],”
Proc. IEEE , vol. 107, no. 5, pp. 858–865, May2019. TABLE VII: Glossary of Biological Terms.
Term DescriptionAction potential Rapid and transient change in electric potential across a membraneATP Adenosine triphosphate. Molecule able to store and transfer chemical energy within a cellAutoinducers Diffusible signal molecules produced by cells to monitor local population changes. They can also have additional functions(e.g., act as antibiotics or toxins)Cytoplasm The gel-like contents of a cell between the outer membrane and the nucleus. Comprised mainly of water, proteins, andsaltsCytoskeleton Complex, dynamic network of filaments spanning the entire cellCytosol The aqueous part of the cytoplasmDNA Deoxyribonucleic acid. Carrier of the hereditary information for the building and maintenance of organismsEndoplasmic reticulum (ER) Continuous membrane system connected to the nucleus. Involved in folding, modification, and transport of proteinsEndosomes Membrane-bound vesicles formed around molecules to facilitate their transport into the cell from the extracellular spaceEnzyme A biologically relevant molecule acting as a catalyst, making chemical reactions possible or greatly increasing their rateExocytosis Active transport of material out of the cell via membrane vesciclesExtracellular matrix Complex and dynamic extracellular network of macromolecules providing structural and chemical support to cellsGolgi apparatus Large organelle of eukaryotic cells responsible for modification, packaging, and transportation of proteinsG-protein-coupled receptors Large group of cell surface receptor proteinsHydrolysis The chemical breakdown of compounds by waterMacromolecule A molecule that consists of a large number of atoms (proteins, nucleic acids, synthetic polymers)Messenger RNA (mRNA) Single-stranded RNA molecule, carrying the information for protein production outside the cell nucleusMicrotubule Dynamic, hollow tubes formed by protein polymers. Part of the cytoskeletonMicroRNA (miRNA) Small non-coding RNA molecules involved in gene regulationNeurotransmitters Chemical messengers that are released into a chemical synapse to convey a message between neuronsOrganelle Cellular structure with specialized functions (e.g., endoplasmic reticulum, mitochondria, golgi apparatus)PCR Polymerase Chain Reaction. Method of rapid replication of a given DNA sample into large numbersPlasmid Genetic structure that can replicate independent of the host’s chromosomePost-transcriptional modification The modification of an mRNA molecule directly after transcription to produce a mature mRNA for protein productionPost-translational modification The modification of proteins after their production in ribosomesPromoter sequence Small DNA sequence preceding a gene that marks where transcription should startProtein Organic compound comprised of one or more macromolecules. Integral to most cellular processesProtein phosphorylation The addition of a phosphate group to an amino acid of a protein. Reversible process crucial for cell signalingRedox reactions Oxidation-reduction reactions involving the transfer of electrons between two chemical speciesRibosomes Protein-synthesizing factories, comprised of ribosomal RNA and associated helper proteinsRNA Ribonucleic acid. A single-stranded biopolymer that is essential for protein production by carrying sequence informationfrom DNA to ribosomesRNA polymerase Enzyme that can bind and follow a strand of DNA, replicating its sequenceSecond messengers Small intracellular molecules relaying information received from first messengers, i.e., cell surface receptorsSignaling pathway A chain of cell components and molecules working in succession to transfer a signalSynapse Small contact site that chemically or electrically links a neuron to other cellsT-cell Type of white blood cell (leukocyte), part of the immune systemTranscription factor A protein that can bind a specific DNA sequence, controlling the expression of a geneTransfer RNA (tRNA) Special RNA molecule involved in protein production within ribosomes. It matches a loose amino acid to mRNA sequence[34] B. D. Unluturk, A. O. Bicen, and I. F. Akyildiz, “Genetically engi-neered bacteria-based biotransceivers for molecular communication,”
IEEE Trans. Commun. , vol. 63, no. 4, pp. 1271–1281, Apr. 2015.[35] A. Marcone, M. Pierobon, and M. Magarini, “Parity-check codingbased on genetic circuits for engineered molecular communicationbetween biological cells,”
IEEE Trans. Commun. , vol. 66, no. 12, pp.6221–6236, Dec. 2018.[36] E. De Leo, L. Donvito, L. Galluccio, A. Lombardo, G. Morabito,and L. M. Zanoli, “Communications and switching in microfluidicsystems: Pure hydrodynamic control for networking labs-on-a-chip,”
IEEE Trans. Commun. , vol. 61, no. 11, pp. 4663–4677, Nov. 2013.[37] L. Birnbaumer, “G proteins in signal transduction,”
Annu. Rev. Phar-macol. Toxicol. , vol. 30, no. 1, pp. 675–705, Apr. 1990.[38] A. Klaus and W. Birchmeier, “Wnt signalling and its impact ondevelopment and cancer,”
Nat. Rev. Cancer , vol. 8, no. 5, pp. 387–398, May 2008.[39] T. Nakano, T. Suda, Y. Okaie, M. J. Moore, and A. V. Vasilakos,“Molecular communication among biological nanomachines: A layeredarchitecture and research issues,”
IEEE Trans. Nanobiosci. , vol. 13,no. 3, pp. 169–197, Sep. 2014. TABLE VIII: Glossary of Communications Terms.
Term DescriptionBinary CSK (BCSK) Information is represented by two concentration levels of a chemical signalBit A basic unit of information in communication systems that represents a logical state with one of two possible values(usually or )Bit error rate (BER) Ratio of the number of bit errors and the total number of transmitted bitsChannel The physical medium or logical connection through which a signal propagatesChannel capacity The tight upper bound on the rate at which information can be reliably transmitted over a communication channeland is the maximum of the mutual information of the input and output of the channelConcentration shift keying (CSK) Information is represented by different concentration levels of a chemical signalDecoding Retrieval of original information symbols from a signalDemodulation The process that uses the observed signals to attempt recovery of the transmitted information symbolsEncoding Conversion of information symbols into a form that can be superimposed onto a signalFrequency shift keying (FSK) Information is represented by different frequencies of a signalLow-density parity-check (LDPC) code A powerful linear error correction codeModulation The process that varies one or more properties (e.g., amplitude, frequency, phase) of the physical carrier signal totransmit informationMolecule shift keying (MoSK) Information is represented by different types of messenger moleculesMutual information Measures the mutual dependence between two random variablesQuadruple CSK (QCSK) Information is represented by four concentration levels of a chemical signalReaction shift keying (RSK) Information is represented by different emission patternsReceiver A device or biological entity capable of observing a signalSingle parity-check (SPC) code A single bit is appended to the end of each transmitted frame. The parity bit is 1 if the data portion of the framehas an odd number of 1’s; otherwise, it is 0Symbol Used to represent a letter, a waveform, or a state in communication systems. Each symbol may encode one orseveral bits. For example, a symbol encodes a single bit in BCSK and two bits in QCSKTransmitter A device or biological entity that physically produces a signal[40] R. Kolter and E. P. Greenberg, “The superficial life of microbes,” Nature , vol. 441, no. 7091, pp. 300–302, May 2006.[41] U. Alon,
An Introduction to Systems Biology . London, U.K.: Chapmanand Hall/CRC, 2006.[42] J. C. Anderson, E. J. Clarke, A. P. Arkin, and C. A. Voigt, “Environ-mentally controlled invasion of cancer cells by engineered bacteria,”
J.Mol. Biol. , vol. 355, no. 4, pp. 619–627, Jan. 2006.[43] S. F. Bush, J. L. Paluh, G. Piro, V. Rao, R. V. Prasad, and A. W.Eckford, “Defining communication at the bottom,”
IEEE Trans. Mol.Biol. Multi-Scale Commun. , vol. 1, no. 1, pp. 90–96, Mar. 2015.[44] L. Felicetti, S. S. Assaf, M. Femminella, G. Reali, E. Alarcon,and J. Sole-Pareta, “The molecular communications markup language(MolComML),”
Nano Commun. Netw. , vol. 16, pp. 12–25, Jun. 2018.[45] H. Zimmermann, “OSI reference model–the ISO model of architecturefor open systems interconnection,”
IEEE Trans. Commun. , vol. 28,no. 4, pp. 425–432, Apr. 1980.[46] V. Cerf and R. Kahn, “A protocol for packet network intercommu-nication,”
IEEE Trans. Commun. , vol. 22, no. 5, pp. 637–648, May1974.[47] L. Felicetti, M. Femminella, G. Reali, T. Nakano, and A. V. Vasilakos,“TCP-like molecular communications,”
IEEE J. Sel. Areas Commun. ,vol. 32, no. 12, pp. 2354–2367, Dec. 2014.[48] B. Alberts, A. D. Johnson, J. Lewis, D. Morgan, M. Raff, K. Roberts,and P. Walter,
Molecular Biology of the Cell , 6th ed. New York, NY,USA: Garland Science, 2015.[49] A. Goldsmith,
Wireless Communications . Cambridge, U.K.: Cam-bridge University Press, 2005.[50] B. P. Lathi and R. A. Green,
Essentials of Digital Signal Processing .Cambridge, U.K.: Cambridge University Press, 2014.[51] K. Darchini and A. S. Alfa, “Molecular communication via micro-tubules and physical contact in nanonetworks: A survey,” pp. 73–85,Jun. 2013.[52] L. Felicetti, M. Femminella, G. Reali, and P. Li`o, “Applications ofmolecular communications to medicine: A survey,”
Nano Commun.Netw. , vol. 7, pp. 27–45, Mar. 2016. [53] Y. Chahibi, “Molecular communication for drug delivery systems: Asurvey,”
Nano Commun. Netw. , vol. 11, pp. 90–102, Mar. 2017.[54] D. Carmean, L. Ceze, G. Seelig, K. Stewart, K. Strauss, and M. Willsey,“DNA data storage and hybrid molecularelectronic computing,”
Proc.IEEE , vol. 107, no. 1, pp. 63–72, Dec. 2018.[55] T. Nguyen, T. S. Jones, P. Fontanarrosa, J. V. Mante, Z. Zundel,D. Densmore, and C. Myers, “Design of asynchronous genetic circuits,”
Proc. IEEE , vol. 107, no. 7, pp. 1356–1368, Jul. 2019.[56] E. Kim, J. Li, M. Kang, D. L. Kelly, S. Chen, A. Napolitano,L. Panzella, X. Shi, K. Yan, S. Wu, J. Shen, W. E. Bentley, and G. F.Payne, “Redox is a global biodevice information processing modality,”
Proc. IEEE , vol. 107, no. 7, pp. 1402–1424, Jul. 2019.[57] J. M. Jornet, Y. Bae, C. R. Handelmann, B. Decker, A. Balcerak,A. Sangwan, P. Miao, A. Desai, L. Feng, E. K. Stachowiak, and M. K.Stachowiak, “Optogenomic interfaces: Bridging biological networkswith the electronic digital world,”
Proc. IEEE , vol. 107, no. 7, pp.1387–1401, Jul. 2019.[58] B. D. Unluturk, M. S. Islam, S. Balasubramaniam, and S. Ivanov,“Towards concurrent data transmission: Exploiting plasmid diversityby bacterial conjugation,”
IEEE Trans. Nanobiosci. , vol. 16, no. 4, pp.287–298, Jun. 2017.[59] A. Lanzavecchia, “Dynamics of T lymphocyte responses: Intermedi-ates, effectors, and memory cells,”
Science , vol. 290, no. 5489, pp.92–97, Oct. 2000.[60] P. Compeau and P. Pevzner,
Bioinformatics Algorithms: An ActiveLearning Approach , 3rd ed. La Jolla, CA, USA: Active LearningPublishers, 2018.[61] R. Ruiz, F. de la Cruz, and R. Fernandez-Lopez, “Negative feedbackincreases information transmission, enabling bacteria to discriminatesublethal antibiotic concentrations,”
Sci. Adv. , vol. 4, no. 11, p.eaat5771, Nov. 2018.[62] S. M. Cretoiu and L. M. Popescu, “Telocytes revisited,”
Biomol.Concepts , vol. 5, no. 5, pp. 353–369, Jan. 2014.[63] D. Cretoiu, B. M. Radu, A. Banciu, D. D. Banciu, and S. M. Cre- toiu, “Telocytes heterogeneity: From cellular morphology to functionalevidence,” Semin. Cell Dev. Biol. , vol. 64, pp. 26–39, Apr. 2017.[64] B. S. Donahue and R. F. Abercrombie, “Free diffusion coefficient ofionic calcium in cytoplasm,”
Cell Calcium , vol. 8, no. 6, pp. 437–448,Dec. 1987.[65] W. H. Bossert and E. O. Wilson, “The analysis of olfactory commu-nication among animals,”
J. Theor. Biol. , vol. 5, no. 3, pp. 443–469,Nov. 1963.[66] A. Tempestini, C. Monico, L. Gardini, F. Vanzi, F. S. Pavone, andM. Capitanio, “Sliding of a single lac repressor protein along DNA istuned by DNA sequence and molecular switching,”
Nucleic Acids Res. ,vol. 46, no. 10, pp. 5001–5011, Mar. 2018.[67] B. Fagrell, A. Fronek, and M. Intaglietta, “A microscope-televisionsystem for studying flow velocity in human skin capillaries,”
Am. J.Physiol. Circ. Physiol. , vol. 233, no. 2, pp. H318–H321, Aug. 1977.[68] T. Gervais and K. F. Jensen, “Mass transport and surface reactions inmicrofluidic systems,”
Chem. Eng. Sci. , vol. 61, no. 4, pp. 1102–1121,Feb. 2006.[69] D. Turner, C. Chang, K. Fang, P. Cuomo, and D. Murphy, “Kinesinmovement on glutaraldehyde-fixed microtubules,”
Anal. Biochem. , vol.242, no. 1, pp. 20–25, 1996.[70] W. C. Cole, J. B. Picone, and N. Sperelakis, “Gap junction uncouplingand discontinuous propagation in the heart. a comparison of experi-mental data with computer simulations,”
Biophys. J. , vol. 53, no. 5,pp. 809–818, May 1988.[71] M. Tkach, J. Kowal, and C. Th´ery, “Why the need and how to approachthe functional diversity of extracellular vesicles,”
Philos. Trans. R. Soc.B Biol. Sci. , vol. 373, no. 1737, p. 20160479, Jan. 2018.[72] D. Bausch-Fluck, U. Goldmann, S. M¨uller, M. van Oostrum, M. M¨uller,O. T. Schubert, and B. Wollscheid, “The in silico human surfaceome,”
Proc. Natl. Acad. Sci. , vol. 115, no. 46, pp. E10 988–E10 997, Nov.2018.[73] T. D. Wyatt,
Pheromones and Animal Behaviour . Cambridge, U.K.:Cambridge University Press, 2003.[74] L. Plener, N. Lorenz, M. Reiger, T. Ramalho, U. Gerland, and K. Jung,“The phosphorylation flow of the Vibrio harveyi quorum-sensing cas-cade determines levels of phenotypic heterogeneity in the population,”
J. Bacteriol. , vol. 197, no. 10, pp. 1747–1756, May 2015.[75] J. Grote, D. Krysciak, and W. R. Streit, “Phenotypic heterogeneity, aphenomenon that may explain why quorum sensing does not alwaysresult in truly homogenous cell behavior,”
Appl. Environ. Microbiol. ,vol. 81, no. 16, pp. 5280–5289, Aug. 2015.[76] M. Rebolleda-G´omez and C. W. Wood, “Unclear intentions: Eaves-dropping in microbial and plant systems,”
Front. Ecol. Evol. , vol. 7,pp. 1–13, Oct. 2019.[77] A. Einstein,
Investigations on the Theory of the Brownian Movement .New York, NY, USA: Dover Publications, 1956.[78] T. Nakano, T. Suda, M. Moore, R. Egashira, A. Enomoto, andK. Arima, “Molecular communication for nanomachines using intercel-lular calcium signaling,” in
Proc. IEEE Conf. Nanotechnol. , Jul. 2005,pp. 478–481.[79] H. C. Berg,
Random Walks in Biology . Princeton, NJ, USA: PrincetonUniversity Press, 1993.[80] W. Hundsdorfer and J. Verwer,
Numerical Solution of Time-DependentAdvection-Diffusion-Reaction Equations . Berlin, Heidelberg: Springer,2003.[81] B. Kirby,
Micro- and Nanoscale Fluid Mechanics: Transport in Mi-crofluidic Devices . Cambridge, U.K.: Cambridge University Press,2010.[82] H. Bruus,
Theoretical Microfluidics . London, U.K.: Oxford Univ.Press, 2008.[83] C. S. Kumar,
Microfluidic Devices in Nanotechnology: FundamentalConcepts . Hoboken, NJ, USA: Wiley, 2010.[84] L. Back, J. Radbill, Y. Cho, and D. Crawford, “Measurement andprediction of flow through a replica segment of a mildly atheroscleroticcoronary artery of man,”
J. Biomech. , vol. 19, no. 1, pp. 1–17, Jan.1986.[85] J. A. Jensen,
Lectures Notes on Medical Imaging Systems–Lecture5: Blood Flow in the Human Body . Lyngby, Denmark: TechnicalUniversity of Denmark, 2018.[86] R. F. Probstein,
Physicochemical Hydrodynamics: An introduction .Hoboken, NJ, USA: Wiley, 1994.[87] W. Wicke, T. Schwering, A. Ahmadzadeh, V. Jamali, A. Noel, andR. Schober, “Modeling duct flow for molecular communication,” in
Proc. IEEE GLOBECOM , Dec. 2018, pp. 206–212. [88] D. Bi, Y. Deng, M. Pierobon, and A. Nallanathan, “Chemical reactions-based microfluidic transmitter and receiver design for molecular com-munication,”
IEEE Trans. Commun. , pp. 1–16, 2020.[89] E. Yariv, G. Ben-Dov, and K. D. Dorfman, “Polymerase chain reactionin natural convection systems: A convection-diffusion-reaction model,”
Europhys. Lett. , vol. 71, no. 6, pp. 1008–1014, Sep. 2005.[90] T. M. Squires, R. J. Messinger, and S. R. Manalis, “Making it stick:Convection, reaction and diffusion in surface-based biosensors,”
Nat.Biotechnol. , vol. 26, no. 4, pp. 417–426, Apr. 2008.[91] M. T. van Genuchten and W. J. Alves,
Analytical Solutions ofthe One-dimensional Convective-dispersive Solute Transport Equation .Washington, D.C., USA: US Department of Agriculture, AgriculturalResearch Service, 1982, no. 1661.[92] R. Phillips, J. Kondev, J. Theriot, H. G. Garcia, and N. Orme,
PhysicalBiology of the Cell , 2nd ed. New York, NY, USA: Garland Science,2013.[93] S. M. Smith, R. Renden, and H. von Gersdorff, “Synaptic vesicleendocytosis: fast and slow modes of membrane retrieval,”
TrendsNeurosci. , vol. 31, no. 11, pp. 559–568, Nov. 2008.[94] T. M. Nieuwenhuizen, S. Klumpp, and R. Lipowsky, “Random walks ofmolecular motors arising from diffusional encounters with immobilizedfilaments,”
Phys. Rev. E , vol. 69, no. 6, p. 061911, Jun. 2004.[95] S. Hiyama, Y. Isogawa, T. Suda, Y. Moritani, and K. Sutoh, “A designof an autonomous molecule loading/transporting/unloading system us-ing DNA hybridization and biomolecular linear motors,” in
Proc. Eur.NanoSyst. , Dec. 2005, pp. 75–80.[96] S. Hiyama, T. Inoue, T. Shima, Y. Moritani, T. Suda, and K. Sutoh,“Autonomous loading, transport, and unloading of specified cargoes byusing DNA hybridization and biological motor-based motility,”
Small ,vol. 4, no. 4, pp. 410–415, Apr. 2008.[97] N. Farsad, A. W. Eckford, S. Hiyama, and Y. Moritani, “On-chip molec-ular communication: Analysis and design,”
IEEE Trans. Nanobiosci. ,vol. 11, no. 3, pp. 304–314, Sep. 2012.[98] N. Farsad, A. W. Eckford, and S. Hiyama, “Design and optimizing ofon-chip kinesin substrates for molecular communication,”
IEEE Trans.Nanotechnol. , vol. 14, no. 4, pp. 699–708, Jul. 2015.[99] S. Balasubramaniam and P. Li`o, “Multi-hop conjugation based bacteriananonetworks,”
IEEE Trans. Nanobiosci. , vol. 12, no. 1, pp. 47–59,Mar. 2013.[100] G. Castorina, L. Galluccio, and S. Palazzo, “On modeling informationspreading in bacterial nano-networks based on plasmid conjugation,”
IEEE Trans. Nanobiosci. , vol. 15, no. 6, pp. 567–575, Sep. 2016.[101] T. Nakano, Y. Okaie, S. Kobayashi, T. Koujin, C.-H. Chan, Y.-H.Hsu, T. Obuchi, T. Hara, Y. Hiraoka, and T. Haraguchi, “Performanceevaluation of leader-follower-based mobile molecular communicationnetworks for target detection applications,”
IEEE Trans. Commun. ,vol. 65, no. 2, pp. 663–676, Feb. 2017.[102] M. J. Schnitzer, “Theory of continuum random walks and applicationto chemotaxis,”
Phys. Rev. E , vol. 48, no. 4, pp. 2553–2568, Oct. 1993.[103] M. Gregori, I. Llatser, A. Cabellos-Aparicio, and E. Alarc´on, “Physicalchannel characterization for medium-range nanonetworks using flagel-lated bacteria,”
Comput. Netw. , vol. 55, no. 3, pp. 779–791, Feb. 2011.[104] V. Petrov, D. Moltchanov, S. Balasubramaniam, and Y. Koucheryavy,“Incorporating bacterial properties for plasmid delivery in nano sensornetworks,”
IEEE Trans. Nanotechnol. , vol. 14, no. 4, pp. 751–760, Jul.2015.[105] J. C. S´aez, V. M. Berthoud, M. C. Bra˜nes, A. D. Mart´ınez, andE. C. Beyer, “Plasma membrane channels formed by connexins: Theirregulation and functions,”
Physiol. Rev. , vol. 83, no. 4, pp. 1359–1400,Oct. 2003.[106] D. A. Goodenough and D. L. Paul, “Gap junctions,”
Cold Spring Harb.Perspect. Biol. , vol. 1, no. 1, p. a002576, Jul. 2009.[107] M. M. Dewey and L. Barr, “The American association for the ad-vancement of science,”
J. Clin. Endocrinol. Metab. , vol. 10, no. 10,pp. 1361–1362, Oct. 1950.[108] P. N. T. Unwin and G. Zampighi, “Structure of the junction betweencommunicating cells,”
Nature , vol. 283, no. 5747, pp. 545–549, Feb.1980.[109] S. Maeda, S. Nakagawa, M. Suga, E. Yamashita, A. Oshima, Y. Fu-jiyoshi, and T. Tsukihara, “Structure of the connexin 26 gap junctionchannel at 3.5 ˚A resolution,”
Nature , vol. 458, no. 7238, pp. 597–602,Apr. 2009.[110] E. B. Trexler, M. V. Bennett, T. A. Bargiello, and V. K. Verselis,“Voltage gating and permeation in a gap junction hemichannel.”
Proc.Natl. Acad. Sci. , vol. 93, no. 12, pp. 5836–5841, Jun. 1996. [111] S. Oh, C. K. Abrams, V. K. Verselis, and T. A. Bargiello, “Stoichiome-try of Transjunctional Voltage–Gating Polarity Reversal by a NegativeCharge Substitution in the Amino Terminus of a Connexin32 Chimera,” J. Gen. Physiol. , vol. 116, no. 1, pp. 13–32, Jul. 2000.[112] B. C. Bennett, M. D. Purdy, K. A. Baker, C. Acharya, W. E. McIntire,R. C. Stevens, Q. Zhang, A. L. Harris, R. Abagyan, and M. Yeager, “Anelectrostatic mechanism for Ca2+-mediated regulation of gap junctionchannels,”
Nat. Commun. , vol. 7, no. 1, p. 8770, Apr. 2016.[113] C. Peracchia, “Chemical gating of gap junction channels,”
Biochim.Biophys. Acta - Biomembr. , vol. 1662, no. 1-2, pp. 61–80, Mar. 2004.[114] A. Moreno and A. Lau, “Gap junction channel gating modulatedthrough protein phosphorylation,”
Prog. Biophys. Mol. Biol. , vol. 94,no. 1-2, pp. 107–119, May 2007.[115] J. L. Solan and P. D. Lampe, “Spatio-temporal regulation of connexin43phosphorylation and gap junction dynamics,”
Biochim. Biophys. Acta- Biomembr. , vol. 1860, no. 1, pp. 83–90, Jan. 2018.[116] T. Nakano, T. Suda, T. Koujin, T. Haraguchi, and Y. Hiraoka, “Molec-ular communication through gap junction channels,” in
Trans. Comput.Syst. Biol. X . Berlin, Heidelberg: Springer, 2008, vol. 5410, pp. 81–99.[117] D. Kilinc and O. B. Akan, “An information theoretical analysis ofnanoscale molecular gap junction communication channel betweencardiomyocytes,”
IEEE Trans. Nanotechnol. , vol. 12, no. 2, pp. 129–136, Mar. 2013.[118] A. O. Bicen, I. F. Akyildiz, S. Balasubramaniam, and Y. Koucheryavy,“Linear channel modeling and error analysis for intra/inter-cellularCa2+ molecular communication,”
IEEE Trans. Nanobiosci. , vol. 15,no. 5, pp. 488–498, Jul. 2016.[119] J. O. Brunkard, A. M. Runkel, and P. C. Zambryski, “The cytosol mustflow: intercellular transport through plasmodesmata,”
Curr. Opin. CellBiol. , vol. 35, pp. 13–20, Aug. 2015.[120] W. J. Nicolas, M. S. Grison, and E. M. Bayer, “Shaping intercellularchannels of plasmodesmata: The structure-to-function missing link,”
J.Exp. Bot. , vol. 69, no. 1, pp. 91–103, Jan. 2018.[121] K. Bell and K. Oparka, “Imaging plasmodesmata,”
Protoplasma , vol.248, no. 1, pp. 9–25, Jan. 2011.[122] S. Amsbury, P. Kirk, and Y. Benitez-Alfonso, “Emerging models onthe regulation of intercellular transport by plasmodesmata-associatedcallose,”
J. Exp. Bot. , vol. 69, no. 1, pp. 105–115, Jan. 2018.[123] J. Guglielmini, L. Quintais, M. P. Garcill´an-Barcia, F. de la Cruz, andE. P. C. Rocha, “The repertoire of ice in prokaryotes underscores theunity, diversity, and ubiquity of conjugation,”
PLoS Genet. , vol. 7, no. 8,p. e1002222, Aug. 2011.[124] G. Gu´edon, V. Libante, C. Coluzzi, S. Payot, and N. Leblond-Bourget,“The obscure world of integrative and mobilizable elements, highlywidespread elements that pirate bacterial conjugative systems,”
Genes(Basel). , vol. 8, no. 11, p. 337, Nov. 2017.[125] E. Cabez´on, J. Ripoll-Rozada, A. Pe˜na, F. de la Cruz, and I. Arechaga,“Towards an integrated model of bacterial conjugation,”
FEMS Micro-biol. Rev. , vol. 39, no. 1, pp. 81–95, Sep. 2014.[126] L. S. Frost, R. Leplae, A. O. Summers, and A. Toussaint, “Mobilegenetic elements: The agents of open source evolution,”
Nat. Rev.Microbiol. , vol. 3, no. 9, pp. 722–732, Sep. 2005.[127] C. Smillie, M. P. Garcillan-Barcia, M. V. Francia, E. P. C. Rocha,and F. de la Cruz, “Mobility of plasmids,”
Microbiol. Mol. Biol. Rev. ,vol. 74, no. 3, pp. 434–452, Sep. 2010.[128] L. Thoma and G. Muth, “The conjugative DNA-transfer apparatus ofstreptomyces,”
Int. J. Med. Microbiol. , vol. 305, no. 2, pp. 224–229,Feb. 2015.[129] G. Waksman, “From conjugation to T4S systems in Gram-negativebacteria: A mechanistic biology perspective,”
EMBO Rep. , vol. 20,no. 2, p. e47012, Feb. 2019.[130] X. Bellanger, S. Payot, N. Leblond-Bourget, and G. Gu´edon, “Con-jugative and mobilizable genomic islands in bacteria: Evolution anddiversity,”
FEMS Microbiol. Rev. , vol. 38, no. 4, pp. 720–760, Jul.2014.[131] C. Ambroset, C. Coluzzi, G. Gu´edon, M.-D. Devignes, V. Loux,T. Lacroix, S. Payot, and N. Leblond-Bourget, “New insights into theclassification and integration specificity of streptococcus integrativeconjugative elements through extensive genome exploration,”
Front.Microbiol. , vol. 6, no. 1, pp. 1–21, Jan. 2016.[132] P. Li`o and S. Balasubramaniam, “Opportunistic routing through conju-gation in bacteria communication nanonetwork,”
Nano Commun. Net. ,vol. 3, no. 1, pp. 36–45, Mar. 2012.[133] A. Rustom, R. Saffrich, I. Markovic, P. Walther, and H. Gerdes,“Nanotubular highways for intercellular organelle transport,”
Science ,vol. 303, no. 5660, pp. 1007–1010, Feb. 2004. [134] N. M. Sherer and W. Mothes, “Cytonemes and tunneling nanotubulesin cell-cell communication and viral pathogenesis,”
Trends Cell Biol. ,vol. 18, no. 9, pp. 414–420, Sep. 2008.[135] S. Gurke, J. F. V. Barroso, and H.-H. Gerdes, “The art of cellularcommunication: Tunneling nanotubes bridge the divide,”
Histochem.Cell Biol. , vol. 129, no. 5, pp. 539–550, May 2008.[136] L. Edelstein, K. Fuxe, M. Levin, B. O. Popescu, and J. Smythies,“Telocytes in their context with other intercellular communicationagents,”
Semin. Cell Dev. Biol. , vol. 55, pp. 9–13, Jul. 2016.[137] M.-S. Faussone-Pellegrini and M. Gherghiceanu, “Telocyte’s contacts,”
Semin. Cell Dev. Biol. , vol. 55, pp. 3–8, Jul. 2016.[138] C. Giorgi, A. Danese, S. Missiroli, S. Patergnani, and P. Pinton,“Calcium dynamics as a machine for decoding signals,”
Trends CellBiol. , vol. 28, no. 4, pp. 258–273, Apr. 2018.[139] M. Kuscu and O. B. Akan, “On the physical design of molecularcommunication receiver based on nanoscale biosensors,”
IEEE Sens.J. , vol. 16, no. 8, pp. 2228–2243, Apr. 2016.[140] J. W. Putney, “Capacitative calcium entry,”
J. Cell Biol. , vol. 169, no. 3,pp. 381–382, May 2005.[141] T. F. Martin, “Tuning exocytosis for speed: Fast and slow modes,”
Biochim. Biophys. Acta - Mol. Cell Res. , vol. 1641, no. 2-3, pp. 157–165, Aug. 2003.[142] C. T. Chou, “A Markovian approach to the optimal demodulationof diffusion-based molecular communication networks,”
IEEE Trans.Commun. , vol. 63, no. 10, pp. 3728–3743, Oct. 2015.[143] H. Arjmandi, A. Ahmadzadeh, R. Schober, and M. Nasiri Kenari,“Ion channel based bio-synthetic modulator for diffusive molecularcommunication,”
IEEE Trans. Nanobiosci. , vol. 15, no. 5, pp. 418–432, Jul. 2016.[144] H. Ramezani and O. B. Akan, “Information capacity of vesicle releasein neuro-spike communication,”
IEEE Commun. Lett. , vol. 22, no. 1,pp. 41–44, Jan. 2018.[145] A. G. Gilman, “G proteins: Transducers of receptor-generated signals,”
Annu. Rev. Biochem. , vol. 56, no. 1, pp. 615–649, Jun. 1987.[146] V. Syrovatkina, K. O. Alegre, R. Dey, and X. Y. Huang, “Regulation,signaling, and physiological functions of G-proteins,”
J. Mol. Biol. , vol.428, no. 19, pp. 3850–3868, Sep. 2016.[147] D. E. Clapham, “Calcium signaling,”
Cell , vol. 131, no. 6, pp. 1047–1058, Dec. 2007.[148] B. Lindner, “Mechanisms of information filtering in neural systems,”
IEEE Trans. Mol. Biol. Multi-Scale Commun. , vol. 2, no. 1, pp. 5–15,Jun. 2016.[149] K. H. Srivastava, C. M. Holmes, M. Vellema, A. R. Pack, C. P. H.Elemans, I. Nemenman, and S. J. Sober, “Motor control by preciselytimed spike patterns,”
Proc. Natl. Acad. Sci. , vol. 114, no. 5, pp. 1171–1176, Jan. 2017.[150] F. Dinc, B. C. Akdeniz, A. E. Pusane, and T. Tugcu, “Impulse responseof the molecular diffusion channel with a spherical absorbing receiverand a spherical reflective boundary,”
IEEE Trans. Mol. Biol. Multi-Scale Commun. , vol. 4, no. 2, pp. 118–122, Mar. 2019.[151] Ankit and M. R. Bhatnagar, “Molecular channel characterization for arectangular container with reflecting and absorbing boundaries,”
IEEECommun. Lett. , vol. 24, no. 2, pp. 234–238, Feb. 2020.[152] J. Jaeger and H. Carslaw,
Conduction of Heat in Solids . London,U.K.: Oxford Univ. Press, 1986.[153] A. Noel, K. C. Cheung, and R. Schober, “Diffusive molecular com-munication with disruptive flows,” in
Proc. IEEE ICC , Jun. 2014, pp.3600–3606.[154] L. Zha, Y. Deng, A. Noel, M. Elkashlan, and A. Nallanathan,“Transceiver observations in asymmetric and symmetric diffusivemolecular communication systems,” in
Proc. IEEE GLOBECOM , Dec.2018, pp. 206–212.[155] A. Noel, K. C. Cheung, and R. Schober, “A unifying model for externalnoise sources and ISI in diffusive molecular communication,”
IEEE J.Sel. Areas Commun. , vol. 32, no. 12, pp. 2330–2343, Dec. 2014.[156] A. Noel, D. Makrakis, and A. Hafid, “Channel impulse responsesin diffusive molecular communication with spherical transmitters,” in
Proc. Bienn. Symp. Commun. , Apr. 2016.[157] C. R. Kahn, “Membrane receptors for hormones and neurotransmitters,”
J. Cell Biol. , vol. 70, no. 2, pp. 261–286, Aug. 1976.[158] S. S. Andrews, “Accurate particle-based simulation of adsorption,desorption and partial transmission,”
Phys. Biol. , vol. 6, no. 4, p.046015, Nov. 2009.[159] S. Mangan, A. Zaslaver, and U. Alon, “The coherent feedforward loopserves as a sign-sensitive delay element in transcription networks,”
J.Mol. Biol. , vol. 334, no. 2, pp. 197–204, Nov. 2003. [160] S. S. Shen-Orr, R. Milo, S. Mangan, and U. Alon, “Network motifs inthe transcriptional regulation network of Escherichia coli,” Nat. Genet. ,vol. 31, no. 1, pp. 64–68, May 2002.[161] T. I. Lee, “Transcriptional regulatory networks in Saccharomycescerevisiae,”
Science , vol. 298, no. 5594, pp. 799–804, Oct. 2002.[162] R. Milo, “Network motifs: Simple building blocks of complex net-works,”
Science , vol. 298, no. 5594, pp. 824–827, Oct. 2002.[163] B.-H. Koo, H. J. Kim, J.-Y. Kwon, and C.-B. Chae, “Deep learning-based human implantable nano molecular communications,” in
Proc.IEEE ICC , Jun. 2020, pp. 1–7.[164] G. Raposo and W. Stoorvogel, “Extracellular vesicles: Exosomes,microvesicles, and friends,”
J. Cell Biol. , vol. 200, no. 4, pp. 373–383,Feb. 2013.[165] N. Ferruz and G. De Fabritiis, “Binding kinetics in drug discovery,”
Mol. Inform. , vol. 35, no. 6-7, pp. 216–226, Jul. 2016.[166] N. Allbritton, T. Meyer, and L. Stryer, “Range of messenger actionof calcium ion and inositol 1,4,5-trisphosphate,”
Science , vol. 258, no.5089, pp. 1812–1815, Dec. 1992.[167] C. Cremer and T. Cremer, “Considerations on a laser-scanning-microscope with high resolution and depth of field,”
Microsc. Acta ,vol. 81, no. 1, pp. 31–44, Sep. 1978.[168] X. Hao, C. Kuang, Z. Gu, Y. Wang, S. Li, Y. Ku, Y. Li, J. Ge, andX. Liu, “From microscopy to nanoscopy via visible light,”
Light Sci.Appl. , vol. 2, no. 10, p. e108, Oct. 2013.[169] H. Zhou, C. Qin, R. Chen, Y. Liu, W. Zhou, G. Zhang, Y. Gao, L. Xiao,and S. Jia, “Quantum coherent modulation-enhanced single-moleculeimaging microscopy,”
J. Phys. Chem. Lett. , vol. 10, no. 2, pp. 223–228,Jan. 2019.[170] S. Sridharan Weaver, Y. Li, L. Foucard, H. Majeed, B. Bhaduri, A. J.Levine, K. A. Kilian, and G. Popescu, “Simultaneous cell traction andgrowth measurements using light,”
J. Biophotonics , vol. 12, no. 3, pp.1–11, Mar. 2019.[171] W. Lee, N. Kalashnikov, S. Mok, R. Halaoui, E. Kuzmin, A. J. Putnam,S. Takayama, M. Park, L. McCaffrey, R. Zhao, R. L. Leask, andC. Moraes, “Dispersible hydrogel force sensors reveal patterns of solidmechanical stress in multicellular spheroid cultures,”
Nat. Commun. ,vol. 10, no. 1, p. 144, Dec. 2019.[172] R. W. Taylor, R. G. Mahmoodabadi, V. Rauschenberger, A. Giessl,A. Schambony, and V. Sandoghdar, “Interferometric scattering mi-croscopy reveals microsecond nanoscopic protein motion on a live cellmembrane,”
Nat. Photonics , vol. 13, no. 7, pp. 480–487, Jul. 2019.[173] D. Axelrod, “Total internal reflection fluorescence microscopy,” in
Encycl. Cell Biol.
Amsterdam: Elsevier, 2016, vol. 2, pp. 62–69.[174] J. D. Keighron, A. G. Ewing, and A. S. Cans, “Analytical tools tomonitor exocytosis: A focus on new fluorescent probes and methods,”
Analyst , vol. 137, no. 8, pp. 1755–1763, Feb. 2012.[175] A. V. Kibble, R. J. Barnard, and R. D. Burgoyne, “Patch-clampcapacitance analysis of the effects of α -snap on exocytosis in adrenalchromaffin cells,” J. Cell Sci. , vol. 109, no. 9, pp. 2417–2422, Jun.1996.[176] M. Lindau, “High resolution electrophysiological techniques for thestudy of calcium-activated exocytosis,”
Biochim. Biophys. Acta - Gen.Subj. , vol. 1820, no. 8, pp. 1234–1242, Aug. 2012.[177] P. Thorn, “Measuring calcium signals and exocytosis in tissues,”
Biochim. Biophys. Acta - Gen. Subj. , vol. 1820, no. 8, pp. 1179–1184,Aug. 2012.[178] X. Liu, Y. Tong, and P.-P. Fang, “Recent development in amperometricmeasurements of vesicular exocytosis,”
TrAC Trends Anal. Chem. , vol.113, pp. 13–24, Apr. 2019.[179] A. Manz, N. Graber, and H. Widmer, “Miniaturized total chemicalanalysis systems: A novel concept for chemical sensing,”
SensorsActuators B Chem. , vol. 1, no. 1-6, pp. 244–248, Jan. 1990.[180] D. Psaltis, S. R. Quake, and C. Yang, “Developing optofluidic technol-ogy through the fusion of microfluidics and optics,”
Nature , vol. 442,no. 7101, pp. 381–386, Jul. 2006.[181] P. Minzioni, R. Osellame, C. Sada, S. Zhao, F. G. Omenetto, K. B.Gylfason, T. Haraldsson, Y. Zhang, A. Ozcan, A. Wax, F. Mugele,H. Schmidt, G. Testa, R. Bernini, J. Guck, C. Liberale, K. Berg-Sørensen, J. Chen, M. Pollnau, S. Xiong, A.-Q. Liu, C.-C. Shiue, S.-K.Fan, D. Erickson, and D. Sinton, “Roadmap for optofluidics,”
J. Opt. ,vol. 19, no. 9, p. 093003, Sep. 2017.[182] D. J. Harrison, K. Fluri, K. Seiler, Z. Fan, C. S. Effenhauser, andA. Manz, “Micromachining a miniaturized capillary electrophoresis-based chemical analysis system on a chip,”
Science , vol. 261, no. 5123,pp. 895–897, Aug. 1993. [183] X. Mao, S.-C. S. Lin, C. Dong, and T. J. Huang, “Single-layerplanar on-chip flow cytometer using microfluidic drifting based three-dimensional (3d) hydrodynamic focusing,”
Lab Chip , vol. 9, no. 11, p.1583, Mar. 2009.[184] M. I. Lapsley, I.-K. Chiang, Y. B. Zheng, X. Ding, X. Mao, and T. J.Huang, “A single-layer, planar, optofluidic Mach-Zehnder interferom-eter for label-free detection,”
Lab Chip , vol. 11, no. 10, p. 1795, Apr.2011.[185] J. S. Mak, S. A. Rutledge, R. M. Abu-Ghazalah, F. Eftekhari, J. Irizar,N. C. Tam, G. Zheng, and A. S. Helmy, “Recent developments inoptofluidic-assisted Raman spectroscopy,”
Prog. Quantum Electron. ,vol. 37, no. 1, pp. 1–50, Jan. 2013.[186] X. Fan, I. M. White, S. I. Shopova, H. Zhu, J. D. Suter, and Y. Sun,“Sensitive optical biosensors for unlabeled targets: A review,”
Anal.Chim. Acta , vol. 620, no. 1-2, pp. 8–26, Jul. 2008.[187] N.-T. Huang, H.-l. Zhang, M.-T. Chung, J. H. Seo, and K. Kurabayashi,“Recent advancements in optofluidics-based single-cell analysis: Op-tical on-chip cellular manipulation, treatment, and property detection,”
Lab Chip , vol. 14, no. 7, pp. 1230–1245, Dec. 2014.[188] J. Estelrich, M. J. S´anchez-Mart´ın, and M. A. Busquets, “Nanoparticlesin magnetic resonance imaging: From simple to dual contrast agents,”
Int. J. Nanomedicine , vol. 10, pp. 1727–1741, Mar. 2015.[189] D. Maresca, A. Lakshmanan, M. Abedi, A. Bar-Zion, A. Farhadi, G. J.Lu, J. O. Szablowski, D. Wu, S. Yoo, and M. G. Shapiro, “Biomolecularultrasound and sonogenetics,”
Annu. Rev. Chem. Biomol. Eng. , vol. 9,no. 1, pp. 229–252, Jun. 2018.[190] A. Farhadi, G. H. Ho, D. P. Sawyer, R. W. Bourdeau, and M. G.Shapiro, “Ultrasound imaging of gene expression in mammalian cells,”
Science , vol. 365, no. 6460, pp. 1469–1475, Sep. 2019.[191] Q. A. Pankhurst, J. Connolly, S. K. Jones, and J. Dobson, “Applicationsof magnetic nanoparticles in biomedicine,”
J. Phys. D. Appl. Phys. ,vol. 36, no. 13, pp. R167–R181, Jul. 2003.[192] M. Arruebo, R. Fern´andez-Pacheco, M. R. Ibarra, and J. Santamar´ıa,“Magnetic nanoparticles for drug delivery,”
Nano Today , vol. 2, no. 3,pp. 22–32, Jun. 2007.[193] C. Sun, J. S. Lee, and M. Zhang, “Magnetic nanoparticles in MRimaging and drug delivery,”
Adv. Drug Deliv. Rev. , vol. 60, no. 11,pp. 1252–1265, Aug. 2008.[194] S. Kisseleff, R. Schober, and W. H. Gerstacker, “Magnetic nanoparticlebased interface for molecular communication systems,”
IEEE Commun.Lett. , vol. 21, no. 2, pp. 258–261, Feb. 2017.[195] H. Unterweger, J. Kirchner, W. Wicket, A. Ahmadzadeh, D. Ahmed,V. Jamali, C. Alexiou, G. Fischer, and R. Schober, “Experimentalmolecular communication testbed based on magnetic nanoparticles induct flow,” in
Proc. IEEE SPAWC , Jun. 2018, pp. 1–5.[196] I. F. Akyildiz and J. M. Jornet, “Electromagnetic wireless nanosensornetworks,”
Nano Commun. Netw. , vol. 1, no. 1, pp. 3–19, Mar. 2010.[197] D.-K. Lee, J.-H. Kang, J.-S. Lee, H.-S. Kim, C. Kim, J. Hun Kim,T. Lee, J.-H. Son, Q.-H. Park, and M. Seo, “Highly sensitive andselective sugar detection by terahertz nano-antennas,”
Sci. Rep. , vol. 5,no. 1, p. 15459, Dec. 2015.[198] L. Grebenstein, J. Kirchner, R. S. Peixoto, W. Zimmermann, F. Irn-storfer, W. Wicke, A. Ahmadzadeh, V. Jamali, G. Fischer, R. Weigel,A. Burkovski, and R. Schober, “Biological optical-to-chemical signalconversion interface: A small-scale modulator for molecular commu-nications,”
IEEE Trans. Nanobiosci. , vol. 18, no. 1, pp. 31–42, Jan.2019.[199] N. Farsad, D. Pan, and A. Goldsmith, “A novel experimental platformfor in-vessel multi-chemical molecular communications,” in
Proc. IEEEGLOBECOM , Dec. 2017, pp. 1–6.[200] E. Osmekhina, C. Jonkergouw, G. Schmidt, F. Jahangiri, V. Jokinen,S. Franssila, and M. B. Linder, “Controlled communication betweenphysically separated bacterial populations in a microfluidic device,”
Commun. Biol. , vol. 1, no. 97, pp. 1–7, Dec. 2018.[201] T. Luo, R. Zheng, J. Song, L. Lin, and H. Yan, “A small-scale modu-lator of electric-to-biological signal conversion for synthetic molecularcommunications,” in
Proc. IEEE ICC , Jun. 2020, pp. 1–7.[202] G. C. Ellis-Davies, “Caged compounds: Photorelease technology forcontrol of cellular chemistry and physiology,”
Nat. Methods , vol. 4,no. 8, pp. 619–628, Aug. 2007.[203] D. Needham, G. Anyarambhatla, G. Kong, and M. W. Dewhirst, “Anew temperature-sensitive liposome for use with mild hyperthermia:Characterization and testing in a human tumor xenograft model,”
Cancer Res. , vol. 60, no. 5, pp. 1197–1201, Mar. 2000.[204] K. Kono, T. Miyoshi, Y. Haba, E. Murakami, C. Kojima, andA. Harada, “Temperature sensitivity control of alkylamide-terminated poly(amidoamine) dendrimers induced by guest molecule binding,” J.Am. Chem. Soc. , vol. 129, no. 23, pp. 7222–7223, Jun. 2007.[205] F. Liu, V. Kozlovskaya, S. Medipelli, B. Xue, F. Ahmad, M. Saeed,D. Cropek, and E. Kharlampieva, “Temperature-sensitive polymer-somes for controlled delivery of anticancer drugs,”
Chem. Mater. ,vol. 27, no. 23, pp. 7945–7956, Nov. 2015.[206] K. Ulbrich, K. Hol´a, V. ˇSubr, A. Bakandritsos, J. Tuˇcek, and R. Zboil,“Targeted drug delivery with polymers and magnetic nanoparticles:Covalent and noncovalent approaches, release control, and clinicalstudies,”
Chem. Rev. , vol. 116, no. 9, pp. 5338–5431, May 2016.[207] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Modulationtechniques for communication via diffusion in nanonetworks,” in
Proc.IEEE ICC , Jun. 2011, pp. 1–5.[208] A. Biral and A. Zanella, “Introducing purely hydrodynamic networkingmechanisms in microfluidic systems,” in
Proc. IEEE ICC Work. , Jun.2013, pp. 798–803.[209] D. Hathcock, J. Sheehy, C. Weisenberger, E. Ilker, and M. Hinczewski,“Noise filtering and prediction in biological signaling networks,”
IEEETrans. Mol. Biol. Multi-Scale Commun. , vol. 2, no. 1, pp. 16–30, Jun.2016.[210] R. Suderman, J. A. Bachman, A. Smith, P. K. Sorger, and E. J. Deeds,“Fundamental trade-offs between information flow in single cells andcellular populations,”
Proc. Natl. Acad. Sci. , vol. 114, no. 22, pp. 5755–5760, May 2017.[211] P. J. Thomas and A. W. Eckford, “Capacity of a simple intercellularsignal transduction channel,”
IEEE Trans. Inf. Theory , vol. 62, no. 12,pp. 7358–7382, Dec. 2016.[212] J. Selimkhanov, B. Taylor, J. Yao, A. Pilko, J. Albeck, A. Hoffmann,L. Tsimring, and R. Wollman, “Accurate information transmissionthrough dynamic biochemical signaling networks.”
Science , vol. 346,no. 6215, pp. 1370–1373, Dec. 2014.[213] D. Ellison, A. Mugler, M. D. Brennan, S. H. Lee, R. J. Huebner, E. R.Shamir, L. A. Woo, J. Kim, P. Amar, I. Nemenman, A. J. Ewald, andA. Levchenko, “Cell-cell communication enhances the capacity of cellensembles to sense shallow gradients during morphogenesis,”
Proc.Natl. Acad. Sci. , vol. 113, no. 6, pp. E679–E688, Feb. 2016.[214] A. Mugler, A. Levchenko, and I. Nemenman, “Limits to the precisionof gradient sensing with spatial communication and temporal integra-tion,”
Proc. Natl. Acad. Sci. , vol. 113, no. 6, pp. E689–E695, Feb.2016.[215] S. Fancher and A. Mugler, “Fundamental limits to collective concen-tration sensing in cell populations,”
Phys. Rev. Lett. , vol. 118, no. 7,p. 078101, Feb. 2017.[216] S. Mukherjee and B. L. Bassler, “Bacterial quorum sensing in complexand dynamically changing environments,”
Nat. Rev. Microbiol. , vol. 17,no. 6, pp. 371–382, Jun. 2019.[217] M. Pierobon, Z. Sakkaff, J. L. Catlett, and N. R. Buan, “Mutualinformation upper bound of molecular communication based on cellmetabolism,” in
Proc. IEEE SPAWC , Jul. 2016, pp. 1–6.[218] Z. Sakkaff, J. L. Catlett, M. Cashman, M. Pierobon, N. R. Buan,M. B. Cohen, and C. A. Kelley, “End-to-end molecular communicationchannels in cell metabolism: An information theoretic study,” in
Proc.ACM NANOCOM , Sep. 2017, pp. 1–6.[219] Z. Sakkaff, A. Immaneni, and M. Pierobon, “Estimating the molecularinformation through cell signal transduction pathways,” in
Proc. IEEESPAWC , Jun. 2018, pp. 1–5.[220] Z. Sakkaff, B. D. Unluturk, and M. Pierobon, “Applying molecularcommunication theory to multi-scale integrated models of biologicalpathways,” in
Proc. ACM NANOCOM , Sep. 2019, pp. 1–2.[221] T. V. Martins, J. Hammelman, S. Marinova, C. O. Ding, and R. J.Morris, “An information-theoretical approach for calcium signalingspecificity,”
IEEE Trans. Nanobiosci. , vol. 18, no. 1, pp. 93–100, Jan.2019.[222] N. A. Abbasi and O. B. Akan, “An information theoretical analysis ofhuman insulin-glucose system toward the internet of bio-nano things,”
IEEE Trans. Nanobiosci. , vol. 16, no. 8, pp. 783–791, Dec. 2017.[223] H. Awan, R. S. Adve, N. Wallbridge, C. Plummer, and A. W. Eckford,“Communication and information theory of single action potentialsignals in plants,”
IEEE Trans. Nanobiosci. , vol. 18, no. 1, pp. 61–73, Jan. 2019.[224] ——, “Information theoretic based comparative analysis of differentcommunication signals in plants,”
IEEE Access , vol. 7, pp. 117 075–117 087, Aug. 2019.[225] M. Cook, D. Soloveichik, E. Winfree, and J. Bruck, “Programmabilityof chemical reaction networks,” in
Algorithmic Bioprocesses . Berlin,Heidelberg: Springer Verlag, 2009, pp. 543–584. [226] R. Weiss, S. Basu, S. Hooshangi, A. Kalmbach, D. Karig, R. Mehreja,and I. Netravali, “Genetic circuit building blocks for cellular compu-tation, communications, and signal processing,”
Nat. Comput. , vol. 2,no. 1, pp. 47–84, Mar. 2003.[227] R. W. Bradley and B. Wang, “Designer cell signal processing circuitsfor biotechnology,”
N. Biotechnol. , vol. 32, no. 6, pp. 635–643, Dec.2015.[228] R. Sarpeshkar, “Analog synthetic biology,”
Philos. Trans. R. Soc. AMath. Phys. Eng. Sci. , vol. 372, no. 2012, p. 20130110, Mar. 2014.[229] H. Jiang, M. D. Riedel, and K. K. Parhi, “Digital logic with molecularreactions,” in
Proc. IEEE/ACM ICCAD , Nov. 2013, pp. 721–727.[230] L. Ge, Z. Zhong, D. Wen, X. You, and C. Zhang, “A formal combina-tional logic synthesis with chemical reaction networks,”
IEEE Trans.Mol. Biol. Multi-Scale Commun. , vol. 3, no. 1, pp. 33–47, Mar. 2017.[231] D. Bi and Y. Deng, “Digital signal processing for molecular commu-nication via chemical reactions-based microfluidic circuits,”
Submittedto IEEE Commun. Mag. , Sep. 2020.[232] ——, “Microfluidic QCSK transmitter and receiver design for molec-ular communication,” arXiv:2005.01353 , May 2020.[233] T. Song, S. Garg, R. Mokhtar, H. Bui, and J. Reif, “Analog computationby DNA strand displacement circuits,”
ACS Synth. Biol. , vol. 5, no. 8,pp. 898–912, Aug. 2016.[234] C. T. Chou, “Chemical reaction networks for computing logarithm,”
Synth. Biol. , vol. 2, no. 1, p. ysx002, Jan. 2017.[235] L. Cardelli, M. Tribastone, and M. Tschaikowski, “From electriccircuits to chemical networks,”
Nat. Comput. , vol. 19, no. 1, pp. 237–248, Mar. 2020.[236] A. Tamsir, J. J. Tabor, and C. A. Voigt, “Robust multicellular computingusing genetically encoded NOR gates and chemical ’wires’,”
Nature ,vol. 469, no. 7329, pp. 212–215, Jan. 2011.[237] B. Wang, R. I. Kitney, N. Joly, and M. Buck, “Engineering modular andorthogonal genetic logic gates for robust digital-like synthetic biology,”
Nat. Commun. , vol. 2, no. 1, p. 508, Sep. 2011.[238] B. H. Weinberg, N. T. Pham, L. D. Caraballo, T. Lozanoski, A. Engel,S. Bhatia, and W. W. Wong, “Large-scale design of robust geneticcircuits with multiple inputs and outputs for mammalian cells,”
Nat.Biotechnol. , vol. 35, no. 5, pp. 453–462, May 2017.[239] M. N. Win and C. D. Smolke, “Higher-order cellular informationprocessing with synthetic RNA devices,”
Science , vol. 322, no. 5900,pp. 456–460, Oct. 2008.[240] J. Kim, Y. Zhou, P. D. Carlson, M. Teichmann, S. Chaudhary, F. C.Simmel, P. A. Silver, J. J. Collins, J. B. Lucks, P. Yin, and A. A. Green,“De novo-designed translation-repressing riboregulators for multi-inputcellular logic,”
Nat. Chem. Biol. , vol. 15, no. 12, pp. 1173–1182, Dec.2019.[241] J. J. Tabor, H. M. Salis, Z. B. Simpson, A. A. Chevalier, A. Levskaya,E. M. Marcotte, C. A. Voigt, and A. D. Ellington, “A synthetic geneticedge detection program.”
Cell , vol. 137, no. 7, pp. 1272–81, Jun. 2009.[242] R. Daniel, J. R. Rubens, R. Sarpeshkar, and T. K. Lu, “Synthetic analogcomputation in living cells,”
Nature , vol. 497, no. 7451, pp. 619–623,May 2013.[243] J. J. Y. Teo, S. S. Woo, and R. Sarpeshkar, “Synthetic biology: Aunifying view and review using analog circuits,”
IEEE Trans. Biomed.Circuits Syst. , vol. 9, no. 4, pp. 453–474, Aug. 2015.[244] J. Ang, E. Harris, B. J. Hussey, R. Kil, and D. R. McMillen, “Tuningresponse curves for synthetic biology,”
ACS Synth. Biol. , vol. 2, no. 10,pp. 547–567, Oct. 2013.[245] C. T. Chou, “Designing molecular circuits for approximate maximuma posteriori demodulation of concentration modulated signals,”
IEEETrans. Commun. , vol. 67, no. 8, pp. 5458–5473, Aug. 2019.[246] ——, “Molecular circuits for decoding frequency coded signals innano-communication networks,”
Nano Commun. Netw. , vol. 3, no. 1,pp. 46–56, Mar. 2012.[247] M. Egan, T. Q. Duong, and M. D. Renzo, “Biological circuits fordetection in MoSK-based molecular communication,”
IEEE Access ,vol. 7, pp. 21 094–21 102, Feb. 2019.[248] H. Awan and C. T. Chou, “Impact of receiver molecular circuits onthe performance of reaction shift keying,” in
Proc. ACM NANOCOM ,Sep. 2015, pp. 1–6.[249] ——, “Molecular communications with molecular circuit-based trans-mitters and receivers,”
IEEE Trans. Nanobiosci. , vol. 18, no. 2, pp.146–155, Apr. 2019.[250] C. Zhang, L. Ge, X. Zhang, W. Wei, J. Zhao, Z. Zhang, Z. Wang, andX. You, “A uniform molecular low-density parity check decoder,”
ACSSynth. Biol. , vol. 8, no. 1, pp. 82–90, Jan. 2019. [251] National Human Genome Research Institute, “DNA strands.”[Online]. Available: https://commons.wikimedia.org/wiki/File:DNA {} strands.png[252] M. G. T. A. Rutten, F. W. Vaandrager, J. A. A. W. Elemans, and R. J. M.Nolte, “Encoding information into polymers,” Nat. Rev. Chem. , vol. 2,no. 11, pp. 365–381, Nov. 2018.[253] L. Ceze, J. Nivala, and K. Strauss, “Molecular digital data storage usingDNA,”
Nat. Rev. Genet. , vol. 20, no. 8, pp. 456–466, Aug. 2019.[254] O. Milenkovic, R. Gabrys, H. M. Kiah, and S. H. Tabatabaei Yazdi,“Exabytes in a test tube,”
IEEE Spectr. , vol. 55, no. 5, pp. 40–45, May2018.[255] S. Shrivastava and R. Badlani, “Data storage in DNA,”
Int. J. Electr.Energy , vol. 2, no. 2, pp. 119–124, Jun. 2014.[256] P. Y. De Silva and G. U. Ganegoda, “New trends of digital data storagein DNA,”
Biomed Res. Int. , vol. 2016, pp. 1–14, Sep. 2016.[257] L. Adleman, “Molecular computation of solutions to combinatorialproblems,”
Science , vol. 266, no. 5187, pp. 1021–1024, Nov. 1994.[258] Y.-J. Chen, C. N. Takahashi, L. Organick, C. Bee, S. D. Ang, P. Weiss,B. Peck, G. Seelig, L. Ceze, and K. Strauss, “Quantifying molecularbias in DNA data storage,”
Nat. Commun. , vol. 11, no. 1, p. 3264, Dec.2020.[259] Y. Erlich and D. Zielinski, “DNA fountain enables a robust and efficientstorage architecture,”
Science , vol. 355, no. 6328, pp. 950–954, Mar.2017.[260] L. Organick, S. D. Ang, Y. J. Chen, R. Lopez, S. Yekhanin,K. Makarychev, M. Z. Racz, G. Kamath, P. Gopalan, B. Nguyen,C. N. Takahashi, S. Newman, H. Y. Parker, C. Rashtchian, K. Stewart,G. Gupta, R. Carlson, J. Mulligan, D. Carmean, G. Seelig, L. Ceze,and K. Strauss, “Random access in large-scale DNA data storage,”
Nat.Biotechnol. , vol. 36, no. 3, pp. 242–248, Mar. 2018.[261] C. N. Tennyson, H. J. Klamut, and R. G. Worton, “The humandystrophin gene requires 16 hours to be transcribed and is cotranscrip-tionally spliced,”
Nat. Genet. , vol. 9, no. 2, pp. 184–190, Feb. 1995.[262] C. N. Takahashi, B. H. Nguyen, K. Strauss, and L. Ceze, “Demonstra-tion of end-to-end automation of DNA data storage,”
Sci. Rep. , vol. 9,no. 1, pp. 1–5, Dec. 2019.[263] C. Roggo, C. Picioreanu, X. Richard, C. Mazza, H. van Lintel, andJ. R. van der Meer, “Quantitative chemical biosensing by bacterialchemotaxis in microfluidic chips,”
Environ. Microbiol. , vol. 20, no. 1,pp. 241–258, Jan. 2018.[264] B. Wang, M. Barahona, and M. Buck, “A modular cell-based biosensorusing engineered genetic logic circuits to detect and integrate multipleenvironmental signals,”
Biosens. Bioelectron. , vol. 40, no. 1, pp. 368–376, Feb. 2013.[265] R. Lo, P.-Y. Li, S. Saati, R. N. Agrawal, M. S. Humayun, andE. Meng, “A passive MEMS drug delivery pump for treatment ofocular diseases,”
Biomed. Microdevices , vol. 11, no. 5, pp. 959–970,Oct. 2009.[266] R. Tietze, S. Lyer, S. D¨urr, T. Struffert, T. Engelhorn, M. Schwarz,E. Eckert, T. G¨oen, S. Vasylyev, W. Peukert, F. Wiekhorst, L. Trahms,A. D¨orfler, and C. Alexiou, “Efficient drug-delivery using magneticnanoparticles - biodistribution and therapeutic effects in tumour bearingrabbits,”
Nanomedicine Nanotechnology, Biol. Med. , vol. 9, no. 7, pp.961–971, May 2013.[267] G. M. Whitesides, “The origins and the future of microfluidics,”
Nature ,vol. 442, no. 7101, pp. 368–373, Jul. 2006.[268] K. M. Polizzi, “What is synthetic biology?” in
Methods Mol. Biol.
Totowa, NJ, USA: Humana Press, 2013, vol. 1073, pp. 3–6.[269] C. Kumar,
Microfluidic Devices in Nanotechnology: Applications .Hoboken, NJ, USA: John Wiley & Sons, Inc., May 2010.[270] S. Kwakye and A. Baeumner, “A microfluidic biosensor based onnucleic acid sequence recognition,”
Anal. Bioanal. Chem. , vol. 376,no. 7, pp. 1062–1068, Aug. 2003.[271] J. Mairhofer, K. Roppert, and P. Ertl, “Microfluidic systems forpathogen sensing: A review,”
Sensors , vol. 9, no. 6, pp. 4804–4823,Jun. 2009.[272] Y.-A. Chen, A. D. King, H.-C. Shih, C.-C. Peng, C.-Y. Wu, W.-H.Liao, and Y.-C. Tung, “Generation of oxygen gradients in microfluidicdevices for cell culture using spatially confined chemical reactions,”
Lab Chip , vol. 11, no. 21, p. 3626, Oct. 2011.[273] J. Reboud, G. Xu, A. Garrett, M. Adriko, Z. Yang, E. M. Tukahebwa,C. Rowell, and J. M. Cooper, “Paper-based microfluidics for DNAdiagnostics of malaria in low resource underserved rural communities,”
Proc. Natl. Acad. Sci. , vol. 116, no. 11, pp. 4834–4842, Mar. 2019.[274] R. P. Saha, S. Samanta, S. Patra, D. Sarkar, A. Saha, and M. K. Singh,“Metal homeostasis in bacteria: The role of ArsR-SmtB family of transcriptional repressors in combating varying metal concentrations inthe environment,”
BioMetals , vol. 30, no. 4, pp. 459–503, Aug. 2017.[275] N.-T. Nguyen, S. A. M. Shaegh, N. Kashaninejad, and D.-T. Phan,“Design, fabrication and characterization of drug delivery systemsbased on lab-on-a-chip technology,”
Adv. Drug Deliv. Rev. , vol. 65,no. 11-12, pp. 1403–1419, Nov. 2013.[276] C. Park, J. Kim, S. Jang, H.-G. Woo, Y. C. Ko, and H. Sohn, “Smartparticles for noble drug delivery system,”
J. Nanosci. Nanotechnol. ,vol. 10, no. 5, pp. 3375–3379, May 2010.[277] H. Gensler, R. Sheybani, P.-Y. Li, R. L. Mann, and E. Meng, “Animplantable MEMS micropump system for drug delivery in smallanimals,”
Biomed. Microdevices , vol. 14, no. 3, pp. 483–496, Jun. 2012.[278] A. S. Khalil and J. J. Collins, “Synthetic biology: Applications comeof age,”
Nat. Rev. Genet. , vol. 11, no. 5, pp. 367–379, May 2010.[279] K. B. Yesin, K. Vollmers, and B. J. Nelson, “Modeling and controlof untethered biomicrorobots in a fluidic environment using electro-magnetic fields,”
Int. J. Rob. Res. , vol. 25, no. 5-6, pp. 527–536, May2006.[280] G. Dogangil, O. Ergeneman, J. Abbott, S. Pane, H. Hall, S. Muntwyler,and B. Nelson, “Toward targeted retinal drug delivery with wirelessmagnetic microrobots,” in
Proc. IEEE/RSJ IROS , Sep. 2008, pp. 1921–1926.[281] S. Fusco, F. Ullrich, J. Pokki, G. Chatzipirpiridis, B. ¨Ozkale, K. M.Sivaraman, O. Ergeneman, S. Pan´e, and B. J. Nelson, “Microrobots:A new era in ocular drug delivery,”
Expert Opin. Drug Deliv. , vol. 11,no. 11, pp. 1815–1826, Nov. 2014.[282] F. Ullrich, C. Bergeles, J. Pokki, O. Ergeneman, S. Erni, G. Chatzipir-piridis, S. Pan´e, C. Framme, and B. J. Nelson, “Mobility experimentswith microrobots for minimally invasive intraocular surgery,”
Investig.Opthalmology Vis. Sci. , vol. 54, no. 4, p. 2853, Apr. 2013.[283] B. L. Bassler and M. B. Miller, “Quorum sensing,” in
The Prokaryotes ,M. Dworkin, S. Falkow, E. Rosenberg, K.-H. Schleifer, and E. Stacke-brandt, Eds. New York, NY: Springer, 2006, pp. 336–353.[284] E. G. Ruby, “Lessons from a cooperative, bacterial-animal association:The Vibrio fischeri-Euprymna scolopes light organ symbiosis,”
Annu.Rev. Microbiol. , vol. 50, no. 1, pp. 591–624, Oct. 1996.[285] B. K. Hammer and B. L. Bassler, “Quorum sensing controls biofilmformation in Vibrio cholerae,”
Mol. Microbiol. , vol. 50, no. 1, pp. 101–104, Aug. 2003.[286] T. R. de Kievit and B. H. Iglewski, “Bacterial quorum sensing inpathogenic relationships,”
Infect. Immun. , vol. 68, no. 9, pp. 4839–4849, Sep. 2000.[287] M. Kleerebezem, L. E. N. Quadri, O. P. Kuipers, and W. M. DeVos, “Quorum sensing by peptide pheromones and two-componentsignal-transduction systems in Gram-positive bacteria,”
Mol. Micro-biol. , vol. 24, no. 5, pp. 895–904, Jun. 1997.[288] K. Papenfort and B. L. Bassler, “Quorum sensing signal-responsesystems in Gram-negative bacteria,”
Nat. Rev. Microbiol. , vol. 14, no. 9,pp. 576–588, Sep. 2016.[289] S. Brameyer, H. B. Bode, and R. Heermann, “Languages and dialects:Bacterial communication beyond homoserine lactones,”
Trends Micro-biol. , vol. 23, no. 9, pp. 521–523, Sep. 2015.[290] S. W. Fuchs, K. A. J. Bozh¨uy¨uk, D. Kresovic, F. Grundmann, V. Dill,A. O. Brachmann, N. R. Waterfield, and H. B. Bode, “Formation of1,3-cyclohexanediones and resorcinols catalyzed by a widely occuringketosynthase,”
Angew. Chemie Int. Ed. , vol. 52, no. 15, pp. 4108–4112,Apr. 2013.[291] A. O. Brachmann, S. Brameyer, D. Kresovic, I. Hitkova, Y. Kopp,C. Manske, K. Schubert, H. B. Bode, and R. Heermann, “Pyronesas bacterial signaling molecules,”
Nat. Chem. Biol. , vol. 9, no. 9, pp.573–578, Sep. 2013.[292] S. A. Joyce, A. O. Brachmann, I. Glazer, L. Lango, G. Schw¨ar,D. J. Clarke, and H. B. Bode, “Bacterial biosynthesis of a multipotentstilbene,”
Angew. Chemie Int. Ed. , vol. 47, no. 10, pp. 1942–1945, Feb.2008.[293] P. Stoodley, D. DeBeer, and Z. Lewandowski, “Liquid flow in biofilmsystems,”
Appl. Environ. Microbiol. , vol. 60, no. 8, pp. 2711–2716,Aug. 1994.[294] P. S. Stewart, “Diffusion in biofilms,”
J. Bacteriol. , vol. 185, no. 5, pp.1485–1491, Mar. 2003.[295] J. Engebrecht and M. Silverman, “Identification of genes and geneproducts necessary for bacterial bioluminescence,”
Proc. Natl. Acad.Sci. , vol. 81, no. 13, pp. 4154–4158, Jul. 1984.[296] A. L. Schaefer, D. L. Val, B. L. Hanzelka, J. E. Cronan, and E. P.Greenberg, “Generation of cell-to-cell signals in quorum sensing: Acylhomoserine lactone synthase activity of a purified Vibrio fischeri LuxI protein.” Proc. Natl. Acad. Sci. , vol. 93, no. 18, pp. 9505–9509, Sep.1996.[297] W.-L. Ng and B. L. Bassler, “Bacterial quorum-sensing networkarchitectures,”
Annu. Rev. Genet. , vol. 43, no. 1, pp. 197–222, Dec.2009.[298] J. Engebrecht, K. Nealson, and M. Silverman, “Bacterial biolumines-cence: Isolation and genetic analysis of functions from Vibrio fischeri,”
Cell , vol. 32, no. 3, pp. 773–781, Mar. 1983.[299] A. M. Stevens, K. M. Dolan, and E. P. Greenberg, “Synergistic bindingof the Vibrio fischeri LuxR transcriptional activator domain and RNApolymerase to the lux promoter region.”
Proc. Natl. Acad. Sci. , vol. 91,no. 26, pp. 12 619–12 623, Dec. 1994.[300] E. A. Meighen, “Molecular biology of bacterial bioluminescence,”
Microbiol. Rev. , vol. 55, no. 1, pp. 123–142, Mar. 1991.[301] L. M. Mashburn-Warren and M. Whiteley, “Special delivery: Vesicletrafficking in prokaryotes,”
Mol. Microbiol. , vol. 61, no. 4, pp. 839–846, Aug. 2006.[302] K. Morinaga, T. Yamamoto, N. Nomura, and M. Toyofuku, “Paracoccusdenitrificans can utilize various long-chain N-acyl homoserine lactonesand sequester them in membrane vesicles,”
Environ. Microbiol. Rep. ,vol. 10, no. 6, pp. 651–654, Dec. 2018.[303] C. Schwechheimer and M. J. Kuehn, “Outer-membrane vesicles fromGram-negative bacteria: Biogenesis and functions,”
Nat. Rev. Micro-biol. , vol. 13, no. 10, pp. 605–619, Oct. 2015.[304] L. Brown, J. M. Wolf, R. Prados-Rosales, and A. Casadevall, “Throughthe wall: Extracellular vesicles in Gram-positive bacteria, mycobacteriaand fungi,”
Nat. Rev. Microbiol. , vol. 13, no. 10, pp. 620–630, Oct.2015.[305] M. Toyofuku, “Bacterial communication through membrane vesicles,”
Biosci. Biotechnol. Biochem. , vol. 83, no. 9, pp. 1599–1605, Sep. 2019.[306] C. M. Waters and B. L. Bassler, “Quorum sensing: Cell-to-cell com-munication in bacteria,”
Annu. Rev. Cell Dev. Biol. , vol. 21, no. 1, pp.319–346, Nov. 2005.[307] M. Schuster, C. P. Lostroh, T. Ogi, and E. P. Greenberg, “Identification,timing, and signal specificity of Pseudomonas aeruginosa quorum-controlled genes: A transcriptome analysis,”
J. Bacteriol. , vol. 185,no. 7, pp. 2066–2079, Apr. 2003.[308] G. C´arcamo-Oyarce, P. Lumjiaktase, R. K¨ummerli, and L. Eberl,“Quorum sensing triggers the stochastic escape of individual cells fromPseudomonas putida biofilms,”
Nat. Commun. , vol. 6, no. 1, p. 5945,May 2015.[309] Y. C. Yong and J. J. Zhong, “A genetically engineered whole-cellpigment-based bacterial biosensing system for quantification of N-butyryl homoserine lactone quorum sensing signal,”
Biosens. Bioelec-tron. , vol. 25, no. 1, pp. 41–47, Sep. 2009.[310] V. Thiel, R. Vilchez, H. Sztajer, I. Wagner-D¨obler, and S. Schulz,“Identification, quantification, and determination of the absolute con-figuration of the bacterial quorum-sensing signal autoinducer-2 by gaschromatography-mass spectrometry,”
ChemBioChem , vol. 10, no. 3, pp.479–485, Feb. 2009.[311] C. Zhang and B.-C. Ye, “Real-time measurement of quorum-sensingsignal autoinducer 3OC6HSL by a FRET-based nanosensor,”
Biopro-cess Biosyst. Eng. , vol. 37, no. 5, pp. 849–855, May 2014.[312] R. P. Novick, S. J. Projan, J. Kornblum, H. F. Ross, G. Ji, B. Kreiswirth,F. Vandenesch, S. Moghazeh, and R. P. Novick, “The agr P2 operon: Anautocatalytic sensory transduction system in Staphylococcus aureus,”
Mol. Gen. Genet. MGG , vol. 248, no. 4, pp. 446–458, Aug. 1995.[313] P. C. Seed, L. Passador, and B. H. Iglewski, “Activation of thePseudomonas aeruginosa lasI gene by LasR and the Pseudomonasautoinducer PAI: An autoinduction regulatory hierarchy,”
J. Bacteriol. ,vol. 177, no. 3, pp. 654–659, Feb. 1995.[314] S. L. Svenningsen, K. C. Tu, and B. L. Bassler, “Gene dosage com-pensation calibrates four regulatory RNAs to control Vibrio choleraequorum sensing,”
EMBO J. , vol. 28, no. 4, pp. 429–439, Feb. 2009.[315] L. Feng, S. T. Rutherford, K. Papenfort, J. D. Bagert, J. C. van Kessel,D. A. Tirrell, N. S. Wingreen, and B. L. Bassler, “A Qrr noncod-ing RNA deploys four different regulatory mechanisms to optimizequorum-sensing dynamics,”
Cell , vol. 160, no. 1-2, pp. 228–240, Jan.2015.[316] N. R. Naylor, R. Atun, N. Zhu, K. Kulasabanathan, S. Silva, A. Chat-terjee, G. M. Knight, and J. V. Robotham, “Estimating the burden ofantimicrobial resistance: A systematic literature review,”
Antimicrob.Resist. Infect. Control , vol. 7, no. 58, pp. 1–17, Dec. 2018.[317] M. P. Schultz, J. A. Bendick, E. R. Holm, and W. M. Hertel, “Economicimpact of biofouling on a naval surface ship,”
Biofouling , vol. 27, no. 1,pp. 87–98, Jan. 2011. [318] D. G. Lloyd and D. E. Whitworth, “The myxobacterium Myxococcusxanthus can sense and respond to the quorum signals secreted bypotential prey organisms,”
Front. Microbiol. , vol. 8, no. 1, pp. 138–144, Mar. 2017.[319] H.-C. Flemming and S. Wuertz, “Bacteria and archaea on earth andtheir abundance in biofilms,”
Nat. Rev. Microbiol. , vol. 17, no. 4, pp.247–260, Apr. 2019.[320] R. M. Donlan and W. Costerton, “Biofilms: Survival mechanisms ofclinically relevant microorganisms,”
Clin. Microbiol. Rev. , vol. 15,no. 2, pp. 167–193, Apr. 2002.[321] T. R. Pintelon, C. Picioreanu, M. C. van Loosdrecht, and M. L. Johns,“The effect of biofilm permeability on bio-clogging of porous media,”
Biotechnol. Bioeng. , vol. 109, no. 4, pp. 1031–1042, Apr. 2012.[322] P. S. Stewart and M. J. Franklin, “Physiological heterogeneity inbiofilms,”
Nat. Rev. Microbiol. , vol. 6, no. 3, pp. 199–210, Mar. 2008.[323] P. D. P´erez and S. J. Hagen, “Heterogeneous response to a quorum-sensing signal in the luminescence of individual Vibrio fischeri,”
PLoSOne , vol. 5, no. 11, p. e15473, Nov. 2010.[324] D. Bronesky, Z. Wu, S. Marzi, P. Walter, T. Geissmann, K. Moreau,F. Vandenesch, I. Caldelari, and P. Romby, “Staphylococcus au-reus RNAIII and its regulon link quorum sensing, stress responses,metabolic adaptation, and regulation of virulence gene expression,”
Annu. Rev. Microbiol. , vol. 70, no. 1, pp. 299–316, Sep. 2016.[325] A. I. Psarras and I. G. Karafyllidis, “Simulation of the dynamics ofbacterial quorum sensing,”
IEEE Trans. Nanobiosci. , vol. 14, no. 4,pp. 440–446, Jun. 2015.[326] A. E. Pereda, “Electrical synapses and their functional interactions withchemical synapses,”
Nat. Rev. Neurosci. , vol. 15, no. 4, pp. 250–263,Apr. 2014.[327] A. L. Hodgkin and A. F. Huxley, “A quantitative description ofmembrane current and its application to conduction and excitation innerve,”
J. Physiol. , vol. 117, no. 4, pp. 500–544, Aug. 1952.[328] P. Nelson,
Biological Physics: Energy, Information, Life . San Fran-cisco, CA, USA: W. H. Freeman and Company, 2008.[329] S. Lotter, A. Ahmadzadeh, and R. Schober, “Synaptic channelmodeling for DMC: Neurotransmitter uptake and spillover inthe tripartite synapse,” pp. 1–36, May 2020. [Online]. Available:https://arxiv.org/abs/2005.09108[330] N. A. Abbasi, D. Lafci, and O. B. Akan, “Controlled informationtransfer through an in vivo nervous system,”
Sci. Rep. , vol. 8, no. 1,p. 2298, Dec. 2018.[331] K. Deisseroth, “Optogenetics,”
Nat. Methods , vol. 8, no. 1, pp. 26–29,Jan. 2011.[332] C. Houghton and J. D. Victor, “Measuring representational distances:The spike-train metrics approach,” in
Vis. Popul. Codes . Cambridge,MA, USA: MIT Press, 2011, ch. 8, pp. 213–244.[333] H. Ramezani and O. B. Akan, “A communication theoretical modelingof axonal propagation in hippocampal pyramidal neurons,”
IEEE Trans.Nanobiosci. , vol. 16, no. 4, pp. 248–256, Jun. 2017.[334] K. Aghababaiyan, V. Shah-Mansouri, and B. Maham, “Axonal channelcapacity in neuro-spike communication,”
IEEE Trans. Nanobiosci. ,vol. 17, no. 1, pp. 78–87, Mar. 2018.[335] M. Veletic, P. A. Floor, Y. Chahibi, and I. Balasingham, “On the upperbound of the information capacity in neuronal synapses,”
IEEE Trans.Commun. , vol. 64, no. 12, pp. 5025–5036, Dec. 2016.[336] K. Aghababaiyan, V. Shah-Mansouri, and B. Maham, “Joint optimiza-tion of input spike rate and receiver decision threshold to maximizeachievable bit rate of neuro-spike communication channel,”
IEEETrans. Nanobiosci. , vol. 18, no. 2, pp. 117–127, Apr. 2019.[337] M. Veletic and I. Balasingham, “Synaptic communication engineeringfor future cognitive brain-machine interfaces,”
Proc. IEEE , vol. 107,no. 7, pp. 1425–1441, Jul. 2019.[338] B. A. Bilgin, E. Dinc, and O. B. Akan, “DNA-based molecularcommunications,”
IEEE Access , vol. 6, pp. 73 119–73 129, Nov. 2018.[339] J. W. Little, S. B. Zimmerman, C. K. Oshinsky, and M. Gellert, “Enzy-matic joining of DNA strands, II. An enzyme-adenylate intermediatein the dpn-dependent DNA ligase reaction.”
Proc. Natl. Acad. Sci. ,vol. 58, no. 5, pp. 2004–2011, Nov. 1967.[340] H. O. Smith and K. Welcox, “A restriction enzyme from Hemophilusinfluenzae,”
J. Mol. Biol. , vol. 51, no. 2, pp. 379–391, Jul. 1970.[341] K. B. Mullis and F. A. Faloona, “Specific synthesis of DNA in vitrovia a polymerase-catalyzed chain reaction,” in
Recomb. DNA Methodol.
Amsterdam: Elsevier, 1989, pp. 189–204.[342] A. O’Driscoll and R. D. Sleator, “Synthetic DNA,”
Bioengineered ,vol. 4, no. 3, pp. 123–125, May 2013.[343] P. W. K. Rothemund, “Folding DNA to create nanoscale shapes andpatterns,”
Nature , vol. 440, no. 7082, pp. 297–302, Mar. 2006. [344] S. M. Douglas, H. Dietz, T. Liedl, B. H¨ogberg, F. Graf, and W. M. Shih,“Self-assembly of DNA into nanoscale three-dimensional shapes,” Nature , vol. 459, no. 7245, pp. 414–418, May 2009.[345] M. Endo and H. Sugiyama, “DNA origami nanomachines,”
Molecules ,vol. 23, no. 7, p. 1766, Jul. 2018.[346] X. Liu, C.-H. Lu, and I. Willner, “Switchable reconfiguration of nucleicacid nanostructures by stimuli-responsive DNA machines,”
Acc. Chem.Res. , vol. 47, no. 6, pp. 1673–1680, Jun. 2014.[347] S. Surana, J. M. Bhat, S. P. Koushika, and Y. Krishnan, “An au-tonomous DNA nanomachine maps spatiotemporal pH changes in amulticellular living organism,”
Nat. Commun. , vol. 2, no. 1, p. 340,Sep. 2011.[348] S. Ranallo, C. Pr´evost-Tremblay, A. Idili, A. Vall´ee-B´elisle, andF. Ricci, “Antibody-powered nucleic acid release using a DNA-basednanomachine,”
Nat. Commun. , vol. 8, no. 1, p. 15150, Aug. 2017.[349] A. C. Wilkinson, H. Nakauchi, and B. G¨ottgens, “Mammalian tran-scription factor networks: Recent advances in interrogating biologicalcomplexity,”
Cell Syst. , vol. 5, no. 4, pp. 319–331, Oct. 2017.[350] S. Green, M. erban, R. Scholl, N. Jones, I. Brigandt, and W. Bechtel,“Network analyses in systems biology: new strategies for dealing withbiological complexity,”
Synthese , vol. 195, no. 4, pp. 1751–1777, Apr.2018.[351] S. Jeon, S. Kim, S. Ha, S. Lee, E. Kim, S. Y. Kim, S. H. Park, J. H.Jeon, S. W. Kim, C. Moon, B. J. Nelson, J. young Kim, S. W. Yu, andH. Choi, “Magnetically actuated microrobots as a platform for stemcell transplantation,”
Sci. Robot. , vol. 4, no. 30, May 2019.[352] I. Akyildiz, M. Pierobon, S. Balasubramaniam, and Y. Koucheryavy,“The Internet of Bio-Nano Things,”
IEEE Commun. Mag. , vol. 53,no. 3, pp. 32–40, Mar. 2015.[353] K. Han, M. S. Levine, and J. L. Manley, “Synergistic activation andrepression of transcription by drosophila homeobox proteins,”
Cell ,vol. 56, no. 4, pp. 573–583, Feb. 1989.[354] L.-W. Chen, M.-W. Lin, and C.-M. Hsu, “Different pathways leading toactivation of extracellular signal-regulated kinase and p38 MAP kinaseby formyl-methionyl-leucyl-phenylalanine or platelet activating factorin human neutrophils,”
J. Biomed. Sci. , vol. 12, no. 2, pp. 311–319,Mar. 2005.[355] M. Natarajan, K. M. Lin, R. C. Hsueh, P. C. Sternweis, and R. Ran-ganathan, “A global analysis of cross-talk in a mammalian cellularsignalling network,”
Nat. Cell Biol. , vol. 8, no. 6, pp. 571–580, Jun.2006.[356] T. E. Gorochowski, “Agent-based modelling in synthetic biology,”
Essays Biochem. , vol. 60, no. 4, pp. 325–336, Nov. 2016.[357] T. I. Yusufaly and J. Q. Boedicker, “Spatial dispersal of bacterialcolonies induces a dynamical transition from local to global quorumsensing,”
Phys. Rev. E , vol. 94, no. 6, p. 062410, Dec. 2016.[358] H. Ueda, K. Stephens, K. Trivisa, and W. E. Bentley, “Bacteria Floc,but Do They Flock? Insights from Population Interaction Models ofQuorum Sensing,” mBio , vol. 10, no. 3, pp. 1–15, May 2019.[359] M. F. Bear, B. W. Connors, and M. A. Paradiso,