A novel in vitro device to deliver induced electromagnetic fields to cell and tissue cultures
Rea Ravin, Teddy X. Cai, Randall H. Pursley, Marcial Garmendia-Cedillos, Tom Pohida, Raisa Z. Freidlin, Herui Wang, Zhengping Zhuang, Amber J. Giles, Nathan H. Williamson, Mark R. Gilbert, Peter J. Basser
aa r X i v : . [ phy s i c s . b i o - ph ] N ov Manuscript submitted to
Biophysical
Journal
Article
A novel in vitro device to deliver inducedelectromagnetic fields to cell and tissue cultures
Rea Ravin , Teddy X. Cai , Randall H. Pursley , Marcial Garmendia-Cedillos , Tom Pohida , Raisa Z. Freidlin , HeruiWang , Zhengping Zhuang , Amber J. Giles , Nathan H. Williamson , Mark R. Gilbert , and Peter J. Basser † * Indicates equal contribution Celoptics, Inc., Rockville, MD, USA Section on Quantitative Imaging and Tissue Sciences
Eunice Kennedy Shriver
National Institutes of Child Health and HumanDevelopment, National Institutes of Health, Bethesda, MD, USA The Signal Processing and Instrumentation Section, Center for Information Technology, National Institutes of Health,Bethesda, MD, USA Neuro-Oncology Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD,USA National Institute of General Medical Sciences, National Institutes of Health, Bethesda, MD, USA † Correspondence: [email protected] We have developed a novel in vitro instrument that can deliver intermediate frequency (100 – 400 kHz), mod-erate intensity (up to and exceeding 6.5 V/cm pk-pk) electric fields (EFs) to cell and tissue cultures generated using inducedelectromagnetic fields (EMFs) in an air-core solenoid coil. A major application of these EFs is as an emerging cancer treatmentmodality.
In vitro studies by Novocure Ltd. reported that intermediate frequency (100 – 300 kHz), low amplitude (1 – 3 V/cm)EFs, which they called “Tumor Treating Fields (TTFields)”, had an anti-mitotic effect on glioblastoma multiforme (GBM) cells.The effect was found to increase with increasing EF amplitude. Despite continued theoretical, preclinical, and clinical study, themechanism of action remains incompletely understood. All previous in vitro studies of “TTFields” have used attached, capaci-tively coupled electrodes to deliver alternating EFs to cell and tissue cultures. This contacting delivery method suffers from apoorly characterized EF profile and conductive heating that limits the duration and amplitude of the applied EFs. In contrast,our device delivers EFs with a well-characterized radial profile in a non-contacting manner, eliminating conductive heating andenabling thermally regulated EF delivery. To test and demonstrate our system, we generated continuous, 200 kHz EMF with anEF amplitude profile spanning 0 – 6.5 V/cm pk-pk and applied them to exemplar human thyroid cell cultures for 72 hours. Weobserved moderate reduction in cell density ( < < in vitro and in the clinic, althoughthe mechanism of action remains unclear. Existing EF delivery systems use capacitively coupled electrodes placed in directcontact with the specimen holder. This direct contact generates unwanted heat and limits the amplitude and duration ofEF stimulation. We have developed a novel in vitro system capable of continuous thermal regulation and delivery of well-characterized (200 kHz, 0 – 6.5 V/cm pk-pk), electromagnetic fields (EMFs) to cell cultures, paving the way for improvedmechanistic biophysical studies. INTRODUCTION
There is an extensive body of literature describing the effects of low (DC-kHz) and high (above 1 MHz) frequency electric fields(EFs) on living cells (1). Very low frequency EFs ( < Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X.
However, Kirson, et al. (2) reported that EFs at frequencies of 100 −
300 kHz and amplitudes between 1 − ≥ > invitro test system (Inovitro ™ ) designed to deliver EFs to cell cultures using two pairs of capacitively coupled electrodes insulatedby a high dielectric constant ceramic (3), placed orthogonal to each other inside a cell culture dish (3, 27). The electrodepairs are connected to a high voltage sinusoidal waveform generator. The stimulated electrode pair is switched periodically tomaximize the orientational efficiency of EF application (3). While contact between the culture dish and electrodes creates adirect means for EFs to be delivered to cell or tissue cultures, it also results in heat conduction from the electrodes to the dishand subsequently from the dish to the culture media, making continuous temperature control problematic. Joule or Ohmicheating from the EFs themselves also significantly contributes to heating of the conductive cell media.In order to maintain a temperature of 37 ◦ C in the culture dish, EF application with the Inovitro ™ device is performed inan 18 ◦ C incubator (28). Electric current limits and on-off operation are also imposed as necessary to maintain temperature(3, 28). A consequence of keeping the apparatus in this refrigerated environment is a lower surrounding water vapor pressure,which accelerates cell media evaporation. Assuming 95% relative humidity in the incubator, a loss of about 0.2 mL of mediaper day can be expected from a 35 mm diameter dish (see Supp. Sec. 1). As a result, media replacement is required every 24hours according to procedures for operating the Inovitro ™ device (2, 28). This rapid media loss and replacement may haveinadvertent confounding effects, such as causing time-varying hyperosmotic bath conditions. Hyperosmotic conditions canresult in nuclear lamina buckling (29), cell cycle disruption (30), or protein aggregation (31). Another issue is that the geometry Manuscript submitted to Biophysical Journalnduced EMF delivery to cells of the EF delivery device does not lend itself to easily predicting the electric field or current density produced within the tissueor cell culture (12, 14).To overcome some of these limitations associated with delivering EFs via contacting electrodes, here we describe a noveldevice that uses electromagnetic induction to deliver EFs/electromagnetic fields (EMFs) in vitro , like those generated byTranscranial Magnetic Simulation (TMS) devices (32) but in a higher frequency range. More specifically, we propose the useof an air-core double solenoid coil in an LC resonant circuit to induce intermediate frequency EMF in cell or tissue culturesplaced within the coil. This non-contacting method of EMF delivery thermally isolates the current-carrying coil from theculture dishes. Temperature uniformity and osmolarity control of the culture medium is improved. Greater power delivery ismade possible, providing potential access to high, previously inaccessible EF amplitudes. Furthermore, the radial, linearlyvarying amplitude of the induced circumferential EF within the coil allows each culture dish to be exposed to a known andwell-characterized range of EF amplitudes or “doses”. Induced EMFs represent a promising alternative EF delivery methodfor in vitro experimentation and may better elucidate the biophysical mechanisms of action of “TTFields” than the currentstate-of-the-art.
MATERIALS AND METHODS200 kHz induced EMF generation
Figure 1: Experimental apparatus inside incubator. ( a ) Rendering of dish sleeve containing stack of seven 35 mm cell culturedishes. Vertical positions labelled P1 - P5. A fiber optic temperature sensor is placed into the P3 dish. ( b ) Inside view ofmodified incubator and experimental apparatus. (1) 22 cm ×
22 cm Plexiglas window. (2) Fiber optic sensor. (3) Double-wrapped copper coil, 3 turns, ID = = c ) Relative position of control stack, off-axis from the treated sleeve to avoid the 1 / 𝑟 EMF contribution outside the coil – seeEq. (3).EMF generation was accomplished using an industrial 10 kW induction heater system (DP-10-400, RDO Induction LLC,Washington NJ) based on a conventional LC resonant circuit design (with a resonant angular frequency of 250 kHz). Theinduction device was connected to a copper coil built to the following specifications: inner diameter (ID) of 5 cm, height of 8cm, 3 turns, double-wrapped, with vertical orientation (Fig. 1b, 3). The induction system and coil were jacketed and connectedto a water chiller (DuraChill © DCA200, PolyScience, Niles, IL) to remove heat and control coil temperature. In order to placethe coil inside the 95% air, 5% CO incubator at 37 ◦ C (MCO-18M Multigas Incubator, SANYO Electric Co. Ltd., Osaka,Japan), a 22 cm x 22 cm hole (Fig. 1b, 1) was made in the incubator to avoid Joule heating of the metal lining via inductioncurrents passing through the copper pipes connecting the coil to the induction device. The hole was covered and sealed withPlexiglas connected to the incubator. A Plexiglas shelf was placed below the coil (Fig. 1b, 5), replacing the metal shelf of theincubator, again to avoid inductive heating. Plastic sleeves for cell culture dishes were made in-house on a 3D printer (Fig.1a). Each sleeve accommodates seven 35 mm cell culture dishes in a vertical stack, so several culture dishes can be studied inparallel. The middle 5 dishes are situated fully within the coil and their positions are labelled P1 – P5. The dish in the thirdposition from the bottom is referred to as P3, and is shown with a fiber optic temperature sensor placed in its center (Fig.1a). An infrared detector records the coil temperature (Fig. 1b, 4). EMFs were delivered continuously for 72 hours (h) in allexperiments.The expected EF profile generated within the coil by the induction device can be calculated from basic principles ofelectricity and magnetism. The magnetic field within an air-core current-carrying coil is readily approximated by Ampere’s
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X. law, 𝐵 𝑧 ( 𝑡 ) = 𝜇 𝐼 ( 𝑡 ) (cid:18) 𝑁𝐿 (cid:19) (1)where 𝐵 𝑧 ( 𝑡 ) is the axial component of the applied magnetic field, B , 𝜇 is the permeability of a vacuum, 𝐼 is the currentamplitude, 𝑁 is the number of coil turns, and 𝐿 is the length of the coil. The direction of the magnetic field is normal to thedirection of the current and can be determined by the right hand rule as being along the coil axis. Faraday’s law in integralform relates the magnetic field to the electromotive force, 𝜀 , 𝜀 = ∮ 𝑙 E · 𝑑 l = − 𝑑𝑑𝑡 ∬ 𝑆 B · 𝑑 S , (2)where E is the induced EF, S denotes some bounded surface, and 𝑑 l is an infinitesimal arc length. If the surface is chosen to bea circular cross section of the coil with radius 𝑟 from the center of the coil then the following solutions are obtained for | E ( 𝑡 )| , | E ( 𝑟 , 𝑡 )| = | 𝐸 𝜃 ( 𝑟 , 𝑡 )| = (cid:16) 𝑟 (cid:17) 𝑑𝐵 𝑧 ( 𝑡 ) 𝑑𝑡 = 𝜇 (cid:18) 𝑁𝐿 (cid:19) (cid:16) 𝑟 (cid:17) (cid:12)(cid:12)(cid:12)(cid:12) 𝑑𝐼𝑑𝑡 (cid:12)(cid:12)(cid:12)(cid:12) 𝑟 < 𝑅 , (cid:18) 𝑅 𝑟 (cid:19) 𝑑𝐵 𝑧 ( 𝑡 ) 𝑑𝑡 = 𝜇 (cid:18) 𝑁𝐿 (cid:19) (cid:18) 𝑅 𝑟 (cid:19) (cid:12)(cid:12)(cid:12)(cid:12) 𝑑𝐼𝑑𝑡 (cid:12)(cid:12)(cid:12)(cid:12) 𝑟 ≥ 𝑅 , (3)where 𝑅 is the radius of the coil, and 𝐸 𝜃 ( 𝑟 , 𝑡 ) is the azimuthal component of the electric field. By symmetry, the other twocomponents of the electric field vanish. The direction of the EF is anti-parallel to the current flowing in the coils. If the appliedcurrent is sinusoidal, e.g., 𝐼 ( 𝑡 ) = 𝐼 cos ( 𝜔𝑡 ) , then the induced EF will oscillate at the same frequency in the circumferentialdirection. By Eq. (3), our induction system delivering current at 200 kHz results in a circumferential 200 kHz EF profile withinthe coil with a linearly increasing amplitude from the center of the coil. Importantly, the radial profile of the generated EFsmakes each treated dish its own “dose titration” experiment, with little to no EFs dose or amplitude in the center of the dishand high dose at the periphery. The effects of EF amplitude can be isolated from any dish-to-dish experimental confounds byanalyzing radially dependent differences within a single treated dish.Note, the electric field outside of the coil is expected to decay by 1 / 𝑟 according to Eq. (3). The EF in the incubator outsidethe coil is therefore not 0. Because of this decaying EF, the control dish sleeve is placed in the incubator at a distance andvertical height offset (off axis from the coil) such that little to no EF should be present, as shown in Fig. 1c. Temperature monitoring and control
Joule heating remains a major hurdle to delivering increased EF amplitudes while maintaining controlled temperatures,even in the absence of conductive heating from contacting electrodes. Heating of an electrically conductive material by aspatially-varying EF is described by the power delivered per unit volume 𝑑𝑃𝑑𝑉 = J · E = 𝜎 | E | by J = 𝜎 E ( microscopic Ohm ′ s Law ) , (4)where 𝑃 is power, 𝑉 is volume, J is current density, E is the EF, and 𝜎 is the conductivity of the material. In the case of asolenoid coil (cylindrical geometry) surrounding some uniform conductive material, the rate of heat generation, ¤ 𝑞 , deliveredby the current-carrying coil to the material of height, 𝐿 𝑧 , can be calculated from the within-coil part of Eq. (3) and Eq. (4) as ¤ 𝑞 ≡ 𝑃 total = 𝜎 ∭ 𝑉 | E ( 𝑟 )| 𝑑𝑉 = 𝜎 ∫ 𝐿 𝑧 𝑑𝑧 ∫ 𝜋 𝑑𝜃 ∫ 𝑅 𝑟 | E ( 𝑟 )| 𝑑𝑟 = 𝜋𝜎𝐿 𝑧 𝑚 𝑅
16 , (5)where ∭ 𝑉 𝑑𝑉 is a volume integral and 𝑚 is the radial slope of the within-coil part of Eq. (3) using peak-to-peak amplitudessuch that | E ( 𝑟 )| rms = 𝑚𝑟 /( √ ) , 𝑚 : = 𝜇 (cid:18) 𝑁𝐿 (cid:19) (cid:12)(cid:12)(cid:12)(cid:12) 𝑑𝐼𝑑𝑡 (cid:12)(cid:12)(cid:12)(cid:12) pk − pk [ = ] V · m − , (6)where rms is the root-mean-square. According to Eqs. (4), (5), and (6), Joule heat generation increases with the square ofthe delivered EF amplitude. Thus, delivering higher EF amplitudes while maintaining well-controlled temperatures requirescareful regulation of environmental conditions. Not only does the current-carrying solenoid coil require cooling, the cell Manuscript submitted to Biophysical Journalnduced EMF delivery to cells cultures themselves require negative heat flux to the surroundings to balance Joule heating. This negative heat flux is attainedby cooling the coil below the incubator temperature. A complex interplay of incubator temperature, a cooled coil temperature,and various thermal resistances (Fig. S2, Table S2) determines the steady-state temperature profile within the cell culturedishes. See Supp. Sec. 2 for a more complete discussion of heat transfer and expected temperature profiles (Figs. S3 – S5).A data acquisition and analysis system was used to monitor the environmental conditions within the incubator duringexperiments. The system can acquire data from as many as four temperature probes and two electric field sensors. Thedata acquisition software was developed using LabVIEW (National Instruments, Austin, TX). During operation, the systemcollected temperature data from Opsens multi-channel signal conditioners (TempSens TMS-G2-1–100ST-M1, Opsens, Quebec,Canada). Data from sensors was measured every second and transmitted to the system computer via an RS-232 interface. Aninfrared detector was used to monitor the coil temperature. Fiber optic sensors (OTG-A-10-62ST-1, Opsens, Quebec, Canada),rather than traditional thermistors, were used to measure temperature within the cell media to avoid both inductive heatingeffects and any parasitic inductance and capacitance potentially contributed by thermistors. Fiber optic sensors were attachedto dishes (e.g., Fig. 1b, 2) using bone wax (Lukens ™ Osmolarity regulation
Cooler air near the dishes can result in accelerated loss of media due to water vapor pressure gradients (see Supp. Sec. 1). Asa result, osmolarity can change over the course of experiments. We opted to take two preparatory steps to mitigate osmolaritychanges, as opposed to replacing cell media during the course of the experiment. First, we used 3 mL of media rather thanthe conventional 2 mL; increasing the total volume decreases the proportion of media lost. Second, we shaped the lids of alldishes by heating the lids with a heat gun, placing them on a sharpened ring mold, and depressing them while malleable witha steel ball. The end result was a curved or domed lid that is partially submerged in the media. This lid shape mitigated medialoss primarily by decreasing the exposed liquid surface area and thus the area undergoing mass transport with the surroundingair (Fig. S1). The lid also directs condensate back into the media volume. These changes reduced the predicted osmolaritychange by over 60% (Table S1). Note, increased media volume and decreased exposed surface area should not affect dissolvedgas (O , CO , etc.) concentrations in media. According to Henry’s law, a dissolved gas in liquid is only a function of thepartial pressure of the gas above the liquid and temperature, which are both unchanged by these modifications. Osmolaritymeasurements with a Vapro 5520 vapor pressure osmometer (Wescor, Logan, UT) were used to experimentally validate theefficacy of these regulation measures. Human thyroid cell culture
We used a human thyroid cell line (Nthy-ori 3-1 Sigma, Sigma-Aldrich, St. Louis, MO) as an exemplar test cell culture system.Cells were infected with lentivirus green fluorescent protein (GFP) and selected for high expression of GFP over several cycles.Cells were maintained in 75ml flasks in DMEM media (Gibco, Thermo Fisher Scientific, Waltham, MA) supplemented with10% FBS (Gibco), 1% Penicillin Streptomycin (Gibco). GFP-expressing cells were passaged twice a week and cell cultureswere maintained in a 95% air, 5% CO incubator at 37 ◦ C. Before each 72 h experiment, cells were detached from the culturevessel with Accutase © (Sigma-Aldrich) and re-suspended in media at a density of 3.33 × cells/ml. 3 ml of cell suspensionwas placed in each 35 mm culture dish (Corning, Falcon) for a plating density of 1 × cells per dish. Dishes were previouslycoated with Poly-D-Lysine to enhance attachment. Dishes were incubated for 24 h to allow for attachment prior to 72 hexperiments. Markings were made on the external part of the dishes – consistent relative to manufacturer markings – to keepthe orientation of dishes consistent during treatment and imaging.A series of cell counting calibration experiments using different initial seeding densities ranging from 3 × to 1.2 × cells per dish were also performed. Cells were again detached with Accutase © , plated at different densities, incubated for 24h, and then incubated for an additional 72 h. At the end of 72 h, the cell culture dishes were scanned using confocal imaging,described below, and all cells were subsequently lifted and counted using a Countess ™ II FL Automated Cell Counter (Thermo
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X.
Fisher). Final cell densities ranged from ∼
100 to 1200 cells/sq. mm. 3 series of calibration experiments were performed for atotal of 56 cell counts and images. Results from these calibration experiments were used to relate imaging observables to theexperimental cell counts.
In situ confocal imaging and ethidium homodimer-1 and propidium iodide staining
A unique challenge of the study design is the presence of a radial dependence in the experimental conditions. This radialdependence precluded more obvious approaches like lifting and counting all cells and informed our choice to use GFP-expressing cells. Entire dishes were imaged in situ to retain radial information, i.e., what EF amplitude treated cells experiencedduring the 72 h EMF application. Stitched images were acquired and analyzed.At the end of experiments, sleeves with dishes were removed from the coil and dishes were gently removed from thesleeves so as not detach cells. Orientation information was retained using dish markings. All dishes were incubated untilimaged. The dishes were mounted to the stage of a Zeiss LSM 780 laser scanning confocal microscope system. For imaging,an EC Plan-NeoFluor 5x, 0.16 N.A. was used. All the visible parts of the 35 mm culture dish were imaged using stitchingmode, stitching 484 areas of interest with 5% overlap. The final image size was 10, 726 ×
10, 726 pixels (px). For GFP, a 488nm excitation wavelength, a beam splitter of 488/562 nm, and emission filters of 500 - 555 nm were used. Special care wastaken to use non-saturating parameters. Imaging parameters were kept constant across dishes and experiments.In 4 experiments, additional ethidium homodimer-1 (E1169, Thermo Fisher) (EthD-1) and propidium iodide (V13241,Thermo Fisher) (PI) staining was performed following GFP imaging. For PI, at the end of EF application and 5 minutes priorto microscope imaging, 750 𝜇 L of DMEM was removed from the dishes and mixed with 0.5 𝜇 L of 1 mg/mL stock solutionand added back to the dish to a final concentration of 2 × − mg/mL. This procedure was used to avoid agitating all mediain the dish and losing spatial information on floating, or incompletely adhered cells. For EthD-1, an identical procedure wasperformed but for a final concentration of 0.4 𝜇 M. No additional washes were performed to avoid loss of the floating cellpopulation. For both PI and EthD-1, a 561 nm excitation wavelength and emission filters of 578 -739 nm were used. Stitchingwas identical.
Image processing
Two populations were visible in images of the GFP-expressing thyroid cells: a brighter, high fluorescence population, whichmight correspond to cellular debris, mitotic cells, pre-apoptotic cells, or a rounded cell morphology; and a dim, lowerfluorescence population that corresponds to attached cells. The primary population of interest is the attached cell population.Therefore, one goal of image processing is to segment these bright and dim px populations for downstream analyses. Thissegmentation was performed in 3 steps: contrast enhancement and denoising (e.g., Fig. 2a and b, Fig. S7), 3-level intensitythresholding (Figs. 2c, Fig. S8), and classification error correction (Fig. 2d, Fig. S9), outlined below. EthD-1 and PI stainingimages underwent a simpler procedure of contrast enhancement and denoising (Figs. 2e and f) followed by 2-level intensitythresholding (Fig. 2g) and size thresholding (Fig. 2h) to identify stained dead cells. See Supp. Sec. 3 for more details andMATLAB code. After processing, images were binned radially (Fig. 2i) to encode radial information in the analysis. Edgeregions and dish regions occluded by fitments were excluded. The image processing pipeline is summarized in Fig. 2.
Contrast enhancement and denoising
The contrast enhancement step (Figs. 2b and f, Fig. S7, Supp. Sec. 3.1) was based on a morphological transform method(33, 34) (Supp. Sec. 3). An image, 𝐴 , can be contrast enhanced by computing 𝐴 Enhanced = 𝐴 + [ 𝐴 − ( 𝐴 ◦ 𝐵 )] − [( 𝐴 • 𝐵 ) − 𝐴 ] (7)where 𝐵 is a structuring element, ◦ denotes an opening operation, and • denotes a closing operation. Opening is erosionfollowed by dilation while closing is dilation followed by erosion. The middle term on the right-hand side of Eq. (7) is a‘top hat’ of the original image, weighted to brighter regions, and the rightmost term is a ‘bottom hat’, weighted to darkerregions. Morphological contrast enhancement can therefore be conceptualized as ‘adding’ to bright regions, and ‘subtracting’from dark regions (33). All images underwent morphological contrast enhancement using a ‘disk’ shaped structuring elementwith a 3 px radius. Operations were performed using MATLAB 2019a (MathWorks, Natick, MA) and the imtophat() andimbottomhat() functions found in the Image Processing Toolbox. After contrast enhancement, a denoising step was performedwith a 3 × Manuscript submitted to Biophysical Journalnduced EMF delivery to cells
Figure 2: Overview of image processing pipeline. ( a - d ) Processing of GFP images. (a) Raw image. (b) Contrast enhanced image.(c) Segmented image using 3-level intensity thresholding prior to corrections. (cyan arrow) Example of “spot” classificationerror. (green arrow) Example of “halo” classification error. (d) Segmented image after corrections, demonstrating goodagreement with qualitative segmentation of (a). ( e - h ) Processing of PI and EthD-1 images. (e) Raw image. (f) Contrastenhanced image. (g) Segmented image using 2-level intensity thresholding. (h) Counted particles or cells based on sizethresholding. ( i ) Schematic representation of radial binning and exclusion of occluded regions on an example stitched dishimage. Shaded region is included. Image has been linearly contrast enhanced ([0.01, 0.2] → [0.1, 1]) and down sampled to20% resolution to aid visualization. Segmentation via thresholding
To segment the contrast enhanced GFP image, two intensity thresholds were needed: a lower threshold to separate thebackground and dim px population and an upper threshold to separate the bright and dim px populations (Supp. Sec. 3.2).The upper threshold (0.3727) was selected by applying Otsu’s method (35) with 3 clusters to intensity histograms drawnfrom a central region of all control dish images from positions P2, P4, and P5 and then taking the average of the largerof the two returned intensity thresholds. The lower intensity threshold (0.0635) was selected by first filtering the contrastenhanced and denoised image region using a 3 × Classification error correction
After applying thresholds and quantizing filters to cluster the GFP image into background, dim, and bright populations, twotypes of classification errors were corrected (Supp. Sec. 3.3, Fig. S9). The first was a “halo” effect around bright regions causedby intensity drop-off such that the edges of bright regions were incorrectly classified as dim px. The second was a “spot” effectin the center of dim regions caused by intensity peaks. To correct the first type of error, all bright regions exceeding 8 px in sizewith a surrounding px population of at least 60% dim px were expanded by one px layer to fill the surrounding dim px (Figs.2c and d, green arrow), i.e., a flood fill. To correct the second type of error, all small bright regions ( < Radial binning and exclusion
Data was analyzed as a function of radius by constructing radial bins or bands. For GFP images, an arbitrary number ofequal annular width bands, 25, was used to divide the total image area starting from the dish center. The choice of band
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X. number results in approximately 0.72 mm wide bands. The bottom portion of the dish is occluded by a fitment and also showsirregularities. To avoid this area, only an upper portion of the band was kept from band 17 and on. The edge of the dish wasalso excluded due to irregularities and observed temperature differences: bands 23 – 25 were discarded from all analyses (Fig.2i). For the EthD-1 and PI stained images, a similar binning procedure with 12 bands of approximately 1.46 mm annular widewas used, with exclusion of lower regions after band 8, and a complete exclusion from band 10 onward.
Study design
In each experiment, 10 dishes were simultaneously seeded. Dishes were then randomized to either the control or treated dishsleeve/stack. Pairs of dishes were assigned to the same height position (P1 – P5). For the dishes assigned to P3, one dish isdiscarded and a dish with only media is used to continuously probe temperature in the treated dish sleeve. After 24 h incubationfollowed by 72 h of incubation and continuous 200 kHz induced EMF application to the treated dish sleeve, dishes were imagedand processed as described. Using data from cell counting calibration experiments, the imaged dim px density was relatedto cell density using an empirical fit, shown later. Each treated dish’s radially dependent cell density was normalized to itsposition matched (e.g., P2 treated vs. P2 control) control dish. In 4 experiments, additional PI and EthD-1 staining, imaging,and image processing was performed to clarify the role of cell death in results.Each experiment generated 3 replicates of dish pairs at P2, P4, and P5. Pairs of dishes assigned to P1 were discarded fromall analyses due to a substantial temperature difference, discussed later. Each dish pair was treated as a single experimentalobservation of the effect of 200 kHz EFs. Data from Kirson et al. (3) suggest that treatment of proliferating human cells withan EF of 200 kHz and 1 V/cm results in a decrease of roughly 20 ±
10% in cell number after 72 h compared to untreatedcells. Presuming an effect of this magnitude, 18 replicates are needed to obtain a conservative statistical power of 0.95 and aconfidence level of 𝛼 = 𝑁 =
36 replicates. In 2 of theseexperiments (6 replicates), PI staining and imaging was performed after primary imaging. In 2 other experiments, EthD-1staining and imaging was performed for a total of 𝑛 =
12 out of 36 replicates assessing cell death.
Statistical analysis and methods
All statistical analyses were performed in R 3.6.2. The linear fit for the experimentally measured EF amplitude was performedusing the typical linear least squares (lm) approach. The bounded exponential fit for dim px density vs. cell density calibration(of the form: [ Dim Px Density ] ∼ − exp { 𝛽 × [ Cell Density ] + 𝛽 } ) was performed using the Gauss-Newton nonlinear leastsquares (nls) approach. Initial guesses of 𝛽 = − 𝛽 = 𝑡 -tests were performedusing the base R t.test function assuming unequal variances, i.e., Welch’s 𝑡 -test. RESULTSDelivery of 200 kHz EFs up to ∼ Figure 3: Measurements of EF amplitude as a function of pickup coil radius/radial distance at different dish stack positions.Black line is a zero-intercept linear least squares fit using data from all positions ([EF Amplitude] = 4.242 × [Radial Distance],Std. Error = 0.071, Adj. R = 0.9902, p < Manuscript submitted to Biophysical Journalnduced EMF delivery to cells several different radii composed of insulated twisted wire pairs to minimize stray inductance. These pickup loops were hotglued to dishes, aligned centrally, and placed within the sleeve. The generated peak-to-peak voltage (V pk − pk ) was measured witha two-channel oscilloscope (Hantek6022BL, Hantek Electronic Co., Ltd., Qingdao, China) and EF amplitude was determinedby | E | = V pk − pk /( 𝜋𝑟 ) where 𝑟 is the loop radius. Results from these measurements are shown in Fig. 3, and agree withthe expected linear EF amplitude profile predicted within the coil from Eq. (3). 200 kHz oscillation was also observed in allmeasurements, confirming the linearity of the system. Note, EF amplitudes throughout refer to pk-pk values. We obtained amaximal EF amplitude of ∼ Temperature and osmolarity control during 72 h continuous 200 kHz EMF delivery
Figure 4: Quantification of temperature stability and homogeneity. ( a ) Temperature log data for the entire duration of a single72 h experiment from (top) the chiller water loop, (middle) the infrared detector reading of the coil, and (bottom) the fiberoptic sensor placed in the center of the dish at P3. Dashed lines indicate 37 ± ◦ C. Note the fluctuations in coil temperatureand chiller set-point adjustment throughout the experiment. ( b ) Adjusted temperature data from 3 fiber optic probes placedin different positions of the dish at P3 during the first 17.5 h of an experiment. ( c ) Temperature data from fiber optic probesplaced in the centers of dishes P1, P2, P4, and P5, expressed as a difference from a simultaneous reading of a probe in thecenter of the dish P3 with dashed lines indicating 0 ± ◦ C.These cell cultures are typically grown at a temperature of 37 ◦ C. Extensive steps were taken to maintain this desiredtemperature in the treated dish sleeve as growth rate is known to be function of ambient temperature. For the presentexperimental setup, 𝑚 = (slope of the fit Fig. 3), such that 0.1 W ≡ ≈ ± ◦ C, cooling the surrounding air to keep the dish at P3’s temperature at a homogeneous 37 ± ◦ C. Noinadvertent effects like Rayleigh-Bénard convection are expected using this temperature control method (Supp. Sec. 2.5, Eq.S24, Fig. S6). This method of temperature control was able to maintain the desired P3 dish temperature, without interruptionand without fail, for all 72 h experiments. Exemplar temperature logs for the P3 fiber optic sensor, the infrared coil recording,and the chiller loop over the course of one experiment are shown in Fig. 4a.To assess the homogeneity of temperature during a stimulation experiment, measurements from different locations withinthe P3 dish (Fig. 4b) (within-dish variability) and measurements from the center of different dish positions (Fig. 4c) (between-dish variability) were taken. For the measurements between different dish positions, data is presented as a difference from asimultaneous P3 dish measurement. In Fig. 4b, stimulation was started at time =
0. Temperature moves in tandem, demonstratingno time-lag in temperature adjustment. Readings show that temperature within the dish is similar, but the edge of the dish is ≈ ◦ C cooler. This was likely the result of convective heat transfer with the cooler surrounding air (see Supp. Secs. 2.1 - 2.4).The edges of all dishes were excluded in downstream analyses as described in Fig. 2i to ensure the integrity of within-plateinformation. In Fig. 4c, data is shown from segments of experiments that are underway and at steady state thermal conditions.
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X.
Temperature measurements from the centers of P2, P4, and P5 remained within 1 ◦ C of measurements from the center of P3 andwere stable. Measurements from the center of P1, however, were several ◦ C cooler and so P1 was excluded from all analyses.Overall, temperature control was satisfactory and allowed us to retain most data with confidence that temperature differenceswere minimal and unlikely to contribute to observed differences between experimental conditions.Figure 5: Quantification of osmolarity regulation improvements. ( a) Lid shaping process with steel ball impressing a pre-heatedlid placed on a sharpened ring mold. Note exposed surface area reduction in shaped lid. ( b ) Osmolarity changes after a 72 hexperiment in different conditions and configurations with 3 repetitions. Error bars = ± Estimating counted cell density using segmented px density obtained via in situ imaging ofGFP-expressing thyroid cells
GFP images from calibration experiments were processed as described in Fig. 2. Data from these calibration experimentsis expressed as total (whole dish) dim px density (px/px) vs. cell count per area (cells/sq. mm) in Fig. 6a. Grouped datacorresponded well to a bounded exponential fit, allowing for the robust conversion of px densities to approximate cell densities.The bounded exponential form of the fit may arise due to changes in the size and morphology of thyroid cells as theyapproach confluence. To demonstrate that our ‘calibration curve’ was valid for the range of data observed, a probability densitydistribution of dim px densities in all included radial bands of all P2, P4, and P5 control dishes is shown alongside the fit(Fig. 6b). Observed dim px densities fell within the range of the calibration curve. The mean value of this distribution is alsoindicated. The mean dim px density is 0.374 and the corresponding cell density from the fit is 358 cells/sq. mm. Manuscript submitted to Biophysical Journalnduced EMF delivery to cells
Figure 6: Translating dim px density to cell density. ( a ) Data from calibration experiments. Dim px density plotted vs. celldensity determined from cell counts. Bounded exponential fits of individual experiments shown in colored lines. A fit withpooled data ( 𝑛 =
56) is shown in black ( [ Dim Px Density ] = − exp {− × [ Cell Density ] − } , Residual sum-of-squares = 0.1922, Spearman’s 𝜌 = 0.949, p < b ) Distribution of dim px densities in included radial bands of controldishes ( 𝑛 = × Application of 200 kHz induced EMF for 72 h causes moderate EF amplitude-dependent reductionin human thyroid cell density, especially at EF amplitudes exceeding 4 V/cm
A pair of control and treated culture dishes is shown in Figs. 7a – b. The displayed images provide an example of a radiallydependent cell density reduction in the treated dish. Both a central and peripheral region of the control and treated dishes areshown at 100% resolution (Figs. 7b – c, e – f) after the described image processing pipeline (Figs. 2a – d) to demonstratethis radially dependent reduction in dim px density and thus cell density. Conversion from radially binned dim px density to anormalized cell density is also shown in Figs. 7g – h using the fit in Fig. 6.Data from all replicates ( 𝑁 =
36) was converted to normalized control vs. treated cell density curves as shown in Fig.7h. We address these normalized dim px density results first, and bright px density and EthD-1 and PI staining later. Averagecurves with bootstrapped 95% confidence intervals (CI) are shown in Fig. 8a. The bootstrapped difference between curvesis shown in Fig. 8b. Results demonstrate that significant differences in cell density are present throughout the dish, even atEF amplitudes thought to be sub-therapeutic ( < > Reduction in human thyroid cell density is not sufficiently explained by cell death
The observed reduction in cell density can be explained by a reduction in proliferation, an increase in cell death, or somemixture of both mechanisms. To explore their relative roles, bright px density and dead cell density were both quantified,shown in Fig. 9. Bright px density, which may correspond to cellular debris, pre-apoptotic cells, mitotic cells, or a roundedcell morphology, was quantified in the same way as dim px density (Fig. 2d) but considering the brighter (violet) segmentedpopulation. Dead cell density was quantified as described in the methods. Data from experiments with EthD-1 and PI stainingare pooled. Both bright px density (Fig. 9c – d) and dead cell density (Fig. 9a – b) are plotted as a percentage of the controlmean values indicated in Fig. 6 to contextualize the relative contribution of bright px and cell death to the effect observed in
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X.
Figure 7: Example of treated vs. control dish and data processing and representation. ( a – c ) Example of (a) control dish andzoomed, processed regions near the center (b, yellow) and near the periphery (c, magenta). Whole dish image is adjusted as inFig. 2i for visualization. ( d – f ) Corresponding treated dish with the same zoomed regions. An overlay of the applied 200 kHzEMF amplitude and azimuthal direction is shown. ( g – h ) Radially binned data comparing control and treated dishes shown inparts (a - f). (g) Raw dim px density converted to cell density (h, left axis) via the fit in Fig. 6, and subsequently normalizedto the mean of the control curve (h, right axis).Fig. 8. Results indicate that while the treated condition exhibits some qualitative differences in the radial curves for bright pxand dead cell density, these differences are not statistically significant at the replicate number used. Moreover, as a percentageof the control means in Fig. 6, the differences are small ( < Manuscript submitted to Biophysical Journalnduced EMF delivery to cells
Figure 8: Summary of treatment results presented as normalized cell density curves converted from dim px densities. ( a )Control vs. treated curve ( 𝑁 = ∗∗ = p < ∗ ∗ ∗ = p < b ) Differencebetween point-wise control vs. treated normalized cell densities bootstrapped in the same way. Dashed line at 4 V/cm indicatesapparent regime change.Figure 9: Cell death and bright px density in response to treatment. ( a - b ) Dead cell density curves and differences constructedfrom pooled EthD-1 staining ( 𝑛 =
6) and PI staining ( 𝑛 =
6) results processed as described and expressed as a percent of controlmean cell density (358 cells/sq. mm). Ribbons = 95% CI. No point-wise two-sample independent 𝑡 -tests were significant. ( c - d ) Bright cell density curves and difference ( 𝑛 = 𝑡 -tests were significant. DISCUSSION
In this work, we detail the development of methods to deliver moderate amplitude, intermediate frequency EFs using inducedEMFs. We were able to experimentally validate the delivery of a spatially varying EF profile and quantify substantialexperimental improvements in thermal and osmolar regulation. We also report novel results of 72 h application of continuous,200 kHz, 0 to 6.5 V/cm EMF stimulation on a non-cancerous, proliferating human cell line. Results were analyzed using a newlydeveloped image processing pipeline designed specifically to characterize the in situ images necessitated by the applicationof a spatially varying EF “dose” profile. Our results suggest that (1) 200 kHz EMF have a moderate maximal effect (< 25%)on non-cancerous cells in this amplitude range, (2) the effect of 200 kHz EMF on cell density is likely anti-proliferative (i.e.,
Manuscript submitted to Biophysical Journal C C E P T E D M A N U S C R I P T Ravin, R. & Cai, T. X. anti-mitotic) rather than destructive, and (3) the effect on cell density increases with increasing EF amplitude in a higher EFregime (> 4 V/cm).Our exploratory results corroborate and extend prior findings on the effects of “TTFields” on proliferating, non-cancerouscell lines. Kirson et al. (2) reported that baby hamster kidney (BHK) fibroblasts grown in 0.1% fetal calf serum exhibited nodecrease in growth rate during 24 h application of 1.2 V/cm pk, 100 kHz EFs. Similarly, Jo et al. (23) reported little to noreduction in cell count for normal human skin cells (CCD-986sk) after 72 h application of up to 1.5 V/cm pk, 200 kHz EFs.Our results agree with these findings. Within and below the therapeutic EF amplitude range of 1 – 3 V/cm pk-pk, we foundthat human thyroid cells (Nthy-ori 3-1 Sigma) did not exhibit EF amplitude-dependent growth reduction. We then extendprior findings by showing that, for this cell line, an EF amplitude-dependent effect is observed at > ≈
75 kWh). Futuredevelopments should attempt to miniaturize the experimental platform and introduce parallelism to enable combinatorialexperiments. Alternative coil architectures, chiller loops, multi-coil array designs, controllable capacitor banks, etc., are allworthwhile engineering improvements that could enhance the utility, applicability, and efficiency of this platform. Thesedesigns could be motivated by more detailed finite element modeling (FEM) of the various components of this test system,which is beyond the scope of this present work.
CONCLUSION
We report methods to deliver EMFs to cell and tissue cultures using electromagnetic induction, which overcomes priorlimitations associated with the delivery of 200 kHz EFs using contacting electrodes. The proposed strategy for inductive EMF Manuscript submitted to Biophysical Journalnduced EMF delivery to cells delivery eliminates conductive heating of culture dishes and improves temperature control and regulation. Additional designenhancements of our culture dishes improved osmoregulation by mitigating media loss. We report results of stimulation ofhuman thyroid cell line cultures with 200 kHz EMFs exceeding 3 V/cm pk-pk and as high as 6.5 V/cm. Results support amoderate anti-proliferative effect, especially at EF amplitudes exceeding 4 V/cm. The effect is unlikely to be due to cell deathbased on secondary PI and EthD-1 staining. In summary, the 200 kHz EMF delivery method reported here is a carefully vettedand well-characterized in vitro experimental platform for future exploration of the effect of intermediate frequency EMFs onliving (and non-living) systems.
AUTHOR CONTRIBUTIONS
R.R., P.J.B. conceived the study. R.R., P.J.B., T.X.C. designed the study. R.H.P., M.G.C., T.P. contributed to electric fielddelivery and monitoring setup. R.R., H.W., A.G. contributed to cell culture methods. R.R. performed all experiments andacquired data. T.X.C. designed the analysis, analyzed data, and wrote the supplementary material. R.Z.F. contributed to imagedata analysis. T.X.C., R.R., P.J.B. wrote the manuscript. N.H.W. gave technical advice. M.R.G. and Z.Z. provided cancer andcell biology expertise and contributed to the discussion. P.J.B. supervised the project. All authors edited the manuscript.
ACKNOWLEDGMENTS
Light microscopy was performed at the NICHD Microscopy & Imaging Core with the assistance of Dr. Vincent Schram andMrs. Lynne Holtzclaw. We would also like to thank Dr. Nicole Morgan for helpful discussions. The authors declare no conflictsof interest. Support for this study comes from the Intramural Research Program (IRP) of the
Eunice Kennedy Shriver
NationalInstitute of Child Health and Human Development.
SUPPORTING MATERIAL
Supporting material accompanies this article and contains supplemental sections concerning: (1) estimated media loss calcu-lations, (2) heat transfer and temperature profile calculations, (3) image processing and threshold selection implementation.References (38–51) appear in the supporting material. The supporting material can be found by visiting BJ Online at . REFERENCES
1. Cifra, M., J. Z. Fields, and A. Farhadi, 2011. Electromagnetic cellular interactions.
Progress in Biophysics and MolecularBiology
Cancer Research
Proceedings of the NationalAcademy of Sciences of the United States of America : PNAS
Current Medicinal Chemistry
New England Journal of Medicine: NEJM
New EnglandJournal of Medicine: NEJM
Manuscript submitted to Biophysical Journal avin, R. & Cai, T. X. M. Malkin, J. C. Landolfi, F. Payer, M. Mehdorn, R. J. Weil, S. C. Pannullo, M. Westphal, M. Smrcka, L. Chin, H. Kostron,S. Hofer, J. Bruce, R. Cosgrove, N. Paleologous, Y. Palti, and P. H. Gutin, 2012. NovoTTF-100A versus physician’s choicechemotherapy in recurrent glioblastoma: A randomised phase III trial of a novel treatment modality.
European Journalof Cancer
JAMA
The Lancet Oncology
PLOS ONE
International Journal of Environmental Research and Public Health In Physics in medicine andbiology
IEEEReviews in Biomedical Engineering
Annals of the NewYork Academy of Sciences
Biophysical Journal
Scientific Reports
IEEE Transactions on Biomedical Engineering
Cell DeathDiscov
Scientific Reports
CellDeath Discovery Manuscript submitted to Biophysical Journalnduced EMF delivery to cells
22. Giladi, M., M. Munster, R. S. Schneiderman, T. Voloshin, Y. Porat, R. Blat, K. Zielinska-Chomej, P. Hååg, Z. Bomzon,E. D. Kirson, U. Weinberg, K. Viktorsson, R. Lewensohn, and Y. Palti, 2017. Tumor treating fields (TTFields) delay DNAdamage repair following radiation treatment of glioma cells.
Radiat Oncol
Cell Death Discovery
Cell Death Dis
Oncogene
Cancers
Scientific reports
JoVE e55820.29. Finan, J. D., and F. Guilak, 2010. The effects of osmotic stress on the structure and function of the cell nucleus.
Journalof cellular biochemistry
PLOS ONE
Scientific Reports In Digest ofthe 6th International Conference in Medical Electronics and Biological Engineering. IFMBE, 112.33. Hassanpour, H., N. Samadiani, and S. M. M. Salehi, 2015. Using morphological transforms to enhance the contrast ofmedical images.
The Egyptian Journal of Radiology and Nuclear Medicine
IEEE Transactions on Systems, Man, andCybernetics
Scientific Reports
Medical Physics
Gas u. Wasserfach
The Journal of ChemicalPhysics
Manuscript submitted to Biophysical Journal avin, R. & Cai, T. X.
41. Teske, V., E. Vogel, and E. Bich, 2005. Viscosity Measurements on Water Vapor and Their Evaluation.
Journal ofChemical & Engineering Data In Proceedings of 1994 IEEE Industry Applications Society Annual Meeting. volume 2, 1470–1476 vol.2.43. Chen, M.-T., C. Jiang, P. T. Vernier, Y.-H. Wu, and M. A. Gundersen, 2009. Two-dimensional nanosecond electric fieldmapping based on cell electropermeabilization.
PMC Biophysics
The Journal of Chemical Physics
International Journal of Heat and Mass Transfer
Labon a Chip18