Marc Hafner

LINCS Canvas Browser: interactive web app to query, browse and interrogate LINCS L1000 gene expression signatures

For the Library of Integrated Network-based Cellular Signatures (LINCS) project many gene expression signatures using the L1000 technology have been produced. The L1000 technology is a cost-effective method to profile gene expression in large scale. LINCS Canvas Browser (LCB) is an interactive HTML5 web-based software application that facilitates querying, browsing and interrogating many of the currently available LINCS L1000 data. LCB implements two compacted layered canvases, one to visualize clustered L1000 expression data, and the other to display enrichment analysis results using 30 different gene set libraries. Clicking on an experimental condition highlights gene-sets enriched for the differentially expressed genes from the selected experiment. A search interface allows users to input gene lists and query them against over 100 000 conditions to find the top matching experiments. The tool integrates many resources for an unprecedented potential for new discoveries in systems biology and systems pharmacology. The LCB application is available at http://www.maayanlab.net/LINCS/LCB. Customized versions will be made part of the http://lincscloud.org and http://lincs.hms.harvard.edu websites.

Profiles of Basal and Stimulated Receptor Signaling Networks Predict Drug Response in Breast Cancer Lines

Activity of receptor tyrosine kinase networks may serve as an effective means to classify breast cancers and predict their sensitivity to therapeutic drugs. Hybrid Biomarkers Changes in genes only tell part of the story in cancer. Cancer cells also exhibit altered signaling networks and can rewire their signaling pathways in response to either endogenous stimuli or drug therapy. Niepel et al. combined information about breast cancer clinical subtype, genetic status, receptor abundance, and signaling pathway activity to create hybrid biomarkers that predicted the effectiveness of various targeted breast cancer therapies. This approach may prove a better way to customize treatments for cancer patients because it accounts for heterogeneity in tumors with similar genetic alterations and because the multitude of genetic changes observed in cancer appear to result in a smaller set of dysregulated signaling states. Identifying factors responsible for variation in drug response is essential for the effective use of targeted therapeutics. We profiled signaling pathway activity in a collection of breast cancer cell lines before and after stimulation with physiologically relevant ligands, which revealed the variability in network activity among cells of known genotype and molecular subtype. Despite the receptor-based classification of breast cancer subtypes, we found that the abundance and activity of signaling proteins in unstimulated cells (basal profile), as well as the activity of proteins in stimulated cells (signaling profile), varied within each subtype. Using a partial least-squares regression approach, we constructed models that significantly predicted sensitivity to 23 targeted therapeutics. For example, one model showed that the response to the growth factor receptor ligand heregulin effectively predicted the sensitivity of cells to drugs targeting the cell survival pathway mediated by PI3K (phosphoinositide 3-kinase) and Akt, whereas the abundance of Akt or the mutational status of the enzymes in the pathway did not. Thus, basal and signaling protein profiles may yield new biomarkers of drug sensitivity and enable the identification of appropriate therapies in cancers characterized by similar functional dysregulation of signaling networks.

'Glocal' robustness analysis and model discrimination for circadian oscillators

To characterize the behavior and robustness of cellular circuits with many unknown parameters is a major challenge for systems biology. Its difficulty rises exponentially with the number of circuit components. We here propose a novel analysis method to meet this challenge. Our method identifies the region of a high-dimensional parameter space where a circuit displays an experimentally observed behavior. It does so via a Monte Carlo approach guided by principal component analysis, in order to allow efficient sampling of this space. This ‘global’ analysis is then supplemented by a ‘local’ analysis, in which circuit robustness is determined for each of the thousands of parameter sets sampled in the global analysis. We apply this method to two prominent, recent models of the cyanobacterial circadian oscillator, an autocatalytic model, and a model centered on consecutive phosphorylation at two sites of the KaiC protein, a key circadian regulator. For these models, we find that the two-sites architecture is much more robust than the autocatalytic one, both globally and locally, based on five different quantifiers of robustness, including robustness to parameter perturbations and to molecular noise. Our ‘glocal’ combination of global and local analyses can also identify key causes of high or low robustness. In doing so, our approach helps to unravel the architectural origin of robust circuit behavior. Complementarily, identifying fragile aspects of system behavior can aid in designing perturbation experiments that may discriminate between competing mechanisms and different parameter sets.

Growth rate inhibition metrics correct for confounders in measuring sensitivity to cancer drugs

Drug sensitivity and resistance are conventionally quantified by IC50 or Emax values, but these metrics are highly sensitive to the number of divisions taking place over the course of a response assay. The dependency of IC50 and Emax on division rate creates artefactual correlations between genotype and drug sensitivity, while obscuring valuable biological insights and interfering with biomarker discovery. We derive alternative small molecule drug-response metrics that are insensitive to division number. These are based on estimation of the magnitude of drug-induced growth rate inhibition (GR) using endpoint or time-course assays. We show that GR50 and GRmax are superior to conventional metrics for assessing the effects of small molecule drugs in dividing cells. Moreover, adopting GR metrics requires only modest changes in experimental protocols. We expect GR metrics to improve the study of cell signaling and growth using small molecules and biologics and to facilitate the discovery of drug-response biomarkers and the identification of drugs effective against specific patient-derived tumor cells.

Fractional killing arises from cell-to-cell variability in overcoming a caspase activity threshold

When cells are exposed to death ligands such as TRAIL, a fraction undergoes apoptosis and a fraction survives; if surviving cells are re‐exposed to TRAIL, fractional killing is once again observed. Therapeutic antibodies directed against TRAIL receptors also cause fractional killing, even at saturating concentrations, limiting their effectiveness. Fractional killing arises from cell‐to‐cell fluctuations in protein levels (extrinsic noise), but how this results in a clean bifurcation between life and death remains unclear. In this paper, we identify a threshold in the rate and timing of initiator caspase activation that distinguishes cells that live from those that die; by mapping this threshold, we can predict fractional killing of cells exposed to natural and synthetic agonists alone or in combination with sensitizing drugs such as bortezomib. A phenomenological model of the threshold also quantifies the contributions of two resistance genes (c‐FLIP and Bcl‐2), providing new insight into the control of cell fate by opposing pro‐death and pro‐survival proteins and suggesting new criteria for evaluating the efficacy of therapeutic TRAIL receptor agonists.

L1000CDS2: LINCS L1000 characteristic direction signatures search engine

The library of integrated network-based cellular signatures (LINCS) L1000 data set currently comprises of over a million gene expression profiles of chemically perturbed human cell lines. Through unique several intrinsic and extrinsic benchmarking schemes, we demonstrate that processing the L1000 data with the characteristic direction (CD) method significantly improves signal to noise compared with the MODZ method currently used to compute L1000 signatures. The CD processed L1000 signatures are served through a state-of-the-art web-based search engine application called L1000CDS2. The L1000CDS2 search engine provides prioritization of thousands of small-molecule signatures, and their pairwise combinations, predicted to either mimic or reverse an input gene expression signature using two methods. The L1000CDS2 search engine also predicts drug targets for all the small molecules profiled by the L1000 assay that we processed. Targets are predicted by computing the cosine similarity between the L1000 small-molecule signatures and a large collection of signatures extracted from the gene expression omnibus (GEO) for single-gene perturbations in mammalian cells. We applied L1000CDS2 to prioritize small molecules that are predicted to reverse expression in 670 disease signatures also extracted from GEO, and prioritized small molecules that can mimic expression of 22 endogenous ligand signatures profiled by the L1000 assay. As a case study, to further demonstrate the utility of L1000CDS2, we collected expression signatures from human cells infected with Ebola virus at 30, 60 and 120 min. Querying these signatures with L1000CDS2 we identified kenpaullone, a GSK3B/CDK2 inhibitor that we show, in subsequent experiments, has a dose-dependent efficacy in inhibiting Ebola infection in vitro without causing cellular toxicity in human cell lines. In summary, the L1000CDS2 tool can be applied in many biological and biomedical settings, while improving the extraction of knowledge from the LINCS L1000 resource.

A Quantitative Approach to Screen for Nephrotoxic Compounds In Vitro

Nephrotoxicity due to drugs and environmental chemicals accounts for significant patient mortality and morbidity, but there is no high throughput in vitro method for predictive nephrotoxicity assessment. We show that primary human proximal tubular epithelial cells (HPTECs) possess characteristics of differentiated epithelial cells rendering them desirable to use in such in vitro systems. To identify a reliable biomarker of nephrotoxicity, we conducted multiplexed gene expression profiling of HPTECs after exposure to six different concentrations of nine human nephrotoxicants. Only overexpression of the gene encoding heme oxygenase-1 (HO-1) significantly correlated with increasing dose for six of the compounds, and significant HO-1 protein deregulation was confirmed with each of the nine nephrotoxicants. Translatability of HO-1 increase across species and platforms was demonstrated by computationally mining two large rat toxicogenomic databases for kidney tubular toxicity and by observing a significant increase in HO-1 after toxicity using an ex vivo three-dimensional microphysiologic system (kidney-on-a-chip). The predictive potential of HO-1 was tested using an additional panel of 39 mechanistically distinct nephrotoxic compounds. Although HO-1 performed better (area under the curve receiver-operator characteristic curve [AUC-ROC]=0.89) than traditional endpoints of cell viability (AUC-ROC for ATP=0.78; AUC-ROC for cell count=0.88), the combination of HO-1 and cell count further improved the predictive ability (AUC-ROC=0.92). We also developed and optimized a homogenous time-resolved fluorescence assay to allow high throughput quantitative screening of nephrotoxic compounds using HO-1 as a sensitive biomarker. This cell-based approach may facilitate rapid assessment of potential nephrotoxic therapeutics and environmental chemicals.

Effect of Network Architecture on Synchronization and Entrainment Properties of the Circadian Oscillations in the Suprachiasmatic Nucleus

In mammals, the suprachiasmatic nucleus (SCN) of the hypothalamus constitutes the central circadian pacemaker. The SCN receives light signals from the retina and controls peripheral circadian clocks (located in the cortex, the pineal gland, the liver, the kidney, the heart, etc.). This hierarchical organization of the circadian system ensures the proper timing of physiological processes. In each SCN neuron, interconnected transcriptional and translational feedback loops enable the circadian expression of the clock genes. Although all the neurons have the same genotype, the oscillations of individual cells are highly heterogeneous in dispersed cell culture: many cells present damped oscillations and the period of the oscillations varies from cell to cell. In addition, the neurotransmitters that ensure the intercellular coupling, and thereby the synchronization of the cellular rhythms, differ between the two main regions of the SCN. In this work, a mathematical model that accounts for this heterogeneous organization of the SCN is presented and used to study the implication of the SCN network topology on synchronization and entrainment properties. The results show that oscillations with larger amplitude can be obtained with scale-free networks, in contrast to random and local connections. Networks with the small-world property such as the scale-free networks used in this work can adapt faster to a delay or advance in the light/dark cycle (jet lag). Interestingly a certain level of cellular heterogeneity is not detrimental to synchronization performances, but on the contrary helps resynchronization after jet lag. When coupling two networks with different topologies that mimic the two regions of the SCN, efficient filtering of pulse-like perturbations in the entrainment pattern is observed. These results suggest that the complex and heterogeneous architecture of the SCN decreases the sensitivity of the network to short entrainment perturbations while, at the same time, improving its adaptation abilities to long term changes.

Analysis of growth factor signaling in genetically diverse breast cancer lines

BackgroundSoluble growth factors present in the microenvironment play a major role in tumor development, invasion, metastasis, and responsiveness to targeted therapies. While the biochemistry of growth factor-dependent signal transduction has been studied extensively in individual cell types, relatively little systematic data are available across genetically diverse cell lines.ResultsWe describe a quantitative and comparative dataset focused on immediate-early signaling that regulates the AKT (AKT1/2/3) and ERK (MAPK1/3) pathways in a canonical panel of well-characterized breast cancer lines. We also provide interactive web-based tools to facilitate follow-on analysis of the data. Our findings show that breast cancers are diverse with respect to ligand sensitivity and signaling biochemistry. Surprisingly, triple negative breast cancers (TNBCs; which express low levels of ErbB2, progesterone and estrogen receptors) are the most broadly responsive to growth factors and HER2amp cancers (which overexpress ErbB2) the least. The ratio of ERK to AKT activation varies with ligand and subtype, with a systematic bias in favor of ERK in hormone receptor positive (HR+) cells. The factors that correlate with growth factor responsiveness depend on whether fold-change or absolute activity is considered the key biological variable, and they differ between ERK and AKT pathways.ConclusionsResponses to growth factors are highly diverse across breast cancer cell lines, even within the same subtype. A simple four-part heuristic suggests that diversity arises from variation in receptor abundance, an ERK/AKT bias that depends on ligand identity, a set of factors common to all receptors that varies in abundance or activity with cell line, and an “indirect negative regulation” by ErbB2. This analysis sets the stage for the development of a mechanistic and predictive model of growth factor signaling in diverse cancer lines. Interactive tools for looking up these results and downloading raw data are available at http://lincs.hms.harvard.edu/niepel-bmcbiol-2014/.

Deterministic characterization of phase noise in biomolecular oscillators

On top of the many external perturbations, cellular oscillators also face intrinsic perturbations due the randomness of chemical kinetics. Biomolecular oscillators, distinct in their parameter sets or distinct in their architecture, show different resilience with respect to such intrinsic perturbations. Assessing this resilience can be done by ensemble stochastic simulations. These are computationally costly and do not permit further insights into the mechanistic cause of the observed resilience. For reaction systems operating at a steady state, the linear noise approximation (LNA) can be used to determine the effect of molecular noise. Here we show that methods based on LNA fail for oscillatory systems and we propose an alternative ansatz. It yields an asymptotic expression for the phase diffusion coefficient of stochastic oscillators. Moreover, it allows us to single out the noise contribution of every reaction in an oscillatory system. We test the approach on the one-loop model of the Drosophila circadian clock. Our results are consistent with those obtained through stochastic simulations with a gain in computational efficiency of about three orders of magnitude.