Prediction of drug cocktail effects when the number of measurements is limited.

Cocktails of drugs can be more effective than single drugs, because they can potentially work at lower doses and avoid resistance. However, it is impossible to test all drug cocktails drawn from a large set of drugs because of the huge number of combinations. To overcome this combinatorial explosion problem, one can sample a relatively small number of combinations and use a model to predict the rest. Recently, Zimmer and Katzir et al. presented a model that accurately predicted the effects of cocktails at all doses based on measuring pairs of drugs. This model requires measuring each pair at several different doses and uses interpolation to reduce experimental noise. However, often, it is not possible to measure each pair at multiple doses (for example, in scarce patient-derived tumor material or in large screens). Here, we ask whether measurements at only a single dose can also predict high-order drug cocktails. To address this, we present a fully factorial experimental dataset on all drug cocktails built of 6 chemotherapy drugs on 2 cancer cell lines. We develop a formula that uses only pair measurements at a single dose to predict much of the variation up to 6-drug cocktails in the present data, outperforming commonly used Bliss independence and regression approaches. This model, called the pairs model, is an extension of the Bliss independence model to pairs: For M drugs, it equals the product of all pair effects to the power 1/(M-1). The pairs model also shows good agreement with previously published data on antibiotic triplets and quadruplets. The present model can only predict combinations at the same doses in which the pairs were measured and is not able to predict effects at other doses. This study indicates that pair-based approaches might be able to usefully predict and prioritize high-order combinations, even in large screens or when material for testing is limited.

Noise-precision tradeoff in predicting combinations of mutations and drugs.

Many biological problems involve the response to multiple perturbations. Examples include response to combinations of many drugs, and the effects of combinations of many mutations. Such problems have an exponentially large space of combinations, which makes it infeasible to cover the entire space experimentally. To overcome this problem, several formulae that predict the effect of drug combinations or fitness landscape values have been proposed. These formulae use the effects of single perturbations and pairs of perturbations to predict triplets and higher order combinations. Interestingly, different formulae perform best on different datasets. Here we use Pareto optimality theory to quantitatively explain why no formula is optimal for all datasets, due to an inherent bias-variance (noise-precision) tradeoff. We calculate the Pareto front of log-linear formulae and find that the optimal formula depends on properties of the dataset: the typical interaction strength and the experimental noise. This study provides an approach to choose a suitable prediction formula for a given dataset, in order to best overcome the combinatorial explosion problem.

Prediction of multidimensional drug dose responses based on measurements of drug pairs.

Finding potent multidrug combinations against cancer and infections is a pressing therapeutic challenge; however, screening all combinations is difficult because the number of experiments grows exponentially with the number of drugs and doses. To address this, we present a mathematical model that predicts the effects of three or more antibiotics or anticancer drugs at all doses based only on measurements of drug pairs at a few doses, without need for mechanistic information. The model provides accurate predictions on available data for antibiotic combinations, and on experiments presented here on the response matrix of three cancer drugs at eight doses per drug. This approach offers a way to search for effective multidrug combinations using a small number of experiments.

Noise genetics: inferring protein function by correlating phenotype with protein levels and localization in individual human cells.

To understand gene function, genetic analysis uses large perturbations such as gene deletion, knockdown or over-expression. Large perturbations have drawbacks: they move the cell far from its normal working point, and can thus be masked by off-target effects or compensation by other genes. Here, we offer a complementary approach, called noise genetics. We use natural cell-cell variations in protein level and localization, and correlate them to the natural variations of the phenotype of the same cells. Observing these variations is made possible by recent advances in dynamic proteomics that allow measuring proteins over time in individual living cells. Using motility of human cancer cells as a model system, and time-lapse microscopy on 566 fluorescently tagged proteins, we found 74 candidate motility genes whose level or localization strongly correlate with motility in individual cells. We recovered 30 known motility genes, and validated several novel ones by mild knockdown experiments. Noise genetics can complement standard genetics for a variety of phenotypes.

Multiomic signatures of body mass index identify heterogeneous health phenotypes and responses to a lifestyle intervention.

Multiomic profiling can reveal population heterogeneity for both health and disease states. Obesity drives a myriad of metabolic perturbations and is a risk factor for multiple chronic diseases. Here we report an atlas of cross-sectional and longitudinal changes in 1,111 blood analytes associated with variation in body mass index (BMI), as well as multiomic associations with host polygenic risk scores and gut microbiome composition, from a cohort of 1,277 individuals enrolled in a wellness program (Arivale). Machine learning model predictions of BMI from blood multiomics captured heterogeneous phenotypic states of host metabolism and gut microbiome composition better than BMI, which was also validated in an external cohort (TwinsUK). Moreover, longitudinal analyses identified variable BMI trajectories for different omics measures in response to a healthy lifestyle intervention; metabolomics-inferred BMI decreased to a greater extent than actual BMI, whereas proteomics-inferred BMI exhibited greater resistance to change. Our analyses further identified blood analyte-analyte associations that were modified by metabolomics-inferred BMI and partially reversed in individuals with metabolic obesity during the intervention. Taken together, our findings provide a blood atlas of the molecular perturbations associated with changes in obesity status, serving as a resource to quantify metabolic health for predictive and preventive medicine.

DNA-based copy number analysis confirms genomic evolution of PDX models.

Genomic evolution of patient-derived xenografts (PDXs) may lead to their gradual divergence away of their tumors of origin. We previously reported the genomic evolution of the copy number (CN) landscapes of PDXs during their engraftment and passaging1. However, whether PDX models are highly stable throughout passaging2, or can evolve CNAs rapidly1,3, remains controversial. Here, we reassess the genomic evolution of PDXs using DNA-based CN profiles. We find strong evidence for genomic evolution in the DNA-based PDX data: a median of ~10% of the genome is differentially altered between matched primary tumors (PTs) and PDXs across cohorts (range, 0% to 73% across all models). In 24% of the matched PT-PDX samples, over a quarter of the genome is differentially affected by CN alterations. Moreover, in matched analyses of PTs and their derived PDXs at multiple passages, later-passage PDXs are significantly less similar to their parental PTs than earlier-passage PDXs, indicative of genomic divergence. We conclude that PDX models indeed evolve throughout their derivation and propagation, and that the phenotypic consequences of this evolution ought to be assessed in order to determine its relevance to the proper application of these valuable cancer models.

A framework for clinical cancer subtyping from nucleosome profiling of cell-free DNA.

Cell-free DNA (cfDNA) has the potential to inform tumor subtype classification and help guide clinical precision oncology. Here we develop Griffin, a framework for profiling nucleosome protection and accessibility from cfDNA to study the phenotype of tumors using as low as 0.1x coverage whole genome sequencing data. Griffin employs a GC correction procedure tailored to variable cfDNA fragment sizes, which generates a better representation of chromatin accessibility and improves the accuracy of cancer detection and tumor subtype classification. We demonstrate estrogen receptor subtyping from cfDNA in metastatic breast cancer. We predict estrogen receptor subtype in 139 patients with at least 5% detectable circulating tumor DNA with an area under the receive operator characteristic curve (AUC) of 0.89 and validate performance in independent cohorts (AUC = 0.96). In summary, Griffin is a framework for accurate tumor subtyping and can be generalizable to other cancer types for precision oncology applications.

The geometry of clinical labs and wellness states from deeply phenotyped humans.

Longitudinal multi-omics measurements are highly valuable in studying heterogeneity in health and disease phenotypes. For thousands of people, we have collected longitudinal multi-omics data. To analyze, interpret and visualize this extremely high-dimensional data, we use the Pareto Task Inference (ParTI) method. We find that the clinical labs data fall within a tetrahedron. We then use all other data types to characterize the four archetypes. We find that the tetrahedron comprises three wellness states, defining a wellness triangular plane, and one aberrant health state that captures aspects of commonality in movement away from wellness. We reveal the tradeoffs that shape the data and their hierarchy, and use longitudinal data to observe individual trajectories. We then demonstrate how the movement on the tetrahedron can be used for detecting unexpected trajectories, which might indicate transitions from health to disease and reveal abnormal conditions, even when all individual blood measurements are in the norm.

Gut microbiome pattern reflects healthy ageing and predicts survival in humans.

The gut microbiome has important effects on human health, yet its importance in human ageing remains unclear. In the present study, we demonstrate that, starting in mid-to-late adulthood, gut microbiomes become increasingly unique to individuals with age. We leverage three independent cohorts comprising over 9,000 individuals and find that compositional uniqueness is strongly associated with microbially produced amino acid derivatives circulating in the bloodstream. In older age (over ~80 years), healthy individuals show continued microbial drift towards a unique compositional state, whereas this drift is absent in less healthy individuals. The identified microbiome pattern of healthy ageing is characterized by a depletion of core genera found across most humans, primarily Bacteroides. Retaining a high Bacteroides dominance into older age, or having a low gut microbiome uniqueness measure, predicts decreased survival in a 4-year follow-up. Our analysis identifies increasing compositional uniqueness of the gut microbiome as a component of healthy ageing, which is characterized by distinct microbial metabolic outputs in the blood.

Tumor diversity and the trade-off between universal cancer tasks.

Recent advances have enabled powerful methods to sort tumors into prognosis and treatment groups. We are still missing, however, a general theoretical framework to understand the vast diversity of tumor gene expression and mutations. Here we present a framework based on multi-task evolution theory, using the fact that tumors need to perform multiple tasks that contribute to their fitness. We find that trade-offs between tasks constrain tumor gene-expression to a continuum bounded by a polyhedron whose vertices are gene-expression profiles, each specializing in one task. We find five universal cancer tasks across tissue-types: cell-division, biomass and energy, lipogenesis, immune-interaction and invasion and tissue-remodeling. Tumors that specialize in a task are sensitive to drugs that interfere with this task. Driver, but not passenger, mutations tune gene-expression towards specialization in specific tasks. This approach can integrate additional types of molecular data into a framework of tumor diversity grounded in evolutionary theory.

Dynamic proteomics reveals bimodal protein dynamics of cancer cells in response to HSP90 inhibitor.

BACKGROUND: Drugs often kill some cancer cells while others survive. This stochastic outcome is seen even in clonal cells grown under the same conditions. Understanding the molecular reasons for this stochastic outcome is a current challenge, which requires studying the proteome at the single cell level over time. In a previous study we used dynamic proteomics to study the response of cancer cells to a DNA damaging drug, camptothecin. Several proteins showed bimodal dynamics: they rose in some cells and decreased in others, in a way that correlated with eventual cell fate: death or survival. Here we ask whether bimodality is a special case for camptothecin, or whether it occurs for other drugs as well. To address this, we tested a second drug with a different mechanism of action, an HSP90 inhibitor. We used dynamic proteomics to follow 100 proteins in space and time, endogenously tagged in their native chromosomal location in individual living human lung-cancer cells, following drug administration. RESULTS: We find bimodal dynamics for a quarter of the proteins. In some cells these proteins strongly rise in level about 12 h after treatment, but in other cells their level drops or remains constant. The proteins which rise in surviving cells included anti-apoptotic factors such as DDX5, and cell cycle regulators such as RFC1. The proteins that rise in cells that eventually die include pro-apoptotic factors such as APAF1. The two drugs shared some aspects in their single-cell response, including 7 of the bimodal proteins and translocation of oxidative response proteins to the nucleus, but differed in other aspects, with HSP90i showing more bimodal proteins. Moreover, the cell cycle phase at drug administration impacted the probability to die from HSP90i but not camptothecin. CONCLUSIONS: Single-cell dynamic proteomics reveals sub-populations of cells within a clonal cell line with different protein dynamics in response to a drug. These different dynamics correlate with cell survival or death. Bimodal proteins which correlate with cell fate may be potential drug targets to enhance the effects of therapy.

Dynamic Proteomics of Herpes Simplex Virus Infection.

The cellular response to viral infection is usually studied at the level of cell populations. Currently, it remains an open question whether and to what extent cell-to-cell variability impacts the course of infection. Here we address this by dynamic proteomics-imaging and tracking 400 yellow fluorescent protein (YFP)-tagged host proteins in individual cells infected by herpes simplex virus 1. By quantifying time-lapse fluorescence imaging, we analyze how cell-to-cell variability impacts gene expression from the viral genome. We identify two proteins, RFX7 and geminin, whose levels at the time of infection correlate with successful initiation of gene expression. These proteins are cell cycle markers, and we find that the position in the cell cycle at the time of infection (along with the cell motility and local cell density) can reasonably predict in which individual cells gene expression from the viral genome will commence. We find that the onset of cell division dramatically impacts the progress of infection, with 70% of dividing cells showing no additional gene expression after mitosis. Last, we identify four host proteins that are specifically modulated in infected cells, of which only one has been previously recognized. SUMO2 and RPAP3 levels are rapidly reduced, while SLTM and YTHDC1 are redistributed to form nuclear foci. These modulations are dependent on the expression of ICP0, as shown by infection with two mutant viruses that lack ICP0. Taken together, our results provide experimental validation for the long-held notion that the success of infection is dependent on the state of the host cell at the time of infection.IMPORTANCE High-throughput assays have revolutionized many fields in biology, both by allowing a more global understanding of biological processes and by deciphering rare events in subpopulations. Here we use such an assay, dynamic proteomics, to study viral infection at the single-cell level. We follow tens of thousands of individual cells infected by herpes simplex virus using fluorescence live imaging. Our results link the state of a cell at the time of virus infection with its probability to successfully initiate gene expression from the viral genome. Further, we identified three cellular proteins that were previously unknown to respond to viral infection. We conclude that dynamic proteomics provides a powerful tool to study single-cell differences during viral infection.

Promoter activity dynamics in the lag phase of Escherichia coli.

BACKGROUND: Lag phase is a period of time with no growth that occurs when stationary phase bacteria are transferred to a fresh medium. Bacteria in lag phase seem inert: their biomass does not increase. The low number of cells and low metabolic activity make it difficult to study this phase. As a consequence, it has not been studied as thoroughly as other bacterial growth phases. However, lag phase has important implications for bacterial infections and food safety. We asked which, if any, genes are expressed in the lag phase of Escherichia coli, and what is their dynamic expression pattern. RESULTS: We developed an assay based on imaging flow cytometry of fluorescent reporter cells that overcomes the challenges inherent in studying lag phase. We distinguish between lag1 phase- in which there is no biomass growth, and lag2 phase--in which there is biomass growth but no cell division. We find that in lag1 phase, most promoters are not active, except for the enzymes that utilize the specific carbon source in the medium. These genes show promoter activities that increase exponentially with time, despite the fact that the cells do not measurably increase in size. An oxidative stress promoter, katG, is also active. When cells enter lag2 and begin to grow in size, they switch to a full growth program of promoter activity including ribosomal and metabolic genes. CONCLUSIONS: The observed exponential increase in enzymes for the specific carbon source followed by an abrupt switch to production of general growth genes is a solution of an optimal control model, known as bang-bang control. The present approach contributes to the understanding of lag phase, the least studied of bacterial growth phases.