High-throughput deconvolution of 3D organoid dynamics at cellular resolution for cancer pharmacology with Cellos.

Three-dimensional (3D) organoid cultures are flexible systems to interrogate cellular growth, morphology, multicellular spatial architecture, and cellular interactions in response to treatment. However, computational methods for analysis of 3D organoids with sufficiently high-throughput and cellular resolution are needed. Here we report Cellos, an accurate, high-throughput pipeline for 3D organoid segmentation using classical algorithms and nuclear segmentation using a trained Stardist-3D convolutional neural network. To evaluate Cellos, we analyze ~100,000 organoids with ~2.35 million cells from multiple treatment experiments. Cellos segments dye-stained or fluorescently-labeled nuclei and accurately distinguishes distinct labeled cell populations within organoids. Cellos can recapitulate traditional luminescence-based drug response of cells with complex drug sensitivities, while also quantifying changes in organoid and nuclear morphologies caused by treatment as well as cell-cell spatial relationships that reflect ecological affinity. Cellos provides powerful tools to perform high-throughput analysis for pharmacological testing and biological investigation of organoids based on 3D imaging.

Cellos: High-throughput deconvolution of 3D organoid dynamics at cellular resolution for cancer pharmacology.

Three-dimensional (3D) culture models, such as organoids, are flexible systems to interrogate cellular growth and morphology, multicellular spatial architecture, and cell interactions in response to drug treatment. However, new computational methods to segment and analyze 3D models at cellular resolution with sufficiently high throughput are needed to realize these possibilities. Here we report Cellos (Cell and Organoid Segmentation), an accurate, high throughput image analysis pipeline for 3D organoid and nuclear segmentation analysis. Cellos segments organoids in 3D using classical algorithms and segments nuclei using a Stardist-3D convolutional neural network which we trained on a manually annotated dataset of 3,862 cells from 36 organoids confocally imaged at 5 µm z-resolution. To evaluate the capabilities of Cellos we then analyzed 74,450 organoids with 1.65 million cells, from multiple experiments on triple negative breast cancer organoids containing clonal mixtures with complex cisplatin sensitivities. Cellos was able to accurately distinguish ratios of distinct fluorescently labelled cell populations in organoids, with ≤3% deviation from the seeding ratios in each well and was effective for both fluorescently labelled nuclei and independent DAPI stained datasets. Cellos was able to recapitulate traditional luminescence-based drug response quantifications by analyzing 3D images, including parallel analysis of multiple cancer clones in the same well. Moreover, Cellos was able to identify organoid and nuclear morphology feature changes associated with treatment. Finally, Cellos enables 3D analysis of cell spatial relationships, which we used to detect ecological affinity between cancer cells beyond what arises from local cell division or organoid composition. Cellos provides powerful tools to perform high throughput analysis for pharmacological testing and biological investigation of organoids based on 3D imaging.

Computational estimation of quality and clinical relevance of cancer cell lines.

Immortal cancer cell lines (CCLs) are the most widely used system for investigating cancer biology and for the preclinical development of oncology therapies. Pharmacogenomic and genome-wide editing screenings have facilitated the discovery of clinically relevant gene-drug interactions and novel therapeutic targets via large panels of extensively characterised CCLs. However, tailoring pharmacological strategies in a precision medicine context requires bridging the existing gaps between tumours and in vitro models. Indeed, intrinsic limitations of CCLs such as misidentification, the absence of tumour microenvironment and genetic drift have highlighted the need to identify the most faithful CCLs for each primary tumour while addressing their heterogeneity, with the development of new models where necessary. Here, we discuss the most significant limitations of CCLs in representing patient features, and we review computational methods aiming at systematically evaluating the suitability of CCLs as tumour proxies and identifying the best patient representative in vitro models. Additionally, we provide an overview of the applications of these methods to more complex models and discuss future machine-learning-based directions that could resolve some of the arising discrepancies.

Bridging the gap between cancer cell line models and tumours using gene expression data.

Cancer cell line models are a cornerstone of cancer research, yet our understanding of how well they represent the molecular features of patient tumours remains limited. Our recent work provides a computational approach to systematically compare large gene expression datasets to better understand which cell lines most closely resemble each tumour type, as well as identify potential gaps in our current cancer models.

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images.

Histopathological images are a rich but incompletely explored data type for studying cancer. Manual inspection is time consuming, making it challenging to use for image data mining. Here we show that convolutional neural networks (CNNs) can be systematically applied across cancer types, enabling comparisons to reveal shared spatial behaviors. We develop CNN architectures to analyze 27,815 hematoxylin and eosin scanned images from The Cancer Genome Atlas for tumor/normal, cancer subtype, and mutation classification. Our CNNs are able to classify TCGA pathologist-annotated tumor/normal status of whole slide images (WSIs) in 19 cancer types with consistently high AUCs (0.995 ± 0.008), as well as subtypes with lower but significant accuracy (AUC 0.87 ± 0.1). Remarkably, tumor/normal CNNs trained on one tissue are effective in others (AUC 0.88 ± 0.11), with classifier relationships also recapitulating known adenocarcinoma, carcinoma, and developmental biology. Moreover, classifier comparisons reveal intra-slide spatial similarities, with an average tile-level correlation of 0.45 ± 0.16 between classifier pairs. Breast cancers, bladder cancers, and uterine cancers have spatial patterns that are particularly easy to detect, suggesting these cancers can be canonical types for image analysis. Patterns for TP53 mutations can also be detected, with WSI self- and cross-tissue AUCs ranging from 0.65-0.80. Finally, we comparatively evaluate CNNs on 170 breast and colon cancer images with pathologist-annotated nuclei, finding that both cellular and intercellular regions contribute to CNN accuracy. These results demonstrate the power of CNNs not only for histopathological classification, but also for cross-comparisons to reveal conserved spatial behaviors across tumors.

Repeat expansions confer WRN dependence in microsatellite-unstable cancers.

The RecQ DNA helicase WRN is a synthetic lethal target for cancer cells with microsatellite instability (MSI), a form of genetic hypermutability that arises from impaired mismatch repair1-4. Depletion of WRN induces widespread DNA double-strand breaks in MSI cells, leading to cell cycle arrest and/or apoptosis. However, the mechanism by which WRN protects MSI-associated cancers from double-strand breaks remains unclear. Here we show that TA-dinucleotide repeats are highly unstable in MSI cells and undergo large-scale expansions, distinct from previously described insertion or deletion mutations of a few nucleotides5. Expanded TA repeats form non-B DNA secondary structures that stall replication forks, activate the ATR checkpoint kinase, and require unwinding by the WRN helicase. In the absence of WRN, the expanded TA-dinucleotide repeats are susceptible to cleavage by the MUS81 nuclease, leading to massive chromosome shattering. These findings identify a distinct biomarker that underlies the synthetic lethal dependence on WRN, and support the development of therapeutic agents that target WRN for MSI-associated cancers.

Treating Cancer as an Invasive Species.

To cure a patient's cancer is to eradicate invasive cells from the ecosystem of the body. However, the ecologic complexity of this challenge is not well understood. Here we show how results from eradications of invasive mammalian species from islands-one of the few contexts in which invasive species have been regularly cleared-inform new research directions for treating cancer. We first summarize the epidemiologic characteristics of island invader eradications and cancer treatments by analyzing recent datasets from the Database of Invasive Island Species Eradications and The Cancer Genome Atlas, detailing the superior successes of island eradication projects. Next, we compare how genetic and environmental factors impact success in each system. These comparisons illuminate a number of promising cancer research and treatment directions, such as heterogeneity engineering as motivated by gene drives and adaptive therapy; multiscale analyses of how population heterogeneity potentiates treatment resistance; and application of ecological data mining techniques to high-throughput cancer data. We anticipate that interdisciplinary comparisons between tumor progression and invasive species would inspire development of novel paradigms to cure cancer.

CCNE1 amplification is associated with poor prognosis in patients with triple negative breast cancer.

BACKGROUND: Triple negative breast cancer (TNBC) is aggressive with limited treatment options upon recurrence. Molecular discordance between primary and metastatic TNBC has been observed, but the degree of biological heterogeneity has not been fully explored. Furthermore, genomic evolution through treatment is poorly understood. In this study, we aim to characterize the genomic changes between paired primary and metastatic TNBCs through transcriptomic and genomic profiling, and to identify genomic alterations which may contribute to chemotherapy resistance. METHODS: Genomic alterations and mRNA expression of 10 paired primary and metastatic TNBCs were determined through targeted sequencing, microarray analysis, and RNA sequencing. Commonly mutated genes, as well as differentially expressed and co-expressed genes were identified. We further explored the clinical relevance of differentially expressed genes between primary and metastatic tumors to patient survival using large public datasets. RESULTS: Through gene expression profiling, we observed a shift in TNBC subtype classifications between primary and metastatic TNBCs. A panel of eight cancer driver genes (CCNE1, TPX2, ELF3, FANCL, JAK2, GSK3B, CEP76, and SYK) were differentially expressed in recurrent TNBCs, and were also overexpressed in TCGA and METABRIC. CCNE1 and TPX2 were co-overexpressed in TNBCs. DNA mutation profiling showed that multiple mutations occurred in genes comprising a number of potentially targetable pathways including PI3K/AKT/mTOR, RAS/MAPK, cell cycle, and growth factor receptor signaling, reaffirming the wide heterogeneity of mechanisms driving TNBC. CCNE1 amplification was associated with poor overall survival in patients with metastatic TNBC. CONCLUSIONS: CCNE1 amplification may confer resistance to chemotherapy and is associated with poor overall survival in TNBC.

High-resolution deconstruction of evolution induced by chemotherapy treatments in breast cancer xenografts.

The processes by which tumors evolve are essential to the efficacy of treatment, but quantitative understanding of intratumoral dynamics has been limited. Although intratumoral heterogeneity is common, quantification of evolution is difficult from clinical samples because treatment replicates cannot be performed and because matched serial samples are infrequently available. To circumvent these problems we derived and assayed large sets of human triple-negative breast cancer xenografts and cell cultures from two patients, including 86 xenografts from cyclophosphamide, doxorubicin, cisplatin, docetaxel, or vehicle treatment cohorts as well as 45 related cell cultures. We assayed these samples via exome-seq and/or high-resolution droplet digital PCR, allowing us to distinguish complex therapy-induced selection and drift processes among endogenous cancer subclones with cellularity uncertainty <3%. For one patient, we discovered two predominant subclones that were granularly intermixed in all 48 co-derived xenograft samples. These two subclones exhibited differential chemotherapy sensitivity-when xenografts were treated with cisplatin for 3 weeks, the post-treatment volume change was proportional to the post-treatment ratio of subclones on a xenograft-to-xenograft basis. A subsequent cohort in which xenografts were treated with cisplatin, allowed a drug holiday, then treated a second time continued to exhibit this proportionality. In contrast, xenografts from other treatment cohorts, spatially dissected xenograft fragments, and cell cultures evolved in diverse ways but with substantial population bottlenecks. These results show that ecosystems susceptible to successive retreatment can arise spontaneously in breast cancer in spite of a background of irregular subclonal bottlenecks, and our work provides to our knowledge the first quantification of the population genetics of such a system. Intriguingly, in such an ecosystem the ratio of common subclones is predictive of the state of treatment susceptibility, showing how measurements of subclonal heterogeneity could guide treatment for some patients.

Distribution-based measures of tumor heterogeneity are sensitive to mutation calling and lack strong clinical predictive power.

Mutant allele frequency distributions in cancer samples have been used to estimate intratumoral heterogeneity and its implications for patient survival. However, mutation calls are sensitive to the calling algorithm. It remains unknown whether the relationship of heterogeneity and clinical outcome is robust to these variations. To resolve this question, we studied the robustness of allele frequency distributions to the mutation callers MuTect, SomaticSniper, and VarScan in 4722 cancer samples from The Cancer Genome Atlas. We observed discrepancies among the results, particularly a pronounced difference between allele frequency distributions called by VarScan and SomaticSniper. Survival analysis showed little robust predictive power for heterogeneity as measured by Mutant-Allele Tumor Heterogeneity (MATH) score, with the exception of uterine corpus endometrial carcinoma. However, we found that variations in mutant allele frequencies were mediated by variations in copy number. Our results indicate that the clinical predictions associated with MATH score are primarily caused by copy number aberrations that alter mutant allele frequencies. Finally, we present a mathematical model of linear tumor evolution demonstrating why MATH score is insufficient for distinguishing different scenarios of tumor growth. Our findings elucidate the importance of allele frequency distributions as a measure for tumor heterogeneity and their prognostic role.

Modeling oscillations and spiral waves in Dictyostelium populations.

Unicellular organisms exhibit elaborate collective behaviors in response to environmental cues. These behaviors are controlled by complex biochemical networks within individual cells and coordinated through cell-to-cell communication. Describing these behaviors requires new mathematical models that can bridge scales-from biochemical networks within individual cells to spatially structured cellular populations. Here we present a family of "multiscale" models for the emergence of spiral waves in the social amoeba Dictyostelium discoideum. Our models exploit new experimental advances that allow for the direct measurement and manipulation of the small signaling molecule cyclic adenosine monophosphate (cAMP) used by Dictyostelium cells to coordinate behavior in cellular populations. Inspired by recent experiments, we model the Dictyostelium signaling network as an excitable system coupled to various preprocessing modules. We use this family of models to study spatially unstructured populations of "fixed" cells by constructing phase diagrams that relate the properties of population-level oscillations to parameters in the underlying biochemical network. We then briefly discuss an extension of our model that includes spatial structure and show how this naturally gives rise to spiral waves. Our models exhibit a wide range of novel phenomena. including a density-dependent frequency change, bistability, and dynamic death due to slow cAMP dynamics. Our modeling approach provides a powerful tool for bridging scales in modeling of Dictyostelium populations.

From intracellular signaling to population oscillations: bridging size- and time-scales in collective behavior.

Collective behavior in cellular populations is coordinated by biochemical signaling networks within individual cells. Connecting the dynamics of these intracellular networks to the population phenomena they control poses a considerable challenge because of network complexity and our limited knowledge of kinetic parameters. However, from physical systems, we know that behavioral changes in the individual constituents of a collectively behaving system occur in a limited number of well-defined classes, and these can be described using simple models. Here, we apply such an approach to the emergence of collective oscillations in cellular populations of the social amoeba Dictyostelium discoideum. Through direct tests of our model with quantitative in vivo measurements of single-cell and population signaling dynamics, we show how a simple model can effectively describe a complex molecular signaling network at multiple size and temporal scales. The model predicts novel noise-driven single-cell and population-level signaling phenomena that we then experimentally observe. Our results suggest that like physical systems, collective behavior in biology may be universal and described using simple mathematical models.

Two-dimensionality of yeast colony expansion accompanied by pattern formation.

Yeasts can form multicellular patterns as they expand on agar plates, a phenotype that requires a functional copy of the FLO11 gene. Although the biochemical and molecular requirements for such patterns have been examined, the mechanisms underlying their formation are not entirely clear. Here we develop quantitative methods to accurately characterize the size, shape, and surface patterns of yeast colonies for various combinations of agar and sugar concentrations. We combine these measurements with mathematical and physical models and find that FLO11 gene constrains cells to grow near the agar surface, causing the formation of larger and more irregular colonies that undergo hierarchical wrinkling. Head-to-head competition assays on agar plates indicate that two-dimensional constraint on the expansion of FLO11 wild type (FLO11) cells confers a fitness advantage over FLO11 knockout (flo11Δ) cells on the agar surface.

Intrinsic noise of microRNA-regulated genes and the ceRNA hypothesis.

MicroRNAs are small noncoding RNAs that regulate genes post-transciptionally by binding and degrading target eukaryotic mRNAs. We use a quantitative model to study gene regulation by inhibitory microRNAs and compare it to gene regulation by prokaryotic small non-coding RNAs (sRNAs). Our model uses a combination of analytic techniques as well as computational simulations to calculate the mean-expression and noise profiles of genes regulated by both microRNAs and sRNAs. We find that despite very different molecular machinery and modes of action (catalytic vs stoichiometric), the mean expression levels and noise profiles of microRNA-regulated genes are almost identical to genes regulated by prokaryotic sRNAs. This behavior is extremely robust and persists across a wide range of biologically relevant parameters. We extend our model to study crosstalk between multiple mRNAs that are regulated by a single microRNA and show that noise is a sensitive measure of microRNA-mediated interaction between mRNAs. We conclude by discussing possible experimental strategies for uncovering the microRNA-mRNA interactions and testing the competing endogenous RNA (ceRNA) hypothesis.