Deep neural networks identify signaling mechanisms of ErbB-family drug resistance from a continuous cell morphology space.

It is well known that the development of drug resistance in cancer cells can lead to changes in cell morphology. Here, we describe the use of deep neural networks to analyze this relationship, demonstrating that complex cell morphologies can encode states of signaling networks and unravel cellular mechanisms hidden to conventional approaches. We perform high-content screening of 17 cancer cell lines, generating more than 500 billion data points from ∼850 million cells. We analyze these data using a deep learning model, resulting in the identification of a continuous 27-dimension space describing all of the observed cell morphologies. From its morphology alone, we could thus predict whether a cell was resistant to ErbB-family drugs, with an accuracy of 74%, and predict the potential mechanism of resistance, subsequently validating the role of MET and insulin-like growth factor 1 receptor (IGF1R) as drivers of cetuximab resistance in in vitro models of lung and head/neck cancer.

NODAL/TGFß signalling mediates the self-sustained stemness induced by PIK3CAH1047R homozygosity in pluripotent stem cells.

Activating PIK3CA mutations are known 'drivers' of human cancer and developmental overgrowth syndromes. We recently demonstrated that the 'hotspot' PIK3CAH1047R variant exerts unexpected allele dose-dependent effects on stemness in human pluripotent stem cells (hPSCs). In this study, we combine high-depth transcriptomics, total proteomics and reverse-phase protein arrays to reveal potentially disease-related alterations in heterozygous cells, and to assess the contribution of activated TGFß signalling to the stemness phenotype of homozygous PIK3CAH1047R cells. We demonstrate signalling rewiring as a function of oncogenic PI3K signalling strength, and provide experimental evidence that self-sustained stemness is causally related to enhanced autocrine NODAL/TGFß signalling. A significant transcriptomic signature of TGFß pathway activation in heterozygous PIK3CAH1047R was observed but was modest and was not associated with the stemness phenotype seen in homozygous mutants. Notably, the stemness gene expression in homozygous PIK3CAH1047R hPSCs was reversed by pharmacological inhibition of NODAL/TGFß signalling, but not by pharmacological PI3Kα pathway inhibition. Altogether, this provides the first in-depth analysis of PI3K signalling in hPSCs and directly links strong PI3K activation to developmental NODAL/TGFß signalling. This work illustrates the importance of allele dosage and expression when artificial systems are used to model human genetic disease caused by activating PIK3CA mutations. This article has an associated First Person interview with the first author of the paper.

Deep Hidden Physics Modeling of Cell Signaling Networks.

According to the WHO, cancer is the second most common cause of death worldwide. The social and economic damage caused by cancer is high and rising. In Europe, the annual direct medical expenses alone amount to more than €129 billion. This results in an urgent need for new and sustainable therapeutics, which has currently not been met by the pharmaceutical industry; only 3.4% of cancer drugs entering Phase I clinical trials get to market. Phosphorylation sites are parts of the core machinery of kinase signaling networks, which are known to be dysfunctional in all types of cancer. Indeed, kinases are the second most common drug target yet. However, these inhibitors block all functions of a protein, and they commonly lead to the development of resistance and increased toxicity. To facilitate global and mechanistic modeling of cancer and clinically relevant cell signaling networks, the community will have to develop sophisticated data-driven deep-learning and mechanistic computational models that generate in silico probabilistic predictions of molecular signaling network rearrangements causally implicated in cancer.

Global view of the RAF-MEK-ERK module and its immediate downstream effectors.

Small molecule inhibitors of BRAF and MEK have proven effective at inhibiting tumor growth in melanoma patients, however this efficacy is limited due to the almost universal development of drug resistance. To provide advanced insight into the signaling responses that occur following kinase inhibition we have performed quantitative (phospho)-proteomics of human melanoma cells treated with either dabrafenib, a BRAF inhibitor; trametinib, a MEK inhibitor or SCH772984, an ERK inhibitor. Over nine experiments we identified 7827 class I phosphorylation sites on 4960 proteins. This included 54 phosphorylation sites that were significantly down-modulated after exposure to all three inhibitors, 34 of which have not been previously reported. Functional analysis of these novel ERK targets identified roles for them in GTPase activity and regulation, apoptosis and cell-cell adhesion. Comparison of the results presented here with previously reported phosphorylation sites downstream of ERK showed a limited degree of overlap suggesting that ERK signaling responses may be highly cell line and cue specific. In addition we identified 26 phosphorylation sites that were only responsive to dabrafenib. We provide further orthogonal experimental evidence for 3 of these sites in human embryonic kidney cells over-expressing BRAF as well as further computational insights using KinomeXplorer. The validated phosphorylation sites were found to be involved in actin regulation, which has been proposed as a novel mechanism for inhibiting resistance development. These results would suggest that the linearity of the BRAF-MEK-ERK module is at least context dependent.

Comprehensive substrate specificity profiling of the human Nek kinome reveals unexpected signaling outputs.

Human NimA-related kinases (Neks) have multiple mitotic and non-mitotic functions, but few substrates are known. We systematically determined the phosphorylation-site motifs for the entire Nek kinase family, except for Nek11. While all Nek kinases strongly select for hydrophobic residues in the -3 position, the family separates into four distinct groups based on specificity for a serine versus threonine phospho-acceptor, and preference for basic or acidic residues in other positions. Unlike Nek1-Nek9, Nek10 is a dual-specificity kinase that efficiently phosphorylates itself and peptide substrates on serine and tyrosine, and its activity is enhanced by tyrosine auto-phosphorylation. Nek10 dual-specificity depends on residues in the HRD+2 and APE-4 positions that are uncommon in either serine/threonine or tyrosine kinases. Finally, we show that the phosphorylation-site motifs for the mitotic kinases Nek6, Nek7 and Nek9 are essentially identical to that of their upstream activator Plk1, suggesting that Nek6/7/9 function as phospho-motif amplifiers of Plk1 signaling.

Bowhead: Bayesian modelling of cell velocity during concerted cell migration.

Cell migration is a central biological process that requires fine coordination of molecular events in time and space. A deregulation of the migratory phenotype is also associated with pathological conditions including cancer where cell motility has a causal role in tumor spreading and metastasis formation. Thus cell migration is of critical and strategic importance across the complex disease spectrum as well as for the basic understanding of cell phenotype. Experimental studies of the migration of cells in monolayers are often conducted with 'wound healing' assays. Analysis of these assays has traditionally relied on how the wound area changes over time. However this method does not take into account the shape of the wound. Given the many options for creating a wound healing assay and the fact that wound shape invariably changes as cells migrate this is a significant flaw. Here we present a novel software package for analyzing concerted cell velocity in wound healing assays. Our method encompasses a wound detection algorithm based on cell confluency thresholding and employs a Bayesian approach in order to estimate concerted cell velocity with an associated likelihood. We have applied this method to study the effect of siRNA knockdown on the migration of a breast cancer cell line and demonstrate that cell velocity can track wound healing independently of wound shape and provides a more robust quantification with significantly higher signal to noise ratios than conventional analyses of wound area. The software presented here will enable other researchers in any field of cell biology to quantitatively analyze and track live cell migratory processes and is therefore expected to have a significant impact on the study of cell migration, including cancer relevant processes. Installation instructions, documentation and source code can be found at http://bowhead.lindinglab.science licensed under GPLv3.

Dynamic Rearrangement of Cell States Detected by Systematic Screening of Sequential Anticancer Treatments.

Signaling networks are nonlinear and complex, involving a large ensemble of dynamic interaction states that fluctuate in space and time. However, therapeutic strategies, such as combination chemotherapy, rarely consider the timing of drug perturbations. If we are to advance drug discovery for complex diseases, it will be essential to develop methods capable of identifying dynamic cellular responses to clinically relevant perturbations. Here, we present a Bayesian dose-response framework and the screening of an oncological drug matrix, comprising 10,000 drug combinations in melanoma and pancreatic cancer cell lines, from which we predict sequentially effective drug combinations. Approximately 23% of the tested combinations showed high-confidence sequential effects (either synergistic or antagonistic), demonstrating that cellular perturbations of many drug combinations have temporal aspects, which are currently both underutilized and poorly understood.

Dataset for the proteomic inventory and quantitative analysis of the breast cancer hypoxic secretome associated with osteotropism.

The cancer secretome includes all of the macromolecules secreted by cells into their microenvironment. Cancer cell secretomes are significantly different to that of normal cells reflecting the changes that normal cells have undergone during their transition to malignancy. More importantly, cancer secretomes are known to be active mediators of both local and distant host cells and play an important role in the progression and dissemination of cancer. Here we have quantitatively profiled both the composition of breast cancer secretomes associated with osteotropism, and their modulation under normoxic and hypoxic conditions. We detect and quantify 162 secretome proteins across all conditions which show differential hypoxic induction and association with osteotropism. Mass Spectrometry proteomics data have been deposited to the ProteomeXchange Consortium with the dataset identifier PXD000397 and the complete proteomic, bioinformatic and biological analyses are reported in Cox et al. (2015) [1].

Kinome-wide decoding of network-attacking mutations rewiring cancer signaling.

Cancer cells acquire pathological phenotypes through accumulation of mutations that perturb signaling networks. However, global analysis of these events is currently limited. Here, we identify six types of network-attacking mutations (NAMs), including changes in kinase and SH2 modulation, network rewiring, and the genesis and extinction of phosphorylation sites. We developed a computational platform (ReKINect) to identify NAMs and systematically interpreted the exomes and quantitative (phospho-)proteomes of five ovarian cancer cell lines and the global cancer genome repository. We identified and experimentally validated several NAMs, including PKCγ M501I and PKD1 D665N, which encode specificity switches analogous to the appearance of kinases de novo within the kinome. We discover mutant molecular logic gates, a drift toward phospho-threonine signaling, weakening of phosphorylation motifs, and kinase-inactivating hotspots in cancer. Our method pinpoints functional NAMs, scales with the complexity of cancer genomes and cell signaling, and may enhance our capability to therapeutically target tumor-specific networks.

Unmasking determinants of specificity in the human kinome.

Protein kinases control cellular responses to environmental cues by swift and accurate signal processing. Breakdowns in this high-fidelity capability are a driving force in cancer and other diseases. Thus, our limited understanding of which amino acids in the kinase domain encode substrate specificity, the so-called determinants of specificity (DoS), constitutes a major obstacle in cancer signaling. Here, we systematically discover several DoS and experimentally validate three of them, named the αC1, αC3, and APE-7 residues. We demonstrate that DoS form sparse networks of non-conserved residues spanning distant regions. Our results reveal a likely role for inter-residue allostery in specificity and an evolutionary decoupling of kinase activity and specificity, which appear loaded on independent groups of residues. Finally, we uncover similar properties driving SH2 domain specificity and demonstrate how the identification of DoS can be utilized to elucidate a greater understanding of the role of signaling networks in cancer (Creixell et al., 2015 [this issue of Cell]).

Pathway and network analysis of cancer genomes.

Genomic information on tumors from 50 cancer types cataloged by the International Cancer Genome Consortium (ICGC) shows that only a few well-studied driver genes are frequently mutated, in contrast to many infrequently mutated genes that may also contribute to tumor biology. Hence there has been large interest in developing pathway and network analysis methods that group genes and illuminate the processes involved. We provide an overview of these analysis techniques and show where they guide mechanistic and translational investigations.

The hypoxic cancer secretome induces pre-metastatic bone lesions through lysyl oxidase.

Tumour metastasis is a complex process involving reciprocal interplay between cancer cells and host stroma at both primary and secondary sites, and is strongly influenced by microenvironmental factors such as hypoxia. Tumour-secreted proteins play a crucial role in these interactions and present strategic therapeutic potential. Metastasis of breast cancer to the bone affects approximately 85% of patients with advanced disease and renders them largely untreatable. Specifically, osteolytic bone lesions, where bone is destroyed, lead to debilitating skeletal complications and increased patient morbidity and mortality. The molecular interactions governing the early events of osteolytic lesion formation are currently unclear. Here we show hypoxia to be specifically associated with bone relapse in patients with oestrogen-receptor negative breast cancer. Global quantitative analysis of the hypoxic secretome identified lysyl oxidase (LOX) as significantly associated with bone-tropism and relapse. High expression of LOX in primary breast tumours or systemic delivery of LOX leads to osteolytic lesion formation whereas silencing or inhibition of LOX activity abrogates tumour-driven osteolytic lesion formation. We identify LOX as a novel regulator of NFATc1-driven osteoclastogenesis, independent of RANK ligand, which disrupts normal bone homeostasis leading to the formation of focal pre-metastatic lesions. We show that these lesions subsequently provide a platform for circulating tumour cells to colonize and form bone metastases. Our study identifies a novel mechanism of regulation of bone homeostasis and metastasis, opening up opportunities for novel therapeutic intervention with important clinical implications.

Integrative analysis of kinase networks in TRAIL-induced apoptosis provides a source of potential targets for combination therapy.

Tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) is an endogenous secreted peptide and, in preclinical studies, preferentially induces apoptosis in tumor cells rather than in normal cells. The acquisition of resistance in cells exposed to TRAIL or its mimics limits their clinical efficacy. Because kinases are intimately involved in the regulation of apoptosis, we systematically characterized kinases involved in TRAIL signaling. Using RNA interference (RNAi) loss-of-function and cDNA overexpression screens, we identified 169 protein kinases that influenced the dynamics of TRAIL-induced apoptosis in the colon adenocarcinoma cell line DLD-1. We classified the kinases as sensitizers or resistors or modulators, depending on the effect that knockdown and overexpression had on TRAIL-induced apoptosis. Two of these kinases that were classified as resistors were PX domain-containing serine/threonine kinase (PXK) and AP2-associated kinase 1 (AAK1), which promote receptor endocytosis and may enable cells to resist TRAIL-induced apoptosis by enhancing endocytosis of the TRAIL receptors. We assembled protein interaction maps using mass spectrometry-based protein interaction analysis and quantitative phosphoproteomics. With these protein interaction maps, we modeled information flow through the networks and identified apoptosis-modifying kinases that are highly connected to regulated substrates downstream of TRAIL. The results of this analysis provide a resource of potential targets for the development of TRAIL combination therapies to selectively kill cancer cells.

Modulation of the chromatin phosphoproteome by the Haspin protein kinase.

Recent discoveries have highlighted the importance of Haspin kinase activity for the correct positioning of the kinase Aurora B at the centromere. Haspin phosphorylates Thr(3) of the histone H3 (H3), which provides a signal for Aurora B to localize to the centromere of mitotic chromosomes. To date, histone H3 is the only confirmed Haspin substrate. We used a combination of biochemical, pharmacological, and mass spectrometric approaches to study the consequences of Haspin inhibition in mitotic cells. We quantified 3964 phosphorylation sites on chromatin-associated proteins and identified a Haspin protein-protein interaction network. We determined the Haspin consensus motif and the co-crystal structure of the kinase with the histone H3 tail. The structure revealed a unique bent substrate binding mode positioning the histone H3 residues Arg(2) and Lys(4) adjacent to the Haspin phosphorylated threonine into acidic binding pockets. This unique conformation of the kinase-substrate complex explains the reported modulation of Haspin activity by methylation of Lys(4) of the histone H3. In addition, the identification of the structural basis of substrate recognition and the amino acid sequence preferences of Haspin aided the identification of novel candidate Haspin substrates. In particular, we validated the phosphorylation of Ser(137) of the histone variant macroH2A as a target of Haspin kinase activity. MacroH2A Ser(137) resides in a basic stretch of about 40 amino acids that is required to stabilize extranucleosomal DNA, suggesting that phosphorylation of Ser(137) might regulate the interactions of macroH2A and DNA. Overall, our data suggest that Haspin activity affects the phosphorylation state of proteins involved in gene expression regulation and splicing.

Identification of hypoxia-regulated proteins using MALDI-mass spectrometry imaging combined with quantitative proteomics.

Hypoxia is present in most solid tumors and is clinically correlated with increased metastasis and poor patient survival. While studies have demonstrated the role of hypoxia and hypoxia-regulated proteins in cancer progression, no attempts have been made to identify hypoxia-regulated proteins using quantitative proteomics combined with MALDI-mass spectrometry imaging (MALDI-MSI). Here we present a comprehensive hypoxic proteome study and are the first to investigate changes in situ using tumor samples. In vitro quantitative mass spectrometry analysis of the hypoxic proteome was performed on breast cancer cells using stable isotope labeling with amino acids in cell culture (SILAC). MS analyses were performed on laser-capture microdissected samples isolated from normoxic and hypoxic regions from tumors derived from the same cells used in vitro. MALDI-MSI was used in combination to investigate hypoxia-regulated protein localization within tumor sections. Here we identified more than 100 proteins, both novel and previously reported, that were associated with hypoxia. Several proteins were localized in hypoxic regions, as identified by MALDI-MSI. Visualization and data extrapolation methods for the in vitro SILAC data were also developed, and computational mapping of MALDI-MSI data to IHC results was applied for data validation. The results and limitations of the methodologies described are discussed.

Computational approaches to identify functional genetic variants in cancer genomes.

The International Cancer Genome Consortium (ICGC) aims to catalog genomic abnormalities in tumors from 50 different cancer types. Genome sequencing reveals hundreds to thousands of somatic mutations in each tumor but only a minority of these drive tumor progression. We present the result of discussions within the ICGC on how to address the challenge of identifying mutations that contribute to oncogenesis, tumor maintenance or response to therapy, and recommend computational techniques to annotate somatic variants and predict their impact on cancer phenotype.

Global characterization of signalling networks associated with tamoxifen resistance in breast cancer.

Acquired resistance to the anti-estrogen tamoxifen remains a significant challenge in breast cancer management. In this study, we used an integrative approach to characterize global protein expression and tyrosine phosphorylation events in tamoxifen-resistant MCF7 breast cancer cells (TamR) compared with parental controls. Quantitative mass spectrometry and computational approaches were combined to identify perturbed signalling networks, and candidate regulatory proteins were functionally interrogated by siRNA-mediated knockdown. Network analysis revealed that cellular metabolism was perturbed in TamR cells, together with pathways enriched for proteins associated with growth factor, cell-cell and cell matrix-initiated signalling. Consistent with known roles for Ras/MAPK and PI3-kinase signalling in tamoxifen resistance, tyrosine-phosphorylated MAPK1, SHC1 and PIK3R2 were elevated in TamR cells. Phosphorylation of the tyrosine kinase Yes and expression of the actin-binding protein myristoylated alanine-rich C-kinase substrate (MARCKS) were increased two- and eightfold in TamR cells respectively, and these proteins were selected for further analysis. Knockdown of either protein in TamR cells had no effect on anti-estrogen sensitivity, but significantly decreased cell motility. MARCKS expression was significantly higher in breast cancer cell lines than normal mammary epithelial cells and in ER-negative versus ER-positive breast cancer cell lines. In primary breast cancers, cytoplasmic MARCKS staining was significantly higher in basal-like and HER2 cancers than in luminal cancers, and was independently predictive of poor survival in multivariate analyses of the whole cohort (P < 0.0001) and in ER-positive patients (P = 0.0005). These findings provide network-level insights into the molecular alterations associated with the tamoxifen-resistant phenotype, and identify MARCKS as a potential biomarker of therapeutic responsiveness that may assist in stratification of patients for optimal therapy.

In vivo SILAC-based proteomics reveals phosphoproteome changes during mouse skin carcinogenesis.

Cancer progresses through distinct stages, and mouse models recapitulating traits of this progression are frequently used to explore genetic, morphological, and pharmacological aspects of tumor development. To complement genomic investigations of this process, we here quantify phosphoproteomic changes in skin cancer development using the SILAC mouse technology coupled to high-resolution mass spectrometry. We distill protein expression signatures from our data that distinguish between skin cancer stages. A distinct phosphoproteome of the two stages of cancer progression is identified that correlates with perturbed cell growth and implicates cell adhesion as a major driver of malignancy. Importantly, integrated analysis of phosphoproteomic data and prediction of kinase activity revealed PAK4-PKC/SRC network to be highly deregulated in SCC but not in papilloma. This detailed molecular picture, both at the proteome and phosphoproteome level, will prove useful for the study of mechanisms of tumor progression.