A systematic pan-cancer study on deep learning-based prediction of multi-omic biomarkers from routine pathology images.

BACKGROUND: The objective of this comprehensive pan-cancer study is to evaluate the potential of deep learning (DL) for molecular profiling of multi-omic biomarkers directly from hematoxylin and eosin (H&E)-stained whole slide images. METHODS: A total of 12,093 DL models predicting 4031 multi-omic biomarkers across 32 cancer types were trained and validated. The study included a broad range of genetic, transcriptomic, and proteomic biomarkers, as well as established prognostic markers, molecular subtypes, and clinical outcomes. RESULTS: Here we show that 50% of the models achieve an area under the curve (AUC) of 0.644 or higher. The observed AUC for 25% of the models is at least 0.719 and exceeds 0.834 for the top 5%. Molecular profiling with image-based histomorphological features is generally considered feasible for most of the investigated biomarkers and across different cancer types. The performance appears to be independent of tumor purity, sample size, and class ratio (prevalence), suggesting a degree of inherent predictability in histomorphology. CONCLUSIONS: The results demonstrate that DL holds promise to predict a wide range of biomarkers across the omics spectrum using only H&E-stained histological slides of solid tumors. This paves the way for accelerating diagnosis and developing more precise treatments for cancer patients.

Molecular profiling tests are used to check cancers for changes in certain genes, proteins, or other molecules. Results of such tests can be used to identify the most effective treatment for cancer patients. Faster and more accessible alternatives to standard tests are needed to improve cancer care. This study investigates whether deep learning (DL), a series of advanced computer techniques, can perform molecular profiling directly from routinely-collected images of tumor specimens used for diagnostic purposes. Over 12,000 DL models were utilized to evaluate thousands of biomarkers using statistical approaches. The results indicate that DL can effectively detect molecular changes in a tumor from these images, for many biomarkers and tumor types. The study shows that DL-based molecular profiling from images is possible. Introducing this type of approach into routine clinical workflows could potentially accelerate treatment decisions and improve outcomes.

AMPK is a mechano-metabolic sensor linking cell adhesion and mitochondrial dynamics to Myosin-dependent cell migration.

Cell migration is crucial for cancer dissemination. We find that AMP-activated protein kinase (AMPK) controls cell migration by acting as an adhesion sensing molecular hub. In 3-dimensional matrices, fast-migrating amoeboid cancer cells exert low adhesion/low traction linked to low ATP/AMP, leading to AMPK activation. In turn, AMPK plays a dual role controlling mitochondrial dynamics and cytoskeletal remodelling. High AMPK activity in low adhering migratory cells, induces mitochondrial fission, resulting in lower oxidative phosphorylation and lower mitochondrial ATP. Concurrently, AMPK inactivates Myosin Phosphatase, increasing Myosin II-dependent amoeboid migration. Reducing adhesion or mitochondrial fusion or activating AMPK induces efficient rounded-amoeboid migration. AMPK inhibition suppresses metastatic potential of amoeboid cancer cells in vivo, while a mitochondrial/AMPK-driven switch is observed in regions of human tumours where amoeboid cells are disseminating. We unveil how mitochondrial dynamics control cell migration and suggest that AMPK is a mechano-metabolic sensor linking energetics and the cytoskeleton.

Myosin II Reactivation and Cytoskeletal Remodeling as a Hallmark and a Vulnerability in Melanoma Therapy Resistance.

Despite substantial clinical benefit of targeted and immune checkpoint blockade-based therapies in melanoma, resistance inevitably develops. We show cytoskeletal remodeling and changes in expression and activity of ROCK-myosin II pathway during acquisition of resistance to MAPK inhibitors. MAPK regulates myosin II activity, but after initial therapy response, drug-resistant clones restore myosin II activity to increase survival. High ROCK-myosin II activity correlates with aggressiveness, identifying targeted therapy- and immunotherapy-resistant melanomas. Survival of resistant cells is myosin II dependent, regardless of the therapy. ROCK-myosin II ablation specifically kills resistant cells via intrinsic lethal reactive oxygen species and unresolved DNA damage and limits extrinsic myeloid and lymphoid immunosuppression. Efficacy of targeted therapies and immunotherapies can be improved by combination with ROCK inhibitors.

Regional Activation of Myosin II in Cancer Cells Drives Tumor Progression via a Secretory Cross-Talk with the Immune Microenvironment.

ROCK-Myosin II drives fast rounded-amoeboid migration in cancer cells during metastatic dissemination. Analysis of human melanoma biopsies revealed that amoeboid melanoma cells with high Myosin II activity are predominant in the invasive fronts of primary tumors in proximity to CD206+CD163+ tumor-associated macrophages and vessels. Proteomic analysis shows that ROCK-Myosin II activity in amoeboid cancer cells controls an immunomodulatory secretome, enabling the recruitment of monocytes and their differentiation into tumor-promoting macrophages. Both amoeboid cancer cells and their associated macrophages support an abnormal vasculature, which ultimately facilitates tumor progression. Mechanistically, amoeboid cancer cells perpetuate their behavior via ROCK-Myosin II-driven IL-1α secretion and NF-κB activation. Using an array of tumor models, we show that high Myosin II activity in tumor cells reprograms the innate immune microenvironment to support tumor growth. We describe an unexpected role for Myosin II dynamics in cancer cells controlling myeloid function via secreted factors.

Actomyosin contractility and collective migration: may the force be with you.

Collective cell migration is essential during physiological processes such as development or wound healing and in pathological conditions such as cancer dissemination. Cells migrating within multicellular tissues experiment different forces which play an intricate role during tissue formation, development and maintenance. How cells are able to respond to these forces depends largely on how they interact with the extracellular matrix. In this review, we focus on mechanics and mechanotransduction in collective migration. In particular, we discuss current knowledge on how cells integrate mechanical signals during collective migration and we highlight actomyosin contractility as a central hub coordinating mechanosensing and mechanotransduction responses.

Erratum: The metabolic co-regulator PGC1α suppresses prostate cancer metastasis.

This corrects the article DOI: 10.1038/ncb3357.

Modes of invasion during tumour dissemination.

Cancer cell migration and invasion underlie metastatic dissemination, one of the major problems in cancer. Tumour cells exhibit a striking variety of invasion strategies. Importantly, cancer cells can switch between invasion modes in order to cope with challenging environments. This ability to switch migratory modes or plasticity highlights the challenges behind antimetastasis therapy design. In this Review, we present current knowledge on different tumour invasion strategies, the determinants controlling plasticity and arising therapeutic opportunities. We propose that targeting master regulators controlling plasticity is needed to hinder tumour dissemination and metastasis.

The metabolic co-regulator PGC1α suppresses prostate cancer metastasis.

Cellular transformation and cancer progression is accompanied by changes in the metabolic landscape. Master co-regulators of metabolism orchestrate the modulation of multiple metabolic pathways through transcriptional programs, and hence constitute a probabilistically parsimonious mechanism for general metabolic rewiring. Here we show that the transcriptional co-activator peroxisome proliferator-activated receptor gamma co-activator 1α (PGC1α) suppresses prostate cancer progression and metastasis. A metabolic co-regulator data mining analysis unveiled that PGC1α is downregulated in prostate cancer and associated with disease progression. Using genetically engineered mouse models and xenografts, we demonstrated that PGC1α opposes prostate cancer progression and metastasis. Mechanistically, the use of integrative metabolomics and transcriptomics revealed that PGC1α activates an oestrogen-related receptor alpha (ERRα)-dependent transcriptional program to elicit a catabolic state and metastasis suppression. Importantly, a signature based on the PGC1α-ERRα pathway exhibited prognostic potential in prostate cancer, thus uncovering the relevance of monitoring and manipulating this pathway for prostate cancer stratification and treatment.

Reactivation of p53 by a Cytoskeletal Sensor to Control the Balance Between DNA Damage and Tumor Dissemination.

BACKGROUND: Abnormal cell migration and invasion underlie metastasis, and actomyosin contractility is a key regulator of tumor invasion. The links between cancer migratory behavior and DNA damage are poorly understood. METHODS: Using 3D collagen systems to recapitulate melanoma extracellular matrix, we analyzed the relationship between the actomyosin cytoskeleton of migrating cells and DNA damage. We used multiple melanoma cell lines and microarray analysis to study changes in gene expression and in vivo intravital imaging (n = 7 mice per condition) to understand how DNA damage impacts invasive behavior. We used Protein Tissue Microarrays (n = 164 melanomas) and patient databases (n = 354 melanoma samples) to investigate the associations between markers of DNA damage and actomyosin cytoskeletal features. Data were analyzed with Student's and multiple t tests, Mann-Whitney's test, one-way analysis of variance, and Pearson correlation. All statistical tests were two-sided. RESULTS: Melanoma cells with low levels of Rho-ROCK-driven actomyosin are subjected to oxidative stress-dependent DNA damage and ATM-mediated p53 protein stabilization. This results in a specific transcriptional signature enriched in DNA damage/oxidative stress responsive genes, including Tumor Protein p53 Inducible Protein 3 (TP53I3 or PIG3). PIG3, which functions in DNA damage repair, uses an unexpected catalytic mechanism to suppress Rho-ROCK activity and impair tumor invasion in vivo. This regulation was suppressed by antioxidants. Furthermore, PIG3 levels decreased while ROCK1/2 levels increased in human metastatic melanomas (ROCK1 vs PIG3; r = -0.2261, P < .0001; ROCK2 vs PIG3: r = -0.1381, P = .0093). CONCLUSIONS: The results suggest using Rho-kinase inhibitors to reactivate the p53-PIG3 axis as a novel therapeutic strategy; we suggest that the use of antioxidants in melanoma should be very carefully evaluated.

Diverse matrix metalloproteinase functions regulate cancer amoeboid migration.

Rounded-amoeboid cancer cells use actomyosin contractility driven by Rho-ROCK and JAK-STAT3 to migrate efficiently. It has been suggested that rounded-amoeboid cancer cells do not require matrix metalloproteinases (MMPs) to invade. Here we compare MMP levels in rounded-amoeboid and elongated-mesenchymal melanoma cells. Surprisingly, we find that rounded-amoeboid melanoma cells secrete higher levels of several MMPs, including collagenase MMP-13 and gelatinase MMP-9. As a result, rounded-amoeboid melanoma cells degrade collagen I more efficiently than elongated-mesenchymal cells. Furthermore, using a non-catalytic mechanism, MMP-9 promotes rounded-amoeboid 3D migration through regulation of actomyosin contractility via CD44 receptor. MMP-9 is upregulated in a panel of rounded-amoeboid compared with elongated-mesenchymal melanoma cell lines and its levels are controlled by ROCK-JAK-STAT3 signalling. MMP-9 expression increases during melanoma progression and it is particularly prominent in the invasive fronts of lesions, correlating with cell roundness. Therefore, rounded-amoeboid cells use both catalytic and non-catalytic activities of MMPs for invasion.

NMR metabolomics of MTLn3E breast cancer cells identifies a role for CXCR4 in lipid and choline regulation.

The alpha chemokine receptor CXCR4 is up-regulated in certain types of breast cancer. Truncation of the C-terminus of this receptor alters cell morphology and increases invasiveness and metastatic potential. Here, to better understand the effects of CXCR4 expression and truncation in breast cancer cells, we have used high resolution magic angle spinning (HR-MAS) NMR studies of rat breast carcinoma MtLn3E cells to characterize the metabolite complement of cells heterologously expressing human CXCR4 or its C-terminal truncation mutant, Δ34-CXCR4. Notable reductions in choline levels were detected when either cells expressing wild-type CXCR4 or Δ34-CXCR4 were compared with cells containing an empty expression vector. Cells expressing CXCR4-Δ34 had reduced lipid content when compared with either the wild-type CXCR4 expressing cells or those containing the empty expression vector. Taken together, our results show that distinct effects on the metabolite complement can be linked to either CXCR4 expression or CXCR4 regulation. The metabolite markers for these two effects identified in the present study can, in turn, be used to further investigate the role of CXCR4 in metastasis.