Diffusion Tensor Imaging for Characterizing Changes in Triple-Negative Breast Cancer During Neoadjuvant Systemic Therapy.

BACKGROUND: Assessment of treatment response in triple-negative breast cancer (TNBC) may guide individualized care for improved patient outcomes. Diffusion tensor imaging (DTI) measures tissue anisotropy and could be useful for characterizing changes in the tumors and adjacent fibroglandular tissue (FGT) of TNBC patients undergoing neoadjuvant systemic treatment (NAST). PURPOSE: To evaluate the potential of DTI parameters for prediction of treatment response in TNBC patients undergoing NAST. STUDY TYPE: Prospective. POPULATION: Eighty-six women (average age: 51 ± 11 years) with biopsy-proven clinical stage I-III TNBC who underwent NAST followed by definitive surgery. 47% of patients (40/86) had pathologic complete response (pCR). FIELD STRENGTH/SEQUENCE: 3.0 T/reduced field of view single-shot echo-planar DTI sequence. ASSESSMENT: Three MRI scans were acquired longitudinally (pre-treatment, after 2 cycles of NAST, and after 4 cycles of NAST). Eleven histogram features were extracted from DTI parameter maps of tumors, a peritumoral region (PTR), and FGT in the ipsilateral breast. DTI parameters included apparent diffusion coefficients and relative diffusion anisotropies. pCR status was determined at surgery. STATISTICAL TESTS: Longitudinal changes of DTI features were tested for discrimination of pCR using Mann-Whitney U test and area under the receiver operating characteristic curve (AUC). A P value <0.05 was considered statistically significant. RESULTS: 47% of patients (40/86) had pCR. DTI parameters assessed after 2 and 4 cycles of NAST were significantly different between pCR and non-pCR patients when compared between tumors, PTRs, and FGTs. The median surface/average anisotropy of the PTR, measured after 2 and 4 cycles of NAST, increased in pCR patients and decreased in non-pCR patients (AUC: 0.78; 0.027 ± 0.043 vs. -0.017 ± 0.042 mm2 /s). DATA CONCLUSION: Quantitative DTI features from breast tumors and the peritumoral tissue may be useful for predicting the response to NAST in TNBC. EVIDENCE LEVEL: 1 TECHNICAL EFFICACY: Stage 4.

The immunogenomic landscape of resected intrahepatic cholangiocarcinoma.

BACKGROUND AND AIMS: Cholangiocarcinoma (CCA) is a deadly and highly therapy-refractory cancer of the bile ducts, with early results from immune checkpoint blockade trials showing limited responses. Whereas recent molecular assessments have made bulk characterizations of immune profiles and their genomic correlates, spatial assessments may reveal actionable insights. APPROACH AND RESULTS: Here, we have integrated immune checkpoint-directed immunohistochemistry with next-generation sequencing of resected intrahepatic CCA samples from 96 patients. We found that both T-cell and immune checkpoint markers are enriched at the tumor margins compared to the tumor center. Using two approaches, we identify high programmed cell death protein 1 or lymphocyte-activation gene 3 and low CD3/CD4/inducible T-cell costimulator specifically in the tumor center as associated with poor survival. Moreover, loss-of-function BRCA1-associated protein-1 mutations are associated with and cause elevated expression of the immunosuppressive checkpoint marker, B7 homolog 4. CONCLUSIONS: This study provides a foundation on which to rationally improve and tailor immunotherapy approaches for this difficult-to-treat disease.

Perturbation biology links temporal protein changes to drug responses in a melanoma cell line.

Cancer cells have genetic alterations that often directly affect intracellular protein signaling processes allowing them to bypass control mechanisms for cell death, growth and division. Cancer drugs targeting these alterations often work initially, but resistance is common. Combinations of targeted drugs may overcome or prevent resistance, but their selection requires context-specific knowledge of signaling pathways including complex interactions such as feedback loops and crosstalk. To infer quantitative pathway models, we collected a rich dataset on a melanoma cell line: Following perturbation with 54 drug combinations, we measured 124 (phospho-)protein levels and phenotypic response (cell growth, apoptosis) in a time series from 10 minutes to 67 hours. From these data, we trained time-resolved mathematical models that capture molecular interactions and the coupling of molecular levels to cellular phenotype, which in turn reveal the main direct or indirect molecular responses to each drug. Systematic model simulations identified novel combinations of drugs predicted to reduce the survival of melanoma cells, with partial experimental verification. This particular application of perturbation biology demonstrates the potential impact of combining time-resolved data with modeling for the discovery of new combinations of cancer drugs.

Perturbation biology: inferring signaling networks in cellular systems.

We present a powerful experimental-computational technology for inferring network models that predict the response of cells to perturbations, and that may be useful in the design of combinatorial therapy against cancer. The experiments are systematic series of perturbations of cancer cell lines by targeted drugs, singly or in combination. The response to perturbation is quantified in terms of relative changes in the measured levels of proteins, phospho-proteins and cellular phenotypes such as viability. Computational network models are derived de novo, i.e., without prior knowledge of signaling pathways, and are based on simple non-linear differential equations. The prohibitively large solution space of all possible network models is explored efficiently using a probabilistic algorithm, Belief Propagation (BP), which is three orders of magnitude faster than standard Monte Carlo methods. Explicit executable models are derived for a set of perturbation experiments in SKMEL-133 melanoma cell lines, which are resistant to the therapeutically important inhibitor of RAF kinase. The resulting network models reproduce and extend known pathway biology. They empower potential discoveries of new molecular interactions and predict efficacious novel drug perturbations, such as the inhibition of PLK1, which is verified experimentally. This technology is suitable for application to larger systems in diverse areas of molecular biology.

Mutant p53 protects triple-negative breast adenocarcinomas from ferroptosis in vivo.

The TP53 tumor suppressor gene is mutated early in most of the patients with triple-negative breast cancer (TNBC). The most frequent TP53 alterations are missense mutations that contribute to tumor aggressiveness. Here, we used an autochthonous somatic TNBC mouse model, in which mutant p53 can be toggled on and off genetically while leaving the tumor microenvironment intact and wild-type for p53 to identify physiological dependencies on mutant p53. In TNBCs that develop in this model, deletion of two different hotspot p53R172H and p53R245W mutants triggers ferroptosis in vivo, a cell death mechanism involving iron-dependent lipid peroxidation. Mutant p53 protects cells from ferroptosis inducers, and ferroptosis inhibitors reverse the effects of mutant p53 loss in vivo. Single-cell transcriptomic data revealed that mutant p53 protects cells from undergoing ferroptosis through NRF2-dependent regulation of Mgst3 and Prdx6, which encode two glutathione-dependent peroxidases that detoxify lipid peroxides. Thus, mutant p53 protects TNBCs from ferroptotic death.

KRAS Allelic Variants in Biliary Tract Cancers.

Importance: Biliary tract cancers (BTCs) contain several actionable molecular alterations, including FGFR2, IDH1, ERBB2 (formerly HER2), and KRAS. KRAS allelic variants are found in 20% to 30% of BTCs, and multiple KRAS inhibitors are currently under clinical investigation. Objectives: To describe the genomic landscape, co-sequence variations, immunophenotype, genomic ancestry, and survival outcomes of KRAS-mutated BTCs and to calculate the median overall survival (mOS) for the most common allelic variants. Design, Setting, and Participants: This retrospective, multicenter, pooled cohort study obtained clinical and next-generation sequencing data from multiple databases between January 1, 2017, and December 31, 2022. These databases included Princess Margaret Cancer Centre, MD Anderson Cancer Center, Foundation Medicine, American Association for Cancer Research Project GENIE, and cBioPortal for Cancer Genomics. The cohort comprised patients with BTCs who underwent genomic testing. Main Outcome and Measure: The main outcome was mOS, defined as date of diagnosis to date of death, which was measured in months. Results: A total of 7457 patients (n = 3773 males [50.6%]; mean [SD] age, 63 [5] years) with BTCs and genomic testing were included. Of these patients, 5813 had clinical outcome data available, in whom 1000 KRAS-mutated BTCs were identified. KRAS allelic variants were highly prevalent in perihilar cholangiocarcinoma (28.6%) and extrahepatic cholangiocarcinoma (36.1%). Thirty-six KRAS allelic variants were identified, and the prevalence rates in descending order were G12D (41%), G12V (23%), and Q61H (8%). The variant G12D had the highest mOS of 25.1 (95% CI, 22.0-33.0) months compared with 22.8 (95% CI, 19.6-31.4) months for Q61H and 17.8 (95% CI, 16.3-23.1) months for G12V variants. The majority of KRAS-mutated BTCs (98.9%) were not microsatellite instability-high and had low tumor mutational burden (ranging from a median [IQR] of 1.2 (1.2-2.5) to a mean [SD] of 3.3 [1.3]). Immune profiling through RNA sequencing of KRAS and NRAS-mutated samples showed a pattern toward a more immune-inflamed microenvironment with higher M1 macrophage activation (0.16 vs 0.12; P = .047) and interferon-γ expression compared with wild-type tumors. The G12D variant remained the most common KRAS allelic variant in all patient ancestries. Patients with admixed American ancestry had the highest proportion of G12D variant (45.0%). Conclusions and Relevance: This cohort study found that KRAS allelic variants were relatively common and may be potentially actionable genomic alterations in patients with BTCs, especially perihilar cholangiocarcinoma and extrahepatic cholangiocarcinoma. The findings add to the growing data on genomic and immune landscapes of KRAS allelic variants in BTCs and are potentially of value to the planning of specific therapies for this heterogeneous patient group.

Comprehensive Immunogenomic Profiling of IDH1-/2-Altered Cholangiocarcinoma.

PURPOSE: Isocitrate dehydrogenase (IDH)1/2 genomic alterations (GA) occur in 20% of intrahepatic cholangiocarcinoma (iCCA); however, the immunogenomic landscape of IDH1-/2-mutated iCCA is largely unknown. METHODS: Comprehensive genomic profiling (CGP) was performed on 3,067 cases of advanced iCCA. Tumor mutational burden (TMB), PD-L1 expression (Dako 22C3), microsatellite instability (MSI), and genomic loss of heterozygosity (gLOH) as a surrogate marker for homologous recombination deficiency were examined. RNA sequencing of 73 patient samples was analyzed for differences in stromal/immune cell infiltration, immune marker expression, and T-cell inflammation. Tissue microarray arrays were subjected to multiplex immunohistochemistry and colocalization analysis in 100 surgical samples. Retrospective clinical data were collected for 501 patients with cholangiocarcinoma to examine median overall survival (mOS) in IDH1/2+ versus IDHwt. RESULTS: Of 3,067 iCCA cases subjected to CGP, 426 (14%) were IDH1+ and 125 (4%) were IDH2+. IDH1 GA included R132C (69%) and R132L/G/S/H/F (16%/7%/4%/3%/<1%). IDH2 GA occurred at R172 (94.4%) and R140 (6.6%). No significant difference was seen in median gLOH between IDH1+ versus IDHwt iCCA (P = .37), although patterns of comutations differed. MSI-High (P = .009), TMB ≥10 mut/Mb (P < .0001), and PD-L1 positivity were lower in IDH1/2+ versus IDHwt iCCA. Resting natural killer cell population, CD70, and programmed cell death 1 expression were significantly higher in non-IDH1-mutated cases, whereas V-set domain containing T-cell activation inhibitor 1 (B7-H4) expression was significantly higher in IDH1+. No significant difference in mOS was observed between IDH1/2+ versus IDHwt patients. CONCLUSION: Significant differences in GA and immune biomarkers are noted between IDH1/2+ and IDHwt iCCA. IDH1-/2-mutated tumors appear immunologically cold without gLOH. These immunogenomic data provide insight for precision targeting of iCCA with IDH alterations.

Co-clinical Trial of Novel Bispecific Anti-HER2 Antibody Zanidatamab in Patient-Derived Xenografts.

Zanidatamab is a bispecific human epidermal growth factor receptor 2 (HER2)-targeted antibody that has demonstrated antitumor activity in a broad range of HER2-amplified/expressing solid tumors. We determined the antitumor activity of zanidatamab in patient-derived xenograft (PDX) models developed from pretreatment or postprogression biopsies on the first-in-human zanidatamab phase I study (NCT02892123). Of 36 tumors implanted, 19 PDX models were established (52.7% take rate) from 17 patients. Established PDXs represented a broad range of HER2-expressing cancers, and in vivo testing demonstrated an association between antitumor activity in PDXs and matched patients in 7 of 8 co-clinical models tested. We also identified amplification of MET as a potential mechanism of acquired resistance to zanidatamab and demonstrated that MET inhibitors have single-agent activity and can enhance zanidatamab activity in vitro and in vivo. These findings provide evidence that PDXs can be developed from pretreatment biopsies in clinical trials and may provide insight into mechanisms of resistance. SIGNIFICANCE: We demonstrate that PDXs can be developed from pretreatment and postprogression biopsies in clinical trials and may represent a powerful preclinical tool. We identified amplification of MET as a potential mechanism of acquired resistance to the HER2 inhibitor zanidatamab and MET inhibitors alone and in combination as a therapeutic strategy. This article is featured in Selected Articles from This Issue, p. 695.

Recurrent composite markers of cell types and states.

Determining concise sets of genomic markers that identify cell types and states within tissue ecosystems remains challenging. To address this challenge, we developed Recurrent Composite Markers for Biological Identities with Neighborhood Enrichment (RECOMBINE). Validations of RECOMBINE with simulation and transcriptomics data in bulk, single-cell and spatial resolutions demonstrated the method's ability for unbiased selection of composite markers that characterize biological subpopulations. RECOMBINE captured markers of mouse visual cortex from single-cell RNA sequencing data and provided a gene panel for targeted spatial transcriptomics profiling. RECOMBINE identified composite markers of CD8 T cell states including GZMK + HAVCR2 - effector memory cells associated with anti-PD1 therapy response. The method outperformed differential gene expression analysis in characterizing a rare cell subpopulation within mouse intestine. Using RECOMBINE, we uncovered hierarchical gene programs of inter- and intra-tumoral heterogeneity in breast and skin tumors. In conclusion, RECOMBINE offers a data-driven approach for unbiased selection of composite markers, resulting in improved interpretation, discovery, and validation of cell types and states.

Mapping the functional interactions at the tumor-immune checkpoint interface.

The interactions between tumor intrinsic processes and immune checkpoints can mediate immune evasion by cancer cells and responses to immunotherapy. It is, however, challenging to identify functional interactions due to the prohibitively complex molecular landscape of the tumor-immune interfaces. We address this challenge with a statistical analysis framework, immuno-oncology gene interaction maps (ImogiMap). ImogiMap quantifies and statistically validates tumor-immune checkpoint interactions based on their co-associations with immune-associated phenotypes. The outcome is a catalog of tumor-immune checkpoint interaction maps for diverse immune-associated phenotypes. Applications of ImogiMap recapitulate the interaction of SERPINB9 and immune checkpoints with interferon gamma (IFNγ) expression. Our analyses suggest that CD86-CD70 and CD274-CD70 immunoregulatory interactions are significantly associated with IFNγ expression in uterine corpus endometrial carcinoma and basal-like breast cancer, respectively. The open-source ImogiMap software and user-friendly web application will enable future applications of ImogiMap. Such applications may guide the discovery of previously unknown tumor-immune interactions and immunotherapy targets.

A prognostic matrix code defines functional glioblastoma phenotypes and niches.

Interactions among tumor, immune and vascular niches play major roles in driving glioblastoma (GBM) malignancy and treatment responses. The composition, heterogeneity, and localization of extracellular core matrix proteins (CMPs) that mediate such interactions, however, are not well understood. Here, we characterize functional and clinical relevance of genes encoding CMPs in GBM at bulk, single cell, and spatial anatomical resolution. We identify a "matrix code" for genes encoding CMPs whose expression levels categorize GBM tumors into matrisome-high and matrisome-low groups that correlate with worse and better survival, respectively, of patients. The matrisome enrichment is associated with specific driver oncogenic alterations, mesenchymal state, infiltration of pro-tumor immune cells and immune checkpoint gene expression. Anatomical and single cell transcriptome analyses indicate that matrisome gene expression is enriched in vascular and leading edge/infiltrative anatomic structures that are known to harbor glioma stem cells driving GBM progression. Finally, we identified a 17-gene matrisome signature that retains and further refines the prognostic value of genes encoding CMPs and, importantly, potentially predicts responses to PD1 blockade in clinical trials for GBM. The matrisome gene expression profiles may provide biomarkers of functionally relevant GBM niches that contribute to mesenchymal-immune cross talk and patient stratification to optimize treatment responses.

A prognostic matrix code defines functional glioblastoma phenotypes and niches.

Interactions among tumor, immune and vascular niches play major roles in driving glioblastoma (GBM) malignancy and treatment responses. The composition, heterogeneity, and localization of extracellular core matrix proteins (CMPs) that mediate such interactions, however, are not well understood. Here, we characterize functional and clinical relevance of genes encoding CMPs in GBM at bulk, single cell, and spatial anatomical resolution. We identify a "matrix code" for genes encoding CMPs whose expression levels categorize GBM tumors into matrisome-high and matrisome-low groups that correlate with worse and better patient survival, respectively. The matrisome enrichment is associated with specific driver oncogenic alterations, mesenchymal state, infiltration of pro-tumor immune cells and immune checkpoint gene expression. Anatomical and single cell transcriptome analyses indicate that matrisome gene expression is enriched in vascular and leading edge/infiltrative anatomic structures that are known to harbor glioma stem cells driving GBM progression. Finally, we identified a 17-gene matrisome signature that retains and further refines the prognostic value of genes encoding CMPs and, importantly, potentially predicts responses to PD1 blockade in clinical trials for GBM. The matrisome gene expression profiles provide potential biomarkers of functionally relevant GBM niches that contribute to mesenchymal-immune cross talk and patient stratification which could be applied to optimize treatment responses.

Assessment of Response to Neoadjuvant Systemic Treatment in Triple-Negative Breast Cancer Using Functional Tumor Volumes from Longitudinal Dynamic Contrast-Enhanced MRI.

Early assessment of neoadjuvant systemic therapy (NAST) response for triple-negative breast cancer (TNBC) is critical for patient care in order to avoid the unnecessary toxicity of an ineffective treatment. We assessed functional tumor volumes (FTVs) from dynamic contrast-enhanced (DCE) MRI after 2 cycles (C2) and 4 cycles (C4) of NAST as predictors of response in TNBC. A group of 100 patients with stage I-III TNBC who underwent DCE MRI at baseline, C2, and C4 were included in this study. Tumors were segmented on DCE images of 1 min and 2.5 min post-injection. FTVs were measured using the optimized percentage enhancement (PE) and signal enhancement ratio (SER) thresholds. The Mann-Whitney test was used to compare the performance of the FTVs at C2 and C4. Of the 100 patients, 49 (49%) had a pathologic complete response (pCR) and 51 (51%) had a non-pCR. The maximum area under the receiving operating characteristic curve (AUC) for predicting the treatment response was 0.84 (p < 0.001) for FTV at C4 followed by FTV at C2 (AUC = 0.82, p < 0.001). The FTV measured at baseline was not able to discriminate pCR from non-pCR. FTVs measured on DCE MRI at C2, as well as at C4, of NAST can potentially predict pCR and non-pCR in TNBC patients.

Longitudinal dynamic contrast-enhanced MRI radiomic models for early prediction of response to neoadjuvant systemic therapy in triple-negative breast cancer.

Early prediction of neoadjuvant systemic therapy (NAST) response for triple-negative breast cancer (TNBC) patients could help oncologists select individualized treatment and avoid toxic effects associated with ineffective therapy in patients unlikely to achieve pathologic complete response (pCR). The objective of this study is to evaluate the performance of radiomic features of the peritumoral and tumoral regions from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) acquired at different time points of NAST for early treatment response prediction in TNBC. This study included 163 Stage I-III patients with TNBC undergoing NAST as part of a prospective clinical trial (NCT02276443). Peritumoral and tumoral regions of interest were segmented on DCE images at baseline (BL) and after two (C2) and four (C4) cycles of NAST. Ten first-order (FO) radiomic features and 300 gray-level-co-occurrence matrix (GLCM) features were calculated. Area under the receiver operating characteristic curve (AUC) and Wilcoxon rank sum test were used to determine the most predictive features. Multivariate logistic regression models were used for performance assessment. Pearson correlation was used to assess intrareader and interreader variability. Seventy-eight patients (48%) had pCR (52 training, 26 testing), and 85 (52%) had non-pCR (57 training, 28 testing). Forty-six radiomic features had AUC at least 0.70, and 13 multivariate models had AUC at least 0.75 for training and testing sets. The Pearson correlation showed significant correlation between readers. In conclusion, Radiomic features from DCE-MRI are useful for differentiating pCR and non-pCR. Similarly, predictive radiomic models based on these features can improve early noninvasive treatment response prediction in TNBC patients undergoing NAST.

Deep Learning for Fully Automatic Tumor Segmentation on Serially Acquired Dynamic Contrast-Enhanced MRI Images of Triple-Negative Breast Cancer.

Accurate tumor segmentation is required for quantitative image analyses, which are increasingly used for evaluation of tumors. We developed a fully automated and high-performance segmentation model of triple-negative breast cancer using a self-configurable deep learning framework and a large set of dynamic contrast-enhanced MRI images acquired serially over the patients' treatment course. Among all models, the top-performing one that was trained with the images across different time points of a treatment course yielded a Dice similarity coefficient of 93% and a sensitivity of 96% on baseline images. The top-performing model also produced accurate tumor size measurements, which is valuable for practical clinical applications.

Computation of conformational transitions in proteins by virtual atom molecular mechanics as validated in application to adenylate kinase.

Many proteins function through conformational transitions between structurally disparate states, and there is a need to explore transition pathways between experimentally accessible states by computation. The sizes of systems of interest and the scale of conformational changes are often beyond the scope of full atomic models, but appropriate coarse-grained approaches can capture significant features. We have designed a comprehensive knowledge-based potential function based on a C alpha representation for proteins that we call the virtual atom molecular mechanics (VAMM) force field. Here, we describe an algorithm for using the VAMM potential to describe conformational transitions, and we validate this algorithm in application to a transition between open and closed states of adenylate kinase (ADK). The VAMM algorithm computes normal modes for each state and iteratively moves each structure toward the other through a series of intermediates. The move from each side at each step is taken along that normal mode showing greatest engagement with the other state. The process continues to convergence of terminal intermediates to within a defined limit--here, a root-mean-square deviation of 1 A. Validations show that the VAMM algorithm is highly effective, and the transition pathways examined for ADK are compatible with other structural and biophysical information. We expect that the VAMM algorithm can address many biological systems.

A force field for virtual atom molecular mechanics of proteins.

Activities of many biological macromolecules involve large conformational transitions for which crystallography can specify atomic details of alternative end states, but the course of transitions is often beyond the reach of computations based on full-atomic potential functions. We have developed a coarse-grained force field for molecular mechanics calculations based on the virtual interactions of C alpha atoms in protein molecules. This force field is parameterized based on the statistical distribution of the energy terms extracted from crystallographic data, and it is formulated to capture features dependent on secondary structure and on residue-specific contact information. The resulting force field is applied to energy minimization and normal mode analysis of several proteins. We find robust convergence in minimizations to low energies and energy gradients with low degrees of structural distortion, and atomic fluctuations calculated from the normal mode analyses correlate well with the experimental B-factors obtained from high-resolution crystal structures. These findings suggest that the virtual atom force field is a suitable tool for various molecular mechanics applications on large macromolecular systems undergoing large conformational changes.

Precision Combination Therapies Based on Recurrent Oncogenic Coalterations.

Cancer cells depend on multiple driver alterations whose oncogenic effects can be suppressed by drug combinations. Here, we provide a comprehensive resource of precision combination therapies tailored to oncogenic coalterations that are recurrent across patient cohorts. To generate the resource, we developed Recurrent Features Leveraged for Combination Therapy (REFLECT), which integrates machine learning and cancer informatics algorithms. Using multiomic data, the method maps recurrent coalteration signatures in patient cohorts to combination therapies. We validated the REFLECT pipeline using data from patient-derived xenografts, in vitro drug screens, and a combination therapy clinical trial. These validations demonstrate that REFLECT-selected combination therapies have significantly improved efficacy, synergy, and survival outcomes. In patient cohorts with immunotherapy response markers, DNA repair aberrations, and HER2 activation, we have identified therapeutically actionable and recurrent coalteration signatures. REFLECT provides a resource and framework to design combination therapies tailored to tumor cohorts in data-driven clinical trials and preclinical studies. SIGNIFICANCE: We developed the predictive bioinformatics platform REFLECT and a multiomics- based precision combination therapy resource. The REFLECT-selected therapies lead to significant improvements in efficacy and patient survival in preclinical and clinical settings. Use of REFLECT can optimize therapeutic benefit through selection of drug combinations tailored to molecular signatures of tumors. See related commentary by Pugh and Haibe-Kains, p. 1416. This article is highlighted in the In This Issue feature, p. 1397.

BET inhibition induces vulnerability to MCL1 targeting through upregulation of fatty acid synthesis pathway in breast cancer.

Therapeutic options for treatment of basal-like breast cancers remain limited. Here, we demonstrate that bromodomain and extra-terminal (BET) inhibition induces an adaptive response leading to MCL1 protein-driven evasion of apoptosis in breast cancer cells. Consequently, co-targeting MCL1 and BET is highly synergistic in breast cancer models. The mechanism of adaptive response to BET inhibition involves the upregulation of lipid synthesis enzymes including the rate-limiting stearoyl-coenzyme A (CoA) desaturase. Changes in lipid synthesis pathway are associated with increases in cell motility and membrane fluidity as well as re-localization and activation of HER2/EGFR. In turn, the HER2/EGFR signaling results in the accumulation of and vulnerability to the inhibition of MCL1. Drug response and genomics analyses reveal that MCL1 copy-number alterations are associated with effective BET and MCL1 co-targeting. The high frequency of MCL1 chromosomal amplifications (>30%) in basal-like breast cancers suggests that BET and MCL1 co-targeting may have therapeutic utility in this aggressive subtype of breast cancer.

CellBox: Interpretable Machine Learning for Perturbation Biology with Application to the Design of Cancer Combination Therapy.

Systematic perturbation of cells followed by comprehensive measurements of molecular and phenotypic responses provides informative data resources for constructing computational models of cell biology. Models that generalize well beyond training data can be used to identify combinatorial perturbations of potential therapeutic interest. Major challenges for machine learning on large biological datasets are to find global optima in a complex multidimensional space and mechanistically interpret the solutions. To address these challenges, we introduce a hybrid approach that combines explicit mathematical models of cell dynamics with a machine-learning framework, implemented in TensorFlow. We tested the modeling framework on a perturbation-response dataset of a melanoma cell line after drug treatments. The models can be efficiently trained to describe cellular behavior accurately. Even though completely data driven and independent of prior knowledge, the resulting de novo network models recapitulate some known interactions. The approach is readily applicable to various kinetic models of cell biology. A record of this paper's Transparent Peer Review process is included in the Supplemental Information.