Neoadjuvant systemic therapy (NST) is a standard treatment approach for patients with early-stage breast cancer, particularly those with stage II-III disease and aggressive subtypes such as HER2-positive and triple-negative breast cancer. While NST improves surgical outcomes and provides prognostic information, accurately assessing preoperative treatment response remains a clinical challenge. Circulating tumor DNA (ctDNA) has emerged as a promising non-invasive biomarker for monitoring disease dynamics and guiding therapeutic decisions. In this study, we aimed to evaluate whether ctDNA analysis in patients with stage II-III breast cancer (n = 20) could serve as a surrogate for invasive biopsies in molecular profiling and as a tool for monitoring response to NST. At baseline, ctDNA was detectable in the majority of patients by droplet digital (dd)PCR (15/18, 83%) and all patients with longitudinal follow-up had ctDNA clearance after NST (13/13; 100%). A positive correlation was observed between the allele fraction in ctDNA, histologic grade and molecular subtype, suggesting that ctDNA levels may be influenced by tumor biology. None of the three patients with undetectable baseline ctDNA had distant relapse, regardless of whether they achieved pathologic complete response (pCR), compared to 5/15 (33%) with detectable baseline ctDNA. These findings suggest that ctDNA assessment at baseline may provide additional prognostic information to define the risk of patients after NST. While ctDNA shows promise in capturing tumor burden and biological characteristics, its role in predicting pCR and long-term outcomes requires further investigation.

Circulating tumor DNA (ctDNA) has potential as a prognostic factor for predicting relapse in high-risk breast cancer (BC). This study investigates the utility of ctDNA assessment using a tumor-informed assay, in patients with high-risk BC treated with neoadjuvant chemotherapy (NAC). Thirty newly diagnosed patients with various high-risk BC subtypes participated, providing serial blood samples at multiple time points, including baseline, during NAC, and during follow-up. ctDNA was detected at baseline in 29/29 patients for whom an assay panel could be designed, with detection sensitivity reaching 0.0083% (variant allele frequency). Among patients with detectable baseline ctDNA, 94% showed clearance during treatment, correlating with improved outcomes. Additionally, ctDNA detection post-surgery or during follow-up predicted disease recurrence. These findings suggest that serial ctDNA monitoring throughout NAC and follow-up can effectively identify residual disease in BC and correlate with clinical outcomes.

Trastuzumab Deruxtecan (T-DXd) is clinically beneficial in HER2-positive and HER2-low metastatic breast cancer. However, therapeutic resistance emerges over time in most patients, with poorly defined resistance mechanisms. Through a molecular characterization of paired patient specimens before and after T-DXd treatment, we found that 49% cases had major decreases in HER2 expression at progression, and among them, 52% exhibited complete HER2 loss. Using isogenic model systems, we demonstrated that decreases in HER2 expression corresponded to reductions in T-DXd internalization and major increases in drug IC50 value for tumor growth inhibition. We further identified and validated ERBB2 mutations in the trastuzumab binding interface (V597M and P593R) that promoted T-DXd resistance. As a strategy to overcome impaired T-DXd binding and internalization, we tested low-dose combinations of T-DXd with TROP2-directed antibody-drug conjugates (ADC) and found that these could more uniformly deliver DXd payloads and thereby overcome resistance mediated by HER2 loss. SIGNIFICANCE:The mechanisms underlying clinical T-DXd resistance have not been established. We now report on two mechanisms of resistance that converge on loss of target binding and propose combinations of distinct ADCs with shared payloads as a strategy to overcome resistance by enhancing intratumor delivery. See related commentary by O'Meara and Tarantino, p.195.

Antibody-drug conjugates (ADCs) represent a growing therapeutic class in oncology marked by several transformative clinical successes. However, these exceptional outcomes remain restricted to a limited number of tumor types, and ADC development has been marked by frequent clinical setbacks, underscoring persistent challenges in optimal patient selection, biomarker assay standardization, chemical design, and the limited predictive value of existing preclinical models. Despite broader advances in precision oncology, ADC development has largely occurred without validated biomarkers. Emerging evidence indicates that ADC activity extends beyond target antigen expression alone, encompassing tumor-intrinsic features and tumor microenvironment-dependent processes. This challenges the view of ADCs as strictly targeted agents and supports their conceptualization as tumor-ecosystem-targeting therapies governed by multidimensional biological determinants. In this review, we link ADC successes and setbacks to mechanisms of efficacy, resistance, and toxicity, and discuss how biomarkers, ADC combinations, next-generation platforms, and curative-intent strategies should shape precision oncology frameworks.

Abstract Background: Endocrine therapy resistance in estrogen receptor positive metastatic breast cancer remains poorly understood, with many resistant tumors lacking actionable genomic alterations. Emerging evidence suggests epigenetic reprogramming drives resistance through transcription factor regulatory network rewiring, yet comprehensive epigenomic characterization of resistant tumors remains limited. Methods: We performed integrated ATAC-seq and RNA-seq profiling in clinically relevant models of therapy-resistant and sensitive estrogen receptor positive metastatic breast cancer, including patient-derived xenografts harboring diverse genomic alterations, along with sensitive breast cancer cell lines. We applied unsupervised clustering to chromatin accessibility and gene expression data to identify non-genomic clusters. Master regulators were identified by integrating transcription factor binding motif enrichment in accessible chromatin regions with downstream target gene expression networks. Results: We identified five distinct epigenomic clusters of therapy-resistant estrogen receptor positive metastatic breast cancer, each driven by unique master transcription factor regulatory programs. Cluster 1, comprising all sensitive models such as patient-derived xenografts and cell lines, retained luminal hormone-responsive identity. Therapy resistant tumors segregated into four distinct programs: Cluster 2 showed pioneer factor dominance with enhanced chromatin remodeling while maintaining partial luminal features. Cluster 5 represented an ESR1-mutant luminal HER2 hybrid state with ERBB2 amplification, combining altered estrogen receptor signaling with receptor tyrosine kinase activation. Cluster 4 displayed mesenchymal resistance via epithelial-mesenchymal transition. Cluster 3 represented the most dedifferentiated phenotype, an inflammatory cancer stem cell state with FOXA1 loss. Conclusion: Our integrated epigenomic approach reveals that therapy-resistance in estrogen receptor positive metastatic breast cancer is orchestrated by four distinct transcription factor-driven regulatory programs that emerge through non-genetic rewiring rather than genomic alterations alone. These clusters span a spectrum from pioneer factor-mediated chromatin remodeling and hybrid luminal-HER2 states to mesenchymal plasticity and inflammatory stem-like networks. Importantly, these epigenetic clusters may explain mutation-negative resistance and reveal subtype-specific therapeutic vulnerabilities. This molecular framework provides a roadmap for precision medicine approaches tailored to the epigenomic state of resistant metastatic breast cancer. Citation Format: Gizem Yayli-Vokshi, Sandra Cohen, Weiling Li, Hong Shao, Sydney Bowker, Elisa de Stanchina, Pedram Razavi, Sarat Chandarlapaty, Ekta Khurana. Integrated epigenomic profiling reveals distinct transcription factor networks driving therapy resistance in estrogen receptor positive metastatic breast cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3220.

TPS1162 Background: Triple-negative breast cancer (TNBC) refers to a heterogenous group of breast cancers that lack expression of ER, PR, and HER2. Despite recent advances with immunotherapy (IO) and antibody-drug conjugates (ADCs), TNBC remains the most aggressive subtype, with short overall survival in the metastatic setting. Breast tumors with low levels of ER and PR expression (1-10%) clinically behave like TNBC, and clinical management follows the TNBC treatment (tx) paradigm. We and others have identified a subset of ER/PR/HER2-negative breast cancers (BCs) that express the androgen receptor (AR). Enzalutamide (enza), an AR-antagonist, has demonstrated activity in AR-positive metastatic TNBC (Traina et al, JCO 2018). Activation of the glucocorticoid receptor (GR) has been implicated as a mechanism of resistance to AR inhibition in prostate and BCs (Kach et al, Sci Transl Med 2015). Effective therapies for advanced TNBC remain an unmet need, particularly in patients who are ineligible for or progress following a checkpoint inhibitor. This randomized study evaluates the efficacy of enzalutamide or enzalutamide plus the GR antagonist mifepristone (mif) as compared to physician’s choice chemotherapy (TPC). Methods: This is a randomized phase II trial; 201 patients (pts) will be randomized in a 1:1:1 fashion to enza, enza with mif, or TPC (carboplatin, paclitaxel, eribulin, or capecitabine). The primary endpoint (endpt) is progression-free survival (PFS), and the trial is designed to test the hypothesis that PFS in the pooled enza arms is superior to TPC; there is 80% power to detect a hazard ratio (HR) of 0.70, corresponding to an increase in median PFS from 3.5 months (mos) with TPC to 5.0 mos with enza-based tx. Secondary endpts include comparisons of PFS among the 3 arms and evaluation of response rate, clinical benefit rate, duration of response, overall survival, safety, and patient-reported outcomes by arm. Exploratory endpts include correlation of tumor and circulating markers (constitutively active AR variants in circulating tumor cells and cfDNA) with tx response. Eligible pts must have: ECOG 0-2, metastatic measurable or evaluable disease (dz), normal organ function, no history of brain mets, < prior lines of chemotx, any # of prior endocrine txs, no prior anti-AR tx, no prior mif, no concurrent CYP17 inhibitor use. Tumors must test ER/PR low or negative, HER2 negative, AR >10%. Pts with PD-L1+ BC must have received prior IO if not contraindicated. As of December 28, 2025, 32 of 201 pts have been enrolled on study. Clinical trial information: NCT06099769 .

The co-occurrence of germline and somatic oncogenic alterations is frequently observed in breast cancer, yet their combined influence on tumour evolution and therapy resistance remains poorly defined. Through an integrated clinicogenomic analysis of more than 5,800 patients, we show that germline (g) pathogenic variants dictate the evolutionary trajectory of acquired resistance. We specifically find that gBRCA2-associated tumours are uniquely predisposed to develop acquired RB1 loss-of-function alterations, resulting in poor outcomes on standard-of-care frontline CDK4/6 inhibitor (CDK4/6i) combinations. This vulnerability is driven by a dual mechanism: baseline RB1 hemizygosity (heterozygous loss resulting in a single functional RB1 allele), which lowers the evolutionary barrier to biallelic inactivation, and ongoing homologous recombination deficiency, which promotes acquisition of RB1 loss-of-function alterations under the selective pressure of CDK4/6i. Preclinical models from gBRCA2 carriers showed near-uniform resistance to CDK4/6i, with consistent post-treatment Rb loss. Across multiple independent models and in our clinical data, PARP inhibition consistently outperformed CDK4/6i. Our findings suggest that prioritizing PARP inhibition in gBRCA2 carriers may intercept RB1-loss trajectories and delay resistance. More broadly, we establish a predictive framework for forecasting drug-resistant trajectories based on pre-treatment allelic configuration and mutational signatures.

The retinoblastoma protein (Rb) is a tumour suppressor best known for repressing E2F transcription factors and halting cell cycle progression1. In hormone receptor-positive (HR+) breast cancer, CDK4/6 inhibitors activate Rb by preventing its phosphorylation, forming a key component of current endocrine therapy regimens2. How pharmacologically activated Rb remodels chromatin and influences transcription beyond cell cycle arrest remains poorly understood. Here we show that CDK4/6 inhibition induces redistribution of hypophosphorylated Rb to promoters and enhancers. Although Rb predictably binds to cell cycle gene promoters to repress transcription, at other sites, it unexpectedly promotes expression of oestrogen-responsive genes by integrating into oestrogen receptor (ER)-rich transcriptional hubs. CDK4/6 inhibition enhances ER target gene expression in breast cancer cells, patient-derived xenografts and clinical HR+ breast cancer samples in an Rb-dependent manner. This reprogramming is mediated in part by KDM5A, whose interaction with Rb contributes to gene regulation at these loci. Critically, components of this Rb-driven ER transcriptional program are pro-proliferative. In endocrine-sensitive tumours, this effect can be neutralized with anti-oestrogen therapy, explaining therapeutic synergy. In endocrine-resistant settings such as ESR1-mutant breast cancer, the program persists, limiting the therapeutic efficacy of CDK4/6 inhibition. These findings reframe Rb as a dual-function transcriptional regulator that, although enforcing cell cycle arrest, can also activate programs that counteract its tumour suppressor function.

Supplementary Figures S1-S5 Supplementary Tables S1-S5

The androgen receptor (AR) is expressed in 75% of estrogen receptor-positive (ER+) breast cancers (BC). Selective AR modulators (SARMs), like EP0062, present a promising therapeutic strategy for ER + BC, particularly in patients who cannot tolerate endocrine therapy (ET) or whose tumors have developed resistance. We aimed to study the antitumor activity of EP0062 in ER+ patient-derived xenograft (PDX) models. EP0062 displayed comparable antitumor efficacy to selective ER degraders (SERDs), including in PDXs with ESR1, PIK3CA, or PTEN mutations. Tumors sensitive to SARMs were enriched in GATA3 mutations. EP0062 treatment induced AR-target genes across all models tested. A transcriptional signature associated with SARM sensitivity was identified, primarily driven by proliferation-related processes, consistent with a significant decrease in S-phase cell cycle proteins upon treatment in EP0062-sensitive models. In some EP0062-resistant tumors, the combination with palbociclib enhanced the antitumor effect of EP0062, suggesting a potential strategy for metastatic patients with acquired ET resistance.

Abstract Background: The Breast Cancer Index (BCI) is a gene expression-based assay that stratifies patients based on their risk of overall (0-10 years) and late (beyond 5 years) distant recurrence. In addition , BCI can also predict the benefit of extended endocrine therapy in early-stage, HR+ breast cancer. In this study, we assessed the relationships between BCI and risk of distant recurrence and explored associations between BCI classification and the tumor genomic and transcriptomic profiles of HR+ patients who experienced metastatic relapse. Methods: Primary tumors from patients with metastatic or recurrent HR+ breast cancer underwent BCI testing, MSK-IMPACT targeted sequencing (up to 505 cancer genes), and mRNA sequencing. Time to DR (TTDR) was defined as the time from surgery to first distant recurrence. A 5-year cutoff was used to distinguish early versus late DR groups. Kaplan-Meier and Cox proportional hazards analyses were used to assess BCI’s prognostic value. Wilcoxon tests were applied for pairwise comparisons of BCI scores between early and late DR groups, and Fisher’s exact tests were used to identify genomic alterations differing by BCI groups as well as DR groups. Given the limited cohort size, multiple testing correction was not applied. Transcriptomic data (N=91) were analyzed for differentially expressed genes by comparing BCI high-risk vs. low-risk tumors and early vs. late DR groups (|log2FC|>1, adjusted p<0.05). Results: The study included 180 HR+ patients (47% post-menopausal, 61% N+, 65.3% grade 3). Most tumors were classified as BCI high risk (86.7%), and patients classified as high risk by BCI had earlier recurrence than those classified as low risk (median TTDR: 3.0 years vs. 4.7 years; HR = 1.89, 95% CI: 1.20-2.96; p = 0.0053). BCI scores differ significantly between early and late DR groups (p = 0.005). The genomic landscape was consistent with high-risk luminal tumors and showed frequent TP53 mutations (38%). PIK3CA and TBX3 mutations were more common in BCI low-risk tumors (71% vs. 41%, p=0.007; 21% vs. 5%, p=0.01). CDKN2A deletions and MCL1 amplifications were enriched in early DR, whereas ERBB2 and SPOP amplifications as well as NOTCH3, and MAP2K4 mutations were more frequent in late DR (p<0.05). Transcriptomic analysis revealed downregulation of cell cycle, p53 signaling, senescence, and progesterone-mediated oocyte maturation pathways in BCI high-risk tumors. The cell cycle pathway was also downregulated in late recurrences, suggesting reduced proliferative activity associated with delayed relapse. Conclusion: BCI remained a strong prognostic indicator in this HR+ metastatic cohort, with higher scores predicting shorter TTDR. Integrated genomic and transcriptomic profiling highlighted distinct molecular programs associated with BCI classification providing insight into biological mechanisms underlying early versus late relapse in HR+ breast cancer. Citation Format: Hong Zhang, Niloufar Khojandi, Natalia Siuliukina, Julia Ah-Reum An, Darya Dahi, Luca Boscolo Bielo, Subhiksha Nandakumar, Enrico Moiso, Edaise M. da Silva, Mehnaj Ahmed, Lisa Loudon, Konner Nelson, Kevin Murphy, Jade Oghoanina, Mark E. Robson, Sarat Chandarlapaty, George Plitas, Amanda K. L. Anderson, Yi Zhang, Pedram Razavi, Kai Treuner. Associations between Breast Cancer Index and MSK-IMPACT genomic and transcriptomic profiles in HR+ breast cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1188.

e13006 Background: Sacituzumab Govitecan (SG) is a Trop2-targeting antibody drug conjugate (ADC) approved for the treatment of HER2-non amplified metastatic breast cancer (MBC). While changes to receptor target density can be associated with response and resistance in HER2-targeted ADCs, the relationship between dynamic changes in Trop2 (or other ADC targets) and SG resistance is unknown. Methods: Institutional databases were used to identify patients treated with SG for MBC who had paired pre and post treatment biopsies. Paired tissue samples were sectioned and submitted to both mass spectrometry (MS, mPROBE) and multiplex immunofluorescence (mIF, Zeiss). Proteomic quantification of key proteins was compared using the Wilcoxon signed rank test. mIF was described qualitatively. Results: Fourteen patients had tissue pairs from both pre and post-SG that passed quality control and underwent quantitative analysis. Seven patients (50%) had ER+ disease. Median progression free-survival (PFS) on SG was 4.7 months and median overall survival was 15.8 months. Baseline Trop2 expression ranged from 296-6,453amol/ug (median 1,560 amol/ug). At the individual patient level, temporal changes in Trop2 expression varied between -96% to 400% (median -18%) after SG treatment. Similar findings were apparent with TOPO1, where 6/14 (43%) of patients had > 25% loss in TOPO1 expression, and the remaining had stable or increased expression. In aggregate, after adjusting for multiple hypothesis testing, no unidirectional quantitative changes in mean protein expression were observed for Trop2 (-393 amol/ug, p = 0.6), TOPO1 (-200 amol/ug, p = 0.12), TOPO2A (-40 amol/ug, p = 0.2), SLFN11 (66 amol/ug, p > 0.9), HER2 (-53 amol/ug, p > 0.9), HER3 (-16 amol/ug, p = 0.8), or Nectin4 (-127 amol/ug, p = 0.042, q = 0.6). mIF results helped visualize protein changes in Trop2, HER2, HER3, and Nectin4 and corroborated observed changes seen from mass spectrometry. Conclusions: Dynamic patient-level changes in Trop2 and several other putative ADC biomarkers were observed, suggesting heterogeneity in ADC resistance. Further mechanistic investigation will be important to understanding heterogeneity in resistance, and quantitative proteomic methods such as MS and mIF may have utility in characterizing ADC resistance and response.

Abstract Background: Germline BRCA2 mutations substantially increase breast cancer risk, but penetrance varies, indicating a role for genetic modifiers. These modifiers can influence tumor initiation and progression even among individuals with the same BRCA2 variant, and their discovery and characterization can improve risk prediction and therapeutic stratification. For many GWAS-identified BRCA2 modifiers, the biological mechanisms are still unclear. We examined STARD13, a cytoskeletal regulator and putative tumor suppressor identified in prior BRCA2 GWAS signals, may intersect functionally with the Hippo pathway through effects on RhoA-actin dynamics that influence LATS2 activity, which in turn regulates YAP/TAZ-mediated proliferation and genomic stability. Because LATS2 supports genomic integrity and suppresses oncogenic signaling, we hypothesized that disruption of the STARD13 and LATS2 axis may modify BRCA2-associated phenotypes in breast epithelial cells. Methods: BRCA2 mutants were generated in breast epithelial cell lines using CRISPR/Cas9 genome editing. Allele-specific regulatory effects of the SNPs were assessed by luciferase reporter assays. To investigate STARD13 as a genetic modifier, siRNA- and shRNA-mediated knockdown was performed in wild-type and BRCA2 mutant cells. LATS2 expression was quantified by qRT-PCR. Functional assays measuring proliferation, apoptosis, and DNA damage sensitivity evaluated the impact of STARD13 knockdown in different BRCA2 contexts. Results: In wild-type BRCA2 cells, STARD13 knockdown upregulated LATS2 expression, indicating activation of a compensatory tumor suppressor pathway. In contrast, In BRCA2-mutant cells, loss of STARD13 fails to induce the compensatory increase in LATS2 seen in BRCA2-wild-type cells, identifying STARD13 as a modifier of the BRCA2-deficient state. These data implicate LATS2 as a modifier of BRCA2 via STARD13-dependent mechanisms. Ongoing experiments in BRCA2 mutant cell lines and organoids are examining effects on DNA damage, repair, colony formation, gene expression, and responses to PARP inhibitors using shRNA and CRISPR knockouts. Conclusions: Our findings support STARD13 as a potential genetic modifier of BRCA2, influencing the activity of LATS2 tumor suppressor and linking cytoskeletal signaling with DNA repair pathways. The differential regulation of LATS2 in wild-type versus BRCA2 mutant backgrounds suggests a novel axis that may underlie variation in BRCA2 penetrance and cancer risk. These studies aim to elucidate the STARD13-LATS2-BRCA2 interaction network as a determinant of BRCA2 penetrance, as a potential biomarker to allow targeting to decrease penetrance of hereditary breast cancer in affected kindreds. (Supported by Breast Cancer Research Foundation and Niehaus Center for Inherited Cancer Genomics). Citation Format: SHIV PRAKASH VERMA, Mitul Waghmare, Sanchari Bhattacharyya, Catherine Fanjoy, Shao Hong, Xu Zhang, Jonathan Amsalem, Yelena Kemel, Minna Lee, Matthew Buas, Zsofia Stadler, Pedram Razavi, Mark Robson, Sarat Chandarlapaty, Kenneth Offit, Vijai Joseph. STARD13-LATS2 Axis as a potential genetic modifier of BRCA2 in breast cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6817.

The oncogenic impact of somatic driver alterations is shaped by tissue context. Classifying alterations by cancer type and evaluating their context-specific properties requires large cohorts of genomically profiled and clinically annotated tumors. Here, we define cancer type-specific patterns of driver alterations, including 164 newly identified hotspots, in 54,331 tumors from 48,179 patients spanning 448 histological cancer subtypes. One-third of all drivers arose in non-canonical contexts and exhibited distinct features, including increased subclonality, later emergence, and divergent biological properties. Within cancer types, gene fusions and other distinct patterns of co-occurring drivers are indicative of earlier age of disease onset. We also identify ancestry-specific differences in human leukocyte antigen (HLA)-restricted driver neoantigens affecting T cell receptor therapy eligibility, and demonstrate cancer-type-specific patterns of intrinsic resistance via somatic HLA loss. Our findings highlight that functional roles of driver alterations depend on the cancer types and clinical contexts in which they arise.

Supplementary Table S2. Sequencing metrics of the endometrioid endometrial carcinomas subjected to single-nucleus DNA sequencing.

This case series examines the effectiveness of dupilumab as a steroid-sparing therapy among patients with antibody-drug conjugate–related cutaneous toxicities.

Proteins that drive or support human disease phenotypes are attractive molecular targets for precision therapy, yet most are nominated by knockout studies and then targeted with drugs that inhibit core catalytic pockets. These strategies cannot resolve which residues are essential, whether non-catalytic sites offer better selectivity or potency, or identify on-target resistance mechanisms. We introduce a framework that integrates precision genome editing, mechanistically diverse therapeutics, and computational sequence-structure-function analysis to map protein essentiality and potential druggability at single amino acid resolution. Applying this framework across 9 cyclin-dependent kinases (CDKs) and 15 cancer therapeutics-including ATP-competitive inhibitors, PROTACs, and molecular glue degraders-we identify shared and CDK-specific residues critical for cell fitness and drug response, including known resistance mutations and dozens of new variants. The resulting functional maps resolve residue- and mechanism-specific differences in the resistance spectra among agents targeting the same protein. We show that this iterative strategy can also uncover higher order interactions by performing intra- and extragenic epistasis screens to identify residues that mediate on-target and within-family cell fitness and drug resistance. Finally, we find evidence of novel CDK6 mutations in breast cancer patients and concordance between experimental and clinical correlates of response to CDK4/6 inhibitors. By mapping residue-level essentiality and forecasting therapy resistance mutations, target-drug interaction maps could inform clinical treatment and guide design of more selective therapeutic molecules.

The Oncotype DX® Recurrence Score (RS) is an assay for hormone receptor-positive early breast cancer with extensively validated predictive and prognostic value. However, its cost and lag time have limited global adoption, and previous attempts to estimate it using clinicopathologic variables have had limited success. To address this, we assembled 6172 cases across three institutions and developed Orpheus, a multimodal deep learning tool to infer the RS from H&E whole-slide images. Our model identifies TAILORx high-risk cases (RS > 25) with an area under the curve (AUC) of 0.89, compared to a leading clinicopathologic nomogram with 0.73. Furthermore, in patients with RS ≤ 25, Orpheus ascertains risk of metastatic recurrence more accurately than the RS itself (0.75 vs 0.49 mean time-dependent AUC). These findings have the potential to guide adjuvant therapy for high-risk cases and tailor surveillance for patients at elevated metastatic recurrence risk.