Joshua Millstein
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Joshua Millstein has a total of 258 co-authors. Here are the top co-authors by number of shared publications:
- Heinz-Josef Lenz
Division of Medical Oncology, University of Southern California Norris Comprehensive Cancer Center, Los Angeles, California, USA
15 shared publications Wu Zhang
Division of Medical Oncology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, USA
12 shared publications- Francesca Battaglin
Division of Medical Oncology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
10 shared publications - Shivani Soni
Division of Medical Oncology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, USA
10 shared publications Francesca Battaglin
Division of Medical Oncology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, USA
7 shared publications- Priya Jayachandran
Division of Medical Oncology, Norris Comprehensive Cancer Center, Keck School of Medicine, University of Southern California, Los Angeles, USA
7 shared publications
Latest Publications
- Open Access
- 19 October 2026
- Published by Springer Nature in Oncogene
- Vol. 45 (46) , 5048-5056
Activation of the endocytosis pathway stratifies subtypes and therapeutic sensitivity in colorectal cancer
Endocytosis regulates receptor trafficking and signaling, yet its role in colorectal cancer (CRC) remains unclear. We analyzed transcriptomic data from 15,025 CRC tumors to evaluate gene expression in clathrin-mediated endocytosis (CME) and the endosomal sorting complexes required for transport (ESCRT) pathway. Pathway-level signatures were compared across consensus molecular subtypes (CMS) and examined in relation to oncogenic signaling programs, while individual endocytosis-related genes were evaluated for associations with clinical outcomes in an independent randomized trial cohort. Expression of CME and ESCRT pathways varied by CMS, with the highest expression in CMS4 and the lowest in CMS3. Endocytosis signatures correlated with multiple oncogenic signaling pathways, including MAPK, TGF-β, WNT, PI3K, NOTCH, and angiogenesis, suggesting broad links between endocytic activity and tumor biology. In the randomized phase III CALGB/SWOG 80405 trial, low expression of AP2M1 —a core adaptor in the CME complex—was significantly associated with longer overall survival and progression-free survival in patients treated with anti-EGFR agents, but not in those receiving anti-VEGF agents. These findings suggest that endocytic trafficking may be associated with oncogenic signaling and differential therapeutic benefit in patients with metastatic CRC. The endocytosis pathway reflects biologically relevant heterogeneity in CRC and may help guide therapeutic stratification.
- 19 October 2026
- Published by Springer Nature in Oncogene
- Vol. 45 (46) , 5048-5056
Open AccessActivation of the endocytosis pathway stratifies subtypes and therapeutic sensitivity in colorectal cancer
Endocytosis regulates receptor trafficking and signaling, yet its role in colorectal cancer (CRC) remains unclear. We analyzed transcriptomic data from 15,025 CRC tumors to evaluate gene expression in clathrin-mediated endocytosis (CME) and the endosomal sorting complexes required for transport (ESCRT) pathway. Pathway-level signatures were compared across consensus molecular subtypes (CMS) and examined in relation to oncogenic signaling programs, while individual endocytosis-related genes were evaluated for associations with clinical outcomes in an independent randomized trial cohort. Expression of CME and ESCRT pathways varied by CMS, with the highest expression in CMS4 and the lowest in CMS3. Endocytosis signatures correlated with multiple oncogenic signaling pathways, including MAPK, TGF-β, WNT, PI3K, NOTCH, and angiogenesis, suggesting broad links between endocytic activity and tumor biology. In the randomized phase III CALGB/SWOG 80405 trial, low expression of AP2M1 —a core adaptor in the CME complex—was significantly associated with longer overall survival and progression-free survival in patients treated with anti-EGFR agents, but not in those receiving anti-VEGF agents. These findings suggest that endocytic trafficking may be associated with oncogenic signaling and differential therapeutic benefit in patients with metastatic CRC. The endocytosis pathway reflects biologically relevant heterogeneity in CRC and may help guide therapeutic stratification.
- Open Access
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
Long Non-Coding RNA SNHG11 as a Prognostic and Predictive Biomarker in Metastatic Colorectal Cancer: Evidence From CALGB (Alliance)/SWOG 80405 and Real-World Data
Background: Long non-coding RNAs (lncRNAs) are emerging regulators of colorectal cancer (CRC) biology, but their clinical relevance in metastatic colorectal cancer (mCRC) remains poorly defined. We evaluated the prognostic and predictive relevance of selected lncRNAs in patients with mCRC enrolled in the CALGB/SWOG 80405 trial. Methods: We analyzed 433 mCRC patients treated with either bevacizumab (n=226) or cetuximab (n=207) plus first-line chemotherapy. Tumor RNA expression (Illumina HiSeq 2500) of 13 candidate lncRNAs was evaluated. Overall survival (OS) and progression-free survival (PFS) were evaluated across expression tertiles using multivariable Cox models. Findings were externally evaluated in an independent real-world clinico-transcriptomic cohort of >15,000 CRC cases.Findings: Among the lncRNAs analyzed, only SNHG11 remained significantly associated with both OS and PFS after adjustment for multiple tests (Benjamini–Hochberg false discovery rate <0.05). In the combined treatment analysis, high SNHG11 expression was independently associated with longer PFS (hazard ratio [HR] 0.82, 95% CI 0.72–0.93) and OS (HR 0.81, 95% CI 0.71–0.92). A significant treatment interaction was observed for OS (interaction P=0.0082), with high SNHG11 expression associated with greater benefit from cetuximab but not bevacizumab. High SNHG11 expression was enriched in Consensus Molecular Subtype 2 tumors (P15,000 CRC cases), high SNHG11 expression was independently associated with improved OS (30.8 versus 27.8 months; HR 0.93, 95% CI 0.89–0.97) and longer time-on-treatment (7.4 versus 5.8 months; HR 0.86, 95% CI 0.79–0.93) with anti-EGFR therapy, but not bevacizumab.Interpretation: High tumor expression of SNHG11 is associated with improved survival outcomes and enhanced benefit from EGFR inhibition in mCRC, supporting its role as candidate prognostic and predictive biomarker. These findings suggest that SNHG11 identifies a biologically favorable, MYC-activated subtype of mCRC and warrants prospective validation as a tool for therapeutic stratification.
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessLong Non-Coding RNA SNHG11 as a Prognostic and Predictive Biomarker in Metastatic Colorectal Cancer: Evidence From CALGB (Alliance)/SWOG 80405 and Real-World Data
Background: Long non-coding RNAs (lncRNAs) are emerging regulators of colorectal cancer (CRC) biology, but their clinical relevance in metastatic colorectal cancer (mCRC) remains poorly defined. We evaluated the prognostic and predictive relevance of selected lncRNAs in patients with mCRC enrolled in the CALGB/SWOG 80405 trial. Methods: We analyzed 433 mCRC patients treated with either bevacizumab (n=226) or cetuximab (n=207) plus first-line chemotherapy. Tumor RNA expression (Illumina HiSeq 2500) of 13 candidate lncRNAs was evaluated. Overall survival (OS) and progression-free survival (PFS) were evaluated across expression tertiles using multivariable Cox models. Findings were externally evaluated in an independent real-world clinico-transcriptomic cohort of >15,000 CRC cases.Findings: Among the lncRNAs analyzed, only SNHG11 remained significantly associated with both OS and PFS after adjustment for multiple tests (Benjamini–Hochberg false discovery rate <0.05). In the combined treatment analysis, high SNHG11 expression was independently associated with longer PFS (hazard ratio [HR] 0.82, 95% CI 0.72–0.93) and OS (HR 0.81, 95% CI 0.71–0.92). A significant treatment interaction was observed for OS (interaction P=0.0082), with high SNHG11 expression associated with greater benefit from cetuximab but not bevacizumab. High SNHG11 expression was enriched in Consensus Molecular Subtype 2 tumors (P15,000 CRC cases), high SNHG11 expression was independently associated with improved OS (30.8 versus 27.8 months; HR 0.93, 95% CI 0.89–0.97) and longer time-on-treatment (7.4 versus 5.8 months; HR 0.86, 95% CI 0.79–0.93) with anti-EGFR therapy, but not bevacizumab.Interpretation: High tumor expression of SNHG11 is associated with improved survival outcomes and enhanced benefit from EGFR inhibition in mCRC, supporting its role as candidate prognostic and predictive biomarker. These findings suggest that SNHG11 identifies a biologically favorable, MYC-activated subtype of mCRC and warrants prospective validation as a tool for therapeutic stratification.
- Open Access
- 5 July 2026
Activation of the endocytosis pathway stratifies subtypes and therapeutic sensitivity in colorectal cancer
Endocytosis regulates receptor trafficking and signaling, yet its role in colorectal cancer (CRC) remains unclear. We analyzed transcriptomic data from 15,025 CRC tumors to evaluate gene expression in clathrin-mediated endocytosis (CME) and the endosomal sorting complexes required for transpor...
- 5 July 2026
Open AccessActivation of the endocytosis pathway stratifies subtypes and therapeutic sensitivity in colorectal cancer
Endocytosis regulates receptor trafficking and signaling, yet its role in colorectal cancer (CRC) remains unclear. We analyzed transcriptomic data from 15,025 CRC tumors to evaluate gene expression in clathrin-mediated endocytosis (CME) and the endosomal sorting complexes required for transpor...
- Open Access
- 24 August 2025
- Published by Springer Nature in Oncogene
- Vol. 44 (41) , 3924-3938
GABAergic signaling contributes to tumor cell invasion and poor overall survival in colorectal cancer
Alterations in neurotransmitter signaling can influence colorectal cancer (CRC). In a large, randomized Phase III clinical trial (CALGB/SWOG 80405) involving patients with metastatic CRC, high expression of gamma-aminobutyric acid (GABA) pathway gene GAD1 and low expression of ABAT, indicative of a GABAergic environment, were associated with worse progression-free survival and overall survival outcomes. A metastasis map of human cancer cell lines (MetMap) and functional studies using a microfluidic tumor-on-chip platform demonstrated that high GAD1 expression correlates with increased metastatic potential. Knockdown and pharmacological inhibition of GAD1 reduced tumor invasion, while exogenous GABA promoted invasion. Tumor-derived GABA was elevated in Ras-altered tumors. Furthermore, analysis of publicly available data confirmed that higher GAD1 expression is associated with worse outcomes in Ras-mutant tumors. These findings establish a role for GABA signaling in tumor invasiveness, particularly in Ras-altered CRC. This study demonstrates using clinical data to inform new discoveries and highlights the need for advanced preclinical model systems that more accurately reflect human physiology to explore these findings.
- 24 August 2025
- Published by Springer Nature in Oncogene
- Vol. 44 (41) , 3924-3938
Open AccessGABAergic signaling contributes to tumor cell invasion and poor overall survival in colorectal cancer
Alterations in neurotransmitter signaling can influence colorectal cancer (CRC). In a large, randomized Phase III clinical trial (CALGB/SWOG 80405) involving patients with metastatic CRC, high expression of gamma-aminobutyric acid (GABA) pathway gene GAD1 and low expression of ABAT, indicative of a GABAergic environment, were associated with worse progression-free survival and overall survival outcomes. A metastasis map of human cancer cell lines (MetMap) and functional studies using a microfluidic tumor-on-chip platform demonstrated that high GAD1 expression correlates with increased metastatic potential. Knockdown and pharmacological inhibition of GAD1 reduced tumor invasion, while exogenous GABA promoted invasion. Tumor-derived GABA was elevated in Ras-altered tumors. Furthermore, analysis of publicly available data confirmed that higher GAD1 expression is associated with worse outcomes in Ras-mutant tumors. These findings establish a role for GABA signaling in tumor invasiveness, particularly in Ras-altered CRC. This study demonstrates using clinical data to inform new discoveries and highlights the need for advanced preclinical model systems that more accurately reflect human physiology to explore these findings.
- Open Access
- 4 July 2025
- Published by Springer Nature in Oncogene
- Vol. 44 (30) , 2565-2573
Pan-RAS inhibitors and polo-like kinase 1: promising targets in colorectal cancer
RAS is an oncogene that is commonly mutated in colorectal cancer (CRC). It has been considered a negative feature both due to its impact on prognosis and due to the shallow interface of oncogenic Ras for therapeutic targeting. Newer pan-Ras inhibitor strategies include improved direct targeting of RAS, blockade of downstream effectors, immunotherapy approaches, and even the inclusion of anti-EGFR drugs. Polo-like Kinase 1 (PLK1) is a serine/threonine protein kinase that controls multiple aspects of the cell-cycle. It is upregulated in CRC and has become an important therapeutic target in KRAS mutant CRC, with several PLK1 inhibitors currently in various phases of development and testing. As with other targeted therapies, resistance remains a problem and combination strategies may be beneficial. This review discusses pan-RAS inhibitors and PLK1 in the context of CRC. It discusses RAS’ many roles, its associated pathways and relationship to cancer progression, the current status of existing inhibitors, and future strategies for targeting in cancer therapy. The wide-ranging impacts of RAS provide a basis to better understand and fight against CRC.
- 4 July 2025
- Published by Springer Nature in Oncogene
- Vol. 44 (30) , 2565-2573
Open AccessPan-RAS inhibitors and polo-like kinase 1: promising targets in colorectal cancer
RAS is an oncogene that is commonly mutated in colorectal cancer (CRC). It has been considered a negative feature both due to its impact on prognosis and due to the shallow interface of oncogenic Ras for therapeutic targeting. Newer pan-Ras inhibitor strategies include improved direct targeting of RAS, blockade of downstream effectors, immunotherapy approaches, and even the inclusion of anti-EGFR drugs. Polo-like Kinase 1 (PLK1) is a serine/threonine protein kinase that controls multiple aspects of the cell-cycle. It is upregulated in CRC and has become an important therapeutic target in KRAS mutant CRC, with several PLK1 inhibitors currently in various phases of development and testing. As with other targeted therapies, resistance remains a problem and combination strategies may be beneficial. This review discusses pan-RAS inhibitors and PLK1 in the context of CRC. It discusses RAS’ many roles, its associated pathways and relationship to cancer progression, the current status of existing inhibitors, and future strategies for targeting in cancer therapy. The wide-ranging impacts of RAS provide a basis to better understand and fight against CRC.
- 1 July 2025
- Published by American Society of Clinical Oncology (ASCO) in Journal of Clinical Oncology
- Vol. 43 (19) , 2173-2183
Development and Validation of a Novel Prediction Model for Hearing Loss From Cisplatin Chemotherapy
PURPOSE: Cisplatin treats many common tumors but causes permanent and debilitating hearing loss (HL). The objective of this study was to develop and externally validate a predictive model of HL in cisplatin-treated children and adolescent cancer survivors. METHODS: The Pediatric Holistic Evaluation of Auditory Risk (PedsHEAR) model architecture used several machine learning approaches followed by an ensemble predictor. The primary end point was post-treatment communication–affecting HL (International Society of Pediatric Oncology Ototoxicity Scale [SIOP] Grade ≥2). PedsHEAR was developed from a multicenter data set of cisplatin-exposed patients up to 21 years old (1984-2017) and externally validated using data from the Children's Oncology Group ACCL05C1 study (2007-2012) and two combined institutional cohorts (1988-2022). The model predicts post-treatment HL in each patient (probability [%], 95% CI) and classifies patients as low, intermediate, or high risk for HL (probability HL 0.60, respectively). RESULTS: In the training data set (n = 1,115, median age 6.3 years, SIOP Grade ≥2 HL 44%), PedsHEAR demonstrated excellent discrimination (AUC, 0.93 [95% CI, 0.92 to 0.95]) and then successfully validated within the internal (testing; AUC, 0.79 [95% CI, 0.74 to 0.85]) and two external validation cohorts (AUC, 0.74 and AUC, 0.67). In an aggregate validation cohort (n = 631), the model predicted the probability of HL (AUC, 0.76 [95% CI, 0.72 to 0.79]) and classified 22% (141/631), 71% (447/631), and 7% (43/631) of patients as low, intermediate, or high risk for HL. CONCLUSION: PedsHEAR predicted SIOP Grade ≥2 HL in pediatric cisplatin-treated patients. This is the first validated model to successfully predict cisplatin-induced HL in a broadly representative population treated with diverse regimens across a range of treatment settings.
- 1 July 2025
- Published by American Society of Clinical Oncology (ASCO) in Journal of Clinical Oncology
- Vol. 43 (19) , 2173-2183
Development and Validation of a Novel Prediction Model for Hearing Loss From Cisplatin Chemotherapy
PURPOSE: Cisplatin treats many common tumors but causes permanent and debilitating hearing loss (HL). The objective of this study was to develop and externally validate a predictive model of HL in cisplatin-treated children and adolescent cancer survivors. METHODS: The Pediatric Holistic Evaluation of Auditory Risk (PedsHEAR) model architecture used several machine learning approaches followed by an ensemble predictor. The primary end point was post-treatment communication–affecting HL (International Society of Pediatric Oncology Ototoxicity Scale [SIOP] Grade ≥2). PedsHEAR was developed from a multicenter data set of cisplatin-exposed patients up to 21 years old (1984-2017) and externally validated using data from the Children's Oncology Group ACCL05C1 study (2007-2012) and two combined institutional cohorts (1988-2022). The model predicts post-treatment HL in each patient (probability [%], 95% CI) and classifies patients as low, intermediate, or high risk for HL (probability HL 0.60, respectively). RESULTS: In the training data set (n = 1,115, median age 6.3 years, SIOP Grade ≥2 HL 44%), PedsHEAR demonstrated excellent discrimination (AUC, 0.93 [95% CI, 0.92 to 0.95]) and then successfully validated within the internal (testing; AUC, 0.79 [95% CI, 0.74 to 0.85]) and two external validation cohorts (AUC, 0.74 and AUC, 0.67). In an aggregate validation cohort (n = 631), the model predicted the probability of HL (AUC, 0.76 [95% CI, 0.72 to 0.79]) and classified 22% (141/631), 71% (447/631), and 7% (43/631) of patients as low, intermediate, or high risk for HL. CONCLUSION: PedsHEAR predicted SIOP Grade ≥2 HL in pediatric cisplatin-treated patients. This is the first validated model to successfully predict cisplatin-induced HL in a broadly representative population treated with diverse regimens across a range of treatment settings.
- Open Access
- 10 June 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (6) , e0324622
CausNet-partial: ‘Partial Generational Orderings’ based search for optimal sparse Bayesian networks via dynamic programming with parent set constraints
In our recent work, we developed a novel dynamic programming algorithm to find optimal Bayesian networks with parent set constraints. This ‘generational orderings’ based dynamic programming algorithm—CausNet—efficiently searches the space of possible Bayesian networks. The method is designed for continuous as well as discrete data, and continuous, discrete and survival outcomes. In the present work, we develop a variant of CausNet—CausNet-partial—where we introduce the space of ‘partial generational orderings’, which is a novel way to search for small and sparse optimal Bayesian networks from large dimensional data. We test this method both on simulated and real data. In simulations, CausNet-partial shows superior performance when compared with three state-of-the-art algorithms. We apply it also to a benchmark discrete Bayesian network ALARM, a Bayesian network designed to provide an alarm message system for patient monitoring. We first apply the original CausNet and then CausNet-partial, varying the partial order from 5 to 2. CausNet-partial discovers small sparse networks with drastically reduced runtime as expected from theory. To further demonstrate the efficacy of CausNet-partial, we apply it to an Ovarian Cancer gene expression dataset with 513 genes and a survival outcome. Our algorithm is able to find optimal Bayesian networks with different number of nodes as we vary the partial order. On a personal computer with a 2.3 GHz Intel Core i9 processor with 16 GB RAM, each processing takes less than five minutes. Our ‘partial generational orderings’ based method CausNet-partial is an efficient and scalable method for finding optimal sparse and small Bayesian networks from high dimensional data.
- 10 June 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (6) , e0324622
Open AccessCausNet-partial: ‘Partial Generational Orderings’ based search for optimal sparse Bayesian networks via dynamic programming with parent set constraints
In our recent work, we developed a novel dynamic programming algorithm to find optimal Bayesian networks with parent set constraints. This ‘generational orderings’ based dynamic programming algorithm—CausNet—efficiently searches the space of possible Bayesian networks. The method is designed for continuous as well as discrete data, and continuous, discrete and survival outcomes. In the present work, we develop a variant of CausNet—CausNet-partial—where we introduce the space of ‘partial generational orderings’, which is a novel way to search for small and sparse optimal Bayesian networks from large dimensional data. We test this method both on simulated and real data. In simulations, CausNet-partial shows superior performance when compared with three state-of-the-art algorithms. We apply it also to a benchmark discrete Bayesian network ALARM, a Bayesian network designed to provide an alarm message system for patient monitoring. We first apply the original CausNet and then CausNet-partial, varying the partial order from 5 to 2. CausNet-partial discovers small sparse networks with drastically reduced runtime as expected from theory. To further demonstrate the efficacy of CausNet-partial, we apply it to an Ovarian Cancer gene expression dataset with 513 genes and a survival outcome. Our algorithm is able to find optimal Bayesian networks with different number of nodes as we vary the partial order. On a personal computer with a 2.3 GHz Intel Core i9 processor with 16 GB RAM, each processing takes less than five minutes. Our ‘partial generational orderings’ based method CausNet-partial is an efficient and scalable method for finding optimal sparse and small Bayesian networks from high dimensional data.
- 1 April 2025
- Published by American Society of Clinical Oncology (ASCO) in JCO Precision Oncology
- Vol. 9 (9) , e2400459
Association of Androgen Receptor Expression With Tumor Immune Landscape and Treatment Outcomes of Patients With Breast Cancer
PURPOSE: Although estrogen receptor is well studied in breast cancer (BC), the role of androgen receptor (AR) in prognosis and therapy response is less understood. Here, we characterized the clinicopathologic and molecular features of AR gene expression in BC subtypes. METHODS: Ten thousand seven hundred twenty-eight BC samples were tested by next-generation DNA sequencing, whole-transcriptome sequencing, and immunohistochemistry at Caris Life Sciences (Phoenix, AZ). Tumors with AR -high and AR -low RNA expression were stratified by top and bottom quartiles, respectively. Treatment-associated survival was obtained from insurance claims and calculated from treatment start to last contact using Kaplan-Meier estimates. Statistical significance was determined by chi-square and Mann-Whitney U test with P values adjusted for multiple comparisons ( q < .05). RESULTS: AR -low was associated with basal-like tumors. AR -high tumors were associated with increased mutation rates in several genes—namely PIK3CA and CDH1 —across all subtypes, while other associations such as RB1 and MAP3K1 were subtype-dependent. The immune landscape was differentially affected by AR expression in each subtype, but these differences did not correspond to differential responses to immune checkpoint blockade. Patients with AR -high tumors had a longer therapy response for most subtypes, but those with AR -high tumors that were human epidermal growth factor receptor 2–enriched and luminal B trended toward worse chemotherapy or hormone therapy response, respectively. CONCLUSION: Our data suggest a unique molecular profile of AR -high BC that is subtype-specific and generally associated with improved outcomes. Exploration of specific mutations and immune-oncology markers associated with AR- high may aid in molecularly selected clinical trial design for patients with advanced BC.
- 1 April 2025
- Published by American Society of Clinical Oncology (ASCO) in JCO Precision Oncology
- Vol. 9 (9) , e2400459
Association of Androgen Receptor Expression With Tumor Immune Landscape and Treatment Outcomes of Patients With Breast Cancer
PURPOSE: Although estrogen receptor is well studied in breast cancer (BC), the role of androgen receptor (AR) in prognosis and therapy response is less understood. Here, we characterized the clinicopathologic and molecular features of AR gene expression in BC subtypes. METHODS: Ten thousand seven hundred twenty-eight BC samples were tested by next-generation DNA sequencing, whole-transcriptome sequencing, and immunohistochemistry at Caris Life Sciences (Phoenix, AZ). Tumors with AR -high and AR -low RNA expression were stratified by top and bottom quartiles, respectively. Treatment-associated survival was obtained from insurance claims and calculated from treatment start to last contact using Kaplan-Meier estimates. Statistical significance was determined by chi-square and Mann-Whitney U test with P values adjusted for multiple comparisons ( q < .05). RESULTS: AR -low was associated with basal-like tumors. AR -high tumors were associated with increased mutation rates in several genes—namely PIK3CA and CDH1 —across all subtypes, while other associations such as RB1 and MAP3K1 were subtype-dependent. The immune landscape was differentially affected by AR expression in each subtype, but these differences did not correspond to differential responses to immune checkpoint blockade. Patients with AR -high tumors had a longer therapy response for most subtypes, but those with AR -high tumors that were human epidermal growth factor receptor 2–enriched and luminal B trended toward worse chemotherapy or hormone therapy response, respectively. CONCLUSION: Our data suggest a unique molecular profile of AR -high BC that is subtype-specific and generally associated with improved outcomes. Exploration of specific mutations and immune-oncology markers associated with AR- high may aid in molecularly selected clinical trial design for patients with advanced BC.