PD-L1 upregulation as a reflection of distinct immune contexts in luminal and basal breast cancer.

1040 Background: PD-L1 immunohistochemistry (IHC) is an approved biomarker to select patients with metastatic triple-negative breast cancer for immune checkpoint blockade (ICB). However, PD-L1 performs poorly in Luminal breast cancer, suggesting that PD-L1 upregulation may reflect distinct immune contexts across intrinsic subtypes. We hypothesized that PD-L1 expression can arise in either a canonical T-cell–inflamed state or a discordant PD-L1–high, T-cell–non-inflamed state. Discordance between PD-L1 expression and T-cell inflammation highlights the complexity of tumor–immune interactions, and accumulating evidence supports that malignancies can induce immune alterations systemically. To explore these broader immune features with a noninvasive approach, we examined plasma extracellular vesicles (EVs). Methods: We analyzed 1,084 primary invasive breast cancers from TCGA, focusing on Luminal (LumA/LumB) and Basal subtypes (n=867). PD-L1 status was defined using CD274 mRNA tertiles as a transcriptomic proxy for IHC, and a T-cell–inflamed signature (Ayers 18-gene TIS) classified tumors as inflamed (top 20th percentile), consistent with prior transcriptomic studies. Biomarker concordance was defined as the proportion of PD-L1-high tumors that were also TIS-high. Because CD274 is included in the TIS, analyses were interpreted conservatively with planned sensitivity analyses excluding CD274. For translational feasibility, EV profiling workflows were developed using plasma from the PyMT-N mouse model (immune-cold, myeloid-dominant TIME; method-development model, n=1). Plasma EVs from two patients with advanced breast cancer and one healthy control were profiled using proteomic and targeted EV RNA assays. Results: In Basal tumors, PD-L1 expression aligned with T-cell inflammation: 75.6% (65/86) of PD-L1-high tumors were TIS-high. In contrast, Luminal tumors showed reduced concordance, with only 39.4% (80/203) of PD-L1-high tumors classified as TIS-high. Basal PD-L1-high tumors were significantly more likely to be TIS-high than Luminal tumors (OR 4.76; 95% CI 2.70–8.39; p=1.90×10⁻⁸). These findings indicate frequent uncoupling of PD-L1 expression from T-cell inflammation in Luminal breast cancer. In PyMT-N plasma EVs, proteomic profiling detected myeloid-suppressive proteins with minimal T-cell–inflamed signals. In exploratory human plasma EVs, both proteomic and RNA assays captured immune-related signals, supporting EV profiling feasibility. Conclusions: PD-L1 upregulation occurs in distinct immune contexts that differ by intrinsic subtype, with reduced T-cell–inflamed concordance in Luminal breast cancer. This discordance may explain the limited utility of PD-L1 as a stand-alone biomarker in this subtype. Plasma EV profiling may capture systemic immune biology and provide orthogonal information to tumor-based assays, supporting further prospective evaluation.