Compartment-specific B7-H3 staining in salivary gland carcinomas and its implications for biomarker interpretation

Abstract Background B7-H3/CD276 is an emerging therapeutic target in solid tumors and the focus of antibody-based agents such as antibody-drug conjugates. Its expression across salivary gland carcinoma (SGC) histologies and cellular compartments remains incompletely characterized. We assessed cytoplasmic and membranous B7-H3 staining in SGCs and normal salivary gland tissues. Methods B7-H3 immunohistochemistry was performed on tissue microarrays containing 72 malignant SGC cases and 14 normal salivary gland specimens, including 11 matched tumor–normal pairs. Cytoplasmic and membranous staining were scored separately. We compared expression between tumor and normal tissues and across histologic subtypes, and explored associations with clinicopathologic features. Results Cytoplasmic staining was evaluable in 69 tumors and membranous staining in 68. Cytoplasmic B7-H3 H-scores were higher in tumors than in normal tissues in both unpaired (median, 56.3 vs 2.5; p = 0.0006; q = 0.0023) and matched comparisons (median, 70.0 vs 2.5; p = 0.0029; q = 0.0059). Membranous staining was generally lower than cytoplasmic staining, and membranous tumor–normal differences were not statistically significant in either comparison. In pooled exploratory analyses, membranous H-scores showed modest positive correlations with ordinal T category (ρ = 0.292; p = 0.016; q = 0.099) and Ki-67 index (ρ = 0.306; p = 0.015; q = 0.099). Neither association was significant when mucoepidermoid carcinoma and adenoid cystic carcinoma were analyzed separately. Conclusions B7-H3 staining in SGCs differs between cytoplasmic and membranous compartments and across histologic subtypes. Cytoplasmic staining distinguished tumors from normal salivary tissues, whereas the clinical significance of membranous staining remains uncertain. Its potential value as a biomarker requires evaluation in larger, subtype-specific cohorts with outcome and treatment-response data.