Between 2014 and 2023, histologically confirmed cases of HNSCC were retrieved from the pathology archives of the Department of Hospital Pathology, Seoul St. Mary’s Hospital. A total of 68 cases were retrospectively selected, comprising tumors of the oropharynx (palatine tonsil and base of tongue) and oral cavity (tongue proper and floor of mouth). These anatomical sites were selected for their accessibility and rising incidence, while tumors of the hypopharynx, maxilla, larynx were excluded due to limited case numbers and suboptimal PD-L1 staining following decalcification.

Within the oropharyngeal group (39 cases), 14 were stage I, 19 were stage II, five were stage III, and one was stage IV at diagnosis. In the oral cavity group (29 cases), 20 were stage III and nine were stage IV. For each case, complete sets of preoperative biopsy, surgical resection, and lymph node metastasis samples were collected. In most cases (66/68), lymph node metastases were resected during the same surgical procedure as the primary tumor. Hematoxylin and eosin (H&E)-stained slides were reviewed to confirm the presence of carcinoma.

Initial HPV status screening was performed using p16 IHC. Confirmation was then conducted with the PANA RealTyper HPV kit (Panagene, Daejeon, Korea), a multiplex real-time PCR assay performed on DNA extracted from the tumor tissue. In parallel, PD-L1 IHC was performed using the PD-L1 IHC 22C3 pharmDx assay (Agilent/Dako, CA, USA).

Both p16 and PD-L1 22C3 IHC were conducted on 4-µm-thick sections prepared from formalin-fixed, paraffin-embedded (FFPE) blocks. The entire process for the 204 tissue sections (representing complete sets of specimens from 68 patients) were automated on the Ventana BenchMark ULTRA automated staining platform (Ventana Medical Systems, AZ, USA).

Whole-slide images of the PD-L1-stained slides were digitalized using the Philips Intellisite Pathology Solution platform. The images were then exported as high-resolution TIFF files for further bioimage analysis.

QuPath, an open-source bioimage analysis platform, was used for the manual annotation of four distinct cell populations: tumor cells, immune cells, PD-L1-expressing tumor cells, and PD-L1-expressing immune cells (10) ( Figure 1 ).

PD-L1 expression was evaluated using two standardized scoring systems for each specimen. The combined proportion score (CPS) was calculated as the number of PD-L1–positive cells, including tumor cells, lymphocytes, and macrophages, divided by the total number of viable tumor cells and multiplied by 100. The tumor proportion score (TPS) was calculated as the percentage of PD-L1–positive tumor cells among at least 100 viable tumor cells. Necrotic areas and non-neoplastic cellular components were excluded from the analysis. Based on these scores, specimens were classified into three expression categories: negative (CPS <1; TPS <1%), low expression (CPS 1–19; TPS 1–49%), and high expression (CPS ≥20; TPS ≥50%) ( Figure 2 ).

Statistical analyses were performed to investigate the relationships between different variables. The Mann-Whitney U test was used to analyze the association between HPV status (determined by p16 IHC and PCR) and PD-L1 expression, as well as the association between PD-L1 expression and the stage of oropharyngeal cancer. The Kruskal-Wallis test was applied to compare CPS and TPS values across different specimen types (biopsy, resection, and metastasis). A two-tailed significance level of p<0.05 was used to determine statistical significance.

Among the 68 HNSCC cases, primary tumors originated from the oropharynx (57.3%, 39/68) and the oral cavity (42.7%, 29/68). The mean age at diagnosis was 59.1 years, with a male predominance (male 71%, female 29%). In oropharyngeal HNSCCs, HPV positivity was confirmed in 89.7% of cases (35/39) and was significantly associated with p16 IHC (p <0.05). CPS was significantly higher in HPV-positive oropharyngeal cancers in both biopsy and resection specimens (p < 0.05). Notably, CPS scores differed significantly between early-stage (I-II) and advanced-stage (III-IV) oropharyngeal cancers in resection specimens (p <0.05). No oral cavity HNSCCs were HPV-associated (0/29). Other clinicopathological aspects and PD-L1 scores are summarized in Table 1 .

The average CPS and TPS was 42.8 (0.3-100) and 27.8% (0-92.1) in preoperative biopsies, 62.2 (4.1 -100) and 40.7% (0.04-100) in surgical resections, and 47.4 (0.1-100) and 27.9% (0-9-2.7) in metastatic lymph nodes.

In preoperative biopsies, CPS was negative in 3/68 (4.4%), low in 21/68 (30.9%), and high in 44/68 (64.7%). TPS was negative in 9/68 (13.2%), low in 44/68 (64.7%), and high in 15/68 (22.1%). In surgical resections, CPS was negative in 0/68 (0%), low in 9/68 (13.2%), and high in 59/68 (86.8%). TPS was negative in 1/68 (1.5%), low in 42/68 (61.7%), and high in 25/68 (36.8%). In metastatic lymph nodes, CPS was negative in 2/68 (2.9%), low in 20/68 (29.4%), and high in 46/68 (67.7%). TPS was negative in 11/68 (16.2%), low in 40/68 (58.8%), and high in 17/68 (25.0%).

Among PD-L1–positive cases (low and high expression), CPS consistently showed higher expression in oropharyngeal cancers across all specimen types, whereas TPS displayed variable patterns between oropharyngeal and oral cavity tumors.

In preoperative biopsies, CPS positivity was observed in 26/37 (70.3%) oropharyngeal cancers and 18/28 (64.3%) oral cavity cancers, while TPS positivity was found in 8/34 (23.5%) and 7/25 (28.0%), respectively. In surgical resections, CPS positivity was seen in 34/39 (87.2%) oropharyngeal cancers and 25/29 (86.2%) oral cavity cancers, with TPS positivity in 13/39 (33.3%) and 12/28 (42.9%), respectively. In metastatic lymph nodes, CPS positivity was present in 27/38 (71.1%) oropharyngeal cancers and 19/28 (67.9%) oral cavity cancers, while TPS positivity was seen in 12/33 (36.4%) and 5/24 (20.8%), respectively. Comparison bar graphs for oropharyngeal and oral cavity tumors across specimen types are shown in Figure 3 .

In oropharyngeal cancers (n=39), HPV-positive tumors were associated with significantly higher PD-L1 expressions than HPV-negative cases for CPS for biopsy and resection specimens, and for TPS in resection specimens (p <0.05) ( Figure 4 ).

The Kruskal-Wallis test revealed significant differences in both CPS and TPS when comparing preoperative biopsies with surgical resections (p<0.05 for both). Similarly, significant differences were found between surgical resections and metastatic lymph nodes (p<0.05 for both). No statistically significant differences were observed for CPS and TPS between preoperative biopsies and metastatic lymph nodes.

When comparing paired primary tumors (resection specimens) and metastatic lymph nodes, the overall categorical concordance rate for CPS was 72.1% (49/68). Higher CPS values were observed in the primary tumor in 25.0% (17/68) of cases, while higher CPS was found in the lymph node in 2.9% (2/68) of cases. For TPS, the overall concordance rate was 47.1% (32/68), with higher values in the primary tumor and lymph node in 38.2% (26/68) and 14.7% (10/68) of cases, respectively ( Table 2 ).

Figure 5 illustrates case-wise PD-L1 expression trends in CPS and TPS. Average CPS scores in surgical specimens were approximately 20 points higher than corresponding TPS values. One case showed consistently low PD-L1 expression (TPS <1%) across all specimens. Inverse expression patterns, where surgical specimens had lower PD-L1 expression than biopsies or lymph nodes, were observed in a subset of cases (highlighted in orange and blue in Figure 5 ).

Across all specimen types, CPS consistently exceeded TPS, with an average difference of approximately 20 units. This discrepancy reflects the contribution of PD-L1-positive immune cells incorporated into CPS but excluded from TPS. Prior studies have similarly suggested that CPS better represents the tumor microenvironment (11), particularly in surgically resected specimens where immune infiltrates are more comprehensively sampled (12). Notably, inter-specimen concordance was higher for CPS than for TPS (72.1% vs. 48.5%), supporting the notion that CPS provides a more stable and reproducible measure across tumor compartments (13, 14), even though intratumoral heterogeneity remains.

Several studies have highlighted the challenges in evaluating PD-L1 expression in HNSCCs, including variability in expression levels and the difficulty of selecting representative samples (13, 14). Karpathiou et al. reported that PD-L1 expression may decline in archival specimens, potentially compounding heterogeneity in reported results (15). Furthermore, PD-L1 assessment after chemotherapy and/or radiotherapy has yielded conflicting outcomes across studies (16–18).

In other malignancies, such as lung cancer, the consistency of PD-L1 expression between primary tumors and metastases has also been examined. Gradecki et al. observed high concordance between core biopsies and resections in NSCLC, particularly in specimens with diffuse staining involving >50% of tumor cells (19). In contrast, more recent investigations have underscored significant variability in PD-L1 expression across specimen types in both early and advanced NSCLC (6, 20).

Several unique cases were identified. One tongue tumor demonstrated uniformly negative PD-L1 expression across all specimens, suggesting an inherently PD-L1-negative phenotype. Of note, a few cases (oropharynx n=1, oral cavity n=2) displayed inverse patterns, with negative surgical resections but corresponding biopsies or lymph nodes strongly positive. This may reflect heterogeneity of tumors, or clonal selection capable of immune evasion in metastatic sites. These unusual patterns underscore the need for further investigation into the biological behavior of such tumor and their potential clinical implications.

Despite these insights, this study has several limitations. First, it is a retrospective, single-institution study with a relatively small sample size, and the inclusion of a few older archival samples, which may limit generalizability of findings. Second, interobserver variability in PD-L1 scoring, especially for immune cell staining, was not assessed. Third, lymph node specimens collected after resection were not separately analyzed but included as a pooled group, without distinction between normal and metastatic lymph nodes, that could have provided further insight for metastatic potential of tumors. Fourth, we did not correlate PD-L1 expression with clinical outcomes, such as response to immune checkpoint inhibitors, which would be necessary to validate the predictive value of these findings.