Patients

The patient cohort for this study was derived from our previously published work [29]. This retrospective study included 100 patients. Of these, 94 patents were clinicopathologically analyzed, and 6 were enrolled in the Visium analysis. All patients were diagnosed with progressive HNSCCs and underwent surgical resection without prior chemotherapy or radiation therapy at the Department of Otorhinolaryngology, Head and Neck Surgery, Kansai Medical University Hospital, between January 2009 and December 2024. This study was approved by the Institutional Review Board (approval number: 2023024). Additional demographic and clinicopathological characteristics are provided in Online Resource 1 and 2.

Construction of Tissue Microarrays (TMAs) for Clinicopathological Analysis Using IHC

In total, 1,040 cores were collected to construct a TMA for clinicopathological analysis using IHC. This included four cores from each of the 94 HNSCC samples, 94 adjacent SIN samples, and 62 adjacent NOM samples from FFPE-resected HNSCC tissues from 94 patients. Four HNSCC cores were collected from different invasive areas: two from superficial invasive areas (depth of invasion ≤ 5 mm) and two from areas of deeper invasion (depth of invasion > 6 mm). Each FFPE tissue block was sampled with 2.0-mm cores using a tissue-arraying instrument (Azumaya Corporation, Tokyo, Japan). We excluded carcinoma in situ (CIS) from the analysis, and all cores had ≥ 100 epithelial cells. Our previous study [29] provides a detailed summary of the results.

Histopathological and Clinicopathological Analysis for IHC Using TMAs

The clinicopathological features of the patients were described in our previous study [29] and in Online Resource 1. IHC analyses of the TMA tissue sections were performed using antibodies against B7-H4 (D1M8I, 1:200; Cell Signaling Technology, Inc., MA, USA), 73 − 10 (pre-diluted; Leica Biosystems, Newcastle Upon Tyne, UK), CD3 (PS1, pre-diluted; Nichirei Bioscience, Inc., Tokyo, Japan), CD4 (1F6, pre-diluted; Nichirei Bioscience, Inc.), and CD8 (G2B10, 1:20000; Proteintech Group, Inc. IL, USA). For B7-H4, antigen retrieval was performed using ethylenediaminetetraacetic acid or citrate buffer at 95 °C for 1 h, followed by detection using a Histofine Simple Stain MAX-PO® polymer detection system (#NIC-414131 F; Nichirei Bioscience Inc.) and visualization using diaminobenzidine. The 73 − 10, CD3, CD4, and CD8 were visualized as previously described [29].

The expression of B7-H4 and 73 − 10 was evaluated based on the tumor cell score [TC, also known as tumor proportion score (TPS)], immune cell score (IC), and combined positive score (CPS) [30,31,32]. Traditionally, these evaluation methods used the invasive carcinoma components as the numerator and denominator. However, this study used the non-neoplastic epithelium for NOM and the dysplastic components for SIN while maintaining the original method for invasive carcinoma components in HNSCC. TC, IC, and CPS values ≥ 1% were considered positive, and values ≤ 0% were considered negative. The mutually exclusive expression of B7-H4 and 73 − 10 was noted as follows: B7-H4+ and 73 − 10−− or B7-H4−− and 73 − 10+. The percentage of CD4+ and CD8+ tumor-infiltrating lymphocytes (TILs) was measured and classified as low (< 20%) or high (≥ 20%) based on CD3+ areas, and cores with both low CD4 and CD8 expression were classified as low immune-active, whereas all others were classified as highly immune-active. TC, IC, CPS, and TILs were assessed in all cores, and the highest values among the four cores for NOM, SIN, and HNSCC were recorded. To assess the utility of TC in biopsies, the B7-H4 TC of HNSCC was recorded from both superficial and invasive front cores. The PD-L1 TC was summarized from previous reports; the majority (88%, 65/74) showed positivity in both superficial and invasive front cores [29].

Construction of TMAs for Visium Spatial Transcriptomics Analysis

Visium ST assays were performed for a total of 18 TMA cores, including six HNSCC, six SIN, and six NOM cores from six patients with progressive HNSCC [sample 1 (S1) to sample 6 (S6)], as previously reported [29]. The clinicopathological characteristics of the patients are presented in Online Resource 2.

Workflow for the Visium Spatial Transcriptomics Analysis

Visium ST (103 Genomics, Pleasanton, CA, USA) analysis was performed for the spatial gene expression landscape in four annotated clusters: NOM, SIN, HNSCC, and non-evaluable epithelium, which were excluded from the ICI IHC assessment because they lacked nuclei and hemorrhagic area, as previously described [29] and summarized in Online Resource 3. For differential gene expression analysis, the log-fold change threshold was set to 0.25, and p-values < 0.05 were adjusted using Bonferroni correction to control for multiple testing. A minimum expression percentage (min.pct) of 0.1 was used to identify significantly differentially expressed genes in each cluster. The significance of mRNA expression was identified as log2 > 0.25 and p-value < 0.05, which were adjusted using Bonferroni correction for comparison with other groups in the same cases. Mutually exclusive mRNA expression was defined as VTCN1+ and CD274−or VTCN1− and CD274+. The log2 fold changes obtained from the differentially expressed genes and up-regulated top 50 genes (log2 > 0.25 and p-value < 0.05) were used for GO analysis with ToppGene (https://toppgene.cchmc.org/, accessed: 2024/10/05).

Statistical Analyses

To measure the mutual exclusivity of B7-H4 and PD-L1 in HNSCC, chi-squared analysis was performed between the mutually exclusive pattern (+) group [B7-H4+/PD-L1- and B7-H4-/PD-L1+) and the (-) group [B7-H4+/PD-L1 + and B7-H4-/PD-L1-]. The concordant expression of B7-H4 TC between the superficial and invasive front cores was examined using the kappa value. Fisher’s exact test was used to determine the correlations between B7-H4 and PD-L1 expression and between B7-H4 expression and clinicopathological features. The e-cutoff value for tumor immune activity was calculated using the area under the curve against overall survival. Statistical analysis was performed using IBM SPSS software (v20.0; IBM Corp., Armonk, NY, USA). Significance was set at p < 0.05.