Transcriptomic subgrouping of the four SGC subtypes
Transcriptomic subgrouping of the four SGC subtypes revealed distinct patterns. Through a transcriptomic analysis of 94 RNA-seq samples representing four major types of SGCs (18 ACC, 40 MECA, 16 SDC, and 20 MEC; Fig. 1), PCA-based transcriptomic similarity identified two major subgroups: SDC and MEC (Group 1) and ACC and MECA (Group 2) (Fig. 1A). Group 1 SGC subtypes largely shared an overall transcriptomic profile, whereas Group 2 SGC subtypes overlap partially. Specifically, MECA showed larger transcriptomic variance, implying higher intrinsic heterogeneity. The two groups can be mainly differentiated by their second principal component (PC2, y-axis of Fig. 1A), whereas the first principal component (PC1, x-axis of Fig. 1A) was more informative for determining heterogeneity within a group. Similarly, hierarchical clustering of transcriptomic expression showed that ~ 92% of Group 1 was SDC and MEC, whereas ~ 95% of Group 2 was ACC, MECA (Fig. 1B). These results suggest the presence of a higher-level grouping pattern among the SGCs.
Gene-level mapping of the principal components (PC1 and PC2) revealed the functional factors that contributed to subgroup-specific characteristics (Fig. 1C). Among the 50 cancer hallmark gene sets of eight functional categories [28], only seven gene signatures in the immune category (allograft rejection, coagulation, complement, interferon alpha response, interferon gamma response, IL-6/JAK/STAT3 signaling, and inflammatory response) were consistently distinguish between the two groups when mapped to PC2 (Fig. 1C) (mean GSVA score = 1.16). PCA-based clustering and immunological characteristics are major factors for SGC subtyping.
Immune profiling of the four SGC subtypes
We calculated the overall immunity of the four SGC subtypes using computational prediction (ESTIMATE [14]) (Fig. 1D). We found that the normalized immune scores were highly distinctive among the two groups (Group 1: SDC = 0.186 and MEC = 0.2 vs. Group 2: ACC = -1.479 and MECA = -1.011; Wilcoxon Rank Sum p-value < 0.001). Compared to those of the 33 TCGA pan-cancer datasets, the immune scores of Group 2 subtypes were the lowest and they were classified as immune-low cancers. In contrast, the immune scores of Group 1 subtypes were in the 69th percentile, similar to that of cancers with potential immunogenicity, including lung squamous carcinoma [29, 30] and mesothelioma [30, 31] (Fig. 1D). SGC subtypes within subgroups have similar immune scores.
A detailed analysis of the TIME features confirmed the immunological differences between the two groups (Figs. 1E–I). ACC and MECA have lower immune infiltration and three T-cell activity-related (T-cell infiltration, cytolytic activity, and antigen presentation machinery) scores than SDC and MEC (Wilcoxon Rank Sum p-value < 0.0001); however, the scores were not different within the subgroups (Fig. 1E–H). Only the angiogenesis score further differentiated between ACC and MECA (Fig. 1I). Moreover, correlation analysis showed that T-cell population and cytolytic activity are positively correlated and distinct for each group (Fig. 1J), indicating that T-cell-mediated immunity contributes to the immunological differences between the groups.
We further examined individual cell type TIME landscape profiles using six different algorithms: xCell [17], TIMER [18], quantiseq [19], MCP counter [20], Epic [21], and CIBERSORT [22] (Fig. 1K). We observed an overall increase in the number of cells involved in both innate and adaptive immunity against SDC and MEC, including cytotoxic T-cells (CD8+), activated NK cells, B-cells, and macrophages. In contrast, uncharacterized cells, which are considered malignant cells, and common lymphoid/myeloid progenitors showed the opposite or no correlation. These results confirmed the immune-low and immune-high features of Groups 1 and 2 subtypes, respectively, regarding both innate and adaptive immunity.
Association between histological origin and immunogenicity of SGC subtypes
We examined the association between the histological origin and the immunogenicity of the identified subtypes. Previous studies have shown that SGC originated from two of the four major substructures in the salivary gland (Fig. 2A): ACC and MECA are intercalated duct (ID)-derived and SDC and MEC are excretory duct (ED)-derived [32–34]. Cell type decomposition analysis confirmed that the samples used in this study had similar origins (Fig. 2B). In addition, we found that the ACC and MECA samples had a higher proportion of myoepithelial cells (Wilcoxon Rank Sum p-value < 0.001), whereas the SDC and MEC samples exhibited abundant fibroblasts, immune cells, including macrophages and dendritic cells (DCs), and ductal epithelial cells (Wilcoxon Rank Sum p-value < 0.001). These results are consistent with the known cellular compositions of ID (higher rate of cellular division and stem cell properties) and ED (higher number of professional antigen-presenting cells (APCs)) [35–38].
We further analyzed whether histological traits could be the source of the different immunogenicity of the SGCs. The computationally predicted score [25] showed higher cancer stemness in the ID-derived subtype than in the ED-derived subtype (Fig. 2C). Moreover, we observed a positive correlation (R = 0.5, p < 0.001) between intercalated duct cells, which are known to exhibit stem cell properties in salivary glands and potential stemness to differentiate into various glandular cell types, particularly ID-derived subtypes (Fig. 2D). Cell type associations revealed that cancer stem cells fractions correlated negatively with the majority of activating immune cells, including CD8 + T, cytotoxic, and NK T-cells, and positively with suppressive immune cells, including Th2 cells, myeloid-derived suppressor cells, tumor-associated macrophages, and naïve CD8 + T-cells (Fig. 2E). These findings suggest that the higher proportion of stem cells in ID-derived cancer contributed to the abundance of CSC during tumorigenesis, promoting immune escape through myeloid progenitor-derived cells [39, 40].
Pathway-level analysis revealed the potential underlying mechanisms of ID-derived cancer cell stemness (Fig. 2F). GSEA showed the enrichment status of the Wnt (Combined FDR = 1.2e-08), Hippo (4.3e-02), and Smoothened (2.4e-02) oncogenic signaling pathways, along with the regulation of stem cell differentiation (Combined FDR = 1.5e-04) in ID-derived subtypes. In addition, a lower tumor suppressor gene (TSG) score in the ID-derived subtypes (Fig. 2G,) and its strong negative correlation with cancer stemness (Fig. 2H) suggested that TSG alleviation may contribute to the proliferation of CSCs as reported in a previous study [41]. Overall, our analysis showed that distinct immunological traits are affected by the histological origin of the tumors, primarily differentiating SGCs into ID- and ED-derived subtypes.
Prediction of subtype-specific immune evasion mechanisms of SGC
To identify subtype-specific T-cell functionality, we performed a gene set analysis based on TIDE (Fig. 3). CD8 and Merck18 T-cell inflamed signatures, interferon-gamma (a pro-inflammatory molecule), and dysfunctional T-cell signatures (CD274 and T-cell dysfunction scores) increased in the ED-derived subtypes (Fig. 3A). In contrast, the ID-derived subtypes showed increased myeloid-derived suppressor cells and M2-polarized tumor-associated macrophages, which are associated with T-cell exclusion (Fig. 3A). Moreover, the T-cell exclusion score was higher in ID-derived subtypes. Additionally, mixed T-cell functionality was observed in MECA. Overall, these findings provide important insights into tumor immune dysfunction and exclusion mechanisms in different histological subtypes, with ED- and ID-derived subtypes associated with increased dysfunctional T-cell population and excluded T-cell population, respectively.
Moreover, we analyzed the T-cell receptor beta-chain-CDR3 sequences to investigate the ability of T-cells to recognize antigens in the two groups (Fig. 3B) and found that the ED-derived subtype had higher TCR diversity, which was positively correlated with the T-cell infiltration score (Fig. 3C). These findings suggest that the ED-derived subtype exhibits increased sensitivity to antigens and more active T-cell infiltration, indicating that dysfunctional T-cells are not caused by low TCR diversity. In contrast, the ID-derived subtype exhibited low TCR diversity in the excluded T-cell population, indicating a lack of T-cell functionality.
To explore the underlying mechanisms of the observed differences in T-cell functionality between the ED- and ID-derived subtypes, we further analyzed the expression of molecules involved in antigen presentation and recognition and the co-inhibitory molecules that suppress T-cells (Fig. 3D) and found that ED-derived subtypes had a higher expression of signal 1 molecules, including MHC I, MHC II, and TCR, and signal 2 co-stimulatory molecules, such as CD80/CD86, CD40, OX40L, ICOSL, CD28, CD40L, OX40, and ICOS. T-cell-suppressing co-inhibitory molecules were also elevated in ED-derived subtypes. The high expression of co-inhibitory molecules in the ED-derived subtypes, including GAL9, TIM3, CTLA4, CD80, PD-L1, and PD-1, was consistent with the results shown in Fig. 3A (for example, CD274). These findings suggest that co-inhibitory molecules may influence dysfunctional T-cell mechanisms, and that weak signal 1 and co-stimulatory molecules alone may be insufficient to activate the T-cell priming process in ID-derived subtypes. This could result in the exclusion of T-cell populations with low tumor specificity.
Furthermore, we observed a correlation between immune cells, suggesting a malfunction in antigen processing and the presentation mechanism of professional APCs (Fig. 3E). This led us to investigate the specific cell types that may contribute to the observed elevation of co-inhibitory molecules and dysfunctional T cells in ED-derived subtypes despite the high expression of signal 1 and co-stimulatory molecules.
We found that macrophages and DCs are professional APCs that forms cluster based on their antigen-presenting machinery scores and similar phenotypes, including tolerogenic and immature DCs, M1- and M2- macrophages (red boxes). Additionally, our analysis revealed that ED-derived subtypes were associated with a suppressive phenotype of professional APC and that the anti-inflammatory subtypes of macrophages (M2-macrophages) and DCs (immature and tolerogenic [tolDCs]) were strongly associated with immunosuppressive factors, such as co-inhibitory signaling molecules and regulatory T-cells.
Both pro-inflammatory M1-[42] and anti-inflammatory M2-polarized macrophages [43] were elevated in the ED-derived subtypes; however, the ratio of M2 to M1 in the ID-derived subtype was higher (Fig. 3F). This suggests that ID-derived subtypes are primarily affected by the anti-inflammatory phenotype of macrophages compared to ED-derived subtypes. Additionally, DCs are another type of professional APC that maintain the balance between adaptive immunity and the tolerogenic response to tumors during antigen processing and presentation. Our results showed that there were higher numbers of mature and activated DCs and immature and tolDCs in the ED-derived subtypes than in the ID-derived subtypes (Fig. 3G).
Tregs and tolDCs are crucial in creating an environment that induces immunosuppression. It has been reported that Tregs inhibit the DCs co-stimulatory signaling molecules and inhibit DC-mediated pro-inflammatory cytokines, IL-12, and TNF-a [44]. In contrast, Tregs upregulate anti-inflammatory cytokines TGF-β and IL-10, which have tolerogenic properties [45, 46]. Anti-inflammatory factors (TGF-β, IL-10, IL-6, VEGF, IDO, and PGE2), which contribute to immunosuppressive environment are highly expressed in ED-derived subtypes than in ID-derived subtypes (Fig. 3H).
Summarily, we presented a hypothetical model that explains the malfunctioning of professional APCs and the formation of dysfunctional T-cell populations in ED-derived subtypes. Tumor-derived factors inhibit the differentiation and maturation of DCs and promote the accumulation of iDCs [44, 47]. Continuous stimulation of naïve T cells by increased levels of iDCs and tumor-derived factors leads to the activation of Tregs, which subsequently promotes the differentiation of iDCs into tolDCs. This creates an environment that increases co-inhibitory signaling molecules, facilitated by the ability of iDCs to secrete IL-10 and many other immunosuppressive factors. Furthermore, there is a bidirectional relationship between tolDCs and Tregs: Tregs affect the activation of tolDCs and activated tolDCs can increase Tregs through positive feedback loops [44] as illustrated in Fig. 3I.