Immunological subgrouping of MECs
We initially profiled the tumor microenvironment in 20 patients with MEC from Yonsei Head and Neck Cancer Center (Table 1) using ssGSEA with respect to 24 immune and stromal cell types (Figure 1A) (see Materials and methods for analysis procedures and Figure S1 for overall workflow). Hierarchical clustering of the 20 MECs with the inferred cell abundance revealed two subgroups: eight tumors with high immune cell infiltration (referred to as immune-hot) and 12 with low infiltration (immune-cold). The infiltration patterns were concordant among most immune cell types, including myeloid progenitor cells, NK/T lineage cells, and stromal cells, with only a slight variation in B and mast cells. In addition, we confirmed that these clusters were consistently reproduced in an independent analysis with a different measurement algorithm (ESTIMATE (16), see Materials and methods), showing clear immunological discrepancies between the two subgroups (Figure 1A and Table S1) (Wilcoxon rank-sum test, P = 4.5 × 10-4 for immune score and P = 6.0 × 10-4 for stromal score, see Materials and methods). These results represent immunological heterogeneity and the existence of subgroups of MECs.
Further analyses confirmed distinct immune activities between the subgroups. The expression levels of CD8A and CD8B were significantly higher in the immune-hot MECs (Figure 1B, Wilcoxon rank-sum test P = 1.1 × 10-3 for CD8A and P = 3.0 × 10-3 for CD8B) than in the immune-cold MECs, demonstrating distinct activities of cytotoxic T-cells. Similarly, scores for TIS (i.e. the average value of central memory, effector memory CD4+ and CD8+ T cells, and Th1, Th2, Th17, and Treg cells), and overall IIS (17) were high in immune-hot MECs (Wilcoxon rank-sum test P = 3.2 × 10-5 for TIS and P = 6.4 × 10-5 for IIS). Moreover, we noted elevation in the CYT score (P = 4.1 × 10-3) (18) and IFN-γ score (P = 3.0 × 10-4) (36) (see Materials and methods) for the immune-hot group (Figure 1C). After observing high levels of immune infiltration and immunity in the immune-hot group, we further characterised the immunophenotypes of the two MEC subgroups, focusing on the correlation between immune-cell infiltration and cytotoxic functions. IFN-γ, one of the final products of antitumor T-cell activity, was strongly positively correlated with the TIS (Pearson correlation coefficient r = 0.76, P = 1.0 × 10-4) and overall IIS (r = 0.84, P = 3.1 × 10-6; Figure S2B). Together, these results suggested that immune cell infiltration and activity level-based MEC stratification exhibited two robust immunophenotypic subgroups.
Subgroup-associated immunophenotypic characteristics
We then investigated the phenotypic characteristics that confer immunological differences between the immune-hot and -cold subgroups. We initially noted that the TMB and neoantigen burden did not differ between the subgroups (Wilcoxon rank-sum test, P = 5.7 × 10-1 and P = 6.8 × 10-1 for immune-hot and -cold subgroups, respectively). The frequency and pattern of known MEC prognosis associated marker (37), CRTC1-MAML2 fusion, was not significantly different between immune subgroups (6/8 and 6/12 in immune-hot and -cold, respectively, Fisher's exact test, P = 3.7 × 10-1), and, no potential immunogenic neoantigens were predicted by the CRTC1-MAML2 fusion event (see Materials and methods). The schematic diagram of fusion gene structure is illustrated Figure S3. Based on these results, we presumed that the immunological differences between subgroups resulted from the machinery or regulatory mechanisms in the immunity cycle, rather than genome-level aberrations.
Next, to profile the activities of subgroup-associated immune-regulatory mechanisms, expression levels of 78 immunomodulator (IM) genes (7) were used to infer the status of seven immune regulatory categories: antigen presentation, cell adhesion, immune checkpoint receptor, ligand, co-inhibitor, co-stimulator, and others (Figure 2). We found that most of the IM genes and the categories, particularly cell adhesion (SELP, ITGB2, and ICAM1), antigen presentation (HLA Class I/II), co-inhibitor (PD-L1, PD-L2, BTN3A1, and BTN3A2), co-stimulator (CD80 and CD28), and ligand (CXCL9, CXCL10, IL2, IL10, OX40L, CD40LG, and IFNG), were consistently elevated in the immune-high MECs (all, P < 5.0 × 10-2), which implied that the immunologic differences may have risen at the early stages of the immunity cycle, such as the antigen-presenting machinery (APM). Of note, level of the APM (36) was significantly low in the immune-cold subgroup (P = 7.3 × 10-3) compared to that in the immune-hot subgroup (Figure S2A), and was positively correlated with the TIS (r = 0.56, P = 9.5 × 10-3) and IIS (r = 0.67, P = 1.1 × 10-3) (Figure S2C). These results indicate that the higher cytolytic potential in the immune-hot MECs could be a result of the higher level of APM expression with a cytolytic feed-forward loop along with an increased amount of immune cell infiltration, thereby leading to elevated IFN-γ levels and anti-cancer immune activity.
Transcriptomic landscape and hub genes of the immune subgroups
To illustrate the biological properties of immune-hot and immune-cold MECs, we conducted a functional analysis in the biological contexts based on gene expression profiles. We identified 1518 DEGs between the two subgroups, 1076 of which were upregulated in the immune-hot and 442 in the immune-cold MECs (see Materials and methods). Gene ontology (GO)-based enrichment analysis (see Materials and methods) uncovered a global overrepresentation of innate and adaptive immune-related terms in the immune-hot subgroup, including T-cell activation, regulation of leukocyte activation, and regulation of lymphocyte activation (all, adjusted P < 1.0 × 10-2) (Figure S4A top). Likewise, cancer hallmark-based GSE analysis (see Materials and methods) revealed enrichment of inflammatory response and interferon-gamma response in the immune-hot MECs, confirming increased antitumor immune activity (Figure S4B). For the immune-cold MECs, ion transport-related terms, epidermis development, cornification, and chloride transport from GO analysis (all, adjusted P < 1.0 × 10-2) (Figure S4A, bottom), and metabolism-related cancer hallmark, such as fatty acid metabolism and adipogenesis, were enriched (Figure S4C).
Next, to improve understanding of factors that are shaping heterogeneity of immune state in MEC immune subgroups, we sought to assess tumor-intrinsic traits of them by comparing with matched adjacent normal tissues. First, we identified 893 DEGs with tumor-specific expression (see Materials and methods). Of these, 467 DEGs were over- or under-expressed from immune-hot MECs tumor and 326 from immune-cold MECs (full list available in Tables S2-5). As the top-ranked upregulated genes observed only in immune-hot MEC, CXCL13, which is a B cell chemoattractant and an important factor in TLS formation and initiation of secondary lymphoid organ development (38) (adjusted P = 1.55 X 10-10 and log2 fold change 5.89) and DUOX2, responsible for the innate immune response (39) (adjusted P = 2.74 X 10-9 and log2 fold change 5.94), were observed, whereas MUC6 (mucin 6) was top ranked in immune-cold MEC (adjusted P = 1.34 X 10-6 and log2 fold change 5.47). POSTN1 (periostin), which affects tumor microenvironment remodelling during tumor progression, was one of the top-ranking genes overexpressed in both immune subgroups (log2 fold change > 5). Then, also, to integrate biological impact of isoform-level changes in forming polar tumor microenvironment, we measured the isoform switches as mentioned in Kristoffer Vitting-Seerup et al. (32) and we identified 173 and 85 genes that underwent isoform switching in tumorigenesis in the immune-hot and -cold MECs, respectively. Among the 640 genes in immune-hot MECs, we identified T-cell regulatory genes (ADA, IL4R, and NFKBIZ) and lipid metabolism regulatory genes (MLXIPL, LEP, and FASN). From 411 genes from immune-cold MECs, we observed collagen and bone forming related genes (MMP13, COL10A1, and ITGB6) (see Materials and methods; full list available in Tables S6 and S7). These gene- and isoform-level intrinsic features provide a broad understanding of the distinctive intrinsic features that reflect the heterogeneous immunophenotypic state of each immune MEC subgroup.
To better study the functional significance of subgroup-specific genes at the system level, we constructed Protein-protein interaction (PPI) network by the union of DEGs and significant isoform-switching genes (Table S8), which formed 610 nodes and 1429 edges, and 403 nodes and 896 edges for immune-hot and -cold subgroups, respectively (Table S9). We then selected hub genes, important nodes with several interaction partners, with seven different topological analysis algorithms provided by cytoHubba (34), which appeared more than twice in the union of all top 20 genes from each algorithm, resulting in 35 genes (15 upregulated and 20 downregulated) and 31 genes (2 upregulated and 29 downregulated; Table S10) for immune-hot and -cold subgroups, respectively, and then searched for their functionality.
In the immune-hot MECs, the genes that are related to chemokine regulation and ECM (POSTN, COL11A1, MMP13, LUM, VCAN, and BGN) and lipid metabolism regulation (SCD, LPL, MLXIPL, PCK1, and DGAT2) were selected as hub genes (Figure 3A). Overexpression of the ECM-related genes and lipid metabolism-related genes in cancers is associated with cancer progression, immune suppression, and tumor aggressiveness (40, 41). Here, as expected, ECM-related genes were over-expressed in the immune-hot MECs; however, lipid metabolic-regulator genes were under-expressed in the immune-hot MECs, and this particular feature, downregulated lipid metabolism, was also captured in above GSE analysis. Based on the aforementioned findings and inverse correlation between signature scores of fatty acid metabolism and adipogenesis obtained from cancer hallmark genesets with cytotoxic activity (Spearman correlation coefficient r = -0.49, P = 3.1 × 10-2 and r = -0.55, P = 1.4 × 10-2, respectively; Figure 3B), we surmised that under-expressed lipid metabolic regulator genes in immune-hot MEC samples could be novel molecular characteristics and may be linked to agonistic anti-tumor immune activity in this MECs, as their overexpression is often related to a decrease in immune activity (41, 42). In the immune-cold MEC samples, all selected genes, except POSTN and RUNX2, were downregulated and are known to function as chemokines and their receptors (CXCR1, CXCR2, CCL2, CCR3, and CCR10) and in leukocyte migration and adhesion (ELANE, LEP, and SELL) (Figure 3C).
Additionally, POSTN and RUNX2 were recurrently overexpressed in tumor samples from both MEC immune subgroups. POSTN participates in ECM structure formation and organisation (43, 44), and RUNX2 is a key transcription factor for osteoblast differentiation (45). At this point, we do not know their exact roles in MECs; however, based on our observations and previous studies, we conjecture that they could play a critical role in MEC ECM formation. In this analysis, we identified unique and common candidate molecular characteristics that were predicted to play a vital role in immune regulation and intrinsic tumor characteristics of two immunophenotypically distinct MECs. Further studies are warranted to better understand their role in MECs and their potential use as targets for novel classification and therapeutic implications.
Molecular features associated with immuno-oncology therapeutics
Based on the subgrouping and immunological analyses, we explored potential subgroup-specific molecular features that can be employed in the application of current immunotherapy to MECs. First, we examined the efficacy of immune checkpoint inhibitors (ICIs). The analysis of mRNA expression identified significant elevation of PD-1 and PD-L1 levels in the immune-hot MECs (P = 9.6 × 10-3 and P = 3.1 × 10-2) (Figure 4A). In addition, co-inhibitor genes, including SLAMF7 (P = 1.0 × 10-2), BTN3A1 (P = 1.0 × 10-4), and BTN3A2 (P = 9.0 × 10-3), were highly expressed in the immune-hot subgroup. Extended analysis of a collection of immuno-oncological therapy biomarkers (46) with the immune-hot subgroup-specific DEGs identified elevated levels of additional immune checkpoint molecules (TIGIT, 4-IBB, TIM-3, PD-L2, and CTLA-4), T-cell targeted immunomodulators (ICOS, IL10, and OX40), adaptive immune systems (MICB), and cell adhesion molecules (ICAM1) in the immune-hot MECs (Figure 4B and Figure S5A). These results suggest that at least a subgroup of immune-hot MECs underwent T-cell exhaustion by checkpoint molecules and could be sensitive to ICIs. In contrast, in the immune-cold MECs, carbonic anhydrase IX (CA9), a surrogate marker of hypoxia-related response (47), and the Wnt signalling signal transduction regulators Dickkopf-1 (DKK1), which functions to confer tumor growth and metastasis (48), Ring Finger Protein 43 (RNF43), cytokine signalling regulator RANK (TNFRSF11A), and cell tight junction-associated claudin 3 (CLDN3) were highly activated (Figure 4B and Figure S5B). This suggests that different or combined therapeutic strategies are needed as the subgroups based on immune activity show distinct immunosuppressive TME characteristics.