scRNA-seq reveals extensive intratumoral transcriptional heterogeneity of CTA genes in melanoma and lung cancer

Consistent with previous findings,6 22–29 immunohistochemical analysis of selected CTAs (GAGE, MAGE-A, MAGE-C1, NY-ESO-1, PRAME and SSX) revealed a highly variable expression in melanomas (figure 1A and online supplemental figure 1) and lung cancers (online supplemental figure 1). To gain a comprehensive understanding of CTA expression patterns in these malignancies, we conducted targeted CTA scRNA-seq expression profiling of tumor cells. Initially, we examined seven primary cell cultures derived from melanoma tumors, representing a diverse cohort of patients, both male and female, aged 44–85 years (online supplemental table 1). The melanoma cells were collected from primary tumors (YUDOSO, YUPEET, YUTOGS and WW165) and metastatic lesions (YUKSI, YUSIT, YUSIF) harboring different driver mutations (online supplemental table 1) and their identity and purity were verified using a panel of melanocyte markers (online supplemental figure 2). Analysis of a panel of 81 CTA genes revealed a variable expression of CTAs across tumors, with the number of expressed genes ranging from 26 to 37 (figure 1B). Notably, every cancer cell derived from these seven distinct melanomas expressed one or more CTA genes (figure 1C), highlighting the prevalence of CTA expression in melanoma and the potential of these antigens as targets for immune targeting of this disease. However, a noticeable variation in the average number of CTA genes expressed per cell was observed among the tumors, ranging from 7 (YUSIV) to 21 (YUDOSO; figure 1D).

Figure 1

Intratumoral heterogeneity in the expression of CTAs in melanoma and lung cancer. (A) Immunostaining of selected CTA genes in low passage cancer cell cultures from primary melanomas and metastases (brown). Counterstain: hematoxylin (blue). Representative pictures are shown. Magnification×25. (B–D) scRNA-seq analysis of the expression of 81 CTA genes in low passage cancer cell cultures derived from primary tumors (YUDOSO, YUPEET, YUTOGS, WW165) or metastases (YUKSI, YUSIV, YUSIT) of melanomas. Plots show the number of CTA genes expressed in at least 1% of the cancer cells of each tumor (B) the frequency of CTA-positive cells within each tumor (C) and the frequency of CTA genes expressed in individual cancer cells of each tumor (D). (E–G) scRNA-seq analysis of the expression of 78 CTA genes in cancer cells from primary LUADs of five patients (LUAD1, LUAD2, LUAD3, LUAD4, LUAD5). Plots show the number of CTA genes expressed in at least 1% of the cancer cells of each tumor (E) the frequency of CTA-positive cells within each tumor (F) and the frequency of CTA genes expressed in individual cancer cells of each tumor (G). (H) scRNA-seq analysis of low passage cancer cell cultures derived from melanoma primary tumors and metastases. The dot plot shows the frequency and intensity (average expression) of CTA genes expressed in >10% of cells in at least one tumor. (I) scRNA-seq analysis of cancer cells from primary LUAD tumors of five patients. The dot plot shows the frequency and intensity (average expression) of CTA genes expressed in >10% of cells in at least one tumor. (J) Expression frequency of selected CTA genes among cancer cells of melanomas and LUADs. The expression frequencies of all expressed CTA genes are shown in online supplemental figure S1 and S3. CTA, cancer/testis antigens; LUAD, lung adenocarcinoma; scRNA-seq, single-cell messenger RNA sequencing.

There was no discernible difference in the average number of CTAs expressed between primary tumors and metastases (figure 1D). Within each tumor, the average number of CTA genes expressed in individual cells was significantly lower than the total number of CTA genes expressed in the tumor (figure 1B,D). Consistent with this, the expression of individual CTA genes was generally confined to subpopulations of cancer cells (figure 1H and online supplemental figure 3). For example, the prominent tumor target NY-ESO-1 (encoded by CTAG1B) was detected in three tumors (YUDOSO, YUTOGS and YUSIT) out of seven, with frequencies of positive cells at 1%, 48%, and 66% (figure 1J). Similarly, other CTAs with recognized therapeutic potential, such as SSX1, MAGE-A1, MAGE-A10, MAGE-C2, XAGE-1B and PAGE-1, also displayed substantial intratumoral heterogeneity in their expression (figure 1H,J and online supplemental figure 3). Overall, 64 out of 71 detected CTA genes consistently exhibited expression confined to subpopulations of tumor cells.

The remaining seven CTAs showed varying degrees of homogeneity in their expression across melanoma tumors, as illustrated in figure 1H,J and online supplemental figure 1. Particularly noteworthy was PRAME, which exhibited uniform expression in six of the seven melanoma primary tumor cultures (figure 1H,J). Furthermore, MAGEA2, MAGEA3, MAGEA6, and MAGEA12, constituting an evolutionarily conserved subcluster of MAGE-A genes, exhibited homogeneous expression in a subset of metastatic tumors (YUSIT and YUKSI; figure 1H,J and online supplemental figure 3). GAGE2 and GAGE12 were also homogeneously expressed in a subset of melanomas (YUDOSO and YUKSI) and were generally widely expressed in the examined melanomas (figure 1H,J and online supplemental figure 3). The intratumoral patterns of CTA gene expression, as uncovered through scRNA-seq analysis, were largely consistent at the protein level, as demonstrated through immunostainings of selected CTAs, including GAGE, MAGE-A, MAGE-C1, NY-ESO-1 and PRAME (figure 1A). This consistency strongly suggests that the observed differences in CTA expression between cancer cells is not due to temporal variations in gene transcription.

Subsequently, we characterized the intratumoral CTA expression patterns in primary LUAD from five patients, aged 54–8 years, representing both sexes (online supplemental table 1). Notably, the five LUAD tumors exhibited different molecular aberrations (online supplemental table 1). Cancer cells isolated from these tumors underwent focused scRNA-seq to analyze CTA gene expression. To confirm the purity of cancer cells, marker genes for different types of normal lung cells were included in the analysis (online supplemental figure 4). Among the 78 CTA genes analyzed, between 4 and 36 were expressed in the five tumors, respectively (figure 1E). In contrast to the melanoma tumors, there was little overlap in the set of genes activated in the different tumors (figure 1I). The prevalence of CTA expression in LUAD was comparatively more limited than that in melanoma, as the proportion of cells expressing at least one CTA was 100%, 98%, 65% 50% and 21%, respectively (figure 1F). Each cancer cell expressed only a minor subset of the CTA genes expressed by the whole LUAD tumor (figure 1E,G), and most CTAs were limited to subpopulations of cancer cells in tumors (figure 1I,J; online supplemental figure S5). Uniform expression of CTA genes, including GAGE, PRAME and GTSF1, was only observed in a single LUAD tumor (LUAD5; figure 1I and online supplemental figure 5). XAGE-1B, a proposed target in LUAD,33–36 was expressed in up to 84% of the cells, whereas BAGE-2 and PAGE-5 were expressed in up to 41% and 26% of the cells, respectively (online supplemental figure 3). The remaining CTAs, including NY-ESO-1 and MAGE-A family members, were found to be expressed in less than 20% of the cancer cells within the tumors (figure 1I,J and online supplemental figure 5).

Overall, these findings unveiled a remarkably high level of transcriptional heterogeneity among CTA genes in melanoma and LUAD tumors, with uniform CTA expression being a rare phenomenon observed only found in melanoma.

Lack of coordination of CTA expression in tumors generates mosaics of tumor cells with diverse cancer/testis antigen profiles

The co-expression of CTAs has been demonstrated in tumors across cancer types.8 37 To further characterize the landscape of CTA gene expression within melanoma and LUAD tumors and identify potential associations in CTA expression, we conducted a UMAP dimensional reduction analysis of the scRNA-seq data. This analysis identified cellular subsets with differential expression of CTAs and some correlation between the expression of specific CTAs (Figure 2A-E). Nevertheless, these subsets displayed poor separation and a significant overlap in CTA expression, indicating a lack of robust transcriptional disparities and coordinated CTA expression among cancer cells within the same tumor. Indeed, minimal overlap in the expression of distinct CTA gene families was observed in both melanoma and LUAD tumors, consistent with expectations based on chance (figure 3A). Even among distantly-related CTAs, such as those encoding the different types of MAGE families, or CTAs expressed during the same stages of male germ cell maturation (eg, GAGE, MAGE-A, CTAG1B and SSX in spermatogonia),6 23 27 38 co-expression was lacking. This suggested a high degree of variation in the CTA expression profiles among cancer cells within tumors, indicating that melanomas and lung cancers are composed of complex mosaics of tumor cells with diverse antigen profiles. Consistent with this, a direct comparison of the CTA expression profiles of individual cancer cells demonstrated a very high level of diversity (figure 3B,C). Lung cancers were less complex in their CTA expression than melanomas reflecting the lower prevalence of CTA expression (figure 3B,C).

Figure 2

Lack of strong transcriptional disparities in CTA gene expression among cancer cell subsets within tumors. (A–D) Clustering analysis of single-cell CTA gene expression in representative melanomas: YUDOSO (A) YUPEET (B) and YUSIV (C) and LUAD: LUAD1 (D) and LUAD5 (E). Left: UMAP dimensional reduction plots showing identified cell clusters within tumors. Middle: dot plots showing the frequency and relative expression levels of CTA genes with expression >10% in at least one cluster. The fraction of cells within each cluster is indicated. Right: feature plots showing the expression (log1p) of selected CTA genes in identified cell clusters. CTA, cancer/testis antigens; LUAD, lung adenocarcinoma; UMAP, Uniform Manifold Approximation and Projection.

Figure 3

Lack of co-expression among distinct CTA gene families in tumors generates mosaics of cancer cells with diverse antigen profiles. (A–C) Single-cell CTA gene expression in representative melanoma and LUAD tumors. Data for selected tumors are shown. (A) Venn diagrams depict the intratumoral overlap in the expression of selected CTA genes of different gene families. (B) Heat maps show CTA gene expression profiles of individual cancer cells. Gene not expressed=0. Gene expressed=1. Clustering analysis was performed to unveil cancer-cell subsets exhibiting comparable gene expression profiles. (C) Principal component analysis of single cells based on the expression of CTA genes. CTA, cancer/testis antigens; LUAD, lung adenocarcinoma.

To comprehensively investigate potential transcriptional associations among CTAs, we conducted pairwise comparisons of their expression patterns (figure 4A). While the majority of CTAs displayed no significant correlation in their expression, distinct subsets exhibited co-expression within tumors. This intriguing correlation was particularly evident among highly evolutionarily and structurally linked genes, suggesting potential regulation by shared transcription factors or preservation within the same regulatory compartments. For example, MAGEA2, MAGEA3, MAGEA6, and MAGEA12, forming a distinct gene cluster of MAGE-A genes on the X chromosome (figure 4B), demonstrated a remarkable level of overlapping expression among YUPEET tumor cells (figure 4C,D), which was not shared by other MAGE genes (figure 4E). A similar pattern was observed for MAGEA2, MAGEA3, MAGEA6, and MAGEA12 in other tumors (figure 4F) and with other-related CTA genes, such as members of the PAGE family (figure 4G). Infrequent co-expression was also noted among genes from different gene families (figure 4A). Importantly, the identified correlations among CTA-gene subsets were not consistent across tumors. This divergence can be attributed to disparities in chromatin organization among tumors, leading to variations in the regulatory compartments governing CTA genes.

Figure 4

Co-activation of structurally and evolutionary linked CTA genes. (A) Pairwise co-expression analysis of CTA genes in selected melanomas and LUADs. Scales depict the level of co-expression (observed co-expression divided by expected co-expression based on the frequency of expression within tumors). (B) Schematic representation of MAGEA gene family genes on chromosome X. The MAGEA2, MAGEA3, MAGEA6 and MAGEA12 subcluster is highlighted in green. (C) Feature plots of genes belonging to the MAGEA2, MAGEA3, MAGEA6 and MAGEA12 subcluster. Yellow indicates cells with co-expression of indicated genes. (D) Venn diagram depicting overlap in expression of the MAGEA2, MAGEA3, MAGEA6 and MAGEA12 subcluster in the Yupeet melanoma tumor. (E) Venn diagram depicting overlap in expression of distantly-related MAGE genes. (F) Venn diagram depicting overlap in expression of the MAGEA2, MAGEA3, MAGEA6 and MAGEA12 subcluster in the LUAD1 tumor. (G) Venn diagram depicting overlap in expression of PAGE genes in the LUAD2 tumor. CTA, cancer/testis antigens; LUAD, lung adenocarcinoma.

Our findings underscore a remarkable level of complexity in the expression of CTAs in melanomas and lung cancers. The general absence of co-expression of CTAs with coordinated expression in spermatogenesis suggests a lack of bona fide germ cell transcriptional programs in tumors, contrary to previous proposals.1 Instead, it suggests that the activation of CTA genes in cancer cells appears to be largely guided by stochastic processes.