MAEL in human cancers and implications in prognostication and predicting benefit from immunotherapy over VEGFR/mTOR inhibitors in clear cell renal cell carcinoma: a bioinformatic analysis

Study design and clinical cohorts

In total, there are six parts in our study, including (i) MAEL expression in normal tissues (tissue bulk RNA-seq, Human Protein Atlas [HPA]) and cells (single cell RNA-seq, data source: Supplementary Table 1) [17–19], (ii) MAEL mRNA expression in 32 types of cancer cell lines (Cancer Cell Line Encyclopedia [CCLE]) and 33 types of cancer tissues (The Cancer Genome Atlas [TCGA]) and its pan-cancer prognostic effects [20, 21], (iii) mRNA expression of the six transcripts of MAEL among 33 cancer types and its association with DNA methylation and copy number variations (CNVs) [22, 23], (iv) protein expression of MAEL in ccRCC in the HPA database [18] and mRNA expression of MAEL in patient-derived xenografts (GSE83820 and GSE36895) [24, 25], (v) associations of MAEL expression with clinicopathological features, mutations, gene expression, and prognosis in The Cancer Genome Atlas-Kidney Renal Clear Cell Carcinoma (TCGA-KIRC) cohort (n=522) [26] and the International Cancer Genome Consortium (ICGC)-Pan-cancer analysis of whole genomes (PCAWG) Renal Cell Carcinoma-Europe (RECA-EU) cohort (n=64) [27, 28], (vi) association between MAEL expression and sensitivities to VEGFR/mTOR inhibitors in the ccRCC cell lines of the Genomics of Drug Sensitivity in Cancer (GDSC, n=16) dataset [29] and patients with advanced ccRCC treated with first-line sunitinib (E-MTAB-3267, n=53) [30], and (vii) implications of MAEL expression in predicting the benefit from ICI-based therapies over VEGFR/mTOR inhibitors in the JAVELIN Renal 101 trial (phase III, avelumab+axitinib vs. sunitinib, n=726) and the CheckMate-025 trial (phase III, nivolumab vs. everolimus, n=250) [31].

The basic features of these clinical cohorts, including sample sizes, outcomes, clinical settings, the platforms of RNA-seq and immunohistochemical (IHC) staining of programmed cell death-ligand 1 (PD-L1), and treatment, are displayed in Supplementary Table 2. This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) and the REporting recommendations for tumour MARKer prognostic studies (REMARK) reporting guidelines.

Genomic analysis

The genomic alterations of the TCGA-KIRC cohort were downloaded from the University of California Santa Cruz (UCSC) Xena database [32]. Tumor mutational burden (TMB) and intratumoral heterogeneity (ITH) were retrieved from the TCGA pan-cancer article [26]. The Catalogue Of Somatic Mutations In Cancer (COSMIC) database was used as a supplement to the TCGA-KIRC cohort for measuring the mutational rate of MAEL in ccRCCs. Silent mutations were excluded from our study. The genomic locations of the six transcripts of MAEL and their regulation regions (e.g., promoter and enhancer) were illustrated using the Ensembl [33].

Transcriptomic analysis

MAEL expression in normal human tissue bulks and single cells was illustrated using the HPA (https://www.proteinatlas.org/ENSG00000143194-MAEL, for details, see Supplementary Methods) [17–19]. For tissue bulk RNA-seq, there are 107 samples of the nervous system (e.g., cerebral cortex) in the HPA dataset, and we used the median value to represent the MAEL expression in the nervous system. For single-cell RNA-seq, Uniform Manifold Approximation and Projection (UMAP) was used to visualize the different single-cell clusters, and the cell type of each cluster was determined by the expression of cell-type markers.

The expression of the six isoforms of MAEL in cancer tissue bulks and the prognostic effects of MAEL expression in the 33 cancer types of the TCGA database were analyzed using the Gene Expression Profiling Interactive Analysis 2 (GEPIA2, http://gepia2.cancer-pku.cn) [21]. The association between MAEL expression and its DNA methylation level was explored using the MEXPRESS (https://mexpress.be/index.html) [22, 23]. The level of transcriptomic data was measured by log₂(transcripts per kilobase million [TPM]+1) in the present study.

Gene Ontology (GO) analysis was performed on the website (http://geneontology.org/) by using the annotation data set named Protein Analysis Through Evolutionary Relationships (PANTHER) GO-slim biological process [34, 35].

For Gene Set Enrichment Analysis (GSEA), the javaGSEA Desktop Application (GSEA 4.0.1) was used to investigate the gene signatures significantly enriched in the ccRCC samples with higher or lower MAEL expression (cut-off: median value) [36]. The normalized enrichment score (NES) is the primary statistic for assessing the enrichment of gene sets.

Pharmacogenomic analysis

In total, 16 ccRCC cell lines with data on transcriptomics and sensitivities to anti-cancer agents were included for analysis. For each targeted agent, the half-maximal inhibitory concentration levels (IC₅₀) of 16 ccRCC cell lines were scaled according to their geomean (formula: lg[IC₅₀/geomean]). Two-way analysis of variance was used to assess the difference in sensitivities between the cell lines with high and low MAEL expression.

Statistical analysis

To assess the between-group difference, we used (i) the Fisher exact test for categorical variables, (ii) the Mann-Whitney test, t test with Welch correction, the Kruskal-Wallis test, or one-way analysis of variance for continuous variables, and (iii) the Kaplan-Meier (KM) curves, the Log-rank test, and the Cox proportional-hazards regression model (hazard ratio [HR] and 95% confidence interval [CI]) for time-to-event variables. The variables with a p-value below 0.05 in the univariable Cox regression were included in the following multivariable Cox model. The spearman or Pearson correlation was used to test the correlations between continuous variables.

The statistical analyses were performed using IBM SPSS Statistics 22 or R 4.1.3. The nominal level of significance was set at 5%, and all 95% CIs were 2-sided unless otherwise specified. The adjusted P-value (Q-value) was calculated using the Benjamini-Hochberg method.

Data availability

The authors declare that relevant data supporting the findings of this study are available within the paper and its Supplementary Files. Due to ethical and privacy concerns, we are unable to publish the patient-level data in our study, of which readers may contact the corresponding authors for the access for non-commercial purposes.

MAEL expression in normal tissues and single cells

A previous study observed that human MAEL expression was exclusive in testis rather than other tissues, including brain, heart, liver, lung, spleen, kidney, and ovary as determined by Northern blot (5). Here, we analyzed the RNA-seq data of 32 kinds of human tissues. Besides testis, relatively higher MAEL expression was revealed in placenta, heart muscle, epididymis, kidney, and nervous system (e.g., brain and spinal cord; Figure 1A and Supplementary Table 3).

We further analyzed the single-cell transcriptomes of 25 kinds of human tissues and peripheral blood mononuclear cells (PBMCs) to assess the expression level of MAEL in different cell types (Supplementary Table 4). As for the tissues with higher MAEL expression (testis, placenta, heart muscle, kidney, and brain), the UMAP plots and MAEL expression in each single cell cluster are shown in Figure 1B–1E and Supplementary Figure 1, and the corresponding mRNA expression of cell-type markers in different single cell type clusters are displayed in Supplementary Figure 2–6, respectively. In short, MAEL was expressed relatively higher in early spermatids, spermatocytes, spermatogonia, and late spermatids in testis (Figure 1B), fibroblasts and endothelial cells in placenta (Figure 1C), smooth muscle cells and cardiomyocytes in heart muscle (Figure 1D), proximal tubular cells in kidney (Figure 1E), and astrocyte, excitatory neurons, and oligodendrocyte precursor cells in brain (Supplementary Figure 1).

Of note, first, unlike the pattern in testis and placenta where MAEL was expressed “equivalently” in the single-cell clusters with higher MAEL expression (e.g., c-7 in testis and c-15 in placenta, Figure 1B, 1C), MAEL was expressed “sporadically” in the proximal tubular cells in kidney (Figure 1E), suggesting that MAEL might function in a certain subset of proximal tubular cells, probably relevant to the homeostasis of proximal tubule. Second, MAEL expression was extremely low in immune tissues (lymph node, spleen, bone marrow, and thymus, Figure 1A) and the immune cell clusters in other tissues and peripheral blood (Supplementary Table 4), raising the possibility that MAEL might not be involved in the maturation and activation of immune cells.

MAEL in cancer tissues and cancer cell lines

Among the 33 cancer types in the TCGA database (abbreviations, see Supplementary Table 5), MAEL was expressed higher in TGCT, glioblastoma multiforme (GBM), brain lower-grade glioma (LGG), kidney renal papillary cell carcinoma (KIRP, also abbreviated as pRCC), and KIRC (also abbreviated as ccRCC; Figure 2A). These results were consistent with its expression in normal cell types. For instance, unlike pRCC and ccRCC originating from proximal tubular cells with high MAEL expression (Figure 1E), kidney chromophobe carcinoma (KICH) develops from distal tubular cells that did not express MAEL (Figure 1E) and had far lower expression of MAEL than KIRC and KIRP (Figure 2A). Similarly, in the CCLE database (abbreviations, see Supplementary Table 5), high MAEL expression was observed in nervous system tumors and KIRC (Figure 2B).

As for prognostic value, the RFS and OS of two subgroups divided by the median MAEL mRNA level were compared among the 33 cancer types in the TCGA database. Consistent prognostic effects for predicting both RFS and OS (P<0.05) were observed in KIRC, acute myeloid leukemia (LAML), and skin cutaneous melanoma (SKCM, Figure 2C).

Given the level of MAEL mRNA and its prognostic effect, MAEL may play a crucial role in KIRC, compared with other cancer types. We sought to further discover its linkages with clinicopathological features, DNA methylation, genomic alterations, pathway activation, drug sensitivity, and immunotherapy efficacy in clear cell renal cell carcinomas.

Expression of the six transcripts of MAEL and its potential regulatory mechanisms in ccRCCs

MAEL, located in chromosome 1 (166,975,582-167,022,214), has six transcripts, of which MAEL-204 and MAEL-205 are processed transcripts, and the other four transcripts (MAEL-206, MAEL-201, MAEL-202, and MAEL-203) can be translated into proteins (Figure 3A; features of these isoforms are shown in Supplementary Table 6). In particular, MAEL-206 lacks the HMG domain compared with MAEL-201, 202, and 203 (Figure 3A).

The expression of these six transcripts in the 33 cancer types in the TCGA database is shown in Supplementary Figure 7. Among the five cancer types with the highest MAEL expression, TGCT had equivalent expression levels of MAEL-206 and MAEL-202, while the MAEL expression in the other four cancer types was dominated by MAEL-206 (>90%, Figure 3B), indicating the potential role of MAEL-206 in brain and kidney tumors.

Given the distinct expression patterns of MAEL isoforms in KIRC and TGCT, we further explored the DNA methylation level of MAEL and its association with mRNA expression in these two cancers. First, in KIRCs where MAEL expression was dominated by MAEL-206, MAEL expression was negatively correlated with the methylation level of cg08348962 near the promoter of MAEL-206 (P=0.049), while this association was non-significant in TGCTs (P=0.24; Figure 3C and Supplementary Table 7, marked in blue). Second, compared with KIRCs, TGCTs had higher MAEL-202 expression and lower methylation levels in the regions near its promoters (Figure 3C, marked in green); the methylation levels of these regions were negatively correlated with MAEL expression in TCGTs instead of KIRCs (P<0.05 in TCGTs and P>0.20 in KIRCs; Supplementary Table 7, marked in green). These results indicate that DNA methylation may, in part, explain the distinct expression patterns of MAEL isoforms in human cancers.

In addition to DNA methylation, copy number and mutation may affect transcription. MAEL expression was positively correlated with copy number (P<0.001, Supplementary Table 7). No mutational event of MAEL was observed in the ccRCCs of the TCGA and the COSMIC databases, suggesting that its function in ccRCCs might be regulated by expression level instead of the mutant protein.

Clinicopathological and prognostic correlates of MAEL in ccRCCs

Age, sex, race, and pathological TNM stage were not significantly associated with MAEL expression, while the samples with a poor histological grade had lower MAEL expression (P=0.022, Figure 3D and Supplementary Table 7). We further calculated the prognostic effect between MAEL expression and OS at each cut-off value ranging from 20^th to 80^th percentiles and observed that high MAEL expression trended to be associated with long OS at most cut-off values (Figure 3E). When the cut-off was empirically determined as the median value, the HR was 1.50 (low vs. high: 95% CI 1.11–2.03, P=0.008, Figure 3E). The prognostic effect of MAEL expression was independent of covariates including age, laterality, TNM stage, and histological grade (multivariable HR=1.54, 95% CI 1.13–2.09, P=0.006, Figure 3F). A similar association with RFS was also observed (univariable HR=1.43, 95% CI 1.00–2.04, P=0.050; multivariable HR=1.47, 95% CI 1.03–2.11, P=0.034; Supplementary Figure 8). In a small cohort retrieved from the ICGC-PCAWG RECA-EU database (n=64) [27, 28], we also observed a prognostic trend with similar HR (low vs. high: HR=1.58, 95% CI 0.70–3.58, P=0.28; Supplementary Figure 9).

MAEL protein expression in the cytoplasm of partial tumor cells was observed in two ccRCC samples of the HPA database using immunohistochemical staining (Supplementary Figure 10) [18]. This distribution profile is consistent with previous studies in hepatocellular carcinoma [14], ovarian cancer [15], bladder urothelial carcinoma [16], and colorectal adenocarcinoma [13]. Moreover, in the GSE83820 dataset including five ccRCC samples and their PDXs, compared to the primary grafts, the MAEL expression was increased at early passages (passage 1 [P1] vs. P0: P=0.014; P2 vs. P0: P=0.086) and tended to return to the baseline level at P4 (Supplementary Figure 11). A similar trend was observed in another PDX dataset (GSE36895, Supplementary Figure 11). These results indicate the stable expression of MAEL in ccRCC and suggest that MAEL might be involved in clonal evolution and/or immune escape during the early phase of xenograft development.

MAEL was expressed higher in ccRCCs than in normal kidneys (P=0.024, Supplementary Figure 12) indicating its oncogenic role in ccRCC, while higher MAEL expression was associated with a better prognosis. This observation might seem counterintuitive. However, different types of ccRCCs may depend on different oncogenes, and MAEL-dependent ccRCCs may progress more slowly than those dependent on other oncogenes, thus exhibiting a relatively better prognosis.

Genomic, transcriptomic, and pharmacogenomic correlates of MAEL in ccRCCs

In the TCGA-KIRC cohort, MAEL expression was not associated with mutational count (P=0.99) or fraction genome altered (P=0.27). As for commonly mutated genes, high MAEL expression was associated with the mutations in VHL, PBRM1, and SETD2 (P<0.05, Figure 4A), suggesting its linkage with activated angiogenesis [37, 38].

We further analyzed the correlations between MAEL expression and other genes’ expression levels and identified 2,576 positively-correlated genes and 2,904 negatively-correlated genes (Figure 4B and Supplementary Table 8). The positively-correlated genes were enriched in the pathways concerning blood vessel development and Ras protein signal transduction (Q<0.05, Figure 4C) and excluded in the immune-related pathways about B cell, immunoglobulin-mediated immune response, lymphocyte activation, and complement activation (Q<0.05; Figure 4C, marked in bold). On the contrary, the negatively-correlated genes were enriched in the immune-related pathways relating to lymphocyte proliferation, inflammatory response, and cytokine (Q<0.05; Figure 4D, marked in bold).

Inhibitors of vascular endothelial growth factor receptor (VEGFR, e.g., sunitinib and axitinib) and mTOR (e.g., everolimus) exhibit anti-tumor activity in ccRCCs [39–41], largely due to the activation of angiogenesis and the PI3K-AKT-mTOR signaling [42–45]. Compared with the ccRCCs with low MAEL expression (below median), VEGF and mTOR signatures were enriched in those with high MAEL expression (VEGF: NES=2.28, P<0.001; mTOR: NES=1.68, P=0.013; Figure 4E). Among the 16 ccRCC cell lines in the CCLE database, the IC₅₀ values of VEGFR and PI3K-AKT-mTOR inhibitors were lower in those with high MAEL expression than those with low MAEL expression (above median; P<0.001, Figure 4F and Supplementary Table 9). In the E-MTAB-3267 cohort including 53 patients with advanced ccRCC [30], high MAEL expression trended to be linked with favorable PFS on first-line sunitinib (high vs. low: HR=0.56, 95% CI 0.30–1.06, P=0.064, Supplementary Figure 13).

Taken together, MAEL expression was associated with inactivated anti-tumor immunity, activated pathways concerning VEGFR and PI3K-AKT-mTOR, and sensitivities to VEGFR/PI3K-AKT-mTOR inhibitors in ccRCCs.

MAEL expression predicts the benefits from ICI-based therapies over VEGFR/mTOR inhibitors in advanced/metastatic ccRCCs

We further investigated the association of MAEL expression with the benefit from ICI-based therapies over VEGFR/mTOR inhibitors in two large phase III trials, the JAVELIN Renal 101 (avelumab plus axitinib vs. sunitinib) and the CheckMate-025 (nivolumab vs. everolimus).

First, the 726 advanced/metastatic ccRCC patients with available RNA-seq data in the JAVELIN Renal 101 trial (clinicopathological features, see Figure 5A) were randomly separated into a training set (n=484) and a validation set (n=242) with a ratio of 2:1. The difference in the association of a biomarker with survival across treatment arms is the essential proof of its predictive utility [46]. In the training set, for each cut-off value ranging from 20^th to 80^th percentiles, we calculated the treatment effect in the below cut-off and the above cut-off subgroups. The treatment effect was larger in the low MAEL group than the high MAEL group at all cut-offs (Figure 5B). The difference in treatment effect between these two subgroups reached its maximum at the cut-off of 67.4^th percentile (interaction HR=0.54, 95% CI 0.32–0.93, P=0.027, Figure 5B). At this cut-off, the benefit from avelumab plus axitinib over sunitinib was considerable in the low MAEL expression group (HR=0.53, 95% CI 0.38–0.73, P<0.001) while negligible in the high MAEL expression group (HR=0.97, 95% CI 0.63–1.49, P=0.87, Figure 5C). Comparable results were observed in the validation set (low MAEL expression group: HR=0.61, 95% CI 0.40–0.92, P=0.016; high MAEL expression group: HR=1.01, 95% CI 0.47–2.15, P=0.98, Figure 5D).

In the total set of all 726 patients, the interaction effect between MAEL expression and treatment effect was significant (interaction HR=0.56, 95% CI 0.36–0.88, P=0.012, Figure 5E). In the multivariable analysis using the data provided by the JAVELIN Renal 101 researchers, the interaction effect remained significant (multivariable interaction HR=0.58, 95% CI 0.37–0.91, P=0.019, Table 1). Compared with the low MAEL expression group, the high MAEL expression group had more females (P=0.001) and the m1/4 cluster defined by the TCGA Research Network study (P<0.001) [47], lower CD8 densities in tumor center (P=0.044) and total area (P=0.040), higher TMB (P=0.028) and ITH (P=0.022), and lower expression of PDCD1 (P=0.010) and CTLA4 (P=0.038, Figure 5A). Of note, MAEL expression was not associated with the CD274 mRNA expression in tissue bulk (P=0.46) and the PD-L1 protein expression on immune cells (P=0.57, Figure 5A), indicating the irrelevance between the predictive utility of MAEL and PD-L1 expression.

In the CheckMate-025 trial involving 250 advanced/metastatic ccRCC patients with available RNA-seq data (Figure 5F), the curves of treatment effect in the “below cut-off” and “above cut-off” subgroups are shown in Figure 5G. The treatment effect was larger in the low MAEL group than the high MAEL group at all cut-offs (Figure 5G). At the cut-off (67.4^th percentile) derived from the training set of the JAVELIN Renal 101 cohort, the interaction effect was significant (HR=0.55, 95% CI 0.37–0.97, P=0.039, Figure 5G). Nivolumab delivered a significantly higher ORR and longer PFS than everolimus in the low MAEL expression subgroup (ORR: 24.7% vs. 4.6%, P<0.001; PFS: HR=0.62, 95% CI 0.45–0.87, P=0.004), but not in the high Notch-score subgroup (ORR: 12.8% vs. 2.3%, P=0.097; PFS: HR=1.17, 95% CI 0.73–1.87, P=0.51; Figure 5H). The interaction effect remained significant in the multivariable model (multivariable interaction HR=0.51, 95% CI 0.29–0.91, P=0.023, Table 1). High MAEL expression was associated with few CD8⁺ T cells in tumor center (P=0.040), tumor margin (P=0.016), and total area (P=0.026), high TMB (P=0.013), and low expression of PDCD1 (P=0.003) and CTLA4 (P=0.049) rather than CD274 (P=0.12, Figure 5F).

We further analyzed the association between MAEL expression and immune cell signatures in advanced/metastatic ccRCCs. In the three cohorts (JAVELIN, CheckMate, and TCGA), high MAEL expression was consistently linked with low levels of the signatures concerning activated B, CD4⁺ T, CD8⁺ T, and dendritic cells, central memory CD4⁺ and CD8⁺ T cells, effector memory CD8⁺ T cell, immature B cell, macrophage, MDSC, natural killer T cell, regulatory T cell, and type I T helper cell (P<0.10, detailed statistics, see Figure 6A–6C), indicating an anti-inflammatory microenvironment.

Taken together, high MAEL expression was identified as an independent indicator of poor benefits from ICI-based therapies over VEGFR/mTOR inhibitors in advanced ccRCCs, potentially mediated by tumor-infiltrating immune cells and the expression of PD-1 and CTLA-4 rather than PD-L1.