Analysis of gene expression levels and their impact on survival in 31 cancer-types patients identifies novel prognostic markers and suggests unexplored immunotherapy treatment options in a wide range of malignancies

In the present study we addressed the role Immune Checkpoint-Related (ICR) molecules, either as markers of disease or as markers of response to therapy/therapeutic targets, in 31 cancer types listed in Table 1. This analysis was carried out on expression data derived from the public cancer portal GEPIA2 (http://​gepia2.​cancer-pku.​cn/​), which refers to TCGA and GTEx datasets. Namely, CTL4, PD-1 (PCD1), PD-L1 (CD274), LAG3, TIM3 (HAVCR2), OX40 (TNFRSF4), GITR (TNFRSF18), 4-1BB (TNFRSF9) and TIGIT were investigated. Selection of these nine ICR molecules was based on a preliminary analysis, indicating that the selected ICR molecules are all currently investigated in clinical trials involving cancer and non-cancer patients (see Table 1 not shown).

Additional file 1: Figures S1 to S9 show expression levels of the ICR molecules in 31 cancer types vs. the corresponding controls, in 9498 cancer samples and 5540 matched-healthy controls. Numerosity of cancer samples and corresponding controls for each cancer type is reported in Table 1. Boxes in Additional file 1: Figures S1 to S9 highlight the cases where a significantly different expression (p < 0.0001) is found in the given cancer vs. the corresponding controls. Data reported in Additional file 1: Figures S1 to S9 are summarized in Fig. 1 reporting the expression data in a heatmap modality. Cancer types are sorted vertically from the top (higher number of significant differences observed) to the bottom (lower number of significant differences observed). ICR molecules are sorted horizontally from the left (higher number of significant differences found) to the right (lower number of significant differences). Red spots indicate significant up-regulation and blue spots indicate significant down-regulation. Figure 1 shows that Lymphoid Neoplasm Diffuse Large B-cell Lymphoma (DLBC) is at the top first position since DLBC patients show expression of all nine molecules up-regulated in cancer vs. controls. Skin Cutaneous Melanoma (SKCM) is at the second top position, since SKCM patients show 7 out 9 molecules up-regulated. Such 7 molecules represent the “Profile-A” investigated later in Fig. 2A. Pancreatic Adenocarcinoma (PAAD) and Testicular Germ Cell Tumors (TGCT) are at the third and fourth top position, since PAAD and TGCT patients show 6 out of 9 molecules up-regulated. Such two 6-molecules combinations are the “Profile-B” and “Profile-C”, investigated later in Fig. 2B, C. Figure 1 then reports all other cancer types showing 5, 4, 3, 2, 1 and 0 ICR molecules differently expressed, according to the vertical sort. It is noteworthy that 23 cancer types out of 31 show at least 1 significantly different expression. ICRs expression data reported in Fig. 1 and Additional file 1: Figures S1 to S9 were from TCGA datasets analyzed in GEPIA2. Expression data of ICRs in melanoma, in pancreatic adenocarcinoma and in thyroid carcinoma were validated in the EBI expression atlas, from independent datasets, as reported in Additional file 1: Table S3 not shown, Additional file 1: Table S4 not shown and Additional file 1: Table S5 not shown, respectively.

Figure 1 also shows that TIM3 (HAVCR2) and OX40 (TNFRSF4) are at the most left positions; they are differently expressed in 15 and 14 cancers types respectively, out of 31, followed by GITR (TNFRSF18), LAG3 and TIGIT, differently expressed in 12 and 10 and 9 cancers, respectively. The other ICR molecules are then reported according to the left-to-right horizontal sort.

The large number of significant differences reported in Fig. 1 indicate that such ICR molecules may have a potential role in several cancer types, either as diagnostic/prognostic markers or as therapeutic targets. We then hypothesized that ICRs expression may represent a profile able to effectively discriminate cancer samples from the corresponding healthy controls. This hypothesis was verified using the Dimensionality Reduction tool of GEPIA2, which performs a PCA analysis on the combined expression data. PCA analysis of all nine ICRs in Lymphoid Neoplasm Diffuse Large B-cell Lymphoma patients (DLBC) was not possible, due to the lack of a proper control dataset to be used in PCA calculation. However, PCA analysis of the “Profile-A” was possible in melanoma (SKCM) patients. In fact, Fig. 2A shows that the 7 molecules differently expressed in SKCM samples (i.e., TIM3 (HAVCR2), OX40 (TNFRSF4), GITR (TNFRRSF18), TIGIT, LAG3, PD-1 (PDCD1) and CTL4), show a clear separation of patients from the corresponding controls. Figure 2A also shows that Profile-A is not able to discriminate LUAD, PRAD and THCA cancer patients from the corresponding controls, indicating that such profile appears to be specific for melanoma patients (SKCM).

“Profile-B” is made by the 6 molecules differently expressed in Pancreatic adenocarcinoma (PAAD) patients (i.e., TIM3 (HAVCR2), OX40 (TNFRSF4), GITR (TNFRRSF18), TIGIT, LAG3 and CTLA4). PCA analysis of Profile-B expression shows a clear separation of PAAD patients form the corresponding controls (Fig. 2B), while the same profile is not able to discriminate LUAD, PRAD and THCA cancer patients from the corresponding controls, indicating that such profile appears to be specific for PAAD patients. Similarly, when the “Profile-C” expression (i.e., TIM3 (HAVCR2), OX40 (TNFRSF4), GITR (TNFRSF18), TIGIT, LAG3 and PDCD1) was investigated in Testicular Germ Cells Tumor (TGCT) samples, a clear separation from the corresponding controls was found (Fig. 2C), while the same profile does not discriminate LUAD, PRAD and THCA cancer patients from the corresponding controls, indicating that such profile appears to be specific for TGCT patients. Figure 2 shows that the 7 molecules differently expressed in SKCM, and the six molecules differently expressed in PAAD and in TGCT are specific expression-profiles able to clearly discriminate cancers form controls samples, according to PCA analysis, showing a relevant marker potential for specific cancer types.

We then investigated the role the nine ICR molecules may play in the metastatic phase of melanoma, by accessing the expression data in 104 melanoma primary patients vs. 368 melanoma metastatic patients, at the UALCAN public portal (http://​ualcan.​path.​uab.​edu/​analysis.​html). All ICRs investigated, except GITR (TNFRSF18), show a significantly different expression in primary vs. metastatic patients, as reported in Fig. 3, suggesting that such molecules (or drugs targeting these molecules) may have a role in metastatic melanoma diagnosis or control.

We then addressed the hypothesis that the nine ICRs may represent potential suitable therapeutic targets or markers effectively predicting prognosis or response to therapy. We investigated such hypothesis by analyzing how their expression relates to patient survival. This analysis may indicate molecules with a potential therapeutic application, irrespective of their marker ability to identify cancer patients. Table 2 reports survival data expressed as Hazard Ratio (HR) as a function of expression values of each ICR molecules, according to GEPIA2 database. Expression levels were categorized as “high-level” (i.e., above the median value) and “low-level” (i.e., below the median value). In Table 2 only HR ≤ 0.7 or ≥ 1.5 are reported, where HR < 1 indicates improved survival in high-level expressing patients, while HR > 1 indicates improved survival in low-level expressing patients. Data in Table 2 are both vertically and horizontally sorted, similarly to Fig. 1. Surprisingly, horizontal sorting shows that in Skin Cancer Melanoma (SKCM), in Testicular Germ Cell Tumors (TGCT) and in Thymoma (THYM) the expression of all nine ICR molecules is associated with relevant survival improvement. In TGCT and THYM, survival improvement is almost invariantly associated with low-expression levels of all ICRs except PD-1 (PDCD1), while improved survival in SKCM is always associated with high expression levels of all nine molecules. Notably, SKCM and TGCT are in the very top positions both in Fig. 1 (expression data) and in Table 2 (survival data).

On the other hand, Table 2 shows that the highest number of cancer types with improved survival associates with expression levels of TIM3 and OX40 (14 out of 31 cancer types and 13 out of 31 cancer types, respectively). Most interestingly, TIM3, OX40 and TIGIT are in the very left positions both in expression and in survival analyses (Fig. 1 and Table 3, respectively).

Table 2 shows that in several cases HR was found to be smaller than 0.5 or largely higher than 1.5, suggesting a very strong association of the expression to the survival, as for instance for PD-1 (PDCD1) expression in Uterine Corpus Endometrial Carcinoma (UCEC) (see the corresponding Kaplan–Meier plots in Fig. 4A), or for TIGIT expression in Thymoma (THYM) (see the corresponding Kaplan–Meier plots in Fig. 4B). Interestingly, Fig. 4 shows a striking complete survival in patients expressing low-levels of TIGIT as compared to high-levels expressing patients showing a time-dependent loss of survival. Such data indicate that drugs inhibiting TIGIT may exert a potential strong protective role (Fig. 4B) and that PD-1 agonists may have an impact on UCEC patients-survival (Fig. 4A).

Data reported above indicate that expression of several molecules associate with strong survival improvement in melanoma, in Testicular Germ Cell Tumors, in Thymoma as well as in other cancer types, introducing the hypothesis that targeting such molecules may represent an effective therapeutic option in different cancer types.

The nine ICRs showing a relevant survival association in melanoma patients according to data reported in TCGA datasets (Table 2), were all validated by analyzing independent GSE datasets, at the PROGgene V2 portal (http://​www.​progtools.​net/​gene/​index.​php). Figure 5 reports the corresponding Kaplan–Meier survival plots and Table 3 reports the corresponding numerical analysis.

Validation achieved on GSE datasets confirm the interest of all such molecules as potential therapeutic targets in melanoma patients.