IRPS associated with the immune microenvironment of CRC and reflected the immune infiltration status.
In order to establish an immune prognostic scoring system related to immune infiltration, we analyzed 230 immune related signatures. We first implemented ssGSEA in 18 CRC cohorts more than 2470 patients according to 230 immune related signatures, and calculated the normalized immune score (NES) of each immune related signature in each sample. Then, the prognostic value of NES in each CRC patients of 18 cohorts was evaluated by univariate Cox regression analysis. Meta analysis was performed on all immune related signatures in 18 cohorts to evaluate the overall prognosis. Finally, we identified 47 immune related signatures as immune prognostic factors (HR<1 or >1, and P < 0.05, Supplementary Table3). Among the 47 signatures, 45 signatures with HR<1 indicates that the higher NES value associated with longer the survival time of patients, and 2 signatures with HR >1 indicates that the higher NES value is related to the poor prognosis. We can find that both signature PD1_ Data and CTLA4_ Data correlated to immunotherapy are prognostic factor. Among the immune infiltrating cells, these cells including iDC, B cell, dendritic cell and macrophage, as well as T cell are all significant for the prognosis of patients. (Figure 1A).
The IRPS was constructed according to the NES values and the prognostic risk model of 47 immune related signatures. IRPS is closely associated with immune infiltrating cells in 18 CRC cohorts. We have known that T cell (47, 48) and cytotoxic cell(49) play an immune promoting role in the anti-tumor process, whereas there is a significant positive correlation between IRPS and T cell B cell and cytotoxic cell, indicating that IRPS plays an immune promoting role in the anti-tumor process (Figure. 1B, C, Supplementary Figure1). To illustrate this view, we further analyzed the relationship between IRPS and immunosuppressive cells, including Treg cell, Th2 cell Tcm cell and Th17 cell. The results showed that there was a negative correlation between IRPS and Th2 cell and Tcm cell (Figure 1B, C, Supplementary Figure1).
Previous studies have shown that the level of immune infiltration is very important in the antitumor processing. Therefore, we calculated the immune infiltration score，including ESTIMATE score, immune score, stromal score and tumor purity of all samples in 18 CRC cohorts. Then, we analyzed the association between IRPS and immune infiltration score. We observed a significant positive correlation between IRPS and ESTIMATE score, immune score and stromal score (Figure 1D-F, Supplementary Figure2). By contrast, tumor purity was negatively correlated with IRPS (Figure 1G, Supplementary Figure2).
Based on above-mentioned results, we can conclude that the IRPS constructed by the NES score of all immune prognostic signatures is related to the immune microenvironment of CRC, and also may reflect the immune infiltration level of CRC.
Relationship between IRPS and microsatellite status in CRC
We calculated the IPRS of each patient in all cohorts and divided the patients into high IRPS group and low IRPS group according to the median of IRPS values. Subsequently, we compared the overall survival of patients in different IRPS subtypes. We found that there was a significant difference in overall survival between high IRPS group and low IRPS group by univariate Cox regression analysis of the cohort in GSE39582 (Figure 2A). IRPS is a prognostic factor in patients with CRC.
The microsatellite status of different patients is different in 18 cohorts. Therefore, dividing patients into MSS group and MSI subtype according to microsatellite status subtype. We found significant differences of IRPS between MSS subtype and MSI subtype, and IRPS is significantly higher in patients with MSI subtype (Figure 2C, Supplementary Figure3). This may indicate that IRPS has a potential function in guiding the immunotherapy of CRC.
In order to further explore the impact of IRPS on microsatellite status, we analyzed the distinction between high IRPS and low IRPS in MSI and MSS subtypes. We can find that there is significant difference between high IRPS group and low IRPS group in MSI subtype and MSS subtype (Figure 2D, E). Meanwhile, we conducted univariate Cox regression analysis in two subtypes patients and found that IRPS has guiding significance for prognosis in patients with MSS subtypes. However, in the patients with MSI subtype, IRPS has no prognostic significance (Figure 2B).
These results prove that IRPS is a prognostic factor of CRC. At the same time, by analyzing the differences of IRPS in patients with microsatellite status subtypes, we find that the level of IRPS may direct the immunotherapy of patients with CRC.
IRPS instructs immunotherapy for CRC
By analyzing the relationship between IRPS and microsatellite status, we found that there is a non-negligible relationship between IPRS and immunotherapy of CRC. Therefore, we analyzed the interaction between IRPS and immune checkpoint molecules.
As we all known, there are three principal immune checkpoint molecules for immunotherapy, namely PD1, PD-L1 and CTLA4. Obviously, there is a significant positive correlation between IRPS and the molecules (Figure 1B, C and 3A-C), especially in CTLA4 (Figure 3C). Interestingly, we found significant differences in immune checkpoint molecules NES values between high IRPS group and low IRPS group (Figure 3D-F).
We already realized that patients with MSI are mainly characterized by high IRPS (Figure 2B), and the microsatellite status of CRC patients who are effective for immunotherapy is characterized by MSI. Given the relationship between IRPS and immune checkpoint molecules (Figure 3A-C), we are more certain that IRPS is closely related to immunotherapy.
Association between innate immune escape mechanism and IRPS in CRC
The intrinsic immune escape mechanism of tumor to immunotherapy mainly has two aspects, immunogenicity and immune checkpoint molecular expression(50). Several potential prognostic factors, including tumor mutation load (TMB), homologous recombination deficient (HRD), neoantigen load, loss of heterozygosity (LOH), copy number variation (CNV) and single nucleotide variation (SNV), determine tumor immunogenicity. In order to explore the effect of IRPS on innate immune escape of CRC, we analyzed the relationship between IRPS and tumor immunogenicity of CRC in TCGA data.
By comparing the differences of these potential prognosis factors between the high IRPS group and the low IRPS group (Figure 4). The prognostic factors CNV, LOH and HRD were higher in low IRPS subtype than in high IRPS subtype (Figure 4A, D and H). Opposite, the mean value of prognostic factors immunogenic indel, indel, tumor mutation load, neoantigen load and SNV tend to be highly expressed in high IRPS subtype (Figure 4B, C, E, F and G). We further analyzed the correlation between IRPS and prognostic factors. Interestingly, IRPS was negatively correlated with CNV, LOH and HRD (Figure 5A, D and H). Simultaneously, IRPS was positively correlated with immunogenic indel, indel, tumor mutation load, neoantigen load and SNV (Figure 5B, C, E, F and G).
Among the microsatellite instability CRC, tumor mutation load is a prognostic factor, and MSI CRC patients with high tumor mutation load are effective for immunotherapy. In view of the significant differential expression of IRPS in MSI tumors (Figure 2D), and there is a positive correlation between IRPS and tumor mutation load (Figure 5E). We propose a hypothesis that the immunophenotype may be shaped by the genomic alterations of tumors.
Exploring relationship between IRPS and genomic alterations in CRC
To test the hypothesis, we analyzed the associations between IRPS and somatic mutations in the TCGA CRC cohort. Firstly, we calculated the mutation frequency of each gene in the CRC cohort and visualized the mutation frequency of the top 35 genes. We observed that the genes with the highest mutation frequency in CRC were APC and TP53, followed by KRAS and PIK3CA (Figure 6A). These genes with high mutation frequency are common in ten carcinogenic pathways, including cell cycle pathway, Hippo signaling, MYC signaling, NOTCH signaling, oxidative stress response / NRF2, PI-3-Kinase signaling, receptor-tyrosine kinase (RTK) / RAS / MAP-Kinase signaling, TGFβ signaling, TP53 and β-catenin / WNT signaling(45).
Subsequently, we discussed the difference of NES values of 10 oncogenic pathways between high IRPS and low IRPS subtypes in TCGA CRC patients (Figure. 6B, C). It is obvious that the NES values of most of the 10 carcinogenic pathways are higher in low IRPS subtypes, in particular, Hippo signaling, MYC signaling, oxidative stress response / NRF2, PI-3-Kinase signaling and TP53 pathway. Only the NES value of cell cycle pathway and NOTCH signaling is higher in the high IRPS subtype. This seems to verify that somatic mutations in common carcinogenic pathways can affect the immunophenotype of patients with CRC.
The IRPS could predict the immunotherapeutic benefit
According to the previous analysis results, we have known that IRPS is closely related to the immunotherapy of CRC, so we studied the immunotherapy benefit of IRPS in CRC. For doing this, we need to obtain the datasets of CRC immunotherapy firstly, but after searching the published CRC immunotherapy datasets, there is no suitable cohorts related to CRC immunotherapy. Instead, we used melanoma immunotherapy cohort (GSE91061, GSE78220) to test the immunotherapeutic benefits of IRPS.
Taking the median IRPS as the cutoff value, patients in GSE91061 and GSE78220 cohorts were divided into high IRPS subtype and low IRPS subtype. Patients in the high IRPs subtype had longer overall survival compared with patients in low IRPS subtype in the GSE78220 cohort (P-value=0.048, Figure7A) and GSE91061 cohort (P-value=0.016, Figure7E). The violin diagram showed that the IRPS value of patients who responded to immunotherapy increased significantly compared with patients who did not respond to immunotherapy (Figure. 7B, F). The association between immunotherapy response and existing immune activity was analyzed by using the information about immunotherapy response in two melanoma immunotherapy cohorts we collected. We observed that patients in high IRPS subtype responded better to immunotherapy than patients in low IRPS subtype (Figure 7C, G). The waterfall plot of two immunotherapy datasets illustrates the correlation between IRPS value and clinical response status in immunotherapy (Figure 7D, H).
In the cohort of GSE78220, GSE91061, IRPS significantly related to the efficacy of immunotherapy was evaluated by predictive model. By using IRPS value as input parameter of support vector machine (SVM), support vector machine is implemented in R software package e1071. Radial basis function (RBF) is selected as the kernel function of support vector machine. The grid search is used to optimize the penalty parameter C and kernel coefficient gamma of support vector machine. The IRPS subtype of patients is used as the prediction target of the classifier. In the test, the overall accuracy of GSE78220 and GSE91061 cohort in predicting patients' immunotherapy response was 0.875 and 0.976 (Figure 8A, B), respectively. The predictive power of the two immunotherapy cohorts clearly shows that IRPS value is a predictive biomarker of immunotherapy benefits.