Predicting the survival benefit and immune response of immunotherapy for multiple cancers based on mutation gene sets.

This issue of the "Precision Frontier" column shared by Zhao Haitao’s research team published inGenome Medicine(IF =11.12) [1], a mutation signature was constructed and verified to predict the prognosis of patients receiving immunotherapy, and the potential immune response of different subtypes was studied by using multidimensional data.

Research background

Immunocheckpoint inhibitors (ICI) have changed the treatment of many cancers. However, the beneficiaries of ICI treatment are limited, so it is necessary to screen and predict biomarkers to classify patients. At present, many biomarkers, such as tumor mutation load (TMB), have been used as indicative biomarkers in clinic. However, some high TMB patients with gene mutation related to immunotherapy resistance are not sensitive to ICI treatment. Therefore, it is necessary to go beyond TMB and identify the specific genetic determinants of ICI treatment response.

research design

The study included genome and clinical data of 12,647 patients. There were 1572 cases in the training set (immunotherapy patients with 9 kinds of cancer) and 932 cases in the verification set (immunotherapy patients with 5 kinds of cancer). The training set was sequenced by MSK-IMPACT panel containing 468 genes.

The authors identified a set of 11 genes based on mutation, which can divide patients into high-risk and low-risk groups. The mutation of these 11 genes is related to the better response to ICI treatment, and this gene set has been proved to be an independent prognostic factor after ICI treatment.

research results

1. Identify mutant gene sets that can predict the outcome of immunotherapy.

Firstly, the survival differences of 468 genes in the training set were compared between wild type and mutant type, and 98 genes related to OS (global survival) were obtained. After that, LASSO COX regression analysis was further screened and 11 important genes were finally obtained. Through COX regression analysis, the risk score of each patient was quantified on the basis of 11 mutant gene sets. In the training concentration, the OS of patients in the high-risk group is shorter than that of patients in the low-risk group (Figure 1B). In order to study whether the gene set is limited to a specific population or suitable for different populations, the subgroup analysis regardless of age, drug use type and cancer species shows that the gene set is significantly related to the OS of patients treated with ICI (Figure 1C-E).

Figure 1. Generation and Verification of Gene Set Based on Mutation

2. Verification of predictable immunotherapy results

In order to further confirm the value of mutation-based gene set in predicting the results of immunotherapy, it was found that the OS of low-risk group was higher than that of high-risk group (Figure 1F). The research results of predicting ICI treatment response in gene set show that the DCB(durable clinical benefit) of ICI treatment in low-risk group is significantly increased compared with that in high-risk group (Figure 1G), and low-risk patients are more likely to respond to ICI treatment (Figure 1H). These results get the same results in the verification set (Figure 1I, J).

3. Gene set is an independent prognostic predictor of immunotherapy.

Next, the author verifies whether the gene set based on mutation is an independent predictor of immunotherapy response. In the training and validation set, univariate COX regression analysis showed that the gene set was related to OS. After adjusting for drug type, tumor type and TMB, multivariate COX regression analysis showed that the gene set was still an independent predictor, which confirmed the stability of its independent prediction of ICI prognosis (Figure 2A, B). In order to determine which factor has the best predictive performance, C-index is used to compare the performance of mutation-based gene sets with TMB and drug types. The results of C-index show that the gene set based on mutation can predict the prognosis more accurately than TMB and drug type in both training set and validation set (Figure 2C, D).

Fig. 2. Relationship between mutation-based gene sets and other characteristics

4. The clinical benefit of ICI treatment can be predicted based on the mutant gene set, disease stage, CTL and 6-IFN-g gene signature.

Considering the stage of disease, CTL and 6-gene IFN-g signature have been proved to be highly predictive of the response to ICI treatment, the authors speculate that they may play a synergistic role in predicting the response to immunotherapy. The author combined the gene set based on mutation with disease stage, CTL and 6-gene IFN-g signature through Nomotograph, and provided a method for clinicians to quantitatively predict the OS of patients treated by ICI. Fig. 2E is the nomogram of Riaz cohort construction, and the calibration curve of fig. 2F shows the consistency of actual and predicted results, indicating that these signature should be integrated into the predicted nomogram of ICI treatment.

5. Potential external immune landscape of high and low risk groups

In order to further explore the relationship between immune system and mutation-based gene set, the authors conducted a multi-group analysis of TCGA cohort, and the risk score divided TCGA cohort into high and low risk groups (Figure 3A). At the genome level, the proportion of white blood cells, lymphocytes and TIL in the low-risk group was significantly higher than that in the high-risk group (Figure 3B-D). The H&E staining results of TIL ratio are consistent with the above results (Figure 3E). In addition, the proportion of immunostimulatory cells (such as CD8 T cells) in the low-risk group is also significantly higher than that in the high-risk group (Figure 3F). The above results were further tested by Danaher et al.’ s immune infiltration score (figure 3G) and immune characteristic score (figure 3H), and it was found that the abundance of immune cells in low-risk group was higher. Then unsupervised clustering was used to cluster the immune characteristic scores of patients in TCGA cohort, and the results were clustered into two immune infiltration modes (Figure 3I), and the high immune infiltration was significantly enriched in the low-risk group (Figure 3J).

Fig. 3. Immune status of high and low risk population in TCGA cohort.

In addition, the immune signature in the low-risk group was significantly higher than that in the adjacent cancer tissue; On the contrary, the immune signature in the high-risk group was significantly lower than that in the adjacent tissues (Figure 4A). The correlation of immune activity in low-risk group was significantly higher than that in high-risk group (Figure 4B, C). GSEA results showed that 13 pathways were significantly enriched in the low-risk group, including 6 immune-related pathways, such as "cytotoxicity mediated by natural killer cells" (Figure 4D). Tumors in the low-risk group are associated with significantly higher CYT scores, and the number of fibroblasts in the high-risk group is now increasing (Figure 4E, F). According to these results, the low-risk group is rich in immune cells, which can respond to ICI treatment, and fibroblasts may help the high-risk group escape. The expression of chemokines in the low-risk group is higher (Figure 4G, H), so the author speculates that the enrichment of chemokines in the low-risk group may trigger an immune response.

Figure 4. Potential external immune landscape of high and low risk groups　

6. Potential inherent immune landscape of high and low risk groups

Some potential factors determining tumor immunogenicity between the two groups were compared. Mutation, new antigen load, TCR and BCR diversity in low-risk group were significantly higher than those in high-risk group, but CNV load and aneuploidy in high-risk group were higher than those in low-risk group (Figure 5A). This result is consistent with previous studies, that is, tumor aneuploidy is related to immunotherapy and decreased immune escape marker response. In addition, the heterogeneity of tumor in high-risk group is higher than that in low-risk group, which further supports the view that tumor will promote the development of heterogeneity in the presence of cytolytic activity and less active infiltration of immune cells. In order to further understand the mutation process of high and low risk groups, the mutation characteristics were described according to somatic mutation data, and four different mutation patterns were determined in TCGA cohort (Figure 5B). The frequency of these four mutation signals in the low-risk group is significantly higher than that in the high-risk group (Figure 5C). In addition, it was also found that immune checkpoint molecules (such as PD-1, PD-L1 and CTLA4) and costimulatory molecules were more highly expressed in the low-risk group (Figure 5E).

Figure 5. Potential innate immune response and escape landscape in high and low risk groups.　

7. Copy number characteristics of high and low risk groups

Significant differences in chromosome variation were detected in high and low risk groups (fig. 6A). Local amplification peaks of immune genes with good characteristics, such as PD-L1 (9p24.1) and PD-L2 (9p24.1), were observed in the low-risk group (fig. 6B, C). GO function annotation of the specifically amplified genes showed that the low-risk group was significantly enriched in two immune-related biological processes, while the high-risk group was significantly enriched in the biological process of "positive regulation of fibroblast proliferation", but not in any immune-related biological processes (Figure 6E). At the level of mRNA expression in TCGA cohort, the expression of PD-L1 and PD-L2 mRNA in low-risk group increased significantly (Figure 6G), which was consistent with CNV data. This finding suggests that CNVs in tumor contributes to the observed difference in immune infiltration.

Figure 6. Variation of copy number in high and low risk groups

discuss

This study is the first time to use an independent cohort to study the comprehensive mutant gene sets of different tumor types. Through PSM algorithm, hierarchical analysis and multivariate COX regression analysis, the application performance of mutation-based gene sets in different types of tumors was tested, and the results showed that mutation-based gene sets were reliable.

This study has the following innovation and practical application value. First, different types of tumors (such as NSCLC (non-small cell lung cancer), melanoma and renal cell carcinoma) are used, which represent the most common types of cancer treated with ICI. Secondly, the application of multiple biomarker prediction model needs to understand the factors that affect the accuracy and precision of Qualcomm analysis in clinical practice. The risk score formula and threshold of mRNA expression calculation are not suitable for verification with other types of data. Therefore, this study developed a gene set based on mutation to predict the clinical efficacy of ICI treatment. The composition of the above mutations is neither affected by tissue types nor adjusted by any other biomarkers. The risk score formula and threshold of gene set based on mutation can be verified by other tumor analysis methods, such as DNA sequencing and single nucleotide polymorphism microarray analysis. Therefore, mutation-based gene sets are not affected by technological changes, even when different platforms are used in different centers. Thirdly, in practice, gene sets based on mutations avoid exposing patients to potential immune-related adverse reactions when they are unlikely to respond, and enable patients to match to potentially more effective treatments more quickly. Fourthly, the prediction performance of gene set based on mutation is compared with other factors that can predict immunotherapy, and it is found that the prediction performance of gene set based on mutation is better than all these factors.

Several limitations of this study. First of all, because some mutations may be enriched in some tumor types, the initial goal of this study is to create a panel instead of identifying a single gene (such as BRAF), because the former can contain more genes to predict the prognosis of different types of tumors. Secondly, although the immune landscape of 11 genes in the gene set based on mutation has been explored, it is still necessary to clarify the molecular mechanism of each gene affecting immunotherapy in vivo and in vitro functional experiments. Thirdly, the enrichment scores of carcinogenic pathways and the expression patterns of immune checkpoints should also be detected by immunohistochemistry.

tag

The mutation-based gene set proposed in this study is the first comprehensive genomic marker systematically identified, which can be used to evaluate the ICI treatment effect of pan-cancer species. This study is also the largest prognostic model discovery project for cancer patients receiving ICI treatment (whether single drug treatment or combined treatment of anti-PD-1 and anti-CTLA-4). Nomoto, which combines mutation-based gene set with TMB and drug types, can help clinicians choose patients who may have a strong response to ICI treatment. In addition, it also reveals the different immune conditions of high and low risk groups, and specific genomic changes may drive the formation of these microenvironments. In a word, this work puts forward a new tumor classification method, which may guide the decision-making of ICI treatment.　END　

References:

[1]　Long, J., Wang, D., Wang, A. et al. A mutation-based gene set predicts survival benefit after immunotherapy across multiple cancers and reveals the immune response landscape. G enome Med 14, 20 (2022). https://doi.org/10.1186/s13073-022-01024-y

Write Liu Hua Hua

Editing and typesetting SX