Landscape of genetic and transcriptional alterations of PRGs in TNBC
The prevalence of gene alterations of PRGs were investigated in the TCGA-TNBC cohort. As shown in Fig. 1a, somatic mutations of PRGs occurred in 84 of the 99 samples, with a mutation frequency of 84.85%. The most mutated gene was TP53 (approximately 83%), followed by AIM2 and CASP8 (roughly 1%). And Fig. 1b showed that most genes were differentially expressed in TNBC. We performed prognostic analysis for each PRGs and OS-related PRGs were shown in Fig. S2. GZMA, AIM2, NLRP1, NLRP3, NLRP7, GZMB, TNF, IRF1, CASP1, NOD2, IL1B and CASP4 expression were linked to a better OS.
Identification of pyro-clusters in TNBC
The datasets TCGA-TNBC, GSE 58812, and METABRIC-TNBC were combined into a single study cohort. TNBC patients were assigned to two distinct pyroptosis-mediated pattern clusters according to 44 PRGs (Fig. S3). They were designated as pyro-cluster A and B, with 320 and 191 patients. PCA analysis revealed that the two clusters could be significantly clustered according to the pyroptotic transcriptional profile (Fig. 1e). We subsequently performed a survival analysis and the KM curves showed patients in pyro-cluster B had a longer OS (Fig. 1d). Furthermore, the association of mRNA profiles of different pyro-clusters with clinicopathological features were displayed in Fig. 1c. GSVA results indicated that pathways relevant to immunology and inflamed processes were considerably enriched in pyro-cluster B, such as interferon alpha, interferon gamma, and complement (Fig. 2c; Table S3).
Correlations of pyro-clusters with TME in TNBC
The immune score of pyro-cluster B patients was greater than pyro-cluster A, implying that the TME of pyro-cluster B had greater components of immunocytes (Fig. 1f). There were differences in 15 of the 22 lymphocyte subsets throughout pyro-clusters, according to CIBERSORT analysis (Fig. 1g). PD-1 and PD-L1 expression were also found to be greater in pyro-cluster B (Fig. 1h-i).
Identification of pyro-gene clusters and exploration of their correlations with TME
A total of 1778 overlapping DEGs were identified between the two clusters. Function enrichment results were shown in Fig. 2a-b and Table S4-5. Immunology and inflamed pathways were found to be substantially enriched, implying the importance of pyroptosis in TME.
We further conducted unsupervised consensus clustering analysis to identify 3 genomic clusters based on DEGs: pyro-gene cluster A, B and C, comprising 161, 252 and 98 samples (Fig. S4). Diverse pyro-gene clusters had dramatically different clinical outcome, according to prognostic analyses (Fig. 2d). As shown in Fig. 2e, 18 of 22 type of immune cells had remarkable differences in infiltration. Pyro-gene cluster A had more infiltration of anti-cancer lymphocytes compared to cluster B and C. The same result was observed in Fig. 2f, the scores of pyro-gene cluster A patients were higher than those of cluster B and C. The above findings revealed separate immune infiltration features among pyro-gene clusters, with pyro-gene cluster A being immune-inflamed phenotype, cluster B being immune-desert phenotype and cluster C being immune-excluded phenotype.
Construction and validation of the risk score
To measure the degree of pyroptosis-mediated patterns in each patient, a risk score was devised. 603 OS-related genes were filtered ultimately by using univariate Cox regression analysis on DEGs. After LASSO regression analysis, 13 genes remained as candidate genes (Fig. S5). After that, we used multivariate Cox regression analysis to come up with a list of 9 genes. The following algorithm depends on 9 genes coefficients:
Risk score = (− 0.3233977* expression of CD109) + (0.2387019*expression of DPCD) + (− 0.1973012*expression of GTSF1) + (− 0.3239561*expression of KLRC3) + (0.1398530*expression of P4HA1) + (− 0.2835991*expression of PNMAL1) + (0.1631214*expression of SPRED2) + (− 0.1820728 *expression of STAMBPL1) + (− 0.2394826*expression of TMEM176A).
TNBC patients were separated into various risk groups based on their median values in the training set. Figure 3a revealed the distribution and interaction in the two pyro-clusters, three pyro-gene clusters, two risk score groups and survival status. Figure 3b-c illustrated the risk score distributions in the two pyro-cluster and three pyro-gene clusters, demonstrating that risk scores may be linked to immune infiltration features. High-risk groups were more prone to have worse clinical outcomes than low-risk groups, according to ranked dot and scatter plots (Fig. 3d-e). Figure 3f showed the differential expression of 9 genes among risk groups. Expression of 9 genes varied dramatically across risk groups, as well as normal groups (Fig. S6). Two risk groups exhibited discrete aspects, according to PCA analysis (Fig. 3g).
In training group, AUC of the 2-, 4-, 6-, and 8-year ROC were 0.721, 0.729, 0.721, 0.746, respectively, showing good predictive ability (Fig. 3h). Patients of validation and all datasets were also separated into 2 risk sets based on the formula and cut off value implemented to the training set (Fig. S7-9). The PCA analysis and scatter plots were displayed in Fig. S7-9, indicating that patient also could be significantly clustered and predicted. For all dataset patients, Fig. S7-9 confirmed that survival differences were observed among risk sets. AUC of the 2-, 4-, 6-, and 8-year ROC (Fig. S7-9) indicated that the risk score exhibited excellent predictive abilities in the validation and all datasets as well.
Clinical correlation analysis of the risk score
Univariate and multivariate analyses were applied to incorporate OS with risk score and clinicopathological characteristics such age, T stage, menopause status, histopathological type, chemotherapy, and N stage. N stage and risk score were potential predictive indicators.
As shown in Fig. S10, a stratified analysis to scrutinize the prediction performance of risk score in divergent clinical subsets, OS differences were discovered for age (p < 0.0001), T 1–2 (p < 0.0001), T3-4 (p = 0.1), N0 and N1 (p < 0.0001), N2-3 (p = 0.019).
Correlations of risk score with TME in TNBC
M2 macrophages, activated mast cells, Tregs, plasma cells, resting CD4 memory T cells, and resting mast cells were considerably and favorably correlated with risk score, while gamma delta T cells, activated memory CD4 + T cells, CD8 + T cells, M1 macrophages, and naive B cells were unfavorably correlated (Fig. 4a). In addition, a low-risk score was linked to a greater stromal score, immune score, and estimation score (Fig. 4b), which matched the findings in Fig. 3b-c. Among the risk groups, there were also substantial differences in the infiltration of 12 of 22 immune cells (Fig. 4c).
Estimation of the role of risk score in chemotherapy and immunotherapy efficacy
We evaluated several drugs widely utilized in TNBC treatment between different risk groups by comparing the IC50 values and found notable differences between risk groups for t lapatinib, vinorelbine, cisplatin and gemcitabine (Fig. 4d).
In recent investigations, the ability of IPS to predict immunotherapy effectiveness has been demonstrated[16]. We assessed differences in risk groups across diverse treatment groups using IPS retrieved from TCIA. In line with our expectations, low-risk group patients performed better, which supports our hypothesis that risk score could be valuable in evaluating immunotherapy effectiveness (Fig. 4e).
In anti-cancer immunotherapy, immune checkpoint blockers targeting PD-1/CTLA-4 have made a humongous accomplishment, and these immune checkpoints are currently the most widely acknowledged biomarkers for predicting treatment response [17]. Figure 4f showed that 26 of 33 molecules were pronouncedly elevated in low risk set. Together, these observations revealed a correlation between risk score and chemotherapy and immunotherapy efficacy.
Development of a nomogram to predict OS
We created a nomogram that combines risks core and clinical features to predict 2-, 4-, 6-, and 8-year OS rates to make the practical application of risk score straightforward. Risk score, age, menopausal status, T-stage, and N-stage were analyzed as candidate predictors by Cox regression, the risk score and N stage were regarded the ultimate prognostic elements in the nomogram, ultimately (Fig. 5a). In both the training and external validation sets, the calibration chart, as well as the AUC values for OS at 2, 4, 6, and 8 years, proved nomogram's significant explanatory power (Fig. 5b-g). At the same time, we also compared the predictive accuracy of nomogram with the TNM stage for prognosis and nomogram had better predictive power (Fig. 5h) These findings demonstrated that the nomogram had a remarkable capacity to forecast survival time in TNBC patients.