The transcript levels and matched clinical information of three MM populations were collected from the GEO (http://www.ncbi.nlm.nih.gov/geo/) database, namely, GSE136324, GSE57317, and GSE4581 cohorts. Samples were excluded if corresponding survival data were missing. The comprehensive summary of patient characteristics was presented in Table S1 . The relative gene expression was normalized using the “limma” R package. The PRGs were acquired from the REACTOME_PYROPTOSIS and GOBP_PYROPTOSIS gene sets from the MSigDB database (https://www.gseamsigdb.org/gsea/msigdb/), as well as prior reviews (14). A workflow chart describing the samples utilized at each stage of analysis is presented in Figure 1 . The datasets of this study are publicly available from the GEO database.

We selected the GSE136324 cohort as the training cohort (TC). Using univariate analysis, we identified PRGs that were strongly correlated with patient prognosis (p < .05). The least absolute shrinkage and selection operator (LASSO) analysis was employed to obtain an optimal PRG weighting coefficient (15). Subsequently, we performed a 10-fold cross verification to penalize the maximal likelihood predictor. Furthermore, the minimal criteria of the penalized maximal likelihood predictor were employed to acquire the quintessential penalty parameter λ values. The GSE57317 and GSE4581 databases were next chosen as the verification cohorts (VCs). Patients were separated into a high- (HR) or low-risk (LR) cohort, depending on the median TC risk score (RS) ( Figure 2 ). To further verify the validity of the model, the Kaplan-Meier analysis of survival was performed between both groups. Additionally, the time-dependent receiver operating characteristic (ROC) curve analysis was established to reflect the sensitivity and specificity of predictions ( Figure 3 ).

The univariate and multivariate Cox analysis for RS and prognostic indicator were performed in TC ( Figure 4 ). Moreover, a nomogram was built based on Revised International Staging System(R-ISS) and RS with the consistency index (C index) and the curve of calibration ( Figure 5 ).

The GSEAv4.0.2 software (http://software.broadinstitute.org/gsea/login.jsp) and c2.cp.kegg.v7.0.symbols gene sets were employed to elucidate the physiological pathways associated with the HR and LR patient cohorts. NOM P-value <0.05 was deemed significant. Additionally, we explored differences in immune cell infiltration landscapes between the two patient populations, using the signature-identified via the CIBERSORT algorithm ( Figure 6 ).

Bone marrow specimens were accumulated from MM (n=20) and non- hematological malignancy patients (n=20) at the Sun Yat-sen University Cancer Center. Total RNA was extracted with TRIzol (Thermo Fisher Scientific, USA) and corresponding cDNA was synthesized using PrimeScript™ RT Master Mix (Takara Bio, USA), followed by qRT-PCR employing TB Green^® Premix Ex Taq (Takara Bio, USA). Intergroup analysis was done with the Student’s t-test (two-tailed). The employed primer sequences are presented in Table S2 .

The MM cell lines U266 and RPMI 8226 were acquired from the American Type Culture Collection (ATCC, Manassas, VA, USA), and maintained in a 37°C humid chamber with 5% CO₂. RPMI-1640 medium was used with 10% foetal bovine serum and 100 IU/mL of penicillin and streptomycin each (RPMI 1640, FBS and Pen-Strep were from Gibco).

Cell viability was examined via Cell Counting Kit-8 assay (CCK-8) (Dojindo, Japan), following kit directions. Doxorubin (DOX, S1208) and Q-VD-Oph (S7311) were acquired from Selleck Chemicals (Houston, TX, USA). U266 and RPMI 8226 cells were grown in 96-well plates with 10000 cells per well in an incubator with 5% CO2 at 37°C for 24 h. Cells were next exposed to the compound at specified concentrations for 48 h, before introduction of CCK-8 kit reagent, followed by a 2 h incubation. Optical density was recorded at 450 nm with a microplate reader. All experiments were done two or more times. In terms of cell apoptosis, compound-treated (specified concentrations for 48 h) U266 and RPMI 8226 cells were lysed and twice rinsed in chilled PBS. Subsequently, the cells underwent FITC-Annexin V/PI staining (KeyGEN, China), following kit directions. The staining was done for 15 min in the dark before detection and quantification of apoptosis using flow cytometry.

Cells were treated as specified for 48 h, then collected and centrifuged at 1000 rpm for 5 min to obtain pellets, which were then lysed in RIPA buffer, enhanced with protease and phosphatase inhibitors (PHYGENE, SantaCruz Inc, Dallas, TX, USA), for 1-2 h. The lysates underwent further vortex mixing, followed by sonication in an ice-water bath for 5 min on high for 30 s and 1 min intervals. The resulting solution was spun down at 12,000 rpm for 15 min, and, following total protein quantification, 50 µg protein was boiled with sample buffer at 95 ° for 5 min, and electrophoresed in a 10% sodium dodecyl sulphate-polyacrylamide gel (SDS-PAGE) before transfer to polyvinylidene fluoride (PVDF) membrane. As primary antibodies, we employed anti-β-actin (1:1000; CST, Boston, MA, USA), GSDME (1:1000; Abcam, Britain), and as secondary antibody we employed anti-rabbit IgG H&L (HRP) antibody (Abcam, ab205718).

Inter-group analyses were done using Student’s t test or one-way ANOVA. Categorial data was analyzed using the Chisq or Fisher exact test, SPSS software version 25 (IBM Corporation, Armonk, NY, USA). A two-sided P-value <0.05 was deemed significant. All data analyses employed the R software (version 3.6.3 for Windows, http://www.R-project.org). All experiments were done thrice, and the resulting data displayed as means ± SD.

Overall, 986 MM patients with relevant RNA profile and corresponding clinical information from the GEO database were included in our analysis ( Table S1 ). To identify PRGs related to MM prognosis, we specified the GSE136324 cohort(n=866) as the TC, and the GSE57317 (n=55) and GSE4452 (n=65) populations as the VCs. Figure 1 summarizes our research design.

We identified 57 PRGs with matching transcript profiles from all three patient populations. Using univariate analysis, we further identified 19 PRGs that relate to MM overall survival (OS) from the GSE136324 cohort (P<0.05) ( Figure 2A ). Next, we employed LASSO analysis to generate a PRG signature that accurately predicts MM prognosis. According to our penalized maximal likelihood predictor of 100000 bootstrap replicates, a profile of 11 PRGs (namely, AIM2, CASP1, ELANE, GSDMB, GSDMC, IL1B, NLRP1, GZMB, IL1A, CHMP7, and CYCS) was generated, carrying a minimal criteria optimal λ value ( Figures 2B, C ). Figure 2E illustrates the tumor mutation profile of the 11 PRGs. The model equation is provided below:

RS = 0.07 X AIM2 levels + 0.12 X CASP1 levels - 0.06 XELANE levels + 0.06 X GSDMB levels - 0.07 XGSDMC levels - 0.05 X IL1B levels + 0.31 X NLRP1 levels - 0.02 X GZMB levels - 0.16 XIL1A levels + 0.25 X CHMP7 levels + 0.16 X CYCS levels.

The TC patients were assigned to high-risk (HR, n=433) or low-risk (LR, n=433) populations, according to the median threshold. LR patients experienced remarkably prolonged OS time, relative to HR patients (P<0.05) ( Figure 3A ). Next, we tested the model reliability with time-dependent ROC curves. The TC AUCs for the 1-, 3-, 5-, 7-, and 10-year OS were 0.626, 0.621, 0.654, 0.674, and 0.637, respectively ( Figure 3B ). The VC RS was computed based on the aforementioned formula, and the median TC RS was used as the threshold. The curves were next applied to the internal VC, and the subsequent AUCs were 0.778 and 0.715, 0.709 and 0.628, as well as 0.645 and 0.769 for the 1-, 3- and 5-year OS in the GSE57317 and GSE4581 populations, respectively. The PRG signature exhibited an elevated OS predictive accuracy in both VCs. Moreover, the HR patient OS was remarkably worse than the LR patient OS, as evidenced by the Kaplan-Meier analysis. Given these data, our model demonstrated persistent superior performance.

Based on our uni- and multivariate analyses examining RS and other prognostic factors, the RS serves as a stand-alone OS prognostic marker in the GSE136324 population ( Figure 4A ). We further assessed the associations between RS profile and clinicopathological features. We observed marked differences between beta2-Microglobulin (β2-MG) and lactate dehydrogenase (LDH) levels, as well as the International Staging System (ISS) and Revised ISS (R-ISS) ( Figure 4B ). Figures 4C–E illustrates the correlations between clinicopathological characteristics and both datasets as heatmaps, whereby each dataset exhibits a slightly altered, yet consistent, gene expression.

We generated a nomogram to collectively display data from the R-ISS stage and PRG profile of the TC ( Figure 5A ). Calibration curve and consistency index (C-index) assessed the predictability of our model. The merged score AUC was higher than the R-ISS stage, indicating the significance of the nomogram in enhancing OS estimation ( Figures 5B–E ).

GSEA was utilized to explore the physiological roles and related pathways associated with RS. Based on our analysis, a majority of PRGs participated in cell proliferation, metabolism, and immune response pathways ( Figures 6A, C, E ). We next employed CIBERSORT (16) to estimate the differences between 22 distinct forms of tumor-infiltrating immune cells between the LR and HR patients. Relative to HR patients, we demonstrated a marked reduction in tumor infiltration, particularly that of monocytes and B cells in the LR group in the TC (P<0.001, Figure 6 )

Validation of the expression of 11-PRGs in normal tissues and tumor tissues in GSE118985 and our cohort are shown in Figures 7A, B . Based on our results, AIM2, CASP1, GSDMB, NLRP1, CHMP7, and CYCS were upregulated, whereas, ELANE, GSDMC, IL1B, GZMB, and IL1A were downregulated in MM specimens versus controls ( Figure 7B ).

First, we assessed MM cells (U266 and RPMI 8226) viability after exposure to DOX and Q-VD-OPH alone or in combination. MM cells were incubated with varying doses of DOX (0 μM, 0.02 μM, 0.04 μM, 0.08 μM, 0.16μM and 0.32μM) and Q-VD-OPH (40μM) in combination or alone for 48h, and cell viability was examined with CCK8. Based on our results, the combined treatment greatly enhanced cell viability, relative to DOX alone in both cell lines ( Figure 7E ).

Using Annexin V -FITC/PI-staining and flow cytometry, we analyzed U266 and RPMI 8226 cell apoptosis after treatments with DOX (0.08μM) and Q-VD-OPH (40μM) alone or in combination for 48h. We demonstrated that DOX significantly induced cell apoptosis while Q-VD-OPH significantly decreased DOX-induced cell apoptosis in both cell lines ( Figures 7C, D ). These results demonstrate that DOX stimulated apoptosis by activating pyroptosis in U266 and RPMI 8226 cells.

Doxorubin (DOX), a common chemotherapeutic agent, induces pyroptosis via caspase-3-induced slicing of GSDME (GSDME-N) (17). Q-VD-OPH, a pan-caspase suppressor, forms covalent bonds with and permanently sequesters caspase-3 (18).So we detected the GSDME- N and GSDME- F protein expressions in two cell lines incubated with DOX and Q-VD-OPH alone or in combination western blot analysis ( Figure 7F ) to confirm that DOX-induced pyroptosis is drastically reduced by Q-VD-OPH in MM cells. Collectively, these evidences suggest that DOX- induced apoptosis, which play an essential function in cancer treatment, is related to pyroptosis.