We used the Gene Expression Omnibus (GEO) database (https://www.gsea-msigdb.org/gsea/index.jsp/) to collect gene expression profiles and relevant clinical data, and this study employed three different datasets, namely, GSE136337, GSE4204, and GSE24080. In preparation for subsequent analysis, the expression profiles underwent a log2 transformation. For a comprehensive overview of the clinicopathological and survival data, please refer to Table 1 . In this study, platelet-related genes were extracted from the platelet-related genomes available in the Gene Set Enrichment Analysis (GSEA) database (http://www.gsea-msigdb.org/gsea/msigdb). A grand total of 1247 genes were chosen for subsequent investigation. Supplementary Table S1 contains information about the genes that overlap.

We utilized the GSE136337 dataset acquired from the GEO dataset as a training cohort to create the risk score model associated with platelets. In order to determine PRGs with prognostic significance, we performed univariate Cox regression analysis on the genes under consideration. A significance threshold of p < 0.001 was applied to screen for genes with potential prognostic associations. Subsequently, a platelet-related risk model was constructed using Least absolute shrinkage and selection operator (LASSO) Cox regression. The model’s coefficients were derived from the previous step. By applying this model, a platelet-related risk score was calculated for MM patients in each dataset. In particular, the GSE4204 and GSE24080 datasets served as validation cohorts. In order to enhance patient stratification, the individuals in each dataset were divided into cohorts of high-risk and low-risk, depending on the risk score median specific to each dataset.

Survival disparities between different risk groups were compared by generating survival curves and dot plots. The R language package “stringr” was employed to create heat maps, enabling a comparison of platelet-related gene expressions across multiple datasets. Furthermore, receiver operating characteristic (ROC) curves were utilized to confirm the sensitivity and specificity of genes associated with platelets.

Univariable and multivariable Cox regression analyses were utilized to assess the influence of independent prognostic factors on the overall survival in both the training and validation cohorts. In addition, in the GSE136337 dataset, we examined clinical characteristics and compared risk scores between subgroups to identify potential subgroup differences.

To minimize variations arising from different algorithms, we employed multiple algorithms to evaluate the immune microenvironment of the subgroups. Specifically, estimating the proportion of immune and cancer cells (EPIC) (Racle et al., 2017), xCell (Aran et al., 2017), and single-sample genome enrichment analysis (ssGSEA) were utilized for this purpose. Furthermore, the correlations between eight platelet-related genes and immune cells were assessed using the cell-type identification by estimating relative subsets of RNA transcripts (CIBERSORT) method (Newman et al., 2015). Utilizing these methodologies, potential links between platelet-related genes and populations of immune cells can be explored. Additionally, the immune cell microenvironment scores of high-risk and low-risk groups were assessed and compared through the implementation of xCell and immunophenotype score (IPS) techniques (Charoentong et al., 2017). This allowed us to obtain knowledge about the variations in the composition of the immune microenvironment among these subgroups. Moreover, we assessed the differences in immune checkpoint responsiveness between the groups at high risk and low risk, enabling us to evaluate the possible consequences for the response to immunotherapy.

To compare the drug susceptibility between the low- and high-risk groups, the R package “pRRophetic” was utilized, enabling a comprehensive assessment of differences in drug response and sensitivity.

To validate the cellular expression of PRGs, the Cancer Cell Line Encyclopedia database (CCLE) was utilized. The CCLE database can be browsed through this link: https://portals.broadinstitute.org/ccle. In order to examine the interactions between proteins (PPIs) associated with platelet-related genes, we obtained the PPI network linked to PRGs from the Search Tool for the Retrieval of Interaction Gene/Proteins (STRING) database (version 11.5) (https:// www.string-db.org/).This network analysis provided insights into the molecular interactions and potential functional relationships among these genes.

In order to explore potential underlying mechanisms linked to the platelet-related genes, we carried out pathway analysis utilizing the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. We assessed enriched pathways across various datasets by utilizing the Gene Set Enrichment Analysis (GSEA) v4.0.2 software (http://software.broadinstitute.org/gsea/login.jsp). Our analysis considered statistical significance as p < 0.05 and q < 0.25.

Utilizing the results obtained from univariable and multivariable Cox regression analyses, we created a combined nomogram facilitating prognostication of the overall survival rates at 1-year, 3-year, and 5-year for MM patients. This nomogram incorporated age, ISS stage, and the platelet-related risk score as prognostic factors. To validate the performance of the nomogram, calibration curves were plotted to assess its accuracy in predicting patient outcomes. In addition, we assessed the predictive abilities of ISS stage, platelet-related risk score, and the nomogram through the application of time-dependent ROC curves for the survival time points at 1 year, 3 years, and 5 years. Decision curve analysis (DCA) was performed to assess the clinical usefulness of every individual clinical characteristic and the risk score. This analysis allowed us to assess the net benefits of each factor in terms of survival prediction.

The LP-1 and MM1.R cell lines, referred to as MM cell lines, were acquired from Fenghui Biotechnology Co., Ltd in Hunan, China. The cultivation of these cell lines took place in a controlled environment within a humid incubator set at 37°C with 5% CO₂. The study included a total of 25 MM patients and 15 healthy individuals, and their clinical characteristics are presented in Supplementary Table S2. The First Affiliated Hospital of Wenzhou Medical University’s ethical committee granted the study permission. All study participants gave their informed consent, and the research followed the guidelines set forth in the Helsinki Declaration.

Bone marrow puncture was performed on 25 patients and 15 healthy people in the control group. And 5 mL of aspirate was taken from their bone marrow. Bone marrow mononuclear cells (BMMNC) were then isolated using density gradient centrifugation. We employed the Righton DNA&RNA Blood and Tissue Kit (supplied by Righton Bio, Shanghai, China) to perform total RNA isolation from the bone marrow of clinical MM patients and healthy volunteers, adhering to the guidelines provided by the manufacturer. To generate complementary DNA (cDNA), cDNA synthesis kits (obtained from Vazyme, Nanjing, China) were utilized in the subsequent reverse transcription step. To conduct PCR amplification, we employed the Taq Pro Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China) in accordance with the guidelines provided by the manufacturer. For quantifying the expression levels of PRGs, we utilized quantitative reverse transcription PCR (qRT-PCR). To serve as an internal reference gene, β-ACTIN was selected, and the primers specified in Table 2 were applied. To ensure accuracy and reproducibility, each sample was subjected to three repetitions.

For the purpose of these tasks, multiple software programs were utilized to conduct clinical evaluations and statistical analyses. The software programs employed included GraphPad Prism version 9.0.0 by GraphPad Software Inc. in San Diego, CA, United States, SPSS version 25.0 by SPSS Inc. in Chicago, IL, United States, as well as the widely used R software developed by the R Foundation for Statistical Computing in Vienna, Austria. In order to pinpoint the potential PRGs, we carried out univariate Cox regression analysis and LASSO regression analysis. Following that, we conducted multivariate Cox regression analysis to evaluate the predictive worth of the platelet-related risk score and clinical characteristics. We compared the survival rates through the utilization of Kaplan-Meier curves and log-rank test. To examine variables that followed a normal distribution, we utilized the independent t-test to make comparisons between groups. Categorical variables, on the other hand, underwent analysis using the Chi-square test. In situations where the distribution was non-normal, we employed the Mann-Whitney U test to compare two groups. For this particular investigation, we deemed a significance level of p < 0.05 as statistically meaningful. Regarding graphical representations, p-values were denoted as follows: *: p < 0.05; **: p < 0.01; ***: p < 0.001; ****: p < 0.0001. Additionally, the label “ns” conveyed a lack of statistical significance.

In this research, we conducted an examination on the survival information of 1511 individuals diagnosed with multiple myeloma. The data was collected from three datasets, specifically GSE136337, GSE4204, and GSE24080. The Cox regression analysis for uni- and multi-variables included subjects with relevant clinical co-variates from the training cohort (N = 415; GSE136337) and the validation cohort (N = 559; GSE24080). Unfortunately, due to insufficient clinical information, further Cox regression analyses could not be conducted on the second validation cohort (N = 537; GSE4204). Table 1 presents the clinical information for all three datasets, providing a comprehensive overview of the relevant patient characteristics.

In the GSE136337 training cohort, we identified 18 platelet-related genes that showed significant associations with survival through univariable Cox regression analyses (p < 0.001). This is depicted in Figure 2A. To construct the platelet-related risk score, we applied LASSO Cox regression analysis and selected eight genes with high coefficients (Figures 2B–D). Among these genes, Transgelin 2 (TAGLN2), Filamin A (FLNA), Kinesin Family Member 23 (KIF23), Familial hypercholesterolemia (FH), H2B clustered histone 12 like (H2BS1, also known as H2BC12L), and Inhibitor of nuclear factor kappa B kinase regulatory subunit gamma (IKBKG) were identified as high-risk genes, while Chromogenic in situ hybridization (CISH) and Cathepsin W (CTSW) were labeled as low-risk genes. The following formula was used to determine the platelet-related risk score: platelet-related risk score = (0.1023 × expression of TAGLN2) + (0.0277 × expression of FLNA) + (0.1552 × expression of KIF23) + (0.0067 × expression of FH) + (0.0319 × expression of H2BS1) + (0.2144 × expression of IKBKG)—(0.0672 × expression of CISH)—(0.0115 × expression of CTSW). By employing this equation, platelet-related risk scores were computed for every individual in both the training and validation cohorts. Subsequently, employing the median value of each set, the specimens were categorized as high-risk or low-risk groups.

Using Kaplan-Meier curves, we examined the differences in survival rates between the high-risk and low-risk cohorts. Our results revealed that patients categorized as high-risk exhibited inferior survival outcomes in comparison to those classified as low-risk across all datasets (Figure 3A). These findings were further supported by the dot plots (Figure 3C). In the dot plots, dead and alive points represented the survivors and deaths in each dataset respectively. And detailed data have been shown in Table 1. The overall survival time for alive and dead points were elevated in the low-risk group compared to the high-risk group, which showed consistent patterns of worse survival in the high-risk groups in each dataset. In order to assess the precision and accuracy of the platelet-related risk score, a time-dependent ROC analysis was performed. Within the GSE136337 training dataset, the AUC for survival at 1-year, 3-year, and 5-year intervals were found to be 0.627 (95% CI: 0.498–0.756), 0.701 (95% CI: 0.628–0.775), and 0.712 (95% CI: 0.658–0.776) correspondingly, as shown in Figure 3B. Furthermore, we compared the expression patterns of eight PRGs across the three datasets using heat maps. Consistent with the previously discussed formula for platelet-related gene expression, the high-risk group displayed lower expression levels of CISH and CTSW, while the other six genes exhibited an opposite trend (Figure 3D). Notably, the time-dependent ROC curves, heat maps, and dot plots of GSE24080 and GSE4204 demonstrated similar trends to those observed in GSE136337.

We utilized both univariate and multivariate Cox regression analyses to evaluate the forecasting potential of the platelet-related risk score. Additionally, the influence of various clinical factors such as sex, age, albumin, β2-microglobulin, lactic dehydrogenase (LDH), ISS, and R-ISS stage were examined utilizing identical methods in the GSE136337 and GSE24080 datasets (Table 3). In the training dataset, the multivariate analysis revealed a HR of 2.001 (95% CI: 1.467–2.730; p < 0.001) for the platelet-related risk score. Similarly, in the validation dataset, the multivariate analysis showed a HR of 1.530 (95% CI: 1.117–2.097; p = 0.008) for the platelet-related risk score. The findings of the study demonstrated that the platelet-related risk score maintained its independent association with survival outcomes.

Our study conducted an investigation into the correlation between risk scores and various clinical features in GSE136337. Notably, higher levels of LDH, albumin, and β2-microglobulin were consistently observed to be positively correlated with higher risk scores, as illustrated in Figure 4A. Additionally, the study observed a progressive increase in risk scores with higher ISS or R-ISS staging, suggesting a direct relationship between disease severity and the platelet-related risk score.

Remarkably, this study examined the responsiveness of the group at high-risk in comparison to the low-risk group towards different therapeutic agents. The outcomes indicated that the high-risk participants displayed a resistant response to etoposide, methotrexate, lenalidomide, and doxorubicin, as affirmed by the decreased half maximal inhibitory concentration (IC50) values recorded for these drugs (Figure 4B). On the other hand, the group at high-risk exhibited increased sensitivity to AMG.706 (motesanib), axitinib and imatinib, which are vascular endothelial growth factor receptor inhibitors.

These findings provide valuable insights into the relationship between platelet-related risk score and clinical features, further highlighting the potential implications for personalized treatment strategies based on risk stratification.

The group comparison charts, which were created using different algorithms, highlighted the disparities in the immunological microenvironment between the high-risk and low-risk groups.

The low-risk group outperformed the high-risk group in terms of immunological and microenvironment scores, as determined by the xCell technique. The group at high risk exhibited a slightly elevated stromal score while it was not statistically significant. These results imply an increased level of immune cell infiltration in the group at low risk, as well as a greater level of stromal cell infiltration in the high-risk group (Figure 5A).

Furthermore, we compared the differences in immune cells and stromal cells between the high-risk and low-risk groups using the Epic, xCell, and ssGSEA methods (Figures 5E,G,H). Across all three approaches, we consistently noticed an increased extent of immune cell infiltration within the low-risk group. This encompassed the presence of γδ T cells, NK cells, effector CD4⁺ T cells and activated B cells. Intriguingly, we also detected favorable associations between CISH and CTSW expression and distinct immune cell subtypes, including activated NK cells, activated CD4⁺ memory T cells and CD8⁺ T cells, as determined by the application of the CIBERSORT technique (Figure 5D). These cells of the immune system are essential players in the process of combating tumors and are linked to a positive prognosis for patients. These findings further support and validate our initial hypothesis.

We also explored the variance in immune checkpoint molecule expression among the high-risk and low-risk clusters. Remarkably, our findings revealed that the high-risk cohort demonstrated elevated levels of immune checkpoint molecule expression, encompassing PD-L1, PD-L2, and CTLA-4. (Figure 5F). These molecules play an essential function in governing the immune response against tumors and inhibiting the activity of immune cells (Yi et al., 2021; Fan et al., 2022). The potential justification for targeted immunotherapeutic interventions is indicated by the observed increase in PD-L1, PD-L2, and CTLA-4 among individuals in the high-risk group.

Moreover, we utilized the IPS to further assess and compare the immune-related scores between the high-risk and low-risk groups. The IPS incorporates four distinct immune phenotypes: antigen presentation (AP), effector cells (EC), suppressor cells (SC), and checkpoints (CP). The comprehensive measure of tumor immunogenicity, known as the IPS z-score, incorporates the aforementioned scores and enables the prediction of immune checkpoint inhibitor (ICI) therapy response in different cancer types (Givechian et al., 2018). When comparing the groups at high risk and low risk, it was observed that the high-risk group exhibited higher scores for EC, CP, and AZ (Figure 5B). Based on this discovery, it is implied that individuals in the high-risk category might demonstrate heightened receptiveness to ICI therapy. This aligns with the prior analyses conducted on the disparities in immune checkpoint function between these two cohorts. Furthermore, we examined the correlation of IPS scores with eight genes, and we observed a negative correlation between the protective genes (CISH and CTSW) and immunosuppressive cells, as well as a positive correlation with antigen presentation (Figure 5C). These results align with our previous findings, further supporting the notion that platelet-related genes may serve as valuable indicators of the immune status in MM patients.

A thorough examination was carried out to investigate the biological functions linked to both the high-risk and low-risk groups. To delve into the enrichment of KEGG pathways associated with genes related to platelets, GSEA was conducted on each dataset (Figures 6A–C). By performing this analysis, a notable clustering of enriched pathways was observed in the high-risk group, such as proteasome pathway, cell cycle, DNA mending, nucleotide elimination mending, along with the one-carbon reservoir of folate, which display close associations with the progression of tumors.

After examining the Matascape database, we discovered two pathways that exhibited a higher abundance of eight genes related to platelets (Figure 6E). These pathways were cytokine signaling in the immune system and platelet degranulation. Furthermore, using the STRING database, we conducted an analysis of the interactions that occur between the eight PRGs. The results revealed associations between TAGLN2, KIF23, FLNA, and IKBKG, indicating potential interactions and functional relationships between these genes (Figure 6F). The biological mechanisms underlying the high-risk group and the role of platelets in the progression of MM are illuminated by these discoveries. Our understanding of the complex interplay between platelets and tumor development in MM is enhanced by the pathways and protein-protein interactions that have been identified.

The CCLE database revealed that TAGLN2, FH, FLNA, IKBKG, and KIF23 were significantly upregulated at the cellular level. In contrast, CISH and CTSW displayed downregulation in the cellular level. It is worth noting that the expression patterns of these genes except for H2BS1 align with the previously mentioned model formula, indicating their consistency with the platelet-related risk score (Figure 6D). These results offer more proofs for the biological significance and applicability of the discovered platelet-related genes in the setting of multiple myeloma.

Age and platelet-related risk scores were integrated into combined nomograms to predict 1-, 3-, and 5-year survival rates (Figure 7A). Calibration plots were employed to evaluate the performance of the nomograms in forecasting survival rates over the specified timeframes. Notably, the calibration plots of the training cohort demonstrated a high degree of consistency between the predicted and actual values of 1-, 3-, and 5-year (Figure 7B). By incorporating age, ISS, and the platelet-related risk score, the nomogram significantly enhanced the accuracy of 1-, 3-, and 5-year survival predictions in the training cohort. The AUC of 1-, 3-, and 5-year improved from 0.717, 0.639, and 0.668 (using ISS alone) to 0.749, 0.692, and 0.721 (using the nomogram) respectively, indicating the superior performance of the nomogram in survival prediction (Figure 7C). Similarly, the validation dataset (GSE24080) also demonstrated improved prediction accuracy with the nomogram, yielding AUC values of 0.718, 0.687, and 0.692 for 1-, 3-, and 5-year survival respectively, compared to AUC values of 0.677, 0.634, and 0.647 obtained using ISS alone. The performance of the nomogram surpassed other metrics when assessed using DCA curves. Notably, the platelet-related risk score demonstrated superior net gain in survival at 1, 3, and 5 years, further reinforcing its role as a valuable independent prognostic marker (Figure 7D). Using the same method, the nomogram was validated at GSE24080, and the related images are shown in Supplementary Figure S1. In summary, the platelet-related risk score serves as an additional and reliable prognostic marker, complementing the conventional ISS stage.

To further assess the predictive capacity of our identified platelet-related genes in MM, we performed qRT-PCR experiments on MM cell lines and patient samples. In line with the platelet-related risk score formula, the expression levels of TAGLN2, FLNA, KIF23, FH, H2BS1 and IKBKG displayed an upregulation trend in MM1.R and LP-1 MM cell lines (Figure 8A). Conversely, CTSW and CISH exhibited a downregulation trend in these cell lines, consistent with the predicted pattern based on the formula.

Moreover, we extended our analysis to MM patient samples, comparing the expression of the PRGs with normal controls. The results demonstrated that CISH and CTSW expression was lower in MM patient samples compared to the normal controls, while the other six genes exhibited an opposite trend, showing higher expression levels in MM patient samples (Figure 8B).