A novel prognostic tool for triple-negative breast cancer: creating and testing a parthanatos-related gene model

Abstract

Background

Parthanatos is a poly (ADP-ribose) polymerase 1 (PARP1)-dependent programmed cell death pathway. However, the association between parthanatos and triple-negative breast cancer (TNBC) as well as its impact on patient prognosis remains unclear. This study aims to investigate the prognostic value of parthanatos in TNBC and to develop a predictive model for identifying potential prognostic biomarkers.

Methods

Gene expression profiles of TNBC cases were acquired from the publicly available dataset of The Cancer Genome Atlas (TCGA). After reviewing existing literature on parthanatos, we pinpointed 31 genes related with this process, hereafter termed parthanatos-related genes (PRGs). Using correlation analysis, we identified 1,569 messenger RNAs (mRNAs) showing strong associations (correlation coefficient >0.6) with these PRGs, which we defined as parthanatos-related mRNAs (PR-mRNAs). Combining these PR-mRNAs with the initial 31 PRGs yielded a comprehensive gene set of 1,600 genes. Initial screening via univariate Cox analysis identified 27 candidate genes demonstrating significant prognostic value (P<0.05) and a prognostic risk model was developed using least absolute shrinkage and selection operator (LASSO) regression analysis. To systematically evaluate the predictive performance of the constructed model, we conducted comprehensive validation using Kaplan-Meier survival analysis, receiver operating characteristic (ROC) curve analysis and calibration plots. Finally, by analyzing the genes in the model, NECTIN2 was selected for evaluation of its potential biological functions.

Results

This investigation developed a 12 PR-mRNAs signature for prognostic risk stratification in TNBC. The model demonstrated strong predictive accuracy across temporal assessments, with area under the curve (AUC) values exceeding 0.89 in the training set (1-year: 0.943, 3-year: 0.980, 5-year: 0.896) and maintaining robust performance in both validation cohorts (testing set: 1-year 0.968, 3-year 0.795, 5-year 0.876; combined cohort: 1-year 0.947, 3-year 0.922, 5-year 0.887). Bootstrap resampling validation confirmed the model’s stability and reproducibility in clinical outcome prediction. Notably, elevated NECTIN2 expression correlated significantly with poorer overall survival (P<0.01), a finding supported by functional studies showing that NECTIN2 knockdown in MDA-MB-468 cells significantly attenuated both proliferative capacity and migratory potential.

Conclusions

Through systematic analysis, we developed a 12 PR-mRNAs signature that effectively predicts clinical outcomes in TNBC patients. The newly established risk assessment model not only provides a reliable tool for prognostic evaluation but also reveals NECTIN2 as a clinically significant biomarker, with elevated expression correlating with poorer survival outcomes.

Highlight box.

Key findings

• A prognostic model was constructed to explore the association between parthanatos-related molecules and triple-negative breast cancer (TNBC) outcomes, with preliminary investigation of potential biomarkers.

What is known and what is new?

• Parthanatos, a unique form of programmed cell death dependent on poly (ADP-ribose) polymerase 1 activation, contributes to the complex regulation of tumor initiation and progression through its involvement in cell death control and DNA damage response mechanisms. Currently, the prognostic significance of parthanatos in TNBC remains unclear.

• We constructed a prognostic correlation model to investigate their clinical significance and identify candidate prognostic markers.

What is the implication, and what should change now?

• These results offer preliminary insights into TNBC prognostic stratification and parthanatos-related oncogenic mechanisms, while the functional implications of model-identified genes in parthanatos-TNBC interplay require further experimental verification.

Introduction

Globally recognized as a critical women’s health concern, breast cancer manifests in distinct molecular subtypes with varying clinical implications. Particularly noteworthy is triple-negative breast cancer (TNBC), characterized by its lack of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) expression (1). This aggressive variant has become a focal point in oncological research owing to its distinct pathological features and therapeutic challenges. Due to the absence of these targets, patients with TNBC do not benefit from endocrine therapy or anti-HER2 treatment (2). Currently, surgery and chemotherapy remain the core strategies for treating TNBC. However, because TNBC is highly aggressive and exhibits strong heterogeneity, there are significant differences in biological behavior and treatment responses among patients (3). As a result, the overall prognosis of TNBC remains poor.

As a programmed cell death mechanism distinct from classical apoptosis or necrosis, parthanatos represents a unique poly (ADP-ribose) polymerase 1 (PARP1)-dependent pathway characterized by PARP1 activation (4). As a complex programmed cell death mechanism, parthanatos involves several key steps, including the sensing of DNA damage, activation of PARP1, formation of poly (ADP-ribose) (PAR) (5), translocation of HMGB1, and transmission of inflammatory signals (6). Parthanatos has been implicated in the pathogenesis of multiple disease states, demonstrating significant involvement in neurodegenerative disorders (7,8), cellular damage following ischemic events (9,10), and malignant transformation processes. For example, BrA (a glycosylated flavonoid extracted from plants) has demonstrated significant anti-tumor effects on metastatic prostate cancer cells in vitro (11), with PARP-mediated cell death (parthanatos) being one of its potential anti-tumor mechanisms. Additionally, emerging evidence demonstrates that pharmacological inhibition of N-myristoyltransferase (NMT) potentiates the cytotoxic effects of platinum-based chemotherapeutics in lung cancer models through mitochondrial iron accumulation and subsequent induction of parthanatos (12). Existing research has identified significant differences in the expression of parthanatos-related genes (PRGs) between breast cancer and normal breast tissues, with findings suggesting that parthanatos activation may serve as a potential biomarker for PARP inhibitor (PARPi) resistance in breast cancer (13).

Therefore, the present study was designed to systematically explore the functional significance of parthanatos-associated genes in TNBC pathogenesis and disease progression. To achieve this, we collected 31 PRGs reported in previous studies (6,14-16). We screened for messenger RNAs (mRNAs) closely associated with these genes through correlation analysis using TNBC mRNA expression profiles from The Cancer Genome Atlas (TCGA) database, and defined the significantly correlated mRNAs as parthanatos-related mRNAs (PR-mRNAs). Subsequently, we constructed a prognostic risk score model based on PR-mRNAs using Cox regression analysis and least absolute shrinkage and selection operator (LASSO) regression analysis. This model has the potential to be applied in clinical practice, helping to identify high-risk populations, explore potential therapeutic targets, and provide new insights into improving the prognosis of TNBC patients. We present this article in accordance with the MDAR and TRIPOD reporting checklists (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-819/rc).

Methods

Acquisition of TNBC datasets and associated clinical data

In this study, gene expression datasets and detailed clinical information for TNBC were obtained from the TCGA database (https://portal.gdc.cancer.gov/), encompassing a total of 122 TNBC patients. The transcriptome data were extracted in the form of fragments per kilobase million (FPKM). Additionally, clinical and pathological characteristics of the patients were collected, including age, clinical stage [tumor (T), node (N), and metastasis (M) classifications], overall survival (OS) time, and survival status. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Identification of PR-mRNAs

After summarizing existing research, 31 PRGs were identified. The mRNA expression profiles of 122 TNBC patients were integrated with these 31 genes, and genes expressed in more than half of the samples were selected for further analysis. Pearson correlation analysis was then conducted between these genes and the 31 PRG. Using the “limma” R package, 1,569 PR-mRNAs were filtered, and these mRNAs were confirmed as PR-mRNAs based on the criteria: |correlation coefficient| >0.6 and P<0.001.

Construction of the PR-mRNA-based prognostic model

To develop a predictive gene model for survival analysis, we began by merging 1,569 genes identified through correlation studies with 31 known PRGs, resulting in a pool of 1,600 potential markers. They then conducted univariate Cox regression analysis on this combined set, narrowing it down to 27 statistically significant genes (P<0.05) for further model development. During sample preparation, the team excluded cases with OS durations shorter than 30 days from the original 122-patient dataset, retaining 116 eligible samples. These were randomly split into training (n=84) and validation (n=32) cohorts at roughly a 3:1 ratio. Finally, LASSO regression was employed to pinpoint 12 PR-mRNAs with strong prognostic relevance. According to the correlation coefficients calculated by LASSO regression analysis (17), a risk prediction model involving 12 genes was constructed: Risk score = ∑Coef(PR-mRNAs) × Exp(PR-mRNAs). The Coef(PR-mRNAs) represented the regression coefficient obtained from the LASSO regression analysis, while the Exp(PR-mRNAs) represented the expression value of the modeling gene (18). The 12 genes included PARP1, SCGB1D2, NECTIN2, TFB2M, SNAP47, CDK14, SIAE, ING1, ZNF454, MASP1, HMGA2, and IL21. Using the above formula, the risk score for each sample could be calculated, and the median of all sample risk scores was used as the cutoff value to classify samples into high- and low-risk groups.

Evaluating the performance of the risk prognostic model

To explore potential associations between risk stratification and clinical outcomes, we first evaluated the distribution of prognostic scores across the TNBC cohort. Survival disparities between high-risk and low-risk groups were systematically examined through Kaplan-Meier methodology. These analyses were performed using the “survival” and “survminer” packages. The model’s temporal predictive accuracy was validated by constructing receiver operating characteristic (ROC) curves for 1-, 3-, and 5-year survival endpoints. To determine the independent prognostic value of the risk signature, we conducted multivariate Cox regression analyses. Comparative performance evaluation incorporated both ROC curve analysis and concordance index (C-index) calculations, benchmarking the molecular signature against conventional clinical parameters including age, tumor stage, and nodal status. Throughout this process, the analyses were primarily carried out using R packages including “survival”, “survminer”, “timeROC”, “dplyr”, “rms”, and “pec”.

Construction of the nomogram and its validation

For the purpose of survival prediction, a nomogram was formulated based on factors like age, stage, T, N, and risk score, with its accuracy being evaluated through calibration curves (19). R packages utilized for these analyses included “survival”, “regplot”, “rms” and “survcomp”.

Bootstrap internal validation

The model was verified internally by Bootstrap twice, resampling 1,000 times and 5,000 times respectively. The R packages “survival” and “boot” were employed for this assessment.

Functional enrichment analysis

The 12 PR-mRNAs used in model construction were combined with 31 PRGs. Next, the identified genes underwent Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis.

Protein-protein interaction network analysis

The protein-protein interaction (PPI) network analysis of the genes included in the enrichment analysis was conducted using the STRING database.

Assessment of immune infiltration levels and immunotherapy response outcomes

To characterize immune microenvironment differences between risk-stratified TNBC patients, we employed CIBERSORT deconvolution to quantify variations in 22 distinct immune cell populations across high- and low-risk cohorts. Immunotherapeutic response potential was evaluated through systematic comparison of Tumor Immune Dysfunction and Exclusion (TIDE) scores between prognostic groups. Furthermore, we conducted comprehensive immune landscape profiling using single-sample gene set enrichment analysis (ssGSEA) to investigate potential associations between the 12-gene prognostic signature and 28 immune cell subsets. All computational analyses were implemented in R utilizing specialized packages for immune deconvolution (‘e1071’, ‘preprocessCore’), data visualization (‘ggsci’, ‘ggpubr’), statistical computing (‘parallel’, ‘tidyr’), and gene set variation analysis (‘GSVA’, ‘limma’), with data management facilitated by ‘tidyverse’ and ‘data.table’ frameworks.

Assessment of drug sensitivity variation

The Genomics of Drug Sensitivity in Cancer 2 (GDSC2) database (https://www.cancerrxgene.org/) was used to investigate possible differences in drug sensitivity between high- and low-risk groups. For this purpose, the R packages ‘limma’, ‘oncoPredict’, and ‘parallel’ were utilized.

Development and confirmation of MDA-MB-468 cells with stable transfection

To validate the potential prognostic marker NECTIN2 in TNBC, this study constructed a stable cell line with NECTIN2 gene knockdown. Based on previous research, the MDA-MB-468 cell line was identified as a human-derived TNBC cell line with high expression of the NECTIN2 gene (20). Therefore, this study selected the MDA-MB-468 cell line (RRID: CVCL_0419) for constructing a stable NECTIN2-knockdown cell line. The MDA-MB-468 cell line was obtained from Genomeditech (Shanghai, China; procurement contract: GM-CS-144559). The construction and pre-validation of the stable transfected cells were completed by Genomeditech. The resulting stable cell lines were used for subsequent experiments. Pre-validation included quantitative real-time polymerase chain reaction (qRT-PCR) to detect NECTIN2 expression levels and Western blot analysis to confirm the gene silencing efficiency. Upon receiving the pre-validated genetically engineered cells from the Genomeditech, we conducted independent Western blot assays to reconfirm the target protein expression profile. In the aforementioned Western blot experiments, the rabbit anti-NECTIN2 antibody (Cat# ab135246, RRID: AB_2936435, Abcam, Cambridge, UK) was consistently used.

Western blotting analysis of protein expression

The NECTIN2 stable knockdown cell line (shNECTIN2) and its negative control cells (ncNECTIN2) were collected and lysed on ice for 30 minutes using RIPA lysis buffer (R0010, Solarbio, Beijing, China) containing protease inhibitor (CW2200S, CWBIO, Beijing, China), phosphatase inhibitor (CWBIO, CW2383S), and phenylmethylsulfonyl fluoride (PMSF). The lysates were centrifuged at 12,000 rpm for 20 minutes at 4 °C, and the supernatants were collected. Protein concentration was determined using a bicinchoninic acid (BCA) protein assay kit (PC0020, Solarbio). After electrophoresis, proteins were transferred to a polyvinylidene fluoride (PVDF) membrane (IPVH00010, Millipore, MA, USA) and blocked with rapid blocking buffer (C220701, YangGuangBio, Beijing, China). The membranes were incubated overnight at 4 °C with primary antibodies, including rabbit anti-NECTIN2 antibody (Cat# ab135246, RRID: AB_2936435, Abcam) and rabbit anti-GAPDH antibody (Cat# 10494-1-AP, RRID: AB_2263076, Proteintech, Wuhan, China). The membranes were then washed three times with 1× Tris-Buffered Saline with Tween 20 (TBST) (5 minutes each) and incubated with horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG secondary antibody (Cat# ab205718, RRID: AB_2819160, Abcam) for 1 hour at room temperature. After additional TBST washes, protein bands were visualized using an enhanced chemiluminescence (ECL) chemiluminescence kit (Fluorescence, 046-100ML).

Cell viability assessment by Cell Counting Kit-8 (CCK-8) procedure

This investigation utilized the CCK-8 procedure to measure cell viability in a quantitative manner. Cells in the logarithmic growth phase were seeded into 96-well plates at a concentration of 1,000 cells per well, suspended in 100 µL of culture medium. A blank control (containing only medium) was included, with each experimental condition tested in triplicate. At designated time points (0, 24, 48, 72, and 96 hours), the medium was carefully aspirated and replaced with 100 µL of fresh medium containing CCK-8 solution. Following brief agitation, the plates were incubated for 1–4 hours before absorbance readings at 450 nm were taken using a microplate reader.

Detection of cell migration ability by Transwell migration assay

Cells were harvested when they reached approximately 70% confluency. The day before the experiment, cells were starved in serum-free medium for 24 hours. Subsequently, the serum-free medium was aspirated, and the cells were washed with phosphate buffered saline (PBS), trypsinized, and digestion was terminated. After centrifugation, the supernatant was discarded, and the cells were resuspended in serum-free medium. Cell counting was performed, and the cell density was adjusted to 2×10⁵ cells/mL. A 24-well plate with Transwell inserts was prepared, and 600 µL of complete culture medium containing 10% fetal bovine serum (FBS) was added to the lower chamber of each well, while 200 µL of the cell suspension was added to the upper chamber. The plate was incubated in a cell culture incubator for 48 hours. After incubation, non-migrated cells on the upper surface of the Transwell membrane were gently removed with a cotton swab. Migrated cells on the lower surface were fixed with 4% paraformaldehyde for 15 minutes and stained with crystal violet for 10 minutes. Cells were examined microscopically, imaged, and tallied across five randomized fields. The mean migration count was determined.

Statistical analysis

This investigation incorporated multiple analytical approaches to evaluate different biological parameters. Survival probability differences across risk strata were examined through Kaplan-Meier methodology with log-rank testing for statistical comparison. Cellular proliferation dynamics were quantitatively measured via CCK-8 assays, while migratory potential was determined using Transwell chamber systems. The proliferation data underwent two-factor analysis of variance (ANOVA), with Transwell results analyzed by paired Student’s t-tests to ascertain intergroup differences. All computational analyses were conducted using R version 4.4.1 or GraphPad Prism (RRID:SCR_002798) 8.0.2, applying a significance threshold of P<0.05 for all statistical tests.

Results

Detection of PR-mRNAs and establishment of a prognostic model

Through correlation analysis, we identified 1,569 mRNAs that were associated with the 31 PRGs (Table S1), and performed the Sankey diagram drawing of these 1,569 PR-mRNAs and the 31 PRGs to illustrate the relationship of expression flows between them (Figure 1A). Subsequently, we merged the 1,569 mRNAs with the 31 PRGs to obtain a total of 1,600 genes. To identify genes with prognostic significance, we performed a univariate Cox regression analysis on all 1,600 candidate genes. This screening process yielded 27 mRNAs significantly associated with patient outcomes (Figure 1B, P<0.05). After excluding TNBC cases where survival duration was shorter than 30 days from the initial pool of 122 samples, we retained 116 TNBC cases for subsequent analysis. These remaining samples were randomly split into training (n=84) and validation (n=32) cohorts at roughly a 3:1 ratio, with no meaningful differences observed in clinical or pathological features between the two groups. Using LASSO regression on the 27 candidate genes, we further refined our selection to 12 key genes for developing the predictive risk model (Figure 1C,1D). These genes were PARP1, SCGB1D2, NECTIN2, TFB2M, SNAP47, CDK14, SIAE, ING1, ZNF454, MASP1, HMGA2, and IL21. Based on the regression coefficients and mRNA expression levels for each sample, the following formula was used to calculate the risk score for each sample: risk score = (0.966367141501677 × PARP1) + (0.00360875276621275 × SCGB1D2) + (0.424905201978442 × NECTIN2) + (−0.465484466567477 × TFB2M) + (−1.8690783634075 × SNAP47) + (−0.838545224900855 × CDK14) + (0.285854742041385 × SIAE) + (−0.997681594020226 × ING1) + (−0.0174435762657747 × ZNF454) + (3.38820137519607 × MASP1) + (−0.379608127277676 × HMGA2) + (−1.93276257178251 × IL21). Multivariate Cox regression analysis was performed on the 12 PR-mRNAs included in the model, and six genes—PARP1, NECTIN2, SNAP47, CDK14, HMGA2, and IL21—were found to have independent prognostic significance (Figure 1E).

Figure 1.

The identification of PR-mRNAs and model construction. (A) Sankey diagram of parthanatos-related genes and their associated mRNAs. (B) Forest plot of univariate Cox analysis for 27 prognosis-related mRNAs. (C,D) LASSO regression analysis for prognostic model construction. (E) Forest plot of multivariate Cox analysis for the 12 genes obtained from LASSO analysis. H95CI, upper 95% confidence interval limit; HR, hazard ratio; L95CI, lower 95% confidence interval limit; LASSO, least absolute shrinkage and selection operator; mRNAs, messenger RNAs; PR-mRNAs, parthanatos-related mRNAs.

Survival assessment of the prognostic model

The categorization into high- and low-risk groups was based on the median risk score of TNBC samples. We generated risk score distribution diagrams for the training set (Figure 2A), testing set (Figure 2B), and total set (Figure 2C). The data visualizations consistently revealed a clear correlation between elevated risk scores and higher mortality rates. Additionally, Kaplan-Meier survival analyses were generated for the training cohort (Figure 2D), validation set (Figure 2E), and combined population (Figure 2F). Each survival analysis demonstrated significantly poorer outcomes for high-risk patients when stacked up against their low-risk counterparts (P<0.05). To assess the model’s predictive accuracy, ROC curves were subsequently analyzed. As shown in Figure 2G, the training cohort exhibited strong predictive performance, with area under the curve (AUC) scores of 0.943, 0.980, and 0.896 for 1-, 3-, and 5-year intervals, respectively. The test cohort (Figure 2H) yielded similarly robust results, posting AUC values of 0.968, 0.795, and 0.876 across the same timeframes. When analyzing the complete TCGA-TNBC dataset (Figure 2I), the model maintained its precision, achieving AUCs of 0.947, 0.922, and 0.887 for the respective annual benchmarks. These consistently high AUC values underscore the model’s diagnostic reliability and its potential utility in real-world clinical settings.

Figure 2.

Survival assessment of the prognostic model. Risk score distribution plots (A-C), Kaplan-Meier survival curves (D-F), and time-dependent ROC curves (G-I) in the training set, testing set, and the full dataset. AUC, area under the curve; ROC, receiver operating characteristic.

Analysis of the model’s prognostic effectiveness

To determine the independent prognostic value of our risk stratification system, multivariate Cox regression analysis was performed to assess potential associations between the computed risk scores and established clinical parameters (Figure 3A). Subsequent analyses compared the prognostic performance of the risk score with that of clinicopathological factors. Notably, the risk score demonstrated superior predictive accuracy compared to conventional prognostic indicators, achieving an AUC of 0.922 (Figure 3B)—outperforming traditional clinical variables including patient age, disease stage, tumor classification (T stage), and nodal involvement status (N stage). To better assess how the risk score influences prognostic outcomes, we developed a nomogram (Figure 3C) and generated calibration curves (Figure 3D) along with a C-index plot (Figure 3E) to validate its predictive accuracy. The calibration curves for 1-, 3-, and 5-year survival rates aligned nearly perfectly with the ideal reference line. Additionally, the prognostic model achieved a robust C-index of 0.878, demonstrating the nomogram’s strong ability to distinguish between outcomes and its reliability in forecasting both near-term and extended prognosis.

Figure 3.

Model performance validation. The multivariate Cox analysis (A), ROC curve analysis (B), nomogram (C), calibration curve (D), and C-index plot (E) of risk score and clinical pathological characteristics. **, P<0.01. AUC, area under the curve; C-index, concordance index; CI, confidence interval; N, node; OS, overall survival; ROC, receiver operating characteristic; T, tumor.

Internal validation

Internal validation of the model was performed using the Bootstrap method. Two rounds of Bootstrap internal validation were conducted: the first round involved 1,000 resampling iterations, and the second round involved 5,000 resampling iterations. This approach allowed for an assessment of the model’s stability and performance under different levels of resampling. When resampled 1,000 times (Figure 4A), the mean C-index of the model was 0.9088074 and the standard deviation of C-index was 0.02633089. When the sample was resampled 5,000 times (Figure 4B), the mean C-index of the model was 0.9071016 and the standard deviation of C-index was 0.0286076. The consistent C-index values and relatively low standard deviations across both resampling scenarios indicate that the model maintains high robustness and reliability regardless of the number of resampling iterations.

Figure 4.

Bootstrap internal validation. (A) Results after resampling 1,000 times. (B) Results after resampling 5,000 times. C-index, concordance index.

Integration of PRGs and model-incorporated PR-mRNAs for functional analysis

To gain deeper insights into the biological roles and relationships of the genes linked to parthanatos—along with the 12 PR-mRNAs chosen for the prognostic model—we conducted comprehensive GO and KEGG enrichment studies, supplemented by PPI network mapping. First, we merged the 31 PRGs with the 12 PR-mRNAs to create a unified gene pool. This combined dataset was then analyzed using GO and KEGG to pinpoint overrepresented biological processes, cellular structures, molecular functions, and key signaling pathways (Figure 5A,5B). The GO and KEGG enrichment analysis results suggest that the gene set is significantly associated with stress responses, DNA repair, and cell death pathways. The PPI network (Figure 5C) reveals potential interactions within this gene set. Key nodes in the network include PARP1, DCAF10, OTUD1, MCL1, DDB1 and PTEN, suggesting their potential significance in the enriched biological pathways.

Figure 5.

GO and KEGG enrichment analysis and PPI network. (A) GO and KEGG enrichment analysis bubble plot. (B) GO and KEGG enrichment analysis bar chart. (C) PPI network analysis. BP, biological process; CC, cellular component; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; PPI, protein-protein interaction.

Immunological and molecular correlation analysis

To further explore the immunological and molecular characteristics of the samples in relation to the 12 modeling genes and parthanatos process, we conducted a comprehensive analysis as depicted in Figure 6. Initially, a rainbow plot (Figure 6A) was used to display the examination results of how risk scores relate to immune cell infiltration in tumors. To evaluate differential responses to immunotherapy between risk-stratified cohorts, we computed TIDE scores, with violin plot visualization (Figure 6B) revealing significantly lower values in high-risk patients, indicative of enhanced potential immune checkpoint inhibitor efficacy. Concurrent immunogenomic analysis employed bubble plot representation (Figure 6C) to systematically characterize associations between the 12 modeling genes and 28 immune cell populations. Notably, PARP1 demonstrated marked positive correlation with activated CD4⁺ T lymphocytes while showing inverse association with neutrophil infiltration. The NECTIN2 gene displayed significant negative regulation of Th2 cell recruitment. IL-21 exhibited pleiotropic immunomodulatory effects, positively coordinating with activated B cells, immature B lymphocytes, and myeloid-derived suppressor cells, yet negatively regulating memory B cells, Th2, and Th17 populations. Moreover, a heatmap analysis (Figure 6D) uncovered correlations between the 12 modeling genes and 31 PRGs, providing insights into possible regulatory networks and molecular interactions. Finally, the comparative assessment of immune cell infiltration patterns in the high- and low-risk cohorts was visualized through box plot analysis (Figure 6E). Notably, the high-risk cohort demonstrated significantly elevated infiltration rates across several cell types, particularly regulatory T cells (Tregs), monocytes, and activated dendritic cells.

Figure 6.

Immunological and molecular correlation analysis. (A) Rainbow plot of immune cell infiltration in TNBC samples. (B) Violin plot of TIDE scores between high-risk and low-risk groups. (C) Bubble plot of the correlation between model genes and immune cells. (D) Heatmap of the correlation between model genes and parthanatos-related genes. (E) Box plots of immune cell infiltration differences between high-risk and low-risk groups. FDR, false discovery rate; NK, natural killer; TIDE, Tumor Immune Dysfunction and Exclusion; TNBC, triple-negative breast cancer

The relationship between the risk score and drug sensitivity

To evaluate potential therapeutic implications, drug sensitivity analysis was conducted according to risk stratification. Using the “oncoPredict” algorithm in R, computational predictions of pharmacological responses revealed significant variations in drug efficacy between prognostic subgroups. The results indicated that high-risk patients exhibited significantly greater sensitivity to six specific therapeutic agents—AZD7762, ibrutinib, IGF1R_3801, niraparib, VE-822, and WEHI-539—compared to their low-risk counterparts, as illustrated in Figure 7.

Figure 7.

Drug sensitivity analysis. Sensitivity differences of AZD7762 (A), ibrutinib (B), IGF1R_3801 (C), niraparib (D), VE-822 (E), and WEHI-539 (F) between high-risk and low-risk groups.

A preliminary study on the impact of NECTIN2 gene in TNBC

Based on the analysis of 12 candidate genes, we ultimately selected NECTIN2 for further investigation of its role in TNBC. Initial analysis involved 11 matched pairs of TNBC and adjacent normal tissue samples obtained from the TCGA database, with paired comparative analysis demonstrating significantly elevated NECTIN2 expression in tumor specimens relative to normal controls (P<0.001, Figure 8A). Subsequent stratification of TNBC cases into high- and low-NECTIN2 expression subgroups, based on median expression thresholds, revealed a strong association between elevated NECTIN2 levels and adverse OS outcomes in Kaplan-Meier analysis (Figure 8B). To functionally characterize NECTIN2 in TNBC pathogenesis, stable NECTIN2 knockdown models were established using the MDA-MB-468 cell line, with successfully pre-validated by Genomeditech (Shanghai, China) at transcriptional (qRT-PCR, Figure 8C) and protein (Western blot, Figure 8D) levels. Seven shRNA targets (S1-S7) designed by Genomeditech were tested, and S5 was chosen based on qRT-PCR and Western Blot results. We also independently verified the knockdown efficiency of the target gene through Western blot analysis (Figure 8E). Using these stable knockdown cells (shNECTIN2) and their control counterparts (ncNECTIN2), we investigated differences in cellular behavior. Transwell migration assays (Figure 8F,8G) demonstrated significantly reduced migration capacity in shNECTIN2 cells. Additionally, CCK-8 assays (Figure 8H) revealed that the proliferation ability of shNECTIN2 cells was also significantly lower than that of ncNECTIN2 cells.

Figure 8.

Preliminary study of the NECTIN2 gene in TNBC. (A) Paired plot of NECTIN2 expression in matched TNBC samples. (B) Kaplan-Meier survival curve between high and low NECTIN2 expression groups. (C) Pre-validation of knockdown efficiency by qRT-PCR (Genomeditech). (D) Pre-validation of knockdown efficiency by Western blot (Genomeditech). (E) Independent Western blot validation of knockdown efficiency. (F,G) Results of the Transwell migration assay (crystal violet staining; magnification, ×10). (H) Results of the CCK-8 assay. ***, P<0.001; ****, P<0.0001. CCK-8, Cell Counting Kit-8; ncNECTIN2, negative control cells; shNECTIN2, NECTIN2 stable knockdown cell line; qRT-PCR, quantitative real-time polymerase chain reaction; TNBC, triple-negative breast cancer.

Discussion

Parthanatos is a PARP1-dependent and caspase-independent programmed cell death pathway (21). It differs from apoptosis, necrosis, or other known forms of cell death, as its occurrence involves multiple steps and plays an important role in the mediation of tumor diseases (4). Some existing studies indicate that parthanatos might represent a potentially efficacious method for inducing cancer cell death, ultimately facilitating the attainment of anti-tumor objectives. However, current research on parthanatos in TNBC is relatively limited. Therefore, to deepen the understanding of the role of parthanatos in TNBC and identify reliable prognostic markers for TNBC, this study developed a risk prognostic model based on PR-mRNAs.

This investigation systematically identified 31 genes associated with parthanatos through a comprehensive literature review. Leveraging transcriptomic data from TNBC patients in the TCGA database, we identified 1,569 mRNAs exhibiting significant correlations with these PRGs. Subsequent univariate Cox regression followed by LASSO penalized regression analysis enabled the development of a robust 12-gene prognostic signature (comprising PARP1, SCGB1D2, NECTIN2, TFB2M, SNAP47, CDK14, SIAE, ING1, ZNF454, MASP1, HMGA2, and IL21) for TNBC outcomes. The model’s predictive efficacy was validated through multiple approaches. ROC curve analysis revealed strong prognostic discrimination across all datasets, including the training cohort, testing subset, and combined cohort. Consistent with these findings, Kaplan-Meier survival analysis demonstrated markedly worse OS in high-risk patients compared to their low-risk counterparts (P<0.05) in all evaluated datasets, further corroborating the model’s clinical utility for risk stratification.

Among the 12 prognostic mRNAs comprising the established model, PARP1 emerges as a pivotal regulator of parthanatos, with well-documented involvement in tumorigenesis and cancer progression across multiple malignancies, positioning it as a potential therapeutic target (22). As a nuclear protein, PARP1 modulates diverse cellular processes, particularly DNA damage repair pathways (23). This biological function underpins the mechanism of PARPi, which exhibit antitumor activity by impairing DNA repair capacity (23,24). Notably, the synthetic lethality strategy employing PARPi has demonstrated clinical success in treating BRCA1/2-deficient tumors (25). SIAE modulates tumorigenesis and immune evasion through regulation of sialic acid acetylation, with dysregulated expression patterns observed in hematological malignancies such as leukemia and multiple myeloma (26-28). ING1, a well-established tumor suppressor, demonstrates frequent downregulation or loss of expression across multiple carcinomas including head and neck, esophageal, and breast cancers, implicating its critical role in malignant transformation (29). Epigenetic profiling identifies ZNF454 as exhibiting cancer-associated hypermethylation in oral precancerous lesions, lung squamous cell carcinoma, and endometrial malignancies (30,31). Notably, in lung squamous cell carcinoma, ZNF454 hypermethylation correlates with transcriptional silencing and improved clinical outcomes, supporting its potential dual role as both tumor suppressor and prognostic indicator (32). MASP1 encodes a serine protease with multifunctional roles in innate immunity, particularly through lectin pathway activation and coagulation cascade initiation (33). While its precise oncogenic mechanisms remain incompletely characterized, MASP1 demonstrates tumor-suppressive properties in gastric carcinogenesis, evidenced by significant downregulation in malignant versus normal tissue. Furthermore, its expression levels show positive associations with tumor-infiltrating immune populations—including dendritic cells, neutrophils, macrophages, and lymphocytes—suggesting microenvironmental immunomodulatory functions that may influence cancer progression (34). HMGA2 demonstrates consistent upregulation across multiple malignancies, where it orchestrates critical carcinogenic pathways including cell cycle control, proliferative signaling, and epithelial-mesenchymal transition, underscoring its fundamental contribution to tumor initiation and progression (35,36). IL-21, a pleiotropic cytokine secreted predominantly by CD4⁺ T lymphocytes, mediates complex immunomodulatory effects through paracrine action on B cells (37). This cytokine displays context-dependent duality in tumor biology, exhibiting both antitumor immunity and protumorigenic activity. Published research has established that the therapeutic effectiveness of CTLA-4 blockade relies on the accurate control of T cells through IL-21 signaling (38). Genomic profiling confirms SCGB1D2 as a pan-cancer differentially expressed gene. Notably, while demonstrating marked overexpression in ovarian carcinoma, its reduced expression shows association with poorer patient prognosis (39). In breast cancer pathogenesis, SCGB1D2 expression is significantly overexpressed in the samples of histologically normal epithelium (HNEpi) from patients with luminal breast cancer, and this gene also has a relatively high rate of dysregulation in low-grade luminal breast cancer (40). NECTIN2, an immunoglobulin superfamily adhesion molecule, maintains neural homeostasis through astrocyte-neuron interactions and synaptic organization (41). Its oncological relevance is evidenced by clinical associations, where co-expression with Nectin-4 predicts reduced survival in intermediate-grade glioma patients (42). Emerging evidence underscores the clinical significance of NECTIN2 across multiple malignancies. In pancreatic ductal adenocarcinoma, co-expression patterns of Nectin-2 and DDX3 demonstrate strong associations with advanced disease stage and unfavorable clinical outcomes (43). Similar prognostic correlations are observed in gallbladder carcinoma, where immunohistochemical detection of these markers correlates with aggressive clinicopathological features and reduced survival (44). Functional studies in ovarian cancer models reveal that NECTIN2 silencing not only suppresses tumor cell proliferation but also enhances T-cell-mediated antitumor immunity (45). Among other signature genes, TFB2M maintains mitochondrial homeostasis through its essential role in mitochondrial DNA (mtDNA) transcription and chromatin organization (46), while SNAP47 facilitates vesicular trafficking through its predominant localization in the endoplasmic reticulum (ER)-Golgi network and involvement in membrane fusion events (47). The cell cycle regulator CDK14 emerges as a critical oncogenic mediator, with mechanistic studies demonstrating its post-transcriptional regulation by tumor-suppressive miRNA miR-216a in osteosarcoma, where this regulatory axis inhibits malignant phenotypes including proliferation and metastasis (48). This regulatory mechanism is conserved in hepatocellular carcinoma, where miR-1202-mediated CDK14 suppression contributes to its tumor-inhibitory effects (49).

As the most clinically aggressive breast cancer subtype, TNBC presents critical challenges in risk stratification and therapeutic management. This study highlights the prognostic significance of PRGs in TNBC, establishing a 12-gene signature through comprehensive analysis of TCGA datasets that may facilitate early identification of high-risk patients and precision medicine approaches (18). This gene signature not only effectively stratifies prognostic outcomes in TNBC patients, but may also provide molecular targets for developing novel targeted therapies, thereby addressing the current therapeutic limitations in TNBC treatment. Particularly for chemotherapy-resistant or recurrence-prone patient subgroups, this biomarker panel could serve as a valuable reference for individualized therapeutic decision-making. While these molecular markers demonstrate potential clinical utility, several limitations warrant consideration. The exclusive reliance on publicly available genomic data may introduce population selection biases, potentially limiting the model’s generalizability to diverse TNBC patient cohorts. Furthermore, heterogeneity in sequencing methodologies across source studies could impact the reproducibility of gene expression measurements, necessitating standardization protocols for clinical translation. Notably, functional validation remains incomplete, as only one candidate gene underwent detailed phenotypic characterization. Future investigations should prioritize mechanistic studies to elucidate the oncogenic contributions of all signature genes through systematic molecular and cellular experiments.

Conclusions

This study successfully developed a prognostic model for TNBC incorporating 12 PR-mRNAs. The model demonstrated excellent predictive accuracy across different time points. Bootstrap resampling validation further confirmed its stability and reproducibility in clinical outcome prediction. Among these PR-mRNAs, we focused on NECTIN2 for further investigation. Our analysis revealed significantly higher NECTIN2 expression in tumor tissues compared to paired normal tissues (P<0.001). Clinically, TNBC patients with elevated NECTIN2 expression showed poorer OS (P<0.01). Functional studies demonstrated that NECTIN2 knockdown in MDA-MB-468 cells effectively inhibited both proliferation and migration capacities. These findings provide preliminary evidence supporting the prognostic value of parthanatos in TNBC and suggest potential therapeutic targets within this pathway. However, the precise mechanisms of parthanatos-related molecules in TNBC pathogenesis warrant further investigation.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Data Sharing Statement

Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-819/dss

tcr-14-10-7018-dss.pdf ^{(71.3KB, pdf)}

DOI: 10.21037/tcr-2025-819