Recent decades have witnessed a paradigm shift in medical research thanks to bioinformatics. These tools, by exploiting large gene expression databases, such as Gene Expression Omnibus (GEO), have provided an unprecedented opportunity to identify the molecular genetic signatures of diseases. Systematic data analysis can identify differentially altered genes and translate them into functional interaction networks using advanced tools, such as Search Tool for the Retrieval of Interacting Genes/Proteins (STRING).5 This can pave the way for uncovering the mechanisms driving the biological complexity of TNBC.2

This methodology contributes to a comprehensive understanding of the disease by discovering gene hubs (hub genes) that operate not individually but as a synergistic network, thus identifying them and their biological and therapeutic significance3,and by identifying gene signatures with greater predictive power than individual genes, in order to better describe the genetic pathways that influence the effectiveness and efficacy of targeted therapies.6

The importance of this study is that it aims to narrow the scope of analysis from a general list of altered genes to an accurate and effective prognostic gene signature in TNBC. To achieve this, a rigorous workflow was followed, starting with differential expression analysis and identifying immune pathways via Gene Ontology (GO)/Kyoto Encyclopedia of Genes and Genomes (KEGG)7, followed by constructing a robust STRING network to identify gene hubs (including CXCL9, CXCL11, and CD163).3

To confirm the clinical significance of these hubs, survival analysis was performed using Kaplan- Meier plots, the primary statistical tool for correlating detected gene expression with clinical prognosis (relapse-free survival [RFS]).8 This final step aimed to demonstrate that high expression of the combined gene signature represents the strongest prognostic indicator of favorable outcome outperforming the prognostic impact of any single gene.6 This study makes a fundamental scientific contribution by confirming that the ultimate predictive power lies in the functional synergy of the immune network, directing future research towards targeting specific axes (e.g., CXCL11/CXCR3) to bolster the antitumor response rather than a general focus on infiltrating immune cells.4 This could open new avenues for the development of targeted therapies to improve survival and reduce the risk of relapse in TNBC patients.

Gene expression data for TNBC were retrieved from the public GEO database. Specifically, dataset access number GSE53752 was downloaded. The dataset was generated on the GPL7264 Agilent-012097 Human 1A Microarray (V2) platform, which includes 51 primary TNBC tumor samples and 25 adjacent normal tissue samples. Corresponding clinical information, including age, AJCC stage, and histological grade, required for survival analysis was extracted from the GSE53752 sequence matrix file.3 All bioinformatics analyses and reporting followed the Minimum Information on Microarray Experience (MIAME) guidelines regarding data source and accessibility.

Differential expression analysis was performed in the R statistical environment (version 4.3.0) using the limma (Linear Models of Microarray Data) package (version 3.56.2). A comparison was made between 51 TNBC tumor samples and 25 normal breast tissue samples. Significantly upregulated and downregulated genes were identified. Genes were considered differentially expressed (DEGs) based on an absolute log2 fold change (LogFC > 1.0) and a corrected P value. The false discovery rate (FDR) was adjusted using the Benjamini-Hochberg method, with a threshold set at FDR < 0.05. A heatmap was generated to depict the expression patterns of the 20 most significantly different genes, demonstrating a clear clustering between tumor and normal tissue samples.

The list of 20 highly-expressed genes identified in the previous step was used. Enrichment analysis of biological pathways and functions was performed using GO and KEGG tools. The analysis showed strong enrichment of terms related to immune response, inflammation, and chemokine signaling.

To construct a protein-protein interaction (PPI) network using the STRING database (Version 12.0), we selected the top 20 DEGs (10 ascendingly expressed and 10 descendingly expressed) based on the highest absolute value of logarithmic folding variation (LogFC). The parameters were adjusted to include functional and physical interactions, and a minimum confidence score was set to ensure the reliability of the links.

Network strength was confirmed to be nonrandom via the PPI-Enrichment P value <0.05). Hub genes were identified based on the highest interaction degrees, including CXCL9, CXCL11, CD163, and FCGR3A, were as hub nodes in the immune network.

The online statistical tool KM-plotter was used. To ensure specificity, the clinical analysis was restricted to patients with TNBC by defining the 3 receptor statuses (i.e., ER, PR, and HER2) as negative. RFS was chosen as the prognostic measure. This filtering resulted in a final cohort of 530 patients (N=530) used to evaluate genetic prognostication.

To ensure the accuracy of the analysis and to reduce the potential variability resulting from the integrated platforms, the following standardized criteria were used: (1) Probe selection: The “automatic selection of the best probe set” option was used to select the probe with the highest average variability in expression for each gene. )2) Optimal separation threshold: The "automatic selection of best separation threshold" option was used to determine the optimal separation point between the high and low expression groups, achieving the highest statistically significant P value in the rank logarithm test. Individual survival was analyzed for each hub gene separately (CXCL11, CD163, COL10A1) to determine its relationship with poor prognosis. Gene Signature: Pooled survival analysis was performed for CXCL9, CXCL11, and CD163 as a single "signature" using the "Use multiple genes" option, and the best cutoff point was determined to stratify the patients.

The principal component analysis (PCA) plot shows the overall variability in gene expression across all samples. The distinct clustering of normal samples (black) and their clear separation from the majority of tumor samples (red) shows that the difference in gene expression between normal and tumor tissue are robust and statistically significant. This confirms that the dataset was of high quality and ensures that the samples are highly suitable for downstream differential expression analysis (Figure 1).

Differential expression analysis of the transcriptomic data demonstrated a clear distinction between tumor samples and normal tissue samples. A list of the top 20 genes significantly overexpressed in tumors compared to normal tissues was identified. The heatmap of these genes showed highly organized clustering of the samples, confirming the clear genetic variability between malignant and control groups, justifying the selection of these genes for further functional analysis (Figure 2).

The "immune-hot tumor" status was confirmed using ssGSEA. Box-and-whisker plots revealed that immune infiltrate enrichment scores for key cell types (including B cells, CD4+ T cells, and mast cells) were significantly higher in tumor samples compared to normal tissue (P value < 0.05). This shows that upregulated gene expression translates into a microenvironment with robust immune infiltration (Figure 3).

Functional enrichment (GO) analysis revealed that the overexpressed genes were predominantly concentrated in pathways related to immune response, inflammatory response, and chemokine receptor binding. This enrichment confirms that immune-mediated inflammatory pathways represent the most prominent biological alteration in the TNBC tumors studied (Figure 4).

A PPI network (STRING) was used to analyze the interconnection between the top 20 identified genes.

The analysis revealed robust connectivity within the network, suggesting that the genes function as a unified functional pathway. The following genes were identified as major hub genes based on the highest degree connectivity and interaction scores in the network hub genes: CXCL9, CXCL11, CD163 and FCGR3A. The network primarily consists of chemokine genes (CXCL9 and CXCL11) and immune cell markers (CD163) (Figure 5).

All survival analyses were restricted to patients with triple-negative (ER-/PR-/HER2-negative) breast cancer, and RFS was used as the primary outcome measure in accordance with standard epidemiological reporting. Hazard ratios (HRs) were reported uniformly as HR <1.0 were interpreted as indicating a reduce risk of relapse (favorable prognosis) whereas HR >1.0 would indicate an increased risk. This approach ensures a direct and accurate interpretation of the protective role played by specific immune signature in microenvironment of TNBC.

Kaplan-Meier analysis of the CXCL11 gene showed a strong association with favorable prognosis. High expression of the chemokine gene CXCL11 was associated with reduced risk of relapse (HR = 0.64, P  = 0.014). These results indicate that elevated CXCL11 levels are associated with a significantly higher RFS rate, suggesting a protective role for this chemokine in the tumor microenvironment of TNBC.

In contrast to CXCL11, other highly expressed genes failed to demonstrate individual statistical prognostic significance. CD163 did not demonstrate a statistically significant difference in RFS rate (P = 0.4). No statistical association with survival outcome was demonstrated for COL10A1 (P = 0.26).

The combined effect of the pivotal gene network (CXCL9, CXCL11, CD163) was evaluated as a single "signature." This analysis demonstrated the most robust predictive power observed in this study consistent with the standard epidemiological reporting criteria. High expression of the combined gene signature was significantly associated with improved RFS (HR = 0.45 , P = 2.3 × 10⁻⁶(. This led to a 55% reduction in the risk of relapse (calculated as 1−HR) for patients with high signature activation. This finding confirms that the CXCL11/CXCL9/CD163 index constitutes a highly potent and favorable prognostic tool, showing that the synergy between these immune hubs is a decisive factor in predicting TNBC outcome.

To determine whether the predictive value of the genetic fingerprint is independent of known clinical factors, a multivariable Cox proportional hazards model was performed. The model was adjusted for patient age, American Joint Committee on Cancer (AJCC) stage, and histological grade. The results showed that the CXCL11/CXCL9/CD163 index remained highly statistically significant and independently predictive of favorable survival outcomes (adjusted HR = 0.45, P = 2.3 × 10⁻⁶(. In this model, traditional clinical factors (age, stage, and grade) did not achieve statistical significance, confirming that the genetic signature provides superior and independent clinical utility for risk stratification in patients with TNBC.

Our analysis presents an interesting discrepancy that highlights the prognostic mechanisms in TNBC: while high expression of CXCL11 (an attractant signal) emerged as a robust predictor of favorable prognosis (HR = 0.64, P = 0.014), the expression of CD163 (a tumor-associated macrophage[TAM] marker) showed a similar protective trend but was not statistically associated with poor prognosis (HR = 0.86 , P = 0.4).20 This discrepancy raises questions about the quality of the immune infiltrate and suggests that the prognostic power lies in the intensity of the chemical signal (CXCL11), which effectively recruits protective immune cells, rather than the mere density of macrophages.13,20

This finding supports research indicating conflicting results regarding the prognostic role of CD163-positive macrophages in nonmetastatic breast cancer.13 While some studies show that CD163 may be associated with increased tumor size and grade in TNBC, its failure to predict survival in our analysis suggests that overall activation, reflected by high CXCL11 levels, ultimately determines the favorable course of the disease, not just macrophage density.21

Furthermore, the failure of tumor structural components, such as the COL10A1 gene (extracellular matrix) to demonstrate prognostic significance (HR = 1.23, P = 0.26) further reinforces the notion that the functional (immune) pathway, rather than the structural components of the tumor, is the primary driver of patient outcomes and key to prognosis. Overall, our data suggest that targeting or assessing the CXCL11/CXCR3 signaling axis provides superior prognostic value in TNBC, as its activation shows that a robust antitumor response may be more clinically relevant and have a higher prognostic value in TNBC than assessing macrophage density alone.22

Although high expression of COL10A1 (extracellular matrix) in TNBC tumors was identified as part of the list of highly expressed genes in our analysis, it failed to demonstrate any statistical prognostic significance in the survival analysis (HR = 1.23, P = 0.26). This contradicts previous bioinformatic and clinical studies that have reported high COL10A1 expression promotes breast cancer progression and predicts poor prognosis.14 Our results indicate that in the specific microenvironment of TNBC, the predictive impact of the structural extracellular matrix components is diminished by the dominant role of immune-related signals, particularly the CXCL11 axis. This discrepancy suggests that the functional pathway (immune-inflammatory) has a significantly higher prognostic power than the structural components (extracellular matrix) in determining patient outcome.

While COL10A1 may play a role in promoting TNBC progression via pathways, such as Wnt/β-catenin, its direct clinical impact on survival may be overshadowed by the dominant power of the immune signature.16 In our study, the integrated immune-related signature demonstrated the greatest prognostic power (HR = 0.45, P = 2.3 × 10−6), but the CXCL11 alone also showed a strong protective effect (P = 0.014). This finding reinforces the argument that the tumor-enhanced immune response is the dominant factor determining the course of TNBC, and that treatment and prognostic efforts should focus on immunomodulation rather than targeting structural components of the TME.23

The Kaplan-Meier analysis of the Humig gene signature is the most powerful clinical evidence in this study. The analysis demonstrated that the combined effect of the pivotal gene network (CXCL9, CXCL11, CD163) is not just a transient genetic feature, but rather a highly robust and independent favorable prognostic factor in TNBC. The combined gene signature, rather than individual genes, serves as an integrated functional fingerprint of the high-quality of the immune response in a TME that would otherwise be dominated by tumor-promoting inflammation.24

In terms of statistical validity, the log-rank test value for this signature reached P = 2.3 × 10−6. This small value conclusively confirms that the observed improvement in RFS in patients with high expression is not random, but rather a direct and robust consequence of the activity of this network. This is a statistically significant value. As for the clinically significant effect, the HR for the signature was 0.45. This means that high gene activity in this network more than reduces the risk of death or relapse by 55% compared to patients with low expression. Therefore, the prognostic superiority of this integrated signature clearly outweighed the prognostic significance of any single gene analyzed, including the potent gene CXCL11 (which had a HR of 0.64).

Thus, this scheme provides conclusive evidence that the detected inflammation/immune axis does not merely represent an immune infiltration, but rather a molecular signature with a clinically significant effect. This can be used as powerful tool for identifying patients who lack this protective immune response and who may be at higher risk of relapse and as a high-priority therapeutic strategy which can enhance the TME and improve long-term response to treatment.

CXCL11 has the ability to bind to chemokine receptors that play a crucial role in the immune response within the TME. CXCL11 has diverse functions which include inhibiting angiogenesis and regulating the recruitment of antitumor immune cells. Most importantly, CXCL11 contributes to the formation of a defensive immune environment by inhibiting the polarization of M2 phagocytic cells and promoting the infiltration of effector lymphocytes.24 Although CXCL11 may play a dual role in different types of malignant tumors, our results in TNBC definitively confirm its antitumor function. In this specific context, CXCL11 acts as a powerful chemotactic agent that attracts tumor-infiltrating lymphocytes, effectively contributing to curbing tumor development and improving the patient’s chances of survival. CXCL11 and CXCL9 signaling occurs primarily via the CXCR3 receptor. In TNBC, this axis stimulates the JAK/STAT and PI3K/Akt-S6 pathways, which are essential for the activation and recruitment of antitumor immune cells, such as T cells and natural killer cells.25 Instead of promoting tumor growth, these molecular pathways coordinate a strong immune response that inhibits tumor development. This directly explains the good prognosis and low risk of relapse observed in our results.

The PI3K/AKT/mTOR signaling axis, and especially the Akt-S6 node, is a critical regulator of CD8+ T cell fate.26,27 Recent evidence presented by Chen et al.27 confirms that mTOR signaling is essential for the formation and maintenance of CD8+ memory T cells, preventing them from entering a state of immune exhaustion where they lose their ability to fight tumors. In our study, it is likely that high expression of the CXCL11-dependent signature might have enhanced this mTOR-dependent metabolic and signaling efficiency, by promoting the survival and persistence of active, unexhausted T cells within the TME. This pathway provides a mechanistic basis for the strong protective effect (HR = 0.45) and the substantial reduction in RFS observed in our cohort of patients with TNBC.

Furthermore, the inclusion of the phagocytic cell marker CD163 in our immune fingerprint reflects the complexity of the immune landscape. Although CD163+ tumor-associated phagocytic cells are traditionally associated with immunosuppression.28 Our results are consistent with recent data showing that in an environment rich in chemokines (e.g., high levels of CXCL11/CXCR3), these cells can coexist with a dominant antitumor immune response or be reprogrammed by it.29 This is reinforced by immunotherapy models in TNBC, where the polarization of CD4+ and CD8+ T cells via CXC chemokines can effectively overcome the suppressive immune environment and target cancer stem cells, thus preventing late relapses.30 Our integrated fingerprint acted as a powerful protective biomarker, providing a clear automated basis for a significant survival advantage (HR = 0.45) and the decrease in the metastases observed.

While CD163 is a well-established marker for M2 TAMs, which are typically associated with poor outcomes in TNBC, its lack of statistical significance in univariable survival analysis (P = 0.4) warrants further investigation. This lack of independent predictive power may be attributed to several biological and methodological factors.

First, the predictive effect of TAMs may depend heavily on their location within the TME (e.g., around the tumor vs within it), a factor that comprehensive gene expression analysis cannot determine. CD163 expression may only exert a dominant harmful effect only when TAMs are in a position that allows them to interact directly with T cells or metastatic environments.30,31

Second, the independent predictive signal of CD163 may be very weak or masked by other potent immunomodulators. Its strongest predictive power is perhaps only demonstrated when CD163 is incorporated into a composite index (CXCL11 /CXCL9/CD163 index). This suggests that CD163 primarily functions as a contextual factor and a key component of the immunosuppression environment that should be evaluated alongside dominant chemokine signals (CXCL11/CXCL9) to fully understand the true risks associated with relapse. As Markovicio et al.31 explained, certain subsets of tumor-associated immune cells can contribute to antitumor activity through the expression of CXC chemokines. This may explain the lack of a negative univariate correlation in our study. Here, the predictive significance of CD163 is realized when it is assessed within the complete immunological signature, rather than as an independent marker.

This study has inherent limitations to the use of public datasets. First, the cohort of patients with TNBC (n = 533 patients) was pooled from multiple populations using the Kaplan-Meier plotting tool, resulting in variations in patient characteristics with respect to treatment and follow-up durations. Although multivariate logistic regression analysis was performed according to the major clinical variables (age, stage, and grade), demographic factors such as race were unavailable and could not be controlled for. Secondly, the use of comprehensive gene expression data is a major limitation. This methodology prevents the analysis of precise spatial localization and cell-to-cell communication within the tumor microenvironment (e.g., CD163+ tumor-associated phagocytic cell interactions). Further validation using single-cell RNA sequencing or spatial transcriptomics is needed to confirm mechanistic roles.