Abstract Background Anthracycline-containing neoadjuvant chemotherapy (NACT) is the standard treatment for early triple-negative breast cancer (eTNBC); however, it is associated with substantial toxicity. We performed whole transcriptome profiling of baseline tumor biopsies to identify gene networks predictive and prognostic for pathological complete response (pCR) and survival after de-escalated, anthracycline-free NACT in the WSG-ADAPT-TN trial ([66]NCT01815242). Methods eTNBC patients (cT1c-cT4c, cN +) were randomized to 12 weeks of nab-paclitaxel + gemcitabine (n = 182) or nab-paclitaxel + carboplatin (n = 154). The primary endpoint was pCR (ypT0/is, ypN0), and the secondary endpoints included survival and translational research. AmpliSeq RNA sequencing, allowing simultaneous analysis of the expression of > 20,000 genes, was performed in 135 patients. Differentially expressed genes were evaluated in training (n = 67) and validation (n = 68) sets, and a polygenic score (PS) for prediction of pCR (PS:pCR) and a PS for prediction of invasive disease-free survival (PS:iDFS) were found. Results 49/135 (36.3%) patients had pCR; 30 iDFS events occurred during 60-month median follow-up. Immune recruitment and viral defense gene networks were strongly associated with pCR, while metabolic pathways were associated with survival. PS:pCR and PS:iDFS predominantly included immune-related genes. Diagnostic accuracy (ROC AUC) in the validation cohort was 83% for PS:pCR and 64% for PS:iDFS. At optimized cut-off, PS:pCR identified a group with a 67.7% pCR rate (vs. 10.8%; p < .0001), and PS:iDFS detected a group with 79.5% (95%CI 64.1%, 88.8%) 5-year iDFS rate (vs. 55.0%, 95%CI 29.8%, 74.5%; p = .04). Conclusion Polygenic scores incorporating immunoregulatory genes can predict pCR and survival and represent an opportunity to select patients for de-escalated, anthracycline-free NACT. This transcriptome network analysis also identifies potential new targets for personalized medicine approaches in patients without response to NACT. Trial registration [67]NCT01815242. Supplementary Information The online version contains supplementary material available at 10.1186/s12943-025-02275-0. Keywords: Neoadjuvant therapy, Triple negative breast neoplasms, Translational research, Biomedical, Gene regulatory networks, Gene expression profiling, Prognosis Background While early triple-negative breast cancer (eTNBC) is generally regarded as a subtype with poor prognosis, patients with pathological complete response (pCR) after neoadjuvant chemotherapy have excellent survival outcomes [[68]1]. Recently, the practice-changing KEYNOTE-522 trial has established the combination of the immune checkpoint inhibitor (ICI) pembrolizumab with an anthracycline-taxane and platinum-containing sequential chemotherapy as standard of care in patients with high-risk eTNBC [[69]2]. Nevertheless, the molecular characteristics underpinning therapeutic sensitivity in eTNBC remain unclear, necessitating the use of toxic systemic chemotherapy in each patient. Interestingly, several trials (e.g., WSG-ADAPT-TN ([70]NCT01815242), NeoSTOP ([71]NCT02413320), NeoCART ([72]NCT03154749), TBCRC 030 ([73]NCT01982448), WSG-PlanB ([74]NCT01049425)) demonstrated promising tumor response and long-term outcomes after the omission of anthracycline from standard chemotherapy regimens. Such therapy de-escalation is associated with fewer acute and late toxicities related to anthracycline use and a better quality of life. However, such treatment optimization requires precise patient selection to ensure survival at least comparable to that after standard anthracycline-based therapy. Over the recent years, much effort has been made to evaluate biomarkers for tumor response and identify patients at a higher risk of recurrence to allow individualized de-escalated treatment approaches. Thus far, only a few biomarkers, including stromal tumor-infiltrating lymphocytes (sTILs) and expression of programmed cell death protein (PD-1) receptor and its ligand (PD-L1) as well as CD8 expression, showed predictive potential in eTNBC [[75]3–[76]6]. Therefore, an unmet need remains for reliable predictive biomarkers to guide treatment de-escalation. In this study, we sought to identify gene expression patterns underpinning chemotherapy response and associated with survival by evaluating baseline biopsy samples collected within the WSG-ADAPT-TN trial, which aimed to investigate de-escalated, anthracycline-free therapeutic concepts [[77]7]. Results Patients and samples The baseline characteristics of 135 patients included in the analysis (nab-paclitaxel plus gemcitabine, arm A: n = 70; nab-paclitaxel plus carboplatin, arm B: n = 65) are shown in Table S1. The overall pCR rate in the analyzed cohort was 36.3%: 24.3% of patients in arm A (n = 17/70) and 49.2% in arm B (n = 32/65) had a pCR. In total, 30 invasive disease-free survival (iDFS) events occurred during 60 months of median follow-up (arm A: n = 17; arm B: n = 13); 24 iDFS events occurred in non-pCR and 6 in pCR cases. Network analysis for association with pCR and iDFS A gene network analysis on 121 genes whose differential expression was associated with pCR returned significant enrichment for immune system processes (adjusted p = 2.74 × 10^–10), regulation of viral processes (adjusted p = 3.09 × 10^–4), and other immune-associated network effects (Fig. [78]1A). Gene network analysis, performed on 728 genes whose differential expression was associated with iDFS, identified cytoplasmic metabolic pathways as the most enriched (Fig. [79]1B, adjusted p = 2.01 × 10^–11). Interestingly, no pathway networks related to the immune system were associated with iDFS. Analysis of differentially expressed genes associated with pCR and iDFS is shown in Fig. S3. Fig. 1. [80]Fig. 1 [81]Open in a new tab Network analysis for association with pCR and iDFS. A Pathway enrichment analysis for 121 genes whose differential expression was associated with pCR outcomes identified significant enrichment for immune system processes and viral genome regulation. B Pathway enrichment analysis for 728 genes whose differential expression was associated with iDFS outcomes identified enrichment for cytoplasmic and protein-binding cellular processes Development and validation of polygenic score for pCR Polygenic score (PS) for the prediction of pCR (PS:pCR) developed in the discovery cohort included five genes: APOBEC3C, APOBEC3D, LOC285972, IL15, and UBN2. PS:pCR score did not differ between the patients in the validation set (n = 68) stratified according to clinical characteristics (menopausal status, cT, and cN), and it was higher in arm B, in patients with pCR and those with higher sTIL levels at baseline and week 3 (Table S3). PS:pCR score cut-off at the 55th percentile yielded the highest sensitivity (84%) and specificity (79%) with an area under the curve (AUC) of 83% (Fig. [82]2A). Fig. 2. [83]Fig. 2 [84]Open in a new tab Development and validation of polygenic scores for pCR and for iDFS. A A 5-gene signature predicting pCR was performed on a discovery dataset to generate a linear equation, which was then validated in a separate validation cohort. For de-escalated 12-week chemotherapy, a minimal set of 5 genes was identified that achieved an AUC ≥ 80% in the validation cohort. B pCR rates according to PS:pCR score below and above cut-off in the validation set. C Comparison of iDFS between patients with PS:iDFS score above or below cut-off in the validation set In the validation cohort, the pCR rate among the patients with a PS:pCR score below the cut-off was 10.8% (n = 4/37) compared to 67.7% (n = 21/31) in those with a PS:pCR score above the cut-off (p < 0.0001, Fig. [85]2B). In multivariate analysis adjusted for clinical factors, the PS:pCR was positively associated with pCR (OR 4.33; 95%CI 1.94, 9.67; nominal p < 0.001, Table S4). The addition of PS:pCR to clinical factors increased an area under the receiver operating characteristic curve (ROC AUC) for pCR from 0.583 to 0.827. Development and validation of a polygenic score for iDFS Four genes were included in the PS for the prediction of invasive disease-free survival (PS:iDFS): LOC728743, SGPP2, SPP1, and TIGIT. PS:iDFS score was not associated with the clinical factors in the validation set (regarding menopausal status, cT, cN, or pCR); however, the mean score was higher in arm B and in patients with lower sTIL levels at baseline or week 3 (Table S5). PS:iDFS score cut-off at the 72nd percentile yielded the highest sensitivity (47%) and specificity (80%) combination, with an AUC of 64%. In the validation set, survival analysis according to the PS:iDFS score demonstrated 5-year iDFS rates of 55.0% (95%CI 29.8%, 74.5%) and 79.5% (95%CI 64.1%, 88.8%) in the group above and below cut-off, respectively (log-rank p = 0.04, Fig. [86]2C). PS:iDFS was not associated with iDFS in univariate analysis (HR 1.46; 95%CI 0.94, 2.28; p = 0.096). However, multivariate analysis adjusted for clinical factors detected an association between the iDFS and PS:iDFS (HR 1.68; 95%CI 1.00, 2.83; nominal p = 0.049, Table S6). Inclusion of PS:iDFS improved the identification of patients at a higher risk of iDFS event (increase in Harrell’s C coefficient from 64 to 71%). Discussion De-escalated neoadjuvant treatment approaches are intensively investigated in the eTNBC setting to limit acute and late toxicities associated with the use of anthracyclines. The WSG-ADAPT-TN and other trials demonstrated that a short, anthracycline-free neoadjuvant chemotherapy can result in clinically meaningful pCR rates and excellent survival in patients with pCR [[87]7], indicating that a significant proportion of patients could benefit from individualized, de-escalated therapy. While these results are promising, it is critically important to carefully select patients for whom anthracyclines could be safely omitted without compromising long-term survival. To this end, we examined transcriptome profiles of patients from the WSG-ADAPT-TN trial to identify tumor characteristics that predict pCR and survival for anthracycline-free chemotherapy. Our transcriptional analysis identified several genes and signaling networks associated with pCR and survival. Furthermore, a linear regression pipeline identified two minimal predictive sets of genes predictive for pCR and iDFS. Overall, the function of identified genes highlights the importance of various immune mechanisms (see Table S7) for response to chemotherapy and better long-term outcomes in eTNBC. This indicates that de-escalated, anthracycline-free treatment could be used to capitalize on pre-activated, functional immune processes in pre-selected patients. Importantly, our polygenic scores could identify a group of patients with a 67.7% pCR rate and a group with a 79.5% 5-year iDFS rate. Therefore, our approach may potentially be used to precisely select TNBC patients most suitable for de-escalated, anthracycline-free neoadjuvant treatment, thereby allowing individualized therapy in a clinical population historically facing limited therapeutic options. Furthermore, our results in patient groups determined by polygenic scores are comparable with the 64.8% pCR rate and 81.3% 5-year event-free survival (EFS) rate in an unselected cohort receiving pembrolizumab with chemotherapy from KEYNOTE-522 (although that trial was performed in stage II-III cancer, whereas our trial also included stage I cancer patients [[88]2]). It would, therefore, be interesting to test whether our polygenic scores could also identify patients with a higher chance of pCR and improved survival after the now standard-of-care KEYNOTE-522 regimen. The absence of such analysis is a limitation of our study, impacting the generalizability of our findings. The importance of genes related to immune processes in both tumor response and survival was also captured by the HER2DX score in HER2 + early breast cancer. HER2DX assay covers the expression of 27 genes grouped into four gene signatures related to immune response, proliferation, luminal differentiation, and HER2 amplicon. Similarly to our analysis, developing prognostic and predictive classifiers based on HER2DX resulted in two distinct scores for pCR and survival with a central role of immune signatures in both scores [[89]8]. Of note, a relatively small number of genes also makes it feasible to use small gene panel assays in the research setting. Nevertheless, our analysis is only hypothesis-generating, and its results should be confirmed in further clinical trials designed to investigate biomarkers for individualized treatment in eTNBC. Our transcriptome data synergizes well with findings from other analyses in eTNBC. For instance, Pérez-Pena et al. showed that genes regulating T-cell function and modulation of immune response are favorably associated with recurrence-free survival after neoadjuvant chemotherapy [[90]9]. Furthermore, our transcriptome profiling corroborates previous results from the WSG-ADAPT-TN trial, demonstrating the importance of dynamic changes and spatial distribution of inflammatory cells in prediction of tumor response [[91]3–[92]5]. Interestingly, the results implicating the role of immune-related mechanisms in both tumor response and better long-term outcomes in eTNBC are in contrast with our findings from the WSG-ADAPT-HR-/HER2 + trial investigating pertuzumab plus trastuzumab without chemotherapy. In that trial, immune-related mechanisms were associated with survival but not with pCR, which in turn correlated with HER2 expression. Although this differential effect could be due to chemotherapy-free targeted anti-HER2 treatment, our results support the paradigm that activation of the immune system is critical for preventing cancer recurrence. There remains a question regarding what alternative therapy could be used in patients with tumors predicted not to achieve pCR or in those with poor long-term prognosis after anthracycline-free treatment. Longer and more aggressive chemotherapy regimens, including anthracyclines to induce DNA damage, certainly remain an option. Furthermore, pembrolizumab, as a newly established standard of care, continues to be a feasible choice. However, while KEYNOTE-522 revealed increased pCR rates and improved survival after a combination of pembrolizumab with chemotherapy, ICIs in other trials demonstrated only increased tumor response (atezolizumab in IMpassion031), or improved survival (durvalumab in GeparNuevo), or no effect (atezolizumab in NeoTRIPaPDL1) [[93]2, [94]10–[95]12]. This suggests considerable differences in activity between various ICIs and suggests that their efficacy may be limited to specific groups of patients who optimally should be selected upfront. Since clinical parameters in the aforementioned studies do not appear to identify responders, researchers turned their attention toward the investigation of biomarkers associated with improved pCR rates and survival. For example, the expression of tumor-specific major histocompatibility complex class II was shown to associate with pCR after durvalumab plus chemotherapy in the I-SPY2 trial [[96]13]. In another trial investigating durvalumab combined with chemotherapy, an immunomodulatory 27-gene signature was shown to predict pCR [[97]14]. Finally, PD-L1 + IDO + antigen-presenting cells and CD56 + neuroendocrine epithelial cell phenotypes were associated with pCR to atezolizumab plus chemotherapy combination in the NeoTRIPaPDL1 trial [[98]15]. Moreover, that study found that although expression of genes related to immune response and cell proliferation was associated with pCR, these gene signatures were not predictive for atezolizumab activity. Similarly, tumor mutational burden and T-cell inflamed 18-gene expression profile were associated with pCR and EFS in the KEYNOTE-522 trial; however, they were not predictive of pembrolizumab benefit [[99]16]. Collectively, these immune response biomarkers potentially correspond to immune mechanisms identified in our study. However, it remains to be tested whether our polygenic scores could optimize the selection of patients with activated immune response mechanisms for neoadjuvant treatment with ICIs. The development of the polygenic score for iDFS emphasizes both the potential and the limitations of transcriptomic predictors for time-to-event outcomes. We hypothesize that the prediction of time-to-event outcomes, such as iDFS, is more biologically complex compared to the polygenic score for pCR because of the multifactorial nature of long-term recurrence and survival mechanisms. These outcomes are likely influenced by an interplay of tumor biology, host immune response, and treatment effects, among other factors. Consequently, the weaker performance of polygenic time-to-event prediction reflects the inherent difficulty in accurately modeling such outcomes with the current dataset, a challenge further compounded by the smaller cohort size. Despite these limitations, developing a polygenic score for iDFS provides valuable, hypothesis-generating insights into potential biomarkers and pathways that could guide future research. Specifically, our study underscores the need for larger, well-powered clinical trials to validate these findings and further explore the mechanisms driving long-term outcomes in triple-negative breast cancer. Conclusions In conclusion, our study revealed that immune system activation and recruitment prior to therapy in eTNBC appears to be a major transcriptomic characteristic determining pCR and improved survival after anthracycline-free neoadjuvant treatment. Therefore, our results pave the way for future clinical trials investigating de-escalated chemotherapy regimens with or without immunotherapy in carefully selected patients, which could ultimately result in personalized therapeutic approaches guided by molecular profiling. Supplementary Information [100]12943_2025_2275_MOESM1_ESM.docx^ (2MB, docx) Supplementary Material 1: Supplementary Methods. Supplementary Results. Fig. S1. Design of the WSG-ADAPT-TN trial. Fig. S2. Mapping of AmpliSeq reads and normalization of gene expression data. (A) An Integrative Genomics Viewer screenshot of 4 representative samples for a single gene of interest. The AmpliSeq RNA assay was originally designed to be run on the Proton sequencer, but for this experiment, it was adapted for the Illumina Nextseq to achieve better throughput and cost-effectiveness. A low number of cDNA/gDNA fragments can occasionally be present in some samples, as indicated by the red bracket, but as the analysis was based on amplicon read counting, these secondary fragments (bottom panel) are not included when tabulating read counts for gene expression analysis. (B) Comparing log2 transformation to Deseq2 rlog transformation for gene expression analysis. Fig. S3. Identification of differentially expressed genes associated with pCR and iDFS. Gene names in bold were used in downstream analysis to predict pCR (A) and iDFS (B) outcomes using polygenic predictor equations. Upregulated genes positively associated with pCR and immune infiltration have been boxed (A). Significance is shown above each gene (*p < 0.05, **p < 0.005, ***p < 0.0005). RPM, reads per million mapped reads. Fig. S4. Bulk tissue expression plot for upregulated genes associated with pCR. The genes identified as upregulated in patients with pCR exhibit minimal expression in subcutaneous adipose, visceral adipose, and breast mammary tissue but substantially higher expression in EBV-transformed lymphocytes, spleen, and whole blood. Table S1. Baseline characteristics of patients included in the analysis. Table S2. Comparison of patient characteristics between patients included in the analysis and the remaining patients from the WSG-ADAPT-TN trial. Table S3. Univariate associations between the PS:pCR and clinical factors and sTILs. Table S4. Multivariable analysis for pCR in the validation cohort. Table S5. Univariate associations between the PS:iDFS and clinical factors and sTILs. Table S6. Multivariable analysis for iDFS in the validation cohort. Table S7. The function of genes included in PS:pCR and PS:iDFS. Acknowledgements