Immune‐based transcriptomic signature predicts CDK4/6 inhibitor efficacy in HR+/HER2– breast cancer

Dear Editor, Cyclin-dependent kinase 4/6 inhibitors (CDK4/6i) have revolutionised the management of hormone receptor-positive (HR+), HER2-negative (HER2‒) advanced breast cancer (ABC).1, 2 However, the molecular mechanisms driving tumour resistance and subsequent therapies remain a challenge.3 Here, we identified and validated a novel transcriptomic immune-based signature (‘Key Immune Activation’ [KIMA] signature) that predicts efficacy to CDK4/6i in HR+/HER2– ABC. KIMA signature provides a clinically actionable biomarker that has the potential to guide immune-based combination strategies and personalised treatment in ABC patients. We conducted a prospective observational study of 100 HR+/HER2– ABC patients who initiated CDK4/6i therapy in combination with endocrine therapy between March 2018 and April 2022 at the Institut Català d'Oncologia, Badalona. Among them, 91 patients had sufficient clinical follow-up for efficacy evaluation. Pre-treatment tumour samples were available for 55 patients (Figure 1A). Clinical characteristics and intrinsic subtypes were characterised (Table 1). Overall, the clinical characterisation of the cohort revealed a median progression-free survival (PFS) of 14 months (interquartile range [IQR]: 5.6–26.6) and a median overall survival (OS) of 42.8 months (IQR: 20.3–47.3; Figure S1A). As expected, the line of treatment had the most significant impact on PFS, with first-line patients showing better outcomes (first line: 30.4 months vs. second line or later lines: 9.3 months, p = .001). Other clinical factors, such as endocrine resistance and hepatic metastasis, showed consistent trends towards shorter PFS, aligning with results from prior larger studies4 (Figure S1B). Intrinsic subtype5 analysis by prediction analysis of microarray 50 (PAM50) revealed a predominance of luminal tumours, with an increased proportion of HER2-enriched and basal subtypes in metastatic lesions compared to primary tumours (p = .034; Figure S2A). Among first-line-treated patients, luminal-subtype tumours tended towards longer PFS (Figure S2B). Due to clinical heterogeneity, patients were stratified into two efficacy groups (good or bad) based on predefined clinical criteria derived from PFS benchmarks reported in pivotal CDK4/6i trials. Specifically, patients with hormone-sensitive disease were categorised as good efficacy if PFS was ≥24 months, while those with hormone-resistant disease or prior therapies were categorised using lower PFS thresholds (≥12 or ≥7 months) according to clinical context (see section Material and Methods in the Supporting Information and Figure S3). Using this approach, 57% of patients were classified in the good efficacy group (median PFS 31 months) and 43% in the bad efficacy group (median PFS 5.7 months; p < .001; Figure 1B). Baseline clinical characteristics were comparable between groups, supporting the need for molecular biomarkers of efficacy (Table 1). To identify potential tumour prognostic and predictive factors influencing the efficacy of CDK4/6i, we performed a transcriptomic analysis using the NanoString Breast Cancer 360™ (BC360™) panel in patients with matched tumour tissue and clinical data (Figure 1A and Table S1). We compared BC360™ predefined gene signatures between good and bad efficacy groups and identified 19 differentially expressed signatures, mostly enriched in the bad efficacy group (18/19, Figure 1C). Among the overexpressed signatures, we identified substantial enrichment in immune-related pathways, including IDO1 (Indoleamine 2,3-dioxygenase 1), TIGIT (T cell immunoreceptor with Ig and ITIM domains), Treg (Regulatory T cell), TIS (Tumor Inflammation Signature), CD8 (Cluster of differentiation 8 lymphocyte marker), PD-1 (Programmed cell death protein 1), inflammatory chemokines, IFN-γ (Interferon gamma), PD-L1 (Programmed death-ligand 1), PD-L2 (Programmed death-ligand 2) and macrophages. In contrast, the only downregulated signature was related to luminal differentiation. To further assess the clinical impact of these signatures while minimising potential biases from prior treatments, survival analysis was performed in patients receiving CDK4/6i as first-line therapy (n = 31). Univariate Cox survival analysis showed that high expression of IFN-γ, PD-L2, TIGIT, IDO1 and macrophage signatures was associated with shorter PFS (Table S2). Multivariate analysis confirmed the increased expression of IFN-γ signature as an independent predictor of poor outcome (hazard ratio = 21.5, p = .008; Figure 1D and Table S3). Furthermore, using the optimal threshold calculated through receiver operating characteristic (ROC) curve analysis, patients with high IFN-γ signature expression had a significantly worse median PFS (15 months vs. not reached, p = .008), confirming the negative impact of high BC360™ IFN-γ signature expression (Figure 1E). To recognise the immune-associated predictors of efficacy, we analysed differences in single-gene expression, which led to the identification of 43 differentially expressed genes (DEGs). This analysis revealed a significant enrichment of immune-related genes in bad efficacy group (Figures 2A and S4A‒C). We then selected all immune-related DEGs that showed at least a 50% difference between efficacy groups (n = 14). Unsupervised consensus clustering effectively grouped 90% of patients with good treatment efficacy (p = .004; Figure 2B), identifying two gene sets that were found either upregulated (n = 9) or downregulated (n = 5) in patients classified as bad versus good efficacy groups. Principal component analysis further demonstrated that all nine genes contributed to the predictive capacity of the signature (Figure 2C,D). Survival analysis revealed that only the expression levels of the upregulated nine-gene set correlated with differential PFS and OS, leading to the definition of the KIMA signature, which included CXCL10, OAS3, STAT1, CD27, TIGIT, IL2RA, FOXP3, TAP1 and TAP2 genes. High KIMA expression was associated with significantly shorter PFS (15 months vs. not reached, p = .016) and OS (29.9 months vs. not reached, p = .027; Figure 2E,F). In contrast, the downregulated five-gene set (HLA-DQA1, HLA-DQB1, CXCL8, TNFSF10 and IL1B) did not correlate with PFS or OS in either the first-line setting or across all treatment lines (Figure S5A‒C). The predictive value of the KIMA signature was further evaluated in all the patients, irrespective of the line of treatment with similar results, that is, patients from the poor efficacy group presented significantly higher expression levels of the KIMA signature (p = .0022) and shorter PFS and OS (Figure 3A‒C). Multivariate analysis, including relevant clinical factors such as line of treatment, liver metastases and visceral disease (Figure S1B), showed that only the number of line treatment and the KIMA (or IFN-γ) signature retained a significant impact on PFS, further supporting their independent predictive value (Figures 3D and S5D). ROC curve analysis confirmed the robust predictive performance of the KIMA signature (area under the curve [AUC] = .787), showing superior accuracy compared to the BC360™ IFN-γ signature (AUC = .728; Figure 3E). To validate the predictive capacity of the KIMA signature in distinct clinical settings and beyond BC360™ technique, we assessed its performance in the neoadjuvant NeoPalAna study (NCT01723774).6 Consistent with our findings in ABC, not responder patients to neoadjuvant palbociclib displayed significantly elevated KIMA levels (p = .0055; Figure 3F), and the signature demonstrated strong predictive accuracy (AUC = .82 in ROC analysis; Figure 3G). While using NeoPalAna as a validation cohort can be considered a limitation due to the differences in diseases setting, it also suggests that immune-mediated resistance mechanisms may be conserved across disease stages, supporting the robustness and broad applicability of the KIMA signature irrespective of disease scenario. Mechanistically, persistent STAT1-driven IFN-γ signalling characterised high-KIMA tumours, promoting immunosuppressive microenvironments enriched in regulatory T cells (FOXP3 and IL2RA), exhausted T cells (TIGIT and CD27) and immune evasion chemokines (CXCL10)7 (Figure S6). These findings underscore chronic immune activation as a key factor limiting the efficacy of CDK4/6i8 and support KIMA as a clinically actionable biomarker. Notably, increased immune infiltration—typically considered favourable in other subtypes—was associated with poor outcomes in ER+/HER2‒ breast cancer, suggesting distinct immunological dynamics in this context.9, 10 Our data support the hypothesis that breast cancer is immunogenic and might be targetable by immune-modulating therapies, albeit alternative approaches to current immunotherapies directed towards the inhibition of specific immunologic signals should be considered. In this sense, the KIMA signature has multiple potential applications in clinical practice, both to identify CDK4/6i-resistant patients harbouring elevated KIMA signature expression but also to enable the selection of combination therapies tailored to individual immune landscapes. Indeed, resistant patients could benefit from immune-targeted interventions prior to or in combination with CDK4/6i as targeting the IFN pathway (e.g., JAK/STAT inhibitors) or immune checkpoints could help overcome this resistance. In conclusion, the KIMA signature offers critical translational insights into immune-mediated resistance mechanisms in HR+/HER2– ABC. By capturing the interplay between tumour-intrinsic and immune factors, KIMA may help identify patients at risk of poor outcomes and guide personalised treatment strategies. Nevertheless, given that the KIMA signature was identified in a relatively limited cohort, prospective clinical trials are warranted to validate its predictive value and assess its generalisability. Future studies will determine how to integrate immune transcriptomic profiling into clinical practice to reduce resistance and ultimately improve therapeutic outcomes in this prevalent breast cancer subtype. Eudald Felip and Edurne Garcia-Vidal substantially contributed to the study conception and design, methodology, patient data extraction, biological analysis, formal analysis and manuscript writing. Sara Cabrero-de las Heras, Milana Bergamino and Adrià Bernat-Peguera contributed to the biological data analysis and interpretation. Beatriz Cirauqui, Vanesa Quiroga, Iris Teruel, Angelica Ferrando-Díez, Anna Pous, Assumpció López-Paradís, Laia Boronat, Gaston Zatta and Marga Romeo contributed to the collection of clinical data, sample management and experimental procedures. Ricard Mesía, Pedro Luis Fernandez and Bonaventura Clotet contributed to study supervision, resources and critical review of the manuscript. Eva Riveira-Muñoz and Anna Martínez-Cardús contributed to data analysis, interpretation and manuscript revision. Ester Ballana and Mireia Margelí contributed substantially to conceptualisation, methodology planning, formal analysis, manuscript writing and critical revision, funding acquisition and overall study supervision. All authors read, reviewed and approved the final version of the manuscript. We sincerely thank all patients for their invaluable and selfless participation in this study. We also thank the Tumour Biobank at IGTP/Hospital Germans Trias i Pujol for their excellent work managing samples. Additionally, we would like to thank the Oncology Day Hospital team, including nursing and administrative staff, for their unwavering support and dedication throughout this project. We also acknowledge NanoString Technologies for the collaboration agreement established in 2020 and their support with the BC360 platform. This work was funded by the Instituto de Salud Carlos III (ISCIII) through project PI21/00642 (co-funded by European Regional Development Fund/European Social Fund “Investing in your future”), by the project InMaM funded under the Plan Complementario de Biotecnología Aplicada a la Salut coordinated by IBEC within the framework of the Recovery, Transformation and Resilience Plan (C17, Resilience Plan C17.I1) and supported by the European Union—“NextGenerationEU”, by Pfizer (Pfizer Independent Research Grant), and by the Generalitat de Catalunya via AGAUR (2021 SGR 00452). E.F., M.B. and A.B.P. are fellows of ISCIII (JR23/00044, CM22/00101 and CD21/00054, respectively). This research was approved by the Ethics Committee of Hospital Germans Trias i Pujol, and written informed consent was obtained from all patients prior to their enrollment in the research project. The datasets supporting this article's conclusions are included within the article (and its additional files) and available from the corresponding author upon reasonable request. Transcriptomic data and associated clinical information have been deposited in GEO (GSE265870). Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.