Integrative analysis of scRNA-seq, spatial transcriptomics, and machine learning constructs a prognostic model for Pyrotinib resistance in breast cancer

Integrative analysis of scRNA-seq, spatial transcriptomics, and machine learning constructs a prognostic model for Pyrotinib resistance in breast cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrative analysis of scRNA-seq, spatial transcriptomics, and machine learning constructs a prognostic model for Pyrotinib resistance in breast cancer Qiheng Gou, Xueming Xia, Yuxin Xie This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9376197/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Pyrotinib resistance remains a major challenge in the treatment of breast cancer (BRCA), highlighting the need for reliable biomarkers and prognostic models. This study aimed to identify Pyrotinib resistance-related biomarkers and explore their regulatory mechanisms and therapeutic potential. We integrated single-cell RNA sequencing (scRNA-seq) of Pyrotinib-resistant SKBR3 cells, spatial transcriptomics (GSE243275), and bulk RNA-seq (TCGA-BRCA, GSE20685, GSE86374). Differential expression analysis, weighted gene co-expression network analysis (WGCNA), and 10 machine learning algorithms were used to screen candidate genes and construct a prognostic model. Functional enrichment, regulatory network analysis, molecular docking, cell communication, pseudotime trajectory analyses and in vitro experiment were performed for validation. A total of 740 Pyrotinib resistance-related genes and 39 candidate genes were identified. The StepCox [forward] + Ridge model, consisting of HMGB3, TFPI, ACTG2, and JCHAIN, exhibited robust prognostic performance (C-index: 0.61–0.65; AUC ≥ 0.6 across datasets), with high-risk patients showing poorer survival. These genes were validated at the mRNA and protein levels, participated in immune-related pathways, and had distinct chromosomal/subcellular localizations. ABT-737 was identified as a potential targeted drug via molecular docking. Spatial transcriptomics revealed fibroblast-centered cell communication and dynamic biomarker expression during malignant progression. Lastly, JCHAIN, HMGB3, ACTG2, and TFPI were significantly upregulated at both mRNA and protein levels in Pyrotinib-resistant BRCA cells compared with parental control cells. The four-gene model serves as a reliable prognostic tool for Pyrotinib response and BRCA outcomes, providing novel insights into resistance mechanisms and precision therapy strategies. breast cancer pyrotinib resistance spatial transcriptome prognostic model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Novelty and Impact This study uniquely integrates single-cell and spatial transcriptomics with machine learning to construct a four‑gene prognostic model for Pyrotinib resistance in breast cancer. It reveals fibroblast‑centric cell communication, dynamic biomarker expression, and identifies ABT‑737 as a potential targeted therapy. The model offers a robust clinical tool to predict drug response and overcome resistance, advancing precision oncology strategies. 1. Introduction Breast cancer (BRCA) represents the most prevalent malignancy among women worldwide 1 – 3 . Globally, BRCA is responsible for approximately one-third of all malignancies diagnosed in women, with its mortality rate accounting for roughly 15% of the total diagnosed cases 4 , 5 . It is noteworthy that the absolute incidence of BRCA has been on the rise in numerous developing countries, a trend that can be attributed to the dual effects of population expansion and the growing adoption of Western lifestyles 6 . The global epidemiological distribution of BRCA is shaped by a complex interplay of genetic predispositions, environmental exposures, and modifiable lifestyle factors 5 . In recent years, advances in the precise diagnosis of molecular subtypes have underpinned the optimization of BRCA therapeutics, and the integration of novel agents and neoadjuvant treatment regimens has further enhanced the precision of locoregional therapeutic interventions 5 . Nevertheless, disease recurrence driven by acquired or intrinsic therapeutic resistance remains a prevalent clinical issue that imposes substantial challenges on current management strategies 7 . Elucidating the intricate molecular mechanisms that underlie BRCA-associated therapeutic resistance is, therefore, of paramount importance for the development of novel, efficacious, and personalized therapeutic modalities that can either treat or circumvent such resistance in clinical practice. 8 , 9 . Pyrotinib, a novel irreversible pan-ErbB tyrosine kinase inhibitor (TKI), exerts broad and potent antitumor effects in HER2-positive BRCA 10 , 11 . Previous studies demonstrated that Pyrotinib exerts robust antitumor efficacy, as evidenced by its capacity to induce more pronounced cell cycle arrest, suppress the proliferation of SKBR3 BRCA cells, and inhibit in vivo tumor growth in xenograft mouse models, a phenomenon that may be attributed to enhanced autophagy triggered by endoplasmic reticulum stress 12 . Nevertheless, the long-term administration of Pyrotinib confers a risk of acquired resistance, which has emerged as a pivotal factor contributing to therapeutic failure 13 . The lack of reliable biomarkers to predict resistance onset and limited understanding of the underlying molecular and spatial regulatory networks further hinder the optimization of clinical treatment strategies for BRCA patients receiving Pyrotinib. In recent years, the integration of single-cell and spatial transcriptomics has been applied in multiple cancer fields, advancing our understanding of cellular heterogeneity while contextualizing single-cell data in multicellular microenvironments, thus yielding valuable insights into tumor biology 14 . The present study innovatively integrates multi-omics sequencing data, combined with comprehensive bioinformatics analysis and machine learning approaches, as well as a series of functional and regulatory validation methods. The purpose is to systematically screen Pyrotinib resistance-related genes in BRCA, clarify the molecular mechanisms and spatial regulatory characteristics underlying Pyrotinib resistance, construct a reliable prognostic model for predicting Pyrotinib treatment response and BRCA patient outcomes, identify potential targeted therapeutic agents, and provide novel technical support and theoretical basis for overcoming Pyrotinib resistance and optimizing personalized treatment strategies for BRCA. 2. Methods 2.1 Obtainment of BRCA datasets The clinical and survival information of the TCGA-BRCA dataset cohort were acquired from the University of The Cancer Genome Atlas database ( https://www.cancer.gov/ccg/research/genome-sequencing/tcga ) , which comprised 1,118 BRCA samples along with 137 control samples. The single-cell dataset GSE243275, whose corresponding platform is GPL24676, was harnessed for spatial transcriptome analysis, while GSE20685 (platform: GPL570) encompassing 327 BRCA samples and GSE86375 (platform: GPL6244) that consists of 124 BRCA samples as well as 35 adjacent non-cancerous samples were employed to validate the prognostic model and verify the expression levels of relevant genes, all of which were retrieved from the Gene Expression Omnibus (GEO) database ( https://www.ncbi.nlm.nih.gov/geo/ ). For the establishment of a Pyrotinib-resistant progressive SKBR3 cell system and subsequent single-cell RNA sequencing (scRNA-seq; one sample per group), the SKBR3 cell line was divided into five groups: a blank control (P0) and four groups treated with 12 nM, 24 nM, 36 nM, or 48 nM Pyrotinib for 4, 8, 12, or 16 days (designated p4, p8, p12, p16, respectively), followed by single-cell suspension preparation (viability > 90% via trypan blue staining), 10x Genomics library construction, and Illumina NovaSeq 6000 sequencing (paired-end 150 bp reads). 2.2 Processing of the scRNA-seq and GSE243275 dataset Leveraging two datasets, we first applied predefined thresholds to exclude low-quality cells; subsequently, the Seurat (v 5.0.1) package 15 was employed for both quality control and unsupervised clustering analysis. To identify genes exhibiting substantial expression heterogeneity across cells, which are referred to as highly variable genes, we normalized the transcriptomic data using the LogNormalize method, which prioritizes genes with relatively high coefficients of variation. Next, we performed principal component analysis (PCA) to achieve linear dimensionality reduction of the single-cell data, a process that assigned all cells to distinct principal components (PCs) based on weighted contributions. Following this dimensionality reduction step, the Louvain algorithm was used to cluster cells at multiple resolution levels, and the resulting clustering hierarchy was visualized via a clustering tree constructed with the clustree tool. We then further visualized the distribution and biological distances among different cell clusters using uniform manifold approximation and projection (UMAP), the scDblFinder algorithm was final utilized to eliminate all detected doublet artifacts For the scRNA-seq dataset, marker genes for each cell cluster were defined as the top 5 genes with significant expression differences (assessed via the Wilcoxon rank-sum test); these markers were thereafter subjected to Gene Ontology Biological Process (GO-BP) enrichment analysis. In parallel, in time-series analyses, genes with a membership value ≥ 0.7 were designated as Pyrotinib resistance-related genes. For the GSE243275 dataset specifically, we additionally conducted cell annotation via manual curation using public databases; we then applied the CRAD algorithm for spatial transcriptomic cell annotation, followed by the use of Cottrazm to identify tumor malignant regions, and ultimately further evaluated the functional activity of tumor-associated features. 2.3 Identification of Pyrotinib-related candidate genes in BRCA Firstly, for the GSE243275 dataset, we performed differential expression analysis between malignant and non-malignant regions using the Wilcoxon rank-sum test to obtain the first set of differentially expressed genes (DEGs1, [|log 2 FC| > 1, FDR < 0.05]). Next, we applied the DESeq2 (v 1.44.0) package 16 to conduct an additional round of differential expression analysis between tumor and control samples in the TCGA-BRCA cohort, which yielded a second set of differentially expressed genes (DEGs2, [|log 2 FC| > 1, FDR < 0.05]). Furthermore, we implemented weighted gene co-expression network analysis (WGCNA) on tumor samples from TCGA-BRCA, with the preliminary step of excluding outlier samples to ensure data robustness. Subsequently, we used the pickSoftThreshold function and set the scale-free topology fitting index (R 2 ) > 0.85 to screen for soft-thresholding powers that exceeded the red cutoff line. We then constructed a clustered gene dendrogram to identify distinct co-expression modules, from which we extracted the modules showing the strongest positive and negative correlations with the target traits; the core genes within these modules were defined as BRCA-associated hub genes. A combined threshold of Gene Significance (GS) > 0.25 and Module Membership (MM) > 0.25 was applied to validate the reliability of the identified hub genes. Ultimately, we took the intersection of upregulated and downregulated genes among DEGs1, DEGs2, hub genes, and Pyrotinib resistance-related genes to obtain Pyrotinib-related candidate genes in BRCA. Using the ClusterProfiler package, based on the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases, the candidate genes were analyzed as a specific gene set and aPEAR conducted further clustering analysis on all the significantly enriched results. 2.4 Construction of a risk model Firstly, BRCA samples with complete survival information from TCGA database were randomly divided into a training set and a testing set at a ratio of 7:3. Subsequently, combined with the GSE20685 dataset, ten advanced machine learning algorithms, including gradient boosting machine (GBM), least absolute shrinkage and selection operator (Lasso), CoxBoost, partial least squares regression for Cox models (plsRcox), random survival forest (RSF), stepwise Cox regression (StepCox), supervised principal components (SuperPC), survival support vector machine (survival-SVM), Ridge regression, and elastic net (Enet), were employed to identify the prognostic model with the optimal predictive performance based on the candidate genes. The diagnostic value of the constructed prognostic model was then evaluated using receiver operating characteristic (ROC) curves. According to the risk scores calculated based on the prognostic genes in the TCGA-BRCA cohort, BRCA samples were stratified into high-risk and low-risk groups using the median risk score as the cutoff value. Kaplan-Meier (KM) survival curves were plotted to compare the survival status between the two risk groups. Finally, the reliability of the risk model was further validated using ROC curve analysis based on the self-sequenced data of triple-negative BRCA samples included in this study, with a threshold of area under the curve (AUC) > 0.6. Additionally, the predictive ability and generalization performance of the model were assessed through a meta-analysis. 2.5 The acquisition and validation of biomarkers Using the glmnet (v 1.0) package 17 (with α = 0), we constructed a Cox model, which allowed us to identify the corresponding biomarkers in BRCA after parsing the optimal model parameters, and we then validated the diagnostic efficacy of these biomarkers by generating ROC curves and performing inter-group differential expression analyses in both the TCGA-BRCA and GSE86374 datasets, while at the protein level we leveraged the Human Protein Atlas (HPA) database to conduct further validation via immunohistochemical assessments. 2.6 Clinical baseline analysis of biomarkers To further assess the correlation between all biomarkers and common BRCA-related clinical features, all BRCA samples in TCGA-BRAC were stratified into high- and low-expression groups based on the median expression level of each biomarker, followed by the evaluation of inter-group differences in various clinical indicators. For numeric variables, data were presented as mean ± standard deviation and compared between groups using the t-test; for categorical variables, frequencies and percentages were calculated and inter-group comparisons were performed using the chi-square test. After completing the baseline analysis, a baseline table was generated. 2.7 Function enrichment and regulatory network analysis of biomarkers To explore the pathways highly correlated with each biomarker in BRCA, GO and KEGG enrichment analyses were performed on the high- and low-expression groups of each biomarker using the clusterProfiler package. The final results were filtered according to the following criteria: FDR < 0.05, q-value 1. Based on the enrichment results, we further integrated the outcomes of all biomarkers and identified their co-regulated pathways in BRCA by taking the intersection, which were visualized as heatmaps. In addition, to predict the interaction pairs at different levels, we leveraged the STRING database ( https://string-db.org/ ) STRING and GeneMANIA database ( http://genemania.org/ ) using the biomarkers and other candidate genes, followed by the construction of an integrated interaction network using Cytoscape. Subsequently, miRNAs targeting the biomarkers were predicted via the miRTarBase 2025 database ( https://mirtarbase.cuhk.edu.cn/~mirtarbase/mirtarbase_2025/ , Release 10.0 ). LncRNAs with high target-directed miRNA degradation scores that could target the key miRNAs were screened using the starBase database ( https://starbase.sysu.edu.cn/ ) , and key transcription factors (TFs) with high binding strength scores were identified via the JASPAR database ( http://jaspar.genereg.net/ ). Finally, an integrated regulatory network was constructed and visualized using Cytoscape. 2.8 The localization and molecular docking analysis of biomarkers To further characterize the genomic and subcellular localization features of the identified biomarkers, we first performed chromosomal localization analysis using the RCircos (v 1.2.2) package 18 , with annotation data derived from AnnoProbe ( https://github.com/jmzeng1314/AnnoProbe ) and the UCSC.HG38.Human.CytoBandIdeogram dataset ( https://genome.ucsc.edu/ ) , and visualized the results as circular chromosomal localization maps. For subcellular localization analysis of all biomarkers, we employed the RNALocate tool and visualized the corresponding distribution profiles to gain insights into their intracellular localization patterns. Additionally, candidate therapeutic agents targeting the biomarkers were predicted by retrieving drugs with a negative correlation (similarity score < 0) from the L1000FWD database ( https://maayanlab.cloud/L1000FWD/ ). The small-molecule chemical structures of the top 10 candidate drugs were obtained from the PubChem database ( https://pubchem.ncbi.nlm.nih.gov/ ) , while the three-dimensional (3D) macromolecular structures of all biomarkers were retrieved from the PDB database ( https://www.rcsb.org/ ) and AlphaFoldDB ( https://alphafold.ebi.ac.uk/ ). Molecular docking simulations were subsequently conducted using AutoDock Vina, and the docking results as well as force analysis were visualized in detailed 3D structural models via PyMOL. 2.9 Cell communication and pseudotime analysis of spatial transcriptomics Cell-cell communication was determined using the CellChat (v 1.5.0) package 19 . A thicker connecting line between ligands and receptors indicates a stronger potential interaction between cells. In addition, spatial cell communication distribution maps (incoming and outgoing modes) and ligand-receptor co-expression distribution maps within pathways were generated for visualization. Based on the spots corresponding to malignant regions in the spatial transcriptome data, we performed independent subclustering of malignant regions, which yielded a total of 5 distinct subclusters (cluster 0–4). Subsequently, Monocle2 (v 2.24.0) package 20 was employed to conduct spatial pseudotime analysis. Furthermore, leveraging the pseudotime analysis results, we further investigated the dynamic expression patterns of the 4 biomarkers in tumor malignant tissues. 2.10 Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) SKBR3 and BT474 human BRCA cell lines were selected for this study due to their inherent sensitivity to Pyrotinib 21 , 22 . These mycoplasma-free, authenticated cell lines, obtained from the American Type Culture Collection (ATCC), were utilized as parental controls for the derivation of drug-resistant sublines 21 . Total RNA was isolated from 1×10 6 cells per sample, with three biological replicates for each of the four groups: SKBR3 control, SKBR3 drug-resistant, BT474 control, and BT474 drug-resistant, using the Animal Total RNA Isolation Kit (Cat. No. RE-03014, Foregene). Then, 2 µg of total RNA was utilized to synthesize cDNA with SweScript RT I First Strand cDNA Synthesis Kit (Cat. No. G3330, Servicebio). Subsequently, PCR amplification was conducted in a 20-µL reaction system. PCR amplification was performed under the following conditions: an initial pre-denaturation step at 95 ℃ for 5 min, followed by 40 cycles of denaturation at 95 ℃ for 10 s and annealing/extension at 60 ℃ for 30 s. A melting curve analysis was then conducted by ramping the temperature from 60°C to 95 ℃ at a rate of 0.3 ℃ every 15 s. The CT values after amplification were harvested, and β-actin was employed as the internal reference gene. The results were evaluated by 2 −ΔΔCT , and the primer sequences (5'–3') utilized for RT-qPCR were as follows: β-Actin-F, AATCTGGCACCACACCTTCTACAA; β-actin-R, GGATAGCACAGCCTGGATAGACAA; HMGB3-F, CTGCCCAGACTAGCGAAACAA; HMGB3-R, GCCATCCTGACTGAATTGCTT; TFPI-F, GGTCGCGAATGGTTTCCAGGT; TFPI-R, AGCGCCATTCATTCCAACAT; ACTG2-F, GTGTGAAGAGGAGACACCACG; ACTG2-R, CAGATCTGGTGGCAGAGG; JCHAIN-F, GGGAGTCCTGGCGGTTTTTA; JCHAIN-R, TGGAAGTAATCCGGGCACAC. Finally, the expression differences of biomarkers were compared and presented graphically by means of GraphPad Prism 10 software (ANOVA-test, p < 0.05). 2.11 Western blot assay Tumor cells were collected by centrifugation, and total protein was extracted by lysing 1×10 6 cells in 250 µL of RIPA lysis buffer (Cat. No. BL504A, Biosharp). After centrifugation, protein concentrations were determined using a BCA Protein Assay Kit (Cat. No. BL521A, Biosharp). For sample preparation, total protein was mixed with 5× SDS loading buffer at a 4:1 ratio, followed by denaturation at 95 ℃ for 5 min (or 70 ℃ for 5–10 min for membrane proteins) and cooling on ice. For SDS-PAGE, glass plates were cleaned and assembled, separating and stacking gels were cast, and electrophoresis was performed at 80 V for the stacking gel and 120 V for the separating gel. For protein transfer, a PVDF membrane was activated with methanol, and a transfer sandwich was assembled before wet transfer at 300 mA for 30 min (or 200 mA for 1 h, or 25 V overnight). The membranes were then blocked with 5% non-fat milk in TBST for 1 h at room temperature, followed by overnight incubation at 4 ℃ with primary antibodies: anti-GAPDH (rabbit, 1:5000, Cat. No. AF7021, Affinity), anti-HMGB3 (rabbit, 1:1000, Cat. No. HA722033, HUABIO), anti-TFPI (rabbit, 1:1000, Cat. No. HA722199, HUABIO), anti-ACTG2 (rabbit, 1:1000, Cat. No. ER62586, HUABIO), and anti-JCHAIN (rabbit, 1:1000, Cat. No. HA721205, HUABIO). After washing, the membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. Protein expression levels were visualized using an enhanced chemiluminescence (ECL) reagent (Cat. No. BL520B, Biosharp). 2.12 Statistical analysis R (v 4.4.3) language was utilized for all statistical analyses. Wilcoxon test and t-test were used to compare the differences between two groups. The value of p < 0.05 was considered statistically significant. 3. Results 3.1 740 Pyrotinib resistance-related genes were obtained in scRNA-seq dataset Following initial data screening, we quantified the changes in cell counts for the p0, p4, p8, p12, and p16 groups before and after quality control ( Supplementary Fig. 1A and Supplementary Table 1 ), identified the top 10 highly variable genes ( Supplementary Fig. 1B ), extracted 7 PCs after dimensionality reduction (Fig. 1 A), and achieved the optimal clustering resolution of 0.3, which yielded 7 distinct cell clusters (Fig. 1 B- 1 C). The comparison plot of doublet identification distributions reflected the genuine cellular heterogeneity of the samples ( Supplementary Fig. 1C ). In addition, functional enrichment analysis of the marker genes for each cluster revealed that all cell clusters exhibited prominent spatial aggregation characteristics, with the enriched functional pathways being highly consistent with the spatial distribution of the corresponding clusters, thus demonstrating a clear correlation between cluster-specific biological functions and cellular clustering patterns (Fig. 1 D). Further inter-group differential abundance analysis indicated that the SKBR3_type0 subtype was the predominant cell type prior to Pyrotinib treatment; after treatment, the abundance of this subtype decreased significantly, with concomitant differentiation into other subtypes. Among these derivative subtypes, some displayed drug concentration-dependent sensitivity (e.g., type2 and type5), whereas others were associated with drug resistance (e.g., type1, type4, and type6), whose abundances peaked under the p12 treatment condition (Fig. 1 E). Finally, time-series analysis categorized all genes into 12 clusters with distinct temporal expression trends and identified a total of 740 Pyrotinib resistance-related genes (Fig. 1 F). 3.2 A comprehensive single-cell dataset analysis in BRCA Consistent with the scRNA-seq dataset, dimensionality reduction of the GSE243275 dataset yielded 8 PCs, and clustering at the optimal resolution of 0.5 resulted in 16 distinct cell clusters (Fig. 2 A- 2 C). After manual annotation, 7 major cell populations were identified: the abundance of drug resistance-related populations showed a gradual increase across treatment stages, while that of drug-sensitive populations exhibited a progressive decline (Fig. 2 D- 2 E). Additionally, the spatial transcriptomics dataset achieved nearly 100% spot coverage, with all spots assigned to 11 clusters (Fig. 2 F). CRAD results indicated that epithelial cells and fibroblasts were the dominant cell types in the spots (Fig. 2 G), and a strong positive correlation was observed between epithelial cells and fibroblasts (Fig. 2 H). Subsequently, we used Cottrazm to demarcate the malignant tumor region, tumor-normal tissue boundary region, and normal tissue region ( Supplementary Fig. 2 ), the key pathways involved in tumor-related features were further visualized ( Supplementary Fig. 3 ). 3.3 Identification of 39 Pyrotinib-related candidate genes A total of 5,628 DEGs1 were identified in the GSE243275 dataset, including 2,354 significantly upregulated genes and 3,274 significantly downregulated genes (Fig. 3 A). Similarly, 5,347 significantly upregulated genes and 2,867 significantly downregulated genes were obtained in the TCGA-BRCA cohort (Fig. 3 B). Subsequent WGCNA showed no outlier samples (Fig. 3 C), and the optimal soft threshold was determined to be 6 (Fig. 3 D). This analysis yielded 9 co-expression modules and 4,616 hub genes (Fig. 3 E- 3 F). Notably, the GS and MM values of most hub genes were above 0.25, indicating good correlations between hub genes and both traits and modules (Fig. 3 G). Finally, Venn diagram analysis identified 39 Pyrotinib-related candidate genes as the intersection of DEGs, including 17 tumor-upregulated and 22 tumor-downregulated genes (Fig. 3 H). A total of 39 Pyrotinib-associated candidate genes were subjected to enrichment analysis, which yielded 7 KEGG pathways, 113 biological process terms, 43 cellular component terms, and 30 molecular function terms, such as critical biological processes such as regulation of apoptotic process, central nervous system development, and neural precursor cell proliferation (Fig. 3 I- 3 J). 3.4 Prognostic risk model was construct in TCGA-BRCA After C-index validation and evaluation, the StepCox [forward] + Ridge model achieved the highest evaluation values among all models: 0.65002337267734 in the training set, 0.64594127806563 in the testing set, and 0.612130696637739 in the GSE20685 (Fig. 4 A). Under the ROC criterion, the AUC value of this model was greater than or equal to 0.6, thus it was identified as the optimal model ( Supplementary Fig. 4 ). Subsequent KM curves showed that the optimal model achieved good discriminability in the grouping of all three datasets; the survival curve of the high-risk group was significantly lower than that of the low-risk group with no obvious overlap, indicating favorable validation results (Fig. 4 B- 4 D). Time-ROC curve analysis in triple-negative BRCA self-sequencing data demonstrated that the optimal model still had certain diagnostic value in this dataset, with good predictive accuracy at 2 and 3 years (AUC > 0.6) and Meta-analysis also showed that the optimal prognostic model exhibited excellent predictive performance across different datasets (HR > 1), which further confirmed its good stability and effectiveness (Fig. 4 E- 4 F). 3.5 HMGB3, TFPI, ACTG2, and JCHAIN were biomarkers in BRCA Using the glmnet package, we identified that the StepCox [forward] + Ridge model consists of HMGB3, TFPI, ACTG2, and JCHAIN as features. Within the established optimal prognostic model, HMGB3 and TFPI were assigned positive coefficients (0.1348 and 0.1244, respectively), indicating their higher expression levels of these genes correlate with elevated risk scores, whereas ACTG2 and JCHAIN carried negative coefficients (-0.0730 and − 0.1085, respectively), consistent with their role as protective factors (Fig. 5 A, Table 1 ). Importantly, these four genes showed high diagnostic efficacy (AUC > 0.7) and consistently significant differential expression trends in both TCGA-BRCA and GSE86374 datasets (Fig. 5 B– 5 E). Similarly, the protein expression trends of the four genes at the immunohistochemical level were consistent with the conclusions obtained from Wilcoxon test (Fig. 5 F). In summary, we defined HMGB3, TFPI, ACTG2, and JCHAIN as the key genes in this study. Table 1 The distribution of characteristic genes and coefficients of the optimal prognosis model gene coef HMGB3 0.134881202254666 TFPI 0.124479160815218 ACTG2 -0.073082741 JCHAIN -0.108561139 3.6 Analysis of the functions and regulatory characteristics of biomarkers Clinical baseline profiles demonstrated that all the identified biomarkers exhibited certain prognostic phenotypic correlations at multiple levels ( Supplementary Table 2 ). Gene set enrichment analysis results revealed that TFPI was enriched in the largest number of pathways among all the biomarkers (Table 2 ). In BRCA, all the candidate biomarkers were jointly correlated with 24 pathways, among which the immunoglobulin complex pathway showed the strongest correlation (Fig. 6 A- 6 B). Table 2 GSEA enrichment results gene KEGG GO_BP GO_CC GO_MF TFPI 120 1313 154 194 HMGB3 92 593 123 129 JCHAIN 75 752 65 95 ACTG2 77 938 72 109 Protein-protein interaction (PPI) network analysis identified three distinct functional modules centered on the core biomarkers. Specifically, the module anchored by ACTG3 clustered associated proteins including MEOX1, FLNC and SVEP1, which were mainly linked via co-expression and co-localization interactions. The sub-module with TFPI as the hub showed direct protein interactions with CENPA, DPP4 and MAF. Meanwhile, the module centered on HMGB3 incorporated RFC2/RFC4/RFC5, NUPR5 and other functional proteins such as GGT and TK1, which were connected through genetic interactions and shared protein domains. These findings collectively suggested that the candidate biomarkers might exert downstream regulatory functions via intricate protein interaction networks (Fig. 6 C). Furthermore, upstream regulatory molecule prediction was performed using databases including miRTarBase and starBase. A total of 84 miRNAs targeting the core biomarkers, 61 lncRNAs binding to these miRNAs, and 32 key TFs were screened out. The constructed regulatory network visually illustrated the potential upstream regulatory molecules of the core biomarkers and their interaction patterns (Fig. 6 D). Chromosomal localization analysis indicated that HMGB3 was mapped to the X chromosome, JCHAIN to chromosome 4, and both ACTG2 and TFPI to chromosome 2 (Fig. 6 E). Subcellular localization analysis further showed that TFPI was predominantly distributed in the extracellular space and on the cell membrane, JCHAIN was concentrated in the extracellular region, ACTG2 was mainly located in the cytoplasm, and HMGB3 was primarily localized in the nucleus (Fig. 6 F). The final molecular docking results showed that ABT-737 had the strongest binding ability to all four biomarkers simultaneously, making it the most promising targeted drug (Fig. 6 G). The detailed three-dimensional structure diagram is shown in Fig. 6 H. 3.7 Spatial transcriptomic analysis of cell communication and malignant tissue dynamics Using spatial transcriptomic data, 186 cell communication pathways were identified via CellChat. The communication network revealed Fibroblast as the core node, forming extensive connections with T cells, epithelial cells, and other populations (Fig. 7 A). For the 2-AG signaling pathway, T/epithelial cells acted as signal senders, while Fibroblast served as the key receiver, with ligand-receptor co-expression verifying this spatial interaction (Fig. 7 B). Malignant-region spots were clustered into 5 subclusters (cluster 0–4) via Monocle2, and trajectory analysis showed continuous dynamic transitions among clusters (Fig. 7 C). Temporal tracing of 4 biomarkers indicated heterogeneous expression trends across pseudotime: ACTG2/HMGB3 first decreased then increased, while TFPI/JCHAIN gradually declined, reflecting biomarker expression dynamics during malignant progression (Fig. 7 D). 3.8 Upregulation of JCHAIN, HMGB3, ACTG2, and TFPI in Pyrotinib-resistant BRCA cells As shown in Fig. 8 , both RT-qPCR and Western blot analyses demonstrated a consistent and significant upregulation of JCHAIN, HMGB3, ACTG2, and TFPI in Pyrotinib-resistant BRCA cell models. RT-qPCR (Figs. 8 A–D) revealed that the mRNA expression levels of these four genes were markedly elevated in the drug-resistant groups of both SKBR3 and BT474 cell lines compared with their respective control groups (p < 0.001). In line with the transcriptional data, Western blot assays (Fig. 8 E) and subsequent densitometric quantification (Figs. 8 F) confirmed that the protein expression levels of JCHAIN, HMGB3, ACTG2, and TFPI were also significantly increased in the resistance group relative to the control group (p < 0.001), validating the concordant upregulation of these genes at both the mRNA and protein levels in Pyrotinib-resistant cells. 4. Discussion BRCA remains a leading contributor to cancer-associated morbidity and mortality worldwide 23 . While Pyrotinib exerts robust antitumor activity against BRCA, long-term administration often culminates in the emergence of acquired resistance, which severely compromises the efficacy of clinical interventions 24 . This is the first integrative analysis of single-cell, spatial (GSE243275) and bulk RNA-seq data in Pyrotinib-resistant SKBR3 cells, which identifies 740 resistance-related genes and narrows them down to 39 candidates. Through the comparison and validation of ten machine learning algorithms, we constructed a four-gene prognostic model (StepCox[forward]+Ridge) comprising HMGB3, TFPI, ACTG2, and JCHAIN. The identified biomarkers were validated at both mRNA and protein levels, found to participate in immune-related pathways with defined chromosomal and subcellular localizations. Spatial transcriptomic analysis revealed a fibroblast-centered cell communication network and the dynamic expression patterns of these biomarkers during malignant progression. Furthermore, molecular docking identified ABT-737 as a potential targeted agent, providing a theoretical basis for reversing Pyrotinib resistance. At last, Both RT-qPCR and Western blot analyses revealed that JCHAIN, HMGB3, ACTG2, and TFPI were significantly upregulated at both mRNA and protein levels in Pyrotinib-resistant BRCA cells. Compared to previously published prognostic models for BRCA, our four-gene model demonstrates competitive discriminative performance across multiple independent cohorts. For instance, in the pivotal TCGA-BRCA training set, our model achieved a C-index of 0.65, which is comparable to, if not superior than, the performance (C-index typically ranging 0.60–0.80) of widely used multi-gene prognostic tools (such as certain immune-related or proliferation-related signatures) reported in similar cohorts 25 . More importantly, our model is specifically constructed for the context of Pyrotinib resistance. Its core genes (HMGB3, TFPI, ACTG2, JCHAIN) not only show significant associations with overall patient survival but also maintain stable AUC values above 0.6 in the GSE20685 and internal validation cohorts. This indicates its reliable and reproducible predictive capability for identifying high-risk patients with poor response to Pyrotinib treatment. In contrast, many general-purpose BRCA prognostic models are not optimized for resistance to specific targeted agents, particularly Pyrotinib, which may limit their predictive accuracy in relevant subgroups 26 , 27 . Therefore, our model is not intended to replace existing generic classifiers but serves as a crucial complementary tool. It provides the first specifically dedicated assessment framework, systematically validated by multi-omics data and machine learning algorithms, to aid clinicians in evaluating the prognosis and potential resistance risk for HER2-positive BRCA patients undergoing Pyrotinib therapy. The four core biomarkers (HMGB3, TFPI, ACTG2, JCHAIN) uncovered in this study were validated at both mRNA and protein levels and are closely linked to key biological processes in BRCA. Among them, HMGB3 (high mobility group protein B3) exhibits high expression in stem cells and cancer cells while showing minimal transactivation in normal adult tissues, thereby emerging as a promising therapeutic target; notably, it is aberrantly overexpressed and plays a pivotal role in the malignant progression of multiple cancer types 28 , 29 . This finding is consistent with the result of the present study that HMGB3 is defined as a risk factor for BRCA. Previous studies have reported that HMGB3 promotes resistance to anti-PD-1 therapy in triple-negative BRCA by interfering with the IFN-γ-mediated ferroptosis pathway, which suggests its potential involvement in the modulation of the tumor immune microenvironment and highlights its value for the development of combination therapeutic strategies 30 . TFPI, functioning as a tumor suppressor gene, can inhibit BRCA cell proliferation and invasion by suppressing the ERK/p38 MAPK signaling pathway 31 . Our study further confirms that its low expression is associated with poor prognosis, supporting its dual role as both a prognostic marker and a therapeutic target. ACTG2 is associated with cytoskeletal remodeling and invasive phenotypes, and its overexpression may drive malignant progression and metastasis in BRCA 32 . However, in both Wang et al.’s BRCA study and our current work, ACTG2 expression is significantly downregulated in tumor tissues 33 . This apparent discrepancy can be attributed to the context-dependent functional role of ACTG2, which is contingent on tumor molecular subtypes and study cohort characteristics: First, the study 30 reporting pro-oncogenic effects of ACTG2 overexpression likely focused on specific cell line models or cohorts of late-stage, highly aggressive BRCA. In this setting, elevated ACTG2 expression drives metastatic phenotypes by modulating the RhoA/ROCK signaling pathway to promote actin polymerization, thereby enhancing cytoskeletal remodeling and cell migratory capacity. In contrast, our cohort (as well as Wang et al.’s) may be enriched in early-stage cases or subtype tumors, which contexts in which reduced ACTG2 expression confers a protective effect: the downregulation of ACTG2 attenuates aberrant cytoskeletal rearrangement, limiting uncontrolled tumor cell proliferation and invasion. This discrepancy does not represent a contradiction but rather reflects the functional complexity of ACTG2 in BRCA. Its pro-oncogenic or tumor-suppressive role is dependent on tumor stage, molecular subtype, and the gene interaction network, thereby providing a novel molecular basis for the stratified treatment of BRCA. More importantly, JCHAIN has been identified as a key gene with prognostic value in BRCA, which constitutes the core finding of a recent study by Shi et al. 34 . Consistent with the key results of our current research, the sustained downregulation of JCHAIN mRNA expression in tumor samples relative to normal counterparts, coupled with its associations with TNM stage and BRCA subtypes, strongly indicates that JCHAIN is intricately involved in the biological processes driving BRCA progression. Notably, these genes are collectively enriched in pathways such as apoptotic regulation, central nervous system development, and neural precursor cell proliferation, suggesting that Pyrotinib resistance may involve complex biological processes including cell fate remodeling and microenvironmental adaptation. Our spatial transcriptomic analysis via CellChat identified 186 cell communication pathways, with Fibroblasts emerging as the central hub that establishes extensive crosstalk with T cells, epithelial cells, and other cellular components in the BRCA TME 35 . Such a central hub role implies that Fibroblasts may integrate signals from multiple cell types (e.g., pro-inflammatory cues from T cells, oncogenic signals from epithelial cells) and transduce downstream effects by secreting cytokines, remodeling the extracellular matrix (ECM), or regulating metabolic pathways, which consistent with previous reports that cancer-associated Fibroblasts (CAFs) drive BRCA malignancy through paracrine signaling and ECM reshaping 36 , 37 . Notably, the 2-arachidonoylglycerol (2-AG) signaling pathway displayed a distinct cell communication pattern. As a major endocannabinoid, 2-AG is known to regulate immune responses, cell proliferation, and ECM remodeling in cancers, but its cell-type-specific crosstalk in BRCA TME remains undercharacterized. Our findings uncover a novel regulatory axis: tumor-infiltrating T cells may secrete 2-AG-related ligands to modulate Fibroblast activation, extending previous observations that endocannabinoids suppress T cell function by highlighting this indirect anti-tumor immunity-weakening crosstalk 38 , 39 . Meanwhile, cancerous epithelial cells may secrete 2-AG to instruct Fibroblasts to support tumor growth, consistent with reports that epithelial-stromal crosstalk drives CAF activation via lipid signaling 40 . Collectively, the 2-AG axis bridges immune, tumor, and stromal cells, coordinating TME remodeling to favor BRCA progression. Additionally, the dynamic expression changes of the biomarkers along the pseudotime trajectory within malignant regions further corroborate their regulatory roles in tumor evolution. Notably, the expression patterns of JCHAIN, HMGB3, ACTG2, and TFPI observed in our Pyrotinib-resistant cell models appear to differ from those reported in previous transcriptomic analyses, where JCHAIN, ACTG2, and TFPI are frequently downregulated in primary breast tumors compared with normal tissues. This apparent discrepancy may reflect fundamental differences in the biological contexts interrogated: large-scale bioinformatic analyses primarily capture gene expression signatures associated with tumor initiation and progression, which are broadly represented in heterogeneous tissue samples, whereas our in vitro model specifically isolates the adaptive response of SKBR3 and BT474 human breast cancer cells to prolonged Pyrotinib exposure. It is plausible that these genes, which may act as tumor suppressors in the context of primary oncogenesis, are co-opted to promote cell survival and drug resistance under therapeutic pressure. Alternatively, the discrepancy could arise from the absence of a complex tumor microenvironment in our cell culture system, which is known to profoundly shape gene expression in vivo. Collectively, these findings underscore the importance of validating bioinformatic predictions in mechanistic models and highlight the context-dependent roles of these biomarkers in BRCA biology. 5. Conclusions Thus, the four-gene prognostic model constructed in our study provides a reliable tool for assessing Pyrotinib treatment response and prognosis in BRCA patients. Simultaneously, it elucidates resistance-related molecular mechanisms and microenvironmental features, offering new targets and theoretical support for developing individualized treatment strategies, thereby holding significant value for advancing precision medicine in BRCA. This study also has certain limitations: all experiments were performed only in in vitro cell models, and in vivo validation as well as large‑scale clinical cohort studies have not been conducted. Furthermore, the expression and function of these key genes in clinical tumor samples and the potential impact of the tumor microenvironment remain to be elucidated. Future research should employ clinical specimens and in vivo models to verify the expression patterns and functional roles of these genes in pyrotinib resistance. In addition, further studies are warranted to explore the synergistic effects of ABT‑737 combined with pyrotinib, so as to promote the clinical translation of these findings. Abbreviations AUC Area under the curve BRCA Breast cancer CAFs Cancer-associated Fibroblasts ECM Extracellular matrix GO Gene Ontology GS Gene Significance KEGG Kyoto Encyclopedia of Genes and Genomes KM Kaplan-Meier MM Module Membership PCA Principal component analysis PCs Principal components ROC Receiver operating characteristic scRNA-seq Single-cell RNA sequencing TKI Tyrosine kinase inhibitor TFs Transcription factors WGCNA Weighted gene co-expression network analysis Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors have no relevant financial or non-financial interests to disclose. Funding This work was also supported by Health Commission of Sichuan Province Medical Science and Technology Program (Grant No. 24QNMP051) and the “Qimingxing” Research Fund for Young Talents of West China Hospital (HXOMX0082). Author’s contributions All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Qiheng Gou, Xueming Xia and Yuxin Xie. The first draft of the manuscript was written by Qiheng Gou and Yuxin Xi. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Acknowledgements We would like to express our sincere gratitude to all individuals and organizations who supported and assisted us throughout this research. Availability of data and materials The data used in this study were extracted from public databases. References Siegel RL, Kratzer TB, Cancer statistics (2026) Jan-Feb 2026;76(1):e70043. 10.3322/caac.70043 Xiong X, Zheng LW, Ding Y et al (2025) Breast cancer: pathogenesis and treatments. Feb 19(1):49. 10.1038/s41392-024-02108-4 Bray F, Laversanne M, Sung H (2024) Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. May-Jun 74(3):229–263. 10.3322/caac.21834 Xu Y, Gong M, Wang Y, Yang Y, Liu S, Zeng Q (2023) Global trends and forecasts of breast cancer incidence and deaths. Sci Data May 27(1):334. 10.1038/s41597-023-02253-5 Xiong X, Zheng LW, Ding Y et al (2025) Breast cancer: pathogenesis and treatments. Signal Transduct Target Ther Feb 19(1):49. 10.1038/s41392-024-02108-4 Lei S, Zheng R, Zhang S et al (2021) Global patterns of breast cancer incidence and mortality: A population-based cancer registry data analysis from 2000 to 2020. Cancer Commun (Lond) Nov 41(11):1183–1194. 10.1002/cac2.12207 Xu L, Xie Y, Gou Q et al (2024) HER2-targeted therapies for HER2-positive early-stage breast cancer: present and future. Front Pharmacol 15:1446414. 10.3389/fphar.2024.1446414 Li X, Zhang X, Yin S, Nie J (2025) Challenges and prospects in HER2-positive breast cancer-targeted therapy. Critical reviews oncology/hematology Mar 207:104624. 10.1016/j.critrevonc.2025.104624 Veeraraghavan J, De Angelis C, Gutierrez C et al (2025) HER2-Positive Breast Cancer Treatment and Resistance. Adv Exp Med Biol 1464:495–525. 10.1007/978-3-031-70875-6_24 Blair HA, Pyrotinib (2018) First Global Approval. Drugs Nov 78(16):1751–1755. 10.1007/s40265-018-0997-0 Qi X, Shi Q, Xuhong J, Zhang Y, Jiang J (2023) Pyrotinib-based therapeutic approaches for HER2-positive breast cancer: the time is now. Breast cancer research: BCR Oct 3(1):113. 10.1186/s13058-023-01694-5 Liu X, Zhang X, Shao Z et al (2023) Pyrotinib and chrysin synergistically potentiate autophagy in HER2-positive breast cancer. Signal Transduct Target Ther Dec 18(1):463. 10.1038/s41392-023-01689-w Wang X, Li Y, Jin R et al (2023) Single-cell analysis reveals the potential mechanisms of pyrotinib resistance in non-small cell lung cancer. Jan 13(1):17. 10.1038/s41392-022-01226-1 Fan J, Slowikowski K, Zhang F (2020) Single-cell transcriptomics in cancer: computational challenges and opportunities. Exp Mol Med Sep 52(9):1452–1465. 10.1038/s12276-020-0422-0 Gribov A, Sill M, Lück S et al (2010) SEURAT: visual analytics for the integrated analysis of microarray data. BMC Med Genom 3(1):21 Love M, Anders S, Huber W (2014) Differential analysis of count data–the DESeq2 package. Genome Biol 15(550):10–1186 Friedman J, Hastie T, Tibshirani R et al (2021) Package ‘glmnet’. CRAN R Repositary 595:874 Zhang H, Meltzer P, Davis S (2013) RCircos: an R package for Circos 2D track plots. BMC Bioinformatics 14(1):244 Jin S, Plikus MV, Nie Q (2025) CellChat for systematic analysis of cell–cell communication from single-cell transcriptomics. Nat Protoc 20(1):180–219 Trapnell C, Cacchiarelli D, Qiu X, Monocle (2017) Cell counting, differential expression, and trajectory analysis for single-cell RNA-Seq experiments. Bioconductor Kokot A, Gadakh S, Saha I et al (2024) Unveiling the Molecular Mechanism of Trastuzumab Resistance in SKBR3 and BT474 Cell Lines for HER2 Positive Breast Cancer. Curr Issues Mol Biol Mar 21(3):2713–2740. 10.3390/cimb46030171 Liu S, Lan B, Wang Y et al (2024) Pyrotinib and trastuzumab combination treatment synergistically overcomes HER2 dependency in HER2-positive breast cancer: insights from the PHILA trial. EBioMedicine Nov 109:105379. 10.1016/j.ebiom.2024.105379 Muñante B, Paz-Manrique R, Pinto JA, Gomez HL (2025) Clinical Management in BRCA Carriers with Early Breast Cancer. Cancer Control Jan-Dec 32:10732748251377864. 10.1177/10732748251377864 Wang X, Li Y, Jin R et al (2023) Single-cell analysis reveals the potential mechanisms of pyrotinib resistance in non-small cell lung cancer. Signal Transduct Target Ther Jan 13(1):17. 10.1038/s41392-022-01226-1 Cui F, Yan C, Wu J et al (2025) Identification and validation of prognostic genes associated with T-cell exhaustion and macrophage polarization in breast cancer. Front Endocrinol (Lausanne) 16:1556496. 10.3389/fendo.2025.1556496 Xu B, Yan M, Ma F et al (2021) Pyrotinib plus capecitabine versus lapatinib plus capecitabine for the treatment of HER2-positive metastatic breast cancer (PHOEBE): a multicentre, open-label, randomised, controlled, phase 3 trial. Lancet Oncol Mar 22(3):351–360. 10.1016/s1470-2045(20)30702-6 Jiang F, Lai J, Zhuo X et al (2023) HER2-positive breast cancer progresses rapidly after pyrotinib resistance: acquired RET gene fusion and TP53 gene mutation are potential reasons. Anticancer Drugs Nov 1(10):1196–1201. 10.1097/cad.0000000000001506 Wen B, Wei YT, Zhao K (2021) The role of high mobility group protein B3 (HMGB3) in tumor proliferation and drug resistance. Mol Cell Biochem Apr 476(4):1729–1739. 10.1007/s11010-020-04015-y Nemeth MJ, Kirby MR, Bodine DM (2006) Hmgb3 regulates the balance between hematopoietic stem cell self-renewal and differentiation. Proc Natl Acad Sci U S A Sep 12(37):13783–13788. 10.1073/pnas.0604006103 Luo B, Zheng H, Liang G, Luo Y, Zhang Q, Li X (Mar 2025) HMGB3 Contributes to Anti-PD-1 Resistance by Inhibiting IFN-γ-Driven Ferroptosis in TNBC. Molecular carcinogenesis . 64(3):490–501. 10.1002/mc.23861 Xing M, Yang Y, Huang J et al (2022) TFPI inhibits breast cancer progression by suppressing ERK/p38 MAPK signaling pathway. Jul 44(7):801–812. 10.1007/s13258-022-01258-5 Wu Y, Liu ZG, Shi MQ et al (2017) Identification of ACTG2 functions as a promoter gene in hepatocellular carcinoma cells migration and tumor metastasis. Biochemical Biophys Res communications Sep 16(2):537–544. 10.1016/j.bbrc.2017.04.007 Wang Z, Li TE, Chen M, Pan JJ, Shen KW (2020) miR-106b-5p contributes to the lung metastasis of breast cancer via targeting CNN1 and regulating Rho/ROCK1 pathway. Aging (Albany NY) Jan 27(2):1867–1887. 10.18632/aging.102719 Shi Y, Lin L, Zhu X et al (2025) Comprehensive Analysis of JCHAIN as a Potential Prognostic Factor for Breast Cancer and an Indicator for Tumor Microenvironment. Biomedicines Sep 26(10). 10.3390/biomedicines13102366 de Visser KE, Joyce JA (2023) The evolving tumor microenvironment: From cancer initiation to metastatic outgrowth. Cancer cell Mar 13(3):374–403. 10.1016/j.ccell.2023.02.016 Hu D, Li Z, Zheng B et al (May 2022) Cancer-associated fibroblasts in breast cancer: Challenges and opportunities. 42(5):401–434. 10.1002/cac2.12291 Li Y, Wang L, Ma W, Wu J, Wu Q, Sun C (2025) Paracrine signaling in cancer-associated fibroblasts: central regulators of the tumor immune microenvironment. J Transl Med Jun 23(1):697. 10.1186/s12967-025-06744-4 Xiong X, Chen S, Shen J et al (2022) Cannabis suppresses antitumor immunity by inhibiting JAK/STAT signaling in T cells through CNR2. Signal Transduct Target Ther Apr 6(1):99. 10.1038/s41392-022-00918-y Sido JM, Nagarkatti PS, Nagarkatti M (2016) Production of endocannabinoids by activated T cells and B cells modulates inflammation associated with delayed-type hypersensitivity. Eur J Immunol Jun 46(6):1472–1479. 10.1002/eji.201546181 Krauze I, Greb-Markiewicz B, Kłopot A, Maciejewska K, Bryk M, Krzystek-Korpacka M (2025) Neutrophil extracellular traps and cannabinoids: potential in cancer metastasis. Front Oncol 15:1595913. 10.3389/fonc.2025.1595913 Additional Declarations The authors declare no competing interests. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9376197","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":620838489,"identity":"5df7ea98-d483-49fc-b5f6-5ad71133ba23","order_by":0,"name":"Qiheng Gou","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Qiheng","middleName":"","lastName":"Gou","suffix":""},{"id":620838490,"identity":"95e1bbf1-c7d4-4063-9638-d99771291c5c","order_by":1,"name":"Xueming Xia","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Xueming","middleName":"","lastName":"Xia","suffix":""},{"id":620838491,"identity":"82ee853e-eda4-4ec8-a5b3-4869ebdfc984","order_by":2,"name":"Yuxin Xie","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAApUlEQVRIiWNgGAWjYNCCCgYeUrWcIVkLYxspqg1u5Bh/+DmvTsac/QDj5wJitEjOSEuT7N12mMeyJ4FZegYxWvglko8xM247wGNwIIGNmSgfsUkkNn9mnFPHY3D+AZFagLYckGZsYOYxuEGsLZI9z9Ike44dBmp52CxNlBaD48AQ+1FTZ29wPvngZ+JiRyABxmJsIEoD0DMHiFQ4CkbBKBgFIxcAALkyK6VfZImzAAAAAElFTkSuQmCC","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":true,"prefix":"","firstName":"Yuxin","middleName":"","lastName":"Xie","suffix":""}],"badges":[],"createdAt":"2026-04-10 07:52:30","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9376197/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9376197/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106876700,"identity":"b7e830cf-9aeb-4d56-85d0-bcbcb12f185a","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":142651,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSingle-cell transcriptomic profiling reveals SKBR3 cell heterogeneity and identification of Pyrotinib resistance-related genes under Pyrotinib treatment.\u003c/strong\u003ePrincipal component analysis (PCA) scree plot (A), Clustering resolution optimization via clustree (B), UMAP visualization of cell clustering (C), Functional enrichment heatmap of cluster-specific marker genes (D), Differential abundance of cell clusters across Pyrotinib treatment groups (p0/p4/p8/p12/p16, E), Time-series clustering of gene expression dynamics (F).\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/cb2d1870b90e9d5da64252cc.png"},{"id":106876702,"identity":"d5bbe0c9-d2c4-4bfb-8cef-558edb1bc7d3","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":200261,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIntegrated scRNA-seq and spatial transcriptomic profiling of Pyrotinib-treated SKBR3 cells.\u003c/strong\u003e Principal component analysis (PCA) scree plot for the GSE243275 dataset (A), Clustree plot for resolution optimization (B), UMAP visualization of cell clustering (resolution = 0.5, C), Manual annotation of 7 major cell populations (D), Abundance dynamics of annotated cell populations across treatment stages (E), Spatial transcriptomic spot clustering (F), Cell type deconvolution (CRAD) results for spatial spots (G), Correlation matrix of cell type abundances in spatial spots (H).\u003c/p\u003e","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/4463f256e73eb8645bd692ec.png"},{"id":106961443,"identity":"cc1a7651-7737-403e-a79e-db3565627941","added_by":"auto","created_at":"2026-04-15 09:25:35","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":143582,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentification of Pyrotinib-related candidate genes via differential expression and WGCNA.\u003c/strong\u003e Volcano plot of differential expressed genes (DEGs) in GSE243275 dataset (A), Volcano plot of DEGs in TCGA-BRCA cohort (B), Sample clustering for WGCNA outlier detection (C), Soft threshold selection for WGCNA (D), Gene co-expression module clustering (dendrogram, E), Module-trait relationship heatmap (F), Module membership (MM) vs. gene significance (GS) scatter plots (G), Venn diagram of candidate gene intersection (H), GO functional enrichment results for candidate genes (I), KEGG pathway enrichment results for candidate genes (J).\u003c/p\u003e","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/6ffb362efc0f26bbc5de407c.png"},{"id":106876704,"identity":"eb7d47a8-c0e4-4128-82b0-15cd6b0cefc4","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":91028,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePerformance validation of the optimal StepCox [forward] + Ridge prognostic model.\u003c/strong\u003e C-index distribution of the optimal model across datasets (A), Kaplan-Meier (KM) survival curves in Training set (B), Testing set (C), GSE20685 (D), Time-dependent ROC curve for triple-negative BRCA self-sequencing data (E), Meta-analysis of the optimal model’s hazard ratio (HR) across datasets (F).\u003c/p\u003e","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/68e32c50c8e945f59a7aa578.png"},{"id":106876706,"identity":"b636571e-97d6-4c59-839e-ec2ded9a36cf","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":112510,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCharacterization and validation of key genes in the optimal prognostic model. \u003c/strong\u003eFeature coefficient profile and stability of the StepCox [forward] + Ridge model (A), ROC curves for diagnostic efficacy of the four key genes in the TCGA-BRCA cohort (B), GSE86374 dataset (C), Differential expression of the four key genes in TCGA-BRCA (D), GSE86374 (E), Immunohistochemical (IHC) validation of key gene protein expression (F).\u003c/p\u003e","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/c8637847ad39d78a18760ea2.png"},{"id":106876707,"identity":"d8638598-4b6f-4a4c-9383-a9298cd84b0d","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":151544,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional enrichment and protein-protein interaction (PPI) network analysis of key biomarkers.\u003c/strong\u003e Venn diagram of shared enriched pathways among the four key biomarkers (A), Heatmap of normalized enrichment scores (NES) for top shared pathways (B), PPI network and functional module analysis (C), ceRNA-TF regulatory network of core biomarkers (D), Chromosomal localization of core biomarkers (E), Subcellular localization distribution of core biomarkers (F), Heatmap of molecular docking Vina scores (G), Three-dimensional (3D) structure of ABT-737-biomarker complexes (H).\u003c/p\u003e","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/0e019e16e567ae983383aed1.png"},{"id":106961261,"identity":"d9e0ac72-c6ff-4460-b4a4-fa4a0bed9cf7","added_by":"auto","created_at":"2026-04-15 09:24:52","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":102374,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial cell communication and pseudotime trajectory analysis of malignant-region cells.\u003c/strong\u003e Spatial cell communication network (A), Ligand-receptor co-expression analysis for the 2-AG signaling pathway (B), Pseudotime trajectory analysis of malignant-region spots (C), Biomarker expression dynamics along pseudotime (D).\u003c/p\u003e","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/b6f734bdd7081d51c0ad0587.png"},{"id":106876708,"identity":"c1d7fcfa-baa4-4129-b897-f80d86e26609","added_by":"auto","created_at":"2026-04-14 10:37:35","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":78470,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExpression of JCHAIN, HMGB3, ACTG2, and TFPI in Pyrotinib-resistant BRCA cells.\u003c/strong\u003e mRNA expression levels of JCHAIN (A), HMGB3 (B), ACTG2 (C), and TFPI (D) in SKBR3 and BT474, Representative Western blot images of biomarkers in SKRB3-R (SKRB3-Resistantance group), BT474-R (BT474-Resistantance group) and SKRB3-C (SKRB3-Control group), BT474-C (BT474-Control group) (E), Quantification of protein expression levels (F), normalized to GAPDH. Results are representative of three independent experiments. ** indicate p \u0026lt; 0.01, *** indicate p \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/087785da861e4f5f8fabd7ad.png"},{"id":106994339,"identity":"f1ba64f0-2f12-4835-afb3-205def6761bc","added_by":"auto","created_at":"2026-04-15 15:07:48","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2857883,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/45e25208-2f68-434f-adcd-79b9842c1550.pdf"},{"id":106960738,"identity":"af04b615-0dbb-43dd-ba98-4daf4d3dad2d","added_by":"auto","created_at":"2026-04-15 09:22:53","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3177698,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9376197/v1/d2cae07e4fb08e0277b3a526.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eIntegrative analysis of scRNA-seq, spatial transcriptomics, and machine learning constructs a prognostic model for Pyrotinib resistance in breast cancer\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"Novelty and Impact","content":"\u003cp\u003eThis study uniquely integrates single-cell and spatial transcriptomics with machine learning to construct a four‑gene prognostic model for Pyrotinib resistance in breast cancer. It reveals fibroblast‑centric cell communication, dynamic biomarker expression, and identifies ABT‑737 as a potential targeted therapy. The model offers a robust clinical tool to predict drug response and overcome resistance, advancing precision oncology strategies.\u003c/p\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eBreast cancer (BRCA) represents the most prevalent malignancy among women worldwide\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Globally, BRCA is responsible for approximately one-third of all malignancies diagnosed in women, with its mortality rate accounting for roughly 15% of the total diagnosed cases\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e,\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. It is noteworthy that the absolute incidence of BRCA has been on the rise in numerous developing countries, a trend that can be attributed to the dual effects of population expansion and the growing adoption of Western lifestyles\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. The global epidemiological distribution of BRCA is shaped by a complex interplay of genetic predispositions, environmental exposures, and modifiable lifestyle factors\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. In recent years, advances in the precise diagnosis of molecular subtypes have underpinned the optimization of BRCA therapeutics, and the integration of novel agents and neoadjuvant treatment regimens has further enhanced the precision of locoregional therapeutic interventions\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Nevertheless, disease recurrence driven by acquired or intrinsic therapeutic resistance remains a prevalent clinical issue that imposes substantial challenges on current management strategies\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Elucidating the intricate molecular mechanisms that underlie BRCA-associated therapeutic resistance is, therefore, of paramount importance for the development of novel, efficacious, and personalized therapeutic modalities that can either treat or circumvent such resistance in clinical practice.\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003ePyrotinib, a novel irreversible pan-ErbB tyrosine kinase inhibitor (TKI), exerts broad and potent antitumor effects in HER2-positive BRCA\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e,\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Previous studies demonstrated that Pyrotinib exerts robust antitumor efficacy, as evidenced by its capacity to induce more pronounced cell cycle arrest, suppress the proliferation of SKBR3 BRCA cells, and inhibit in vivo tumor growth in xenograft mouse models, a phenomenon that may be attributed to enhanced autophagy triggered by endoplasmic reticulum stress\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Nevertheless, the long-term administration of Pyrotinib confers a risk of acquired resistance, which has emerged as a pivotal factor contributing to therapeutic failure\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The lack of reliable biomarkers to predict resistance onset and limited understanding of the underlying molecular and spatial regulatory networks further hinder the optimization of clinical treatment strategies for BRCA patients receiving Pyrotinib.\u003c/p\u003e \u003cp\u003eIn recent years, the integration of single-cell and spatial transcriptomics has been applied in multiple cancer fields, advancing our understanding of cellular heterogeneity while contextualizing single-cell data in multicellular microenvironments, thus yielding valuable insights into tumor biology\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. The present study innovatively integrates multi-omics sequencing data, combined with comprehensive bioinformatics analysis and machine learning approaches, as well as a series of functional and regulatory validation methods. The purpose is to systematically screen Pyrotinib resistance-related genes in BRCA, clarify the molecular mechanisms and spatial regulatory characteristics underlying Pyrotinib resistance, construct a reliable prognostic model for predicting Pyrotinib treatment response and BRCA patient outcomes, identify potential targeted therapeutic agents, and provide novel technical support and theoretical basis for overcoming Pyrotinib resistance and optimizing personalized treatment strategies for BRCA.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Obtainment of BRCA datasets\u003c/h2\u003e \u003cp\u003eThe clinical and survival information of the TCGA-BRCA dataset cohort were acquired from the University of The Cancer Genome Atlas database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cancer.gov/ccg/research/genome-sequencing/tcga\u003c/span\u003e\u003cspan address=\"https://www.cancer.gov/ccg/research/genome-sequencing/tcga\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, which comprised 1,118 BRCA samples along with 137 control samples. The single-cell dataset GSE243275, whose corresponding platform is GPL24676, was harnessed for spatial transcriptome analysis, while GSE20685 (platform: GPL570) encompassing 327 BRCA samples and GSE86375 (platform: GPL6244) that consists of 124 BRCA samples as well as 35 adjacent non-cancerous samples were employed to validate the prognostic model and verify the expression levels of relevant genes, all of which were retrieved from the Gene Expression Omnibus (GEO) database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e\u003c/p\u003e \u003cp\u003eFor the establishment of a Pyrotinib-resistant progressive SKBR3 cell system and subsequent single-cell RNA sequencing (scRNA-seq; one sample per group), the SKBR3 cell line was divided into five groups: a blank control (P0) and four groups treated with 12 nM, 24 nM, 36 nM, or 48 nM Pyrotinib for 4, 8, 12, or 16 days (designated p4, p8, p12, p16, respectively), followed by single-cell suspension preparation (viability\u0026thinsp;\u0026gt;\u0026thinsp;90% via trypan blue staining), 10x Genomics library construction, and Illumina NovaSeq 6000 sequencing (paired-end 150 bp reads).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Processing of the scRNA-seq and GSE243275 dataset\u003c/h2\u003e \u003cp\u003eLeveraging two datasets, we first applied predefined thresholds to exclude low-quality cells; subsequently, the Seurat (v 5.0.1) package\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e was employed for both quality control and unsupervised clustering analysis. To identify genes exhibiting substantial expression heterogeneity across cells, which are referred to as highly variable genes, we normalized the transcriptomic data using the LogNormalize method, which prioritizes genes with relatively high coefficients of variation. Next, we performed principal component analysis (PCA) to achieve linear dimensionality reduction of the single-cell data, a process that assigned all cells to distinct principal components (PCs) based on weighted contributions. Following this dimensionality reduction step, the Louvain algorithm was used to cluster cells at multiple resolution levels, and the resulting clustering hierarchy was visualized via a clustering tree constructed with the clustree tool. We then further visualized the distribution and biological distances among different cell clusters using uniform manifold approximation and projection (UMAP), the scDblFinder algorithm was final utilized to eliminate all detected doublet artifacts\u003c/p\u003e \u003cp\u003eFor the scRNA-seq dataset, marker genes for each cell cluster were defined as the top 5 genes with significant expression differences (assessed via the Wilcoxon rank-sum test); these markers were thereafter subjected to Gene Ontology Biological Process (GO-BP) enrichment analysis. In parallel, in time-series analyses, genes with a membership value\u0026thinsp;\u0026ge;\u0026thinsp;0.7 were designated as Pyrotinib resistance-related genes. For the GSE243275 dataset specifically, we additionally conducted cell annotation via manual curation using public databases; we then applied the CRAD algorithm for spatial transcriptomic cell annotation, followed by the use of Cottrazm to identify tumor malignant regions, and ultimately further evaluated the functional activity of tumor-associated features.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Identification of Pyrotinib-related candidate genes in BRCA\u003c/h2\u003e \u003cp\u003eFirstly, for the GSE243275 dataset, we performed differential expression analysis between malignant and non-malignant regions using the Wilcoxon rank-sum test to obtain the first set of differentially expressed genes (DEGs1, [|log\u003csub\u003e2\u003c/sub\u003eFC| \u0026gt; 1, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05]). Next, we applied the DESeq2 (v 1.44.0) package\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e to conduct an additional round of differential expression analysis between tumor and control samples in the TCGA-BRCA cohort, which yielded a second set of differentially expressed genes (DEGs2, [|log\u003csub\u003e2\u003c/sub\u003eFC| \u0026gt; 1, FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05]). Furthermore, we implemented weighted gene co-expression network analysis (WGCNA) on tumor samples from TCGA-BRCA, with the preliminary step of excluding outlier samples to ensure data robustness. Subsequently, we used the pickSoftThreshold function and set the scale-free topology fitting index (R\u003csup\u003e2\u003c/sup\u003e)\u0026thinsp;\u0026gt;\u0026thinsp;0.85 to screen for soft-thresholding powers that exceeded the red cutoff line. We then constructed a clustered gene dendrogram to identify distinct co-expression modules, from which we extracted the modules showing the strongest positive and negative correlations with the target traits; the core genes within these modules were defined as BRCA-associated hub genes. A combined threshold of Gene Significance (GS)\u0026thinsp;\u0026gt;\u0026thinsp;0.25 and Module Membership (MM)\u0026thinsp;\u0026gt;\u0026thinsp;0.25 was applied to validate the reliability of the identified hub genes. Ultimately, we took the intersection of upregulated and downregulated genes among DEGs1, DEGs2, hub genes, and Pyrotinib resistance-related genes to obtain Pyrotinib-related candidate genes in BRCA. Using the ClusterProfiler package, based on the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases, the candidate genes were analyzed as a specific gene set and aPEAR conducted further clustering analysis on all the significantly enriched results.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Construction of a risk model\u003c/h2\u003e \u003cp\u003eFirstly, BRCA samples with complete survival information from TCGA database were randomly divided into a training set and a testing set at a ratio of 7:3. Subsequently, combined with the GSE20685 dataset, ten advanced machine learning algorithms, including gradient boosting machine (GBM), least absolute shrinkage and selection operator (Lasso), CoxBoost, partial least squares regression for Cox models (plsRcox), random survival forest (RSF), stepwise Cox regression (StepCox), supervised principal components (SuperPC), survival support vector machine (survival-SVM), Ridge regression, and elastic net (Enet), were employed to identify the prognostic model with the optimal predictive performance based on the candidate genes. The diagnostic value of the constructed prognostic model was then evaluated using receiver operating characteristic (ROC) curves.\u003c/p\u003e \u003cp\u003eAccording to the risk scores calculated based on the prognostic genes in the TCGA-BRCA cohort, BRCA samples were stratified into high-risk and low-risk groups using the median risk score as the cutoff value. Kaplan-Meier (KM) survival curves were plotted to compare the survival status between the two risk groups. Finally, the reliability of the risk model was further validated using ROC curve analysis based on the self-sequenced data of triple-negative BRCA samples included in this study, with a threshold of area under the curve (AUC)\u0026thinsp;\u0026gt;\u0026thinsp;0.6. Additionally, the predictive ability and generalization performance of the model were assessed through a meta-analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 The acquisition and validation of biomarkers\u003c/h2\u003e \u003cp\u003eUsing the glmnet (v 1.0) package\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e (with α\u0026thinsp;=\u0026thinsp;0), we constructed a Cox model, which allowed us to identify the corresponding biomarkers in BRCA after parsing the optimal model parameters, and we then validated the diagnostic efficacy of these biomarkers by generating ROC curves and performing inter-group differential expression analyses in both the TCGA-BRCA and GSE86374 datasets, while at the protein level we leveraged the Human Protein Atlas (HPA) database to conduct further validation via immunohistochemical assessments.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Clinical baseline analysis of biomarkers\u003c/h2\u003e \u003cp\u003eTo further assess the correlation between all biomarkers and common BRCA-related clinical features, all BRCA samples in TCGA-BRAC were stratified into high- and low-expression groups based on the median expression level of each biomarker, followed by the evaluation of inter-group differences in various clinical indicators. For numeric variables, data were presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation and compared between groups using the t-test; for categorical variables, frequencies and percentages were calculated and inter-group comparisons were performed using the chi-square test. After completing the baseline analysis, a baseline table was generated.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Function enrichment and regulatory network analysis of biomarkers\u003c/h2\u003e \u003cp\u003eTo explore the pathways highly correlated with each biomarker in BRCA, GO and KEGG enrichment analyses were performed on the high- and low-expression groups of each biomarker using the clusterProfiler package. The final results were filtered according to the following criteria: FDR\u0026thinsp;\u0026lt;\u0026thinsp;0.05, q-value\u0026thinsp;\u0026lt;\u0026thinsp;0.25, and absolute normalized enrichment score (|NES|)\u0026thinsp;\u0026gt;\u0026thinsp;1. Based on the enrichment results, we further integrated the outcomes of all biomarkers and identified their co-regulated pathways in BRCA by taking the intersection, which were visualized as heatmaps. In addition, to predict the interaction pairs at different levels, we leveraged the STRING database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://string-db.org/\u003c/span\u003e\u003cspan address=\"https://string-db.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e STRING and GeneMANIA database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://genemania.org/\u003c/span\u003e\u003cspan address=\"http://genemania.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e using the biomarkers and other candidate genes, followed by the construction of an integrated interaction network using Cytoscape. Subsequently, miRNAs targeting the biomarkers were predicted via the miRTarBase 2025 database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://mirtarbase.cuhk.edu.cn/~mirtarbase/mirtarbase_2025/\u003c/span\u003e\u003cspan address=\"https://mirtarbase.cuhk.edu.cn/~mirtarbase/mirtarbase_2025/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cb\u003eRelease 10.0\u003c/b\u003e). LncRNAs with high target-directed miRNA degradation scores that could target the key miRNAs were screened using the starBase database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://starbase.sysu.edu.cn/\u003c/span\u003e\u003cspan address=\"https://starbase.sysu.edu.cn/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, and key transcription factors (TFs) with high binding strength scores were identified via the JASPAR database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://jaspar.genereg.net/\u003c/span\u003e\u003cspan address=\"http://jaspar.genereg.net/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e Finally, an integrated regulatory network was constructed and visualized using Cytoscape.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8 The localization and molecular docking analysis of biomarkers\u003c/h2\u003e \u003cp\u003eTo further characterize the genomic and subcellular localization features of the identified biomarkers, we first performed chromosomal localization analysis using the RCircos (v 1.2.2) package\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, with annotation data derived from AnnoProbe (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/jmzeng1314/AnnoProbe\u003c/span\u003e\u003cspan address=\"https://github.com/jmzeng1314/AnnoProbe\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e and the UCSC.HG38.Human.CytoBandIdeogram dataset (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://genome.ucsc.edu/\u003c/span\u003e\u003cspan address=\"https://genome.ucsc.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, and visualized the results as circular chromosomal localization maps. For subcellular localization analysis of all biomarkers, we employed the RNALocate tool and visualized the corresponding distribution profiles to gain insights into their intracellular localization patterns. Additionally, candidate therapeutic agents targeting the biomarkers were predicted by retrieving drugs with a negative correlation (similarity score\u0026thinsp;\u0026lt;\u0026thinsp;0) from the L1000FWD database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://maayanlab.cloud/L1000FWD/\u003c/span\u003e\u003cspan address=\"https://maayanlab.cloud/L1000FWD/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e The small-molecule chemical structures of the top 10 candidate drugs were obtained from the PubChem database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pubchem.ncbi.nlm.nih.gov/\u003c/span\u003e\u003cspan address=\"https://pubchem.ncbi.nlm.nih.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e, while the three-dimensional (3D) macromolecular structures of all biomarkers were retrieved from the PDB database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.rcsb.org/\u003c/span\u003e\u003cspan address=\"https://www.rcsb.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e)\u003c/b\u003e and AlphaFoldDB (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://alphafold.ebi.ac.uk/\u003c/span\u003e\u003cspan address=\"https://alphafold.ebi.ac.uk/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cb\u003e).\u003c/b\u003e Molecular docking simulations were subsequently conducted using AutoDock Vina, and the docking results as well as force analysis were visualized in detailed 3D structural models via PyMOL.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e2.9 Cell communication and pseudotime analysis of spatial transcriptomics\u003c/h2\u003e \u003cp\u003eCell-cell communication was determined using the CellChat (v 1.5.0) package\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. A thicker connecting line between ligands and receptors indicates a stronger potential interaction between cells. In addition, spatial cell communication distribution maps (incoming and outgoing modes) and ligand-receptor co-expression distribution maps within pathways were generated for visualization.\u003c/p\u003e \u003cp\u003eBased on the spots corresponding to malignant regions in the spatial transcriptome data, we performed independent subclustering of malignant regions, which yielded a total of 5 distinct subclusters (cluster 0\u0026ndash;4). Subsequently, Monocle2 (v 2.24.0) package\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e was employed to conduct spatial pseudotime analysis. Furthermore, leveraging the pseudotime analysis results, we further investigated the dynamic expression patterns of the 4 biomarkers in tumor malignant tissues.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e2.10 Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)\u003c/h2\u003e \u003cp\u003eSKBR3 and BT474 human BRCA cell lines were selected for this study due to their inherent sensitivity to Pyrotinib\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. These mycoplasma-free, authenticated cell lines, obtained from the American Type Culture Collection (ATCC), were utilized as parental controls for the derivation of drug-resistant sublines\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Total RNA was isolated from 1\u0026times;10\u003csup\u003e6\u003c/sup\u003e cells per sample, with three biological replicates for each of the four groups: SKBR3 control, SKBR3 drug-resistant, BT474 control, and BT474 drug-resistant, using the Animal Total RNA Isolation Kit (Cat. No. RE-03014, Foregene). Then, 2 \u0026micro;g of total RNA was utilized to synthesize cDNA with SweScript RT I First Strand cDNA Synthesis Kit (Cat. No. G3330, Servicebio). Subsequently, PCR amplification was conducted in a 20-\u0026micro;L reaction system. PCR amplification was performed under the following conditions: an initial pre-denaturation step at 95 ℃ for 5 min, followed by 40 cycles of denaturation at 95 ℃ for 10 s and annealing/extension at 60 ℃ for 30 s. A melting curve analysis was then conducted by ramping the temperature from 60\u0026deg;C to 95 ℃ at a rate of 0.3 ℃ every 15 s. The CT values after amplification were harvested, and β-actin was employed as the internal reference gene. The results were evaluated by 2\u003csup\u003e\u0026minus;ΔΔCT\u003c/sup\u003e, and the primer sequences (5'\u0026ndash;3') utilized for RT-qPCR were as follows: β-Actin-F, AATCTGGCACCACACCTTCTACAA; β-actin-R, GGATAGCACAGCCTGGATAGACAA; HMGB3-F, CTGCCCAGACTAGCGAAACAA; HMGB3-R, GCCATCCTGACTGAATTGCTT; TFPI-F, GGTCGCGAATGGTTTCCAGGT; TFPI-R, AGCGCCATTCATTCCAACAT; ACTG2-F, GTGTGAAGAGGAGACACCACG; ACTG2-R, CAGATCTGGTGGCAGAGG; JCHAIN-F, GGGAGTCCTGGCGGTTTTTA; JCHAIN-R, TGGAAGTAATCCGGGCACAC. Finally, the expression differences of biomarkers were compared and presented graphically by means of GraphPad Prism 10 software (ANOVA-test, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.11 Western blot assay\u003c/h2\u003e \u003cp\u003eTumor cells were collected by centrifugation, and total protein was extracted by lysing 1\u0026times;10\u003csup\u003e6\u003c/sup\u003e cells in 250 \u0026micro;L of RIPA lysis buffer (Cat. No. BL504A, Biosharp). After centrifugation, protein concentrations were determined using a BCA Protein Assay Kit (Cat. No. BL521A, Biosharp). For sample preparation, total protein was mixed with 5\u0026times; SDS loading buffer at a 4:1 ratio, followed by denaturation at 95 ℃ for 5 min (or 70 ℃ for 5\u0026ndash;10 min for membrane proteins) and cooling on ice. For SDS-PAGE, glass plates were cleaned and assembled, separating and stacking gels were cast, and electrophoresis was performed at 80 V for the stacking gel and 120 V for the separating gel. For protein transfer, a PVDF membrane was activated with methanol, and a transfer sandwich was assembled before wet transfer at 300 mA for 30 min (or 200 mA for 1 h, or 25 V overnight). The membranes were then blocked with 5% non-fat milk in TBST for 1 h at room temperature, followed by overnight incubation at 4 ℃ with primary antibodies: anti-GAPDH (rabbit, 1:5000, Cat. No. AF7021, Affinity), anti-HMGB3 (rabbit, 1:1000, Cat. No. HA722033, HUABIO), anti-TFPI (rabbit, 1:1000, Cat. No. HA722199, HUABIO), anti-ACTG2 (rabbit, 1:1000, Cat. No. ER62586, HUABIO), and anti-JCHAIN (rabbit, 1:1000, Cat. No. HA721205, HUABIO). After washing, the membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. Protein expression levels were visualized using an enhanced chemiluminescence (ECL) reagent (Cat. No. BL520B, Biosharp).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e2.12 Statistical analysis\u003c/h2\u003e \u003cp\u003eR (v 4.4.3) language was utilized for all statistical analyses. Wilcoxon test and t-test were used to compare the differences between two groups. The value of p\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.1 740 Pyrotinib resistance-related genes were obtained in scRNA-seq dataset\u003c/h2\u003e \u003cp\u003eFollowing initial data screening, we quantified the changes in cell counts for the p0, p4, p8, p12, and p16 groups before and after quality control (\u003cb\u003eSupplementary Fig.\u0026nbsp;1A and Supplementary Table\u0026nbsp;1\u003c/b\u003e), identified the top 10 highly variable genes (\u003cb\u003eSupplementary Fig.\u0026nbsp;1B\u003c/b\u003e), extracted 7 PCs after dimensionality reduction (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA), and achieved the optimal clustering resolution of 0.3, which yielded 7 distinct cell clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB-\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eC). The comparison plot of doublet identification distributions reflected the genuine cellular heterogeneity of the samples (\u003cb\u003eSupplementary Fig.\u0026nbsp;1C\u003c/b\u003e). In addition, functional enrichment analysis of the marker genes for each cluster revealed that all cell clusters exhibited prominent spatial aggregation characteristics, with the enriched functional pathways being highly consistent with the spatial distribution of the corresponding clusters, thus demonstrating a clear correlation between cluster-specific biological functions and cellular clustering patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eD). Further inter-group differential abundance analysis indicated that the SKBR3_type0 subtype was the predominant cell type prior to Pyrotinib treatment; after treatment, the abundance of this subtype decreased significantly, with concomitant differentiation into other subtypes. Among these derivative subtypes, some displayed drug concentration-dependent sensitivity (e.g., type2 and type5), whereas others were associated with drug resistance (e.g., type1, type4, and type6), whose abundances peaked under the p12 treatment condition (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eE). Finally, time-series analysis categorized all genes into 12 clusters with distinct temporal expression trends and identified a total of 740 Pyrotinib resistance-related genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.2 A comprehensive single-cell dataset analysis in BRCA\u003c/h2\u003e \u003cp\u003eConsistent with the scRNA-seq dataset, dimensionality reduction of the GSE243275 dataset yielded 8 PCs, and clustering at the optimal resolution of 0.5 resulted in 16 distinct cell clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA-\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eC). After manual annotation, 7 major cell populations were identified: the abundance of drug resistance-related populations showed a gradual increase across treatment stages, while that of drug-sensitive populations exhibited a progressive decline (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD-\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE). Additionally, the spatial transcriptomics dataset achieved nearly 100% spot coverage, with all spots assigned to 11 clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF). CRAD results indicated that epithelial cells and fibroblasts were the dominant cell types in the spots (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG), and a strong positive correlation was observed between epithelial cells and fibroblasts (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH). Subsequently, we used Cottrazm to demarcate the malignant tumor region, tumor-normal tissue boundary region, and normal tissue region (\u003cb\u003eSupplementary Fig.\u0026nbsp;2\u003c/b\u003e), the key pathways involved in tumor-related features were further visualized (\u003cb\u003eSupplementary Fig.\u0026nbsp;3\u003c/b\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Identification of 39 Pyrotinib-related candidate genes\u003c/h2\u003e \u003cp\u003eA total of 5,628 DEGs1 were identified in the GSE243275 dataset, including 2,354 significantly upregulated genes and 3,274 significantly downregulated genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Similarly, 5,347 significantly upregulated genes and 2,867 significantly downregulated genes were obtained in the TCGA-BRCA cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Subsequent WGCNA showed no outlier samples (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC), and the optimal soft threshold was determined to be 6 (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). This analysis yielded 9 co-expression modules and 4,616 hub genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE-\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF). Notably, the GS and MM values of most hub genes were above 0.25, indicating good correlations between hub genes and both traits and modules (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eG). Finally, Venn diagram analysis identified 39 Pyrotinib-related candidate genes as the intersection of DEGs, including 17 tumor-upregulated and 22 tumor-downregulated genes (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eH). A total of 39 Pyrotinib-associated candidate genes were subjected to enrichment analysis, which yielded 7 KEGG pathways, 113 biological process terms, 43 cellular component terms, and 30 molecular function terms, such as critical biological processes such as regulation of apoptotic process, central nervous system development, and neural precursor cell proliferation (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eI-\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eJ).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Prognostic risk model was construct in TCGA-BRCA\u003c/h2\u003e \u003cp\u003eAfter C-index validation and evaluation, the StepCox [forward] + Ridge model achieved the highest evaluation values among all models: 0.65002337267734 in the training set, 0.64594127806563 in the testing set, and 0.612130696637739 in the GSE20685 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). Under the ROC criterion, the AUC value of this model was greater than or equal to 0.6, thus it was identified as the optimal model (\u003cb\u003eSupplementary Fig.\u0026nbsp;4\u003c/b\u003e). Subsequent KM curves showed that the optimal model achieved good discriminability in the grouping of all three datasets; the survival curve of the high-risk group was significantly lower than that of the low-risk group with no obvious overlap, indicating favorable validation results (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB-\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). Time-ROC curve analysis in triple-negative BRCA self-sequencing data demonstrated that the optimal model still had certain diagnostic value in this dataset, with good predictive accuracy at 2 and 3 years (AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.6) and Meta-analysis also showed that the optimal prognostic model exhibited excellent predictive performance across different datasets (HR\u0026thinsp;\u0026gt;\u0026thinsp;1), which further confirmed its good stability and effectiveness (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eE-\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e3.5 HMGB3, TFPI, ACTG2, and JCHAIN were biomarkers in BRCA\u003c/h2\u003e \u003cp\u003eUsing the glmnet package, we identified that the StepCox [forward] + Ridge model consists of HMGB3, TFPI, ACTG2, and JCHAIN as features. Within the established optimal prognostic model, HMGB3 and TFPI were assigned positive coefficients (0.1348 and 0.1244, respectively), indicating their higher expression levels of these genes correlate with elevated risk scores, whereas ACTG2 and JCHAIN carried negative coefficients (-0.0730 and \u0026minus;\u0026thinsp;0.1085, respectively), consistent with their role as protective factors (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Importantly, these four genes showed high diagnostic efficacy (AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.7) and consistently significant differential expression trends in both TCGA-BRCA and GSE86374 datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB\u0026ndash;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eE). Similarly, the protein expression trends of the four genes at the immunohistochemical level were consistent with the conclusions obtained from Wilcoxon test (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eF). In summary, we defined HMGB3, TFPI, ACTG2, and JCHAIN as the key genes in this study.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe distribution of characteristic genes and coefficients of the optimal prognosis model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003egene\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecoef\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHMGB3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.134881202254666\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTFPI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.124479160815218\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eACTG2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.073082741\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJCHAIN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.108561139\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Analysis of the functions and regulatory characteristics of biomarkers\u003c/h2\u003e \u003cp\u003eClinical baseline profiles demonstrated that all the identified biomarkers exhibited certain prognostic phenotypic correlations at multiple levels (\u003cb\u003eSupplementary Table\u0026nbsp;2\u003c/b\u003e). Gene set enrichment analysis results revealed that TFPI was enriched in the largest number of pathways among all the biomarkers (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In BRCA, all the candidate biomarkers were jointly correlated with 24 pathways, among which the immunoglobulin complex pathway showed the strongest correlation (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA-\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGSEA enrichment results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003egene\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKEGG\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGO_BP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGO_CC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGO_MF\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTFPI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e120\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1313\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e154\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e194\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHMGB3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e593\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e129\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJCHAIN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e752\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eACTG2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e938\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e109\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eProtein-protein interaction (PPI) network analysis identified three distinct functional modules centered on the core biomarkers. Specifically, the module anchored by ACTG3 clustered associated proteins including MEOX1, FLNC and SVEP1, which were mainly linked via co-expression and co-localization interactions. The sub-module with TFPI as the hub showed direct protein interactions with CENPA, DPP4 and MAF. Meanwhile, the module centered on HMGB3 incorporated RFC2/RFC4/RFC5, NUPR5 and other functional proteins such as GGT and TK1, which were connected through genetic interactions and shared protein domains. These findings collectively suggested that the candidate biomarkers might exert downstream regulatory functions via intricate protein interaction networks (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eC).\u003c/p\u003e \u003cp\u003eFurthermore, upstream regulatory molecule prediction was performed using databases including miRTarBase and starBase. A total of 84 miRNAs targeting the core biomarkers, 61 lncRNAs binding to these miRNAs, and 32 key TFs were screened out. The constructed regulatory network visually illustrated the potential upstream regulatory molecules of the core biomarkers and their interaction patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eD).\u003c/p\u003e \u003cp\u003eChromosomal localization analysis indicated that HMGB3 was mapped to the X chromosome, JCHAIN to chromosome 4, and both ACTG2 and TFPI to chromosome 2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eE). Subcellular localization analysis further showed that TFPI was predominantly distributed in the extracellular space and on the cell membrane, JCHAIN was concentrated in the extracellular region, ACTG2 was mainly located in the cytoplasm, and HMGB3 was primarily localized in the nucleus (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eF).\u003c/p\u003e \u003cp\u003eThe final molecular docking results showed that ABT-737 had the strongest binding ability to all four biomarkers simultaneously, making it the most promising targeted drug (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eG). The detailed three-dimensional structure diagram is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eH.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e3.7 Spatial transcriptomic analysis of cell communication and malignant tissue dynamics\u003c/h2\u003e \u003cp\u003eUsing spatial transcriptomic data, 186 cell communication pathways were identified via CellChat. The communication network revealed Fibroblast as the core node, forming extensive connections with T cells, epithelial cells, and other populations (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA). For the 2-AG signaling pathway, T/epithelial cells acted as signal senders, while Fibroblast served as the key receiver, with ligand-receptor co-expression verifying this spatial interaction (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eMalignant-region spots were clustered into 5 subclusters (cluster 0\u0026ndash;4) via Monocle2, and trajectory analysis showed continuous dynamic transitions among clusters (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eC). Temporal tracing of 4 biomarkers indicated heterogeneous expression trends across pseudotime: ACTG2/HMGB3 first decreased then increased, while TFPI/JCHAIN gradually declined, reflecting biomarker expression dynamics during malignant progression (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e3.8 Upregulation of JCHAIN, HMGB3, ACTG2, and TFPI in Pyrotinib-resistant BRCA cells\u003c/h2\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, both RT-qPCR and Western blot analyses demonstrated a consistent and significant upregulation of JCHAIN, HMGB3, ACTG2, and TFPI in Pyrotinib-resistant BRCA cell models. RT-qPCR (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA\u0026ndash;D) revealed that the mRNA expression levels of these four genes were markedly elevated in the drug-resistant groups of both SKBR3 and BT474 cell lines compared with their respective control groups (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). In line with the transcriptional data, Western blot assays (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eE) and subsequent densitometric quantification (Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eF) confirmed that the protein expression levels of JCHAIN, HMGB3, ACTG2, and TFPI were also significantly increased in the resistance group relative to the control group (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), validating the concordant upregulation of these genes at both the mRNA and protein levels in Pyrotinib-resistant cells.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eBRCA remains a leading contributor to cancer-associated morbidity and mortality worldwide\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. While Pyrotinib exerts robust antitumor activity against BRCA, long-term administration often culminates in the emergence of acquired resistance, which severely compromises the efficacy of clinical interventions\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. This is the first integrative analysis of single-cell, spatial (GSE243275) and bulk RNA-seq data in Pyrotinib-resistant SKBR3 cells, which identifies 740 resistance-related genes and narrows them down to 39 candidates. Through the comparison and validation of ten machine learning algorithms, we constructed a four-gene prognostic model (StepCox[forward]+Ridge) comprising HMGB3, TFPI, ACTG2, and JCHAIN. The identified biomarkers were validated at both mRNA and protein levels, found to participate in immune-related pathways with defined chromosomal and subcellular localizations. Spatial transcriptomic analysis revealed a fibroblast-centered cell communication network and the dynamic expression patterns of these biomarkers during malignant progression. Furthermore, molecular docking identified ABT-737 as a potential targeted agent, providing a theoretical basis for reversing Pyrotinib resistance. At last, Both RT-qPCR and Western blot analyses revealed that JCHAIN, HMGB3, ACTG2, and TFPI were significantly upregulated at both mRNA and protein levels in Pyrotinib-resistant BRCA cells.\u003c/p\u003e \u003cp\u003eCompared to previously published prognostic models for BRCA, our four-gene model demonstrates competitive discriminative performance across multiple independent cohorts. For instance, in the pivotal TCGA-BRCA training set, our model achieved a C-index of 0.65, which is comparable to, if not superior than, the performance (C-index typically ranging 0.60\u0026ndash;0.80) of widely used multi-gene prognostic tools (such as certain immune-related or proliferation-related signatures) reported in similar cohorts\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. More importantly, our model is specifically constructed for the context of Pyrotinib resistance. Its core genes (HMGB3, TFPI, ACTG2, JCHAIN) not only show significant associations with overall patient survival but also maintain stable AUC values above 0.6 in the GSE20685 and internal validation cohorts. This indicates its reliable and reproducible predictive capability for identifying high-risk patients with poor response to Pyrotinib treatment. In contrast, many general-purpose BRCA prognostic models are not optimized for resistance to specific targeted agents, particularly Pyrotinib, which may limit their predictive accuracy in relevant subgroups\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e,\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Therefore, our model is not intended to replace existing generic classifiers but serves as a crucial complementary tool. It provides the first specifically dedicated assessment framework, systematically validated by multi-omics data and machine learning algorithms, to aid clinicians in evaluating the prognosis and potential resistance risk for HER2-positive BRCA patients undergoing Pyrotinib therapy.\u003c/p\u003e \u003cp\u003eThe four core biomarkers (HMGB3, TFPI, ACTG2, JCHAIN) uncovered in this study were validated at both mRNA and protein levels and are closely linked to key biological processes in BRCA. Among them, HMGB3 (high mobility group protein B3) exhibits high expression in stem cells and cancer cells while showing minimal transactivation in normal adult tissues, thereby emerging as a promising therapeutic target; notably, it is aberrantly overexpressed and plays a pivotal role in the malignant progression of multiple cancer types\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. This finding is consistent with the result of the present study that HMGB3 is defined as a risk factor for BRCA. Previous studies have reported that HMGB3 promotes resistance to anti-PD-1 therapy in triple-negative BRCA by interfering with the IFN-γ-mediated ferroptosis pathway, which suggests its potential involvement in the modulation of the tumor immune microenvironment and highlights its value for the development of combination therapeutic strategies\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. TFPI, functioning as a tumor suppressor gene, can inhibit BRCA cell proliferation and invasion by suppressing the ERK/p38 MAPK signaling pathway\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. Our study further confirms that its low expression is associated with poor prognosis, supporting its dual role as both a prognostic marker and a therapeutic target. ACTG2 is associated with cytoskeletal remodeling and invasive phenotypes, and its overexpression may drive malignant progression and metastasis in BRCA\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. However, in both Wang et al.\u0026rsquo;s BRCA study and our current work, ACTG2 expression is significantly downregulated in tumor tissues\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. This apparent discrepancy can be attributed to the context-dependent functional role of ACTG2, which is contingent on tumor molecular subtypes and study cohort characteristics: First, the study\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e reporting pro-oncogenic effects of ACTG2 overexpression likely focused on specific cell line models or cohorts of late-stage, highly aggressive BRCA. In this setting, elevated ACTG2 expression drives metastatic phenotypes by modulating the RhoA/ROCK signaling pathway to promote actin polymerization, thereby enhancing cytoskeletal remodeling and cell migratory capacity. In contrast, our cohort (as well as Wang et al.\u0026rsquo;s) may be enriched in early-stage cases or subtype tumors, which contexts in which reduced ACTG2 expression confers a protective effect: the downregulation of ACTG2 attenuates aberrant cytoskeletal rearrangement, limiting uncontrolled tumor cell proliferation and invasion. This discrepancy does not represent a contradiction but rather reflects the functional complexity of ACTG2 in BRCA. Its pro-oncogenic or tumor-suppressive role is dependent on tumor stage, molecular subtype, and the gene interaction network, thereby providing a novel molecular basis for the stratified treatment of BRCA. More importantly, JCHAIN has been identified as a key gene with prognostic value in BRCA, which constitutes the core finding of a recent study by Shi et al.\u003csup\u003e34\u003c/sup\u003e. Consistent with the key results of our current research, the sustained downregulation of JCHAIN mRNA expression in tumor samples relative to normal counterparts, coupled with its associations with TNM stage and BRCA subtypes, strongly indicates that JCHAIN is intricately involved in the biological processes driving BRCA progression. Notably, these genes are collectively enriched in pathways such as apoptotic regulation, central nervous system development, and neural precursor cell proliferation, suggesting that Pyrotinib resistance may involve complex biological processes including cell fate remodeling and microenvironmental adaptation.\u003c/p\u003e \u003cp\u003eOur spatial transcriptomic analysis via CellChat identified 186 cell communication pathways, with Fibroblasts emerging as the central hub that establishes extensive crosstalk with T cells, epithelial cells, and other cellular components in the BRCA TME\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. Such a central hub role implies that Fibroblasts may integrate signals from multiple cell types (e.g., pro-inflammatory cues from T cells, oncogenic signals from epithelial cells) and transduce downstream effects by secreting cytokines, remodeling the extracellular matrix (ECM), or regulating metabolic pathways, which consistent with previous reports that cancer-associated Fibroblasts (CAFs) drive BRCA malignancy through paracrine signaling and ECM reshaping\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Notably, the 2-arachidonoylglycerol (2-AG) signaling pathway displayed a distinct cell communication pattern. As a major endocannabinoid, 2-AG is known to regulate immune responses, cell proliferation, and ECM remodeling in cancers, but its cell-type-specific crosstalk in BRCA TME remains undercharacterized. Our findings uncover a novel regulatory axis: tumor-infiltrating T cells may secrete 2-AG-related ligands to modulate Fibroblast activation, extending previous observations that endocannabinoids suppress T cell function by highlighting this indirect anti-tumor immunity-weakening crosstalk\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e,\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Meanwhile, cancerous epithelial cells may secrete 2-AG to instruct Fibroblasts to support tumor growth, consistent with reports that epithelial-stromal crosstalk drives CAF activation via lipid signaling\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. Collectively, the 2-AG axis bridges immune, tumor, and stromal cells, coordinating TME remodeling to favor BRCA progression. Additionally, the dynamic expression changes of the biomarkers along the pseudotime trajectory within malignant regions further corroborate their regulatory roles in tumor evolution.\u003c/p\u003e \u003cp\u003eNotably, the expression patterns of JCHAIN, HMGB3, ACTG2, and TFPI observed in our Pyrotinib-resistant cell models appear to differ from those reported in previous transcriptomic analyses, where JCHAIN, ACTG2, and TFPI are frequently downregulated in primary breast tumors compared with normal tissues. This apparent discrepancy may reflect fundamental differences in the biological contexts interrogated: large-scale bioinformatic analyses primarily capture gene expression signatures associated with tumor initiation and progression, which are broadly represented in heterogeneous tissue samples, whereas our in vitro model specifically isolates the adaptive response of SKBR3 and BT474 human breast cancer cells to prolonged Pyrotinib exposure. It is plausible that these genes, which may act as tumor suppressors in the context of primary oncogenesis, are co-opted to promote cell survival and drug resistance under therapeutic pressure. Alternatively, the discrepancy could arise from the absence of a complex tumor microenvironment in our cell culture system, which is known to profoundly shape gene expression in vivo. Collectively, these findings underscore the importance of validating bioinformatic predictions in mechanistic models and highlight the context-dependent roles of these biomarkers in BRCA biology.\u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThus, the four-gene prognostic model constructed in our study provides a reliable tool for assessing Pyrotinib treatment response and prognosis in BRCA patients. Simultaneously, it elucidates resistance-related molecular mechanisms and microenvironmental features, offering new targets and theoretical support for developing individualized treatment strategies, thereby holding significant value for advancing precision medicine in BRCA. This study also has certain limitations: all experiments were performed only in in vitro cell models, and in vivo validation as well as large‑scale clinical cohort studies have not been conducted. Furthermore, the expression and function of these key genes in clinical tumor samples and the potential impact of the tumor microenvironment remain to be elucidated. Future research should employ clinical specimens and in vivo models to verify the expression patterns and functional roles of these genes in pyrotinib resistance. In addition, further studies are warranted to explore the synergistic effects of ABT‑737 combined with pyrotinib, so as to promote the clinical translation of these findings.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eBRCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBreast cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCAFs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCancer-associated Fibroblasts\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eECM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExtracellular matrix\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGene Ontology\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGene Significance\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKEGG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eKyoto Encyclopedia of Genes and Genomes\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eKaplan-Meier\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eModule Membership\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePCA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePrincipal component analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePCs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePrincipal components\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver operating characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003escRNA-seq\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSingle-cell RNA sequencing\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTKI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTyrosine kinase inhibitor\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTFs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTranscription factors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eWGCNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eWeighted gene co-expression network analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was also supported by Health Commission of Sichuan Province Medical Science and Technology Program (Grant No. 24QNMP051) and the \u0026ldquo;Qimingxing\u0026rdquo; Research Fund for Young Talents of West China Hospital (HXOMX0082).\u003c/p\u003e\u003ch2\u003eAuthor\u0026rsquo;s contributions\u003c/h2\u003e \u003cp\u003eAll authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Qiheng Gou, Xueming Xia and Yuxin Xie. The first draft of the manuscript was written by Qiheng Gou and Yuxin Xi. All authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eWe would like to express our sincere gratitude to all individuals and organizations who supported and assisted us throughout this research.\u003c/p\u003e\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e \u003cp\u003eThe data used in this study were extracted from public databases.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSiegel RL, Kratzer TB, Cancer statistics (2026) Jan-Feb 2026;76(1):e70043. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3322/caac.70043\u003c/span\u003e\u003cspan address=\"10.3322/caac.70043\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiong X, Zheng LW, Ding Y et al (2025) Breast cancer: pathogenesis and treatments. Feb 19(1):49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-024-02108-4\u003c/span\u003e\u003cspan address=\"10.1038/s41392-024-02108-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBray F, Laversanne M, Sung H (2024) Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. May-Jun 74(3):229\u0026ndash;263. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3322/caac.21834\u003c/span\u003e\u003cspan address=\"10.3322/caac.21834\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu Y, Gong M, Wang Y, Yang Y, Liu S, Zeng Q (2023) Global trends and forecasts of breast cancer incidence and deaths. Sci Data May 27(1):334. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41597-023-02253-5\u003c/span\u003e\u003cspan address=\"10.1038/s41597-023-02253-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiong X, Zheng LW, Ding Y et al (2025) Breast cancer: pathogenesis and treatments. Signal Transduct Target Ther Feb 19(1):49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-024-02108-4\u003c/span\u003e\u003cspan address=\"10.1038/s41392-024-02108-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLei S, Zheng R, Zhang S et al (2021) Global patterns of breast cancer incidence and mortality: A population-based cancer registry data analysis from 2000 to 2020. Cancer Commun (Lond) Nov 41(11):1183\u0026ndash;1194. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/cac2.12207\u003c/span\u003e\u003cspan address=\"10.1002/cac2.12207\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu L, Xie Y, Gou Q et al (2024) HER2-targeted therapies for HER2-positive early-stage breast cancer: present and future. Front Pharmacol 15:1446414. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fphar.2024.1446414\u003c/span\u003e\u003cspan address=\"10.3389/fphar.2024.1446414\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi X, Zhang X, Yin S, Nie J (2025) Challenges and prospects in HER2-positive breast cancer-targeted therapy. Critical reviews oncology/hematology Mar 207:104624. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.critrevonc.2025.104624\u003c/span\u003e\u003cspan address=\"10.1016/j.critrevonc.2025.104624\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVeeraraghavan J, De Angelis C, Gutierrez C et al (2025) HER2-Positive Breast Cancer Treatment and Resistance. Adv Exp Med Biol 1464:495\u0026ndash;525. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-031-70875-6_24\u003c/span\u003e\u003cspan address=\"10.1007/978-3-031-70875-6_24\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlair HA, Pyrotinib (2018) First Global Approval. Drugs Nov 78(16):1751\u0026ndash;1755. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40265-018-0997-0\u003c/span\u003e\u003cspan address=\"10.1007/s40265-018-0997-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQi X, Shi Q, Xuhong J, Zhang Y, Jiang J (2023) Pyrotinib-based therapeutic approaches for HER2-positive breast cancer: the time is now. Breast cancer research: BCR Oct 3(1):113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13058-023-01694-5\u003c/span\u003e\u003cspan address=\"10.1186/s13058-023-01694-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu X, Zhang X, Shao Z et al (2023) Pyrotinib and chrysin synergistically potentiate autophagy in HER2-positive breast cancer. Signal Transduct Target Ther Dec 18(1):463. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-023-01689-w\u003c/span\u003e\u003cspan address=\"10.1038/s41392-023-01689-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Li Y, Jin R et al (2023) Single-cell analysis reveals the potential mechanisms of pyrotinib resistance in non-small cell lung cancer. Jan 13(1):17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-022-01226-1\u003c/span\u003e\u003cspan address=\"10.1038/s41392-022-01226-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan J, Slowikowski K, Zhang F (2020) Single-cell transcriptomics in cancer: computational challenges and opportunities. Exp Mol Med Sep 52(9):1452\u0026ndash;1465. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s12276-020-0422-0\u003c/span\u003e\u003cspan address=\"10.1038/s12276-020-0422-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGribov A, Sill M, L\u0026uuml;ck S et al (2010) SEURAT: visual analytics for the integrated analysis of microarray data. BMC Med Genom 3(1):21\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLove M, Anders S, Huber W (2014) Differential analysis of count data\u0026ndash;the DESeq2 package. Genome Biol 15(550):10\u0026ndash;1186\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFriedman J, Hastie T, Tibshirani R et al (2021) Package \u0026lsquo;glmnet\u0026rsquo;. CRAN R Repositary 595:874\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang H, Meltzer P, Davis S (2013) RCircos: an R package for Circos 2D track plots. BMC Bioinformatics 14(1):244\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin S, Plikus MV, Nie Q (2025) CellChat for systematic analysis of cell\u0026ndash;cell communication from single-cell transcriptomics. Nat Protoc 20(1):180\u0026ndash;219\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrapnell C, Cacchiarelli D, Qiu X, Monocle (2017) Cell counting, differential expression, and trajectory analysis for single-cell RNA-Seq experiments. Bioconductor\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKokot A, Gadakh S, Saha I et al (2024) Unveiling the Molecular Mechanism of Trastuzumab Resistance in SKBR3 and BT474 Cell Lines for HER2 Positive Breast Cancer. Curr Issues Mol Biol Mar 21(3):2713\u0026ndash;2740. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/cimb46030171\u003c/span\u003e\u003cspan address=\"10.3390/cimb46030171\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu S, Lan B, Wang Y et al (2024) Pyrotinib and trastuzumab combination treatment synergistically overcomes HER2 dependency in HER2-positive breast cancer: insights from the PHILA trial. EBioMedicine Nov 109:105379. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ebiom.2024.105379\u003c/span\u003e\u003cspan address=\"10.1016/j.ebiom.2024.105379\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMu\u0026ntilde;ante B, Paz-Manrique R, Pinto JA, Gomez HL (2025) Clinical Management in BRCA Carriers with Early Breast Cancer. Cancer Control Jan-Dec 32:10732748251377864. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/10732748251377864\u003c/span\u003e\u003cspan address=\"10.1177/10732748251377864\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Li Y, Jin R et al (2023) Single-cell analysis reveals the potential mechanisms of pyrotinib resistance in non-small cell lung cancer. Signal Transduct Target Ther Jan 13(1):17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-022-01226-1\u003c/span\u003e\u003cspan address=\"10.1038/s41392-022-01226-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCui F, Yan C, Wu J et al (2025) Identification and validation of prognostic genes associated with T-cell exhaustion and macrophage polarization in breast cancer. Front Endocrinol (Lausanne) 16:1556496. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fendo.2025.1556496\u003c/span\u003e\u003cspan address=\"10.3389/fendo.2025.1556496\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu B, Yan M, Ma F et al (2021) Pyrotinib plus capecitabine versus lapatinib plus capecitabine for the treatment of HER2-positive metastatic breast cancer (PHOEBE): a multicentre, open-label, randomised, controlled, phase 3 trial. Lancet Oncol Mar 22(3):351\u0026ndash;360. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s1470-2045(20)30702-6\u003c/span\u003e\u003cspan address=\"10.1016/s1470-2045(20)30702-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang F, Lai J, Zhuo X et al (2023) HER2-positive breast cancer progresses rapidly after pyrotinib resistance: acquired RET gene fusion and TP53 gene mutation are potential reasons. Anticancer Drugs Nov 1(10):1196\u0026ndash;1201. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/cad.0000000000001506\u003c/span\u003e\u003cspan address=\"10.1097/cad.0000000000001506\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWen B, Wei YT, Zhao K (2021) The role of high mobility group protein B3 (HMGB3) in tumor proliferation and drug resistance. Mol Cell Biochem Apr 476(4):1729\u0026ndash;1739. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11010-020-04015-y\u003c/span\u003e\u003cspan address=\"10.1007/s11010-020-04015-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNemeth MJ, Kirby MR, Bodine DM (2006) Hmgb3 regulates the balance between hematopoietic stem cell self-renewal and differentiation. Proc Natl Acad Sci U S A Sep 12(37):13783\u0026ndash;13788. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1073/pnas.0604006103\u003c/span\u003e\u003cspan address=\"10.1073/pnas.0604006103\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuo B, Zheng H, Liang G, Luo Y, Zhang Q, Li X (Mar 2025) HMGB3 Contributes to Anti-PD-1 Resistance by Inhibiting IFN-γ-Driven Ferroptosis in TNBC. \u003cem\u003eMolecular carcinogenesis\u003c/em\u003e. 64(3):490\u0026ndash;501. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/mc.23861\u003c/span\u003e\u003cspan address=\"10.1002/mc.23861\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXing M, Yang Y, Huang J et al (2022) TFPI inhibits breast cancer progression by suppressing ERK/p38 MAPK signaling pathway. Jul 44(7):801\u0026ndash;812. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s13258-022-01258-5\u003c/span\u003e\u003cspan address=\"10.1007/s13258-022-01258-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Y, Liu ZG, Shi MQ et al (2017) Identification of ACTG2 functions as a promoter gene in hepatocellular carcinoma cells migration and tumor metastasis. Biochemical Biophys Res communications Sep 16(2):537\u0026ndash;544. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bbrc.2017.04.007\u003c/span\u003e\u003cspan address=\"10.1016/j.bbrc.2017.04.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Li TE, Chen M, Pan JJ, Shen KW (2020) miR-106b-5p contributes to the lung metastasis of breast cancer via targeting CNN1 and regulating Rho/ROCK1 pathway. Aging (Albany NY) Jan 27(2):1867\u0026ndash;1887. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.18632/aging.102719\u003c/span\u003e\u003cspan address=\"10.18632/aging.102719\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi Y, Lin L, Zhu X et al (2025) Comprehensive Analysis of JCHAIN as a Potential Prognostic Factor for Breast Cancer and an Indicator for Tumor Microenvironment. Biomedicines Sep 26(10). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/biomedicines13102366\u003c/span\u003e\u003cspan address=\"10.3390/biomedicines13102366\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Visser KE, Joyce JA (2023) The evolving tumor microenvironment: From cancer initiation to metastatic outgrowth. Cancer cell Mar 13(3):374\u0026ndash;403. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ccell.2023.02.016\u003c/span\u003e\u003cspan address=\"10.1016/j.ccell.2023.02.016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu D, Li Z, Zheng B et al (May 2022) Cancer-associated fibroblasts in breast cancer: Challenges and opportunities. 42(5):401\u0026ndash;434. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/cac2.12291\u003c/span\u003e\u003cspan address=\"10.1002/cac2.12291\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Wang L, Ma W, Wu J, Wu Q, Sun C (2025) Paracrine signaling in cancer-associated fibroblasts: central regulators of the tumor immune microenvironment. J Transl Med Jun 23(1):697. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12967-025-06744-4\u003c/span\u003e\u003cspan address=\"10.1186/s12967-025-06744-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiong X, Chen S, Shen J et al (2022) Cannabis suppresses antitumor immunity by inhibiting JAK/STAT signaling in T cells through CNR2. Signal Transduct Target Ther Apr 6(1):99. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41392-022-00918-y\u003c/span\u003e\u003cspan address=\"10.1038/s41392-022-00918-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSido JM, Nagarkatti PS, Nagarkatti M (2016) Production of endocannabinoids by activated T cells and B cells modulates inflammation associated with delayed-type hypersensitivity. Eur J Immunol Jun 46(6):1472\u0026ndash;1479. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/eji.201546181\u003c/span\u003e\u003cspan address=\"10.1002/eji.201546181\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrauze I, Greb-Markiewicz B, Kłopot A, Maciejewska K, Bryk M, Krzystek-Korpacka M (2025) Neutrophil extracellular traps and cannabinoids: potential in cancer metastasis. Front Oncol 15:1595913. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2025.1595913\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2025.1595913\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"West China Hospital of Sichuan University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"breast cancer, pyrotinib, resistance, spatial transcriptome, prognostic model","lastPublishedDoi":"10.21203/rs.3.rs-9376197/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9376197/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePyrotinib resistance remains a major challenge in the treatment of breast cancer (BRCA), highlighting the need for reliable biomarkers and prognostic models. This study aimed to identify Pyrotinib resistance-related biomarkers and explore their regulatory mechanisms and therapeutic potential. We integrated single-cell RNA sequencing (scRNA-seq) of Pyrotinib-resistant SKBR3 cells, spatial transcriptomics (GSE243275), and bulk RNA-seq (TCGA-BRCA, GSE20685, GSE86374). Differential expression analysis, weighted gene co-expression network analysis (WGCNA), and 10 machine learning algorithms were used to screen candidate genes and construct a prognostic model. Functional enrichment, regulatory network analysis, molecular docking, cell communication, pseudotime trajectory analyses and in vitro experiment were performed for validation. A total of 740 Pyrotinib resistance-related genes and 39 candidate genes were identified. The StepCox [forward] + Ridge model, consisting of HMGB3, TFPI, ACTG2, and JCHAIN, exhibited robust prognostic performance (C-index: 0.61\u0026ndash;0.65; AUC\u0026thinsp;\u0026ge;\u0026thinsp;0.6 across datasets), with high-risk patients showing poorer survival. These genes were validated at the mRNA and protein levels, participated in immune-related pathways, and had distinct chromosomal/subcellular localizations. ABT-737 was identified as a potential targeted drug via molecular docking. Spatial transcriptomics revealed fibroblast-centered cell communication and dynamic biomarker expression during malignant progression. Lastly, JCHAIN, HMGB3, ACTG2, and TFPI were significantly upregulated at both mRNA and protein levels in Pyrotinib-resistant BRCA cells compared with parental control cells. The four-gene model serves as a reliable prognostic tool for Pyrotinib response and BRCA outcomes, providing novel insights into resistance mechanisms and precision therapy strategies.\u003c/p\u003e","manuscriptTitle":"Integrative analysis of scRNA-seq, spatial transcriptomics, and machine learning constructs a prognostic model for Pyrotinib resistance in breast cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-14 10:37:30","doi":"10.21203/rs.3.rs-9376197/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a8e71906-245e-4ede-9b96-d47aa36c7b2a","owner":[],"postedDate":"April 14th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-14T10:37:30+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-14 10:37:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9376197","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9376197","identity":"rs-9376197","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}