Identification of lncRNAs in Breast Cancer Through Bioinformatics Analysis: Implications for Pathogenesis and Prognostic Significance Authors: Yuhan Zheng, Xinggui Chen, Xuan Liu, Yijin Zhou, Pei Yang, Shuhui Li Published: 2026 ABSTRACT ABSTRACTObjectiveBreast cancer (BC) is a disease with a subtle onset, often diagnosed late, and has a high mortality rate. This study aimed to identify long non‐coding RNAs (lncRNAs) linked to BC as prognostic markers using bioinformatics and initial experiments, explore their regulatory pathways, and assess their expression and prognostic significance.MethodsTranscriptome data and clinical details of a BC cohort were obtained from The Cancer Genome Atlas and Gene Expression Omnibus. Bioinformatics methods, including Weighted Gene Co‐expression Network Analysis (WGCNA), Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Gene Set Enrichment Analysis (GSEA), and survival analysis, were used to evaluate lncRNAs, related microRNAs (miRNAs), and target genes. The competing endogenous RNA (ceRNA)‐based regulatory axis of lncRNA‐miRNA‐messenger RNA (mRNA) was identified, and gene expression on this axis in normal human mammary epithelial and BC cell lines was analyzed using quantitative reverse transcription polymerase chain reaction (qRT‐PCR).ResultsWe screened the lncRNA modules most associated with BC traits through bioinformatics analysis and analyzed the immune cell infiltration level of BC samples. Active cluster of differentiation 4+ (CD4+) memory T cells were more infiltrated in BC, whereas resting mast cells were less infiltrated in controls. WGCNA revealed 14 lncRNA modules linked to immune cells and 277 modules associated with BC and immune traits. Univariate Cox analysis identified 12 survival‐related target mRNAs mainly involved in immune processes and cell proliferation. We constructed a putative BC‐related ceRNA network, specifically the long intergenic non‐protein coding RNA 200 (LINC00200)‐microRNA 20b (miR‐20b‐5p)‐high mobility group box 3 (HMGB3) axis, through the identification of survival‐associated lncRNAs and their target mRNAs to identify key regulatory pathways in BC. qRT‐PCR revealed elevated LINC00200 and HMGB3 levels in BC cells, with minor differences in LINC00200 across cell lines. The expression of miR‐20b‐5p decreased in MCF‐7 cells but increased in BT549 and MDA‐MB‐231.ConclusionsThe study is the first to identify LINC00200 as a potential prognostic lncRNA predictive of survival outcomes in BC and to systematically delineate a novel ceRNA regulatory axis comprising LINC00200, miR‐20b‐5p, and HMGB3. Although this axis holds promise as a novel prognostic marker, further functional studies and assessment of diagnostic utility are warranted. Background Breast cancer (BC) has emerged as the most prevalent malignancy among women globally, with its incidence persistently increasing [1]. BC is categorized into several subtypes, including luminal A, luminal B, human epidermal growth factor receptor 2 (HER‐2) overexpression, basal‐like, and other special types [2]. The efficacy of treatment and prognosis for BC are predominantly influenced by its stage and subtype. Nonetheless, existing diagnostic methodologies for BC present certain limitations. In particular, pathological examinations are invasive, while imaging techniques are associated with specific inaccuracies. Furthermore, the sensitivity and specificity of current biomarkers necessitate further investigation, thereby constraining their efficacy in the diagnosis and prognosis of BC [3, 4]. Recently, numerous multigene expression‐based assays have been developed and commercialized, offering prognostic and predictive insights into the risk of BC recurrence and the effects of treatment. However, these assays still fall short of fulfilling clinical requirements. Consequently, there is an urgent need to develop novel biomarkers to enhance the auxiliary diagnosis and prognostic evaluation of BC, with the ultimate aim of improving patient survival rates. Long non‐coding RNAs (lncRNAs) represent a distinct category within the broader class of non‐coding RNAs. Traditionally, lncRNAs were considered to consist of non‐functional sequences generated during transcription. However, recent research has elucidated that lncRNAs may serve critical roles as “signals”, “guides”, “scaffolds”, and “placeholder bindings” throughout transcriptional, post‐transcriptional, and epigenetic processes. Moreover, lncRNAs are implicated in the regulation of various molecular biological processes, including DNA methylation, histone modification, chromosomal rearrangement, and the activation or silencing of target genes [5, 6, 7, 8]. The expression of lncRNAs is both tumor‐specific and stage‐specific, and they have been identified as potential molecular biomarkers for diagnosis, prognosis, and therapeutic response [9]. Presently, lncRNAs have been the subject of extensive research in the context of BC. For instance, long intergenic non‐protein coding RNA 1614 (LINC01614) serves as a diagnostic and prognostic marker, with its elevated expression correlating with poor survival outcomes. Additionally, hox transcript antisense RNA (HOTAIR) promotes BC progression, metastasis, and drug resistance through the microRNA‐129‐5p (miR‐129‐5p)/frizzled class receptor 7 (FZD7) axis, thereby emerging as a potential biomarker and therapeutic target for BC [10]. Furthermore, other lncRNAs including small nucleolar RNA host gene 1 (SNHG1), H19 Imprinted Maternally Expressed Transcript (H19), HAGLR opposite strand lncRNA (HAGLROS), growth arrest specific 5 (GAS5), and myocardial infarction associated transcript (MIAT) have also been shown to significantly associate with BC growth, metastasis, and drug resistance [11, 12, 13, 14, 15]. Advancements in high‐throughput gene sequencing and microarray gene chips have facilitated the establishment of public database platforms, notably The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) [16], which serve as valuable resources for the identification and exploration of biomarkers closely associated with various diseases. This tool has been effectively utilized to pinpoint critical targets linked to BC lesions and their corresponding regulatory mechanisms [17, 18]. WGCNA is a widely employed method in biological research, offering a more comprehensive and systematic understanding of tumorigenesis and progression. The analysis of gene interrelationships across various samples is crucial for understanding genetic patterns. Specifically, identifying a substantial number of genes associated with similar genetic diseases allows for the formation of gene collections. This approach facilitates the identification of gene sets and core genes that are closely linked to clinical phenotypes [19]. WGCNA has demonstrated significant value in investigating the pathogenesis, diagnosis, prognosis, and treatment resistance of BC. In this study, we identified differentially expressed genes associated with BC from the TCGA database. We conducted an analysis of lncRNA modules and immune cells using WGCNA, followed by an assessment of immune cell infiltration. To identify target mRNAs linked to survival risk, we performed univariate Cox regression analysis on each key gene. Additionally, we utilized multiple databases to predict interactions between microRNAs (miRNAs) and target mRNAs, culminating in the construction of a competing endogenous RNA (ceRNA) network comprising lncRNAs, miRNAs, and target mRNAs. Furthermore, the target mRNAs were subjected to enrichment analyses utilizing the KEGG, GO, and Gene Set Enrichment GSEA. Although LINC00200 has been reported in other cancers, its expression, prognostic significance, and functional role in BC remain unexplored. Moreover, while miR‐20b‐5p and HMGB3 have been individually investigated in BC, they have not yet been incorporated into a lncRNA‐miRNA‐mRNA ceRNA regulatory axis. Consequently, this study presents the first systematic integration of these three molecules into a predicted BC‐specific ceRNA network, supported by multi‐cohort bioinformatics analysis and preliminary experimental validation. Materials and Methods Data Collection and Preprocessing Gene expression data and comprehensive clinical information for BC patients were sourced from the TCGA database (https://portal.gdc.cancer.gov/). After excluding samples lacking survival data, the study included 1057 bc samples and 111 control samples, all of which contained expression data for lncRNA, miRNA, and mRNA. These samples serve as a training set for our study. And, we also took datasets from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) as an external validation cohort (GSE134359). It is a dataset from Mexico with 12 control samples and 74 bc samples. Additionally, immune‐related mRNA data comprising a total of 2660 genes were retrieved from the GeneCards database (https://www.genecards.org/). The “venn” R package was utilized to illustrate the intersection of immune‐related genes and BC mRNA. This study employed the analytical workflow shown in Figure 1 to systematically screen for key genes linked to BC prognosis and immunity. The analytical process of the present study. WGCNA We conducted two WGCNA focusing on lncRNAs in the context of BC, utilizing the R package “WGCNA” [20]. In the initial analysis, lncRNAs were examined in relation to traits associated with both the BC cohort and the control group, allowing us to ascertain the correlations between lncRNA modules and these two sets of traits. In the subsequent analysis, we investigated the relationship between lncRNA modules and the infiltration levels of 22 distinct types of immune cells, thereby determining the correlation between lncRNA modules and immune cell infiltration. The gene expression analysis pipeline was implemented as follows. Initial transcripts per million (TPM) values underwent log2‐transformation followed by filtration of highly variable genes (standard deviation > 0.01). Subsequent hierarchical clustering with static tree‐cutting (height = 20,000; minimum cluster size = 10) removed outlier samples. We then constructed signed adjacency and topological overlap matrices (TOM) for co‐expression network analysis. Using dynamic tree‐cutting (depth = 2; minimum module size = 25), we identified gene modules that were further merged when inter‐module correlation exceeded 0.75. Module‐phenotype associations were ultimately assessed through Pearson correlation analysis. All parameters were determined following established WGCNA guidelines to ensure biological validity and analytical reproducibility. Immune Cell Infiltration Analysis The algorithm's source code, accessible via the Cell‐type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT) database website (https://cibersortx.stanford.edu/) [21], along with reference files for 22 distinct immune cell types, was utilized to estimate immune cell infiltration levels in each BC sample. Subsequently, the R package “corrplot” was employed to analyze the correlations among the various immune cell types. The analytical workflow employed ν‐Support Vector Regression (ν‐SVR) integrated with permutation testing to deconvolve mixed cellular samples. Input data comprised a normalized feature matrix (mean‐centered, unit variance‐scaled) and log2‐transformed mixed expression profiles (maximum value < 50), with optional quantile normalization for batch effect mitigation. Systematic ν‐parameter optimization (0.25, 0.5, 0.75) was conducted through root mean square error (RMSE) minimization to select the optimal regression model, followed by computation of cell type specificity weights after zero‐normalization adjustment of negative values. Permutation testing with 1000 iterations established null distributions to statistically validate correlation coefficients. Implementation leveraged the e1071 R package (v1.7–14) with parallel computing acceleration, while strict gene symbol matching ensured biological consistency across feature matrices. Model stability was further verified through 10‐fold cross‐validation, maintaining adherence to standardized computational genomics protocols throughout the pipeline. Survival Analysis Single‐factor Cox regression analysis and Kaplan–Meier survival analysis were conducted utilizing the “survival” package in R. To identify target mRNAs associated with survival risk [22], univariate Cox regression analysis was applied to each gene, and those demonstrating statistical significance were selected for inclusion in the study. Furthermore, to refine the selection of lncRNAs and target mRNAs linked to survival, Kaplan–Meier survival analysis was performed on each gene, with statistically significant genes being incorporated into the study. Integrated clinical metadata informed the stratification of samples into high/low expression cohorts using median expression thresholds derived from gene expression profiles with matched survival data (time‐to‐event endpoints). Survival differences were assessed via log‐rank testing (Mantel‐Cox method), generating significance probabilities (p‐values) for inter‐group divergence. Kaplan–Meier curves were subsequently generated through the survival R package's survfit algorithm, quantifying time‐dependent survival probabilities with associated 95% confidence intervals calculated using the Greenwood variance estimator. Gene–Gene Interaction Analysis To predict the interactions between lncRNAs and miRNAs, we utilized the miRcode database (http://www.mircode.org/) [23]. Additionally, we employed the microRNA Database (miRDB, https://mirdb.org/), microRNA Target Base (miRTarBase, https://mirtarbase.cuhk.edu.cn/), and Target Scanning Database (TargetScan, https://www.targetscan.org/) databases to predict the interactions between miRNAs and their target mRNAs [24, 25, 26]. To ensure the reliability of the predicted interactions, we implemented a stringent triple‐intersection criterion: we retained only those miRNA‐mRNA interactions concurrently predicted by all three databases (miRDB, miRTarBase, and TargetScan) for ceRNA network construction. Notably, miRTarBase comprises experimentally validated interactions (e.g., luciferase reporter assays, CLIP‐seq), whereas miRDB and TargetScan yield computationally predicted interactions based on sequence complementarity and conservation. By mandating consensus across all three databases, our approach integrates both experimental evidence and computational consensus, thereby substantially reducing the false‐positive rate inherent in methods relying solely on computational predictions. Functional Enrichment Analysis We utilized the R package “clusterProfiler” to conduct enrichment analyses, including the KEGG, GO, and GSEA, on the target mRNA [27]. Gene identifiers were systematically mapped to Entrez IDs using the org.Hs.eg.db annotation database (v3.17.0), enabling functional enrichment analysis through hypergeometric distribution testing against KEGG pathways and GO categories (biological processes, molecular functions, cellular components). Statistical significance thresholds were rigorously defined as raw p‐values < 0.05 with Benjamini‐Hochberg false discovery rate (FDR)‐adjusted p‐values < 0.05. For pathway‐level characterization, GSEA was implemented using log2‐fold change (logFC)‐ranked differential expression profiles, normalized through the same annotation framework. The analytical workflow specifically employed the Molecular Signatures Database (MSigDB) C2 KEGG gene set collection (version c2.cp.kegg.v2022.1) to define functional modules, followed by FDR correction of pathway‐level p‐values [28]. Significant enrichment patterns were visualized through normalized enrichment score (NES) trajectories, calculated as weighted Kolmogorov–Smirnov statistics reflecting directional pathway activation (positive NES) versus suppression (negative NES), with critical parameter optimization guided by the clusterProfiler R package (v4.8.1). The results of these enrichment analyses were visualized using the “enrichplot” package. Cell Culture The cell lines used in this study were routinely maintained in our laboratory, authenticated by short tandem repeat (STR) analysis, and regularly confirmed to be free of mycoplasma contamination via polymerase chain reaction (PCR) testing. Normal breast epithelial MCF‐10A cells and BC cell lines BT549, MDA‐MB‐231, and MCF‐7 were obtained from the Cell Bank of the Chinese Academy of Sciences. The MCF‐10A cells were maintained in a complete culture medium specific for human breast epithelial cells (Procell, Wuhan, China), while the BT549, MDA‐MB‐231, and MCF‐7 cell lines were cultured in a complete medium designed for human BC tissue‐derived cells (Procell, Wuhan, China). All cell cultures were incubated in a carbon dioxide incubator set at 5% CO2 and 37°C, with the culture medium being refreshed every 2 days. qRT‐PCR A total of 2 × 106 cells in the exponential growth phase were collected, and 1 mL of TRIzol reagent (Thermo, USA) was added for RNA extraction. Total RNA was isolated following the manufacturer's protocol. Subsequently, the extracted RNA served as a template for first‐strand cDNA synthesis, utilizing either the StarScript III All‐in‐one RT Mix with gDNA Remover (GenStar) or the StarScript III miRNA RT Kit (by Stem‐loop) (GenStar). The quantitative PCR (qPCR) reaction setup consisted of 10 μL of 2× RealStar Fast SYBR qPCR Mix (Low ROX), 0.5 μL each of upstream and downstream primers (both at a concentration of 10 μmol/L), 1 μL of cDNA, and 8 μL of double‐distilled water (ddH2O). The thermal cycling conditions were as follows: an initial pre‐denaturation step at 95°C for 2 min, followed by 40 cycles of denaturation at 95°C for 15 s, annealing at 60°C for 25 s, and extension at 72°C for 30 s. The qPCR was conducted using the ABI 7500 instrument, with the melting curve analysis automatically configured by the device. The primer sequences are presented in Table 1. The relative expression levels of each gene were calculated using the 2−ΔCt method, with β‐actin or U6 serving as internal reference genes. TABLE 1Primer sequences.GenesPrimer sequencesβ‐ActinForward5′‐CTCCATCCTGGCCTCGCTGT‐3′Reward5′‐GCTGTCACCTTCACCGTTCC‐3′LINC00200‐201Forward5′‐GTCCAGCAGCAAAGGTAG‐3′Reward5′‐CAAGAGGTGAGGAAGTG‐3′LINC00200‐202Forward5′‐TTCAGGTAGGGACGGTTC‐3′Reward5′‐GAAGCATTGTCCTCGCAG‐3′LINC00200‐203Forward5′‐CGTTATCAGGGACTGTGCT‐3′Reward5′‐GCAGAGCTGAGAAAGACTC‐3′LINC00200‐204Forward5′‐GCACCTCAACCTCACTCTG‐3′Reward5′‐CACATCTGGAGAAAGCCAC‐3′LINC00200‐205Forward5′‐CTCCTTGCCCAGCACACAG‐3′Reward5′‐GCAGGCTCATCTCATCTG‐3′LINC00200‐206Forward5′‐AGGCAGAGTCTTCCTATG‐3′Reward5′‐GAAGCCAAATCCCTCAAC‐3′HMGB3Forward5′‐GATTCTTCCTGTTCTGTTC‐3′Reward5′‐CTCATACTTCTCCTTCAG‐3′U6Forward5′‐CTCGCTTCGGCAGCACA‐3′Reword5′‐AACGCTTCACGAATTTGCGT‐3′miR‐20b‐5pForward5′‐GCTGCAAAGTGCTCATAGTG‐3′Reward5′‐GCAGGGTCCGAGGTATTC‐3′ Primer sequences. Statistical Analysis The bioinformatics analysis section employs R software (version 4.1.2) for statistical computations. For variables exhibiting non‐normal distributions, the Wilcoxon rank‐sum test was utilized to assess statistical significance. In contrast, the independent t‐test was applied to evaluate differences between two sets of continuous variables with normal distributions. The comparison and analysis of statistical significance between two sets of categorical variables were conducted using either Fisher's exact test or the Chi‐square test. The monotonic correlation between two ordinal variables was assessed using Spearman's rank correlation coefficient. Statistical analyses were performed using SPSS version 24.0. Measurement data were expressed as mean ± standard error, and differences among groups were evaluated using one‐way analysis of variance followed by Tukey's post hoc test. A p‐value of less than 0.05 was considered indicative of statistical significance. *p < 0.05, **p < 0.01, ***p < 0.001. Results WGCNA for the lncRNA Modules Correlated With BC Traits Following the exclusion of genes exhibiting minimal variability (standard deviation < 0.01), we initially performed clustering on the samples from 1057 bc samples and 111 control samples, setting the dendrogram cut‐off height at 20,000. Samples falling below this threshold were subsequently removed (2 bc samples and 30 control samples), and the remaining samples (1055 bc samples and 81 control samples) were re‐clustered, resulting in distinct separation into control and BC groups (Figure 2A). We selected a soft threshold of 5, ensuring that the correlation coefficient R 2 exceeded 0.9 (Figure 2B) and the average connectivity remained below 100 (Figure 2C), thereby optimizing the network to accurately reflect the biological state. We employed the heterogeneity of the TOM algorithm to cluster lncRNAs, integrating similar genes into distinct modules. Those with a dissimilarity height below 0.25 (MEsalmon, MEsienna3, MEpink, MEdarkmagenta and MEsteelblue) were excluded (Figure 2D), and subsequently, analogous modules were consolidated. This process resulted in the identification of 33 lncRNA modules (Figure 2E). We then conducted an analysis to examine the associations between these modules and BC status, as well as control traits (Figure 2F; Table S1). From this analysis, we identified lncRNAs in the MEbrown module containing 991 lncRNAs (correlation coefficient = 0.59, p < 0.001) and the MEwhite module containing 230 lncRNAs (correlation coefficient = 0.67, p < 0.001), which showed statistically significant associations (Table S1). The relationship between these 2 gene modules and tumor composition is positively correlated. WGCNA analysis in BC samples and control samples. (A) Sample clustering dendrograms (BC vs. samples). (B) Scale‐free fit index versus soft thresholds. (C) Average connectivity versus soft thresholds. (D) BC‐specific lncRNA module clustering. (E) lncRNAs co‐expression modules (colored by cluster). (F) Modules‐traits associations (rows, modules; columns, BC/Control). WGCNA for lncRNA Modules Correlated With Immune Cells Comparing the expression levels of these immune cell marker genes (usually mRNA) in 1057 bc samples with the standard dataset from the CIBERSORT official website (Table S2), the CIBERSORT algorithm is used to assess the infiltration levels of the corresponding immune cell types based on the expression levels of the marker genes. Subsequently, we analyzed the correlations among the levels of immune cell infiltration. Our findings revealed a positive correlation between active memory CD4+ T cells and CD8+ T cells (correlation coefficient = 0.48, p < 0.05), as well as a negative correlation between resting mast cells and T follicular helper (Tfh) cells (correlation coefficient = −0.34, p < 0.05) (Figure 3A). Following the exclusion of genes exhibiting low variability (standard deviation < 0.2), we initially performed clustering on the samples from 1.057 bc samples and 111 control samples. Additionally, we conducted clustering of lncRNAs using the heterogeneity‐based TOM algorithm, subsequently integrating similar genes into distinct modules. A soft threshold of 5 was selected, ensuring that the correlation coefficient R 2 exceeded 0.9 (Figure 3B) and the average connectivity remained below 100 (Figure 3C), thereby aligning the constructed network with the biological state. Those with a shear height below 0.25 were excluded (MEmidnightblue, MEpink, MEsalmon, MEtan, MEbrown and MEgreenyellow; Figure 3D). Subsequently, similar modules were consolidated (Figure 3E). This process resulted in the identification of 14 lncRNA modules (Table S3). Within the MEgreen module, we identified 950 lncRNAs exhibiting the largest absolute correlation coefficients with active CD4 memory T cells (correlation coefficient = 0.49, p < 0.001). Additionally, 3181 lncRNAs were identified in the MEturquoise module, characterized by a correlation coefficient of −0.49 with resting mast cells (p < 0.001), as potential candidate genes related to immune cells (Figure 3F; Table S3). WGCNA levels in BC samples with immune cell infiltration. (A) Heat map immune cell correlations (22 cell types). (B) Scale‐free fitting index versus soft thresholds. (C) Average connectivity versus soft thresholds. (D) BC‐specific lncRNA module dendrogram. (E) lncRNA co‐expression modules (color‐coded clusters). (F) Module‐immune cell associations (rows: Modules; columns: Immune cells). Identification of Key Genes in BC and Creation of a ceRNA Network From previous WGCNAs, we identified 277 bc‐ and immune cell‐associated lncRNAs in 1057 bc samples by crossing 1121 bc‐associated lncRNAs with 4131 immune cell‐associated lncRNAs (Figure 4A; Table S4). Additionally, we intersected 19,938 target mRNAs related to BC with 2660 immune‐related genes, resulting in 1759 immune‐related target mRNAs pertinent to BC (Figure 4B; Table S5). To further refine the scope of our target mRNA analysis, we performed a univariate Cox regression analysis, identifying 12 target mRNAs (RE1 silencing transcription factor (REST)—Interferon Regulatory Factor 1 (IRF1), C‐C motif chemokine ligand 1 (CCL1), heat shock protein family A (Hsp70) member 8 (HSPA8), interferon gamma (IFNG), TNF receptor associated factor 1 (TRAF1), mitogen‐activated protein kinase 9 (MAPK9), nuclear receptor subfamily 1 group H member 3 (NR1H3), mitogen‐activated protein kinase kinase 6 (MAP2K6), HMGB3, Janus kinase 2 (JAK2), and CKLF like MARVEL transmembrane domain containing 4 (CMTM4)) that were statistically significant and correlated with survival risk (Figure 4C; Table S4). We used the miRcode database to predict the interactions between lncRNAs and existing miRNAs in BC samples, identifying 9 lncRNAs (zinc and ring finger 3 antisense RNA 1 (ZNRF3‐AS1), adenosine deaminase RNA specific B2 antisense RNA 1 (ADARB2‐AS1), long intergenic non‐protein coding RNA 243 (LINC00243), major histocompatibility complex Class I F antisense RNA 1 (HLA‐F‐AS1), LINC00200, LINC00160, zinc finger protein 295 antisense RNA 1 (ZNF295‐AS1), Helicobacter pylori responsive 1 (HPYR1), and ribonucleotide reductase catalytic subunit M1 antisense RNA 1 (RRM1‐AS1)) and 11 miRNAs (human miRNA (hsa‐miR)‐20b‐5p, hsa‐miR‐125a‐5p, hsa‐miR‐301b‐3p, hsa‐miR‐17‐5p, hsa‐miR‐24‐3p, hsa‐miR‐761, hsa‐miR‐613, hsa‐miR‐125b‐5p, hsa‐miR‐425‐5p, hsa‐miR‐3619‐5p, and hsa‐miR‐135a‐5p) that have interaction relationships. Subsequently, we used the miRDB database, miRTarBase database, and TargetScan database to predict the interactions between miRNAs and target mRNAs associated with BC risk, identifying 12 mRNAs that have interaction relationships (CCL1, REST, IRF1, HSPA8, IFNG, TRAF1, MAPK9, NR1H3, MAP2K6, HMGB3, JAK2, and CMTM4). Finally, the ceRNA network of lncRNA‐miRNA‐target mRNA was constructed (Figure 4D). Key Genes screening and ceRNA network in breast cancer (BC). (A) WGCNA module (BC‐related vs. immune‐related genes). (B) Overlap between immune genes and BC target mRNAs (Venn diagram). (C) Univariate Cox analysis of immune‐linked mRNA (forest plot). (D) ceRNA regulatory network map (lncRNA‐miRNA‐target mRNA interactions). Enrichment Analysis of Key Target mRNAs in BC To investigate the roles and associated biological pathways of pivotal target mRNAs in 1057 bc samples, we conducted enrichment analyses using KEGG, GO, and GSEA methodologies. Our findings revealed significant enrichment in several KEGG pathways, including those related to BC, the TNF signaling pathway, the prolactin signaling pathway, differentiation of Th1 and Th2 cells, differentiation of Th17 cells, programmed death‐ligand 1 (PD‐L1) expression in cancer, and the PD‐1 checkpoint pathway (Figure 5A). Within the enriched GO pathway analysis, the biological processes identified encompass the gamma interferon‐mediated signaling pathway, the regulation of interleukin‐17 production, and the regulation of nitric oxide synthase biosynthesis, among others. In terms of molecular function, the pathways include protein tyrosine kinase activity, G protein‐coupled receptor binding, protein serine/threonine/tyrosine kinase activity, and cytokine receptor binding (Figure 5B). In the GSEA, the control group exhibited enrichment in pathways such as the adipocytokine signaling pathway, cytochrome P450 metabolism of xenobiotics, neuroactive ligand‐receptor interaction, peroxisome proliferator activated receptor gamma (PPAR) signaling pathway, and retinol metabolism (Figure 5C). Conversely, the BC group demonstrated enrichment in pathways related to antigen processing and presentation, cell cycle regulation, DNA replication, proteome dynamics, and pyrimidine metabolism. These findings indicate that target mRNA in BC is more significantly implicated in immune processes and cellular proliferation (Figure 5D). Functional enrichment analysis of key breast cancer (BC) genes. (A) KEGG pathway enrichment bar plot. (B) GO term enrichment bubble chart (biological processes, molecular functions). (C) Control group GSEA ridge plot (baseline gene set activity). (D) BC group GSEA ridge plot (disease‐specific pathway enrichment). Identification and Analysis of lncRNAs Linked to Survival Outcomes in BC To further identify lncRNAs associated with survival outcomes in 1057 bc, we stratified the BC samples into high and low expression groups based on the median expression levels of nine lncRNAs (each group has 528 or 529 samples). Subsequently, we performed a survival analysis using the Kaplan–Meier method. The analysis revealed that, apart from LINC00200 and LINC00160, the differences in survival between the high and low expression groups for ADARB2‐AS1, HLA‐F‐AS1, HPYR1, LINC00243, RRM1‐AS1, ZHF295‐AS1, and ZNRF‐AS1 were not statistically significant. To our knowledge, this is the first study to report the significant overexpression of LINC00200 in BC and is associated with an unfavorable prognosis. Both LINC00200 (p = 0.038) and LINC00160 (p = 0.041) were identified as high‐expression groups associated with poorer prognosis (Figure 6A–I). These molecules represent candidate molecules for exploratory survival analysis. Our comparative analysis of the expression levels of nine lncRNAs between 111 the control samples and the 1057 bc samples revealed elevated expression of LINC00200 and LINC00160 in the latter. These findings indicate a correlation between LINC00200 (p < 0.001) and LINC00160 (p < 0.001) expression and patient prognosis in BC, with elevated levels serving as predictors of reduced survival outcomes (Figure 7A–I). Moreover, we obtained the same results from the analysis of an external cohort containing 12 control samples and 74 bc samples, especially LINC00200 (p < 0.001) and LINC00160 (p < 0.001) were both highly expressed in BC samples (Figure 7J–R). The novel association between LINC00200 and BC was subsequently validated using an independent GEO cohort (GSE134359). From the preceding analytical phase, we identified the regulatory pathways associated with LINC00200 and LINC00160, subsequently uncovering seven corresponding target mRNAs (JAK2, HSPA8, MAPK9, REST, HMGB3, CCL1 and MAP2K6; Figure 8A). In general, lncRNAs impede miRNA‐mediated degradation of target mRNAs by suppressing miRNA expression. Given that the aforementioned analysis indicates LINC00200 and LINC00160 function as oncogenic genes, it follows that the target mRNAs indirectly regulated by these lncRNAs are also oncogenic in nature (Figure 8B). Furthermore, we conducted an analysis of gene expression across various BC subtypes in 1057 bc samples. The findings revealed that, relative to the 111 control samples, the expression of LINC00200 was elevated in the Luminal A, Luminal B, and Her‐2 subtypes, with the Luminal B subtype exhibiting the highest expression levels (Figure 8C). Moreover, in the external validation cohort (12 control group samples and 74 bc samples), we observed that LINC00200 and HMGB3 were also the same as in the training set (Figure 8E,F). In the case of HMGB3, all four subtypes demonstrated increased expression levels compared to the control group, with the Her‐2 subtype showing the most pronounced expression (Figure 8D). Survival analysis of breast cancer samples grouped according to 9 key lncRNAs expression levels. (A–I) Kaplan–Meier survival analyses were conducted for 9 key lncRNAs including LINC00200, LINC00160, ADARB2‐AS1, HLA‐F‐AS1, HPYR1, LINC00243, RRM1‐AS1, ZHF295‐AS1, and ZNRF‐AS1. Expression analysis of nine key lncRNAs in breast cancer (BC) versus controls. (A–I) Differential expression of lncRNAs in BC versus control groups: LINC00200, LINC00160, ADARB2‐AS1, HLA‐F‐AS1, HPYR1, LINC00243, RRM1‐AS1, ZNF295‐AS1 and ZNRF3‐AS1. (J–R) Validation in an external cohort confirmed consistent expression patterns for 9 key lncRNAs: LINC00200, LINC00160, ADARB2‐AS1, HLA‐F‐AS1, HPYR1, LINC00243, RRM1‐AS1, ZNF295‐AS1 and ZNRF3‐AS1. Regulatory mechanism and gene expression in breast cancer (BC) subtypes. (A) LINC00160/LINC00200‐ miRNAs‐mRNA interaction network. (B) Schematic of the lncRNA‐miRNA‐mRNA regulatory axes. (C) LINC00200 expression BC subtypes versus control (training set). (D) HMGB3 expression in BC subtypes versus control (training set). (E) LINC00200 expression validation (external cohort). (F) HMGB3 expression validation (external cohort). Identification of Target mRNAs Linked to Survival and Elucidation of Key Regulatory Pathways in BC To further identify target mRNAs associated with survival in BC, we categorized the 1057 bc samples into high and low expression groups based on the median expression levels of seven target mRNAs. Subsequently, we performed a survival analysis using the Kaplan–Meier method. The analysis revealed that, with the exception of HMGB3 (p = 0.026) and MAP2K6 (p = 0.03), the differences in survival between the high and low expression groups for CCL1, JAK2, HSPA8, MAPK9, and REST were not statistically significant (Figure 9A–G). The cohort exhibiting elevated HMGB3 expression demonstrated a poorer prognosis, whereas the cohort with increased MAP2K6 expression exhibited a more favorable prognosis. Comparative analysis of the expression levels of seven target mRNAs between the 111 control samples and the 1057 bc group samples revealed that HMGB3 (p < 0.001) was significantly overexpressed in the BC group, whereas MAP2K6 (p < 0.001) was underexpressed (Figure 10A–G). Similarly, we also analyzed 12 control samples and 74 bc samples from the external validation set, and the expression of genes HMGB3 (p < 0.001) and MAP2K6 (p < 0.001) was consistent with the results of the training set (Figure 10H–N). These findings suggest that HMGB3 functions as an oncogene, while MAP2K6 acts as a tumor suppressor gene. In conclusion, our study identifies the predicted LINC00200‐miR‐20b‐5p‐HMGB3 axis as a candidate regulatory pathway in BC. Survival analysis of breast cancer samples grouped according to mRNA expression levels of seven key targets. (A–G) Kaplan–Meier survival analyses were conducted for 7 key target mRNAs: HMGB3, MAP2K6, CCL1, JAK2, HSPA8, MAPK9, and REST. Expression analysis of seven key target mRNAs in breast cancer (BC) versus controls. (A–G) Differential expression of lncRNAs in BC versus control groups: HMGB3, MAP2K6, CCL1, JAK2, HSPA8, MAPK9, and REST. (H–N) Validation in an external cohort confirmed consistent expression patterns for 7 key target mRNAs: HMGB3, MAP2K6, CCL1, JAK2, HSPA8, MAPK9, and REST. Expression of the Key Genes in Various BC Cells Lines The expression levels of LINC00200 transcripts, miR‐20b‐5p, and HMGB3 in MCF‐10A cells, as well as in three distinct BC cell lines, were quantitatively assessed using qRT‐PCR. As illustrated in Figure 11, the expression levels of various LINC00200 transcripts exhibited significant variability across the different BC cell types. Notably, LINC00200‐201 and LINC00200‐202 were most abundantly expressed in MDA‐MB‐231 cells, with LINC00200‐202 also demonstrating upregulation in BT549 cells. The expression levels of LINC00200‐203, LINC00200‐204, and LINC00200‐205 were found to be highest in MCF‐7 cells. Additionally, LINC00200‐203 exhibited elevated expression in MDA‐MB‐231 and BT549 cells, while LINC00200‐205 showed increased expression in MDA‐MB‐231 cells. Conversely, LINC00200‐206 was downregulated in both MDA‐MB‐231 and BT549 cells. The downstream molecule miR‐20b‐5p, associated with LINC00200, was downregulated in MCF‐7 cells but upregulated in BT549 and MDA‐MB‐231 cells. Furthermore, the expression of HMGB3, the downstream target of miR‐20b‐5p, was elevated in both MCF‐7 and MDA‐MB‐231 cells. The expression levels of (A) LINC00200‐201, (B) LINC00200‐202, (C) LINC00200‐203, (D) LINC00200‐204, (E) LINC00200‐205, (F) LINC00200‐206, (G) miR‐20b‐5p, and (H) HMGB3 were analyzed using qRT‐PCR (n = 3). Discussion In this study, we identified lncRNA modules strongly correlated with BC characteristics. Analysis of immune cell infiltration revealed elevated levels of active CD4 memory T cells in BC samples, while quiescent mast cells showed reduced infiltration in control samples. Fourteen lncRNA modules linked to immune cells were discovered. Further intersection analyses highlighted modules associated with both BC traits and immune regulation. To further investigate the clinical significance, univariate Cox regression identified 12 survival‐related mRNAs, and enrichment analyses (KEGG, GO, and GSEA) demonstrated their predominant involvement in immune responses and cell proliferation. Key lncRNAs influencing BC survival were pinpointed, along with their target mRNAs and regulatory pathways. Notably, the LINC00200‐miR‐20b‐5p‐HMGB3 axis emerged as a predicted pivotal regulatory axis in BC. qRT‐PCR validation confirmed differential expression of LINC00200, miR‐20b‐5p, and HMGB3 in BC cell lines. While LINC00200 has been implicated in other cancers, its expression pattern, prognostic value, and functional role in BC remain unreported. Thus, our finding that LINC00200 is overexpressed in BC and associated with poor survival represents a novel discovery specific to this malignancy. Furthermore, although miR‐20b‐5p and HMGB3 have been individually studied in BC, those studies focused on the isolated role of each molecule. No previous study has integrated these two molecules with an upstream lncRNA into a ceRNA network within the BC landscape. This integrated approach provides a new mechanistic perspective on BC pathogenesis. Tumor immunity critically influences BC development and outcomes, where enhanced CD4+/CD8+ T cell infiltration correlates with improved prognosis [29], while CD4+ T cells directly modulate tumor progression and the clinical responses [30, 31]. lncRNAs significantly regulate genetic functions and participate in various biological processes, including tumor immunology, by affecting antigen release, presentation, immune activation, and immune cell infiltration [32]. Those influencing immune cell infiltration are termed immune‐associated lncRNAs (irlncRNAs) [33]. Although few irlncRNAs are known, research shows lncRNA TCL6 affects the tumor microenvironment by modulating cell invasion [34], while lncRNA MIAT is linked to 13 types of tumor‐infiltrating immune cells [35]. Tan et al. found that BC patients in the high‐risk group had a higher distribution and quantity of immune cells compared to the low‐risk group [36]. Our study revealed key immune‐related findings in BC. Analysis of TCGA and GEO datasets showed distinct immune cell infiltration patterns. Active CD4+ memory T cells and CD8+ cytotoxic T cells were positively associated with BC presence, whereas resting mast cells and T follicular helper cells showed negative associations with BC presence. BC samples exhibited higher CD4+ memory T cell infiltration but lower resting mast cell levels compared to healthy controls, highlighting the role of T cell and mast cell activity in BC progression. Fourteen lncRNA modules associated with BC immune cells were identified, along with 277 modules related to BC trait and immune cells. Twelve survival‐related mRNAs, including REST, IRF1, CCL1, HSPA8, IFNG, TRAF1, MAPK9, NR1H3, MAP2K6, HMGB3, JAK2, and CMTM4, were identified as key prognostic markers. Functional analyses revealed their critical roles in immune responses and BC cell proliferation. Research indicates that CCL1 serves as the primary factor for the attraction of regulatory T cells in BC [37]. Additionally, HSPA8 demonstrates a significant correlation with various immune cell subsets and is linked to a poor prognosis in cases of triple‐negative BC [38]. Furthermore, low expression levels of NR1H3 are associated with macrophage invasion and a decreased survival rate in BC patients [39]. Discs large MAGUK scaffold protein associated protein 5 (DLGAP5) has been shown to regulate the proliferation, migration, invasion, and cell cycle of BC cells via the JAK2/signal transducer and activator of transcription 3 (STAT3) signaling pathway [40]. Moreover, STAT4 is implicated in the promotion of PD‐L1 expression and serves as a predictor of response to immunotherapy in BC through the interleukin‐12 receptor (IL‐12R)/JAK2/STAT3 axis [41]. In addition to REST, numerous other target mRNAs have been identified in the context of immune cell infiltration in BC [42, 43, 44, 45, 46]. These findings align with the results of a bioinformatic analysis of target mRNA enrichment, indicating that upstream lncRNAs may contribute to BC progression and immune response. Kaplan–Meier analysis revealed that elevated LINC00200 and LINC00160 levels correlated with worse outcomes, hinting at their oncogenic roles. Investigation into their regulatory pathways identified HMGB3 and MAP2K6 as key mRNA targets with significant associations with patient survival. Specifically, high HMGB3 expression indicates poor prognosis, while high MAP2K6 suggests better outcomes, implying HMGB3 is an oncogene and MAP2K6 a tumor suppressor. Alluvial diagrams indicate LINC00200 may modulate HMGB3 via miR‐20b‐5p, suggesting its role in BC and tumor immunity. Further experimental validation, such as functional assays and in vivo studies, is needed to confirm these interactions. Current research highlights the interaction and mutual regulation between lncRNAs and miRNAs, influencing gene expression. Typically, lncRNAs bind to miRNAs, suppressing their expression and allowing mRNA to function. Occasionally, miRNAs can target and reduce lncRNAs [47, 48]. Our bioinformatics analysis identifies the predicted LINC00200/miR‐20b‐5p/HMGB3 axis as a candidate regulatory pathway in BC. Prior research links LINC00200 to tumor occurrence, growth, and spread. Lin et al. found that LINC00200 was overexpressed in gastric cancer tissues, correlating with poor differentiation, lymph node metastasis, increased cell proliferation, reduced apoptosis, and resistance to oxaliplatin chemotherapy [49]. Chen et al. [50]. found that LINC00200 is upregulated in neuroblastoma, enhancing its proliferation, migration, and invasion. Our study detected overall elevated expression of LINC00200's five transcripts in HR‐positive and triple‐negative BC cells, with variations across cell lines. PCR validation aligned with bioinformatics data, reinforcing LINC00200's potential as a BC therapeutic target. Notably, LINC00200‐206 expression increased in MCF‐7 cells but decreased in BT549 and MDA‐MB‐231 cells, suggesting subtype‐specific roles. Further investigation is warranted to clarify these findings. Many studies indicate that miRNAs show abnormal expression in various human tumors, affecting gene expression post‐transcriptionally in both normal and disease states. These miRNAs can act as oncogenes suppressors or promote tumor growth by silencing tumor‐suppressing mRNAs or not adequately silencing oncogenic mRNAs [51]. MiR‐20b‐5p is overexpressed in several cancers, promoting cell growth and spread in liver, gastric, laryngeal cancers, and T cell acute lymphoblastic leukemia but acts as a tumor suppressor in colon cancer by targeting cyclin D1 [52, 53, 54, 55, 56]. Reduced levels of miR‐20b‐5p, affected by lncRNA HOTAIR, can increase retinoblast proliferation and reduce apoptosis [57]. Thus, miR‐20b‐5p can function as both an oncogene and a tumor suppressor, impacting tumors through different genes. Research on miR‐20b‐5p in BC is scarce; however Xia et al. [58]. found it enhances malignancy in BC cells and stem cells, likely by regulating cyclin D1 (CCND1) and E2 promoter binding factor transcription factor 1 (E2F1) expression. Laborda‐Illanes et al. reported higher miR‐20b‐5p levels in metastatic BC patients [59]. Xu et al. [60] proposed that miR‐20b‐5p functions as a tumor suppressor by inhibiting the metastasis of BC. This conclusion was further supported by Sueta et al. [61], who observed significantly reduced serum levels of miR‐20b‐5p in patients with recurrent BC. qRT‐PCR showed decreased miR‐20b‐5p and increased LINC00200 expression in MCF‐7 cells, consistent with our predictions. In contrast, elevated miR‐20b‐5p levels were observed in BT549 and MDA‐MB‐231 cells, suggesting potential subtype variation in BC. Further studies are needed to confirm the regulatory pathways of miR‐20b‐5p. Numerous studies have investigated the role of HMGB3 in BC. Elevated expression of HMGB3 serves as an independent prognostic marker for reduced survival in BC patients. HMGB3 exhibits tumorigenic properties in BC and is involved in lymphatic vascular infiltration [62]. Gu et al. [63] demonstrated that silencing HMGB3 impedes BC cell proliferation and tumor growth through its interaction with hypoxia‐inducible factor 1α (HIF‐1α). miR‐142‐3p/sh‐RNA inhibits HMGB3, regulating autophagy and triggering apoptosis via ROS buildup and mitochondrial dysfunction, thereby reducing BC cells' tumorigenic potential [64]. Moreover, miR‐205 regulation of HMGB3 decreases BC cell proliferation and invasion [65]. qRT‐PCR and target mRNA expression analyses revealed elevated HMGB3 expression in MCF‐7 and MDA‐MB‐231 cells, as well as in BC tissues, compared to normal mammary epithelial cells and breast tissues. These findings indicate that HMGB3 functions as an oncogene and may serve as a potential therapeutic target in BC. In this study, we developed a predicted ceRNA network associated with BC, specifically the LINC00200‐miR‐20b‐5p‐HMGB3 regulatory axis, through biogenic analysis. Preliminary validation using qRT‐PCR confirmed the involvement of LINC00200, hsa‐miR‐20b‐5p, and HMGB3 in the pathogenesis of BC. In summary, LINC00200, miR‐20b‐5p, and HMGB3 could serve as exploratory survival‐associated candidates and potential diagnostic biomarkers for breast cancer. This study provides the first evidence that LINC00200 is overexpressed in breast cancer and associated with poor prognosis. Moreover, we systematically proposed LINC00200, miR‐20b‐5p, and HMGB3 as a predicted ceRNA regulatory axis in BC. The LINC00200‐miR‐20b‐5p‐HMGB3 axis is implicated in BC pathogenesis and represents a promising exploratory prognostic indicator; however, further functional validation is warranted. We acknowledge that several aspects of this study merit further investigation. First, the proposed LINC00200‐miR‐20b‐5p‐HMGB3 ceRNA axis remains predictive in nature, as our qRT‐PCR experiments confirmed the expression of individual components but did not directly demonstrate the regulatory relationship. Further functional experiments would help to substantiate this axis. Second, our survival analyses were based on nominal p‐values without correction for multiple testing; therefore, these molecules might best be interpreted as exploratory survival‐associated candidates and potential prognostic biomarkers at this stage. Third, while the external GEO cohort provided valuable validation of differential expression, it did not extend to survival associations or independent prognostic assessment. Future studies incorporating multiple‐testing corrections, larger independent cohorts with survival follow‐up, and functional validation of the ceRNA axis will further strengthen the clinical relevance of our findings. Author Contributions Yuhan Zheng: conceptualization, investigation, visualization, writing – original draft, data curation, methodology. Xinggui Chen: project administration, data curation, methodology. Xuan Liu: software, formal analysis, data curation. Yijin Zhou: software, data curation, formal analysis. Pei Yang: software, data curation, formal analysis. Shuhui Li: writing – review and editing, supervision, validation, funding acquisition, resources. Funding The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: lncRNAs of 33 WGCNA modules associated with BC traits. Table S2: Reference expression levels of marker genes used for assessing immune cell infiltration levels in the CIBERSORT algorithm. Table S3: lncRNAs of WGCNA modules associated with immune cells. Table S4: lncRNAs derived from tumor‐associated and immune cell‐associated WGCA and their intersection genes. Table S5: Immune‐related genes, all mRNAs in the BC samples, and the intersection genes of the two.