All human breast cancer tissue samples were obtained from the Affiliated Hospital of Guizhou Medical University. The prospective collection of diagnostic core (needle) biopsy specimens used for the establishment of patient-derived primary cancer cell lines (PCCLs) was approved by the Ethics Committee of Guizhou Medical University (Approval No. 2019-048), and written informed consent was obtained from all donors. For the exploratory discovery analysis, we restricted the PCCL set to four HR+/HER2-negative lines (BC4, BC5, BC10, BC13) with complete NACT and follow-up information, to minimize inter-subtype heterogeneity and align in-vitro phenotypes with clinically annotated sensitivity/non-response. The relevant clinical annotations are reported in Tables 1–3. Consequently, PCCL RNA-seq served as a hypothesis-generating layer that was cross-validated in an independent postoperative FFPE IHC cohort and public datasets.

For IHC analysis, we retrospectively retrieved formalin-fixed, paraffin-embedded (FFPE) tumor tissues obtained from surgical resection after treatment at the Affiliated Hospital of Guizhou Medical University. Clinical and imaging data (including RECIST v1.1 assessments of non-pCR cases: stable disease, recurrence or progression) were collected for patients followed for up to 3 years after surgery (18). This retrospective part of the study was approved by the Ethics Committee of the Affiliated Hospital of Guizhou Medical University (Approval No. 2024-489), and, owing to its retrospective nature and the use of de-identified data, the requirement for written informed consent was waived.

Breast tumor tissues (BC4, BC5, BC10, and BC13) were obtained from diagnostic biopsy specimens of four patients with HR+/HER2− breast cancer (Luminal B type, HER2 non-amplified). The tissues were placed in D-Hank’s balanced salt solution containing 500 U/mL penicillin and 500 μg/mL streptomycin, transported to the laboratory at low temperature, and stored at 4 °C. After washing the specimens twice with sterile D-Hank’s balanced salt solution (about 3 mL each time), the tissues were minced and added to D-Hank’s balanced salt solution containing 140 U/mL type II collagenase (17101015, Thermo Fisher Scientific, USA). Digestion was performed on a shaker at 37°C and 100 rpm for approximately 1 h. Subsequently, the cells were centrifuged at 1000 rpm for 10 min, the supernatant was discarded, and the cell suspension was cultured in BCMI medium (SFCtech, Beijing, China) at 37°C and 5% CO₂. After cell proliferation, the medium was changed to high-glucose DMEM medium. After at least two trypsin-EDTA/TrypLE digestions and mild mechanical dissociation to remove stromal cells, the expression of ERα, PR, and HER2 in PCCL (BC4, BC5, BC10, BC13) and their corresponding breast cancer tissues was verified by staining. The results demonstrated that the expression of molecular markers in primary breast cancer cells was consistent with that in the corresponding tissues.

To predict the sensitivity of breast cancer cells to docetaxel, represented by docetaxel, the proliferation ability of four PCCL cells was assessed using the CCK-8 method. Breast cancer cells were seeded in 96-well plates and cultured for 24 hours. Different doses of docetaxel (5000 nM, 1250 nM, 312 nM, 78 nM, 19.5 nM, 4.88 nM, 1.22 nM, and 0 nM) were used for culture. Each concentration was replicated three times and cultured for 72 hours. Cytotoxicity was determined using a CCK-8 kit, with absorbance measured at 450 nm after 1 hour of incubation. The average cell viability at each drug concentration was determined by normalization with the negative control value, and the proliferation curve was plotted to determine the half-inhibitory concentration (IC₅₀) value.

Quantitative real-time PCR (qRT-PCR). Total RNA was extracted from cells at ~70–75% confluence using TRIzol (Thermo Fisher Scientific, USA; Cat# TSP413), quantified on NanoDrop, and reverse-transcribed with the SynScript®III RT SuperMix (Thermo Fisher Scientific; Cat# TSK314S) following the manufacturer’s protocol. qRT-PCR was performed on a QuantStudio StepOne Plus (Applied Biosystems) using ArtiCan^CEO SYBR qPCR Mix (Cat# TSE401). Relative expression was calculated by the 2^{^-ΔΔCt} method with ACTB as the endogenous control. Primer sequences are listed in Table 4.

Immunohistochemical analysis was performed on tumor tissue samples from 54 patients with luminal B-type HER2-negative breast cancer. After fixation with 4% paraformaldehyde, dewaxing, and hydration, antigen retrieval was carried out using citrate buffer. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide, and nonspecific binding sites were blocked with 5% normal goat serum. Slides were incubated with primary antibodies against Ki67, E-cadherin, N-cadherin, and COL1A2 at 4°C overnight, washed with TBST, incubated with HRP-labelled secondary antibodies, and developed with DAB. Haematoxylin was employed for counterstaining. Two pathologists evaluated the immunostaining results in a blinded fashion. Positive signals were predominantly localized in the extracellular matrix. Five high-power fields were randomly selected, and the positive rate of 200 tumor cells was determined. Scoring was based on staining intensity (0–3 points) and the percentage of positive cells (0–4 points), with the product result determining the positive grade: 0 points for grade 0 (negative), 1–2 points for grade 1, 3–4 points for grade 2, and 4 or more points for grade 3 (positive).

Continuous variables were summarized as mean ± SD for normally distributed data or median (IQR) otherwise. Normality was assessed with the Shapiro–Wilk test and homoscedasticity with Levene’s test. Between-group comparisons used Student’s t test; when assumptions were violated, the Mann–Whitney U test was applied. For repeated measurements on the same samples (e.g., pre–post or paired designs), paired t tests (or Wilcoxon signed-rank tests for non-normal data) were used. IC₅₀ values were obtained by fitting four-parameter logistic (4PL) dose–response curves using GraphPad Prism 10 (nonlinear least squares), and differences in IC₅₀ between docetaxel-resistant and -sensitive groups were compared with the appropriate (unpaired) t test or its nonparametric alternative. Categorical variables were presented as n (%), and group differences were evaluated with the χ² test or Fisher’s exact test as appropriate. Associations between clinicopathological characteristics and gene mutation/status and analyses of clinical predictors of the primary outcome were performed in SPSS 27.0 using logistic regression; for RECIST analyses, SD/PD was coded as 1 and CR/PR as 0; results are reported as odds ratios (ORs) with 95% confidence intervals (CIs). Linear correlations were assessed with the Pearson correlation coefficient (or Spearman when distributions were non-normal). Where applicable, multiple testing was controlled using the Benjamini–Hochberg false discovery rate (FDR) procedure. All tests were two-sided, with P < 0.05 considered statistically significant. Unless otherwise specified (e.g., visualization in volcano plots), inference was based on FDR-adjusted P-values.

The breast cancer diagnostic biopsy samples included in the study were from patients with invasive ductal carcinoma, which was histologically confirmed as luminal B type, HER2 non-amplified, and clinically staged II-III. All patients received at least 8 cycles of AC-T neoadjuvant chemotherapy (Table 4). To exclude the potential impact of anthracyclines and cyclophosphamide on the clinical evaluation of docetaxel, imaging data were collected at three stages: initial admission, after anthracycline and cyclophosphamide treatment, and before surgery (i.e., after completing docetaxel treatment). RECIST version 1.1 was used to evaluate tumor downstaging. The results showed that imaging of patients BC4 and BC13 demonstrated significant tumor shrinkage after anthracycline and cyclophosphamide treatment, with the sum of the longest diameters of the target lesions decreasing by approximately 33.0% and 37.9%, respectively. These were evaluated as PR. In the subsequent neoadjuvant treatment, the tumor size of patient BC4 tended to stabilize with docetaxel treatment and was evaluated as SD, while patient BC13 remained highly sensitive to docetaxel. During docetaxel treatment, tumor changes in patients BC5 and BC13 were significantly different. Patient BC5 was resistant to anthracycline and cyclophosphamide, with the sum of the longest diameter of the tumor decreasing by only 0.06%. However, under docetaxel treatment, the tumor shrank by about 41.3%, showing high sensitivity. In contrast, patient BC10 responded flatly to both treatments, with the tumor size decreasing by only 17.8% during docetaxel treatment. According to the Miller-Payne grading system for evaluating the effect of neoadjuvant chemotherapy, patients BC5 and BC13 reached G5, patient BC4 was G3, and patient BC10 was G2.

After collecting breast cancer tissue from breast surgery, we successfully established 4 primary breast cancer cell lines that could be passaged multiple times. It was observed that the BC4, BC5, and BC13 cell lines proliferated faster, and the cells were dispersed, whereas the BC10 cell line proliferated more slowly, and the cells were tightly bound (Figure 2A). To identify the histological consistency between primary cell lines and primary breast cancer tumor tissues, we performed IHC and immunofluorescence staining of ERα, PR, and HER2 in the four primary cell lines and their corresponding breast cancer tissues. The results showed that the expression of these molecular markers in the primary cell lines was consistent with that in the corresponding breast cancer tissues (Figure 2).

To predict the chemosensitivity of breast cancer cells to docetaxel, we used the CCK-8 assay to detect 4 primary breast cancer cell lines (Figure 3A).Under the same concentration gradient of docetaxel treatment, the average IC₅₀ value of BC10 was significantly higher than those of the other cell lines, while the average IC₅₀ value of BC5 was the lowest. This indicates that BC10 is relatively insensitive to docetaxel treatment, while BC5 shows higher sensitivity (Figure 3B). It is worth noting that BC10 was still assessed as G2 by the Miller-Payne classification after extended treatment cycles. In addition, BC5 was resistant to anthracyclines and cyclophosphamide in neoadjuvant chemotherapy but showed high sensitivity to docetaxel, which is consistent with the results of in vitro experiments.

To explore the resistance mechanism of docetaxel, we performed whole genome RNA sequencing on primary breast cancer cell lines from the docetaxel-resistant group (BC4, BC10) and the sensitive group (BC5, BC13) to analyze gene expression differences and investigate the resistance mechanism of docetaxel. After differential expression analysis of RNA sequencing data, 2603 genes were found to be significantly differentially expressed between the resistant and sensitive groups (|log2FC| > 1, P < 0.05), with up-regulated genes marked in red and down-regulated genes marked in green (Figure 3D).The Venn diagram showed that there were 508 differentially co-expressed genes between the resistant and sensitive groups (Figure 3C).

In addition, we constructed a protein-protein interaction (PPI) network using the STRING online database and imported it into Cytoscape v.3.9.1 for analysis. Based on the changes in protein/gene activity and their regulatory mechanisms, we identified 10 key regulatory factors: CDH2, SNAI2, KRT19, THY1, CLDN1, DSP, CLDN4, SPP1, ITGA1, and COL1A2 (Figure 3E).The GEPIA database was further used to analyze the differential expression of the above genes in breast cancer. The results showed that COL1A2, DSP, KRT19, SPP1, and THY1 were upregulated in breast cancer, and the expression differences were statistically significant (Figure 3F).

The breast cancer transcriptome dataset from the TCGA database was exported, and the TCGA-BRCA data were divided into normal and tumor groups. Differential genes between normal breast tissue and breast cancer tissue were analyzed. The conditions were |log2FC| > 1, P < 0.05, and 4226 differentially co-expressed genes were identified (Figures 4A, B). The first 500 breast cancer survival-related genes obtained from GEPIA2, the 700 taxane (docetaxel) resistance-related genes obtained from GeneCards, and the TCGA-BRCA differentially co-expressed genes were jointly represented in a Venn diagram (Figure 4C). The intersecting genes were identified: PMM2, COL1A2, KRT5. After comparing the 10 significantly differentially co-expressed genes (CDH2, SNAI2, KRT19, THY1, CLDN1, DSP, CLDN4, SPP1, ITGA1, and COL1A2) screened from PCCL cell lines, we found that COL1A2 was a differentially co-expressed gene that simultaneously satisfied the above data sets.

COL1A2 was validated and analyzed (Figures 4D–F), and we found that COL1A2 expression was significantly upregulated in breast cancer (P < 0.001). Kaplan-Meier survival curves of COL1A2 high and low expression groups were plotted. It was observed that overall survival (OS) was significantly lower in the high expression group compared to the low expression group (P < 0.001). The median value was used for estimation, and the median survival of patients with high COL1A2 expression was nearly 2 years shorter than that of patients with low COL1A2 expression. GDSC drug sensitivity data were used to analyze the sensitivity of COL1A2 to docetaxel. We found that when COL1A2 was highly expressed, the IC₅₀ of docetaxel was significantly higher than when COL1A2 was lowly expressed (P < 1×10^-15), indicating that high expression of COL1A2 can induce cells to have stronger resistance to docetaxel. Therefore, we believe that the expression of COL1A2 may serve as a potential biomarker to predict the response of tumor patients to docetaxel chemotherapy, and high expression of COL1A2 may indicate poor docetaxel treatment efficacy in patients.

To confirm the differential expression of COL1A2 in docetaxel-resistant and -sensitive cell lines, we measured the expression level of COL1A2 mRNA in BC4 and BC10 cell lines, which are resistant to docetaxel (DTX), at standardized doses using qRT-PCR analysis. The results showed that the expression of COL1A2 in resistant cell lines (BC4, BC10) was significantly higher than in sensitive cell lines (BC5, BC13) (Figure 4D).

To study the effect of COL1A2 on drug resistance in patients with luminal B HER2-negative breast cancer receiving neoadjuvant chemotherapy (NACT), we performed immunohistochemistry (IHC) to detect the expression level of COL1A2 in primary tumor tissues from 54 patients treated with the AC-T regimen. The results (Table 5) showed that among the 22 patients with Miller-Payne grades G1 and G2, 54.55% (12/22) had increased COL1A2 expression; while among G3 patients, 40.00% (6/15) had increased COL1A2 expression, and only 11.76% (2/17) of patients with G4/G5 had increased COL1A2 expression. Additionally, increased COL1A2 expression was mainly found in the cytoplasm and extracellular matrix of cancer tissues. It is worth noting that, in patients with Miller-Payne grades G1 and G2, the neoadjuvant chemotherapy (NACT) treatment effect was better in areas with low COL1A2 expression. These findings suggest that COL1A2 expression may be associated with the response and resistance of breast cancer patients to neoadjuvant chemotherapy, providing an important experimental basis for subsequent studies.

Interestingly, we observed the IHC experimental results and found that in areas with high COL1A2 expression, the extracellular matrix of tumor cells showed strong staining. Tumor cells surrounded by type I collagen deposition responded less to chemotherapy, whereas in areas with low COL1A2 expression, the staining was more uniform, and the tumor cells responded better to chemotherapy (Figures 5E–G).

We collected 54 clinical tissue samples, all of which had complete treatment follow-up data and gene expression data. The patient characteristics were as follows: 31 patients (57.40%) were under 50 years old at diagnosis, more than 60% of the patients (33 patients) were premenopausal at diagnosis, and 11 patients (20.37%) were locally advanced at diagnosis. Lymph node staging: N0 (28 patients, 51.85%), N1 (14 patients, 25.93%), N2/N3 (11 patients, 20.37%). Clinical staging: stage I (5 patients, 9.26%), stage II (27 patients, 50.00%), stage III/IV (22 patients, 40.74%). After receiving neoadjuvant chemotherapy, the ratio of patients who achieved pathological complete remission (pCR) to those who did not achieve pCR was approximately 1:3. The 3-year follow-up showed that the ratio of disease-free recurrence to disease-related recurrence (including local recurrence and distant metastasis) was approximately 3:1.

To further evaluate the clinical significance of COL1A2 in docetaxel chemotherapy resistance in breast cancer, we divided patients into two stages according to the imaging B-ultrasound results at different stages of NACT treatment and the RECIST guidelines: anthracycline combined with cyclophosphamide treatment stage and docetaxel treatment stage. According to the RECIST criteria, the treatment effect was divided into a sensitive group (CR, PR) and a resistant group (SD, PD), and the relationship between COL1A2 expression and clinical efficacy during neoadjuvant therapy in patients with primary breast cancer was evaluated (Table 6). The results showed that during docetaxel treatment, COL1A2 was highly expressed in 17 patients (85%) in the resistant group, while only 3 patients (15%) in the sensitive group were lowly expressed. Moreover, the expression level of COL1A2 was significantly negatively correlated with the remission rate of docetaxel treatment (χ²=11.153, P = 0.001), while during the treatment of anthracycline combined with cyclophosphamide, the expression level of COL1A2 was not significantly correlated with its remission rate (χ²=0.374, P = 0.572). The IHC score was converted into a violin plot to evaluate the efficacy of docetaxel treatment during NACT, the total efficacy of treatment, the Miller-Payne grading evaluation of pathology after neoadjuvant treatment, and the pCR assessment. It was found that the median COL1A2 score decreased in order according to PD~CR; G1~G5; non-pCR~pCR status (Figures 5I–L).

Univariate logistic regression analysis showed that in HR+/HER2− breast cancer patients, high COL1A2 expression levels were significantly associated with age (OR = 3.6; 95%CI: 1.129-11.478; P = 0.03), menopausal status (OR = 7.0; 95%CI: 1.993-24.581; P = 0.002), tumor size (T4 vs The expression level of COL1A2 was significantly correlated with the remission rate during docetaxel treatment (OR = 8.608; 95%CI: 1.325-55.928; P = 0.024). In addition, high expression of COL1A2 was significantly associated with poor response rate (OR = 7.356; 95%CI: 1.055-51.316; P = 0.044) and recurrence-free survival rate of docetaxel (DTX) chemotherapy (Table 7).