Optimizing the ER Cutoff in HER2‐Positive Breast Cancer: ER ≥ 50% Predicts Low pCR Rates and Resistance to Antibody‐Drug Conjugates in the Neoadjuvant Setting

ABSTRACT

Background

Estrogen receptor (ER)‐positive/HER2‐positive breast cancer demonstrates significantly lower rates of pathological complete response (pCR) to neoadjuvant HER2‐targeted therapies compared to ER‐negative/HER2‐positive disease. However, the optimal ER positivity cutoff for clinically meaningful patient stratification remains undefined.

Methods

We analyzed a retrospective cohort of 741 HER2‐positive breast cancer patients treated with neoadjuvant chemotherapy plus dual HER2 blockade (trastuzumab and pertuzumab) at Sun Yat‐sen University Cancer Center. The optimal ER cutoff for predicting pCR was determined by ROC analysis with Youden index and validated by bootstrap resampling (1000 iterations). Transcriptomic data from TCGA and SCAN‐B cohorts were analyzed to characterize biological differences between ER subgroups. Drug response was predicted using the oncoPredict algorithm, and cancer dependency was assessed using DepMap CRISPR screening data.

Results

ER ≥ 50% positivity was identified as the optimal predictive cutoff (bootstrap 95% CI: 25.0%–77.5%) and was an independent predictor of significantly lower pCR rates in multivariate analysis (OR = 0.27; 95% CI: 0.19–0.40; p < 0.001). Transcriptomic analysis revealed that ER ≥ 50% tumors are characterized by activated estrogen response signaling, downregulated cell cycle and immune pathways. Predicted resistance to trastuzumab, T‐DM1, and T‐DXd was consistently enriched in this subgroup, consistent with lower HER2 and CD16A expression observed in ER ≥ 50% tumors. CCND1, a canonical transcriptional target of ESR1, was significantly upregulated in ER ≥ 50% tumors across both cohorts, and ER + /HER2 + cell lines exhibited significantly higher CCND1 and CDK4 dependency scores in DepMap CRISPR screening (p < 0.05), supporting activation of the ESR1‐CCND1‐CDK4/6 axis in this subgroup.

Conclusions

ER ≥ 50% positivity defines a clinically and biologically distinct HER2‐positive subgroup with poor response to standard neoadjuvant therapy and predicted resistance to ADCs, consistent with lower HER2 and CD16A expression in this subgroup. The convergent transcriptomic and functional evidence for ESR1‐CCND1 axis activation provides mechanistic support for combining CDK4/6 inhibitors with endocrine and anti‐HER2 therapies to improve outcomes in this resistant subgroup.

1. Introduction

Human epidermal growth factor receptor 2 (HER2)‐positive breast cancer represents approximately 20% of all breast malignancies, 55%–70% of which are also estrogen receptor (ER)‐positive [1, 2, 3]. Compared to ER⁻/HER2⁺ tumors, ER + /HER2 + breast cancer generally presents with lower tumor grade and distinct metastatic patterns, favoring bone over lung metastasis [2, 3]. Additionally, ER+/HER2+ cases are more frequently categorized as Luminal A or B rather than HER2‐enriched subtypes, correlating with superior overall and progression‐free survival [2, 3, 4].

Neoadjuvant therapy is a cornerstone in the management of HER2 + breast cancer, where achieving a pathological complete response (pCR) is strongly associated with improved prognosis [5, 6]. Although patients with ER⁺/HER2⁺ breast cancer have better overall survival, their pCR rates are significantly lower than those with ER‐/HER2⁺ disease, a discrepancy partly attributed to crosstalk between ER and HER2 signaling [7]. While high ER positivity is known to predict lower pCR rates [1, 3, 8], the definition of “high” ER positivity varies across studies [1, 8, 9, 10]. Consequently, the optimal ER cutoff for stratifying patients in the neoadjuvant setting remains undefined.

The emergence of antibody‐drug conjugates (ADCs) has further transformed the treatment landscape of HER2‐positive breast cancer. Trastuzumab deruxtecan (T‐DXd) has demonstrated superior efficacy over T‐DM1 in the residual disease setting (DESTINY‐Breast05) and is now established as a standard‐of‐care option in both early and metastatic HER2‐positive disease [11, 12]. However, subgroup analyses from DESTINY‐Breast11 suggest that hormone receptor co‐expression may attenuate ADC efficacy [13], raising important questions about whether a biologically meaningful ER threshold could predict ADC resistance and guide treatment escalation strategies in this population. Defining such a threshold therefore has direct implications not only for neoadjuvant chemotherapy optimization but also for the rational deployment of ADC‐based regimens.

In this study, we comprehensively analyzed a large clinical cohort of 741 HER2‐positive breast cancer patients treated with neoadjuvant dual HER2 blockade to determine the optimal ER positivity cutoff for predicting pCR. We further utilized TCGA and SCAN‐B genomic and transcriptomic data to characterize the biological differences between ER‐high and ER‐low HER2‐positive tumors, assessed predicted resistance to chemotherapy and antibody‐drug conjugates, and identified novel therapeutic vulnerabilities using CRISPR screening data from the Cancer Dependency Map. Together, these analyses aim to establish a biologically‐anchored ER threshold with direct implications for treatment stratification in HER2‐positive breast cancer.

2. Method

2.1. Patient Selection

Female patients with HER2‐positive breast cancer who received chemotherapy plus trastuzumab and pertuzumab‐based neoadjuvant therapy followed by subsequent breast surgery at Sun Yat‐sen University Cancer Center (SYSUCC) before July 10th, 2025, were retrospectively enrolled in this study. All patients were 18 years of age and had histologically confirmed invasive breast cancer. Patients presenting with distant metastasis or occult breast cancer were excluded. This retrospective study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of Sun Yat‐sen University Cancer Center (SYSUCC). Written informed consent was waived by the Ethics Committee because the study used pseudonymized data and provided participants with the opportunity to opt out.

2.2. Clinicopathological Assessments

Demographics data, clinical and pathological tumor stages (according to the 8th edition of the American Joint Committee Cancer Staging Manual [14]), and the specific neoadjuvant treatment regimens were retrospectively collected. Estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki‐67 levels were evaluated in pre‐treatment core biopsies and at surgery by SYSUCC pathologists, adhering to the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guidelines and the International Ki67 in Breast Cancer Working Group recommendations [15, 16, 17]. Pathological complete response (pCR) was defined as the absence of residual invasive cancer in the breast and axillary lymph nodes (ypT0N0 or ypTisN0).

2.3. Differential Expression and Pathway Analysis

Gene expression data from The Cancer Genome Atlas (TCGA) were downloaded from Gene Expression Omnibus (GEO, accession number GSE62944) [18]. Survival data and PAM50 intrinsic subtypes for the TCGA cohort were retrieved from Liu et al. and our previous study, respectively [19, 20]. Gene expression and mutation data of the Sweden Cancerome Analysis Network‐Breast (SCAN‐B) were retrieved from Staaf et al. [21] and Brueffer et al. [22].

Differential gene expression analysis was performed using R package DESeq2 (version 1.44.0). Gene set enrichment analysis (GSEA) was executed using R package ClusterProfiler (version 4.14.3) with default settings. 50 hallmark gene sets from the Molecular Signature Database (MsigDB, version 7.5.1) were used for GSEA. Trastuzumab resistance, T‐DM1 resistance, and T‐DXd resistance signatures were retrieved from GEO with accession numbers GSE50948, GSE243375 and GSE269495, respectively [23, 24, 25].

2.4. Prediction of Drug Response

The R package oncoPredict (version 1.2) was used to predict clinical drug response in TCGA and SCAN‐B cohorts [26]. Gene expression and response data for 545 drugs across 829 cell lines were downloaded from cancer therapeutics response portal (CTRP) and utilized as the training data. Default settings of the “calcPhenotype” function in oncoPredict were used, with the following exceptions: minNumSamples = 20, powerTransformPhenotype = FALSE, and removeLowVaringGenesFrom = homogenizeData.

2.5. Curation of Cancer Dependency Map Database

The gene expression and CERES dependency scores of 4 ER⁺/HER2⁺ and 12 ER⁻/HER2⁺ breast cancer lines were retrieved from the Cancer Dependency Map project (DepMap, version 25Q2). The CERES dependency score (scored 0–1, where higher values indicate greater essentiality) measures the effect size of knocking out a gene, where a higher score indicates greater gene essentiality for cell survival. A dependency score cutoff of 0.5 was employed to select genes essential for the survival of ER+/HER2+ cell lines.

Genes selected had to be expressed in 80% of the ER ≥ 50% samples from TCGA and SCAN‐B. These were then screened in the Drug‐Gene Interaction Database (DGIdb 4.0) to identify druggable genes [27]. Genes previously reported as essential for normal cells were discarded [28, 29]. The remaining genes were queried against the OpenTargets website to distinguish established targets from novel ones [30].

2.6. Statistical Analysis

All statistical analyses were performed using R software (version 4.4.1). Clinicopathological characteristics were summarized using frequencies and percentages for categorical variables, and median with interquartile range (IQR) for continuous variables. The optimal ER cutoff for predicting pCR was determined using R package pROC (version 1.18.5) based on Youden index (J) method [31]. To assess the stability of the derived cutoff, bootstrap resampling (1000 iterations) was performed using the R package boot (version 1.3–31). In each iteration, a new ROC curve was constructed from a resampled dataset of the same size as the original cohort (with replacement), and the Youden‐optimal cutoff was recalculated. The 95% confidence interval (CI) of the optimal cutoff was estimated using the percentile method. A clinical cutoff of 50% was considered statistically equivalent to the Youden‐derived threshold of 55% if the 95% CI of the bootstrap distribution encompassed 50%. The two‐tailed Wilcoxon rank‐sum test and chi‐squared test were used to evaluate the significance of difference for continuous and categorical variables, respectively. Univariate and multivariate logistic regression analyses were performed using the glm function in R package stats (version 4.4.1) to identify independent predictors of pCR. Kaplan–Meier survival curves were plotted using R package survminer (version 0.5.0), and the log‐rank test was used to assess survival significances. A two‐sided p < 0.05 was considered statistically significant.

3. Results

3.1. Clinicopathological Characteristics of the Cohort

A total of 741 HER2‐positive breast cancer patients who received neoadjuvant dual HER2 blockade (trastuzumab and pertuzumab) with chemotherapy were enrolled in the study. The overall pathological complete response (pCR, defined as ypT0/TisN0) rate for the cohort was 42.1%. The median age at diagnosis was 49.9 years (IQR 40.6–55.8, Table 1). 90.7% of the tumors had tumor size > 2 cm. 90.6% of the patients presented with axillary node involvement.

TABLE 1.

Clinicopathological characteristics of the 741 patients in SYSUCC.

Characteristics	No. of patients (%)
Age (years), median (IQR)	49.9 (40.6–55.8)
BMI (kg/m2), median (IQR)	23.4 (21.5–25.6)
Menopausal status, n (%)
Premenopausal	416 (56.1)
Postmenopausal	325 (43.9)
Pretreatment clinical T stage, n (%)
T1	69 (9.3)
T2	463 (62.5)
T3	145 (19.6)
T4	64 (8.6)
Pretreatment clinical lymph node status, n (%)
N0	70 (9.4)
N1	244 (32.9)
N2	197 (26.6)
N3	230 (31.0)
Pretreatment stage, n (%)
IA	4 (0.5)
IIA	82 (11.1)
IIB	178 (24.0)
IIIA	215 (29.0)
IIIB	32 (4.3)
IIIC	230 (31.0)
ER
< 1%	320 (43.2)
≥ 1%	421 (56.8)
< 50%	466 (62.9)
≥ 50%	275 (37.1)
PR
< 1%	343 (46.3)
≥ 1%	398 (53.7)
Pretreatment Ki67, median (IQR)	40% (30%–60%)
Neoadjuvant therapy regimens, n (%)
TCbHP*6	448 (60.5)
THP*6	32 (4.3)
EC*4‐THP*4	198 (26.7)
ECHP*4‐THP*4	51 (6.9)
Others	12 (1.6)
ypCR
Yes	312 (42.1)
No	429 (57.9)

Based on immunohistochemistry, 56.8% (421/741) of the tumors were ER‐positive (ER percentage ≥ 1%), with 65.3% (275/421) of these showing ER positivity ≥ 50%. The most common neoadjuvant regimens were paclitaxel‐carboplatin with trastuzumab and pertuzumab (TCbHP × 6, 60.5%), and doxorubicin‐cyclophosphamide followed by paclitaxel with trastuzumab and pertuzumab (EC × 4−THP × 4, 26.7%).

3.2. Determination of the Optimal ER Cutoff for Predicting pCR

ER‐positive tumors had significantly lower pCR rate (48.5% vs. 70.3%, p < 0.001) than ER‐negative tumors (Figure 1a). We compared the pCR rates across different ER positivity ranges against ER‐negative tumors. PCR rates were significantly lower in tumors with ER positivity in the 60–69, 80–89 and 90–100 ranges (Figure 1b).

FIGURE 1.

Determination of the optimal cutoff of ER positivity for predicting pCR. (a) Pathological complete response (pCR) rates stratified by traditional ER status (≥ 1% vs. < 1%). Statistical significance was assessed using the Chi‐squared test. (b) pCR rates across different ranges of ER positivity. The percentage of pCR is labeled for each group. Groups showing a significantly different pCR rate compared to the ER‐negative group (< 1%) are indicated by an asterisk (*). (c) Receiver Operating Characteristic (ROC) curve analysis of ER positivity as a predictor of pCR following neoadjuvant therapy. The optimal cutoff was determined using the Youden index (J) method. (d) pCR and non‐pCR patient counts across various ER positivity cutoffs. The significance of differences was assessed using the Chi‐squared test.

To formally determine the optimal prognostic threshold, we utilized a Receiver Operating Characteristic (ROC) curve analysis. Based on the Youden Index method, 55% (bootstrap 95% confidence interval: 0.25–0.775) was identified as the statistically derived cut‐point (Figure 1c). The confidence interval encompassed the clinically established threshold of 50%, indicating that these two cutoffs are statistically equivalent. Furthermore, the number of patients classified by the ER ≥ 50% and ER ≥ 55% thresholds was comparable (Figure 1d), and 50% represents a more widely adopted and practically actionable threshold in both clinical and pathological practice. Therefore, 50% was selected as the final clinically pragmatic threshold for all subsequent analyses.

Patients in the ER ≥ 50% group, compared to the ER < 50% group, were characterized by lower median age, lower percentage of menopausal status, and earlier clinical tumor stage (Table S1). The ER ≥ 50% tumors also exhibited lower Ki67 levels and higher PR positivity (Table S1).

Univariate logistic regression analysis demonstrated that pCR was significantly correlated with menopausal status, clinical T stage, clinical N stage, ER positivity, PR positivity and neoadjuvant therapy regimen (Table 2). Multivariate analysis identified ER positivity (ER ≥ 50% vs. ER < 50%) as an independent negative predictor of pCR, with an odds ratio of 0.27 (95% CI, 0.19–0.40; p < 0.001). Other significant negative predictors included advanced clinical T stage (T4 vs. T1; OR = 0.37; 95% CI, 0.17–0.78; p = 0.010), advanced clinical N stage (N2 vs. N0; OR = 0.44; 95% CI, 0.23–0.82; p = 0.011), and the EC × 4−THP × 4 regimen (vs. TCbHP × 6; OR = 0.59; 95% CI, 0.41–0.85; p = 0.005).

TABLE 2.

Univariate and multivariate analyses of factors predicting pCR after neoadjuvant therapy. Factors with p < 0.05 were included in the multivariate analysis.

Characteristics	Univariate analysis		Multivariate analysis
	OR (95% CI)	p	OR (95% CI)	p
Age (years), median (IQR)	1.00 (0.99–1.02)	0.701
BMI (kg/m2), median (IQR)	0.99 (0.95–1.03)	0.619
Menopausal status, n (%)
Premenopausal	Ref	—	Ref	—
Postmenopausal	1.37 (1.02–1.84)	0.038	1.24 (0.89–1.72)	0.211
Pretreatment clinical T stage, n (%)
T1	Ref	—	Ref	—
T2	0.97 (0.57–1.61)	0.896	0.93 (0.52–1.62)	0.791
T3	0.89 (0.49–1.58)	0.683	0.81 (0.43–1.52)	0.513
T4	0.41 (0.20–0.82)	0.013	0.37 (0.17–0.78)	0.010
Pretreatment clinical lymph node status, n (%)
N0	Ref	—	Ref	—
N1	0.70 (0.39–1.22)	0.215	0.69 (0.36–1.27)	0.243
N2	0.44 (0.24–0.78)	0.006	0.44 (0.23–0.82)	0.011
N3	0.55 (0.30–0.96)	0.040	0.50 (0.26–0.93)	0.032
Pretreatment stage, n (%)
IA	Ref	—
IIA	NA	0.975
IIB	NA	0.975
IIIA	NA	0.974
IIIB	NA	0.972
IIIC	NA	0.974
ER
< 50%	Ref	—	Ref	—
≥ 50%	0.23 (0.16–0.31)	< 0.001	0.27 (0.19–0.40)	< 0.001
PR
< 1%	Ref	—	Ref	—
≥ 1%	0.35 (0.26–0.48)	< 0.001	0.70 (0.48–1.01)	0.058
Pretreatment Ki67, median (IQR)	1.98 (0.91–4.33)	0.086
Neoadjuvant therapy regimens, n (%)
TCbHP*6	Ref	—	Ref	—
THP*6	0.82 (0.40–1.74)	0.595	0.78 (0.35–1.79)	0.550
EC*4‐THP*4	0.46 (0.33–0.64)	< 0.001	0.59 (0.41–0.85)	0.005
ECHP*4‐THP*4	0.68 (0.38–1.23)	0.201	0.68 (0.36–1.28)	0.226
Others	0.56 (0.17–1.82)	0.324	0.64 (0.18–2.28)	0.485

3.3. Genomic and Transcriptomic Comparison Between ER‐High and ER‐Low Groups

To investigate the biological differences underlying the reduced pCR rate in the ER ≥ 50% group, we compared genomic and transcriptomic features using TCGA and SCAN‐B datasets. Patients with ER‐high positivity exhibited better overall survival (Figure 2a,b). The molecular subtype distribution differed significantly, with the ER < 50% group being predominantly HER2‐enriched, a subtype significantly less frequent in the ER ≥ 50% group (Figure 2c,d). Conversely, the proportions of Luminal A and Luminal B subtypes were significantly higher in the ER ≥ 50% group. ER ≥ 50% tumors showed a significantly lower frequency of TP53 alteration (24.3% vs. 57.1%, p < 0.001, excluding nonsense mutations) (Figure 2e, Figure S1a). No significant difference was observed in the alteration frequency of PIK3CA, mTOR and AKT1 (Figure S1e, Figure S1a,b).

FIGURE 2.

Comparison of genomic and transcriptomic features between ER ≥ 50% and ER < 50% subgroups. (a, b) Kaplan–Meier survival analysis of Overall Survival (OS) for patients stratified by ER ≥ 50% (high) and ER < 50% (low) groups in The Cancer Genome Atlas (TCGA) and Sweden Cancerome Analysis Network‐Breast (SCAN‐B) cohorts. Differences were analyzed using the Log‐rank test. (c, d) Distribution of PAM50 intrinsic molecular subtypes for ER ≥ 50% (high) and ER < 50% (low) tumors in TCGA and SCAN‐B. (e) Comparison of somatic mutation profiles for the top 20 altered genes in TCGA tumors, stratified by ER ≥ 50% and ER < 50% status. Genes with a significantly different alteration frequency between the two groups are marked with an asterisk (*). Statistical significance was determined using the Chi‐squared test. (f, g) Gene Set Enrichment Analysis (GSEA) plots illustrating the differential enrichment of 50 hallmark pathways between the ER ≥ 50% and ER < 50% groups in the TCGA and SCAN‐B cohorts. Top 20 pathways ranked by adjusted p‐value were shown. (h–k) Comparison of messenger RNA (mRNA) expression levels for ESR1, ERBB2, CCND1, and CD16A in TCGA and SCAN‐B cohorts. Statistical significance was assessed using the Wilcoxon rank‐sum test.

Pathway analysis demonstrated that ER ≥ 50% tumors had activated estrogen response pathways (Figure 2f,g, Tables S2 and S3). Cell cycle related pathways (Hallmark E2F Targets and Hallmark G2M checkpoint) and immune related pathways (Hallmark Interferon Gamma Response and Hallmark Inflammatory Response) were inhibited (Figure 2f,g, Figure S1d). The Hallmark PI3K AKT mTOR Signaling pathway, whose activation was reported to be associated with resistance to anti‐HER2 therapy, was observed to be downregulated in ER ≥ 50% group (Figure 2f,g, Figure S1c), despite no significant difference in phosphorylated protein levels of PI3K, mTOR, and AKT (Figure S1i). Differential gene expression analysis found ER ≥ 50% group had higher expression of ESR1 and CCND1, a canonical transcriptional target of ESR1, in both TCGA and SCAN‐B cohorts. ER ≥ 50% group also had lower expression of HER2 and FCGR3A/CD16A (Figure 2h–k, Figure S1e–h).

To address the possibility that the observed biological differences reflect intrinsic molecular subtype rather than ER expression per se, we performed a subtype‐stratified analysis within the HER2‐enriched PAM50 subtype in the SCAN‐B cohort (ER ≥ 50%: n = 159; ER < 50%: n = 199). Within this subtype, ER ≥ 50% tumors continued to show significantly activated ESR1 signaling, higher CCND1 expression, and suppressed E2F targets and G2M checkpoint pathways (Figure S2a,b) demonstrating that the transcriptomic differences associated with ER ≥ 50% persist independently of molecular subtype.

3.4. Prediction of Drug Response and Resistance

Using oncoPredict to predict patients' drug response based on gene expression data, we found that ER‐high patients were more sensitive to endocrine therapies (Fulvestrant and Tamoxifen, Figure 3a,b). Conversely, the ER ≥ 50% group was predicted to be more resistant to taxane chemotherapy (docetaxel and paclitaxel, Figure 3a,b), which is consistent with the lower pCR rate observed in our clinical cohort (Table 2). Furthermore, patients in the ER‐high group were predicted to be more resistant to tyrosine kinase inhibitors, lapatinib and neratinib (Figure 3a,d, Figure S3).

FIGURE 3.

Prediction of resistance to chemotherapy and Antibody‐Drug Conjugates (ADCs) in the ER‐high group. (a) Venn diagram illustrating the overlap of predicted drug sensitivity and resistance for ER ≥ 50% tumors between the TCGA and SCAN‐B cohorts, based on oncoPredict modeling. (b–d) Comparison of predicted sensitivity (IC50) to representative chemotherapy agents (b), endocrine therapy (c), and tyrosine kinase inhibitors (d) between ER ≥ 50% (high) and ER < 50% (low) groups. Statistical analysis was performed using the Wilcoxon rank‐sum test. (e–g) Gene Set Enrichment Analysis (GSEA) of resistance signatures for trastuzumab (e), T‐DM1 (f), and T‐DXd (g) in TCGA and SCAN‐B cohorts, showing enrichment in the ER ≥ 50% group.

Investigating resistance to antibody‐drug conjugates (ADCs), we found the ER ≥ 50% group was significantly enriched for resistance signatures against trastuzumab, T‐DM1, and T‐DXd resistant signature (Figure 3e–g, Table S4). This suggests that ER‐high patients may exhibit higher resistance to ADCs compared to the ER‐low group, and was consistent with the low expression of HER2 and CD16A in ER‐high patients, which are two established resistance mechanisms to ADCs [32, 33].

3.5. ESR1‐CCND1 Axis Activation and Exploratory Dependency Analysis in ER ⁺/HER2 ⁺ Breast Cancer

To identify potential therapeutic targets, we analyzed the Cancer Dependency Map (DepMap) database, searching for essential genes in ER⁺/HER2⁺ breast cancer cell lines [34]. We identified 82 genes with median dependency scores > 0.5 in ER+/HER2+ breast cancer cell lines that were predicted to be nonessential for normal cells (Figure 4a). Of these genes, 40 were targets of approved or experimental drugs (Table S5).

FIGURE 4.

Prioritization of CDK4 as a novel therapeutic target for ER‐high HER2‐positive breast cancer. (a) Schematic illustrating the pipeline for novel target selection using the Cancer Dependency Map (DepMap) and DGIdb (Drug‐Gene Interaction Database). (b) CERES dependency scores for the 82 top candidate genes in 4 ER⁺/HER2⁺ cell lines compared to 12 ER⁻/HER2⁺ cell lines. Genes are ranked by the median dependency score in the ER⁺/HER2⁺ cell lines. Genes with significantly different dependency scores between the two subgroups are marked with an asterisk (*). Statistical analysis was performed using the Wilcoxon rank‐sum test. (c) Comparison of the CERES dependency scores for CDK4 and CCND1 between ER⁺/HER2⁺ and ER⁻/HER2⁺ cell lines. Statistical significance was assessed using the Wilcoxon rank‐sum test. (d) Schematic of Functional Dependency (Graphical Abstract Panel). Upper panel: ER‐low/HER2‐enriched tumors exhibit high proliferation, leading to sensitivity to mitotic inhibitors (Taxanes). Lower panel: ER‐high/Luminal tumors are characterized by low proliferation (low Ki67) but maintain cell survival via “oncogene addiction” to the ER‐Cyclin D1‐CDK4 axis, rendering them resistant to chemotherapy but uniquely vulnerable to CDK4/6 inhibition.

ERBB2 and ESR1 were ranked among the top genes by median dependency score in ER⁺/HER2⁺ cell lines, validating the approach's ability to capture known drivers (Figure 4b,c). CDK4 was ranked as the top gene in the list, and ER⁺/HER2⁺ cell lines showed a significantly higher dependency score on CDK4 and CCND1, the obligate activating subunit of CDK4, than ER⁻/HER2⁺ cell lines (Figure 4b,c), suggesting functional reliance on ESR1‐CCND1‐CDK4 for survival in this subgroup. The list also included multiple ribosome protein genes (RPL30, RPL35A, etc.) and mitochondrial electron transport chain complex genes (NDUFA1, NDUFA4, etc.), although dependency scores for these genes did not significantly differ between the ER⁺ and ER⁻ cell lines (Figure 4b).

4. Discussion

In this retrospective cohort of 741 HER2‐positive breast cancers, we confirm that ER ≥ 50% positivity was inversely associated with response to neoadjuvant therapy, aligning with findings from numerous studies [1, 3, 8]. Our overall pCR rate of 42.1% (for the whole cohort) is consistent with previously reported outcomes for HER2‐positive disease treated with dual blockade [35]. Specifically, our data show that ER⁺/HER2⁺ tumors (pCR rate 48.5%) have significantly lower response rates than ER⁻/HER2⁺ tumors (pCR 70.3%), reflecting known variations attributable to chemotherapy regimens, tumor stage, and HER2 blockade strategy across different trials [1, 3, 8, 35].

A major contribution of this study is the determination of an optimal ER cutoff. Previous studies have used varying thresholds (e.g., 1%, 10%, 30%, 50%, and 70%) [1, 3, 8, 9, 10]. Our comprehensive ER analysis identified 50% as the clinically pragmatic threshold, based on the Youden index, for predicting significantly lower pCR rates in patients receiving neoadjuvant dual HER2 blockade with chemotherapy. This finding is reinforced by the clear biological divergence observed in the TCGA and SCAN‐B datasets: the ER ≥ 50% group exhibited better long‐term survival and distinct PAM50 intrinsic subtypes (predominantly Luminal A/B), compared to the ER < 50% group (predominantly HER2‐enriched), strongly supporting the use of 50% to delineate two clinically and molecularly distinct entities.

A critical clinical implication of establishing the 50% cutoff lies in the management of the ‘gray zone’—tumors with low‐to‐intermediate ER expression (1%–49%). While current ASCO/CAP guidelines classify these tumors as ER‐positive based on a 1% threshold [15], our multi‐omics validation reveals that the ER < 50% population is biologically distinct from the ER ≥ 50% group. Specifically, the ER < 50% cohort was predominantly comprised of the HER2‐enriched intrinsic subtype, a profile historically associated with high sensitivity to HER2‐targeted blockade and chemotherapy. Consequently, our findings suggest that patients with ER 1%–49% phenocopy the ER‐negative population and are likely to derive substantial benefit from standard neoadjuvant chemotherapy plus dual blockade. Recognizing this distinction is vital to ensure that ‘low ER’ patients are not denied effective standard regimens, while identifying that the challenge of chemo‐resistance is specific to the ER ≥ 50% “true luminal” subgroup, which requires the novel therapeutic escalation proposed in this study.

To understand the poor response of the ER ≥ 50% subgroup, we performed integrated genomic and transcriptomic analysis. As anticipated, the ER ≥ 50% tumors displayed activation of estrogen response signaling pathways. Furthermore, these tumors showed significant downregulation of cell cycle‐related pathways (Hallmark E2F Targets and Hallmark G2M checkpoint). Notably, the ER ≥ 50% group showed a significantly higher expression of CCND1 and decreased expression of HER2, both factors previously linked to resistance to HER2‐targeted therapy [36, 37]. Furthermore, subtype‐stratified analysis within the HER2‐enriched PAM50 subtype confirmed that the transcriptomic differences associated with ER ≥ 50% persist independently of molecular subtype, suggesting that ER quantification provides biological information beyond PAM50 classification. As PAM50 testing remains unavailable in many clinical settings, particularly in China, IHC‐based ER quantification offers a practical and universally accessible approach to identifying this biologically distinct subgroup.

Unexpectedly, the PI3K/AKT/mTOR signaling pathway—frequently associated with anti‐HER2 resistance—was downregulated at the transcriptomic level in the ER ≥ 50% group. This is likely attributable to the lower overall HER2 expression in this luminal‐like subgroup, as HER2 signaling is a primary driver of PI3K pathway [38]. Crucially, our drug response prediction confirmed these biological differences, showing that ER‐high tumors were resistant not only to taxanes (consistent with the low pCR rates) but also to multiple TKIs (lapatinib, neratinib).

With ADCs emerging as a standard for HER2‐positive breast cancer treatment [11, 12, 13, 39], we investigated potential predicted resistance using the oncoPredict computational algorithm. We found that the ER ≥ 50% group had enriched predicted resistant signatures for trastuzumab, T‐DM1, and T‐DXd. This finding is consistent with the lower pCR rates reported in HR‐positive patients in the DESTINY‐Breast11 trial [13]. The predicted resistance to ADCs in this subgroup is consistent with lower HER2 and FCGR3A/CD16A expression observed at the transcriptomic level, which may contribute to reduced ADC target density and impaired ADCC activity, respectively [32, 33]. Experimental validation of these associations is warranted in future studies.

The suboptimal efficacy of current HER2‐targeted therapies in the ER‐high subgroup highlights an urgent clinical need for novel therapeutic strategies. Consistent with the high ESR1 activity in ER ≥ 50% tumors, CCND1—a canonical transcriptional target of ESR1 and a key activator of CDK4/6‐mediated cell cycle entry—was significantly upregulated in the ER ≥ 50% group in both TCGA and SCAN‐B cohorts. Furthermore, ER⁺/HER2⁺ breast cancer cell lines exhibited significantly higher CCND1 and CDK4 dependency scores compared to ER⁻/HER2⁺ lines in the DepMap CRISPR screen (p < 0.05), though this analysis was based on a limited number of cell lines (4 ER⁺/HER2⁺ vs. 12 ER⁻/HER2⁺) and should be interpreted as exploratory. Together, these findings support a model in which ESR1‐driven CCND1 overexpression sustains CDK4/6 pathway activity, rendering ER ≥ 50%/HER2 + tumors potentially vulnerable to CDK4/6 inhibition. The DepMap findings are presented as exploratory and supportive of this model, with the primary evidence being the consistent CCND1 upregulation across two independent transcriptomic cohorts. A distinct biological paradox emerged in our analysis of the ER ≥ 50% group, consistent with ESR1‐driven CCND1 upregulation sustaining CDK4/6 pathway activity (Figure 4d): ER‐high tumor cells sustain viability through the ESR1‐CCND1‐CDK4/6 axis despite low proliferative activity, rendering them resistant to taxane‐based chemotherapy but potentially sensitive to CDK4/6 inhibition.

This mechanistic rationale is consistent with emerging clinical data. Trials such as PATINA and MonarchHER2 have demonstrated that adding CDK4/6 inhibitors to anti‐HER2 and endocrine therapy significantly improves PFS in the metastatic setting [40, 41]. Furthermore, neoadjuvant trials (NA‐PHER2, MUKDEN 01, and MUKDEN 01 plus) have shown promising pCR rates (27%–58%) using chemo‐free strategies that combine endocrine therapy, CDK4/6 inhibitors, and anti‐HER2 therapy [42, 43, 44]. Our findings identify the ESR1‐CCND1‐CDK4 axis as a key biological feature of ER ≥ 50%/HER2+ tumors, providing mechanistic support for the combination of endocrine therapy and CDK4/6 inhibitors alongside anti‐HER2 treatment in this subgroup.

Our study has several limitations. Its retrospective, single‐center nature introduces potential selection and confounding bias. We acknowledge that the proportion of patients receiving anthracycline‐based sequential regimens (EC × 4−THP × 4) was lower in the ER ≥ 50% group (21.2% vs. 36.0%); however, the neoadjuvant regimen was included as a covariate in the multivariate analysis, and ER ≥ 50% remained an independent predictor of lower pCR rates (OR = 0.27; 95% CI: 0.19–0.40; p < 0.001), indicating that the observed difference is not solely attributable to regimen imbalance. Although the TCGA and SCAN‐B analyses support the biological distinction of the ER 50% cutoff, clinical validation in an independent, prospectively collected neoadjuvant cohort is required to confirm this threshold's utility for predicting pCR. Furthermore, the wide bootstrap confidence interval (25.0%–77.5%) of the Youden‐derived cutoff reflects the inherent uncertainty in data‐driven threshold estimation. We note that this variability is in part attributable to the bimodal distribution of ER positivity in our cohort, where the majority of tumors clustered at either the low end (≤ 10%) or high end (≥ 80%), with relatively few patients in the intermediate range, limiting the precision of cutoff estimation in the intermediate zone. The adopted 50% cutoff should be prospectively validated in larger, multi‐center cohorts before routine clinical implementation. The predictive value of the ER ≥ 50% threshold requires prospective validation in independent, multi‐center cohorts before clinical implementation. We primarily focused on ER, though PR can modulate ER function. While ER ≥ 50% was the main predictor of poor response, the high pCR rate observed in the small subgroup of ER ≥ 50%/PR‐negative tumors (71.4% vs. 30.1% in ER ≥ 50%/PR‐positive) warrants further dedicated investigation into PR's role in the ER‐high setting. Finally, our mechanistic predictions based on CRISPR screening utilized a limited number of cell lines, necessitating further in vivo and clinical investigation to confirm the efficacy and feasibility of combining CDK4/6 inhibitors with anti‐HER2 and endocrine therapies. Specifically, the DepMap analysis included only 4 ER⁺/HER2⁺ cell lines, which limits statistical power; the CCND1 and CDK4 dependency findings should therefore be interpreted as hypothesis‐generating and require validation in larger experimental models.

5. Conclusion

In summary, our study, based on a large clinical cohort, establishes ER positivity ≥ 50% as the optimal and independently predictive cutoff for identifying HER2‐positive breast cancer patients with significantly lower pathological complete response (pCR) rates following neoadjuvant dual HER2 blockade. This ER‐high subgroup possesses a distinct luminal‐like biology, characterized by ESR1 activation, CCND1 upregulation, and predicted resistance to both taxanes and ADC therapies. Mechanistically, the ESR1‐CCND1 axis emerges as a key therapeutic vulnerability. These findings provide mechanistic support for the clinical investigation of CDK4/6 inhibitors combined with endocrine therapy in the standard anti‐HER2 treatment regimen for patients with ER ≥ 50% tumors.

Author Contributions

Tian Du: conceptualization, data curation, formal analysis, investigation, visualization, writing – original draft, writing – review and editing, funding acquisition. Min Lin: data curation, formal analysis, writing – review and editing. Gehao Liang: conceptualization, formal analysis, investigation, writing – review and editing. Yan Wang: investigation, writing – review and editing. Hao Wu: investigation, writing – review and editing. Zixuan Zhao: investigation, writing – review and editing, supervision, funding acquisition. Luhao Sun: investigation, writing – review and editing. Jun Tang: supervision, funding acquisition, conceptualization, writing – review and editing.

Funding

This work was supported by the National Natural Science Foundation of China, 82373378. Fostering Program for NSFC Young Applicants (Tulip Talent Training Program) of Sun Yat‐sen University Cancer Center, 2026yfd13. The Science and Technology Project in Guangzhou, 2024A04J4151.

Disclosure

During the preparation of this work, the authors used Gemini to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Ethics Statement

This retrospective study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board (IRB, approval number B2026‐628‐01) of Sun Yat‐sen University Cancer Center (SYSUCC). Written informed consent was waived by the Ethics Committee because the study used pseudonymized data and provided participants with the opportunity to opt out.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Genomic and pathway analysis. (A) Somatic mutation profiles of ER ≥ 50% and ER < 50% tumors in the SCAN‐B cohort. (B) Mutation status of key PI3K/AKT/mTOR pathway components (PIK3CA, mTOR, AKT1) in the TCGA cohort. (C, D) GSEA analysis illustrating downregulation of the Hallmark PI3K/AKT/mTOR signaling and Hallmark E2F Targets pathways in ER ≥ 50% tumors in TCGA and SCAN‐B. (E–G) mRNA expression levels of CDK4, CDK6 and FASN in TCGA and SCAN‐B. (H, I) Comparison of total or phosphorylated protein levels of key signaling molecules in the TCGA cohort, determined by RPPA (Reverse Phase Protein Array). Statistical significance for (E–G) was assessed using the Wilcoxon rank‐sum test.

Figure S2: Comparison of transcriptomic features between ER ≥ 50% and ER < 50% subgroups in SCAN‐B HER2‐enriched tumors. (A) Gene Set Enrichment Analysis (GSEA) plots illustrating the differential enrichment of 50 hallmark pathways between the ER ≥ 50% (n = 159) and ER < 50% (n = 199) groups in the SCAN‐B HER2‐enriched cohorts. Top 20 pathways ranked by adjusted p‐value were shown. (B) Comparison of messenger RNA (mRNA) expression levels for CCND1 in SCAN‐B HER2‐enriched cohorts. Statistical significance was assessed using the Wilcoxon rank‐sum test.

Figure S3: Comparison of predicted sensitivity (IC50) to the tyrosine kinase inhibitor neratinib between ER ≥ 50% (high) and ER < 50% (low) groups. Statistical analysis was performed using the Wilcoxon rank‐sum test.

Table S1: Comparison of clinicopathological characteristics between ER ≥ 50% and ER < 50% groups.

Table S2: Differential expression analysis results comparing ER ≥ 50% to ER < 50% tumors in TCGA and SCAN‐B.

Table S3: GSEA results of 50 Hallmark signatures comparing ER ≥ 50% to ER < 50% tumors in TCGA and SCAN‐B.

Table S4: ADC resistance signatures used in the study.

Table S5: Annotation of 16 HER2 positive cell lines, list of essential genes in normal cell and dependency scores for CCND1, CDK4 and other top‐ranked genes in the DepMap analysis.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.