Pathway activity profiling can predict neoadjuvant endocrine therapy response in HR+ HER2- postmenopausal early stage breast cancer.

Introduction
About 70% of breast cancers are estrogen receptor positive (ER+) [1, 2] and up to 45% of patients do not respond to neoadjuvant endocrine therapy (NET) initially and/or long-term [3]. Signal transduction pathway activity profiling appears promising to predict therapy response more adequately compared to immunohistochemistry (IHC), since this directly relates to tumor biology and pathway signaling as understood from the current paradigm about cancer genomics and dynamics [4]. Previously, OncoSIGNal signal transduction profiling (InnoSIGN BV, The Netherlands), a Bayesian network model using a mRNA-based PCR test, identified ER -IHC positive patients with low ER-pathway activity signaling (ER-PAS) to be associated with primary endocrine resistance [5, 6].
ER, progesterone receptor (PR) and Ki67 IHC are the standard biomarkers available to predict responders and non-responders to NET [7]. However, ER-IHC stains the nuclear protein which may not reflect activity of the pathway on which the tumor growth relies [4, 8]. Other concerns with ER-IHC percentage score are the difficulty of reproducibility at low ER expression and thresholding intensity variation within and between laboratories [9, 10]. Alternatives such as (modified) Allred score [11] or the H-score [12] are not standardly used, partly due to labor intensity.
Proposed explanations for primary or secondary endocrine therapy resistance are ESR1 mutations [13], tumor heterogeneity with ER- clones through conversion from ER + to ER- by for example alternative splicing [14, 15], up- or downregulation of coregulator proteins [2, 16] and/or changes in downstream proteins such as PI3K, AKT, mTOR [13], all potentially affecting the activity of signal transduction pathways.
Therefore, OncoSIGNal signal transduction profiling for ER pathway activity, using mRNA, is studied to predict clinical response to neoadjuvant letrozole in postmenopausal patients, with IHC ER+, HER2- breast cancer treated in the NEOLBC study, who received 2 weeks of letrozole followed by letrozole monotherapy if Ki67 at 2 weeks (Ki67-2 W) was < 1% [17]. Additionally, other pathways such as androgen receptor (AR), MAPK, PI3K, Hh, Notch and TGFβ were analyzed to obtain better insights in possible resistance mechanisms and uncover actionable targets for precision therapy.

Material and methods

Study patients
From 2018 to 2021 the NEOLBC (NCT03283384) included 161 postmenopausal patients with hormone receptor positive (ER-IHC ≥ 50%, PR any), HER2 negative, stage II/III breast cancer as previously described [17]. In brief, the main inclusion criteria were measurable disease, WHO performance status 0–2 and adequate bone marrow, liver and renal function [17]. A biopsy was taken at baseline and after two weeks of neoadjuvant letrozole. If at the two-week biopsy Ki67 IHC was < 1%, i.e. complete cell cycle arrest (CCCA), patients continued letrozole for 6–9 months until surgery. If Ki67-2 W ≥ 1% patients were randomized 1:1 to receive ribociclib with letrozole or standard chemotherapy, see Fig. 1 for the NEOLBC study design overview.

From 96 of the 161 patients formalin-fixed paraffin embedded (FFPE) material was available, of which 30 patients with Ki67-2 W < 1% continued with letrozole as NET until surgery, available samples per timepoint are displayed in Fig. 2. ER-IHC was scored as a percentage of ER-positive tumor cells and categorized into < 10%, 10–50%, 50–75%, 75–90% and > 90%. Clinical response was measured by RECIST 1.1, Miller and Payne score 1–5, pathological complete response (pCR) and CCCA (at two weeks and surgery). All patients gave written informed consent. The study (NCT03283384) was conducted in accordance with the Declaration of Helsinki (October 2013) and approved by the Ethics Committee of the Leiden-The Hague-Delft hospitals in agreement with the Dutch law for medical research involving human subjects.

RNA extraction and assays
Baseline, two week and surgical resection (≥ 30% epithelial tumor cell content) FFPE material of 89 NEOLBC patients was available for real-time quantitative reverse transcription-PCR (RT-qPCR) OncoSIGNal pathway assays (InnoSIGN BV, The Netherlands). After annotation and microdissection, total mRNA was extracted according to the manufacturer’s protocol (VERSANT® Tissue Preparation Reagents kit, Siemens, Germany). RT-qPCR was performed using the SuperScriptTM III PlatinumTM One-Step qRT-PCR kit (Invitrogen, ThermoFisher Scientific, USA). Commercially available G7 OncoSIGNal 96-well PCR plates were processed with a CFX96 Real-Time PCR Detection System (Bio-Rad, USA) for a total of 275 samples. Internal quality control of reference genes confirmed sufficient input for pathway analysis; 2 samples that failed quality control were excluded from pathway analysis.
Subsequently signal transduction pathway activity was estimated for ER, AR, MAPK, PI3K, Hedgehog (Hh), Notch and TGFβ pathways using Bayesian network computational models which infer activity of the corresponding transcription factor complex from the expression of pathway-specific target genes, as described in detail before [5, 18]. The assays present functional pathway activity scores on a normalized 0–100 scale, where 0 corresponds to the lowest and 100 to the highest odds in favor of an active pathway. Since pathway activity is tissue/cell type dependent, a pathway specific reference range (0-100) is established for normal pathway activity based on the pathway distribution in reference tissue, in line with the method described by Keizer et al. [19]., For pathways AR, MAPK, PI3K, Hh, Notch and TGFβ normal epithelial breast tissue obtained from reduction mammoplasty was used as reference material. For the ER pathway score, ER-IHC-negative breast cancer tissue was used as reference. The 95th percentile score of 42 of the reference distribution determined that scores ≤ 42 correspond were considered non-elevated, i.e. within normal range and scores > 42 (including the measurement confidence interval), were therefore considered elevated ER pathway activity in breast tumors of this study cohort.

Statistical data analysis
ER-PAS normality distribution was assessed with Kolmogorov-Smirnov or Shapiro-Wilk test before either paired T-test or Wilcoxon signed rank test were applied, same for independent T-test and Mann Whitney U test. After assessment of sphericity assumption by Mauchly’s test, a repeated measurements ANOVA was used to compare ER pathway activity and ER-IHC between baseline, two-weeks and 26 weeks at surgery samples. Mann Whitney U tests were used for associations with CCCA, clinical and pathological response. Receiver operating characteristic (ROC) and area under the curve (AUC) analysis was used to analyze predictive value of ER-PAS at baseline and two-weeks. Statistical analysis was performed using Statistical Package for Social Sciences (IBM SPSS, version 24.0 and 25.0), Armonk, NY, USA: IBM Corp and R 4.2.1 (http://cran.r-project.org).

Results

ER pathway activity score is more dynamic than ER-IHC
Despite all baseline samples being ≥ 50% ER-IHC, ER-PAS varied considerably with a range of 32–77 at baseline as shown in Fig. 3A. 20% (18/89) of ER-IHC positive samples showed low ER-PAS (32–45), comparable to triple negative breast cancer tissue. Most samples (65/89) were ER-IHC 100%, while ER-PAS varied from 32 to 76. There was no correlation between baseline ER-PAS and ER-IHC (ρ = 0.12; p = 0.27).

Moreover, ER-IHC remained similar between the baseline and two-week biopsy (Z = 0.71, p = 0.48), while ER-PAS showed a reduction in 73/82 (89%) of patients with a mean difference (MD) of -11.03 (95%CI -5.67, -16.4; p < 0.001) between baseline compared to two weeks (mean baseline 55.5, SD 10.3; mean two weeks 44.2, SD 7.5), shown in Fig. 3B, C.
The patients who continued NET in the letrozole monotherapy arm showed continued ER pathway inhibition from two weeks until the surgical resection at 26 weeks (MD -6.59; 95%CI -3.25,-9.92; p < 0.001). The ER-PAS MD from baseline until surgery was − 17.6 ER-PAS (95%CI -12.31,-22.93; p < 0.001), displayed in Fig. 4A. In contrast, ER IHC percentage category did not differ significantly across baseline, 2-week, and surgery timepoints in Fig. 4C (Friedman test, χ2 = 4.50, p = 0.105; n = 11 patients with complete data) (interestingly, the group treated with chemotherapy after randomization - no hormonal therapy - showed initial decrease in ER-PAS after two weeks letrozole, MD -10.44; 95%CI -7.47, -13.4; p < 0.001), but then ER-PAS increased again from two-weeks during chemotherapy (MD 6.17; 95%CI 1.61–10.72; p = 0.03) until ER-PAS was similar to baseline at surgery around 26 weeks shown in Fig. 4B (MD baseline and surgery − 5.01; 95%CI 0.18, -9.84; p = 0.13). Across all timepoints the chemotherapy group (n = 24) showed no significant difference in ER-PAS (ANOVA; F = 0.81, p = 0.45). Thus, ER-IHC is not correlated to ER-PAS, while ER-PAS appears to be a more dynamic and specific marker in response to letrozole therapy and pathway activity.

ER-PAS as predictive factor for NET
There was a trend between baseline ER-PAS and clinical response using the pre-surgery MRI in the letrozole monotherapy group where patients with complete response (CR) appeared to have a higher ER-PAS at baseline compared to those with stable disease (SD) (p = 0.03), as shown in Fig. 5A. There was a trend between baseline ER-PAS and clinical response, given these mixed results in partial responders (PR).
ROC analysis of ER-PAS baseline (n = 24) and two-weeks (n = 21) with RECIST score showed an AUC of 0.704 and 0.730 respectively in the letrozole-group, demonstrating that ER-PAS is an acceptable prognostic factor with an AUC of ≥ 0.7 in this group, further details are available in supplementary Fig. 1, along with potential clinically prognostic cut-off points for ER-PAS at baseline and two-weeks in the letrozole treated subgroup. Due to the trial design, it was not possible to compare predictability for clinical response between CCCA and ER-PAS. There was no association between ER-PAS and pathological response by Miller-Payne in the letrozole group (Fig. 5B).

Both CCCA and non-CCCA group showed ER-PAS decrease after 2 weeks, both p < 0.001 (see Fig. 5A). Although baseline ER-PAS was similar between CCCA (letrozole group) and non-CCCA (randomized groups) p = 0.12, 83% (15 / 18) of the patients with non-elevated ER-PAS (< 42) at baseline had non-CCCA (Ki67 ≥ 1%) after two weeks letrozole therapy (see Fig. 5C) despite ER-IHC positivity (Chi-square 0.158). Ki67 distribution at baseline and surgery are available in supplementary Fig. 2A, B. At closer look, baseline Ki67 IHC showed a statistically significant, but negligible inverse correlation with ER-PAS (Spearman ρ = −0.03, p = 0.004) and there was no association between Ki67 at surgery and baseline ER-PAS (Spearman ρ = −0.14, p = 0.207), added in supplementary Fig. 2C, D.

Targetable alternative pathways in patients with non-elevated ER-PAS and non-CCCA
ER pathway showed most prominent pathway activity score reduction after two weeks of letrozole treatment in the total study cohort. In addition, Hh showed significant pathway inhibition (p < 0.0001), followed by TGFβ (p = 0.011) and PI3K (p = 0.03) while the other pathways maintained similar activity during two-week letrozole treatment (Fig. 6A). Most of the non-elevated ER-PAS patients from Fig. 5C had at least two other pathways that showed increased activity, such as increased PI3K, MAPK and Hh compared to the responder group, shown in the table of Fig. 6B. The pathway profiles of the other patients with elevated ER-PAS, are shown in supplementary Table 1. Increased pathway activity on group level is shown in supplementary Table 2. In summary, OncoSIGNal pathway activity profiling shows potential in improving patient stratification of and identifying alternative actionable pathways for patients at risk for early NET resistance.

Discussion
The current study shows that ER-PAS, both at baseline and 2 weeks, is a more dynamic marker that appears to reflect variable NET response more accurately compared to ER-IHC staining ER-PAS varied considerably at baseline and decreased upon NET, and it remained decreased when letrozole treatment was continued. In contrast, ER-IHC was > 50% at baseline and remained unchanged after 2 weeks letrozole and surgery at 26 weeks. This discrepancy could at least partially be attributed to several biological compensation mechanisms, where the tumor still expresses the ER receptor while the pathway is inactive or repressed, as mentioned before [2, 20].
Additionally, our data suggests that non-elevated ER-PAS at baseline predicted a non-CCCA after two weeks of letrozole (non-responders). This finding is in line with a previous study of our group [6] where, in an ER-IHC+ breast cancer cohort higher ER-PAS correlated to improved NET response [6]. Moreover, 20% of the baseline samples showed low ER-PAS, similar to that of TNBC tissue, which raises the question if these patients (with ≥ 50% ER-IHC) positive would benefit from NET at all [21]. Finally, ER-PAS at baseline and at 2 weeks alone hold predictive value of AUC of 0.7 and 0.73 respectively for clinical response measured by pre-surgery MRI (MRI RECISTv1.1 score). Based on these results ER-PAS could be useful as a clinically relevant negative predictor for resistance to NET or clinical response. Nevertheless, validation in a larger cohort is needed, since the letrozole monotherapy group had a relatively small sample size (n = 27).
Interestingly, ER-PAS increased in the chemotherapy arm after letrozole discontinuation, which is in contrast to the other treatment arms using letrozole. This reactivation of ER signaling highlights a critical insight: in ER-positive breast cancer, continuous inhibition of the ER pathway is essential, as both ER and IGF1/insulin pathway signaling are key drivers of tumor growth [22]. These results support the hypothesis of the NEOLBC trial by de Groot et al. that endocrine therapy with ribociclib may be able to replace chemotherapy in patients who show KI67 ≥ 1% (non-CCCA) after two weeks of NET [17]. Other literature also suggested that endocrine therapy could be a useful treatment addition in the neoadjuvant setting for a select group of ER+ patients to improve the rate of breast conserving treatment/surgery [20, 23].
RT-qPCR assays are a fast and reliable readout of pathway activation that measures expression levels of direct target genes of the respective pathway-associated transcription factor. In contrast, immunohistochemistry of ER, PR, HER2 measures mostly the presence of the signaling proteins (not transcription factors) and therefore do not provide quantitative information on the functional pathway activity. Another study demonstrated this with ESR1 RT-PCR assay, and showed that low-level expression was a determinant for tamoxifen resistance in ER+ breast cancer as well, in line with our results [24]. On the other hand, multiple predictive tools that measure gene expression as well, such as Oncotype DX [25] and MammaPrint [26] have been developed and it would be interesting to compare the predictive value of ER-PAS with these gene expression profile tests.
18 patients in the study cohort with baseline non-elevated ER-PAS showed a switch towards more PI3K, MAPK and Hh pathway activity, suggesting a shift toward non–ER-dependent signaling [27, 28]. Crosstalk between ER and growth factor–driven pathways such as PI3K and MAPK is well established in breast cancer, and activation of these pathways has been implicated in endocrine resistance [29, 30].
Activating PIK3CA mutations occur in approximately up to 40% of primary breast tumors, with higher prevalence in ER-positive subtypes [31, 32]. OncoSIGNal uses RT-qPCR–based expression panels, reflecting downstream transcriptional functional output of these genetic mutations as well. This distinction is relevant because PIK3CA mutations, do not consistently translate into elevated pathway activity, and the measured PI3K-PAS may capture both mutation-driven and non-genomic PI3K signaling. However, mutation status of PIK3CA was not available in this cohort. The question remains whether the elevated activity of alternative pathways can be attributed to genomic predisposition or early adaptive responses.
This reflects the novelty and relevance of OncoSIGNal, where escape/resistance mechanisms could be identified and where low ER pathway activity levels at baseline can be used to identify endocrine responders from non-responders. Altogether these results highlight the need to improve patient stratification for NET resistance and clinical relevance of OncoSIGNal pathway activity profiling. Before using PAS as diagnostic tool for actual clinical recommendations, further validation should be performed due to the small sample size in this current study.

Conclusion
HR + IHC staining does not always reflect an active ER signal transduction pathway activity, which may explain why not all patients respond equally well to NET. Baseline ER pathway activity appears to be a more reliable and dynamic indicator of NET responsiveness than IHC staining. OncoSIGNal could improve identification of patients at risk for early NET resistance, while alternative pathway profiling could lead to other targeted treatment options for patients with ER-positive breast cancer.

Supplementary Information
Below is the link to the electronic supplementary material.