Abstract

Introduction

Retrospective analyses of prospective breast cancer (BC) trials have shown comparable efficacy between chemotherapy administered in the adjuvant versus neoadjuvant settings [1]. Excellent response, defined as achieving pathologic complete response (pCR) to neoadjuvant chemotherapy (NACT), varies across breast cancer (BC) subtypes and is strongly influenced by the treatment regimen used. During the analyzed trials, pCR rates were approximately 50% and 30% for patients with triple-negative and HER2-positive disease, respectively [2]. In recent developments, incorporating immune checkpoint inhibitors into NACT for TNBC resulted in pCR rates of 64.8% [3], while employing dual anti-HER2 blockade yielded pCR rates of 66.2% in HER2-positive disease, depending on hormone receptor status [4]. Excellent response serves as a prognostic indicator for patient survival, especially in triple-negative and HER2-positive disease [5], and neoadjuvant therapy can serve as an in vivo assay for chemotherapy response.

Prediction of therapy response is of clinical interest offering the potential to customize treatment approaches and enhance response rates. With the emergence of new therapies and refined treatment protocols, there arises the question of whether de-escalating local and/or systemic treatment is viable for patients with a strong likelihood of achieving pCR [6]. For example, PET-based imaging can be used to predict pCR in patients with HER2-positive BC during neoadjuvant treatment with dual anti-HER2 treatment [7].

The conventional method for identifying response markers involves correlating genomic measurements from pre-therapeutic samples with clinical outcomes [8]. On-treatment tissue samples offer the opportunity for microscopic confirmation of response, biomarker examination, and tissue provision for translational research. However, the reliability of biopsy procedures and histopathological assessment for predicting on-treatment pCR remains uncertain. The RESPONDER trial examines if vacuum-assisted biopsies can be used to predict response in the breast with a false negative rate below 10% [9].

In neoadjuvant aromatase inhibition for HR+BC, gene expression analysis of on-treatment samples has been shown to predict treatment response and patient survival [10]. Ki-67 immunohistochemistry can indicate the need to switch to neoadjuvant chemotherapy if Ki-67 levels remain elevated during endocrine treatment alone [11, 12]. In the context of neoadjuvant chemotherapy (NACT), we assessed on-treatment response using ultrasound in the GeparTrio (G3) [13] and GeparQuinto (G5) [14] trials. In G3, we could demonstrate that response-guided switch of chemotherapy regimens can improve patient outcome.

During neoadjuvant chemotherapy (NACT), on-treatment samples can be utilized to uncover molecular mechanisms linked to therapy response, such as immune and proliferation signatures [15] and to pinpoint potential markers of resistance/response through comparative gene expression analysis between responders and non-responders [16].

Aim of this retrospective-prospective biomarker study was to evaluate the frequency of residual cancer cells in on-treatment samples from neoadjuvant clinical chemotherapy trials for BC, and to correlate their presence with response to treatment.

We also included the evaluation of two biomarkers: tumor-infiltrating lymphocytes and the proliferation marker Ki-67. Pre- and post-treatment levels of tumor-infiltrating lymphocytes (TILs) can predict response to chemotherapy and survival [17, 18] and chemotherapy may trigger or amplify a cytotoxic immune response [19]. High levels of the proliferation marker Ki-67 can predict a better response to neoadjuvant chemotherapy but also a poorer long-term prognosis due to more aggressive tumor biology [20]. During neoadjuvant aromatase inhibition, on-treatment Ki-67 evaluation also predicts patient outcomes [12]. We hypothesized that an increase in TILs or a decrease in Ki-67 could predict patient outcome after completion of chemotherapy.

Methods

Results

Frequencies of residual cancer cells and their association with patient and tumor characteristics (Fig. 1)

Open in a separate window

Fig. 1

Frequency of on-treatment biopsies without tumor cells according to tumor characteristics. On-treatment biopsies without tumor cells were more frequent in triple-negative and HER2-positive breast cancers (A), in biopsies from G6 trial patients (B), and in smaller tumors (C). There was no statistical association with lymph node status (D), histological grading (E), tumor-infiltrating lymphocytes (F) or histology (G)

Residual cancer cells were present in biopsies of 207 patients (tu+; 70%) after 4 of 8 cycles (G4, G5) and 2 of 6 cycles chemotherapy (G6), respectively. 90 biopsies showed no residual disease (tu−; 30%). The highest frequencies of tu− biopsies were observed in patients with HER2+ (49%) and TNBC (38%) BC (Fig. 1). The highest frequency of tu− patients was observed among patients of the G6 trial (all patients had triple-negative or HER2-positive disease). The frequency of tu− patients was also higher in patients with small tumors. There was no statistically significant association with lymph node status, histological grading, TILs or histologic subtype (Fig. 1).

Frequencies of residual cancer cells and their association with response to treatment (Figs. 2, ​,33)

Open in a separate window

Fig. 2

Association of cancer cells in on-treatment samples with pathological complete response after completion of chemotherapy (pCR,  A) and disease-free survival (B). The rate of pCR was 8% in patients with evidence of cancer cells on-treatment (tu+), but 50% when the biopsy was negative for cancer (tu−) (A). In the Kaplan–Meier analysis of patients with HR−/HER2− BC, residual cancer on-treatment (tu+) and residual disease (RD) after completion of chemotherapy were associated with shorter survival (B)

Open in a separate window

Fig. 3

Distribution of patients with pCR or non-pCR after completion of chemotherapy according to on-treatment biopsies with (tu+) or without (tu−) residual cancer cells. Of the 90 on-treatment samples without residual cancer cells (blue), 50% had a pCR at the end of treatment and 50% had a non-pCR. Of the 207 on-treatment samples with residual cancer (red), 8.2% had a pCR and 91.8% had a non-pCR. NPV=negative predictive value; PPV=positive predictive value; FPR=false positive rate; FNR=false negative rate

In tu− patients a pCR was observed in 50% (45/90) (Fig. 2A). In contrast, only 17 of 207 (8%) patients with positive biopsies (tu+) had a pCR after completion of the full treatment course, and 92% had residual diseases (190/207). A similar association could be observed in the different BC subtypes (Fig. 2A). The distribution of patients with pCR or non-pCR after completion of chemotherapy according to on-treatment biopsies with (tu+) or without (tu−) residual cancer cells is demonstrated in a Sankey plot (Fig. 3) in detail. Sensitivity to predict residual disease was 0.81 (specificity 0.72). The positive and negative predictive values were 0.92 and 0.50, respectively.

In univariate Cox regression analyses, the absence of tumor cells in on-treatment biopsies was associated with a lower risk of relapse in patients with triple-negative disease (Table 2). However, the effect was not statistically significant when adjusted for pCR in a bivariate model (Table 3). The relationship between the presence of residual disease during and/or after chemotherapy and patient survival was demonstrated in a Kaplan–Meier analysis (Fig. 2B). Patients with residual cancer cells during chemotherapy (tu+) and residual disease (RD) after completion of the full course show the highest risk of relapse.

Table 2

Univariate Cox regression models to predict disease-free survival according to residual cancer in on-treatment biopsies during neoadjuvant chemotherapy

Subtype	Covariate	Hazard ratio (95% CI)	P
HR−/HER2−	tu− vs. tu+	0.402 (0.163–0.987)	0.047
HR+/HER2−	tu− vs. tu+	1.039 (0.428–2.525)	0.933
HER2+	tu− vs. tu+	1.77 (0.577–5.43)	0.318

Open in a separate window

Table 3

Bivariate Cox regression models to predict disease-free survival according to residual cancer in on-treatment biopsies during neoadjuvant chemotherapy and after treatment

Subtype	Covariate	Hazard ratio (95% CI)	P
HR−/HER2−	tu− vs. tu+	0.544 (0.217–1.369)	0.196
	pCR vs. no pCR	0.282 (0.083–0.964)	0.043
HR+/HER2−	tu− vs. tu+	1.14 (0.442–2.939)	0.787
	pCR vs. no pCR	0.689 (0.149–3.18)	0.633
HER2+	tu− vs. tu+	2.114 (0.589–7.584)	0.251
	pCR vs. no pCR	0.704 (0.197–2.516)	0.589

Open in a separate window

Patients with false negative predictions, i.e. those without residual tumor cells on-treatment but residual disease after completion of treatment were analyzed (Fig. S1). A higher frequency of false negative predictions was observed in HR+/HER2− disease. There was no statistically significant association with tumor stage, nodal status, grade or TILs. Patients with false positive predictions, i.e. those with residual cancer cells on-treatment, but with pCR after chemotherapy were demonstrated in Fig. S2. This was more frequently observed in HR−/HER2− cases, in the GeparSixto trial, in smaller tumors and in patients with negative clinical lymph node status.

Dynamic change of TILs and Ki-67 and its association with patient outcome (Fig. 4)

Open in a separate window

Fig. 4

The number of tumor-infiltrating lymphocytes TILs (A) and Ki67 levels (B) in the on-treatment biopsies are plotted against the baseline sample. Only samples with residual cancer in the on-treatment biopsies are shown. The association of dynamic changes in TILs (ΔTILs) and Ki-67 (ΔKi-67) with pathological complete response (pCR) and disease-free survival (DFS) was evaluated using logistic regression and Cox regression analyses, respectively (C, D). An increase in TILs was associated with the likelihood of pCR across tumor subtypes (C) and with a lower risk of relapse in patients with triple-negative disease (D). No decrease in Ki-67 was associated with a lower likelihood of pCR across subtypes and a higher risk of relapse in patients with triple-negative disease. HR=hazard ratio, OR=odds ratio, CI=confidence interval

Paired TIL data was available for 207 patients (all samples with residual cancer on-treatment), paired Ki-67 data was available for 196 patients. There was no association between the dynamic change in Ki-67 and dynamic change of TILs (Fig. S3).

In patients with residual invasive cancer cells in the on-treatment biopsy an increase of TILs in a subset of patients was observed, while only a few patients showed a decrease (Fig. 4A). An increase of TILs was associated with a higher probability of pCR in the overall study cohort, but not within the BC subtypes (Fig. 4C). It was associated with a lower risk of relapse in patients with triple negative disease.

The proliferation index as measured by Ki-67 immunohistochemistry decreased in most patients during chemotherapy (Fig. 4B). The lack of decrease in Ki-67 was associated with a low probability of pCR in all patients and was associated with a higher risk of relapse in patients with triple negative disease (Fig. 4C).

In bivariate Cox regression analyses in patients with triple-negative disease adjusted for pCR, the dynamic change of TILs was statistically significantly associated with DFS. The change of Ki-67 was not significantly associated with relapse-free survival in a bivariate model adjusted for pCR (Table 4).

Table 4

Bivariate Cox regression models to predict disease-free survival according to pCR and dynamic change in TILs or Ki-67, respectively

	Covariate	Hazard ratio (95% CI)	P
HR−/HER2−	Delta TILs	0.979 (0.959–1.000)	0.048
	pCR vs. no pCR	0.190 (0.025–1.418)	0.105
HR−/HER2−	Delta Ki-67	1.019 (0.996–1.043)	0.098
	pCR vs. no pCR	0.211 (0.028–1.600)	0.132

Open in a separate window

Discussion

In this retrospective research study, we analyzed on-treatment biopsies obtained during neoadjuvant chemotherapy (NACT) for breast cancer (BC).

30% of on-treatment biopsies showed no cancer cells. However, if residual cancer cells were detected, achieving a pathologic complete response (pCR) post-treatment was less likely. This suggests the potential for early treatment adjustment, including alternative chemotherapy agents or targeted therapies, to enhance response rates, particularly within the framework of clinical trials. In the GeparTrio and GeparQuinto neoadjuvant trials [13, 14], treatment was adjusted according to evaluation of on-treatment response using ultrasound and led to improved patient’s survival in GeparTrio. Biopsy procedures might be an additional tool to identify tumors prone to treatment failure early on treatment in future trials.

If no cancer cells were detected during treatment, this information could not reliably predict chemotherapy outcome, as the rate of pCR was 50% in this group in the current study. This suggests that sampling error of the residual disease may have led to a false negative on-treatment sample. False negative prediction was more frequent in HR+/HER2− disease, reflecting a tumor biology with a lower a priori probability for pCR. False positive prediction (cancer cells on-treatment but pCR after therapy) was more frequent in HR−/HER2− tumors, in the GeparSixto trial and in smaller tumors, reflecting patients with a higher a priori probability of pCR.

As our study collected on-treatment samples for translational research rather than for predicting pCR or guiding treatment decisions, its comparability to studies focused on on-treatment pCR prediction may be limited (reviewed in [6]). The reported negative predictive values of these studies vary, ranging from 71% for core needle biopsy (CNB) to as high as 95% for vacuum-assisted biopsy, which offers a larger specimen for analysis. However, with its false negative rates reaching 49.3%, presents a considerable margin for error [28, 29]. However, these investigations aimed to evaluate pathologic complete response (pCR) after completion of neoadjuvant chemotherapy (NACT) before surgery, utilizing minimally invasive methods to identify patients potentially eligible for surgery omission. These trials did not assess the capability to predict pCR early during treatment.

In a report of 40 patients from the ISPY2 trial, the presence of cancer cells in mid-treatment biopsies was associated with a 20% pCR rate, while the pCR rate in patients without invasive cancer cells was 90% [30].

Regarding survival, patients with triple-negative tumors and residual disease both on- and post-treatment had the highest risk of relapse. Patients without the evidence of cancer cells on-treatment but residual disease after the completion of the full course of treatment had a lower risk. This observation could be explained by the fact that the biopsy procedure is more likely to miss smaller tumors or tumors with only a minimal amount of residual disease.

Pre-treatment levels of TILs can be used to predict response to chemotherapy and patient’s survival in BC [17]. Chemotherapy might induce or augment a cytotoxic immune response [19] and an influx of TILs during treatment is associated with better response [31]. Moreover, their presence in surgical specimens is associated with better outcome [18] and a gene signature for prediction of TILs in post-treatment samples is predictive of survival [32].

In this study, we observed an increase of TILs in a subset of patients and only a few cases showed a decrease. An increase of TILs was associated with a higher probability of pCR in all patients and with a lower risk of relapse in patients with TNBC reflecting their known predictive value in triple-negative disease. This observation is particularly interesting, as it can identify patients early on treatment that still harbor invasive tumor residuals, where continuing standard therapy might be the option that is superior to a change in treatment plan. Further investigation in this group of patients could help to refine adapting tumor response into treatment plans and could ultimately allow to move the risk assessment after NACT to an earlier timepoint.

The marker of tumor cell proliferation Ki-67 can be used to predict response to NACT and patient’s survival [20, 33]. High levels are typically associated with better response to cytotoxic treatment but shorter long-term outcome due to more aggressive tumor biology. In the context of neoadjuvant aromatase inhibition, on-treatment evaluation of Ki-67 can predict patient outcome [11, 12].

In this study, most patients showed a decrease in Ki-67 index and this was associated with a higher probability of response in triple-negative disease. It was also associated with a lower risk of relapse in triple negative disease, but not in other subtypes. Bivariate survival analyses demonstrated that these effects were probably due to the association with pCR and its strong association with survival in this subtype.

From a translational research perspective, these observations suggest that on-treatment biopsies could be valuable for studying mechanisms of therapy resistance and predicting failure to achieve pathologic complete response (pCR) after neoadjuvant chemotherapy (NACT). However, they are not suitable for identifying markers of chemotherapy sensitivity, as highly sensitive tumors would not contain residual cancer cells in on-treatment biopsies. To address this, further analyses should focus on revisiting naive biopsies from patients who achieved early pCR. On a molecular level, NACT induces global changes in gene expression [34]. Examples of alterations are genes involved in proliferation, epithelial-mesenchymal transition and metabolic processes [35]. Analysis of serial biopsy samples during chemotherapy allows the characterization of mechanisms of early response and adaption to therapy. In such a study, a decreased expression of genes related to immune response and proliferation could be observed and the downregulation of cell-cycle inhibitors was associated with worse response [15]. The use of on-treatment biopsies for patient stratification should be further explored in the context to clinical trials and should be a part of the study protocol.

Several limitations of the study warrant consideration. Firstly, it's important to note that the biopsy procedure was not strictly standardized according to the study protocol. Detailed records of the sampling procedure were not available in the database of these older clinical trials. This limits the interpretation of the results, as bias due to patient selection (e.g. due to different or absent tumor visibility on ultrasound) or different sampling procedures (e.g. different number and/or diameter of biopsies) cannot be ruled out. At the time of the clinical trials used in this study, no clips were placed in the primary tumor area prior to treatment. The accuracy of the biopsy procedure may be limited. Additionally, variations in the timing of the biopsy procedure across the three trials were inevitable due to differences in study design. In GeparQuattro and GeparQuinto, on-treatment samples were obtained after 4 of 8 cycles, in GeparSixto after 2 of 6 cycles. Moreover, GeparQuattro and GeparQuinto switched from anthracycline to taxane therapy following the biopsy, whereas GeparSixto continued with the same regimen (concurrent taxane/anthracycline). The collection of on-treatment biopsies was not a mandatory part of the study protocol with a potential selection bias and comparably small samples sizes in subgroup analyses. The study was primarily designed to collect material for translational research purposes. It must be considered that patients within this trial had been treated before 2012 which possibly limits accuracy of on-treatment biopsy due to less experienced examiners and less refined examination instruments. Regimen not matching current standards thereby limiting the chances of achieving a pCR, are also able to influence false negative rates and predictive values of on treatment biopsies. Immune checkpoint inhibitors were not part of neoadjuvant chemotherapy in patients with triple-negative breast cancer, and patients with HER2-positive disease did not receive dual blockade.

In summary, our findings show that on-treatment biopsies can effectively predict non-pCR across breast cancer (BC) subtypes when residual cancer is present. This discovery presents potential avenues for tailoring therapy concepts in future clinical trials, such as implementing de-escalation strategies for responders and exploring experimental treatments for non-responders.

A reliable method for predicting chemotherapy response during treatment could greatly impact patient management and improve treatment strategies. By monitoring response, treatment plans can be tailored to individual patients, and if a patient is unlikely to respond well to the initial regimen, adjustments can be made to optimize therapeutic outcomes by switching to more effective drugs or modifying the dose. Early prediction of non-response allows for discontinuing ineffective treatments, avoiding unnecessary toxicity and complications, and better allocation of healthcare resources.

For example, ISPY2 trial is evaluating a combination of mid-treatment MRI and core biopsies to predict response and guide de-escalation of neoadjuvant treatment [36].

Predicting response helps plan the extent of surgery more accurately, potentially reducing the need for extensive procedures like mastectomy. Early prediction of treatment success can improve overall prognostication and support informed decision-making for post-neoadjuvant treatment and observation strategies.

Moreover, analyzing sequential biopsies could be instrumental in identifying molecular markers of therapy resistance to enhance our understanding of tumor biology and its adaptation to therapy. These advancements can ultimately improve overall treatment paradigms, improving the precision and effectiveness, leading to better patient outcomes. Further research is warranted to determine whether more standardized or extensive sampling procedures, or their combination with additional clinicopathological features, could enhance the sensitivity of on-treatment biopsy procedures.

Supplementary Information

Acknowledgements

Abbreviations

Author contributions

The analysis was designed by the members of the translational research subcommittee of the German Breast Group (BS, TK, MvM, FM, CS, ES, PAF, VM, SL and CD). BS and VN has analysed the data. BS and VN had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. All authors interpreted the data. The first draft of the manuscript was written by BS. The decision to submit the manuscript for publication was made by all authors. All authors contributed to the review of the manuscript. No persons other than the listed authors contributed to the writing of the manuscript.

Funding

Open Access funding enabled and organized by Projekt DEAL. This analysis was funded by GBG. The GeparQuattro trial received funding support from Roche and Sanofi-Aventis. The GeparQuinto trial received funding support from GlaxoSmithKline, Novartis, Roche and Sanofi Aventis. The GeparSixto trial received funding support from Cephalon, GlaxoSmithKline and Roche. The funders had no access to the study database and were not involved in the analysis and interpretation of the results. No grant number applicable.

Availability of data and materials

All data needed for this analysis and results generation is included in this published article.

Declarations

Footnotes

References