Circulating tumor DNA in neoadjuvant-treated breast cancer reflects response and survival.

BACKGROUND
Pathologic complete response (pCR) to neoadjuvant chemotherapy (NAC) is strongly associated with favorable outcome. We examined the utility of serial circulating tumor DNA (ctDNA) testing for predicting pCR and risk of metastatic recurrence.


PATIENTS AND METHODS
Cell-free DNA (cfDNA) was isolated from 291 plasma samples of 84 high-risk early breast cancer patients treated in the neoadjuvant I-SPY 2 TRIAL with standard NAC alone or combined with MK-2206 (AKT inhibitor) treatment. Blood was collected at pretreatment (T0), 3 weeks after initiation of paclitaxel (T1), between paclitaxel and anthracycline regimens (T2), or prior to surgery (T3). A personalized ctDNA test was designed to detect up to 16 patient-specific mutations (from whole exome sequencing of pretreatment tumor) in cfDNA by ultra-deep sequencing. The median follow-up time for survival analysis was 4.8 years.


RESULTS
At T0, 61 of 84 (73%) patients were ctDNA-positive, which decreased over time (T1-35%; T2-14%; T3-9%). Patients who remained ctDNA-positive at T1 were significantly more likely to have residual disease after NAC (83% non-pCR) compared to those who cleared ctDNA (52% non-pCR; OR 4.33, P=0.012). After NAC, all patients who achieved pCR were ctDNA-negative (n=17, 100%). For those who did not achieve pCR (n=43), ctDNA-positive patients (14%) had significantly increased risk of metastatic recurrence (HR 10.4; 95% CI, 2.3-46.6); interestingly, patients who did not achieve pCR but were ctDNA-negative (86%) had excellent outcome, similar to those who achieved pCR (HR 1.4; 95% CI, 0.15-13.5).


CONCLUSIONS
Lack of ctDNA clearance was a significant predictor of poor response and metastatic recurrence, while clearance was associated with improved survival even in patients who did not achieve pCR. Personalized monitoring of ctDNA during NAC of high-risk early breast cancer may aid in real-time assessment of treatment response and help fine-tune pCR as a surrogate endpoint of survival.


INTRODUCTION
][10] However, it is not yet known if failure to clear ctDNA during therapy could provide guidance for escalation of treatment to prevent early disease recurrence. 11eoadjuvant chemotherapy (NAC) has become a standard-of-care for patients with locally advanced breast cancer. 124][15] Second, NAC may downstage tumors and thus improve chances of breast-conserving surgery. 12,16,17][18][19][20] Pooled analysis by Cortazar and colleagues 21 has shown that patients who achieved a pathologic complete response (pCR, or the absence of residual cancer in the breast and lymph nodes after NAC) have significant survival advantage over those who did not.Standard NAC alone or in combination with other agents has resulted in pCR for 10%-50% of patients depending on subtype. 21Data from the I-SPY 2 TRIAL, a multicenter phase II trial that evaluates investigational drugs in combination with standard NAC (paclitaxel followed by anthracycline treatment), 22 have shown that pCR in women with molecularly high-risk stage II or III tumors, whether from standard or targeted therapies, unequivocally conferred a survival advantage [hazard ratio (HR) of 0.2]. 23hile pCR accurately identifies patients with low risk of relapse, studies have shown that predicting early metastatic recurrence in those with residual disease (non-pCR) is less robust. 21,23For example, survival analysis in the I-SPY 2 TRIAL (median follow-up of 3.8 years) showed that the 3-year distant disease-free survival (DRFS) of patients who achieved pCR was 95%. 23Among non-pCR patients, 22% of experienced metastatic recurrence.In this study, we evaluated the potential role of ctDNA as a biomarker for monitoring of response to NAC and assessed the additive value of ctDNA to further stratify patients with residual disease to predict early metastatic recurrence.We hypothesized that early changes in ctDNA are predictive of response to NAC and that ctDNA dynamics during NAC as well as ctDNA status (positive versus negative) at each time point are associated with patient outcomes.To address these hypotheses, we performed a correlative study in the I-SPY 2 TRIAL to detect ctDNA in serial plasma samples collected before, during, and after NAC. 246][27] The test involves multiplex polymerase chain reaction amplification followed by ultra-deep sequencing to detect tumor-specific mutations (i.e.ctDNA) in cell-free DNA (cfDNA).This approach enables more accurate monitoring of disease burden than prefixed driver mutation panels, as each test reflects tumor heterogeneity at the individual patient level. 5,8,28

Patients
We performed a retrospective ancillary ctDNA study on prospectively collected samples from high-risk early breast cancer patients enrolled in the multicenter neoadjuvant I-SPY 2 TRIAL (NCT01042379).Women with !2.5-cm stage II/III breast cancer were eligible.Patients were screened for metastatic disease by imaging (computed tomography or positron emission tomography) prior to enrollment, and those with de novo metastatic disease were excluded.Restaging scans were not performed after NAC prior to surgery.Eligibility was limited to patients with a Mamma Print high score, and thus the trial was enriched for those with increased risk of metastatic recurrence within 5 years after diagnosis.Patients received standard NAC combined with MK-2206 (AKT inhibitor) or standard NAC alone.Detailed descriptions of the design, eligibility, and study assessments in the I-SPY 2 TRIAL have been reported previously. 22,29Institutional Review Boards at all participating institutions approved the protocol.All patients signed informed consent to allow research on their biospecimen samples.

ctDNA analysis
Detailed description of the clinical samples and the methods for ctDNA analysis 26,27,30

Statistical analysis
To determine the cutoff for ctDNA positivity, a large set of negative control samples (w1000) was preprocessed to build a background error model.For each target variant identified in the plasma, a confidence score was calculated based on the depth of read for mutant and reference alleles. 25In addition, simulation studies were performed as previously described 10,26,27 to determine limits of detection and quality control thresholds for stringent assessment of ctDNA results (see supplementary Methods, available at https://doi.org/10.1016/j.annonc.2020.11.007 and supplementary Figures S4 and S5, available at https://doi.org/10.1016/j.annonc.2020.11.007).A plasma sample with at least two variants with a confidence score above a predefined threshold (0.97) was defined as ctDNA positive.
Logistic regression was used to assess association between pCR and ctDNA clearance.Survival curves were generated by KaplaneMeier analysis and compared using log-rank test.Cox regression analysis was used to estimate HR and 95% confidence interval (CI).Survival data were available for 75 of the 84 patients.Detailed description of the study design and the statistical methods can be found in the supplementary Methods, available at https://doi.org/10.1016/j.annonc.2020.11.007.
We also evaluated the absolute ctDNA levels (i.e.mean tumor molecules per ml of plasma) in the different groups stratified according to these same clinical variables and observed the same trend.The mean tumor molecules per ml in TNBC patients was significantly higher compared with that of HRþ/HER2À patients (Figure 2C).Significantly higher levels of ctDNA were also observed for clinical T-stage T3/T4 versus T1/T2 and MammaPrint high 2 versus high 1.

ctDNA positivity decreases with distinct dynamics during NAC
In the population as a whole, ctDNA positivity decreased during the course of NAC, from 73% before treatment (T0), to 35% at 3 weeks (T1), to 14% at the inter-regimen time point (T2), and down to 9% after NAC (T3) (Figure 3A).Similarly, the absolute ctDNA levels decreased over time (Figure 3B).Although, on average the ctDNA positivity decreased with time, at the individual patient level, five main patterns were observed.Figure 3C shows ctDNA positivity as a function of time during treatment for 58 of the 84 patients who had complete serial data available at all four time points: Patients with undetectable ctDNA at T0 who remained undetectable at T3 (n ¼ 20, 34.5%); patients who, respectively, cleared at T1 (n ¼ 20, 34.5%), at T2 (n ¼ 9, 15.5%), or at T3 (n ¼ 4, 6.9%); or patients who remained ctDNA positive after NAC (T3) (n ¼ 5, 8.6%).

Clearance dynamics of ctDNA is associated with NAC response
We evaluated ctDNA clearance as a predictor of response to NAC.The rates for pCR across subtypes were 13.8%, 47.4%, and 27.8% for HRþ/HER2e, HER2þ, and TNBC, respectively.As much as 56 patients who were ctDNA positive at T0 had a corresponding T1 plasma measurement (Figure 4A), and of these, 29 (52%) remained ctDNA positive at T1, 3 weeks after the initiation of treatment.As much as 83% of patients who did not clear their ctDNA at T1 had residual disease at surgery (24/29 non-pCR) compared with 52% in patients who cleared ctDNA at T1 (14/27 non-pCR).This association was significant (odds ratio 4.33, P ¼ 0.012, adjusted for subtype and treatment received).Among the 39 non-pCR patients who had undetectable ctDNA after NAC, 17 (43%) were ctDNA negative at baseline, 10 (26%) cleared ctDNA by T1, and 12 (31%) cleared ctDNA by

Annals of Oncology
T2 or T3.The positive predictive value (PPV) of the test (for predicting non-pCR) increased with time (Figure 4B).

Clinical events are frequent in patients with detectable ctDNA
Survival data were available for 75 of the 84 patients, with a median follow-up of 4.8 years (range 0.5-6.3years).In this period, 8 had local recurrences and 10 experienced distant metastases, of whom 8 died (Figure 5A).Detectable ctDNA in at least one time point was observed in 6 of the 8 patients (75%) with local recurrence, 9 of the 10 patients (90%) who had distant recurrence, and in all 8 patients who died (100%).Of note, a patient who experienced brain metastasis did not have detectable ctDNA at all time points.

ctDNA dynamics is significantly associated with metastatic recurrence
We examined whether ctDNA dynamic patterns (Figure 3C) were associated with DRFS, the secondary endpoint of the I-SPY 2 TRIAL.Of 58 patients with ctDNA data at all time points, 54 had follow-up information.Patients who had cleared ctDNA at T1, T2, or T3 (n ¼ 29) had similar risk of metastatic recurrence compared with those who were ctDNA negative at T0 (n ¼ 20; HR 2.1; 95% CI 0.22-20.2;Figure 3D).Patients who did not clear ctDNA at T3 (n ¼ 5) had a significantly higher risk of metastatic recurrence (HR 22.4; 95%, CI 2.5-201, P < 0.001).
ctDNA at T1, T2, and T3 but not T0 is associated with increased risk of metastatic recurrence Next, we examined whether ctDNA status (positive or negative) at different time points was associated with DRFS (supplementary Figure S6, available at https://doi.org/10.1016/j.annonc.2020.11.007).At baseline (T0), ctDNApositive patients had increased risk of metastatic recurrence, but this association did not reach statistical significance (HR 4.11; 95% CI 0.52-32.4).By contrast, ctDNA positivity at 3 weeks after initiation of therapy (T1; HR 4.5; 95% CI 1.2-17.4),between regimens (T2; HR 5.4; 95% CI 1.3-22.5),and after NAC (T3; HR 11.5; 95% CI 2.9-46.1)was significantly associated with increased risk of metastatic recurrence.(A) Overview of patient and tumor characteristics according to ctDNA status at baseline (T0).(B) Proportion of ctDNA-positive and ctDNA-negative patients at baseline (T0) according to clinical characteristics.P values were calculated using Fisher's exact test.(C) Mean tumor molecules per ml of plasma according to clinical characteristics.Distributions were compared using Wilcoxon rank sum (binary variable) or KruskaleWallis (ternary variable) tests.HER2, human epidermal growth factor receptor 2; HR, hormone receptor; NAC, neoadjuvant chemotherapy; pCR, pathological complete response; TNBC, triple-negative breast cancer.Of the 58 patients, 54 had survival data.Patients who cleared ctDNA at T1, T2, or T3 were combined into one group and their survival was compared with that of patients who did not clear ctDNA at T3 and those who were ctDNA negative at T0 (reference group).AC, anthracycline; HR, hazard ratio.

DISCUSSION
In this study, we examined the role of personalized ctDNA as a predictive biomarker for response and outcome in the neoadjuvant setting.The cohort included early-stage breast cancer patients with high risk of recurrence and who were treated with standard NAC alone or combined with MK-2206 (AKT inhibitor) treatment in the I-SPY 2 TRIAL.
ctDNA studies in the neoadjuvant setting in breast cancer have recently been reported 28,[31][32][33] (supplementary Table S4, available at https://doi.org/10.1016/j.annonc.2020.11.007).Two of the four studies were limited to a particular breast cancer subtype (i.e.only TNBC 28 or only HER2þ 31 ).Rothé and colleagues 31 observed that ctDNA detection before NAC was associated with decreased likelihood of achieving a pCR.McDonald and colleagues 32 showed that nonresponding patients have higher ctDNA levels after NAC compared with those who achieved a pCR.Two of the studies examined association between ctDNA and survival, 28,31 but none was able to demonstrate the prognostic impact of residual ctDNA after NAC.To our knowledge, our work represents the most comprehensive study on ctDNA detection in all subtypes, before, during, and after NAC and examined for the first time its association with response and survival in early breast cancer.
Here, we report on the use of a personalized ctDNA test informed by each patient's tumor genotype.We found that ctDNA is frequently detected in untreated high-risk early stage population (w70% of patients).The patterns of change in ctDNA during NAC were significantly correlated with risk of metastatic recurrence.We also found that ctDNA testing early during NAC (at 3 weeks) provided potentially actionable information as persistent ctDNA identified patients who were unlikely to achieve a pCR, whereas clearance was associated with improved response.
ctDNA positivity rate at baseline was significantly different among breast cancer subtypes (HRþ/HER2À: 52%, HER2þ: 82%, TNBC: 89%).We speculate that the lower rate of ctDNA positivity in HRþ breast cancer compared with HER2þ and TNBC is due in part to the lower proliferation rates (lower expression of Ki67 34 ) in this subtype, as was observed by Abbosh and colleagues in lung cancer. 25lucidating molecular and genomic factors predictive of ctDNA presence in the blood may shed on light on the biology of ctDNA release and clearance during treatment.
We examined whether ctDNA status at different time points was associated with risk of metastatic recurrence.We found that ctDNA-positive patients at T1, T2, and T3 (but not at T0) had significantly inferior DRFS compared with those who were ctDNA negative.We observed that PPV and HR increased with time, indicating that the last time point (i.e. after NAC before surgery) may be most informative for risk stratification of patients, and thus potentially help guide treatment in the adjuvant setting.
Our study showed that ctDNA status after NAC can potentially stratify patients who did not achieve pCR into low-and high-risk groups (Figure 5C).We found that clearance of ctDNA after NAC was associated with improved survival even in patients who did not achieve pCR.If validated, these findings could have a profound impact on treatment management in the neoadjuvant and adjuvant settings.
Recent clinical studies in breast cancer have shown that additional adjuvant therapy for nonresponders to NAC can lead to improvements in patient outcomes. 35,36Future studies should take into account the potential confounding effects of adjuvant treatment on the prognostic performance of ctDNA and other biomarkers analyzed in the neoadjuvant setting.
Differences in prognostic value of pCR by subtype have been reported, including its poor association with prognosis in HRþ breast cancer. 21,37Survival analysis in I-SPY 2 involving 950 patients has shown that pCR and subtype (including HRþ) were strongly associated with DRFS. 23In this subset, the individual prognostic impact of pCR and subtype was not observed, perhaps due to the modest sample size.Our exploratory survival analysis did show that ctDNA after NAC was a strong prognostic factor for DRFS.Further studies in larger cohorts are warranted to examine the contributions of ctDNA, pCR, and subtypes in predicting outcomes of patients who received NAC.
The I-SPY 2 schema includes the collection of serial magnetic resonance imaging (MRI) data during NAC to assess tumor response. 38,39We have previously analyzed paired ctDNA and MRI data collected at the same time points in the same cohort as this present study. 40We found that MRI-based functional tumor volumeda clinically established measure of residual disease in the breast 39,41 dwas significantly correlated with ctDNA levels at all time points. 40Furthermore, we found that ctDNA status after NAC improved the performance of functional tumor volume as predictor of metastatic recurrence and death.ctDNA testing could therefore serve as complementary tool to MRI for risk stratification of patients post-NAC.
A number of technologies for detection of ctDNA have been developed and are described in detail in a recently published review. 42Our approach provides several advantages over other methods of ctDNA analysis.The upfront whole-exome sequencing of primary tumors enables personalized selection of ctDNA targets that is independent of driver status.Our assay simultaneously tracks up to 16 patient-specific somatic variants and thus offers a more robust representation of the heterogeneity of a patient's tumor. 26,27,30By contrast, other methods such as droplet digital polymerase chain reaction 8 or BEAMing 43 can track only one to a few somatic variants.Our ctDNA test does have certain limitations including the inability to detect new second primary cancers which are often genetically unrelated to the original cancer 44 ; also, it will miss novel somatic variants that arise during tumor evolution in response to therapy-mediated selection pressures. 45lonal hematopoiesis of indeterminate potential mutations are potential sources of false positives in sequencing analyses of cfDNA. 46,47The ctDNA detection approach used in this study filters out clonal hematopoiesis of indeterminate potential mutations by focusing only on tumor-specific mutations that were initially detected by whole-exome sequencing of paired pretreatment tumor and germline DNA.
In the light of our findings, novel paradigms for ctDNAdirected treatment can be envisioned in future clinical trials.The current I-SPY 2 schema provides patients a single therapeutic opportunity to achieve a pCR 48 (Figure 1A).In the next iteration of the trial, patients will be given options to receive additional treatment to improve their chances of achieving a pCR, that is, if the initial agent does not result in a predicted complete response.For example, the decision to switch therapy for a patient without an early clinical or imaging response to a novel therapeutic agent would be supported if the patient fails to clear ctDNA.By contrast, patients who clear their ctDNA could continue treatment.Furthermore, information from ctDNA testing after NAC may help guide clinical decisions on whether to escalate or de-escalate treatment in the adjuvant setting.For example, if clearance of ctDNA is confirmed as a predictor of low risk of metastatic recurrence, such information can support treatment de-escalation.
Analysis of pooled serial circulating tumor cell (CTC) data obtained during neoadjuvant treatment of early breast cancer revealed that the prognostic impact of CTCs was the strongest at pretreatment (prior to NAC) compared with other time points. 49By contrast, our study showed that ctDNA status after NAC appeared to be the most important time point for prognostication.While both blood-based biomarkers display prognostic impact, their clinical value may not be redundant. 50Further studies that contemporaneously assess ctDNA and CTCs in the neoadjuvant setting are needed to elucidate the relative contributions of each biomarker in predicting response and outcome.The focus of this study was to examine the clinical significance of ctDNA monitoring in the neoadjuvant setting.The detection of minimal residual disease after surgery is of great clinical importance and can provide a unique opportunity for treatment redirection to delay metastatic recurrence and improve patient outcomes.We have now expanded our studies to include postsurgical monitoring of ctDNA in the adjuvant setting with the focus on residual disease detection after surgery and recurrence prediction.
In summary, our study shows promise that early response prediction by highly sensitive ctDNA analysis in high-risk early breast cancer patients may facilitate a timely and judicious change in treatment to improve patients' chances of achieving favorable long-term outcomes.The I-SPY 2 TRIAL provides an excellent platform to investigate how personalized ctDNA testing can complement imaging 51 and pathologic evaluation 52 of tumor response to fine-tune pCR as a surrogate endpoint for improved survival.Dynamic monitoring of ctDNA during NAC could facilitate evaluation of new agents by providing an early endpoint of treatment efficacy.Response over time as measured by imaging and ctDNA in the setting of early (pCR) and late (DRFS) outcomes will provide a robust framework for elucidating the potential clinical utility of ctDNA in the neoadjuvant setting.
(A) Diagram showing the study schema of the I-SPY 2 TRIAL.Prior to study entry, tumor biopsy from each patient is analyzed to assess hormone receptor and human epidermal growth factor receptor 2 status and MammaPrint scores.Blood samples are collected at the following time points: T0, baseline/pretreatment; T1, 3 weeks after initiation of therapy; T2, between two treatment regimens [paclitaxel AE MK-2206 and anthracycline (AC)]; and T3, after neoadjuvant chemotherapy (NAC) prior to surgery.(B) Flow chart showing patients and samples evaluated in the study and sample performance at different quality control (QC) points.(C) Schema of the methods for ctDNA analysis.pCR, pathological complete response.

2 AFigure 2 .
Figure 2. Association between circulating tumor DNA (ctDNA) and clinicopathologic characteristics.(A)Overview of patient and tumor characteristics according to ctDNA status at baseline (T0).(B) Proportion of ctDNA-positive and ctDNA-negative patients at baseline (T0) according to clinical characteristics.P values were calculated using Fisher's exact test.(C) Mean tumor molecules per ml of plasma according to clinical characteristics.Distributions were compared using Wilcoxon rank sum (binary variable) or KruskaleWallis (ternary variable) tests.HER2, human epidermal growth factor receptor 2; HR, hormone receptor; NAC, neoadjuvant chemotherapy; pCR, pathological complete response; TNBC, triple-negative breast cancer.

Figure 3 .
Figure 3. Circulating tumor DNA (ctDNA) dynamics over the course of neoadjuvant chemotherapy.(A) Proportion of patients according to ctDNA positivity based on number of samples available per time point.(B) Mean tumor molecules per ml of plasma across time points.(C) Patients with complete ctDNA data for four time points (n ¼ 58) grouped according to observed patterns of ctDNA clearance or nonclearance.(D) Survival in patients grouped according to ctDNA clearance.Of the 58 patients, 54 had survival data.Patients who cleared ctDNA at T1, T2, or T3 were combined into one group and their survival was compared with that of patients who did not clear ctDNA at T3 and those who were ctDNA negative at T0 (reference group).AC, anthracycline; HR, hazard ratio.

6 PPVFigure 4 .Figure 5 .
Figure 4. Association of circulating tumor DNA (ctDNA) with response to neoadjuvant chemotherapy and its positive predictive value.(A) Sankey plot showing ctDNA dynamics (clearance or non-clearance) early during treatment versus response [pathologic complete response (pCR) or no pCR].Analysis was focused on patients who were ctDNA positive at baseline (T0) and had corresponding ctDNA testing results at T1, 3 weeks after initiation of therapy.(B) Positive predictive value (PPV) of ctDNA positivity in predicting failure to achieve pCR.PPV is the proportion of patients with a positive ctDNA test (at a specific time point) with residual cancer after neoadjuvant chemotherapy (NAC).

Table 1 .
Multivariate Cox regression analysis to determine association between ctDNA positivity after NAC (T3) and distant disease-free survival (DRFS) while controlling for pCR and subtype