The clinical significance and biology of circulating tumor DNA in high-risk early-stage HER2-negative breast cancer receiving neoadjuvant chemotherapy

SUMMARY

Circulating tumor DNA (ctDNA) analysis may improve early-stage breast cancer treatment via non-invasive tumor burden assessment. To investigate subtype-specific differences in the clinical significance and biology of ctDNA shedding, we perform serial personalized ctDNA analysis in hormone receptor (HR)-positive/HER2-negative and triple-negative breast cancer (TNBC) patients receiving neoadjuvant chemotherapy (NAC) in I-SPY2. ctDNA positivity rates before, during, and after NAC are higher in TNBC than in HR-positive/HER2-negative patients. Early clearance of ctDNA 3 weeks after treatment initiation predicts a favorable response to NAC in TNBC only. Whereas ctDNA-positivity associates with reduced distant recurrence-free survival in both subtypes. Conversely, ctDNA-negativity after NAC correlates with improved outcomes, even in patients with extensive residual cancer. Pretreatment tumor mRNA profiling reveals associations between ctDNA shedding and cell cycle and immune-associated signaling. Based on these findings, the I-SPY2 trial will prospectively test ctDNA for utility in redirecting therapy to improve response and prognosis.

GRAPHICAL ABSTRACT

eTOC Blurb

Magbanua et al. examine the dynamics of ctDNA in plasma of high-risk early-stage breast cancer patients receiving neoadjuvant chemotherapy. Understanding the predictive and prognostic value of ctDNA and biology of ctDNA shedding in different breast cancer subtypes can inform the use of ctDNA for treatment selection to improve patient outcomes.

INTRODUCTION

Circulating tumor DNA (ctDNA) analysis offers a non-invasive approach to assessing tumor burden during treatment. Monitoring ctDNA can potentially improve the management of early-stage breast cancer by replacing standard-of-care tissue biopsy for DNA-based biomarker evaluation. ^1,2 Breast cancer is, however, a collection of many diseases; thus, studies that seek to demonstrate the clinical impact of ctDNA need to account for the existing heterogeneity across different breast tumor types.

Patients with early-stage, locally advanced breast cancer can benefit from neoadjuvant chemotherapy (NAC, treatment given before surgery), mainly because tumor response during and after treatment can be measured and used to inform subsequent treatment decisions. ³ Patients who achieve a pathologic complete response (pCR), defined as the absence of invasive cancer in the breast and regional lymph nodes, have a significantly decreased risk of metastatic recurrence and death compared to those with residual cancer after NAC. ^4–6 For patients who do not achieve pCR, understanding the relationship between residual disease and risk of recurrence presents the opportunity to evaluate new therapies following surgery in the post-neoadjuvant setting. ^7,8

However, response rates to NAC vary among receptor subtypes. ^9,10 For example, results from I-SPY2—a neoadjuvant trial in high-risk early-stage breast cancer evaluating the efficacy of standard chemotherapy with or without investigational drugs—show that among 10 early arms of the trial, the maximum estimated pCR rate is the highest in the hormone receptor (HR)-negative/HER2-positive group at 74%. ⁹ In contrast, triple-negative breast cancer (TNBC) and HR-positive/HER2-negative groups have lower maximum estimated pCR rates of 60% and 30%, respectively. ⁹ Given that the objective is for each patient to achieve a pCR, there is an unmet need—especially in HER2-negative breast cancer—for developing early predictors of response, in part to de-escalate therapy and limit toxicity for patients who achieve pCR early during treatment or to switch treatment strategy in the case of early non-response.

ctDNA monitoring shows great promise for such a biomarker. However, given the differences in biology, response, and clinical histories between HR-positive/HER2-negative and TNBC, we hypothesize that the predictive and prognostic value of ctDNA varies between the two subtypes. Although recent studies in early-stage breast cancer have investigated the clinical significance of serial ctDNA assessment during NAC; several focus on a single breast cancer subtype—TNBC ^11–13 or HER2-positive breast cancers ¹⁴; others have insufficient sample size to evaluate the predictive value of ctDNA across breast cancer subtypes ^15–18; or do not examine the prognostic significance of ctDNA. ^16,19

In this study, we address these limitations by examining the predictive and prognostic value of ctDNA in a relatively large cohort representing two major breast cancer subtypes: HR-positive/HER2-negative and TNBC. We present a follow-up study to our previous work investigating the clinical significance of ctDNA in 84 patients with high-risk early-stage breast cancer receiving NAC. ¹⁵ In our expanded cohort of 283 patients (145 HR-positive/HER2-negative and 138 TNBC) and 1,024 serial blood samples, we evaluate the predictive and prognostic value of ctDNA. In addition to pCR as a response endpoint ⁴, we examine the relationship of ctDNA to residual cancer burden (RCB), a measure of the extent of residual disease in the breast and axillary lymph nodes following NAC. ^20,21

We also hypothesize that gene expression analysis could elucidate mechanisms that govern ctDNA shedding ²², as it is unclear why some breast tumors shed high amounts of ctDNA while others have undetectable levels. In addition, tumor size and breast cancer subtype appear to be associated with ctDNA shedding ^15,23–25, but these factors cannot fully explain differences in the concentrations of ctDNA in the blood of breast cancer patients. To this end, we leverage I-SPY2 trial data to examine genes and pathways associated with ctDNA shedding before treatment in these two subtypes. Findings from this study—both subtype-specific clinical associations and related to the biology of ctDNA shedding—may help maximize and fine-tune the use of ctDNA as a biomarker of response and survival in patients with high-risk early-stage breast cancer receiving NAC.

RESULTS

Patients and samples

The study involved high-risk early-stage breast cancer patients in the neoadjuvant trial, I-SPY2. A total of 295 HER2-negative patients with available pretreatment biopsies were included in this biomarker study (Figure S1A, Table 1 and STAR Methods). Plasma samples were collected at 4 time points—pretreatment (T0), 3 weeks after initiation of treatment (T1), at 12 weeks, between paclitaxel-based and anthracycline (AC) regimens (T2), and after NAC before surgery (T3) (Figure S1B). ctDNA was detected in plasma using a personalized and tumor-informed test (STAR Methods). Samples that did not pass quality control measures were excluded from the analysis (Figure S1A and STAR Methods). Of the 295 patients, 283 (96%) were evaluable, with 1,024 plasma samples. ctDNA and clinical data are available in Table S1.

Table 1. Clinicopathologic characteristics of patients grouped according to breast cancer subtype.

Hormone receptor-positive/HER2-negative (HR+HER2−) and triple-negative breast cancer (TNBC).

		All patients  (n=283)			HR+HER2−  (n=145, 51.2%)		TNBC (n=138, 48.8%)
		n	%		n	%	n	%	Fisher p-value
Clinical T stage									0.6777
T1/T2		170	60.1		82	56.6	88	63.8
T3/T4		75	26.5		39	26.9	36	26.1
Missing		38	13.4		24	16.6	14	10.1
Clinical N stage									0.4348
Node-negative		118	41.7		55	37.9	63	45.7
Node-positive		118	41.7		62	42.8	56	40.6
Missing		47	16.6		28	19.3	19	13.8
Grade									<0.0001
1/2		68	24		56	38.6	12	8.7
3		156	55.1		62	42.8	94	68.1
Missing		59	20.8		27	18.6	32	23.2
MammaPrint score									<0.0001
High 1		120	42.4		104	71.7	16	11.6
High 2		163	57.6		41	28.3	122	88.4
Treatment arm									0.4105
Paclitaxel		136	48.1		72	49.7	64	46.4
Paclitaxel + PD-1 inhibitor		44	15.5		27	18.6	17	12.3
Paclitaxel + MK-2206		39	13.8		17	11.7	22	15.9
Paclitaxel + Ganitumab		14	4.9		8	5.5	6	4.3
Irinotecan + Talazoparib		13	4.6		4	2.8	9	6.5
Paclitaxel + Ganetespib		12	4.2		4	2.8	8	5.8
Paclitaxel + AMG 386		8	2.8		3	2.1	5	3.6
Paclitaxel + PD-1 inhibitor 8-Cycle		8	2.8		4	2.8	4	2.9
Paclitaxel + ABT 888 + Carboplatin		5	1.8		4	2.8	1	0.7
Paclitaxel + Neratinib		3	1.1		2	1.4	1	0.7
SGN-LIV1A		1	0.4		0	0	1	0.7
Pathologic complete response (pCR) *									0.0683
pCR		53	18.7		21	14.5	32	23.2
no pCR		230	81.3		124	85.5	106	76.8
Residual cancer burden (RCB)									0.0010
RCB-0		56	19.8		22	15.2	34	24.6
RCB-I		31	11		10	6.9	21	15.2
RBC-II		115	40.6		61	42.1	54	39.1
RBC-III		75	26.5		51	35.2	24	17.4
Missing		6	2.1		1	0.7	5	3.6
Age (mean)									0.1440
		49.2			50.1		48.2

^*Per I-SPY2 protocol, patients who received study treatment but switched to non-protocol therapy, did not go to surgery, or withdrew consent before surgery were considered to have a non-pCR for efficacy and biomarker analysis (see Methods).

Clinical correlates of ctDNA

Consistent with previous observations ^15,26, pretreatment (Figure S2A) and concentration (Figure S2B)—expressed as mean tumor molecules (MTM) per mL of plasma—were higher in TNBC than in HR-positive/HER2-negative patients. ctDNA positivity remained significantly higher in TNBC across all time points (Figure S2A).

Furthermore, we found that higher pretreatment (T0) ctDNA positivity and concentration were associated with clinicopathologic variables that confer aggressiveness, e.g., larger tumor and node-positivity in both subtypes and higher grade and MammaPrint score in the HR-positive/HER2-negative group (Figure S3).

ctDNA concentration is more strongly associated with pCR and RCB in TNBC as compared to HR-positive/HER2-negative patients

We examined the relationship of ctDNA concentration (MTM/ml) with response to NAC. In this cohort, the pCR rates were 14.5% (21/145) in HR-positive/HER2-negative and 23.2% (32/138) in TNBC patients (Table 1).

We compared the ctDNA concentrations in patients grouped according to response to NAC. In the HR-positive/HER2-negative group, no significant differences in the distribution of ctDNA concentrations across all time points were observed between patients who had pCR vs. no pCR (Figure 1A). However, in TNBC, ctDNA concentrations were significantly higher at all time points in patients who did not achieve pCR compared to those who did (Figure 1B). Similar findings were observed when ctDNA concentrations were compared across RCB classes in both subtypes (Figure 1C and 1D).

Figure 1. Predictive value of ctDNA in high-risk early-stage HER2-negative breast cancer receiving neoadjuvant chemotherapy (NAC).

A-D. Distribution of ctDNA concentration expressed as mean tumor molecules per ml (MTM/mL) of plasma across time points during NAC in hormone receptor-positive/HER2-negative (HR+HER2−, left panel) and triple-negative breast cancer (TNBC, right panel) patients grouped according to (A, B) pathologic complete response (pCR) and (C, D) residual cancer burden (RCB) class. ctDNA was analyzed in plasma collected at pretreatment (T0), 3 weeks after treatment initiation (T1), 12 weeks after treatment initiation between paclitaxel-based treatment and anthracycline regimens (T2), and after NAC before surgery (T3). For each box plot, the center line represents the median value (50th percentile); the box contains the 25th to 75th percentiles of the data distribution; the whiskers represent the 5th and 95th percentiles; and the dots beyond the upper and lower bounds are outliers; E-J. Patients who were ctDNA-positive at T0 were classified into 3 groups based on ctDNA dynamics from T0 to T2. Distribution of ctDNA dynamic patterns in E. HR+HER2− and F. TNBC patients with complete ctDNA data from T0-T2. Association of ctDNA dynamics with (G, H) pCR and (I, J) RCB. RCB was divided into 4 classes: RCB-0 (no residual disease equivalent to pCR), RCB-I (minimal burden), RCB-II (moderate burden), and RCB-III (extensive burden). Logistic regression was used to estimate the odds ratio (OR) and 95% confidence interval of achieving (G, H) pCR or (I, J) RCB-0/I.

Early ctDNA clearance is a significant predictor of pCR and RCB in TNBC but not in HR-positive/HER2-negative patients

We investigated whether ctDNA clearance was predictive of pCR or RCB in patients who were ctDNA-positive at T0 (Figure 1E–J). Complete ctDNA data for the first 3 time points were available from 83 (57.2%) HR-positive/HER2-negative and 85 (61.6%) TNBC patients. In both groups, similar proportions of patients cleared early at T1 (ctDNA+/−/−), 27.3% in HR-positive/HER2-negative, and 22.4% in the TNBC group (Figure 1E and 1F). Among the HR-positive/HER2-negative patients, the largest group, representing about half, consisted of those who cleared their ctDNA late at T2 (ctDNA+/+/− ; n=40, 48.2 %), while in the TNBC group, the largest group consisted of those who remained positive until T2 (ctDNA+/+/+, n=42, 49.4%).

There was no significant association between ctDNA clearance and pCR in HR-positive/HER2-negative patients (p=0.92) (Figure 1G). In contrast, the proportion of patients who achieved pCR was significantly higher in TNBC patients who cleared their ctDNA early by T1 (ctDNA+/−/−) compared to those who cleared at T2 (ctDNA+/+/−) or remained positive for ctDNA until T2 (ctDNA+/+/+) (p=0.0002) (Figure 1H). Early clearance was a significant predictor of pCR in the TNBC group (Odds ratio 13.06, 95% CI: 3.54–57.95) but not in HR-positive/HER2-negative group (Figure 1G and 1H). Similar results were observed when associations between ctDNA clearance and RCB were examined (Figure 1I and 1J). Early clearance was a significant predictor of RCB-0/I in the TNBC group (Odds ratio 19.00, 95% CI: 4.98–89.06) (Figure 1J) but not in HR-positive/HER2-negative group (Figure 1I). Altogether, our analysis revealed a significant predictive value of early ctDNA clearance in TNBC patients but not in HR-positive/HER2-negative patients.

ctDNA positivity before, during, and after NAC is a significant negative prognostic factor

To evaluate the prognostic value of ctDNA, we examined its correlation with distant recurrence-free survival (DRFS, STAR Methods).

ctDNA positivity at all time points was significantly associated with inferior DRFS in both subtypes (all log-rank p<0.05, Figure S4). The correlation between ctDNA positivity and poor DRFS was strongest at T2 (at 12 weeks) and T3 (after NAC before surgery) (all log-rank p<0.0001). ctDNA positivity remained a significant negative prognostic factor for DRFS in bivariate Cox proportional hazards regression models that adjusted for pCR, except for T1 in patients with TNBC (Figure S5). No DRFS events were observed in TNBC patients who tested ctDNA-negative at T0, and thus no accurate estimates of hazard ratio and 95% confidence interval were calculated.

ctDNA status after NAC refines the prognostic value of pCR and RCB

We examined whether ctDNA status after NAC (T3) could further refine risk stratification by pCR. Patients were classified into 4 groups using ctDNA status at T3 (ctDNA-positive or ctDNA-negative) and whether the patient achieved a pCR or not.

In HR-positive/HER2-negative patients, the stratification yielded only 3 groups because all patients who achieved pCR were ctDNA-negative at T3. As expected, these patients had the most favorable DRFS (Figure 2A). For those who failed to achieve pCR, DRFS was significantly worse in ctDNA-positive patients compared to ctDNA-negative (Hazard ratio 5.89; 95% CI 2.68–12.98). Similar results were observed in TNBC: ctDNA-positive patients had a significantly increased risk of relapse and death compared to ctDNA-negative (Hazard ratio 3.79; 95% CI 1.87–7.68, Figure 2B). One patient with TNBC who achieved a pCR was ctDNA-positive at T3; this patient experienced a recurrence one year and 9 months after study entry.

Figure 2. Prognostic value of ctDNA in high-risk early-stage HER2-negative breast cancer receiving neoadjuvant chemotherapy (NAC).

A-D. Survival curves of patients grouped according to combined ctDNA status after NAC (ctDNA-positive or ctDNA-negative at T3) and (A, B) pCR (pCR or no pCR) or (C, D) binarized RCB (RCB-0/I or RCB-II/III) in hormone receptor-positive/HER2-negative subset (HR+HER2−, left panel) and triple-negative breast cancer (TNBC, right panel). RCB was divided into 2 classes: RCB-0 (no residual disease equivalent to pCR) and RCB-I (minimal burden) vs. RCB-II (moderate burden) and RCB-III (extensive burden). Hazard ratios (HR) and 95% confidence intervals (CI) shown were estimated from Cox proportional hazards regression models; E-H. Distribution of ctDNA dynamic patterns in E. HR+HER2− and F. TNBC patients with complete ctDNA data for all time points. Survival curves of G. HR+HER2− and H. TNBC patients grouped according to ctDNA dynamics. Distant recurrence-free survival (DRFS) was the survival endpoint. P-values were calculated using the log-rank test.

We repeated the same analyses to assess whether combining ctDNA status at T3 and dichotomized RCB class (RCB-0/I vs. RCB-II/III) could improve risk stratification. In the HR-positive/HER2-negative group, patients classified as RCB-II/III who tested ctDNA-positive at T3 had a significantly higher risk of metastatic recurrence and death than RCB-II/III patients who tested negative (Hazard ratio 5.65; 95% CI 2.45–12.99) (Figure 2C). Survival analysis in TNBC showed similar results: ctDNA-positivity at T3 was associated with significantly worse outcomes (Hazard ratio 3.84; 95% CI 1.70–8.66, Figure 2D). In addition, all 3 TNBC patients with RCB-0 (n=1) and RCB-I (n=2) who tested positive for ctDNA at T3 experienced metastatic recurrence.

In summary, among non-responders (no pCR or RCB-II/III), a ctDNA-negative test at T3 was significantly associated with improved survival compared to those who tested positive for ctDNA. These results suggest that risk stratification using established prognostic factors, such as pCR and RCB, especially in patients with poor response to NAC, can be improved by adding information from ctDNA testing after NAC.

ctDNA dynamics during NAC are prognostic in both subtypes

We then examined the prognostic significance of ctDNA dynamics in each subtype. Patients with available ctDNA data for all 4 time points were classified into 5 groups based on ctDNA dynamics (STAR Methods). The distributions of patients based on ctDNA dynamics are shown in Figures 2E and 2F. The largest group among HR-positive/HER2-negative patients was Group 1, i.e., patients who tested negative at T0 and remained negative (31%, Figure 2E); and among TNBC patients, was Group 4, i.e., patients who were ctDNA positive at T0 and cleared at T3 (31.4%, Figure 2F).

Consistent with the results from previous analysis examining the prognostic impact of ctDNA status at each time point, we observed that in both subtypes, patients who were ctDNA-negative at T0 and consistently remained negative (Group 1) had the most favorable survival (Figure 2G and 2H). In this group, 2 of the 35 HR-positive/HER2-negative patients experienced a metastatic recurrence (Figure 2G). Of the 35, 74.3% (26/35) did not achieve a pCR. In the same group in TNBC, none of the 8 patients relapsed or died (Figure 2H); 50% (4/8) of them did not achieve a pCR. Consistent with previous analyses, we observed that failure to clear ctDNA (Group 5) was associated with the poorest DRFS. Interestingly, in the TNBC group, patients who cleared their ctDNA after paclitaxel-based treatment (Group 3) had favorable survival, with only 1 metastatic recurrence observed after a median follow-up of 4 years and 2 months (Figure 2H). Of the 17 patients in Group 3, 13 (76.5%) did not achieve a pCR.

Genes expression correlates of ctDNA shedding

Next, we determined genes and pathways associated with ctDNA shedding by performing differential expression and gene set enrichment analysis (GSEA) of pretreatment tumor gene expression (T0) between ctDNA-positive and ctDNA-negative patients (Figure S6). Our analysis revealed 153 genes significantly differentially expressed (DE) [Benjamini-Hochberg (BH) adjusted p<0.05] and >2,000 genes nominally DE (p<0.05) in the HR-positive/HER2-negative subtype (Figure S6A and S6B), and ~900 nominally DE genes in TNBC (Figure S6A and S6C).

In the HR-positive/HER2-negative subtype, GSEA revealed enrichment of 28 Molecular Signatures Database (MSigDB) Hallmark gene sets (56%) significantly associated with ctDNA status at baseline (BH adjusted p<0.05), with one other nominally enriched (p<0.05) (Figure 3A, top panel). Of these, 11 gene sets, including metabolic and estrogen response gene sets, had higher expression levels in ctDNA-negative patients. In ctDNA-positive patients, we observed significant enrichment of 18 gene sets—including higher expression levels of cell cycle and proliferation gene sets, inflammatory cytokines (IL6, TNF-alpha), and inflammatory response immune-associated gene sets. In TNBC, GSEA identified nominal enrichment of 5 gene sets (p<0.05) (Figure 3A, bottom panel), including immune-associated Interferon-alpha response and TGF-beta signaling.

Figure 3. Differentially expressed genes and enriched gene sets associated with ctDNA shedding at pretreatment.

A. GSEA of MSigDB Hallmark gene sets in hormone receptor-positive/HER2-negative subset (HR+HER2−) (top) and TNBC (bottom). Gene sets with significant (BH adjusted p<0.05) and nominally significant (p<0.05) enrichment in ctDNA-positive or ctDNA-negative patients are depicted with p-value significance and normalized enrichment scores (NES). B-C. Gene expression heatmap of leading-edge genes in enriched immune-associated gene sets: B. IL6 JAK STAT3 signaling, inflammatory response, and TNF-alpha signaling via NF-κB in HR+HER2−; and C. Interferon alpha response, and TGFB signaling in TNBC. D. STRING protein-protein interaction (PPI) of genes that are commonly differentially expressed (p<0.05) in both HR+HER− and TNBC with concordant fold change direction across subtypes. The PPI for the 99 genes comprising the main network is shown with community membership, differential expression direction, degree and type of connectivity, and PPI confidence level annotated. Each community’s STRING PPI KEGG functional enrichment is listed.

To explore the impact of proliferation, we next performed differential expression analysis in the HR-positive/HER2-negative subtype, adjusting for MammaPrint status, and found fewer DE genes detected (Figure S6C). The 3 genes that remained significantly DE (adj. p<0.05) after adjusting for MammaPrint status include N92541, CYBRD1, and RCBTB1, all upregulated in ctDNA-negative patients. Despite adjustment for MammaPrint status, GSEA still captured gene sets associated with cell cycle processes (Figure S6D). These analyses suggest a strong association between proliferation and ctDNA release in the HR-positive/HER2-negative subtype.

Immune cells in the tumor microenvironment have been proposed to be involved in the shedding of ctDNA into blood ²⁷. We performed leading-edge analysis (LEA) ²⁸ to identify genes (referred to as leading-edge genes) with the highest impact on the enrichment signal of immune response pathways associated with ctDNA shedding. In the HR-positive/HER2-negative subtype, LEA identified key genes within enriched pathways in ctDNA-positive patients, including 32 genes involved in IL6/JAK/STAT3 signaling (e.g., CCL7, BAK1, TNFRSF12A); 76 genes involved in Inflammatory response (e.g., PVR, CCL7, ADRM1); and 68 genes involved in TNF-alpha signaling via NF-kB (e.g., VEGFA, ATF3, HBEGF) (Figure 3B, Figure S6Ei–iii). In TNBC, LEA identified key genes, including 45 genes (e.g., LY6E, TRIM26, TRIM14) in Interferon-alpha response enriched in ctDNA-positive patients; and 16 genes (e.g., SMAD1, LTBP2, TRIM33) in TGF-beta signaling enriched in ctDNA-negative patients (Figure 3C, Figure S6Fi–ii).

Next, we considered the 190 genes that were commonly DE (p<0.05) across subtypes with concordant up- or down-regulation associated with ctDNA release (Figure S6G). Protein-protein interaction (PPI) network analysis identified 99 genes that form the main PPI network with 9 gene communities (Figure 3D). PPI functional enrichment of each gene community (STRING Database) identified the association of 5 communities with KEGG pathways (Figure 3D). These results suggest that ctDNA shedding at baseline in both HR-positive/HER2-negative and TNBC tumors was associated with active carbon and amino acid metabolism, high proliferation, and high levels of immune activity.

DISCUSSION

In this study involving 283 patients, we examined the clinical significance and gene expression correlates of ctDNA shedding in HER2-negative breast cancers (HR-positive/HER2-negative and TNBC), representing the majority of breast cancer cases. ²⁹

CtDNA levels at all time points were higher in TNBC than in HR-positive/HER2-negative patients. The higher ctDNA positivity rate in TNBCs may be due to the high proliferation and cell turnover rates in this subtype ³⁰, both of which are associated with increased ctDNA shedding. ²⁴

Moreover, ctDNA positivity and dynamics were more strongly associated with response to NAC in TNBC than in HR-positive/HER2-negative. This may be due to inherent differences in the biology, response to treatment, and disease trajectories between the two subtypes. ^9,10,31,32 For example, we found that early clearance of ctDNA at 3 weeks after initiation of therapy was predictive of pCR and RCB in TNBC but not in the HR-positive/HER2-negative subset. These results can inform decisions for treatment redirection (escalation/de-escalation) in the neoadjuvant setting to increase the likelihood of patients achieving a pCR. Switching to more effective therapy in TNBC patients who do not clear ctDNA at 3 weeks may improve treatment response.

We examined the prognostic significance of serial ctDNA status and dynamics. In both subtypes, ctDNA positivity at pretreatment was significantly associated with an increased risk of metastatic recurrence or death, confirming previous observations. ²⁶ In most cases, ctDNA positivity at different time points remained a significant negative prognostic factor for DRFS even after adjustment for the effects of pCR. These results suggest that ctDNA reflects therapeutic efficacy and could serve as an early surrogate marker of survival.

To assess whether ctDNA can refine risk stratification by pCR and RCB, we grouped patients based on ctDNA status after NAC (ctDNA-positive or ctDNA-negative at T3) and either pCR status (pCR or no pCR) or binarized RCB class (RCB-0/I or RCB-II/III). In both subtypes, patients who were ctDNA-positive after NAC and had a poor response to NAC (no pCR or RCB-II/III) showed the worst survival outcome of all the groups compared. Most importantly, our analysis revealed that ctDNA negativity after NAC is associated with favorable DRFS, even in those who did not achieve pCR or those with moderate or extensive residual cancer burden (RCB-II/III). Residual cancers that do not shed detectable levels of ctDNA may differ in biology from those that do and may represent a less aggressive type of cancer with lower metastatic potential. These findings could help inform patient decisions on whether to have additional treatment after primary NAC.

We next examined the prognostic significance of ctDNA dynamics in patients with ctDNA data for all time points. As expected, patients in both subtypes who did not clear their ctDNA after NAC (T3) had the poorest survival outcomes, whereas patients who tested ctDNA-negative at all time points appeared to have the best survival outcomes. We speculate that patients who were ctDNA-negative at all time points and did not achieve pCR may have pretreatment and residual tumors that are less aggressive and proliferative and, therefore, are less likely to recur. Thus, patients with undetectable levels of ctDNA may be eligible for treatment de-escalation in both the neoadjuvant and adjuvant settings. If our observations are confirmed in a larger cohort, consistent negative tests for ctDNA before, during, and after NAC could be an early surrogate of favorable survival in both subtypes.

Our exploratory analysis of pretreatment gene expression revealed common and unique patterns potentially associated with ctDNA shedding across and within subtypes. Though our power to detect DE genes was limited in TNBC due to an imbalance between the ctDNA-positive vs. ctDNA-negative groups, PPI enrichment of commonly DE genes (p<0.05) identified cell cycle genes associated with ctDNA shedding in both HR-positive/HER2-negative and TNBC subtypes, suggesting an important role of tumor cell proliferation in ctDNA release. Indeed, associations of cell cycle-related gene sets with ctDNA release were particularly significant in the HR-positive/HER2-negative subtype, even after adjustment for MammaPrint status. Transcriptional analysis of urothelial tumors from ctDNA-positive patients also showed upregulation of cell-cycle genes. ²² Additionally, gene expression analysis suggested a key role of specific immune response pathways in ctDNA release, consistent with preclinical studies demonstrating the potential role of immune cells in the shedding of DNA into circulation. ^33,34

Recent studies in early-stage breast cancers have shown that ctDNA before ^17,26 and after NAC ^{11–13,15,17,18} were prognostic of poor survival. Because our study analyzed ctDNA at 4 time points, we demonstrated the prognostic value, not only of pretreatment and post-NAC ctDNA but also of ctDNA detection as early as 3 and 12 weeks after initiation of treatment. Moreover, we classified patients into groups using the results from serial ctDNA tests and demonstrated the prognostic significance of ctDNA dynamics. In addition, our study showed that combining information from ctDNA status after NAC and treatment response could improve risk stratification, especially in patients with no pCR. In contrast, none of the previous studies have examined the prognostic significance of ctDNA dynamics and combined ctDNA and response information,

A limitation of this study is the heterogeneity in the treatment received by patients. Approximately 50% of patients received standard chemotherapy (paclitaxel + AC), while the rest received various investigational drugs combined with paclitaxel. We speculate that ctDNA, like radiographic imaging, could serve as a measure of tumor burden regardless of treatment received. Another limitation is that the patient-specific mutations selected for the design of each personalized ctDNA assay were based on the tumor mutational profiles of tissue biopsy taken before treatment. Less dominant clones in the untreated tumor could be selected in response to treatment resulting in the emergence of new mutations that the initial assay would miss. Due to the small sample size available for analysis, we could not evaluate the predictive and prognostic significance of ctDNA in patients with HER2-positive disease, representing approximately 15% of breast cancers. ²⁹ We also acknowledge limitations in the sensitivity and specificity of the ctDNA assay used in the study. For example, some patients with residual cancer tested negative for ctDNA after NAC. This represents a potential issue with the performance of the assay and also suggests that a ctDNA test alone may not provide reliable information regarding disease status and patient outcomes. Thus, improvements in assay sensitivity are needed, and combining ctDNA test results with other tests (e.g., imaging and pathology) could improve the detection of residual disease and the prediction of metastatic recurrence. Importantly, with the potential for false negatives and false positives, caution should be exercised when utilizing ctDNA information to de-escalate or escalate treatment.

Here, we present the results of a large study that examined the predictive and prognostic value of ctDNA as well as the biology of ctDNA shedding in early-stage HER2-negative breast cancer receiving NAC. Our findings can inform the design of future clinical trials that seek to evaluate the clinical utility of ctDNA in the neoadjuvant setting. To this end, we plan to expand our ctDNA analyses to include 1000 patients enrolled in the I-SPY2 trial to validate the findings described in this report, expand investigations across all breast cancer subtypes (including HER2-positive patients), and compare ctDNA trajectories in patients who received immunotherapy agents vs. those who did not.

The I-SPY2 trial will prospectively test ctDNA for utility in redirecting therapy to improve patient outcomes. Results from ctDNA tests, combined with those from magnetic resonance imaging ^15,25 and pathology of on-treatment tumor biopsy, will guide early therapy changes (treatment escalation) in non-responding patients to increase the likelihood of achieving a pCR. Also, information from these tests will aid decisions regarding treatment de-escalation for patients predicted to have a pCR by providing an option for early surgery to limit exposure to toxicity of unnecessary treatments.

STAR METHODS

Detailed methods are provided in the online version of this paper and include the following:

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources or data should be directed to the I-SPY Data Access and Publications Committee coordinator (ispy2dapc@quantumleaphealth.org).

Materials availability

This study did not generate new unique reagents.

Data and Code Availability

Transcriptomic data used in this study are available in NCBI’s Gene Expression Omnibus (GEO) GSE194040 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE194040). ctDNA and clinical data are available in Table S1. ctDNA and clinical data, Related to Figures 1 and 2. No original code was developed for this work.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

The I-SPY2 Trial.

I-SPY2 (NCT01042379; IND 105139) is an ongoing, open-label, multicenter adaptive, randomized phase 2 platform trial of neoadjuvant therapy for high-risk early-stage breast cancer. The overview, design, patient eligibility, and oversight of the trial have been previously described in detail. ^40–43 I-SPY2 evaluates multiple experimental treatments against a common control arm, in parallel. The study stratifies patients within 8 subtypes based on HR, HER2, and MammaPrint scores, and the combinations of these subtypes define 10 biomarker signatures. An adaptive algorithm based on Bayesian probabilities of benefit vs. control is used to randomize within a subtype. ^40–42 The primary endpoint of the trial is pCR (ypT0/is, ypN0), defined as no residual invasive cancer in either breast or lymph nodes, evaluated at the time of surgery. Adaptive randomization in I-SPY2 preferentially assigns patients to trial arms according to continuously updated Bayesian probabilities of pCR rates within each biomarker signature. The demonstration of statistical superiority in pCR rate (vs. the control arm) for any of the 10 pre-defined biomarker signatures determines the “graduation” of an experimental arm.

I-SPY2 is an intent-to-treat trial. Per protocol, patients who received study treatment but switched to non-protocol therapy, withdrew consent before surgery, or did not receive surgery were considered to have non-pCR. Because response in those situations cannot be attributed to the treatment the patient was initially randomized to, a non-pCR was assigned for efficacy and biomarker analyses. Response to study treatment (and non-protocol therapy) was also assessed at surgery using the residual cancer burden (RCB) method. ^20,21 While RCB is a pathology-based approach to measure the amount of residual invasive cancer in the breast and regional lymph nodes, pCR (per I-SPY2 protocol) is a combination of pathology and trial rules/administration geared to evaluating the efficacy of a specific treatment arm. For example, when a patient assigned to an experimental arm drops out of the trial protocol, the patient is considered to have non-pCR even if, by pathology, the patient eventually achieves pCR/RCB-0 using the RCB method.

In I-SPY2, the observed pCR rates are often lower than those reported by other studies. One contributing factor is that, per the I-SPY2 protocol (discussed above), any patient who receives non-protocol therapy is considered non-pCR. Another factor is the use of the RCB method for assessing pCR, which entails a more comprehensive evaluation of the resected surgical specimens and typically results in increased identification of cases with minimal residual disease (non-pCR).

Patients.

The ctDNA study was conducted in the context of the I-SPY2 trial. Patients had MammaPrint high tumors and therefore are at high risk of metastatic recurrence within five years after diagnosis. ⁴⁴

This study involved patients with HER2-negative tumors. Breast cancer is generally classified into one of the 4 receptor subtypes based on hormone receptor (HR; estrogen and progesterone receptor) and HER2 expression. The HER2-negative receptor subtypes—HR-positive/HER2-negative and HR-negative/HER2-negative (also known as triple-negative breast cancer or TNBC)—represent about 85% of all breast cancers [73% HR-positive/HER2-negative and 12% TNBC]. ²⁹ Receptor subtypes have distinct biological characteristics that are reflected in differences in clinical outcomes ⁴⁵ and recommended treatment modalities in the early-stage setting; ³ for example, HR-positive/HER2-negative tumors are treated with endocrine therapy with or without chemotherapy, while HER2-positive tumors are treated with HER2-targeted drugs in combination with chemotherapy. Early-stage TNBC (stage II and higher) is most often treated with the checkpoint inhibitor, pembrolizumab, in the neoadjuvant setting in combination with chemotherapy, followed by pembrolizumab post-surgery. ⁴⁶

In the I-SPY2 trial, HER2-negative patients in the control arm of I-SPY2 received paclitaxel followed by AC. Investigational regimens are given in combination with paclitaxel or as a replacement for paclitaxel. The trial limits eligibility to women >18 years with stage II or III breast cancer >2.5cm and high MammaPrint score; ⁴⁷ these are patients at high risk of metastatic recurrence within 5 years after diagnosis. The I-SPY2 protocol was approved by Institutional Review Boards at all participating institutions and all patients signed written informed consent. Patients included in the current study were HER2-negative, with pretreatment tumor biopsy specimens available and whose plasma samples were analyzed for ctDNA. These patients were enrolled in I-SPY2 between March 2010 and July 2018. All patients provided written informed consent for subsequent use of their specimens for research purposes.

Of the 283 evaluable patients, 26.5% (75) were stage T3/T4; 41.7% (118) were node-positive; 55.1% (156) had grade 3 disease; 57.6% (163) were MammaPrint (ultra) high-risk 2; and 48.1% (136) received standard NAC, while the rest received NAC combined with an investigational drug (Table 1). Clinicopathologic characteristics and treatment assignment were balanced between subtypes except for a significantly higher proportion of patients with grade 3 (68.1% vs. 42.8%) and MammaPrint (ultra) high-risk 2 tumors (88.4% vs. 28.3%), and RCB-0 (24.6% vs. 15.2%) in the TNBC group compared with the HR-positive/HER2-negative group (all Fisher’s exact test p<0.05).

Clinical samples.

Blood samples for ctDNA analysis were collected at pretreatment (T0), 3 weeks after treatment initiation (T1), and 12 weeks after treatment initiation between paclitaxel-based treatment and anthracycline regimens (T2), and after NAC before surgery (T3). Tumor biopsy for whole exome sequencing (WES) was collected at pretreatment (T0). Cell-free DNA (cfDNA) isolated from plasma served as the input for ctDNA analysis. Pretreatment core biopsies were subjected to WES, and germline DNA isolated from buffy coats was also sequenced and used as matched normal control. Peripheral blood was collected into EDTA-containing tubes and centrifuged at 1100–1300 g for 20 minutes at room temperature. The buffy coat and plasma were then aspirated and immediately frozen. Pretreatment core biopsies were collected from the primary breast tumor and mounted in Tissue-Tek^® O.C.T.^™ embedding media. A section from the frozen tissue was stained with hematoxylin and eosin (H&E) for pathologic evaluation to assess the percent tumor content. For tissue samples that met the tumor content requirement (30%), 8–10 cryosections of 30 μM thickness were further collected as ribbons using a microtome into cryovials. All samples (buffy coat, plasma, and tissue) were immediately frozen and stored at −80 °C until further processing.

Clinical samples were collected and stored at participating I SPY 2 trial sites. Samples were then shipped overnight on dry ice to the I-SPY Laboratory at the University of California San Francisco for accessioning and storage at −80 °C. Buffy coat and tissue samples were shipped to a commercial vendor for WES and plasma samples were shipped to Natera, Inc. for ctDNA analysis.

METHOD DETAILS

Quality control and analytic dataset.

Of the 295 patients with available pretreatment biopsies, 283 (96%) tumors were successfully analyzed by WES. In total, 1,024 plasma samples from 283 HER2-negative patients (145 HR-positive/HER2-negative and 138 TNBC) with WES data comprised the final analytic dataset (Figure S1A). This dataset includes ctDNA data from 65 HER2-negative patients (223 plasma samples) from previously published work. ¹⁵ Of the 283 patients, 251 were part of the I-SPY2-990 mRNA/RPPA Data Resource ⁹ with pretreatment gene expression.

ctDNA analysis.

ctDNA was analyzed in plasma collected at pretreatment (T0), 3 weeks after treatment initiation (T1), 12 weeks after treatment initiation between paclitaxel-based treatment and AC regimens (T2), and before surgery (T3) using the Signatera^™ test, a bespoke multiplex polymerase chain reaction (PCR) next generation sequencing-based assay. The test detects up to 16 patient-specific somatic mutations selected from WES data of pretreatment biopsies. The methods for ctDNA analysis, including cfDNA extraction, quantification, library preparation, provenance testing, WES workflow, and bioinformatics pipeline, have been previously described in detail. ¹⁵ Briefly, WES data derived from pretreatment core biopsies and matched normal blood samples from each patient were analyzed to select a set of 16 personalized (patient-specific), somatic, clonal, single nucleotide variants (SNVs) for multiplex PCR testing. ^{15,22,48–51} The multiplex PCR primers were designed and synthesized to track ctDNA in a patient’s plasma. Plasma samples with at least 2 of the 16 variants detected were defined as ctDNA-positive. ctDNA concentration was reported as mean tumor molecules (MTM) per ml of plasma ¹⁵. The MTM per ml of plasma was calculated by dividing the total number of mutant molecules by the number of targets detected. The number of mutant tumor molecules per mL of plasma was calculated as follows:

$$\frac{cfDNA\hspace{0.17em}extracted\left(ng\right)}{1\hspace{0.17em}ng}\times \frac{1000\hspace{0.17em}pg}{3.3\hspace{0.17em}pg\hspace{0.17em}per\hspace{0.17em}hGE}\times \frac{haploid genome equivalent\left(hGE\right)\hspace{0.17em}\hspace{0.17em}\times \hspace{0.17em}\hspace{0.17em}Variant Allele Fraction\left(VAF\right)}{Plasma volume for extraction\left(mL\right)}$$

The personalized and tumor-informed ctDNA test used in this study has been clinically validated in breast cancer ^15,25,51 and other cancer types. ^22,52,53 Bratman and colleagues used the same ctDNA technology to show that pretreatment ctDNA concentration and ctDNA dynamics were significantly correlated with response and survival in advanced cancer patients treated with pembrolizumab. ⁴⁸ Others have demonstrated that the same ctDNA test can guide patient selection to identify urothelial cancer patients who will benefit from adjuvant atezolizumab. ²²

Gene expression analysis.

Full transcriptome gene expression data from pretreatment biopsy were generated for each I-SPY2 patient on an Agilent microarray. The expression data were used to calculate the MammaPrint Score, which is required to assess patient eligibility for the trial. In a recent publication from our group, gene expression data were used to classify each patient into one of the 5 response-predictive subtypes. ⁹ Here, we used the gene expression data to discover genes and pathways associated with pretreatment ctDNA positivity.

Of the 283 patients, 251 had tumor gene expression (GSE194040) ⁹ and ctDNA data at pretreatment. Gene expression between ctDNA-positive and ctDNA-negative patients at pretreatment for each subtype: HR-positive/HER2-negative (n=131, 66% ctDNA-positive) and triple-negative breast cancer (TNBC, n=120, 91% ctDNA-positive) was evaluated independently for differential expression (limma v3.48.3), ³⁵ gene set enrichment (fgsea v1.18.00), ³⁶ and protein-protein interaction (STRING v11.5). ³⁷

QUANTIFICATION AND STATISTICAL ANALYSIS

This study examined the clinicopathologic and molecular correlates of ctDNA positivity. Given the differences in the biology and clinical histories between HR-positive/HER2-negative and TNBC, we hypothesized that the predictive and prognostic value of ctDNA may vary between the two subtypes. The analyses performed in the study are outlined in Figure S1B.

Statistical analysis of the clinical and gene expression data

The response endpoints used in the study were pCR and RCB, the latter of which is divided into 4 classes: RCB-0 (no residual disease equivalent to pCR), RCB-I (minimal burden), RCB-II (moderate burden), and RCB-III (extensive burden). The survival endpoint was distant recurrence-free survival (DRFS), defined as the time interval between the date of patient consent for treatment and the date of clinical diagnosis of metastatic recurrence or death by any cause. A p <0.05 was considered significant for the statistical tests described below.

ctDNA vs. clinicopathologic characteristic.

Association between ctDNA status and categorical variables was assessed using Fisher’s exact test. For continuous clinical variables, a t-test or analysis of variance (ANOVA) was performed to compare means between groups. The ctDNA concentration (MTM/ml) was compared across clinicopathologic groupings using the Wilcoxon rank sum test (2 groups) or the Kruskal-Wallis test (3 or more groups).

ctDNA vs. response.

A Fisher’s exact test was used to compare the proportions of patients who achieved pCR vs. those who did not, or by RCB class, stratified based on their c ctDNA dynamics. To assess the association between ctDNA dynamics and response, patients who tested ctDNA-positive at T0 were grouped according to ctDNA status at T0, T1, and T2: early clearance at T1 (3 weeks after initiation of treatment, ctDNA+/−/−), late clearance at T2 (at 12 weeks, after paclitaxel-based treatment, ctDNA+/+/−) and no clearance at T2 (ctDNA +/+/+). We also used logistic regression to estimate odds ratios and 95% confidence intervals of achieving a pCR (or RCB-0/I).

ctDNA vs. survival.

Of the 283 patients, 272 (96.1%) had DRFS data, of whom 74 (27.2%) experienced metastatic recurrence or death. The median follow-up was 3.10 years (range 0.46–7.6) for the HR-positive/HER2-negative group and 3.12 years (range 0.31–7.91) for the TNBC group. In the HR-positive/HER2-negative group, 32/142 (22.5%) experienced a DRFS event with 28 distant recurrences and 4 deaths, while in the TNBC group, 42/130 (32.3%) experienced a DRFS event with 33 distant recurrences and 9 deaths.

We investigated whether ctDNA positivity at different time points was associated with DRFS. We also examined whether ctDNA status after NAC (T3) could further refine risk stratification by pCR and RCB class (see main text).

Furthermore, we examined the prognostic significance of ctDNA dynamics in each subtype. Patients with available ctDNA data for all 4 time points [115 (79.3%) HR-positive/HER2-negative and 88 (63.8%) TNBC patients] were classified into 5 groups based on ctDNA dynamics: Group 1 includes patients who were ctDNA-negative at T0 and remained negative until T3; Groups 2, 3, and 4 include patients who were ctDNA-positive at T0 and cleared their ctDNA at T1, T2, or T3, respectively; and Group 5 includes those where positive at T0 and did not clear ctDNA at T3. We excluded 2 patients from each subtype from the analysis because they did not fall into one of the 5 groups based on ctDNA dynamics. For the HR-positive/HER2-negative subtype, one patient was ctDNA−/+/−/− and the other ctDNA−/−/−/+; and for the TNBC subtype: 2 patients were ctDNA−/−/+/−. A total of 110 of 115 HR-positive/HER2-negative and 83 of 88 TNBC patients had follow-up data for survival analysis.

Hazard ratios and 95% confidence intervals were estimated using Cox proportional hazards regression analysis. Survival curves were plotted using Kaplan-Meier analysis, and p-values were calculated using a log-rank test. In multivariable analysis, we chose pCR as a covariate based on recent findings in the I-SPY2 trial, showing a strong prognostic impact of pCR on survival in neoadjuvant-treated patients. ⁴ The R package “survival” was used for Cox proportional hazards model, Kaplan-Meier survival analysis, and log-rank tests.

Differential expression analysis.

Differential expression analysis on the global transcriptome data (m=19,134 genes, GSE194040) ⁹ between ctDNA-positive and ctDNA-negative patients at pretreatment was performed in limma (v3.48.3) with no covariates. ³⁵ Additionally, to explore the impact of proliferation in the HR-positive/HER2-negative subtype, Differential expression analysis was also performed with adjustment for MammaPrint status.

Linear modeling was performed on the log2-transformed expression data; the empirical Bayes moderation of computed statistics was applied with intensity-trend allowed for the prior variance. Significantly differentially expressed (DE) genes were defined as Benjamin-Hochberg adjusted p<0.05. For TNBC subtypes with unbalanced groups, we also report genes with nominally significant DE at p<0.05. Genes with log2 fold changes (lfc) were upregulated in ctDNA-positive patients, and those with negative lfc were upregulated in ctDNA-negative patients. Genes were considered commonly DE between HR-positive/HER2-negative and TNBC subtypes if they were at least nominally DE (p<0.05) in each subtype and had concordant direction of log2 fold changes.

DE genes were visualized via hierarchical clustering heatmaps (pheatmap v1.0.12) using hclust Ward’s clustering criterion (ward.D2) agglomerative method with Euclidean distances as the distance metric. Protein-protein interactions (PPI) and network analysis were performed using the STRING database (v.11.5, https://string-db.org/) with minimum interaction confidence of 0.4. PPI network was visualized in igraph (v.1.2.6), ⁵⁴ and only the largest fully connected main network is visualized. To identify gene communities—defined as natural divisions of densely connected subgroups or community structures determined algorithmically ⁵⁵—community detection was performed on this main network using igraph::cluster_optimal algorithm. ⁵⁶ Each community was analyzed in STRING for functional network enrichment (FDR p<0.05), and representative KEGG pathways were annotated.

Gene set enrichment analysis.

Gene set enrichment analysis (GSEA) of Molecular Signatures Database Hallmark (H, m=50) gene sets was performed in fgsea (v1.18.00) in HR-positive/HER2-negative and TNBC subtypes. ³⁶ Fast GSEA was performed on ordered DE t-statistics using 1000 permutations with minimum and maximum gene set sizes set to 15 and 500, respectively. Significantly enriched gene sets were defined as those with enrichment BH adjusted at p<0.05; nominally enriched gene sets were defined at p<0.05. Gene sets with positive normalized enrichment scores (NES) were enriched in ctDNA-positive patients, and those with negative NES were enriched in ctDNA-negative patients. Enrichment plots were generated for immune-associated gene sets with at least nominal enrichment (p<0.05) in either subtype. Leading-edge genes—core member genes that contribute to enrichment score—belonging to these immune-associated gene sets were visualized via hierarchical clustering heatmaps (pheatmap v1.0.12) using hclust Ward’s clustering criterion (ward.D2) agglomerative method with Euclidean distances as the distance metric.

ADDITIONAL RESOURCES

Patient-level data are available in Table S1. ctDNA and clinical data, Related to Figures 1 and 2.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Plasma samples	I-SPY2 Trial	https://clinicaltrials.gov/ct2/show/NCT01042379
Tumor biopsy before treatment	I-SPY2 Trial	https://clinicaltrials.gov/ct2/show/NCT01042379
Critical commercial assays
Signatera™	Natera	https://www.natera.com/oncology/signatera-advanced-cancer-detection/
Deposited data
Transcriptomic	9	Gene Expression Omnibus (GEO) GSE194040 (mRNA) https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE194040
Patient-level ctDNA and clinical data	This study	Table S1. ctDNA and clinical data, Related to Figures 1 and 2.
Software and algorithms
limma v.3.48.3	35	https://bioconductor.org/packages/release/bioc/html/limma.html
fgsea v.1.18.	36	https://bioconductor.org/packages/release/bioc/html/fgsea.html
STRING v11.5	37	https://string-db.org/
stats R package (v.3.6.3)	38	https://stat.ethz.ch/R-manual/R-devel/library/stats/html/stats-package.html
survival R package (v.3.1–12)	39	https://CRAN.R-project.org/package=survival

Highlights.

• Early clearance of ctDNA in triple-negative patients associates with good response

• ctDNA dynamics during neoadjuvant chemotherapy predict clinical outcomes

• ctDNA-negativity associates with improved survival despite having residual cancer

• Expression analysis reveals pathways associated with ctDNA shedding