Patient clinicopathologic characteristics and ctDNA features
130 female patients with stage II–III TNBC received NAC followed by surgery were prospectively enrolled in this study (Table 1, Figure S1A). The median age of the study cohort was 44 years (range 25–65 years), with the median follow-up at 24.2 months (range 6.0 to 45.3 months) as of November 1st, 2023. Most patients were stage II (80.8%) and clinically lymph node positive (56.2%). After NAC, 44 (33.9%) patients achieved total pCR. 15 patients experienced recurrence (12 distant metastasis and 3 local relapse) and 2 patients died of disease during follow-up.
Overall, 130 tumor tissues (including 123 core needle biopsy samples and 7 postoperative tumor samples when biopsy samples were insufficient for sequencing) were successfully sequenced using a 1021 cancer related-gene panel previously reported (Table S1)10. A total of 625 blood samples, including 122 baseline (pre-NAC) samples, 113 post-NAC (pre-surgery) samples, 118 post-surgery samples taken 1–8 weeks after surgery or radiation and 272 follow-up samples (every 3–6 months after surgery) were subjected to parallel ultra-deep sequencing (>10,000x) using the same 1021 panel followed by tumor-guided ctDNA analysis (Fig. 1A, Figure S1B).
In total, 1759 variants in 436 genes were detected in tumor tissue samples of 130 patients, with 2–76 variants (median 10 variants) per sample. The top 5 altered genes were TP53, MYC (CNV), PIK3CA, MCL1 (CNV), MDM4 (CNV) (Figure S2).
Among 122 patients with baseline blood samples before any treatment, 96 (78.7%) of them had detectable ctDNAs. A total of 485 variants in 227 genes were detected, with a median of 3 variants per sample (range 1 to 39). TP53, PIK3CA, PTEN, LRP1B, RB1 were the most frequently altered genes (Fig. 1B). Positive baseline ctDNA was significantly correlated with high Ki67 (≥ 50%) (p = 0.01), but not with age, tumor size, lymph node status (Fig. 1C). Nevertheless, larger tumors, positive lymph nodes and high Ki67 were associated with higher baseline ctDNA concentration (measured by maximum variant allele fraction, MVAF) (Fig. 1D).
Dynamic ctDNA response after NAC
It was reported that NAC could effectively clear ctDNA and thus decrease relapse22. In our cohort, the ctDNA positive rate dropped sharply from 79–23% (26/113) at the post-NAC timepoint (Fig. 2A). Patients with negative ctDNA at the post-NAC timepoints were more likely to achieve pCR or higher Miller-Payne grade than those with positive ctDNA at the post-NAC timepoint (Fig. 2B, Figure S3A) indicating better local and systemic response to NAC in ctDNA negative patients. A significant drop of ctDNA concentration was also observed after NAC (p < 0.001, Fig. 2C). We then compared ctDNA concentration changes between pCR and non-pCR patients. There were similar ctDNA concentrations between pCR and non-pCR patients at the baseline (p = 0.76, Fig. 2D). However, non-pCR patients had significantly higher post-NAC ctDNA concentrations than pCR patients (Fig. 2D). Explorative analysis was performed to delineate ctDNA mutation profile and their change after NAC between patients with pCR and non-pCR. All TP53 mutations and PIK3CA mutations in blood were eradicated by NAC in pCR patients, while about 24% and 8% of non-pCR patients still had detectable post-NAC TP53 mutations and PIK3CA mutations respectively (Fig. 2E), suggesting that continued presence of TP53 or PIK3CA mutations indicates non-pCR status.
Furthermore, the post-surgery ctDNA positive rate further dropped to 16.9% (20/118), indicating the removal of primary tumor can further decrease ctDNA positivity in certain patients.
ctDNA status at different timepoints of treatment and the prognosis of TNBC
To determine the optimal timing of ctDNA sampling during the treatment of eTNBC, we analyzed the ctDNA positivity at different timepoints and correlated them with the EFS and distant recurrence free survival (DRFS).
The patients with positive baseline ctDNA showed a trend of worse EFS (Fig. 3A, p = 0.21, HR = 3.39, 95% CI: 0.44–26.3) and DRFS (Fig. 3B, p = 0.08, HR not calculable) than the ones with negative baseline ctDNA, but the differences were not statistically significant. Nevertheless, it is worthy of noting that none of the patients with negative baseline ctDNA had distant metastasis, indicating this subgroup of patients (21% of total, clinical stage: 13 IIA, 11 IIB and 2 IIIC) has a very low systemic tumor burden and very low-risk of distant metastasis.
The presence of positive ctDNA post-NAC and pre-surgery, as well as the ones post-surgery, were associated with significantly worse EFS (Fig. 3C, p = 0.002, HR = 4.34, 95% CI: 1.51–12.44; Fig. 3E, p = 0.002, HR = 4.63, 95% CI: 1.60-13.38) and DRFS (Fig. 3D, p < 0.001, HR = 6.18, 95% CI: 1.95–19.58; Fig. 3F, p < 0.001, HR = 6.25, 95% CI: 2.01–19.42), indicating their strong value in predicting the prognosis of eTNBC patients.
Post-surgery ctDNA represents the earliest MRD status and has been shown as a crucial marker to guide adjuvant chemotherapy in a randomized trial of early colon cancer23. However, it’s noteworthy that among the 12 eTNBC patients who had post-surgery ctDNA testing and later developed distant metastasis, only six of them had positive ctDNA at the post-surgery timepoint. Thus, the sensitivity of post-surgery ctDNA to predict metastasis was only 50%, indicating this single post-surgery timepoint is not good enough to select most high-risk eTNBC patients. This result is in line with the complexity of ctDNA dynamics throughout the management of eTNBC and underscore the need for a more comprehensive approach in using ctDNA to assess eTNBC prognosis.
Higher threshold of baseline ctDNA better stratifies the prognosis of eTNBC patients
Although the detection of baseline ctDNA in eTNBC patients could be as high as 73–100%9,16,17,19,20, yet only ~ 30% of such patients will develop relapse24. Furthermore, 21% of eTNBC patients had negative ctDNA in our study and none of them developed metastasis during follow-up, indicating there may be a threshold of baseline ctDNA that could distinguish high-risk from low-risk TNBC patients.
Using the data from our previous study20, a ROC model was constructed to explore the optimal cutoff value to distinguish high-risk and low-risk patients. The ROC model was statistically significant (AUC: 0.73, 95% CI: 0.5424–0.9176) with the optimal cutoff value being 1.1% of MVAF (Fig. 4A). Next, the cutoff value was applied into the present study. The patients with higher baseline ctDNA (defined as MVAF higher than 1.1%, 53/122, 43.4%) demonstrated significantly worse EFS (Fig. 4B, p = 0.014, HR = 4.49, 95% CI = 1.21–16.60) and DRFS (Fig. 4C, p = 0.002, HR = 13.02, 95% CI = 1.65–102.8), than the patients with lower baseline ctDNA (defined as MVAF lower than 1.1%, 69/122, 56.6%). The sensitivity and AUC of baseline ctDNA at 1.1% threshold were 90% and 0.75 respectively, better than any single timepoint without threshold (Table S2). For external validation, we employed the I-SPY2 cohort17 that enrolled eTNBC patients with similar clinical characteristics (Figure S3B). Per the calculation formula provided in the I-SPY2 study, the cutoff value of 1.1% MVAF was converted to 28.27 mean tumor molecules per milliliter (MTM/mL) using the median cfDNA extraction concentration (ng/mL) of the study. In this external validation, the patients with baseline ctDNA concentration higher than 28.27 MTM/mL had significantly worse DRFS (Fig. 4D, p = 0.002, HR = 2.55, 95% CI = 1.37–4.73) than the patients with ctDNA lower than 28.27 MTM/mL in the I-SPY2 trial. The specificity and AUC with threshold in I-SPY2 were better than the ones without threshold (Table S2), demonstrating the value of such a threshold of ctDNA.
Systemic tumor burden with pathologic response to NAC identify TNBCs with different relapse risk
It is known that effective NAC can eradicate not only local tumor, but also ctDNA in eTNBC patients. To reflect the efficacy of NAC and surgery, the results of ctDNA from baseline and post-surgery timepoints were integrated to build a systemic tumor burden model. The patients with high systemic tumor burden (63/119, 52.9%), defined as high baseline ctDNA (more than 1.1% MVAF) or detectable ctDNA after surgery, had significantly inferior EFS (Fig. 4E, p = 0.029, HR = 3.76, 95% CI = 1.05–13.49) and DRFS (Fig. 4F, p = 0.004, HR = 11.05, 95% CI = 1.43–85.63) than other patients with low systemic tumor burden (56/119, 47.1%). The systemic tumor burden model also exhibited a high sensitivity of 91.7% for DRFS (Table S2). Moreover, systemic tumor burden remained independently prognostic in a multivariate Cox model (Table S3) after adjusted for clinical characteristics and pathologic response.
Many studies had shown non-pCR status after NAC is strongly associated with worse EFS and overall survival in TNBC patients25–27. The patients who achieved pCR in our study also had markedly better EFS and DRFS (Figure S3C,D). Thus, we tried to further refine prognosis stratification by combining systemic tumor burden with pathologic response. Not surprisingly, the patients with high systemic tumor burden and non-pCR status had the worst EFS and DRFS (Fig. 4G, H), suggesting this subgroup (10 distant metastasis out of 42 patients) is in urgent need of more effective adjuvant therapy after surgery. Among the 21 patients with high systemic tumor burden but reached pCR, only one patient had distant recurrence, indicating excellent response to NAC can significantly enhance the prognosis of TNBC patients with high systemic tumor burden. Among the 37 patients with low systemic tumor burden and non-pCR status, 3 patients had recurrence including 2 local relapses in the ipsilateral breast after breast conserving surgery and 1 distant metastasis. Moreover, the 19 patients with low systemic tumor burden and reached pCR had 100% EFS and DRFS (Fig. 4G, H), suggesting this subgroup is a highly curable population. Taken together, the combinational use of systemic tumor burden with pCR status after NAC provides an excellent tool to separate TNBC patients with high or low-risk of metastasis right after surgery, which may help to tailor personized adjuvant treatment.
MRD surveillance after surgery precedes clinical imaging to detect relapse
Positive ctDNA after surgery represents un-eradicated tumor cells in the patients and the presence of MRD, which is associated with worse prognosis13,19. Next, we integrated longitudinal follow-up timepoints with the post-surgery timepoint. Patients remained ctDNA negative during follow-up (at least 2 timepoints) were defined as MRD negative, while patients with detectable MRD at any timepoints after surgery were defined as MRD positive. 82 of 130 patients had longitudinal MRD results, of which 44 were negative and 38 were positive. Remarkably, patients with negative MRD had significantly better EFS (Fig. 5A, p < 0.01, HR = 7.70, 95% CI = 1.71–34.83) and 100% DRFS (Fig. 5B, p < 0.001) than patients with positive MRD. The positive predictive value (PPV) and negative predictive value (NPV) of longitudinal MRD for DRFS were 29% and 100% respectively. MRD remained an independent prognostic factor in a multivariate Cox regression model after adjusting for clinicopathological characteristics (Table S4). Interestingly, two patients who had a local relapse in the ipsilateral breast after breast conserving surgery remained MRD negative during follow-up (Fig. 5C), suggesting that MRD surveillance is more effective in monitoring distant recurrence but not local relapse.
Because MRD surveillance is quite expensive, it is valuable if we can further narrow down the subpopulation that needs close monitor. Among the 38 MRD positive patients, 31 were high systemic tumor burden and 10 of them had distant metastasis (Table S5). The sensitivity, PPV and NPV of systemic tumor burden are markedly higher than baseline ctDNA with the threshold, indicating that systemic tumor burden is a good marker in identifying MRD positive patients.
We then examined the temporal patten of ctDNA MRD and the timing of recurrence during follow-up. 15 of 130 (11.5%) patients had regional or distant recurrence as of November 1st, 2023 (Fig. 5C), among which 14 patients had post-surgery ctDNA samples. 11 out of the 14 (78.6%) patients had positive MRD before clinical relapse. The median lead time between MRD positivity and clinical relapse were 3.4 months (range 0.3 to 12.4 months). MRD monitoring is a good tool for post-surgery surveillance of distant recurrence and may provide a window of opportunity to escalate the systemic treatment while micro-metastasis starts to appear.