Breast cancer (BC) is well documented to be the most prevalent malignancy and the leading cause of cancer-related mortality among women worldwide. According to GLOBOCAN 2023 data, more than 2.3 million new cases and approximately 670,000 deaths have been reported annually. Currently, developing countries exhibit higher age-standardized incidence rates (ASR, 34.8 per 100,000) than developed nations (ASR, 34.4 per 100,000), and concerning trends toward earlier age at onset and higher relapse ratios have been observed¹. Although multidisciplinary therapies combining surgery, radiotherapy, chemotherapy, endocrine therapy, and targeted agents have significantly improved outcomes, 30%–40% of patients with early-stage BC have elevated recurrence risk, and the 5-year overall survival rate for metastatic BC remains below 30%^2,3. Two recurrence peaks have been observed at 2–3 years and 5–7 years in hormone receptor (HR)-positive BC, whereas recurrence peaks generally occur at 2–3 years in triple-negative BC (TNBC) and human epidermal growth factor receptor 2 (HER2)-positive BC with visceral and brain metastases⁴. The combination of novel CDK4/6 inhibitors (e.g., abemaciclib and ribociclib) combined with endocrine therapy decreases the 5-year recurrence risk in clinical high-risk HR+/HER2− patients^2,5–7, and the prognosis of HER2-positive BC has markedly improved in patients treated with dual-targeted regimens and antibody-drug conjugates^8,9. Nevertheless, these breakthroughs require precise recurrence risk stratification tools for individualized and precision management^9–11. Current post-operative monitoring relies primarily on imaging modalities (ultrasound, mammography, and CT/MRI)¹² and serum biomarkers (CEA and CA15-3), which have low sensitivity and specificity^13,14. Although tissue biopsy remains the gold standard, its invasive nature limits repeatability and cannot capture tumor spatiotemporal heterogeneity^14,15. Minimal residual disease (MRD) testing via circulating tumor DNA (ctDNA) addresses these gaps by enabling ultra-sensitive detection of residual tumor cells and prognostic stratification, and providing a lead-time advantage. MRD can further guide precision interventions: ctDNA clearance correlates with therapeutic efficacy, and rising levels prompt regimen adjustments. Plasma ESR1 and PIK3CA mutations in BC are required to guide ESR1-proteolysis targeting chimeras (PROTACs)¹⁶, oral selective estrogen receptor degraders (SERDs)¹⁷, PI3K/AKT inhibitors¹⁸, and other targeted treatments. MRD detection, by bridging the critical divide between population-based adjuvant therapy and individualized management, has transformed relapse monitoring from reactive diagnostics to proactive intervention.

Contemporary risk stratification integrates clinicopathological prognostic models (e.g., ADJUVANT!) and RNA assay-based molecular subtypes (PAM50). Multigene assays (Oncotype DX and MammaPrint) analyzing 21-/70-gene expression profiles enable chemotherapy de-escalation in selected low-risk patients, yet have demonstrated limited benefits for patients with HR-positive, HER2-negative BC and low tumor burden in axillary lymph nodes^19,20. Consequently, their applicability is restricted to a specific subset of patients. Tissue-based biomarkers (e.g., Ki-67 and TP53 mutations) provide insights into tumor proliferation and aggressiveness but are limited by sampling bias and heterogeneity-induced variability²¹. Through multi-omics analysis (genomics, transcriptomics, metabolomics, and proteomics) of 579 patients, the first Fudan Classification (SNF subtypes) was proposed for luminal BC with varying sensitivity to targeted therapies or immune checkpoint inhibitors²². An AI model using digital pathology (H&E slides) has been developed for rapid subtyping, and clinical trials to validate these precision strategies are ongoing. Spatial heterogeneity in immune cell topology has been identified to predict responses to SHR-A1811 (accuracy: 86%)¹⁰. Antigen-presenting mast cells play a critical role in immune therapy resistance²³. Nonetheless, current approaches rely solely on tissue-based multi-omics analysis and have not yet incorporated liquid biopsy for multidimensional classification-guided precision therapy^3,24,25.

Liquid biopsy has revolutionized non-invasive MRD monitoring through ctDNA²⁶, circulating tumor cells (CTCs)²⁷, cell-free RNA²⁸, tumor-educated platelets²⁹, and exosomes^30–32. ctDNA demonstrates superior clinical utility to CTCs and exosomes in post-operative recurrence monitoring, because of its enhanced sensitivity³³, temporal resolution, and analytical robustness³⁴. Although CTC detection remains constrained by low abundance in early-stage disease and technical challenges in cell capture/viability, ctDNA assays can identify tumor-specific mutations even at MRD levels through highly sensitive droplet digital polymerase chain reaction (ddPCR)^30,35–37 or next-generation sequencing (NGS)-based methods³⁸. Similarly, exosome-based profiling faces biological heterogeneity and standardization challenges in isolation/analysis. In contrast, the relatively short half-life of ctDNA enables near real-time monitoring of tumor dynamics and can capture early molecular relapse weeks to months before radiographic recurrence³⁹. This temporal advantage, combined with standardized quantification of tumor-derived genomic alterations (e.g., methylation signatures and somatic mutations)⁴⁰, positions ctDNA as a dynamic biomarker for longitudinal risk stratification and therapy response assessment⁴¹. Moreover, its compatibility with high-throughput sequencing facilitates comprehensive tumor evolution analysis, which remains unattainable via single-cell CTC characterization or fragmented exosomal RNA profiling. Consequently, ctDNA might potentially be the most promising liquid biopsy tool for guiding personalized adjuvant interventions. ctDNA analysis provides a molecular snapshot of the tumor’s genetic landscape. This method reveals specific mutations and alterations, which are crucial for identifying targetable genomic changes and monitoring the emergence of resistance during targeted therapies. CTCs are intact, viable cancer cells that can be enumerated to assess tumor burden, and to conduct protein expression analysis and even ex vivo culturing for drug susceptibility testing. The combined application of ctDNA and CTCs transcends a single diagnostic lens by providing multidimensional characterization of the disease, thereby enhancing the accuracy of prognosis, monitoring, and treatment management (Figure 1).

Importantly, ctDNA detection based on whole-exome sequencing (WES) or target NGS panels enables recurrence prediction 7.9–11 months earlier than imaging techniques. NGS-based personalized MRD panels can monitor clonal evolution and resistance mutations (ESR1 and PIK3CA), and CTC phenotyping can aid in predicting metastatic potential. Meta-analyses have demonstrated a 40% reduced disease-free survival in patients with ≥5 CTCs than those with <5 CTCs per 7.5 mL of blood. Prospective trials (c-TRAK-TN, PADA-1, and SERENA-6) have concluded that ctDNA-guided therapeutic intensification (e.g., pembrolizumab augmentation) or strategy switching (replacing aromatase inhibitors with fulvestrant or camizestrant) in metastatic BC extends the median progression-free survival⁴². These advancements have marked a paradigm shift from population-based approaches to MRD-driven precision intervention in BC management.

Recent advances in high-throughput sequencing and bioinformatics have made MRD detection clinically useful, and ctDNA-based methods have emerged as critical tools for monitoring residual cancer burden. Two main methods, tumor-informed (patient-specific) and tumor-naive strategies, each with distinct operational frameworks and clinical implications, are used for contemporary MRD testing⁴³. The tumor-agnostic or strategy, under the tumor-naive umbrella, refers to assays designed to detect a pre-defined, fixed set of genomic alterations (e.g., mutations or methylation patterns) that are common across many cancer types (Figure 2).

The tumor-informed strategy starts with deep sequencing of the patient’s primary tumor tissue, typically via custom-targeted multiple panels or WES, to identify somatic mutations unique to the individual’s malignancy. A personalized ctDNA assay is subsequently designed to monitor these tumor-derived variants in serial blood samples through ultra-deep sequencing (≥ 30,000× coverage or ≥ 100,000× coverage)^44,45. Notably, this approach achieves exceptional sensitivity (detection limits as low as 0.001% variant allele frequency, VAF) by minimizing background interference from clonal hematopoiesis or germline polymorphisms. Moreover, its adaptability to tumor evolution enables monitoring of clonal dynamics during therapy and provides critical insights into emerging resistant subpopulations⁴⁶. However, because this method is limited by its dependency on high-quality tumor tissue, it has limited applicability in cases with inadequate or degraded biopsy specimens. The requirement for WES analysis and custom panel design poses challenges in rapid clinical deployment, because it extends turnaround times (typically 2–4 weeks) and elevates costs.

In the context of ultra-low ctDNA abundance monitoring after early-stage tumor resection, transitioning from WES to a tumor-informed whole-genome sequencing (WGS) baseline strategy significantly enhances MRD detection performance. WGS achieves this enhancement through comprehensive genomic coverage, which captures additional individualized variant sites (e.g., non-coding region mutations and structural variations) and consequently increases the number of personalized monitoring targets. Concurrently, by integrating high-depth sequencing with noise suppression algorithms, it effectively attenuates background noise and increases the signal-to-noise ratio for low-frequency signals. These synergistic effects elevate detection sensitivity to the parts per million (ppm) level (1–8 ppm) and therefore enable precise early detection and dynamic tracking of post-operative trace MRD⁴⁷. However, because WGS generates 50–100 times more raw data than WES, it requires substantial analytical resources and significantly higher costs; as a result, its widespread clinical adoption is limited.

The NeXT Personal MRD platform has provided a major advancement in BC monitoring, through its innovative use of WGS to detect ctDNA at unprecedented sensitivity and specificity. This ultrasensitive approach, which enables identification of residual disease at levels as low as 1 ppm, significantly expands the window for early intervention and improves patient outcomes. By leveraging tumor-informed ctDNA panels, the platform ensures high specificity, thus minimizing false positives and enabling evaluation of each patient’s disease status with enhanced precision. Studies have highlighted the prognostic relevance of ctDNA detection, by demonstrating its strong association with relapse risk and survival outcomes. In an early-stage non-small cell lung cancer (NSCLC) cohort, NeXT Personal MRD, compared with the WES-based > 100-plex PCM™ monitoring strategy and the 16-plex Signatera strategy (179 d vs. 119 d vs. 70 d), significantly prolonged the lead time before radio-relapse. The NeXT Personal MRD method was used to stratify the preoperative ctDNA-negative population (below 0.008%) in the TracerX cohort (100+ sites) monitored by Invitae PCM into 2 groups (ultra-low MRD+ and ctDNA-negative populations) with different overall survival. The recurrence risk and mortality in ctDNA-negative patients approached those in the cured postoperative population but were significantly higher in patients with ultra-low MRD positivity [between 0.0001% and 0.008%; RFS: P = 4.4e-5; overall survival (OS): P = 0.0017]. In patients with early BC, the NeXT Personal WGS-powered ultrasensitive ctDNA MRD assay detected ctDNA in 98% of patients at diagnosis, and significantly outperformed exome-based and dPCR-based assays. This outstanding sensitivity enabled earlier detection of MRD and improved the lead time to clinical relapse (median lead time of 15 months with the WGS-based MRD method). The assay’s prognostic value was also demonstrated: ctDNA detection was strongly associated with shorter RFS and OS than those without ctDNA detection. The assay identified a subset of patients with low-level MRD who cleared ctDNA and did not relapse during long-term follow-up (64/64), thus highlighting its potential for risk stratification and guiding treatment decision-making⁴⁵. Overall, the NeXT Personal MRD platform has the potential to transform BC management, by enabling earlier detection of relapse, more accurate assessment of disease status, and improved patient outcomes through personalized treatment approaches.

The PhasED-seq method leverages the unique characteristics of phased variants (PVs), defined as 2 or more mutations occurring within a single DNA fragment, located on the same DNA strand, in tumors⁴⁸. Through design of a specific sequencing panel, PhasED-seq enriches and detects these PVs, and therefore achieves ctDNA detection with enhanced sensitivity. PVs exhibit highly consistent genomic distribution in B-cell malignancies, probably because of physiological and aberrant somatic hypermutations. PhasED-seq uses a sequencing panel targeting PVs, which can capture PVs from tumor or plasma samples for monitoring residual disease. Its advantages include its low background signal and enhanced DNA molecule recovery efficiency. Compared with traditional single-nucleotide variant (SNV)-based methods, PhasED-seq achieves a superior detection sensitivity capable of identifying ctDNA at the one-in-a-million level⁴⁸. Consequently, PhasED-seq can detect low-burden residual disease. In clinical settings, PhasED-seq has demonstrated remarkable results in patients with B-cell lymphoma. Specifically, PhasED-seq has detected ctDNA that could not be identified by Cancer Personalized Profiling by Deep Sequencing (CAPP-Seq) and was associated with poor prognoses. PhasED-seq is applied to solid tumors through a personalized approach. WGS of tumor and healthy tissues is conducted to identify patient-specific PVs, and a personalized sequencing panel targeting these PVs is designed and used to sequence plasma samples for residual disease monitoring. Notably, PhasED-seq has also demonstrated promising performance in solid tumors. In lung adenocarcinoma and BC, PhasED-seq has detected ctDNAs that CAPP-Seq could not identify and provided more accurate prediction of disease progression.

Structural variations (SVs), prevalent cancer genome features that include breakpoints and rearrangements, are tumor-specific and reflect tumor biology. Given the high inter-individual variability of SV breakpoints, this assay detects low ctDNA levels on the basis of SVs from tumor tissue WGS, and the design of a personalized panel effectively decreases the false positives commonly encountered in SNV-based detection approaches^49,50. In one study, ctDNA was detected in blood samples from 96% of participants at baseline, with a median VAF of 0.15%. All participants with detectable ctDNA post-surgery or during follow-up experienced recurrence, and the median lead time was 417 days⁴⁹. On the basis of tumor WGS, Signatera Genome MRD has revealed a strong association between ctDNA negativity and improved RFS and OS among 126 patients tested within the MRD window period (within 3 months after surgery and before the initiation of adjuvant therapy). The hazard ratios for RFS were 10.0 in the overall patient population and 17.0 in patients with extracranial recurrence⁵¹. The MONSTAR-SCREEN-3 study, using ultra-sensitive WGS combined with ctDNA analysis, has demonstrated that ctDNA is cleared in patients achieving pathological complete response (pCR) but persists in patients without pCR. Patients who were landmark MRD negative after surgery had no recurrence within 6 months, thus indicating excellent short-term prognosis, whereas MRD-positive patients who did not receive adjuvant therapy faced an 80% recurrence risk within 6 months; these findings confirmed MRD as a valid predictor of treatment efficacy and prognosis⁵².

The tumor-informed WGS-based MRD strategy is considered a highly sensitive ctDNA detection method offering enhanced precision for disease monitoring and prognosis assessment.

The tumor-agnostic approach uses fixed, predefined gene panels targeting recurrent driver mutations, methylation signatures, or fragmentomic patterns to detect ctDNA directly from plasma without prior tumor sequencing. Techniques such as multiplex PCR or hybrid capture-based NGS enable standardized workflows, deliver results within 7 days, and are ideal for large-scale screening. This approach broadens applicability to patients with inaccessible tumors (inoperable cases or long-interval samples) and integrates multi-omic enhancements. Moreover, use of methylation markers improves sensitivity in low-mutation-burden cancers by leveraging dense epigenetic alterations.

Despite operational efficiency, tumor-agnostic assays exhibit lower sensitivity (typical limit of detection: 0.1% VAF) than tumor-informed methods, because of restricted genomic coverage. The use of fixed panels increases the risk of false positives arising from clonal hematopoiesis of indeterminate potential (CHIP)-derived mutations and might result in suboptimal performance in tumors with low mutation burdens⁵³. Platforms such as Guardant Reveal and FoundationOne Liquid CDx have demonstrated utility for early relapse prediction in NSCLC and colorectal cancer (CRC)^54,55, although their prognostic accuracy remains inferior to that of tumor-informed assays.

CAPP-Seq, an ultrasensitive tumor-agnostic ctDNA detection technology developed in 2014, was introduced in response to the urgent clinical need for dynamic monitoring of solid tumors³⁵. Conventional ctDNA detection faced limitations arising from insufficient sensitivity to capture trace DNA released by early-stage tumors or residual disease (typically accounting for less than 0.01% of cell-free DNA). The technological breakthrough, achieved by integrating cancer gene databases such as COSMIC to design cancer-specific “selectors” targeting high-frequency mutation regions, achieves a sensitivity of 0.02% mutant allele frequency, equivalent to detecting one tumor DNA molecule among 10,000 normal DNA molecules. This method enables early diagnosis and staging, as evidenced by a detection rate of 50% patients with stage I NSCLC and 100% of patients with stage II–IV NSCLC, with a specificity of 96%. Furthermore, ctDNA levels significantly correlate with tumor volume (R² = 0.89, P = 0.0002), and the assay can assist in residual lesion identification⁵⁶. For treatment monitoring and prognosis, ctDNA dynamics change earlier than radiographic responses, as illustrated in a stage IIIB lung cancer case in which ctDNA elevation during apparent radiographic remission preceded clinical recurrence by 7 months. Postoperative ctDNA positivity is associated with a significantly shorter progression-free survival than that with ctDNA-neagtive, and provides a mean recurrence warnings 5.2 months earlier than imaging modalities. Key applications of CAPP-Seq include early diagnosis (detecting mutations with allele frequencies as low as 0.02% to improve early-stage detection), treatment monitoring (dynamic ctDNA quantification enabling timely intervention), prognosis assessment (postoperative ctDNA positivity identifying high-risk patients requiring intensive therapy), and resistance management (detecting resistance mutations to optimize targeted therapies)⁵⁷.

The core trade-off between sensitivity (favoring tumor-informed methods) and accessibility (favoring tumor-naive methods) emphasizes the need for context-specific implementation. Standardization gaps persist, and the consensus recommendation is ultra-deep sequencing (> 30,000× coverage for agnostic panels) to mitigate technical noise^35,58. Emerging solutions include integrating multi-modal biomarkers (e.g., methylation or fragmentomics)⁵⁵ and leveraging machine learning algorithms to enhance specificity⁵⁹. Tumor-informed and tumor-naive MRD detection are complementary paradigms⁶⁰. Some Chinese companies have integrated tumor-informed and tumor-agnostic MRD strategies to balance sensitivity and specificity, and have used fusion mutation monitoring to further optimize the detection of treatment-driven alterations. As multi-omic refinements mature and prospective data accumulate, these technologies are anticipated to play decisive roles in guiding therapy de-escalation, recurrence surveillance, and drug development across oncology.

Positivity criteria vary across various MRD detection technologies. Signatera^® (Natera), a tumor-informed multiplex PCR assay, customizes a personalization panel of as many as 16 somatic mutations, with a median of 11 (range 8–16), derived from WES of tumor tissue. MRD positivity is defined as the detection of ≥ 2 tumor-specific mutations, with a LoD95 of 17 ppm^61,62. In contrast, the Signatera Genome assay uses WGS to identify and monitor as many as 64 patient-specific somatic mutations from tumor tissue, a number representing an expansion beyond its original 16-mutation panel. The assay uses mPCR with a sensitivity of low single-digit ppm levels and maintains the same validated chemistry as the original Signatera platform. With customized primers, serial blood-based ctDNA detection achieves MRD positivity determination via a proprietary sample-calling algorithm that compares target confidences against sample-level noise and demonstrates a LoD95 of 5–9 ppm⁵¹.

PCM (Invitae) uses a similar tumor-informed multiplex PCR approach targeting a maximum of 50 variants and a median of 33 personalized mutations (range: 11–50). Positivity is defined with a Poisson distribution-based statistical model (P < 0.001) to distinguish tumor-derived signals from background noise^63,64. Likewise, RaDaR™ (Inivata) customizes as many as 48 tumor-specific variants (median 48) and requires either an estimated VAF > 0.001% or detection of ≥ 2 tumor-specific mutations to indicate MRD positivity^65,66. ArcherDx’s AMP assay, used in the TRACERx cohort, simultaneously monitors 150 clonal and subclonal mutations (median 197 variants) and applies a Poisson model (P < 0.01) to differentiate tumor signals from noise⁶⁷. CAPP-Seq uses a fixed panel design covering 95% of NSCLC driver genes, coupled with integrated digital error suppression (iDES) technology to minimize sequencing background noise. The iDES system, by suppressing average noise levels from 0.03% to 0.0015%, enables reliable detection of true low-frequency mutations. The presence or absence of ctDNA has been assessed with a Monte Carlo-derived detection algorithm and a threshold value of ≤ 0.05 (validated in previous research) as the criterion for positivity. Specimens exhibiting values exceeding this threshold were designated as ctDNA-negative, whereas those meeting or falling below the cutoff were classified as ctDNA-positive³⁵.

Novel technologies include MAESTRO (Broad Institute), which enriches thousands of low-frequency mutations (< 0.1% VAF) for 10-fold signal-to-noise enhancement. Because MRD positivity is defined by the detection of > 1 tumor-specific mutation, false negatives are minimized while reasonable specificity is maintained. To further enhance accuracy, use of the duplex sequencing consensus (DSC)/single-strand consensus (SSC) ratio as a noise-filtering metric ensures specificity in chimeric mutation detection. This ratio represents the proportion of DSC reads supporting a mutation relative to the SSC reads. Empirical evidence has demonstrated that at ratios below 0.15, non-mutant libraries display poor reproducibility during replicate capture, thereby indicating high background noise. A DSC/SSC ratio threshold of ≥ 0.15 is applied to exclude low-confidence variants^68,69.

The PhasED method implements personalized hybrid capture panels targeting PVs to analyze plasma-derived cell-free DNA (cfDNA) and achieves high unique molecular depth through ultra-deep sequencing. The tumor fraction is calculated as the ratio of sequencing reads containing predefined PVs to the total reads covering PV genomic positions. A sample is classified as ctDNA-positive when at least one predefined PV is detected above the established threshold, and all calls require statistical significance (P < 0.01) after background noise correction⁴⁸.

The tumor-informed NeXT Personal MRD platform uses WGS to achieve ultra-sensitive detection at single-molecule resolution. The ctDNA burden is quantified by integration of variant allele signals across all patient-specific genomic targets after rigorous noise filtration. Sample-level positivity is established through a stringent statistical framework in which tumor-derived molecular counts are aggregated to surpass background noise expectations with Poisson-distributed significance (P ≤ 0.001)⁴⁵.

The determination of optimal MRD positivity thresholds requires careful consideration of the specific MRD monitoring strategy, the number and characteristics of personalized genomic variant targets, and the assay’s background noise suppression ability (Table 1). When properly balanced, these parameters enable enhanced detection of MRD-positive cases without compromising test sensitivity or specificity, thereby improving the identification of patients at high recurrence risk. This optimized approach provides clinicians with substantially longer lead times and consequently facilitates timely therapeutic intervention. However, indiscriminate lowering of positivity thresholds to maximize detection rates might increase the false-positive results, and expose patients to unnecessary psychological distress and anxiety without conferring clinical benefit^70,71.

The sensitivity, specificity, cost, and clinical utility of various MRD monitoring strategies significantly vary across technologies and applications (Table 2). Tumor-informed personalized panels based on WGS have demonstrated the highest sensitivity and specificity. However, this approach is costly because of the high-depth sequencing, and the need for tumor tissue for baseline WGS profiling limits its accessibility. Tumor-agnostic methods offer moderate sensitivity and lower specificity than personalized panels, but are more affordable and widely available. Cost-effectiveness analyses have suggested that NGS might offset expenses through treatment de-escalation in MRD-negative cases, although the upfront costs and tissue dependency remain barriers. Overall, sensitivity and clinical utility are maximized by tumor-informed WGS-based or WES-based methods, whereas tumor agnostic fixed panels balance affordability and broad applicability.

ctDNA-based MRD detection has emerged as a transformative tool in early-stage BC management, by enabling dynamic risk stratification and personalized therapeutic interventions⁷² (Figure 3). Unlike traditional imaging, ctDNA analysis provides molecular-level insights into residual tumor burden. Postoperative ctDNA positivity has been demonstrated to confer a 35.8-fold higher recurrence risk and significantly shorter disease-free survival (DFS) than that with ctDNA-negative^73,74. This prognostic power is subtype-dependent: ctDNA positivity rates correlate with biological aggressiveness, as observed in 28.6% of HR+ tumors, 18.8% of TNBC, and 5.0% of HER2+ tumors⁷³. In a study by Shaw et al.,⁶² Signatera™ testing every 6 months in high-risk patients achieved a sensitivity and specificity of 88% and 95%, and a median lead time exceeding 10 months. In another study, RaDaR™ monitoring every 6–12 months has shown 100% sensitivity for metastatic recurrence (median lead time 12.4 months), although one localized recurrence was missed (overall sensitivity 85.7%)⁶². In contrast, Guardant Reveal™ detected 85% of metastatic relapses but only 14% of local recurrences in early BC, and single-timepoint testing at approximately 2 years post-chemotherapy identified 34% of distant recurrences. Guardant Infinity™ methylation analysis of 83 patients with TNBC/luminal BC further corroborated a strong association between postoperative ctDNA positivity and recurrence. The PCM study confirmed that ctDNA detection preceded imaging by 11.7 months and achieved a specificity of 100%, thus highlighting its potential for early intervention⁶⁴. Technologies such as WGS-based NeXT Personal MRD can achieve 98% sensitivity at diagnosis and track ultra-low ctDNA levels as low as 2.19 ppm⁴⁵.

The detection of MRD in the adjuvant setting can aid in identifying patients at high recurrence risk who might benefit from more aggressive adjuvant therapies, additional chemotherapy, targeted therapy, or prolonged treatment duration. In contrast, patients with undetectable MRD might be considered for treatment de-escalation, to decrease the risk of unnecessary toxicity. Moreover, longitudinal MRD monitoring can provide valuable insights into the effectiveness of adjuvant treatment. Undetectable MRD or achieving ctDNA clearance might indicate a favorable treatment response, whereas the emergence or persistence of MRD might suggest therapeutic resistance and poor prognosis.

Personalized multiple-mutation sequencing (e.g., Signatera or RaDaR), compared with ddPCR, significantly extends lead times and improves sensitivity. The MRD positive ratio has been found to be significantly higher in the TNBC population than the HR+ population, regardless of post-surgery time^62,73. In TNBC, combining baseline ctDNA thresholds with postoperative dynamic monitoring optimizes risk stratification: longitudinal MRD monitoring has achieved 88%–100% sensitivity and 100% specificity for distant metastasis, and a median lead time of 8.9 months^73,75. For HR+ BC in the late adjuvant phase, the MRD detection rate was 10% with a median lead time of 12.4 months, and all distant metastatic recurrences were preceded by MRD detection⁷⁶.

The MonarchE study has evaluated MRD status via Signatera ctDNA testing in patients with HR+/HER2− high-risk early BC after adjuvant therapy⁷². The results revealed a strong association between ctDNA positivity and relapse risk: in a subgroup of 910 patients, baseline ctDNA-positive patients had an 80% relapse rate (vs. 23% in the negative group), whereas ctDNA detection at any time during treatment was predictive of poor prognosis. Among persistently ctDNA-positive patients, 87% experienced invasive DFS events, as compared with only 15% of persistently ctDNA-negative patients⁷⁷. Interestingly, patients with ctDNA conversion from negative to positive during treatment experienced a high relapse risk of 93%, and ctDNA positivity preceded radiographic relapse by 9–19 months, thus providing a critical window for therapeutic intervention.

The PADA-1 trial pioneered ctDNA ESR1 mutation-guided therapy switching in HR+ metastatic disease. Patients switching to fulvestrant + palbociclib after ESR1 mutation detection had a median PFS of 11.9 months, as compared with 5.7 months in the control group⁷⁸. The SERENA-6 trial pioneered ctDNA-guided dynamic monitoring of ESR1 mutations to direct therapy switching to oral SERD camizestrant while maintaining the original CDK4/6 inhibitor before radiographic progression. The median progression-free survival was significantly extended (16.0 vs. 9.2 months; HR = 0.44), and the safety profiles were manageable^17,42. Similar strategies are being evaluated in early-stage disease. For example, the c-TRAK-TN trial has leveraged ctDNA to guide adjuvant therapy intensification. This prospective phase II study evaluated ctDNA-based MRD monitoring to guide immunotherapy intensification with pembrolizumab (PD-1 inhibitor) in early-stage TNBC⁷⁹. The study enrolled 208 patients with high-risk TNBC (non-pCR after neoadjuvant therapy or tumors > 2 cm with nodal positivity) and used personalized digital PCR for ctDNA surveillance every 3–12 months. The 12-month ctDNA positivity rate was 27.3% (44/161), and the baseline rate was 14.3% (23/161). ctDNA positivity strongly predicted relapse: 23 of 32 ctDNA-positive patients (71.9%) had concurrent clinical recurrence at detection, thus confirming MRD as a robust prognostic biomarker. None of the 5 ctDNA-positive patients without recurrence receiving pembrolizumab (200 mg IV every 3 weeks for 1 year) achieved ctDNA clearance, and all eventually relapsed. These findings indicated the limited efficacy of pembrolizumab monotherapy in these patients. Irreversible micrometastases were likely to have been present in most ctDNA-positive cases (23/32), thereby narrowing the intervention window. MRD assay sensitivity requires improvement (current specificity 96%–100%, sensitivity 19%–93%), and the monitoring frequency should be optimized (e.g., every 3 months) to enable earlier intervention⁷⁹. Although c-TRAK-TN prospectively validated ctDNA-MRD as a reliable tool for identifying patients with high-risk TNBC, pembrolizumab monotherapy did not improve the outcomes. Therefore, earlier intervention or combination therapies must be investigated in future studies.

In the neoadjuvant setting, dynamic MRD assessment serves as a real-time biomarker of therapeutic response. Baseline ctDNA detection can further identify patients with inferior DFS and OS, and therefore has a role in upfront risk stratification⁸⁰. Among patients with early BC, the prevalence of pre-neoadjuvant therapy (pre-NAT) ctDNA positivity is higher in hormone receptor-negative (HR-negative) than HR-positive cancers. Overall, 47.7%–78.7% of patients exhibited MRD positivity^{32,64,81–86} (Table 3). Notably, within the TNBC subgroup, ctDNA positivity rates ranged from 63.4% to 91%^80,87,88. WGS-based MRD detection demonstrated a positivity rate as high as 98% at diagnosis, with a median lead time of 15 months, and no recurrence was observed in patients with undetectable ctDNA⁴⁵. MRD detection with a tumor-agnostic approach yielded a relatively lower positivity rate but was associated with a shorter lead time^53,59,89. Persistent ctDNA detection mid-therapy correlates with residual cancer burden (RCB) and poor outcomes, whereas ctDNA clearance predicts pCR and improved survival, particularly in HER2-negative disease^75,82. Therapeutic decision-making is increasingly informed by ctDNA kinetics. During neoadjuvant therapy, the I-SPY2 trial used Signatera ctDNA testing to dynamically monitor 283 patients with high-risk early-stage HER2-negative BC (HR+/HER2−: 145; TNBC: 138). The ctDNA positivity at baseline, and during the treatment (3/12 weeks) and pre-surgery (T3) timepoints, significantly correlated with shorter distant relapse-free survival (DRFS) (P = 0.02 to P < 0.0001)⁸⁵. Patients with ctDNA-negative status at T3 exhibited markedly improved DRFS, particularly among non-pCR patients. The HR+/HER2- group exhibited a 5.89-fold higher relapse risk in patients with ctDNA-positive than those with ctDNA-negative (95% CI: 2.68-12.98), while the TNBC group displayed a 3.79-fold higher risk in patients with ctDNA-positive than those with ctDNA-negative (95% CI: 1.87-7.68). In patients with TNBC, ctDNA clearance at 3 weeks (T1) was strongly predictive of pCR (OR = 13.06; 95% CI: 3.54–57.95) and low residual cancer burden (RCB-0/I) (OR = 19.00; 95% CI: 4.98–89.06). In contrast, this association was absent in HR+/HER2− subtypes⁹⁰. Of note, ctDNA status can refine risk stratification beyond pCR/RCB. One patient with TNBC who achieved pCR but remained ctDNA-positive relapsed at 21 months, whereas ctDNA-negative non-pCR patients experienced favorable outcomes, thus supporting ctDNA integration into post-treatment evaluation⁸⁷. In a study by Elliott et al.^44,49 (n = 119 patients with early BC receiving neoadjuvant therapy), tumor-informed RaDaR™ monitoring of 48 mutations demonstrated that failure to clear ctDNA after neoadjuvant therapy predicted inferior outcomes, particularly in HER2-negative patients (HR = 14.4). ctDNA dynamics strongly correlated with pathological response, and patients achieving pCR (RCB-0) did not experience clinical recurrence, regardless of mid-treatment ctDNA status. Among patients with suboptimal response (RCB 1/2/3), those with ctDNA clearance during therapy showed significantly superior RFS to those with non-clearance (P = 0.011)⁸⁷. Postoperative or longitudinal ctDNA detection demonstrated even stronger prognostic relevance and was associated with significantly shorter RFS than that with ctDNA undetection (P < 0.0001)⁸⁵. Early ctDNA clearance in TNBC might therefore be used to guide de-escalation, whereas post-surgery positivity warrants intensified therapy.

The relationship between ctDNA positivity rates and biological aggressiveness is particularly evident in the comparison between TNBC and HER2+ subtypes. TNBC demonstrates higher ctDNA positivity rates pre-, intra-, and post-NAC than HR+/HER2− disease, given its aggressive biology, including rapid proliferation and a propensity for early metastasis. In contrast, HER2+ tumors (with a positivity rate of ~5% in some cohorts) potentially show lower ctDNA shedding because of targeted therapy efficacy or different biological drivers, although this finding does not diminish their potential aggressiveness. Technologies such as tumor-informed PCR or NGS assays (e.g., Signatera) are optimal for TNBC, because of their high sensitivity and ability to track rapid dynamics, whereas HR+ tumors, with their lower shedding rates, might benefit from broader panel-based NGS to capture heterogeneous mutations over time. Practical recommendations should also consider clinical context: for NAC-treated high-risk patients, we advise baseline, mid-treatment, pre-surgery, and post-surgery time points, with subtype-guided adjustments; e.g., TNBC might require additional testing every 3–6 months for the first 3 years. Ultimately, the testing frequency should balance prognostic value against patient burden. We propose standardized windows based on risk (e.g., high-risk TNBC and HER2+: 0, 3, 6, 12, 18, or 24 months; HR+: 0, 6, 12, 18, or 24 months annually thereafter). Future trials should validate these intervals prospectively.

Emerging interventional clinical trials investigating the clinical efficacy of MRD in BC adjuvant and neoadjuvant settings are also underway. The TRAK-ER, DARE, and LEADER trials were aimed at determining the clinical utility of MRD-directed therapies and assisting in the development of more effective treatment strategies to eliminate MRD in HR+ BC^8,91. The TRAK-ER trial (NCT04985266) dynamically monitors ctDNA every 3 months in patients with intermediate-to-high-risk HR+ cancers post-neoadjuvant therapy. Those with ctDNA positivity are randomized to either standard endocrine therapy or an intensified regimen, and an early intervention window (11.7 months ahead of imaging) is being explored⁹². The DARE trial assessed intensified therapy with CDK4/6 inhibitors plus endocrine therapy for ctDNA-positive high-risk HR+/HER2− postoperative patients and validated targeted escalation to delay recurrence in the MRD-positive high-risk population⁹³. The TREAT ctDNA trial (NCT05512364)⁹¹ randomized patients with ER+/HER2− stage II–III cancers with MRD positivity to standard endocrine therapy or elacestrant (a novel oral SERD)³⁰. Using Signatera tumor-informed assays (30% higher sensitivity), elacestrant targets ESR1 mutations in high risk patients with MRD-positivity, with distant metastasis-free survival as the primary endpoint. The LEADER part trial adapts the PADA-1 model by switching to elacestrant plus CDK4/6 inhibitors upon detection of MRD positivity in patients with early HR+ high-risk cancers, to optimize sequential adjuvant therapy^8,94,95. Regarding PARPi-targeted therapy, the ZEST trial evaluates the efficacy and safety of niraparib vs. placebo in 800 patients with HR+/HER2− germline BRCA mutation or high-risk TNBC, and uses the Signatera assay to detect MRD and evaluate the optimal frequency of ctDNA monitoring, independently of relapse timing^30,96 (Table 4).

The pathophysiological mechanisms of MRD in BC are complex and elusive. Tumor cells that persist after treatment might have unique metabolic and regulatory characteristics. Elucidating these pathophysiological mechanisms is crucial for developing effective strategies to target and eliminate MRD^97,98 (Figure 4). Exploratory analysis of pre-treatment gene expression profiles has revealed both shared and subtype-specific patterns potentially associated with ctDNA shedding⁶¹. Preoperative ctDNA-positive patients (also referred to as “shedders”) exhibit a distinct profile characterized by 3 defining features: high invasiveness, manifesting as larger tumor volume, predominance of solid nodules, and high histologic grade; high metabolic activity, evidenced by significantly elevated PET-CT SUVmax values, correlating with aggressive tumor proliferation and high Ki67 index⁹⁹; and high genomic instability, which positively correlates with ctDNA shedding driven by chromosomal abnormalities, and upregulated proliferation and cell cycle/DNA repair pathway signaling^61,67. Tumor heterogeneity can lead to false-negative MRD results if the resistant cells responsible for recurrence originate from a sub-clone not captured by the original biopsy, because of sampling bias, or from a sub-clone that acquired new mutations after treatment. Although the power to detect differentially expressed genes in TNBC was limited by a sample size imbalance between the ctDNA-positive and ctDNA-negative groups, protein-protein interaction enrichment analysis of common differentially expressed genes in both the HR+/HER2− and TNBC subtypes identified cell cycle genes associated with ctDNA shedding, thus suggesting that tumor cell proliferation plays a key role in ctDNA release⁸⁵. In the HR+/HER2− subtype, the association between cell cycle-associated gene sets and ctDNA shedding remained statistically significant even after adjustment for MammaPrint status¹⁰⁰. Transcriptomic analysis of urothelial tumors from ctDNA-positive patients further revealed upregulation of cell cycle genes and hypoxia-related pathways. Additionally, gene expression profiling implicated specific immune response pathways in ctDNA release, in agreement with preclinical evidence indicating that immune cells might facilitate DNA shedding into the circulation. The tumor microenvironment, including interactions with immunosuppressive cells, stromal cells, the extracellular matrix, and elevated vascular permeability, can influence the fate of ctDNA release¹⁰¹.

Genomically, landmark MRD profiling is predictive of poor prognosis and a potential need for intensive anti-cancer therapy. Tumor-informed sequencing has identified actionable targets (e.g., PIK3CA, ESR1, and DNA repair defects) in 77.8% of post-operative MRD-positive patients with therapies matched via databases such as OncoKB. Patients with detectable postoperative MRD exhibited greater enrichment in somatic mutations and amplifications of genes such as MAP3K1, ATM, FLT1, GNAS, POLD1, SPEN, and WWP2 in tumor tissue than those with undetectable MRD^102,103. GSEA of RNA sequencing data from tumors in MRD-positive patients indicated elevated frequencies of cell cycle-related pathway mutations, DNA repair disorder, and AID-mediated somatic hypermutation signatures, thus implying potential mechanistic links to MRD persistence in BC^102,104. Non-Tie I/II mutations (variants of unknown clinical signature) or subclonal mutations from the tumor tissue were associated with a significantly higher recurrence risk than observed in MRD-negative patients. These findings highlight the prognostic value of monitoring these traditionally overlooked non-driver mutations during MRD surveillance⁴⁶.

Standardizing MRD detection and interpretation of the findings remains a major challenge. A universal consensus regarding the optimal detection methods is lacking, including the selection between tumor-informed and tumor-agnostic approaches, as well as the specific assays within each approach. This lack of standardization poses challenges in comparing results across studies and translating research findings into clinical practice. Heterogeneity in MRD methods, encompassing blood sample collection, storage, transport, sequencing depth, variant-calling algorithms, and positivity thresholds, compromises the reproducibility of results across laboratories. Discrepancies in detection techniques (e.g., PCR vs. NGS) further exacerbate inter-laboratory variability and can undermine clinical decision-making. Tumor-agnostic panels targeting recurrent driver genes (e.g., TP53, PIK3CA, and ESR1) are particularly vulnerable, and the reported false-positive rates have exceeded 15% in some cohorts⁴². The integration of fragmentomics, methylation signatures, and CHIP mutation databases has been used to filter background noise^72,98. To address the potential interference of CHIP with MRD detection, the investigators explored its underlying mechanisms and proposed corresponding solutions. The study revealed that the infiltration of CHIP-associated mutations (e.g., in DNMT3A, TET2, ASXL1) into tumors, a phenomenon known as TI-CH, is not only associated with an unfavorable prognosis but may also interfere with ultra-sensitive ctDNA testing by contributing to background noise. To address this, the researchers employed tumor-normal paired sequencing. This approach allowed them to rigorously distinguish true somatic tumor mutations from CHIP-derived mutations by using the normal tissue sample as a genetic baseline. Furthermore, they applied tumor purity correction to exclude low-frequency CHIP signals and integrated filtering based on known CHIP-associated gene databases (e.g., COSMIC CHIP tags) for accurate mutation annotation¹⁰⁵. Finally, ethical considerations in MRD research and treatment cannot be overlooked. The high cost of MRD detection, particularly tumor-informed approaches requiring WES or WGS of primary tumors, remains a major impediment to widespread use¹⁰⁶. Economic evaluation of MRD monitoring has revealed a paradigm shift in cost-effectiveness by enabling risk-stratified treatment decisions, particularly through decreasing overtreatment in low-risk populations while maintaining clinical outcomes. Key studies such as the DYNAMIC trial in colorectal cancer have demonstrated that MRD-guided strategies can significantly decrease unnecessary chemotherapy (from 28% under conventional surveillance to 15% in MRD-positive patients) without compromising 5-year RFS¹⁰⁷. This precision approach not only lowers direct treatment costs (e.g., chemotherapy agents, administration, and hospitalization) but also mitigates indirect costs associated with treatment-related adverse events (e.g., toxicity management and productivity loss). However, the cost-benefit balance hinges on MRD testing expenses, which vary by method: tumor-informed assays have higher initial costs because of WES and personalized panel designs, whereas tumor-agnostic assays eliminate tissue biopsy costs but might sacrifice sensitivity in low-VAF situations. The design of MRD intervention trials must incorporate strategies to address participants’ comprehension of complex diagnostic information, particularly by validating the informed consent process. Patients must be explicitly provided with a clear understanding of both the technical capabilities and inherent limitations associated with advanced MRD detection technologies. Such clarity is critical, given the evolving clinical implications of MRD status to guide therapy escalation/de-escalation. Large-scale clinical trials are warranted to validate the clinical utility of MRD-directed therapies and to establish standardized treatment algorithms. By following this strategic management, MRD can be effectively integrated into BC management, thereby leading to improved patient outcomes.

As liquid biopsy technologies evolve, MRD-guided strategies have potential to transform the paradigm of BC management from empirical approaches to dynamic, biomarker-driven precision oncology. Despite its transformative potential, MRD integration in BC faces challenges in cost, standardization, and biological specificity. Tumor-informed and agnostic strategies each offer distinct advantages. Future adaptive trials based on MRD are highly anticipated to serve as a cornerstone for precision oncology.