The role of serologic biomarkers “liquid biopsy” in the fight against breast cancer: From biological foundations to clinical impact

Summary

Liquid biopsy, analyzing circulating tumor-derived components, including ctDNA, circulating tumor cells, microRNAs, and exosomes, has emerged as a key tool in precision oncology. This review synthesizes evidence from the past five years, covering biological mechanisms, analytical platforms, clinical applications, and implementation barriers. Evidence from multiple clinical studies suggests ctDNA shows higher sensitivity for metastatic disease monitoring than conventional biomarkers such as CA15-3 and circulating tumor cells. For minimal residual disease detection, tumor-informed assays demonstrate sensitivity of 88%–100% and provide a median lead time of 8–15 months before clinical relapse. Despite these advances, early-stage cancer detection remains suboptimal, cost-effectiveness evidence is limited, and disparities in access are increasing. We critically evaluate current technologies and translational gaps and highlight the need for standardized clinical integration pathways. Overall, liquid biopsy is transitioning from a complementary diagnostic tool toward a longitudinal framework for cancer management, although substantial methodological and systemic challenges remain.

Graphical abstract

Oncology; Diagnostics; Therapeutics

Introduction

Breast cancer remains a major global health challenge. In the United States alone, more than 316,000 new invasive cases and over 42,000 deaths are projected in 2025.¹ Incidence continues to rise worldwide, including a concerning increase among younger women, highlighting the limitations of age-based screening strategies. Despite substantial advances in surgery, radiotherapy, systemic therapy, and early detection, three fundamental limitations in current diagnostic paradigms remain unresolved.² First, tissue biopsy suffers from inherent spatiotemporal sampling bias. A single lesion sampled at a single time point cannot capture the genomic heterogeneity of multifocal primary tumors or metastatic disease. Clinically actionable mutations may be missed if they arise in unsampled regions.³ Second, early detection remains inadequate. Mammography, while widely adopted, performs poorly in dense breasts, generates high false-positive rates, and depends on infrastructure that is often inaccessible in low-resource settings.⁴ Third, conventional surveillance detects recurrence only after substantial tumor burden accumulates. Radiographic imaging typically identifies lesions once they contain billions of cancer cells, long after micrometastatic disease has already developed.⁵

Since the first observation of circulating cell-free DNA (cfDNA) in 1948 and the identification of tumor-specific mutations in blood during the 1990s, liquid biopsy has gradually evolved from a conceptual supplement to tissue biopsy into a versatile clinical tool.⁶ In contrast to the static and spatially restricted nature of tissue sampling, repeated liquid biopsy testing enables a dynamic assessment of disease, allowing investigators to capture molecular material shed from multiple tumor sites, to follow temporal clonal evolution under therapeutic pressure, and to detect recurrence months before radiographic progression, all through a minimally invasive approach.⁷ However, the biological release of tumor-derived material into circulation is highly heterogeneous. Variations in cellular proliferation, apoptosis and necrosis, vascular accessibility, tumor burden, and microenvironmental constraints all influence circulating tumor DNA (ctDNA) shedding.⁸ These biological differences shape assay design and determine platform performance, underscoring the need for context-specific interpretation rather than a one-size-fits-all approach.⁹

We searched PubMed and Web of Science for articles published between 2015 and 2025 using the keywords “liquid biopsy,” “breast cancer,” “ctDNA,” “CTCs,” “miRNA,” and “MRD,” with priority given to prospective clinical trials, meta-analyses, and regulatory-approved assays. This review links the mechanisms of biomarker release with technical determinants of assay sensitivity and evaluating both successful applications and existing limitations. It also highlights real-world implementation challenges. This review emphasizes that mechanistic understanding, technological development, and integration into clinical workflows must converge to enable the effective and equitable application of liquid biopsy in breast cancer management.

Biological mechanisms and technical platforms: Understanding performance determinants

ctDNA and other analytes enter circulation through apoptosis, necrosis, cellular turnover, and potentially active secretion; however, the efficiencies of these processes vary widely among tumor types, and probably among different subtypes of the same cancer. It is plausible that highly proliferative tumors—such as triple-negative or HER2-positive breast cancers—might shed more circulating DNA because of increased cell turnover, whereas hormone receptor-positive tumors may shed less, posing greater challenges for early detection or minimal residual disease monitoring.¹⁰ Tumor vascularization, stromal composition, metastatic burden, and the physical proximity of tumor cells to the circulatory system further affect whether shed material reaches peripheral blood.¹¹ Biological factors on the host side—such as hepatic and renal clearance rates or age-related clonal hematopoiesis—can obscure true tumor signal and complicate assay interpretation. Together, these variables form a biologically complex landscape in which analytical sensitivity is at once critically important and inherently limited by the amount of tumor-derived material that enters the bloodstream¹² (Figure 1).

Figure 1.

ctDNA is released via tumor cell apoptosis, necrosis, and proliferation, and can be sensitively detected by dPCR and NGS

CTCs are captured by CellSearch and single-cell technologies for prognosis and drug sensitivity testing. Exosomes and miRNAs are promising for early detection and risk stratification but lack standardized isolation methods.

Tumor tissue samples and liquid biopsy represent two complementary approaches for assessing genomic alterations in cancer. Tissue-based genomic profiling primarily relies on next-generation sequencing (NGS), whole-exome sequencing (WES), and whole-genome sequencing (WGS) technologies,¹³ enabling comprehensive detection of somatic mutations, copy-number variations, and structural rearrangements. Fluorescence in situ hybridization (FISH) and immunohistochemistry (IHC) are widely applied in clinical practice for detecting gene amplification and validating protein expression. In contrast, liquid biopsy approaches include droplet digital PCR (ddPCR), methylation profiling, fragmentomics, and circulating tumor cell (CTC)-based assays, which enable minimally invasive and real-time monitoring of tumor genomic alterations.¹⁴ Tissue biopsy generally provides deeper genomic profiling, whereas liquid biopsy enables real-time assessment of tumor evolution; therefore, the two approaches are complementary. In liquid biopsy analyses, Digital PCR represents one of the earliest and most robust approaches, achieving extremely high analytical sensitivity by partitioning DNA into thousands of micro-reactions.¹⁵ Next-generation sequencing (NGS) platforms expanded the field substantially by enabling simultaneous interrogation of dozens to hundreds of genes.¹⁶ Panel-based approaches offer broad genomic coverage that supports treatment decision-making, yet their sensitivity remains constrained in low-shedding tumors unless sequencing depth is dramatically increased. As an evolution of this paradigm, tumor-informed personalized assays directly address the challenge of detecting extremely low ctDNA levels by identifying patient-specific mutations through initial tumor sequencing and subsequently tracking dozens to thousands of variants in plasma.¹⁷

ctDNA detection: Technical evolution and performance benchmarking

ctDNA has become the most extensively investigated analyte in liquid biopsy, largely because its quantitative and qualitative features directly reflect tumor burden, genomic evolution, and therapeutic response. The evolution of ctDNA detection technologies illustrates how the field has progressively pushed analytical sensitivity to overcome the inherently low abundance of circulating DNA fragments, particularly in early-stage and hormone receptor-positive breast cancers. Digital PCR laid the foundation by partitioning DNA molecules into thousands of micro-reactions, enabling highly precise quantification of known variants with sensitivities reaching 0.001% variant allele frequency. Its speed and low cost make it ideally suited for monitoring established hotspots, although its design inherently limits analysis to predefined mutations^18,19 (Figure 1).

The advent of NGS expanded ctDNA profiling by enabling simultaneous interrogation of broader genomic regions at high depth.²⁰ Improvements such as molecular barcoding and sophisticated error-correction algorithms have significantly reduced background noise and allowed reliable detection of extremely low-frequency variants. These panel-based approaches strike a balance between genomic breadth and analytical depth and have become central to treatment selection and resistance monitoring.²¹ However, their sensitivity still decreases when ctDNA concentration falls below the detection threshold, a challenge frequently encountered in early breast cancer.¹⁶ Tumor-informed assays represent a conceptual leap by tailoring ctDNA detection to the unique mutational landscape of each patient. Instead of screening for a limited set of recurrent mutations, these personalized assays track dozens to thousands of tumor-specific variants simultaneously, dramatically increasing the probability of detecting even minute amounts of residual disease.²² Platforms capable of monitoring extensive variant repertoires routinely achieve sensitivities as low as 1–3 ppm and identify recurrence months before radiographic evidence becomes apparent. The increased sensitivity, however, comes with higher cost, longer turnaround times, and the requirement for initial tumor tissue sequencing, which may not always be available.²³ As the field progresses, multi-signal approaches that incorporate fragmentation patterns, nucleosome positioning, and methylation signatures are being integrated into ctDNA analysis to capture tumor-associated changes that extend beyond point mutations.²⁴ Early studies suggest that these orthogonal signals can complement mutation-based detection, especially in cases where the abundance of ctDNA is exceedingly low. Collectively, the trajectory of ctDNA technology reflects ongoing efforts to reconcile the limitations imposed by tumor biology with the need for clinically meaningful sensitivity across a broad spectrum of disease settings.²⁵

CTCs: From enumeration to functional characterization

CTCs represent another clinically relevant component of liquid biopsy and provide a fundamentally different layer of biological insight compared with ctDNA. Unlike fragmented DNA, CTCs are intact, viable cells that retain genomic, transcriptomic, and phenotypic features of the primary tumor, making them valuable for understanding metastatic dissemination and therapeutic vulnerability.²⁶ The CellSearch system established the first standardized and FDA-cleared platform for CTC enumeration, using epithelial markers to isolate tumor cells from peripheral blood and demonstrating strong prognostic significance in metastatic breast cancer. Patients with elevated baseline CTC counts consistently exhibit worse outcomes, reinforcing the role of CTC burden as an indicator of tumor aggressiveness and metastatic potential.²⁷

Despite their prognostic value, CTC detection is inherently limited by their rarity, often fewer than a handful per several milliliters of blood—which restricts sensitivity, particularly in early-stage disease where ctDNA typically offers better detectability. Advances in microfluidics, immunomagnetic enrichment, and high-volume apheresis-based collection systems have sought to address this challenge by processing significantly larger blood volumes and thus capturing more cells. These approaches have enabled isolation of CTCs even from early breast cancer patients, suggesting that technical innovations may gradually close the sensitivity gap. Perhaps the most transformative aspect of CTC research lies in the increasing ability to characterize these cells functionally.^28,29 Single-cell sequencing has revealed dynamic and heterogeneous molecular phenotypes within CTC populations, reflecting clonal evolution and treatment-induced shifts. Some studies have successfully cultured CTCs ex vivo, creating patient-specific model systems for drug testing and personalized therapy selection.³⁰ Although these functional applications remain technically demanding and far from routine clinical use, they highlight the unique potential of CTCs to illuminate metastatic biology and to serve as living biomarkers capable of guiding individualized treatment strategies.

MicroRNAs and exosomes: Underexplored territory

Beyond ctDNA and CTCs, microRNAs and exosomes constitute two additional analyte classes that expand the molecular breadth of liquid biopsy.³¹ Circulating microRNAs are stable, abundant, and often dysregulated in cancer, making them attractive candidates for early detection and disease classification. Numerous studies have demonstrated their diagnostic potential in breast cancer, with some microRNAs—particularly miR-21—repeatedly highlighted as promising biomarkers.^32,33,34 However, the field has been slowed by substantial methodological heterogeneity. Differences in sample processing, RNA extraction, platform selection, and normalization strategies have produced inconsistent results across studies, and systematic reviews have identified considerable risk of bias. Until standardized protocols are established, the clinical translation of microRNA-based assays will likely remain limited^35,36 (Figure 1).

Exosomes, in contrast, offer a more complex representation of tumor biology. These nanoscale vesicles carry DNA, RNA, proteins, lipids, and metabolites, reflecting the cellular state of their tumor of origin and facilitating intercellular communication.^37,38 Their abundance and stability provide a rich reservoir of molecular information, and early studies suggest that exosomal RNA and protein signatures may outperform some traditional serum biomarkers.³⁹ Yet, like microRNAs, the lack of consensus on optimal isolation and characterization methods has slowed their clinical adoption.⁴⁰ Current approaches range from ultracentrifugation to precipitation reagents and microfluidic platforms, each with distinct biases and limitations, making comparisons across studies challenging.⁴¹

Despite these barriers, microRNAs and exosomes represent important avenues for broadening liquid biopsy capabilities. Their complementary nature—capturing regulatory signals, secretory processes, and intercellular communication—offers insights that ctDNA alone cannot provide.⁴²

Clinical applications: Evidence hierarchy and performance metrics

Clinical applications in metastatic breast cancer

In metastatic breast cancer, liquid biopsy has emerged as one of the areas with relatively robust clinical evidence and increasingly mature applications. Compared with early-stage disease, patients with metastatic tumors typically exhibit higher tumor burden and more consistent ctDNA shedding, enabling ctDNA-based assays to demonstrate greater sensitivity and reproducibility for disease monitoring and molecular subtyping.¹⁶ It is important to emphasize the conceptual distinction between prognostic value and clinical utility: the former refers to a biomarker’s ability to predict disease outcomes, whereas the latter focuses on whether it can guide therapeutic decisions and ultimately improve patient prognosis.⁴³ In the context of metastatic disease, ctDNA analysis has begun to demonstrate tangible clinical decision-making value. For example, hotspot mutations in the ligand-binding domain (LBD) of the ESR1 gene, such as Y537S and D538G, are recognized as key molecular mechanisms driving acquired resistance to aromatase inhibitors (AIs) in ER-positive metastatic breast cancer. Studies indicate that ESR1 mutations predominantly occur in patients with prior AI exposure, whereas they are relatively rare in treatment-naive individuals, suggesting that these alterations arise as an adaptive response to therapeutic selective pressure.⁴⁴ Similarly, another clinical cohort study reached comparable conclusions: ESR1 mutations are associated with poorer treatment outcomes, with mutation-positive patients exhibiting significantly shorter progression-free survival during subsequent AI therapy, indicating limited responsiveness to these agents. During monitoring and treatment, plasma ctDNA detection shows high concordance with tissue biopsy results and can identify the emergence of mutations before radiographic progression, thereby enabling early detection of endocrine resistance and providing a noninvasive approach for dynamic monitoring of ESR1 mutations.⁴⁵ Moreover, dynamic changes in ESR1 mutations within ctDNA—such as persistent detection or increasing allele frequency—are generally associated with poor prognosis, whereas clearance of the mutation indicates a favorable treatment response.⁴⁶ For instance, upon detection of ESR1 mutations in ctDNA, switching from an aromatase inhibitor to fulvestrant combined with a CDK4/6 inhibitor has been shown to prolong progression-free survival.⁴⁷ Although multiple studies have shown that ctDNA testing has high concordance in detecting ESR1 mutations, discrepancies still exist between studies.^48,49 These differences may be related to testing platforms, sample handling, and mutation characteristics. Variations in NGS panels, digital PCR, and other methods can affect sensitivity and specificity, while blood collection timing, processing, and cfDNA levels may further influence results. In addition, low-abundance mutations and spatial tumor heterogeneity may lead to discordance between ctDNA and tissue biopsy. Therefore, ESR1 mutation results should be interpreted cautiously in clinical practice, taking platform differences and clinical context into account.

Similar evidence has also been observed in studies related to PIK3CA mutations.⁵⁰ Aberrations in the PI3K/AKT/mTOR pathway are relatively common in HR⁺/HER2^- breast cancer, with approximately 40% of patients harboring PIK3CA mutations.^51,52 A global, randomized, double-blind, placebo-controlled phase 3 clinical trial demonstrated that in patients with PIK3CA-mutant tumors, the PI3Kα inhibitor alpelisib combined with fulvestrant significantly prolonged progression-free survival, whereas no substantial benefit was observed in patients with wild-type tumors, indicating that the therapeutic effect is mutation-specific.⁵² In addition, an innovative multicenter, platform-based clinical trial screened over 1,000 patients with metastatic breast cancer using ctDNA and assigned them to corresponding targeted therapy cohorts based on distinct driver gene mutations.^53,54 The study demonstrated that ctDNA testing showed high concordance with tissue biopsy for identifying PIK3CA mutations and provided results in a shorter time frame, thereby enhancing clinical decision-making efficiency. Patients selected via ctDNA screening exhibited clinically meaningful objective response rates (ORR) and disease control with the corresponding targeted therapies, supporting the practical utility of ctDNA testing in patient selection and therapy matching.⁵³ The clinical value of ctDNA testing hinges on its ability to distinguish predictive utility from prognostic value. This paradigm is fundamentally established by the pivotal EMERALD trial,⁵⁵ the first phase 3 study to prospectively validate ESR1 mutations as a bona fide therapeutic target in metastatic breast cancer. By demonstrating the superior efficacy of elacestrant—a novel oral SERD—over standard endocrine therapy specifically in the ESR1-mutated cohort, EMERALD transitioned ESR1 assessment into a predictive tool for selecting next-generation endocrine agents. To summarize the latest evidence that ctDNA can guide therapeutic decision-making in metastatic breast cancer, Table 1 lists landmark clinical trials.

Table 1.

Key ctDNA trials in breast cancer

Feature	PADA-1	SOLAR-1	EMERALD
Population	HR+/HER2− advanced breast cancer receiving first-line AI + palbociclib	HR+/HER2− advanced breast cancer after prior endocrine therapy	HR+/HER2− mBC progressing on 1–2 lines of ET (including a CDK4/6i)
Biomarker	ESR1 mutation	PIK3CA mutation	ESR1 mutation
Assay	ctDNA (ddPCR)	tissue or ctDNA (therascreen®)	ctDNA (Guardant360 CDx)
Study design	phase 3 randomized trial	phase 3 randomized trial	phase 3 randomized trial
Primary endpoint	PFS after randomization	PFS in PIK3CA-mutant cohort	PFS in ESR1-mutant cohort and overall population
Key results	early switch to fulvestrant + palbociclib improved PFS	alpelisib + fulvestrant improved median PFS; no benefit in wild-type	elacestrant improved PFS; 30% reduction in risk of death/progression in ESR1-mutant group
Clinical significance	demonstrates ESR1 as a predictive biomarker enabling early treatment switch	led to FDA approval; establishes predictive (not prognostic) utility	first phase III trial to prospectively validate ESR1 as a bona fide therapeutic target; led to first oral SERD approval

In addition, dynamic changes in ctDNA have also demonstrated important prognostic value in metastatic breast cancer. In the plasmaMATCH trial and other longitudinal studies, reductions in ctDNA levels after treatment initiation were associated with prolonged progression-free survival, whereas persistent ctDNA detection or increasing variant allele frequency often indicated early therapeutic resistance and poor clinical outcomes.⁵⁶ Similar findings have been reported in studies monitoring ESR1 mutations: patients with decreased levels of ESR1-mutant ctDNA after treatment switching tended to show better responses to endocrine therapy, whereas persistent mutations were associated with shorter progression-free survival⁵⁷ (Figure 2).

Figure 2.

Core clinical applications of liquid biopsy in breast cancer

MRD detection, which can predict recurrence 8–15 months in advance; Treatment monitoring and response assessment, capturing dynamic molecular changes through ctDNA/CTC analysis; Early detection and screening in high-risk populations; Resistance mechanism detection, using ctDNA to identify mutations such as ESR1 and PIK3CA, enabling real-time molecular profiling and treatment adaptation

Early-stage disease and MRD detection

Compared with its well-evidenced and treatment-guiding applications in metastatic breast cancer, the use of liquid biopsy in early-stage disease remains primarily limited to prognostic assessment, particularly in the context of minimal residual disease (MRD) detection.^58,59 In early-stage breast cancer, MRD detection is used to enable early identification of recurrence risk at the molecular level.^60,61 Even after curative surgery or systemic therapy, patients may harbor extremely low levels of residual tumor cells or their molecular traces, undetectable by conventional imaging. In this context, ctDNA analysis serves as a molecular monitoring tool, and positive ctDNA detection has been shown to reliably predict tumor recurrence.⁶² In ctDNA analysis, tumor-informed approaches are designed based on patient-specific mutations from the primary tumor, offering high sensitivity and the ability to detect low-frequency ctDNA, but they incur higher costs and longer turnaround times. In contrast, tumor-naive approaches do not require tumor tissue, relying on standardized mutation panels; these methods are faster, more cost effective, and amenable to large-scale application, but exhibit relatively lower sensitivity. Thus, each strategy involves trade-offs between sensitivity, cost, turnaround time, and clinical applicability.⁶³ The pioneering study by Garcia-Murillas et al. demonstrated that post-surgical ctDNA–positive patients had a 12-fold higher risk of recurrence compared with ctDNA-negative individuals. Molecular recurrence was detected on average 7.9 months earlier than conventional imaging, establishing the conceptual foundation for “lead time” in ctDNA monitoring.⁶⁰ Subsequently, Coombes et al. demonstrated using ultra-sensitive sequencing that ctDNA monitoring during follow-up could achieve 100% specificity, with some molecular signals preceding clinical recurrence by up to two years.²³ In high-risk subtypes, the BRE12-158 trial led by Radovich demonstrated that triple-negative breast cancer patients who did not achieve pathological complete response (pCR) after neoadjuvant chemotherapy and were ctDNA-positive post-surgery exhibited significantly reduced 24-month distant disease-free survival.⁶² However, unlike the well-established clinical applications in metastatic breast cancer, MRD detection currently remains primarily a prognostic tool. Although the cTRAK trial did not demonstrate significant benefit from immunotherapy in MRD-positive patients, it marked a pivotal shift toward investigating how molecular signals can be translated into treatment decisions through randomized controlled trials, representing a critical step for the clinical implementation of liquid biopsy.⁵⁸ To date, most studies remain observational, lacking randomized controlled trials that directly validate the critical causal chain of “detection—intervention—improved outcomes.” Moreover, the potential risk of overtreatment from premature intervention adds further uncertainty to the clinical application of MRD. At present, the field largely remains at the stage of “molecular detection—recurrence prediction,” with translation to clinical utility still dependent on high-quality interventional studies for verification (Figure 2).

Non-ctDNA biomarkers in clinical stratification: CTCs, miRNAs, and multi-marker integration

CTCs and miRNAs also play important roles in the clinical stratification of breast cancer patients. In early-stage breast cancer, even CTCs ≥1 cell per 7.5 mL can identify patients at significantly increased risk of recurrence. In addition, molecular profiling of CTCs can provide real-time insights into tumor biology, thereby enabling the classification of patients into distinct therapeutic benefit subgroups.⁶⁴ However, the clinical application of CTCs remains limited by their rarity and variability in detection methodologies.

MiRNAs, as stable circulating molecular biomarkers, have also shown promising potential in risk stratification. In triple-negative breast cancer (TNBC), circulating miRNA expression profiles have been shown to distinguish between recurrent and non-recurrent patients. For example, Kujala et al. identified 10 differentially expressed miRNAs through sequencing analysis, among which upregulation of miR-21-5p and downregulation of miR-16-5p and miR-26b-5p were independently associated with poorer recurrence-free survival. These miRNA signatures can stratify patients into high- and low-risk groups and provide guidance for decisions regarding the intensity of adjuvant therapy.⁶⁵ Nevertheless, variability across detection platforms and the lack of standardized procedures remain key barriers to their clinical translation (Figure 2).

However, a single biomarker is often insufficient to comprehensively capture tumor heterogeneity, whereas an integrative strategy combining CTCs, miRNAs, and ctDNA can provide a more complete molecular profile of tumors. Tayeb et al.by synthesizing 11 systematic reviews, found that CTCs exhibit high specificity making them more suitable for confirmatory assessment, whereas miRNAs demonstrate moderate to high sensitivity, making them more appropriate for screening and risk stratification.⁶⁶ Representative biomarkers such as miR-21 show strong discriminatory performance in distinguishing breast cancer patients from healthy controls. In this context, a higher-quality clinical decision-making process requires the integration of multiple liquid biopsy biomarkers with multi-modal analytical approaches and computational algorithms. By incorporating weighted contributions of different biomarkers, risk modeling, and machine learning-based data integration, a synergistic interpretation of multi-parameter information can be achieved.⁶⁷ This approach not only improves the robustness of single biomarker detection but also leverages the complementary relationships and potential interactions among biomarkers to better capture tumor heterogeneity and dynamic changes, thereby serving as a key component in multi-biomarker integrative interpretation. For example, CTCs primarily reflect tumor cell phenotypic and quantitative characteristics, whereas ctDNA is used to assess potential genomic mutation profiles. On this basis, machine learning-based multimodal integration was applied to jointly model CTC-derived phenotypic features and ctDNA-associated mutational information. Feature fusion and classification algorithms were then employed to construct a predictive model for endocrine therapy resistance⁶⁸ (Figure 2).

Artificial intelligence integration: From pattern recognition to clinical decision support

AI-enhanced biomarker discovery and interpretation

Machine learning models are well suited to identify subtle and nonlinear patterns across large molecular datasets, enabling more accurate distinction between tumor-derived signals and the extensive biological background noise inherent to plasma-based assays.⁶⁹ By integrating multimodal information—ranging from genomic and transcriptomic alterations to proteomic and metabolomic signatures—AI is able to construct composite biomarker profiles that outperform single-marker approaches and reveal relationships that conventional statistical methods often miss.⁷⁰ A major advance in this area comes from the incorporation of explainable AI, which has begun to address long-standing concerns about the “black-box” nature of deep learning.⁷¹ Methods such as Shapley additive explanations (SHAP) values and attention mechanisms allow investigators to visualize which features drive model predictions, an essential requirement for high-stakes clinical decisions.⁷² At the same time, federated learning frameworks offer a practical solution to the privacy challenges inherent to large-scale data aggregation, enabling models to train across multiple institutions⁷³ (Figure 3).

Figure 3.

AI facilitates multi-omics integration, biomarker discovery, and model development, supporting real-time treatment assessment, prognostic prediction, and individualized dose adjustment

At the same time, key challenges remain, including data bias, insufficient validation across diverse populations, and the lack of established regulatory standards.

Beyond biomarker discovery, AI has begun to influence how liquid biopsy information is synthesized and translated into clinical decisions. Decision support systems that integrate ctDNA dynamics with imaging findings, pathological features, and patient-specific clinical variables are steadily improving the ability to characterize disease status in real time. Such systems can identify early molecular evidence of treatment response or resistance, allowing clinicians to adapt management strategies well before changes become apparent on radiographic assessment.⁷⁴ These predictive models have the potential to substantially accelerate therapeutic evaluation by linking early ctDNA trends to long-term outcomes such as progression-free or overall survival. As datasets grow, AI-enabled decision tools may support more nuanced treatment strategies, including determining the optimal duration of therapy, identifying patients who may benefit from treatment escalation,⁷⁵ and guiding de-escalation in individuals who achieve deep molecular response.⁷⁶ These developments highlight how AI can transform liquid biopsy into a continuously informative tool that guides personalized cancer care.

The clinical translation of these technological pathways has been substantiated by multiple clinically validated platforms. The MRD-enhanced detection of genomic events (EDGE) platform employs a machine learning-guided whole-genome sequencing approach that achieved 301-fold signal-to-noise enrichment, enabling ultrasensitive detection of minimal residual disease across multiple cancer types with lead times up to 14 months before radiographic recurrence. In advanced melanoma and small cell lung cancer, plasma-only ctDNA dynamics at week 3 significantly correlated with progression-free and overall survival.⁷⁷ Regulatory-approved assays further demonstrate clinical translation. The Guardant360 CDx assay, approved as a companion diagnostic for osimertinib and sotorasib in non-small cell lung cancer, integrates machine learning-based bioinformatics for mutation detection with a limit of detection of 0.2% and per-sample specificity exceeding 98%. Clinical bridging studies confirmed that patients selected by this AI-enhanced liquid biopsy derived equivalent treatment benefit as those selected by tissue testing.^78,79

Challenges and limitations

Despite its promise, the integration of artificial intelligence into liquid biopsy workflows faces several important challenges. AI models are highly dependent on data quality, and inconsistencies in sample processing, sequencing platforms, or annotation standards can propagate substantial noise into predictive algorithms. Systematic biases in training datasets risk producing models that perform unevenly across populations. External validation also remains a significant barrier. Algorithms that perform well within the institutions that developed often show reduced accuracy when applied to independent cohorts, reflecting differences in patient characteristics, assay conditions, and clinical workflows. Finally, regulatory frameworks have not yet fully adapted to the complexity of AI-driven diagnostics, particularly those that update continuously as new data are incorporated⁸⁰ (Figure 3).

Barriers to implementation: The clinical translation gap

Scientific and biological barriers

Liquid biopsy is progressively transitioning from a technological platform to clinical application; however, its translation requires systematically traversing three critical levels of evidence: analytical validity, clinical validity, and clinical utility.⁸¹ This continuum—from technical performance to demonstrated patient benefit—represents the fundamental pathway for any in vitro diagnostic to achieve routine clinical implementation. Despite strong analytical performance across multiple platforms, current progress remains unbalanced. Targeted approaches enable highly sensitive detection of predefined variants, whereas untargeted strategies allow broader mutation discovery at the cost of reduced sensitivity and specificity.^{82,83,84,85,86,87,88} Most assays can reliably detect ctDNA, CTCs, or methylation signatures and show consistent associations with tumor burden, treatment response, and recurrence risk, thereby supporting clinical validity; however, translation into clinical utility remains limited. Existing evidence is largely derived from observational studies and real-world datasets. While large-scale analyses have demonstrated high concordance between platforms such as Guardant360 and orthogonal genotyping methods,⁸⁹ head-to-head comparisons reveal limited complete agreement across platforms, raising concerns regarding reproducibility and analytical validity.⁹⁰ More importantly, although ctDNA-based assays correlate with clinical outcomes in advanced disease, there is insufficient evidence that their use improves survival or guides effective interventions. A joint review by the American Society of Clinical Oncology and the College of American Pathologists highlighted the limited evidence supporting clinical utility, particularly in early detection, treatment monitoring, and minimal residual disease assessment.⁹¹ This gap between clinical validity and clinical utility constitutes the so-called “valley of death” in liquid biopsy research,⁹² reflecting the lack of high-level evidence linking technological capability to improved patient outcomes. Contributing factors include insufficiently powered prospective studies, limited randomized controlled trials, high costs, and long follow-up requirements, as well as weak commercial incentives for non-proprietary biomarkers.⁹¹ Beyond evidence gaps, biological heterogeneity further constrains performance. ctDNA shedding varies by tumor subtype, burden, vascularization, and treatment dynamics; for example, the I-SPY2 study showed significantly lower ctDNA positivity rates in hormone receptor-positive breast cancer compared to triple-negative disease, often below detection thresholds even in metastconstrainngs. In addition, clonal hematopoiesis of indeterminate potential (CHIP) can introduce false-positive mutations, with one validation study reporting that 17.3% of detected variants originated from hematopoiesis rather than tumors.

Implementation and socioeconomic barriers

Liquid biopsy faces substantial practical and socioeconomic barriers in clinical implementation. A major challenge is the lack of standardization across pre-analytical, analytical, and bioinformatic workflows. Variability in blood collection, processing time, centrifugation protocols, DNA extraction, library preparation, and computational pipelines can significantly affect ctDNA yield and assay performance.⁹³ In addition, variability in sensitivity, specificity, and reproducibility across different assay platforms further underscores the importance of standardized workflows.^94,95,96 Economic barriers are also substantial. Tumor-informed assays require high upfront-sequencing costs, while serial plasma monitoring imposes ongoing financial burdens. Even tumor-agnostic approaches, such as NGS combined with digital PCR, can place considerable strain on healthcare systems when used for longitudinal tumor surveillance.⁹⁷ Health equity represents a critical yet under-addressed dimension of implementation. Populations with the highest cancer burden are often underrepresented in research and underserved in clinical practice. Structural barriers, including limited insurance coverage, inadequate access to specialized care, and low participation in clinical trials, exacerbate these disparities. Addressing these challenges requires coordinated, multilevel strategies, including validation of assays in diverse populations, equitable reimbursement policies, expanded clinical infrastructure, and enhanced clinician training. Ultimately, ensuring equitable access is not an optional consideration but a prerequisite for the responsible and effective integration of liquid biopsy into routine cancer care.⁹⁸

Future directions: From therapy guidance to continuous surveillance

In recent years, the application of liquid biopsy in therapeutic strategies has gradually expanded from resistance monitoring to several promising clinical directions. For example, multi-cancer early detection technologies can screen for multiple cancers lacking routine screening modalities from a single blood draw, enabling effective early-stage detection and showing potential to reshape cancer screening paradigms. Meanwhile, a meta-analysis of 13 trials including 380 patients showed that ctDNA clearance was highly associated with pathologic complete response, with a sensitivity of approximately 98%, while patients without clearance had a markedly lower likelihood of achieving pathologic complete response. Real-world data show that among patients with metastatic cancer who had failed standard treatments, 88% received treatment guidance through liquid biopsy, with 64% matching FDA-approved targeted therapies, highlighting the broad applicability of liquid biopsy in precision treatment decision-making. Technologically, advances in nanotechnology and molecular diagnostics have ushered in a new era of precision liquid biopsy.⁹⁹ Microfluidic chips, nanowire arrays, nano sensors, and CRISPR-based platforms have enabled the highly integrated combination of sample preparation, amplification, detection, and analysis, facilitating rapid point-of-care testing with femtomolar-level sensitivity.^100,101,102 Meanwhile, single-cell multi-omics approaches and ex vivo culture of patient-derived CTCs enable simultaneous genomic, transcriptomic, and proteomic analyses, uncovering previously unrecognized biological heterogeneity and functional dynamics. These methods provide new avenues for revealing tumor heterogeneity and guiding personalized drug screening. Although early studies have demonstrated their feasibility, significant challenges remain in translating these technologies into routine clinical practice^103,104 (Figure 4).

Figure 4.

Nanotechnology (including microfluidic chips and nanosensors) enables the integration of sample preparation, amplification, detection, and analysis into a unified workflow

CRISPR-based diagnostic platforms offer femtomolar-level sensitivity. Single-cell multi-omics and ex vivo CTC culture allow simultaneous analysis of genomic, transcriptomic, and drug-response features, strengthening precision diagnostics and personalized therapy.

Emerging liquid biopsy technologies are transforming cancer management, shifting the paradigm from episodic disease detection to continuous molecular surveillance.¹⁰⁵ Regular liquid biopsy assessments could be integrated into routine health check-ups to establish individualized molecular baselines, detect aberrant signatures before overt tumor formation, and guide precision interventions, including enhanced surveillance, chemoprevention strategies, lifestyle modifications, and early enrollment in clinical trials.¹⁰⁶ Regular liquid biopsy assessments could be integrated into routine health check-ups to establish individualized molecular baselines, detect aberrant signatures before overt tumor formation, and guide precision interventions, including enhanced surveillance, chemoprevention strategies, lifestyle modifications, and early enrollment in clinical trials. However, realizing continuous health management driven by liquid biopsy requires coordinated efforts across multiple stakeholders. Rigorous validation studies across diverse populations are needed to establish clinical reliability, the industry must develop scalable and cost-effective platforms while committing to post-approval evaluation. Beyond individual health systems, global equity must be considered, including the simplification of technologies for resource-limited settings, capacity-building initiatives, and mechanisms to address the imbalance between research investment and global disease burden.

Acknowledgments

This work was supported by grants from the National Natural Science Foundation of China (grant no.: 82360469), grants from the Natural Science Foundation of Gansu Province (24JRRA937, 25JRRA1301), the Excellent Introduced Talents and Doctoral Startup Fund Project of Gansu Provincial Maternity and Child Care Hospital (GMCCH2024-2-7), the Talent Training Project of Gansu Provincial Maternity and Child Care Hospital (GMCCH2025-2-3-04), the Excellent Introduced Talents and Doctoral Startup, and the Major Project of Scientific and Technological Innovation in Gansu Provincial Health Industry (GSWSQNPY2025-12). All images were created in BioRender (BioRender.com).

Author contributions

Y. Li, conceptualization, data curation, investigation, visualization, writing – original draft, writing – review and editing, software, and methodology; Y.W., data curation, formal analysis, investigation, methodology, software, visualization, writing – original draft, and writing – review and editing; R.W., methodology, investigation, and data curation; Y. Liu, investigation, data curation, writing – review and editing, and methodology; J.P., data curation, conceptualization, funding acquisition, investigation, methodology, writing – review and editing, and supervision.

Declaration of interests

The authors declare no competing interests.