Multiscale mechanobiology of circulating tumor cells: microfluidics and atomic force microscopy

1 Introduction: Why CTC mechanics matter

Cancer remains a major global health burden, with an estimated 20.0 million new cases and 9.7 million deaths worldwide in 2022 (Bray et al., 2024). Global Burden of Disease estimates attributed 250 million disability-adjusted life years (DALYs) to cancer in 2019, with 96.9% arising from years of life lost due to premature mortality (Global Burden of Disease Cancer Collaboration, 2022). Lung, breast, colorectal, and prostate cancers were the four most commonly diagnosed cancers worldwide, and lung cancer accounted for 18.7% of cancer deaths in 2022 (Bray et al., 2024). The scale of this burden reinforces the need for improved methods to detect and characterize disseminated tumor cells involved in cancer progression.

Cancer progression is a process of somatic evolution in which genetic alterations and non-genetic plasticity generate tumor cell variants with different capacities for survival, proliferation, invasion, and treatment resistance. Variants better adapted to prevailing tissue conditions or treatment pressures preferentially survive and expand (Chaffer and Weinberg, 2011; Greaves and Maley, 2012). Within this broader process, the metastatic cascade comprises local invasion, vascular entry, survival during circulatory transport, arrest within distant vascular beds, extravasation, and subsequent persistence or metastatic outgrowth in secondary tissues (Figure 1a) (Follain et al., 2020; Ring et al., 2023). Each stage exposes individual tumor cells and multicellular clusters to distinct mechanical constraints, including matrix stiffness and confinement, fluid shear, cellular and nuclear deformation, cluster cohesion, and endothelial adhesion (Kumar and Weaver, 2009; Wirtz et al., 2011).

In many carcinomas, epithelial-mesenchymal plasticity contributes to cancer progression, particularly metastatic dissemination, by enabling transitions among epithelial, hybrid epithelial/mesenchymal, and mesenchymal states (Haerinck et al., 2023). Epithelial-mesenchymal transition (EMT)-associated programs can reduce cell-cell adhesion, reorganize the cytoskeleton, and promote migratory and invasive behavior, thereby facilitating detachment, local invasion, and intravasation (Kalluri and Weinberg, 2009; Zhang et al., 2025; Abusamra et al., 2026). Metastatic dissemination does not require a complete EMT followed by mesenchymal-to-epithelial transition (MET). Tumor cells may retain epithelial features or occupy hybrid epithelial/mesenchymal states and can circulate individually or within multicellular clusters (Aceto et al., 2014; Padmanaban et al., 2019; Haerinck et al., 2023). Following intravasation, individual circulating tumor cells (CTCs) and CTC clusters must survive fluid shear, immune attack and anchorage loss-induced anoikis before arresting and extravasating into distant tissues (Kurma and Alix-Panabieres, 2023). At secondary sites, MET-associated programs may promote the reacquisition or reinforcement of epithelial features that support disseminated tumor cell survival, proliferation, and metastatic outgrowth (Kim et al., 2017).

Metastatic competence therefore has both mechanical and molecular dimensions. Evidence from cancer cell models, engineered microenvironments, and the limited studies of patient-derived CTCs indicates that progression through the metastatic cascade involves context-dependent remodeling of cellular and nuclear deformability, apparent elasticity, viscoelasticity, adhesion, cytoskeletal organization, and actomyosin contractility. These properties shape cellular responses to matrix confinement, fluid shear, endothelial contact, and the mechanical environments of distant tissues. Hypoxia and other microenvironmental cues can further modify these mechanical states (Figure 1b) (Perea Paizal et al., 2021; Ildiz et al., 2023; Agrawal et al., 2025; Mierke, 2025; Belousova et al., 2026). Cell surface organization, including adhesion receptor distribution and glycocalyx architecture, adds another level of regulation by influencing cell-cell and cell-matrix interactions, immune recognition, vascular arrest, and extravasation (Offeddu et al., 2021; Chastney et al., 2025). CTC mechanics should consequently be understood as a dynamic, multiscale phenotype shaped by molecular state, cellular architecture, mechanical history, and microenvironmental exposure.

CTCs are scarce and phenotypically heterogeneous, and estimates of their abundance vary widely across studies. Reported counts depend on cancer type, disease stage, blood volume analyzed, preanalytical handling, detection platform, and the criteria used to define a CTC (Dai et al., 2025). Individual CTCs and CTC clusters are distinct analytical units and should be reported separately because a cluster event can contain multiple tumor cells and may also include immune or stromal companion cells (Aceto, 2020). Population-averaged molecular or mechanical measurements can further obscure subpopulations with distinct deformability, adhesion, or metastatic potential, creating a need for methods that resolve cell-to-cell and subcellular variation while preserving molecular and functional information (Lekka et al., 2012).

CTC clusters present an additional analytical challenge because isolation, immobilization, or dissociation can alter or disrupt their multicellular architecture, cell-cell junctions, companion cell associations, and regional mechanical organization (Yang et al., 2024). CTC clusters have been detected before surgery in early-stage breast cancer, and in both early and metastatic breast cancer cohorts (Krol et al., 2021; Reduzzi et al., 2021). Together, these considerations motivate analytical workflows that recover and position rare viable CTCs while preserving their mechanical state for subsequent spatially resolved characterization.

Microfluidics and atomic force microscopy (AFM) address complementary aspects of this analytical challenge. Microfluidic platforms can isolate, enrich, position, and manipulate rare CTCs on the basis of physical properties, including size, deformability, density, and hydrodynamic behavior, or through biological recognition of cell surface antigens (Saghezi et al., 2025; Xu et al., 2025). These procedures can preferentially recover particular phenotypes and expose cells to shear, confinement, affinity capture, release, and substrate attachment, potentially altering the mechanical properties measured downstream. AFM can characterize viable cancer cells under controlled aqueous conditions, resolving surface architecture and mechanical heterogeneity at single-cell and subcellular scales without requiring exogenous labels for the mechanical measurement. Functionalized AFM probes additionally enable quantification of specific molecular recognition forces at the cell surface (Muller et al., 2021). However, its label-free mechanical measurements do not establish tumor identity, which requires complementary immunophenotypic, molecular, or genomic evidence.

Recent reviews of CTC microfluidics have examined physical and affinity-based isolation, single-cell and cluster capture, integrated detection, and clinical translation (Anwar et al., 2024; Qiao et al., 2024; Yang et al., 2024; Saghezi et al., 2025; Xu et al., 2025; Kumari et al., 2026; Li et al., 2026). In parallel, AFM reviews have addressed medical applications, cancer liquid biopsy, molecular analysis, mechanophenotyping, and biomarker development (Krawczyk-Woloszyn et al., 2024; Li, 2024; Marcuello et al., 2025; Persano et al., 2025; Stylianou, 2026). Complementary reviews have also examined AFM in combination with fluorescence and super-resolution microscopy, Raman spectroscopy, and other single-cell analytical methods (Miranda et al., 2021; Li, 2023a; Khadem et al., 2024).

Workflows linking rare-cell isolation to AFM characterization of the same cells remain comparatively underexplored. Patient-derived evidence includes membrane filtration followed by AFM mapping directly on the capture membrane (Chen et al., 2013; Osmulski et al., 2014; Huang et al., 2016) and affinity-based microfluidic capture followed by AFM analysis of prostate CTCs (Deliorman et al., 2020; Glia et al., 2020). Marker-independent microfluidic separation coupled with deep learning-guided AFM measurements has been demonstrated in model mixtures of cultured cancer cells and blood cells (Qi et al., 2025).

Together, these studies establish technical feasibility, although generalizability across tumor types and CTC phenotypes remains unresolved. However, no AFM study directly characterizing intact patient-derived CTC clusters was identified. Current understanding of cluster mechanics must therefore be inferred from individual clinical CTCs, microfluidic cluster isolation studies, and cultured multicellular models.

This review addresses that interface through a multiscale examination of the molecular and cellular determinants of CTC mechanics, the mechanical phenotypes relevant to circulation and metastatic dissemination, microfluidic isolation and handling, AFM-based characterization, and emerging combined AFM-microfluidic workflows. Throughout, we distinguish four evidence categories: (1) patient-derived CTCs, including individual cells and clusters; (2) spiked blood samples or model rare-cell mixtures; (3) cultured cancer cell models, including engineered microenvironments; and (4) enabling technologies whose application to CTCs remains prospective. Particular emphasis is placed on how isolation and immobilization can influence subsequent mechanical measurements and on the unresolved challenge of characterizing intact CTC clusters as multicellular mechanical units.

2 Multiscale mechanobiology of CTC dissemination

CTC behavior emerges from coupled changes across molecular, subcellular, cellular, and multicellular scales. Epithelial-mesenchymal plasticity, cytoskeletal and nuclear remodeling, membrane organization, mechanical heterogeneity, and cluster composition jointly influence how tumor cells enter, withstand, and exit the circulation. Figure 2 integrates representative evidence linking these processes to CTC-relevant mechanical states and multimodal phenotyping.

2.1 Molecular and structural determinants of CTC mechanics

Genetic alterations and molecular signaling establish the biological context in which mechanical adaptation occurs during tumor progression (Greaves and Maley, 2012; Weinberg, 2013). Growth factor production, receptor overexpression, impaired tumor-suppressor signaling, and resistance to programmed cell death support uncontrolled proliferation, tissue invasion, and adaptation to biological and mechanical stressors. Mitogenic and paracrine factors, including transforming growth factor-α, platelet-derived growth factor, and hepatocyte growth factor, also mediate communication among tumor, stromal, endothelial, and immune cells (Weinberg, 2013). The effects of transforming growth factor-β (TGF-β), on the other hand, are context dependent. It can restrain proliferation during early tumorigenesis, whereas in advanced disease it can promote EMT, invasion, immune evasion, extracellular matrix remodeling, and metastatic progression (Kalluri and Weinberg, 2009).

Hypoxia-inducible signaling, extracellular matrix remodeling, and altered cell-cell adhesion can disrupt epithelial organization and increase tumor cell motility. Loss of E-cadherin-mediated adhesion weakens intercellular junctions and facilitates EMT-associated detachment, invasion, and vascular entry (Kalluri and Weinberg, 2009; Weinberg, 2013). However, EMT is not an all-or-none transition (Fontana et al., 2024). Tumor cells can occupy epithelial, mesenchymal, and intermediate or hybrid states, generating diverse combinations of adhesive, proliferative, migratory, and invasive behavior (Genna et al., 2020). This epithelial-mesenchymal plasticity is illustrated in Figure 2a, together with representative epithelial, mesenchymal, and mixed transcript profiles identified in patient-derived breast CTCs (Yu et al., 2013).

Mechanical regulation also arises through reciprocal interactions between tumor cells and the extracellular matrix. Matrix stiffening and collagen crosslinking can strengthen integrin-mediated adhesion, cytoskeletal tension, and growth factor signaling (Paszek et al., 2005; Levental et al., 2009). Glycocalyx enlargement can further promote integrin clustering and mechanically reinforced signaling at the cell surface (Paszek et al., 2014). Downstream pathways involving integrin-focal adhesion signaling, Yes-associated protein, and transcriptional coactivator with PDZ-binding motif (YAP/TAZ) connect extracellular matrix stiffness and cell surface organization with cytoskeletal tension, nuclear responses, and cellular mechanical phenotype (Li and Wang, 2020; Wang et al., 2023). These pathways link molecular state to measurable properties, including apparent elasticity, adhesion, deformability, and viscoelasticity.

Detachment from the extracellular matrix introduces an additional survival barrier. Normal epithelial cells depend on integrin-mediated adhesion for survival signaling, and its loss can trigger anoikis (Kanchanawong and Calderwood, 2023; Corsini et al., 2025). Tumor cells that acquire anoikis resistance can survive in suspension while adapting to altered cytoskeletal tension and adhesion signaling during hematogenous dissemination (Perea Paizal et al., 2021; Corsini et al., 2025). Following vascular arrest and extravasation, partial or complete MET can restore epithelial characteristics that support proliferation and metastatic outgrowth (Yu et al., 2013). CTC mechanics should therefore be interpreted as a context-dependent readout of reversible molecular and structural states across the metastatic cascade.

2.2 Mechanical phenotypes across the metastatic cascade

Cancer cells frequently differ from nonmalignant cells in stiffness, deformability, adhesion, viscoelasticity, and cytoskeletal organization, with substantial variation among cancer types, cellular states, and levels of metastatic potential (Suresh, 2007; Kumar and Weaver, 2009; Swaminathan et al., 2011; Wirtz et al., 2011). Stiffness describes resistance to applied load, deformability reflects whole-cell shape change under flow or confinement, adhesion governs interactions with other cells and extracellular components, and viscoelasticity captures time-dependent deformation and recovery (Yu et al., 2022; Li, 2024; Agrawal et al., 2025; Belousova et al., 2026). Mechanical softening or stiffening may reflect active remodeling, passive responses to mechanical stress, or pre-existing cellular states. Such mechanical phenotypes should therefore be interpreted within their specific biological and experimental context unless a direct mechanistic relationship is demonstrated through controlled or longitudinal experiments.

Patient-derived tumor cells collected from malignant body cavity fluids illustrate the mechanical heterogeneity associated with metastatic disease. Their mean stiffness was about 70% lower than that of benign cells obtained from the same sample types, consistent with greater deformability (Cross et al., 2007). The broad and distinct distributions shown in Figure 2b demonstrate that population averages can obscure mechanically different subpopulations. These cells were not identified as CTCs. The comparison nevertheless indicates that metastatic tumor cells from body cavity fluids can be less stiff than benign cells from comparable specimens.

Mechanical phenotype is dynamic and context-dependent. A softer state may facilitate invasion and passage through confined spaces, whereas fluid shear, rigid extracellular matrices, and therapy can induce cytoskeletal reinforcement or resistance-associated remodeling (Suresh, 2007; Wirtz et al., 2011; Agrawal et al., 2025). Changes induced by shear or substrate stiffness can also persist after cells enter a new environment, indicating that mechanical history may influence subsequent behavior (Cambria et al., 2024; Lee and Holle, 2024; Meng et al., 2025). Reduced stiffness should therefore not be interpreted as a universal or permanent feature of metastatic cells.

Circulatory shear also modifies molecular state, adhesion, and nuclear organization. In cancer cell models, shear exposure increased the expression of stemness-related genes, including ALDH1, CD44, and CD133, and brain metastasis-associated genes, including COX2 and ST6GALNAC5. c-Jun N-terminal kinase activation provides a potential link between mechanical stimulation, EMT-associated transcription, and circulatory adaptation (Jin et al., 2018; Jin et al., 2022). Increasing shear was accompanied by nuclear expansion in suspended SK-BR-3 and MCF-7 cells (Figure 2c) (Xu et al., 2022). Responses depended on cellular state, with epithelial-like SK-BR-3 cells showing more prominent changes in EMT-associated transcription than mesenchymal-like MDA-MB-231 cells (Cognart et al., 2020). Shear exposure can also modify cancer cell adhesion to extracellular matrix during drug treatment (Spencer and Baker, 2016). These findings remain to be confirmed in patient-derived CTCs.

Physical confinement imposes a distinct challenge because the nucleus is often the largest and least deformable structure passing through narrow spaces. Migration through microfluidic constrictions can alter nuclear shape and chromatin organization, as illustrated in Figure 2d. Narrow pore migration induced partially reversible chromatin condensation in cancer cell models. The overall abundance of highly condensed chromatin recovered after deformation, while its spatial distribution remained altered (Blazquez-Romero et al., 2026). Because chromatin organization contributes to nuclear mechanics, persistent redistribution may influence subsequent cellular behavior (Stephens et al., 2017; Xie et al., 2018; Deliorman et al., 2023b; De Corato and Gomez-Benito, 2024).

Together, these observations support mechanical phenotype as a preclinical candidate marker for CTC classification and a functional indicator of invasive behavior, although prospective clinical validation remains limited (Yu et al., 2022). Simulations further suggest that modest differences in cell size and membrane elasticity can affect CTC velocity, deformation, shear exposure, vascular transit, and arrest (Nan et al., 2023). The variation among individual cells highlights the importance of preserving distribution-level information and identifying mechanical phenotypes.

2.3 Single CTCs and CTC clusters as distinct biomechanical units

CTC heterogeneity includes molecular state, mechanical phenotype, and multicellular organization. Individual CTCs can display epithelial, hybrid epithelial/mesenchymal, or mesenchymal characteristics associated with different proliferative, invasive, and treatment-response profiles (Yu et al., 2013; Genna et al., 2020). In orthotopic breast cancer models, mesenchymal CTCs exhibited greater invasive and metastatic capacity, while intermediate-state CTCs showed stronger proliferative and colony-forming behavior (Jin et al., 2022). These findings indicate that CTC phenotypes do not form a simple linear hierarchy from lower to higher metastatic potential.

CTCs can circulate individually, as homotypic clusters composed of tumor cells, or as heterotypic clusters containing tumor cells together with blood-derived or stromal companion cells (Figure 2e). Associated cells may include platelets, neutrophils, other leukocytes, and stromal cells (Pereira-Veiga et al., 2022; Yang et al., 2024). Platelets can protect CTCs and promote invasive signaling, while neutrophil-associated clusters can support cell cycle progression and metastatic seeding (Labelle et al., 2011; Szczerba et al., 2019; Zhan et al., 2023).

Although less abundant than individual CTCs, CTC clusters exhibited 23- to 50-fold greater metastatic seeding efficiency in mouse models (Aceto et al., 2014). Detection of clusters in patient blood has also been associated with poor clinical outcomes (Rostami et al., 2019; Zhan et al., 2023). Their metastatic advantage has been linked to cell-cell adhesion, collective protection from shear and anoikis, and maintenance of cellular states associated with self-renewal, plasticity, tumor initiation, and stress resistance (Aceto et al., 2014; Szczerba et al., 2019; Zhan et al., 2023).

However, a cluster cannot be represented adequately as a collection of independent cells. Its mechanical behavior depends on cellular composition, junctional cohesion, three-dimensional (3D) geometry, regional stiffness, and interactions between peripheral and mechanically shielded cells. Dissociation disrupts this organization and can eliminate cluster-dependent information on cell-cell adhesion, cooperative survival, and mechanical heterogeneity (Aceto et al., 2014; Sarioglu et al., 2015; Yang et al., 2024). CTC clusters should therefore be treated as distinct multicellular mechanical units during isolation and characterization.

2.4 The need for single-cell multimodal analysis

CTC metastatic potential cannot be inferred reliably from molecular identity or a single mechanical property alone. Biochemical methods provide information on protein expression, signaling, marker localization, and genomic or transcriptomic state, although they generally do not connect these features with mechanical variation in the same viable cell (Budhbaware et al., 2024). Population-level mechanical measurements can similarly obscure rare subpopulations with distinct stiffness, deformability, adhesion, viscoelasticity, or transport behavior (Nan et al., 2023). Single-cell multimodal analysis is therefore needed to link tumor identity and molecular state with functional mechanical phenotype.

AFM and microfluidics address complementary aspects of this requirement. AFM provides spatially resolved measurements of local elasticity, adhesion, viscoelasticity, and surface organization, while microfluidics supports CTC enrichment, positioning, manipulation, and controlled exposure to fluidic conditions (Hao et al., 2020; Deliorman et al., 2023c; Qi et al., 2025; Saghezi et al., 2025; Xu et al., 2025). Marker-based imaging can establish phenotypic identity, and sequencing can characterize genomic and transcriptomic states (Qiao et al., 2024; Lege et al., 2026). Raman and surface-enhanced Raman spectroscopy can provide molecular fingerprints associated with proteins, lipids, nucleic acids, and phenotypic states, while single-cell mass spectrometry can provide complementary information on lipid and metabolic composition (Ide et al., 2014; Hassanzadeh-Barforoushi et al., 2024; Sarcinelli et al., 2024; Zhan et al., 2025a; Zhan et al., 2025b; Zhou et al., 2025; Fan et al., 2026; Zhan et al., 2026; Zhang and Shen, 2026). Together, these approaches could connect CTC identity, molecular state, membrane composition, and functional mechanics.

Multimodal AFM studies remain uncommon, although several coordinated strategies illustrate their value. Across HT29, Caco-2, HCT116, SW837, and SW480 colorectal cancer cell lines, AFM-derived topography, apparent elasticity, and viscoelasticity were compared with membrane microviscosity measured by fluorescence lifetime imaging microscopy and lipid composition determined by time-of-flight secondary ion mass spectrometry (Figure 2f). Confocal microscopy provided additional information on actin organization. More motile, mesenchymal-like cell lines were stiffer and exhibited more viscous plasma membranes, while lipid analysis indicated that membrane microviscosity reflected contributions from multiple membrane components (Efremov et al., 2023). A later application of this framework showed that MCF-7 breast cancer cells were softer than nonmalignant MCF-10A mammary epithelial cells, yet exhibited greater plasma membrane microviscosity, increased sphingomyelin and saturated fatty-acid signals, and reduced polyunsaturated fatty-acid signals (Shimolina et al., 2026). These studies connected mechanical, imaging, and chemical measurements across matched cell populations, although every modality was not applied sequentially to the same individually tracked cell.

Coordinated measurements can also clarify specific mechanisms. Spatially co-registered confocal imaging and AFM indentation revealed progressive stiffening of MDA-MB-231 breast cancer cells during invasion into 3D collagen I matrices. Inhibition of Rho-associated coiled-coil containing protein kinase, a regulator of actomyosin contractility and cytoskeletal tension, reduced this stiffening, implicating actomyosin contractility in the mechanical response to invasion (Staunton et al., 2016). In BxPC-3 pancreatic cancer cells, AFM and Laurdan-based two-photon microscopy were combined to examine how aquaporin-mediated transport influences membrane organization, mechanics, and motility. Silencing aquaporin-3, aquaporin-5, or both channels impaired migration and weakened cell-cell adhesion. Reduced stiffness, greater deformation, and increased membrane fluidity were primarily associated with aquaporin-5 silencing, alone or together with aquaporin-3 (Silva et al., 2022).

The destructive nature of sequencing and mass spectrometry makes preservation of cell identity across modalities technically demanding. Spatially registered workflows should therefore perform viable cell imaging and AFM before destructive molecular analysis. Reliable spatial registration, controlled recovery of selected cells, and linkage of every measurement to the same cell remain important requirements for extending multimodal analysis from cancer cell models to patient-derived CTCs.

2.5 AFM among single-cell mechanophenotyping methods

Single-cell mechanophenotyping spans high-throughput population screening and spatially resolved measurements, as summarized in Table 1. Real-time deformability cytometry and hydrodynamic stretching quantify whole-cell deformation under flow across large cell populations, supporting rapid label-free characterization of mechanical phenotypes (Gossett et al., 2012; Otto et al., 2015). Recent deformability-cytometry platforms can extract multiple contour- and deformation-based parameters and apply machine learning to classify cell populations (Zhou et al., 2024). These approaches provide population-level distributions without applying localized forces or resolving subcellular mechanical heterogeneity.

TABLE 1. Comparison of single-cell mechanophenotyping methods.

Throughput categories are relative and depend on device architecture, imaging rate, sample preparation, and analysis criteria.

Microconstriction channels measure cell entry, transit, deformation, and recovery under confinement. Entry behavior is strongly influenced by deformability, whereas transit also reflects cell-channel friction and adhesion. Cancer cell lines with greater metastatic potential have shown faster entry into constrictions, and parallelized devices can support high-throughput mechanical classification (Byun et al., 2013). Notably, a parallel constriction-based platform integrating deformability cytometry with deep learning achieved accuracies of 92.4% for invasiveness classification and 89.5% for distinguishing cancer cells from leukocytes. It also distinguished cells before and after hypoxia treatment and was evaluated using breast, lung, and bladder cancer cell lines (Hua et al., 2023). These measurements model aspects of confined vascular and tissue transit, although direct application to patient-derived CTCs remains limited.

Micropipette aspiration provides controlled measurements of whole-cell deformation, cortical tension, and time-dependent mechanical response (Hochmuth, 2000). In-silico modeling of HT29 and SW48 colorectal cancer cells suggested that increasing aspiration pressure induces pressure-dependent stiffening and attenuates increases in deformation rate, equilibrium time, and aspirated length (Ghoytasi et al., 2024). Optical stretching applies contact-free laser-induced stresses and has distinguished cells according to malignant transformation and metastatic competence (Guck et al., 2005). A local optical stretching implementation for adherent HeLa cervical cancer cells and NIH/3T3 fibroblast cells extracted viscoelastic parameters using forces of a few piconewtons and deformations of a few nanometers (Janik et al., 2025). These methods provide complementary information, although their throughput, cell attachment requirements, and suitability for rare patient-derived CTCs vary.

Magnetic twisting cytometry applies controlled torque to ligand-coated ferromagnetic beads bound to cell surface receptors, enabling local measurements of cytoskeletal mechanics and receptor-mediated force transmission. Its foundational demonstration showed integrin-dependent, force-induced stiffening that depended on the integrity of actin microfilaments, microtubules, and intermediate filaments (Wang et al., 1993). Subsequent work quantified local cytoskeletal viscoelasticity using this approach (Puig-de-Morales-Marinkovic et al., 2007), and recent reviews summarize its applications in cancer cell mechanics (Mierke, 2025). Magnetic rotational spectroscopy is a related, mechanistically distinct method that uses intracellular magnetic wires to measure cytoplasmic rheology. Compared with non-malignant MCF-10A cells, tumorigenic MCF-7 and highly metastatic MDA-MB-231 breast cancer cells exhibited significantly lower cytoplasmic elastic moduli. MCF-7 cells had a substantially longer cytoplasmic relaxation time than MDA-MB-231 cells, indicating slower stress relaxation (Dessard et al., 2024).

AFM occupies a complementary niche characterized by lower throughput and substantially greater spatial and functional resolution. It combines surface imaging, localized force control, subcellular mechanical mapping, adhesion measurements, and molecular recognition using functionalized probes. Foundational AFM work showed that normal and cancerous bladder epithelial cells could be distinguished at the single-cell level based on their elastic properties (Lekka et al., 1999). This spatial resolution enables analysis of mechanical heterogeneity within selected cells that most flow-based approaches do not resolve.

Direct AFM studies of patient-derived CTCs nevertheless remain limited. A pioneering study measured elasticity, deformation, and adhesion in 130 prostate CTCs isolated by membrane filtration from 17 patients (Osmulski et al., 2014). CTCs from patients with castration-resistant disease were softer, more deformable, and more adhesive than those from patients with castration-sensitive disease. Subsequent affinity-based microfluidic capture enabled AFM analysis of prostate CTCs from patients with localized and metastatic disease, with metastatic CTCs exhibiting lower elasticity and greater deformability (Deliorman et al., 2020). Together, these findings suggest that CTC mechanical phenotype may vary with disease status, although the evidence remains restricted to small prostate cancer cohorts. Direct AFM characterization of intact patient-derived CTC clusters was not identified in the reviewed literature. Cluster-preserving microfluidic enrichment and controlled positioning could help extend AFM studies to larger patient cohorts, additional tumor types, and intact clusters.

3 Microfluidic isolation and handling for CTC mechanophenotyping

Microfluidic isolation influences both the CTC phenotypes recovered and the physical state in which cells reach downstream analysis. Shear, confinement, capture, release, and substrate attachment can influence cell viability, cluster integrity, adhesion state, and mechanical phenotype. Platform selection must therefore consider isolation performance together with compatibility with subsequent characterization. Figure 3 summarizes representative physical, affinity-based, cluster-preserving, open-space, and AI-assisted strategies for CTC recovery and downstream analysis.

3.1 Requirements for AFM-Ready CTC isolation

CTC analysis is constrained by the rarity and phenotypic heterogeneity of tumor cells within a large background of hematologic cells (Nagrath et al., 2007; Pantel and Alix-Panabieres, 2010; Lin et al., 2018; Dai et al., 2025). Effective isolation must therefore enrich rare cells, limit leukocyte carryover, and preserve viability and molecular integrity for downstream functional analysis (Bhat et al., 2022; Kumari et al., 2026).

Conventional preprocessing methods, including red blood cell lysis, density gradient centrifugation, and membrane filtration, can reduce sample complexity. Their performance may be limited by cell loss, leukocyte contamination, mechanical stress, and physical overlap between CTCs and blood cells (Anwar et al., 2024; Saghezi et al., 2025). In this regard, microfluidic systems provide controlled flow, low reagent consumption, spatially defined capture, and coordinated enrichment, staining, imaging, molecular analysis, and cell release (Qiao et al., 2024; Wu et al., 2024; Farahinia et al., 2025; Xu et al., 2025). Separation can exploit physical differences in size, deformability, density, hydrodynamic behavior, acoustic response, or electrical polarizability, as well as biological differences in surface-antigen expression (Gascoyne and Shim, 2014; Abusamra et al., 2024; Li et al., 2026).

The intended downstream analysis should guide platform selection. Label-free approaches reduce dependence on surface marker expression and broaden the recovery of epithelial, hybrid, mesenchymal, and treatment-altered phenotypes (Yu et al., 2013; Genna et al., 2020). Affinity-based approaches can improve specificity and provide marker-associated information when suitable targets are available (Nagrath et al., 2007; Lege et al., 2026). Hybrid platforms combine physical pre-enrichment with biological capture to reduce sample complexity while improving specificity and recovery (Abusamra et al., 2024; Qiao et al., 2024).

For AFM, enrichment alone is insufficient. Recovered cells must remain viable, stable during measurement, optically identifiable, and physically accessible to the cantilever. Flow, confinement, surface attachment, staining, and release may also alter the mechanical properties intended for measurement (Deliorman et al., 2020). CTC clusters require low shear handling that preserves cell-cell junctions, companion cells, and multicellular architecture (Aceto et al., 2014; Sarioglu et al., 2015; Glia et al., 2021; Yang et al., 2024). An AFM-ready workflow should therefore be evaluated in terms of recovery, purity, viability, immobilization, cantilever access, handling history, and preservation of the mechanical phenotype of interest.

3.2 Label-free physical separation

Most evidence supporting label-free microfluidic separation derives from cultured cell models and spiked blood samples, with patient-derived validation limited to specific platforms. These platforms enrich cells according to physical properties without requiring predefined antigen recognition, thereby reducing marker selection bias and broadening recovery across epithelial, hybrid, mesenchymal, and treatment-altered phenotypes (Yu et al., 2013; Genna et al., 2020; Anwar et al., 2024; Qiao et al., 2024). Representative approaches based on filtration, inertial or Dean flow, acoustophoresis, and dielectrophoresis are shown in Figure 3a.

Membrane and microfilter systems separate cells primarily according to size and deformability using porous membranes, microfabricated constrictions, or filter arrays. Their conceptual simplicity supports integration into compact devices, while clogging, leukocyte retention, mechanical stress, and loss of small or highly deformable CTCs remain important limitations (Anwar et al., 2024; Qiao et al., 2024; Li et al., 2026). These concerns become particularly relevant when viable cells or mechanically fragile clusters are required for downstream analysis. Lateral filter arrays and serpentine channel designs have reduced clogging and achieved reported capture efficiencies of 82%–93%, with viability exceeding 92% in the evaluated cell line models (Wu et al., 2024).

Hydrodynamic and inertial methods direct cells into different streamlines or outlets according to size, deformability, and equilibrium position within a microchannel. Deterministic lateral displacement, inertial focusing, spiral channels, and Dean-flow systems support continuous processing of comparatively large sample volumes without molecular labels (Anwar et al., 2024; Qiao et al., 2024; Wu et al., 2024; Li et al., 2026). Recovery depends on channel geometry, flow rate, sample preparation, and the physical overlap between CTCs and leukocytes. The shear and confinement experienced during separation should also be considered when subsequent measurements aim to characterize native cell deformability or mechanically induced adaptation.

Acoustophoretic platforms use acoustic fields to separate cells according to size, density, and compressibility. Their contact-free operation can support viable cell recovery, although performance depends on acoustic contrast, channel configuration, and flow conditions (Anwar et al., 2024; Saghezi et al., 2025; Xu et al., 2025). Dielectrophoresis uses nonuniform electric fields to distinguish cells through membrane capacitance, cytoplasmic properties, and frequency-dependent polarizability (Shim et al., 2013; Gascoyne and Shim, 2014). 3D electrode architectures and dielectrophoretic field-flow fractionation can improve field coverage and reduce voltage requirements, while throughput, buffer composition, electrode design, and Joule heating remain important constraints (Mansor et al., 2025; Najafipour et al., 2025).

Physical selection alone does not establish tumor identity because CTCs and blood cells may overlap in size, deformability, acoustic mobility, and dielectric properties (Harouaka et al., 2013). Recovered events should therefore remain classified as candidate CTCs until confirmed through immunophenotypic, molecular, or genomic analysis. Label-free methods are particularly valuable as broad pre-enrichment steps followed by cell identification, spatial positioning, and targeted mechanical characterization (Edd et al., 2022; Qiao et al., 2024).

3.3 Affinity-based and hybrid isolation

Affinity-based platforms capture CTCs through specific biological interactions, most commonly antibody-antigen binding. Epithelial cell adhesion molecule (EpCAM) remains widely used to enrich carcinoma-derived CTCs that retain epithelial features (Nagrath et al., 2007; Anwar et al., 2024). The CTC-Chip established affinity capture using EpCAM-coated microposts, while the herringbone HB-Chip introduced passive microvortices to increase contact between flowing cells and antibody-functionalized surfaces (Nagrath et al., 2007; Stott et al., 2010).

Dependence on EpCAM can underrepresent CTCs that partially or completely lose epithelial markers during EMT (Pantel and Alix-Panabieres, 2010; Yu et al., 2013). Multiplexed strategies consequently combine epithelial, mesenchymal, tissue-associated, and tumor-associated markers. Expanded surface marker panels have improved live CTC capture in metastatic triple-negative breast cancer (Lege et al., 2026), while cell surface vimentin has been investigated as an additional target for mesenchymal-like CTC populations (Kanayama et al., 2024). Tetrahedral DNA frameworks carrying aptamers against EpCAM, vimentin, and epidermal growth factor receptor provide another approach for capturing heterogeneous phenotypes while preserving viability for culture and molecular analysis (Zhang et al., 2026).

Magnetic beads, aptamers, nanostructured interfaces, graphene oxide, and NanoVelcro-like surfaces can increase ligand accessibility, effective capture area, and multivalent contact with the cell membrane (Abusamra et al., 2024; Anwar et al., 2024; Qiao et al., 2024; Li et al., 2026). NanoVelcro-like platforms use nanowires, nanofibers, or nanopillars to increase the effective surface area available for ligand presentation and promote multivalent interactions with the CTC membrane. Magnetic systems additionally enable controlled cell manipulation and recovery, although affinity-based platforms remain susceptible to marker selection bias, nonspecific leukocyte binding, and variability in antigen abundance (Macaraniag et al., 2025).

Hybrid enrichment combines physical and biological separation mechanisms within a sequential workflow. The CTC-iChip exemplifies this approach by integrating deterministic lateral displacement, inertial focusing, and immunomagnetic sorting for high-throughput enrichment of viable CTCs (Figure 3b).

In positive-selection mode, magnetically labeled cells expressing selected tumor-associated antigens are enriched. In negative-depletion mode, magnetically labeled leukocytes are removed, retaining a broader population of unlabeled candidate CTCs for downstream molecular analysis (Ozkumur et al., 2013; Karabacak et al., 2014). Antibody-conjugated magnetic nano/hybrid microgels provide a related strategy for selective cell labeling, magnetic capture, and in situ monitoring within microfluidic devices (Seyfoori et al., 2023).

3.4 Cluster-preserving isolation and biophysical profiling

CTC clusters impose distinct isolation requirements. Their larger size and asymmetry can facilitate label-free recovery, while their deformability, cohesion, and cellular composition determine how they interact with microfluidic structures. Excessive shear or restrictive geometries can fragment clusters, disrupt cell-cell junctions, or remove associated blood and stromal cells, thereby altering the multicellular state intended for downstream analysis (Aceto et al., 2014; Sarioglu et al., 2015; Yang et al., 2024).

Several dedicated platforms have been developed for this purpose. The Cluster-Chip uses bifurcating traps to capture aggregates according to size and deformability, with capture exceeding 90% for large clusters under the reported low shear conditions (Sarioglu et al., 2015). Cluster-Wells combines more than 100,000 microwells with 15 μm × 15 µm bottom micromesh openings for rapid aggregate-selective screening (Figure 3c). The device recovered viable clusters containing two to more than 100 cells from one ovarian cancer patient, whereas sequencing involved clusters from two prostate cancer patients (Boya et al., 2022). Deterministic lateral displacement has also been adapted for continuous recovery of viable clusters according to aggregate size and asymmetry (Liu et al., 2021; Fan et al., 2025).

Cluster-safe operating limits depend on device geometry and cluster properties. Cluster-Chip processed blood at 2.5 mL h^-1, with a peak local flow speed of 70 μm s^-1 at the bifurcating traps, preserving cluster integrity under low-shear capture (Sarioglu et al., 2015). An upper non-damaging limit was not determined. Cluster-Wells was evaluated at 25–750 mL h^-1 (Boya et al., 2022), with dissociation evident only above 500 mL h^-1. At 250 mL h^-1, recovery remained high for larger clusters, although doublet recovery declined. 25 mL h^-1 provided broad recovery and viable post-capture release.

PillarX extends cluster isolation into combined physical and molecular profiling (Figure 3d). Its pillar array separates clusters according to geometry and deformability, followed by affinity-based characterization using functionalized magnetic nanoparticles. Cohesive clusters were less deformable and more frequently retained within the pillar array, whereas less cohesive clusters could rearrange during transit. Evaluation in mouse models and samples from patients with metastatic breast cancer linked cluster geometry, deformability, cohesion, and epithelial marker state within the same workflow (Green et al., 2022).

3.5 Open-space capture and cell recovery

Open-space microfluidic probes deliver and aspirate fluids locally over exposed substrates without enclosing captured cells within conventional channels (Deliorman et al., 2023a; Ali et al., 2026). This configuration supports spatially addressable capture, controlled exposure, and direct access for imaging or manipulation (Glia et al., 2026). These features are particularly relevant to AFM, which requires optical localization and unobstructed positioning of the cantilever over an individual cell or cluster.

The fully 3D-printed herringbone microfluidic probe (HB-MFP) combines open-space operation with multiplexed affinity capture (Figure 3e). Herringbone structures generate microvortices that increase cell contact with regions functionalized using antibodies against EpCAM, prostate-specific membrane antigen, and prostate-specific antigen. The device captured individual CTCs and clusters containing as many as 40–50 cells from blood samples of patients with prostate cancer (Glia et al., 2021). Its physically separate, antibody-functionalized glass substrate permits multiplex immunocytochemical identification and enumeration. Direct transfer of captured CTCs to an AFM stage, followed by AFM analysis involving one clinical CTC from a prostate cancer patient, was demonstrated as a proof-of-concept (Glia et al., 2020). The modular surface chemistry offers potential extension to other tumor types through the selection of appropriate capture ligands, subject to validation in relevant patient cohorts (Deliorman et al., 2023c).

Digital microfluidics provides a complementary route for controlled cell recovery and molecular processing. A digital microfluidic-mass spectrometry platform combined wettability-based trapping with hydrodynamic cell distribution to isolate individual CTCs for multiplex mutation analysis (Ruan et al., 2022). On-chip lysis and amplification within nanoliter droplets can reduce sample loss and contamination during rare-cell processing. Because these procedures are destructive, they must follow viable cell imaging and AFM when mechanical and molecular measurements are intended to describe the same recovered cell.

Rapid prototyping and 3D printing are expanding both closed-channel devices and open-space probe microfluidic architectures. These fabrication methods support complex geometries, integrated valves, functionalized capture regions, and modular components that can be difficult to produce through conventional soft lithography (Bendre et al., 2024; Saghezi et al., 2025; Glia et al., 2026; Kumari et al., 2026). Their principal value for CTC mechanophenotyping lies in combining efficient fluid handling with optical and physical access to selected viable cells.

Potential compatibility with AFM varies among cluster isolation platforms. Cluster-Chip and Cluster-Wells preserve viable intact clusters, although their use before AFM would require cluster-preserving recovery, transfer to a flat and exposed substrate, and stable immobilization. PillarX could provide upstream information on cluster geometry, deformability, and cohesion, provided that individual clusters can be recovered traceably for subsequent AFM. Among these platforms, HB-MFP offers the most direct route to AFM because capture occurs on an exposed substrate that can be handled independently and transferred to the AFM stage. However, AFM characterization of intact patient-derived CTC clusters has not yet been demonstrated with any of these systems. Future measurements will require preservation of cluster composition and cell-cell junctions, selection of an appropriate AFM tip geometry, stable positioning, and measurement of local cluster thickness. Table 2 summarizes these platform-specific considerations.

TABLE 2. Representative microfluidic platforms for CTC isolation and downstream AFM analysis.

Evidence categories indicate the sample types examined and do not imply clinical validation.

3.6 AI-assisted detection and workflow control

Artificial intelligence (AI) is increasingly incorporated into microfluidic CTC platforms for candidate cell detection, functional phenotyping, and device optimization. Machine learning and deep learning models can analyze bright-field, phase, holographic, scattering, fluorescence, impedance, or cell trajectory data to extract morphological, optical, mechanical, and transport-related features (Anwar et al., 2024; Qiao et al., 2024; Li et al., 2025; Saghezi et al., 2025). These capabilities can reduce manual image analysis and, in label-free implementations, dependence on molecular labels.

Morphological and mechanical overlap between CTCs and leukocytes prevents AI classification alone from establishing tumor origin. AI-identified events should remain candidate CTCs until confirmed through immunophenotypic, molecular, or genomic evidence. A clinical pilot combined inertial microfluidic preprocessing, in-flow digital holographic microscopy, real-time deep learning, and dual-channel immunofluorescence (Figure 3f). Integrating holographic and immunofluorescence data produced higher candidate CTC counts in patients with metastatic castration-resistant prostate cancer than in healthy controls, and reduced the false-positive rate to 1 cell per mL in a pilot involving 13 patients and 8 controls (Mallery et al., 2026).

Label-free optical scattering combined with deep learning has been used to detect homotypic and heterotypic CTC clusters in spiked whole blood, with reported purity of 72%, sensitivity of 35.3%, and a false alarm rate of 0.78 events min^-1 (Vora et al., 2024). Models based on nuclear morphodynamics have supported label-free, noninvasive estimation of individual CTC viability (Zeng et al., 2026). Convolutional neural networks have distinguished TGF-β-induced EMT-like morphological states in individual MCF-7 breast cancer cells and associated these states with differences in drug sensitivity (Yang et al., 2025). A workflow combining nonfluorescent eosin-methylene blue suspension staining with pretrained convolutional neural networks distinguished red blood cells, leukocytes, and cancer cells with reported test accuracies of 95.34%–100% under the selected experimental conditions. The analysis was restricted to images with clearly defined cell borders (Al-Bashir et al., 2026).

Machine learning can additionally assist device design and operational control. Computational fluid dynamics combined with supervised learning has been used to predict particle trajectories and critical separation diameters in deterministic lateral displacement systems (Sarowar et al., 2026). In a standing surface acoustic wave system, surrogate modeling and multi-objective optimization have similarly been used to tune buffer flow rate, voltage, and acoustic pressure while balancing recovery, lateral displacement, and energy consumption. Reported optimized conditions produced 100% recovery in the investigated model system, although validation using patient-derived samples remains necessary (Kouhkord and Naserifar, 2025).

Within integrated workflows, AI can localize candidate cells, link imaging and AFM coordinates, guide cantilever targeting and region-of-interest selection, optimize operating conditions, and automate force curve quality control and mechanical phenotyping. AI-selected cells should remain candidate CTCs until tumor origin is confirmed by immunophenotypic, molecular, or genomic analysis. Robust deployment requires annotated datasets spanning patients, protocols, instruments, and clinical sites (Li et al., 2025), together with donor-level separation, disease controls, external cohorts, and prospective validation. Reporting should include recovery or sensitivity, purity or specificity, false-positive events per unit blood volume, throughput, viability, uncertainty, and decision thresholds.

3.7 Downstream compatibility and remaining challenges

CTC microfluidic platforms must balance recovery, purity, viability, throughput, and compatibility with subsequent analysis (Descamps et al., 2022; Anwar et al., 2024). The optimal balance depends on the analytical objective. Enumeration may prioritize recovery and standardized identification, whereas culture, sequencing, drug testing, and AFM require viable cells, minimal mechanical perturbation, and suitable recovery or attachment.

Cluster-oriented workflows must additionally preserve cell-cell junctions, companion cells, and 3D organization. Blood handling, anticoagulant selection, processing time, flow conditions, capture chemistry, release procedures, and attachment substrates can influence both recovery and the measured phenotype (Qiao et al., 2024; Saghezi et al., 2025). Label-free enrichment broadens phenotypic coverage while requiring independent confirmation of tumor identity. Affinity capture improves specificity while introducing marker selection and receptor-mediated attachment effects.

4 AFM characterization of CTC-relevant mechanical phenotypes

4.1 Operational modes for CTC characterization

AFM enables surface imaging and localized mechanical characterization of viable, hydrated cells at single-cell and subcellular scales (Persano et al., 2025). Measurements can resolve topography, apparent elasticity, adhesion, deformation, and time-dependent mechanical behavior that would be spatially averaged by population-level methods (Lekka et al., 2012). AFM has been applied to individual patient-derived CTCs, although the available evidence remains limited (Chen et al., 2013; Osmulski et al., 2014; Huang et al., 2016; Deliorman et al., 2020; Glia et al., 2020).

In conventional optical lever AFM, a laser reflects off the cantilever onto a position-sensitive photodetector, while a piezoelectric scanner precisely controls lateral and vertical movement between the probe and sample (Figures 4a,b). Deflection is converted into force following calibration of the optical lever sensitivity and cantilever spring constant. During force spectroscopy in liquid, the probe approaches, contacts, and indents the cell to generate a force-distance curve. The approach and retraction segments yield measurements of deformation, adhesion, and elasticity (Figures 4c,d). Spatial mapping reveals subcellular mechanical heterogeneity and nanoscale architecture. Liquid operation preserves cell hydration and viability, supporting analysis of CTCs immobilized on exposed, optically accessible substrates.

Contact imaging maintains continuous tip-surface interaction while the probe scans across the cell. This mode can resolve cellular shape, roughness, and membrane-associated surface features. It also generates lateral forces that may distort soft membranes, alter local mechanical properties, or displace weakly attached cells (Muller and Dufrene, 2008). Affinity-captured CTCs therefore require conservative loading to limit the “push-aside” displacement or deformation during probing (Deliorman et al., 2020).

Dynamic imaging uses an oscillating cantilever that interacts intermittently with the cell surface. Reduced contact time and lower lateral loading make this mode better suited to viable cells and mechanically fragile structures (Garcia and Perez, 2002; Dufrene et al., 2017; Muller et al., 2021). Dynamic imaging may consequently be preferable for initial localization and topographical assessment of captured CTCs. Its proposed use for intact CTC clusters will require sufficient cantilever access and stable attachment without flattening or disrupting the multicellular architecture.

Although dynamic imaging reduces lateral loading, its application to intact patient-derived CTC clusters has not been demonstrated. Clinical AFM studies have examined individual CTCs, whereas cluster-related mechanical evidence is derived mainly from cultured multicellular models. Measurements of patient-derived clusters will require preservation of cluster composition and cell-cell junctions, AFM-accessible capture or transfer surfaces, stable positioning, probe geometry suited to irregular and multilayered topography, conservative loading, and measurement of local cluster thickness. These platform-specific considerations are summarized in Table 2.

Force spectroscopy adds quantitative mechanical information by recording cantilever responses during approach, indentation, and retraction (Krieg et al., 2019; Magazzù and Marcuello, 2023). The approach curve records deformation under applied load and supports estimation of apparent elastic or viscoelastic properties, while the retraction curve can reveal adhesion forces, membrane tether formation, discrete rupture events, and mechanical hysteresis. Repeating these measurements across predefined positions generates maps of surface topography, apparent elasticity, adhesion, and related interaction parameters, revealing mechanical heterogeneity across cellular and subcellular regions.

Force-volume mapping extends this principle by acquiring a complete approach-retraction curve at each position within a predefined measurement grid. Analysis of these spatially registered curve arrays generates corresponding maps of surface topography, apparent elasticity, adhesion, and other interaction parameters, revealing mechanical heterogeneity across cellular and subcellular regions. Measurements acquired at different loading rates or dwell times can further assess time-dependent and viscoelastic responses (Butt et al., 2005; Garcia, 2020; Muller et al., 2021).

Probe geometry determines the contact area, spatial resolution, and local strain imposed on the cell. Sharp probes provide high lateral resolution and are suited to localized membrane or subcellular measurements, although their concentrated loading increases sensitivity to contact point selection and indentation depth. Spherical or colloidal probes distribute force over a broader area and reduce local strain, providing measurements that better represent regional or whole-cell responses at the cost of spatial resolution (Dimitriadis et al., 2002). Indentation curves are analyzed using contact mechanics models appropriate to the probe geometry and indentation regime. Hertz- and Sneddon-type models are commonly used, although their assumptions and ranges of validity must be carefully considered (Kontomaris et al., 2025a).

Contact mechanics models only approximate the behavior of living cells. Hertz formulations for spherical or parabolic probes and Sneddon formulations for conical probes assume a known, rigid indenter and a homogeneous, isotropic, linearly elastic, non-adhesive half-space undergoing frictionless, small-strain contact on a locally flat surface. Spherical Hertz analysis further requires the indentation depth to be much smaller than the probe radius.

To limit substrate effects, indentation is commonly restricted to less than 10% of the local sample thickness. Finite-thickness corrections should be considered when deeper indentation is required (Dimitriadis et al., 2002; Krieg et al., 2019; Magazzù and Marcuello, 2023). Rate dependence or approach-retraction hysteresis indicates viscoelastic behavior and may require a time-dependent model. Moduli obtained from living cells should therefore be reported as apparent values.

For spherical contacts with measurable adhesion, the Derjaguin-Muller-Toporov model is used for relatively stiff, weakly adhesive contacts with small probe radii, whereas the Johnson-Kendall-Roberts model is suited to compliant, strongly adhesive contacts with larger probe radii. The Maugis model covers the intermediate regime (Cohen and Kalfon-Cohen, 2013). These models do not fully describe membrane tethers, discrete receptor-ligand ruptures, surface roughness, viscoelasticity, or active cellular remodeling. Probe geometry and radius, local sample thickness, indentation depth, assumed Poisson ratio, loading rate, fitting range, adhesion, and model choice should therefore be reported.

Substrate rigidity can bias AFM-derived apparent stiffness when the deformation field extends through the sample to the underlying support. A rigid substrate can cause stiffness overestimation in thin regions, whereas deformation of a substrate softer than the sample can lead to underestimation when force curves are analyzed using a standard Hertz model (Dimitriadis et al., 2002; Rheinlaender et al., 2020). This issue would be particularly relevant for intact CTC clusters because their irregular, multilayered geometry produces regional differences in thickness and substrate contact.

Thin peripheral regions may be more susceptible than thicker central regions. Measurements should therefore include local height data, restrict indentation to less than 10% of local thickness where possible, compare groups on identical substrates, and apply finite thickness or cell-substrate corrections when needed. Substrate stiffness, surface chemistry, cluster thickness, and measurement location should be reported, and the resulting moduli interpreted as apparent values.

4.2 Advanced AFM capabilities for cancer research

Advanced AFM modes extend conventional force spectroscopy through faster spatial mapping, multiparametric mechanical contrast, and time-resolved observation of cancer cell behavior. Representative applications are summarized in Figure 5. Quantitative imaging builds on the force-volume mapping by performing rapid, controlled approach-retraction cycles at each pixel under a defined maximum force, improving measurement speed while limiting lateral loading on soft biological samples (Proksch et al., 2016; Garcia, 2020; Garcia and Tejedor, 2025). This spatial resolution is important because malignant transformation and metastatic progression produce heterogeneous mechanical phenotypes that cannot be represented by a single elasticity value (Plodinec et al., 2012; Lekka, 2016).

Multifrequency AFM expands the mechanical contrasts available during the same experiment by analyzing cantilever responses at several frequencies, enabling topographical, elastic, viscous, and dissipative information to be obtained simultaneously (Cartagena and Raman, 2014; Cartagena-Rivera et al., 2015; Proksch et al., 2016). Multiplexed viscoelastic mapping across melanoma progression revealed softer, more fluid-like behavior near the nucleus and a stiffer, more viscous leading edge, demonstrating regional organization of mechanical states associated with migration (Figure 5a) (Parvini et al., 2025). Amplitude modulation-frequency modulation AFM similarly connected F-actin organization with spatial variations in elasticity and viscosity in living Huh-7 liver cancer cells. Acidic conditions promoted more highly oriented F-actin within lamellipodia and altered local viscoelasticity in association with increased migration (Chen et al., 2021). Under changing glucose conditions, hierarchical F-actin remodeling was accompanied by viscoelastic changes across multiple cell passages, with stronger responses in the cytoplasm than in the nucleus (Li et al., 2024).

High-speed AFM adds temporal resolution, enabling nanoscale biological dynamics to be visualized under near-physiological conditions (Ando et al., 2006; Lim and Wong, 2026). In studies of the EML4-ALK oncoprotein, high-speed AFM revealed reversible assembly of monomers, dimers, and trimers, with subsecond structural rearrangements in the v5 dimer and ALK inhibitor-induced compaction of the intrinsically disordered region (Figure 5b) (Han et al., 2026). In HCT116 colorectal cancer cells, high-speed AFM visualized changes within the nuclear pore inner channel during drug-induced cell death (Mohamed et al., 2017). These capabilities could be extended to CTCs, although direct application to patient-derived CTCs has not been reported.

AFM measurements can connect mechanical phenotype with mechanotransduction, the process through which cells convert physical cues into biochemical signaling (Sevgi et al., 2026). MG-63 osteosarcoma cells were softer and had less nuclear localization of the mechanosensitive transcriptional regulators YAP and TAZ than mesenchymal stem cells and normal human osteoblasts. Disruption of F-actin with cytochalasin D further reduced cell stiffness and nuclear YAP/TAZ localization, indicating that cytoskeletal organization and tension contribute to YAP/TAZ signaling (Wang et al., 2016). In lung cancer cells and xenograft models, ionizing radiation induced lncRNA CRYBG3, disrupted F-actin organization, reduced stiffness, and promoted cytoplasmic retention and inactivation of YAP/TAZ. These changes were accompanied by reduced proliferation, invasion, and metastatic activity (Zheng et al., 2022). A 3D osteosarcoma scaffold model showed a different stiffness response, with increased nuclear YAP/TAZ localization as scaffold stiffness decreased. This finding highlights the influence of dimensionality, cell spreading, scaffold architecture, and cellular context on mechanotransduction (Molina et al., 2019).

AFM can also characterize individual extracellular vesicles, including exosomes, extending mechanophenotyping from cells to nanoscale analytes relevant to liquid biopsy (Sorkin et al., 2018; Yurtsever et al., 2021). Nanoindentation and force mapping can quantify vesicle stiffness, deformation, and membrane mechanics while resolving size, morphology, and surface-associated nanodomains. These measurements depend on vesicle size and composition, adsorption to the substrate, probe geometry, loading conditions, and the mechanical model used for analysis (Vorselen et al., 2018; Ridolfi et al., 2020). In cultured non-small cell lung cancer models, A549-derived extracellular vesicles were stiffer than vesicles from PC9 and gefitinib-resistant PC9/GR cells, and a deep learning model classified A549-derived vesicles with 96% accuracy (Park et al., 2025). Osteosarcoma-derived exosomes similarly exhibited cell-of-origin-dependent mechanical properties (Yurtsever et al., 2021). This mechanistic interpretation remains based on cultured cell models. These findings suggest that AFM-derived vesicle mechanics could complement molecular liquid biopsy analysis, although clinical application requires standardized vesicle preparation and validation in patient-derived samples.

4.3 Subsurface and nuclear mechanical mapping

Nuclear mechanics is central to cancer cell migration because the nucleus is often the largest and least deformable structure traversing confined tissue spaces. Its mechanical response depends on chromatin organization, the nuclear lamina, and forces transmitted by the surrounding cytoskeleton (Hobson et al., 2021). Actin remodeling and altered actomyosin contractility also modify cancer cell stiffness and deformability, thereby influencing migration and invasion (Suresh, 2007). Lamin A/C stabilizes the nucleus, and its abundance scales with tissue stiffness (Swift et al., 2013).

Measuring nuclear mechanics inside an intact cell remains challenging because surface indentation integrates contributions from the plasma membrane, cell cortex, cytoplasm, perinuclear cytoskeleton, and substrate (Guilluy et al., 2014). Nuclear isolation provides more direct mechanical access, although it changes the measured state (Deliorman et al., 2023b). Nanoendoscopy-AFM provides direct intracellular access by introducing a nanoneedle probe into living cells (Figure 5c) (Penedo et al., 2021; Ichikawa et al., 2023). In PC9 lung cancer cells, nuclear elasticity increased after serum depletion and decreased when TGF-β was added, paralleling changes in histone H4 lysine 20 trimethylation (Ichikawa et al., 2025). In HeLa cells, measurement protocols did not significantly alter cell division intervals, supporting minimally disruptive intracellular investigation under defined conditions (Hosain et al., 2025). Application of these approaches to patient-derived CTCs remains prospective and would require careful validation of cell viability and identity.

Subsurface AFM offers a nonpenetrating route for detecting intracellular structures through mechanical contrast. Elastic contrast primarily reflects near-surface cortical and actin-rich structures, whereas inelastic or dissipative contrast can reveal deeper structures such as nucleoli (Guerrero et al., 2019). In living MDA-MB-231 breast cancer cells, multi-harmonic AFM using 10–15 µm high-aspect ratio carbon tips was combined with spinning-disk confocal microscopy and computational modeling. The long tips reduced hydrodynamic artifacts and enabled actin-rich structures and nucleoli to be mapped with a spatial resolution of 100–200 nm (Efremov et al., 2022).

Nuclear and subsurface signals must be interpreted in the context of cytoskeletal force transmission. Actin, microtubules, and vimentin form mechanically coupled networks, allowing perturbation of one system to reorganize the others. In T24 and 5637 bladder cancer cells, low-dose docetaxel altered microtubules together with actin and vimentin, producing cell line-dependent changes in stiffness, nuclear morphology, and migration (Zemla et al., 2025). These cytoskeletal contributions highlight the value of resolving local nuclear and subsurface mechanical changes following genetic, biochemical, or pharmacological perturbation (Li, 2023b; Marcuello et al., 2025). Such changes should be interpreted alongside cytoskeletal organization and treatment state.

4.4 AFM bioprobes for molecular recognition and single-cell manipulation

AFM bioprobes extend cancer cell mechanophenotyping into molecular recognition and force-controlled single-cell manipulation. Functionalizing the probe with antibodies, peptides, lectins, viral particles, or other recognition molecules enables specific receptor-ligand and glycan-mediated interactions to be quantified at single-cell and single-molecule scales (Dufrene et al., 2017). These measurements complement stiffness, adhesion, and viscoelasticity by connecting cell surface composition with mechanical phenotype (Garcia and Tejedor, 2025).

Molecular recognition AFM uses tips functionalized to localize specific binding sites while recording surface topography. Simultaneous topography and recognition imaging detects binding-induced changes in the cantilever oscillation signal, whereas AFM records an approach-retraction curve at each pixel to map binding probability, rupture force, and work of detachment (Hinterdorfer and Dufrene, 2006). Flexible linkers, such as polyethylene glycol, improve ligand accessibility and reduce nonspecific interactions (Wildling et al., 2011).

Reliable interpretation requires calibrated probes, controlled loading rate and contact time, minimal indentation, and specificity controls using receptor blocking, nonbinding ligands, or unfunctionalized probes. Loading rate dependence and multivalent rupture events must also be distinguished from single-bond interactions. In CTCs, these methods could complement mechanical mapping by revealing the spatial distribution and accessibility of receptors, glycans, and adhesion molecules associated with metastatic behavior.

Functionalized probes can resolve heterogeneity in cancer-associated receptors and glycans (Table 3). Nanoparticle-modified cantilevers coated with streptavidin were used to measure multipoint binding to biotinylated antibodies targeting HER2 and HER3 in paraffin-embedded sections of SKOV3, AU565, and MCF-7 cells. The resulting adhesion force distributions distinguished cell lines with different receptor-expression levels and enabled antigen densities within regions smaller than 30 nm to be inferred (Kitamura et al., 2025).

TABLE 3. AFM bioprobe and fluidic probe strategies relevant to CTC biomechanics research.

Evidence categories identify the experimental systems evaluated in each study and do not imply clinical validation.

Lectin-functionalized probes similarly quantify carbohydrate-mediated adhesion Figure 5d, top panel). Glycosylation influences cancer cell adhesion, signaling, and metastatic behavior (Zhang et al., 2010). Lectin-coated colloidal probes applied to LNCaP and PC3 prostate cancer cells revealed distinct adhesion force and work of detachment distributions. Measurements of a patient-derived CTC further demonstrated heterogeneous membrane-associated adhesion patterns (Glia et al., 2020).

Viral particles provide another class of recognition probe (Figure 5d, bottom panel). AFM tips functionalized with reovirus virions revealed serotype-dependent binding to cell surface glycans in malignant and non-malignant breast epithelial cells. T3 reovirus showed greater binding to malignant cells, consistent with their increased expression of α-sialylated glycans (Mohammed et al., 2021). Notably, at the single-molecule scale, unfolding force fingerprints combined with statistical clustering and proteomics have enabled abundant proteins to be identified in isolated native plasma membranes. This approach has not been established as a cancer assay, yet it demonstrates how AFM measurements can connect mechanical fingerprints with molecular identity (Galvanetto et al., 2022).

Fluidic force microscopy (FluidFM) extends AFM from molecular recognition to controlled delivery, aspiration, and extraction (Table 3). Microchannels integrated within the cantilever combine force-sensitive positioning with pressure-driven fluid manipulation, enabling localized injection, cytoplasmic sampling, organelle transfer, and single-cell force spectroscopy (Meister et al., 2009; Li et al., 2022). Recent cancer-focused applications include characterization of MDA-MB-231 breast cancer cells with altered connexin 43 expression, where Cx43 downregulation produced a softer phenotype and Cx43 upregulation markedly increased cell-cell adhesion (Habli et al., 2024).

FluidFM-assisted transplantation of cancer cell mitochondria into fibroblasts induced cancer-associated fibroblast features (Cangkrama et al., 2025). The Live-seq method extracts small cytoplasmic biopsies for transcriptomic analysis while preserving cell viability, enabling sequential profiling of the same cells before and after stimulation (Figure 5e) (Chen et al., 2022). Although these capabilities have not yet been applied to patient-derived CTCs, they offer a route to connect mechanical measurements with molecular state in individually tracked cells.

Additive manufacturing enables customized AFM probe geometries that are difficult to achieve using conventional fabrication methods (Göring et al., 2016; Alsharif et al., 2018; Glia et al., 2022). The 3D-printed FluidTIP integrates microfluidic structures with commercial AFM cantilevers and supports high-resolution imaging, localized fluid delivery, and cell manipulation (Glia et al., 2023). Its use for targeted CTC manipulation, retrieval, or localized mechanochemical perturbation has yet to be demonstrated and will require validation with rare patient-derived cells.

4.5 Therapy-associated mechanical phenotypes and drug resistance

AFM is increasingly used to characterize how cancer cells respond mechanically to therapy. Chemotherapy and ionizing radiation can remodel the cytoskeleton, focal adhesions, membrane organization, and viscoelastic behavior, producing measurable changes in apparent stiffness, adhesion, and mechanical heterogeneity (Kontomaris et al., 2025b; Persano et al., 2025). These mechanical signatures provide functional insight into treatment response, drug-induced mechanical adaptation, and resistance-associated phenotypes.

Mechanical responses depend on the cancer model, treatment, exposure duration, and measurement parameter. Following low-dose docetaxel treatment, 5637 bladder cancer cells progressively stiffened over 72 h, reaching an increase of up to 41%. T24 cells showed smaller and parameter-dependent changes that included softening and stiffening (Zemla et al., 2025). In lung cancer models, ionizing radiation reduced tumor stiffness through CRYBG3-associated disruption of F-actin and nuclear exclusion of YAP/TAZ, linking treatment-induced softening with altered mechanotransduction and metastatic behavior (Zheng et al., 2022). These contrasting responses indicate that a decrease or increase in stiffness cannot serve as a universal marker of therapeutic response.

Drug resistance can also be accompanied by mechanical reinforcement and cytoskeletal remodeling. Doxorubicin-resistant MCF-7/ADR breast cancer cells were stiffer and less deformable than sensitive MCF-7 cells and exhibited increased motility and vinculin expression (Park, 2018). Further analysis revealed increased actin stress fibers and focal adhesion kinase activity in MCF-7/ADR cells.

Pharmacological inhibition of focal adhesion kinase increased treatment sensitivity, connecting the resistance-associated mechanical phenotype with a potentially actionable pathway (Choi and Park, 2022). In ovarian cancer, cisplatin-resistant SKOV3-CisR and OVCAR5-CisR cells were stiffer and displayed more organized actin stress fibers than their sensitive counterparts (Sharma et al., 2012). A2780cis cells similarly exhibited increased stiffness and a bimodal modulus distribution, suggesting mechanically distinct subpopulations within the resistant population (Seo et al., 2015).

AFM has been evaluated as a marker of treatment response across eight epithelial ovarian cancer cell lines. Treatment with 2c, an experimental anticancer compound with F-actin-depolymerizing activity, produced an overall decrease in elastic modulus after 7 h. Representative baseline and post-treatment distributions for TYKNU, OVCAR4, and HEY cells are shown in Figure 5f. Cell lines with greater resistance and bimodal stiffness distributions generally exhibited smaller mechanical changes, while baseline stiffness was positively associated with the 2c half-maximal inhibitory concentration. The association was strongest for the stiffer subpopulation of cell lines with bimodal distributions (Tierno et al., 2023). Together, these findings show that treatment response and resistance may involve stiffening, softening, altered adhesion, or broader and multimodal mechanical distributions. Distribution-level and time-resolved analysis may therefore provide more information than a single mean local elasticity. Current evidence remains dominated by cell line and experimental tumor models (Krawczyk-Woloszyn et al., 2024).

The representative treatment- and resistance-associated mechanical signatures summarized in Table 4 derive from established cancer cell lines and preclinical experimental models. Treatment-induced mechanical changes have not yet been evaluated directly in patient-derived CTCs. Translation to clinical treatment monitoring will therefore require longitudinal analysis of patient-derived samples, standardized acquisition and analysis protocols, and prospective correlation with therapeutic outcomes (Schillers et al., 2017).

TABLE 4. Representative AFM evidence linking cancer cell biomechanics to treatment response and drug resistance.

Summarized studies are preclinical and do not directly evaluate patient-derived CTCs. Apparent elasticity depends on probe geometry, loading conditions, analysis model, and cellular state and should therefore not be compared directly across studies.

5 Integrating AFM and microfluidics for CTC analysis

5.1 Direct analysis of patient-derived CTCs

Despite their complementary capabilities, AFM and microfluidics have rarely been combined within a coordinated workflow for analyzing patient-derived CTCs. The clearest direct demonstration is the AFM-Chip, a reversibly assembled microfluidic device that captures prostate CTCs on an antibody-functionalized glass substrate (Figure 6a). Following capture and immunostaining, the polydimethylsiloxane layer is removed, leaving the cells on the same substrate and directly accessible to AFM. Proof-of-principle measurements were performed on five CTCs from two patients: three CTCs from one patient with localized prostate cancer and two from one patient with metastatic disease. The measured metastatic CTCs were softer and more deformable. Mean probe-cell adhesion forces were similar between the two cell sets, whereas multiple adhesion events were less frequent among the metastatic CTCs (Deliorman et al., 2020). A subsequent protocol described device fabrication, multiplexed affinity capture, and AFM characterization procedures in greater detail (Deliorman et al., 2022).

5.2 Automated mechanophenotyping in model rare-cell mixtures

Other studies have coupled label-free microfluidic separation with automated downstream AFM measurements. Contraction-expansion microchannels were used to separate model tumor cells from other blood cells, followed by deep learning-based optical recognition and automated indentation on a separate substrate where cells were transferred (Figure 6b). Representative force curves demonstrated autonomous targeting and calculation of cellular elasticity. The workflow was evaluated using microspheres, mixtures of cancer cell lines, and model samples containing cancer cells within a background of blood cells, establishing technical feasibility without patient-derived CTC validation (Qi et al., 2025).

A related high-throughput workflow combined autonomous optical recognition, automated stage movement, sequential indentation, and microfluidic sorting to characterize mechanically heterogeneous cell mixtures (Figure 6c). In spiked blood experiments, MCF-7 breast cancer cells were mixed with HaCaT immortalized keratinocytes at HaCaT input ratios of 1:1, 10:1, and 50:1. Following microfluidic enrichment, 200 cells per condition were analyzed by AFM, with fluorescence prestaining used to validate cell identity. The resulting bimodal elasticity distributions changed with mixture composition, demonstrating population-sensitive mechanical profiling (Xiao et al., 2025). These experiments should therefore be interpreted as model cell studies demonstrating automated mechanical profiling, not as clinical identification of CTCs.

5.3 Enabling models of mechanical conditioning

AFM-compatible microfluidic models demonstrate that prior physical exposure can alter the mechanical phenotype subsequently measured. Increased flow and substrate stiffness produced persistent stiffening in MCF-10A epithelial cells and C2C12 myoblasts after transfer to static culture, providing a model of mechanical memory that requires validation in cancer cells and patient-derived CTCs (Figure 6d) (Meng et al., 2025). Flow conditioning reduced adhesion between gastric cancer cell probes and mesothelial cells, showing that prior fluid exposure can modify AFM-derived cell-cell adhesion measurements (Figure 6e) (Wei et al., 2024). This model did not involve vascular endothelial cells or patient-derived CTCs.

Following passage through five or ten periodic constrictions, recovered and expanded MHCC-97L populations were about 15% and 30% softer, respectively, than non-migrating controls (Figure 6f) (Ma et al., 2018). Because AFM measurements were performed after transit selection, recovery, and expansion, the study could not distinguish confinement-induced softening from preferential passage of pre-existing softer cells. Collectively, these studies show that microfluidic handling forms part of the CTC measurement workflow because shear, substrate stiffness, adhesion conditioning, and confinement can alter or select the mechanical phenotypes subsequently measured by AFM. With the preceding platforms, they illustrate complementary integration strategies, including same substrate analysis, post-sorting indentation, automated population profiling, controlled mechanical conditioning, cell-cell adhesion measurements, and post-confinement phenotyping (Table 5).

TABLE 5. Representative AFM-microfluidic integration strategies and enabling architectures.

The AFM-Chip study included patient-derived CTCs. The remaining studies demonstrate enabling workflow components in model systems. Mechanically or optically selected events require independent confirmation of tumor origin.

6 Clinical translation and future directions

The evidence supporting multiscale CTC mechanobiology is distributed unevenly across microfluidics, AFM, and their integration. Microfluidic technologies have progressed from cultured cell and spiked sample experiments to patient sample studies, standardized enrichment systems, and clinically oriented workflows. AFM evidence remains concentrated in cancer cell models, tissues, and relatively small studies of individual patient-derived CTCs. Integrated AFM-microfluidic systems are at an earlier stage and have primarily established technical feasibility. Translation should therefore be evaluated according to an evidence hierarchy extending from analytical performance to biological validation, clinical validation, and demonstrated clinical utility.

6.1 Microfluidic CTC platforms: Clinical progress and limitations

Clinical translation of microfluidic CTC platforms should be considered in relation to CellSearch. CellSearch remains an established, non-microfluidic clinical benchmark for EpCAM-based CTC enumeration in metastatic breast, colorectal, and prostate cancers (Allard et al., 2004; Cristofanilli et al., 2004; de Bono et al., 2008; Farace et al., 2011; Bidard et al., 2014). More recent studies reinforce its prognostic role. In a correlative prognostic analysis within the phase III SWOG S1216 trial, baseline CellSearch counts in 503 evaluable participants distinguished markedly different survival outcomes among men with metastatic hormone-sensitive prostate cancer and added prognostic information to established clinical factors (Goldkorn et al., 2024). The PREDICT Global Pooled Analysis of 4,436 patients with metastatic breast cancer further demonstrated that baseline and early serial CellSearch measurements provided prognostic information across hormone-receptor/HER2 subtypes and treatment types (Janni et al., 2025).

Several microfluidic platforms have progressed to evaluation in patient cohorts. The Cellsway platform combines label-free enrichment with downstream immunophenotypic identification. In a study of 20 patients with localized prostate cancer and 10 healthy controls, CTCs were detected in 85% of patient samples, with a median of two CTCs per 7.5 mL of blood, while no CTCs were detected in the controls (Kilercik et al., 2024). A subsequent study evaluated an optimized Cellsway workflow using 51 patients with stage IV non-small cell lung cancer and 20 healthy controls. CTCs were detected in 47% of patients, and the assay achieved 95% specificity at a threshold of one CTC per 7.5 mL (Ozgur et al., 2026). These studies demonstrate expanding patient sample evaluation across disease settings, although external validation and outcome-based studies remain necessary.

The FDA-cleared Parsortix PC1 system uses a microfluidic separation cassette to enrich and harvest cells according to size and deformability from the blood of patients with metastatic breast cancer. Its clearance applies to enrichment and recovery for subsequent user-validated assays, whereas the device alone does not identify, enumerate, or characterize CTCs (Templeman et al., 2023). Parsortix-based CTC enrichment was subsequently adapted to in-cassette cytokeratin staining, single-cell recovery, and genomic analysis. In an exploratory cohort of 12 patients with metastatic prostate cancer, CTCs were detected in 67% of patients, and selected cells were recovered for copy-number analysis (Loppke et al., 2024). The study supports workflow feasibility and downstream analytical compatibility without establishing clinical utility for prostate cancer management.

Larger patient studies demonstrate how microfluidic enrichment can progress beyond analytical recovery. The label-free Labyrinth system was evaluated in 167 participants undergoing low dose computed tomography assessment and 31 healthy volunteers. At a threshold of one CTC per mL, sensitivity and specificity were 84.4% and 82.4%, respectively in the positive cohort. Specificity was 85.9% when all patients with lung nodules were considered (Jin et al., 2024). In a study of 32 patients with mediastinal neuroblastoma and seven healthy children, CTC counts differed by disease stage, risk group, metastatic status, and clinical response. These findings suggest preliminary associations, while further validation is needed to assess diagnostic or predictive performance (Wang et al., 2025).

The clinical value of microfluidics may ultimately depend on the analysis it enables after enrichment. Automated microfluidic immunomagnetic enrichment was used to collect 273 CTC samples from 117 patients with metastatic prostate cancer. Of these, 146 samples from 70 patients met the high tumor purity criterion for transcriptional analysis. RNA sequencing identified four phenotypes, with luminal B-like and neuroendocrine phenotypes associated with shorter overall survival from initial CTC sampling. In a prospective substudy of 37 patients receiving radioligand therapy, a pretreatment luminal B-like phenotype was associated with earlier progression (Sharifi et al., 2025). This study provides an important example of outcome-linked molecular characterization enabled by microfluidic CTC enrichment. It also establishes a relevant evidentiary benchmark for evaluating whether mechanical phenotyping can contribute additional clinical information.

Persistent barriers include variable performance across laboratories, limited adaptability to heterogeneous patient samples, incomplete agreement on CTC identification criteria, and insufficiently defined clinical scenarios (Li et al., 2026). Recovery and positivity rates obtained using different blood volumes, preprocessing methods, tumor types, disease stages, marker panels, and confirmation criteria cannot be compared directly. Clinical studies should therefore report blood volume, processing time, recovery, purity, viability, throughput, false-positive events per unit volume, and the criteria used to establish tumor identity.

6.2 Clinical potential of AFM mechanophenotyping

AFM has been used to investigate cancer cell mechanics for more than 2 decades (Lekka et al., 1999), yet it has not entered routine cancer diagnostics (Stylianou, 2026). Major translational barriers include low throughput and technically demanding measurements, while interlaboratory comparisons are affected by differences in calibration, acquisition, and analysis procedures (Schillers et al., 2017; Wang et al., 2026).

The clinical progress of CellSearch and emerging microfluidic platforms establishes an important translational benchmark for AFM: mechanical measurements must provide reproducible information that adds value beyond CTC enumeration and molecular profiling. Enumeration quantifies CTC abundance, while molecular assays identify mutations, transcriptional states, and potential therapeutic targets. AFM adds a functional dimension through measurements of local elasticity, adhesion, viscoelasticity, nanoscale surface architecture, and receptor-specific interactions (Cartagena and Raman, 2014; Dufrene et al., 2017; Mohammed et al., 2021). These measurements provide functional readouts influenced by cytoskeletal organization, membrane properties, and cell surface interactions, although their molecular determinants require immunophenotypic or genomic analysis confirmation (Sharma et al., 2012; Mohammed et al., 2021).

Initial clinical validation should determine whether baseline AFM measurements or longitudinal mechanical changes are associated with progression-free survival, metastatic progression, treatment response, or the emergence of resistant cellular subpopulations. Such associations would establish clinical validity. Demonstrating clinical utility would require evidence that AFM-derived measurements improve patient stratification, treatment selection, or therapeutic monitoring beyond established CTC counts and molecular profiles, and ultimately that their use contributes to better-informed clinical decisions or patient outcomes.

AFM studies of patient-derived CTCs remain exploratory and have focused on individual cells (Chen et al., 2013; Osmulski et al., 2014; Huang et al., 2016; Deliorman et al., 2020; Glia et al., 2020), with no identified study directly characterizing intact clusters. This translational gap is particularly significant because CTC clusters demonstrate markedly higher metastatic seeding efficiency in preclinical models (Aceto et al., 2014), and their mechanical behavior may differ fundamentally from that of individual CTCs.

Extending AFM to molecularly confirmed patient-derived CTC clusters is therefore a key next step toward defining their mechanical phenotype and functional relevance. Several strategies could facilitate this: (1) adapting open-space microfluidic platforms (e.g., HB-MFP) for cluster capture on removable substrates that permit AFM access; (2) developing gentle cluster release and transfer protocols that preserve intercellular junctions; (3) optimizing AFM imaging modes (e.g., dynamic imaging) to minimize lateral forces on fragile multicellular structures; (4) implementing local thickness measurements to correct for substrate artifacts; and (5) validating these approaches in model cluster systems before clinical application.

6.3 Standardization and clinical validation roadmap

Standardization must encompass the entire analytical sequence. Microfluidic variables include blood collection, anticoagulant, processing time, preprocessing, flow conditions, capture chemistry, release, and attachment. AFM measurements additionally depend on probe geometry, spring constant and optical lever calibration, loading rate, indentation depth, contact point detection, fitting assumptions, and substrate properties.

The standardized nanomechanical AFM procedure provides a foundation for reducing calibration-related variability and improving interlaboratory comparability of elasticity measurements (Schillers et al., 2017). Clinical implementation will require a streamlined extension that also addresses CTC enrichment, identification, immobilization, force curve quality control, and inter-instrument reproducibility. Because curves obtained from the same cell and cells obtained from the same patient are not statistically independent, analysis should preserve the hierarchy of patients, experiments, cells, and force curves. Additional curves cannot substitute for independent biological replication across patients (Lazic et al., 2018).

A translational roadmap can therefore be organized into five validation stages supported by an AI layer (Figure 7). We refer to this framework as the CTC Mechanobiology Assessment Pathway (CTC-MAP).

• Analytical validation: Establish recovery, sensitivity, repeatability, measurement uncertainty, inter-instrument agreement, and predefined quality control criteria for microfluidic enrichment and AFM analysis.
• Biological validation: Compare molecularly confirmed CTCs with leukocytes, healthy donor samples, and relevant disease controls, while determining how enrichment, attachment, and measurement conditions influence mechanical phenotype.
• Clinical validation: Evaluate predefined diagnostic, prognostic, or predictive endpoints through blinded analyses, independent cohorts, and prospective multisite studies, including whether AFM-derived measurements add information beyond established clinical variables.
• Workflow validation: Demonstrate traceable coordination of CTC enrichment, tumor identity confirmation, AFM measurement, selective cell recovery, and downstream molecular analysis.
• Clinical utility: Establish that AFM-derived measurements provide reproducible and actionable information that improves clinical decision making beyond CTC enumeration, imaging, molecular profiling, and existing clinical variables.

The AI support layer can assist all five stages through optical cell localization and targeting, cell tracking and measurement prioritization, contact point standardization, force curve quality control and outlier detection, mechanical map analysis, and assessment of data and model generalizability. AI-classified cells should remain designated as candidate CTCs until tumor identity is independently confirmed. Independent validation also remains essential.

The likely clinical role of integrated AFM-microfluidic platforms will be complementary to established liquid biopsy methods. Their distinctive contribution is the ability to connect cellular identity and molecular state with functional mechanical behavior. Progression from a research platform to a clinical tool will depend on demonstrating that this additional information improves clinically relevant assessment and decision making. AI can accelerate this translation by automating labor-intensive steps, improving reproducibility, and enabling analysis of larger datasets. Its translational use should remain supported by independent confirmation of tumor identity and rigorous clinical validation.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Some schematics were generated using a generative AI tool. In each case, this was disclosed in the corresponding figure caption.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Xiangchun Xuan, Clemson University, United States

Reviewed by: Carlos Marcuello, Instituto de Nanociencia y Materiales de Aragón (INMA), Spain

Zhijian Liu, Dalian Maritime University, China