Abstract

   Lung cancer metastasis is a leading cause of cancer‐related mortality,
   necessitating innovative approaches for early prediction and
   personalized clinical management. A novel strategy is present to
   predict lung cancer metastasis risk by combining single‐cell metabolic
   profiling of circulating tumor cells (CTCs) with a self‐developed CTC
   sorting and capture platform, enabling high‐efficiency, high‐viability
   CTC isolation from blood. Using nanoelectrospray ionization‐atmospheric
   pressure chemical ionization mass spectrometry, single‐cell metabolomic
   profiling on 301 CTCs derived from patients and animal models are
   performed. 390 unique metabolites are identified and discovered
   distinct metabolic signatures associated with different metastatic
   potentials (brain and bone). Based on these metabolic profiles, a
   classification model that categorizes CTCs into subgroups with distinct
   metastatic risks are constructed. The model outperformed traditional
   clinical indicators and total CTC counts, achieving AUCs of 0.74 (brain
   metastasis) and 0.92 (bone metastasis). Prospective validation
   confirmed its metabolite‐based classification accuracy for one‐year
   metastasis risk prediction. This study highlights the potential of
   single‐cell metabolomics to uncover novel therapeutic targets and
   prognostic markers, advancing liquid biopsy from quantitative counting
   to qualitative analysis. The approach represents a significant
   advancement in precision medicine for lung cancer management, offering
   a personalized strategy for predicting metastasis risk and guiding
   clinical treatment.

   Keywords: circulating tumor cells, lung cancer metastasis, metabolic
   profiling, personalized clinical management, single‐cell analysis
     __________________________________________________________________

   A self‐developed CTC sorting and capture platform enabled single‐cell
   metabolomics of patient/animal CTCs, revealing brain/bone
   metastasis‐specific metabolic signatures. A classification model
   developed from these signatures predicted lung cancer metastasis risk,
   outperforming traditional markers and total CTC counts (AUCs:89
   brain = 0.74; bone = 0.92). Prospective validation confirmed its 1‐year
   prediction accuracy.

   graphic file with name ADVS-12-e08878-g004.jpg

1. Introduction

   Lung cancer is a major global health burden, causing a large number of
   cancer‐related deaths each year.^[ [52]^1 ^] The high metastatic
   tendency of lung cancer is a crucial factor influencing its prognosis,
   as distant metastasis is the main obstacle to long‐term survival.^[
   [53]^2 ^] Despite advancements in imaging and histopathological
   diagnostics, early detection of metastatic propensity remains
   challenging. Current methods such as positron emission tomography (PET)
   and computed tomography (CT) lack the sensitivity to predict metastasis
   before clinical symptoms appear. This highlights the urgent need for
   noninvasive biomarkers to identify high‐risk patients and guiding
   personized treatments.^[ [54]^3 ^]

   Circulating tumor cells (CTCs), which detach from primary tumors and
   circulate in the bloodstream, are key drivers of metastasis and hold
   promise for liquid biopsy.^[ [55]^4 ^] The presence and count of CTCs
   have been clinically linked to poor prognosis^[ [56]^5 , [57]^6 ^] and
   higher metastasis risk^[ [58]^7 ^] in various cancers, particularly in
   lung cancer.^[ [59]^8 ^] However, its clinical utility is limited by
   CTC heterogeneity and the inability to distinguish subpopulations with
   different metastatic potentials.^[ [60]^9 ^] Conventional CTC detection
   methods like epithelial cell adhesion molecules (EpCAM)‐based
   enrichment, preferentially capture epithelial‐like CTCs while
   neglecting mesenchymal phenotypes,^[ [61]^10 ^] which are vital for
   metastasis.^[ [62]^11 ^] Moreover, existing platform often compromise
   cell viability during enrichment,^[ [63]^12 , [64]^13 ^] limiting
   downstream functional and molecular analyses.^[ [65]^14 ^] These
   challenges emphasize the need for technologies that ensure both high
   capture efficiency and cellular integrity.^[ [66]^15 , [67]^16 ^]

   Metabolism plays a pivotal role in cancer progression and metastasis,^[
   [68]^17 , [69]^18 ^] with metabolic reprogramming being a hallmark of
   metastatic cells.^[ [70]^19 , [71]^20 ^] During metastasis, CTCs
   undergo metabolic alterations that enable them to survive in the
   circulation and eventually colonize distant organs.^[ [72]^21 , [73]^22
   ^] This metabolic plasticity is a key factor underlying the
   differential metastatic potential observed among CTCs.^[ [74]^23 ^]
   Therefore, understanding the metabolic profile of individual CTCs could
   provide insights into their metastatic capabilities and offer a novel
   approach to stratify patients based on their risk of developing
   metastasis. Traditional bulk metabolomics masks cellular heterogeneity,
   failing to resolve metabolic variations in individual CTCs.^[ [75]^24
   ^] Single‐cell metabolomics (SCM) has recently emerged as a powerful
   tool to resolve this metabolic heterogeneity at a high resolution. For
   example, Abouleila et al. utilized live single‐cell mass spectrometry
   (LSC‐MS) combined with a microfluidics‐based enrichment technique to
   identify unique metabolic signatures that differentiate CTCs from
   lymphocytes and distinguish CTCs from different cancer types,
   highlighting the potential of CTC metabolomics for cancer diagnosis and
   treatment.^[ [76]^25 ^] Neumann et al. employed high‐resolution MS
   (HRMS) to analyze lipid heterogeneity across over 30 000 individual rat
   cerebellar cells, demonstrating the power of single‐cell MS to uncover
   lipid diversity and cellular heterogeneity without relying on bulk
   sampling.^[ [77]^26 ^] Zhang et al. employed integrated multiomic
   analysis to uncover amino acid dysregulation in hepatocellular
   carcinoma cells at single‐cell resolution, emphasizing how single‐cell
   technologies can decode tumor‐immune dynamics for personalized
   therapies.^[ [78]^27 ^] However, three critical challenges still remain
   in translating SCM into clinically applicable CTC analysis tools:
   First, current methods (e.g., laser capture microdissection^[ [79]^28
   ^] or microfluidic isolation^[ [80]^29 ^]) require cell fixation or
   lysis, which disrupts real‐time metabolic dynamics. Second, the rarity
   of CTCs (1‐100 cells/mL blood) and low metabolite abundance together
   with transient metabolite instability demand platforms with high
   single‐cell capture efficiency (>90%), analytical sensitivity (at the
   amol level) and clinical sample compatibility. Most critically, most
   SCM studies currently rely on cultural cell lines or mouse models,
   leaving significant gaps in the metabolomic databases of clinically
   derived CTCs. In particular, specific metabolic biomarkers for CTCs
   with different metastatic organs have not yet been systematically
   established.

   To address these gaps, we proposed three key questions: 1) Can a CTC
   enrichment platform achieve high capture efficiency while preserving
   metabolic activity for single‐cell analysis? 2) Does single‐cell
   metabolomic heterogeneity distinguish CTC subsets with distinct
   metastatic potentials? 3) Can metabolic signatures predict lung cancer
   metastasis risk more accurately than conventional biomarkers?

   Herein, we present a self‐developed CTC sorting and capture platform
   that integrates negative enrichment with a Parylene C micropore‐arrayed
   membrane, optimized for minimal cell damage (viability > 87%) and high
   recovery rates (>89%). Coupled with nanoelectrospray
   ionization‐atmospheric pressure chemical ionization (nanoESI‐APCI) MS,
   we performed single‐cell metabolomic profiling of CTCs from lung cancer
   patients and metastatic variant cell lines (bone metastasis [Bom] and
   brain metastasis [Brm]). Our platform's non‐destructive release
   mechanism, enabled by a polyvinyl alcohol (PVA)‐coated membrane,
   ensures cell viability and compatibility with downstream culture—a
   significant improvement over existing technologies.

   By analyzing 181 single cells, we identified 390 metabolites and
   uncovered pathway‐specific differences: Bom cells exhibited enriched
   arginine biosynthesis, while Brm cells showed elevated phenylalanine
   metabolism. These findings were translated into a metabolite‐based
   classification model, stratifying CTCs into three subgroups (CTC1,
   CTC2, and CTC3). Prospective clinical validation revealed CTC1 and CTC2
   as robust predictors of brain and bone metastasis, respectively,
   outperforming total CTC counts and traditional biomarkers (AUC:
   0.74–0.92). Our findings indicate that metabolic heterogeneity within
   individual CTCs can predict the metastatic potential of lung cancer,
   enabling more personalized clinical management. Moreover, the metabolic
   markers identified may serve as potential therapeutic targets^[ [81]^30
   ^] and enhance our understanding of cancer metastasis (Figure [82] 1
   ).^[ [83]^31 ^]

Figure 1.

   Figure 1
   [84]Open in a new tab

   Schematic workflow of predicting lung cancer distant metastasis using a
   self‐developed CTC sorting and capture platform coupled with
   single‐cell MS metabolomic analysis. First, individual CTCs were
   efficiently isolated from whole blood samples obtained from lung cancer
   patients and animal models. Next, single‐cell metabolomic profiling was
   performed using nanoelectrospray ionization‐atmospheric pressure
   chemical ionization (nanoESI‐APCI). Subsequently, metabolite screening
   was conducted in human lung cancer cell lines with distinct metastatic
   potentials, including Brm variant cells, Bom variant cells, and the
   PC9‐CTC cells, to identify differentially abundant metabolites.
   Finally, the identified metastasis‐associated metabolites were used to
   classify CTC subgroups. These clinically relevant CTC subgroups with
   metastatic potential were correlated with clinical data to develop a
   predictive classification system for lung cancer metastasis.

2. Results and Discussion

2.1. Design and Performance Evaluation of the CTC Sorting and Capture
Platform

   As described in the Experimental section, we developed a CTC sorting
   and capture platform using an integrated approach that combines
   negative enrichment and physical separation techniques. To evaluate the
   platform's sorting performance, healthy human peripheral blood samples
   were spiked with PC9 and NCI‐H1975 lung cancer cell lines as rare cell
   models and processed through the platform for sorting and capture (CTC
   sorting and capture workflow shown in Figure [85] 2A). The viability of
   the captured PC9 and NCI‐H1975 cells was evaluated using live/dead cell
   staining assays. The results showed that the viability of captured PC9
   and NCI‐H1975 cells was 81.1% ± 2.9% and 78.7% ± 3.0%, respectively
   (Figure [86]2B). Representative images of live/dead rare cells and
   background cells are shown in Figure [87]2C.

Figure 2.

   Figure 2
   [88]Open in a new tab

   Performance evaluation of the self‐developed CTC sorting and capture
   platform. A) Schematic illustration of individual CTC enrichment using
   the self‐developed platform. B) Cell viability of PC9 and NCI‐H1975
   cells enriched by the platform compared to control groups. Statistical
   analysis was performed using unpaired t‐tests (two‐tailed) for each
   cell line, with Šidák correction for multiple comparisons. C) Viability
   assessment of captured PC9 and NCI‐H1975 cells via propidium iodide
   (PI) staining (viable cells: green fluorescence; dead cells: red‐green
   overlay). D) Bright‐field and confocal fluorescence images of PC9 cells
   cultured in situ for 48 h post‐filtration, demonstrating platform
   compatibility with cell culture. E) Release efficiency of PC9 cells
   from a PVA‐coated micropore‐arrayed filtration membrane. F) Recovery
   rate of rare cells from 5 mL of peripheral blood using the platform.
   G) Fluorescence images of rare cells and leukocytes recovered from
   peripheral blood. B, E, F) Data presented as mean ± SD (n = 3).

   Previous studies have demonstrated that the edge‐to‐edge spacing of
   micropores is a critical factor affecting the viability of recovered
   cells.^[ [89]^32 ^] In this study, we used a Parylene C
   micropore‐arrayed filtration membrane with a pore size of 8 µm and pore
   spacing of 4 µm. The optical and scanning electron microscopy (SEM)
   images of this membrane are provided in Figure [90]S1 (Supporting
   Information). Compared to other microporous membranes, the smaller pore
   spacing in our design enhances cell viability post‐capture, thereby
   significantly improving the platform's overall performance.

   To further investigate the impact of the release process on cell
   viability, we evaluated the viability of cells after release from the
   platform. The results revealed that released cells maintained a
   viability of 87.2% ± 3.5% (Figure [91]S2, Supporting Information), with
   no significant difference compared to the control group. This confirms
   that the release operation does not adversely affect cell viability.
   Additionally, to validate the long‐term viability of captured tumor
   cells, PC9 cells were cultured in situ for 48 h postcapture and
   release. As shown in Figure [92]2D, the cells exhibited strong adhesion
   and spreading morphology, indicating robust cell viability over an
   extended period. The release efficiency of the platform was also tested
   by capturing and releasing PC9 cells suspended in PBS. The results
   showed a release efficiency of 97.2% ± 0.6% (Figure [93]2E), confirming
   the platform's high capability for cell recovery.

   To assess the platform's efficiency in recovering rare tumor cells from
   complex cellular backgrounds, fluorescently labeled PC9 or NCI‐H1975
   cells were spiked into 5 mL of healthy human peripheral blood at
   concentrations of 2 to 100 cells mL^−1. Postcapture cell counts
   revealed a recovery efficiency of 89.2% ± 1.3% when 100 rare cells per
   mL were spiked (Figure [94]2F). Representative images of captured
   fluorescently labeled cells are shown in Figure [95]2G.

   Collectively, the platform enables efficient isolation of rare CTCs
   from peripheral blood while supporting their in vitro culture, thereby
   providing a basis for downstream analyses such as drug screening and
   genomic sequencing. Additionally, its capability for in situ culture of
   captured CTCs offers critical insights into their molecular and
   functional characteristics, advancing tumor metastasis research. A key
   innovation of this platform is the high‐efficiency, nondestructive
   release mechanism for captured CTCs,^[ [96]^33 ^] which achieved by
   a PVA coated micropore‐arrayed filtration membrane. This design allows
   gentle cell release at 37 °C while prevents issues such as deformation
   or physical entrapment. This platform outperforms existing CTC
   enrichment methods,^[ [97]^34 , [98]^35 , [99]^36 , [100]^37 ^]
   especially under low‐abundance conditions (≤100 cells mL^−1 of blood).
   It enables rapid and sensitive enrichment of rare CTCs while
   maintaining high viability, fulfilling the essential requirements for
   liquid biopsy and multi‐omics analyses.

2.2. Isolate CTCs in Animal Models and Clinical Specimens

   To validate the technical utility of our self‐developed CTC sorting and
   capture platform, we established an orthotopic lung cancer xenograft
   model for CTC isolation. As illustrated in Figure [101] 3A,
   luciferase‐expressing PC9 cells were orthotopically injected into the
   left lung parenchyma of highly immunodeficient SRG rats (which have
   functionally deleted Rag2 and Il2rg genes on the Sprague‐Dawley
   background^[ [102]^38 ^]) to generate primary tumors. Tumor progression
   was longitudinally monitored via IVIS Spectrum imaging on days 5, 10,
   and 15 after the injection (Figure [103]3B), and tumor size was
   measured to validate luminescence intensity (Figure [104]S3, Supporting
   Information). After confirming lung tumor formation by imaging results,
   1 mL of peripheral blood was collected via jugular vein puncture on day
   16 for CTC isolation using self‐developed platform. Captured CTCs
   exhibited staining patterns of CK or EpCAM+/GFP+/Hoechst+, while
   remaining white blood cells (WBCs) displayed CK or EpCAM‐/GFP‐/Hoechst+
   characteristics (Figure [105]3C). Through in vitro culture, we
   successfully established a CTC cell line (PC9‐CTC) with stable
   proliferative capacity. At the CTC collection timepoint, no
   extrapulmonary metastases were observed in the animal models,
   indicating that this CTC cell line represents circulating tumor cells
   shed from the primary lesion prior to metastatic dissemination.

Figure 3.

   Figure 3
   [106]Open in a new tab

   Isolate CTCs from animal models and clinical samples. A) Schematic of
   PC9‐CTC cell line derivation from orthotopic lung cancer rat models. B)
   Bioluminescence imaging of lung cancer rat models. C) Fluorescence
   images of CTCs and WBCs isolated from the peripheral blood of lung
   cancer rat models. D) Representative fluorescence images of CTCs
   (CK/EpCAM+) and WBCs (CD45+) captured from the peripheral blood of lung
   cancer patients. E) Distribution of CTC counts in 5 mL peripheral blood
   samples from 31 non‐small‐cell lung cancer patients. Statistical
   analysis: Data E is presented as mean ± SD (LUAD, n = 23; LUSC, n = 8).
   Significance was determined by two‐tailed Student's t‐test.

   Further functional assays were conducted to evaluate the metastatic
   potential of PC9‐CTC cells. Transwell assays demonstrated that PC9‐CTC
   cells exhibited higher migration and invasion capacity than parental
   PC9 cells (Figure [107]S4A,B, Supporting Information). Additionally,
   5‐ethynyl‐2′‐deoxyuridine (EdU) assays revealed faster proliferation
   rates in PC9‐CTC cells than PC9 cells (Figure [108]S4C, Supporting
   Information). Further investigation of PC9‐CTC's epithelial‐mesenchymal
   transition (EMT) characteristics in the lung cancer organoid model
   uncovered a hybrid epithelial‐mesenchymal phenotype (Figure [109]S5,
   Supporting Information). This suggests that EMT‐mediated acquisition of
   mesenchymal‐like properties contributes to the enhanced migratory and
   invasive capacities of PC9‐CTC cells.^[ [110]^39 ^]

   After validating the platform's efficacy in animal models, we conducted
   preliminary clinical evaluations using peripheral blood samples from 31
   lung cancer patients (clinical information of the patients is provided
   in Table [111]S1, Supporting Information). Captured cells were
   immunostained with anti‐CK, anti‐EpCAM, and anti‐CD45 antibodies. CTCs
   were identified as CK or EpCAM+/CD45‐/Hoechst+, while WBCs were
   identified as CK or EpCAM‐/CD45+/Hoechst+ (Figure [112]3D). No CTCs
   were detected in negative controls from five healthy volunteers and
   five patients with smoking history and benign pulmonary nodules (all
   pathologically confirmed after resection), confirming the assay's
   specificity (Figure [113]S6, Supporting Information). CTC counts varied
   substantially among patients, with an average of 3.9 ± 3.5 CTCs per
   5 mL of whole blood (Figure [114]3E). No significant correlation was
   observed between CTC counts and pathological subtypes. These results
   highlight the platform's high sensitivity for CTC detection in both
   animal models and clinical specimens, demonstrating its potential for
   practical liquid biopsy applications.

   The PC9‐CTC cell line derived from lung cancer animal models exhibited
   enhanced invasive and migratory capacities compared to parental PC9
   cells, highlighting CTCs’ role in metastasis.^[ [115]^40 ^] This
   supports the platform's utility for in vitro CTC cultivation and
   functional analysis, although optimized culture conditions are still
   needed for clinical CTC studies. While the exclusive use of female rats
   in this initial phase ensured model standardization, it may limit the
   generalizability of our findings. Given the known influence of
   biological sex on cancer research outcomes, we will incorporate male
   SRG rats in ongoing experiments to systematically evaluate sex‐specific
   effects on lung cancer model establishment, tumor progression,
   treatment response, and underlying mechanisms. In patient samples, CTC
   identification mainly relied on CK and EpCAM markers, which may
   overlook mesenchymal‐like CTCs and underestimate CTC heterogeneity.^[
   [116]^41 ^] While our label‐free enrichment approach aims to capture
   diverse CTC phenotypes, epithelial‐biased identification criteria
   likely undercount actual CTC numbers. Future studies should develop
   broader CTC marker panels with improved specificity.

   Our CTC sorting and capture platform demonstrates superior sensitivity
   in both animal models and clinical specimens, fulfilling clinical needs
   for noninvasive tumor diagnosis and therapeutic monitoring. Downstream
   molecular characterization of CTCs could elucidate associations between
   CTC phenotypic profiles and tumor aggressiveness, potentially revealing
   novel prognostic biomarkers and therapeutic molecular targets. Beyond
   rigorous analytical performance, this platform exhibits significant
   potential for clinical translation, promising to advance liquid biopsy
   technologies and improve cancer patient care.

2.3. Single‐Cell Metabolomic Profiling Uncovers Metastasis‐Specific Metabolic
Heterogeneity in CTCs

   To investigate the metabolic reprogramming of CTCs and its association
   with organ‐specific metastatic potential, we performed single‐cell
   metabolomic MS analysis on three cell populations: PC9‐CTC cells
   (isolated from orthotopic lung cancer rat models), Brm cells, and Bom
   cells (in prior studies,^[ [117]^42 ^] we established Brm and Bom lung
   cancer cell lines through intracardiac injection of PC9 cells into
   immunodeficient mice). These three types of cells represent circulating
   tumor cell populations shed from the primary site but not yet causing
   metastasis (PC9‐CTC), as well as CTC groups with high potential for Brm
   and Bom. Metabolic screening of three human lung adenocarcinoma cell
   lines with distinct metastatic tropisms identified potentially brain‐
   or bone‐tropic metabolic signatures. A total of 181 single cells (58
   PC9‐CTC, 60 Bom, and 63 Brm cells) were analyzed using the
   nanoESI‐APCI‐HRMS platform, resulting in the identification of 390
   unique metabolites (Table [118]S2, Supporting Information).

   Pairwise partial least squares discriminant analysis (PLS‐DA) revealed
   distinct clustering patterns among Bom, Brm, and PC9‐CTC cells
   (Figure [119] 4A), with permutation tests confirming model robustness
   (Figure [120]S7, Supporting Information). Based on the variable
   importance in projection (VIP) values derived from the PLS‐DA models,
   an initial set of differential metabolites (VIP ≥ 1) was defined,
   followed by rigorous filtering via volcano plot analysis with criteria
   of fold change (FC) ≥ 1.5 and p ≤ 0.05 (Figure [121]4B). Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis
   was performed to identify the metabolic pathways associated with the
   differential metabolites (Figure [122]S8, Supporting Information). The
   analysis revealed significant differences in several key metabolic
   pathways among the cell types. For instance, the arginine and proline
   metabolic pathway was more prominent in Bom cells compared to Brm
   cells, indicating its potential role in bone metastasis.^[ [123]^43 ^]
   Differential metabolites were also observed in tyrosine metabolism
   between Bom and Brm cells, suggesting distinct metabolic adaptations in
   these metastatic variants. In the comparison between Bom and PC9‐CTC
   cells, the arginine biosynthesis pathway was found to be more
   prominent,^[ [124]^44 ^] and significant differences were observed in
   glycine, serine, and threonine metabolism.^[ [125]^45 ^] In the
   comparison between Brm and PC9‐CTC cells, phenylalanine, tyrosine, and
   tryptophan biosynthesis was more pronounced in Brm cells, revealing
   potential metabolic signatures associated with brain metastasis. A
   heatmap visualizing the differential metabolites corresponding to the
   identified KEGG pathways (Figure [126]4C) clearly illustrated the
   distinct metabolic profiles of Bom, Brm, and PC9‐CTC cells,
   highlighting the metabolic heterogeneity associated with different
   metastatic potentials.

Figure 4.

   Figure 4
   [127]Open in a new tab

   Metabolic profiles of CTCs and their association with metastatic
   potential. A) PLS‐DA plots of Bom/Brm, Bom/PC9‐CTC, and Brm/PC9‐CTC
   cell lines (Bom/Brm: R^2X (cum) = 0.534, R^2Y (cum) = 0.890, Q^2
   (cum) = 0.671; Bom/PC9‐CTC: R^2X (cum) = 0.598, R^2Y (cum) = 0.858, Q^2
   (cum) = 0.584; Brm/PC9‐CTC: R^2X (cum) = 0.739, R^2Y (cum) = 0.875, Q^2
   (cum) = 0.748). B) Volcano plots of selected metabolites (VIP ≥ 1) in
   Bom versus Brm, Bom versus PC9‐CTC, and Brm versus PC9‐CTC cell lines.
   Red and blue dots represent increased and decreased metabolites,
   respectively (FC ≥ 1.5, p < 0.05; significance determined by
   Mann‐Whitney U test). C) Heatmap of the relative intensities of
   differential metabolites in Bom, Brm, and PC9‐CTC cell lines.

   The distinct metabolic profiles observed in CTCs and metastatic
   variants provide valuable insights into the metabolic adaptations that
   drive metastasis. Our results suggest that targeting specific metabolic
   pathways could inhibit metastatic progression by disrupting the
   adaptive capabilities of tumor cells. Future studies should focus on
   validating these metabolic signatures in larger cohorts and exploring
   their potential as diagnostic biomarkers or therapeutic targets.

2.4. Metabolite‐Driven Classification Model Identifies CTC Subgroups with
Distinct Metastatic Risks

   To establish a clinically practicable framework for stratifying
   metastatic risk, we developed a metabolite‐based classification model
   utilizing the metabolic heterogeneity of CTCs. As described previously,
   metabolic reprogramming in CTCs reflect their adaptive capabilities and
   organotropism during metastasis.^[ [128]^46 ^] We hypothesized that
   intrinsic metabolic differences among CTC subpopulations could serve as
   discriminative markers to categorize CTCs into functionally distinct
   subgroups.

   To assess the clinical relevance of the identified metabolites, we
   collected blood samples from 30 patients with lung adenocarcinoma,
   including 10 with brain metastasis, 10 with bone metastasis, and 10
   with nonmetastatic lung cancer. Using our CTC sorting and capture
   platform, a total of 120 CTCs was isolated from these patients. The
   primary clinical characteristics and biomarker levels of the patients
   are summarized in Tables [129]S3 and [130]S4 (Supporting Information).
   In the training cohort, the number of CTCs detected per 5 mL of blood
   ranged from 2 to 11, with a frequency distribution of 2 CTCs (10%), 3
   CTCs (33.3%), 4 CTCs (30.0%), and ≥5 CTCs (26.7%). Notably, there was
   no significant difference in total CTC counts between the nonmetastasis
   group (median: 3; range: 2–5) and the metastasis group (median: 4;
   range: 2–11) (p = 0.1170).

   The isolated CTCs were aliquoted into 200 µL Eppendorf tubes containing
   1 µL of 150 mM ammonium formate solution and stored at −80 °C. Prior to
   MS analysis, samples were lyophilized under nitrogen flow,
   reconstituted in 1 µL of methanol containing deuterated phenylalanine
   (d5‐Phe, 1 µg mL^−1) as an internal standard and subjected to
   nanoESI‐APCI‐HRMS. The obtained raw MS data were normalized to internal
   standard intensity and aligned with reference m/z values of the 390
   previously identified metabolites. To resolve CTC heterogeneity, we
   applied t‐distributed stochastic neighbor embedding (t‐SNE), a
   nonlinear dimensionality reduction algorithm, to visualize metabolic
   similarities among CTCs. As shown in Figure [131] 5A, CTCs from
   different patient cohorts exhibited partial overlap but formed three
   tentative clusters, reflecting both interpatient variability and
   intrinsic metabolic divergence.

Figure 5.

   Figure 5
   [132]Open in a new tab

   The heterogeneity of CTCs and their potential utility in identifying
   metastatic risk in lung cancer patients. A) t‐SNE visualization of CTCs
   from patients. B) Silhouette analysis based on dimension reduction
   using t‐SNE. C) t‐SNE visualization of CTCs labeled by hierarchical
   clustering. D) Distributions of CTC subgroups in patients. E) Heatmap
   of CTC subgroup counts in patients. Statistical analysis: Data in D is
   presented as mean ± SD (n = 10). Significance was determined by one‐way
   Analysis of Variance.

   To objectively define CTC subgroups, we performed silhouette analysis
   on t‐SNE embeddings, which identified three optimal clusters
   (Figure [133]5B). Hierarchical clustering confirmed this partitioning,
   categorizing CTCs into three subgroups: CTC1 (n = 44), CTC2 (n = 42),
   and CTC3 (n = 34) (Figure [134]5C; Table [135]S5, Supporting
   Information). Subgroup distributions across patients revealed striking
   clinical correlations: CTC1 was predominantly enriched in brain
   metastasis patients, CTC2 was notably prevalent in bone metastasis
   patients, and CTC3 showed no significant association with metastatic
   status (Figure [136]5D; Table [137]S6, Supporting Information). These
   findings indicate that higher CTC1 and CTC2 counts are associated with
   brain and bone metastasis, respectively. In non‐metastatic patients,
   increased CTC1 levels may suggest a potential risk of brain metastasis,
   while elevated CTC2 levels may indicate a possible risk of bone
   metastasis. Conversely, predominant CTC3 counts are likely to imply a
   lower short‐term metastatic risk. Figure [138]5E displays the
   distribution of each CTC subgroup within individual patients,
   indicating that the metastasis‐associated CTC1 and CTC2 subgroups are
   indeed more broadly distributed in the corresponding metastatic
   populations. To validate these observations, Uniform Manifold
   Approximation and Projection (UMAP) was applied as an alternative
   dimensionality reduction method. UMAP revealed a CTC distribution
   pattern similar to that observed with t‐SNE. Subsequent silhouette
   analysis and hierarchical clustering of UMAP embeddings further
   confirmed that CTC subgroup distributions closely matched those derived
   from t‐SNE (Figure [139]S9, Supporting Information). These
   complementary analyses collectively reinforce the robustness and
   interpretation of our findings. The presence of CTC1 or CTC2 subgroups
   in patients with primary tumors could serve as an early warning of
   tumor metastasis, enabling timely adjustments to therapeutic
   strategies.

   These outcomes underscore the potential of CTC subgroup analysis as a
   valuable tool for predicting metastatic risk and guiding personalized
   treatment strategies in patients with lung cancer.

2.5. CTC Subgroup Abundance Predicts Organ‐Specific Metastasis Risk in Lung
Cancer

   Emerging evidence indicates that CTCs can be detected even in the early
   stages of lung cancer, and their clustering forms often possess a
   higher metastatic capacity compared to individual CTCs. Moreover, CTCs
   exhibit remarkable adaptability through diverse phenotypes, enabling
   them to facilitate metastasis and develop resistance to therapies.
   However, the specific subgroups of CTCs and their correlation with
   different metastatic risks in lung cancer have not been fully explored.
   This study aimed to investigate the association between CTC subgroups
   and lung cancer metastasis risk, as well as their correlations with
   clinical characteristics and conventional tumor biomarkers. The results
   demonstrated no significant associate between metastasis risk and
   factors such as sex, age, primary lung cancer status, or biomarkers
   including NSE, CEA, CA‐125, and CYFRA21‐1 (Figure [140] 6A). However,
   elevated CTC1 subgroup counts (p = 0.0416) and CTC2 subgroup counts
   (p < 0.0001) were strongly linked to the risks of brain and bone
   metastasis, respectively (Figure [141]6B). Receiver Operating
   Characteristic (ROC) curve analysis determined the optimal cutoff
   thresholds for CTC1 and CTC2 subgroups as 1.5 based on Youden's index
   (Figure [142]6C). Patients with CTC1 or CTC2 counts exceeding these
   thresholds exhibited a poorer prognosis and higher metastasis risk
   (Figure [143]6A). In terms of predictive performance, the area under
   the ROC curve (AUC) for CTC1 in predicting brain metastasis and CTC2 in
   predicting bone metastasis were 0.74 and 0.92, respectively. These
   values significantly outperformed the predictive performance of total
   CTC counts (Figure [144]6C). Univariate and multivariate logistic
   regression analyses further confirmed that CTC subgroup counts are
   independent predictors of lung cancer metastasis (Tables [145]S7
   and [146]S8, Supporting Information).

Figure 6.

   Figure 6
   [147]Open in a new tab

   Predictive potential of metastasis‐associated CTC subgroups for distant
   lung cancer metastasis. A) Correlation between patient metastasis
   status (distant metastasis vs non‐distant metastasis), clinical
   characteristics, biomarker levels (NSE, CEA, CA‐125, and CYFRA21‐1),
   and CTC subgroup counts (CTC1, CTC2, total CTCs) in the test cohort.
   B) Comparison of CTC1 and total CTC counts between brain
   metastasis‐positive and ‐negative groups, and CTC2/total CTC counts
   between bone metastasis‐positive and ‐negative groups in the training
   cohort. Box plots show medians (horizontal lines), 25th–75th
   percentiles (boxes), and ranges (whiskers). Statistical significance
   was determined by two‐tailed Student's t‐test. C) ROC analysis of CTC1,
   CTC2, and total CTC counts for predicting brain or bone metastasis in
   the training cohort. D) Distribution of CTC1, CTC2, and CTC3 counts in
   non‐distant metastasis patients. E) Changes in tumor biomarkers (NSE,
   CEA, CA‐125, and CYFRA21‐1) during follow‐up for five patients with a
   high risk of distant metastasis (P1, P2, P5, P6, and P9). The cut‐off
   values for each biomarker are indicated by red lines. F) Representative
   CT and ECT images of the patients (P1, P5, and P6) showing metastatic
   lesions (red arrows) detected during follow‐up.

   Additionally, a prospective study was conducted on 10 lung cancer
   patients without distant metastasis in the cohort. Basic clinical
   information is summarized in Table [148]S3 (Supporting Information).
   The median CTC count was 3/5 mL of blood (range: 2–5). Subgroup
   classification based on CTC metabolic phenotypes revealed that cases 1,
   2, and 5 had CTC1 counts above the threshold (indicating elevated brain
   metastasis risk), while cases 6 and 9 had elevated CTC2 counts
   (indicating increased bone metastasis risk) (Figure [149]6D). During
   the 1‐year follow‐up, cases 1 and 5 developed brain metastasis at 9 and
   11 months, respectively, and case 6 developed bone metastasis at 6
   months. The remaining 7 patients showed no signs of extrapulmonary
   metastasis (Figure [150]6F), validating the feasibility of
   metabolic‐based CTC subgroup classification for metastasis prediction.
   In contrast, conventional tumor biomarkers showed no significant
   prognostic correlations (Figure [151]6E).

   In clinical practice, CTCs, as a critical component of “liquid biopsy”,
   have demonstrated significant potential in the diagnosing, prognosing,
   and monitoring solid tumors.^[ [152]^47 ^] However, most studies focus
   only on the correlation between CTC counts and prognosis, ignoring the
   substantial heterogeneity of CTCs across patients and disease stages,^[
   [153]^48 ^] which limits their broader clinical utility. Single‐cell
   sequencing technologies offer unprecedented opportunities to
   comprehensively dissect CTC heterogeneity, elucidate molecular
   mechanisms of metastasis, and identify novel therapeutic strategies.^[
   [154]^12 , [155]^49 ^]

   Metabolic reprogramming is a key driver of tumor progression and a key
   contributor to intra‐ and intercellular heterogeneity.^[ [156]^50 ^]
   Thus, single‐cell metabolomic profiling of CTCs may uncover the link
   between their heterogeneity and metastatic propensity.^[ [157]^51 ^]
   This study employed single‐cell metabolomics to classify CTCs into
   distinct subgroups based on characteristic metabolites, revealing that
   CTC1 and CTC2 subgroup abundances closely correlated with patient
   prognosis. In prospective validation, metabolic phenotype‐based CTC
   classification outperformed total CTC counts and conventional
   biomarkers in predicting metastasis within 1 year, suggesting
   single‐cell metabolomics as a powerful complement to CTC enumeration
   for enriched prognostic insights. However, the molecular mechanisms
   driving these metabolic alterations remain unclear and require further
   investigation.

   While our study has demonstrated the potential of single‐cell
   metabolomics to uncover metastasis‐specific metabolic heterogeneity in
   CTCs and to stratify CTCs into subgroups with distinct metastatic
   risks, it is important to acknowledge several limitations in this
   study. First, the deeper metabolic regulatory mechanisms underlying
   brain‐ and bone‐tropic metastasis in lung cancer CTCs remain to be
   fully elucidated. Further mechanistic studies are warranted to explore
   the metabolic drivers of organotropic metastasis and their clinical
   implications. Second, the relatively small sample size of our study
   cohort may limit the statistical power and generalizability of our
   findings. Third, while we specifically enrolled patients with brain or
   bone metastases, some cases also presented with concurrent lymph node
   involvement, which warrants further investigation regarding its
   potential influence on distant metastasis prediction. Additionally, the
   absence of longitudinal data limits our ability to assess the dynamic
   changes in CTC metabolites during disease progression and treatment.
   Incorporating longitudinal sampling will be essential to elucidate
   temporal dynamics of CTC metabolism and its relationship with
   metastasis.

   Future studies will focus on validating and expanding the clinical
   utility of our CTC metabolite‐based classification system. One
   promising direction is to apply this analysis to a prospective cohort
   of locally advanced lung cancer patients. This will allow us to
   evaluate whether the identified metabolomic signatures can predict
   site‐specific metastasis before it occurs. Such studies could provide
   valuable insights into the early metastatic events and help identify
   high‐risk patients for specific metastasis, enabling timely
   interventions and personalized treatment strategies. Additionally, we
   aim to explore the dynamic changes in CTC metabolites during disease
   progression and treatment, which may further enhance our understanding
   of the metabolic mechanisms underlying metastasis and improve the
   predictive accuracy of our model. Another important area for future
   research is the integration of multi‐omics approaches with single‐cell
   metabolomics. Combining metabolomic data with genomic, transcriptomic,
   and proteomic profiles could provide a more comprehensive understanding
   of the biological mechanisms driving metastasis.^[ [158]^52 , [159]^53
   , [160]^54 ^] This integrative approach may identify novel therapeutic
   targets and biomarkers that can be translated into clinical practice.

   Taken together, this study established a metabolic phenotype‐based CTC
   classification system by integrating CTC enrichment, single‐cell
   metabolomics, and metastasis‐specific metabolic profiles. This marks a
   critical step toward precise prediction of lung cancer metastasis risk.
   Although a larger cohort validation and functional characterization of
   key metabolites are needed, this approach offers a novel perspective
   for integrating metabolomics into clinical decision‐making. With
   advancing technologies, single‐cell molecular profiling is set to
   assume a growing role in oncology. Elucidating the interplay between
   CTC heterogeneity and disease progression will drive liquid biopsy from
   quantitative counting to qualitative analysis, advancing precision
   medicine. Furthermore, methodologies to identify functional CTC
   subgroups at the single‐cell level may open new avenues for cancer
   biology, facilitating the discovery of prognostic markers and
   therapeutic targets.

3. Conclusion

   This study introduces a novel approach to classify CTCs based on their
   metabolic profiles, offering a more precise prediction of lung cancer
   metastasis. By integrating a self‐developed CTC sorting and capture
   platform with single‐cell metabolic MS analysis, we have successfully
   established a classification system that categorizes CTCs into distinct
   subgroups associated with metastatic potential. This innovative method
   not only demonstrates superior predictive power compared to traditional
   clinical indicators and total CTC counts but also highlights the
   potential of single‐cell metabolic profiling in uncovering new
   therapeutic targets and prognostic markers.

   The key findings and contributions of this study are as follows: i)
   Development of a platform for high‐viability capture and effective
   release of rare CTCs from peripheral blood, which supports downstream
   analyses; ii) Identification of differential metabolites linked to
   metastatic potential through single‐cell metabolic profiling; iii)
   Establishment of a model that categorizes CTCs into subgroups with
   distinct metastatic risks, validated using clinical specimens; iv)
   Prospective follow‐up confirmed the accuracy of this metabolite‐based
   classification system in predicting metastasis risk.

   In summary, this work represents a significant advancement in liquid
   biopsy and precision medicine, offering a more personalized approach to
   managing lung cancer. Future research will focus on validating this
   approach in larger cohorts and exploring the underlying molecular
   mechanisms. This study paves the way for integrating single‐cell
   metabolomics into clinical decision‐making to enhance patient care.

4. Experimental Section

   The detailed procedures of certain experiments are provided in the
   Supporting Experimental Section.

Materials and Reagents

   A 0.25% Trypsin‐EDTA, Antibiotic‐Antimycotic (100X), Dulbecco's
   Modified Eagle's Medium (DMEM), Fetal Bovine Serum (FBS) and phosphate
   buffered saline (PBS) were purchased from Meilunbio (Dalian, China).
   Ammonium formate was purchased from Sigma Aldrich (St. Louis, MO).
   Ultrapure water was produced by Milli‐Q water purification system
   (Millipore, Bedford, MA).

Clinical Characteristics of Patients with Lung Cancer

   This study was conducted at the Second Affiliated Hospital of Dalian
   Medical University from March 2023 to May 2024, with approval from the
   Institutional Ethics Committee (Approval No. 128, 2023; Second
   Affiliated Hospital of Dalian Medical University). All participants
   provided written informed consent. A total of 61 non‐small cell lung
   cancer (NSCLC) patients were enrolled, of whom 31 were used for
   clinical validation of the CTC sorting and capture platform, and 30
   were for constructing a single‐cell metabolic profiling model of lung
   cancer CTCs. Inclusion criteria for patients in the metabolic profiling
   cohort were as follows: 1) Age ≥ 18 years; 2) Pathologically confirmed
   lung adenocarcinoma; 3) Imaging‐confirmed brain or bone metastasis; 4.
   No history of other malignancies or major comorbidities within the past
   five years. After excluding cases with CTC counts < 2, patients were
   stratified into three groups based on metastatic status: brain
   metastasis group (n = 10), bone metastasis group (n = 10), and
   nonmetastatic group (n = 10). Patients in the non‐metastatic group
   underwent prospective follow‐up for one year, with quarterly
   assessments by specialists to document metastasis occurrence and
   location. Clinical characteristics collected included age, sex, TNM
   stage, and serum biomarkers (NSE, CEA, CA‐125, CYFRA21‐1).

Cells and Cell Culture

   Human lung adenocarcinoma cell lines PC9 (RRID: CVCL_B260) and
   NCI‐H1975 (RRID: CVCL_1511) were purchased from the Chinese Academy of
   Medical Sciences (Beijing, China). Brain metastasis‐derived sublines
   (Brm) and bone metastasis‐derived sublines (Bom) were previously
   established by the research group through experimental protocols.^[
   [161]^42 ^] Parental PC9, NCI‐H1975, Brm, Bom, and circulating tumor
   cell sublines (PC9‐CTC) were cultured in RPMI 1640 medium (Meilunbio,
   Dalian, China) supplemented with 10% FBS and 1% penicillin/streptomycin
   (P/S). All cell lines were maintained at 37 °C in a humidified
   incubator with 5% CO₂. All cell lines tested negative for Mycoplasma
   contamination and were authenticated by short tandem repeat (STR)
   analysis upon acquisition. Prior to cryopreservation, cells received
   Mycoplasma prophylactic treatment; following resuscitation, they were
   maintained in prophylaxis‐free culture medium for 5–7 days to confirm
   the absence of prophylactic effects on experimental outcomes.

CTC Sorting and Capture

   For CTC sorting, fresh whole blood samples were treated with RosetteSep
   Human CD45 Depletion Cocktail (STEMCELL Technologies, USA). The samples
   were mixed thoroughly and incubated at room temperature for 20 min to
   promote specific crosslinking between leukocytes and erythrocytes. The
   treated blood samples were then diluted 1:1 with buffer (Hank's
   Balanced Salt Solution containing 2% FBS). The diluted blood was
   carefully layered onto a pre‐prepared density gradient medium and
   centrifuged at 1200 g for 20 min. After centrifugation, the cell layer
   at the interface between the density gradient medium and diluted plasma
   was gently aspirated using a sterile pipette. For CTC capture, a
   micropore‐arrayed filtration membrane was secured in a custom‐designed
   holder and pretreated with 75% ethanol, followed by thorough rinsing
   with Hank's Balanced Salt Solution. The enriched cell suspension was
   transferred onto the membrane surface using a pipette. After complete
   filtration, CTCs and residual blood cells were retained on the
   membrane. Captured cells were gently washed with buffer to remove
   nonspecifically adhered blood cells. Finally, the membrane was removed
   from the holder, inverted, and immersed in buffer at 37 °C for 10 min.
   As the PVA coating dissolved, the captured cells were released into the
   buffer. All procedures were performed in a laminar flow hood to
   maintain sterility. Sample processing was completed within 2 h to
   maximize cell viability. Processed samples were either immediately
   analyzed or temporarily stored at 4 °C.

Sorting and Capture platform Performance Toward Rare Tumor Cells

   To evaluate the platform's ability to maintain the viability of rare
   tumor cells during peripheral blood processing, healthy donor blood
   samples were spiked with GFP‐labeled PC9 and NCI‐H1975 cells
   (100 cells mL^−1) and processed through the CTC sorting and capture
   platform. Captured cells were then subjected to in situ staining using
   a commercial live/dead cell viability kit (Solarbio Science &
   Technology, Beijing, China), with propidium iodide (PI, dead cell
   marker, red fluorescence). Rare cells displaying red fluorescence were
   observed and counted under a fluorescence microscope to calculate
   short‐term survival rates. A control group (untreated spiked cells) and
   the experimental group (processed cells) were maintained under
   identical conditions for comparison. To assess long‐term viability,
   captured PC9 cells were cultured in situ for 48 h, followed by
   evaluation of their morphology, adhesion, and spreading behavior.

   To validate the recovery efficiency of the CTC sorting and capture
   platform for target cells in complex cellular backgrounds, healthy
   donor whole blood samples (5 mL) were spiked with varying numbers (10,
   50, 125, 250, and 500) of PC9 and NCI‐H1975 cells to test recovery
   rates at concentrations of 2, 10, 25, 50, and 100 mL. For cell counts
   exceeding 100, spiking was performed via serial dilution, while 50‐cell
   aliquots were precisely added using a microscope‐guided micropipette
   and micromanipulator (Eppendorf, Hamburg, Germany). To facilitate
   fluorescence microscopy‐based observation and recovery rate
   quantification, PC9 and NCI‐H1975 cells were pre‐labeled with
   fluorescent markers. During spiking, tumor cells were resuspended in
   RPMI 1640 medium supplemented with 10% FBS to maintain viability during
   the precise, time‐sensitive pipetting process.

Construction of Lung Cancer Model

   A lung cancer animal model was established using female SRG rats (6–8
   weeks old). PC9‐GFP‐Luc cells (expressing green fluorescent protein and
   luciferase) were employed to monitor tumor growth via bioluminescence
   imaging. A cell suspension containing 5 × 10^6 cells/rat in 50 µL was
   mixed 1:1 with Matrigel matrix (Corning, New York, USA) and injected
   into the lungs of immunodeficient rats via orthotopic implantation. In
   vivo analysis of lung colonization was performed using bioluminescence
   imaging technology, with images acquired every 5 days starting 5 days
   postinjection via the IVIS imaging system. Peripheral blood was
   collected via the jugular vein prior to euthanasia. All animal
   experiments were approved by the Dalian Medical University Animal
   Ethics Committee (AEE23117). Following peripheral blood collection, all
   experimental animals were euthanized under anesthesia.

Detection of CTCs in Blood Samples

   To validate the effectiveness of the CTC sorting and capture platform
   in practical applications and enable downstream CTC analysis,
   peripheral blood samples from lung cancer animal models and non‐small
   cell lung cancer (NSCLC) patients was processed. To confirm the
   specificity of the capture and staining protocols, blood samples were
   collected from healthy volunteers (n = 5) and patients (n = 5) with a
   smoking history and pathologically confirmed benign pulmonary nodules
   (post‐resection) as negative controls. Rat (2 mL) and human (5 mL)
   peripheral blood samples were drawn into EDTA‐containing vacuum blood
   collection tubes. For patient samples, unfixed CTCs were labeled with
   Alexa Fluor 594‐conjugated pan‐cytokeratin (CK, 1:100), Alexa Fluor
   594‐conjugated anti‐human CD326 (EpCAM, 1:100), Alexa Fluor
   488‐conjugated CD45 (1:100), and Hoechst 33 342 (1:100). Cells were
   observed and counted using fluorescence microscopy, with CTCs defined
   as CK+/EpCAM+, CD45−, and Hoechst+. For subsequent single‐cell
   metabolomic analysis, individual CTCs from patients were isolated under
   direct microscopic visualization using a micropipette and
   micromanipulator (Eppendorf, Hamburg, Germany) and transferred to
   sample preservation tubes. All procedures were completed within 2 h to
   ensure cell viability.

Mass Spectrometric Single‐Cell Metabolomics Analysis

   The nanoelectrospray ionization‐atmospheric pressure chemical
   ionization (nanoESI‐APCI) analysis procedure was conducted as described
   in previous publication.^[ [162]^55 ^] Briefly, cells were cultured in
   6 cm dishes until they reached 80% confluence. The culture medium was
   then removed, and the cells were washed twice with 4 mL PBS to remove
   residual medium, followed by the addition of 5 mL PBS. Afterward, the
   dish was transferred to the stage of a CKX53 microscope (Olympus,
   Tokyo, Japan), where a nanoESI micropipette preloaded with 10 µL of
   150 mm ammonium formate solution was mounted on a 3‐D micromanipulator
   (MP‐225, Sutter Instrument, Novato, CA). The micropipette was obtained
   by pulling a borosilicate glass capillary tube (I.D. 0.86 mm, O.D.
   1.5 mm) to form a tip with 1 µm via a P‐1000 puller (Sutter Instrument,
   Novato, CA). Under microscopic guidance, the tip was carefully moved
   toward the cells. Once the tip contacted the cell surface, an air pump
   was activated to create a negative pressure at the tail of the
   micropipette for the extraction of cytoplasm. The extraction was
   maintained for 90 s. After completion, the micropipette was unloaded
   and transferred to the homemade nanoESI‐APCI platform, where the APCI
   source and pulsed high‐voltage were activated to perform nanoESI‐APCI
   analysis on a QE‐HF mass spectrometer. The MS parameters were set as
   follows: the capillary temperature was set to 275 °C, with both gas
   flow and voltage set to 0. The RF‐lens was adjusted to 50. Data
   acquisition was performed in full‐scan mode with a mass range of
   70–1050 m z^−1.

Data Processing and Statistical Analysis

   Custom R or Python programs were developed to process raw single‐cell
   metabolomics data, including peak alignment, stable ion feature
   selection, and metabolite identification. Detailed information is
   provided in the Supporting Information.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   Y.X., X.H., and Y.Y. contributed equally to this work. Three authors
   were involved in experimental design, experimental execution, data
   collection, statistical analyses, and manuscript writing. J.W., W.L.,
   and C.Y. collected all biological samples. X.S. critically revised the
   manuscript for important intellectual content and approved the version
   to be published. W.Q. and L.W. assisted with data interpretation. M.L.
   and Y.J. assisted with the CTC sorting and capture. Q.W., G.X., C.H.
   and W.W. are the guarantors of this work and, as such, had full access
   to all the data in the study and take responsibility for the integrity
   of the data and the accuracy of the data analysis.

Supporting information

   Supporting Information
   [163]ADVS-12-e08878-s001.docx^ (4.1MB, docx)

Acknowledgements