Single-cell metabolic fingerprints discover a cluster of circulating tumor cells with distinct metastatic potential

Metastasis is the primary cause of mortality in most cancers, including colorectal cancer (CRC), where approximately half of patients develop metastatic disease. Conventional imaging methods (MRI/CT) have limited sensitivity for early metastatic lesions, underscoring the need for better predictive markers. Circulating tumor cells (CTCs) are shed from primary tumors into the bloodstream and are used clinically as indicators of metastatic risk, yet simple CTC enumeration may be insufficient because it ignores CTC heterogeneity. The study aims to determine whether single-cell metabolic phenotyping of CTCs can stratify metastatic potential more accurately than total CTC counts and clinical biomarkers, and to develop a practical classifier for predicting metastasis risk in CRC patients.

Prior work has established the clinical relevance of CTCs in several cancers and highlighted technological advances in liquid biopsy and single-cell analyses. Single-cell mass spectrometry has emerged for profiling metabolites at cellular resolution, yet quantitative single-cell metabolomics has been hampered by challenges in controlled subpicoliter extraction, normalization of nonbiological variation, and assay sensitivity/repeatability. Previous studies also link metabolic reprogramming, including amino acid and glutathione metabolism and the Warburg effect, to metastatic processes. These insights motivate a targeted, quantitative single-cell metabolomics approach to capture functionally relevant CTC heterogeneity for metastasis prediction.

Study design integrated discovery and validation. 1) Discovery in cell lines: Two pairs of CRC cell lines with differing metastatic potential (SW480 vs. SW620; HT-29 vs. COLO 205) underwent UPLC-HRMS untargeted metabolomics in positive and negative ion modes. After QC, filtering, and normalization, PCA and OPLS-DA identified differentially abundant metabolic features (|Log2FC| > 1.0, p < 0.05, VIP > 1), which were annotated using MS/MS spectra and standards. Intersections across pairs yielded candidate metabolites subjected to KEGG pathway and metabolite set enrichment analyses, prioritizing pathways in amino acid metabolism, glutathione metabolism, and the Warburg effect. Fourteen metabolites (better covered in negative mode) were selected for single-cell analysis. 2) Single-cell quantitative mass spectrometry platform: A home-built platform used electro-osmosis-assisted extraction with nanocapillaries (~100 nm tip) to withdraw controlled subpicoliter volumes from single cells. Extracted volume was linearly tunable with applied voltage (0 to −4 V) and extraction time (5–60 s), achieving ~60% extraction efficiency and ~120 nL estimated volume. To correct nonbiological variation, isotope-labeled internal standards were co-extracted; signal ratios restored linearity in calibration (e.g., glucose ratio R² ~0.97). Targeted MRM transitions were optimized for 14 metabolites (two transitions each where applicable). Using 5% BSA as surrogate matrix, LODs, LLOQs, and linear ranges were established; intraday and interday CVs (approximately 8–17.5%) met general bioanalytical validation criteria. 3) Clinical CTC cohorts and single-CTC analysis: CRC patients were enrolled and subjected to CTC enrichment and identification. Single CTCs were isolated, micro-sampled, and quantified for target metabolites via MRM. Eleven metabolites were robustly quantified per CTC (including glutamic acid, lactic acid, aspartic acid, malic acid, glutamine, glutathione, etc.). 4) Machine learning classifier: To account for heterogeneity, unsupervised non-negative matrix factorization (NMF) on single-CTC metabolite abundances defined two subgroups (K = 2). Logistic regression on the resulting components identified a 4-metabolite fingerprint (glutamic acid, lactic acid, aspartic acid, malic acid) to compute a risk score per CTC. An optimal cutoff (0.420, Youden index) dichotomized CTCs into C1 (low risk) and C2 (high risk). Patient-level predictors included total CTC count, C1 count, and C2 count. 5) Validation and functional assays: Associations between subgroup counts and metastasis were evaluated in training and test cohorts, with ROC/AUC, sensitivity, specificity, and accuracy. Additional functional validation included adding the four metabolites to cells in transwell assays (promoting migration), establishing C1- and C2-derived CTC lines (C2 showed higher proliferation/migration by CCK-8 and transwell assays), and a CTC-derived explant (CDX) model (C2 enhanced metastasis in vivo). Detailed analytical methods (sample prep, LC-MS parameters, data processing with MS-DIAL, MetaboAnalyst, and statistical analyses) are provided.

- Untargeted metabolomics in CRC cell lines revealed 19 shared metabolites differing between low- and high-metastatic lines, enriched in amino acid, glutathione, and Warburg-effect-related pathways. Fourteen metabolites were prioritized; 11 were robustly quantified at the single-CTC level. - Single-CTC analyses showed pronounced heterogeneity (e.g., glutathione 0.22–4.33 nM within patients; intrapatient RSD up to 124.3%). - NMF identified two CTC clusters. A 4-metabolite fingerprint (glutamic acid, lactic acid, aspartic acid, malic acid) produced a risk score and separated CTCs into C1 (low risk) and C2 (high risk) at a cutoff of 0.420. - In the training cohort, patient metastasis status associated weakly with total CTC counts (p = 0.0207; AUC = 0.681) but strongly with C2 CTC count (p < 0.0001). C2 count achieved sensitivity 78.8%, specificity 96.3%, accuracy 86.7%, and AUC 0.927 for predicting metastasis, outperforming total CTC count and clinical indices. C1 CTC count was inversely associated (p = 0.0004). - In the test cohort, the classifier again separated CTCs into two subgroups; total CTC and C1 counts did not significantly associate with metastasis, while C2 count retained strong predictive value with favorable ROC metrics (consistent with training cohort). Reported overall performance included accurate prediction in 57.9% (33/57) of metastatic patients among enrolled cases, with false-positive and false-negative rates of approximately 19.0% and 6.1% among CTC-positive patients. - Functional assays supported biological relevance: the four metabolites enhanced migration in transwell assays; C2-derived CTC lines exhibited higher proliferation and migration; and in a CDX model, C2 subgroup promoted metastasis in vivo.

The study demonstrates that incorporating single-CTC metabolic heterogeneity markedly improves metastasis risk prediction over total CTC counts or bulk metabolic profiles. By focusing on a biologically coherent, knowledge-reduced set of metabolites linked to amino acid/glutathione metabolism and the Warburg effect, the NMF-driven 4-metabolite fingerprint isolates a CTC subgroup (C2) with distinct metastatic potential. This aligns with literature implicating glutamic acid, aspartic acid, lactic acid, and malic acid in cancer progression and metabolic plasticity. The classifier generalized to an independent test cohort and showed superiority to standard clinical indices (CEA, CA19-9) and total CTC enumeration. Functional experiments corroborated the mechanistic plausibility that the identified metabolic phenotype confers pro-metastatic traits. These findings suggest that specific CTC subpopulations, rather than aggregate counts, drive metastatic risk and can be captured via quantitative single-cell metabolomics combined with heterogeneity-aware machine learning.

A quantitative single-cell mass spectrometry platform and machine learning pipeline identified a metabolically defined CTC subgroup (C2) that strongly predicts metastatic risk in colorectal cancer. A simple 4-metabolite fingerprint (glutamic acid, lactic acid, aspartic acid, malic acid) stratifies individual CTCs and, when aggregated as C2 CTC count, outperforms total CTC counts and common clinical biomarkers. Functional assays and in vivo models support the biological relevance of this metabolic phenotype. Future work should expand clinical validation with larger, multi-center cohorts, dissect mechanisms via pathway perturbations and isotope tracing/flux analysis, and compare metabolic profiles of CTCs with primary and metastatic lesions to refine biomarkers and therapeutic targets.

- Mechanistic underpinnings linking the four-metabolite signature to metastatic behavior are not fully delineated; further pathway-level perturbation and flux studies are needed. - Cohort sizes are moderate; broader, multi-center validation with higher statistical power is required. - Some enrolled patients were CTC-negative and not analyzed by the molecular typing system, limiting generalizability and contributing to false-negative/false-positive estimates. - Single-cell sampling and quantitation, while validated, may still be susceptible to residual technical variability and selection biases in CTC isolation. - The classifier was derived from CRC cohorts; applicability to other cancers remains to be tested.