CTC Clusters Predict Breast Cancer Metastasis

Clinical associations of single and clustered CTCs in breast cancer. At the Northwestern University Circulating Tumor Cell Core, we established a clinical protocol of blood CTC tests for patients with stage III and IV breast cancer from 2016 to 2025 (N = 1,529) using the CellSearch platform coupled with image scans (Figure 1A and Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/JCI193521DS1). In addition to the enumeration report of single CTCs (DAPI+CK+CD45–), we manually reviewed the immunofluorescence images of EpCAM bead–precipitated cells for the identification of homotypic CTC clusters (≥2 CTCs/cluster) and heterotypic CTC clusters with CD45+DAPI+ WBCs (Figure 1, A and B, and Supplemental Figure 1). Of these tests, 43.23% were CTC positive (≥5 single CTCs within 7.5 mL blood). Of CTC-positive biospecimens (N = 661), 44.02% contained homotypic clusters (at least 1 cluster), whereas 75.49% were positive for heterotypic clusters (at least 1 cluster) (Figure 1B). The range, mean, and median counts were 0–17,427; 47; and 0 for single CTCs; 0–2,265; 3; and 0 for homotypic clusters; and 0–1,017; 6; and 2 for heterotypic clusters with a higher detection frequency of heterotypic clusters than homotypic clusters but lower than single CTCs (Figure 1C). Most heterotypic CTC-WBC clusters contained only 1 immune cell with 1 tumor cell (Supplemental Figure 1B).

CTC frequencies in the blood biopsies of breast cancer patients and their clinical associations. (A) Top: Schematic of CTC analysis via CellSearch with blood specimens drawn from patients with breast cancer (n = 1,529). Bottom: Representative CellSearch images of single CTC (left), homotypic CTC-CTC cluster (middle), or heterotypic CTC-WBC cluster (right) with merge channels of CK (green) and DAPI (magenta) as well as a single CD45 channel. Scale bars: 10 μm. (B) Frequency of CellSearch-detected CTC+ tests/scans (≥5 CTCs within 7.5 mL blood) among breast cancer patient biopsies (left; n = 1,529) and frequencies of homotypic CTC-CTC clusters (middle) and heterotypic (right) CTC-WBC clusters among CTC+ biospecimens (n = 661). (C) Counts of single CTCs and homotypic and heterotypic CTC clusters per 7.5 mL blood in 1,529 CellSearch tests. The range, mean, and median are 0–17,427; 47; and 0 for single CTCs; 0–2,265; 3; and 0 for homotypic clusters; and 0–1,017; 6; and 2 for heterotypic clusters. ****P < 0.0001 for any 2-group comparison using Wilcoxon’s matched-pair signed rank test. (D) Cox proportional hazard model odds ratio plot with 95% CI for risk of single CTCs (black), homotypic CTC-CTC clusters (red), and heterotypic CTC-WBC clusters (blue) among subtypes of breast cancer and self-identified racial groups of the patients. Filled squares highlight significant features calculated using Wald’s test (P < 0.05). (E) Scatter plots of single versus homotypic clusters and single versus heterotypic clusters with Pearson’s correlation coefficient and 2-tailed P value. (F) Kaplan-Meier survival curves of patients positive for single CTCs (≥5), homotypic clusters (≥1), or heterotypic clusters (≥1) versus the patients with negative results. Log-rank (Mantel-Cox) test P values and hazard ratio (HR) are displayed. (G) Kaplan-Meier survival curves of patients with breast cancer, divided by race (Black and White) and heterotypic cluster status. Log-rank (Mantel-Cox) test P value is shown.

We performed a Cox proportional hazard analysis to determine which patient variables would be risk factors for the detection of single CTCs, homotypic clusters, or heterotypic clusters (Figure 1D). Compared with the White group, self-identified Black or African American patients had a higher risk specifically for heterotypic clusters with no significant difference for single CTCs or homotypic clusters. When luminal A breast cancer subtype served as the reference control, luminal B, HER2-enriched, and triple-negative breast cancer (TNBC) showed higher risks of single CTCs. In contrast, only TNBC and luminal B had higher risks for homotypic and heterotypic clusters.

Single CTC counts positively correlated with both cluster types, having a stronger correlation with homotypic cluster count (R2 = 0.9624) than heterotypic cluster count (R2 = 0.5915) (Figure 1E), implying that additional factors derived from the WBCs might contribute to heterotypic cluster formation. Based on the clinical follow-up data, Kaplan-Meier analysis of these patients demonstrated what we believe to be new prognostic values of heterotypic CTC-WBC clusters and confirmed those of single CTCs (5) and homotypic CTC clusters (6, 17), all of which correlate with unfavorable survival (Figure 1F). Notably, when stratified by race and the status of heterotypic clusters, Black patients positive for heterotypic clusters had the shortest overall survival versus the other 3 groups (Figure 1G). Shortened survival when stratified by race was not observed when analyzing single CTC status, but we also observed worse overall survival in Black patients that were positive for homotypic CTC clusters, compared with White patients (Supplemental Figure 2, A–C). These data provide what we believe to be novel insights into the clinical utility of CTCs, especially heterotypic CTC-WBC clusters associated with breast cancer outcomes.

DPT cells are enriched in heterotypic CTC-WBC clusters and promote seeding. We continued to examine the WBC composition in heterotypic clusters and their functional contribution to CTC-mediated seeding. Because CellSearch is limited to 4-channel immunofluorescence staining with only 1 channel open for customized marker analysis, we utilized established flow cytometry (15, 32) and imaging flow approaches (ImageStream and BD CellView) to characterize WBCs in heterotypic CTC clusters, based on expanded channels and cellular images (Figure 2A). After red blood cell lysis, we profiled the blood cells of 26 patients with breast cancer using extensive markers (33, 34) (plus forward and side scatter channels) to identify T cells, B cells, NK cells, monocytes, and neutrophils in single WBCs and heterotypic clusters (Supplemental Figure 3 and Supplemental Table 2). Compared with single WBCs in circulation, heterotypic CTC clusters included overrepresented T cells (32.6%), NK cells (12.6%), and B cells (4.7%), underrepresented neutrophils (30.2%), and monocytes (19.9%) without significant changes (Figure 2, B–D). The most striking identification was CD4+CD8+ DPT cells (14.2%) with a 140-fold enrichment in clusters versus a rare presence in WBCs (0.1%), in which double positivity was confirmed via ImageStream imaging cytometry (Figure 2, B–D). Additionally, the single-positive CD8+ and CD4+ T cells made up 13.2% and 5.6% of the heterotypic clusters, respectively (Figure 2, B–D).

DPT cells are 140-fold enriched in CTC-WBC clusters compared with single WBCs. (A) Schematic of analyzing the frequency of broad classes of immune cell compositions in the blood biopsies of breast cancer patients via flow cytometry and ImageStream. N = 26 patients (n = 1,402 CTC-WBC clusters). (B) Frequency of immune cells (neutrophils, monocytes, NK cells, and B cells) and different T cell populations (DPT, CD4+ T, and CD8+ T) in patient blood (WBCs) (left pie chart) and CTC-WBC clusters (right pie chart). N = 26 patients (n = 698 CTC-T cell clusters). (C) Left: Flow dot plots of CD3+ T cells in single WBCs (top) and heterotypic CTC-WBC clusters (bottom). Top right: ImageStream photos of CD8+CD4+ DPT, CD4+T, and CD8+T cells. Bottom right: Representative images of CTC-DPT cluster via ImageStream imaging cytometry. Scale bars: 10 μm. (D) Frequency of subset T cells, neutrophils, monocytes, NK cells, and B cells from individual patients. Multiple Wilcoxon’s tests, *P < 0.05. N = 26. (E) MFI of various T cell markers (CD44, CD62L, CD45RO, CCR7, TIM-3, PD-1, CD25, and TIGIT) in human DPT cells of breast cancer patients compared with CD4+ and CD8+ single-positive T cells, as detected by flow cytometry. Friedman’s test with Dunn’s multiple-comparison test. N = 8. (F) Phenotypic characterization of human DPT cells in breast cancer patients compared with CD4+ and CD8+ single-positive T cells, including naive (CD62L+CD44–), memory (CD45RO+), central memory (CD45RO+CCR7+), effector memory(CD45RO+CCR7–), terminal effector (PD-1+TIM3–), progenitor-exhausted (TIM3+PD-1–), and terminal-exhausted cells (PD-1+TIM3+). N = 8 patients. Friedman’s test with Dunn’s multiple-comparison test. N = 8 breast cancer patients. *P < 0.05. (G) CTC-WBC clusters containing DPT cells (normalized counts) in patients who received anti–PD-1 treatment (pembrolizumab) before liquid biopsy. N = 20 patients who did not receive it (no anti–PD-1), and N = 6 patients who received it (+anti–PD-1). Mann-Whitney unpaired 2-sided t test.

While thymocytes go through a double-positive stage in T cell development and could become fully mature T cells (28), we found that DPT cells were relatively rare in the periphery in the blood analyses (shown above). In collaboration with the HuBMAP consortium (35, 36), we obtained spatial immunofluorescence staining images of human spleens and identified rare DPT cells close to the germinal centers (Supplemental Figure 4). We further characterized human DPT cell features in the blood biopsies of breast cancer patients using flow cytometry (Supplemental Figure 5A). DPT cells displayed a marker profile closely resembling that of CD4+ T cells. However, compared with CD8+ T cells, they had slightly higher expression levels of CD44 (activation), CD62L (trafficking; also known as L selectin), and exhaustion marker TIM3 (Figure 2E). In line with this, DPT cells had a similar distribution as CD4+ T cells but portrayed higher frequencies than CD8+ T cells in memory (CD45RO+) subsets (Figure 2F). Nevertheless, DPT showed higher frequencies of progenitor-exhausted phenotypes (TIM3+PD-1–) relative to single-positive T cells (Figure 2F). This suggests that DPT cells in cancer patients have an activated, exhausted, and immunosuppressive phenotype. Based on this information, we reanalyzed our flow cytometry data of CTC-DPT clusters to see if patients that received pembrolizumab (anti–PD-1 antibody) in the treatment schedule before the liquid biopsy blood sample had reduced CTC-DPT clusters. While not significant, there was a trend toward reduced CTC-DPT clusters in patients who had received pembrolizumab, suggesting that an immune checkpoint blockade may affect DPT cells and CTC-DPT clusters (Figure 2G).

Next, we sought to determine the role of DPT cells in heterotypic cluster-mediated metastatic seeding during experimental lung colonization in syngeneic mouse models (Figure 3A). After splenocytes were harvested from BALB/c mice, DPT cells were isolated via high-speed sorting (see Methods). These cells were premixed at 4:1 ratio with murine TNBC 4T1 cells (Luc2-tdTomato/L2T labeled) (37) and seeded in a poly(2-hydroxyethyl methacrylate)–coated (poly-HEMA–coated) 96-well plate for 6-hour cluster formation ex vivo (Supplemental Figure 6A). Mixed cells and clusters were collected gently and immediately infused into the tail veins of BALB/c recipient mice, and mice were immediately imaged using bioluminescent imaging. Compared with 3 control groups of 4T1 singles, 4T1 mixture with homotypic clusters, and 4T1-splenocyte mixture, 4T1-DPT mixture seeded to the lungs with the highest efficiency and strongest bioluminescence signal at 16, 24, and 96 hours (Figure 3, B and C). H&E staining of the mouse lungs 6 days after injection validated the better colonization by 4T1-DPT clusters compared with other groups (Figure 3, D and E).

DPT tumor cell clustering promotes metastasis formation in an experimental metastasis assay in vivo. (A) Schematic depicting experimental design of mouse DPT isolation, clustering with L2T-labeled 4T1 tumor cells ex vivo (controls groups including 4T1 cells [singles and clusters], and 4T1 and mouse splenocytes), tail vein infusion, and lung colonization monitored via bioluminescence imaging of L2T+ 4T1 cells and histology validation. (B and C) Representative images (B) and quantification (C) of in vivo bioluminescent signals in mouse lungs of L2T+ 4T1 tumor cells after clustering and tail vein injection. Kruskal-Wallis test with multiple comparison; n = 3 mice per group. (D and E) H&E staining images with inserted regions of control or micrometastasis in the mouse lungs (D) and quantification of metastasis lesions of lung sections (E) on day 6 after infusion of 4 groups of cells: 4T1 singles, 4T1 homoclusters, 4T1-DPT clusters, and 4T1-splenocytes. Arrows point to metastatic lesions. Scale bars: 100 μm. Three-way ANOVA with Dunn’s multiple-comparison test was used for P value calculations. N = 3 for 4T1-DPT and 4 for all other groups. (F) Repeated experiment of DPT-4T1 clustering–promoted metastatic seeding and colonization with single CD4+ and CD8+ T cell controls in mix clustering with 4T1 cells. Data in graphs represent mean ± SEM. Unpaired 2-tailed t test; *P < 0.05, ****P < 0.001. N = 6 biological replicates. (G) Enriched pathways upregulated (left) or downregulated (right) in 4T1 cells after being incubated with DPT cells versus those with splenocytes for 6 hours, identified via the Enrichr database (https://maayanlab.cloud/Enrichr/) (WikiPathways [WP]) based on scRNA-Seq (10X Genomics) data. The respective UMAP plots for each group are provided in Supplemental Figure 6B.

As single-positive T cells were observed to be clustered with CTCs in patient blood, we also compared the seeding efficiency of the DPT-4T1 mixture with CD4-4T1 and CD8-4T1 mixtures. Similarly, infusion of DPT-4T1 cells resulted in the highest lung bioluminescent signal of disseminated tumor cells 6 days after tail vein injection compared with these 2 groups and other controls (Figure 3F and Supplemental Figure 7). These data demonstrate that DPT cells of heterotypic clusters promote CTC-mediated pulmonary seeding and colonization.

To elucidate DPT interaction–mediated effects on tumors, we conducted single-cell RNA-Seq (scRNA-Seq) of 4T1 tumor cells incubated with sorted DPTs or splenocytes for 6 hours of clustering. Tumor cells and immune cells from each condition (DPT or splenocytes) were collected together, stained with multiplexing hashtag antibodies for downstream identification, analyzed using scRNA-Seq, and plotted using Uniform Manifold Approximation and Projection (UMAP). The 4T1 tumor cells that were clustered ex vivo with either DPT cells or splenocytes did not separate strongly from each other on the UMAP; however, differentially expressed genes were identified between the tumor cells in these 2 groups (Supplemental Figure 6B and Supplemental Table 3). Using the pathway analysis tool Enrichr (38), we identified DPT-upregulated pathways in tumor cells (pluripotency, mRNA processing, nuclear receptors, and ErbB signaling) as well as markedly downregulated pathways in immune activation–related phagocytosis; IL-3, IL-5, and IL-1 signaling; and chemokine signaling (Figure 3G and Supplemental Table 4). These data suggest that the DPT interaction promotes stemness-related pluripotency and immune evasion of tumor cells.

ITGB1 and ITGA4 are drivers of heterotypic T cell clustering with tumor cells. A previous study reported the presence of intratumoral DPT cells (39), yet circulating DPT cells are rarely studied compared with other WBCs. To bridge this gap, we performed scRNA-Seq of human peripheral WBCs from 19 breast cancer patients and 12 healthy controls (Figure 4A and Supplemental Table 5). To compare different patients, samples were multiplexed using hashtag oligonucleotides (HTOs). Each patient was labeled with a unique HTO per sequencing run for downstream identification. Additionally, since our samples were collected over the course of several months, the dataset was integrated to control batch effects among sequencing runs that contained multiple hash-tagged patient samples per run/batch and to facilitate comparative analysis across datasets (40). After performing quality control to remove low-quality cells and potential doublets followed by dataset integration, 8 broad immune cell types were identified in both cohorts (Figure 4B and Supplemental Figure 8, A and B). When comparing overall immune cell representation and expected abundance based on cancer status, several immune cell types demonstrated differences, including overrepresented T, B, and NK cells in healthy controls along with more abundant intermediate monocytes in breast cancer patients (Figure 4C and Supplemental Figure 9A, left). However, when the patients were stratified by their CTC status into 3 groups, WBCs from CTC-positive patients (≥5 CTCs) showed a higher proportion of T cells and classical monocytes than the other 2 cancer groups with low (1–5 CTCs) or no CTCs (Figure 4C, Supplemental Figure 8C, and Supplemental Figure 9A, right).

scRNA-Seq reveals enrichment of rare T cell subsets (DPTs) in the blood of breast cancer patients, dependent on CTC status. (A) Schematic depicting the isolation of human WBCs and subsequent scRNA-Seq. (B) UMAP plots of 47,234 single WBCs from 19 breast cancer patients (n = 35,401 cells) and 12 healthy control liquid biopsies (n = 11,833 cells), with broad immune cell subsets annotated. (C) Correlation plots depicting χ2 test residuals to determine over- or underenrichment of each immune cell subset from the WBCs of breast cancer patients versus healthy controls (top) and by CTC status (bottom). Dot size corresponds to the absolute value of correlation coefficients, and color corresponds to χ2 residuals. (D) UMAP plots of T cell subsets (n = 20,930 cells). (E) Correlation plots of over- or underenrichment of each T cell subset from the WBCs of CTC-positive, -low, and -negative cancer patients and healthy controls. (F) DPT cells highlighted on T cell subset UMAP plots, split by CTC status. (G) UMAP plot of mouse T cells, including single-positive and DPT cells collected from BALB/c mouse splenocytes. (H) Volcano plot depicting most differentially expressed genes in DPT cells versus all other T cells in mouse splenic T cells. (I) Venn diagram showing the overlap between adhesion molecule genes expressed in total human T, mouse T, human DPT, and mouse DPT cells and a list of 32 genes shared among 4 groups. (J) Violin plots of mouse Itgb1 mRNA expression in mouse T cells isolated from splenocytes, as measured by scRNA-Seq. Kruskal-Wallis test P values are provided. (K) Bar graphs of Cd29 (encoded by Itgb1) and Cd49d (encoded by Itga4) expression (MFI) in mouse DPT, CD4+, and CD8+ cells isolated from BALB/c mouse splenocytes. One-way ANOVA with Tukey’s multiple-comparison test P values provided. N = 6 mice.

To further understand the T cell subset in general, and DPT cells specifically, we isolated the T cell cluster, reperformed dimensionality reduction, and annotated the resulting T cell subsets. Here, we identified 8 subsets of T cells based on their transcript expression, such as naive T cells (TCF7), CD8+ effector (NKG7), memory and effector memory (IL7R), and Tregs (FOXP3) (Figure 4D and Supplemental Figure 8D). While healthy controls had an enrichment in the abundance of effector memory T cells, CTC-positive cancer patients had overrepresented naive T cells (Figure 4E, Supplemental Figure 8, E and F, and Supplemental Figure 9B, left). In contrast, CTC-low and -negative patients possessed a relatively high abundance of circulating effector and memory T cells (Figure 4E and Supplemental Figure 9B, right), which might be able to reject CTCs upon encountering them. Consistent with the literature (41), Tregs were slightly overrepresented in cancer patients in our dataset analyses (Figure 4E).

Based on transcript expression cutoff values of CD4 and CD8A, we identified DPT cells (~500 cells) that were distributed in the CD8+ effector, memory (mostly CD4+), and naive T subsets with a clear enrichment in CTC-positive patients (Figure 4F and Supplemental Figure 8G). Using a public scRNA-Seq database of peripheral blood cells from patients with TNBC (42), we extracted additional DPT cells for combined profiling with our data (a total of 1,454 DPT cells), alongside 1,000 single-positive CD4+ or CD8+ T cells from each dataset. In this combined dataset, we observed the heterogeneity of DPT cells, as some are present in scRNA-Seq clusters with CD8+ T cells, while others more closely cluster with CD4+ T cells (Supplemental Figure 8H, Supplemental Figure 10, A and B, and Supplemental Figure 11A). Furthermore, we conducted scRNA-Seq on sorted mouse splenic single-positive T and DPT cells. DPT cells are abundant in multiple clusters with relatively distinct profiles from single-positive T cells on the UMAP plot (Figure 4G). In the mouse dataset, DPT and single-positive T cells from tumor-naive or tumor-bearing mouse spleens were sorted and stained with HTOs separately. DPT or single-positive T cell identity for each cluster was determined based on both gene expression and abundance of each hashtag label present in those clusters (Supplemental Figure 10, C–E). In general, we observed negligible differences between DPT and T cells from tumor-bearing and tumor-naive mice. While DPT cells dominated some clusters (Supplemental Figure 10E), there was some shared gene expression between single-positive T and DPT cells, likely owing to the heterogeneity and plasticity of DPT cells originating from single-positive T cells. Compared with single-positive T cells, the mouse DPT cells show the strongest enrichment for immunosuppressive genes Foxp3 and Ctla4, activation gene Il2ra (Cd25), and integrin Itgb1, along with moderately enriched Itga4 (Figure 4H and Supplemental Figure 11B).

In search of the molecular candidates contributing to DPT-CTC clustering, we analyzed expression of genes within the KEGG cell adhesion molecules pathway genes (n = 598) (43) and identified 32 genes with overlapping expression among mouse and human T and DPT cells, including integrins ITGB1 and ITGA4, SELL, and CD2, among others (Figure 4I). Notably, mouse Itgb1 (protein product Cd29) and Itga4 (protein product Cd49d), which together form the integrin heterodimer, very late antigen 4 (VLA-4), were among the most enriched molecules in DPT cells compared with single-positive T cells (Figure 4, H–J, Supplemental Figure 11C, and Supplemental Figure 12A). The higher expression levels of integrin proteins Cd29 and Cd49d were also observed in DPT cells relative to single-positive T cells from mouse spleen and blood, as demonstrated via flow cytometry analyses (Figure 4K and Supplemental Figure 12, B–D). Consistently, a similar but moderate enrichment of human ITGB1 expression was also observed in blood DPT cells versus CD4+ and CD8+ T cells, based on scRNA-Seq data of breast cancer patients (Supplemental Figure 7E).

To determine the functional importance of these candidate genes in heterotypic T cell interaction with tumor cells, we conducted genetic modulations by transient transfection of Jurkat T cells (or breast cancer cells) that stably express Cas9 with 2 individual synthetic guide RNAs (sgRNAs) for each gene. We first knocked out ITGB1 and SELL in Jurkat cells and confirmed their depletion via flow cytometry (Supplemental Figure 12F, left panels). As sgRNAs were unavailable for CD2, we alternatively knocked out CD58, the cognate ligand for CD2, in MDA-MB-231 breast cancer cells (Supplemental Figure 12F, right panel).

We then performed a surrogate screening test for heterotypic T tumor cell clustering in vitro, with 40,000 Jurkat cells (red) and 10,000 MDA-MB-231 cells (green) added to cell suspensions in a poly-HEMA–coated 96-well plate (Supplemental Figure 12, H and J). As monitored by Incucyte time-lapse imaging of clustering, among all tested gene modulations, ITGB1-KO in Jurkat cells via 2 separate sgRNAs showed the strongest inhibition of heterotypic T tumor cell interactions with differences present as early as 3 hours (Supplemental Figure 12, G and H). One sgRNA for SELL-KO in Jurkat cells also moderately reduced heterotypic interactions, whereas another SELL sgRNA-mediated KO in Jurkat cells or CD58-KO in MDA-MB-231 cells did not significantly impact heterotypic clustering (Supplemental Figure 12, G–J). Therefore, ITGB1 (VLA-4 in partnership with ITGA4) became the top candidate to be further investigated in mouse models in vivo.

VCAM1 is required for CTC-DPT clustering in a spontaneous metastasis model. To identify the ligand(s) in tumor cells that interacts with candidate receptors in DPT cells, we expanded the scRNA-Seq data with a publicly available dataset of peripheral WBCs from breast cancer patients (44) that include a subset of CTCs among the WBC populations (Figure 5A). This dataset includes 138 CTCs detected from 13 metastatic breast cancer patients and 14 CTCs detected from 1 local breast cancer patient (GSE139495). We analyzed the expression of predicted ligands on CTCs that would bind to adhesion molecules identified to be highly expressed in DPT cells (Figure 4I). From the heat map of predicted ligands in CTCs that could interact with the top adhesion molecules expressed in DPTs, we found 2 top candidates, ICAM1 and VCAM1, with higher expression in CTCs from the patients with metastatic breast cancer than those from patients with local disease (Figure 5, B and C). Our previous work demonstrated that ICAM1 promotes cancer stemness and homotypic CTC cluster formation (14). We analyzed the mass spectrometry proteomic datasets of treatment-naive human breast primary tumors (45) (n = 122) and found a higher expression of the VCAM1 protein in Black patients with breast cancer versus non-Black patients (Figure 5D). As VCAM1 is a known ligand of VLA-4 and is involved in breast cancer metastasis (46, 47), we hypothesized that VCAM1 contributes to tumor cell–T cell interactions.

Targeting VCAM1 inhibits spontaneous lung metastasis and CTC-DPT clusters in vivo. (A) UMAP plot of CTCs and WBCs analyzed from a scRNA-Seq dataset of patients with breast cancer. (B and C) Heatmap of ligand adhesion molecules (B) and gene expression bar graph of VCAM1 and control genes GAPDH, MKI67, and EPCAM via scRNA-Seq (C) in CTCs isolated from the patients with local and metastatic (Met) breast cancer (BC). Two-tailed unpaired t test. N = 138 for Met BC group and 14 for local BC group. (D) Plot of proteomic VCAM1 in nontreated breast cancer tumors of Black and non-Black patients. Unpaired, 1-tailed nonparametric t test. N = 14 for Black patients and 108 for non-Black. Black line on graph indicates median value. (E) Left: Flow histograms of mouse 4T1 tumor cells, WT, and Vcam1-KO. Right: Schematic depicting experimental design of orthotopic implants of eGFP+ 4T1 (WT and Vcam1-KO) tumors in BALB/c mice and subsequent analyses of tumor burden, CTCs, and lung metastasis. (F) Bar graphs of 4T1 primary tumor volumes, WT and Vcam1-KO, on day 9 (NS = not significantly changed, P > 0.05) prior to eGFP immunogen–triggered immune attacks in mice. N = 16 tumors. Two-tailed unpaired t test. (G) Bar graphs of lung metastatic signals of eGFP+ 4T1 cells detected via ex vivo fluorescence imaging of dissected lungs (left) and flow cytometry of dissociated eGFP+ cells from the mouse lungs (right). N = 3 for WT and 6 for KO. Unpaired, 2-tailed t test P values are displayed. (H) Bar graphs showing CTC-WBC clusters, CTC-T cell clusters, and CTC-DPT cell clusters in peripheral blood of mice with 4T1-NT control or VCAM1-KO tumors. N = 11 for WT and 14 for KO. Unpaired, 2-tailed t test P values are displayed.