Pericytes promote metastasis by regulating tumor local vascular tone and hemodynamics

Abstract

Hemodynamics are important for survival and extravasation of circulating tumor cells, whereas the effects of hemodynamics on tumor cell intravasation remain mostly unknown. Here, we show that colorectal cancer patients with liver metastasis are characterized by increased diameter and blood flow in the primary tumor compared with non-metastatic patients. A subpopulation of NKX2-3^high tumor pericytes (TPCs) in the primary tumor is associated with hematogenous metastasis of patients. Mechanistically, long noncoding RNA NEAT1-enriched extracellular vesicles induce NKX2-3 upregulation in TPCs. NKX2-3 suppresses calcium influx in TPCs via PDE1C/cAMP/PKA signaling axis, inducing tumor vasodilation and increasing blood flux and vascular leakage. Genetic deletion of Nkx2-3 or pharmacological blocking the transcriptional activity of NKX2-3 in TPCs with designed peptide induce tumor local vasoconstriction and decrease blood flow to mitigate tumor metastasis. These findings reveal that TPCs-regulated vasodilation and hemodynamics facilitate tumor metastasis, and provide a potential prognostic marker and therapeutic strategy for tumor metastasis.

Tumor pericytes are located surrounding blood vessels, and can regulate tumor vascular structure. Here, the authors discover that a subpopulation of NKX2-3 high tumor pericytes modulates vasodilation and hemodynamics to promote metastasis.

Introduction

Distant metastasis is the primary cause of cancer death, and hematogenous metastasis is the main mechanism of tumor distant metastasis¹. Hematogenous metastasis mainly includes the steps of tumor cell intravasation into blood vessels, establishment of circulating tumor cells (CTCs), tumor cell extravasation into secondary organs and colonization of metastatic foci². Among these steps, intravasation of tumor cells into blood vessels is an early step in hematogenous metastasis³. Existing studies have focused mainly on the effects of tumor cells, endothelial cells (ECs) and macrophages on tumor cell intravasation⁴. However, the mechanisms underlying tumor cell intravasation remain elusive, and there is still a lack of effective therapeutic drugs and strategies⁵. Exploring the mechanisms mediating tumor cell intravasation from a new perspective could be beneficial for the identification of novel therapeutic targets.

Accumulating evidence has shown that hemodynamic parameters, including blood flow and fluid shear stress, play an important role in the regulation of hematogenous metastasis^6,7. The high blood flow velocity and the corresponding increase in fluid shear stress can dissociate CTC clusters and lead to tumor cell anoikis^8,9. In addition, mechanical forces make it difficult for disseminated tumor cells to attach to the vascular lumen in metastatic lesions, which inhibits tumor cell extravasation¹⁰. However, the effects of local hemodynamics in primary tumor tissues on tumor cell intravasation are still poorly studied.

Tumor vessels are neocapillaries with chaotic, tortuous, and incomplete structures, which are mainly composed of ECs and pericytes¹¹. Pericytes are contractile cells that are wrapped around ECs and embedded within the walls of capillaries, which have been demonstrated to play an important role in maintaining capillary stability and regulating hemodynamics including blood flow and velocity in non-tumor diseases such as cerebrovascular disease and retinopathy^12,13. Specifically, pericytes regulate microcirculatory blood flow by modulating vascular tone and diameter^14,15. In tumors, as the gatekeeper, tumor pericytes (TPCs) regulate tumor metastasis mainly by modulating vascular structure including integrity and permeability of tumor vessels¹⁶. However, the effects of TPCs on tumor local vascular tone and hemodynamics remain unclear.

Here, we found that vascular diameter and blood flow were significantly increased in the primary tumor of colorectal cancer liver metastasis (CRCLM) patients. NKX2-3, a homeobox transcription factor (TF)¹⁷, was highly-expressed in TPCs derived from CRCLM patients and positively associated with hematogenous metastasis of cancer patients. The overexpression of NKX2-3 in TPCs was induced by tumor derived extracellular vesicles (EVs) via long noncoding RNA (lncRNA) NEAT1/miR-769-5p pathway. NKX2-3 suppressed TPC contraction by hindering the activation of calcium channels via a PDE1C/cAMP/PKA-dependent signaling axis, thereby inducing tumor local vasodilation and increasing blood flow and vessel leakage. Conditional knockout of Nkx2-3 or specific blockade of the transcriptional activity of NKX2-3 in TPCs by peptides restored the local vasoconstriction and hemodynamics, thus ameliorating hematogenous metastasis of tumor cells.

Results

Hemodynamics in primary tumor are associated with CRC metastasis

To investigate the association of hemodynamic changes with tumor metastasis in clinic, doppler ultrasound images for blood flow in the primary tumor of CRC patients with or without metastasis were collected, and the signal intensity of blood flow was calculated by Open CV. Our results showed that blood flow was significantly increased in CRC patients with liver metastasis compared with that in non-metastatic patients (Fig. 1a), further analysis of blood flow signal in the primary tumor of patients at different stages revealed that the blood flow signals were increased in patients at stage IV compared with that in patients at stage I/II (Supplementary Fig. 1a). Additionally, the signal intensity of blood flow achieved moderate performance in predicting CRCLM, with the area under the receiver operating characteristic (ROC) valued of 0.69 (Fig. 1b). Similar results were obtained by computed tomography (CT)-based radiomics. The signal of blood flux was significantly increased in the primary tumor of CRC patients with liver metastasis at stage IV (Fig. 1c, Supplementary Fig. 1b) and the area under ROC was 0.74 (Fig. 1d), indicating that blood flow signal achieved a fair predictive value of CRCLM. Further correlation analysis revealed that both signal strength value (Supplementary Table 1) and small area emphasis value (Supplementary Table 2) were significantly correlated with CRCLM, rather than other clinical parameters. The velocity and flux of blood flow are always determined by vascular tone and structure¹⁸. Immunofluorescence analysis revealed that microvessel diameter, length and surface area in the primary tumor were significantly increased in CRCLM patients, while no significant difference was observed in microvessel density (Fig. 1e). Additionally, the integrity of vascular basement membrane was attenuated in the primary tumor of CRCLM patients (Fig. 1f), which was accompanied with augmented hypoxia (Fig. 1g). Pearson correlation analysis indicated microvessel diameter rather than microvessel density was positively correlated, whereas the basement membrane integrity was negatively correlated with the blood flow signal (Fig. 1h–j). However, there was no significant correlation between blood flow signal and tumor hypoxia (Fig. 1k). These data suggest that the increased blood flow in the primary tumor is closely associated with CRCLM. However, the regulatory effects of hemodynamic changes on tumor metastasis remain unclear.

Fig. 1. Hemodynamic changes in primary tumor are associated with colorectal cancer liver metastasis.

a Doppler ultrasound images of blood flow signal derived from the primary tumor tissues of NM CRC (n = 45) and LM CRC patients (n = 45). The quantification of blood flow signal is shown (right). b ROC curve analysis of the blood flow signal in CRC patients (n = 90). c Representative CT images and quantification for blood flow signal in NM CRC (n = 49) and LM CRC patients (n = 44). d ROC curve analysis of the blood flow signal (small area emphasis value) in CRC patients (n = 93). Yellow circles indicate primary tumor region. e Immunofluorescence analysis of microvessel density, diameter, length, and surface area in the primary tumor of NM CRC (n = 49) and LM CRC patients (n = 44). Scale bar, 20 μm. f Immunofluorescence analysis of laminin⁺/VWF⁺ area per field in the primary tumor of NM CRC (n = 49) and LM CRC patients (n = 44). Scale bar, 20 μm. g Immunofluorescence analysis of CA9⁺/tumor area in the primary tumor regions derived from NM CRC (n = 49) and LM CRC patients (n = 44). Scale bar, 20 μm. Pearson correlation analysis of microvessel density (h), microvessel diameter (i), % of laminin⁺/VWF⁺ area (j), and % of CA9⁺/tumor area (k) with the signal of blood flow (small area emphasis value) in (c) (n = 93). Each sample in the violin plots represents data from an individual patient. P values were determined by two-tailed unpaired t-test, n denotes the number of patients. AUC area under curve, CRC colorectal cancer, CT computed tomography, LM CRC liver-metastatic colorectal cancer, NM CRC non-metastatic colorectal cancer, ROC receiver operating characteristic.

NKX2-3 in TPCs is positively correlated with tumor metastasis

Pericytes are wrapped around capillaries and known to regulate constriction and blood perfusion of vessels¹⁹. Therefore, we speculate whether tumor pericytes are involved in the regulation of tumor local vascular tone and hemodynamics, leading to tumor metastasis. TPCs derived from the primary tumor tissues of CRC patients with or without liver metastasis were analyzed by scRNA-seq in our previous study²⁰. In total, TPCs were divided into 13 distinct clusters with significantly enriched genes (Supplementary Fig. 2; Supplementary Fig. 3a), among which, cluster 3 had the most significant increase in cell abundance, and was almost exclusively identified in CRCLM samples, therefore termed as metastasis-associated pericytes (MAPs) (Fig. 2a; Supplementary Fig. 3b). Gene ontology (GO) analysis revealed that the upregulated genes enriched in MAPs were associated with immune response, extracellular matrix organization and blood vessel development (Fig. 2b). To further investigate the mechanisms underlying TPC gene expression, regulon activities to each TPC cluster were identified (Supplementary Fig. 3c), while NKX2-3 ranked the highest in MAPs (Fig. 2c) and was highly-expressed in MAPs (Fig. 2d). Next, the expression of NKX2-3 was compared between TPCs derived from non-metastatic (TPC_NM) and liver metastatic (TPC_LM) CRC patients. First, the specificity of NKX2-3 antibody was determined using IgG (Supplementary Fig. 4a, b). The expression of NKX2-3 was significantly increased in TPC_LM compared with that in TPC_NM and other cells in tumor microenvironment (Fig. 2e, f; Supplementary Fig. 4c)_. In addition, the proportion of NKX2-3⁺ TPCs was elevated in the primary tumor sections derived from CRC patients with liver metastasis (Fig. 2g). However, NKX2-3 was undetectable in TPCs of normal tissues adjacent to tumors (Supplementary Fig. 4d) and hepatic metastatic nodules derived from CRCLM patients (Supplementary Fig. 5). Moreover, NKX2-3 was selectively expressed in TPCs-wrapped capillaries rather than Myh11⁺SM22α⁺ smooth muscle cells (SMCs) covered on arterials (Supplementary Fig. 6). Correlation analysis revealed that NXK2-3⁺ TPC ratio was significantly correlated with CRCLM (Supplementary Table 3), and the NKX2-3⁺ TPC ratio had high diagnostic efficiency for CRCLM at the optimal cutoff value of 44% (Fig. 2h). Additionally, increased NKX2-3⁺ TPC ratio was associated with worse overall survival (OS) (Fig. 2i) and disease-free survival (DFS) (Fig. 2j) in CRC patients. Univariate analysis indicated that M0/M1 stage and NKX2-3⁺ TPC ratio were identified as factors significantly correlated with prognosis (Supplementary Table 4). Similar results were observed in primary tumor of breast cancer patients with pulmonary metastases (Supplementary Fig. 7a) and lung cancer patients with hematogenous metastasis (Supplementary Fig. 7b), NKX2-3⁺ TPC ratio exhibited a high value for predicting hematogenous metastasis of lung cancer (Supplementary Fig. 7c), and increased NKX2-3⁺ TPC ratio was associated with poor OS and DFS in lung cancer patients (Supplementary Fig. 7d, e). These data suggest that NKX2-3 in TPCs could be a risk factor for tumor hematogenous metastasis and associated with poor prognosis.

Fig. 2. NKX2-3 is highly expressed in MAPs and related to colorectal cancer liver metastasis.

a t-SNE visualization of TPC subsets derived from primary tumor tissues of CRC patients with (n = 2) or without (n = 2) liver metastasis. b GO analysis of MAPs-enriched genes. c Dot plots of the regulon activities in MAPs. d t-SNE visualization of NKX2-3 expression in all TPC clusters. e Western blot assay for NKX2-3 expression in TPC_NM and TPC_LM. f qPCR analysis of NKX2-3 expression in TPC_NM and TPC_LM (n = 3, n denotes the number of independent experiments). g Immunofluorescence staining and quantification of the NKX2-3⁺ TPC ratio in tumor sections from CRC patients (n = 93). The white arrows indicate NKX2-3 staining in TPCs. Scale bar, 20 μm. Each sample in the violin plots represents data from an individual patient. h ROC curve analysis of the NKX2-3⁺ TPC ratio in CRC patients (n = 93). Kaplan–Meier analysis of OS (i) and DFS (j) in CRC patients with a high or low NKX2-3⁺ TPC ratio (optimal cutoff value of 44%, n = 93). P values were determined by two-tailed unpaired t-test (f, g), log-rank (Mantel-Cox) test (i, j), n denotes the number of patients. AUC area under curve, CRC colorectal cancer, GO gene ontology, MAPs metastasis-associated pericytes, OS overall survival, DFS disease-free survival, ROC receiver operating characteristic, TPC tumor pericyte. Western blot samples derive from the same experiment but different gels for NKX2-3, another for GAPDH was processed in parallel.

NKX2-3 regulates TPC contraction by suppressing calcium signaling

Given that NKX2-3^highTPCs were associated with tumor metastasis, we sought to investigate the function and underlying mechanisms of NKX2-3 in TPCs. Chromatin immunoprecipitation (ChIP)-seq was conducted on TPC_NM^NKX2-3 (Supplementary Fig. 8a, b), and the peak genes of NKX2-3 in TPCs were mainly enriched in calcium signaling pathway (gene number was 110) (Fig. 3a). NKX2-3 overexpression in TPCs suppressed the calcium influx stimulated by L-type Ca²⁺ agonist, (S)-(-)-Bay K8644 (Fig. 3b), while knockdown of NKX2-3 in TPCs increased cytosolic calcium concentration (Supplementary Fig. 8c, d; Supplementary Fig. 9a). Then, RNA-seq was performed on TPC_LM^gNKX2-3 vs TPC_LM^gNC, and 637 down-regulated genes were found in TPC_LM^gNKX2-3. To investigate the molecular mechanisms by which NKX2-3 in TPCs regulates calcium signaling, the overlapping genes between NKX2-3 peak genes-associated with calcium signaling pathway (110) and the down-regulated genes in TPC_LM^gNKX2-3 (637) were defined by Venn diagram. Five overlapping genes including PDE1A, PDE1C, SLC8A1, CAMK2A and CAMK2D were found between these two lists (Fig. 3c). The scRNA-seq data also showed that the expression of PDE1C and PDE1A was higher in the MAP cluster (Supplementary Fig. 9b). Among these genes, the promotors of PDE1C and PDE1A could be bound to NKX2-3 (Fig. 3d) and regulated by NKX2-3 (Fig. 3e; Supplementary Fig. 9c). Interestingly, only knockdown of PDE1C in TPC_LM significantly induced L-type calcium currents (Fig. 3f; Supplementary Fig. 9d–g). PDE1C is a phosphodiesterase that hydrolyzes cAMP to AMP and subsequently suppresses PKA-induced calcium transits^21,22(Fig. 3g). Our results showed that NKX2-3 overexpression in TPCs induced the degradation of cAMP, which was reversed by siRNA targeting PDE1C and accompanied by increased calcium transits (Fig. 3h, i; Supplementary Fig. 9h). Furthermore, inhibiting the activity of PKA with H89 2HCl in TPC_NM^{NKX2-3+siPDE1C} reduced calcium currents (Fig. 3j). Similarly, overexpression of PDE1C hindered cAMP production induced by NKX2-3-knockdown in TPCs, and the cytosolic calcium concentration was decreased in TPC_LM^{siNKX2-3+PDE1C} compared with TPC_LM^{siNKX2-3+Vector} (Supplementary Fig. 9i–k). However, stimulation of cAMP production with forskolin in TPC_LM^{siNKX2-3+PDE1C} increased calcium currents, which could be reversed by treatment with H89 2HCl (Supplementary Fig. 9l). These data indicate that NKX2-3 in TPCs suppresses calcium currents through the PDE1C/cAMP/PKA signaling axis.

Fig. 3. NKX2-3 in TPCs inhibits cell contraction through suppressing the PDE1C/cAMP/PKA/Ca²⁺ axis.

a KEGG analysis of NKX2-3 peak genes derived from ChIP-seq in TPC_NM^NKX2-3. b Fluo-4AM fluorescence images and intensities in TPC_NM^Vector and TPC_NM^NKX2-3. Scale bar, 20 μm. c Venn diagram showing the number of overlapped genes derived from NKX2-3-regulated genes and NKX2-3 peak genes associated with calcium signaling pathway. RNA-seq was performed on TPC_LM^gNKX2-3 (n = 3) vs TPC_LM^gNC (n = 3). d ChIP-qPCR analysis for the binding of NKX2-3 to the promoters of the indicated genes. e qPCR analysis for gene expression in NKX2-3-knockdown TPCs_. f Differential changes of Fluo-4AM intensity in PDE1C-knockdown TPC_LM. g Schematic diagram showing the intracellular PDE1C/cAMP/PKA/calcium signaling axis. Created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp. h ELISA analysis of the cAMP concentration in NKX2-3-overexpressing TPCs with PDE1C-knockdown. i Fluo-4AM fluorescence analysis of intracellular calcium in NKX2-3-overexpressing TPCs with PDE1C-knockdown. j Fluo-4AM fluorescence assay for the intracellular calcium in TPC_NM^{NKX2-3+siPDE1C} treated with or without H89 2HCl (10 μM). k Immunofluorescence analysis of F-actin expression in TPC_NM^Vector and TPC_NM^NKX2-3 (n = 3, n denotes the number of independent experiments). TPC_NM^NKX2-3 were transfected with siPDE1C and treated with verapamil (5 μM). Scale bar, 20 μm. l Representative images and quantification of collagen contraction induced by the indicated TPC_NM^Vector and TPC_NM^NKX2-3. m Western blot analysis of MLC2 and p-MLC2^Ser15 expression in TPC_NM^Vector and TPC_NM^NKX2-3 as indicated in k. Data are presented as mean ± SEM (n = 3, n denotes the number of samples each group). P values were determined by Benjamini-Hochberg method following a two-sided hypergeometric test (a); by two-tailed unpaired t-test (d); by one-way ANOVA followed by Tukey’s post hoc test (e, h, l). BayK, (S)-(-)-Bay K8644; KEGG, kyoto encyclopedia of genes and genomes; NC negative control, TPC tumor pericyte. Western blot samples derive from the same experiment but different gels for MLC2, another for p-MLC2^Ser15, and another for GAPDH was processed in parallel.

Intracellular calcium oscillation is a key regulator of perivascular cell contraction, which facilitates the phosphorylation of myosin light chain (MLC) and induces vasoconstriction^23,24. Indeed, our results demonstrated that NKX2-3-overexpression suppressed TPC contraction, as indicated by the shape transition from fusiform to rhomboid (Fig. 3k), decreased collagen contraction ability (Fig. 3l), and p-MLC2^Ser15 expression compared with TPC_NM^Vector (Fig. 3m). As expected, PDE1C knockdown facilitated TPC contraction compared with that of TPC_NM^NKX2-3+siNC, while pharmacological inhibition of the PKA activity or calcium channel (Supplementary Fig. 10a) reversed the contraction of TPC_NM^{NKX2-3+siPDE1C} (Fig. 3k–m; Supplementary Fig. 10b-d). In contrast, NKX2-3 knockdown facilitated TPC contraction, which could be reversed by PDE1C overexpression (Supplementary Fig. 10e, f). However, stimulation of cAMP in TPC_LM^{gNKX2-3+PDE1C} accelerated TPC contraction compared with that of untreated TPCs, which was suppressed by H89 2HCl and re-activated by the Ca²⁺ agonist (Supplementary Fig. 10g–i). Nonetheless, NKX2-3 in TPCs had negligible effects on the contraction of ECs (Supplementary Fig. 11). Taken together, these data indicate that NKX2-3 suppresses TPC contraction by hindering the activity of L-type calcium channel via a PDE1C/cAMP/PKA-dependent signaling axis.

Pericyte-specific deletion of Nkx2-3 inhibits tumor metastasis

Given that NKX2-3 in TPCs was associated with tumor metastasis, the effects of pericyte-NKX2-3 on tumor metastasis and its related underlying mechanisms were further studied in vivo. Cspg4-CreERT mice were crossed with ROSA26-loxP-stop-loxP-tdTomato mice to generate lineage tracing mice (PC^lin mice), and then PC^lin mice were crossed with Nkx2-3^flox/flox mice to generate mice with pericyte-Nkx2-3 specific knockout (PC^lin-KO mice) (Supplementary Fig. 12; Supplementary Fig. 13a). Tamoxifen administration resulted in labeling of TPCs with tdTomato fluorescence, which was similar between PC^lin and PC^lin-KO mice (Supplementary Fig. 13b), moreover, injection of tamoxifen led to depletion of NKX2-3 in TPCs derived from PC^lin-KO mice (Supplementary Fig. 13c). To investigate the effects of NKX2-3 in TPCs on CRC metastasis, MC-38-luc-LM3 cells were injected into the cecum wall of PC^lin and PC^lin-KO mice and detected by bioluminescence imaging. Depletion of Nkx2-3 in TPCs had negligible effects on non-tumor tissues (Supplementary Fig. 13d) and tumor growth, whereas significantly suppressed CRCLM (Fig. 4a; Supplementary Fig. 14a), as indicated by the decreased liver-metastatic foci (Fig. 4b) and CTCs (Supplementary Fig. 14b, c). Furthermore, pericyte-Nkx2-3 deletion suppressed pulmonary metastasis of melanoma (Supplementary Fig. 15). The effects of NKX2-3 in TPCs on lymph node metastasis was also investigated, while no significant differences in lymph node metastasis were observed between PC^lin and PC^lin-KO group (Supplementary Fig. 16). These data reveal that NKX2-3 in TPCs facilitates tumor hematogenous metastasis.

Fig. 4. Pericyte-specific deletion of Nkx2-3 facilitates tumor local vasoconstriction and inhibits tumor metastasis.

a Bioluminescence assay for tumor growth in PC^lin and PC^lin-KO mice orthotopic injected with MC-38-luc-LM3 cells. The luminescence of primary tumor was quantified. b H&E staining and quantification for the number and area of liver metastatic foci in tumor-bearing PC^lin and PC^lin-KO mice. The yellow and black dotted lines indicate the metastatic foci. Scale bar, 1 mm. c Representative images of orthotopic tumors post tissue clearing and stained tumor vessels derived from PC^lin and PC^lin-KO mice (n = 3, n denotes the number of independent experiments). Scale bar, 2 mm. The white arrows indicate tumor vessels with different diameters. d Immunofluorescence analysis of microvessel density and diameter in orthotopic tumors. Scale bar, 25 μm. The white lines indicate vessel diameters. e Schematic diagram of the vasoconstriction assay performed by two-photon excited fluorescence microscopy (left, created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp). Representative images and quantification of tumor vessel diameter after exposure to U46619 for 10 min (right). Scale bar, 5 μm. The white arrows and lines indicate vessel diameters. f Schematic depicting the experimental design, whereby PC^lin and PC^lin-KO mice were intravenously injected with AAV expressing GCaMP6 to indicate intracellular calcium signaling, and tumor vessels were exposed to U46619 and detected by intravital imaging. Created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp. g Representative images and quantification of GCaMP6 expression in TPCs from PC^lin-GCaMP6 and PC^{lin-KO-GCaMP6} mice treated with U46619. Scale bar, 20 μm. Yellow circles indicate GCaMP6 expression in TPCs, white arrows and lines indicate vascular diameter in mouse tumor. Data are presented as mean ± SEM (n = 6, n denotes the number of samples each group). P values were determined by two-tailed unpaired t-test. PC pericyte.

Since NKX2-3 in TPCs suppressed cell contraction. We therefore investigated whether NKX2-3 in TPCs induced tumor metastasis by modulating vascular tone. Light-sheet and confocal fluorescence microscopy were applied to analyze vascular morphology and structure. Conditional deletion of Nkx2-3 had negligible effects on pericyte coverage (Supplementary Fig. 17a) and vessel density (Fig. 4d). However, the architecture of vascular network in the primary tumor of PC^lin-KO mice was different from that in PC^lin mice, as characterized by a significantly reduced mean vascular diameter compared to that in PC^lin mice (Fig. 4d), which might be resulted from the reduced proportion of vascular with large diameter, accompanied by an increased proportion of small-diameter vessels (Supplementary Fig. 17b), similar results were obtained by optical coherence tomography angiography (OCTA) analysis (Supplementary Fig. 18a). Then, vasoconstriction and vasodilation in the primary tumor of mice were examined by two-photon microscopy. U46619 is thromboxane A (2) analog, 9,11-dideoxy-9α,11α-methanoepoxy prostaglandin F2α, serving as a vasoconstrictor for capillaries¹⁹. Our results showed that the capillary diameter was reduced by ~20% in PC^lin-KO mice after U46619 treatment; however, no obvious change was observed in that from PC^lin mice (Fig. 4e), indicating that NKX2-3 in TPCs induces vasodilation. Moreover, the intracellular calcium level of TPCs was determined by calcium indicator GCaMP6, which was driven by the promoter of Cspg4 in TPCs (Fig. 4f). Our results showed that the response of intracellular Ca²⁺ to U46619 was significantly increased in PC^{lin-KO-GCaMP6} mice compared with PC^lin-GCaMP6 mice (Fig. 4g). Taken together, NKX2-3 in TPCs induces vasodilation and tumor metastasis by suppressing calcium channel activity.

NKX2-3 in TPCs increases blood flow and vessel permeability

Since pericyte contraction and relaxation can either directly regulate blood flow dynamics through vasoconstriction or dilation¹⁴ or indirectly affect vascular leakage via opening the endothelial junctions²⁵, we sought to determine whether NKX2-3 in TPCs exerts effects on hemodynamics and vessel permeability in primary tumor. Blood velocity was assessed by two-photon imaging, which was significantly attenuated in PC^lin-KO mice compared with PC^lin mice (Fig. 5a). Additionally, tumor blood velocity and vascular diameter were significantly decreased in PC^lin-KO mice after exposed to U46619 (Supplementary video 1) compared with treatment before (Supplementary video 2); however, no significant difference was observed in PC^lin mice before (Supplementary video 3) and post-treatment with U46619 (Fig. 5b; Supplementary Fig. 18b, c; Supplementary video 4). In addition, MRI imaging was conducted on PC^lin and PC^lin-KO mice, our results showed that deletion of Nkx2-3 in TPCs induced a reduction in perfusion, as indicated by the decreased D* and f values (Fig. 5c). Consistent with the MRI experiments, laser speckle contrast imaging showed that microvascular blood flow was weakened in PC^lin-KO mice compared with PC^lin mice (Fig. 5d). Deletion of Nkx2-3 in TPCs augmented the tight junction between ECs, as indicated by the increased expression of ZO-1 (Fig. 5e). Moreover, vessel permeability was significantly lower in PC^lin-KO mice than that in PC^lin mice (Fig. 5f; Supplementary Fig. 18d). To further investigate whether vasodilation observed in PC^lin mice is favorable for tumor cell intravasation, microfluidic vascular chip models (Fig. 5g) and mouse vessel organoids (Supplementary Fig. 19a) were constructed. Our results showed that compared to TPC_NM^Vector, TPC_NM^NKX2-3 significantly facilitated transmigration of HCT116 cells through vascular barrier (Fig. 5h). Similarly, the number of intravasated MC-38 cells was significantly increased in vessel organoids derived from PC^lin mice compared with those from PC^lin-KO mice (Supplementary Fig. 19b). Additionally, in vivo experiments showed that the green fluorescent protein (GFP)-tagged tumor cells were quickly passed through the vascular wall and entered the circulation post-treatment with a vasodilator (Fig. 5i), indicating that vasodilation facilitates tumor cell intravasation and metastasis. Taken together, these results indicate that NKX2-3 in TPCs is a key regulator of capillary blood flow and velocity as well as vessel leakiness, thus paving a way for tumor cell metastasis.

Fig. 5. NKX2-3 in TPCs facilitates tumor metastasis by increasing blood flow and vessel permeability.

a Representative two-photon images showing the dextran signal in blood vessels derived from PC^lin and PC^lin-KO mice (the area denoted with a yellow box is enlarged). The midline of each vessel (yellow line) was chosen as the measurement plane for a continuous line scan (up), and the direction of blood flow is indicated by red arrow. Representative images of the line scan and quantification of blood velocity (bottom). Scale bar, 100 μm.b Intravital imaging to evaluate the change of RBC velocity and vessel diameter post-U46619 treatment in orthotopic tumors derived from PC^lin and PC^lin-KO mice. Scale bar, 20 μm. c MRI images of tumors derived from PC^lin and PC^lin-KO mice (left). Quantification of the f and D* values is shown (right). The white rings indicate the tumors. Scale bar, 10 mm. d Laser speckle contrast imaging and quantification of blood flux in PC^lin and PC^lin-KO mice. The yellow and white circles indicate the tumor area. Scale bar, 2 mm. e Immunofluorescence staining and quantification of ZO-1 (red) expression on CD31 positive endothelial cells derived from orthotopic tumor sections. Scale bar, 20 μm. f Immunofluorescence analysis of FITC-dextran 40 kDa (green) surrounding vessels labeled with CD31 in orthotopic tumor sections. Scale bar, 20 μm. g Schematic diagram describing the microfluidic vessel chip. Created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp. h Immunofluorescence assay for the transendothelial migration of HCT116 cells. White dotted lines indicate the porous membrane between the top channel and bottom channel. Data are presented as mean ± SEM (n = 3, n denotes the number of samples each group). Scale bar, 200 μm. i Intravital imaging evaluating tumor cell intravasation in PC^lin-KO mice before and after exposure to 10 μM nitroprusside (n = 3, n denotes the number of independent experiments). Scale bar, 20 μm. Data are presented as mean ± SEM (n = 6, n denotes the number of samples each group). P values were determined by two-tailed unpaired t-test. PC pericyte, RBC red blood cell.

Tumor-derived EVs upregulate NKX2-3 expression in TPCs through lncRNA NEAT1/miR769-5p

Accumulating evidence has demonstrated that EVs are important mediators of intercellular communication in the tumor microenvironment, while lncRNAs are selectively packaged into EVs and taken up by recipient cells²⁶. LncRNA NEAT1 was found to be one of the highly-expressed lncRNAs in MAPs (Fig. 6a). NEAT1 upregulated the expression of NKX2-3 both in the mRNA and protein level (Fig. 6b, c; Supplementary Fig. 20a–c). LncRNAs may function as competing endogenous RNAs (ceRNAs) to sponge microRNAs (miRNAs), thereby suppressing the functions of miRNAs and reducing their regulatory effects on target mRNAs²⁷. Bioinformatic analysis of the miRNA recognition sequence in NEAT1 was performed with miRDB, miRTarBase, TargetScan and miRWalk databases, and a recognition sequence for miR-769-5p was identified. LncRNA NEAT1 could bind to the miR-769-5p (Supplementary Fig. 20d) and suppress the level of miR-769-5p (Fig. 6d). Similarly, miR-769-5p bound to NKX2-3 (Fig. 6e, f) and miR-769-5p mimic reversed the NEAT1-induced NKX2-3 upregulation (Fig. 6g, h). On the contrary, miR-769-5p inhibitor induced the expression of NKX2-3 (Supplementary Fig. 20e, f) and reversed the downregulation of NKX2-3 mediated by NEAT1 knockdown (Supplementary Fig. 20g, h). These results indicate that NEAT1 in MAPs induces NKX2-3 expression by regulating miR-769-5p.

Fig. 6. Tumor derived EVs induce the expression of NKX2-3 in TPCs through the lncRNA NEAT1/miR-769-5p signaling axis.

a Box plot for the level of NEAT1 in all TPC subsets. Box plots show the median (center line), 25th/75th percentiles (box bounds), smallest/largest values within 1.5×IQR (whiskers), and outliers beyond whiskers (dots). b, c qPCR and Western blot analysis for NKX2-3 expression in NEAT1-overexpressing TPC_NM. d miR-769-5p level in NEAT1-overexpressing TPC_NM. e Schematic representation of the miR-769-5p binding sties in NKX2-3. f Luciferase reporter assay evaluating the binding of the miR-769-5p mimic to NKX2-3. g qPCR analysis of miR-769-5p and NKX2-3 in TPCs treated with the miR-769-5p mimic. h Protein expression of NKX2-3 in the indicated TPCs. i Representative TEM phase images of DLD-1 EVs (n = 3, n denotes the number of independent experiments). Scale bar, 100 nm. j Enrichment diameter distribution of DLD-1 EVs. k Immunofluorescence analysis for F-actin in TPC_NM treated with DLD-1 EVs. Scale bar, 10 μm. l, m Gene and protein expression in TPC_NM treated with DLD-1 EVs. n Protein expression in Rab27A-knockdown DLD-1 cells. Western blot samples derive from the same experiment but different gels for Rab 27A, another for GAPDH was processed in parallel. o Schematic diagram of the co-culture model (left). Gene expression in TPC_NM treated with EVs derived from Rab27A-knockdown DLD-1 cells (right). Created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp. p FISH analysis of NEAT1 expression in primary tumor of CRC patients. The white arrows indicate NEAT1 staining in TPCs. Scale bar, 20 μm. Each sample in the violin plots represents data from an individual patient (n = 15 each group). q qPCR analysis of NEAT1 in the blood EVs derived from CRC patients (n = 5 each group). Data are presented as mean ± SEM (n = 3, n denotes the number of samples each group). P values were determined by two-tailed unpaired t-test (b, d, f, l, p, q); one-way ANOVA followed by Tukey’s post hoc test (g, o). CTL control, EVs extracellular vesicles, FISH fluorescence in situ hybridization, IQR interquartile range, Mut mutant, NC negative control, TPC tumor pericyte, Wt wildtype. Western blot samples derive from the same experiment but different gels for NKX2-3, another for GAPDH was processed in parallel.

To further investigate whether the overexpression of NEAT1 in MAPs is caused by EVs derived from tumor cells. EVs were purified and identified from DLD-1 cells (Fig. 6i, j; Supplementary Fig. 20i). EVs released from NEAT1-overexpressing DLD-1 cells could be taken up by TPCs (Fig. 6k; Supplementary Fig. 20j), leading to the decreased miR-769-5p (Fig. 6l) and increased NKX2-3 level in a dose-dependent manner (Fig. 6m). EVs derived from NEAT1-knockdown DLD-1 cells showed the opposite effects (Supplementary Fig. 20k–m). Additionally, interference with the production of EVs in DLD-1 cells by Rab27A knockdown decreased the expression of NEAT1 and NKX2-3 in TPCs, while the level of miR-769-5p was increased (Fig. 6n, o). Subsequently, the level of NEAT1 was detected in the primary tumor and blood circulation of human CRC, compared to non-metastatic CRC patients. We found that NEAT1 was highly enriched in tumor cells and TPCs (Fig. 6p), as well as the EVs of blood circulation (Fig. 6q) of CRCLM patients. Correlation analysis revealed that NEAT1 expression in TPCs was positively correlated with that in CRC cells (Supplementary Fig. 20n), as well as microvessel diameter (Supplementary Fig. 20o) and blood flow (Supplementary Fig. 20p). However, NEAT1 expression in TPCs was negatively associated basement membrane integrity (Supplementary Fig. 20q). Taken together, these data indicate that CRC cell-derived NEAT1 induces NKX2-3 expression in TPCs via acting as a ceRNA for miR-769-5p.

Blocking the transcription activity of NKX2-3 suppresses tumor metastasis

We next investigated whether blocking the binding of NKX2-3 to PDE1C can suppress tumor metastasis. Oligopeptides derived from transcription factors can competitively bind to their target DNA sequences, thereby blocking TF binding to DNAs²⁸. Therefore, it might be an effective method to inhibit the binding of NKX2-3 and PDE1C. First, the structural of NKX2-3 was predicted by AlphaFold DB, and the predicted local similarity to NKX2-3 showed that, sequences with p of local distance difference test (pLDDT) > 90 were residues 152-206 (Fig. 7a; Supplementary Fig. 21a). Corresponding to the sequences with pLDDT >90, six oligopeptides (termed as NB1 to NB6) were synthesized (Fig. 7b). Among them, NB2 has the highest binding affinity to PDE1C (Supplementary Fig. 21b), and further blocked the binding of NKX2-3 with the promoter of PDE1C (Fig. 7c; Supplementary Fig. 22a). However, NB2 had negligible effects on NKX2-3 expression (Supplementary Fig. 22b). Therefore, NB2 was selected as the candidate oligopeptide. Fibroblast activation protein α (FAPα), a type II integral membrane serine protease²⁹, was overexpressed in TPC_NM^NKX2-3 but not in HCT116 cells (Supplementary Fig. 22c, d). FAPα possesses endopeptidase activity that can specifically cleave N-terminal benzyloxy carbonyl–blocked (Z-blocked) Gly-Pro (Z-GP) dipeptide-linked substrates³⁰. The specific expression of FAPα in TPCs and dipeptide substrate hydrolytic activity make it an ideal target for TPC-targeting prodrug strategy. To improve the delivery efficiency of NB2 to TPCs, we synthesized a FAPα-activated prodrug, Z-GP-NB2, by coupling NB2 to Z-GP (Fig. 7d). The hydrolysis of Z-GP-NB2 into NB2 was mediated by FAPα in TPCs (Supplementary Fig. 22e–g), and then NB2 penetrated into the TPC nucleus (Supplementary Fig. 22h). NB2 had negligible effects on tumor cell viability and migration (Supplementary Fig. 23a–c), however, both NB2 and the prodrug Z-GP-NB2 restored cAMP production (Supplementary Fig. 23d) and L-type calcium transits (Supplementary Fig. 23e) hindered by NKX2-3 in TPCs, thus facilitating TPC contraction (Supplementary Fig. 23f–h). The effects of Z-GP-NB2 on TPC contraction was dependent on NKX2-3 expression (Supplementary Fig. 23i), but Z-GP-NB2 had negligible effects on the NEAT1-induced NKX2-3 upregulation (Supplementary Fig. 23j).

Fig. 7. Targeting NKX2-3^high TPCs inhibits CRCLM by inducing vasoconstriction.

a Predicted protein structure of NKX2-3 by AlphaFold DB. b Amino-acid sequences encoding oligopeptides NB1-NB6. c MST analysis for the interaction of PDE1C with NB2 and NKX2-3 (n = 3). d Structure of Z-GP-NB2. e Schematic diagram of the animal experiments. Created in BioRender. Pan, J. (2025) https://BioRender.com/life0kp. f Representative images of whole livers and H&E analysis of liver metastatic foci derived from mice bearing HCT116 orthotopic xenografts (n = 6). The yellow and black dotted lines indicate the metastatic foci. Scale bar, 1 mm. g Representative images and quantification of vessel diameter after exposure to U46619 for 10 min (right). Scale bar, 5 μm. The white arrows and lines indicate vessel diameters (n = 6). h Laser speckle assay for blood flow in orthotopic tumors derived from mice treated with or without Z-GP-NB2 (n = 6). i MRI examination of tumors derived from HCT116 xenograft-bearing mice with indicated treatment. Quantification of K^trans values is shown (n = 3). Data are presented as mean ± SEM, n denotes the number of samples each group. P values were determined by one-way ANOVA followed by Tukey’s post hoc test. I.P. intraperitoneal, TPC tumor pericyte.

We further investigated the effects of NB2 and Z-GP-NB2 in vivo. Both of them had negligible effects on non-tumor tissues including heart, spleen and mucosal high endothelial venules (HEVs), however, NB2 exerted toxic effects on mouse kidney and lung tissues (Supplementary Fig. 24a, b). Therefore, Z-GP-NB2 was used to examine the anti-metastatic effects and mechanisms. Z-GP-NB2 inhibited liver metastasis of HCT116 xenografts induced by TPC_NM^NKX2-3 (Fig. 7e, f; Supplementary Fig. 24c). Nonetheless, Z-GP-NB2 had negligible effects on the body weight of tumor-bearing mice (Supplementary Fig. 24d). Additionally, compared to vehicle group, Z-GP-NB2 treatment induced vasoconstriction (Fig. 7g) accompanied by a decreased vascular diameter (Supplementary Fig. 25a), reduced blood flow (Fig. 7h; Supplementary Fig. 25b, c) and blood velocity (Supplementary Fig. 25d–f, Supplementary Fig. 26). Z-GP-NB2 had negligible effects on pericyte coverage (Supplementary Fig. 27a), but attenuated tumor hypoxia (Supplementary Fig. 27b) and increased the integrity of tumor vessels (Fig. 7i; Supplementary Fig. 27c–f). Taken together, these data demonstrate that inhibiting the transcription activity of NKX2-3 in TPCs could ameliorate vessel function and suppress tumor metastasis.

Discussion

The structure and function of tumor blood vessels, especially the hemodynamics, play an important role in tumor hematogenous metastasis⁶. As an important component of tumor vessels, it is still unclear whether TPCs can regulate local hemodynamics in primary tumors and subsequently influence tumor cell intravasation and circulation. NKX2-3 controls the development of salivary glands, teeth and the small intestine, which is also required for the self-renewal of hematopoietic stem cells and B cell maturation^17,31. Recent studies have indicated that NKX2-3 confers susceptibility to gut inflammation³², hematopoietic malignancies³³ and may contribute to CRC through Wnt signaling³⁴. Nonetheless, the expression and function of NKX2-3 in TPCs are uncharacterized. This study revealed a population of NKX2-3^high TPCs facilitated tumor metastasis through inducing tumor local vasodilation and regulating hemodynamics.

Tumor vessels are different from those in healthy tissues, which maintained a tortuous and dilated capillary-like structure^35,36. Here, we found that initial vascular diameter was also different between non-metastatic and liver metastatic CRC, which was due to the effects of NKX2-3 in TPCs. NKX2-3^high TPCs augmented vascular dilation and leakiness, as well as increasing blood flow and tumor hypoxia. Tumor hypoxia is affected by various factors including blood flow and geometry of the vessel, tumor vessels turned to be irregular and leaky in metastatic cancer patients, thus impaired oxygen supply and leading to hypoxia condition³⁷. Concerning vascular leakiness and greater blood flow^38,39, we suggested it could be caused by the fact that, NKX2-3 overexpression induced TPC relaxation, leading to vasodilation and increased blood flow. Meanwhile, TPC relaxation loosened the tight junctions between endothelial cells and increased vascular permeability, thus the tumor vessels in metastatic patients have greater leakiness and blood flow.

The mechanisms underlying pericyte contraction and the subsequent promotion of vasoconstriction have been extensively studied⁴⁰. It has been demonstrated that increased Ca²⁺ influx in pericytes increases the protein level of p-MLC2^Ser15 and RhoA, leading to pericyte contraction^23,24. Although multiple agonists of Ca²⁺ channel including acetylcholine and norepinephrine⁴¹, have been applied to induce vasoconstriction, however, due to the widespread distribution of Ca²⁺ channels, the existing agonists cannot specifically target TPCs. Therefore, it is necessary to activate intracellular Ca²⁺ influx and induce local vasoconstriction with peptides specifically targeting NKX2-3^high TPCs.

Tumor microenvironment modulates the aberrant expression of genes and proteins within cells via various mechanisms⁴². Here, we found tumor-derived EVs induced NKX2-3 upregulation. However, EVs might not the sole determinants of NKX2-3 expression in TPCs. Hypoxia is a potent regulator for tumor metastatic cascade⁴³, and our results showed that hypoxic condition significantly increased the expression of NKX2-3 in TPCs (Supplementary Fig. 28), which is consistent with a previous study indicating that pericyte relax when oxygen is insufficient⁴⁴. NKX2-3 expression in TPCs might be also related to genes associated with liver-metastasis in tumor cells, including TPC53⁴⁵, VEGF⁴⁶ and CXCR4³¹, but not KRAS (Supplementary Fig. 29). Additionally, the expression of NKX2-3 in TPCs might be associated with tumor microsatellite status⁴⁷ (Supplementary Fig. 30) and immune checkpoint blockade response⁴⁸ (Supplementary Fig. 31). Taken together, the elevated expression of NKX2-3 in TPCs could be a consequence of synergistic effects from various factors within the tumor microenvironment.

In conclusion, the present study found that NKX2-3^high TPCs facilitated tumor cell intravasation by inducing local vasodilation in tumor tissues and blocking the transcriptional activity of NKX2-3 in TPCs significantly inhibited tumor hematogenous metastasis. This study revealed that the TPCs play an important role in regulating tumor vascular tone and affecting tumor cell intravasation and hematogenous metastasis, providing a potential prognostic marker and therapeutic target for inhibiting tumor hematogenous metastasis at an early stage.

Methods

Inclusion and ethics

All animal experiments were approved by the Institute of Experimental Animal Ethics Committee of Jinan University (Guangzhou, China). The approval ID for the use of animals were 00365755 and 00379905. The human specimens used in this study were approved by the Clinical Ethics Committee of the First Affiliated Hospital of Jinan University (Guangzhou, China), and written informed consent was received from participants prior to inclusion in the study. The approval ID was JNUKY-2023-0067.

Human samples and specimens

Computed tomography (CT) and ultrasound doppler images of human CRC, tumor samples of human CRC, breast cancer and lung cancer patient were collected from the First Affiliated Hospital of Jinan University (Full patients’ information was provided in Supplementary Data 1–8). Radiotherapy or chemotherapy were not administered to patients before the operation, and patients with preoperative infections were excluded. All the immunofluorescence images of human specimens were confirmed by the qualified pathologists in pathology department; they help to confirm the field of view and assist in statistics. Informed consent was obtained from all patients whose data were included in this retrospective analysis. Since the study involved only the review of existing records and no new patient participation, financial compensation for participation was not provided. We fully adhere to SAGER guidelines and our study design was not related to gender.

Animals

Male C57BL/6JGpt mice (5–6 weeks, 18–20 g), BALB/c nude mice (5–6 weeks, 20–22 g), Rosa26-CAG-LSL-Cas9-tdTomato mice (B6/JGpt-Rosa26^{tm1(CAG-LSL-Cas9-tdTomato)}/Gpt; RRID: IMSR_GPT:T002249), Cspg4-CreERT2 mice (B6/JGpt-Cspg4^{em1Cin(CreERT2-P2A)}/Gpt; RRID: IMSR_GPT:T006187), and Nkx2-3-flox mice (B6/JGpt-Nkx2-3^em1Cflox/Gpt; T016620) were acquired from GemPharqmatech Co., Ltd. (Nanjing, Jiangsu, China). All mice were housed in a specific pathogen-free (SPF) facility at Jinan University Laboratory Animal Center under a 12-h light/12-h dark cycle, with controlled temperature (22–24 °C) and humidity (40–60%). Animals had ad libitum access to autoclaved food and water and were maintained in individually ventilated cages (IVCs) with corn cob bedding and environmental enrichment. To enact pericyte-specific lineage tracing and Nkx2-3 knockout, mice carrying tamoxifen-induced Cre recombinase driven by a pericyte-specific Cspg4 promoter (Tg^{Cspg4-CreERT2}) were crossed with mice carrying a Cre response reporter gene (tandem dimer Tomato (tdT)) inserted at the ROSA26 locus (ROSA^tdT/+) to allow pericyte-specific lineage tracing (PC^lin mouse), Therefore, NG2⁺ pericytes were selectively labeled with red fluorescence in PC^lin mouse. PC^lin mice were further bred with mice carrying the Nkx2-3 allele (Nkx2-3^flox/flox) to generate tamoxifen-inducible pericyte-specific Nkx2-3 knockout mice (PC^lin-KO). The genotypes of transgenic mice were identified by PCR assay using specific primers (The primer sequences were listed in Supplementary Table 5). To induce lineage marker expression and Nkx2-3 knockout by Cre recombinase, MC-38-luc-LM3 allografts-bearing mice (7–8 weeks, weight 22–25 g) received 10 mg/kg of tamoxifen (resolved in peanut oil) every three days for 3 times. The recombination efficiency was assessed by immunofluorescence assay. For calcium detection, mice were intravenously (i.v.) administered with AAV-CMV bGlobin-FLEX-Gcamp6-WPRE-hGH polyA virus (Genepharma, Suzhou, China). The titer for the virus was set at 5 × 10¹¹ each mouse. 40 days after the viral infection, mice were intragastrically injected with tamoxifen to induce Gcamp6 expression in pericytes. In this study, to control for experimental bias, we exclusively used male mice. As sex differences were not a primary focus of this study, a single-sex cohort design was adopted to eliminate sex-related confounders, precluding the need for sex-disaggregated analysis.

Cell lines and cell culture

Mouse colorectal cancer cell line MC-38 (Cat. BNCC337716) was purchased from BeNa Culture Collection (Beijing, China). MC-38 cells were infected with lentivirus harboring luciferase (Genechem, Shanghai, China) to generate the MC-38-luc cells, which were then selected with puromycin (2 μg/mL) for 2 days. Additionally, MC-38 cells were infected with lentivirus harboring GFP (Genechem) to generate MC-38-GFP cells, which were selected with puromycin in the same pattern. Human CRC cell lines HCT116 (RRID: CVCL_0291), DLD-1 (RRID: CVCL_0248), HT-29 (CVCL_0320), Caco2 (RRID: CVCL_0025), and human microvascular endothelial cell-1 (HMEC-1, RRID: CVCL_0307) were obtained from the American Type Culture Collection (Manassas, Virginia, USA) and cultured in DMEM (Cat. 11965092; Gibco, Waltham, MA, USA) with 10% FBS (Cat. FSP500, ExCell Bio, Shanghai, China) and 1% penicillin-streptomycin (PS, Cat. 15140122, Gibco). Lymphatic endothelial cell (LEC) was obtained from iCell (Shanghail, China). LEC and HMEC-1 cells were cultured in Endothelial Cell Medium (ECM, Cat. 1001, Sciencell research laboratories, Corte Del Cedro Carlsbad, CA) containing 5% FBS, 1% endothelial cell growth supplement (ECGS), and 1% PS. Tumor pericyte (TPC), cancer associated fibroblast (CAF), and smooth muscle cell (SMC) were primary cells isolated by ourselves. All cells were maintained at 37 °C in a humidified atmosphere containing 5% CO₂ and have been routinely tested negative for mycoplasma contamination and authenticated to have no cross-contamination by STR analysis. TPCs under hypoxia condition were incubated in a modular incubator chamber (Billups-Rothenberg, CA, USA) filled with 1% O₂, 5% CO₂ and 94% N₂ at 37 °C.

ScRNA-seq data analysis

The scRNA-seq data of TPCs from CRC patients without (n = 2) and with (n = 2) liver metastasis was obtained from the GEO database (GSE199726)²⁰. The 10x Genomics Cell Ranger software suite (v.3.1.0) with human reference GRCh38 was used to process the scRNA-seq Fastq files. The R package Seurat (v.4.4.0)⁴⁹ was used to analyze and visualize single-cell whole-transcriptome gene expression. Briefly, only cells with >200 and <3000 genes detected, of which <10% were mitochondrial genes, were kept for subsequent analysis. The expression count data for genes detected in each sample were first normalized (NormalizeData) as log₁₀(CPM) values. The normalized data within each age group were then merged with the function IntegrateData, and finally merged into a master data object. The expression values of all genes in all cells were subjected to principal-component analysis, and the first fifteen principal components were used for clustering (FindClusters, resolution of 0.5), UMAP projection (RunUMAP) and visualization (DimPlot). The clustering on the master transcriptome data yielded 13 clusters, and the top 10 cluster-specific expressed genes were extracted (FindAllMarkers) and used to annotate the cell types.

Radiomic analysis

For CT-based radiomics (Supplementary Fig. 32), all CT images of human CRC were preprocessed, including intensity normalization and resampling with the same voxel size and Z-score standardization. Volumes of interest (VOIs) were created by manually tracing tumors on CT using 3D Slicer software (version 5.6.2) and radiomic features were extracted by Python software package, PyRadiomics⁵⁰. The machine learning model was constructed based on FeAture Explorer software, while feature reduction was performed using Pearson correlation coefficient (PCC) values > 0.990 and intraclass correlation coefficient (ICC) ≥  0.8, ensuring that each feature was independent. Further, the number of features was reduced by recursive feature elimination (RFE) and the top 5 contributive features were selected based on Least absolute shrinkage and selection operator (LASSO) analysis⁵¹. The Shapley Additive exPlanations (“SHAP”) package, an alternative explainability method grounded in game theory, was used to assess the contribution of each feature to the signal of blood flow⁵², in which the small area emphasis value was the most important.

Ultrasound doppler analysis

In ultrasound Doppler images, red and blue regions conventionally represent blood flow, with red indicates blood flow toward the transducer, while blue indicates blood flow away from the transducer, Open Source Computer Vision Library (OpenCV, https://opencv.org) was utilized for image processing. First, the color space was converted from BGR to HSV to facilitate precise segmentation of color-coded areas. Then specific HSV color ranges were defined for both red and blue flow regions, applying these ranges to create binary masks that isolated the areas of interest. These masks were applied to extract the corresponding blood flow regions from the original image. To further improve the segmentation accuracy, OpenCV’s edge detection algorithms were employed, refining the boundaries of the extracted regions. Finally, within these segmented regions, the average saturation and brightness values were calculated to evaluate the blood flow signal strength, resulting in an objective quantification of blood flow signal within the ultrasound Doppler images.

Isolation and identification of tumor pericytes (TPCs)

TPCs were isolated by microdissection combined with pericyte medium-based approach (MPMA)⁵³. Freshly obtained human colon cancer tissues were immediately stored in cold DMEM with PS and processed within 1 h. The tissues were washed twice with pre-cooled PBS containing 0.1% gentamicin, 0.1% ciprofloxacin, and 0.1% kanamycin in a sterile hood. After removing fibrous and adipose tissues, the tumor was cut into small pieces and fixed in a sylgard-coated dish. For TPC isolation, ascending capillaries with a diameter of 20 µm were carefully dissected from the mucosa. After removing residual adipose tissue, the capillaries were transferred to a 6-well plate with pericyte medium (PM, Cat. 1201, Sciencell research laboratories). TPCs migrated out within 14 days, and those cells were further authenticated and characterized by STR profiling multi-amplification kit, transmission electron microscopy and immunofluorescence assay, confirming pericyte markers and high nuclear/cytoplasmic ratio⁵³. TPCs were maintained in PM supplemented with 2% FBS, 1% pericyte growth supplement (PGS), and 1% PS.

Isolation of smooth muscle cells (SMCs)

SMCs were isolated by microdissection combined with SMC medium-based approach, which is similar to MPMA. For SMC collection, arteries (diameter ≈ 150 µm) were gently separated under a stereomicroscope. The attached adipose tissues were removed, and the separated arteries were transferred to fresh dishes containing SMC medium. SMCs migrated from the vascular samples to the culture plate within 14 days and were further incubated in SMC medium (SMCM, Cat. 1101, Sciencell research laboratories) containing 2% FBS, 1% smooth muscle cell growth supplement (SMCGS), and 1% PS.

Isolation of cancer associated fibroblasts (CAFs)

CAFs were isolated from the primary tumor of liver-metastatic colorectal cancer patients according to previous study⁵⁴. The primary tumor tissues were sectioned into small pieces (~1 mm³), and incubated within 60 mm dishes. The tissues were maintained in the DMEM/F12 medium (Cat. 12634010; Gibco, Waltham, MA, USA) containing 10% FBS and 1% PS and cultured in 5% of CO₂ for at least 2 weeks to let CAFs spread and expand. The culture media was renewed 2–3 times per week. When CAFs reached 60% of confluency, they were detached with trypsin and further cultured.

Regulatory network analysis

TPC cluster-specific gene regulatory network (regulon) identification was performed using the SCENIC workflow⁵⁵. Briefly, the SCENIC analysis included four steps: (1) the single-cell gene expression count table for cells from each cluster was first fed into SCENIC with a list of 1390 known human TFs (https://scenic.aertslab.org/), and sets of genes that are co-expressed (either positively or negatively) with TFs were identified by random forest models; (2) putative target genes in each co-expressed module were then subjected to cis-regulatory motif discovery analysis, and only modules with significant motif enrichment in the ±500 bp or ±10 kb region (i.e., extended region) around the transcription sart site (TSS) for the corresponding TF were kept for further analysis; (3) the AUC ell algorithm of SCENIC was used to score the activity of each regulon in each cell, and regulons with average AUC score ≥0.1 were retained; and (4) the AUC scores were used as regulon activities, and the top 5 regulons for each TPC cluster (excluding cluster 10, 11 and 12 due to small number of cells) were visualized as heatmap.

Cell infection and transfection

Lentivirus-carrying vector or NKX2-3-overexpressing plasmid, lentivirus-carrying PDE1C-overexpressing plasmid and lentivirus-carrying NEAT1-overexpressing plasmid were obtained from VectorBuilder (Guangzhou, China). TPCs were infected with indicated lentivirus for 48 h and selected by puromycin (2 μg/mL). The survival cells were collected and validated by qPCR or western blot. For NEAT1 knockdown assay, TPC_LM were infected with lentivirus-carrying shNC (Cat. VB151023-10034) or shNEAT1 (Cat. VB210317-1271svu) to generate TPC_LM^shNC or TPC_LM^shNEAT1, respectively. CRISPER/dCas9 technology was used to knockout NKX2-3 in TPCs, Guide RNAs specific to NKX2-3 (gNKX2-3) or scrambled control (gNC) were designed using the online CRISPR design tool (http://crispr.mit.edu/). Lentivirus-carrying gRNA and dCas9 were purchased from VectorBuilder. TPCs were infected with lentivirus-carrying dCas9 for 48 h and selected by hygromycin (100 μg/mL). Then, the survival TPCs were infected with lentivirus-carrying gNC or gNKX2-3 for 48 h and selected by puromycin (2 μg/mL). For calcium imaging assay, TPC_LM was transfected with siRNAs targeting NKX2-3 or a negative control (NC) for 48 h. Transfection was performed using Lipofectamine™ 3000 (Cat. L3000015, Invitrogen, Carlsbad, CA, USA), according to the manufacturer’s instructions. The siRNA sequences used were listed in Supplementary Table 6.

Oligonucleotide and siRNA transient transfection

Cells were transfected with scrambled control siRNA or specific siRNA targeting indicated genes using Lipofectamine™ 3000, according to the manufacturer’s instructions. In the case of miRNA mimic and inhibitor transfection, TPCs were transfected with has-miR-769-5p mimic or has-miR-769-5p inhibitor by Lipofectamine™ 3000. After 48 h, cells were acquired for further analysis. The siRNAs, miRNA mimics and inhibitors were designed and synthesized by Suzhou Genepharma Co.,Ltd, and their sequences were listed in Supplementary Table 7.

Construction of MC-38-luc-LM3 cells and MC-38-GFP-LM3 cells

MC-38-luc-LM3 cells and MC-38-GFP-LM3 cells were established according to a previous study²⁰. Briefly, 1 × 10⁵ MC-38-luc cells or 1 × 10⁵ MC-38-GFP cells were mixed with 100 µL of Matrigel (Cat. 354248, Corning, NY) and injected into the spleen of male C57BL/6JGpt mice. Mouse liver metastasis was detected by in vivo imaging system until the metastatic foci formed. Then, the metastatic tumor cells in liver were collected using a mouse tumor dissociation kit (Cat. 130-096-730, Miltenyi Biotec, Bergisch Gladbach, Germany) and further selected by puromycin (2 μg/mL), and the survival tumor cells were referred to LM1 cells. LM2 cells were obtained from the liver metastatic foci following the same pattern. Tumor cells isolated from the liver metastatic foci in the third round were termed as MC-38-luc-LM3 cells and MC-38-GFP-LM3 cells, which were applied for the subsequent experiments.

Construction of MC-38-luc-LNM cells

MC-38-luc-LNM cells were established in our lab according to previous study⁵⁶. Metastatic lymph node (LN) tissues were harvested from MC-38-luc allografts-bearing mice that developed LN metastasis. Metastatic LN tissues were cut into small pieces (~1 mm³), and digested with the tumor tissue dissociation kit as described above. Single-cell suspension of digested tissues was filtered through a 100 μm filter and selected with anti-CD326 microbeads (Cat. 130-061-101, Miltenyi Biotec.). The sorted cells were termed as MC-38-luc-lymph node metastatic (LNM) cells.

ROC curve analysis

ROC curve analysis was based on the data collected from the first affiliated hospital of Jinan University and analyzed by ourselves, which was created in Graphpad Prism software(https://www.graphpad.com/guides/prism/latest/statistics/stat_howto_roc.htm), and the raw data has been uploaded in the Source Data file.

Animal studies

For colorectal cancer liver metastasis model, MC-38-luc-LM3 cells or MC-38-GFP-LM3 cells (1 × 10⁵) were suspended in 100 µL of Matrigel and injected into the cecum wall of PC^lin and PC^lin-KO mice, which were previously anesthetized with isoflurane inhalation. Mouse tumor growth and metastasis were detected by in vivo imaging system. When liver metastasis was detected in mice, mouse tumor and liver were collected for further pathological examination. For colorectal cancer lymph node metastasis model, MC-38- LNM cells (1 × 10⁵) were suspended in 100 µL of Matrigel and injected into the cecum wall of PC^lin and PC^lin-KO mice, which were previously anesthetized with isoflurane inhalation. Mouse metastasis was detected by in vivo imaging system and H&E staining. For melanoma metastasis model, B16-F10 cells (5 × 10⁵) were suspended in 100 µL of PBS and subcutaneously injected into PC^lin and PC^lin-KO mice. Four weeks later, the lung tissues were acquired and applied for hematoxylin and eosin (H&E) analysis. The subcutaneous tumor maximum volume was 2000 mm³ and authorized by the Committees on Animal Research and Ethics. For co-injection assay, HCT116 cells, TPC_NM^Vector and TPC_NM^NKX2-3 were collected and counted. HCT116 cells (4 × 10⁵) were mixed either with TPC_NM^Vector or TPC_NM^NKX2-3 (1.6 × 10⁶) respectively and suspended in 100 µL of Matrigel. Then, the mixed cells were injected into the cecum wall of nude mice that were anesthetized previously. Three weeks later, mice administered with HCT116 cells and TPC_NM^NKX2-3 were randomly divided into saline, NB2 and Z-GP-NB2 treatment group. Mice were intraperitoneally injected with saline, NB2 (3 mmol/kg) or Z-GP-NB2 (3 mmol/kg) every other day for two weeks. Two weeks later, mice were sacrificed, and the mouse tumor and liver tissues were acquired for further investigation.

In vivo cell tracking

Each mouse was injected intraperitoneally with 3 mg of D-luciferin (Cat. 40901ES01, Yeason Biotechnology, Shanghai, China). 5 min later, the mice were anesthetized with 1% pentobarbital sodium and detected using the IVIS Lumina LT imaging system (PerkinElmer, Waltham, MA, USA).

Co-culture system

TPCs were co-cultured with DLD-1 cells using a 0.4 μm-pore size transwell chamber (Corning Costar, NY). The transfected DLD-1 cells (2 × 10⁵) were cultured in the upper chamber, while TPC (3 × 10⁵) supplemented with 600 μL PM were seeded in the bottom chamber of transwell. After 24-h culture, TPCs were collected for subsequent analysis.

Verapamil, forskolin, (S)-(-)-Bay K8644, H89 2HCl and NB2 treatment

Verapamil (Cat. S4202), forskolin (Cat. S2449), (S)-(-)-Bay K8644 (Cat. E4039) and H89 2HCl (Cat. S1582) were purchased from Selleck (Houston, TX, USA) and dissolved in DMSO. Verapamil (5 μM), forskolin (25 μM), (S)-(-)-Bay K8644 (10 μM), H89 2HCl (10 μM), NB1-6 (20 μM) and Z-GP-NB2 (20 μM) were added to the TPCs as indicated and cultured for 24 h. Then, TPCs were collected for further analysis.

Cell contraction assay

Contractility of TPCs and ECs were evaluated by Cell Contraction Assay kit (Cat. CBA-5021, CELL BIOLABS, CA, USA) according to the manufacturer’s instructions. Cells (5 × 10⁶) were harvested and mixed with prepared collagen solution (1:4 v/v). A 0.5 mL of cell-collagen mixture was added to each well of a 24-well plate and incubated at 37 °C for 1 h. After collagen polymerization, 1.0 mL medium was carefully added to the top of each collagen gel lattice and incubated for 2 days, the collagen gel size (contraction index) was measured at 48 h and quantified with Image J.

Calcium Imaging

TPCs with indicated treatment were seeded in a 96-well plate with 8000 cells per well. The next day, TPCs were washed with the calcium-free PBS (KCl 2.7 mM, KH₂PO₄ 2.0 mM, NaCl 137 mM, Na₂HPO₄ 10 mM) twice and finally incubated with Kreb’s solution (NaCl 119 mM, KCl 4.7 mM, NaHCO₃ 25 mM, KH₂PO₄ 1.2 mM, MgCl₂•2H₂O 1.0 mM, CaCl₂•2H₂O 2.5 mM, D glucose 11.1 mM) supplemented with (S)-(-)-Bay K8644 (10 μM). (S)-(-)-Bay K8644 was used to activate the L-type calcium channel and intracellular influx of calcium derived from the Kreb’s solution. The calcium influx before and after (S)-(-)-Bay K8644 (10 μM) treatment was examined under the Zeiss LSM 800 confocal microscope and analyzed by GraphPad Prism.

Two-photon imaging

For vasoconstriction and vasodilation experiments, mice were anesthetized with isoflurane inhalation and anesthesia was confirmed by the lack of a withdrawal reflex to a paw pinch. A small fragment was cut from mouse tumor and stained with 0.1 mg/ml of Lectin (Cat. L32478, Invitrogen, Carlsbad, CA, USA) for 10 min. Then, the tumor fragment was carefully fixed by agarose and immersed in oxygenated Krebs solution as described above. U46619 (Cat. D1827, Sigma) was used as a vasoconstrictor for tumor vessels^57–59. The change of vessel diameter after U46619 treatment was observed under Zeiss LSM780 microscope with the two-photon laser tuned to 920 nm. blood flow analysis was performed¹⁴. Tumor-bearing mice were intravenously injected with FITC-dextran prior to imaging. Intravenous injection of dextran revealed the vascular architecture as well as individual red blood cells (RBCs), which appeared as shadows flowing with the fluorescent plasma. Then, mouse tumor was exposed to 4× objective and rapid line scan (735 lines per second) was used to analyze the blood velocity, the line-scan bisected the central axis of the capillary lumen to track the movement of blood cells, RBC moving through the capillary appeared as oblique shadows from which we determined the tumor blood velocity(Δx/Δt).

Light-sheet imaging

Tissue clearing assay was performed according to the manufacturer’s instructions of CytoVista™ Tissue Clearing/Staining kit (Cat. V11324, Invitrogen). Briefly, mice were anesthetized with 1% pentobarbital sodium and perfused by 4% paraformaldehyde (PFA). Then, mouse tumor was acquired and cut into tissue blocks at a thickness of 4 mm, followed by the immersion fixation in 4% PFA overnight. The fixed tumor tissues were washed and permeabilized by increasing concentration of methanol at 4 °C, bleached with ice-cold 5% H₂O₂ in methanol containing 20% DMSO, and then incubating at 4 °C overnight to reduce the background fluorescence. After re-hydrating, permeabilization and blocking, tissues were incubated with anti-CD31 antibody (RRID: AB_2161028, R&D Systems) at a dilution of 1:300 for 10 days, followed by the Alexa Fluor 546-Donkey anti-Rabbit IgG antibody (RRID: AB_2534016, Invitrogen) at a dilution of 1:500 for 10 days. Then, the tumor tissue blocks were washed and dehydrated with increasing concentration of methanol at 4 °C. Subsequently, the samples were transferred to CytoVista™ Tissue Clearing Enhancer and imaged with Lattice Light Sheet 7 microscope (ZEISS, Jena, Germany).

Pericyte coverage analysis

In immunofluorescence assay, α-SMA was used to label TPCs and CD31 was used to label tumor vessels. Pericyte coverage was evaluated according to previous study¹⁶, the α-SMA positive area and CD31 positive area were respectively quantified using Image J, and then the ratio of α-SMA positive area to CD31-positive area was calculated as pericyte coverage.

MRI scanning

Mice were anesthetized with 1% pentobarbital sodium, and a 24-gauge catheter was placed in the tail vein for contrast agent administration. MRI was conducted on a 3.0 T system (GE Healthcare Signa HDxt, Milwaukee, WI) using an eight-channel wrist coil. The imaging protocol included T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), dynamic contrast-enhanced MRI (DCE-MRI), and intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI). For DCE-MRI, a 3D fast spoiled gradient-recalled echo (3D-FSPGR) sequence was used. Images were acquired every 3 seconds before and after intravenous injection of 0.1 mmol/kg Gd-DTPA (Magnevist), followed by a 0.3 mL saline flush. IVIM-DWI was performed using a single-shot echo-planar imaging sequence with diffusion gradients applied in three orthogonal directions at 12 b-values.

MRI analysis

All data was analyzed at the post-processing workstation (AW 2.0, GE Healthcare). The regions of interest were drawn by outlining the entire tumor boundary in the contrast enhancement images. The IVIM data were analyzed using the Functool MADC program, and a biexponential model was used to calculate the f and D*⁶⁰. The DCE-MRI data were analyzed using the Cinetools program, and the volume transfer coefficient (K^trans) was calculated based on the Tofts model⁶¹.

Laser speckle contrast imaging

Tumor-bearing mice were anesthetized with isoflurane and subjected to laparotomy to expose orthotopic intestinal tumors. Tumor perfusion was quantitatively assessed using a laser speckle contrast imaging system (Moor Instruments, Devon, UK). The system was equipped with a near-infrared laser (785 nm, 80 mW) positioned at a fixed distance of 10 cm from the tissue surface. For precise spatial localization, a visible guidance laser (650 nm, 3 mW) was utilized to demarcate the region of interest. Imaging parameters included a field of view of 1.0 × 1.0 cm to ensure dynamic monitoring of perfusion characteristics.

Prediction and synthesis of oligopeptides

The structure of human NKX2-3 (Q8TAU0) was predicted with AlphaFold DB on https://swissmodel.expasy.org/. The sequences characterized pLDDT > 90 in the local similarity were considered as helix domain, which were used for the synthesis of candidate oligopeptides (NB1-NB6). Z-GP-NB2 was prepared according to previously described methods²⁹. Oligopeptides (NB1-NB6, Z-GP-NB2 and NB2-FITC) with purity >  99% were ordered from ChinaPeptides Co., Ltd. (Shanghai, China). Peptides used in the experiments were dissolved in saline to a final concentration of 20 mM stock solutions and storage at −80 °C no more than two weeks.

Hydrolysis of Z-GP-NB2

Briefly, Z-GP-NB2 was first diluted in serum-free DEME medium at gradient concentrations from 0.3375 to 20 μM and then subjected to AB SCIEX Triple Quad 4500 LC-MS/MS System (Applied Biosystems, CA, USA). The ion peak data of Z-GP-NB2 was collected and then the standard curves were drawn with the peak area of Z-GP-NB2 as Y axis and their concentrations (C_Z-GP-NB2) were X axis. HCT116 cells, TPC_NM^NKX2-3, TPC_NM^NKX2-3+siNC, TPC_NM^{NKX2-3+siFAPα} (2 × 10⁵) were placed in a 6-well plate and cultured overnight. The next day, the culture medium was replaced with serum-free DMEM containing 20 μM of Z-GP-NB2. The supernatants were collected after the cells were cultured for 2 h at 37 °C, and centrifuged at 18,000 × g for 15 min. After filtration, the peak areas of Z-GP-NB2 were analyzed by liquid chromatograph-mass spectrometer (LC-MS)/MS System and the concentrations of Z-GP-NB2 were calculated according to the standard curve. The hydrolysis rate was equal to 100%-C_Z-GP-NB2/20 × 100%.

Vessel permeability assay

PC^lin and PC^lin-KO mice were intravenously injected with 1 mg of FITC-labeled Dextran-40 kDa (Cat. 53379, Sigma). After 10 min, mice were anesthetized with 1% pentobarbital and perfused with 4% PFA. Tumors were acquired, frozen and sectioned at a thickness of 6 μm. Tumor vessels were stained by anti-CD31 antibody (RRID: AB_2161028, R&D Systems), followed by the Alexa Fluor 546-Donkey anti-Rabbit IgG antibody (RRID: AB_2534016, Invitrogen). The fluorescence area of dextran over CD31 staining was calculated by Image J, which indicates vessel permeability.

Microfluidic vascular chip models

The lower and upper channels of microfluidic chip (LNP-Bl, FluidicLab) were pre-coated with 50 μg/mL fibronectin (Yeasen) for 1 h to allow the cells adhere to the surface. To mimic tumor vessels, GFP-transfected TPCs (2 × 10⁴, green fluorescence) were added to the bottom channel and cultured for 4 h, and then, endothelial cells (1 × 10⁵) were added to the TPCs-coated channel and further cultured for 4 h to form endothelial layer. Unadhered cells were removed by gentle wash with PBS. The prepared microfluidic chip was placed at 37 °C atmosphere containing 5% CO₂ overnight. The next day, PKH26-labeled HCT116 cells (red fluorescence) were added to the upper channel with a directional flow at bottom channel and upper channel, respectively. Tumor cells and mimetic blood vessels were observed, and images were acquired with a high-connotation imaging scanning (Revvity, Opera Phenix Plus).

Mouse vascular organoids

The primary tumors of PC^lin and PC^lin-KO mice were obtained and then tumor capillaries were micro dissected from the tissues. The separated capillaries were embedded in the Matrigel and placed in a transwell chamber to prepare vascular organoid. After maintaining at 37 °C in a humidified atmosphere containing 5% CO₂ for 48 h, MC-38 cells (5 × 10³) were added to the vascular organoid and further culturing for 48 h. At the end of experiments, the whole Matrigel was fixed with 4% PFA and subjected to H&E staining.

Isolation and identification of circulating tumor cells (CTCs)

CTC isolation and identification were performed according to a previous study^20,62. Blood samples obtained via cardiac puncture from tumor-bearing mice were collected in heparin-coated tubes to prevent coagulation. After erythrocyte lysis, the remaining cells were resuspended in DMEM containing 20% FBS and cultured in 12-well plates. Within 12 h of plating, adherent cells were examined to exclude potential tumor cell proliferation. CTC identification was performed by immunofluorescence staining using epithelial cell adhesion molecule (EpCAM) as a cancer cell marker and CD45 as a leukocyte marker, with visualization by confocal microscopy. Cells exhibiting EpCAM-positive/CD45-negative staining were classified as CTCs and manually quantified, with results expressed as CTCs per milliliter of whole blood.

RNA-seq analysis

Total RNA was extracted using TRIzol reagent (Cat. 15596018, Invitrogen, Carlsbad, CA, USA). High-throughput transcriptome sequencing and subsequent bioinformatics analysis were performed by Genedenovo Biotechnology Co., Ltd. (Guangzhou, China). In brief, the RNA concentration and integrity were evaluated by NanoDrop 2000 (NanoDrop, Wilmington, DE, USA). Then, the samples were inspected by agarose gel electrophoresis, Nanodrop and Agilent 2100. The fragmented RNA pieces were used to construct library by Hieff NGS^® Ultima Dual-mode mRNA Library Prep Kit (12309ES, Yeasen) and sequenced with Illumina Novaseq™ 6000 (Genedenovo Biotechnology Co., Ltd., Guangzhou, China). Then, StringTie and edgeR were used to evaluate the expression levels of all transcripts. The differentially expressed mRNAs and genes were picked with log2 (fold change) >1 or log2 (fold change) <−1 and with statistical significance (P  < 0.05) by R package-edgeR.

Quantitative polymerase chain reaction (qPCR) assay

Total RNA was extracted using E.Z.N.A.® Total RNA Kit (Cat. R6834-02, Omega Bio-Tek, Norcross, GA, USA), and the concertation of RNA was identified by Nanodrop Lite micro spectrophotometer. Then, RNA (1 μg) was reversely transcribed into complementary DNA (cDNA) using the SweScript RT I First Strand cDNA Synthesis Kit (Cat. A234-10, Genestar, Beijing). qPCR analysis was performed by 2×RealStar Fast SYBR qPCR Mix (Cat. ZA301-101S, Genestar). The PCR products were quantified with a LightCycler 480 PCR system (Roche, Mannheim, Germany) and the data was further analyzed using the 2-ΔΔCT method [ΔCt = C_target – Ct_ACTB and Δ (ΔCt) = ΔCt_sample – ΔCt_control]. The primers used for PCR were listed in Supplementary Table 8.

Enzyme-linked immunosorbent assay (ELISA)

The level of cAMP in the conditioned medium of TPCs was examined by cAMP ELISA Kit (Cat. E-EL-0056c, Elabscience, Wuhan, China) according to the manufacturer’s instructions.

Chromatin immunoprecipitation (ChIP) and ChIP-Seq

ChIP assay was performed using the SimpleChIP® Enzymatic Chromatin IP Kit (Cat. 9003, Cell Signaling Technology, MA, USA) following the instructions of the manufacturer. Briefly, about 4 × 10⁶ TPC_NM^NKX2-3 were collected and washed. Chemical crosslinking of protein-DNA was performed using 1% formaldehyde and then stopped by glycine (125 mM). TPCs were washed twice with cold PBS and scraped into cold PBS containing PIC. Afterwards, TPC suspension was centrifuged and the pellet was resuspended in buffer A + DTT + PIC + PMSF followed by incubation on ice for 10 min. TPC nuclei was pelleted and resuspended in buffer B + DTT + micrococcal nuclease and incubated for 20 min at 37 °C with frequent mixing to digest DNA (average DNA length of 150–900 bp). Then, the digested chromatin was diluted and 10 μL diluted sample was saved as 2% Input. The remaining lysate was incubated with anti-NKX2-3 antibody (RRID: AB_2680088, Sigma, Shanghai, China) to immunoprecipitate chromatin overnight at 4 °C with rotation and followed by ChIP-grade protein G magnetic beads incubation. The magnetic beads were washed and the eluted DNA was purified and analyzed by qPCR to determine the binding of NKX2-3 to indicated genes. The primers used for ChIP assay were listed in Supplementary Table 9. For ChIP-sequencing, immunoprecipitated DNA was used to construct sequencing libraries following the protocol provided by the NEXTflex® ChIP-Seq kit (Cat. NOVA-5143-02, BioScientific, TX, USA) and sequenced on Illumina Xten with PE 150 method (LC-Bio Technology CO., Ltd., Hangzhou, China). For data analysis, Trimmomatic (version 0.38) was used to filter out low-quality read. MACS2 software (version 2.1.1.20160309) was used to call peaks by default parameters (bandwidth, 300 bp; model fold, 5, 50; q value, 0.05). The peak will be linked to a gene if its summit is closest to the TSS of that gene. The EasyGO gene ontology enrichment analysis tool (http://bioinformatics.cau.edu.cn/easygo) was used to conduct GO enrichment analysis.

Immunofluorescence analysis

For tissues, tumor tissues embedded with paraffin were sectioned, deparaffinized, rehydrated, and subjected to antigen retrieval using sodium citrate antigenic repair solution (Cat. P0086, Beyotime). Then, tumor sections were permeabilized by 0.1% Triton^TM X-100 (Cat. MKBW8386V, Sigma) and blocked with QuickBlock^TM immunostaining blocking solution (Cat. ST797, Beyotime), followed by incubation with the corresponding primary antibodies overnight at 4 °C (The information of antibodies was listed in Supplementary Table 10). The next day, tumor slices were washed with PBS and incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. The slides were then washed and visualized with iF488-Tyramide, iF555-Tyramide, or iF647-Tyramide using TSAPLus Fluorescence Kits (Cat. G1236, Servicebio), and the cell nucleus was stained with 1 μg/mL DAPI (Cat. D9524, Servicebio) for 15 min. For cell immunofluorescence, TPCs (5 × 10⁴) were seeded in the glass bottom of culture dish and cultured for 24 h. The next day, cells were fixed, permeabilized, and incubated with Alexa Fluor™ 594-phalloidin (Cat. A12381, Thermo) for 1 h. Cell nucleus were labeled by DAPI (1 μg/mL). Images were obtained using Zeiss LSM 800 confocal microscope and analyzed by Image J software.

Fluorescence in situ hybridization (FISH) assay

The level of NEAT1 was analyzed according to the manufacturer’s instructions of RNA FISH Kit (Cat. lnc1M-2, RiboBio, Guangzhou, China). Briefly, human CRC specimens were deparaffined, permeabilized by proteinase K, and blocked by 5% BSA. Then, NEAT1 probe designed by RiboBio Co., Ltd. was diluted by and incubated with tumor slice at 37 °C. After 16-h incubation, tumor sections were washed and applied for further immunofluorescence assay. Images were obtained using Zeiss LSM 800 confocal microscope and analyzed by Image J software.

Western blot assay

Cells with indicated treatment were lysed with RIPA lysis buffer containing phosphatase inhibitors (Cat. C001, TargetMOI, MA, USA) and protease inhibitors cocktail (Cat. C002, TargetMOI) for 0.5 h on ice. After centrifugate at 12,000 × g at 4 °C for 15 min, the supernatants were collected and the concentration of proteins was determined using BCA Protein Assay Kit (Cat. MA0082-3, Meilunbio, Dalian, China). Equal amounts of proteins (20 μg) were isolated using SDS-PAGE gel and transferred to polyvinylidene fluoride (PVDF) membranes (Cat. IPVH00010, Millipore, Boston, MA, USA). The PVDF membranes were then blocked with 5% BSA for 1 h and incubated with primary antibodies a dilution of 1:1000 overnight at 4 °C. The next day, the membranes were incubated with HRP-conjugated secondary antibody at room temperature for 1 h and the blots were detected by Amersham Imager 600 (GE, Boston, MA, USA). The details of primary and secondary antibodies were listed in Supplementary Table 11.

Intravital microscopy imaging

Tumor-bearing mice were intravenously injected with 25 mg/kg Evans Blue (Cat. E2129, Sigma) and then anesthetized with isoflurane inhalation. The anesthetized mouse was placed on a customized stage for tumor immobilization and colon tumor was exposed to intravital microscopy (IVM-MS2, IVIM Technology, Korea) equipped with 25× water immersion objective. Fluorescent events in the GFP-tagged tumor intravasation, blood flow and GCaMP6 expression in TPCs were visualized and recorded at different positions of primary tumor. Excitation was tuned at 620 nm and emitted fluorescence was collected with 2 non-descanned detectors: Channel 1 (503-558) for GFP and GCaMP6, and Channel 2 (641-708) for Evans Blue.

Optical coherence tomography angiography

Mice were anesthetized with isoflurane inhalation and placed on a black board. Then mouse tumor was exposed and detected by the Monitoring System of Vascular Microcirculation in Vivo (OptoProbe Science LTD, UK). The 3D OCT scanning range of the monitoring system is 10 mm ×10 mm × 2 mm and the time used for scanning OCTA data is about 60 s.

Extracellular vesicle isolation, identification and fluorescent labeling

Extracellular vesicles (EVs) were isolated from the blood of colorectal cancer patients and the conditioned medium of DLD-1 cells respectively. For patient blood samples, EVs were isolated from patients using Exosupur columns (Cat. ES9P11e, Echobiotech, Beijing, China), and RNA of EVs were extracted with Exosomal RNA Purification Kit (Cat. 5202050, SIMEN, Hangzhou, China) according to the instruction description. For DLD-1 cells, EVs were isolated by sequential centrifugation. Briefly, the conditioned medium of DLD-1 cells were centrifuged at 10,000 × g for 30 min (4 °C), filtered through a 0.22 μm syringe filter and ultracentrifuged at 120,000 × g for 1.5 h (4 °C) in a Optima XPN-100 (Beckman Coulter, CA, USA). Pellets were resuspended in PBS and ultracentrifuged again at 120,000 × g for 1.5 h (4 °C). The final pellets were resuspended in 50 μL cold PBS. The isolated vesicles were identified by Western blot using anti-HSP70, anti-CD9, anti-CD63, anti-TSG101 and anti-GM130⁶³. Vesicle morphology was analyzed by transmission electron microscopy and the concentration of DLD-1 cells-derived EVs was measured in terms of their protein content using BCA Protein Assay Kit⁶⁴. The diameter size and concentration of vesicles were determined by Nano flowcytometry (FCM)⁶⁵. EVs were fluorescently labeled with the PKH67 (Cat. MINI67, Sigma). Briefly, EVs were mixed with PHK67 dye for 5 min. Incubation was stopped by adding EV-free FBS and ultracentrifuged at 120,000 × g for 3 min. The remaining pellets were resuspended in 100 μL PBS.

In vitro capture of labeled extracellular vesicles

TPCs (1 × 10⁴) were seeded in the bottom glass of the culture dish and stained with Alexa Fluor™ 594-phalloidin for 1 h. Then, TPCs were washed with PBS twice and incubated with 1 μg/mL PKH67-labeled EVs or PBS as a negative control. After 24-h incubation, TPCs were washed with PBS and fixed by 4% PFA for 10 min. cell nucleus were stained by DAPI (1 μg/mL) for 15 min. Images were obtained using Zeiss LSM 800 confocal microscope.

Dual-luciferase reporter assay

TPCs (1.5 × 10⁵) were seeded into 6-well plates and cultured for 12 h until the confluence reaches to 60–70%. NKX2-3 3′ UTR WT and NKX2-3 3′ UTR mutant plasmids; NEAT1 3′ UTR WT and NEAT1 3′ UTR mutant plasmids were constructed in advanced. TPCs were transiently co-transfected with miR-769-5p mimic or miR-769-5p CTL together with NKX2-3-Wt, NKX2-3-Mut, NEAT1-Wt or NEAT1-Mut reporter plasmids using Lipofectamine 3000. After 48 h, Dual Luciferase Reporter Gene Assay Kit (Cat. RG027, Beyotime) was used to detect firefly and Renilla luciferase activities and recorded using a microplate reader (BioTek Instruments, VT, USA).

H&E staining

Formalin-fixed tissues were embedded in paraffin and sectioned at thickness of 6 μm. Following deparaffinized, the sections were treated with antigen retrieval solution and incubated with hematoxylin followed by counterstaining with eosin. Images were photographed by Olympus BX 53 inverted microscope and analyzed with Image J software.

Microscale thermophoresis (MST)

The binding affinity of NKX2-3 and NB2 to downstream PDE1C promoter was determined by MST. NKX2-3 protein was purchased from Proteintech Group (Cat. Ag25216, Wuhan, China) and labeled with a Monolith NT Protein Labeling Kit RED-NHS (Cat. MO-L011, Nanotemper Technologies, Munchen, Germany). NB2-FITC was synthesized by Chinapeptides CO., Ltd. NKX2-3 and NB2-FITC were diluted to 20 nM with PBS. PDE1C was diluted at half-concentrations with PBS. In the NKX2-3 binding experiment, the final concentration of PDE1C ranged from 12.2 nM to 400 μM, whereas in the NB2-FITC binding experiment, the concentration of PDE1C ranged from 0.763 nM to 25 μM. Then, each PDE1C dilution was mixed with one volume of labeled NKX2-3 or NB2-FITC dilution. In a competitive assay, NKX2-3 (20 nM) was premixed with NB2 (400 nM) at room temperature before PDE1C treatment. The final concentration of PDE1C ranged from 48.8 nM to 1.6 mM. After incubating in the dark at room temperature for 20 min, 16 mixtures were loaded into the standard Monolith NT.115 Capillaries (Cat. MO-K022, Nanotemper Technologies) and subjected to MST measurements. The value of the dissociation constant (Kd) was determined by fitting the curve using the Kd Model of the MO. Affinity Analysis software (version 2.3, Nano Temper Technologies).

Isothermal titration calorimetry (ITC)

The interaction of the peptides NB1-6 to downstream PDE1C promoter was examined by a MicroCal PEAQ-ITC instrument (Malvern Panalytical, Malvern, UK). Each ITC titration consisted of 19 successive injections. NB1-6 (20 μM) was load into the syringe respectively and titrated into cell containing 250 μL of PDE1C (2 μM). The reference power was set to 10 μcal/s. Data analysis was preformed using the one site-binding model from MicroCal PEAQ-ITC analysis software version 1.30 (Malvern Panalytical, Malvern, UK).

Statistics and reproducibility

All experiments were performed at least three independent times. Statistical analyses were carried out using two-tailed unpaired t-test or Mann–Whitney test between two groups and significant differences between more than two groups were evaluated using one-way ANOVA followed by Tukey’s post hoc test. OS and DFS curves were plotted using the Kaplan–Meier method and compared using the log-rank test. The area under ROC curve was analyzed by logistic regression model. GraphPad Prism 8.0 software was used for statistical analyses (GraphPad Software, Inc., San Diego, CA). The data are represented as mean values ± SEM. P < 0.05 was considered as significant difference.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.F.C., W.C.Y., D.M.Z., and O.J.L. designed and supervised the experiments, and revised the manuscript. M.F.C., X.B.L., O.J.L., and S.S.Y. wrote the manuscript and analyzed the data. M.F.Y, M.F.C., X.B.L., S.S.Y., X.Y.W., Q.M., D.Z., J.A.W., X.T.Z., M.Q., S.W., and J.W.X. performed experiments. S.H. and Z.S.T. preprocessed the images of CT and ultrasound Doppler. W.F.M. conducted CT radiomic analysis, S.Z. performed doppler-based radiomic analysis, O.J.L. analyzed the data of ScRNA-sequencing. J.H.P., D.D.H., S.H.Q., and Z.Z. collected human tumor tissues and assessed clinical samples. J.H.P. created the Biorender images. M.F.Y., J.Q.Z., M.H.H., and C.R.W. revised the manuscript critically. The order of the co–first authors was determined on the basis of their efforts and contributions to the manuscript.

Peer review

Peer review information

Nature Communications thanks Peter Balogh, who co-reviewed with Zoltan Kellermayer, and Valerie LeBleu, Byungheon Lee, Dan Xie, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data generated or analyzed during this study are included within the article, Supplementary Information, the Source Data file. ScRNA-seq data of the TPCs was downloaded from GEO accession GSE199726. ScRNA-data of PD1 resistant (non-responders) and sensitive (responders) tumors was downloaded from GSE236581. ScRNA data of microsatellite instability-high (MSI-H) and microsatellite stability (MSS) tumors was downloaded from GSE178341. RNA-seq and ChIP-seq data were deposited in GEO through numbers GSE252596 and GSE248500. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-62475-6.