Bioinformatics analysis of differentially expressed exosomal circRNAs in human GC tissues

To analyse the differential expression of circRNAs in exosomes between gastric cancer and normal gastric mucosal epithelial tissues, we isolated exosomes from gastric cancer tissues and gastric mucosal epithelial tissues from healthy individuals via ultracentrifugation. As shown in Fig. 1a, typical extracellular vesicles with double membranes and diameters ranging from 30 to 150 nm were observed using electron microscopy. Nanoparticle tracking analysis (NTA) revealed that most of the diameters of the exosomes were greater than 100 nm (Fig. 1b). The detection of an endoplasmic reticulum membrane protein (calnexin) and signature exosomal markers (TSG101, CD81, Alix and CD9) verified the purity of the extracted exosomes (Fig. 1c).

Fig. 1

Bioinformatics analysis of differentially expressed exosomal circRNAs in human GC tissues. A Extracellular vesicles with double membranes, ranging in diameter from 30 to 150 nm, extracted from GC tissue and normal control were observed using electron microscopy. Scale bar: 500 nm (left), 100 nm (right). B Nanoparticle tracking analysis (NTA) revealed that the diameters of most purified exosomes were greater than 100 nm. C Extracted exosomes and cell lysate were subjected to Western blot analysis for signature exosomal markers (TSG101, CD81, Alix and CD9) and an endoplasmic reticulum membrane protein (calnexin), respectively. D The origin and distribution of all 33,997 circRNAs detected by exosomal circRNA sequencing. E Relative expression of the 9 circRNAs at the transcriptional level in 40 pairs of GC and adjacent normal tissues from GC patients. F circMAN1A2 (hsa_circ_0000118) was spliced from exons 2,3,4,5 of the precursor mRNA transcribed from the MAN1A2 gene located on chromosome 1, forming a circular transcript of 553 nt in length. Sanger sequencing confirmed the head-to-tail back-splicing of circMAN1A2. G The expression levels of circMAN1A2 in six different human GC cellular exosomes (HGC27 exo, MKN28 exo, KATOIII exo, SNU1 exo, MKN45 exo, and AGS exo). Data were normalized to the expression levels of circMAN1A2 in exosomes derived from normal gastric mucosal tissue. H qRT-PCR analysis of the level of circMAN1A2 and linear MAN1A2 mRNA after treatment with RNase R in HGC27 and AGS. I Relative levels of circMAN1A2 and MAN1A2 mRNA were measured by qRT-PCR in HGC27 treated with Actinomycin D for different periods of time. J The divergent primers detected circMAN1A2 in cDNA but not in gDNA, GAPDH was used as a negative control. K Relative levels of GAPDH (positive control for cytoplasmic fraction), U6 (positive control for nuclear fraction), circMAN1A2, and MAN1A2 mRNA from cytoplasmic and nuclear fractions in HGC27. L Fluorescence in situ hybridization (FISH) was conducted to confirm the sub-cellular localization of circMAN1A2 in HGC27 and AGS. Scale bar: 20 μm. M Relative circMAN1A2 expression at the transcriptional level in 80 paired GC and adjacent tissues. N Overall survival analysis based on the circMAN1A2 expression level in 80 GC patients. The median circMAN1A2 expression level was used as a cutoff. O The comparison of circMAN1A2 expression levels in plasma exosomes of GC patients and individuals without GC. P ROC curves of plasma exosomal circMAN1A2, serum CEA, and the combination of the two indicators (circMAN1A2: AUC = 0.685, 95% CI: 0.563–0.806, p = 0.008; CEA: AUC = 0.696, 95% CI: 0.575–0.818, p = 0.005; circMAN1A2 + CEA: AUC = 0.798, 95% CI: 0.693–0.903, p < 0.001). Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

Exosomal circRNA sequencing was subsequently performed. A total of 33,997 circRNAs were identified, and the origin and distribution of these circRNAs are shown in Fig. 1d. It can be seen that most of the circRNAs detected are formed from exons by back-splicing. The start–end length distributions of the genes corresponding to these circRNAs were queried using circAtlas (http://circatlas.biols.ac.cn) (Supplementary Fig. 1a). The chromosomal locus distributions of these circRNAs are shown in Supplementary Fig. 1b. Then, we focused on circRNAs enriched in GC-derived exosomes. After obtaining the intersection, we identified 75 circRNAs (Supplementary Fig. 1c). These 75 circRNAs were distributed on chromosomes 1, 3, 10 and 13 (Supplementary Fig. 1d). A total of 9 circRNAs were identified in more than five databases (circAtlas, Uniform, circBase, circRNADb, deepbase2, and circpedia2). Relative expression of these 9 circRNAs at the transcriptional level in 40 pairs of GC and adjacent normal tissues were quantified using qRT-PCR (Fig. 1e and Supplementary Fig. 1e, f). The results demonstrated that the difference in the expression of hsa_circ_0000118 between GC and adjacent normal tissues was the most significant (p = 0.0026). Then, we examined the expression levels of hsa_circ_0000118 in exosomes derived from 20 pairs of GC and adjacent normal tissues and found that its expression was elevated in GC-derived exosomes (p = 0.0074, Supplementary Fig. 1g). Therefore, we further explored the role of hsa_circ_0000118 in the biological progression of GC.

Identification of circMAN1A2 and the clinical features of circMAN1A2

Hsa_circ_0000118 (termed “circMAN1A2” in this study) was spliced from exons 2, 3, 4 and 5 of the precursor mRNA transcribed from the MAN1A2 gene located on chromosome 1 (UCSC data in NCBI). CircMAN1A2 formed a circular transcript of 553 nt (Exon 2, 256 nt; Exon 3, 97 nt; Exon 4, 119 nt; Exon 5, 81 nt). We confirmed the head-to-tail back-splicing of circMAN1A2 using Sanger sequencing (Fig. 1f).

Next, circMAN1A2 expression levels were validated in six different human GC cell lines (HGC27, MKN45, KATOIII, SNU1, MKN28, and AGS) and two gastric mucosal epithelial tissues from healthy individuals. The expression levels of circMAN1A2 were elevated to varying degrees in GC cells and exosomes (Fig. 1g and Supplementary Fig. 2a). Notably, the most significant differences in circMAN1A2 expression levels were observed in both the cells and exosomes derived form the HGC27 and AGS cell lines. These two cell lines were selected for subsequent studies.

We performed several experiments to identify the circular characteristics of circMAN1A2. First, as shown in Fig. 1h, circMAN1A2 was resistant to RNase R exonuclease digestion compared with the linear form of MAN1A2. Next, we used actinomycin D, an inhibitor of transcription, to measure the changes in the half-life of circMAN1A2 and MAN1A2 mRNAs in both the HGC27 and AGS cell lines. The results suggested that circMAN1A2 is more stable than MAN1A2 mRNA (Fig. 1i and Supplementary Fig. 2b). Next, we designed divergent and convergent primers to identify circMAN1A2. By employing complementary DNA (cDNA) and genomic DNA (gDNA) extracted from HGC27 and AGS cells as templates, we found that circMAN1A2 was amplified exclusively from cDNA but not gDNA, indicating that the possibility of genomic rearrangements or trans-splicing was excluded (Fig. 1j). These experiments clearly confirmed the circular structure of circMAN1A2.

We subsequently explored the subcellular localization of circMAN1A2 in HGC27 and AGS cells. Following nucleocytoplasmic separation, we probed the relative circMAN1A2 expression levels in the cytoplasm and nucleus of the cells using qRT-PCR. The results revealed that circMAN1A2 was predominantly expressed in the cytoplasm of HGC27 and AGS cells (Fig. 1k and Supplementary Fig. 2c). The fluorescence in situ hybridization (FISH) assay further confirmed that circMAN1A2 is mainly localized in the cytoplasm (Fig. 1l).

We performed further qRT-PCR to determine the relative circMAN1A2 expression levels in paired cancer and adjacent tissues from 80 GC patients. The results confirmed greater circMAN1A2 expression levels in cancer tissues compared with adjacent tissues (Fig. 1m; p = 0.0010). Next, we integrated and analysed the clinical information of these patients. Patients were divided into high and low expression groups according to the median circMAN1A2 expression level. Statistical analysis revealed that circMAN1A2 expression levels correlated with lymph node metastasis and TNM stage (Table 1; Lymph node metastasis: p = 0.024, TNM stage: p = 0.006). Subsequently, using Kaplan–Meier survival analysis, we found that patients with high circMAN1A2 expression levels had a significantly worse prognosis (Fig. 1n; HR = 2.917, 95% CI = 1.306–6.516, p = 0.0120).

Table 1 Correlation between circMAN1A2 expression and the clinicopathologic parameters of 80 GC patients

To assess the diagnostic value of circMAN1A2, we collected blood from 46 GC patients and 28 individuals without GC (non-GC). qRT-PCR revealed that circMAN1A2 expression levels in the plasma exosomes of GC patients were significantly elevated compared with that of non-GC individuals (Fig. 1o; p = 0.0091). The receiver operating characteristic (ROC) curve results suggested that plasma exosomal circMAN1A2 in combination with CEA, a commonly used clinical serum tumour marker for GC, is instructive for the diagnosis of GC patients (Fig. 1p; circMAN1A2: AUC = 0.685, 95% CI: 0.563–0.806, p = 0.008; CEA: AUC = 0.696, 95% CI: 0.575–0.818, p = 0.005; circMAN1A2 + CEA: AUC = 0.798, 95% CI: 0.693–0.903, p < 0.001).

In summary, circMAN1A2 expression was confirmed to be upregulated in GC tissues and plasma exosomes. High circMAN1A2 expression is correlated with poor prognosis in GC patients. Thus, circMAN1A2 exhibits potential as a clinically valuable biomarker for GC.

CircMAN1A2 promotes GC cell proliferation and migration in vitro

To study the biological role of circMAN1A2 in GC progression, we constructed three small interfering RNAs (siRNAs) that target the back-splicing region of circMAN1A2. Among them, si-circ-1 and si-circ-3 successfully inhibited circMAN1A2 expression in HGC27 and AGS cells without affecting the expression levels of the linear form of MAN1A2 (Supplementary Fig. 3a, b). Additionally, we transfected both GC cell lines with a circMAN1A2 overexpression plasmid. qRT-PCR was used to verify the transfection efficiency (Supplementary Fig. 3c).

We performed colony formation assays (Fig. 2a, b and Supplementary Fig. 4a, b), CCK-8 assays (Fig. 2c, d, e and Supplementary Fig. 4c, d, e) and EdU incorporation assays (Fig. 2f, g and Supplementary Fig. 4f, g) to explore the effects of circMAN1A2 expression on the proliferative capacity of GC cells. The results showed that the proliferation capacities of HGC27 and AGS cells transfected with siRNAs were significantly inhibited. In contrast, the proliferation capacity of HGC27 and AGS cells transfected with the circMAN1A2 overexpression plasmid was significantly increased. Moreover, wound healing experiments and Transwell assays were employed to explore the effects of circMAN1A2 on the migratory capacity of GC cells (Fig. 2h, i, j, k and Supplementary Fig. 4 h, i, j, k). The results indicated that circMAN1A2 promoted the migratory ability of GC cells; conversely, silencing circMAN1A2 expression significantly suppressed GC cell migratory capacity. We subsequently used flow cytometry to explore whether circMAN1A2 interferes with the cell cycle of HGC27 and AGS cells. The results revealed that interfering with circMAN1A2 expression decreased the percentage of S phase cells and increased the percentage of G0/G1 phase cells, and the opposite result was observed after circMAN1A2 overexpression (Fig. 2l, m and Supplementary Fig. 4l, m).

Fig. 2

CircMAN1A2 promotes GC cell proliferation and migration in vitro. A, B The colony formation assay was performed to evaluate proliferation ability after upregulating or downregulating circMAN1A2 in HGC27 cells. C-E The CCK8 assay was performed to evaluate proliferation ability after upregulating or downregulating circMAN1A2 in HGC27 cells. F, G The EdU incorporation assay was performed to evaluate proliferation ability after upregulating or downregulating circMAN1A2 in HGC27 cells. Scale bar: 100 μm. H, I The wound healing experiment was performed to evaluate migration ability after upregulating or downregulating circMAN1A2 in HGC27 cells. Scale bar: 500 μm. J, K Transwell assay was performed to evaluate migration ability after upregulating or downregulating circMAN1A2 in HGC27 cells. Scale bar: 100 μm. L, M The effect of circMAN1A2 on modulating HGC27 cell cycle progression was evaluated by flow cytometry assay. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

To further validate whether circMAN1A2 also exerts the biological function through entry into exosomes, we modulated the expression level of circMAN1A2 in HGC27 cells and then extracted cellular exosomes from different groups. We examined the extracted exosomes for circMAN1A2 expression levels (Supplementary Fig. 5a). Then, extracted exosomes were added to HGC27 cells for coculture. Colony formation, CCK-8, and EdU incorporation assays showed that the proliferative capacity of HGC27 cells was significantly reduced when coculturing with exosomes extracted form circMAN1A2 knockdown HGC27 cells. In contrast, exosomes derived from circMAN1A2-overexpressing HGC27 cells promoted the proliferation of HGC27 cells (Supplementary Fig. 5b, c, d, e, f). Furthermore, wound healing and Transwell assays demonstrated that coculturing HGC27 cells with exosomes from circMAN1A2-overexpressing HGC27 cells significantly enhanced cell migration, whereas co-culturing with exosomes from circMAN1A2-silenced HGC27 cells weakened cell migration (Supplementary Fig. 5 g, h, i, j). These findings suggest that exosomal circMAN1A2 promotes the malignant biological behaviors of GC cells.

Collectively, these findings confirmed that circMAN1A2 promotes GC cell proliferation and migration in vitro.

CircMAN1A2 promotes GC proliferation and metastasis in vivo

We then performed animal experiments to determine whether circMAN1A2 could promote the malignant biological behaviours of GC in vivo.

HGC27 and AGS cells with stable circMAN1A2 knockdown by sh-circMAN1A2 were injected subcutaneously into nude mice, and the size of the subcutaneous tumours was observed weekly. Four weeks later, the nude mice were euthanized, and the subcutaneous tumours were removed. The size and weight of the tumours were measured for comparison of the differences between the groups. The results showed that the knockdown of circMAN1A2 limited the growth of subcutaneous xenograft tumours. In contrast, GC cells transfected with the plasmid overexpressing circMAN1A2 were able to promote subcutaneous tumour growth (Fig. 3a, b, c).

Fig. 3

CircMAN1A2 promotes GC proliferation and metastasis in vivo. A-C The designated tumour cells were injected subcutaneously into nude mice, and xenograft tumours were collected 28 days later. Tumour volume was measured and calculated every week. Tumour weight was measured after 28 days. D, E Immunohistochemistry staining and IHC scores of Ki-67 in respective xenograft tumour tissues. Scale bar: 50 μm. F, G The representative Bioluminescence images of liver metastases after injecting tumour cells into the spleen of nude mice and quantification of these Bioluminescence images. H The representative photographs of liver tissues. White arrows point to liver metastases. I Representative images of HE staining of liver tissues. Black arrows point to liver metastases. Scale bar: 3 mm. J Liver index (liver weight/ body weight) of each group were calculated. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

We then performed immunohistochemical staining. Ki-67 (a proliferation marker) staining was used to detect the proliferative capacity of the subcutaneous tumours (Fig. 3d). By calculation of IHC scores, we found that tumours with stable knockdown of circMAN1A2 showed weaker Ki-67 staining, while tumours in the cirMAN1A2-overexpression group had stronger Ki-67 staining. Thus we concluded that circMAN1A2 promotes GC proliferative capacity in vivo (Fig. 3e).

Next, we stably transfected the luciferase plasmids into the above cells to investigate the effect of circMAN1A2 on tumour metastasis. We constructed a liver metastasis (LM) model of GC in nude mice via the injection of GC cells into the spleen. Four weeks after surgery, we examined liver metastases in the nude mice via an IVIS Spectrum Imaging System (Fig. 3f, g). Then, we euthanized the nude mice, removed the liver tissues, and observed the number of liver metastases with the naked eye (Fig. 3h). Subsequently, the liver tissues were subjected to paraffin-embedded sectioning and H&E staining. By microscopic observation of H&E sections, we found that the number of GC-LM lesions was significantly suppressed after circMAN1A2 expression was knocked down. In contrast, when circMAN1A2 was overexpressed, the LM capacity increased (Fig. 3i, j). These results suggest that circMAN1A2 enhances GC metastasis. We further confirmed this finding using IHC staining of E-cadherin (the tumour aggressiveness marker) in subcutaneous tumours (Supplementary Fig. 6a, b).

Based on the above experiments, we confirmed that circMAN1A2 plays a biological role in promoting GC progression in vivo.

CircMAN1A2 is packaged into exosomes via hnRNPA2B1 and delivered to T cells, which suppresses immune properties

We then investigated how circMAN1A2 is packaged into exosomes in GC cells and how it functions through exosomes.

First, we analysed the mass spectrometry (MS) data of circMAN1A2 pull-down experiments in HGC27 cells (Supplementary Fig. 7a). HnRNPA2B1, which has been reported to package a variety of RNAs into exosomes, attracted our attention (Supplementary Fig. 7b) [21, 22]. RIP-qPCR and RNA pull-down experiments were performed in HGC27 and AGS cells to validate the interaction between circMAN1A2 and hnRNPA2B1 (Supplementary Fig. 7c, d). Following the knockdown of hnRNPA2B1 expression in HGC27 and AGS cells, circMAN1A2 expression decreased significantly in the extracted exosomes (Supplementary Fig. 7e). These findings further confirmed that circMAN1A2 is encapsulated in exosomes through interactions with hnRNPA2B1.

Exosomes secreted by tumour cells can influence immune infiltrating cells in the tumour microenvironment and thus modulate antitumour immunity [23, 24]. CD8+ T cells have long been considered among the most representative antitumour cells. Therefore, to investigate whether GC-derived exosomal circMAN1A2 affects CD8+ T cells, we performed further experiments. Exosomes extracted from HGC27 cells were stained with PKH67 and then cocultured with Jurkat cells. Exosomes can be taken up by Jurkat cells, suggesting that GC-derived exosomes may enter T cells to affect the tumour immune microenvironment (Fig. 4a). Therefore, we extracted exosomes from different groups of GC cells with circMAN1A2 overexpression or silencing and cocultured them with Jurkat cells. ELISAs verified that, compared with those derived from control cells, exosomes derived from circMAN1A2-overexpressing GC cells inhibited IL-2 secretion from Jurkat cells (Fig. 4b). Flow cytometry analysis revealed that the secretion of T-cell effector indicators, such as IFN-γ and TNF-α, was suppressed when Jurkat cells were cocultured with exosomes extracted from GC cells overexpressing circMAN1A2. In contrast, inhibition of circMAN1A2 expression in GC cells resulted in a significant increase in effector secretion by T cells (Fig. 4c, d). After coculturing the above treated Jurkat cells with GC cells for 24 h, we measured the degree of apoptosis in the GC cells using flow cytometry. Jurkat cells in the circMAN1A2-exo group inhibited the apoptosis of GC cells, whereas those in the sh-circMAN1A2-exo group exhibited the opposite trend (Fig. 4e, f). TUNEL staining further revealed that Jurkat cells, after being co-cultured with exosomes derived from circMAN1A2-silenced GC cells, showed a significant increase in TUNEL-positive GC cells when co-cultured with corresponding GC cells. In contrast, Jurkat cells co-cultured with exosomes from circMAN1A2-overexpressing GC cells exhibited a notable reduction in TUNEL-positive GC cells following co-culture with corresponding GC cells. These findings provided further evidence that circMAN1A2 inhibited the effector function of T cells (Fig. 4g, h).

Fig. 4

CircMAN1A2 is packaged into exosomes and delivered to T cells, suppressing the antitumour immunity. A HGC27-derived exosomes, labeled with PKH-67(green), were found to be taken up by Jurkat cells. Scale bar: 100 μm. B ELISA assays evaluated the secretion levels of IL-2 from Jurkat cells under the influence of GC-derived exosomal circMAN1A2. C, D Representative images of the flow cytometry assay showed the percentage of CD8 + IFN-γ + and CD8 + TNF-α + Jurkat cells under the influence of GC-derived exosomal circMAN1A2. E, F GC cells were cocultured with Jurkat cells for 24 h and then the apoptosis rates of GC cells were evaluated by flow cytometry, using annexin V-FITC and propidium iodide (PI) double labeling. G, H Representative images of TUNEL staining of GC cells after cocultured with different treatments of Jurkat cells. Scale bars: 20 μm. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

From the above experiments, we confirmed that circMAN1A2 is encapsulated in exosomes through hnRNPA2B1 and that T cells can take up GC-derived exosomes. GC-derived exosomal circMAN1A2 affects the immune activation and effector function of T cells, thus regulating the antitumour immune response.

CircMAN1A2 exerts its biological function by binding to SFPQ in GC cells and T cells

Based on the above findings, we preliminarily verified that circMAN1A2 promotes the malignant biological behaviour of GC and inhibits the antitumour immunity of T cells. Therefore, we explored specific molecular mechanisms involved.

It has been reported that circRNAs function mainly by acting as competing endogenous RNAs (ceRNAs) and interacting with RNA-binding proteins (RBPs). We performed RNA pull-down experiments in HGC27 and Jurkat cells. AGO2 protein was not detected, suggesting that ceRNA may not be the main mechanism by which circMAN1A2 functions (Supplementary Fig. 8a). Next, we subjected the products of RNA pull-down in Jurkat cells to mass spectrometry (Fig. 5a). Combined with the mass spectrometry results from HGC27 cells (Supplementary Fig. 7a), SFPQ proteins were found to be present in both HGC27 and Jurkat cells (Fig. 5b).

Fig. 5

CircMAN1A2 exerts its biological function by binding to SFPQ in GC cells and T cells. A Protein bands detected by silver stain for mass spectrometry of the circMAN1A2-protein complex pulled down by sense or anti-sense circMAN1A2 in Jurkat cells. B The typical SFPQ peptide was identified in circMAN1A2-enriched proteins based on MS analysis. C RNA pull-down and Western blot assays were performed to confirm the interaction between SFPQ and circMAN1A2 in Jurkat and HGC27 cells. D RIP and qRT-PCR assays showed the interaction between SFPQ and circMAN1A2 in Jurkat and HGC27 cells, using IgG and SFPQ antibodies, n = 3. E Immunofluorescence-FISH was conducted to confirm the sub-cellular localization of circMAN1A2 and SFPQ in HGC27 and Jurkat cells. Scale bar: 20 μm. F The EdU incorporation assay was performed to evaluate proliferation ability after downregulating SFPQ in HGC27 and AGS cells. Scale bar: 100 μm. G The colony formation assay was performed to evaluate proliferation ability after downregulating SFPQ in HGC27 and AGS cells. H Transwell assay was performed to evaluate migration ability after downregulating SFPQ in HGC27 and AGS cells. Scale bar: 100 μm. I The effect of SFPQ on modulating GC cell cycle progression was evaluated by flow cytometry assay. J Representative images of the flow cytometry assay showed the percentage of CD8 + IFN-γ + and CD8 + TNF-α + Jurkat cells after downregulating SFPQ. K ELISA assays evaluated the secretion levels of IL-2 from Jurkat cells after downregulating SFPQ. L HGC27 cells were cocultured with Jurkat cells (downregulating SFPQ or not) for 24 h and then the apoptosis rates of HGC27 cells were evaluated by flow cytometry, using annexin V-FITC and propidium iodide (PI) double labeling. M Representative images of TUNEL staining of HGC27 cells after cocultured with different treatments of Jurkat cells. Scale bars: 20 μm. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

SFPQ, splicing factor proline- and glutamine-rich, is a ubiquitous RNA-binding protein that is localized primarily in the nucleus and cytoplasm [25]. Dysregulation of SFPQ is often observed in neurological disorders such as Alzheimer's disease (AD) and amyotrophic lateral sclerosis (ALS) [26, 27]. However, reports suggest that SFPQ plays a significant role in tumourigenesis and progression by regulating the activation and repression of transcription, regulating splicing, and affecting the innate immune response [28, 29]. Therefore, we further explored its role in the biological progression of GC.

Using TIMER (Tumour Immune Estimation Resource, http://timer.cistrome.org/), we found that SFPQ was significantly negatively correlated with CD8 + T-cell infiltration (Supplementary Fig. 8b). RPIseq prediction revealed that SFPQ exhibited high binding potential with circMAN1A2 (Supplementary Fig. 8c) [30]. Then, we performed RNA pull-down experiments and RIP-qPCR experiments and found that circMAN1A2 interacts with SFPQ in both GC cells and Jurkat cells (Fig. 5c, d). Immunofluorescence-FISH confirmed the localization of circMAN1A2 and SFPQ in the cytoplasm of HGC27 cells and Jurkat cells (Fig. 5e). After interfering with SFPQ expression (Supplementary Fig. 9a), HGC27 and AGS cell proliferation and migration abilities were found to be inhibited by EdU incorporation assays, colony formation assays and Transwell assays (Fig. 5f, g, h and Supplementary Fig. 9b, c, d). HGC27 and AGS cell proliferation cycles were significantly arrested as detected by flow cytometry assays when silencing SFPQ expression (Fig. 5i and Supplementary Fig. 9e). In Jurkat cells, after downregulating SFPQ, the percentages of CD8 + IFN-γ + and CD8 + TNF-α + cells were increased (Fig. 5j and Supplementary Fig. 9f, g), and the secretion levels of IL-2 were elevated (Fig. 5k). When cocultured with SFPQ-silencing Jurkat cells, the apoptosis rates of HGC27 cells were found to be elevated compared with controls (Fig. 5l and Supplementary Fig. 9 h). TUNEL staining of HGC27 cells further showed the increase in TUNEL-positive cell numbers when cocultured with SFPQ-silencing Jurkat cells (Fig. 5m and Supplementary Fig. 9i), which indicated that the effector function was increased after interfering with SFPQ expression in Jurkat cells. Therefore, we suggested that SFPQ is a downstream molecule of circMAN1A2 in both HGC27 and Jurkat cells.

We subsequently aimed to investigate the structural regions of circMAN1A2 that interact with SFPQ. We utilized the catRAPID fragment module (http://s.tartaglialab.com/page/catrapid_group), an online site for RNA–protein binding prediction. In circMAN1A2, the 51–104, 126–179, 226–352 and 378–429 nt regions exhibit the potential to bind to SFPQ (Supplementary Fig. 10a). On the basis of these four regions, we designed four different circMAN1A2 truncation probes (Supplementary Fig. 10b). RNA pull-down assays were conducted with the full-length and truncated probes of circMAN1A2 to identify the specific areas that bind to SFPQ. We detected SFPQ proteins in the RNA pull-down assays of the experimental groups numbered 1, 3, 5, and 6 but not in the group numbered 4 (Supplementary Fig. 10c). Thus, we inferred that SFPQ mainly binds to the 226–352 nt region of circMAN1A2.

We subsequently constructed circMAN1A2-ΔSFPQ and verified that it lost the ability to bind to SFPQ via RIP-qPCR experiments in HEK293T cells (Fig. 6a). A series of cellular experiments, such as colony formation assays, Transwell assays and cell cycle assays, revealed that circMAN1A2-ΔSFPQ does not affect the malignant biological progression of GC cells (Fig. 6b, c, d, e, f, g, h, i).

Fig. 6

CircMAN1A2 interacts with SFPQ to promote GC progression and suppress immune activation and tumour killing effects of T cells. A RIP and qRT-PCR assays showed the interaction between SFPQ and circMAN1A2 but not circMAN1A2-△SFPQ in HEK293T cells, using IgG and SFPQ antibodies, n = 3. B, C The EdU incorporation assay was performed to evaluate proliferation ability after upregulating circMAN1A2 or circMAN1A2-△SFPQ in HGC27 cells. Scale bar: 100 μm. D, E The colony formation assay was performed to evaluate proliferation ability after upregulating circMAN1A2 or circMAN1A2-△SFPQ in HGC27 cells. F, G Transwell assay was performed to evaluate migration ability after upregulating circMAN1A2 or circMAN1A2-△SFPQ in HGC27 cells. Scale bar: 100 μm. H, I The effects of circMAN1A2 and circMAN1A2-△SFPQ on modulating HGC27 cell cycle progression were evaluated by flow cytometry assay. J ELISA assays evaluated the secretion levels of IL-2 from Jurkat cells after cocultured with exosomes extracted from circMAN1A2 or circMAN1A2-△SFPQ overexpressed HGC27 cells. K-M Representative images of the flow cytometry assay showed the percentage of CD8 + IFN-γ + and CD8 + TNF-α + Jurkat cells after cocultured with exosomes extracted from circMAN1A2 or circMAN1A2-△SFPQ overexpressed HGC27 cells. N, O HGC27 cells were cocultured with Jurkat cells (cocultured with exosomes extracted from circMAN1A2 or circMAN1A2-△SFPQ overexpressed HGC27 cells) for 24 h and then the apoptosis rates of HGC27 cells were evaluated by flow cytometry, using annexin V-FITC and propidium iodide (PI) double labeling. P, Q Representative images of TUNEL staining of HGC27 cells after cocultured with different treatments of Jurkat cells. Scale bars: 20 μm. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

Next, we extracted exosomes from HGC27 cells overexpressing circMAN1A2 or overexpressing circMAN1A2-ΔSFPQ and cocultured them with Jurkat cells. The levels of secreted IL-2 in Jurkat cells were detected via ELISA, and the levels of secreted indicators of CD8 + T cells, such as IFN-γ and TNF-α, were detected using flow cytometry analysis. The results showed that circMAN1A2-ΔSFPQ does not have the ability to suppress CD8 + T-cell immune activation (Fig. 6j, k, l, m). In addition, the apoptosis levels of GC cells cocultured with the circMAN1A2-exo group Jurkat cells were reduced, whereas no significant differences were noted between the circMAN1A2-ΔSFPQ group and the control group (Fig. 6n, o). TUNEL staining also confirmed that circMAN1A2-ΔSFPQ reverses the effect of GC cell apoptosis inhibition by Jurkat cells in the circMAN1A2 group (Fig. 6p, q).

From the above experiments, we concluded that circMAN1A2 promotes GC progression by binding to SFPQ through its 226–352 nt region and that GC-derived exosomal circMAN1A2 inhibits immune activation and tumour killing effects by binding to SFPQ in T cells through its 226–352 nt region.

Single-cell analysis confirmed that SFPQ promotes GC progression by regulating the cell cycle of GC cells and suppressing the antitumour immunity of T cells

To investigate the mechanism of SFPQ function in GC and T cells, we employed single-cell analysis, which is able to assess the distribution of expression levels of each gene in different cell populations and can address cell-specific changes in the transcriptome. We integrated two large single-cell RNA sequencing (scRNAseq) datasets from the GEO database related to gastric cancer (GSE183904, GSE150290). A series of data quality control measures were first performed, including data integration, normalization and batch-effect correction. These procedures were followed by dimensionality reduction and clustering of the integrated data and cell identification on the basis of the expression of classical marker genes. The integrated data were clustered into 23 subpopulations and then classified into 10 cell types after cell type annotation (Fig. 7a, b). We selected the epithelial cell and T-cell subpopulations for further dimensionality reduction and clustering and visualized the distribution of SFPQ genes among them (Fig. 7c, d). We categorized epithelial cells and T cells into high and low SFPQ expression groups on the basis of their expression levels. We subsequently detected the differentially expressed genes (DEGs) and performed KEGG enrichment analyses. Among the enriched pathways, as shown in Fig. 7e, the DEGs in the high SFPQ expression group were enriched in “regulation of cell cycle phase transition” in the epithelial cell subpopulation. Moreover, the DEGs in the low SFPQ expression group were enriched in the “T-cell receptor (TCR) signalling pathway” in the T-cell subpopulation (Fig. 7f).

Fig. 7

Single-cell analysis confirms that SFPQ promotes GC progression by regulating the G1/S phase transition of cell cycle and suppresses antitumour immunity of T cells by inhibiting the TCR signalling pathway. A Plotting of 23 cell clusters in single-cell RNA sequencing. B Plotting of 10 cell types after cell type annotation in single-cell RNA sequencing. C Distribution of SFPQ genes among epithelial cells. D Distribution of SFPQ genes among T cells. E Gene Set Enrichment Analysis of the differentially expressed genes in the SFPQ high expression group epithelial cell subpopulation. F Gene Set Enrichment Analysis of the differentially expressed genes in the SFPQ low expression group T cell subpopulation. G, H The expression levels of key proteins in G1/S phase progression (CCND1, CDK4, CDK6 and phosphorylated retinoblastoma) among GC cells with different treatments. I The phosphorylation levels of the key signalling proteins in the TCR signalling pathway among Jurkat cells with different treatments. J-L The designated GC cells were injected subcutaneously into huPBMC-NCG mice and different drugs were injected meanwhile. Xenograft tumours were collected 28 days later. Tumour volume was measured and calculated every week. Tumour weight was measured after 28 days. M, N Multiplex immunohistochemical (mIHC) detected the expression levels of CD8 and GZMB in tumour tissues of different groups. Scale bar: 50 μm. Graph represents mean ± SD; *p < 0.05, **p < 0.01, and ***p < 0.001

The cell cycle phase transition is tightly coordinated to ensure efficient cell cycle progression and genomic stability. Disruption of the cell cycle phase transition can lead to various diseases, including malignant tumours. Together, our findings confirmed that circMAN1A2 promotes the cell cycle progression of GC cells and that the GC cell cycle is arrested after circMAN1A2 expression is silenced (Fig. 2l and Supplementary Fig. 4 l). Similarly, cell cycle blockage in GC cells was observed when SFPQ was silenced (Fig. 5i). Interestingly, we found that changes in various phases of the cell cycle occur mainly in the G1 and S phases. Therefore, we hypothesized that SFPQ promoted the GC cell cycle by regulating the G1/S phase transition.

During the G1/S phase transition, Cyclin D1 (CCND1) is activated and binds Cyclin-dependent kinase 4/6 (CDK4/6), which phosphorylates the retinoblastoma protein (Rb). Then, the E2F transcription factor is released, contributing to the progression from G1 to S phase. Here, we examined the expression levels of key proteins involved in G1/S phase progression (CCND1, CDK4, CDK6 and phosphorylated retinoblastoma). We found that CDK4 and phosphorylated retinoblastoma (pRb) were upregulated after SFPQ overexpression. After SFPQ expression was silenced, CDK4 and pRb expression levels were significantly decreased (Fig. 7g, h). Thus, we hypothesized that SFPQ catalyses the G1/S phase transition and promotes GC cell proliferation.

The TCR signalling pathway plays indispensable roles in T-cell development, differentiation and the antitumour immune response. Recently, owing to the rise of cell therapies and immunotherapies, the TCR signalling pathway has attracted considerable attention in the field of cancer therapy. On the basis of the results of our single-cell analysis, we hypothesized that high SFPQ expression leads to blockade of the activation of the TCR signalling pathway, which subsequently affects the antitumour immunity of T cells. LCK, ZAP70, PLCγ1 and LAT phosphorylation are critical in the TCR signalling pathway [31,32,33]. We transfected a plasmid overexpressing SFPQ into Jurkat cells and stimulated the TCR using an anti-CD3/CD28 antibody. As shown in Fig. 7i, we found that the phosphorylation levels of these key signalling proteins in the SFPQ-overexpressing group were almost all reduced, whether at the basal level or under TCR stimulation at different time points. Therefore, we suggest that SFPQ affects the antitumour activity of T cells by inhibiting TCR signalling pathway activation.

Next, we verified that circMAN1A2 promotes GC progression in vivo by inhibiting TCR signalling pathway activation. We transplanted human peripheral blood mononuclear cells (PBMCs) into the NCG of severely immunodeficient mice and constructed a huPBMC-NCG mouse model for subsequent validation in animal experiments. Using subcutaneous tumour loading experiments, we found that, compared with those in the control group, the sizes of the xenograft tumours in the circMAN1A2-overexpressing group were significantly greater. However, anti-PD1 treatment reversed this trend, and the growth of subcutaneous tumours was further inhibited by the addition of the TCR receptor agonist LYP-IN-3 (Fig. 7j, k, l). Multiplex immunohistochemistry (mIHC) revealed that LYP-IN-3 increased the amount of GZMB secreted by cytotoxic T cells (Fig. 7m, n). The above experiments further suggest that circMAN1A2 affects the biological progression of GC by inhibiting the TCR signalling pathway.

Overall, we conclude that circMAN1A2 exerts its biological function by binding to SFPQ. Specifically, circMAN1A2 regulates the malignant biological behaviour of GC by contributing to the G1/S phase transition of the cell cycle. And in T cells, it regulates antitumour immune activity by inhibiting TCR signalling pathway activation.

CircMAN1A2 interacts with the RRM1 domain of SFPQ and inhibits ubiquitin–proteasome-mediated degradation of SFPQ

We then explored the specific mechanisms by which circMAN1A2 interacts with SFPQ.

First, we identified the specific domains of SFPQ necessary for the interaction with circMAN1A2. Acco