circRNA-CREIT was downregulated in TNBC

To identify the dysregulated circRNAs in human breast cancer, we profiled circRNA transcripts from six paired breast cancer tissues and adjacent normal mammary tissues using RNA-sequencing (RNA-seq) analysis. As shown in Fig. 1A, B, there were 108 differentially expressed circRNAs between the two groups with the criteria of fold change > 1.5 and p value < 0.05. Then, five dysregulated circRNAs (hsa_circ_0001346, hsa_circ_0002484, hsa_circ_0000231, hsa_circ_0009043 and circRNA-CREIT (circBase ID: hsa_circ_0001798)) with high abundance were selected for further validation. Among them, the expression of circRNA-CREIT showed the most remarkable difference between the breast cancer tissues and paired normal counterparts (Fig. 1C, D and Additional file 1: Fig. S1A). The ISH assay also verified the significant downregulation of circRNA-CREIT in breast cancer tissues (Fig. 1E).

Fig. 1

CircRNA-CREIT was downregulated in TNBC. A A volcano plot showing the upregulated and downregulated circRNAs in breast cancer tissues. B Circos plot indicating the differentially expressed circRNAs. The outermost circle shows the chromosomal distribution of the circRNAs. The second circle shows the expression levels of the indicated circRNAs. The third circle indicates the circBase ID of the circRNAs. The fourth circle shows the logFC of the indicated circRNAs between the two groups. The innermost circle shows the logp values. C qRT-PCR assay was performed to detect the expression of circRNA-CREIT in the 6 paired breast cancer tissues and their normal counterparts. D Expression of circRNA-CREIT in 75 breast cancer tissues and normal mammary tissues detected by qRT-PCR assay. E ISH of circRNA-CREIT in breast cancer tissues and paired normal mammary tissues. Scale bars = 200 μm. F Kaplan–Meier survival analysis of circRNA-CREIThigh and circRNA-CREITlow patients. G Expression of circRNA-CREIT in breast cancer cell lines with different hormone receptor statuses. H (upper) Diagram showing the locus of circRNA-CREIT in the genome. (lower) Schematic illustration showing that exons 6–7 of human SPIDR circularize to form circRNA-CREIT. The black arrow represents the back-splicing site of circRNA-CREIT confirmed by Sanger sequencing. I Convergent and divergent primers were used to validate the loop structure of circRNA-CREIT. J The expression of circRNA-CREIT, linear SPIDR and β-actin in TNBC cells with or without RNase R treatment. K qRT-PCR analysis of circRNA-CREIT expression in cDNA reverse transcribed with random hexamer or oligo (dT) primers. L Relative RNA levels of circRNA-CREIT and linear SPIDR after actinomycin D treatment detected by qRT-PCR. M Detection of circRNA-CREIT expression in cytoplasmic and nuclear fractions of RNAs extracted from TNBC cells. N RNA FISH assay for circRNA-CREIT with the nucleus distinguished by DAPI. Scale bars = 10 μm. ns no significance; *p < 0.05; **p < 0.01; ***p < 0.001 compared with the controls

Next, 244 breast cancer patients were divided into high circRNA-CREIT expression group (n = 131) and low circRNA-CREIT expression group (n = 113). Kaplan–Meier plotter demonstrated that high expression of circRNA-CREIT was associated with a favorable prognosis (Fig. 1F). Multivariate analyses revealed that circRNA-CREIT was an independent predictive factor for the prognosis of breast cancer patients (Additional file 1: Table S4). Moreover, higher levels of circRNA-CREIT were significantly correlated with a lower pathological grade, a lower rate of lymph node metastasis and a smaller tumor size (Additional file 1: Fig. S1B-D). We then detected the expression of circRNA-CREIT in breast cancer cell lines and found that circRNA-CREIT was much lower in TNBC than in the non-TNBC subtype (Fig. 1G). Collectively, we speculated that circRNA-CREIT might play a suppressive role in the progression of breast cancer, especially in the TNBC subtype.

Characteristics of circRNA-CREIT in TNBC

CircRNA-CREIT is derived from the exons 6–7 of the SPIDR (scaffold protein involved in DNA repair) gene located on human chromosome 8. The back-splicing site was confirmed by Sanger sequencing in TNBC cells (Fig. 1H). To validate the circular structure of circRNA-CREIT, we designed convergent and divergent primers to amplify circRNA-CREIT and linear SPIDR, respectively. As shown in Fig. 1I, circRNA-CREIT could only be amplified from cDNA rather than gDNA, indicating that circRNA-CREIT was a back-splicing product of the pre-mRNA. Additionally, circRNA-CREIT was more resistant to RNase R degradation than the linear form of SPIDR mRNA (Fig. 1J, Additional file 1: Fig. S1E). Moreover, circRNA-CREIT could not be efficiently reverse-transcribed with oligo (dT) primers, consistent with its loop structure lacking a poly-A tail (Fig. 1K and Additional file 1: Fig. S1F). The actinomycin D assay demonstrated that the half-life of circRNA-CREIT was much longer than that of the linear SPIDR transcript, which verified the high stability of circRNA-CREIT (Fig. 1L and Additional file 1: Fig. S1G). Additionally, qRT‒PCR and fluorescence in situ hybridization (FISH) assays showed the predominant cytoplasmic distribution of circRNA-CREIT (Fig. 1M, N and Additional file 1: Fig. S1H).

The expressions of circRNAs usually depend on the expression levels of their host genes and back-splicing rate [22]. To explore the mechanisms underlying circRNA-CREIT downregulation, we detected the SPIDR levels in tumor tissues. As shown in Additional file 1: Fig. S2A, B, SPIDR was significantly increased in breast cancer tissues compared to their normal counterparts, which was not responsible for the downregulation of circRNA-CREIT. Therefore, we determined to investigate the roles of back-splicing in circRNA-CREIT biogenesis. It was previously reported that DExH-box helicase 9 (DHX9), a well-known RNA helicase, could bind with the inverted repeat Alu (IRAlu) sequence located in the flanking introns to unwind IRAlu pairs and reduce the biogenesis of circRNAs [23]. Our results demonstrated that DHX9 overexpression led to the downregulation of circRNA-CREIT, while silencing DHX9 increased its levels (Additional file 1: Fig. S2C). In addition, by analyzing the data from the TCGA and GEO databases, we found that DHX9 was significantly upregulated in breast cancer, especially in the TNBC subtype, and was related to the poor overall survival of patients (Additional file 1: Fig. S2D-G). Then the sequences of the flanking introns were compared to the Alu sequence using the BLAST tools. We found twenty Alu elements in intron 5 and five Alu elements in intron 7. Previous studies revealed that the Alu element targeted by DHX9 had a much shorter distance to its closest potential pairing partner than that not targeted [24]. Therefore, we selected the closest complementary IRAlu pair (Alu19 and Alu 21) for further investigation (Additional file 1: Fig. S2H). RIP assays confirmed the interaction of the DHX9 protein and the two Alu elements (Additional file 1: Fig. S2I).

CircRNA-CREIT enhanced DOX sensitivity of TNBC in vitro and in vivo

To explore the potential roles of circRNA-CREIT in TNBC, we constructed circRNA-CREIT-overexpressing plasmids and three different short hairpin RNAs (shRNAs) targeting the junction region. The overexpression and knockdown efficiencies were validated in TNBC cells (Additional file 1: Fig. S3A, B). We found that shRNA-3 had a better knockdown effect, and it was used in our subsequent studies. Moreover, the expression of host gene SPIDR was not influenced by the change of circRNA-CREIT levels (Additional file 1: Fig. S3C, D). Then we applied RNA-seq to screen differentially expressed genes in circRNA-CREIT overexpressing cells (Additional file 1: Fig. S4A). Gene Ontology (GO) analysis and Gene Set Enrichment Analysis (GSEA) for differentially expressed genes revealed that circRNA-CREIT might be closely associated with the regulation of cellular response to various environmental stresses, especially chemical-induced stresses and the process of apoptosis (Fig. 2A and Additional file 1: Fig. S4B). DOX is the first-line chemotherapeutic agent in TNBC treatment and has potent antitumor effects [25]. Thus, DOX was used in our following experiments. By conducting qRT-PCR and ISH assays, we found that circRNA-CREIT was significantly downregulated in chemotherapy-resistant breast cancer tissues and doxorubicin resistant TNBC cells (MDA-MB-231/DOXR) (Fig. 2B, C and Additional file 1: Fig. S4C, D). As shown in Fig. 2D and Additional file 1: Fig. S5A, circRNA-CREIT overexpression obviously improved the chemosensitivity of TNBC cells to DOX, with a significant decrease in the IC50 value, whereas reduced circRNA-CREIT expression levels had the opposite effect. The results of the colony formation assay were consistent with the above findings (Additional file 1: Fig. S5B). In addition, we found that circRNA-CREIT overexpression notably enhanced DOX-induced apoptosis, while circRNA-CREIT knockdown led to decreased cell apoptosis (Fig. 2E and Additional file 1: Fig. S5C, D).

Fig. 2

CircRNA-CREIT significantly enhanced the chemosensitivity of TNBC cells in vitro and in vivo. A GO analysis of differentially expressed genes in circRNA-CREIT-overexpressing MDA-MB-231 cells based on RNA-seq data. B Violin plot showing circRNA-CREIT expression in chemosensitive and chemoresistant breast cancer tissues, detected by qRT-PCR assays. C ISH assay of circRNA-CREIT in chemosensitive and chemoresistant breast cancer tissues. Scale bars = 200 μm. D The impact of circRNA-CREIT overexpression on the cytotoxic effects of DOX is shown. Cell viability was detected by MTT assays. The IC50 values with the 95% CIs are presented. E Western blotting was performed to detect the expression of apoptosis pathway markers. F Images of the xenograft tumors. Scale bars = 10 mm. G Growth curves of xenograft tumors in different groups after treatment. H The tumor weights of the xenograft tumors. I Representative images of IHC staining for Ki67 and cleaved caspase-3 in different groups. Scale bars = 100 μm. J Representative morphologies of breast cancer patient-derived organoids (PDOs) treated with increasing DOX concentrations. Scale bars = 200 μm. K The IC50 values and 95% CIs of the PDOs are shown. Cell viability was measured by CCK8 assays. L Heatmap showing the IC50 value and relative circRNA-CREIT expression level of each PDO. Three independent experiments were conducted for each result. *p < 0.05; **p < 0.01; ***p < 0.001 compared with the controls

To evaluate the effect of circRNA-CREIT on tumor chemosensitivity in vivo, tumor xenograft models were constructed in female nude BALB/c mice. The results showed that the tumor growth rate and tumor weight were significantly decreased by circRNA-CREIT overexpression in both the vehicle control group and the DOX-treated group (Fig. 2F–H). Immunohistochemistry analysis for Ki67 and cleaved caspase-3 was performed to detect the proliferative rate and apoptosis level in the xenograft tumors. Decreased proliferation and elevated apoptosis levels were observed in the circRNA-CREIT overexpression group, especially in the DOX-treated group (Fig. 2I). Conversely, circRNA-CREIT knockdown led to accelerated tumor growth, reduced chemosensitivity of tumors to DOX and downregulated apoptosis (Additional file 1: Fig. S5E-H). We then constructed a patient-derived organoid (PDO) model to detect the correlation between circRNA-CREIT expression and PDO sensitivity to doxorubicin. The results showed that PDOs with higher expression levels of circRNA-CREIT tended to be more sensitive to DOX and had a lower IC50 values (Fig. 2J–L and Additional file 1: Fig. S5I).

GO and GSEA analyses (Fig. 2A, and Additional file 1: Fig. S6A) also suggested circRNA-CREIT might play a role in cell proliferation, cell adhesion and epithelial–mesenchymal transition (EMT), and hence, we tested the function of circRNA-CREIT in the above biological processes. The results demonstrated that circRNA-CREIT overexpression could significantly inhibit cell proliferation, migration, invasion and EMT in TNBC cells, in contrast to the effect of circRNA-CREIT knockdown (Additional file 1: Fig. S6B-L).

CircRNA-CREIT directly interacted with double-stranded RNA-activated protein kinase (PKR)

To elucidate the molecular mechanisms by which circRNA-CREIT influences the chemosensitivity, we performed RNA pulldown assays, followed by electrophoresis, Coomassie blue staining and mass spectrometry (MS). Among the proteins identified by MS, PKR, which plays important roles in sensing and responding to multiple types of cellular stresses and cancer progression [26], aroused our interest and was identified as a candidate protein interacting with circRNA-CREIT (Fig. 3A). Then NPDock was used to predict the molecular docking between circRNA-CREIT and PKR protein, indicating the physical interaction between the two molecules (Fig. 3B). The binding between circRNA-CREIT and PKR was verified by independent RNA pulldown and immunoblot assays (Fig. 3C). To determine the specific region of circRNA-CREIT that mediated the association with PKR, the secondary structure of circRNA-CREIT with the minimum free energy (MFE) was predicted utilizing the online tool RNAfold (Fig. 3D, upper). Three truncated circRNA-CREIT probes representing three different stem-loops were constructed (Fig. 3D, lower). As shown in Fig. 3E, stem-loop #2 could efficiently pull down endogenous PKR protein in TNBC cells, whereas stem-loops #1 and #3 scarcely bound with PKR. FISH combined with immunofluorescent (IF) assays showed the colocalization of endogenous circRNA-CREIT and PKR in the cytoplasm, which further verified the direct interaction between circRNA-CREIT and PKR (Fig. 3F). Furthermore, we performed RNA immunoprecipitation (RIP) assays and observed marked enrichment of circRNA-CREIT using anti-PKR antibodies (Fig. 3G). To determine which domain of PKR was responsible for its association with circRNA-CREIT, we constructed truncated PKR mutants (Fig. 3H). RIP assays showed that deletion of the dsRNA-binding motif (dsRBM1-2) at the N-terminus significantly abolished the interaction between PKR and circRNA-CREIT (Fig. 3I).

Fig. 3

CircRNA-CREIT physically interacted with PKR. A CircRNA-CREIT probes and control probes were biotinylated and incubated with MDA-MB-231 cell lysates for RNA pull-down assays. (left) Photograph presenting Coomassie brilliant blue staining for the proteins precipitated in the RNA pull-down assays. The orange arrow indicates the size of the PKR protein. (right) Two segments of PKR proteins identified by mass spectrometry (MS). B Graphical representation of the molecular docking between circRNA-CREIT and the PKR protein using NPDock. C Western blotting of independent RNA pull-down assays verified the specific association of PKR protein with circRNA-CREIT using MDA-MB-231 cells. D The secondary structure of circRNA-CREIT was predicted by the online tool RNAfold web server (lower). (upper) Mountain plot representing the minimum free energy (MFE, red), the thermodynamic ensemble (green) and the centroid structures (blue) of circRNA-CREIT. CircRNA-CREIT was divided into three truncates representing three stem loop structures. E Western blotting analysis of PKR pulled down by different circRNA-CREIT truncates. F FISH and IF assays showing the colocalization of PKR and circRNA-CREIT. Scale bars = 10 μm. G RIP assay verifying the binding between PKR and circRNA-CREIT. H Diagrams of full-length (FL) PKR proteins and truncates with domain depletion. I (left) Western blotting analysis of PKR with full-length and truncated PKR proteins in the lysates of the HEK-293 T cells. (right) RIP assay for circRNA-CREIT enrichment in cells transfected with flag-tagged PKR (FL) overexpression vectors and truncated PKR expression vectors. Three independent experiments were conducted for each result. ns, no significance; *p < 0.05; ***p < 0.001 compared with the controls

CircRNA-CREIT enhanced the degradation of PKR via K48-linked polyubiquitination

We next investigated the impact of circRNA-CREIT on PKR expression. The results showed that the mRNA levels of PKR were not changed by circRNA-CREIT overexpression or knockdown (Additional file 1: Fig. S7A). However, elevated circRNA-CREIT expression led to significant downregulation of PKR protein, while circRNA-CREIT knockdown enhanced PKR expression (Fig. 4A). Further cycloheximide (CHX) treatment revealed that circRNA-CREIT knockdown increased the half-life of PKR protein, in contrast to the effect of circRNA-CREIT overexpression (Fig. 4B, C). Therefore, we hypothesized that circRNA-CREIT played an important role in PKR degradation.

Fig. 4

CircRNA-CREIT promoted PKR degradation via the ubiquitin–proteasome system. A Western blotting for PKR protein levels after circRNA-CREIT overexpression or knockdown. B, C TNBC cells with circRNA-CREIT overexpression or knockdown were treated with cycloheximide (CHX) for the indicated times. Western blotting analysis (upper) and statistical analysis (lower) of PKR levels upon CHX treatment are presented, with the level at 0 h as a control. D Western blotting analysis of PKR protein levels regulated by circRNA-CREIT with or without MG132 treatment. E Effects of circRNA-CREIT overexpression on the ubiquitination of PKR proteins. MDA-MB-231 cells were cotransfected with circRNA-CREIT overexpression plasmids and HA-tagged ubiquitin expression plasmids or the corresponding empty vectors. The cell lysates were incubated with anti-PKR or anti-IgG antibodies and protein A/G magnetic beads. The proteins precipitated in the co-IP assay were analyzed by Western blotting. IB immunoblot. F Western blotting assay showing the upregulation of K48-linked ubiquitination in circRNA-CREIT-overexpressing MDA-MB-231 cells. IB: immunoblot. HA-Ub-K48only: the cells were transfected with plasmids expressing HA-tagged ubiquitin with all lysines mutated except K48. HA-Ub-K63only: the cells were transfected with plasmids expressing HA-tagged ubiquitin with all lysines mutated except K63. G Identification of the candidate E3 ligases of PKR by bioinformatics prediction and MS analysis for RNA pull-down products, illustrated by a Venn diagram. H The interaction between PKR and HACE1 proteins was predicted by the ZDOCK server. The predicted structure was visualized by Discovery Studio software. I, J The interaction between circRNA-CREIT and HACE1 was verified by RNA pull-down assays (I) and RIP assay (J). Three independent experiments were conducted for each result. ns no significance; **p < 0.01 compared with the controls

In mammalian cells, the ubiquitin–proteasome system (UPS) participates in the degradation of most proteins [27]. To determine whether the UPS mediated the influence of circRNA-CREIT on PKR degradation, we treated cells with the proteasome inhibitor MG132. We found that MG132 abolished the effect of circRNA-CREIT overexpression on PKR protein levels (Fig. 4D). Then, co-IP assays were performed to detect the ubiquitination level of PKR proteins. We observed that the ubiquitination of endogenous PKR was markedly upregulated by circRNA-CREIT overexpression (Fig. 4E). K48- and K63-linked polyubiquitin chains are two main types of polyubiquitin linkage in mammalian cells [28]. K48-linked polyubiquitin usually induces degradation of substrate proteins and K63-linked polyubiquitin is correlated with protein stabilization or activation [29]. We observed that the K48-linked rather than K63-linked polyubiquitin of PKR was enhanced by circRNA-CREIT overexpression (Fig. 4F). These findings validated that circRNA-CREIT mediated PKR degradation through the K48-linked ubiquitin–proteasome pathway.

We next sought to determine which E3 ligases were involved in the regulatory effect of circRNA-CREIT on PKR degradation. The MS results revealed that there were several E3 ligases in the precipitation pulled down by circRNA-CREIT probes. Then, we predicted the potential E3 ligase of PKR by combining bioinformatics analysis on the Ubibrowser platform and the above MS results. The HECT domain and ankyrin repeat containing E3 ubiquitin protein ligase 1 (HACE1) was selected as the candidate (Fig. 4G). By applying molecular docking with ZDOCK (version 3.0.2) software, HACE1 was predicted to bind with PKR protein and the result was visualized by Discovery Studio (version 4.5) (Fig. 4H). In addition, circRNA-CREIT was verified to bind with HACE1, as indicated by RNA pulldown and RIP assays (Fig. 4I, J).

CircRNA-CREIT enhanced HACE1-PKR interaction by serving as a scaffold

Circular RNA can modulate the degradation process of downstream proteins by acting as a scaffold for that protein and its E3 ligase [30, 31]. According to the above results, we hypothesized that circRNA-CREIT might function in this way. As expected, HACE1 overexpression significantly inhibited PKR protein levels, and stable HACE1 knockdown increased PKR expression in TNBC cells (Fig. 5A). Additionally, HACE1 knockdown efficiently prolonged the half-life of PKR (Fig. 5B, C). Co-IP assays showed that cells with HACE1 overexpression exhibited increased PKR ubiquitylation, especially K48-linked ubiquitylation as expected (Fig. 5D, E). The above results provide essential evidence that HACE1 acts as an E3 ligase of PKR. Then, co-IP assays were performed to assess the interaction of PKR and HACE1. We observed that endogenous PKR could be immunoprecipitated by flag-tagged HACE1, and endogenous HACE1 could be immunoprecipitated by flag-tagged PKR (Fig. 5F). Moreover, exogenous co-IP assays demonstrated that PKR-flag could be easily detected in the anti-myc-HACE1 immunoprecipitates and vice versa (Fig. 5G, H). The interaction between PKR and HACE1 could be remarkably enhanced by circRNA-CREIT (Fig. 5I). Thus, circRNA-CREIT acted as a scaffold that could bring PKR and HACE1 together, leading to increased degradation of PKR.

Fig. 5

CircRNA-CREIT enhanced the binding of HACE1 and PKR proteins. A Western blotting for PKR protein levels after HACE1 overexpression or knockdown in TNBC cells. B, C TNBC cells stably transfected with sh-HACE1 plasmids or empty vectors were treated with CHX for the indicated times. Western blotting (B) and statistical analysis (C) of PKR protein levels are shown, with the level at 0 h as a control. D The impact of HACE1 on the PKR ubiquitination level was verified by co-IP assays and subsequent Western blotting. IB: immunoblot. E K48-linked ubiquitination of PKR was enhanced by HACE1 overexpression. IB: immunoblot. HA-Ub-K48only: the cells were transfected with plasmids expressing HA-tagged ubiquitin with all lysines mutated except K48. HA-Ub-K63only: the cells were transfected with plasmids expressing HA-tagged ubiquitin with all lysines mutated except K63. F HEK-293 T cells transfected with HACE1-Flag or PKR-Flag were lysed, immunoprecipitated with anti-Flag and then subjected to Western blotting assays using anti-PKR or anti-HACE1, respectively. G, H HEK-293 T cells co-transfected with PKR-Flag and HACE1-Myc or the corresponding empty vectors were lysed, immunoprecipitated with anti-MYC (G) or anti-Flag (H), and subjected to Western blotting analysis. I HEK-293 T cells were cotransfected with PKR-Flag, HACE1-Myc and circRNA-CREIT overexpression vectors or the corresponding empty plasmids. Co-IP assays validated that circRNA-CREIT increased the PKR-Flag level precipitated by HACE1-Myc. Three independent experiments were conducted for each result

CircRNA-CREIT attenuated DOX-induced stress granule formation and released RACK1 trapped by stress granules via the PKR/eIF2α axis

According to previous reports, PKR is an important kinase that phosphorylates eIF2α at serine residue 51 and triggers the formation of stress granules (SGs) [32]. We examined whether circRNA-CREIT played its roles via the modulation of PKR expression. We found that TNBC cells overexpressing circRNA-CREIT exhibited significantly lower IC50 values of DOX than control cells, and this effect could be abrogated by PKR overexpression (Fig. 6A). In addition, we noticed that circRNA-CREIT knockdown and PKR overexpression played a synergistic role in elevating DOX resistance in TNBC cells (Additional file 1: Fig. S8A). Colony formation assays also confirmed the above findings (Fig. 6B and Additional file 1: Fig. S8B), and we concluded that PKR mediated the effect of circRNA-CREIT on chemosensitivity in TNBC cells.

Fig. 6

CircRNA-CREIT attenuated SG formation via the PKR/eIF2α axis. A PKR reversed the effects of circRNA-CREIT in promoting chemosensitivity. Cell viability was detected MTT assays. The IC50 and the 95% CI of cells with different treatments are shown. B Colony formation assays presenting the roles of PKR in mediating the functions of circRNA-CREIT. C Immunofluorescence staining for the SG marker EIF3A under DOX treatment (for 24 h). Representative images are shown and the percentage of cells with SGs and the number of SGs per cell were calculated. D Western blotting assay showing that circRNA-CREIT inhibited p-eIF2α expression and circRNA-CREIT knockdown increased p-eIF2α levels in TNBC cells. E Immunofluorescence staining for p-eIF2α after circRNA-CREIT overexpression or knockdown with or without DOX treatment (for 24 h). Quantitative analyses were performed using Image J, and the groups treated with empty vectors and PBS were used as controls. F Subcellular localization of RACK1 and endogenous SG markers EIF3A and EIF4G1 under DOX treatment. Cells were transfected with RACK1-pmCherry-C1 plasmids, treated with DOX for 24 h and subjected to immunofluorescence. Colocalization analysis for RACK1 and SG markers along the indicated line was performed by ImageJ. G Western blotting analysis of the co-IP assay showing that the interaction between RACK1 and MTK1 was blocked by DOX treatment. H Co-IP assay and the subsequent Western blotting assay verified that circRNA-CREIT restored the binding of RACK1 and MTK1. Scale bars = 20 μm. Three independent experiments were conducted for each result. *p < 0.05; **p < 0.01, ***p < 0.001 compared with the controls

Considering the notable roles of SGs in cancer progression, especially in enhancing drug resistance of cancer cells [32], we hypothesized that circRNA-CREIT might exert its effect on TNBC chemosensitivity by regulating the formation of SGs. By visualizing the SG marker proteins EIF3A, G3BP1 and G3BP2, we found that DOX treatment could efficiently trigger the formation of SGs (Additional file 1: Fig. S8C). As expected, circRNA-CREIT overexpression attenuated SG formation under DOX treatment, and this effect was rescued by PKR overexpression (Fig. 6C). In contrast, circRNA-CREIT knockdown remarkably enhanced the formation of DOX-induced and hypoxia-induced SGs (Additional file 1: Fig. S9A, B). Next, we examined whether circRNA-CREIT had an effect on the phosphorylation level of eIF2α. The results showed that circRNA-CREIT overexpression significantly reduced the phosphorylation of eIF2α, whereas circRNA-CREIT knockdown promoted eIF2α phosphorylation in TNBC cells (Fig. 6D, E and Additional file 1: Fig. S9C, D). DOX treatment triggered the phosphorylation of eIF2α in a time-dependent manner, and this effect could be enhanced by circRNA-CREIT knockdown (Additional file 1: Fig. S9E). Furthermore, we found that DOX could induce higher expression levels of p-eIF2α and more stress granules in DMDA-MB-231/DOXR than in MDA-MB-231 cells (Additional file 1: Fig. S10A, B). Therefore, there was a close relationship between DOX resistance and the SG formation in TNBC cells.

RACK1, a scaffold protein with multiple functions, plays an important role in activating stress-induced apoptosis by interacting with MTK1 in the cytoplasm. However, under specific cellular stresses, such as arsenite treatment, RACK1 can be sequestered into SGs and the interaction between RACK1 and MTK1 is blocked [13]. We then examined the influence of circRNA-CREIT on the subcellular locations of RACK1 under DOX treatment. We observed that in unstimulated TNBC cells, RACK1 was mainly diffusely distributed in the cytoplasm (data not shown). After treatment with DOX, most RACK1 proteins were transferred to the nucleus, while some condensed into granules in the cytoplasm. Intriguingly, circRNA-CREIT knockdown remarkably enhanced the nucleation of RACK1 proteins in the cytoplasm, many of which showed colocalization with SG markers (Fig. 6F and Additional file 1: Fig. S10C). These results suggested that circRNA-CREIT knockdown significantly promoted the recruitment of RACK1 by DOX-induced SGs. As shown in Fig. 6G, the binding of RACK1 and MTK1 was scarcely detected after DOX treatment, and was dramatically rescued by circRNA-CREIT overexpression (Fig. 6H). These results indicated that circRNA-CREIT could enhance stress-induced apoptosis and improve the chemosensitivity of TNBC cells by preventing RACK1 from being sequestered by SGs.

The small molecule ISRIB targeting SGs greatly enhanced DOX sensitivity in synergy with circRNA-CREIT

It has been well established that the small molecule ISRIB can efficiently reverse the effects of eIF2α phosphorylation and trigger SG disassembly [19]. Our study showed that the combined utilization of the circRNA-CREIT overexpression vector and ISRIB could play a synergistic role and further improve the drug sensitivity of TNBC cells (Fig. 7A). On the other hand, ISRIB notably rescued the chemoresistance caused by circRNA-CREIT knockdown (Fig. 7B). The colony formation assays further corroborated these results (Fig. 7C, D). The synergistic function of circRNA-CREIT and ISRIB in reversing DOX resistance was also confirmed using DOX-resistant cells (Additional file 1: Fig. S11A, B).

Fig. 7

ISRIB exerted synergistic roles with circRNA-CREIT in improving chemosensitivity of TNBC cells. A Chemosensitivity to DOX of MDA-MB-231 cells treated with circRNA-CREIT, ISRIB alone or in combination. Cell viability was detected with MTT assays. The IC50 and the 95% CI of cells with different treatments are shown. B Chemosensitivity to doxorubicin of MDA-MB-231 cells treated with sh-circRNA-CREIT, ISRIB alone or in combination was detected by MTT assays. The IC50 and the 95% CI of cells with different treatments are shown. C, D Colony formation assays showed the synergistic roles of circRNA-CREIT and ISRIB in increasing cell chemosensitivity. E A schematic diagram indicating the experimental process of constructing the subcutaneous xenograft model and drug administration in female nude mice. F Images of xenograft tumors after the indicated treatment. Scale bars = 10 mm. G Growth curves and relative weights of the xenograft tumors in the four groups. The group treated with DMSO was used as a control. H IHC staining for Ki67 and cleaved caspase-3 expression in the xenograft tumors of different groups. Scale bars = 100 μm. Three independent experiments were conducted for each result. *p < 0.05; **p < 0.01, ***p < 0.001 compared with the controls

To examine the role of ISRIB in chemoresistance in vivo, we subcutaneously injected MDA-MB-231 cells into female nude mice (Fig. 7E). Treatment with DOX alone or ISRIB alone could inhibit the tumor growth in vivo, while the combination of DOX and ISRIB had a dramatic synergistic effect on attenuating tumor growth (Fig. 7F, G). IHC staining for Ki67 and cleaved caspase-3 also validated the above findings (Fig. 7H). The results suggest that ISRIB is a promising drug that could be used in combination with chemotherapy to treat TNBC.

Exosomal circRNA-CREIT endowed TNBC cells with improved chemosensitivity and a decreased proliferation rate

Recent studies demonstrated that circular RNAs could be packaged into exosomes and transmit drug resistance to other cells by exosomes [