Genome stability disrupted by the circRBM39(4,5,6)–RPA2 axis represses breast tumorigenesis

Significance

Circular RNAs (circRNAs) are natural outputs of transcription and RNA processing, which have been implicated in physiology and human diseases. We have identified circRBM39(4,5,6) as a tumor suppressor in breast cancer (BC), which directly interacts with replication protein A2 (RPA2) and facilitates its HERC2-mediated ubiquitination to disrupt the homologous recombination repair. Exogenous circRBM39(4,5,6) injection represses breast tumorigenesis and increases DNA damage, sensitizing BC cells to poly (ADP-ribose) polymerase inhibitors (PARPi) and rendering them susceptible to synthetic lethality. Our study expands the knowledge of circRNA-based synthetic lethality in human cancer.

Abstract

Genome instability (GI) is a hallmark of cancer. GI is associated with accumulative DNA damage and tumor-specific defects in DNA repair. Multiple antitumor drugs have been developed to promote persistent GI and to drive DNA damage beyond a threshold that tumor cells can survive. However, regulatory roles of circular RNAs (circRNAs) in GI and DNA damage remain elusive. Through circRNA profiling and a small interfering RNA (siRNA)-mediated screen, we have identified a circRNA termed circRBM39(4,5,6), which is significantly decreased in breast cancer (BC) and disrupts DNA damage response. CircRBM39(4,5,6) inhibits breast tumorigenesis through increasing DNA damage and promoting sensitivity to the antitumor drug poly Adenosine Diphosphate (ADP-ribose) polymerase inhibitor (PARPi). Mechanistically, cytoplasmic circRBM39(4,5,6) interacts with replication protein A2 (RPA2), an essential DNA repair factor, through a 7-nt AG-rich motif and destabilizes RPA2 protein by facilitating HECT and RCC1-like domain 2 (HERC2)-mediated ubiquitination, thus repressing homologous recombination repair. Importantly, administration of in vitro synthesized circRBM39(4,5,6) substantially inhibits breast tumorigenesis, sensitizes BC cells to PARPi, and renders them susceptible to synthetic lethality (SL) in distinct breast tumorigenesis mouse models. Our findings highlight the interplay between circRNA and GI in cancers, and the SL of circRBM39(4,5,6) to PARPi provides strategies for RNA-based BC therapeutics.

Genome instability (GI) is a hallmark of cancer and a major driving force for tumorigenesis (1). On the other aspect, tumor cells have evolved sophisticated DNA damage repair (DDR) pathways to counteract the GI-associated DNA damage, including homologous recombination (HR) (2–4). A series of antitumor therapies and drugs trigger persistent and elevated GI that cause DNA damage beyond tumor cell survival (3, 4). Understanding the molecular mechanisms affecting DNA repair activity is critically important for anticancer drug discovery or therapeutic approach exploration (4). An efficient system to monitor HR activity in living cells dynamically is the DR-GFP reporter (5). Briefly, this reporter comprises two mutated GFP genes, an inactivated version (SceGFP) and an internal HR template (iGFP). The upstream SceGFP contains a SceI endonuclease site engineered with in-frame stop codons, and the downstream iGFP is an internal GFP fragment that lacks its own start and stop signals. When the SceI enzyme introduces a DNA double-strand break (DSB) at the SceGFP site, HR machinery can repair the break using iGFP as a homologous template, restoring the functional GFP expression.

Breast cancer (BC) is the most common malignant tumor and the second leading cause of cancer-related death among women worldwide (6–9). Estrogen receptor (ER)-positive tumors represent the most common subtype (~70%), and triple-negative BC (TNBC) is the most aggressive form (~15%) (10). GI is prevalent in BC cells and frequently accompanied by dysfunctional DDR pathway, which influences the malignancy, metastatic potential, and therapeutic response (11–13). For example, mutations in BRCA1/2 genes, which encode proteins involved in DSB repair, are well-characterized genetic susceptibility factors in BC (14, 15). Additionally, mutations in other DDR-related genes such as ATM and CHEK2 are also associated with an increased risk of BC (16, 17). TNBC exhibits higher GI, which might partially explain its more aggressive behavior and poorer prognosis (18). GI also provides new aspects for BC treatment, with the principle of synthetic lethality (SL) being a widely adopted strategy to improve therapeutic outcomes (19, 20). SL occurs when simultaneous disruption of two genes or pathways results in cell death, whereas the perturbation of either component alone is insufficient to induce lethality (21). Poly (ADP-ribose) polymerase (PARP) inhibitors (PARPis), which impede the base excision repair pathway, cause SL in BRCA-mutated cancers, where HR repair is defective (22–24). The successful application of PARPi in BRCA-mutated BC patients exemplifies an SL-based targeted therapeutic strategy (24). Therefore, understanding molecular mechanisms underlying GI provides an important theoretical foundation and potential therapeutic targets for prognosis and treatment in BC.

Circular RNAs (circRNAs) are natural products of eukaryotic transcription and RNA processing, which are mainly generated by backsplicing (25–28). They are highly conserved and widely expressed with cell type– and tissue-specificity (29–32). Most circRNAs localize to cytoplasm in an exportin 2&4–dependent mechanism (33, 34). CircRNAs often function as miRNA sponges, transcriptional regulators, alternative splicing modulators, RNA-binding protein (RBP) interactors, and templates for translation (35–40). CircRNAs have emerged as crucial regulators in various biological processes and human diseases (41–45). For example, extensive nuclear circRNA accumulation caused by insufficient exportin 4 results in increased levels of circRNA–DNA hybrids (ciR-loop) and DNA damage, thereby contributing to neurological defects and male infertility (33). CircHIPK3, which is widely upregulated in diverse cancers, binds to the BRCA1 mRNA via RNA–RNA interaction to increase BRCA1 translation by competing for FMRP’s binding (44–45).

To investigate functional roles of circRNAs in GI, we integrated high-throughput RNA-sequencing (RNA-seq) of BC specimens and a low-throughput siRNA-mediated circRNA screen with the DR-GFP reporter. We have identified circRBM39(4,5,6), which is significantly downregulated in BC and increases DNA damage. CircRBM39(4,5,6) inhibits breast tumorigenesis by disrupting HR repair, and synergizes with PARPi. With a series of molecular, cellular, and biochemical assays, we further elucidated the underlying mechanism, where circRBM39(4,5,6) interacts with HR-related protein RPA2 in the cytoplasm and decreases its protein level through promoting its ubiquitination.

Results

High-Throughput RNA-seq and Low-Throughput Screen Identify circRBM39(4,5,6) to Disrupt Genome Stability in BC.

To explore critical roles of circRNAs in GI, we have integrated high-throughput RNA-seq of BC specimens and a low-throughput screen with siRNAs against circRNAs. First, we performed ribosomal RNA (rRNA)-depleted RNA-seq of three-paired BC and adjacent nontumor (NT) specimens to identify circRNA candidates. A total of 6,662 circRNAs with backsplicing junction (BSJ) ≥2 were detected, and differentially expressed (DE) analysis revealed 209 significantly dysregulated circRNAs (71 upregulated, 138 downregulated) in BC (Fig. 1A and Dataset S1). We chose top 10 upregulated and top 10 downregulated circRNAs for further siRNA-mediated screening (SI Appendix, Fig. S1 A and B and Dataset S2) (46). Second, we established a reporter system in MDA-231 cells, termed MDA-231-DR-GFP, in which a site-specific DSB generated by the SceI endonuclease underwent HR-mediated repair to produce a functional GFP protein (Fig. 1B). The MDA-231-DR-GFP stable cell line is monoclonal with one copy of the genome-integrated reporter vector, which ensures consistent HR efficiency (SI Appendix, Fig. S1C). For functional screening via the MDA-231-DR-GFP reporter, we designed two independent siRNAs targeting BSJs of each candidate circRNA. We performed fluorescence-activated cell sorting (FACS) to examine GFP levels in MDA-231 and MDA-231-DR-GFP cells following SceI expression upon individual knockdown (KD) of these 20 circRNAs (SI Appendix, Fig. S1D). We observed that circRBM39(4,5,6) KD resulted in the most significant increase in GFP production compared to other circRNAs (Fig. 1B). CircRBM39(4,5,6) (hsa_circ_0008817 in circBase) was also the most significantly downregulated candidate (P-value = 3.37E-4 from RNA-seq) in BC, which was chosen for further investigation (47).

Fig. 1.

Identification of circRBM39(4,5,6) to disrupt genome stability in BC. (A) Volcano plots showing differentially expressed circRNAs in three-paired BC specimens and adjacent nontumor (NT) tissues. Red and blue dots indicate significantly upregulated and downregulated circRNAs, respectively. (B) Schematic illustration of the MDA-231-DR-GFP reporter to detect HR (Left). siRNA-mediated circRNA screen to detect HR using MDA-231-DR-GFP cells (Right). Percentage of GFP-positive cells in nontransfected and SceI-transfected MDA-231-DR-GFP cells treated with siNC or indicated siRNAs targeting circRNA BSJs. MDA-231 cells with or without SceI transfection also used as controls. UnT, untransfection; siNC, siRNA with scrambled sequences; BSJ, backsplicing junction. iGFP, an internal GFP fragment lacking its own start and stop signals. (C) Northern blotting of circRBM39(4,5,6) with probes against exon sequence or BSJ region. RNAs were from MDA-231 and MCF-7 cells. rRNA bands indicate equal loading. (D) RNA smFISH of circRBM39(4,5,6) (green) in MDA-231 and MCF-7 cells. Nuclei (blue) were stained with DAPI. (Scale bar, 20 μm.) (E) Percentage of GFP-positive cells in indicated MDA-231-DR-GFP cells upon circRBM39(4,5,6) KD or OE. Two shRNAs against the circRBM39(4,5,6) BSJ (sh#1, sh#2) were used to KD circRBM39(4,5,6). The plasmid (circ OE) to overexpress circRBM39(4,5,6) contains its endogenous flanking sequences including inverted Alu pairs. EIB, exon–intron boundary; shRBM39, shRNA targeting RBM39 mRNA but not targeting circRBM39(4,5,6); circ Con, the control minigene for circRBM39(4,5,6) OE. (F) The γH2AX IF staining of indicated MDA-231 and MCF-7 cells after IR and the recovery for 1, 12, or 24 h. (Scale bar, 20 μm.) gRNA-NC, the gRNA control; gRNA-BSJ, gRNA antisense to the circRBM39(4,5,6) BSJ. (G) Quantification of γH2AX-positive MDA-231 and MCF-7 cells in indicated groups. γH2AX-positive cells were defined as ≥5 foci per nucleus, quantified by Image J. (H) Proteomics revealing differentially expressed proteins in control or circRBM39(4,5,6) OE MDA-231 cells. Red and blue dots indicate significantly increased and decreased proteins, respectively. (I) Gene Ontology analysis of 618 dysregulated proteins in MDA-231 cells upon circRBM39(4,5,6) OE. In B, E, and F, data are shown as mean ± SD from three independent biological experiments. In E and G, P-values are from two-tailed Student’s t test and one-way ANOVA test for the comparison of two and three groups, respectively.

The genomic structure shows that circRBM39(4,5,6) is composed of exons 4-6 from the RNA-binding motif protein 39 (RBM39) gene, and consists of 315 nucleotides (SI Appendix, Fig. S1E). PCR amplification and northern blotting confirmed the existence and full length of circRBM39(4,5,6) in MDA-231 and MCF-7 cells (Fig. 1C and SI Appendix, Fig. S1E). The estimated copy numbers of circRBM39(4,5,6) were 323, 118, 209, and 200 in MCF10A (normal breast epithelial cells), MDA-231, MCF-7, and MDA-453 cells, respectively (SI Appendix, Fig. S1F). Copy numbers of more than 100 are high for circRNAs or even for mRNAs. We further revealed the existence of circRBM39(4,5,6) in 4T1 cells, a murine TNBC cell line (SI Appendix, Fig. S1G). Circularized exons of circRBM39(4,5,6) are ~92.5% identity between humans and mice, demonstrating its high conservation in mammals (SI Appendix, Fig. S1H).

The function of noncoding RNA is closely associated with its subcellular localization (48–50). We found that circRBM39(4,5,6) preferentially localized to the cytoplasm analyzed by single-molecular fluorescence in situ hybridization (smFISH) (Fig. 1D). Next, we generated two independent short hairpin RNAs (shRNAs) (sh#1, sh#2) against the circRBM39(4,5,6) BSJ to KD this circRNA (Fig. 1E). We also constructed a minigene for circRBM39(4,5,6) overexpression (OE), which contains circularized exons and its endogenous flanking intronic sequences including inverted Alu pairs (Fig. 1E). To exclude the effect of linear RNAs, we constructed a mutated minigene as the control (circ Con), where 5’ backsplice site utilized for circRNA biogenesis was mutated at exon–intron boundary of circularized exons (51). The control minigene abolished circRBM39(4,5,6) formation but still generated the corresponding linear transcript (Fig. 1E). We observed no detectable GFP production in MDA-231-DR-GFP cells without SceI expression. CircRBM39(4,5,6) KD but not RBM39 KD increased GFP production in MDA-231-DR-GFP cells after SceI expression, whereas circRBM39(4,5,6) OE led to a significant decrease in GFP production (Fig. 1E and SI Appendix, Fig. S1I). These data indicate that circRBM39(4,5,6) disrupts DDR in BC cells.

CircRBM39(4,5,6) Increases DNA Damage in BC Cells.

To investigate functional roles of circRBM39(4,5,6) in DDR, we generated MDA-231, MCF-7, and MDA-453 cells with stable circRBM39(4,5,6) KD/OE, or RBM39 KD. CircRBM39(4,5,6) KD or OE had no significant effects on RBM39 mRNA levels in BC cells (SI Appendix, Fig. S2A). We then assessed the γH2AX foci, a surrogate marker for DNA damage, in these stable BC cells after the exposure to ionizing radiation (IR) (52). Without IR and at 24 h after radiation, circRBM39(4,5,6) KD but not RBM39 KD led to significant inhibition of γH2AX foci compared to control cells (Fig. 1 F and G and SI Appendix, Fig. S2 B and C). To exclude possible off-target effects of shRNA-mediated KD, we utilized the RfxCas13d system to KD circRBM39(4,5,6) in BC cells (53, 54). Similarly, RfxCas13d-BSJ-sgRNA-mediated circRBM39(4,5,6) KD reduced γH2AX signals in BC cells at 24 h after radiation (Fig. 1 F and G and SI Appendix, Fig. S2 B and C). Conversely, circRBM39(4,5,6) OE resulted in elevated γH2AX signals in BC cells at 24 h after radiation (SI Appendix, Fig. S2 D and E).

Under the treatment of camptothecin (CPT), a DNA topoisomerase I inhibitor to induce replication-associated DSBs (55), significant γH2AX foci were exacerbated in BC cells upon circRBM39(4,5,6) OE (SI Appendix, Fig. S2F). Meanwhile, CPT-induced RAD51 foci, a well-established marker for HR repair, were obviously abrogated upon circRBM39(4,5,6) OE (SI Appendix, Fig. S2F). CHK1 phosphorylation at Ser317 is an essential DDR signal timely responsive to DNA damage (56). The significantly decreased phosphorylated CHK1 level and the significantly increased γH2AX level were observed in BC cells upon circRBM39(4,5,6) OE after CPT treatment (SI Appendix, Fig. S2G).

Then, we performed proteomics with three independent replicates in the circRBM39(4,5,6)-stable OE and control MDA-231 cells (Fig. 1H). DE analysis identified 266 upregulated and 352 downregulated proteins upon circRBM39(4,5,6) OE (Fig. 1H and Dataset S3). These 618 proteins were subjected to GO term analysis using DAVID (57). Within top 10 significantly enriched GO terms were 4 DNA damage, 3 mitosis, 1 cell division, 1 mitochondria, and 1 chromosome segregation categories (Fig. 1I). Taken together, these results demonstrate that circRBM39(4,5,6) increases DNA damage in BC cells.

CircRBM39(4,5,6) Suppresses Breast Tumorigenesis via Increasing DNA Damage.

We then investigated the clinical significance of circRBM39(4,5,6) in 50-paired collected BC and NT specimens. The circRBM39(4,5,6) level was significantly decreased in BC compared to NT, and was negatively related to tumor size and advanced TNM stage (III-IV) (Fig. 2A and SI Appendix, Fig. S3 A and B). The mRNA level of its parental gene RBM39, a classic RBP that participates in both steroid hormone receptor-mediated transcription and alternative splicing (58), showed no significant difference between BC and NT specimens (SI Appendix, Fig. S3C). TCGA database analysis demonstrated that the RBM39 mRNA level had no significant correlations to overall survival of BC patients (SI Appendix, Fig. S3D). Collectively, circRBM39(4,5,6) but not RBM39 mRNA might function as a tumor suppressor in BC.

Fig. 2.

CircRBM39(4,5,6) suppresses breast tumorigenesis via increasing DNA damage. (A) Relationship between circRBM39(4,5,6) levels and clinicopathologic factors in BC patients. The connected lines indicated individual pairs of BC and NT specimens. (B) Clonogenic assays and quantification for MDA-231 and MCF-7 cells with two shRNAs (sh#1, sh#2) against the circRBM39(4,5,6) BSJ. shCtrl, shRNA with scrambled sequences. (C) 3D spheroid culture assays in BC cells upon circRBM39(4,5,6) KD. Spheroid diameters were quantified by Image J. (Scale bar, 100 μm.) (D) Transwell assays in BC cells upon circRBM39(4,5,6) KD. Migrated cells were quantified by Image J. (Scale bar, 100 μm.) (E) Cleaved PARP and γH2AX protein levels of BC cells upon circRBM39(4,5,6) OE with or without olaparib treatment. (F) Comet assays detecting DNA damage in control or circRBM39(4,5,6)-overexpressed BC cells with or without olaparib treatment. The percentage of DNA in tail was quantified (N = 50 cells) using the CASP software. (Scale bar, 20 μm.) (G) Images of xenograft tumors and quantification in BABL/c female nude mice (N = 6 per group) with subcutaneous injection of the circRBM39(4,5,6)-stable KD (sh#1) or wild-type MDA-231 cells. (H) Representative IHC staining and quantification of Ki-67, E-cadherin (E-cad), N-cadherin (N-cad), RAD51, and γH2AX proteins on subcutaneous tumors (N = 6 per group) upon circRBM39(4,5,6) KD (sh#1) or OE. Positive cells for Ki-67, RAD51, and γH2AX were defined with ≥5 foci per nucleus quantified by Image J. N-cad and E-cad signals were quantified by average optical density (AOD) with Image J. (Scale bar, 40 μm.) (I) Representative H&E staining and quantification of lung metastases in female nude mice (N = 3 per group) with intravenous injection of MDA-231 cells upon circRBM39(4,5,6) KD (sh#1) or OE. (Scale bar, 100 μm.) For B–F, data are shown as mean ± SD from three independent biological experiments. For B–D, P-values are from the one-way ANOVA test; for A and G–I, P-values are from two-tailed Student’s t test.

Next, we examined biological functions of circRBM39(4,5,6) at cellular levels. In MCF10A cells, a nontumorigenic immortalized human mammary epithelial cell line, clonogenic and Transwell assays revealed that circRBM39(4,5,6) KD significantly promoted cell proliferation and migration (SI Appendix, Fig. S3E). Furthermore, circRBM39(4,5,6) KD resulted in decreased levels of the epithelial marker E-cadherin (E-cad) and increased levels of mesenchymal markers N-cadherin (N-cad) and Vimentin (Vim) (SI Appendix, Fig. S3F), indicating that circRBM39(4,5,6) KD led to transformative phenotypes in MCF10A cell culture. CircRBM39(4,5,6) KD significantly promoted, while its OE inhibited cell proliferation of MDA-231, MCF-7, and MDA-453 cells (Fig. 2B and SI Appendix, Fig. S3 G–I). A three-dimensional (3D) spheroid model highly mimicking the tumor microenvironment was further subjected to evaluate cell growth (59). Gain- and loss-of-function revealed that circRBM39(4,5,6) significantly inhibited spheroid size and cell migration in BC cells (Fig. 2 C and D and SI Appendix, Fig. S3 J–M). Given that circRBM39(4,5,6) inhibits HR repair in BC, we determined its roles in PARPi sensitivity. With the treatment of olaparib, a well-recognized PARPi, we observed significant increases in cleaved PARP and γH2AX levels in BC cells upon circRBM39(4,5,6) OE (Fig. 2E and SI Appendix, Fig. S4A). Comet assays revealed that the combination of circRBM39(4,5,6) OE and olaparib induced substantially more DNA damage and inhibited cell growth more effectively than either treatment alone (Fig. 2F and SI Appendix, Fig. S4 B–D).

To investigate the tumor-suppressive role of circRBM39(4,5,6) in vivo, we established a xenograft mouse model by subcutaneous injection of MDA-231 cells, which were selected owing to the high aggressiveness and metastatic potentials (60, 61). CircRBM39(4,5,6) KD resulted in obviously larger tumors, while circRBM39(4,5,6) OE remarkably inhibited breast tumor growth, as reflected by smaller tumor size and decreased tumor weight (Fig. 2G). Immunohistochemistry (IHC) signals of the proliferation marker Ki-67, N-cad, and RAD51 were significantly increased upon circRBM39(4,5,6) KD, while their levels were remarkably decreased upon circRBM39(4,5,6) OE (Fig. 2H). Meanwhile, circRBM39(4,5,6) KD resulted in lower E-cad and γH2AX levels, while circRBM39(4,5,6) OE increased their levels (Fig. 2H). With a lung metastasis model using tail vein injection of MDA-231 cells, we found that MDA-231 cells with circRBM39(4,5,6)-stable KD formed more lung metastatic nodes than the controls, while circRBM39(4,5,6) OE resulted in fewer lung metastases (Fig. 2I). Together, we concluded that circRBM39(4,5,6) repressed breast tumorigenesis through increasing DNA damage, and sensitized BC cells to PARPi.

CircRBM39(4,5,6) Directly Interacts with Cytoplasmic RPA2.

Since cytoplasmic circRNAs always served as miRNA sponges, we wondered the potential of circRBM39(4,5,6) as miRNA sponges (44). RNA immunoprecipitation (RIP) of Ago2, which mediated the circRNA–miRNA interaction, revealed no or low detectable enrichment of circRBM39(4,5,6) in MDA-231 cells (SI Appendix, Fig. S4E). We further confirmed the low translational potential for circRBM39(4,5,6) with ribosome profiling in MDA-231 cells (SI Appendix, Fig. S4F). These results implied that circRBM39(4,5,6) might function with mechanisms beyond miRNA sponges and encoding polypeptides.

Then, we performed an RNA pull-down assay to identify the circRBM39(4,5,6)-interacting proteins (SI Appendix, Fig. S4G). We first designed various biotinylated antisense probes with different lengths that target circRBM39(4,5,6) BSJ to examine the probe specificity and efficiency. The 30-nt probe (Oligo1) spanning 15-nt upstream and 15-nt downstream BSJ specifically captured circRBM39(4,5,6) but not RBM39 mRNA in MDA-231 cells (SI Appendix, Fig. S4H). As the length increased, the amount of captured circRBM39(4,5,6) also increased, however, RBM39 mRNA was increasingly coenriched (SI Appendix, Fig. S4H). Considering the specificity, we selected the Oligo1 probe for further pulldown experiments. The purified circRBM39(4,5,6)-protein complexes were subjected to mass spectrometry (MS). We totally identified 15 and 25 circRBM39(4,5,6)-interacting proteins in MDA-231 and MCF-7 cells, respectively (Fig. 3A). 8 proteins (RPA2, hnRNPK, PTBP1, hnRNPD, hnRNPH1, hnRNPC, hnRNPAB, and G3BP2) are shared in MDA-231 and MCF-7 cells, and among them, RPA2 is the most prominent circRBM39(4,5,6)-interacting protein in both BC cells (Fig. 3A). The purified circRBM39(4,5,6)-protein complex was also subjected to SDS-PAGE, followed by silver staining (Fig. 3B). The most significant band ~30 kDa was excised and identified as RPA2 with MS (Fig. 3B). Then, we selected four overlapped circRBM39(4,5,6)-interacting proteins (RPA2, hnRNPK, PTBP1, hnRNPD), and confirmed their interactions with circRBM39(4,5,6) via RNA pulldown in MDA-231 and MCF-7 cells (Fig. 3C and SI Appendix, Fig. S4I). Although RPA2, hnRNPK, PTBP1, and hnRNPD mainly reside in the nucleus, a substantial portion of these four proteins localized to the cytoplasm in MDA-231 cells, which could potentially interact with cytoplasmic circRBM39(4,5,6) (SI Appendix, Fig. S5A). We performed the endogenous RIP assays for RPA2, hnRNPK, PTBP1, and hnRNPD using the cytoplasmic fraction, and confirmed the interactions between circRBM39(4,5,6) and these four proteins in the MDA-231 cytoplasm (Fig. 3D) (29, 62–64). The circRBM39(4,5,6)’s effects on protein levels of RPA2, hnRNPK, PTBP1, and hnRNPD were further examined. We observed that circRBM39(4,5,6) KD significantly increased the RPA2 protein level, but had no significant effect on three other proteins in MDA-231 cells, and circRBM39(4,5,6) OE significantly decreased the levels of RPA2 protein but not three other proteins (Fig. 3E). Given that RPA2 is the most abundant circRBM39(4,5,6)-interacting protein, and circRBM39(4,5,6) reduced the RPA2 protein level, the circRBM39(4,5,6)–RPA2 interaction was chosen for further investigation. RPA2 is an ssDNA binding protein and its loading on ssDNA promotes efficient DNA repair (65). The endogenous RPA2 RIP further confirmed the circRBM39(4,5,6)–RPA2 interaction in the cytoplasm and whole cell lysates (WCL), but not in the nucleus of BC cells (SI Appendix, Fig. S5 B–D). The RPA2 protein level was also markedly decreased in proteomics of MDA-231 cells upon circRBM39(4,5,6) OE (SI Appendix, Fig. S5E). CircRBM39(4,5,6) KD and OE significantly increased and decreased RPA2 protein levels in MCF-7 cells, respectively (SI Appendix, Fig. S5F). These findings indicate that circRBM39(4,5,6) interacts with RPA2 in the cytoplasm and is negatively correlated with the RPA2 protein level.

Fig. 3.

CircRBM39(4,5,6) directly interacts with RPA2 in the cytoplasm. (A) Heatmap of circRBM39(4,5,6)-interacting proteins by MS analysis of biotin-labeled circRBM39(4,5,6) pulldown in BC cells. (B) Potential circRBM39(4,5,6)-interacting proteins identified through SDS-PAGE gels followed by silver staining in indicated BC cells. The band indicated by the red arrow was subjected to MS and identified as RPA2. (C) RPA2 and indicated proteins were co-pulled down by circRBM39(4,5,6) pull-down in BC cells. Scr, biotin-labeled oligonucleotide with scrambled sequences; Oligo1, 30-nt biotin-labeled oligonucleotide antisense to the circRBM39(4,5,6) BSJ; RBM39-AS, biotin-labeled oligonucleotide against RBM39 mRNA but not targeting circRBM39(4,5,6). (D) The specific association of four indicated proteins with circRBM39(4,5,6) determined based on their RIP using MDA-231 cytoplasmic fraction. MecciND1, circFOXK2(2,3), circFCHO2(17-21), and circLARP1B(2-4) were positive controls for the RIP of RPA2, hnRNPK, PTBP1, and hnRNPD, respectively. RBM39 and ACTB mRNAs were negative controls for RT-PCR. (E) The RPA2 and three other protein levels in MDA-231 cells upon circRBM39(4,5,6) KD or OE. (F) HDOCK algorithm predicts the RPA2-binding sites within circRBM39(4,5,6). (G) The association of RPA2 with circRBM39(4,5,6) examined by RNA pulldown in MDA-231 cells transfected with biotin-labeled FL and truncated form (Δ55-61) of circRBM39(4,5,6). (H) The association of RPA2 with circRBM39(4,5,6) examined by RPA2 IP with the cytoplasmic fraction of MDA-231 cells transfected with biotin-labeled FL and truncated form (Δ55-61) of circRBM39(4,5,6). (I) Representative images of circRBM39(4,5,6) FISH and RPA2 IF in MDA-231 cells with overexpression of circRBM39(4,5,6) FL (circ OE) and truncated form (Δ55-61) (circ Mut) for 24 h. A permeabilization condition was used to reduce nuclear RPA2 signals, thereby highlighting cytoplasmic signals in MDA-231 cells. Boxed areas are enlarged. The colocalization between circRBM39(4,5,6) (R, red) and RPA2 (G, green) is shown (N = 30 randomly selected areas). (Scale bar, 1 μm.) circ Con, the control minigene for circRBM39(4,5,6) OE; G/R, proportion of green-red signal colocalization; R/G, proportion of red-green signal colocalization. P-values are from the one-way ANOVA test.

To map RPA2-binding sites in circRBM39(4,5,6), we predicted the RPA2-circRBM39(4,5,6) docking with the HDOCK algorithm (66). A 7-nt sequence (UAAGGAA) (No. 55-61 from BSJ) in circRBM39(4,5,6) was identified as the RPA2-binding sites (Fig. 3F and SI Appendix, Fig. S5G). Then, we generated in vitro synthesized biotin-labeled full-length (FL) and truncated form (Δ55-61) of circRBM39(4,5,6) with the permuted intron–exon (PIE) system (SI Appendix, Fig. S5H) (27). Biotin-labeled circRBM39(4,5,6) FL but not the Δ55-61 (Mut) form could pulldown the RPA2 protein (Fig. 3G). Further RIP assays revealed that RPA2 could co-pulldown the biotinylated circRBM39(4,5,6) FL rather than the Δ55-61 form (Fig. 3H). We also generated biotin-labeled circRBM39(4,5,6) mutant with 55-61 sequences scrambled (SCR55-61) and found that this mutant failed to interact with RPA2 in MDA-231 cells (SI Appendix, Fig. S5I). Furthermore, cytoplasmic RPA2 RIP-seq in MDA-231 cells revealed that circRBM39(4,5,6) but not RBM39 mRNA was highly enriched in the RPA2 RIP sample compared to the input (SI Appendix, Fig. S5J). To explore why RBM39 mRNA did not interact with RPA2 despite sharing the same sequence, we analyzed available ribosome profiling sequencing (Ribo-seq) datasets from BC cells and specimens. The 5’ portion of RBM39 mRNA exhibited extensive ribosome occupancy (SI Appendix, Fig. S5J), suggesting that RBM39 mRNA was actively engaged in translation and that the putative RPA2-binding sites in the 5’ portion of this mRNA were therefore inaccessible for RPA2 binding. In addition to circRBM39(4,5,6), 70 transcripts (8 circRNAs and 62 linear transcripts) were also identified as cytoplasmic RPA2 targets (Dataset S4). Among these 71 targets, 27 (5 circRNAs and 22 linear transcripts) with “UAAGGAA” motif accounted for 38.03% of the total transcripts bound, and 44 targets lacked the motif (SI Appendix, Fig. S5K). Immunofluorescence (IF) analysis revealed that most RPA2 localized in the nucleus, and ~14.7% of RPA2 localized to the cytoplasm in MDA-231 cells, making visualization of cytoplasmic RPA2 signals challenging (SI Appendix, Fig. S5L). By using conditions that reduce the nuclear permeability to lower nuclear signals and thus to highlight cytoplasmic RPA2 signals (29), more than 50% circRBM39(4,5,6) signals overlapped with RPA2 protein signals were observed in the MDA-231 cytoplasm (Fig. 3I). Furthermore, OE of circRBM39(4,5,6) but not the Δ55-61 form (circ Mut) enhanced the circRBM39(4,5,6)–RPA2 colocalization in the MDA-231 cytoplasm (Fig. 3I). We also estimated the copy numbers of total RPA2 proteins, and found ~2,648 and ~2,717 copies in MDA-231 and MCF-7 cells, respectively (SI Appendix, Fig. S5M). Considering that ~14.7% of RPA2 was cytoplasmically localized, we estimated that there were ~389 molecules of RPA2 protein in the cytoplasm of MDA-231 cells. The molecular ratio of circRBM39(4,5,6) to cytoplasmic RPA2 is about 118:389 (~1:3.3) (SI Appendix, Figs. S1F and S5 L and M), which would be sufficient for the circRNA to exert its regulation on RPA2 protein. Collectively, we demonstrated that circRBM39(4,5,6) interacted with a substantial portion of cytoplasmic RPA2 protein via a 7-nt AG-rich motif.

CircRBM39(4,5,6) Promotes RPA2 Ubiquitination Mediated by HERC2.

Next, we wondered how circRBM39(4,5,6) reduced the RPA2 protein level, and found that circRBM39(4,5,6) had no significant effect on the levels, stability, and translation of RPA2 mRNA in MDA-231 cells (SI Appendix, Fig. S6 A–C). These results suggested that circRBM39(4,5,6) repressed the RPA2 protein level in a posttranslational manner. As ubiquitination is one of the most common posttranslational modifications and is tightly associated with protein stability (67), we examined whether circRBM39(4,5,6) affected RPA2 ubiquitination. The RPA2 protein half-life was dramatically prolonged upon circRBM39(4,5,6) KD in BC cells (SI Appendix, Fig. S6D). OE of circRBM39(4,5,6) but not the Δ55-61 form significantly shortened the RPA2 protein half-life in BC cells (Fig. 4A). We further found that circRBM39(4,5,6) KD significantly decreased RPA2 polyubiquitination levels in BC cells (SI Appendix, Fig. S6E). OE of circRBM39(4,5,6) but not the Δ55-61 form significantly increased RPA2 polyubiquitination levels in BC cells (Fig. 4B).

Fig. 4.

CircRBM39(4,5,6) promotes HERC2-mediated RPA2 ubiquitination. (A) The RPA2 protein levels of MDA-231 and MCF-7 cells treated with 100 μg/mL cycloheximide (CHX) for indicated times upon circRBM39(4,5,6) FL or the Δ55-61 (circ Mut) form OE. circ Con, the control minigene for circRBM39(4,5,6) OE; P-values, from two-way ANOVA test. (B) RPA2 ubiquitination in MDA-231 and MCF-7 cells treated with 25 μM MG-132 for 6 h upon circRBM39(4,5,6) FL or the Δ55-61 form OE. (C) The association of HERC2, RFWD3, and PRP19 with circRBM39(4,5,6) from biotin-labeled RNA pulldown in MDA-231 and MCF-7 cells. (D) The RPA2–HERC2 interaction examined by RPA2 co-IP in BC cells using cytoplasmic fraction upon OE of FL or the Δ55-61 form of circRBM39(4,5,6). (E) The RPA2 protein levels in MDA-231 and MCF-7 cells upon circRBM39(4,5,6) FL or the Δ55-61 form (circ Mut) OE with or without RPA2 OE. EV, empty vector. (F and G) Clonogenic (F) and Transwell (G) assays in indicated groups of MDA-231 and MCF-7 cells. (Scale bar, 100 μm.) (H) Under 2 μM CPT treatment for 1 h, γH2AX IF staining of indicated MDA-231 and MCF-7 cells. Positive cells were defined with ≥5 foci per nucleus quantified by Image J. (Scale bar, 20 μm.) For A, B, and F–H, data are shown as mean ± SD from three independent biological experiments; for B, F, and G, P-values are from the one-way ANOVA test.

Then, we asked how circRBM39(4,5,6) promoted RPA2 ubiquitination. HERC2, RFWD3, and PRP19 are three previously reported E3 ubiquitin ligases that specifically trigger RPA2 degradation (68–70). Among them, only HERC2 interacted with circRBM39(4,5,6) in the RNA pull-down assays of BC cells (Fig. 4C). Moreover, RPA2 KD reduced the interaction between circRBM39(4,5,6) and HERC2 in BC cells (SI Appendix, Fig. S6F). OE of circRBM39(4,5,6) but not the Δ55-61 form markedly enhanced the HERC2–RPA2 interaction in BC cell cytoplasm (Fig. 4D). Collectively, our results suggested that circRBM39(4,5,6) promoted RPA2 polyubiquitination mediated by HERC2.

RPA2-Binding and RPA2 Are Essential for the Antitumor Effects of CircRBM39(4,5,6).

Then, we performed rescue experiments in BC cells with stable KD of circRBM39(4,5,6). γH2AX staining revealed that OE of circRBM39(4,5,6) but not the Δ55-61 form (circ Mut) reversed the shortened DDR process caused by circRBM39(4,5,6) KD in BC cells (SI Appendix, Fig. S7 A–C). OE of circRBM39(4,5,6) but not the Δ55-61 form also reversed the promotive effects on cell growth and migration caused by circRBM39(4,5,6) KD in BC cells (SI Appendix, Fig. S7 D–G). All these results indicated that the RPA2-binding site of circRBM39(4,5,6) was essential for its tumor-suppressive roles in BC cells.

To validate functional roles of the circRBM39(4,5,6)–RPA2 axis in BC, we knocked down RPA2 in BC cells with or without circRBM39(4,5,6) KD. RPA2 KD significantly suppressed colony formation and cell migration in BC cells, which was consistent with the known tumor-promotive effects of RPA2 (SI Appendix, Fig. S8 A–C) (71). RPA2 KD also reversed the promotive roles in BC cell growth and migration caused by circRBM39(4,5,6) KD (SI Appendix, Fig. S8 B and C). OE of circRBM39(4,5,6) but not the Δ55-61 form decreased RPA2 protein levels, and inhibited BC cell growth and migration (Fig. 4 E–G). RPA2 OE reversed the suppressive effects caused by OE of circRBM39(4,5,6) but not the Δ55-61 form (Fig. 4 F and G and SI Appendix, Fig. S8D). RPA2 OE rescued the promotive effects of CPT-induced γH2AX signals caused by OE of circRBM39(4,5,6) rather than the Δ55-61 form in BC cells (Fig. 4H). These findings demonstrated that RPA2-binding and RPA2 were essential for the antitumor effects of circRBM39(4,5,6) in BC cells.

CircRBM39(4,5,6) with PARPi Synergistically Inhibits Breast Tumorigenesis In Vivo.

To examine the in vivo antitumor effects of circRBM39(4,5,6) and its combination with PARPi, we established a subcutaneous tumorigenesis model using wild-type MDA-231 cells in female nude mice (Fig. 5A). We first synthesized circRBM39(4,5,6) or the control circRNA (circgfp contains 315 nucleotides of GFP sequence) in vitro. Currently, there is no standard dosage in the field of noncoding RNA or circRNA-mediated therapeutic studies. Based on the dosage of RNA vaccines in mouse models, such as 5 µg gDE7 mRNA-LNP HPV vaccine and 10 µg GPC3-based circRNA vaccine against hepatocellular carcinoma (72, 73), we selected a median dose of 7 μg circRBM39(4,5,6) or circgfp, which were encapsulated by liposome and were administrated to mice by the intratumoral injection (Fig. 5A). More than twofold increases in circRBM39(4,5,6) levels were observed in tumors with circRBM39(4,5,6) administration compared with circgfp injection (Fig. 5B). Neither circRBM39(4,5,6) injection or olaparib treatment alone, nor their combination induced whole body weight loss (SI Appendix, Fig. S8E). The tumor size and weight were significantly decreased in mice treated with circRBM39(4,5,6) compared to the control circgfp (Fig. 5C). The combination of circRBM39(4,5,6) with olaparib further enhanced the reduction in tumor size and weight, indicating that circRBM39(4,5,6) could partially promote therapeutic efficacy of PARPi in breast tumorigenesis (Fig. 5C). CircRBM39(4,5,6) administration alone resulted in significantly decreased Ki-67, N-cad, and RPA2 IHC signals, and increased E-cad and γH2AX IHC signals (Fig. 5D). The combination of circRBM39(4,5,6) and olaparib exhibited a greater decrease in Ki-67, N-cad, and RPA2 signals, and more increase in E-cad and γH2AX signals compared to circRBM39(4,5,6) injection alone (Fig. 5D).

Fig. 5.

Administration of circRBM39(4,5,6) plays synergistic antitumor effects with PARPi in breast tumorigenesis models. (A) Schematic illustration for the subcutaneous model with circRNA administration and olaparib as single-agents or in combination. s.c., subcutaneous injection; i.p., intraperitoneal injection; i.t., intratumoral injection. (B) RT-qPCR revealing the circRBM39(4,5,6) levels in tumor tissues (N = 6 per group) with circRBM39(4,5,6) administration and olaparib (Ola) as single-agents or in combination. circgfp, the control circRNA containing 315-nt GFP sequence. (C) Xenograft tumor image and weight in breast tumorigenesis mice (N = 6 per group) with circRBM39(4,5,6) and olaparib as single-agents or in combination. (D) Representative IHC staining of Ki-67, N-cad, E-cad, RPA2, and γH2AX in subcutaneous mice (N = 6 per group) with circRBM39(4,5,6) and olaparib as single-agents or in combination. (Scale bar, 40 μm.) (E) Schematic illustration for the establishment of BC PDX model with circRNA administration and olaparib as single-agents or in combination. s.c., subcutaneous injection; i.p., intraperitoneal injection; i.t., intratumoral injection. (F) PDX tumor image and weight in PDX mice (N = 5 per group) with circRBM39(4,5,6) administration and olaparib as single-agents or in combination. (G) Representative IHC staining of Ki-67, N-cad, E-cad, RPA2, and γH2AX proteins in PDX mice (N = 5 per group) with circRBM39(4,5,6) administration and olaparib as single-agents or in combination. (Scale bar, 40 μm.) For D and G, positive cells for Ki-67, RPA2, and γH2AX were defined with ≥5 foci per nucleus quantified by Image J. N-cad and E-cad signals were quantified by AOD with Image J. For B–D, F, and G, P-values are from the one-way ANOVA test.

We also evaluated the therapeutic effects of circRBM39(4,5,6) or PARPi as single-agents or combination therapy using a TNBC patient-derived xenograft (PDX) model, which more faithfully preserves the characteristics and heterogeneity of primary tumors (Fig. 5E and SI Appendix, Fig. S8 F–H) (74, 75). In vitro synthesized circRBM39(4,5,6) or circgfp encapsulated by lipid nanoparticles (LNPs) was administrated to mice by the intratumoral injection, and more than fourfold increases in circRBM39(4,5,6) levels were observed in PDX tumors with circRBM39(4,5,6) injection (SI Appendix, Fig. S8I) (76). CircRBM39(4,5,6) administration significantly decreased PDX tumor size and weight compared to the control circgfp, and the decrease was further enhanced in combination with olaparib (Fig. 5F). CircRBM39(4,5,6) and olaparib in combination led to more reduced Ki-67, N-cad, and RPA2 levels compared to either circRBM39(4,5,6) or olaparib alone (Fig. 5G). CircRBM39(4,5,6) administration also induced E-cad and γH2AX levels in PDX tumors, which were further increased by PARPi combination (Fig. 5G). Collectively, these results demonstrated that circRBM39(4,5,6) administration with PARPi synergistically inhibited breast tumorigenesis in vivo.

The Tumor-Suppressive Effects of CircRBM39(4,5,6) in Multiple Cancer Cells.

We wondered whether circRBM39(4,5,6) exerted tumor-suppressive effects across distinct cancer types. We transiently transfected circRBM39(4,5,6) OE or the control constructs into representative cells of five cancer types, including colon cancer (HT29), gastric cancer (AGS), ovarian cancer (A2780), cervical cancer (SiHa), and osteosarcoma (U2OS) (SI Appendix, Fig. S8J). These cancer cell growth and migration were significantly inhibited upon circRBM39(4,5,6) OE (SI Appendix, Fig. S9 A–C). Additionally, circRBM39(4,5,6) OE significantly decreased RPA2 protein levels and increased γH2AX levels (SI Appendix, Fig. S9D). These results indicated that circRBM39(4,5,6) exerted important antitumor roles in various cancer types.

Discussion

We have identified circRBM39(4,5,6) as a tumor-suppressor to inhibit breast tumorigenesis through disrupting DDR (SI Appendix, Fig. S9E). CircRBM39(4,5,6) interacts with cytoplasmic RPA2 and promotes its HERC2-mediated ubiquitination, thus compromising HR repair. Importantly, exogenous circRBM39(4,5,6) suppresses breast tumorigenesis and sensitizes BC cells to PARPi. The identification of a circRNA that disrupts DDR and promotes PARPi sensitivity provides insights into therapeutic targets for BC.

Although several circRNAs exert their functions coupled with their cognate genes (35, 77), circRBM39(4,5,6) disrupts genome stability to suppress breast tumorigenesis independently of its parental gene. RBM39 has been reported to interact with the MLL1 complex to facilitate BC cell transcription and proliferation (78). Circular and linear transcripts derived from the single gene locus exhibit opposite effects on cancer proliferation, highlighting an interesting area for further exploration. Precise regulation also indicates the balance in levels of circRNA and the linear counterpart.

We have demonstrated that circRBM39(4,5,6) interacts with the nonclassical RBP RPA2 in the cytoplasm (Fig. 3). Although RPA2 is a well-known ssDNA-binding protein, it is also reported to bind to some RNAs, including mecciND1 (62). With multiple lines of evidence, we have found that circRBM39(4,5,6) binds to cytoplasmic RPA2 via a 7-nt AG-rich motif. Although many RNA molecules in cells possess the 7-nt motif, they do not bind to RPA2. One possibility is that their secondary and/or tertiary structures prevent the motif from being exposed, or these RNA molecules are shielded by ribosomes or RBPs, just like RBM39 mRNA (SI Appendix, Fig. S5J), which makes the motif unavailable for RPA2 binding. Some RNAs and RPA2 protein may have distinct temporal and spatial expression patterns or cellular localization, which also hinders RPA2 binding to these RNAs. For those 44 RPA2 targets without the 7-nt motif (SI Appendix, Fig. S5K), other functional motifs or certain structures may be responsible for their interactions with RPA2, and this requires further exploration. We have quantified the molecular ratio of circRBM39(4,5,6) to cytoplasmic RPA2 protein as ~1:3.3, which is sufficient for the circRBM39(4,5,6)–RPA2 interaction in the cytoplasm to reduce considerable RPA2 protein levels. There is a modest but significant changes (~10%) of the circRBM39(4,5,6)–RPA2 colocalization between circRBM39(4,5,6) OE and Mut groups (Fig. 3I). However, under the same conditions, more than ~twofold changes in RPA2 polyubiquitination levels are observed in BC cells between the two groups (Fig. 4B). Also, biochemical evaluations revealed that RPA2 could not bind to circRBM39(4,5,6) when the 7-nt motif was missing (Fig. 4 G and H). The underlying mechanisms for this difference between circRBM39(4,5,6)–RPA2 colocalization and regulatory efficacy require more investigations. In addition to RPA2, circRBM39(4,5,6) interacts with various RBPs in BC cells (Fig. 3). While circRBM39(4,5,6) does not affect their protein levels, it is possible that interactions between circRBM39(4,5,6) and these RBPs are essential for its function. Whether circRBM39(4,5,6) binds to these RBPs to regulate their downstream targets requires further investigations.

RNA-based therapeutics have shown their potential in treating human cancers (79). Herein, we employ delivery amounts of circRNA exceeding physiological levels to establish a proof-of-concept for circRBM39(4,5,6) as an exogenous therapeutic molecule. With synthesized circRBM39(4,5,6) to treat breast tumors by the intratumoral injection, we have indeed found obvious antitumor effects of this circRNA in different BC models. Although intratumoral injection demonstrates its safety and potency, it still has several limitations, including its limited accessibility (80). The circRNA dose and delivery approach require further optimization. Additionally, small-molecule inhibitors targeting DDR, such as PARPi, hold promise to effectively treat multiple cancers including BC (23, 81). We have demonstrated that circRBM39(4,5,6) and PARPi synergistically inhibit breast tumorigenesis in vivo (Fig. 5), providing the molecular basis of the combinatorial strategy of circRNA and PARPi in biomedical research. The synergistic effects of circRBM39(4,5,6) and PARPi in inhibiting breast tumorigenesis raises the possibility that circRBM39(4,5,6) might predict the therapeutic efficacy of PARPi-based targeted therapy in BC treatment. Recently, the oncogenic circHIPK3 was reported to interact with BRCA1 mRNA via RNA–RNA interaction, thereby increasing BRCA1 translation. Depletion of circHIPK3 promotes DNA damage and sensitizes cancer cells to DNA-damage-inducing agents (49). By contrast, our data demonstrate that circRBM39(4,5,6) suppresses HR repair in a posttranslation manner.

In conclusion, we have identified circRBM39(4,5,6) as a tumor suppressor in BC and other cancer types. It disrupts genome stability by inhibiting RPA2-mediated HR repair, sensitizes BC cells to PARPi, and renders them susceptible to SL. To our knowledge, circRBM39(4,5,6) is the only reported tumor suppressor circRNA to be synthetically lethal to PARPi, providing strategies for BC therapy.

Materials and Methods

Clinical Samples Preparation.

All fresh tumors and adjacent nontumor specimens were collected from BC patients in the First Affiliated Hospital of Jiaxing University, which was approved by the Human Research Ethics Committee of First Affiliated Hospital of Jiaxing, Jiaxing University (2024-KY-197). Written informed consent was obtained from each patient or their families, as appropriate, for this study. All clinical samples were rinsed with PBS after operation, and then were cut into small pieces for storage at −80 °C until the need for analysis. All primers and oligonucleotides were listed in Dataset S2 and more details of this section are included in SI Appendix.

Statistical Analysis.

As indicated in figure legends, the statistical significance was assessed by Student’s t test when two sets of data were compared, or one-way ANOVA with a Tukey’s post hoc test when more than two sets of data were compared. For SI Appendix, Fig. S3D, log-rank test was used; for SI Appendix, Fig. S5K, the chi-square test was used; for Fig. 4A and SI Appendix, Fig. S6 B–D, two-way ANOVA followed by two-tailed Dunnett’s multiple-comparisons test was used. For Student’s t tests, the values used in the graphs represent averages of three independent experiments, with error bars showing SD. After ANOVA with F-tests, the statistical significance and P-values were evaluated with Student’s t tests. P-value below 0.05 was considered as a significant difference.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Data, Materials, and Software Availability

RNA-sequencing data have been deposited in the GEO database under accession number GSE289776 (82). All plasmids have been sequenced for confirmation and further information about these plasmids is available upon request. All other data are included in the manuscript and/or supporting information.