SUMO1 modification of KHSRP regulates tumorigenesis by preventing the TL-G-Rich miRNA biogenesis

To detect whether KHSRP can be SUMOylated in cells, we co-transfected His-SUMO1, Flag-Ubc9 and EBG-SENP1 with (Fig. 1a) or without (Fig. 1b) HA-KHSRP plasmid into 293T cells, and performed the SUMOylation assay by using the method of Ni²⁺-NTA resin pull-down [28]. The results showed that both exogenous and endogenous KHSRP were SUMOylated even with only His-SUMO1. The SUMO1 modification of KHSRP was strongly enhanced by the E2 conjugating enzyme Ubc9, whereas it was attenuated with the addition of the de-SUMOylation enzyme SENP1. Next, we preformed an in vitro E.coli-based SUMOylation assay [29] and showed that GST-KHSRP was SUMOylated in the E.coli transformed with pE1E2SUMO1 and GST-KHSRP but not in the E.coli transformed with GST or GST-KHSRP alone (Fig. 1c). More importantly, we further proved that SUMO1 modification of KHSRP occurred naturally in 293T cells by the method of immunoprecipitation (IP). Endogenous SUMO1-KHSRP was detected only in the immunoprecipitated complexes with anti-KHSRP antibody but not with normal IgG (Additional file 1: Fig. S1). Collectively, these results demonstrate that KHSRP can be modified by SUMO1 in cells and in vitro.

According to the prediction of SUMOplot™ Analysis Program, KHSRP has several putative SUMOylated sites (Additional file 2: Fig. S2). To determine which lysines (Ks) of KHSRP are the major sites for SUMOylation, we mutated those sites with R (arginine) replacing K, respectively. Wild type (WT) or mutant HA-KHSRP construct with His-SUMO1 were co-transfected into 293T cells for the Ni²⁺-NTA precipitation SUMOylation assays. The results showed that the mutation K87R clearly reduced the SUMOylation level of KHSRP compared with HA-KHSRP-WT and other point mutations including -K359R, −K628R (Fig. 1d), −K244R, −K251R, −K435R, −K473R and -K494R (Additional file 3: Fig. S3). However SUMO1 modification was not removed completely in KHSRP K87R mutant, we speculated there might exist other sites apart from the eight putative sites predicted by SUMOplot™ Analysis Program. Taken together, these results suggest that K87 is a major SUMO1 modification site of KHSRP.

Since hypoxia can regulate the SUMOylation level of distinct SUMO targets [26, 30], we investigated whether oxidative stress can regulate the SUMOylation level of KHSRP. Indeed, we found that the SUMO1 modification level of exogenous KHSRP was significantly enhanced while those of KHSRP-K87R was less increased by hypoxia (1% O₂) treatment for 6 and 12 h (Fig. 2a). In contrast to hypoxia, the treatment with 100 μM of hydrogen peroxide (H₂O₂) strongly decreased the SUMO1 modification level of KHSRP at 1.5 and 3 h whereas slightly affected that of KHSRP-K87R (Fig. 2b).

We also found that EGF slightly downregulated SUMO1 modification of KHSRP (Fig. 2c, lane 4), which is consistent with a rapid alteration in SUMOylation of target proteins under EGF stimulation [31, 32]. Since activated AKT by growth signals can phosphorylate KHSRP at serine 193 (S193) [33], we wanted to test whether this phosphorylation influences SUMO1 modification of KHSRP. Interestingly, the SUMO1 modification level of KHSRP was attenuated when AKT1 phosphorylation was induced by the treatment with insulin (Fig. 2d, lane 3). On the contrary, SUMO1 modification of KHSRP was significantly increased by inhibition of AKT1 phosphorylation via the treatment with LY294002 (Fig. 2d, lane 4). Moreover, the phospho-site mutant KHSRP-S193A significantly increased the SUMO1 modification level of KHSRP (Fig. 2e, lanes 4, 5), whereas the phosphomimetic mutant KHSRP-S193D almost abolished its SUMOylation (Fig. 2E, lanes 6, 7), as expectedly. Thus, the above results suggest that SUMO1 modification of KHSRP can be regulated by external signaling pathways, especially PI3K/AKT1 signal pathway.

Hypoxia is linked to poor patient outcomes and has a negative impact on the effectiveness of tumor treatment, such as surgery and radiotherapy [34]. The signaling pathway induced by EGF or insulin are also associated with tumor cell proliferation and migration [35, 36]. KHSRP knockdown in glioblastoma multiforme (GBM) cells promotes migration and causes multifocal tumor in a mouse model. 70% of GBM patients (n = 548) with high expression levels of KHSRP survive long after surgery [37]. KHSRP knockdown in osteosarcoma cell U2OS significantly upregulates cell proliferation [7]. All these suggest that KHSRP behaves as a tumor suppressor. Since our above data have proven that KHSRP SUMOylation could be regulated by hypoxia and growth factors, we questioned whether K87-SUMOylation of KHSRP is involved in tumorigenesis and cancer progression. To this end, we constructed stable cell lines, in which endogenous KHSRP was firstly stably knockdown by a short hairpin RNA targeting KHSRP 3’UTR (shKHSRP) in the lentiviral vector pLKO.1 system, and then HA-KHSRP-WT and HA-KHSRP-K87R were re-introduced by the lentiviral-expressing system. The expression levels of endogenous KHSRP and re-expressed HA-KHSRP-WT or HA-KHSRP-K87R were assessed by Western blotting (Additional file 4: Fig. S4). To explore whether SUMOylation of KHSRP affects the transforming potential of each stable DU145 cell lines, we performed a soft-agar colony-forming assay. We found that knockdown of KHSRP enhanced the capacity of anchorage-independent growth, whereas re-expression of the HA-KHSRP-WT suppressed soft-agar colony formation in DU145 cells as expectedly (Fig. 3a). But interestingly, re-expression of the SUMO-site mutant HA-KHSRP-K87R more significantly inhibited the anchorage-independent growth compared with the HA-KHSRP-WT (Fig. 3a). These results indicate that SUMOylation inhibits the tumor-suppressive role of KHSRP.

In order to assess whether SUMOylation of KHSRP influences the migration capacity of tumor cells, we performed a RTCA-migration assay [38]. The results showed that DU145-shKHSRP cells displayed a sharper curve in cell index and a higher slope value of migration compared to that of DU145-shRNA-Ctrl cells, which indicated that KHSRP suppressed tumor cell migration. Re-expression of KHSRP-WT or KHSRP-K87R mutant in DU145-shKHSRP cells both decreased the migratory capability and the latter more powerfully (Fig. 3b). Furthermore, we used the method of 3D growth cell cultures on extracellular matrix to mimic the in vivo conditions to investigate the invasive ability of tumor cells [39]. DU145-shKHSRP cells grew diffusely and displayed a scattered morphology, showing the better capability to invade into extracellular matrix than that of DU145-shRNA control cells, which revealed that KHSRP knockdown increased the invasive ability. On the contrary, ectopic re-expression of KHSRP-WT or KHSRP-K87R in DU145-shKHSRP cells decreased cell penetrating into the matrix; in particular, cells re-expressing KHSRP-K87R grew into tighter and round colonies (Fig. 3c), which suggested the disruption of KHSRP SUMOylation potentially decreased the invasive ability of tumor cells. Aggressive cancer cells forming de novo vascular networks is defined as tumor cell vasculogenic mimicry (VM), which is considered to be an indicator of the aggressive, metastatic phenotype [40]. Therefore, we also assessed whether KHSRP SUMOylation influences VM formation with above four stable DU145 cell lines. The VM assays showed that knockdown of KHSRP in DU145 cells promoted formation of vasculogenic networks on the matrigel compared to that of DU145-shRNA control cells, while re-expression of either KHSRP-WT or KHSRP-K87R greatly attenuated the formation of pipe-like structures on the matrigel and the latter almost abolished the VM formation (Fig. 3d). Moreover, to investigate whether KHSRP SUMOylation affects xenograft tumor growth in vivo, DU145 stable cell lines were inoculated subcutaneously into the backs of nude mice. Tumor growth measurement was performed at day 15, 21, 27, and 32 after injection, showing that tumors of the DU145-shKHSRP group grew most quickly, whereas tumors of the DU145-shKHSRP re-expressing KHSRP-K87R group grew more slowly than other groups (Additional file 5: Fig. S5). Tumors were weighed after killing the nude mice at 5 weeks after injection (Fig. 3e), showing the similar pattern of results as in above soft-agar colony formation, RTCA-migration, 3D culture growth and VM formation. Moreover, SUMO1 modification of KHSRP was confirmed in the xenograft tumors from the KHSRP-WT group but not in the KHSRP-K87R group (Fig. 3f).

To pursue the probability of an correlation between KHSRP SUMOylation and prostate cancer, we further extracted data from the database of “The Cancer Genome Atlas Research Network” (TCGA). Gleason score is a risk stratification factor and prognostic factor, which is positively related to clinical recurrence and metastasis of prostate cancer [41, 42]. The data of KHSRP expression levels together with the associated clinical data revealed that the KHSRP expression levels for 435 prostate cancer patients (Additional file 6: Table S1) was not related to Gleason score (Pearson correlation with r value = 0.056; P = 0.242). But among 212 prostate cancer patients with high expression levels of KHSRP (median value >8732 FPKM, see in Table S1), the expression levels of Ubc9, which is only E2 conjugating enzyme for SUMOylation (Fig. 1a-b), was positively correlated with Gleason score (Pearson correlation with r value = 0.147^*; P = 0.033). We also detected the SUMO1 modification of KHSRP in tumors and paracancerous tissues of gastric and colorectal cancer, respectively. The results demonstrated that KHSRP was expressed in tumors higher than that in paracancerous tissues, and SUMO1 modification of KHSRP could be detected only in tumors (Additional file 7: Fig. S6).

Taken together, these data demonstrate that SUMOylation of KHSRP promotes tumorigenesis and cancer progression.

Since KHSRP is a component of the Drosha-DGCR8 multiprotein complex [7], we speculated that KHSRP SUMOylation might affect the interaction between KHSRP with the Drosha-DGCR8 complex. 293T cells co-transfected HA-Drosha and Flag-tagged KHSRP-WT or KHSRP-K87R were lysed for immunoprecipitation with anti-Flag antibody and then immunoblotted with anti-HA antibody. The result showed that the Drosha band pulled-down by the SUMO-site mutant KHSRP-K87R was stronger than that by the KHSRP-WT (Fig. 4a). Similarly, we found that the interaction of DGCR8 with the mutant KHSRP-K87R was also stronger than that of DGCR8 with KHSRP-WT (Fig. 4b). Since KHSRP knockdown can abrogate the interaction between Drosha and some special pri-miRNAs, which harbor short G-rich stretches in the terminal loop, thus influencing the biogenesis of those miRNAs [7], and SUMOylation of KHSRP potentially affected its interaction with the Drosha-DGCR8 complex, we raised a question whether KHSRP SUMOylation regulates the biogenesis of TL-G-Rich miRNAs. Therefore, next we performed the high-throughput deep sequencing for above DU145 stable cell lines. In DU145-shRNA-Control and DU145-shRNA-KHSRP cells there were 368 miRNAs, in which the expression level of each miRNA >10 reads per million (RPM) (Additional file 8: Table S2). We found that a total of 151 miRNAs were downregulated in DU145 shKHSRP cell lines compared to those in DU145-shRNA-Control cell lines (Fig. 4c; Additional file 9: Table S3). The biogenesis of these miRNAs was considered to be regulated by KHSRP. Consistently, let-7 family and other miRNAs as KHSRP-dependent miRNAs whose precursor terminal loops harbor short G-rich stretches [7, 8], were included in the above set of miRNAs. In particular, of 151 KHSRP-dependent miRNAs, there were 51 miRNAs (including let-7 family let-7i, let-7e, let-7g and miR-98 etc.) that were upregulated in DU145-shRNA-KHSRP cell lines stably re-expressing KHSRP-K87R compared to those of re-expressing KHSRP-WT (Fig. 4c and d; Additional file 10: Table S4), indicating that the mutant KHSRP-K87R promotes the production of a subset of miRNAs. To validate the sequencing results, some miRNAs including let-7i-5p, miR-98-5p, miR-182-5p and miR-183-5p were chosen for validation by using the quantitative RT-PCR. Indeed, the relative expression levels of all these miRNAs were higher in KHSRP-K87R cells compared with cells expressing KHSRP-WT (Fig. 4e). Thus, above results indicate that SUMOylation of KHSRP may reduce its interaction with the pri-miRNA/Drosha-DGCR8 complex, thereby decreasing the biogenesis of TL-G-Rich miRNAs.

Since SUMOylation regulates the translocation of many proteins between nucleus and cytoplasm [25, 43, 44], we questioned whether SUMOylation of KHSRP regulates its subcellular localization. The amino acid (aa) sequences from 109 to 122 ‘KRQLEDGDQPESKK’ of KHSRP has been identified as a bipartite NLS (nuclear localization sequence) [11]. However GFP-KH1–4 even with the deletion of NLS still can localize in both the nucleus and the cytoplasm [11, 33], so it is still not very clear about the mechanisms in regulation of the nuclear localization of KHSRP. We performed the nuclear/cytosolic fractionation assays and showed that the deletion of NLS increased the localization of KHSRP in the cytoplasm compared with KHSRP-WT (Fig. 5a, lanes 1, 3), which mostly localized in the nucleus (lanes 2, 4), whereas the SUMO-site mutant KHSRP-K87R almost completely localized in the nucleus and barely in cytoplasm (lanes 5, 6). The cytoplasmic fraction percentage of KHSRPΔNLS, −WT and -K87R was ~22%, ~10% and 0.5%, respectively (Fig. 5a, right panel). The mutant HA-KHSRP-K87R even co-transfected with GFP-SUMO1 still existed only in the nucleus; however in this case of HA-KHSRP-WT co-transfected with GFP-SUMO1, the strong translocation of KHSRP from the nucleus to the cytoplasm in about 39% cells was observed (Fig. 5b and Additional file 11: Fig. S7). These results suggest that SUMO1 modification of KHSRP may control its translocation between nucleus and cytoplasm.

To further confirm SUMOylation of KHSRP affecting its translocation, we generated a mimic SUMOylated KHSRP construct Flag-SUMO1-KHSRPΔN, which the N-terminal (1–67 aa) of KHSRP was replaced by Flag-tagged SUMO1(2–96 aa), to simulate the high SUMO1 modification status of KHSRP. HeLa cells transfected with Flag-KHSRPΔN or Flag-SUMO1-KHSRPΔN (Additional file 12: Fig. S8) were lysed for the nuclear/cytosolic fractionation assay, showing that the cytoplasmic fraction of Flag-SUMO1-KHSRPΔN had higher percentage by ~45% whereas Flag-KHSRPΔN had lower percentage by ~10% (Fig. 5c). Being consistent with this, the distribution in both the cytoplasm and the nucleus of SUMO1-KHSRPΔN was obviously observed in 31% cells whereas that of KHSRPΔN was observed in 5.8% cells. As expectedly, the distribution in both cytoplasma and nucleus of KHSRPΔNLS was in 85% cells (Fig. 5d). Since hypoxia and LY294002 could promote KHSRP SUMOylation, we wanted to detect whether the cytoplamsic/nuclear localization of KHSRP is changed under hypoxia environment and LY294002 stimulation. Consistent with above results, we found that both hypoxia (Additional file 13: Fig. S9) and LY294002 (Additional file 14: Fig. S10) promoted cytoplasmic localization of endogenous KHSRP. These results supported that SUMOylation of KHSRP promotes its cytoplasmic localization by probably facilitating its nuclear export.

Next we attempted to validate the above hypothesis that SUMOylation of KHSRP influences the nuclear localization of KHSRP, we constructed SENP1- (a deSUMOylation enzyme) and Ubc9- (a SUMOylation E2) knockdown HeLa stable cell lines (Additional file 15: Fig. S11) to detect the translocation change of endogenous KHSRP under the status of high-SUMOylation (shSENP1) or low-SUMOylation (shUbc9). The nuclear/cytosol fractionation assays displayed that the fraction of KHSRP in cytoplasm was much higher in HeLa-shSENP1 cells whereas not easily detected in either HeLa-shCtrl or HeLa-shUbc9 cells (Fig. 5e). The cytoplasmic fraction of endogenous KHSRP in HeLa-shSENP1 had higher percentage by ~18.8% whereas that in HeLa-shCtrl and HeLa-shUBC9 had lower percentage by ~10.4% and ~3.8%, respectively (Fig. 5e, right panel). The immunofluorescent staining results also showed that there was a small percentage of KHSRP existed in the cytoplasm in addition to most of KHSRP in the nucleus in HeLa-shSENP1 cells, while KHSRP seemed to be observed almost in the nucleus in HeLa shCtrl or HeLa-shUbc9 cells (Fig. 5f). The distribution in both the cytoplasm and the nucleus of KHSRP was observed in 8% HeLa-shSENP1 cells while that was observed in about 2% HeLa-shRNA Ctrl and shUbc9 cells (Fig. 5f, bottom panel). These data revealed that high SUMOylation of KHSRP promotes its nuclear export to the cytoplasm. Taken together, above results demonstrate that SUMOylation of KHSRP increases its cytoplasmic localization by promoting its nuclear export.

Since the KHSRP activity in miRNA biogenesis is promoted by the kinase ATM mediated- phosphorylation through the enhanced interaction of KHSRP and TL-G-Rich pri-miRNAs [8], we questioned whether SUMOylation of KHSRP influences the interaction between KHSRP and pri-miRNAs. To this end, we firstly predicted the secondary structure of pri-miRNAs with the RNAstructure V5.3 software, which is based on the minimum free energy principle according to the RNA primary sequence [45]. Most of these pri-miRNAs, whose mature miRNAs were upregulated by re-expression of KHSRP-K87R compared to those of re-expression of KHSRP-WT in DU145-shKHSRP stable cell lines (Fig. 4d; Additional file 10: Table S4), harbored short G-rich stretches in their terminal loops. For instances, as like pri-let-7a-1 [9, 10], the secondary structures of pri-miRNAs including pri-let-7a-3, pri-let-7g, pri-let-7i, pri-let-7e, pri-miR-98 and pri-miR-182 contained G-rich stretches (Fig. 6a).

Next, we performed the RIP (RNA-immunoprecipitation) assays [20, 26] to confirm the effect of SUMOylation on the interaction between KHSRP and pri-miRNAs. 293T cells transfected with pri-let-7a-1 and Flag-tagged KHSRPΔN, or SUMO1-KHSRPΔN were lysed for RIP (Fig. 6b, right panel) and followed by qRT-PCR analysis, showing that the interaction between pri-let-7a-1 and SUMO1-KHSRPΔN was decreased and consequently the mature let-7a biogenesis was downregulated compared to that of KHSRPΔN (Fig. 6b). Consistently with this, by using the same RIP assay we also found that the binding of pri-let-7a-1 or pri-let-7a-3 with KHSRP-K87R was significantly higher compared to that with KHSRP-WT, and accordingly the biogenesis of mature let-7a was much more in KHSRP-K87R than in KHSRP-WT (Fig. 6c-d). Above results demonstrate that SUMOylation of KHSRP inhibits its interactions with TL-G-Rich miRNAs precursors, consequently reducing the biogenesis of those miRNAs.

Human embryonic kidney 293T, 293FT, HeLa and prostate cancer DU145 cell lines were cultured in Dulbecco’s modified Eagle’s medium (DMEM, obtained from Hyclone) containing 10% fetal calf serum (Biowest, Kansas, MO, USA), 1% penicillin and streptomycin (Invitrogen). All cell lines were cultured at 37 °C in a 5% CO₂ humidified incubator. Cell transfection was performed using Lipofectamine 2000 (Invitrogen).

Antibody against KHSRP (#A302–021) was from Bethyl Laboratories, Inc. (Montgomery, UK). Tubulin-Alpha mouse Mcab (#66031–1-Ig), His-Tag mouse antibody (#66005–1-Ig), GST mouse antibody (#66001–1-Ig) were from Proteintech™. GAPDH (#ab37168), SUMO1 (Y299, #ab32058) antibodies were from Abcam. Mouse antibodies against Flag (M2, #F1804), HA (16B12, #MMS-101P) were from Sigma. Rabbit antibodies against HA (Y-11, #sc805), Ubc9 (H-81, #sc-10,759) and mouse antibody against Lamin B1 (8D1, #sc-56,144) were from Santa Cruz Biotech. Rabbit antibody against SUMO1 (C9H1, #4940S) was from Cell Signal Technology. Alexa Fluor® 568 (Rabbit, #A11011), 488 (Rabbit, #A11008), 488 (Mouse, #A11001) were from Invitrogen.

Protein G Plus/Protein A agarose suspension (#IP05) were purchased from Calbiochem. Puromycin (#P8833), hydrogen peroxide (H₂O₂) solution (#H1009), insulin (#I-5500), LY294002 (#L9908) and EGF (#E9644) were from Sigma.

The pEGFP-6xHis-FLKHSRP plasmid was from Addgene [11]. The FLKHSRP (full-length KHSRP) was amplified by PCR, and subcloned into the pEF-5HA vector with EcoRI and XbaI sites, the pCMV-Tag2B vector with EcoRI and XhoI sites, the pGEX-4T-1 vector with EcoRI and XhoI sites, respectively. Point mutations (K87R, K359R, K628R, S193A, S193D, K244R, K251R, K435R, K473R and K494R) and nuclear localization sequence (NLS) deletion mutant of KHSRP (KHSRPΔNLS) were carried out by using KOD-plus-mutagenesis Kit (TOYOBO) according to the manual. The pEF-5HA-KHSRPΔN (N-terminal aa 1–67 was deleted) was amplified from Flag-KHSRPΔN which was kindly provided by Michael G. Rosenfeld’s Lab [7] by using KOD plus (TOYOBO) and subcloned into the pEF-5HA vector with EcoRI and XbaI sites. The KHSRP cDNA was amplified from pCMV-Tag2B-KHSRP with the primer containing KOZAK (GCCACC) and HA tag sequence, and then subcloned into the lentiviral vector CD513B (System Biosciences) carrying EGFP and puromycin genes. The shRNA anti-KHSRP were obtained from Sigma and sub-cloned into pLKO.1 vector.

Flag-SUMO1-KHSRPΔN plasmid was constructed by two steps. Firstly, KHSRPΔN was amplified from pEF-5HA-KHSRPΔN by using KOD plus (TOYOBO) with 10% DMSO, and then subcloned into the pCMV-Tag2B vector with SalI and XhoI sites. Secondly, SUMO1 (amino acids 2–96) was amplified from His-SUMO1 and inserted into of pCMV-Tag2B-KHSRPΔN with HindIII and SalI sites. All above plasmids were verified by sequencing. Primers used for constructions are shown in Additional file 16: Table S5.

(A) SUMOylation of KHSRP was analyzed in 293T cells by using the method of Ni²⁺-NTA beads with His-tagged SUMO1, as described previously [29, 50, 51]. (B) In vitro SUMOylation assay in E.coli system with pE1E2SUMO1 was performed as previously described [20, 29]. Briefly, pGEX-4T-1-KHSRP-WT was co-expressed with or without pE1E2SUMO1 plasmid in E.coli BL21 (DE3) respectively, and then lysed by using B-PER Protein Extraction Reagent (#78248, Thermo Fisher, USA) and incubated with Glutathione sepharose 4B (GE healthcare) at 4 °C overnight. The beads bound proteins were washed for three times with lysis buffer, and subjected to Western-bot for analysis of SUMO1-modified GST-KHSRP. (C). The method of immunoprecipitation (IP) was also used to detect the SUMO1 modification of endogenous KHSRP. Briefly, 293T cells were lysed in NEM-RIPA buffer (50 mM Tris–HCl pH 7.4, 150 mM NaCl, 1% NP-40, 20 mM N-ethylmaleimide and one complete protease inhibitor cocktail). Cell lysates (1 mg) were used for immunoprecipitation. To detect the endogenous SUMO1-KHSRP in 293T cells, 5 μl of KHSRP antibody or normal IgG (as a control) was used for immunoprecipitation. Tissues were lysed in NEM-RIPA with 0.1% SDS and 5 mM EDTA, which was according to our previous study [28] and SUMO1 antibody was used for immunoprecipitation.

Nuclear and cytosolic fractions were extracted by Nuclear/Cytosol fractionation kit according to the manual (Nuclear/Cytosol Fractionation Kit, Catalog #K266, BioVision, BioVision Incorporated, Milpitas, CA 95035 USA). 3 × 10⁶ of HeLa cells were harvested. One-tenth of these cells were harvested by SDS lysis as the Input for the protein expression by Western-blot, and nine-tenth of cells were extracted by Nuclear/Cytosol fractionation kit. One-fifth of both nuclear and cytoplasmic fraction was used for Western-blotting with indicated antibody.

Immunofluorescent staining was performed as described previously [28, 50]. Briefly, HeLa cells were seeded into the Poly-Lysine coated slides overnight, and then transfected with plasmids expressing Flag-KHSRPΔN, Flag-SUMO1-KHSRPΔN or Flag-KHSRPΔNLS, respectively. After 48 h, cells were fixed with 4% paraformaldehyde for 30 min and then permeabilized with 0.2% Triton X-100 for 1 h, and then incubated in the primary antibody anti-Flag (M2) (dilution 1:800) at 4 °C overnight. Cells were washed three times with PBS and then incubated in the secondary antibody (Alexa 488 anti mouse, dilution 1:500) in blocking solution for 2 h. The cells were then washed five times with PBS. DAPI (4′,6-diamidino-2-phenylindole) was added to visualize the nucleus. Images were taken with Nikon microscope.

In another experiment, HeLa cells were transfected with HA-KHSRP-WT or HA-KHSRP-K87R with GFP-SUMO1. Cells were incubated with the primary antibody anti-HA (Rabbit, dilution 1:250), with the secondary antibody Alexa 568 anti rabbit (dilution 1:500). DAPI staining was to visualize the nucleus. The images of DAPI and anti-HA were merged. The green signals indicating the expression of GFP-SUMO1 were directly observed and taken by Nikon microscope.

To detect the endogenous KHSRP in HeLa cells, KHSRP antibody (Rabbit, dilution 1:500) was used as primary antibody and Alexa 488 anti-rabbit antibody (dilution 1:500) was used as secondary antibody.

293T cells transfected with HA-Drosha and Flag-KHSRP-WT or -K87R, or cells transfected with Flag-DGCR8 and HA-KHSRP-WT or -K87R respectively were lysed in RIPA buffer (50 mM Tris-HCl pH = 7.4, 150 mM NaCl, 1% NP40, 10% glycerol and a complete protease inhibitor cocktail (Roche)). 1 mg of total extracted proteins were incubated with 25 μl of protein A/G agarose and 2 μg of anti-Flag antibody at 4 °C overnight. Then the beads were washed with RIPA buffer for three times, and followed by Western blot analysis.

The method of high-throughput sequencing was previously described [52]. Briefly, total RNAs from DU145 stable cell lines were extracted by Trizol (Life Technologies, Carlsbad, CA). The RNAseq library of miRNA was prepared using NEBNext® Multiplex Small RNA Library Prep Set for Illumina (NEB, Beverly, MA). Modified RNAs were reversely transcribed and then PCR amplified with specific primers corresponding to the adapters. The amplified products were resolved in 6% PAGE and the bands corresponding to ~140 bp were isolated. The libraries were quality controlled with a Bioanalyzer 2100 (Agilent, Santa Clara, CA) and sequenced by Nextseq 500 (Illumina, San Diego, CA) on a 75 bp single-end run.

The raw sequencing reads from the small RNA libraries were mapped to the human genome (hg19) using the mapper.pl script included in miRDeep2 program with the following parameters: -l,-k,-e,-h,-j,-m. This clipped the adapter sequence from each read while keeping only reads no shorter than 18 nt. Specifically, the mapper.pl package takes Bowtie2 as the mapping engine and generated a collapsed set of non-redundant reads along with the mapped genomic locations. Then the quantifier.pl script was used to calculate the expression level of known microRNAs from the annotation of miRBase (Release 21). Additionally, the normalized abundance (RPM, Reads Per Million) for each known microRNAs was calculated as following:

Where Ri represents the count of reads mapped to the genomic region of a given microRNA i. And Nj represents the total count of reads in the small RNA library of sample j.

The RNA immunoprecipitation assay (RIP) was performed as previously described [20, 26]. Briefly, 48 h after transfection with the indicated plasmids, one-tenth of these cells in 10-cm plate cultured were reserved as the Input for qRT-PCR analysis, and nine-tenth of cells were lysed in RIP buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.4, 1% NP40, 1 mM DTT, 100 units/ml RNase inhibitor (Fermentas), 400 μM VRC (New England BioLabs), Protease inhibitor cocktail (Roche)). After incubated on ice for 1 h, 1/50 of lysates were used for Western blot to examine the expression of KHSRP, the others were incubated with 40 μl of protein A/G agarose and 4 μg of anti-Flag antibody at 4 °C overnight. After immunoprecipitation, the beads were washed with RIP-lysis buffer for three times and then 1/10 of the beads were immunoblotted for the efficiency analysis of immunoprecipitation. The remaining beads were used to extract RNA by Trizol reagent (Invitrogen).

Quantitative real-time PCR was performed with SYBR® Green PCR Master Mix (#4309155, Applied Biosystems, USA) to analyze the fold changes of pri-miRNAs immunoprecipitated by KHSRP and mature miRNAs biogenesis. GAPDH and U6 levels were used for normalization of pri-miRNAs and mature miRNAs PCR and the pri-miRNAs immunoprecipitated by KHSRP was normalized by Input of pri-miRNAs.

The soft-agar colony forming assay was performed as described previously [20, 26, 50]. Briefly, this assay was performed in 6-well plates with a base of 2 ml of medium containing 5% FBS with 0.6% Bacto agar (Amresco). Cells were seeded in 2 ml of medium containing 5% FBS with 0.35% agar at 2 × 10³ (for DU145 cells) cells/well and layered onto the base. 2 ml of DMEM with 10% FBS was covered on the top of agar gel. The photographs of colonies growing in the plates were taken at day 21. The number of colonies was scored by photoshop CS5.

The procedure was carried out as previously described [20, 26]. Briefly, 4 × 10⁴ of each of serum-starved DU145 stable cells were suspended in 100 μl of serum-free medium, and then cell suspension was added into the pre-equilibrated upper chamber of CIM-plate. The lower chamber was filled with 160 μl of complete DMEM containing 10% FBS. The kinetic cell index of migration was recorded every 15 min for 24 h and then calculated by RTCA software v1.2 (Roche Applied Science).

The vasculogenic mimicry experiment of DU145 stable cells were carried out using μ-Slide Angiogenesis Kit (IBIDI) according to the procedure as previously described [39, 53]. DU145 cells at the density of 5 × 10³ in 50 μl of growth medium were plated in each well pre-coated with matrix™ (Millipore). Pictures were taken with Nikon microscope 20 h later.

The 3D cell culture assay was performed as described before [39]. Generally, 5 μl of 3D matrix™ (Millipore) and 5 μl of cell solution (2 × 10³ cells) were mixed and added into the inner well of μ-Slides (IBIDI) and covered with complete cell culture medium. The μ-slides were incubated at 37 °C and 5% CO₂ cell incubator for 7 days and pictures were taken with Nikon microscope.

The experiment of xenograft tumor model was conducted as previously described [20, 50]. Stable DU145 cell lines (2.5 × 10⁶) were injected subcutaneously into 5-week-old male BALB/c nude mice (n = 5) individually. 15 days after injection, the tumors were measured every 6 days. All mice were sacrificed at 35 days and the tumors were dissected, photographed and weighed. All animal studies were conducted with the approval and guidance of Shanghai Jiao Tong University Medical Animal Ethics Committees.

Experiments were performed at least three times, and representative results were shown. All data are presented as means ± s.e.m. for qPCR, RTCA migration, mouse xenograft model and soft agar colony forming assay. Statistical analysis was calculated with Microsoft Excel analysis tools. Differences between individual groups are analyzed using the t-test (two-tailed and unpaired) with triplicate or quadruplicate sets. A value of P < 0.05 was considered statistically significant and P-value <0.05 was marked with (*), < 0.01 with (**) or <0.001 with (***). TCGA data was analyzed by Pearson Correlation method by the program SPSS 22.