A VEGF-A/SOX2/SRSF2 network controls VEGFR1 pre-mRNA alternative splicing in lung carcinoma cells

The splice variant sVEGFR1-i13 is a truncated version of the cell membrane-spanning VEGFR1 receptor that is devoid of its transmembrane and tyrosine kinase domains. We recently showed the contribution of sVEGFR1-i13 to the progression and the response of squamous lung carcinoma to anti-angiogenic therapies. In this study, we identify VEGF165, a splice variant of VEGF-A, as a regulator of sVEGFR1-i13 expression in these tumors, and further show that VEGF165 cooperates with the transcription factor SOX2 and the splicing factor SRSF2 to control sVEGFR1-i13 expression. We also demonstrate that anti-angiogenic therapies up-regulate sVEGFR1-i13 protein level in squamous lung carcinoma cells by a mechanism involving the VEGF165/SOX2/SRSF2 network. Collectively, our results identify for the first time a signaling network that controls VEGFR1 pre-mRNA alternative splicing in cancer cells.

(e) Western-blot analyses of VEGF 165 and sVEGFR1-i13 in MGH7 or H2170 cells as indicated. Actin was used as a loading control. Numbers represent the quantification of VEGF 165 or sVEGFR1-i13 signal intensities relative to actin signal using Image J software. The value 1 was arbitrarily assigned to the pcDNA3 condition signal. All western blot experiments were performed at least three times. Illustrations of a representative result are presented for each condition. (f) Mean levels ± SD of VEGF 165 immunohistochemical scores according to sVEGFR1-i13 status in squamous cell lung carcinoma, where SQLC are sub-divided in two classes representing tumors with high or low levels of sVEGFR1-i13 compared to normal lung tissues 4 . Statistical analyses were performed using a non parametric Mann-Whitney test (*p < 0.05; **p < 0.01; ***p < 0.001).
Scientific RepoRts | (2019) 9:336 | DOI: 10.1038/s41598-018-36728-y In endothelial cells, several signals controlling sVEGFR1-i13 expression have been identified. It has been shown that VEGF-A upregulates sVEGFR1-i13 level by a mechanism depending on VEGFR2 15,16 . A cooperative role between the arginine demethylase and lysine hydroxylase JMJD6 (JuMonJi Domain containing-protein 6) and the splicing factor U2AF65 was also reported to control sVEGFR1-i13 expression 17 . Moreover, a NOTCH1 decoy variant which reduces NOTCH1 signaling was shown to increase sVEGFR1-i13 levels and to inhibit angiogenesis in retinas and tumors 18 . Up to now, the molecular mechanisms that regulate sVEGFR1-i13 expression in cancer cells have not been described. In this study, we identify a VEGF 165 /SOX2/SRSF2 network that controls sVEGFR1-i13 expression in squamous lung carcinoma cells. Importantly, this network also contributes to sVEG-FR1-i13 accumulation in response to anti-angiogenic therapies.

Results
VeGF 165 controls sVEGFR1-i13 expression in lung cancer cells. It was previously shown that VEGF-A up-regulates sVEGFR1-i13 but not full-lenght VEGFR1 expression in human vascular endothelial cells 16 . To test whether VEGF-A controls sVEGFR1-i13 in tumor cell lines, we treated MGH7 and H2170 squamous lung carcinoma cells for 24 hours with various recombinant splice variants of VEGF-A, namely rhVEGF 121 , rhVEGF 165 and rhVEGF 189 . In both cell lines, an increase of sVEGFR1-i13 but not VEGFR1 mRNA level was observed upon treatment with rhVEGF 165 only (Fig. 1b). Similar results were obtained when sVEGFR1 levels were quantified by ELISA assay in cellular extracts (Fig. 1c). To confirm these data, we transiently transfected MGH7 and H2170 cells with a plasmid encoding VEGF 165 . We showed that sVEGFR1-i13 mRNA (Fig. 1d) and protein (Fig. 1e) levels are upregulated in cells overexpressing VEGF 165 as compared to control cells. In contrast, full-lenght VEGFR1 mRNA was not affected by VEGF 165 (Fig. 1d). Taken together, these data demonstrated that VEGF 165 regulates VEGFR1 pre-mRNA splicing in favor of its truncated splice variant sVEGFR1-i13 in lung tumor cells. To confirm the link between VEGF 165 and sVEGFR1-i13, we took advantage of a retrospective Non Small Cell Lung Carcinoma (NSCLC) cohort in which we previously performed VEGF 165 , VEGFR1 and sVEG-FR1-i13 immunohistochemical stainings 4,19 . In agreement with our results in cell lines, NSCLC patients with high VEGF 165 scores were those with high sVEGFR1-i13 level (Fig. 1f, p = 0.007). Of note, no relationship between VEGF 165 scores and VEGFR1 immunostainings was observed in these patients (data not shown).
VeGF 165 cooperates with SOX2 to regulate sVEGFR1-i13 expression in lung cancer cells. It was recently reported that VEGF-A controls the expression of the transcription factor SOX2 in breast and lung cancer cells 20 . As SOX2 is amplified in about 30% of squamous lung carcinoma patients 21 , we asked whether it plays a role in the regulation of sVEGFR1-i13 by VEGF 165 . We first showed that MGH7 and H2170 cells express detectable levels of SOX2 protein (Fig. 2a). When SOX2 was neutralized by siRNA, the increase of sVEGFR1-i13 mRNA and protein levels following treatment with rhVEGF 165 (Fig. 2b,c) or after transfection with a plasmid encoding VEGF 165 (Fig. 2d,e) was prevented. VEGFR1 mRNA levels were never affected by the knock-down of SOX2, whatever the conditions. Interestingly, a decrease of sVEGFR1-i13 protein level was observed when SOX2 was neutralized (Fig. 2b,d), thereby indicating that SOX2 might also control sVEGFR1-i13 expression in the absence of VEGF 165 stimulation. We noticed an increase of SOX2 protein (Fig. 2d) and mRNA (Fig. 2f) levels in cells transfected with the plasmid encoding VEGF 165 , consistent with previous data demonstrating that VEGF-A controls SOX2 expression in lung cancer cells 20 . As a whole, our data provided the first evidence that a VEGF 165 / SOX2 signaling network regulates sVEGFR1-i13 expression in squamous lung carcinoma cells.
In SQLC patients, inhibitory alterations of NOTCH signaling are frequent. In addition, in transgenic lung tumor mouse models, SOX2 binds to Notch1 and Notch2 regulatory regions leading to a significant reduction of Notch1 and Notch2 transcripts 22 . As it was recently shown that inhibition of NOTCH signaling by a NOTCH1 decoy variant increases sVEGFR1-i13 level in endothelial cells 18 , we thus tested whether NOTCH signaling regulates sVEGFR1-i13 expression in SQLC cells. To do so, MGH7 or H2170 cells were treated with FLI-06, a gamma-secretase inhibitor which blocks NOTCH processing and trafficking. As compared to control untreated cells, we did not observe any significant variation in sVEGFR1-i13 protein ( Supplementary Fig. 1a), or mRNA ( Supplementary Fig. 1b), o) levels. Rather a global increase in both VEGFR1 and sVEGFR1-i13 mRNA levels was observed following FLI-06 treatment ( Supplementary Fig. 1b).
A VeGF 165 /SOX2/SRSF2 signaling network controls sVEGFR1-i13 expression in lung tumor cells. Then, we investigated the mechanism by which SOX2 controls VEGFR1 pre-mRNA splicing. Serine Arginine Rich (SR) proteins belong to a family including twelve members that are critical splicing factors involved in constitutive and alternative pre-mRNA splicing 23 . We previously demonstrated that the SR proteins, SRSF1, SRSF2 and SRSF6, are up-regulated in NSCLC patients compared to normal lung tissues 24 . More recently, we observed an heterogeneous immunostaining of sVEGFR1-i13 in the same series of tumors 4 . Therefore, we looked for a putative relationship between sVEGFR1-i13 status and SR proteins levels. As shown in Fig. 3a, NSCLC patients with high sVEGFR1-i13 immunostaining scores were also those with high level of SRSF2 protein (p = 0.005). In contrast, no relationship was found between sVEGFR1-i13 scores and SRSF1 or SRSF6 status (patients with high versus low level; data not shown). To confirm these results, we took advantage of an Affymetrix dataset published in a cohort of 130 SQLC patients which contained two probe sets that distinguish between sVEGFR1-i13 and full-length VEGFR1 mRNAs (Gene Omnibus data set GSE4573 25 ). We found that SQLC patients with high SRSF2 mRNA level (>50 th percentile) also have high sVEGFR1-i13 mRNA levels (Fig. 3b, p = 0.01). These results were confirmed by Spearman correlation analysis ( Supplementary Fig. 2a, r = 0.5089, p < 0.0001). In contrast, no correlation was observed between SRSF2 and VEGFR1 mRNA levels (data not shown). To study whether SRSF2 was involved in the control of sVEGFR1-i13 by VEGF 165 , we transfected MGH7 cells with mismatch or Srsf2 siRNA, and studied sVEGFR1-i13 expression by western blotting following rhVEGF 165 treatment. The knock-down of SRSF2 significantly prevented the increase of sVEGFR1-i13 protein    (Fig. 3d) levels induced by rhVEGF 165 . These data highly suggested that a VEGF 165 /SRSF2 network controls sVEGFR1-i13 expression in SQLC. It was shown that the OSN (OCT4, SOX2, NANOG) transcription complex controls the expression of SRSF2 26 . Therefore, we postulated that SRSF2 may be located downstream of the VEGF 165 /SOX2 network to regulate the expression of sVEGFR1-i13. In agreement with this hypothesis, the neutralization of SOX2 using siRNA prevented the SRSF2 accumulation in cells treated with rhVEGF 165 (Fig. 3e). In addition, SQLC patients with high levels of SOX2 mRNA (>50 th percentile) were also those with high SRSF2 mRNA levels in the GSE4573 dataset (Fig. 3f, p = 0.0006). These results were confirmed by Spearman correlation analysis in the GSE4573 cohort ( Supplementary Fig. 2b, r = 0.5342, p < 0.0001) as well as in another SQLC cohort (GSE68793; Supplementary  Fig. 2c, r = 0.30, p = 0.0004). Of note, no significant correlation was observed between SRSF2 and SOX2 mRNA levels in two distinct lung ADC cohorts ( Supplementary Fig. 2d,e). Taken together, these data demonstrated that a VEGF 165 /SOX2/SRSF2 signaling network controls sVEGFR1-i13 expression in SQLC patients.
Anti-angiogenic therapies activate the VEGF 165 /SOX2/SRSF2 network to control sVEGFR1-i13 expression. We recently demonstrated that anti-angiogenic therapies induce sVEGFR1-i13 expression (mRNA and protein) in squamous lung carcinoma cell lines and murine tumorgrafts 4 . Therefore, we asked whether the VEGF 165 /SOX2/SRSF2 network was involved. We first showed that bevacizumab, a monoclonal antibody targeting VEGF-A, or KI8751 or SU5416, two VEGFR tyrosine kinase inhibitors, induced the accumulation of intra-cellular VEGF 165 protein in MGH7 and H2170 cells (Fig. 4a,b). The increase of intracellular VEGF 165 was also observed in squamous lung carcinoma murine tumorgrafts treated with sunitinib, a VEGFR TKI, or DC101, an antibody against murine VEGFR2 (Fig. 4c). In these tumorgrafts, we previously showed an increase of sVEGFR1-i13 upon treatment with anti-angiogenic therapies 4 . Importantly, neutralization of SOX2 prevented the increase of intra-cellular sVEGFR1-i13 protein and mRNA levels in response to anti-angiogenic therapies in MGH7 (Fig. 4d,e) and H2170 cells (Fig. 4f,g). Conversely, the knock-down of SOX2 did not modify VEGFR1 mRNA levels, whatever the treatments (Fig. 4e,g). NOTCH1/NOTCH2 mRNA levels also did not vary upon anti-angiogenic treatments ( Supplementary Fig. 1c), indicating that NOTCH signaling is not involved in the regulation of sVEGFR1-i13 expression in response to anti-angiogenic therapies. Of note, we were not able to test the effects of VEGF 165 knock-down in these conditions as we did not find siRNAs selectively targeting VEGF 165 only (data not shown).
Moreover, in both cell lines, the accumulation of sVEGFR1-i13 following anti-angiogenic treatments was associated with an increase of SRSF2 protein level and was significantly prevented in cells deprived of SRSF2, both at the protein (Fig. 5a,c,e) and mRNA (Fig. 5b,d) levels. The knock-down of SRSF2 did not modulate VEGFR1 mRNA levels, whatever the conditions. Furthermore and consistent with a role of SOX2 in regulating SRSF2 expression, SOX2 knock-down prevented SRSF2 accumulation in response to anti-angiogenics in both cell lines (Fig. 5f). More importantly, by performing chromatin immunoprecipitation experiments, we finally showed that SOX2 directly binds to the SRSF2 promoter and that SU5416 treatment increases this binding (Fig. 5g). As a whole, these data demonstrated that a VEGF 165 /SOX2/SRSF2 network controls VEGFR1 pre-mRNA splicing towards sVEGFR1-i13 expression in lung cancer cells treated with anti-angiogenic therapies (Fig. 6).

Discussion
To our knowledge, nothing is known regarding the molecular mechanisms that regulate the expression of VEGFR1 splice variants in cancerous cells. Hence, high levels of sVEGFR1s have been previously reported in plasma, serum or tissues of many types of cancer such as colorectal cancer, breast cancer, glioblastoma and lung cancer [27][28][29][30] . Such increase has been mainly correlated with poor prognosis, but the molecular mechanisms behind this regulation have not been clearly elucidated. In addition, several clinical trials have reported variation in circulating levels of sVEGFR1 following anti-angiogenic therapies, and a high level of sVEGFR1s was correlated with a poor therapeutic response [5][6][7][31][32][33][34][35] . Again, none of these studies has investigated intra-tumoral levels of sVEGFR1s, nor pointed out to a a specific mechanism for sVEGFR1 generation. Recently, we demonstrated that the sVEG-FR1-i13 splice variant is up-regulated in lung cancer patients, and contributes to the progression and the escape of squamous lung carcinoma from anti-angiogenic therapies 4 . In this study, we demonstrate that a signaling network involving the VEGF 165 , SOX2 and SRSF2 proteins controls the expression of sVEGFR1-i13 in these tumors. More importantly, we demonstrate that this signaling network also controls the expression of sVEGFR1-i13 in response to anti-angiogenic therapies. These results identify for the first time upstream regulators of VEGFR1 pre-mRNA splicing in cancer cells.
In endothelial cells, the expression of sVEGFR1-i13 is up-regulated by VEGF-A 15,16 , hypoxia 36 or by decreased expression of the oxygen-sensing hydroxylase JMJD6 which controls the hydroxylation of the splicing factor U2AF65 17,37 . Beside VEGF-A, we also studied the impact of hypoxia on sVEGFR1-i13 expression and the status of JMJD6 in our cellular models, but we did not find any significant effect in cells treated or not with anti-angiogenics (data not shown). We focused on SOX2 because its amplification occurs in about 30% of SQLC patients 21 , and because VEGF-A was shown to control SOX2 expression in cancer cells 20 . In this study, we identified SOX2 as an upstream regulator of sVEGFR1-i13 in SQLC cells. In addition, we showed that SOX2 regulates the expression of the SR protein SRSF2, an ubiquitous splicing factor that plays a critical role in both constitutive and alternative pre-mRNA splicing 23 . We previously demonstrated that SRSF2 protein is overexpressed in SQLC patients as compared to normal lung tissues 24 . In this study, we observed an association between SRSF2 and sVEGFR1-i13 protein levels in the same series of patients (Fig. 3a, p = 0.005). Interestingly, SRSF2 was recently shown to be part of a WNT5a-dependent signaling network controlling sVEGFR1 expression in aged mice exposed to ischemic stress 38 . In addition, we found a correlation between SRSF2 and SOX2 mRNA levels in two Gene Omnibus cohorts, namely GSE4573 ( Supplementary Fig. 2b, p < 0.0001) and GSE68793 ( Supplementary  Fig. 2c, p = 0.0004). This was consistent with SOX2 and SRSF2 being closely connected in SQLC tumors. In  agreement, we showed that SRSF2 is a direct target gene of SOX2 in SQLC cell lines. SOX2, together with OCT4 and NANOG, is part of the transcription OSN complex, and SRSF2 is an OCT4 target gene required for pluripotency in human pluripotent stem cell (hPSC) 39 . Therefore, these and our results highly suggest that SRSF2 could be an important mediator of the OSN complex function in non transformed and transformed cells. In this setting, it remains to characterize the upstream signals that control the activation of the VEGF 165 /SOX2/SRSF2 pathway in lung tumors. We recently observed a correlation between sVEGFR1-i13 and β1 integrin expression in SQLC cell lines and primary tumors 4 . Therefore, one possibility is that β1 integrin senses the extracellular matrix to activate the VEGF 165 /SOX2/SRSF2 pathway. We are currently testing this hypothesis. We found that sVEGFR1-i13 accumulates in SQLC cells in response to anti-angiogenic therapies by a mechanism that requires SOX2 and SRSF2 proteins. In pre-clinical SQLC murine models, resistance of tumor cells to anti-angiogenic therapies has been associated with the accumulation of Cancer Stem Cells markers (CSC) 40 . In addition, it was previously shown that VEGF-A and SOX2 proteins cooperate to promote CSC self-renewal in lung cancer cells 20 and that SRSF2 is the most enriched splicing factor in human pluripotent stem cells 26 . It has been suggested that high levels of sVEGFR1 contribute to tumor escape from anti-angiogenic therapies by decreasing baseline microvascular density 32,34 . Based on our results, it is thus tempting to speculate that high levels of sVEGFR1-i13 may also be part of a VEGF/SOX2/SRSF2 axis involved in CSC self-renewal post-treatment.
To conclude, we highlight VEGF-A, SOX2 and SRSF2 proteins as regulators of VEGFR1 pre-mRNA alternative splicing in SQLC, and demonstrate that anti-angiogenic therapies affect this network. Because treatment of SQLC patients remains very challenging, our data offer a new signaling pathway to be explored in these patients for potential therapeutic avenues.

Patients, tissue and subcutaneous tumorgrafts samples. Seventy-seven human NSCLC and 17
matched normal lung parenchymas were included in this study. Tumors consisted of 41 lung adenocarcinoma (ADC) and 36 SQLC. Tumor tissues and normal lung parenchyma, taken away from the bulk of the tumor, were collected from resection of lung tumors, and stored for scientific research in a biological resource repository (Centre de Ressources Biologiques, CHU Albert Michallon, Grenoble Hospital) following national ethical guidelines. All patients enrolled in this trial provided written informed consent. Tissue banking and research conduct was approved by the Ministry of Research (approval AC-2010-1129) and by the regional IRB (CPP 5 Sud Est). For histological classification, tumor samples were fixed in formalin, and diagnosis was made on paraffin-embedded material using the WHO VII th classification of lung criteria 42 . For each case, one section from the most representative block was chosen. These sections always contained more than 70% tumor cells. Immunohistochemical stainings of VEGF 165 and sVEGFR1-i13 were performed as previously described 4 . Sections from UN-SCC680 subcutaneous tumorgrafts were recovered from previous experiments 40 and stained for VEGF 165 . For automatic quantification of VEGF 165 staining in mouse models, sections were scanned using a ZEISS AxioImager M2 automated slide scanner with 5X magnification and the images were analyzed using Image J software. Threshold values were adjusted until masked brownpixels correlated with positive immunostaining or with total area of the digitized tissue. The percentage of positive areas was then calculated for each staining. The Affymetrix datasets GSE4573 published in a cohort of 130 SQLC patients and containing two probe sets that distinguish between sVEGFR1-i13 and full-length VEGFR1 mRNAs 25 and GSE68793 were recovered from Gene Omnibus (GEO).

RNA interference.
The two siRNA specifically targeting sVEGFR1-i13 were: sVEGFR1-i13 (1)  ELISA assays. ELISA assays were performed in duplicate in 96-wells plates using a Quantikine sVEGFR1 kit (R&D Systems). Manipulations were carried out according to manufacturer's instructions. Briefly, 1.5 × 10 6 cells/ well were seeded in 6-wells plates and treated or not for different times. The concentration of sVEGFR1 in the supernatants was calculated from the absorbance value compared to the standard curve and expressed in pg/ml.

Antibodies and immunoblotting.
Immunoblotting experiments were performed as previously described 43 . The antibodies used were: anti-actin from Sigma, anti-tubulin (clone B512, sc-23948) from Santa Cruz, anti-SOX2 (AB5603) and anti-phospho-VEGFR1-Tyr1213 (cat#07-758) from Millipore and anti-SRSF2 (clone 4F-11) from Euromedex. The specific anti-sVEGFR1-i13 was generated against a peptide mapping in the unique C-terminus 13 . We previously checked that this antibody recognizes sVEGFR1-i13 protein in our cells by using siRNA targeting retained intron 13 in sVEGFR1-i13 4 . The anti-VEGF 165 antibody raised against the six terminal amino acids (CDKPRR sequence) and a sixteen amino acids sequence targeting terminal part of VEGF 165 encompassing exons 7 and 8a was produced by Covalab (Villeurbanne, France). We previously checked that this antibody recognizes a recombinant VEGF 165 but not VEGF 165 b protein 19 .
Scientific RepoRts | (2019) 9:336 | DOI:10.1038/s41598-018-36728-y Chromatin immunoprecipitation assay. Chromatin immunoprecipitation experiments were performed in MGH7 and H2170 cells treated or not (NT) with 10 µM SU5416 for 24 hours. ChIP experiments were performed using the ChIP-IT R Express Magnetic Chromatin Immunoprecipitation kit from Active Motif (La Hulpe, Belgium) according to manufacturer's instructions. Briefly, cells were formaldehyde cross-linked and chromatin was isolated and sonicated using a Bioruptor apparatus. An equal amount of chromatin (30 µg) was precleared, immunoprecipitated with either a polyclonal antibody specific for SOX2 (D6D9 XP R , ChIP formulated, Cell Signaling) or unrelated rabbit IgG, overnight at +4 °C, washed and reverse cross-linked. One-twentieth of the immunoprecipitated chromatin was analyzed for the presence of SRSF2 promoter DNA by Q-PCR using primers that flanked two potential SOX2 consensus binding sites (TTGT) at (−169; −165) and (−216; −212) positions on the promoter. A sequence corresponding to the GAPDH promoter was used as a negative control in SOX2 ChIP. Q-PCR studies were performed using using iTaq ® qPCR Universal SYBR Green Supermix (Bio-Rad). Input DNA sample corresponding to 1% of immunoprecipitated chromatin was analyzed in parallel in order to normalized the results of each ChIP DNA sample to the corresponding input DNA sample. The primers used were as follow: SRSF2 forward 5′-AAGGTTTCATTTCCGGGTGG-3′; SRSF2 reverse 5′-GGGACACTGGGAAAGGCCA-3′; GAPDH forward 5′-AGCTCAGGCCTCAAGACCTT-3′ and GAPDH reverse 5′-AAGAAGATGCGGCTGACTGT-3′.
Statistical analyses. The statistical analyses were performed using Statview software (Abacus Concepts).
Descriptive analyses comparing continuous and two-level categorical variables were carried out using the Mann-Whitney U test. P values < 0.05 were considered significant.

Data Availability Statement
All data generated or analysed during this study are included in this published article (and its Supplementary Information Files).