Respective roles of Pik3ca mutations and cyproterone acetate impregnation in mouse meningioma tumorigenesis

Despite their rarity, PIK3CA mutations in meningiomas have raised interest as potentially targetable, ubiquitous mutations owing to their presence in sporadic benign and malignant tumors but also in hormone-related cases. Using new genetically engineered mouse models, we here demonstrate that Pik3ca mutations in postnatal meningeal cells are sufficient to promote meningioma formation but also tumor progression in mice. Conversely, hormone impregnation, whether alone or in association with Pik3ca and Nf2 mutations, fails to induce meningioma tumorigenesis while promoting breast tumor formation. We then confirm in vitro the effect of Pik3ca mutations but not hormone impregnation on the proliferation of primary cultures of mouse meningeal cells. Finally, we show by exome analysis of breast tumors and meninges that hormone impregnation promotes breast tumor formation without additional somatic oncogenic mutation but is associated with an increased mutational burden on Pik3ca-mutant background. Taken together, these results tend to suggest a prominent role of Pik3ca mutations over hormone impregnation in meningioma tumorigenesis, the exact effect of the latter is still to be discovered.


INTRODUCTION
Meningiomas are the most common primary tumors of the central nervous system in adults and are mostly benign (WHO grade I) [1]. In recent years, there has been a significant effort to unravel the molecular biology of meningiomas at genetic, transcriptomic, and epigenetic levels [2][3][4][5][6]. While the studies have led to a better understanding of meningioma tumorigenesis, no additional treatment has yet emerged following those discoveries. Mutations in the NF2 gene remain the most frequent driver event in meningiomas [7], and the various biological roles of merlin [8], the gene product of NF2, make it difficult to generate an effective treatment. We therefore decided to focus our efforts on a less frequent but more easily targetable driver of meningioma tumorigenesis, the PIK3CA gene, whose role in the disease has recently increased.
In WHO grade I meningiomas, PIK3CA mutations occur in 4.7% of all tumors and 7.4% of non-NF2 meningioma cases [9]. In a majority of cases (68%) [4], PIK3CA mutations co-occur with TRAF7 mutations raising doubt on whether PIK3CA mutations alone may be sufficient to promote meningioma tumorigenesis. More recently, PIK3CA mutations in meningioma have gained interest due to their prominent role in hormone-associated meningioma tumorigenesis. Meningiomas are considered hormone-sensitive tumors due to their association with the female gender (sex ratio, F/M 2:1) [10], the presence of progesterone receptors at their surface [11], and the rapid changes in meningioma size observed during pregnancy [12].
Following several case reports and series on the association between synthetic anti-androgenic progestin intake (cyproterone acetate, chlormadinone acetate, nomegestrol acetate) and meningioma, a recent epidemiology study confirmed the increased risk of developing meningiomas after prolonged anti-androgenic progestin treatment [13]. We further characterized this association by demonstrating a shift in the mutational landscape of progestin-associated meningiomas compared with a series of unselected tumors [14]. Progestin-associated meningiomas were prominently PIK3CA-mutant skull base tumors. The tumorigenesis mechanisms underlying the association between progestin intake and PIK3CA mutations are still unknown.
In order to determine if PIK3CA mutations alone were sufficient to promote benign and aggressive meningioma tumorigenesis and to determine the associated role of hormone intake, we created a novel mouse model of Pik3ca-mutant meningioma. We first show that patterns of expression of progesterone receptors in mouse are comparable to human meninges. We then demonstrate that, in mouse, Pik3ca mutations in postnatal meningeal cells are sufficient to promote meningioma formation, as opposed to cyproterone acetate impregnation, which is not sufficient to promote meningioma formation in mice, whether alone or in association with Pik3ca and Nf2 mutations. We confirm these results by in vitro analysis of the effect of cyproterone acetate and Pik3ca mutations on primary cultures of mouse dural and arachnoid cells. Finally, we show progestin impregnation promotes tumor formation without additional somatic mutation but is associated with an increased mutational burden when combined with Pik3ca mutations.

Progesterone receptors expression in mouse meninges
In order to determine if mouse was a reliable model to test the effect of hormone impregnation compared to humans, we decided to study the expression of progesterone receptors in mouse meninges. The main progesterone receptor, Pgr, is a cytoplasmic protein that dimerizes and translocates to the nucleus to activate Progesterone Response Elements (PRE) in DNA upon progesterone binding. Nuclear staining of Pgr is therefore considered a reliable marker of Pgr activation, and Pgr is expressed in a majority of meningiomas. In normal meninges, Pgr expression has been observed in normal human [15,16] and rabbit [17] arachnoid tissue. Apart from Pgr, other transmembrane receptors have been more recently associated with progesteronedependent effects, such as Pgrmc1 (Progesterone receptor membrane component 1). Following its description in rat meninges [18], we decided to study its expression in mouse meninges.
We first analyzed nuclear progesterone receptor expression in adult mice based on Hybridization data from Allen Atlas (Fig. 1, upper panel). As opposed to Pgds, a robust marker of the meninges, both Pgr and Pgrmc1 were not expressed in mouse meninges except in scattered places. As Allen Atlas images do not encompass leptomeninges, we conducted an additional immunohistochemical analysis of progesterone receptors Pgr and Pgrmc1 in mouse meninges. We show that Pgrmc1 is expressed in arachnoid cells starting E18 both at the skull base and the convexity. On the opposite, we were not able to demonstrate any nuclear Pgr staining both in the arachnoid and the dura mater of developing and adult mice (Fig. 1, lower panel). Bulk-RNA sequencing data from neonatal dura mater confirms the expression of Pgrmc1 and Pgrmc2 in mouse meninges but not Pgr [19], as in single-cell analyses of normal human dura mater [20]. Due to the similarity of expression profiles of progesterone receptors in mouse and humans, at least on a transcriptomic level, we considered mouse could be a reliable model of hormone impregnation.
Pik3ca mutation in postnatal meningeal cells is sufficient to promote meningioma formation in mice To investigate the role of Pik3ca mutation in meningeal tumorigenesis, we had previously generated a cohort of PGDSCre;Pik3ca mice [21] and observed that prenatal activation of Pik3ca in PGDS-positive cells was not sufficient to promote meningioma tumorigenesis in mice. We therefore sought to analyze the effect of postnatal Pik3ca activation and generated a first cohort of 20 Pik3ca flox/flox mice injected with adCre (mean volume: 3 μL) between postnatal day 2 and 5 ( Table 1 and Fig. 2). Due to the intravascular and subcutaneous diffusion of adCre, we observed 95% of early head hypertrophy associated with subcutaneous vascular malformations, leading to early mortality (mean survival: 1.6 months) ( Supplementary Fig. 1). Upon histological analysis, we nevertheless observed 30% of meningiomas, including one mouse with two distinct meningiomas of different histological subtypes (Fig. 3A) and one Grade II tumor [22] with brain invasion and several mitoses in a 9-month-old mouse (Fig. 3C, D). Interestingly, we also observed one case of intraparenchymal cavernoma (Fig. 3B).
We also decided to analyze the effect of Pik3ca activation in the meninges of adult mice. A cohort of 22 Pik3ca flox/flox mice was therefore injected with adCre (mean volume: 6 μL) at 6 weeks, either at the convexity (15 mice) or at the skull base (7 mice). We observed 50% of meningiomas, including one mouse harboring a convexity fibroblastic meningioma and skull base transitional meningioma (Fig. 3E, F). While meningiomas were mostly meningothelial in the postnatal model (5/6, 83%), they were prominently fibroblastic in the adult model (7/11, 63%). Taken together, these results demonstrate that postnatal expression of a mutant form of Pik3ca in the meninges is sufficient to promote meningioma formation of different histological grades and subtypes in contrast to prenatal expression.
Cyproterone acetate intake fails to promote meningioma formation in mice, either alone or in combination with Pik3ca or Nf2 mutations In order to study the role of progestin intake on meningioma tumorigenesis, we treated a cohort of 29 wild-type mice with cyproterone acetate (CPA) for 9 months. After a mean follow-up of 11 months, we did not find any meningioma on histological analysis but observed uterus hypertrophy (Fig. 4A) in all mice, thus demonstrating CPA impregnation, as previously described in breast cancer mouse models [23]. Interestingly, we also discovered mammary tumors in 90% of mice, defined as muco-epidermoïd carcinomas of the breast (Fig. 4B, C) and one spinal intramedullary epidermoid cyst (Fig. 4D). In order to determine if progestin intake and Pik3ca activation could synergize to promote meningioma formation in mice, we treated a cohort of 22 PGDSCre; Pik3ca H1047R mice with CPA during 8 months. Similar to the first group with CPA impregnation, mice were sacrificed after a mean follow-up of 8.3 months due to mammary tumors. No meningioma was found on histological diagnosis. As meningioma frequency was low in PGDSCre; Pik3ca H1047R mice, we sought to determine if CPA intake might synergize with postnatal activation of Pik3ca mutations. In order to avoid the effects of intravascular and subcutaneous diffusion of adCre, we lowered the injected volume to 1 μL and generated two cohorts of adCre; Pik3ca H1047R mice with (13 mice) and without (14 mice) CPA impregnation. The rate of subcutaneous vascular malformations was lowered to 62% in the control group and 23% in the CPA-treated group. The survival of mice was nonetheless reduced (Fig. 2B), and 10 out of 13 mice received only 3 months of CPA impregnation, compared to 9 months in previous cohorts. While the frequency of meningiomas in the control group was similar to our previous cohort (29% vs. 30%), it was much lower (8%) in the CPA-treated group.
Finally, we postulated that progestin intake might contribute to the growth of pre-existing benign tumors. We therefore decided to treat a cohort of 17 PGDSCre het ; Nf2 del2/flox mice with cyproterone acetate compared to a control cohort of 19 PGDSCre het ; Nf2 del2/flox mice [24]. On histological analysis after a mean follow-up of 9.5 months, we found 88% of fibroblastic skull base meningiomas in the CPA-treated group and 89% of fibroblastic skull base meningiomas in the control group. The mean tumor size was 0.2294 mm 2 (median: 0.118 mm 2 ; range: 0.033-0.718) in the CPA-treated group and 0.305 in the control group (median: 0.256 mm 2 ; range: 0.067-0.823). We did not find any statistically significant difference in tumor size between the two groups. Collectively, these results indicate that cyproterone acetate failed to promote meningioma formation and/or progression in mice, whether it be on a Pik3ca-or an Nf2-related genetic background.

Pik3ca mutations but not cyproterone acetate increase proliferation in meningeal cells in vitro
To analyze the effect of Pik3ca activation in the meninges according to their nature, we established primary arachnoid and dura mater cultures from skull base meninges of Pik3ca H1047R mice. AdCre infection of primary cells in vitro was assessed by detection of GFP fluorescence (Fig. 5C). No fluorescence was expressed in the adLacZ-infected control cells (data not shown). Proliferation assays showed that Pik3ca H1047R expression induced a proliferative advantage both in arachnoidal and dural cells at day 9 (p < 0.05, t-test) (Fig. 5A, B). Increased proliferation was associated with activation of the Pi3k-Akt-mTor pathway through increased phosphorylation of Akt and S6RP (Fig. 5D) Table 1). Cyproterone acetate was unable to promote increased cellular proliferation of dural cells, either alone or in combination with Pik3ca H1047R expression (Fig. 5E, G). In arachnoidal cells, however, cyproterone acetate increased cellular proliferation alone and in combination with CPA without reaching statistical significance (Fig. 5F, H).

Exome analysis of hormone and Pik3ca-related tumors and meninges in mice
In order to determine if progestin impregnation alone was sufficient to promote mutagenesis in mouse tissues, we decided to perform exome analysis on mouse meninges treated with CPA compared to control meninges and in CPA-related mammary tumors either on a wild-type or a Pik3ca-mutant background. Five tumors and two meninges were analyzed. We did not find any oncogenic mutation in mouse meninges. In the mammary tumor developed on a wild-type background, no oncogenic mutation was observed. Interestingly, in mammary tumors developed when combined with Pik3ca mutations, we found additional mutations already described in human breast tumors, including mutations in Map2k4 and Ctnnb1 described in the COSMIC database ( Fig. 6 and Supplementary Table 2).

DISCUSSION
We demonstrate in this new genetically engineered mouse model that Pik3ca H1047R alone may promote meningioma formation in mice, as in previously reported mouse models [22,25,26]. Interestingly, we show that PGDS-driven prenatal expression of Pik3ca H1047R failed to promote meningioma formation as opposed to previously published models with Nf2 bi-allelic inactivation and Smo activating mutation [24,26]. This result might be linked to the fact that only postnatal differentiated meningeal cells are sensitive to Pik3ca H1047R expression and underlines that heterozygous mutant Pik3ca remains a mild oncogene, raising the possibility that an additional mutation in a co-regulator of the pathway, as observed in ovarian cancer [27], or in a functionally independent gene, as observed with TRAF7 in meningiomas [4,9], might be required for mutant Pik3ca to promote full transformation in vivo. We also observe that in older animals, Pik3ca H1047R may trigger meningioma progression towards higher histological grades, similar to human cases. In the literature, 86 anaplastic meningiomas have been tested for PIK3CA mutations [3,9,[28][29][30][31][32]: mutations were found in only three cases (3.4%) but 7.8% of NF2independent tumors. In human anaplastic PIK3CA-mutant meningiomas, PIK3CA mutations are often found alone [28,29], without any other identifiable genomic event explaining malignant meningioma progression. The underpinnings of this progression mechanism remain unknown.
As PIK3CA mutations are more frequent in progestin-related meningiomas [14], we sought to determine if progestin impregnation in mice might synergize with Pik3ca mutations to promote meningioma formation. We discovered that both in vitro and in vivo, cyproterone acetate impregnation had only a mild effect on meningeal proliferation and no synergistic effect with Pik3ca mutation. Cyproterone acetate also failed to promote meningioma tumorigenesis initiation in wild-type mice. We tried to determine if this result was related to a different pattern of expression of progesterone receptors in mice compared to humans and found that Pgrmc1 was ubiquitously expressed in mouse meninges starting E18 while Pgr was not expressed in mouse meninges at any time point. Previous reports found that in rabbit, progesterone receptor immunoreactivity was present in arachnoid cells, while the cells of dura mater remained negative [17]. In human, most studies did not find any expression of progesterone receptor mRNA in normal adult pachymeninges [16,33] but also in normal arachnoid and leptomeningeal cell line [34]. In conclusion, the expression of progesterone receptors in mouse meninges does not seem to differ dramatically from human meninges and therefore does not explain the absence of meningiomas under progestin impregnation.
Those results are in line with previously published studies on models of breast cancer, in which progestins failed to promote tumor growth both in vivo and in vitro [23]; other studies have clarified that only androgenic progestins have a proliferative effect on breast epithelial cells [35]. The oncogenic effect of CPA was first a subject of debate as it was shown to be a hepatocarcinogen in the rat [36] and mouse [37]. However, no liver tumors have been reported to date in humans under CPA, and later studies demonstrated the existence of foci of preneoplastic cells in the liver of rats [38]. We here show that CPA-induced mammary tumors are devoid of oncogenic mutations but acquire additional mutations when occurring on Pik3Ca-mutant background. Taken together, the results suggest that CPA is not a carcinogen per se, as it does not induce mutations nor stimulate cell proliferation in wild-type cells.
This further complicates the definition of epidermoid and muco-epidermoid lesions observed both in the mammary gland and the central nervous system in our models, even more with no breast tumors reported to date, despite prolonged treatments in previous studies (104 weeks [37]). In the breast, it has nonetheless been demonstrated that progestin impregnation initiates a luminal to myoepithelial switch in a subset of tumor cells characterized by overexpression of certain cytokeratins, mainly CK5 and CK6 [23]. Our hypothesis is that progestins might induce such a morphologic transition in different mouse tissues and induce epidermoid cyst formation by keratin production following the epithelial patterning of progenitor cells. The morphologic transition specific to meningioma tissue remains to be determined.
Taken together, these results suggest that anti-androgenic progestins do not have an oncogenic effect per se but rather modify the tumor cellular architecture, thus explaining how progestin-associated meningiomas may present a "hormone addiction" and may rapidly decrease in size after progestin withdrawal [39]. This effect might therefore require pre-existing mutations in the target tissue, as already demonstrated in the normal endometrial epithelium [40] and breast epithelium [41], particularly for the PIK3CA gene. The rate and number of those mutations might be influenced by an underlying genetic susceptibility, as reflected by the prevalence of multiple hormone-associated meningiomas in identical twins [42,43]. Additional individual genomic, transcriptomic, and genome-wide association studies are therefore mandatory to explore this hypothesis in human tumors, but mouse studies will once again guide and complete discoveries on human disease.
In conclusion, this meningioma mouse modeling and in vitro study further explores the relationships between hormones and cancer as encountered in breast and uterine tumors by demonstrating that in mouse, progestins tend to modify cellular architecture rather than promote cell proliferation while Pik3ca mutations alone may promote tumorigenesis and tumor progression.

MATERIALS AND METHODS Mice
The following mouse strains were used in this study: R26-Pik3ca H1047R [44], PGDSCre [24], and Nf2 flox/flox [25]. All animal care and experimentation reported herein were conducted in compliance with the guidelines and with the specific approval of the Institutional Animal Care and Use Committee of the French Department of Agriculture (Reference number: #11977-2017100216257454v9). As hormone-related meningiomas are almost exclusively found in women, only female mice were used for all experiments except for CPA-treated AdCre; Pik3ca H1047R mice, for which individuals from both sexes were used due to the high initial mortality.

Viral injections
For postnatal infections, Cre recombinase was targeted to the mouse leptomeninges by direct injection of 1-3 μL (5* 10 10 to 1*10 11 plaque-forming units) of Ad5CMV Cre-GFP (AdCre) (University of Iowa Gene Transfer Vector Core) suspension into the subarachnoidal space on postnatal day 2 to 5 by subdural approach [25]. Injections in adult mice were realized in the subdural space either at the convexity or at the skull base under general anesthesia. The mouse head was stabilized in small animal stereotaxic frame. For the convexity approach, a burr hole was performed in the frontal region using a 27G needle after skin incision. With a 10 μL Hamilton Syringe (Hamilton, Switzerland), 6-8 μL of Ad5CMV Cre-GFP (AdCre) was infused over the course of 1 min in the frontal subarachnoid space. For skull base injections, a burr hole was drilled 1.5 mm anterior and 1.5 mm to the right of the bregma, with a needle slowly inserted downward about 5 mm until reaching the floor of the temporal fossa.

Treatment with cyproterone acetate
In order to establish the optimum dose for CPA impregnation in mice, we first established the standard dose used in humans as 17 mg/kg/month (50 mg/day, 20 days per month in a 60 kg woman). The presumed toxic dose in mice was previously defined as 2 mg/kg/month [37]. We then decided to use 90-day release pellets of 10 mg, corresponding to a dose of 165 mg/kg/month in mice and an equivalent dose of 13.4 mg/kg/month in humans, in order to mirror the doses used in women despite the increased risk of toxicity. CPA-treated mice received treatment from 1.5 months until 10.5 months using the 90-day release CPA pellets of 10 mg (Innovative Research of America, Inc, ref# NC-114). Pellets were implanted every 3 months, according to the company recommendations behind the neck under general anesthesia using Isoflurane.

Histopathology
Mice were monitored closely and killed when seriously ill or at 12 months. To preserve the arachnoidal and dural layers, the head was fixed in formalin in toto, decalcified, and sliced coronally before embedding in paraffin as previously described [25]. The terminology used for the description of the meningothelial lesions in the mouse models on HES staining is based on the WHO classification of human tumors with adaptations [22]. The term "meningothelial hyperplasia" is not used, as in humans, where it refers to the proliferation of reactive, normal arachnoidal cells. "Meningothelial proliferation" refers to very small (microscopic) lesions composed of meningothelial cells that represent early tumor formation. "Meningioma" refers to a larger meningothelial lesion with features similar to a WHO Grade I meningioma in humans.

Arachnoidal cells culture
Primary dural and arachnoidal cell cultures were established from FVB wild-type and R26-Pik3ca H1047R mice according to the team protocol [26]. The skull base arachnoidal tissue was surgically removed at the ventral surface of the brainstem, and the basal dural tissue was removed from the clivus. The meninges were then mechanically dissociated using a surgical blade in 12-well Petri dishes and cultured in Dulbecco's modified Eagle's medium (DMEM; Gibco/Invitrogen, Carlsbad, CA, USA) containing 10% fetal calf serum (FCS) and Penicillin/streptomycin 1%.

In vitro viral infection
After 10-12 days of primary culture, cells were passaged in 12-well plates at a concentration of 10,000 cells/plates. Cells were infected after 24 h either with Ad5CMVCre-eGFP, called adCre (group called "Pik3ca H1047R -adCre treated") or with Ad5CMVntLacZ (REF #VVC-U of Iowa-1917, University of Iowa, Viral Vector Core, Iowa City, IA, USA), called adLacZ (group called "Pik3ca H1047R -adLacZ treated") at 50 pfu/cells for 24 h. Expression of GFP was verified using a fluorescence microscope to assess infection of adCre.

In vitro hormonal treatment
Cyproterone acetate (CPA) (PubChem substance ID 24278309) was purchased from Merck in powdered form soluble in dimethyl sulfoxide (DMSO) solution. In order to test CPA toxicity in meningeal cells, the cells were exposed to different concentrations of CPA according to previous toxicity work on hepatocytes [45], and apoptosis tests were done. We decided to use CPA at maximal concentration with a low apoptosis rate, 10 µM. Each infected group was divided into two groups of hormonal treatment, either with CPA at 10 µM or with the same volume of DMSO. Meningeal cells were treated after the second passage in 12-well plates at a concentration of 10,000 cells/plates every 24 h for 7 days until the third passage in 25 cm 2 flasks at a concentration of 20,000 cells/flasks in order to have a long time of hormonal exposure. Proliferation assays were then conducted after third passages in triplicate and hormonal treatment was  continued all along this assay. The cells that were not used for proliferation assays were centrifuged at 300 × g at the end of the second passage, and cell pellets were stored at −80°C before DNA extraction.

In vitro functional assays
Proliferation assay: Arachnoidal and dural cells were seeded after infection in triplicate in 25-cm 2 flasks at the concentration of 20,000 cells/flasks and counted manually using a Malassez cell counting chamber 1, 4, 7 and 9 or 10 days after plating. Senescence assay: SA β-galactosidase staining was performed on infected and treated meningeal cells [46]. Apoptosis assay: After infection, meningeal cells were passaged, and we applied a standardized method of immunocytochemistry using Cleaved Capspase-3 antibody (Asp 175 ) (Cell Signaling Technologies, ref# 9661, dilution 1/200) with Di Aminido Phenyl Indol (DAPI).

Immunohistochemistry
Immunohistochemical stainings were performed on 5-μm paraffin sections. For antigen retrieval, after pressure boiling in 1 mM EDTA (pH 9.0 during 1 h for PGR, pH 6.0 during 30 min for PGRMC1), sections were blocked in 10% serum followed by incubation with primary antibodies (anti-PGR A/B; Abcam, ab63605, dilution 1/50 anti-PGRMC1; Cell Signaling Technology, #13856S, dilution 1/50). Secondary antibodies were used at 1:250 dilution and detected using the Vectastain ABC system and DAB substrate kit (Vector Labs, Burlingame, CA, USA). Normal mouse endometrium was used as a positive control.

Exome analysis
Blood and tumor DNA were extracted using the QIAmp DNA Micro kit (Qiagen). DNA was quantified and qualified using Tapestation. DNA libraries' preparations were realized following the manufacturer's recommendations (Library Preparation Enzymatic Fragmentation Kit 2.0 from TWIST). One capture reaction was performed with Mouse Exome Panel from TWIST and eight libraries in equimolar conditions. Final post capture reactions were purified and equimolar pooled, then sequenced on ILLUMINA Novaseq 6000 with SP-200 cartridge (2x800Millions of 100 bases reads), corresponding to 2x50Millions of reads per sample after demultiplexing. For the bioinformatics analysis, the DRAGEN TM somatic pipeline v3.10.4 (Illumina) was used to align the reads to the mouse reference genome (mm10) and perform the calling of the small variants (<50 bases) in tumornormal mode with each tumor sample compared against a unique control of similar genetic background (FVB). Only the variants passing the default filters were considered for analysis and annotated with the Ensembl Variant Effect Predictor (VEP) v105. We then performed an additional selection and retained only variants present in the protein-coding segments of the gene both in mouse and human and described in cancer according to the literature. The remaining variants were included if the mutation were described in the COSMIC database (cancer.sanger.ac.uk) [47] and/or the OncoKB database [48]. For mouse meninges, CPA-treated meninges were compared to control DMSO-treated meninges, and only variants expressed in CPA-treated meninges were retained.

Statistical analysis
Student's t-tests were used in proliferation assays, all tests were two-sided, and a p-value of ≤0.05 was considered to be statistically significant. Statistical analyses were performed using GraphPad Prism version 5.0 for MacIntosh (GraphPad software, La Jolla, California, USA).

DATA AVAILABILITY
All data supporting the findings of this study are available within the paper and its Supplementary Information.