Biallelic CCM3 mutations cause a clonogenic survival advantage and endothelial cell stiffening

Abstract CCM3, originally described as PDCD10, regulates blood‐brain barrier integrity and vascular maturation in vivo. CCM3 loss‐of‐function variants predispose to cerebral cavernous malformations (CCM). Using CRISPR/Cas9 genome editing, we here present a model which mimics complete CCM3 inactivation in cavernous endothelial cells (ECs) of heterozygous mutation carriers. Notably, we established a viral‐ and plasmid‐free crRNA:tracrRNA:Cas9 ribonucleoprotein approach to introduce homozygous or compound heterozygous loss‐of‐function CCM3 variants into human ECs and studied the molecular and functional effects of long‐term CCM3 inactivation. Induction of apoptosis, sprouting, migration, network and spheroid formation were significantly impaired upon prolonged CCM3 deficiency. Real‐time deformability cytometry demonstrated that loss of CCM3 induces profound changes in cell morphology and mechanics: CCM3‐deficient ECs have an increased cell area and elastic modulus. Small RNA profiling disclosed that CCM3 modulates the expression of miRNAs that are associated with endothelial ageing. In conclusion, the use of CRISPR/Cas9 genome editing provides new insight into the consequences of long‐term CCM3 inactivation in human ECs and supports the hypothesis that clonal expansion of CCM3‐deficient dysfunctional ECs contributes to CCM formation.

upregulated transcript in human TF-1 erythroleukaemia cells upon GM-CSF deprivation and was reported as CCM disease gene in 2005. 7,8 Pull-down and co-immunoprecipitation studies demonstrated that CCM3 forms a ternary complex with CCM1 and CCM2 in vitro and acts in intracellular networks with GCKIII serine/threonine kinases and other molecules. 9 Its inactivation in ECs is associated with altered autophagy, 10 impairment of the DLL4-Notch pathway, 11 activation of RhoA, 4 MEKK3-KLF2/4, 12-14 BMP/TGF-ß 15 signalling and increased exocytosis of angiopoietin 2. 16 Interestingly, the possibilities of the CRISPR/Cas9 system have not been used in these studies.
Since the clarification of the fundamental DNA interference mechanism and its first application for precise genome editing in mammalian cells, [17][18][19] the CRISPR/Cas9 system has become a versatile research tool. However, its efficiency is cell type-specific. Notably, CRISPR/Cas9-mediated gene disruptions in umbilical vein ECs, microvascular ECs, coronary artery ECs or ECs derived from cord blood colony-forming cells have almost exclusively been realized with viral delivery systems or plasmid transfections so far. [20][21][22][23][24][25][26] While the occurrence of potential off-target mutations may limit the use of both techniques, the delivery of the CRISPR/Cas9 components as ribonucleoprotein (RNP) complex can reduce off-target effects due to its short intracellular half-life. 27 Yu and co-workers have demonstrated that CRISPR/Cas9 RNPs can be used for genome editing in ECs. 28 Its efficiency has been reported to be significantly lower than in easy-to-transfect cells though.
In this study, we used a crRNA:tracrRNA:Cas9 RNP-based model of human cavernous malformations and siRNA transfections in parallel to study the molecular and functional effects of long-term CCM3 deficiency in comparison to its acute knockdown. Hereby, we demonstrate that CCM3 inactivation in human ECs induces a clonogenic survival advantage of functionally impaired CCM3 -/-ECs over time.

| Real-time deformability cytometry (RT-DC)
and immunofluorescence staining RT-DC analysis was performed as described before. 30

| 3D spheroid sprouting, migration, tube formation and proliferation assays
Spheroid formation and sprouting assay was performed as described previously. 32 ImageJ software was used for data analysis. WST-1 proliferation assay (Roche) was performed following manufacturer's instructions. For migration analyses, cell monolayers were scratched with a 200 µl pipet tip on a 24-well plate. Images were made after 0 and 7 hours. The relative migrated area was calculated using ImageJ software. Tube formation was analysed using a Matrigel assay (Corning, Kaiserslautern, Germany). After gel matrix incubation at 37°C for 60 minutes, 20 000 cells/well were seeded in 96-well plates. Tube formation was imaged after 18 hours and quantified with the angiogenesis analyzer for ImageJ (http://image.bio.methods.free.fr/Image J/).

| Statistical analysis
Data were analysed with the GraphPad Prism software (v7.0a, GraphPad Software, La Jolla, USA) and presented as mean and standard deviation (SD) if not stated otherwise. Two-tailed, one sample t tests were used for normalized data. Two or more groups were compared with a two-tailed, Student's t test or one-and two-way ANOVA, respectively. Dunnett or Šidák corrections were applied for multiple comparisons. False discovery rate (FDR; q) was used for analysis of small RNA-sequencing data. Cell area and elastic moduli measurements were evaluated with the software ShapeOut (Zellmechanik Dresden) using linear mixed model comparisons. This approach allows to estimate the statistical significance of large datasets with multiple biological replicates. Briefly, linear mixed models separate RT-DC data into two contributionsa fixed effect summarizing the impact of the knockout on cell mechanical properties and a random effect due to systematic or random measurement bias.
After defining an additional model that lacks the fixed effect term a maximum likely hood ratio is calculated and the P value is derived. A P < 0.05 or q-value <0.05 was regarded as statistically significant.

| Efficient CRISPR/Cas9-mediated CCM3 gene disruption in human ECs
To compare the efficiencies of viral and non-viral CRISPR/Cas9 delivery systems for CCM3 gene disruption in human ECs, we either transduced CI-huVECs with lentiviral particles or transfected them with crRNA:tracrRNA:Cas9 RNPs. Locus-specific effects were excluded since the same target sequence in the first coding exon of CCM3 was used in both approaches [crRNA (e3); Figure 1A]. Fourteen days after lentiviral vector transduction and puromycin SCHWEFEL ET AL. | 1773 selection, CCM3 mutant allele frequencies of 46%-66% were detected by T7EI cleavage assay and amplicon deep sequencing (Figure S1). In the non-viral approach, an average T7EI cleavage efficiency of 43% (range: 13%-70%) was observed 12 days after RNPtransfection without any selection for CRISPR/Cas9-edited cells (Figure 1B). Amplicon deep sequencing demonstrated total CCM3 mutant allele frequencies of up to 63% ( Figure 1C). As expected for non-homologous end-joining repair, the majority of variants induced by viral and non-viral CRISPR/Cas9 genome editing were small frameshift mutations located near the PAM site (±6 bp). The 1-bp duplication c.90dupT was one of the predominant alleles whereas only very few deletions or insertions of more than 20 bp were identified ( Figure 1C and Figure S1).
Even in samples with low initial editing efficiencies, we observed CCM3 indel frequencies of up to 88% after four to eight weeks ( Figure 1D). Furthermore, we noticed a marked shift in the mutational spectrum of crRNA:tracrRNA:Cas9 RNP-treated cells (Figure 1E). While CCM3 mRNA expression was not reduced, CCM3 protein expression was completely abolished ( Figure 1F,G). This observation is consistent with a previous CRISPR/Cas9 study targeting the CIITA locus in human ECs. Despite high genome editing efficiencies, no reduced CIITA transcript levels were found. 26 Since offtarget effects are a concern in CRISPR/Cas9 genome editing experiments, we screened potential off-target loci but found no CRISPR/ Cas9-induced variants (Table S1). In addition, we tested a second crRNA with a target sequence located in exon 6 which is shared by all CCM3 transcript variants and encodes for a part of the FAT-homology domain of CCM3. Increasing CCM3 mutant allele frequencies were also found in this second approach. From day 12 to 18 after crRNA:tracrRNA:Cas9 RNP transfection, indel frequencies had more than doubled from 17% (range: 8%-24%) to 38% (range: 27%-48%).
To exclude cell-type-specific effects, we also used primary HUVECs and hCMEC/D3 cells. Consistent with our observations in CI-huVECs, the relative number of CCM3 indel alleles increased in both cell types after crRNA:tracrRNA:Cas9 RNP-mediated genome editing ( Figure S2).

| Increased clonogenicity of CCM3 -/-CI-huVECs
To support the hypothesis that CCM3-deficient ECs have a survival benefit, 58 CI-huVEC clones were established by limiting dilution within three weeks. These were derived from two crRNA:tracrRNA:Cas9 RNP-treated cultures which had CCM3 mutant allele frequencies of 64% and 77%, respectively. Amplicon sequencing demonstrated compound heterozygosity for two distinct loss-of-function variants or homozygosity for a truncating CCM3 mutation in 39 clones (67%) (Figure 2A). Of note, no cells with CCM3 wild-type alleles were found. All remaining 19 clones (33%) were compound heterozygous for a CCM3 loss-of-function mutation and an in-frame variant. Again, CCM3 mRNA levels were not reduced in CCM3 -/-CI-huVECs with biallelic loss-of-function variants whereas CCM3 protein expression was completely abolished ( Figure 2B,C). Only minor compensatory changes in CCM1 and CCM2 mRNA expression were observed ( Figure S3). CCM3-deficient CI-huVECs maintained their endothelial identity with strong expression of platelet endothelial cell adhesion molecule-1 (PECAM-1/CD31) but low levels of smooth muscle protein 22-alpha (SM22α) ( Figure 2D).
The discrepancy between the observed and the expected number of CCM3 wild-type clones indicated a cell protective effect of CCM3 long-term deficiency. In line with this observation, CCM3 -/-CI-huVECs demonstrated a significantly reduced activity of caspase 3 which is an indicator of apoptosis under standard culture conditions and also upon staurosporine treatment ( Figure 2E). Moreover, senescence-associated β-galactosidase activity was slightly reduced in CCM3 -/-CI-huVECs under serum starvation ( Figure 2F). Conversely, proliferation of serum-starved CCM3 -/-CI-huVECs was increased ( Figure 2G).

CI-huVECs
Little is known about the long-term effects of CCM3 deficiency in human ECs since most studies had focused on its acute inactivation so far. 11,16,33 When we analysed the angiogenic properties of CCM3 -/-CI-huVECs, we observed lower cell migration rates and significantly reduced endothelial sprouting although the expression of VEGFR2 and its phosphorylation after VEGF-A treatment were upregulated twofold ( Figure 3F,G and Figure S3). The elongation of endothelial sprouts was already reduced under basal culture conditions and did neither respond to VEGF-A nor FGF-2 treatment. Even excessive stimulation with 250 ng/mL VEGF-A or FGF-2 did not restore normal sprouting (data not shown). The effect of CCM3 long-term deficiency on sprout number was less pronounced (Figure 3G). The reduced angiogenic properties were not restricted to clonal CCM3 -/-CI-huVECs but were also observed in mixed crRNA: tracrRNA:Cas9 RNP-treated CI-huVECs that had not been cloned by limiting dilution. Upon genome editing of both CCM3 target loci in CI-huVECs [crRNA (e3) + (e6)], significantly impaired sprout F I G U R E 1 Efficient CCM3 gene disruption in CI-huVECs. A, The exon-intron structure of the CCM3 gene and its RefSeq transcripts are schematically depicted. CRISPR/Cas9 target sequences are highlighted in green and red. e = exon. B, Twelve days after crRNA(e3):tracrRNA: Cas9 transfection, T7EI assays indicated estimated indel rates of 6%-45% in three independent replicates (I-III). C, The mutational spectrum identified in a representative replicate 12 d after crRNA(e3):tracrRNA:Cas9 transfection is shown as sequence alignment. The PAM sequence is highlighted in green and nucleotide changes are marked in red. WT = reference allele. An increase of the total indel frequency (D) and a shift of the mutational spectrum (E) were observed after transfection. Mutations with variant frequencies ≤2% are summarized as "others". Digital PCR revealed no difference in CCM3 transcript expression (F) whereas no CCM3 protein was detectable 20-23 d after crRNA(e3):tracrRNA: Cas9 transfection. Western blot results of three independent crRNA(e3):tracrRNA:Cas9-treated cell pools are given in the lower panel (G). Data are presented as mean and SD (n = 3). ctrl = control. Two-tailed, Student's t test was used for statistical analysis: **P < 0.01 SCHWEFEL ET AL.
| 1775 elongation and migration were found ( Figure 4A,C). While tube formation after 6 hours was unaffected (data not shown), the stability of tube-like structures after 18 hours was significantly reduced in CCM3-deficient CI-huVECs ( Figure 4B). This instability could not be rescued by short-term CCM3 re-expression in CCM3 -/-CI-huVECs ( Figure S6). Of note, forced CCM3 expression in wild-type CI-huVECs also had an anti-angiogenic effect when compared to cells transduced with the recombinant GFP control adenovirus ( Figure S6). This result was not completely unexpected since it is well known that overexpression and knockout of genes encoding for multiprotein complex components can cause the same phenotype due to a negative impact on stoichiometry or interaction of the other binding  3-4). Student's t test, one sample t test, linear mixed models (RT-DC) or two-way ANOVA with Šidák's multiple comparisons test were used for statistical analyses: *P < 0.05; **P < 0.01; ****P < 0.0001 partners. 34 Furthermore, adenoviral transduction alone proved to be pro-angiogenic, as it has been described in a previous report. 35 Impaired sprouting and migration were only observed upon longterm CCM3 inactivation. No such endothelial dysfunctions were found in CI-huVECs from the same passage whose CCM3 expression had been transiently silenced by siRNA transfection ( Figure S5). The chronic effects of CCM3 inactivation on the angiogenic phenotype of CI-huVECs were also verified in primary HUVECs and hCMEC/D3 cells. Endothelial sprouting was significantly reduced in HUVECs that had been transduced with a lentiviral CRISPR/Cas9 vector and also F I G U R E 4 Reduced angiogenic response of CI-huVECs after CCM3 gene disruption. A, crRNA:tracrRNA:Cas9 RNP-treated CI-huVECs demonstrated reduced sprout lengths under basal culture conditions and stimulation with 25 ng/mL VEGF-A or FGF-2. Scale bar ≙ 200 µm. Decreased number of meshes in tube formation assay (B) and migration rates (C) were also observed. Scale bars ≙ 1 mm. nc crRNA = nontargeting control crRNA, ctrl = untreated control. Data are presented as mean and SD (n = 3). One-way or Two-way ANOVA with Dunnett's multiple comparisons test and One-sample t test were used for statistical analysis: *P < 0.05; ****P < 0.0001 in hCMEC/D3 cells upon crRNA:tracrRNA:Cas9 RNP-induced CCM3 gene disruption ( Figure S7 and S8).

| Altered expression pattern of ageingassociated miRNAs in CCM3 −/− ECs
MicroRNAs (miRNAs) have become promising therapeutic targets in cancer, cardiovascular and infectious research. 36 Figure 5A). Since not only the mature miRNAs but in most cases also their corresponding precursor miRNAs (hsa-mir-493, hsamir-335, hsa-mir-216a, hsa-mir-217) were found to be regulated in the same direction, we considered miRNA biogenesis rather than miRNA degradation or turn-over to be controlled by CCM3.

Prediction of miRNA-target interactions and pathway analysis
with the miRNet algorithm 37 demonstrated an enrichment of various gene ontology gene sets. Notably, "vasculature development" (P = 0.006) and "ageing" (P = 0.007) were among the most significantly enriched gene ontology biological process terms. For example, the miR-217 level that has previously been reported as upregulated in senescent ECs 38 was found to be reduced by more than 95% in CCM3 -/-CI-huVECs when compared to wild-type cells (129 vs.

RPM)
. Since ageing appears to be a major biological process influenced by CCM3 deficiency, we analysed the expression level of the identified miRNAs in high-passage, CRISPR/Cas9 RNP-treated and wild-type HUVECs and found miR-139-5p and miR-139-3p to be significantly upregulated in CCM3-deficient HUVECs ( Figure 5C and Figure S9).

| miR-139-5p inhibition is not a treatment option for CCM3 deficiency
The observation that miR-139-5p was also upregulated more than twofold in CI-huVECs upon acute CCM3 inactivation prompted us to address its function in CCM3-deficient ECs in more detail ( Figure 5C).
Since the proangiogenic CXC chemokine receptor 4 (CXCR4) gene is a validated target of miR-139-5p, 39 we first analysed its expression in CI-huVECs upon long-term CCM3 inactivation. As expected, CXCR4 transcript levels were significantly reduced in CCM3 -/-CI-huVECs. Because re-expression of wild-type CCM3 restored CXCR4 levels, a regulatory CCM3/CXCR4 axis was assumed ( Figure 5D). In order to further delineate the role of miR-139-5p in this axis, we silenced its signalling by transfection of a specific miRNA inhibitor ( Figure 5E). While negative regulation of CXCR4 by miR-139-5p was intact in wild-type CI-huVECs, it was compensated in CCM3 -/-ECs ( Figure 5F). In agreement with this observation, neither apoptosis induction nor sprouting, endothelial network formation or migration could be restored by miR-139-5p inhibition ( Figure 5G

| DISCUSSION
The lack of approved CCM therapies highlights that our understanding of the signalling cascades by which CCM3 controls endothelial quiescence and supports blood-brain barrier (BBB) integrity is still incomplete. For the first time, we have therefore used the CRISPR/ Cas9 technology to generate an in vitro knockout model which mimics complete CCM3 inactivation by a second somatic mutation in heterozygous mutation carriers. 1,2 Using the non-viral and selectionfree CRISPR/Cas9 RNP delivery approach, we observed CCM3 indel frequencies in human ECs that were significantly higher than those described for other target genes so far. 28 Thus, we have generated a model which enabled us to address the molecular and functional effects of chronic CCM3 inactivation in human ECs. Our results illustrate that crRNA:tracrRNA:Cas9 RNPs, which minimize the risk of off-target effects, can be used as a powerful and straightforward tool for genome editing in ECs.
With NGS-based tracking of CCM3 mutant allele frequencies after genome editing and use of a limiting dilution cloning assay, we have clearly shown that CCM3 -/-ECs have a clonal survival advantage. In particular, activation of the caspase 3 apoptotic cascade was significantly impaired in CCM3 -/-CI-huVEC clones that harbour biallelic loss-of-function variants. The effects of CCM3 on cell death can be variable in different cell types and micro-environmental conditions but our observations emphasize that CCM3 modulates apoptosis not only in serum-starved or staurosporine-treated cells but also in non-stressed ECs. [40][41][42][43][44][45] Of note, wild-type CI-huVECs displayed signs of replicative senescence and growth arrest upon long-term culture. In contrast, CCM3 -/-CI-huVEC clones proliferated well to high passages and had slightly lower SA-β-gal activities under serum starvation which is in agreement with observations from CCM3-silenced primary human coronary artery ECs after replicative stress. 46 Taken together, these results suggest that clonal expansion of mutant ECs contributes to CCM pathobiology. 47 Since CCM3 facilitates activation of pro-apoptotic cascades in ECs, it is reasonable to conclude that its chronic inactivation counteracts the removal of dysfunctional CCM3 -/-ECs. Consequently, not SCHWEFEL ET AL. to promote vascular permeability. 48 As a logical consequence, the ability of CCM3 -/-ECs to form organized spheroids in 3D culture was found to be markedly impaired.
We also demonstrated that chronic CCM3 inactivation compromises the angiogenic properties of ECs whereas its acute silencing does not. Notably, increased migration and sprouting, 11,16,33 but also impaired endothelial network formation of CCM3-silenced ECs have been reported. [49][50][51] Upon chronic CCM3 ablation in human ECs, we observed dramatically reduced endothelial sprouting and migration despite strong activation of VEGFR2 signalling. However, the impact of this signalling cascade on CCM biology remains controversial.
CCM3 inactivation has been shown to block VEGF signalling by destabilization of VEGFR2 in vitro and in vivo. 50 On the contrary, increased VEGFR2 activation and mRNA expression have been observed in human pulmonary artery ECs upon CCM1 inactivation and in HUVECs upon CCM3 silencing, respectively. 11,52 The discrepancy between the activation status of CCM3 -/-CI-huVECs and their F I G U R E 5 Deregulation of miR-139-5p and CXCR4 in CCM3 -/-CI-huVECs. A, Volcano plot of up-or downregulated mature miRNAs in CCM3 -/-CI-huVECs. The log 2 fold change (x-axis) is plotted against the non-adjusted P-value (y-axis). miRNAs with FDR-adjusted q-values <5% are marked in red. B, qPCR validation of the 3p and/or 5p strands of deregulated miRNAs found in CCM3 -/-CI-huVECs. Ctrl = wild-type CI-huVECs. C, Relative miR-139-5p quantification in HUVECs (passage 18) after crRNA(e3):tracrRNA:Cas9 transfection and in CI-huVEC following siRNA-mediated CCM3 inactivation. D, CXCR4 expression was significantly reduced in CCM3 -/-CI-huVECs and could be restored by CCM3 reexpression. E, Significant downregulation of miR-139-5p expression upon miR-inhibitor transfection. F, Relative CXCR4 mRNA expression after miR-139-5p inhibition in wild-type (left) and CCM3 -/-CI-huVECs (right). G, Sprouting was unaffected by acute miR-139-5p downregulation. Scale bar ≙ 200 µm. Data are presented as mean and SD (n = 3-4). Two-way ANOVA with Šidák's multiple comparisons test and Student's t test were used for statistical analysis: *P < 0.05; **P < 0.01; ****P < 0.0001 in vitro angiogenic behaviour indicates that chronic CCM3 inactivation triggers compensatory mechanisms. This hypothesis is in line with the observation that different pathways become sequentially activated after Ccm3 ablation in ECs of Ccm3 ECKO mice. 53 miRNAs can orchestrate complex endothelial networks and we hypothesized that these small non-coding RNAs might be part of the adaptive response of ECs to chronic CCM3 depletion. We therefore profiled the miRNA expression pattern of CCM3 -/-CI-huVECs and demonstrated for the first time that CCM3 regulates the expression of miRNAs that are associated with ageing and vascular development. miR-216a and miR-217 have been reported to be upregulated during endothelial ageing. Notably, forced expression of miR-217 and miR-216a induces premature endothelial senescence. 38,54 Their strong downregulation in CCM3 -/-CI-huVECs fits well to the higher replicative potential of CCM3-deficient ECs.
Another deregulated miRNA, miR-139-5p, which was significantly upregulated upon acute and chronic CCM3 inactivation in CI-huVECs is known to control CXCR4 gene expression on a posttranscriptional level. 39 CXCR4 is a tip cell-enriched receptor whose activation promotes endothelial migration and filopodia development. 39,55 It is also a key component of Notch-controlled sprouting angiogenesis and needs to be tightly regulated for proper vascular maturation. 56,57 Notably, its expression is not only reduced in CCM3 -/-CI-huVECs but also in brain microvascular ECs of Kri-t1 ECKO mice. 58 Rescue of CXCR4 levels by adenoviral CCM3 re-expression in CCM3 -/-CI-huVECs indicates that CCM3 positively regulates CXCR4 expression. Our study therefore represents an independent experimental approach suggesting that CXCR4 downregulation might be a common consequence of CCM1 and CCM3 inactivation. However, this regulatory axis is only partially controlled by miR-139-5p. While negative regulation of CXCR4 by miR-139-5p was validated in wild-type CI-huVECs, it was found to be compensated after chronic CCM3 inactivation. Consequently, neither CXCR4 expression nor the dysfunctional angiogenic properties of CCM3 -/-CI-huVECs could be rescued by miR-139-5p inhibition alone. Taken together, these results support the hypothesis that miR-139-5p may modulate phenotypic variability 59 but is not a promising therapeutic target within CCM pathogenesis.
In conclusion, our genome editing study demonstrates that longterm CCM3 inactivation rather than its acute knockdown induces a clonogenic survival benefit and disturbances of endothelial network formation and angiogenic properties. This novel in vitro model may be useful to identify new therapeutic targets to block clonal expansion of mutant ECs upon chronic CCM3 depletion.