Endothelial cell clonal expansion in the development of cerebral cavernous malformations

Cerebral cavernous malformation (CCM) is a neurovascular familial or sporadic disease that is characterised by capillary-venous cavernomas, and is due to loss-of-function mutations to any one of three CCM genes. Familial CCM follows a two-hit mechanism similar to that of tumour suppressor genes, while in sporadic cavernomas only a small fraction of endothelial cells shows mutated CCM genes. We reported that in mouse models and in human patients, endothelial cells lining the lesions have different features from the surrounding endothelium, as they express mesenchymal/stem-cell markers. Here we show that cavernomas originate from clonal expansion of few Ccm3-null endothelial cells that express mesenchymal/stem-cell markers. These cells then attract surrounding wild-type endothelial cells, inducing them to express mesenchymal/stem-cell markers and to contribute to cavernoma growth. These characteristics of Ccm3-null cells are reminiscent of the tumour-initiating cells that are responsible for tumour growth. Our data support the concept that CCM has benign tumour characteristics.

C erebral cavernous malformation (CCM) is a vascular disease that is characterised by capillary-venous cavernomas. These malformations are almost exclusively located in the central nervous system and they can cause micro bleeds that can lead to epileptic seizures and cerebral haemorrhage [1][2][3][4][5] . CCM occurs in familial and sporadic forms. Familial cases follow autosomal-dominant inheritance due to loss-of-function mutations in one of the three genes known as CCM1/KRIT1, CCM2/malcavernin or CCM3/PDCD10. The inherited forms of CCM develop multiple vascular lesions that tend to increase in number and size with age. In contrast, the sporadic form of CCM is characterised by only a single lesion. This difference in the lesion burden between familial and sporadic CCM suggests that CCM pathogenesis might follow a two-hit, biallelic molecular mechanism that is similar to that seen for tumour-suppressor genes. Moreover, a remarkable feature of familial CCM is that, although all of the endothelial cells have a heterozygote loss-offunction mutation to one of the three CCM genes, the malformations are only found in a few localised regions of the brain microcirculation. Furthermore, it has been shown that, for human sporadic cavernomas, only a small fraction of endothelial cells have a null mutation for the CCM genes [6][7][8][9] . Considering that the double hit is a rare event, this suggests that a small number of mutated endothelial cells appear to be enough to trigger the malformations.
In our previous studies, we reported that in mouse models of CCM and in human patients the endothelial cells lining cavernomas have different features than the surrounding endothelial cells of the same vessel. Specifically, the endothelial cells in the lesions show a mixed phenotype that combines both endothelial and mesenchymal features in a way similar to endothelial cells that are undergoing endothelial-to-mesenchymal transition (EndMT). Most importantly, these cells also express a relatively large set of stem cell markers (e.g., Cd44, Id1, Slug, Klf4, Sca1) [10][11][12][13][14] . These characteristics are reminiscent of the so-called tumour-initiating cells or tumour stem cells that are responsible for tumour growth and metastatic differentiation [15][16][17] .
We show here that cavernomas originate from clonal expansion of a small number of Ccm3 −/− endothelial cells that express EndMT and progenitor markers. These cavernoma-initiating cells can attract the surrounding wild-type endothelial cells and induce their mesenchymal transition, coinciding with a strong and timedependent increase in the size of the cavernomas. Our findings support the concept that CCMs have benign tumour characteristics and are consistent with the observations of mosaicism in human cavernomas. Furthermore, data indicate that a minimal proportion of Ccm3-null endothelial cells can induce large-sized malformations, as in the human disease 8,9 . This concept is also in agreement with the fact that Ccm3 is a tumour suppressor 18,19 and its deletion may be correlated to benign brain tumours 20 .

Results
Cavernomas have clonal origin. To follow the clonal expansion of endothelial cells, we took advantage of the Cdh5(PAC)-Cre-ER T2 /R26R-Confetti mouse that carries the stochastic and multicolour reporter Brainbow2.1 in the R26 locus (R26R-Confetti 21,22 ) and the inducible Cre-ER T2 recombinase that is driven by the endothelial-specific promoter of VE-cadherin (Cdh5(PAC)-Cre-ER T2 ) 23 . After a single injection of tamoxifen, this Cre-mediated recombination promotes endothelial cells to randomly express any one of the four fluorescent proteins to form a mosaic (Fig. 1a, b). To study whether the CCM lesions have a clonal origin, the Cdh5(PAC)-Cre-ER T2 /R26R-Confetti mice were crossed with Ccm3 f/f mice (Cdh5(PAC)-Cre-ER T2 /Ccm3 f/f /R26R-Confetti), which develop CCM lesions upon tamoxifen injection that resemble the human pathology 24 .
Injection of tamoxifen on postnatal day 1 (P1) induced deletion of the Ccm3 gene and expression of one of the four fluorescent proteins in an endothelium-specific manner. By P8, the retina showed vascular malformation at the front, with large areas of clonal expansion (Fig. 1a). In the cerebellum, where most of the cavernomas were formed in this model (Fig. 1b, f), the majority of the small lesions appeared to be composed of cells of the same colour, which thus suggested their clonal origin. Larger lesions had a more complex composition, with clonal areas surrounded by regions with endothelial cells of mixed colours (Fig. 1b-f and Supplementary . This suggested that, after the first clonal growth, the adjacent lesions might fuse or that surrounding cells might be recruited into the lesion. The clonal expansion presupposes an increased cell proliferation of Ccm3 −/− cells within the lesions, as we have previously shown in vivo 24 . Ccm3 is known to have a pivotal role in regulating cell survival and cell death, and anti-apoptotic [25][26][27] as well as pro-apoptotic [28][29][30][31] functions have been reported in different cell types. Nevertheless, whether the increase in cell proliferation of endothelial cells lining the cavernomas is directly dependent on loss of Ccm3 is not completely understood. Here we show that the loss of Ccm3 is sufficient to increase the proliferation rate of endothelial cells and to drive the entrance into the S-phase, while the re-expression of the gene decreased cell proliferation to wild-type level (see Supplementary Figs. 1, 2, 13 and 14 for more details).
In parallel, we have tested the activated caspase 3 protein levels in both Ccm3 +/+ and Ccm3 −/− conditions and we have not seen any significant difference, arguing that the deletion of Ccm3 could not be sufficient to inhibit the endothelial cell apoptosis under physiological conditions.
Large cavernomas are mosaics. This 'fast progression' acute mouse model of Ccm3 −/− CCM is lethal in around 12 days, and it does not allow detailed analysis of the temporal steps of cavernoma formation. To overcome this limitation, we generated a slow progression model (i.e. chronic) of Ccm3 −/− CCM by giving a low dose of tamoxifen to these Cdh5(PAC)-Cre-ER T2 /Ccm3 f/f mice ( Fig. 2 and Supplementary Movie 7), and we analysed them at different developmental stages (i.e. P8, P14 and P30). Under this condition, the mice reached adulthood, and as in the human CCM disease, only a small proportion of the cells (<15%) showed Ccm3 deletion (Supplementary Figs. 3a,11 and 12).
In the acute model, the lesions were preferentially located in the cerebellum and retina, while in the chronic model they were also in other brain regions, including the hippocampus and the olfactory bulb, similar to the human disease 32,33 (Fig. 2e).
While the number of lesions remained substantially constant over time, at P14 and P30 the mean area, and thus the total lesioned area, strongly increased ( Fig. 2f-h).
Taking all of these characteristics into account, the chronic model appears to be more suitable for studies on the progression of human cavernomas.
In the chronic model of the Cdh5(PAC)-Cre-ER T2 /Ccm3 f/f / R26R-Confetti mice, not all the endothelial cells lining the large cavernomas were Confetti positive; i.e. Ccm3 −/− . In contrast, most of the endothelial cells were Confetti negative, and therefore they had not undergone Ccm3 recombination ( The stochastic nature of the recombination in the Confetti system cannot exclude that, by chance, two adjacent cells express the same colour without deriving by the same cell. In order to assess the frequency of this event, and to better quantify the clonal expansion within the lesions, we performed a more quantitative analysis on cerebellum from Cdh5(PAC)-Cre-ER T2 / Ccm3 f/f /R26R-Confetti mice and their Ccm3 +/+ counterpart after injection of low tamoxifen at P2 (Fig. 3e-g, Supplementary  Fig. 4). We considered a clone any group of contiguous cells expressing the same colour within a vessel and counted the number of cells that composed the clone. In the vessels of Ccm3 +/+ mice, the average size of clones was 1.14 ± 0.016 cells, with 88% of single cells and no clone >4 cells. Likewise, in normal vessels of the Ccm3 −/− , the average size of clones was 1. 19  The quantification also showed that there is no different distribution of the four fluorophores between lesions and normal vessels and that the CFP is slightly under represented among the four fluorescent proteins (Fig. 3e).
While this paper was under revision, others reported on clonal expansion of endothelial cells under pathological conditions using different mathematical approaches 34,35 .
Fluorescence-activated cell sorting (FACS) analysis of endothelial cells isolated from the Cdh5(PAC)-Cre-ER T2 /Ccm3 f/f /R26R-Confetti mice treated with low tamoxifen showed that only 11.0 ± 7.4% endothelial cells underwent recombination of Ccm3 ( Supplementary Fig. 3a). In addition, quantitative real-time polymerase chain reaction (RT-qPCR) confirmed that endothelial cells positive for Confetti also showed recombination for the Ccm3 gene, while the Confetti-negative cells expressed the same level of Ccm3 transcript as the wild-type endothelial cells ( Supplementary Fig. 3b). Overall, these data support the concept that, after the initial clonal expansion of a few endothelial cells that have undergone recombination, further development of the cavernomas was promoted by recruiting Ccm3 +/+ cells that had not undergone recombination (i.e. wild-type endothelial cells).
In the acute model, a large proportion of the endothelial cells (>80%) had biallelic inactivation of one of the Ccm genes, but only the cells that lined the cavernomas showed EndMT markers 24,36 . Double immunostaining for endothelial (ERG1, COLLAGEN IV, PECAM1) and EndMT markers (i.e., FSP1, pSMAD3, ID1, FN1, SCA1, αSMA, KLF4) in the chronic model showed that at early stage (P8) small lesions are composed mostly by endothelial cells that expressed higher levels of EndMT markers if compared to cells of surrounding normal vessels, while at later stages (P14), in larger cavernomas only a subset of endothelial cells underwent EndMT (Fig. 4).
At P8, almost 100% of the endothelial cells that lined the cavernomas were KLF4-positive (i.e. bona fide endothelial cells undergoing EndMT), which then declined during the increase in the size of the cavernomas to 70% KLF4-positive cells at P14 and 40% KLF4-positive cells at P30 (Fig. 4b).
To further confirm this data, we crossed Cdh5(PAC)-Cre-ER T2 / Ccm3 f/f mice with a different reporter mouse, for R26-EYFP, which is widely used for cell tracking 37 . With this reporter, mosaicism of the recombination of the endothelial cells lining the lesions was confirmed, with a progressive increase of EYFP-negative, i.e. Ccm3 +/+ , endothelial cells being recruited into the large cavernomas (Fig. 5a). Moreover, coimmunostaining demonstrated that the cells lining the cavernomas and undergoing EndMT transition had an endothelial origin ( Fig. 5b-f), as they also expressed the EYFP. The wild-type endothelial cells that were recruited into the lesions also overexpressed EndMT markers, including KLF4, ID1, pSMAD3, SCA1 and αSMA ( Fig. 5b-f).
To investigate the mechanism(s) that regulates the cross-talk between these Ccm3 −/− and Ccm3 +/+ cells in greater depth, we developed a set of in vitro assays. First, we co-cultured Ccm3 +/+ and Ccm3 −/− immortalised endothelial cells. When cultured alone, both the Ccm3 +/+ and Ccm3 −/− cells formed a typical endothelial monolayer ( Supplementary Fig. 5c, d). However, when the Ccm3 −/− cells were mixed with the Ccm3 +/+ cells they progressively formed spheroids while the Ccm3 +/+ endothelial cells still grew in a monolayer (Fig. 7a, b; Supplementary Fig. 6). These spheroids were composed of 26.8 ± 3.19% Ccm3 +/+ cells. In the areas of close contact, the Ccm3 +/+ cells became integrated into the Ccm3 −/− spheroids. Time-lapse movies confirmed that the wild-type (i.e. Ccm3 +/+ ) cells were actively recruited into the Ccm3 −/− spheroids, rather than just being entrapped through physical interactions (Supplementary Movie 9). Moreover, when the established spheroids were detached and placed onto a new Ccm3 +/+ cell monolayer, these new wild-type cells were actively recruited and entered the spheroids (Fig. 7c-e; Supplementary Fig. 7). Time-lapse movies of wound-healing experiments showed the different migration behaviours of these Ccm3 +/+ and Ccm3 −/− cells ( Supplementary Fig. 8a, b, and Supplementary Movie 10). These also showed that, as soon as the Ccm3 −/− cells induce EndMT in Ccm3 +/+ endothelial cells. As well as being recruited by the Ccm3 −/− cells, the Ccm3 +/+ cells were stimulated to undergo EndMT. The Ccm3 +/+ cells that were isolated from these spheroids expressed higher levels of the EndMT markers Fsp1, Cd44 and Fn1 and decreased levels of the endothelial-specific Claudin5, as compared to non-recruited Ccm3 +/+ cells that remained as monolayer at the bottom (Fig. 7g). To exclude that the upregulation of EndMT markers was merely due to the fact that they grew in three-dimensional  (3D), we cultured Ccm3 +/+ alone as spheroids on a methylcellulose matrix and compared their gene expression to the same cells cultured as monolayer ( Supplementary Fig. 9). After 7 days, the cells cultured as spheroids showed no increase of Fn1 and Claudin5 and a slight increase of Cd44, which was not comparable to the one induced by the co-culture with Ccm3 −/− cells.
Only Fsp1 showed a significant upregulation under 3D conditions. Therefore, we can assess that the molecular modifications of recruited Ccm3 +/+ cells are mostly induced by the presence of Ccm3 −/− cells rather than by the 3D culture itself. Moreover, the induction of EndMT by the co-culture was reversible, as once the Ccm3 +/+ cells isolated from the spheroids were put back in culture alone, they reverted to the expression level of the Ccm3 +/+ cells cultured as monolayer ( Supplementary Fig. 9). We also asked whether the EndMT phenotype induced by the deletion of Ccm3 could be rescued by re-expression of the gene. To this aim, we infected Ccm3 −/− cells with a lentiviral vector that expresses the human CCM3 gene fused to EGFP (Supplementary Fig. 1), to generate a reconstituted cell line (Ccm3 −/− -hCCM3-EGFP). Interestingly, the Ccm3 −/− -hCCM3-EGFP cells reverted to wild-type phenotype in terms of proliferation rate (Supplementary Fig. 1), morphology, adherens junctions and expression of EndMT markers ( Supplementary Fig. 1).
Furthermore, conditioned medium from these Ccm3 −/− cells increased the migration rates of Ccm3 +/+ cells (Fig. 7f) and induced them to upregulate their EndMT markers (Fig. 7h), compared to the negative control of conditioned medium from Ccm3 +/+ cells.
Taken together, these data support the concept that Ccm3 −/− cells can attract Ccm3 +/+ cells and induce them to undergo EndMT, as it has also been reported in vivo. The data on the effects of this conditioned medium here (Fig. 7f) suggested that secreted soluble factors are involved in the Ccm3 −/− -cell-mediated recruitment of the Ccm3 +/+ endothelial cells in cooperation with direct interaction of Ccm3 −/− and Ccm3 +/+ cells.
We reported previously that Ccm1 −/− cells produce high amounts of BMP6 and show significant phosphorylation of SMAD2/3 and SMAD1/5 36,38 also under resting conditions. Furthermore, stimulation of wild-type endothelial cells with BMP6 induced upregulation of their EndMT markers, which was inhibited by the BMP pathway general inhibitor DMH1 36,39 . Last but not least, BMP6 is a strong inducer of endothelial chemotaxis 40,41 .
Overall, these data are suggestive of a role of BMP6, possibly in combination with other inflammatory cytokines, in the attraction of wild-type endothelial cells in the cavernomas by Ccm3 −/− cells.

Endothelial progenitors are involved in cavernoma formation.
We have shown that cavernomas originate from clonal expansion of Ccm3 −/− endothelial cells that co-express endothelial and mesenchymal/stem cell markers and trigger the recruitment of Ccm3 +/+ endothelial cells. Recently, vessel-resident endothelial progenitors have been identified and characterised on the basis of the expression of different markers 10,12,42,43 , which included Cd157/Cd200 44 , Procr 45 and Peg3/PW1 46 . These progenitors coexpressed endothelial and mesenchymal/stem cell markers and underwent clonal expansion to generate new blood vessels 34 .
Taking into consideration the strong similarities with other tissue stem cells, it is tempting to speculate that the early steps of cavernoma formation are due to the clonal expansion of resident endothelial progenitors. This is supported by the endothelial cells in the early lesions, which express most of the stem/progenitor markers, including Id1, Sca1, Cd44 and Klf4, and more interestingly, also the specific endothelial progenitor marker Protein C Receptor (Procr; Fig. 8a, b). Furthermore, it has been shown that the number of endothelial progenitors declines with the age of these mice, as higher numbers in neonates and almost undetectable levels in adults, 46 which is consistent with the decline in the number of CCM malformations that form when Ccm3 is deleted at later time points after birth 24,47 . Here we confirmed by FACS analysis that Procr is expressed in a small population, 4.4 ± 1.5%, of brain endothelial cells in wild-type mice at P2 (Fig. 8d), which is consistent with what is already reported in other tissues 45 . Moreover, RNAseq analysis made on endothelial cells from the wild-type brains 48 showed that the expression of Procr decreases over time after birth (Fig. 8e).
To further test this hypothesis, we asked whether inactivation of Ccm3 in progenitor cells would specifically induce their clonal expansion and early steps of CCM development. To this end, we crossed the Ccm3 f/f mice with the Procr CreERT2-IRES-tdTomato/+ knock-in mouse model, in which a CreERT2-IRES-tdTomato cassette is driven by the promoter of Procr 45,49 . The Procr CreERT2-IRES-tdTomato/+ /Ccm3 f/f mouse model allowed induction of the tamoxifen-dependent Ccm3 deletion specifically in Procr + endothelial progenitors.
After recombination, the mice developed lesions both in the retina and the cerebellum (Fig. 8c), in a way that recapitulates the formation of lesions in the chronic model. By crossing the Procr CreERT2-IRES-tdTomato/+ /Ccm3 f/f mouse model with the Fig. 2 The slow progression model of cerebral cavernous malformation (CCM) develops large lesions. A chronic model of CCM was generated by treating Cdh5(PAC)-Cre-ER T2 /Ccm3 f/f /R26R-Confetti mice with low-dose tamoxifen. a Scheme of treatment with tamoxifen at P2 and analysis at P8, P14 and P30. b Representative photographs of whole brains from chronic P8, P14 and P30 mice; scale bar: 100 μm. c Representative tiling of a cerebellum at P14 showing the distribution of lesions; upper panel shows a projection from a 1-mm-thick section; lower panels show three-dimensional reconstruction of corresponding regions. Lower left panel was rotated by 90°; vessels were stained for Podocalyxin; scale bars: 1000 μm lower magnification, 300 μm higher magnification. d Representative confocal image of P30 retina stained for Isolectin B4 (black vessels) showing large cavernomas at the front; scale bar: 500 μm lower magnification, 100 μm higher magnification. e Representative confocal images of P30 brain stained for Isolectin B4 (black vessels) showing cavernomas in the hippocampus and the olfactory bulb; 1000 μm lower magnification, 500 μm higher magnification. f, g Quantification of total number and mean area of lesions in chronic P8, P14 and P30 mice. Data are means ± SE; p < 0.001 among groups (one-way analysis of variance (ANOVA)); *p < 0.01 (Tukey's post hoc test). h Lesions were divided into four groups according to their size, as shown. Data are means ± SE. *p < 0.001 among groups (one-way ANOVA); p < 0.01 (Tukey's post hoc test); a total of 1698 lesions have been counted from 5 animals at P7, 4 animals at P14 and 9 at P30. Source data are provided as a Source Data file R26-EYFP reporter, we confirmed that the lesions were composed of EYFP + cells, thus originating from Procr + endothelial progenitors (Fig. 8f). Finally, the endothelial cells lining the cavernomas overexpressed EndMT markers if compared to the cells of surrounding normal vessels (Fig. 8g-j), thus resembling the pathogenesis already shown by the Cdh5 (PAC)-Cre-ER T2 /Ccm3 f/f model. These data thus support the idea that endothelial progenitors are responsible for the triggering of cavernoma formation upon Ccm3 deletion.

Discussion
Taken together, our data explain some of the crucial aspects of the development and evolution of CCM, while also opening novel therapeutic opportunities. Using a slowly evolving Ccm3 −/− model system, we have demonstrated that, upon induction of homozygote Ccm3 mutation, only a few of the endothelial cells undergo clonal expansion and attract Ccm3 +/+ cells, which, in turn, contribute to the growth of the malformations. This is in agreement with the human disease, where following a double-hit process 6-9 , a few heterozygote endothelial cells acquire homozygosity for the Ccm gene inactivation and trigger the formation of cavernomas. Consistently in both mice and humans, the endothelial cells lining the lesion appear as a mosaic, i.e. formed by both mutated and wild-type cells. Therefore, an important event in the evolution of the cavernomas appears to be the attraction of Ccm3 +/+ cells by Ccm3 −/− cells, which, in turn, strongly contribute to the growth of the malformation. An important question here is whether or not the occurrence of the malformations is a stochastic event, i.e. all endothelial cells have the potential to undergo clonal expansion and EndMT upon Ccm3 mutation in a random way. Alternatively, once mutated in Ccm3, only a subset of the cells might trigger the formation of lesions. This second possibility has been described in cancer, where tumour-initiating cells are responsible for metastatic dissemination and clinical relapse in a variety of cancers 43,50 . Tumour-initiating cells have similarities to normal tissue stem cells and undergo epithelial-to-mesenchymal transition. It is therefore attractive to hypothesise that, also in CCM, the cavernomas are triggered by a few mutated endothelial cells with characteristics and markers specific of endothelial-resident progenitors, such as Cd157 + , Procr + and PW1 + . As in tumours, these cells would then recruit wild-type cells, which will, in turn, contribute to the growth of the cavernomas. Therefore, with these cavernoma-initiating cells normally present along the vessels, we propose a model in which upon loss of Ccm3 they undergo clonal expansion to form malformations. In a further step, the mutated cells recruit wild-type cells, which, in turn, strongly contribute to cavernoma growth (Fig. 9). Taking advantage of the Procr CreERT2-IRES-tdTomato/+ /Ccm3 f/f mouse model, we have also shown that resident endothelial progenitors can trigger the formation of cavernomas upon deletion of Ccm3. We cannot completely exclude that these cavernoma-initiating cells are recruited systemically, as a bone marrow transplantation is not possible on new-born mice. However, the data presented here support the idea that cavernomas originate from Procr + progenitors, which have been described to reside in pre-existing vessels rather than being recruited 45 from circulating cells.
Broadly, our findings are suggestive of new possibilities for therapeutic intervention for patients with CCM, such as through inhibition of endothelial chemokines, to prevent the recruitment of wild-type endothelial cells or, more drastically, through identification and suppression of these cavernoma-initiating cells.

Methods
Antibodies and reagents. Tamoxifen was from Sigma-Aldrich, and Vectashield without and with DAPI were from Vector Laboratories. The following primary Immunohistochemistry and confocal microscopy. For the retinal vasculature, mouse eyes were fixed in 4% paraformaldehyde (PFA) for 2 h prior to dissection. Whole-mount retinas were incubated overnight with primary antibodies diluted in 10% donkey serum, 1% bovine serum albumin (BSA) and 0.5% Triton-X100 in phosphate-buffered saline (PBS) at 4°C. After washes with 0.01% Triton-X100 in PBS, the retinas were incubated with the fluorophore-conjugated antibodies (Invitrogen) for 4 h at room temperature, washed, post-fixed with 4% PFA and mounted with Vectashield.
Immediately after dissection, mouse brains were fixed in 4% PFA overnight and then included in low-temperature melting agarose and 100-μm-thick sections were cut (VT1200s vibratome; Leica, Wetzlar, Germany). Free-floating sections were incubating overnight with the primary antibodies diluted in 5% donkey serum, 2% BSA and 0.3% Triton-X100 in PBS at 4°C. After washes with 0.01% Triton-X100 in PBS, the sections were incubated with the fluorophore-conjugated antibodies (Invitrogen) for 4 h at room temperature, washed, post-fixed with 4% PFA and mounted with Vectashield.
Confocal microscopy was performed using a confocal microscope (TCS SP5, Leica), with the ImageJ software (NIH, New York, NY) used for image analysis.
Lesion quantification. For lesion quantification, 100-μm thick vibratome sections of brain were prepared and stained for endothelial markers as described above. For each brain, at least 10 sections were cut, and z-stacks were acquired with a ×10 objective. After maximal projection, the numbers and the areas of the lesions were manually measured by a blinded operator, using the ImageJ software.
Passive tissue clearing. Mice of strain and age as indicated in the text were anaesthetized by intraperitoneal injection of Avertin (20 mg/kg) and perfused with 4% PFA in PBS. The brains were carefully dissected out from the skull and postfixed overnight by immersion in 4% PFA at 4°C. The next day brains were washed in PBS and processed for sectioning.
Passive tissue clearing was carried out as previously described 51 , with some modifications. Briefly, vibratome coronal sections of cerebellum (1-mm thick) were incubated overnight at 4°C in hydrogel solution (4% Acrylamide, 0.25% VA-044 in PBS). Sections were then transferred in cold microwave reaction vials and hydrogel solution was replaced with fresh solution. Tubes were sealed using a cap with rubber septum. Before starting the polymerization, hydrogel solution was degassed by bubbling nitrogen through the septum for 2 min on ice. Samples were then placed at 37°C to activate polymerization. After 2 h, excess acrylamide was removed and sections were incubated overnight in clearing solution (8% sodium dodecylsulfate (SDS), 0.01% Sodium Azide in water, pH 8.5) at 37°C. The clearing solution was then replaced and passive clearing was continued for 72 h at 37°C. The cleared sections were then washed in PBS containing 0.1% Triton X100 and 0.01% Sodium Azide (washing buffer) for 48 h, with at least four changes of washing buffer. To enhance both clearing and washing efficacy, the buffer-tosection volume ratio was always >50.
Immunostaining was performed at room temperature as follows. Sections were incubated with the indicated antibody in washing buffer supplemented with 5% donkey serum for 72 h. Samples were then washed for 30 h and incubated, for 72 h, with the appropriate secondary antibody (Alexa 647 conjugated) diluted in washing buffer. Finally, samples were washed for 30 h.
Sections were then incubated and, subsequently, mounted in refractive index matching solution (88% w/v Hystodenz in water) for confocal imaging with a Leica SP8 microscope. Three-dimensional reconstructions were created using the Arivis Vision4D software and movies were assembled with Photoshop.
Brain endothelial cell isolation and FACS sorting. Endothelial cells were isolated from the brains using a dissociator (gentleMACS; 130-093-235; Miltenyi) and the Adult Brain Dissociation Kits (130-107-677; Miltenyi), following the manufacturer's protocol. After dissociation, the cells were re-suspended in PBS with 2% foetal bovine serum and 2 mM EDTA and incubated for 30 min with an APCconjugated anti-Cd31 antibody for FACS sorting. The sorting was performed on a MoFlo (Beckman Coulter Inc., Brea, CA, USA) cell sorter.
Indirect measure of cell proliferation: cell counting. The endothelial cells Ccm3 +/+ , Ccm3 −/− and Ccm3 −/− to which has been re-added the human Ccm3 gene were plated at 60,000/cm 2 and counted at different time points (24, 48, 72 h) to assess their proliferation. After having been harvested, the cells were stained with the Erythrosin B and counted at an optical microscope through the Burker's chamber. Fig. 8 Endothelial progenitors are involved in cavernomas formation. a In situ hybridization for Procr using the RNAScope assay with chronic Ccm3 −/− cerebral cavernous malformation (CCM) mouse brains. Red, Procr probe; grey, endothelial cells stained with a combination of anti-COLLAGEN IV and anti-ERG1 antibodies. Black and green panels show endothelial cells with high expression of Procr, blue panel shows cells with no expression of Procr. b Quantification of the percentage of endothelial cells that express Procr within the lesions; n = 10 lesions from 3 animals. c Procr CreERT2-IRES-tdTomato/ + /Ccm3 f/f mice received tamoxifen injection at P1 and were analysed at P30. Representative images of the cerebellum stained for PECAM1 and wholemount retina stained for Isolectin B4. d Fluorescence-activated cell sorting analysis on endothelial cells isolated from wild-type mice at P2. The plot shows the percentage of Procr + cells among the Cd45 − /Cd31 + endothelial cells. Data are means ± SE; n = 3 animals. e RNAseq analysis on endothelial cells isolated from the brain of wild-type mice at different ages, showing the reduction of Procr expression during development. Data are means ± SE from 3 animals each group. Source data are provided as a Source Data file. f Procr CreERT2-IRES-tdTomato/+ /Ccm3 f/f /R26-EYFP mice received tamoxifen injection at P1 and were analysed at P30. Representative images of whole-mount retina stained for Isolectin B4 and EYFP showing endothelial cells lining the lesion that underwent recombination. g-j Representative confocal images of brain sections double stained for PECAM1 (red) and the mesenchymal markers (green) KLF4 (g), SCA1 (h), FN1 (i) and αSMA (j). White arrows, cells lining the cavernoma; yellow arrows, cells in normal vessels. Scale bars, 500 μm (c and f main panels); 100 μm (a main panel, c and f magnifications), 50 μm (g-j main panels); 20 μm (a, g-j magnifications)

Recombined
Not recombined Cavernoma initiating cell Step 1 Clonal expansion of recombined cells Step 2 Enlargement of the lesion by attraction of non recombined cells Fig. 9 Model for the onset and growth of cavernomas. Cavernoma-initiating cells undergo clonal expansion in response to Ccm3 loss and start to form cavernomas. In the further step, recombined cells recruit normal cells, which in turn, contribute to cavernoma expansion FACS analysis: Ki67 and PI staining. The protocol is intended for 3 × 10 6 cells and adapted from Vignon et al. 52 . After having been harvested, the cells were washed in PBS 1× and spun down at 500 × g for 5 min. The pellet recovered was resuspended in PBS 1× and fixed by adding pure cold ethanol 100% (1 PBS 1×:3 EtOH 100%), dropwise while vortexing. The cells were left in fixative at least for 30 min on ice. Following two washes with PBS 1×, 1% FBS SA and 0,25% Triton X-100 (PFT), the cells were stained in 300 μl of anti-mouse/rat Ki67 (Invitrogen by Thermo Fisher Scientific, clone SolA15), diluted 1:100 in PFT, for 5 h at room temperature under constant rotation. A control tube was prepared with 300 μl of anti-rat 647 diluted 1:400 in PFT. After two washes with PFT, the cells were stained in 300 μl of PFT with anti-rat 647 antibody from Jackson, diluted 1:400, followed by incubation for 1 h at room temperature, in the dark, under constant rotation. Cells were washed twice with PBS 1×, centrifuged for 5 min at 500 × g and resuspended in 1 ml of the intercalating fluorescent dye PI (2.5 μg/ml) plus RNase A (250 μg/ml). After an overnight incubation at 4°C, FACS analysis was executed on the Attune NxT Life Technologies machine.
FACS analysis: BrdU and PI staining. The protocol is intended for 3 × 10 6 cells. While growing, the cells have been pulsed in MCDB131 medium containing 33 μM of BrdU. After having been harvested, the cells were washed in PBS 1× and spinned down at 500 × g for 5 min. The pellet recovered was resuspended in PBS 1× and fixed by adding pure cold ethanol 100% (1 PBS 1× : 3 EtOH 100%), dropwise while vortexing. The cells were left in fixative at least for 30 min on ice. Following one wash with PBS 1×/BSA 1%, the cells were resuspended in 1 ml of denaturing solution (2 N HCl) and incubated at room temperature for 25 min. After addition of 3 ml of 0.1 M Sodium Borate (Na 2 B 4 O 7 ) and incubation for 2 min at room temperature, the cells were spinned down at 500 × g for 5 min and washed twice with PBS 1×/ BSA 1%. The pellet was resuspended in 100 μl of anti-BrdU (purified mouse anti-BrdU, BD Biosciences, cod.347580), diluted 1:5 in PBS 1×/BSA 1%, for 2 h at room temperature, light protected, under constant rotation. A control tube was prepared with 300 μl of anti-mouse 488 diluted 1:400 in PBS 1×. After one wash with PBS 1×/BSA 1%, the cells were stained in 300 μl of PBS 1× with antimouse 488 antibody from Jackson, diluted 1:400, followed by incubation for 1 h at room temperature, in the dark, under constant rotation. Cells were then washed once with PBS 1×/BSA 1%, centrifuged for 5 min at 500 × g and resuspended in 1 ml of the intercalating fluorescent dye PI (2.5 μg/ml) plus RNase A (250 μg/ml). After an overnight incubation at 4°C, FACS analysis was executed on the Attune NxT Life Technologies machine.
For experiments with BMP6 and inhibitors, the cells were starved overnight, pre-incubated with either 1 μM K02288 53 or 1 μM DMH1 36,39 for 1 h before stimulation with 100 ng/ml BMP6. The BMP6 and inhibitors were replaced daily over 5 days, and then the cells were lysed and analysed. In the same way, for the migration experiments, the cells were starved, pre-incubated with the inhibitors and then stimulated with BMP6. Just before stimulation, a wound was made, and after 48 h the cells were fixed and analysed.
For the experiments with conditioned medium, the cells were starved overnight and then incubated with complete medium that had been conditioned for 48 h by either the Ccm3 +/+ or Ccm3 −/− cell cultures. The medium was replaced after 3 days, and the cells analysed after a further 3 days.
For the cell migration experiments, the cells were starved, pre-incubated with 1 μM DMH1 in medium for 1 h and then incubated with the conditioned medium from Ccm3 −/− or Ccm3 +/+ cell cultures either without or with 1 μM DMH1. Just before stimulation, a wound was made, and after 48 h the cells were fixed and analysed.
Cloning, recombinant lentiviruses and gene transduction. The plasmid encoding LifeAct-EGFP was a kind gift from Professor Wedlich-Soldner 54 . To construct LifeAct-mCherry, the EGFP tag was excised by double digestion with the AgeI and BsrgI restriction enzymes, and the mCherry was ligated using the same sites. LifeAct-EGFP and LifeAct-mCherry were subsequently sub-cloned into the lentiviral vector (pFUGW). The plasmids containing the LifeAct-mCherry or LifeAct-EGFP fusion fragments were double-digested with XbaI and NheI, and subsequently the LifeAct-EGFP or LifeAct-mCherry fusions were gel-purified and sub-cloned into the lentiviral vector pFUGW using XbaI cloning sites. Recombinant lentiviruses were produced by triple transfection of the packaging 293T cells with the lenti FUGW carrying the gene of interest, packaging vectors (pCMV delta R8.2) and the envelope vector (pMD2.G) carrying the VSV glycoprotein (purchased from Addgene, Cambridge, MA 02139, USA). The virus was concentrated by ultracentrifugation at 50,000 × g for 90 min at 4°C, and the pellet was stored at −80°C until use.
Generation of cell lines. To generate Ccm3 +/+ and Ccm3 −/− cells expressing either LifeAct-EGFP or LifeAct-mCherry, Ccm3 +/+ and Ccm3 −/− were transduced with the recombinant lentiviruses. In brief, the recombinant lentiviruses were re-suspended in serum-free MCDB-131 medium and added to the cells for 1 h at 37°C.
To generate Ccm3 −/− cells re-expressing mGFP-tagged CCM3, Ccm3 −/− were transduced with the recombinant lentivirus Lenti ORF clone mGFP-tagged PDCD10 (OriGene Technologies Inc). In brief, the recombinant lentiviruses were re-suspended in serum-free MCDB-131 medium and added to the cells for 1 h at 37°C. To increase the number of the cells, the cells were then passaged four times.
Ccm3 +/+ and Ccm3 −/− cell co-cultures. Ccm3 +/+ and Ccm3 −/− cells that expressed either LifeAct-EGFP or LifeAct-mCherry were mixed as single-cell suspensions (1:1) and plated and cultured under standard conditions. Usually the spheroids were analysed after 7 days. For cell isolation, the spheroids were removed from the monolayer following gentle shaking of the plate and collected and trypsinised. The single-cell suspensions obtained were immediately sorted by FACS for EGFP and mCherry expression, and then the cells were lysed and analysed for gene expression. For the experiments of spheroids detachment and transfer onto established monolayers, first the labelled Ccm3 −/− cells were co-cultured with non-labelled Ccm3 +/+ cells. After 7 days, the established spheroids were detached by shaking and placed on the top of an established monolayer of labelled Ccm3 +/+ cells.
Generation of endothelial spheroids. Endothelial cell spheroids were generated as previously described 36 .
Briefly, 3000 lung-derived immortalised endothelial cells were suspended in culture medium containing 0.20% (w/v) methyl cellulose and seeded in nonadherent round-bottom 96-well plates (Greiner). Under these conditions, all suspended cells contribute to the formation of one single rounded endothelial cell spheroid per well. These standardised spheroids were cultured for 6 days, then were harvested from 96-well plate, washed with PBS and processed for total RNA extraction.
Wound-healing assay with two cell types and live imaging. Ccm3 +/+ and Ccm3 −/− cells that expressed either LifeAct-EGFP or LifeAct-mCherry were seeded in each side of an ibidi silicon insert and grown till confluence. Time-lapse images were acquired either immediately after the insert removal or after cells were allowed to close the wound. Live-cell imaging was performed with Leica TCS SP8 confocal (Leica, Germany) under 37°C and 5% CO 2 . Images were acquired using either HC PL APO 1.3 oil ×40 or HCX PL APO ×63 1.4 oil objective lenses (Leica, Germany), respectively.
Western blotting. The cells were lysed in boiling modified Laemmli sample buffer (2% SDS, 20% glycerol, 125 mM Tris-HCl, pH 6.8). Equal amounts of protein were loaded onto gels and separated by SDS-polyacrylamide gel electrophoresis and then transferred to nitrocellulose membranes (Protran; Whatman). After blocking with 5% BSA and incubations with the primary and HRP-linked secondary antibodies, specific binding was quantified using a chemiluminescence system (GE Healthcare).
Statistical analysis. Statistical significance was determined using a one-way analysis of variance (ANOVA) tests followed by Tukey's post hoc analysis. For the growth curve experiment, two-way ANOVA test were followed by Tukey's post hoc analysis for multiple comparison within each time point.
In Fig. 7h and Supplementary Figs. 6e and 9, the statistical significance was determined using a non-paired two-tailed t test, assuming unequal variances.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.