Endothelial TLR4 and the microbiome drive cerebral cavernous malformations

Abstract

Cerebral cavernous malformations (CCMs) are a cause of stroke and seizure for which no effective medical therapies yet exist. CCMs arise from the loss of an adaptor complex that negatively regulates MEKK3-KLF2/4 signalling in brain endothelial cells, but upstream activators of this disease pathway have yet to be identified. Here we identify endothelial Toll-like receptor 4 (TLR4) and the gut microbiome as critical stimulants of CCM formation. Activation of TLR4 by Gram-negative bacteria or lipopolysaccharide accelerates CCM formation, and genetic or pharmacologic blockade of TLR4 signalling prevents CCM formation in mice. Polymorphisms that increase expression of the TLR4 gene or the gene encoding its co-receptor CD14 are associated with higher CCM lesion burden in humans. Germ-free mice are protected from CCM formation, and a single course of antibiotics permanently alters CCM susceptibility in mice. These studies identify unexpected roles for the microbiome and innate immune signalling in the pathogenesis of a cerebrovascular disease, as well as strategies for its treatment.

Extended Data Figure 1. CCM formation in resistant iECre;Ccm2^fl/fl animals is stimulated by abscess formation and LPS

a, Resistance to CCM formation is maintained in a C57BL/6J strain background. iECre;Ccm2^fl/fl animals were back-crossed 7 generations onto a C57BL/6J background and gene deletion induced at P1 with visual hindbrain assessment at P10. N=7. Scale bars, 1 mm. b, Retinal CCM formation is stimulated by gram negative bacterial infection. Retinas of P17 resistant iECre;Ccm2^fl/fl littermates are shown. The sample shown below is from the animal that developed the spontaneous gram negative abscess shown in Fig. 1c. Scale bars, 500 μm. c–d, Administration of LPS does not drive CCM formation in Cre-negative neonatal mice. LPS was administered intravenously to Ccm2^fl/fl and iECre;Ccm2^fl/fl littermates as shown in Figure 1g, and hindbrains assessed at P17 visually (c) and histologically (H&E staining, d). N≥3 per group. Scale bars, 1 mm (c) and 100 μm (d). e, LPS induces myosin light chain activation in CCM-deficient brain endothelial cells. Phospho-myosin light chain (pMLC) and PECAM staining of hindbrains from P5 LPS- or vehicle-injected resistant iECre;Ccm2^fl/fl littermates. Dotted lines trace the purkinje cell layer. N≥4 per group. Scale bars, 50 μm. f, Tlr4 expression does not differ between CCM susceptible and resistant animals. Tlr4 expression was measured using qPCR in cerebellar endothelial cells isolated from the indicated animals at P10. Error bars shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. n.s. indicates p>0.05.

Extended Data Figure 2. Analysis of immune cells in P6 and P11 Krit1^fl/fl and iECre;Krit1^fl/fl brains

a, Gating strategy for B cells, NK cells, γδ T cells, CD4 T cells, CD8 T cells, eosinophils, neutrophils and monocytes/macrophages from cerebrum and cerebellum is shown. Cellular surface markers used were as follows: Neutrophils (CD45⁺, CD11b⁺, Ly6-G⁺), Eosinophils (CD45⁺, CD11b⁺, CD11c⁻, Ly6G⁻, Siglec-F⁺, SSChi), Monocyte/Macrophage (CD45⁺, CD11b⁺, CD11c⁻, Ly6G⁻, Siglec-F⁻, SSClo), NK cells (CD45⁺, CD11b⁻, CD19⁻, NK1.1⁺), B cells (CD45⁺, CD11b⁻, NK1.1⁻, CD19⁺), γδ T cell (CD45⁺, CD11b⁻, NK1.1⁻, CD19⁻, CD3⁺, TCRγδ⁺), CD4 T cell (CD45⁺, CD11b⁻, NK1.1⁻, CD19⁻, CD3⁺, TCRγδ⁻, CD8⁻, CD4⁺), CD8 T cell (CD45⁺, CD11b⁻, NK1.1⁻, CD19⁻, CD3⁺, TCRγδ⁻, CD4⁻, CD8⁺). b, The number of B cells, NK cells, γδ T cells, CD4 T cells, CD8 T cells, eosinophils, neutrophils, and monocytes/macrophages isolated from P6 cerebrum (top) and cerebellum (bottom) is shown for susceptible Krit1^fl/fl and iECre;Krit1^fl/fl littermates. N≥6 per group. No significant differences were detected. c, The number of B cells, NK cells, γδ T cells, CD4 T cells, CD8 T cells, eosinophils, neutrophils, and monocytes/macrophages isolated from P11 cerebrum (top) and cerebellum (bottom) is shown for susceptible Krit1^fl/fl and iECre;Krit1^fl/fl littermates. N≥6 per group. d–e, Frequency of RORγt⁺ CD4 T cells isolated from P6 and P11 cerebellum. N≥6 per group. Error bars of all graphs shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. *indicates p<0.05. Note that there is significant immune cell presence in the cerebellum of susceptible iECre;Krit1^fl/fl animals at P11 but not at P6.

Extended Data Figure 3. Changes in the volume of CCM lesions are not accompanied by changes in total brain volume

The indicated total brain volumes were measured using microCT imaging. a–b, Brain volumes corresponding to the genetic rescue experiments shown in Fig. 2c–f, respectively. c, Brain volumes corresponding to the C-section/germ free fostering experiment shown in Fig. 4b–c. d, Brain volumes corresponding to the intergenerational antibiotic experiment shown in Fig. 6f–h and j. n.s. indicates p>0.05.

Extended Data Figure 4. Lineage tracing of the Cdh5(PAC)-Cre^ERT2 (iECre) transgene in neonatal mice

a–c, R26-LSL-RFP, R26-Cre^ERT2;R26-LSL-RFP, and Cdh5(PAC)-Cre^ERT2;R26-LSL-RFP neonates were induced with doses of tamoxifen on P1+2 (two total doses) and CD45⁺;RFP⁺ hematopoietic cell numbers in the spleen and peripheral blood assessed at P10. N≥5 per group. Error bars shown as s.e.m. and significance determined by one-way ANOVA with Holm-Sidak correction for multiple comparisons. ***indicates p<0.001; n.s. indicates p>0.05. Note that the number of labeled hematopoietic cells in Cdh5(PAC)-Cre^ERT2;R26-LSL-RFP animals is indistinguishable from R26-LSL-RFP negative control animals, while >90% of CD45+ cells were RFP⁺ in R26-Cre^ERT2;R26-LSL-RFP positive control animals. d, Anti-RFP and anti-PECAM immunostaining of P10 hindbrains from Krit1^fl/fl;R26-LSL-RFP negative control and iECre;Krit1^fl/fl;R26-LSL-RFP was performed to identify Cre⁺ descendants at the site of CCM formation. Note that all RFP⁺ cells in iECre;Krit1^fl/fl;R26-LSL-RFP animals are PECAM⁺, consistent with endothelial-specific Cre activity. Asterisk indicates CCM lesion. Results are representative of ≥ 3 per group. Scale bars, 100 μm.

Extended Data Figure 5. The Slco1c1(BAC)-Cre^ERT2 (iBrECre) transgene expresses selectively in brain endothelial cells and iBrECre-driven deletion of Krit1 confers CCM formation in neonatal mice

a, R26-LSL-RFP, Cdh5(PAC)-Cre^ERT2;R26-LSL-RFP and Slco1c1(BAC)-Cre^ERT2;R26-LSL-RFP neonates were induced with tamoxifen injection on P1+2 (two total doses). Immunostaining for RFP and PECAM was performed at P10 in the indicated tissues. Results are representative of at least three animals per group and three independent experiments. Scale bars, 100 μm. Note the presence of RFP⁺ PECAM⁺ cells in the brain, small intestine, cecum, colon and liver of Cdh5(PAC)-Cre^ERT2;R26-LSL-RFP animals, but only in the brain of Slco1c1(BAC)-Cre^ERT2;R26-LSL-RFP animals. b, Visual (top) and corresponding microCT (bottom) images of brains from susceptible iBrECre;Krit1^fl/+ and iBrECre;Krit1^fl/fl P10 animals. Arrow indicates CCM lesions in the cerebrum. Scale bars, 1 mm. c, H&E staining of cerebellum (hindbrain) from the indicated animals (left). H&E staining of cerebrum (forebrain) from the indicated animals (middle). KLF4 and PECAM immunostaining from the indicated animals (right). Scale bars, 50 μm. Asterisks denote CCM lesions. N≥5 per group.

Extended Data Figure 6. CCM formation can be stimulated by IL-1β or poly(I:C) treatment

a, Schematic of the experimental design in which littermates receive a retro-orbital injection of the indicated cytokine or TLR ligand at P5 and P10 prior to tissue harvest and analysis at P17. b–m, Visual images and volumetric quantitation of CCM lesions in the hindbrains of P17 iECre;Ccm2^fl/fl littermates injected with the indicated cytokines, TLR ligands, or vehicle control are shown. Error bars shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. *indicates p<0.05; n.s. indicates p>0.05. Scale bars, 1 mm.

Extended Data Figure 7. 16s rRNA sequencing results from susceptible and resistant Krit1^fl/fl and Ccm2^fl/fl dams

a, Heat map showing relative abundance of bacterial taxa (right) identified in susceptible (blue) and resistant (salmon) Krit1 (ccm1, purple) and Ccm2 (ccm2, green) animals (top). b, Box plots of bacterial taxa that demonstrated significant differential abundance in susceptible versus resistant animals and the relative abundance of those taxa. c, Boxplot of the Firmicutes [Ruminococcus] taxon that displayed significant differential abundance between Krit1 and Ccm2 genotypes. Note that the relative abundance of Bacteroidetes s24-7 is anywhere from 10 to 10,000-fold greater than any other taxon. Significance (p<0.05) for b and c determined by linear mixed effects modeling with Benjamini-Hochberg correction for multiple comparisons.

Extended Data Figure 8. Blockade of CCM formation by the TLR4 antagonist LPS-RS

a, Schematic of the experimental design in which iECre;Krit1^fl/fl littermates receive retro-orbital injections of the TLR4 antagonist LPS-RS. b, Visual (left) and microCT (right) images of hindbrains from vehicle or LPS-RS injected animals. c–d, Quantitation of CCM lesion and brain volume in iECre;Krit1^fl/fl littermates treated with vehicle or LPS-RS. Error bars shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. **indicates p<0.01; n.s. indicates p>0.05. All scale bars, 1 mm.

Extended Data Figure 9. CCM formation is stimulated by spontaneous abscess formation and not blocked by vancomycin

a, P10 hindbrains from Generation 3/Post ABX iECre;Krit1^fl/fl littermates in the longitudinal antibiotic experiment described in Fig 6e–l. The animal with a large CCM lesion burden on the far right was found to have an abdominal abscess (circle, “absc”) and splenomegaly (arrow, lower right). Scale bar, 1 mm. b, Schematic of the experimental design in which cohoused, lesion susceptible iECre;Krit1^fl/fl mating pairs were used to test the acute effect of vancomycin treatment on CCM formation. Offspring were studied after receiving maternal vehicle or vancomycin administered from E14.5 to P11. c–d, Visual images of hindbrains from representative offspring following vehicle or vancomycin antibiotic treatment. Scale bars, 1 mm. e–f, Volumetric quantitation of CCM lesions and brain volumes in iECre;Krit1^fl/fl littermates treated with vehicle or vancomycin. g–h, Relative quantitation of total neonatal gut bacterial load measured by qPCR of bacterial universal 16S or Firmicutes-specific rRNA gene copies. N≥6 per group. Error bars of all graphs shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. n.s. indicates p>0.05. ****indicates p<0.0001; n.s. indicates p>0.05.

Extended Data Figure 10. CCM formation is conferred to the offspring of resistant animals by fostering to Swiss-Webster mothers

a, Schematic of the experimental design in which timed matings of resistant iECre;Krit1^fl/fl and resistant iECre;Ccm2^fl/fl mating pairs were used to generate E19.5 offspring delivered by natural birth and raised by the birth mother or C-section/fostered to conventional Swiss-Webster foster mothers. b–c, Visual images of hindbrains from P10 resistant iECre;Krit1^fl/fl and iECre;Ccm2^fl/fl offspring following natural delivery and nursing by resistant mothers or after C-section/fostering to Swiss-Webster mothers. N≥6 per group.

Figure 1. CCM formation is stimulated by gram negative bacterial infection and intravenous LPS injection

a, Lesion formation in susceptible and resistant iECre;Ccm2^fl/fl mice at P17. Dotted lines trace cerebellar white matter. Asterisks; CCM lesions Scale bars, 1 mm (left) and 100 μm (right). b, Hindbrains of resistant iECre;Ccm2^fl/fl littermates without (top) and with (bottom) spontaneous abdominal gram negative abscess. Scale bars, 1 mm. Arrows; CCM lesions. c, The bacterial abscess (“absc”) identified in (b) contains gram negative bacteria (arrows). Scale bars, 4 mm (top) and 10μm (bottom). d, CCM formation in resistant iECre;Ccm2^fl/fl littermates following injection with a live B. fragilis/autoclaved cecal contents mixture (B. fragilis) or ACC alone (ACC vehicle). Scale bars, 1 mm. e–f, Resistant iECre;Ccm2^fl/fl responders exhibit splenic abscesses and increased spleen weight compared with non-responders. g, CCM formation in resistant iECre;Ccm2^fl/fl mice following vehicle or LPS treatment. Scale bars, 1 mm. h–i, Quantitation of lesion and total brain volumes. Error bars shown as s.e.m. and significance determined by one way ANOVA with Holm-Sidak correction for multiple comparisons or unpaired, two-tailed t-test. **** indicates p<0.0001; ***indicates p<0.001; n.s. indicates p>0.05.

Figure 2. CCM lesion formation requires endothelial TLR4/CD14 signaling

a, Injection of LPS at P5 drives CCM formation by P6 in susceptible iECre;Krit1^fl/fl littermates. Scale bars, 1mm (white) and 50 μm (yellow). Arrows and arrowheads; CCM lesions. Dotted lines; cerebellar white matter. b, Gene expression in cerebellar endothelial cells isolated from the indicated littermates at P6. N≥3 per group. c–f, Genetic rescue of CCM formation with endothelial loss of TLR4 or global loss of CD14. Visual appearance of CCM lesions (above), corresponding microCT images (below), and lesion volume quantitation. Scale bars, 1mm. Error bars shown as s.e.m and significance determined by one-way ANOVA with Holm-Sidak correction for multiple comparisons. ****indicates p<0.0001; ***indicates p<0.001; **indicates p<0.01; n.s. indicates p>0.05.

Figure 3. Increased TLR4 or CD14 expression is associated with higher lesion number in familial CCM patients

a, SNPs in the 5′ genomic regions of TLR4 and CD14 associated with increased lesion numbers in familial CCM patients are shown relative to the transcriptional start site (TSS). b–c, Normalized microarray measurement of TLR4 and CD14 expression in whole blood cells from individuals in the general population with the indicated TLR4 rs10759930 and CD14 rs778587 genotypes. d, Representative MRI images of KRIT1 Q455X patients with raw lesion count and TLR4/CD14 SNP genotypes (RA; risk allele). e–f, Sex and age adjusted log(lesion burden) in KRIT1 Q455X patients with indicated genotypes. Error bars shown as 95% confidence intervals and significance determined by one-way ANOVA with Holm-Sidak correction for multiple comparisons. ****indicates p<0.0001; ***indicates p<0.001; **indicates p<0.01; *indicates p<0.05.

Figure 4. CCMs fail to form in most germ-free mice

a, Experimental design in which offspring of susceptible Krit1^fl/fl females were fostered to conventional or germ-free Swiss-Webster mothers. b, Hindbrains from P10 offspring fostered in conventional (4/8 shown, top) or germ-free conditions (8/8 shown, bottom). c, Lesion volume quantitation of iECre;Krit1^fl/fl hindbrains following C-section/fostering in conventional or germ-free conditions. d, Relative quantitation of neonatal gut bacterial load measured by qPCR of bacterial 16S rRNA gene copies. e, Relative quantitation of Krit1 mRNA in lung endothelial cells (LEC) measured by qPCR. Red boxes indicate values for the single germ-free animal with significant lesions. Scale bars, 1 mm. Error bars shown as s.e.m. and significance determined by unpaired, two-tailed Student’s t-test. ****indicates p<0.0001; n.s. indicates p>0.05.

Figure 5. CCM susceptibility is associated with increased levels of gram negative Bacteroidetes s24-7

a–c, Visual and microCT images of hindbrains from susceptible (top) and resistant (bottom) iECre;Krit1^fl/fl and iECre;Ccm2^fl/fl animals and susceptible iECre;Krit1^fl/fl animals fostered to conventional Swiss-Webster (SW) mothers. Scale bars, 1 mm. d–e, Quantitation of lesion and brain volumes. Error bars shown as s.e.m. and significance determined by one-way ANOVA with Holm-Sidak correction for multiple comparisons. f–g, Principle Coordinates Analysis (PCoA) of unweighted and weighted UniFrac bacterial composition distances from the feces of susceptible and resistant Krit1^fl/fl and Ccm2^fl/fl mothers. P-values compare bacterial compositions in all resistant to all susceptible animals using PERMANOVA. h, Relative abundance boxplot of Bacteroidetes s24-7 in susceptible or resistant Krit1^fl/fl or Ccm2^fl/fl mothers and conventional SW foster mothers. Significance determined by linear mixed effects model with Benjamini-Hochberg correction for multiple comparisons. ****indicates p<0.0001; ***indicates p<0.001; n.s. indicates p>0.05. Note: Conventional SW data from Figs. 4 and 5 are the same experiment.

Figure 6. Preventing CCM formation by TLR4 antagonism and microbiome manipulation

a, Model of pathogenesis in which gram negative bacteria (GNB) in the gut are the source of LPS that enters circulating blood, activating luminal, brain endothelial TLR4 receptors. LPS-TLR4 stimulation drives MEKK3-KLF2/4 signaling to induce CCMs. b, Visual and microCT images of hindbrains from vehicle or Tak242 injected animals. c–d, Lesion and brain volume quantitation. e, Intergenerational experiment in which susceptible iECre;Krit1^fl/fl mating pairs were used to test the acute and chronic effects of antibiotic treatment on CCM formation. f–h, Visual and corresponding microCT images of hindbrains from offspring of three generations from one mating pair, representative of three pairs. i, Relative quantitation of neonatal gut bacterial load measured by qPCR of the bacterial 16S rRNA gene. N=4 per group. j, Lesion volume quantitation. k–l, Relative quantitation of Bacteroidetes s24-7 (s24-7) load in the neonatal gut measured by qPCR of the s24-7 rRNA gene (two distinct primer sets). N≥8 per group. All scale bars, 1 mm. Error bars shown as s.e.m. or boxplot and significance determined by one-way ANOVA with Holm-Sidak correction for multiple comparisons. ****indicates p<0.0001; ***indicates p<0.001; **indicates p<0.01; *indicates p<0.05; n.s. indicates p>0.05.