Permissive microbiome characterizes human subjects with a neurovascular disease cavernous angioma

Abstract

Cavernous angiomas (CA) are common vascular anomalies causing brain hemorrhage. Based on mouse studies, roles of gram-negative bacteria and altered intestinal homeostasis have been implicated in CA pathogenesis, and pilot study had suggested potential microbiome differences between non-CA and CA individuals based on 16S rRNA gene sequencing. We here assess microbiome differences in a larger cohort of human subjects with and without CA, and among subjects with different clinical features, and conduct more definitive microbial analyses using metagenomic shotgun sequencing. Relative abundance of distinct bacterial species in CA patients is shown, consistent with postulated permissive microbiome driving CA lesion genesis via lipopolysaccharide signaling, in humans as in mice. Other microbiome differences are related to CA clinical behavior. Weighted combinations of microbiome signatures and plasma inflammatory biomarkers enhance associations with disease severity and hemorrhage. This is the first demonstration of a sensitive and specific diagnostic microbiome in a human neurovascular disease.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Fecal microbiotas are different in CA and non-CA cohorts.

a Organization of microbiome species in non-CA and CA cohorts. Co-occurrence network analyses were performed at the species level, as determined by metagenomic shotgun sequencing data analysis (n = 27 non-CA, n = 122 CA, keystone species are labeled in red). b Multi-variate differential abundance analyses of metagenomic shotgun data at species level. Species with significantly different abundance (ANCOM analysis, followed by two-sided Mann–Whitney U test with Benjamini–Hochberg false discovery rate (FDR) correction for multiple testing, pFDR < 0.01) and medium relative abundance ≥0.25% in either group are presented as box-whisker plots (blue boxes: non-CA cohort; red boxes: CA cohort). c Identification of key species (medium relative abundance ≥0.1% in either group) by random forest analysis. Species identified in both multi-variate and random forest analyses are shown in green. d ROC curve was identified based on best-weighted combination of common bacterial species identified by multi-variate and random forest analyses (AUC = 0.826, specificity = 0.667, sensitivity = 0.925). e α-Diversity analyses of fecal samples by Shannon and Simpson indices based on 16 S rRNA gene amplicon sequencing data, presented as box-whisker plots (n = 250 non-CA, n = 115 CA, Kruskal–Wallis one-way analysis of variance test; blue boxes: non-CA cohort; red boxes: CA cohort). f Relative abundance of Gram-negative and Gram-positive bacteria in non-CA and CA cohorts (n = 250 non-CA, n = 115 CA, ANCOM analysis, followed by two-sided Mann–Whitney U test, blue boxes: non-CA cohort, red boxes: CA cohort). g Multi-variate differential abundance taxonomic analyses of 16S rRNA gene amplicon sequencing data. ESVs with significantly different relative abundances (pFDR ≤ 0.01) and medium relative abundance of ≥1% in either group are presented as box-whisker plots (n = 250 non-CA, n = 115 CA, ANCOM analysis, followed by two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction; blue boxes: non-CA cohort; red boxes: CA cohort). In box plots, bounds of boxes show interquartile range (IQR), top and bottom whiskers demonstrate maximum and minimum, lines in the middle of the box indicate median, and stars show mean of the data. + signs indicate outliers.

Fig. 2. Fecal LPS synthesis pathway is upregulated in CA patients.

a Metagenomic shotgun data were used to compare relative gene abundance of bacterial metabolic pathways. Relative abundances of significantly different metabolic pathways are shown (n = 27 non-CA, n = 121 CA, two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction). b Relative abundance of significantly different LPS synthesis pathway-related genes between non-CA and CA individuals (two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction). c Batch-corrected LPB content in peripheral plasma of non-CA and CA individuals (n = 16 for non-CA, n = 47 for CA individuals, two-tailed unpaired t test with Welch’s correction). Average data are presented as mean ± s.e.m. Blue: non-CA individuals, red: CA patients.

Fig. 3. CA patient subpopulations can have different microbiota.

a Organization of microbiome species in non-aggressive and aggressive patients. Co-occurrence network analyses were performed at species level, as determined by metagenomic shotgun data analysis (n = 45 non-aggressive patients, n = 62 aggressive patients, keystone species are labeled in red). b Multi-variate differential abundance analysis of metagenomic shotgun data between non-aggressive and aggressive patients at species level. Species with significantly different abundance are presented as box-whisker plots (ANCOM analysis followed by two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction, blue boxes: non-aggressive patients, red boxes: aggressive patients). c Identification of key species by random forest analysis. d ROC curve was identified based on best-weighted combination of all bacterial species identified by multi-variate and random forest analyses (AUC = 0.778, specificity = 0.786, sensitivity = 0.660). e α-Diversity analyses of fecal samples of CA patients with non-aggressive and aggressive disease by Shannon and Simpson indices based on 16S rRNA gene amplicon sequencing data (n = 43 non-aggressive patients, n = 58 aggressive patients, Kruskal–Wallis one-way analysis of variance test, blue boxes: non-aggressive patients, red boxes: aggressive patients). f Multi-variate differential abundance taxonomic analyses between non-aggressive and aggressive CA patients based on 16S rRNA gene amplicon sequencing results. ESVs with significantly different relative abundances are presented as box-whisker plots (ANCOM analysis followed by two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction, blue boxes: non-aggressive patients, red boxes: aggressive patients). g Multi-variate differential abundance analysis of metagenomic shotgun data between non-CASH and CASH patients at the species level. Species with significantly different abundance are presented as box-whisker plots (n = 100 non-CASH patients, n = 13 CASH patients, ANCOM analysis followed by two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction, green boxes: non-CASH patients, orange boxes: CASH patients). h Identification of key species by random forest analysis. i ROC curve was identified based on best-weighted combination of all bacterial species identified by multi-variate and random forest analyses (AUC = 0.682, specificity = 0.933, sensitivity = 0.432). j α-Diversity analyses of fecal samples of CA patients with non-CASH and CASH disease by Shannon and Simpson indices based on 16S rRNA gene amplicon sequencing data, presented as box-whisker plots (n = 93 non-CASH patients, n = 13 CASH patients, Kruskal–Wallis one-way analysis of variance test, green boxes: non-CASH patients, orange boxes: CASH patients). k Multi-variate differential abundance taxonomic analyses between non-CASH and CASH patients based on 16S rRNA gene amplicon sequencing results. ESVs with significantly different relative abundances are presented as box-whisker plots (ANCOM analysis followed by two-sided Mann–Whitney U test with Benjamini–Hochberg FDR correction, green boxes: non-CASH patients, orange boxes: CASH patients). In box plots, bounds of boxes show IQR, top and bottom whiskers demonstrate maximum and minimum, lines in the middle of the box indicate median, and stars show mean of the data. + signs indicate outliers.

Fig. 4. Plasma and fecal microbiome as CA biomarkers.

a Correlation between individual bacterial species and circulating factors. Correlation between species identified by multi-variate differential abundance and random forest analyses (Figs. 1 and 3, Supplementary Fig. 2) and individual plasma biomarkers. Degree of correlation is color coded (also see Supplementary Table 1, Pearson’s correlation testing; red: positive correlation; blue: negative correlation). b Correlation between combination of species identified based on clinical questions and circulating factors. Correlation between combination of bacterial species identified (in Figs. 1 and 3, Supplementary Fig. 2) and individual plasma biomarkers. Degree of correlation is color coded (also see Supplementary Table 1, Pearson’s correlation testing, red: positive correlation, blue: negative correlation). c–e Comparison of best ROC curve determined by combination of bacterial species identified in Fig. 3 and Supplementary Fig. 2, individual circulating factors, and bacterial species and individual circulating factors. c Distinction between sporadic/solitary and familial/multifocal CA patients (green: bacterial species identified in Supplementary Fig. 2 (B. dorei), blue: LPB, red: combined bacterial species and LPB). d Distinction between non-aggressive and aggressive CA patients (green: bacterial species identified in Fig. 3b, c, blue: IL-10, red: combined bacterial species and IL-10). e Distinction between non-CASH and CASH patients (CASH patient, green: bacterial species identified in Fig. 3g, h, blue: CRP, red: combined bacterial species and CRP). IFNγ interferon-γ, TNF tumor necrosis factor; CRP C-reactive protein, TLR4 toll-like receptor 4, VEGF vascular endothelial growth factor, THBS1 thrombospondin 1, TM thrombomodulin.