Prevotella

Hierarchical clustering of the 13,160 taxonomic profiles using Bray-Curtis distances and ward linkage was first employed to define the vaginal CSTs (Fig. 1). This analysis recovered the canonical five CSTs as described in Ravel et al. [7 ] but went further in delineating subtypes among the five CSTs. Cluster selection was made using the cutree function from the R stats package (version 3.6.0) on the dendrogram produced by hierarchical clustering. Cluster numbers from 2 to 20 were produced and then evaluated using the Davies Bouldin score. Support was found for nine clusters (Supplemental Figure 3c). For the L. crispatus-dominated CSTs, we were able to distinguish between communities which had mostly just L. crispatus (CST I-A) and those that had a lower, moderate relative abundance of the species (CST I-B). The same paradigm was observed for L. iners-dominated communities. Communities dominated by L. gasseri and L. jensenii more uncommon and were therefore not split into sub-CSTs. We were also able to distinguish three non-CSTs with a paucity of lactobacilli: CST IV-A, which contained Ca. L. vaginae, G. vaginalis, A. vaginae, and Prevotella; CST IV-B which contained G. vaginalis, A. vaginae, and Prevotella; and CST IV-C which was characterized by a paucity of Lactobacillus spp., G. vaginalis, Ca. L. vaginae, and A. vaginae. A second round of hierarchical clustering was performed (Bray-Curtis distances, ward linkage) on just the CST IV-C communities to further split this diverse collection of communities into additional sub-CSTs. Cutree was used on the resulting dendrogram to split IV-C into five subtypes, four of which had a characteristic phylotype and one which had a more even taxonomic composition. This decision improved assignments to IV-C, enhanced the interpretation of these communities, and resulted in a better Davies Bouldin score (Supplemental Figure 3). Reference centroids were constructed by averaging the relative abundances of each phylotype across the samples in training dataset which were included in each of the 13 sub-CSTs. Shannon diversity of samples assigned to each sub-CST was calculated using the log base 2.

DNA was prepared from frozen sputum and bacterial 16S ribosomal subunit (16S rRNA) genes were sequenced as part of a study we described previously [5 (link),8 (link)]. Details are available in [5 (link),8 (link)] and in the supporting information section [S1 Methods]. Original sff files with metadata are deposited in NCBI Sequence Read Archive (NCBI BioProject ID PRJNA423040).
For sequence analysis of the 631 samples included in the present study, the mothur (v.1.29) software package was used following the standard operating procedures (https://www.mothur.org/wiki/454_SOP). The total number of reads for each sample was rarefied (an average of 1000 subsampling iterations, rounded to the nearest whole number) to 547, the smallest number of reads obtained in the sample set, to control for differences in sequencing depth before alpha diversity measures were calculated. The dominant genus was defined as the most abundant genus observed in the sample. Prevalence was calculated as the number of samples with nonzero relative abundance divided by the total sample size. Non-parametric Shannon dissimilarity (Shannon diversity) was used to describe community richness and evenness [12 ].
OTUs representing Pseudomonas, Achromobacter, Burkholderia, Haemophilus, Staphylococcus, and Stenotrophomonas were categorized as “typical CF pathogens.” OTUs representing the obligate anaerobic genera Actinomyces, Fusobacterium, Porphyromonas, Prevotella, and Veillonella, and the facultative anaerobic genera Gemella, Granulicatella, Rothia, and Streptococcus were collectively referred to as “anaerobic genera.” Each of these nine anaerobic genera accounted for at least 1% mean relative abundance when present. Collectively, the six typical CF pathogen genera and nine anaerobic genera represented 94.3% of all sequencing reads (60.7% of reads represented typical CF pathogens; 33.6% represented anaerobic genera); the remaining 5.7% of DNA sequencing reads represented other genera.
To accommodate the longitudinal design of this study (89 subjects contributed more than one sample), generalized estimating equations (GEE) were used, with a gamma probability distribution, log link function, and unstructured working correlation matrix. Alpha was set at 0.05 (2-tailed) and analyses were conducted in SPSS (v24).

Gut microbial time series data were collected from 20 women each of whom donated stool samples for over a month, with a sampling frequency close to one sample per day (Vandeputte et al., submitted) [26 ]. These women also reported data on their menstrual cycle. For each sample, enterotype assignments were carried out as in Vandeputte et al. [27 (link)] with Dirichlet multinomial clustering. Samples were assigned to Bacteroides 1, Bacteroides 2, Ruminococcaceae, or Prevotella.
Progression through the menstrual cycle was rescaled to 28 days (the average length of a menstrual cycle) for all women. For days where there was more than one sample, only the first sample was used. Taxa present in less than 50% of participants were discarded from the analysis. Association networks were constructed with fastLSA v1.0 [28 (link)] with data rarefied to 10,000 sequences per sample, with correlations inferred across a delay of three time points (α = 0.05). Set sizes were analyzed with anuran, by generating 20 networks per observed network and resampling 100 different groups from these. Positive controls were generated 20 times, with a core size equal to 20% of the union of edges at 10% prevalence (edges present in at least two networks) and at 50% prevalence (edges present in at least ten networks). Set sizes and centralities with a p value below 0.05 for comparisons to values from random networks were considered significantly different from the random networks. The anuran toolbox was also used to assess the effect of increasing the number of participants.
The Walktrap community finding algorithm [29 ], implemented in the igraph R package v1.2.6 [30 ], was used to cluster the inferred CAN as the lack of negative edges in the CAN suggested that random walks could sufficiently identify clusters. To visualize enterotype-specific patterns of relative abundance, we computed the mean relative abundance of taxa per individual. We then took the median relative abundances across all individuals who belonged predominantly to the Ruminococcaceae enterotype, an enterotype previously linked to lower stool moisture [27 (link)], and subtracted from these all other median relative abundances, giving an estimate of taxa that had high abundance in the Ruminococcaceae enterotype compared to other enterotypes.
For the case study on the sponge microbiome, QIIME-processed data were downloaded from Moitinho et al. [31 (link)]. Samples with fewer than 1000 counts were removed and the samples were rarefied to even depth at 1034 sequences. After rarefaction, the abundance data were first filtered for 20% taxon prevalence across all samples, then once more to ensure 20% prevalence across different orders. Counts for removed taxa were retained to preserve the sample sums. After excluding host orders with fewer than 50 samples, 10 orders remained. CoNet v1.1.1 with renormalisation was then used to infer association networks (Faust and Raes [2 ]). Edges were generated with Pearson correlation, Spearman correlation, mutual information, Bray–Curtis dissimilarity, and Kullback–Leibler distance. Edges were included if at least one method reached significance; only edges with a combined Q-value below 0.05 (estimated using a combination of permutation and bootstrapping) were retained. The CoNet CANs were inferred with anuran generating 20 negative control random networks per host order and resampling these 100 times. For the positive controls, 20 network groups were generated with a core size equal to 20% of the union of edges at 20% prevalence (edges present in at least two networks) and at 50% prevalence (edges present in at least five networks). Set sizes and centralities with a p value below 0.05 for comparisons to values from random networks were considered significantly different from the random networks. CoNet networks were compared to FlashWeave networks [7 (link)]. FlashWeave v0.16.0 was run as FlashWeave-S (sensitive set to true and heterogeneous to false), with all other settings set to the default. To compare FlashWeave networks to CoNet networks, anuran generated five randomized networks per order-specific network and resampled these five times.
Prior research indicated that microbial abundance was a significant driver of community structure in sponges [32 (link)]. Therefore, taxa in the CAN were compared to taxa reported as indicators of high microbial abundance (HMA) or low microbial abundance (LMA) [32 (link)]. CAN network clusters were identified with manta v1.0.0 [33 ], as this algorithm has been designed to handle negative edges in the CAN. To run the clustering algorithm, default settings were used, except the number of iterations and permutations, which was set to 200. A Chi-squared test was used to compare HMA–LMA predictions to CAN cluster assignments (α = 0.05).

Twenty-four germ-free (GF) C57BL/6 mice (6 weeks of age) were supplied by the Chinese Academy of Medical Sciences (Beijing, China) and used for behavioral testing following treatment with fecal bacteria derived from individuals with DS and non-DS volunteers, as well as a preparation of Prevotella copri. All animal experimental procedures adhered to guidelines approved by the Chinese Academy of Medical Sciences [Permit No. SYXK (Beijing)-2018–0019].
Mice were housed in a standard room at 22 ± 1°C with a humidity of 55 ± 5%, under a 12 h light/dark cycle at the Chinese Academy of Medical Sciences Animal Facility. After acclimation for 7 days, the mice were randomly divided into four groups of six mice and received the following treatments twice a week for 42 days. (1) The control group was treated with intragastric gavage of 100 μL phosphate-buffered saline. (2) The HC group was treated with intragastric gavage of 100 μL of fecal bacteria pooled from 23 non-DS volunteers. (3) The DS group was treated with intragastric gavage of 100 μL of fecal bacteria pooled from 17 participants with DS. (4) The Prevotella group was treated with intragastric gavage of 100 μL of Prevotella copri (10⁸ CFU). After 42 days of treatment, the mice were underwent a series of behavior tests including sucrose preference test, open field test, and forced swimming test, as described below. All mice were then anesthetized with isoflurane and blood samples were collected by heart puncture after a 12-h fast following the accomplishment of all behavior tests.

DNA was extracted by the QIAmp Fast DNA Stool Mini Kit, following the manufacturer’s protocol (QIAGEN) and 1 pg was used with previously described primers (13 (link)) to amplify the 16s rRNA gene of Bacteroides, Lactobacillus and Prevotella by quantitative PCR using Applied Biosystems 7900HT systems.