Insular Resting State Functional Connectivity is Associated with Gut Microbiota Diversity

Abstract

The gut microbiota has recently gained attention as a possible modulator of brain activity. A number of reports suggest that the microbiota may be associated with neuropsychiatric conditions such as major depressive disorder, autism, and anxiety. The gut microbiota is thought to influence the brain via vagus nerve signaling, among other possible mechanisms. The insula processes and integrates these vagal signals. To determine if microbiota diversity and structure modulate brain activity, we collected fecal samples and examined insular function using resting state functional connectivity (RSFC). Thirty healthy participants (non-smokers, tobacco smokers, and electronic cigarette users, n=10 each) were studied. We found that the RSFC between the insula and several regions (frontal pole left, lateral occipital cortex right, lingual gyrus right, and cerebellum 4, 5 and vermis 9) were associated with bacterial microbiota diversity and structure. In addition, two specific bacteria genera, Prevotella and Bacteroides, were specifically different in tobacco smokers and also associated with insular connectivity. In conclusion, we show that insular connectivity is associated with microbiome diversity, structure, and at least two specific bateria genera. Furthemore, this association is potentially modulated by tobacco smoking, although the sample sizes for the different smoking groups were small and this result needs validation in a larger cohort. While replication is necessary, the microbiota is a readily accesible therapeutic target for modulating insular connectivity, which has previously been shown to be abnormal in anxiety and tobacco use disorders.

Graphical Abstract

We show that tobacco smokers have gut bacteria different to non-smokers and electronic cigarette users. The composition of gut bacteria (including bacteria prevalence specifically associated with tobacco smoking) correlates with the functional connectivity of the brain insula. The connectivity of the insula, especially with frontal regions, may be important in control of emotions and may be associated with drug abuse, making the microbiota a possible therapeutic target in psychiatry.

Introduction

The gut microbiota was hypothesized to affect brain function more than a century ago (Philips, 1910). Recently, a series of reports have shown that these relationships are likely to be stronger than expected. The brain communicates bi-directionally with the gut microbiota through immune, neuroendocrine, and neural pathways (Dinan & Cryan, 2017). The main neural line of communication is the vagus nerve, which connects viscera with the brain, ultimately reaching the insula, a brain region important for information integration. Specifically the middle insula signals gustatory and interoceptive information (Avery et al., 2015). Several published reports show that the gut microbiota likely plays a role in mood regulation, cognition, and visceral pain (Bercik et al., 2012; Collins et al., 2012; Mayer et al., 2014). A major impetus for this research is the increasing evidence that many substance use and psychiatric disorders and animal models of these disorders are associated with substantial alterations in the gut microbiota, and the ready accessibility of microbiota for a broad range of therapeutic interventions using anti-bacterials and pro-biotics to induce specific microbiota changes (Cryan & Dinan, 2012; Leclercq et al., 2014; Chi et al., 2017).

Tobacco smoking is a substance use disorder with widespread biological effects, including differences in the gut microbiota of two important bacterial taxa - Bacteroides and Prevotella. Quitting smoking has been associated with increased bacterial diversity in the gut, as different bacterial types become more abundant (Biedermann et al., 2013). Many components of tobacco smoke might alter the gut microbiota, but one component that can be isolated in humans is nicotine. Electronic cigarette (eCig) users ingest nicotine alone without the broader array of chemicals in tobacco smoke. In a recent pilot study, smokers had higher relative abundance of Prevotella and lower Bacteroides when compared to eCig users and healthy controls (Stewart et al., 2018).

These smoking-induced shifts in bacterial taxa, particularly changes in Bacteroides and Prevotella may have consequences for amino acid and neurotransmitter concentrations in the gut, which could then alter vagal nerve transmissions to the brain and the insula particularly. Changes in Bacteroides would be expected to modulate gut production of amino acids such as serotonin, catecholamines and glutamate, because Bacteroides constitute about 25% of all gut microbiota and provide amino acids and vitamins from dietary proteins (Hooper et al., 2002). In this pilot study, we aimed to address if gut microbiota among tobacco smokers, eCig users, and controls is associated with brain connectivity.

We focused on the insula, a brain region long known to be involved with both gustatory and interoceptive sensations that reach the brain through the vagus nerve (Craig, 2002; Maffei et al., 2012). Within the insular cortex, the middle insula integrates visceral interoceptive and gustatory information for homeostatic regulation (de Araujo et al., 2012; Avery et al., 2015). Thus, visceral information from the gut reaches the middle insula. We hypothesized that gut microbiota composition may influence the information travelling through the vagus to the brain, and therefore that the middle insula region might integrate interoceptive information about the specific taxa in the gut microbiota which was transmitted via the vagus nerve to the brain. We specifically hypothesized that middle insular connectivity would be modulated through changes in the gut microbiota.

Our study recruited healthy non-smokers, tobacco-smokers, and eCig users (n=10 each) and studied their gut microbiota composition. We then related microbiota parameters to brain insular resting state functional connectivity (RSFC). Our goal was to determine if there is an association between microbiome composition such as the observed different fecal proportion of Bacteroides and Prevotella and potentially reduced microbiota diversity and insular connectivity measures. In addition, we exploratorily studied possible differences between the smoking groups.

Methods

Participants.

Thirty participants were recruited using fliers and word of mouth (Table 1). The Baylor College of Medicine Institutional Review Board approved all aspects of the study. Participants were screened for MR safety and gave signed, informed consent. Tobacco smokers had an FTND >4 (Fagerstrom test for Nicotine Dependence (Heatherton et al., 1991)) and smoked a minimum of 10 cigarettes per day. eCig smokers reported daily use for at least the last six months.

Table 1.

Sample characteristics.

	Non-Smokers (N=10)	Ecig users (N=10)	Tobaco smokers (N=10)
Age (n.s.)	32 ± 2	30 ± 3	37 ± 3
Sex (% male)	90	90	100
Cigarettes/day	-	-	15 ± 2
FTND	-	-	4.8 ± 0.5
Ecig conc. (mg)	-	10 ± 2	-
Ecig vol. /day	-	11 ± 3	-
Ecig years use	-	2.7 ± 0.5	-
Weight (n.s.)	166 ± 10	174 ± 14	169 ± 7

Imaging.

Participants were scanned in a 3T Siemens Trio MR scanner in the Core for Advanced MR Imaging at Baylor College of Medicine in Houston, TX, USA. A 4.5 min structural MPRAGE sequence (TE = 2.66 ms, TR = 1200 ms, flip angle = 12°, 256 × 256 matrix, 160 one mm axial slices at 1×1×1 mm voxels) was collected, followed by a 5 min resting state scan (TE = 40 ms, TR = 2 s, flip angle = 90, 3.4×3.4×4 mm voxels). During the resting state scan a large “X” was displayed on the screen and participants were instructed to relax, with eyes open or closed according to their preference. We did not control for when each participant had eyes open or closed.

RSFC data were pre-processed using the CONN Functional Connectivity Toolbox (Whitfield-Gabrieli & Nieto-Castanon, 2012). The preprocessing pipeline consisted of realignment to the first time series image, slice-timing correction, structural segmentation and normalization to the MNI EPI template, functional normalization, ART-based outlier detection (Artifact Detection Tool, https://www.nitrc.org/projects/artifact_detect) using default settings, and smoothing with an 8 mm full width at half maximum Gaussian smoothing kernel.

For RSFC the primary region of interest (ROI) was the middle insula (Figure 1), with other seed regions considered exploratory to avoid multiple comparison issues. The insula ROIs were based on (Cauda et al., 2011) by pooling the spheres defined as anterior, middle, and posterior in their report into insula-constrained non-spherical ROIs. Anterior cingulate cortex and striatum were defined as in the Automated Anatomical Labeling (AAL) atlas (Tzourio-Mazoyer et al., 2002). The habenula was manually defined for each participant as two individual bilateral voxels using the T1 image as guide, as in (Curtis et al., 2017). Functional images were down-sampled to 3×3×3 to mitigate partial volume issues. RSFC data were then analyzed. White matter, cerebrospinal fluid, realignment, and scrubbing were considered variables of no interest. Fisher’s z-transformed correlation coefficients between the different seeds for each subject were used in statistical analyses. We subdivided the insula into its constituent parts: anterior, mid and inferior (left and right) for a total of 6 insular ROIs (Figure 1). Associations between microbiome diversity and structure and the right and left middle insula were the a priori hypotheses, while analyses of other insular regions and other areas (see below) were exploratory.

Figure 1.

Regions of interest (ROI). The left and right insular lobes were divided into right (r) and left (l) anterior, middle, and posterior parts. ROIs were drawn in AFNI, based on brain landmarks and coordinates published in previous literature. Insula ROIs were created in MNI space and entered into the CONN functional connectivity toolbox as seeds of interest for analyses, with middle insula as a priori hypothesis. We studied RSFC from the subdivided insula to the rest of the brain using seed-to-voxel analysis.

Following pre-processing, the RSFC data were de-noised using linear regression and band-pass filtering to remove potential confounds from the BOLD signal. Functional connectivity was then calculated as bivariate Fisher’s z-transformed correlation coefficients between different seed ROIs (middle insula as a priori hypothesis ROI, and anterior insula, inferior insula, anterior cingular gyrus, habenula, and striatum as exploratory ROIs) using a weighted generalized linear model. We chose those exploratory regions because of their roles in data integration and in reward processing (Posner et al., 2007; Hikosaka, 2010; Salas et al., 2010). For RSFC, we used a peak p value of 0.001 and a cluster false discovery rate (FDR) corrected p value of 0.05. Thus, p values are always FDR corrected to account for the number of voxels in the brain. In some cases we used an additional Bonferroni correction because we tested more than one hypothesis: For example, for our primary hypothesis (right and left middle insula) we used a Bonferroni correction of 8 because that was the total number of hypotheses tested: right and left middle insula, and four microbiota characteristics (operational taxonomic units (OTUs), Shannon’s index, and the first two weighted UF components). For exploratory data we did not use Bonferroni correction and simply used the FDR corrected p values for the clusters. For the tobacco-associated bacteria we Bonferroni corrected for 4 comparisons (right and left and two types of bacteria). Thus, any p value labeled “uncorrected” is in fact FDR corrected in the imaging analysis software, while p values labeled “corrected” are Bonferroni corrected for the total number of hypotheses tested.

Microbiota

DNA from 125 mg of fresh fecal samples was extracted using the AllPrep Bacterial kit (MoBio) following manufactures’ instructions. The bacterial 16S rRNA gene (V4 region) was amplified by PCR with barcoded Illumina adapter-containing primers 515F and 806R (Caporaso et al., 2012), and sequenced with the 2×250 bp cartridges in the MiSeq platform (Illumina). The read pairs were de-multiplexed and reads were merged using USEARCH v7.0.1090 (Edgar, 2010). Merging allowed zero mismatches and a minimum overlap of 50 bases, and merged reads were trimmed at the first base with a Q ≤ 5. A quality filter was applied to the resulting merged reads and those containing above 0.5% expected errors were discarded. Sequences were stepwise clustered into OTUs at a similarity cutoff value of 97% using the UPARSE algorithm (Edgar, 2010). Chimeras were removed using USEARCH v7.0.1090 and OTUs were mapped to the SILVA Database (Quast et al., 2013)(containing only the 16S V4 region) using USEARCH v7.0.1090. Abundances were recovered by mapping the merged reads to the UPARSE OTUs. Samples were rarefied to 4,000 reads. The full results of the microbiota analysis have been published (Stewart et al., 2018). The weighted UF distances were plotted by principal coordinate analysis (PCoA) to provide information on how samples group according to the overall microbiome profiles. The first two components from the PCoA ordination were used in subsequent brain analysis. Significant genera between groups were identified by the Kruskal-Wallis test and corrected for multiple comparisons using FDR correction. P values <0.05 after adjustment were considered significant.

Results

Middle insular RSFC to frontal and cerebellar areas were associated with gut microbiota

We first showed that Shannon diversity and weighted UniFrac (wUF) distances (principal component 1–3 from PCoA ordination) were affected by tobacco status (Table 2). Therefore, tobacco status was treated as a covariate of no interest in this analysis. In accordance with our original hypothesis, we studied RSFC between the middle insula and the whole brain in 30 healthy controls, using measures of microbiota structure and diversity as regressors of interest and smoking status as a covariate of no interest. We found that the connectivity between the left middle insula and the left frontal pole was positively associated with the number of OTUs (p_uncorrected=0.00019, p_corrected=0.0015). In addition, the connectivity between the left middle insula and the cerebellar vermis 9 region was (nominally, not surviving Bonferroni correction) associated with weighted UF principal component 2 (p_uncorrected=0.030, p_corrected=0.24) (Figure 2).

Table 2.

Measures of bacterial diversity and bacterial classes.

	Non-Smokers	Ecig Users	Tobacco Smokers	One-way Anova p-value
OTUs	96.7 ± 7.8	85.0 ± 6.3	91.1 ± 6.9	0.508
Shannons	3.23 ± 0.09	3.06 ± 0.07	2.62 ± 0.18	0.004
Weighted UF1	−0.04 ± 0.02	−0.04 ± 0.01	0.08 ± 0.03	8.13e–04
Weighted UF2	0.002 ± 0.014	−0.007 ± 0.03	0.004 ± 0.010	0.898
Bacteroides	374.6 ± 72.4	341.8 ± 47.0	119.6 ± 35.9	0.005
Prevotella	47.0 ± 47	0.0 ± 0.0	533.1 ± 156	6.16e–04

Figure 2.

Significant clusters resulting from seed-to-voxel resting state functional connectivity analyses using middle insula areas as seeds and measures of microbiota structure and diversity as regressors of interest. A) Cluster (−18 +48 +12) – significant regression of OTUs, using left (l) middle insula as seed; 131 voxels, frontal pole left. B) Cluster (+06 −56 −30) – significant regression of weighted UF component 2, using l middle insula as seed; 45 voxels, vermis 9.

RSFC in other insular areas, but not anterior cingulate, habenula, or striatum were associated with gut microbiota

In an exploratory analysis, we studied several ROIs to determine whether RSFC in other regions was also associated with bacterial structure and diversity, using smoking status as a covariate of no interest. We studied the anterior cingulate cortex, habenula, striatum, and anterior and inferior insular regions. Interestingly, only the additional insular regions showed association with microbiota diversity. With number of OTUs as a regressor of interest, we found a significant cluster for the connectivity between the right inferior insula and the right occipital cortex. With the wUF component 2 as a regressor of interest, we found significant clusters for connectivity between the right anterior insula and the right lingual gyrus as well as two clusters between the left anterior insula and the cerebellum lobes 4 and 5 (Figure 3, all FDR corrected, Bonferroni uncorrected p <0.05). No other ROIs or regressors rendered significant results.

Figure 3.

Significant clusters resulting from seed-to-voxel resting state functional connectivity analyses using additional insula ROIs as exploratory areas of interest and measures of microbiota structure and diversity as regressors of interest. A) Cluster (+12 −80 +50) – significant regression of OTUs, using right (r) inferior insula as seed. 223 voxels, lateral occipital cortex, superior division right. B) Cluster (+18 −68 −10) – significant regression of weighted UniFrac (wUF) component 2, using r anterior insula as seed. 232 voxels, lingual gyrus right. C) Cluster (+24 −70 −14) - significant regression of wUF component 2, using l anterior insula as seed. 125 voxels, cerebelum 4, 5 left. D) Cluster (−20 −52 −18) - significant regression of wUF component 2, using l anterior insula as seed. 125 voxels, cerebelum 4, 5 left.

Bacterial genera significantly different in tobacco smokers were associated with insular RSFC

We found that tobacco smokers had lower relative abundance of Bacteroides and higher relative abundance of a Prevotella (Table 2). We then compared the relative abundance of Bacteroides and Prevotella with insular connectivity by evaluating RSFC between the left and right anterior, medial, and inferior insula and the rest of the brain. Bacteroides and Prevotella relative abundance were included as regressors of interest while tobacco status was treated as a covariate of no interest. For Bacteroides, we found a significant cluster for the connectivity between the left anterior insula and the left operculum (cluster size = 420 voxels, p_uncorrected=0.00030, p_corrected=0.0012) with higher Bacteroides associated with lower RSFC). For Prevotella, we found a significant cluster for the connectivity between the left anterior insula and the right occipital cortex (cluster size = 222 voxels, p_uncorrected=0.016, p_corrected=0.064, not surviving Bonferroni correction) and also between the right inferior insula and the middle frontal gyrus (cluster size = 30 voxels, p_uncorrected=0.0082, p_corrected=0.033) with higher Prevotella associated with higher RSFC) (Figure 4).

Figure 4.

Significant clusters resulting from seed-to-voxel functional resting state connectivity analyses using insula ROIs, Bacteroides and Prevotella measures as regressors of interest, and tobacco status as covariate of no interest. A) Cluster (−56 −28 +12) – significant regression of Bacteroides, using l anterior insula as seed. 134 voxels, parietal operculum cortex left. B) Cluster (+32 −74 +56) – significant regression of Prevotella, using l anterior insula as seed. 184 voxels, lateral occipital cortex, superior division right. C) Cluster (+34 +30 +46) - significant regression of Prevotella, using r inferior insula as seed. 211 voxels, middle frontal gyrus right. Tobacco status was used as covariate of no interest.

Discussion

In this study, we found that middle insular RSFC was associated with bacterial structure and diversity in the gut. Specifically, connectivity between the middle insula and the frontal pole was positively correlated with the number of OTUs and connectivity between the middle insula and the cerebellum vermis 9 was positively correlated with the bacterial profiles by wUF (this cerebellar result did not survive Bonferroni correction and must be taken as preliminary). Supporting this connection, the frontal pole, and specifically the lateral portion, is strongly connected to the insula (Gilbert et al., 2010). The frontal pole may underlie important cognitive functions, including control over multitasking, episodic memory retrieval, and mentalizing (Gilbert et al., 2010). Thus, if the frontal pole exerts inhibitory control over the insula, increased RSFC between the frontal pole and the insula could result in lower insula activation during specific events. Although the occipital cortex connectivity appeared as possibly related to the microbiota in both the exploratory study and the Prevotella specific experiment, neither result is statistically significant when multiple comparisons are taken into account. However, the intrinsic connectivity of the lateral occipital cortex, superior division right has been shown to be decreased, together with insular intrinsic connectivity, after 3,4-methylenedioxymethamphetamin ingestion (Walpola et al., 2017). Thus, it is possible that the insula/lateral occipital cortex connectivity may play a role in substance use disorders in general.

Whether diversity per se or specific types of bacteria are associated with our results is matter of speculation. However, Bacteroides are responsible for the production of key amino acids and neurotransmitters, thus fewer Bacteroides could make a contribution to the supply of neurotransmitters in the brain. This relative deficit in neurotransmitters would then be expressed as alterations in insular activity and connectivity to adjust the brain’s homeostasis in response to this deficit. More broadly, a more diverse gut microbiota may be linked to an increased functional capacity in many body systems including the brain. Further work is necessary to determine the potential functional consequences of higher bacterial richness in the gut on brain activity.

The gut microbiota can interact with the brain through immune, neuroendocrine, and neural pathways (Dinan & Cryan, 2017). In terms of neural pathways, the middle insula is known to be important for the evaluation and integration of interoceptive information (Avery et al., 2015). To avoid multiple comparison problems currently being debated in the brain imaging literature (Eklund et al., 2016; Cox et al., 2017), we circumscribed the study to the middle insula as an a priori ROI, with a few other regions known to integrate several kinds of neural information (anterior and inferior insula, anterior cingulate cortex, habenula, and striatum) as exploratory. Similarly, we focused on measures of bacterial structure and diversity (and on two specific genera shown to be different in tobacco smokers compared to non-smokers) to avoid likely false positives from studying a myriad of bacteria.

Given how few other studies are available comparing gut microbiota and brain function, we felt it necessary to present a larger characterization of possible connections. However, we note that these should be considered exploratory only, and leave their confirmation or refutation to future work. In these exploratory hypotheses, the only areas where bacterial structure and diversity correlated with RSFC were the anterior and inferior insula (the fact that in these exploratory results we did not used corrections for the number of comparisons explored must be noted). RSFC in other signal integration brain regions (e.g. anterior cingulate, habenula, striatum) did not vary with bacterial diversity. This finding supports a specialized role for the insula as a region that integrates interoceptive information from the vagal nerve and other brain regions. A lack of findings outside the insula suggests that our results are unlikely to be false positives, as could be the case if many unrelated areas exhibited correlations between RSFC and bacterial diversity or microbiota structure.

Outside of the current study, other recent reports suggest a relationship between the gut microbiota and observable differences in functional and structural brain imaging measures. For example, in a sample of women consuming either fermented or non-fermented milk products for 4 weeks (i.e., fermented products likely to cause changes in the gut microbiota), intake of fermented milk products was associated with reduced task-related response of a distributed functional network containing affective, viscerosensory, and somatosensory cortices, including the insula (Tillisch et al., 2013). In another report aimed at differentiating microbiota and brain imaging parameters in obese and non-obese participants, the Shannon diversity index correlated with fractional anisotropy in hippocampus, hypothalamus, and caudate (Fernandez-Real et al., 2015). In addition, when patients with irritable bowel syndrome viewed negative emotional stimuli, treatment with probiotic bifidobacterium longum NCC3001 (versus placebo) reduced depression scores and altered brain activity in several brain regions, including amygdala and fronto–limbic regions (Pinto-Sanchez et al., 2017). Gut microbiota has also been related to ventral striatum responses to reward anticipation in a sample of adolescent ADHD patients (Aarts et al., 2017). Finally, in a group of healthy women, cluster analysis showed two groups (interestingly, higher Prevotella and higher Bacteroides), which could be differentiated by their brain morphometry characteristics (Tillisch et al., 2017).

Although the sample size for the three smoking conditions (non-smokers, tobacco smokers, and e-cig users) was small, we were able to replicate previous work showing differences in gut microbiota among smokers (Benjamin et al., 2012). Thus, our preliminary evidence suggests that these microbiota differences were not directly associated with nicotine but likely with the many other components of tobacco smoke. Given that the insula has been shown to play a very important role in tobacco cessation and possibly in other addictions as well (Droutman et al., 2015), tobacco-related alterations in bacteria genera and their association with insular connectivity also may be relevant for the development of tobacco cessation therapies.

Several limitations to this report should be noted. First, this study involved only 30 participants, which was sufficient for preliminary analysis, but replication and validation are necessary. Second, we used a large voxel size (3.4 × 3.4 × 4 mm) and it is possible that the insular subdivision data suffers from partial volume effects. Although this would not change the broad conclusions of the manuscript, finer resolution may be able to yield greater insights into the roles of different divisions within the insula. Next, we used only 5 minutes of resting state, while it has been shown that 7 minutes may render more reliable results with lower within-subject variability (Tomasi et al., 2016). All but two of the participants were men, and a replication of these experiments on women is necessary. Finally, extensive dietary information was not available.

In conclusion, we showed that insular resting state functional connectivity is associated with microbiota structure, bacterial diversity, and specific bacteria genera in the gut. While validation in larger cohorts is necessary, these data open the door to future studies aimed at modulating microbiota to help treat several insula-related neuropsychiatric diseases.

Abbreviations

AAL

Automated Anatomical Labeling

eCig

electronic cigarette

FDR

false discovery rate

OTUs

operational taxonomic units

PCoA

principal coordinate analysis

RSFC

resting state functional connectivity

ROI

region of interest

wUF

weighted UniFrac