Effects of Visceral Interoception on Topological Properties of the Brain – A Graph Theory Analysis of Resting state fMRI

Abstract

Recent neuroimaging studies have employed graph theory as a data-driven approach to describe topological organization of the brain under different neurological disorders or task conditions and across life span. In this exploratory study, we tested whether subtle differences in interoception related to intravesical fullness can alter brain topological architecture in healthy participants. 17 right-handed women underwent a series of resting state fMRI scans that included catheterization and partial bladder filling. Using a whole brain regions of interest (ROIs), we computed several graph theory metrics to assess the efficiency of brain-wide information exchange. Results showed that brain network’s topological properties significantly changed in many brain regions when we binary compared different interoceptive resting state conditions. Notably, we observed changes in global efficiency in the salience network, the central executive network, anterior dorsal attention network and the posterior default-mode network (DMN) as bladder became full and interoceptive signals intensified. Moreover, degree (the number of connections for each node), and betweenness centrality (how connected a particular region is to other regions) differed between the empty bladder, the catheterized empty bladder, and the catheterized and partially filled bladder. Comparing resting state data before and after an interoceptive task (repeated intravesical infusion and drainage) further showed increased average path length for the salience networks and decreased clustering coefficient of the DMN. These results suggest visceral interoception influences brain topological properties of resting state networks.

I. INTRODUCTION

The human brain is a complex system whose topological organization can be represented as a structural connectome of interconnected axonal pathways [1], or a functional connectome of synchronized neural activity [2]. Brain connectome can be mapped using non-invasive techniques such as fMRI [3] and further characterized through graph theory analytic techniques [4]. fMRI studies that applied graph theory on resting state data have consistently demonstrated non-trivial topological properties of brain [4].

Graph theoretical approaches model the human brain connectome as a collection of nodes linked by edges. In fMRI data, the nodes can represent functionally-defined brain regions of interest (ROIs) while the edges can be estimated by thresholded temporal correlations between ROI timeseries[4]. Previous studies have shown that local and global measures of brain networks undergo topological changes under different neurological disorders or task conditions and across life span [3], [5]. Little research, however, has investigated the effect of visceral interoception on topological properties of the human brain. Interoception is the processing and integration of body afferent signals[6]. Visceral interoception can be defined as the sensations and corresponding bodily feelings that arise from the viscera including organ pressure and distention (fullness) [6]. Rather than ascertaining whether individual brain regions are impacted by interoception, here we analyzed resting state fMRI data of healthy women with graph theory and asked whether the architecture of communication patterns across the brain was altered under different interoceptive conditions.

II. MATERIALS AND METHODS

A. Subjects and Imaging Protocol

Seventeen healthy women (mean age 31.12 ± 8.7 years) participated in this study. Only women were recruited because of the known gender difference in brain response to viscerosensory stimulation [7]. Written informed consent was obtained before the MRI scan and the study protocol was approved by the Institutional Review Board of the local ethics committee and conformed to the ethical guidelines of the Declaration of Helsinki. As shown in Fig. 1, a high-resolution T1-weighted image was obtained to yield an anatomical underlay for the fMRI results. The first resting state fMRI scan (REST1) was then collected during rest with eyes open (i.e., subjects were asked not to think of anything in particular, and to passively look at the fixation point). This initial acquisition was performed with an empty bladder. The second (REST2) and third (REST3) resting state fMRI scans were also collected during eyes open rest conditions but after transurethral catheterization, and after intravesical infusion of 100 ml body-temperature (i.e., 37°C) saline solution, respectively. Next fMRI scan was collected during a visceral stimulation task involving repetitive intravesical infusion and drainage of saline solution using an MR-compatible visceral stimulation device. Task data were analyzed separately [8]. The fourth and final resting state fMRI data (REST4) were collected under similar condition to REST3 except that they were acquired after the task.

Fig. 1.

Neuroimaging protocol

B. Imaging Data Acquisition and Analysis

Magnetic resonance imaging was performed with a 3T Ingenia scanner with a 15-channel head coil. Structural images were acquired with 3D T1-TFE sequence (TR/TE = 6.9/3.1 ms, matrix size = 256 × 256, 180 slices, 1 mm isotropic resolution, scan duration = 304.6 s). T2*-weighted functional images were acquired with GE-EPI sequence: TR/TE/flip angle = 2 s/40 ms/80°, image matrix size in plane = 96 × 96, FoV = 240 mm, slice thickness = 3 mm, inter-slice gap = 1 mm, 34 axial slices in ascending order.

Images were preprocessed using SPM12 (Wellcome Trust Centre, London, UK). Following MRI quality check, raw functional images were slice-time corrected, realigned (motion corrected), unwarped (correcting in part for distortions attributable to head motion), coregistered to subjects corresponding T1-weighted structural scan, normalized to a standard template (i.e., Montreal Neurological Institute [MNI] reference space), and spatially smoothed (FWHM = 6 mm). Artifact Detection Tool (ART) was used for outlier detection of global signal and motion and CompCor method [9] was applied to identify principal components associated with each subject’s segmented white matter and cerebrospinal fluid. These components were entered as confounds along with realignment parameters and other nuisance covariates mentioned above in a first-level analysis and were regressed out. Residual BOLD timeseries was detrended linearly and band-pass filtered (0.008–0.09Hz). A connectivity matrix was generated for each subject and each REST condition by calculating the correlation coefficient between 32 ROIs included in the CONN toolbox [10]. These 32 ROIs (Table 1) span across the following eight networks: the default-mode network (DMN), the sensorimotor network (SMN), the visual network (VN), the salience network (SN), the dorsal attention network (DAN), the frontoparietal network otherwise known as central executive network (CEN), the language networks (LN), and the cerebellar network (CN). Three types of graph theoretical measures were computed for each subject and for each REST condition: (i) functional integration (global efficiency and average path length),(ii) functional segregation (clustering coefficient and local efficiency), and (iii) centrality (degree and betweenness centrality). Following recommendations, for every ROI, these six measures were calculated adopting a conventional 2-sided p−value corrected using the Bonferroni correction for multiple comparisons to identify statistically significant correlations between ROIs and a two-sided cost value of 0.15 as adjacency matrix threshold for network edges [10]. Uncorrected p < 0.05 was used for exploratory purposes.

TABLE 1.

Thirty-two ROIs from eight networks used as nodes

Network	Hemisphere	Nodes	Coordinate (x y z)
DMN		MPFC	(1, 55, −3)
	L	LP	(−39, −77, 33)
	R	LP	(47, −67, 29)
		PCC	(1, −61, 38)
SMN	L	Lateral	(−55, −12, 29)
	R	Lateral	(56, −10, 29)
		Superior	(0, −31, 67)
VN		Medial	(2, −79, 12)
		Occipital	(0, −93, −4)
	L	Lateral	(−37, −79, 10)
	R	Lateral	(38, −72, 13)
SN		ACC	(0, 22, 35)
	L	AI	(−44, 13, 1)
	R	AI	(47, 14, 0)
	L	RPFC	(−32, 45, 27)
	R	RPFC	(32, 46, 27
	L	SMG	(−60, −39, 31)
	R	SMG	(62, −35, 32)
DAN	L	FEF	(−27, −9, 64)
	R	FEF	(30, −6, 64)
	L	IPS	(−39, −43, 52)
	R	IPS	(39, −42, 54)
CEN	L	LPFC	(−43, 33, 28)
	L	PPC	(−46, −58, 49)
	R	LPFC	(41, 38, 30)
	R	PPC	(52, −52, 45)
LN	L	IFG	(−51, 26, 2)
	R	IFG	(54, 28, 1)
	L	pSTG	(−57, −47, 15)
	R	pSTG	(59, −42, 13)
CN		Anterior	(0, −63, −30)
		Posterior	(0, −79, −32)

Abbreviations: L: left; R: right; MPFC: medial prefrontal cortex; LP: lateral parietal; PCC: posterior cingulate cortex; ACC: anterior cingulate cortex; AI: anterior insula; RPFC: rostrolateral prefrontal cortex; SMG: supramarginal gyrus; FEF: frontal eye fields; IPS: intraparietal sulcus; LPFC: lateral prefrontal cortex; PPC: posterior parietal cortex; LPFC: lateral prefrontal cortex; IFG: inferior frontal gyrus; pSTG: posterior superior temporal gyrus.

III. RESULTS

Alterations of 32 nodes on eight brain networks by graph topological metrics measurement are shown in Fig. 3. Red and blue nodes correspond to positive and negative T-values, respectively. The corresponding values of different graph metrics are listed in Table 2. The bold values in Table 2 are statistically significant at Bonferroni adjusted p < 0.05.

Fig. 3.

Alterations of selected network ROIs by graph topological metrics A REST2 > REST1, B REST3 > REST1, C REST4 > REST1, D REST3 > REST2, E REST4 > REST2, and F REST4 > REST3 . Red and blue nodes correspond to the positive and negative T-values, respectively (Punc < 0.05).

TABLE 2.

THE GRAPHIC METRICS OF DIFFERENT INTEROCEPTIVE RESTS

	Measure	ROI	Beta	T	dof	p-unc
A REST2 > REST1	Global Efficiency
	Average Path Length
	Clustering Coefficient	SMN.Superior	−0.22	−2.42	12	0.0322
	Local Efficiency	SMN.Superior	−0.27	−3.42	12	0.0050
		DAN.EFE(R)	−0.15	−2.45	9	0.0370
	Degree
	Betweenness Centrality	VN.Lateral (R)	−0.04	−2.25	16	0.0390
	Cost
B REST3 > REST1	Global Efficiency
	Average Path Length
	Clustering Coefficient	LN.IFG (L)	0.21	3.02	14	0.0091
	Local Efficiency	LN.IFG (L)	0.19	2.26	14	0.0399
	Degree
	Betweenness Centrality	SN.SMG (R)	−0.02	−3.44	16	0.0033
			−0.03	−2.29	16	0.0362
	Cost
C REST4 > REST1	Global Efficiency	DMN.LP (R)	0.07	2.85	16	0.0116
		VN.Lateral (R)	0.05	2.27	16	0.0373
	Average Path Length	SN.SMG (L)	0.35	2.26	16	0.0378
		VN.Lateral (L)	0.31	2.17	16	0.0456
	Clustering Coefficient	DMN.LP (R)	−0.2	−3.43	15	0.0037
		CEN.PPC(L)	0.22	3.02	13	0.0098
		VN.Lateral (R)	−0.26	−2.55	15	0.0222
	Local Efficiency	DMN.LP (R)	0.24	3.36	13	0.0051
		SMN.Superior	−0.16	−2.91	15	0.0107
		CEN.PPC(L)	−0.21	−2.28	15	0.0376
	Degree	DMN.LP (R)	1.06	2.66	16	0.0169
		VN.Lateral (R)	1.53	2.52	16	0.0227
		SN.AI (R)	−1.24	−2.32	16	0.0338
	Betweenness Centrality	CEN.PPC(L)	−0.06	−2.62	16	0.0187
		SN.AI (L)	0.06	2.56	16	0.0208
		VN.Lateral (R)	0.06	2.27	16	0.0376
	Cost	DMN.LP (R)	0.03	2.66	16	0.0169
		VN.Lateral (R)	0.05	2.52	16	0.0227
		SN.AI (R)	−0.04	−2.32	16	0.3382
D REST3 > REST2	Global Efficiency	SN.SMG (R)	−0.05	−3.04	16	0.0077
		VN.Occipital	−0.12	−2.64	16	0.0177
		CEN.LPFC(L)	−0.07	−2.57	16	0.0206
		CEN.PPC(R)	−0.06	−2.43	16	0.0273
		DAN.FEF(L)	−0.09	−2.25	16	0.0386
		DMN.LP (R)	−0.08	−2.19	16	0.0436
		CEN.LPFC(R)	−0.07	−2.15	16	0.0471
	Average Path Length
	Local Efficiency	DMN.LP (R)	0.21	2.66	14	0.0187
		SN.ACC	−0.13	−2.32	14	0.0356
		SMN. Lateral (R)	−0.17	−2.26	15	0.0389
	Degree
	Betweenness Centrality	DMN.LP (R)	−0.05	−2.22	16	0.0413
		SMN. Lateral (R)	0.05	2.15	16	0.0469
	Cost
E REST4 > REST2	Global Efficiency	VN.Occipital	−0.13	−2.8	16	0.0128
		SN.SMG (L)	−0.03	−2.69	16	0.0161
		SN.RPFC(L)	−0.03	−2.68	16	0.0164
		VN.Medial	−0.07	−2.2	16	0.0429
	Average Path Length	SN.SMG (L)	0.27	2.16	16	0.0459
	Clustering Coefficient
	Local Efficiency	SN.RPFC(R)	−0.16	−2.13	15	0.0498
	Degree	VN.Lateral (R)	1.82	4.41	16	0.0004
		VN. Medial	−1.18	−2.85	16	0.0116
		DMN.LP (R)	0.82	2.25	16	0.0389
	Betweenness Centrality	VN.Lateral (R)	0.07	4.35	16	0.0005
		VN.Lateral (L)	0.03	2.47	16	0.0252
		CEN.PPC(L)	−0.08	−2.27	16	0.0371
		SMN. Lateral (R)	0.03	2.2	16	0.0427
	Cost	VN.Lateral (R)	0.06	4.41	16	0.0004
		VN.Medial	−0.04	−2.85	16	0.0116
		DMN.LP (R)	0.03	2.25	15	0.0389
F REST4 > REST3	Global Efficiency	DMN.LP (R)	0.12	3.68	16	0.0021
		VN.Lateral (R)	0.09	3.14	16	0.0063
		DMN.LP (L)	0.07	2.19	16	0.0441
	Average Path Length	SN.RPFC(L)	0.43	3.7	16	0.0019
		SN.ACC	0.34	3.38	16	0.0037
		SN.RPFC(R)	0.32	2.5	16	0.0236
		SN.SMG (L)	0.39	2.38	16	0.0301
		SN.AI (R)	0.29	2.27	16	0.0372
	Clustering Coefficient	DMN.LP (R)	−0.28	−4.85	15	0.0002
		VN.Lateral (R)	−0.21	−2.44	16	0.0263
	Local Efficiency	DMN.LP (R)	−0.21	−3.53	15	0.0031
		VN.Lateral (R)	−0.18	−2.14	16	0.0485
	Degree	DMN.LP (R)	1.76	4.44	16	0.0004
		VN.Lateral (R)	1.59	3.32	16	0.0004
	Betweenness Centrality	VN.Lateral (R)	0.06	3.76	16	0.0017
		DMN.LP (R)	0.06	3.28	16	0.0047
		SN.SMG (R)	0.09	2.36	16	0.0313
		CEN.PPC(L)	−0.03	−2.36	16	0.0313
	Cost	DMN.LP (R)	0.06	4.44	16	0.0133
		VN.Lateral (R)	0.05	3.32	16	0.0688

IV. DISCUSSION AND CONCLUSION

Here we examined brain network topology in healthy women and explored the effect of visceral interoception (bladder fullness and catheterization) on resting state fMRI data. At the whole-brain level, we measured the functional integration, segregation, and centrality. Comparing REST3 (catheterized and partially-filled bladder) > REST 2 (catheterized and empty bladder), we observed decreased global efficiency in the right supramarginal gyrus of the SN (p = 0.0077), dorsolateral prefrontal cortex of the CEN (p = 0.0206), right posterior parietal cortex of the CEN (p = 0.0273), anterior DAN (left frontal eye field, p = 0.0386), and posterior DMN (right lateral parietal node, p = 0.0436). We also detected a decrease in the local efficiency of the anterior cingulate cortex of the SN (p = 0.0356) and right lateral SMN (p = 0.0389) (Fig. 3). Global efficiency measures the exchange of information across the whole network and the corresponding local efficiency is the inverse of the average shortest path connecting neighbors of nodes. The difference in the global and local efficiency can be interpreted as a change in the rate of information transfer across specific brain regions such as the frontoparietal and temporal regions as bladder fills up and interoceptive signals intensify. Moreover, degree (the number of connections for each node), and betweenness centrality (how connected a particular region is to other regions) differed between empty bladder, catheterized bladder, and catheterized bladder infused with 100 ml saline. Comparing before and after task conditions (i.e., F: REST4 > REST3), reveals increased average path length for all nodes of the salience networks (Fig 4) and decreased clustering coefficient (related to the functional specificity of regional brain areas) of DMN (Bonferroni-corrected p < 0.05).

Fig. 4.

Illustration of a increased average path length across SN nodes for REST4 (after task) > REST3 (before task) at at a liberal threshold of p < 0.05, uncorrected. Larger red nodes indicate stronger positive T-values. SN: salience network; ACC: anterior cingulate cortex; AI: anterior insula; RPFC: rostrolateral prefrontal cortex; SMG: supramarginal gyrus.

In conclusion, the present findings suggest alterations of topological properties of brain networks under different interoceptive rest conditions.

Fig. 2.

Illustration of a decrease in global efficiency when we compared REST3 (catheterized, partially filled bladder) to REST2 (catheterized, empty bladder) at a liberal threshold of p < 0.05, uncorrected. Larger blue nodes indicate stronger negative T-values. CEN: central executive network; DAN: dorsal attention network; DMN: default-mode network (DMN); SN: salience network; VN: visual network. For the ROI abbreviations see Table 1.