Impact of Intravesical Cold Sensation on Functional Network Connectivity Estimated Using ICA at Rest & During Interoceptive Task

Abstract

Afferent nerves that carry interoceptive signals from the viscera to the brain include Aδ and C-fibers. Previously, we examined the effects of detrusor distention (conveyed mainly by Aδ fibers) on the static functional network connectivity (FNC) of the brain using independent component analysis (ICA) of fMRI time series. In the present study, we investigate the impact of intravesical cold sensation (thought to be conveyed by C-fibers) on brain FNC using similar ICA approach. Thirteen healthy women were scanned on a 3.0T MRI scanner during a resting state scan and an intravesical cold sensation task fMRI. High dimensional ICA (n = 75) were used to decompose the fMRI data into several intrinsic connectivity networks (ICNs) including the default-mode (DMN), subcortical (SCN; amygdala, thalamus), salience (SN), central executive (CEN), sensorimotor (SMN), and cerebellar/brainstem (CBN) networks. Results demonstrate significant FNC differences in several ICN pairs primarily between the SCN and cognitive networks such as CEN, as well as between SN and CBN and DMN when intravesical cold water condition was compared to rest (FDR-corrected p-value of 0.05). Significant increases in FNC between CBN and between SMN were also observed during interoceptive condition. The results indicate significant impact of Aδ and C-fiber-originated interoceptive signals on the brain connectivity when compared to the baseline rest.

I. INTRODUCTION

The primary afferent fibers that transmit signals that represent the physiological status of the body (i.e., interoception) from the viscera to the central nervous system include Aδ and C-fibers [1]. These afferent fibers terminate on projection neurons in the lamina I of the spinal dorsal horn. The secondary neurons then project to the thalamus through ascending lateral spinothalamic tract [1]. Neuroimaging studies have shown that interoceptive signals eventually relay from the thalamus to higher areas in the brain including the insula and prefrontal cortices [2].

Aδ and C-fibers that innervate visceral organs such as bladder produce feelings from the body including pain, organ distention (fullness), abd temperature sensation [3]. The myelinated Aδ fibers, which are larger in diameter (2–5 μm) and have greater conduction velocities (2 to 30 m/sec) act as mechanoreceptors transmitting signals related to bladder distension and thus indicate the degree of bladder filling (i.e., urgency to void). The unmyelinated C-fibers, which are smaller (0.2–1.5 μm) with slower conduction velocities (<2 m/sec), are thought to transmit interoceptive signals caused by infection, thermal (e.g., cold) or chemical stimuli [4].

Previously we applied spatial independent component analysis (ICA) as a data-driven analysis technique [5] on blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) data to investigate the effect of intravesical distention (i.e., Aδ fiber activation) on the functional network connectivity (FNC [6]) of spatially distributed brain regions [7]. In this study, we investigate the effect of intravesical cold sensation, which is thought to activate C-fibers in the viscera, on the ICA-derived brain networks using similar approach. On the basis of differential afferent fiber activation and corresponding feelings from the bladder, we hypothesize that bladder cooling recruits a different cerebral network and thus different FNC patterns will be observed compared to bladder filling when each condition is compared to the baseline rest. The present analysis approach takes advantage of the multivariate statistical method (ICA-based MANCOVA [8]) of BOLD fMRI data to capture connectivity variability due to interoceptive sensations of a visceral organ.

II. MATERIALS AND METHODS

A. Participants and Imaging Protocol

Data were collected from 13 healthy right-handed female participants (mean age = 31.7, SD = 8.4 years). Each participant provided informed consents approved by the local Institutional Review Board, and underwent comprehensive clinical evaluation prior to their MRI scan. During the resting state fMRI scan, participants were asked to maintain fixation on a central cross and refrain from thinking of anything in particular or sleeping. The interoceptive task was adapted from previous studies of viscerosensory stimulation studies that consists of controlled filling and emptying of the bladder through intravesical catheter alternating with rest and ratings of pain and desire to void [9], [10]. To activate C-fibers, intravesical infusion was performed with cold saline solution (about 4 – 8°C) using custom-built MR-compatible devices [11], [12]. The fMRI paradigm (Fig. 1) consisted of 8 identical blocks. Each block began with intravesical infusion of 100 ml cold saline (24 s), followed by ratings of urgency (desire to void) and pain (15 s), passive intravesical drainage (60 s), and ended with rest (7 – 9 s).

Fig. 1.

Imaging protocol for the intravesical cold sensation fMRI

B. Neuroimaging Data Acquisition

BOLD fMRI was performed on a 3.0T Ingenia scanner using a gradient-echo echo-planar imaging (GE-EPI) sequence using the following parameters: TR/TE = 2000/40 ms, image matrix in plane = 96 × 96, flip angle = 80°, FOV = 240 mm, slice thickness = 3 mm, inter-slice gap = 1 mm, 34 axial slices in ascending order). Resting-state scan was 6 min, 42 s allowing collection of 200 volumes. Intravesical task fMRI scan was 15 minutes during which 500 volumes were collected. A 3D T1-weighted Ultrafast Gradient Echo (TFE) structural scan (TR/TE = 6.9/3 ms, matrix size = 256 × 256, voxel size = 1.0 mm³; 180 sagittal slices) was also acquired for coregistration purposes and to exclude any gross structural pathology.

C. Neuroimaging Data Preprocessing

Following MR image quality assessment in MRIQC [13], data were preprocessed using SPM12 (Wellcome Department of Cognitive Neurology, UK) in MATLAB (version 2016a,The Mathworks Inc., USA). Preprocessing pipeline included slice time correction, motion correction/realignment, coregistration of functional and structural data, and spatial normalization into the Montreal Neurological Institute (MNI) reference space. The data were spatial smoothing with 8 mm full-width half-maximum [FWHM] Gaussian kernel and data intensity was normalized and converted to percent signal change units to improve the accuracy and test-retest reliability of ICA [8].

D. Independent Component Analysis

Group ICA of fMRI Toolbox (GIFT; Medical Image Analysis Lab, University of New Mexico, Albuquerque, New Mexico, USA) [5] was used to decompose the preprocessed fMRI data to 75 independent components (ICs). In the first step, subject-specific data reduction PCA with a standard economy-size decomposition retained 100 principal components (PCs). In the second step, subject-specific reduced data were concatenated across time and group data reduction retained 75 PCs using the expectation-maximization algorithm. Spatial ICA was estimated using the infomax algorithm and repeated 20 times in ICASSO. In the final step, estimated ICs at the group level were back-reconstructed into subject-specific spatial maps and time series using GICA3 back-reconstruction method [8]). GICA3 method in GIFT software is similar to dual regression algorithm in FSL MELODIC pipeline. One-sample t-test was performed using SPM12 on IC spatial maps on a voxel-wise basis to determine regions of peak activation clusters across participants (p < 0.01, family-wise error [FWE]-corrected). ICs with ICASSO cluster quality index (Iq) below 0.8 were eliminated. Other artifactual ICs were identified based on their head motion-related spatial maps, localization of IC peak activation to the cerebrospinal fluid and ventricles, and time series dominated by high frequency spectral power. Established network templates (e.g., [8]) was also used to further identify the components that resembled known ICNs.

E. Functional Network Connectivity Analysis

Connectivity analysis was performed on IC time series with the MANCOVAN toolbox implemented in GIFT [8]. For each subject, the corresponding time series were detrended, despiked with AFNIs 3dDespike and bandpass filtered at [0.017–0.15 Hz]. A constrained maximal time-lagged correlation method was used to compute Pearson’s correlation values [6]. Let ρ be the correlation between two time series, $\overline{X}$ and $\overline{Y}$ of dimension T × 1 units, where T is the number of time points in the time series. Let i_o be the starting reference of the two time series, and Δi the non-integer change in time (in units of seconds). Assume $\overline{X}$ at initial reference time point $i_o\left({\overline{X}}_{i_o}\right)$, and $\overline{Y}$ circularly shifted Δi units from reference point of $i_o\left({\overline{Y}}_{i_o+\Delta i}\right)$ then ${\rho }_{i_o+\Delta i}$ at this configuration of the two time series can be calculated as (1):

$${\rho }_{i_o+\Delta i}=\frac{\left({\overline{X}}_{i_o}^{\text{T}}\right)\left({\overline{Y}}_{i_o+\Delta i}\right)}{\sqrt{\left({\overline{X}}_{i_o}^{\text{T}}{\overline{X}}_{i_o}\right)}\times \sqrt{\left({\overline{Y}}_{i_o+\Delta i}^{\text{T}}{\overline{Y}}_{i_o+\Delta i}\right)}}$$ (1)

Fisher r-to-z transformation was applied to normalize the FNC correlation values. To determine the significant differences in FNC between the resting state and interoceptive task, paired Student’s t-test was performed at the false discovery rate [FDR]-corrected threshold of 0.05.

III. RESULTS

Thirty four ICs were identified as ICNs (Fig. 2): (i) default mode networks (DMN; ICs 23, 31, and 58), (ii) subcortical networks (SCN, ICs 12, 25 and 43), (iii) salience networks (SN; ICs 2, 9, 18, 47, and 67), (iv) cognitive and attention networks (CAN) including the self-referential network (IC 3), frontoparietal central executive networks (CENs; ICs 41 and 48), ventral and dorsal attention networks (ICs 62 and 64, respectively), and frontal networks (ICs 50 and 51), (v) auditory network (AUD; IC 56), (vi) sensorimotor networks (SMN; ICs 26, 37, and 66), (vii) visual networks (VN; ICs 19, 29, 35, and 40), and (viii) cerebellar/brainstem network s(CBN; ICs 1, 7, 8, 14, 24, 46, and 59). Significant FNC differences between the rest and interoceptive task conditions at p < 0.05 (FDR corrected) are shown for all ICNs (Fig. 3a), and for network average (Fig. 3b). The changes in functional connectivity are also displayed in the connectogram (Fig. 4).

Fig. 2.

Special maps of thirty four ICNs displayed at the three most informative slices in neurological convention (right = right).

Fig. 3.

Significant effects of rest – intravesical cold sensation (FDR corrected-p < 0.05) A All ICNs B Network average.

Fig. 4.

Connectogram showing significant effects of rest – intravesical cold sensation on FNC (FDR corrected-p < 0.05)

IV. Discussion & Conclusions

Our goal was to characterize the connectivity changes due to interoceptive task involving intravesical infusion of cold saline to see how the connectivity patterns differ from detrusor distention. Similar to detrusor distention, the majority of FNC differences that survived the FDR-threshold involve the affective and cognitive-motivational ICNs. For example, several cognitive networks such as CEN showed differential FNC during both interoceptive tasks compared to rest. CEN, which includes dorsolateral prefrontal cortex and posterior parietal regions, is involved in executive function such as working memory, set switching, and inhibitory control [14]. However, there were significant FNC changes involving SCN during intravesical cold sensation. We observed changes in FNC between IC 12 (hippocampus-amygdala) and IC 58 (DMN), and IC 14 (CBN), and between IC 26 (thalamus) and IC 21 (prefrontal network). Several CBN also showed significant FNCs that included connectivity within CBN (i.e., FNC between ICs 7 and 24) as well as between CBN nd SCN as mentioned above (i.e., FNC between ICs 12 and 14).

C-fibers in bladder are regarded as silent. However, under certain neurological (e.g., spinal cord injury) or pathological conditions (e.g., overactive bladder) changes in C-fiber sensitization occurs. Understanding the differences of Aδ vs. C-fiber activation on the brain networks can help to shed light on the neural correlates of bladder dysfunctions, such as increased urination frequency, urgency, and pain.