Neuroscience Letters 577 (2014) 83–88

Contents lists available at ScienceDirect

Neuroscience Letters journal homepage: www.elsevier.com/locate/neulet

Serial processing in primary and secondary somatosensory cortex: A DCM analysis of human fMRI data in response to innocuous and noxious electrical stimulation Mina Khoshnejad a,c,d , Mathieu Piché d,e,g , Soha Saleh f , Gary Duncan b,c , Pierre Rainville b,c,d,∗ a

Department of neuroscience Université de Montréal, Montreal, QC, Canada Department of Stomatology, Université de Montréal, Montréal, QC, Canada c Groupe de recherche sur le système nerveux central (GRSNC), Université de Montréal, Montréal, QC, Canada d Centre de recherche de l’Institut universitaire de gériatrie de Montréal (CRIUGM) Université de Montréal, Montréal, QC, Canada e Centre de Recherche en Neuropsychologie et Cognition (CERNEC), Université de Montréal, Montréal, QC, Canada f Department of Biomedical Engineering, Lebanese International University, Beirut, Lebanon g Department of Chiropractic, Université du Québec à Trois-Rivières, Trois-Rivières, QC, Canada b

h i g h l i g h t s • • • •

We explored parallel versus serial mode of processing in human somatosensory cortex. DCM was applied to fMRI brain responses to cutaneous electrical stimulation. Results support serial processing from S1 to S2 at innocuous and noxious intensities. Connectivity patterns change as a function of pain at the more intense level.

a r t i c l e

i n f o

Article history: Received 15 December 2013 Received in revised form 21 May 2014 Accepted 6 June 2014 Available online 13 June 2014 Keywords: Electrical stimulations Somatosensory cortex Serial processing DCM Intensity encoding

a b s t r a c t The anatomy of the somatosensory system allows both serial and parallel information flow but the conditions involving each mode of processing is a matter of debate. In this functional magnetic resonance imaging (fMRI) study, cutaneous electrical stimulation was applied to human volunteers at three intensities (low-innocuous, moderate-noxious and high-noxious) to investigate interactions between contralateral primary and secondary somatosensory cortices (S1c and S2c), and between contralateral and ipsilateral S2 (S2c and S2i), using dynamic causal modeling (DCM). Our results are consistent with serial processing with a key role of the direct input to S1c for all three intensity levels. The more intense stimulus also induced significantly more interactions between S2i and S2c, consistent with an increase in inter-hemispheric integration associated with the additional recruitment of nociceptive inputs. However, stronger pain reports were also associated with reduced information flow from S1c to S2c at both the moderate (r = −0.81, p = 0.004) and the high stimulation level (r = −0.63, p = 0.037). These findings suggest that the connectivity pattern driven by innocuous inputs is modified by the additional activation of nociceptive afferents. © 2014 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

∗ Corresponding author at: Département de Stomatologie, Faculté de médecine dentaire, Université de Montréal, Montréal QC, Canada H3W1W4. Tel.: +1 514 343 6111x3935. E-mail address: [email protected] (P. Rainville). http://dx.doi.org/10.1016/j.neulet.2014.06.013 0304-3940/© 2014 Elsevier Ireland Ltd. All rights reserved.

Somatosensory areas S1 and S2 are the two principal cortical areas implicated in somatosensation. Anatomical evidence provides support for both serial and parallel modes of processing in the somatosensory cortices. For example, studies of non-human primates show that different thalamic nuclei project in parallel to S1 and S2 [1]. On the other hand, anatomical evidence in primates also show that S1 is connected to S2 via reciprocal cortico-cortical

84

M. Khoshnejad et al. / Neuroscience Letters 577 (2014) 83–88

connections [11], and therefore, pain and tactile information can be conveyed to S2 by this indirect serial pathway from the thalamus via S1. Consistent with anatomical studies, human functional brain imaging demonstrates robust activation of S1 and S2 to innocuous tactile stimulation [15,16,19,21,24]. Robust activation of S1 and S2 has also been reported during nociceptive stimulation, regardless of the stimulus modality [2,6]. Activation levels in both S1 and S2 generally code for the intensity of tactile [8,28] and noxious stimulations [5,18,23,27], but basic analyses of brain activation studies provide no information on the interactions between the regions activated. Human imaging studies have used effective connectivity approaches to investigate functional integration among somatosensory regions. One MEG study using Granger Causality found causal influence of S1 on S2 only for tactile-related but not for pain-related activations (laser stimuli), supporting serial and parallel processing for tactile and pain information, respectively [22]. However, a fMRI study [17] using dynamic causal modeling (DCM) suggests that parallel processing dominates for both tactile (innocuous electrical pulses) and noxious heat (laser) stimuli. Another study [12] points to serial processing of vibrotactile information in the somatosensory cortices. Yet, another one [4] shows parallel input to S1 and S2 as well as modulation in the serial pathway from S1c to S2c using pressure stimuli. Taken together, these studies provide evidence for both serial and parallel processing that may be dependent upon a variety of factors including the type and intensity of stimuli. Another aspect that has yet to be examined in studies using connectivity approaches is the interactions with ipsilateral sensory regions. Although pain and tactile-related activation is predominantly observed in S1 and S2 contralateral to the stimulus, ipsilateral activations are also frequently observed, especially in S2 [5], consistent with studies showing the existence of large number of cortical somatosensory neurons with bilateral receptive fields [9,10], as well as lesion studies showing the involvement of ipsilateral regions in somatosensation [14,25]. The origin of ipsilateral activations is debated but early anatomical investigations on non-human primates strongly support the hypothesis that callosal connections are largely, although maybe not exclusively, responsible for sensory activation of ipsilateral cortical regions [13]. The inclusion of ipsilateral sites in effective connectivity models may provide further insight on somatosensory integration. The purpose of the present study was to test a number of causal hypotheses regarding functional interactions between S1 and S2, based on brain activity measured using fMRI in response to transcutaneous electrical stimulation at three intensity levels. Our causal hypotheses examined the relationship between S1 and S2 contralateral to the stimulus (S1c and S2c), as well as inter-hemispheric connections of S2c with ipsilateral S2 (S2i). DCM analyses allowed testing parallel versus serial processing of somatosensation and assessing the effect of stimulus intensity and pain perception on patterns of connectivity. 2. Materials and methods All experimental procedures conformed to the standards set by the latest revision of the Declaration of Helsinki and were approved by the Research Ethics Board of “Institute Universitaire de Geriatrie de Montréal”. Eleven healthy volunteers participated in the study (three males and eight females; mean age: 26.9 years; SD: 4.7). All participants gave written informed consent, and received a monetary compensation for their participation. Detailed methodological information is provided in a previous report of this study [21]. Here, we briefly outline the stimulation and fMRI acquisition methods and we describe the DCM analysis in more details.

2.1. Stimulation paradigm Transcutaneous electrical stimulation was delivered with a Grass S48 square-pulse stimulator (Astro-Med Inc., West Warwick, RI, USA) connected to a custom-made constant-current stimulus-isolation unit. The stimulation consisted of a 30 ms train of 10 × 1 ms pulses (333 Hz) delivered over the retro-maleolar path of the right sural nerve. The intensity of the stimulus was adjusted individually according to the subject’s RIII reflex thresholds (mean ± SD: 10.4 ± 4.5 mA). Low, moderate and high intensity stimulation (80%, 120%, and 150% of the individual RIII-threshold, respectively), were applied in separate scans (order counterbalanced between subjects). Note that responses produced by low intensity stimulation of the sural nerve (a cutaneous nerve) reflect innocuous A-beta activity while moderate and high levels reflect the additional, but not exclusive, recruitment of nociceptors. In each functional scan, 40 stimuli were delivered with a pseudorandom ISI of 6, 9, 12 or 15 s. Participants rated pain intensity after each run on a visual analog scale (0–100 VAS). The innocuous stimulus condition (low level) was never rated as painful and the moderate and high stimulus intensities were rated (mean ± SD) 31.8 ± 12.3 and 53.9 ± 22.7, respectively. 2.2. fMRI acquisition Imaging data was acquired at the “Centre de recherche de l’Institut de gériatrie de Montréal” on a 3 T Siemens Trio scanner (Munich, Germany) using a CP head coil. The anatomical scans were T1-weighted high-resolution scans [repetition time (TR): 13 ms; echo time (TE): 4.92 ms; flip angle: 25◦ ; field of view: 256 mm; voxel size: 1 × 1 × 1.1 mm]. The functional scans used a blood oxygen level-dependent (BOLD) protocol with a T2* weighted gradient echo-planar imaging sequence (TR: 3.0 s with an intervolume delay of 500 ms; TE: 30 ms; flip angle: 90◦ ; 64 × 64 matrix; 130 volume acquisitions). The scanning planes were oriented parallel to the anterior–posterior commissure line and covered the entire brain (41 contiguous 5-mm-thick slices; voxel size, 3.44 × 3.44 × 5 mm). Electrical stimuli were always administered during the 500 ms inter-volume delay, as described in a previous report [21]. 2.3. GLM analysis The GLM analysis was conducted using SPM2 running in Matlab version 7.1 to select the proper ROIs for DCM analysis in the target areas S1c, S2c and S2i (Section 2.3.1). Pre-processing included slice-time correction and realignment. Anatomical and functional images were then spatially normalized to a standard stereotaxic space using the MNI template. Subsequently, functional images were spatially smoothed using a Gaussian kernel twice the voxel size (FWHM: 7 × 7 × 10 mm) and temporally filtered using a highpass filter with a cut-off period of 128 s. They were then corrected for serial autocorrelation using the AR (1) correction. Stimulusrelated activity was identified individually with an event-related design by convolving each stimulus with a canonical hemodynamic response function. The shock-evoked responses were assessed by a random-effect one-sample t-test, using images of stimulus-evoked responses from each subject and a whole-brain FDR-corrected threshold of p < 0.05. 2.3.1. ROI definition The ROIs were selected based on (a) the GLM group maps, (b) the location of peak voxels in individual runs, and (c) the individual anatomical landmarks. The group analysis showed significant shock-evoked activation in the three target areas (S1c, S2c and S2i) and in each of the three stimulation conditions. The activation maps

M. Khoshnejad et al. / Neuroscience Letters 577 (2014) 83–88

of each individual for each stimulation level were then thresholded at a permissive p < 0.05 to identify activated voxels within the target anatomical structures and in the vicinity of the group activation peak. The peak activation voxel in each area and for each run was selected as the center of the ROI. All three regions were activated for high, and moderate intensity electric stimulations but two subjects did not show activation for low intensity stimulation for at least one of the three regions at p < 0.05. Therefore these two subjects were excluded in the DCM analysis for low intensity stimulation. Time–course data were determined by the average of the first eigenvariates of the stimulus-evoked GLM responses across voxels contained within a radius of 5 mm from the center of the ROI. The coordinates of the ROIs are given in Supplementary Table S1. 2.4. Effective connectivity analysis using dynamic causal modeling (DCM) The time course of each ROI in each condition and subject provided the input for DCM analysis. In brief, DCM utilizes a causal model by which neuronal activity in a given region causes changes in neuronal activity in other regions via a pattern of interregional connections and in its own activity by local selfregulatory connections. Additionally, any of these connections can be modulated by experimental manipulation such as sensory inputs. The connections are parameterized, and are estimated using a Bayesian scheme. The estimated connectivity parameters are used to infer the strength of neuronal communication among modeled brain regions. Finally, several competing hypotheses about integration among modeled brain regions can be compared, within the Bayesian framework [7,26]. Eq. (1) summarizes how DCM models the neural dynamic among interacting regions. According to this model, the change in neural activity (z) at any given region is explained in terms of connectivity parameters (A, B and C) as well as the experimental input (u). dz = A + (Bz + C)u dt

(1)

A (intrinsic connectivity) contains fixed connectivity parameters among the modeled regions. B (modulatory connectivity) parameterizes the modulation of intrinsic connections by any experimental manipulations (here the stimulus). Lastly C (extrinsic connectivity) depicts the strength of direct driving inputs to the modeled system [7]. 2.4.1. DCM hypotheses We tested a set of models of integration between S1c, S2c and S2i. We assumed reciprocal intrinsic connections (A parameters) between S1 and S2 in the hemisphere contralateral to the stimulation as well as reciprocal callosal connection between homologous regions of S2 (Fig. 1). Intrinsic self-connections were also included in all models as this is an assumption in DCM [7]. Based on the description of dynamics of the interacting regions, the sensory inputs can act directly on some specific anatomical regions (C parameters), or they can evoke response through modulation of coupling among different anatomical regions (B parameters). We explored three families of models based on where the driving input could directly exert its effects: (1) only to S1c (Family 1), (2) to S1c and S2c (Family 2), or (3) to S1c, S2c and S2i (Family 3). The three tested families, allowed examining the issue of serial versus parallel mode of processing, as we directly tested whether the thalamo-cortical input to S2c is redundant, given the input coming from S1c (Family 1), or if a parallel input to S2c is indeed necessary to explain the responses in S2c (Family 2). Similarly, we tested whether ipsilateral activation in S2 is better accounted for by including a direct input to S2i (Family 3) as compared to indirect input from S2c (Families 1 and 2).

85

In terms of changes in coupling between the selected ROIs in response to the electrical stimulation, we explored 6 basic models (see Fig. 1B) all sharing the same structure. These models included combinations of connections between S1c and S2c and between S2c and S2i. Six additional models tested the additional changes in self-recursive connectivity within each of the ROIs. In total, we investigated 12 × 3 = 36 models. The same 36 models (i.e. 12 models per family), were estimated separately for each level of stimulations; low, moderate and high. 2.4.2. Bayesian model selection and comparison For each tested model, DCM computes its statistical evidence using Bayesian scheme. Based on the estimated model evidence of each model, Bayesian model selection (BMS) calculates the probability of each model being more likely than any other tested model (exceedance probability (EP)). The model with the highest EP is considered as the best model. For comparing model families, all models within a family are averaged using Bayesian model averaging (BMA) and the exceedance probabilities are calculated for each model family [20]. BMS using random effect (RFX) was performed to determine the best model family for each intensity level (low, moderate and high). BMA values were used to ascertain which parameter is expressed consistently across subjects for each level of stimulation using one sample t-test. In addition BMA values were used to compare the strength of connectivity parameters, across the three intensity levels using ANOVA. In order to relate our findings more directly to pain-related processes, the between-subject correlation of connectivity parameters with pain ratings was tested at the two noxious levels using exploratory Pearson correlation. 3. Results 3.1. GLM analysis The classical GLM analysis confirmed that the electrical stimulus produced a brain activation pattern consistent with previous imaging studies of painful and non-painful sensations. The group activation map of the target areas is depicted in Fig. 1A and shows the location of peak activation in S1c, S2c and S2i used as a reference to determine and extract individual ROIs values (see Section 2.3.1, ROI definition and Supplementary Table S1). 3.2. DCM model comparison and connectivity parameters The results show that for all three levels of stimulation, Family 1 clearly wins over Families 2 and 3 with exceedance probability above 0.9 (Fig. 2A). This suggests that direct sensory input only to S1c for all three levels of stimulations is sufficient for response estimation in accordance with serial mode of processing. Among the 12 models tested within Family 1, there was no clear evidence of an optimal model for low intensity, but model 1 had the highest EP of 0.17. For moderate intensity, model 1 surpassed the other models with EP of 0.41 while all the other models had EP