Attention modulates the gating of primary somatosensory oscillations

Abstract

Sensory gating (SG) is a well-studied phenomenon in which neural responses are reduced to identical stimuli presented in succession, and is thought to represent the functional inhibition of primary sensory information that is redundant in nature. SG is traditionally considered pre-attentive, but little is known about the effects of attentional state on this process. In this study, we investigate the impact of directed attention on somatosensory SG using magnetoencephalography. Healthy young adults (n = 26) performed a novel somato-visual paired-pulse oddball paradigm, in which attention was directed towards or away from paired-pulse stimulation of the left median nerve. We observed a robust evoked (i.e., phase-locked) somatosensory response in the time domain, and three stereotyped oscillatory responses in the time-frequency domain including an early theta response (4–8 Hz), and later alpha (8–14 Hz) and beta (20–26 Hz) responses across attentional states. The amplitudes of the evoked response and the theta and beta oscillations were gated for the second stimulus, however, only the gating of the oscillatory responses was altered by attention. Specifically, directing attention to the somatosensory domain enhanced SG of the early theta response, while reducing SG of the later alpha and beta responses. Further, prefrontal alpha-band coherence with the primary somatosensory cortex was greater when attention was directed towards the somatosensory domain, supporting a frontal modulatory effect on the alpha response in primary somatosensory regions. These findings highlight the dynamic effects of attentional modulation on somatosensory processing, and the importance of considering attentional state in studies of SG.

1. Introduction

Sensory gating (SG) is a robust phenomenon whereby neural responses to identical stimuli are reduced when presented in rapid succession. This phenomenon has been widely studied in the auditory (Adler et al., 1982; Davies et al., 2009; Weiland et al., 2008) and somatosensory (Cheng et al., 2015b; Hsiao et al., 2013; Kurz et al., 2017; Spooner et al., 2018; Spooner et al., 2019; Thoma et al., 2007; Wiesman et al., 2016) systems, and is commonly interpreted as representing the “filtering” of redundant stimulus features at an early level of processing. Traditionally, SG has been considered a pre-attentive process in the human brain (Grunwald et al., 2003). Despite this consideration, very few studies have examined whether differences in attentional state directly modulate SG. This is surprising, as a number of studies have reported robust interactions between neuropsychological measures of attention function and SG, such that reduced attentional capacity is related to reductions in SG. For instance, stronger sensory gating has been linked with reduced distractibility and faster reaction times on the continuous performance task of sustained attention (Jones et al., 2016; Karper et al., 1996; Lijffijt et al., 2009a), as well as enhanced performance on the Posner attentional orienting task (Karper et al., 1996), the Stroop cognitive interference task (Wan et al., 2008), and the Attention Network Task (Jones et al., 2016; Wan et al., 2008). Beyond these indirect links to neuropsychology, only a handful of studies (Gjini et al., 2011; Golubic et al., 2019; Kho et al., 2003; Rosburg et al., 2009) have examined the neural dynamics of SG across differing attentional states. Generally, these studies have found no significant effect of attention on the gating of primary sensory responses, however, none of these studies have examined this potential effect in the somatosensory domain or comprehensively examined the role of neural oscillations. An enhanced understanding of the potential effects of attention on SG is essential to better understand the basic neurophysiology of the human brain, as well as to aid in interpretation of aberrant SG in patient populations.

It also remains unknown whether the gating of the evoked (i.e., phase-locked) and multi-spectral oscillatory neural responses serving somatosensory processing (Bardouille et al., 2010; Haegens et al., 2011; Haegens et al., 2012; Hari and Forss, 1999; Hlushchuk and Hari, 2006; Huttunen et al., 2008; Jones et al., 2010; Kanno et al., 2003; Schnitzler et al., 1995; Thoma et al., 2007; van Ede et al., 2010; van Ede et al., 2014; Wiesman et al., 2016) are differentially impacted by attention. In general, early-latency evoked and low-frequency theta synchronizations are thought to index the processing of incoming somatosensory stimulus information in a “bottom-up” manner (Andersen and Lundqvist, 2019; Hari and Forss, 1999; Hlushchuk and Hari, 2006; Hsiao et al., 2013; Huttunen et al., 2008; Kanno et al., 2003; Schnitzler et al., 1995; Thoma et al., 2007; Wiesman et al., 2016). In contrast, later-latency desynchronizations in the alpha and beta frequencies following somatosensory stimulation have been robustly tied to the “top-down” processing of this information in relation to context-specific task demands, and appear to be modulated by the direction of attention towards the somatosensory domain (Bardouille et al., 2010; Dockstader et al., 2010; Haegens et al., 2011; Haegens et al., 2012; Jones et al., 2010; van Ede et al., 2010; van Ede et al., 2014; Whitmarsh et al., 2014). In light of these previous findings, it seems likely that the gating of these differing responses would be affected by attention in opposing directions. In addition, the direction of such effects would provide clarification regarding the functional nature of these responses.

A significant volume of research has also been devoted to the study of SG in clinical populations and as a function of healthy aging. Perhaps most auspiciously, both auditory and somatosensory gating have been found to be aberrant in patients with schizophrenia (Adler et al., 1992; Adler et al., 1998; Adler et al., 1982; Cromwell et al., 2008; Thoma et al., 2007), and somatosensory gating deficits have been reported in cerebral palsy (Kurz et al., 2017) and HIV-associated cognitive dysfunction (Spooner et al., 2018). These aberrations have widely been interpreted to represent an inability by these patients to suppress non-salient sensory information, which could then lead to common disease sequelae such as degraded perception and even hallucinations. In addition, somatosensory gating is often found to decrease as age increases in healthy adults (Cheng et al., 2015a; Cheng et al., 2015b; Cheng and Lin, 2013; Spooner et al., 2019), suggesting a degradation of somatosensory processing as age progresses. Importantly, in the majority of clinical populations commonly studied with SG paradigms (Ally et al., 2006; Kurz et al., 2017; Lijffijt et al., 2009b; Spooner et al., 2018; Thomas et al., 2010), attentional deficits are also consistently reported. This is problematic, as the effects of directed attention on SG are not well studied, particularly in the somatosensory domain.

In this study, we investigate the interaction between directed attention and somatosensory gating, as measured with magnetoencephalography (MEG). Twenty-six healthy young adults performed a novel somato-visual paradigm designed expressly for this purpose during an MEG recording, whereby alertness was held constant and attention was either directed towards or away from a paired-pulse somatosensory stimulation applied to the left median nerve. We hypothesized that SG would be significantly altered when attention was directed away from the somatosensory stimuli, relative to when it was directed towards the stimuli. Specifically, we predicted that attention would enhance SG of neural somatosensory responses that are known to index early, “bottom-up” stimulus processing, including the initial evoked broadband and early theta-frequency responses (Hari and Forss, 1999; Hlushchuk and Hari, 2006; Huttunen et al., 2008; Kanno et al., 2003; Thoma et al., 2007; Weiland et al., 2008; Zakharova et al., 2016). Conversely, we predicted that attention would reduce SG of somatosensory responses thought to represent “top-down” integration with executive systems, and in particular neural activity in the alpha and beta bands, where attention has been repeatedly found to have a robust influence (Bardouille et al., 2010; Dockstader et al., 2010; Haegens et al., 2011; Haegens et al., 2012; Jones et al., 2010; van Ede et al., 2010; van Ede et al., 2014; Whitmarsh et al., 2014). Along these lines, we also expected that coherent neural activity in the alpha and beta frequency bands might facilitate communication between prefrontal attention cortices and primary somatosensory cortex.

2. Materials and Methods

2.1. Participants

We enrolled 26 healthy young adults (mean age = 24.04 years; SD = 3.22 years; range = 19–31 years; 12 males/14 females) for participation in this study. Exclusionary criteria included any medical illness affecting CNS function, any neurological disorder, history of head trauma, any non-removable metal implant that would adversely affect data acquisition, and current substance abuse. The Institutional Review Board at the University of Nebraska Medical Center reviewed and approved this investigation. After complete description of the study, written informed consent was acquired from each participant. All participants had normal or corrected-to-normal vision. All participants completed the same experimental protocol.

2.2. Experimental Paradigm

Participants were seated in a custom-made nonmagnetic chair with their head positioned within the MEG sensor array. During the scan, participants performed a novel somato-visual oddball paradigm, aimed at systematically dividing attention between the somatosensory and visual domains during paired-pulse somatosensory stimulation (Figure 1). This novel paradigm is the first to systematically manipulate attention towards and away from the somatosensory domain during concurrent paired-pulse stimulation, and was developed expressly for the purpose of this study. Stimuli from these two sensory modalities were presented in alternation, and a small proportion of the stimuli from each modality were temporal “oddballs,” which were utilized to monitor behavior and ensure that attention was directed towards either the visual or somatosensory domain. The visual stimulus consisted of a right-lateralized circle centered on the horizontal axis and to the right of a centrally-presented fixation crosshair. In 80 of the 88 total visual trials, this stimulus was presented for a duration of 500 ms, and for the other eight “oddball” trials it was presented for 1000 ms. The somatosensory stimulus consisted of a paired-pulse delivered using unilateral electrical stimulation to the median nerve of the left hand. For each participant, 80 paired-pulse trials were collected using an inter-stimulus interval of 500 ms, while the remaining eight “oddball” somatosensory trials used an inter-stimulus interval of 1000 ms. Visual and somatosensory trials were presented in alternation for a total of 160 trials in a single block (i.e., 80 somatosensory and 80 visual). The inter-pair interval (IPI) between somatosensory paired-pulses was 5300 ± 400 ms (randomly jittered to prevent anticipatory effects; not accounting for the additional 500 ms present on the eight oddball trials out of the total 88 visual trials) and the inter-modality interval (IMI) between visual and somatosensory stimuli was 2400 ± 200 ms. Each participant performed two blocks of the experiment (i.e., 320 total trials), and the only difference between the two blocks was the instructions given (i.e., “respond to somatosensory oddballs” versus “respond to visual oddballs”). In the “visual” block, participants responded to only the visual oddballs, and were told to ignore the task-irrelevant somatosensory stimuli. Conversely, in the “somatosensory” block, participants were told to respond only to the somatosensory oddballs, and to ignore the task-irrelevant visual stimuli. Importantly, participants were required to fixate on the centrally-presented crosshair and keep their left arm still for the entirety of both blocks. The order of the blocks was counterbalanced across participants. Participants used a MEG-compatible five-finger response pad to respond to the occurrence of the oddball stimuli, using their right index finger. Total MEG recording time was approximately 20 minutes per participant.

Figure 1. Somato-visual oddball task design.

Each participant performed one counterbalanced 88-trial blocks of the experiment per attention condition (i.e., 176 total trials). The task consisted of interspersed somatosensory paired-pulse (inter-stimulus interval (ISI): 500 ms) and visual stimuli (duration: 500 ms), separated by a variable inter-modality interval (IMI) of 2400 ± 200 ms. Eight of the total 88 stimuli per modality were oddballs (somatosensory ISI: 1000 ms; visual duration: 1000 ms), and participants only responded to oddballs in one modality per block, depending on the condition. These conditions only differed in the instructions given (i.e., “respond to the somatosensory oddballs” versus “respond to the visual oddballs”), and the visual fixation was present for the entire duration of the task. IPI: interpair interval.

For the somatosensory stimuli, mild electrical stimulation was delivered using external cutaneous stimulators connected to a Digitimer DS7A constant-current stimulator system (Digitimer Limited, Letchworth Garden City, UK). Each pulse was comprised of a 0.2 ms constant-current square wave that was set to 10% above the motor threshold required to elicit a subtle twitch in the thumb, and the same stimulation amplitude was used in both blocks for each participant. The rationale for stimulating above the motor threshold in this study was twofold: (1) this approach allows us to establish that the median nerve is, in fact, being specifically targeted by the stimulation (i.e., by observing a subtle twitch of the thumb), and (2) this allows us to ensure that this nerve-specificity does not fail or become significantly degraded during the course of the ~20 minute experiment. This second point is especially important here, since the same stimulation site and intensity was used for each participant across the counterbalanced blocks, so as to allow for statistical comparison across the attention conditions. A 500 ms inter-stimulus interval (ISI) between the pulses was chosen, as this is the interval most commonly used in previous research, and is known to elicit robust SG responses (Adler et al., 1982; Arpin et al., 2017; Bak et al., 2011; Hsiao et al., 2013; Kurz et al., 2017; Spooner et al., 2018; Spooner et al., 2019; Wiesman et al., 2016). Custom visual stimuli were programmed in Matlab (Mathworks, Inc., Massachusetts, USA) using Psychophysics Toolbox Version 3 (Brainard, 1997) and back-projected onto a semi-translucent non-ferromagnetic screen at an approximate distance of 1.07 meters, using a Panasonic PT-D7700U-K model DLP projector with a refresh rate of 60 Hz and a contrast ratio of 4000:1.

2.3. MEG Data Acquisition

Our MEG data acquisition, structural coregistration, preprocessing, and sensor-/source-level analyses closely followed the analysis pipeline of previous manuscripts (Kurz et al., 2017; Spooner et al., 2018; Spooner et al., 2019; Wiesman et al., 2016). All recordings were conducted in a one-layer magnetically-shielded room with active shielding, based on measurements taken from several magnetometers within the MEG array, engaged for environmental noise compensation. Neuromagnetic responses were sampled continuously at 1 kHz with an acquisition bandwidth of 0.1– 330 Hz using a 306-sensor Elekta/MEGIN MEG system (Helsinki, Finland) equipped with 204 planar gradiometers and 102 magnetometers. Participants were monitored during data acquisition via real-time audio-video feeds from inside the shielded room. Each MEG dataset was individually corrected for head motion and subjected to noise reduction using the signal space separation method with a temporal extension (correlation limit: .950; correlation window duration: 6 seconds; Taulu and Simola, 2006). Only data from the gradiometers were used for further analysis.

2.4. Structural MRI Processing and MEG Coregistration

Preceding MEG measurement, four coils were attached to the participant’s head and localized, together with the three fiducial points and scalp surface, using a 3-D digitizer (Fastrak 3SF0002, Polhemus Navigator Sciences, Colchester, VT, USA). Once the participant was positioned for MEG recording, an electric current with a unique frequency label (i.e., 293, 307, 314, and 321 Hz) was fed to each of the coils. This induced a measurable magnetic field and allowed each coil to be localized in reference to the sensors throughout the recording session. Since coil locations were also known in head coordinates, all MEG measurements could be transformed into a common coordinate system. With this coordinate system, each participant’s MEG data were co-registered with structural T1-weighted MRI data using BESA MRI (Version 2.0) prior to source-space analysis. Structural MRI data were aligned parallel to the anterior and posterior commissures and transformed into Talairach space. Following source analysis (i.e., beamforming), each participant’s 4.0 × 4.0 × 4.0 mm functional images were also transformed into Talairach space using the transform that was previously applied to the structural MRI volume and spatially resampled.

2.5. MEG Preprocessing, Time-Frequency Transformation, and Sensor-Level Statistics

Cardiac and blink artifacts were removed from the data using signal-space projection (SSP), which was subsequently accounted for during source reconstruction (Uusitalo and Ilmoniemi 1997). The continuous magnetic time series was then filtered between 0.5 – 200 Hz plus a 60 Hz notch filter, and divided into 2500 ms epochs, with the baseline extending from −500 to 0 ms prior to the onset of the somatosensory paired-pulse stimuli. It should be noted that only the “short” duration (i.e., 500 ms) paired pulse somatosensory trials were considered in this analysis, and the oddball trials were excluded entirely. The visual stimulation trials were also excluded. Epochs containing artifacts were rejected using a fixed threshold method, supplemented with visual inspection. Briefly, in MEG, the raw signal amplitude is strongly affected by the distance between the brain and the MEG sensor array, as the magnetic field strength falls off sharply as the distance from the current source increases. To account for this source of variance across participants, as well as actual variance in neural response amplitude, we used an individually-determined threshold based on the signal distribution for both signal amplitude and gradient to reject artifacts. Across all participants, the average amplitude threshold was 947.79 (SD = 157.74) fT and the average gradient threshold was 135.42 (SD = 35.01) fT/s. Across the group, an average of 71.48 (SD = 1.93) trials per participant per condition (out of 80 possible trials) were used for further analysis. Importantly, none of our statistical comparisons were compromised by differences in trial number nor artifact thresholds, as none of these metrics significantly differed across attention conditions (trial number: p = .479, BF₀₁ = 3.81; amplitude threshold: p = .291, BF₀₁ = 2.86; gradient threshold: p = .404, BF₀₁ = 3.48).

The epochs remaining after artifact-rejection were averaged across trials to generate a mean time series per sensor, and the specific time windows used for subsequent source analysis were determined by statistical analysis of the sensor-level time series across all conditions and the entire array of gradiometers. Each data point in the time series was initially evaluated using a mass univariate approach based on the general linear model. To reduce the risk of false positive results while maintaining reasonable sensitivity, a two-stage procedure was followed to control for Type 1 error. In the first stage, paired-sample t-tests were conducted to test for differences from baseline at each data point and the output time series of t-values was thresholded at p < .001 to define time-points containing potentially significant responses across all participants. In stage two, the time points that survived the threshold were clustered with temporally and/or spatially neighboring time points that were also above the threshold (p < .001), and a cluster value was derived by summing all of the t-values of all data points in the cluster. Nonparametric permutation testing was then used to derive a distribution of cluster values and the significance level of the observed clusters (from stage one) were tested directly using this distribution (Ernst 2004; Maris and Oostenveld 2007). For each comparison 10,000 permutations were computed to build a distribution of cluster values, and the time windows of phase-locked, time-domain data that were non-exchangeable with baseline across all participants according to these permutation analyses were used to guide subsequent time domain source level analysis.

To investigate the oscillatory responses commonly associated with somatosensory processing, we next transformed the same post-artifact-rejection epochs into the time-frequency domain using complex demodulation (Hoechstetter et al., 2004; Kovach and Gander, 2016; Papp and Ktonas, 1977). Briefly, complex demodulation works by first transforming the signal into the frequency space, using a Fast Fourier Transform (FFT). This results in a frequency spectrum, inherently containing the same power and cross spectrum information as the original signal. From here, this frequency spectrum is (de)modulated in a step-wise manner to adopt the center frequency of a series of complex sinusoids with increasing carrier frequencies, in a process termed “heterodyning.” These resulting signals are then low-pass filtered to reduce spectral leakage, and thus the nature of this filter inherently determines the time and frequency resolution of the resulting data. For this study, the time-frequency analysis was performed with a frequency-step of 2 Hz and a time-step of 25 ms between 4 and 100 Hz, using a 4 Hz lowpass finite impulse response (FIR) filter with a full-width half maximum (FWHM) in the time domain of ~115 ms. Importantly, prior to this time-frequency transformation, we regressed out the time-domain averaged evoked signal from the single-trial data, in order to avoid any “contamination” of the oscillatory data by the evoked response, which is the focus of the time-domain analysis. The resulting spectral power estimations per sensor were averaged over trials to generate time-frequency plots of mean spectral density, which were normalized by the baseline power of each respective bin, calculated as the mean power during the −500 to 0 ms time period. The time-frequency windows used for the time-frequency domain source analysis were again determined by means of a paired-sample cluster-based permutation test against baseline across all participants and the entire frequency range (4 – 100 Hz), with an initial cluster threshold of p < .001 and 10,000 permutations.

2.6. MEG Source Analysis

Time domain source images were computed using standardized low resolution brain electromagnetic tomography (sLORETA; regularization: Tikhonov .01%; Pascual-Marqui, 2002). The resulting whole-brain maps were 4-dimensional estimates of current density per voxel, per time sample across the experimental epoch. These data were normalized to the sum of the noise covariance and theoretical signal covariance, and thus the units are arbitrary. Using the temporal clusters identified in the sensor-level analysis, these maps were averaged over time following each somatosensory stimulation (i.e., 25 – 70 ms and 525 – 570 ms after the onset of the first stimulation) and across both attention conditions. The resulting maps were then grand-averaged across the two stimulations to determine the peak voxel of the time-domain neural response to the stimuli across participants. From this peak, the sLORETA units were extracted per stimulation and attention condition to derive estimates of the time-domain response amplitude for each participant. This approach of using a collapsed localizer to identify the spatial focus of the neural response, regardless of attention condition or stimulation, was taken to avoid circularity in subsequent hypothesis testing (Kriegeskorte et al., 2009). Importantly, this approach is highly unlikely to be systematically biased by differences in the signal-to-noise ratio in the subsequent tests, as the number of trials were matched across all relevant conditions of interest, and each response was present in both attention conditions (Figure S1).

Time-frequency resolved beamformer source images were computed using the dynamic imaging of coherent sources (DICS; regularization: singular value decomposition .0001%; Gross et al., 2001) approach, which uses the time-frequency averaged cross-spectral density to calculate voxel-wise estimates of neural power and/or coherence. Following convention, we computed noise-normalized, source power per voxel in each participant using active (i.e., task) and passive (i.e., baseline) periods of equal duration and bandwidth. The use of active and passive periods with comparable durations and bandwidths is essential, as it ensures that the dual-state beamformer is not biased by the inclusion of different amounts of data in the computation of one versus the other. Such images are typically referred to as pseudo-t maps, with units (pseudo-t) that reflect noise-normalized power differences (i.e., active vs. passive) per voxel. This approach generated three-dimensional participant-level pseudo-t maps per attention condition and stimulation (i.e., the first or second stimulation in the pair), for each time-frequency cluster identified in the sensor-level analysis. As with the time-domain source analysis, the resulting images were next grand-averaged (i.e., across attention condition and stimulation number) and used to derive peak voxel locations for each time-frequency response. Using these peak voxel locations, virtual sensor data were computed by applying the sensor-weighting matrix derived through the forward computation to the preprocessed signal vector, which yielded a time series corresponding to the location of interest. These virtual sensor data were then decomposed into time-frequency space and averaged across the previously identified time-frequency extents (i.e., used in the beamformer analysis) for each response, within each attention condition. This resulted in amplitude estimates of each time-frequency domain response per participant.

To address hypotheses regarding fronto-somatosensory connectivity in the time-frequency domain, peak voxels identified in the DICS power analysis were used as seeds for computation of whole-brain cortico-cortical coherence (again using DICS), reflecting time-frequency-resolved connectivity between these seeds and all other voxels in the brain. Similarly to the power analysis, coherence maps computed from active periods were normalized to coherence maps from passive periods, resulting in whole-brain estimates of percent-change in coherence from baseline for each participant, stimulation, and attention condition. These whole-brain cortico-cortical coherence images were compared voxel-wise, and corrected for multiple comparisons using a similar cluster-based permutation approach as detailed in the Sensor-Level Statistics section (i.e., initial cluster threshold of p < .001; 10,000 permutations). Importantly, due to the persistent concern regarding amplitude confounds in MEG measures of functional connectivity (e.g., coherence; Schoffelen and Gross, 2009), we also used peak-voxel data from these coherence maps to compute repeated-measures ANOVAs of the same conditional connectivity differences, above and beyond the effects of amplitude at both sources. All reported clusters for the coherence analysis are thus significant above and beyond the effects of amplitude.

2.7. Statistical Analyses and Software

To examine the effects of attention condition on SG, a gating ratio (stimulation 2/stimulation 1) was derived per participant for each attention condition, and a repeated-measures ANOVA model was used to test for significant differences in this ratio (i.e., as [1] a function of attention condition and [2] neural response). It is important to note that this ratio was used to test hypotheses, rather than modeling somatosensory gating as a within-participant contrast, since such a model would test the effect of gating as (S2 – S1), rather than (S2/S1). The prior is problematic for two reasons: (1) it is less comparable to previous literature in the field that typically uses the ratio instead, and (2) it biases participants with a higher overall amplitude of response (regardless of stimulation) towards artificially-high gating estimates, whereas the ratio provides a better control for this confound. Simple effects testing for differences in gating ratio between attention conditions for each response was then used to guide interpretation of the initial RM-ANOVA results (Bonferroni correction: p = .050/4 responses = .0125). For similar reasons, significant effects of gating (i.e., regardless of attention condition) were tested on these data using one-sample t-tests against the null hypothesis of stimulation 2/stimulation 1 = 1. This gating analysis was performed on each of the four responses (i.e., one time domain and three time-frequency domain). Additionally, since this initial analysis suggested no effect of attention on SG in the time domain response identified at the sensor-level, an exploratory analysis was conducted whereby time-varying estimates of evoked response amplitude across the epoch were extracted from the peak voxel identified by the sLORETA analysis described above. Using these data, time-varying estimates of SG were computed (stimulation 2/stimulation 1), and cluster-based permutation testing was used to determine whether a significant effect of attention was present during any time window across the time period ranging from 0 to 400 ms post-stimulation using a liberal threshold (initial cluster threshold: p < .500; final significance threshold: p < .200; 10,000 permutations). To test whether attention condition significantly modulated connectivity between the primary somatosensory cortex and other cortical regions during sensory processing, we averaged the cortico-cortical coherence images across stimulations 1 and 2 within each attention condition and participant, and tested these images against each other using voxel-wise paired-samples t-tests corrected for multiple comparisons using cluster-based permutation testing (10,000 permutations). All primary data preprocessing, coregistration, and sensor- and source-level analyses were performed in the Brain Electrical Source Analysis software suite (BESA Research v6.1 and BESA MRI v2.0). Cluster-based permutation testing on sensor-array and whole-brain cortico-cortical coherence data was performed in BESA Statistics (v2.0), and all parametric statistics were computed in JASP (JASP-Team, 2018). Multiple comparisons correction for parametric statistics used the Bonferroni approach, with a corrected significance threshold set to p = .0125 (p = .050/4 tests). To complement our initial frequentist statistical approach, Bayesian analysis was also performed in JASP, using a zero-centered Cauchy distribution with a default scale of 0.707. Finally, due to the upper limit of 100% on our accuracy data, a bounded logistic regression model was also performed on these data in R (Team, 2017) using glmer.

3. Results

All participants performed well on the somato-visual oddball task (Figure 1), with a mean accuracy of 95.19% correct overall (SD = 7.14%; 95% CI: [92.45, 97.93]). Performance did not significantly differ by attention condition (attend somatosensory: mean = 96.63%, SD = 6.66%; attend visual: mean = 93.27%, SD = 10.14%) when using either an unbounded (p = .090, BF₀₁ = 1.25) or a bounded (p = 1.00) distribution. Importantly, no participant identified the oddball stimuli at a rate of <65%, indicating that attention was being effectively directed towards the relevant stimulus modality across all participants.

3.1. Neural responses to paired-pulse stimulation

Prior to determining the spatial origin of the neural responses to each stimulation, we first identified neural response windows that were not exchangeable with the baseline based on our cluster-based permutation testing (see Section 2.5). In the time domain, this revealed one temporally-defined (25 – 70 ms post-stimulus; p < .001) cluster after each somatosensory stimulation in sensors over right somato-motor regions (Figure 2A). For the time-frequency data, three spectrally- and temporally-distinct clusters were identified following each somatosensory stimulation. These included an early increase in theta activity (4 – 8 Hz, 0 – 250 ms post-stimulus; p < .001), and later decreases in both alpha (8 – 14 Hz, 175 – 475 ms post-stimulus; p < .001) and beta (20 – 26 Hz, 100 – 350 ms post-stimulus; p < .001) activity (Figure 2B). In addition, see Figure S2 for sensor-level topographic representations of each of these clusters per stimulation. Imaging of all four of these responses (i.e., one time domain and three time-frequency domain) revealed robust activity in the contralateral postcentral gyrus (Figure 2), just posterior to the motor-hand knob (Yousry et al., 1997). From these source-level images, peak voxel time series data were then extracted and averaged over the same time/time-frequency windows per stimulation and attention condition for subsequent hypothesis testing (i.e., for effects of gating and attention; see Section 2.6– 2.7 for more details).

Figure 2. Neural responses in primary somatosensory cortex.

(A) The time domain average of data from a representative sensor over right somato-motor cortices (MEG1132), with a time domain source image averaged across both stimulations and attention conditions overlaid on the plot. (B) A grand-averaged spectrogram from the same sensor (MEG1132). Note that the phase-locked (evoked) signal has been regressed out. Time is indicated in ms on the x-axis and frequency is indicated in Hz on the y-axis, with percent change from baseline indicated by the color bar above. The white dashed lines represent the onset of each of the two stimulation pulses. Below this plot are frequency-resolved source images of each time-frequency cluster identified in the sensor-level data (again grand-averaged over stimulations and attention conditions). The response amplitude (in pseudo-t) for each cluster is indicated by the color scale bars in between.

3.2. Interactions between SG and directed attention on primary somatosensory neural responses

Source reconstruction of each of these neural responses indicated that all four were centered on the primary somatosensory cortex. Next, we examined the SG effect (stimulation 2/stimulation 1) on each of these source-level responses, as well as the impact of directed attention (i.e., toward or away from the somatosensory stimuli) on this gating. The evoked (i.e., phase-locked) response exhibited significant gating (t(25) = −10.40, p < .001; Figure 3, right), such that the amplitude was reduced in response to the second stimulation. The theta (t(25) = −6.38, p < .001; BF₁₀ = 16,087.22, error % = 5.23 × 10⁻⁸) and beta (t(25) = 2.95, p = .007; BF₁₀ = 6.51, error % = 7.26 × 10⁻⁴) responses, but not the alpha response (t(25) = 1.77, p = .090; BF₀₁ = 1.25, error % = 3.65 × 10⁻⁵), also exhibited significant SG when collapsing across both attention conditions (Figure 4). Similar to the evoked response, the absolute amplitude of theta activity decreased in response to the second stimulation of the pair. For the beta response, this effect was reversed, such that the absolute amplitude was higher in response to the second stimulation as compared to the first. However, it should be noted that since the beta response was a desynchronization (i.e., decrease from pre-stimulus levels), this SG effect should be interpreted as a weakened response to the second stimulation relative to the first, which again would be interpreted as a gating effect.

Figure 3. Phased-locked somatosensory responses are gated and attention-invariant.

Time domain responses were robustly gated (right; purple), but this gating was not modulated by attention (left; blue versus red). Box and whisker plots represent the gating ratio (stimulation 2 amplitude/stimulation 1 amplitude) per attention condition (red and blue) and for the condition-averaged gating ratio (purple). Each plot includes the individual data points, median (horizontal line), mean (white x), first and third quartile (box), and local minima and maxima (whiskers). Points falling outside of the whiskers are more than 1.5 times the interquartile range above or below the third and first quartiles, respectively, and are plotted as such for visualization purposes. These data were included in all analyses. **p < .001

Figure 4. Oscillatory somatosensory responses are gated in the theta and beta bands.

Box and whisker plots represent the gating ratio (stimulation 2 amplitude/stimulation 1 amplitude) per oscillatory response. Each plot includes the individual data points, median (horizontal line), mean (white x), first and third quartile (box), and local minima and maxima (whiskers). Points falling outside of the whiskers are more than 1.5 times the interquartile range above or below the third and first quartiles, respectively, and are plotted as such for visualization purposes. These data were included in all analyses. It should be noted that for theta activity, lower ratios indicate a stronger gating of this response, whereas for alpha and beta activity lower values mean reduced gating, or even an enhancement of the response to the second stimulation. Significant gating effects were observed, regardless of attention condition, for the theta and beta oscillations, but not the alpha response. *p < .01 **p < .001

A two-way repeated-measures ANOVA (within-participants contrasts of response [4 levels] and attention condition [2 levels]) indicated significant main effects of condition (F(1,25) = 6.45, p = .018) and frequency (F(3,75) = 58.56, p < .001), as well as a significant condition-by-response interaction (F(3,75) = 4.032, p = .010), on SG. Post-hoc simple-effects testing indicated that gating of the evoked response was not significantly affected by attention condition (t(25) = −0.89, p = .380; Figure 3, left). Post hoc Bayesian analysis of this effect gave moderate evidence for the null hypothesis, suggesting that gating of the evoked response is indeed attention-invariant (BF₀₁ = 3.36, error % = 0.03). Exploratory analyses following a similar analytical pipeline (i.e., imaging with sLORETA) examining the qualitative response components at 21, 35, 52, and 103 ms (Figure S3), as well as the later components extending from 70 – 130 ms and 150 – 200 ms, indicated that (1) there was no distinct response peak in secondary somatosensory cortices (SII) in any of these time windows, and (2) that the gating of the response at the primary somatosensory peak did not significantly differ as a function of attentional condition in any time window, even when extremely liberal thresholds for significance were used (cluster-based permutation test; initial cluster threshold: p < .500; final significance threshold: p < .200; 10,000 permutations; Figure S4).

Interestingly, the SG of all three oscillatory responses was altered by directed attention (Figure 5). Specifically, the theta response exhibited lower SG ratios when attention was directed toward the somatosensory stimuli, relative to when it was directed away (t(25) = −3.03, p = .006; BF₁₀ = 7.74, error % = 4.25 × 10⁻⁴). The direction of this effect (Figure 6) indicates that theta response gating was stronger when attention was directed towards the somatosensory domain. Similarly, attention towards somatosensory stimulation resulted in significantly lower alpha (t(25) = −2.98, p = .006; BF₁₀ = 6.98, error % = 5.90 × 10⁻⁴) and beta (t(25) = −3.04, p = .005; BF₁₀ = 7.89, error % = 4.00 × 10⁻⁴) SG ratios relative to when attention was directed away. However, note that since these alpha and beta responses were desynchronizations (i.e., decreases relative to baseline), lower SG values reflect reduced gating, or even response enhancement. Indeed, inspection of the underlying data for the alpha and beta effects (Figure 6) indicated that the amplitude of these responses were increased for the second stimulation when attention was directed towards the somatosensory domain, suggesting a potential enhancement of these responses by “top-down” attentional modulation. Importantly, no significant main effect of attention was found on the neural response amplitude to the stimulations for any of the four neural responses (evoked: p = .981, BF₀₁ = 4.83; theta: p = .618, BF₀₁ = 4.29; alpha: p = .949, BF₀₁ = 4.82; beta: p = .256, BF₀₁ = 2.63). For enhanced interpretation of the interactions between attention and SG, see Figure 6 for comprehensive participant-level data per attention condition and stimulation.

Figure 5. Directed attention modulates the gating of theta, alpha, and beta somatosensory oscillations.

Box and whisker plots represent the sensory gating ratio (stim 2 amplitude/stim 1 amplitude) per attention condition. Each plot includes the individual data points, median (horizontal line), mean (white x), first and third quartile (box), and local minima and maxima (whiskers). Points falling outside of the whiskers are more than 1.5 times the interquartile range above or below the third and first quartiles, respectively, and are plotted as such for visualization purposes. These data were included in all analyses. It should be noted that for the theta synchronization response, higher values indicate reduced gating, whereas for the alpha and beta desynchronization responses higher values indicate enhanced gating. Thus, directing attention toward somato-sensation enhanced theta-band SG, while the opposite was true for both alpha and beta oscillations. *p < .01

Figure 6. Individual participant-level data per attention condition and stimulation for each neural response.

Box and whisker plots represent the absolute response amplitude (in a.u. or nAm) per attention condition and stimulation. Each plot includes the individual data points, median (horizontal line), mean (white x), first and third quartile (box), and local minima and maxima (whiskers). Points falling outside of the whiskers are more than 1.5 times the interquartile range above or below the third and first quartiles, respectively, and are plotted as such for visualization purposes. These data were included in all analyses. Importantly, there were no main effects of attention on any of these four neural responses to somatosensory stimulation.

The direction of attention effects on the gating of these multi-spectral responses suggested that the early theta component may be an early, “bottom-up” response, while the later alpha and beta oscillations were potentially modulated by “top-down” control. Essentially, since the somatosensory stimulus was only task-relevant in the “attend somatosensory” condition, a declining response to the second stimulus during directed attention (such as in the theta response) indicates an earlier alerting component of stimulus processing. In contrast, an increasing response to the second somatosensory stimulus during directed attention (such as in the alpha and beta responses) indicates heightened processing of this relevant stimulus. To test this possibility further, we computed whole-brain cortico-cortical coherence maps for each oscillatory response, attention condition, and stimulation in each participant, and averaged these maps across stimulations to derive voxel-wise coherence estimates per attention condition and oscillatory response per participant. These maps represent whole-brain coherence between the neural response of interest (e.g., the primary somatosensory alpha response) and the activity across the rest of the brain within the same time-frequency window. We then tested these whole-brain maps for effects of attention condition on coherence in both the alpha and beta frequencies, and corrected for multiple comparisons using cluster-based permutation testing (initial cluster threshold: p < .001; final significance threshold: p < .025; 10,000 permutations). Intriguingly, coherence between the primary somatosensory alpha response and the left dorsolateral prefrontal cortices was significantly increased when attention was directed towards the somatosensory stimuli, relative to when it was directed away (Figure 7; p = .004). Further, alpha coherence between the primary somatosensory response and the right cuneus was also modulated by attention, such that connectivity between these regions decreased when attention was directed toward the somatosensory domain (p = .021). No significant cortico-cortical coherence differences were observed for the beta response.

Figure 7. Directed attention modulates inter-regional somatosensory alpha coherence.

The images on the left of the dashed line in (A) represent whole-brain alpha-band coherence (in % change from baseline) with the primary somatosensory cortex as the seed for each attention condition. Scale bars are shown above the maps. Maps on the right of the dashed line represent a voxel-wise paired-samples t-test of this alpha coherence between the two attention conditions (Left-DLPFC: p = .008, corrected; Right-Cuneus: p = .010, corrected), with the color bars to the right indicating uncorrected voxel-wise significance. The box and whisker plots on the far right represent the condition-wise coherence differences at the peak difference voxel from the overlaid maps (bottom). The features of each plot matches those in the box and whisker plots above. These data were included in all analyses. We found that coherence between somatosensory and prefrontal cortices was sharply increased during the attend somatosensory condition, while such coherence was sharply decreased between the cuneus and somatosensory cortices in the same condition.

4. Discussion

In this study, we used a novel somato-visual oddball task and whole-brain MEG to investigate the impact of directed attention on SG in the somatosensory domain. We found that attention toward somatosensation significantly altered the gating of all three population-level neural oscillatory responses to the paired-pulse stimuli, and that this gating effect differed according to the spectro-temporal profile of the response. Specifically, SG of the early theta response was increased when attention was directed towards the somatosensory domain, while gating of the alpha and beta responses was decreased in the same attentional state. Importantly, this attention effect on SG was not present for the evoked (i.e., phase-locked) somatosensory response. Further, all of these attentional effects were the most robust in frequencies strongly tied to somatosensory processing in previous studies (Bardouille et al., 2010; Dockstader et al., 2010; Haegens et al., 2011; Haegens et al., 2012; Jones et al., 2010; van Ede et al., 2010; van Ede et al., 2014; Whitmarsh et al., 2014). These findings, as well as their implications and future directions for study, are discussed at length below.

The current findings have important implications for understanding the basic functional role of the spectrally distinct somatosensory responses. Alpha and beta oscillations in the primary somatosensory cortices have been tied to anticipatory and attentional processing (Bardouille et al., 2010; Dockstader et al., 2010; Haegens et al., 2011; Haegens et al., 2012; van Ede et al., 2010; van Ede et al., 2014), and so it is unsurprising that SG in these frequencies was robustly affected by directed attention. What is perhaps more surprising, is that the effects of attention on the gating of neural oscillatory responses reversed direction depending on the spectro-temporal profile of the response in question. While attention enhanced gating of the early theta response, it decreased gating of the later alpha and beta responses. This broadly supports the conceptualization of this early theta component as representing low-level stimulus recognition and feature encoding (Andersen and Lundqvist, 2019; Hari and Forss, 1999; Hlushchuk and Hari, 2006; Schnitzler et al., 1995; Wiesman et al., 2016). Essentially, as such gating is thought to represent the “filtering” of irrelevant stimulus information at an early stage, it is intuitive that enhanced attention towards this stimulus would translate to more effective gating. In other words, since the stimulus properties (e.g., amplitude, pulse-width) were identical for both stimulations, additional processing of these properties would be unnecessary or even detrimental, and this effect would only be accentuated when the timing (but not the stimulus properties themselves) were relevant. On the other hand, the reduction in gating of the later beta and alpha responses as a function of directed attention indicates that these responses are representative of modulatory feedback and (at least in this case) temporal processing, as the timing of the second stimulus was more salient in the “attend somatosensory” condition.

Further supporting this notion, alpha coherence between the prefrontal cortex and the primary somatosensory cortices was higher when attention was directed towards the somatosensory domain. This points to a prefrontal modulator of the alpha-somatosensory response and, interestingly, this effect was specific to this frequency band. This finding is in line with previous reports of a prefrontal modulator of somatosensory processing (Staines et al., 2002; Wilson et al., 2015; Yamaguchi and Knight, 1990), but importantly, to our knowledge is the first empirical evidence of direct prefrontal-somatosensory modulation (i.e., prior studies showed only co-activation). Somatosensory alpha coherence with the right cuneus was also significantly decreased when attention was directed towards the somatosensory domain. This finding is in line with a vast literature supporting parieto-occipital alpha desynchronizations as an active dis-inhibition of visual processing circuits during specific visual tasks (Bonnefond and Jensen, 2012; Handel et al., 2011; Janssens et al., 2018; Jensen et al., 2002; Jensen and Mazaheri, 2010; Klimesch, 2012; Klimesch et al., 2007; Wiesman et al., 2018; Wiesman and Wilson, 2019). Given these previous findings, the relative decrease in somato-visual connectivity observed during the attend somatosensory condition represents a “decoupling” of the somatosensory and visual processing circuits, in order to facilitate more effective performance on the somatosensory task.

The evidence provided herein for no attentional effect on SG of phase-locked (i.e., evoked) primary somatosensory responses is also highly informative. Although more recent studies have begun to focus on the oscillatory neural dynamics of SG (Cheng et al., 2015b; Kurz et al., 2017; Spooner et al., 2018; Spooner et al., 2019; Wiesman et al., 2016), historically, the vast majority of this literature has centered around time domain analysis of the evoked components. While these studies have provided a foundational understanding of the neurophysiological bases of SG, it is clear from this study and others that SG of evoked responses is only one part of a complex series of neurophysiological phenomena at play. Indeed, our findings align well with previous investigations that often find no significant effect of attention on SG of early evoked responses (Gjini et al., 2011; Kho et al., 2003; Rosburg et al., 2009). Our study expands this into the somatosensory domain, and provides the first evidence for the null hypothesis of no attentional effect, using post hoc Bayesian analysis.

In addition to the significance for understanding the population-level neurophysiology of somatosensory processing, the implications of this research for previous and future studies of clinical populations should also be addressed. Given the vast number of studies that have reported SG alterations in patient groups, the fact that most of these studies did not control for attentional state across participants raises important concerns. Basically, we systematically modulated attention and found robust effects on SG across three well-documented oscillatory somatosensory responses. Thus, it is possible, perhaps even probable, that the known attentional differences in many psychiatric and neurologic disorders may have incidentally affected previous findings. Supporting the likelihood that attentional differences might be partly responsible for these effects, SG has been repeatedly tied to neuropsychological tests of attention function (Karper et al., 1996; Lijffijt et al., 2009a; Wan et al., 2008), and select components of auditory SG have been found to be modulated by attention (Gjini et al., 2011; Golubic et al., 2019; Kho et al., 2003; Rosburg et al., 2009). The current results extend this potential confound into the somatosensory domain, and also provide evidence for spectrally-specific differences in the nature of the attention effect on SG. Future studies are certainly warranted to better understand the actual impact of attentional differences on SG. Further, future studies investigating SG in populations which vary in attentional abilities (e.g., patient populations or aging samples) should attempt to either control for these potential confounds, or to investigate the impact of attentional abilities on key SG metrics. With all of this said, it is notable that many previous patient-based studies have focused on evoked responses, which to at least some degree should be reassuring, as we did not observe attentional effects on these responses.

Despite its novelty, this study is not without limitations. First, although we were successful in directing participant’s attention towards and away from the somatosensory domain, the attentional load required for this task was likely only moderate. Future studies might systematically increase the attentional load towards the somatosensory domain in a step-wise manner, which would show whether the attentional effects on SG observed here reach any type of functional plateau. Second, although participants did respond to the stimuli presented in this study, these responses were only to the oddball stimuli, and thus there was not enough behavioral data for a thorough analysis. Additional research is necessary to determine how these attentional effects might affect perception and discrimination of somatosensory stimuli. Third, although we found sufficient evidence for no effect of attention on the gating of somatosensory responses in the primary somatosensory cortex, we are not as confident that such an effect does not exist in secondary somatosensory regions (SII). Initial exploratory analyses indicated no such effect in the later evoked components usually attributed to this region, and no distinct SII peak could be identified using our methods, however, it remains a possibility that such an effect might be identified using more targeted methodologies and analytical approaches. Indeed, MEG has even been suggested to be a poor method for measurement of SII activity, which is highly variable between participants (Hari and Forss, 1999; Hsiao et al., 2013; Torquati et al., 2003; Wiesman et al., 2016). Finally, we were able to identify a prefrontal modulator of somatosensory dynamics in this study, supporting our hypothesis that the later alpha response represents “top-down” processing of the stimulus. However, conversely, we would also predict that a similar pattern of coherence with “bottom-up” regions would exist for the earlier theta response (i.e., from thalamic inputs). We found no such pattern of coherence, and though it is possible that this connectivity does not exist, it seems more likely that the limited sensitivity of MEG to deeper brain structures might have played a limiting role. Regardless, these findings have important implications for advancing our basic understanding of somatosensory neurophysiology, as well as for our interpretation of previous research in clinical and aging populations.