Neural correlates of tactile detection: a combined magnetoencephalography and biophysically based computational modeling study

Abstract

Previous reports conflict as to the role of primary somatosensory neocortex (SI) in tactile detection. We addressed this question in normal human subjects using whole-head magnetoencephalography (MEG) recording. We found that the evoked signal (0-175 ms) showed a prominent equivalent current dipole that localized to the anterior bank of the postcentral gyrus, area 3b of SI. The magnitude and timing of peaks in the SI waveform were stimulus amplitude dependent and predicted perception beginning at approximately 70 ms after stimulus. To make a direct and principled connection between the SI waveform and underlying neural dynamics, we developed a biophysically realistic computational SI model that contained excitatory and inhibitory neurons in supragranular and infragranular layers. The SI evoked response was successfully reproduced from the intracellular currents in pyramidal neurons driven by a sequence of lamina-specific excitatory input, consisting of output from the granular layer (approximately 25 ms), exogenous input to the supragranular layers (approximately 70 ms), and a second wave of granular output (approximately 135 ms). The model also predicted that SI correlates of perception reflect stronger and shorter-latency supragranular and late granular drive during perceived trials. These findings strongly support the view that signatures of tactile detection are present in human SI and are mediated by local neural dynamics induced by lamina-specific synaptic drive. Furthermore, our model provides a biophysically realistic solution to the MEG signal and can predict the electrophysiological correlates of human perception.

Figure 1.

Localization of SI activity. Estimated SI ECD localizations (blue dots) and orientations (blue lines) overlaid on the subjects' structural MRI brain images. The response evoked by stimulus to the third digit of the right hand was localized in the anterior bank of the contralateral postcentral gyrus, the position of area 3b. A, B, Example localizations and orientations from two subjects using the inverse solution technique described in Materials and Methods. C, Example localization for one of two subjects for which the SI ECD was placed (see Materials and Methods).

Figure 2.

Model SI network architecture. Ten PNs and 3 INs were included per layer. Excitatory (dark green) and inhibitory (red) synaptic connections were set as depicted. Bold outlined dendrites were contacted. A, Local synapses. Within-layer PN-to-PN synapses (not shown) were also present on dark green outlined dendrites. Each set of synaptic weights had a Gaussian spatial profile (Table 2). B, Spiking patterns evoked by somatic injected current (1 nA, 100 ms; no synaptic input). C, Connection pattern of output from the GR. The black arrow is only schematic, because lemniscal thalamic input was not explicitly modeled. D, Connection pattern of exogenous input to the SGR, presumably from a higher-order cortical and/or nonspecific thalamic neurons. The output from GR and input to SGR were modeled as spike train generators with a predetermined temporal profile and synaptic strength (Table 3).

Figure 3.

SI evoked responses. A, SI ECD responses from individual subjects (n = 7; baseline, mean 0–20 ms subtracted for display purposes) for suprathreshold-level (left) and threshold-level (right) stimuli. B, Average of suprathreshold-level (red curve) and threshold-level (blue curve) stimuli over all subjects. Consistent peaks emerged in the grand averages from 0 to 175 ms as labeled. Early peaks were not observed in the threshold response. C, Colored curves, Difference between suprathreshold- and threshold-level responses from individual subjects. Black curve, Mean difference over all subjects. Dark gray curve, Mean difference over subjects excluding subject with smallest difference (red curve). Light gray curve, Mean difference over subjects excluding subject with largest difference (cyan curve). The mean differences show that the stimulus amplitude differences were not driven by the response of an outlier subject.

Figure 4.

The magnitude and timing of the SI evoked response predict perception. A, Threshold-level stimulus SI evoked responses averaged over perceived (dark blue) and nonperceived (light blue) trials for varying stimulus amplitudes that were dynamically maintained at 50% perceptual threshold (trials = 100; n = 7). On perceived trials, the onset slope from the M70 to the M100 peak was larger, and the magnitudes of the M100 and M135 peaks and area under the curve between them were larger. B, Average threshold-level responses comparing perceived and nonperceived trials that had equal stimulus amplitudes (trials = 19 ± 9; n = 6; for details, see Results). On perceived trials, the magnitude of the M70 peak and onset slope from M70 to M100 were larger in this comparison. C, Average threshold-level responses sorted by stimulus amplitude. Dark green, Larger stimulus amplitudes; light green, smaller stimulus amplitudes (trials = 100 trials; n = 6). There were no statistically significant differences in this comparison. Error bars represent SEM.

Figure 5.

Simulated SI evoked responses for suprathreshold- and threshold-level stimuli. A, The timing and location of synaptic inputs sets SI ECD polarity in the model. Red curve, Suprathreshold response. Output from activity in the GR at ∼25 ms reproduces the initial M25-M35-M50 peaks, exogenous SGR input at ∼70 ms created the subsequent M70 and M100 peaks, and second later GR output at ∼135 ms induced the M135 peak. Input times were selected over trials from Gaussian distributions displayed schematically in green with arrows marking the earliest input times (Table 3) (average of 100 trials). Blue curve, Threshold response. Decreasing the synaptic strengths by 50% (GR), 25% (SGR), and 7% (late GR) reproduced the waveform for the threshold-level stimulus (compare with Fig. 3 B). B, Top, Contributions to the net current dipole during the suprathreshold response separated by layer. Bottom, Example of the activity from an individual L2/3 (left) and L5 (right) PN on a single trial; top traces, separate contributions from the basal (green) and apical (blue) dendritic compartments to the total current dipole (red) produced by the neuron; bottom traces, somatic membrane potential showing action potentials (black). C, Contributions to the net current dipole during the threshold response separated by layer. B, C, Schematics of network architecture drawn at each peak and arrows describe the direction of the net intracellular current flow within the pyramidal neurons that determines the polarity of the peak.

Figure 6.

Simulated SI evoked responses for perceived and nonperceived trials (compare with Fig. 4 A). The statistically significant differences in the SI evoked responses for perceived (dark blue) versus nonperceived (light blue) trials were reproduced in the model by decreasing the mean latency and increasing the synaptic input strength of the SGR input and late GR output. For perceived trials, the mean SGR input and late GR output latencies were each decreased by 5 ms (to 65 and 130 ms, respectively) and their strengths were increased by 5% and 30%, respectively (Table 3) (average of 100 trials). The initial GR output was fixed. The green arrows and schematic Gaussians mark the distributions and earliest input times for perceived trials. These manipulations reproduced the increase in M70 magnitude, the larger onset slope to the M100 peak, the larger magnitude of the M100 and M135 peaks, and the greater mean area under the curve between them. Nonperceived trials were simulated with default parameters that produced the threshold-level response in Figure 5 A.