Hierarchical predictive system for the neural integration of incoming somatosensory stimulations

• ¹ Université libre de Bruxelles (ULB), UNI – ULB Neuroscience Institut, Belgium

Introduction: Recent works have highlighted the role of a hierarchical predictive coding system for the processing of incoming auditory stimuli that can be evidenced by studying specific evoked neural responses. A first unconscious and automatic level of processing is assessed by the mismatch negativity response (MMN) that is observed when a deviant auditory stimulus is detected in a sequence of identical stimuli. This response is thought to reflect the cortical response to a violation of incoming auditory stimulus prediction occurring on a brief time frame basis. Higher hierarchical levels of auditory processing that depend on attention and conscious awareness are reflected by the P300 and the contingent negative variation (CNV) responses. The P300 is elicited by more global and integrative violations of incoming auditory stimulus expectations occurring over longer timescales. The CNV reflects top-down expectation and is characterized by a progressive baseline drift arising when an incoming stimulus or pattern is expected to occur because identical stimuli/sequences have happened before and been learned. These cortical evoked responses associated with auditory stimulus processing reflect the neural integration of sensory stimuli occurring in the extra-personal space. The existence of such hierarchical predictive coding system for the processing of incoming somatosensory stimuli has not been studied so far and such investigation would bring further insights into how the human brain integrates sensory stimuli occurring in the personal (bodily) space. The aim of this work is therefore to determine if the same pattern of hierarchical predictions as assessed by the MMN, P300 and CNV responses exists for the processing of somatosensory stimuli. Subjects and methods: Somatosensory evoked magnetic fields were recorded (band-pass: 0.1–330Hz, sampling rate: 1kHz) using whole-scalp magnetoencephalography (MEG; Elekta Oy, Finland) in 16 healthy adult subjects (aged 29 +/-3 years, 7 females). Subjects underwent a unilateral tactile oddball paradigm adapted from Chennu et al. (2013), in which standard stimuli (S) corresponded to a pneumatic tactile stimulation (covering area: 1cm2, intensity: 2 bars, duration: 50ms) of the right index fingertip and deviant stimuli (D) to a pneumatic tactile stimulation simultaneously applied to the fingertip and the middle phalanx of the right forefinger. Blocks of five tactile stimuli were presented using an inter-stimulus interval of 500ms. Blocks either comprised five successive standards (SSSSS) or four standards followed by a deviant (SSSSD). In the experiment, 120 blocks were successively presented in a semi-randomized order (inter-block interval: 800 ms) during which the first 20 blocks were always SSSSD and in the following 100 blocks, 20 SSSSS blocks were randomly intermingled. Each deviant stimulus in SSSSD blocks, by breaking a sequence of four identical stimuli, produced a local violation that was expected to generate a somatosensory MMN (sMMN). Subjects were further informed that the first twenty blocks (SSSSD) of the experiment fixed the pattern they had to remember as the rule. They were asked to count the number of blocks that violated the learned rule (i.e., SSSSS blocks), a global violation detection process supposed to elicit a P300 response. Finally, baseline drifts corresponding to the CNV were investigated within (between the first and the fifth stimulus of successive SSSSD block) and across (between SSSSD blocks occurring immediately after and before a rule violation) blocks. MEG data were preprocessed and analyzed using methods adapted from Chennu et al. (2013). MMN and CNV responses were assessed at the individual level, while the P300 response was evaluated at the group-level to optimize the signal-to-noise ratio. Neural generators of MMN and P300 responses were identified using conventional equivalent current dipole (ECD) modeling tools (Elekta Neuromag Oy) and spherical conductor models determined from subjects’ individual MRI. ECDs were superimposed on the co-registered subjects’ MRI. Source strength waveforms were then obtained over the considered epochs. Statistical significance of MMN and P300 responses were assessed in the sensor space using a non-parametric clustering approach (Maris & Robert 2007). Paired t-tests were applied on sources intensity for each of the above comparisons. Results were considered significant at p<0.05. Baseline drifts were assessed using the least-squares linear fit approach. Results: Local violations elicited in 14/16 subjects a significant sMMN response, peaking at 75-150 ms post-deviants, with a neural generator located at secondary somatosensory cortex contralateral to the stimulation. At the group-level, global violations elicited a significant P300 response, peaking at 350-430 ms post-deviants, with neural generators located at the temporo-parietal junction bilaterally. Significant baseline drifts were found in all subjects both within and across blocks with maximum drifts observed over central sensors. Conclusion: This study demonstrates the existence of a hierarchical processing of somatosensory stimuli consistent with the predictive coding theory. This paradigm confirms that predictive coding is a common feature of sensory integration whatever the stimulus modality and its occurrence in personal or extrapersonal space. By exploring neural integration of sensory stimuli that occur in the personal space, this paradigm could be clinically useful to better understand the pathophysiology of disorders affecting personal space perception such as personal neglect, conversion disorders or cortico-basal dementia.