Discovering gain modulation dynamics with dynamic causal modelling

1935 

Abstract Submission 

Johan Medrano¹, Noor Sajid¹, Stephanie Mellor¹, Robert Seymour¹, Karl Friston¹, Peter Zeidman¹

¹Wellcome Centre for Human Neuroimaging, University College London, London, United Kingdom

Johan Medrano  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Noor Sajid  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Stephanie Mellor  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Robert Seymour  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Karl Friston  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Peter Zeidman  
Wellcome Centre for Human Neuroimaging, University College London
London, United Kingdom

Gain modulation is an important mechanism related to sensory attenuation or enhancement, e.g., in visual [1,5] or auditory [2,3,4] processing. Discrete changes in synaptic gain can be related to experimental manipulations using dynamic causal modelling (DCM), a statistical framework for inferring the underlying causes of observed data. Here, we use a generic set of temporal basis functions to appropriately model slow continuous dynamics of gain modulation in DCM. Our approach relies on the separation of temporal scales between the dynamics of brain electrophysiological responses and synaptic gain [8]. Our simulations show that fast effects can be modelled using existing dynamical models while slow effects can only be retrieved using these temporal basis functions. Next, we empirically evaluate this approach using magnetoencephalographic (MEG) data from an auditory roving oddball paradigm collected with Optically-Pumped Magnetometers (OP-MEG) and demonstrate effective recovery of slow modulation of synaptic gain. Our formulation provides a generic method to recover modulatory dynamics from neural activity.

In DCM, a matrix (i.e., design matrix) is used to map modulatory effects to experimental conditions i.e., relating differences in brain responses to changes in population gain and effective connectivity. We propose to capture the smooth, slow variations of synaptic efficacy between experimental conditions using a generic set of temporal basis functions in the design matrix that can efficiently represent the slow modulatory dynamics. Briefly, each of the modulatory effect matrices now separately models the weight of each of the temporal basis functions. This allows us to retrieve the time course of the slow fluctuations in synaptic gain by projecting the posterior estimates of the modulatory matrices onto the temporal basis functions. Our approach extends the existing DCM formulation and was implemented using standard machinery in SPM.

Face validity: Using a two-region model of electrophysiological responses, we simulated evoked responses for repetition of a stimulus, with slow fluctuations of the synaptic gains over time. Our analysis assumed that conditions were temporally ordered. We used a DCM for event-related potentials [7] with a Fourier basis set to retrieve slow parameter trajectories. Our approach led to successful recovery, i.e., the posterior trajectories attributed a high probability to the true trajectories (Figure 1).
Application to gain modulation in auditory repetition with OP -MEG: We investigate the modulatory dynamics in an auditory roving oddball paradigm using this model. Specifically, we asked how changes in evoked response from the first presentation of a tone (deviant) onwards can be explained by the slow evolution of synaptic efficacy. For this, we use OP-MEG data from Ref [8], pre-processed following Ref [10]. Here, two participants listened for approximately 1140 tones dispatched in sequences of 2 to 11 repeated tones with 160 deviants (i.e., tone switches). To model this, we used a DCM for ERP model with a 4th-order cosine set to model the slow evolution of synaptic parameters. Consistent with previous studies [2, 6], our results show that changes in evoked response were explained by a decrease in synaptic gain after the first deviant (Figure 2).,We found this effect to be stronger in the right primary auditory cortex, providing a mechanistic hypothesis for the observed repetition suppression.

We introduce a new method to model gain modulation, building upon the existing DCM formulation for modelling brain dynamics . We empirically validate it using OP-MEG data. Our results illustrate the relevance of our formulation for recovering slow modulatory dynamics without explicit priors about the shape of the gain modulation. More generally, our method may be useful way to link fast neural activity with slower changes in behaviour, which is of relevance for natural, real-world experiments.

Bayesian Modeling ²

Connectivity (eg. functional, effective, structural)

EEG/MEG Modeling and Analysis

Methods Development ¹

MEG

Other - Dynamic causal modelling; OP-MEG; Gain modulation