Dynamic Modulation of Information Flow from Occipitotemporal Cortex According to Cognitive Demands

1488 

Abstract Submission 

Vicky He^1,2, Bahman Tahayori¹, David Vaughan^1,2, Graeme Jackson^1,2, David Abbott^1,2, Chris Tailby^1,2, for the Australian Epilepsy Project Investigators¹

¹The Florey Institute of Neuroscience and Mental Health, Melbourne, Australia, ²University of Melbourne, Melbourne, Australia

Vicky He  
The Florey Institute of Neuroscience and Mental Health|University of Melbourne
Melbourne, Australia|Melbourne, Australia

Bahman Tahayori  
The Florey Institute of Neuroscience and Mental Health
Melbourne, Australia

David Vaughan, PhD  
The Florey Institute of Neuroscience and Mental Health|University of Melbourne
Melbourne, Australia|Melbourne, Australia

Graeme Jackson  
The Florey Institute of Neuroscience and Mental Health|University of Melbourne
Melbourne, Australia|Melbourne, Australia

David Abbott, PhD  
The Florey Institute of Neuroscience and Mental Health|University of Melbourne
Melbourne, Australia|Melbourne, Australia

Chris Tailby, PhD  
The Florey Institute of Neuroscience and Mental Health|University of Melbourne
Melbourne, Australia|Melbourne, Australia

for the Australian Epilepsy Project Investigators  
The Florey Institute of Neuroscience and Mental Health
Melbourne, Australia

Neurobiological models of cognition hold that the degree to which information flows between different brain areas is modulated as a function of cognitive demands (Park and Friston 2013). Psychophysiological interaction analysis (PPI) is a regression-based method for evaluating such modulations in functional magnetic resonance imaging (fMRI) data (Friston et al. 1997). Here we applied PPI to investigate the task dependent modulation of connectivity from the ventral occipitotemporal cortex (vOT, also known as the Visual Word Form Area) during execution of a language task (pseudoword rhyming) versus a visuospatial task. We hypothesised that there will be reweighting of information flow from vOT according to task demand. More specifically, the vOT will interact more strongly with perisylvian language areas during pseudoword rhyming than during visual pattern matching.

Ninety-four Australian Epilepsy Project (AEP) participants completed a block design fMRI task contrasting rhyming blocks (whether two visually presented pseudowords would rhyme if pronounced aloud) against pattern matching blocks (whether two patterns composed of forward and backslashes matched). First level analysis was applied using the iBrain Toolbox (Abbott and Jackson 2001) with SPM12 (Ashburner et al. 2021). The PPI regression includes the main effect of task, the main effect of the seed (vOT) time course, the interaction between the two, and nuisance regressors. The PPI was implemented using the gPPI (generalised PPI) toolbox (McLaren et al. 2012), which incorporates a deconvolution procedure in forming the interaction (Gitelman et al. 2003). Given some debate around whether mean centring the task regressor in the interaction term is necessary (see Di, Reynolds, and Biswal 2017; 'NITRC: Generalized PPI Toolbox: Task Time Course Not Mean-Centered' n.d.) we modified the code in the toolbox to carry out PPI both with and without (default setting) mean-centring. Finally, we carried out a systematic search on papers that used PPI from 2018 to 2022 to examine whether mean-centring has become the standard in the field.

Figure 1a shows the SPM-t map of the group level analysis on the interaction term with recommended mean-centring. As hypothesised, the vOT connects more strongly with classical language areas (left inferior frontal gyrus and superior temporal sulcus) during the rhyming task. On the other hand, there is greater connectivity between the vOT and the right intraparietal sulcus in the visuospatial task. Figure 1b shows the results without mean-centring, where there is extensive interaction with a peak at the seed location. This indicates a misspecification of the model, as the seed should not appear significant in the interaction (because the main effect of the seed is already included in the regression). Through the systematic search, we found that most papers do not include sufficient information about how the PPI model is implemented, with a number of published figures showing interaction effects within the seed region suggesting model misspecification.

Our findings utilising the mean-centred PPI method are consistent with the idea of dynamic modulation of information flow according to cognitive demands (Park and Friston 2013). Using comparable visual stimuli as input, connectivity from left vOT was increased to key language areas during a linguistic task, but to visuospatial processing areas of the right hemisphere during visuospatial judgments. These findings are consistent with models of the reading system (Sandak et al. 2004) and language (Hickok and Poeppel 2007). Furthermore, we support the observations of Di and colleagues (Di, Reynolds, and Biswal 2017), that with deconvolution, it is essential for the task regressor to be mean centred to avoid spurious results.

Reading and Writing ²

Connectivity (eg. functional, effective, structural) ¹

fMRI Connectivity and Network Modeling

Data analysis

FUNCTIONAL MRI

Language

Modeling

Statistical Methods