Reebal W. Rafeh¹ (), Geoffrey N. Ngo¹, Matthew Liaw¹, Lyle E. Muller¹, Ali R. Khan¹, Ravi S. Menon¹, Taylor W. Schmitz¹, Marieke Mur¹; ¹Western University

Population receptive field (pRF) mapping is a powerful fMRI tool for fine-grained mapping of cortical responses to visual stimulus features. Over the past decade, multiple studies have shown a non-linear relationship between pRF responses and simultaneous stimulus presentation (Kay et al., 2013; Kupers et al., 2024). Simultaneous stimuli presented within a pRF drive sub-additive responses relative to individually presented stimuli. However, there is no clear mechanistic explanation for this compression. We propose that the normalization model (Reynolds & Heeger, 2009) provides a parsimonious mechanistic account of pRF response compression, where suppressive interactions between differentially tuned populations within fMRI voxels drive response non-linearities. In this 7T frequency-tagged fMRI (ft-fMRI; Ngo, Rafeh et al. 2024) experiment, we investigate sub-voxel dynamics by stimulating distinct neural subpopulations within pRFs at temporally dissociable frequencies. Participants (n=7) distributed their attention across two simultaneously presented visual checkerboard stimuli oscillating at 0.125 Hz and 0.2 Hz. Our analyses revealed that distinct populations of visual cortical neurons synchronized to either one (singular) or both (multiplexing) frequencies. A separate pRF mapping experiment confirmed that multiplexing populations had visual field preferences that overlapped with both stimuli, whereas singular populations overlapped with only one of the two stimuli. Adapting the normalization model to our dynamic frequency-tagging paradigm, we predicted that mutual suppression between neuronal subpopulations within multiplexing pRFs should generate stronger non-linear intermodulation components in their timeseries. Confirming this prediction, multiplexing populations exhibited significantly stronger intermodulation amplitudes than singular populations. Furthermore, directing attention to either of the two stimuli significantly reduced intermodulation amplitude, corroborating our dynamic normalization model. Together, our findings support the normalization model as a mechanistic account of pRF response compression and introduce a novel ft-fMRI marker for suppressive interactions at a sub-voxel scale. In addition, our dynamic normalization model offers a framework for understanding neuronal computations in ft-fMRI paradigms.

Acknowledgements: Natural Sciences and Engineering Research Council