Higher and deeper: Bringing layer fMRI to association cortex

Abstract

Recent advances in fMRI have enabled non-invasive measurements of brain function in awake, behaving humans at unprecedented spatial resolutions, allowing us to separate activity in distinct cortical layers. While most layer fMRI studies to date have focused on primary cortices, we argue that the next big steps forward in our understanding of cognition will come from expanding this technology into higher-order association cortices, to characterize depth-dependent activity during increasingly sophisticated mental processes. We outline phenomena and theories ripe for investigation with layer fMRI, including perception and imagery, selective attention, and predictive coding. We discuss practical and theoretical challenges to cognitive applications of layer fMRI, including localizing regions of interest in the face of substantial anatomical heterogeneity across individuals, designing appropriate task paradigms within the confines of acquisition parameters, and generating hypotheses for higher-order brain regions where the laminar circuitry is less well understood. We consider how applying layer fMRI in association cortex may help inform computational models of brain function as well as shed light on consciousness and mental illness, and issue a call to arms to our fellow methodologists and neuroscientists to bring layer fMRI to this next frontier.

Predictive coding during action observation - a depth-resolved intersubject functional correlation study at 7T

“…While in primary sensory regions such as V1 both these conditions appear to be met, with feed-forward signals peaking in Layer IV while feedback signals alter activity elsewhere (Self et al 2019), the connectivity pattern between parietal and premotor regions in monkeys are, as we mentioned, more complex, with some frontal input to parietal regions terminating throughout all layers in a lateral connectivity pattern while others are multilayered, avoiding layer VI, albeit with a pattern that differs from the prototypical pattern described in early visual regions (Felleman and Van Essen 1991;Rozzi et al 2006;Shipp 2007;Gerbella et al 2010Gerbella et al , 2011. While studies like ours have thus been argued to be necessary to start exploring whether the prevalence of predictive feed-back signals in deep layers observed in the visual system applies elsewhere (Finn et al 2020), there is also urgent need to perform the kind of laminar recordings that have benchmarked the foundations of this approach in the visual system (Self et al 2019) across posterior parietal and premotor regions.…”

“…Despite these limitations, we hope that this study will serve as a proof of concept that depth-resolved analyses can be performed during naturalistic viewing in higher cognitive brain regions using intersubject correlation-based approaches, and that we may contribute to what Finn and colleagues called for: "layer fMRI is now at a point where we can expand from tightly controlled experiments in sensory cortex with clear hypotheses-which were necessary to show feasibility of the technique-to more exploratory, data-driven investigations of functional dynamics both across the cortical hierarchy as well as within higher-order regions themselves. […] Data acquired during naturalistic stimulation-e.g., movie watching-lend itself to both connectivity and activation analyses" (Finn et al 2020). Ultimately such studies may literally contribute to a deeper understanding of social cognition.…”

“…Recent work in particular confirms that activations in deep layers is indeed measurable when feedback predictions about upcoming sensory events are likely to occur in the visual system (Kok et al 2016;de Lange et al 2018;Aitken et al 2020), while activation in superficial layers seem to be measured when there is a mismatch between predicted and observed input and attentional signals (de Lange et al 2018). Unfortunately, experiments testing whether deep layers carry signals that could be related to predictions have so far been restricted to early visual cortices and we do not know whether the same might be observed in parietal and premotor regions (Finn et al 2020). Therefore here we acquired brain activity at 7T with an isotropic resolution of 0.8 mm in 9 participants, planning a partial-volume acquisition to include PFt, BA44 and LatOcc cortex of the left hemisphere (Supplementary Figure 1), while they observed intact and scrambled actions to test the validity of predictive coding models of action observation, specifically by testing for increased stimulus-dependent information in deep layers of PFt when seeing intact vs scrambled sequences.…”

“…Pioneering studies in the early visual cortex have shown that acquiring fMRI at submillimeter resolution at high-field strength (≥7T) can indeed test predictive coding hypotheses (Kok et al 2016;Lawrence et al 2018Lawrence et al , 2019aFinn et al 2020;Aitken et al 2020). Recent work in particular confirms that activations in deep layers is indeed measurable when feedback predictions about upcoming sensory events are likely to occur in the visual system (Kok et al 2016;de Lange et al 2018;Aitken et al 2020), while activation in superficial layers seem to be measured when there is a mismatch between predicted and observed input and attentional signals (de Lange et al 2018).…”

LayNii: A software suite for layer-fMRI

“…It’s Github site welcomes >1400 unique visitors per year and its code is being cloned >100 times per month. It found particular application for layer-fMRI studies in the motor cortex (Huber et al 2017), in sensory cortex (Yu et al 2019), DLPFC (Finn et al 2019), across association cortices (Finn et al 2020) for columnar imaging in the motor cortex (Huber et al 2020b), in layer-specific functional connectivity mapping ( Huber et al 2020a), for mental imaginary layer-fMRI (Persichetti et al 2020), for methods development of new sequences (Beckett et al 2020; Chai et al 2019; Guidi et al 2020), for visual layer-fMRI (Zamboni et al 2020), for model-based removal of vein effects in layer-fMRI-EEG (Marsh et al 2020) and for methods debugging of human 9.4T layer-fMRI (Huber et al 2018). In the early days of LayNii, its programs were shaped and optimized by continuous interactions with its users.…”