Neural and vascular resting-state cortical network dynamics

2037 

Abstract Submission 

Lisa Meyer-Baese¹, Michael Bian², Dieter Jaeger², Shella Keilholz³

¹Georgia Institute of Technology and Emory University, Atlanta, GA, ²Emory University, Atlanta, GA, ³Georgia Institute of Technology, Atlanta, GA

Lisa Meyer-Baese  
Georgia Institute of Technology and Emory University
Atlanta, GA

Michael Bian  
Emory University
Atlanta, GA

Dieter Jaeger  
Emory University
Atlanta, GA

Shella Keilholz  
Georgia Institute of Technology
Atlanta, GA

Dynamic resting state fMRI (rsfMRI) is altered in a wide range of otherwise indistinguishable disease states, revealing potential biomarkers of neurological and psychiatric disease (Brier et al., 2012; Grieder, Wang, Dierks, Wahlund, & Jann, 2018). Yet we do not know how this activity varies across spatiotemporal scales and how it relates directly to changes in neural activity. Wide-field optical imaging as a technique allows us to image both neural and hemodynamic activity across dorsal cortex in awake mice to better understand the dynamics of these resting state networks. Here we look at dynamic changes in resting-state cortical activity as measured with a voltage fluorescent sensor that is optimized to capture the subthreshold voltage membrane activity and hemodynamics across dorsal cortex (Akemann et al., 2012). This provides insight into how the dynamics of time-varying neural activity relate to dynamic changes in hemodynamics.

Awake mice (n = 5) expressing the VSFP-Butterfly 1.2 voltage-based fluorescent sensor in excitatory neurons in all layers of cortex were imaged using our wide-field imaging set-up. From this imaging technique we get voltage and hemodynamic activity per pixel, this data is processed separately through the pipeline (Fig 1). All frames were aligned to the Allen Brain Mouse Atlas (Fig 2. C), a mask of the 2D Allen Atlas was created in MATLAB and applied which set all pixels outside of cortex to zero. Further masking was performed to remove the vasculature and artifacts from the glass coverslip by cropping out the midline and part of anterior and posterior cortex (Fig 2. A) Analysis was done on a group level using the pipeline displayed (Fig 1). Briefly, data was concatenated and the dimensionality was reduced, using spatial non-negative matrix factorization (NMF). Only spatial maps reflecting cortical activity were selected to reconstruct the original data (Ren & Komiyama, 2021). Data was then embedded into a 2D subspace using t-stochastic neighborhood embedding (t-SNE) creating a continuous embedding space. Segmenting this embedding using the inverse watershed transform resulted in distinct cortical states that can be assigned to each time point in the imaging data.

A total of 13 cortical states for the voltage activity and 9 cortical states for the hemodynamic activity were identified by segmenting the respective 2D density embeddings for all animals (Fig 2, E and F). We are showing the top 6 states for each of the two signals sorted by percent dwell time (Fig 2., B and D). The obtained cortical states represent symmetrical, activation/deactivation of explicit cortical areas. This includes somatosensory areas closer to the midline, prefrontal cortex, and more lateral auditory/visual sensory areas. We found in both cases one dominant state (state 1) which represented low levels of positive activity across most of dorsal cortex aside from the somatosensory areas close to the midline which were slightly negative. Looking at the transition probabilities for different states for both signals the probability of transitioning to state 1 from any other state was the highest. Additionally, we noticed that the probability of staying within a given state was higher for the hemodynamic signal than for the voltage.

A challenge involved in interpreting time-varying rsfMRI connectivity metrics is that we know little about how the dynamics of neural activity relate to hemodynamics across spatiotemporal scales. We've found on average that dynamic changes in resting state cortical voltage activity can be summarized by a total of 13 states while hemodynamic activity is represented by 9 states. All states exhibited bilateral symmetry, a common feature of resting state activity. Observed differences across states were defined by the activation/deactivation of distinct cortical areas. For both signals, one dominant resting state existed with the probability of remaining in the same state being higher for the hemodynamics.

Methods Development ²

Task-Independent and Resting-State Analysis ¹

Cortex

Data analysis

Machine Learning

Optical Imaging Systems (OIS)

Other - Dynamics