Project Grant F31MH143476

A COMPUTATIONAL MODEL LINKING MICROCIRCUIT PROPERTIES TO CORTEX-WIDE ACTIVITY DURING DECISION MAKING - PROJECT SUMMARY/ABSTRACT ORGANISMS CONSTANTLY MAKE DECISIONS AS THEY MOVE THROUGH THEIR ENVIRONMENT AND GATHER EVIDENCE FOR OR AGAINST TAKING CERTAIN ACTIONS. THE PROCESSES OF ACCUMULATING INFORMATION AND COMMITTING TO A CHOICE ARE ASSOCIATED WITH DISTINCT PATTERNS OF WIDESPREAD CORTICAL ACTIVITY. BUT HOW DO THESE PATTERNS EMERGE AND HOW ARE THEY SHAPED BY CIRCUIT PROPERTIES AT MULTIPLE SPATIAL SCALES? ONE FEATURE OF CORTICAL NETWORKS CRITICAL FOR UNDERSTANDING THESE PATTERNS OF ACTIVITY IS PROPOSED BY THEORETICAL WORK: SPONTANEOUS TIMESCALE, A MEASUREMENT OF ACTIVITY PERSISTENCE WITHIN A NETWORK. DATA FROM OUR LAB SHOWS THAT, DURING SHORT-TERM MEMORY, THE PATTERN OF LARGE-SCALE CORTICAL RECRUITMENT MIRRORS ITS GRADIENT OF SPONTANEOUS ACTIVITY TIMESCALES, WHICH INCREASE ALONG THE ANATOMICAL HIERARCHY, FROM POSTERIOR TO FONTAL AREAS. THIS HIERARCHY OF TIMESCALE IS IMPLICATED IN DECISION MAKING, WHERE PRIMARY SENSORY CORTEX HAS SHORTER TIMESCALES THAT LEND IT THE CAPABILITY TO INTEGRATE SENSORY INFORMATION, AND ASSOCIATION AREAS HAVE LONGER TIMESCALES NECESSARY FOR INFORMATION MAINTENANCE. THIS HIERARCHY CO-VARIES WITH GRADIENTS IN CORTICAL CYTOARCHITECTURE AND CELLULAR PROPERTIES, SUCH AS LEVELS OF EXCITATORY RECURRENCE, PROPORTIONS OF INHIBITORY CELL TYPES, AND INTRINSIC MEMBRANE TIME CONSTANTS OF INDIVIDUAL NEURONS. HOWEVER, THE RELATIVE CONTRIBUTIONS OF THESE PROPERTIES TO SETTING NETWORK TIMESCALE AND CORTICAL RECRUITMENT PATTERNS DURING DECISION MAKING REMAIN UNKNOWN. THIS KNOWLEDGE GAP WILL BE ADDRESSED BY CREATING A COMPUTATIONAL MODEL OF SPONTANEOUS AND TASK-DRIVEN ACTIVITY ACROSS THE CORTEX, CONSTRAINED BY KNOWLEDGE OF AREA-LEVEL MICROCIRCUITRY AND BY FUNCTIONAL DATA FROM MULTIPLE CORTICAL CELL TYPES AND AREAS. SPECIFICALLY, IN AIM 1, WIDEFIELD CA2+ IMAGING DATA WILL BE USED TO MEASURE THE SPONTANEOUS ACTIVITY TIMESCALES OF DIFFERENT EXCITATORY AND INHIBITORY GENETICALLY DEFINED CELL TYPES IN THE CORTEX. IN AIM 2, MULTI-AREA RECURRENT NEURAL NETWORKS (RNNS) WILL BE CONSTRUCTED, JOINTLY CONSTRAINED BY THE NEURAL DATA AND OTHER KNOWN BIOLOGICAL PROPERTIES OF CORTICAL CIRCUITRY, SUCH AS LONG-RANGE CONNECTIVITY AND THE CYTOARCHITECTONIC PROPERTIES ABOVE. THESE RNNS WILL BE TRAINED TO MATCH BOTH SPONTANEOUS AND TASK-DRIVEN CORTICAL ACTIVITY. THE LATTER HAS BEEN PREVIOUSLY MEASURED IN OUR LAB DURING A NEW DECISION-MAKING TASK FOR MICE NAVIGATING IN VIRTUAL REALITY, WHICH DISSOCIATES SHORT-TERM MEMORY FROM OTHER COGNITIVE PROCESSES. PERFORMANCE OF THE TASK-SOLVING MODEL WILL BE COMPARED WHEN INITIALIZED RANDOMLY OR WITH THE SPONTANEOUS MODEL, WHERE IT IS HYPOTHESIZED THAT THE ORGANIZATION OF SPONTANEOUS TIMESCALE WILL BE LEVERAGED SO THAT THE MODEL CONVERGES MORE QUICKLY DURING TRAINING. FINALLY, IN SILICO PERTURBATIONS WILL BE PERFORMED TO RELATE NETWORK STRUCTURE TO ITS ORGANIZATION OF SPONTANEOUS TIMESCALES AND FUNCTIONAL ROLES DURING DECISION-MAKING. THIS MODEL WILL LEAD TO PROPOSED CIRCUIT MECHANISMS THAT LINK NETWORK STRUCTURE TO ITS COGNITIVE FUNCTION.