Flexible recruitment of memory-based choice representations by human medial-frontal cortex

Abstract

Decisions in complex environments rely on flexibly utilizing past experience as required by context and instructions. This process depends on medial frontal cortex (MFC) and the medial temporal lobe (MTL), but it remains unknown how these structures interact during memory retrieval. We recorded single neurons in MFC and MTL while human subjects switched between making memory and categorization-based decisions. Here we show that MFC rapidly implements changing task demands by utilizing different subspaces of neural activity during different types of decisions. Choices requiring memory retrieval selectively engaged phase-locking between thetafrequency band oscillations in MTL and MFC neurons. Choice neurons signaled decisions independent of output modality. In contrast, no effect of task demands was seen locally in the MTL. This work reveals a mechanism for selectively engaging memory retrieval and shows that unlike perceptual decision-making, memory-related information is only represented in frontal cortex when choices require it.

A representational similarity analysis of cognitive control during color-word Stroop

“…One speculative interpretation is that, during the relatively late phase of response selection and execution, control networks may lose modular structure as the circuitry collectively converges on a behavioral choice. This interpretation accords with the fact that “choice axes” are encoded within multiple key nodes of frontoparietal decision circuitry (e.g., in macaque LIP: Roitman & Shadlen, 2002; in macaque caudal DLPFC: Mante et al, 2013; in human DACC and pre-SMA: Minxha et al, 2020). This account could be addressed using the enriched experimental design described above, identifying when and where choice coding is emphasized over the course of a trial.…”

“…Via RSA, rule-dependent coding of target features could now be isolated from rule-independent coding of stimulus hue, word, and response-related information, in addition to coding of abstract task rules (Figure 5, A, B; see also Hall-McMaster, Muhle-Karbe, Myers, & Stokes, 2019). By incorporating an additional manipulation of response modality, this response-related information could be further decomposed into abstract choice (effector independent) versus motoric coding (Figure 5, A, C; a la Minxha, Adolphs, Fusi, Mamelak, & Rutishauser, 2020). In these expanded designs, the interplay of the additional distinct coding schemes — within or between brain regions, with more classical neural correlates of control (e.g., incongruency response), or with behavioral measures — could then be properly modeled, perhaps jointly (Turner, Forstmann, & Steyvers, 2019) at the between-subjects or single-trial (Kikumoto & Mayr, 2020) levels.…”

“…Now, coding of true abstract choice options or decision variables, independent of effector, could be separated from “lower-level” motor-related (effector-dependent) coding. For an elegant example of a similar manipulation, see Minxha et al (2020).…”

Inferring brain-wide interactions using data-constrained recurrent neural network models

“…We next applied the method to cellular resolution, multiregion, spiking electrophysiology recordings from humans. Five participants performed a set of two memory tasks in eight interleaved blocks (Figure 6a-c, see Methods) 40 . In the first, participants categorized images based on high level sensory features.…”

“…Using the same procedure as in the monkey dataset, we constructed pseudopopulations from neurons recorded from between two and five sessions in each participant. Since some participants had few recorded neurons from either hippocampus or amygdala, we combined them for later analyses 40 . Memory retrieval is believed to be mediated by interactions between frontal cortices and the H/A 40 .…”