Exercise accelerates place cell representational drift

eTOC Blurb

Representational drift is broadly observed in the brain but the mechanisms that drive it are unknown. Here, de Snoo et al. show that voluntary exercise, a behavioral intervention associated with neural circuit remodeling, is sufficient to accelerate representational drift in CA1 place cells.

Stable neural ensembles are often thought to underlie stable learned behaviors and memory. Recent longitudinal experiments, however, that tracked the activity of the same neurons over days to weeks have shown that neuronal activity patterns can change over extended timescales even if behaviors remain the same – a phenomenon termed representational drift.¹ Here we test whether neural circuit remodeling (defined as any change in structural connectivity) contributes to representational drift. To do this, we tracked how hippocampal CA1 spatial representations of a familiar environment change with time in conventionally housed mice relative to mice housed with a running wheel. Voluntary exercise is an environmental stimulus that promotes hippocampal circuit remodeling, primarily via promoting adult neurogenesis in the dentate gyrus. Adult neurogenesis alters structural connectivity patterns since the integration of adult-generated granule cells (abGCs) is a competitive process where new input/output synaptic connections may co-exist and/or even replace existing synaptic connections.² Comparing the spatial activity of downstream hippocampal CA1 place cells in the same familiar environment over 2 weeks, we found that the activity of place cells in exercise mice exhibited accelerated representational drift compared to control mice, suggesting that hippocampal circuit remodeling may indeed drive representational drift.

Decades of behavioral neuroscience research have shown that repeated behaviors are accompanied by reliable neuronal activity. However, as technological advances have made it possible to track the activity of the same neurons for days to weeks, representational drift has been observed in many brain regions including the primary visual cortex,³ posterior parietal cortex,⁴ anterior piriform cortex,⁵ and the hippocampus.^6,7 While frequently observed, little is known about the underlying environmental and neurophysiological mechanisms of representational drift. One possibility is that representational drift is the consequence of dynamic turnover of synaptic connections.⁸ In this view, neuronal firing properties inevitably change in response to changes in their physical inputs. However, no studies to date have directly examined whether manipulations that induce synaptic turnover affect the rate of representational drift.

To explore this possibility, we implanted transgenic Thy1-GCaMP6f mice with gradient index (GRIN) lenses above the dorsal CA1 of the hippocampus (Fig 1A,B) to image the activity of individual pyramidal neurons, and then manipulated rates of synaptic turnover by randomly assigning them to either control (conventional housing; N = 6) or exercise (continuous home-cage access to a voluntary running wheel; N = 5) conditions. Voluntary exercise is associated with neural circuit reorganization in the hippocampus primarily via upregulation of neurogenesis,² and we confirmed that exercise was proneurogenic in a separate cohort of mice (Fig 1E). To allow sufficient time for abGCs to functionally integrate into the hippocampal circuitry,² exercise was initiated 3 weeks before imaging. During imaging, we measured rates of representational drift in the spatial representations formed by the mice as they navigated the same 91 cm linear track over 2 weeks (Fig 1C & Fig S1A). To rule out potential contributions of novelty or learning to representational drift, mice were extensively pre-trained on the linear track for 2 weeks before beginning imaging (Fig 1D).

Figure 1. Exercise enhances place cell representational drift.

(A,B) Transgenic Thy1-GCamP6f mice were implanted with GRIN lens above their dorsal CA1 to image pyramidal neuron activity. (C,D) Mice were trained to run on a 91 cm linear track and after a 2-week period of familiarization their neuronal activity was recorded during track running every other day for 2 weeks. Comparisons were made between control (conventionally housed; N = 6) and exercise (continuous home-cage access to a voluntary running wheel; N = 5) mice. (E) Representative images (left) and quantification (right) of adult-born neurons (EdU⁺NeuN⁺) from control and exercise mice (p < 0.001, unpaired t-test). (F) In control and exercise mice, a comparable proportion of the recorded CA1 neuronal populations were active on any given day, and the propensity of active cells to have place fields was equivalent. (G) Spatial activity curves of the same place cells over 2 weeks, sorted by their peak activity on day 1. Place cells in both control and exercise mice change their spatial activity patterns over time. (H) Pearson correlation of a place cell’s spatial activity during a given session compared to its activity during another session N days apart (mean ± 95% confidence interval). Place cell spatial activity from exercise mice exhibits a faster decline in correlation with time (p < 0.01 for the interaction Group x Days apart, linear mixed effects model with fixed effects for Group, Days apart, and their interaction, and random effects for Mouse and Place cell).

We recorded 1972 and 1546 unique CA1 neurons (329±53 and 309±32, mean ± SEM per mouse) from control and exercise mice, respectively. On any given day, both groups had similar proportions of active neurons (Fig 1F & Fig S1B) and place cells (Fig 1F & Fig S1C,D), and both groups formed populations of place cells with spatial activity fields that spanned the entire length of the linear track (Fig 1G & Fig S1E,F), indicating similar same-day spatial representations. When we sorted the activity of the same place cells on subsequent days by their peak activity on day 1, we found that the spatial activity of place cells from control and exercise mice becomes more dissimilar with time (Fig 1G & Fig S1E,F), reflecting representational drift. To quantify the rates of representational drift across groups, we calculated the Pearson correlation of each place cell’s spatial activity curve on any given day to its spatial activity curve during sessions N days later (Fig 1H). Correlations declined faster in exercise, compared to control, mice indicating greater rates of representational drift (Fig 1H., p < 0.01 for the interaction of Group × Days Apart, linear mixed effects model with fixed effects for Group, Days Apart, and their interaction, and random effects for Mouse and Place Cell). This difference in representational drift is unlikely to be explained by differences in behavior since control and exercise mice had similar running speed and completed similar numbers of trials throughout the imaging sessions (Fig S2).

It has been proposed that representational drift may serve a computational purpose by updating the population of active neurons in which new information can be incorporated while avoiding catastrophic interference.¹ While observations of representational drift have been frequently reported,^1,3–5 studies characterizing factors that impact the rate of drift have been relatively rare. Schoonover et al. 2021 showed that the frequency at which an odor is presented, rather than any change in salience (e.g., by pairing the odor with a foot shock), affects the drift rate of odor representations in the anterior piriform cortex.⁵ It is possible that dynamic environments generate more uncertainty about the external world and necessitate greater representational drift to effectively sample the learning space and result in the retention of generalized semantic information while discarding episodic specificity. Our results are in accordance with this idea and additionally suggest a possible explanation – that dynamic environments increase synaptic turnover (e.g., via promoting adult neurogenesis) which in turn increases representational drift of the neuronal population. This is consistent with the observations of accelerated forgetting of contextual fear memories⁹ and the enhanced ability to learn conflicting information¹⁰ following exercise. In the future, more selective manipulations of adult neurogenesis can test whether (1) interventions that prevent neurogenesis-mediated circuit remodeling block exercise-induced accelerations in representational drift and (2) if interventions that promote neurogenesis-mediated circuit remodeling, in the absence of exercise, are sufficient to accelerate representational drift.

Supplementary Material

1

Document S1. Experimental Procedures and Two Figures