Sensory experience steers representational drift in mouse visual cortex

Abstract

Representational drift-the gradual continuous change of neuronal representations-has been observed across many brain areas. It is unclear whether drift is caused by synaptic plasticity elicited by sensory experience, or by the intrinsic volatility of synapses. Here, using chronic two-photon calcium imaging in primary visual cortex of female mice, we find that the preferred stimulus orientation of individual neurons slowly drifts over the course of weeks. By using cylinder lens goggles to limit visual experience to a narrow range of orientations, we show that the direction of drift, but not its magnitude, is biased by the statistics of visual input. A network model suggests that drift of preferred orientation largely results from synaptic volatility, which under normal visual conditions is counteracted by experience-driven Hebbian mechanisms, stabilizing preferred orientation. Under deprivation conditions these Hebbian mechanisms enable adaptation. Thus, Hebbian synaptic plasticity steers drift to match the statistics of the environment.

Fig. 1. Preferred orientation of V1 neurons drifts over time.

a Example field of view from one mouse. Top: Average GCaMP6s (green) and mRuby2 (red) fluorescence, scale bar 100 µm. Bottom: imaging timeline of six mice, white squares are days with imaging sessions. b Correlation of the pairwise signal correlation (PSC) matrices compared across all imaging sessions of one mouse. Red: exponential decay fit ($y=0.31+0.43e^{-0.09*x}$). c Left: responses of an example neuron on 3 days. Single-trial responses, gray, average black. Gray bars: stimulus window of 5 s, grating directions indicated above. Scale bar: 100 ΔF/F. Right: polar plots of the responses on the left, with gray lines indicating mean response. Preferred orientation (PO) in blue with 95% confidence intervals (CI) as blue dashed lines. d Left: POs 1 day apart. Two-sided circular-circular Pearson’s correlation r = 0.968 (p < 1 × 10⁻¹⁶), n = 781 PO changes from 169 neurons from six mice. Right: POs 20 days apart. r = 0.800 (p < 1 × 10⁻¹⁶), n = 360 PO changes from 170 neurons from six mice. Red: significant changes, gray: non-significant changes. e Percentage of concurrently tuned cells that significantly changed their PO vs. interval length. Six individual mice as gray lines, with black line as mean with error bars as S.E.M. One-way ANOVA F(61)14.98, p = 5.62 × 10⁻¹⁴. Asterisks indicate Dunnett’s post hoc test (1–2 days vs. all), p < 0.05. f Cumulative probability distributions of the absolute size of PO changes (|∆PO|; drift magnitude) for different intervals. g Median drift magnitude for all PO changes in black and only significant PO changes in red. Error bars are bootstrapped 95% CIs. Two-sided Kruskal–Wallis test on all PO changes (χ²(12) 864, p = 2.9 × 10⁻¹⁷⁷, n = 468–4194 PO changes), and on only significant PO changes (χ²(12) 122, p = 2.4 × 10⁻²⁰, n = 188 to 347 PO changes). Asterisks indicate Bonferroni corrected two-sided Mann–Whitney U tests between 1 and 2 days vs. all other intervals p < 0.05. h Cumulative probability distribution of drift magnitude. Cyan: 1–2 day intervals, n = 2147 PO changes from five mice; magenta: 19–20 day intervals, n = 657 PO changes from five mice; black: changes between still trials and running trials within sessions, n = 152 PO changes from five mice. Two-sided Mann–Whitney U test for within-session changes between running and still trials compared to changes across short time intervals (U = 9.78 × 10⁶, p = 1.04 × 10⁻⁵), or long time intervals (U = 1.11 × 10⁶, p = 2.13 × 10⁻⁸). Source data are provided as a Source Data file.

Fig. 2. Orientation deprivation leads to PO convergence.

a Photograph of a visual scene in a mouse cage taken without (top) and with (bottom) cylinder lens goggles (inset). Note that only one orientation is present in the bottom image. b Distributions of POs relative to the experienced orientation (at 0°; relative PO: rPO) before (gray) and after (green) 28d of orientation deprivation. Cell numbers are shown as a percentage of the initial total cell count (835 neurons from seven mice). Mean across mice with S.E.M error bars. c Median convergence (∆|rPO|) for different time intervals under normal visual conditions (black, six mice) and orientation-deprived visual conditions (orange, eight mice deprived for seven days; green, seven mice deprived for 28 days). Positive ∆|rPO| indicates changes towards the experienced orientation. Error bars are bootstrapped 95% CIs. d PO convergence could result from neurons whose PO is similar to the experienced orientation (black triangle) drifting less than those with more dissimilar POs. e Alternatively, neurons may drift towards the experienced orientation, while drift magnitude is unaffected. f Initial PO difference from experienced orientation is uncorrelated with drift magnitude (|∆rPO|) after 28-day deprivation. Two-sided Spearman’s correlation r = −0.034 (p = 0.950, n = 414 neurons from seven mice). g Median convergence (∆|rPO|) before and after shuffling drift magnitudes. Error bars are bootstrapped 95% CIs. Two-sided Wilcoxon signed rank z = 0.807, p = 0.419, (n = 414 neurons from seven mice). h Initial PO difference from experienced orientation plotted against convergence (∆|rPO|). n = 414 neurons from seven mice. Gray areas indicate impossible values. i Median convergence (∆|rPO|) before and after shuffling drift directions. Error bars are bootstrapped 95% CIs. Two-sided Wilcoxon signed rank z = 4.19, p = 4.94 × 10⁻⁵ (n = 414 neurons from seven mice). j Distributions of POs relative to the experienced orientation (0°): initial (gray), after 28d orientation deprivation (green), and after 21d of recovery (blue). Cell numbers are shown as a percentage of the initial total cell count. N = 7 mice. Mean across mice with S.E.M error bars. k Relative change in cell numbers from initial to post recovery, for two PO bins (±0° to ±45° and ±45° to ±90° from the experienced orientation). Gray lines are seven individual mice, black line is mean. Two-sided one-sample T tests were used to compare within bin changes (t(6)−0.309, p = 0.768 and t(6)0.998, p = 0.357). l Median PO convergence during orientation deprivation vs. during recovery with 95% CI (n = 335 neurons from seven mice). Source data are provided as a Source Data file.

Fig. 3. Network model indicates PO drift is a trade-off between Hebbian plasticity and synaptic volatility.

a Model schematic: two-layer network with $n$ orientation-tuned presynaptic neurons (pre) connected topographically with weights $\left(w\right)$ to $n$ orientation-selective postsynaptic neurons (post). $n$ = 500. b Synaptic plasticity rule: weights $\left(w\right)$ are updated in proportion to the sum of stimulus-driven changes ($H$; scaled by k) and synaptic volatility ($\xi$), both scaled by a propensity factor proportional to initial weight ($\rho \left(w\right)$). c POs of postsynaptic neurons in the model 20 days apart, under baseline stimulus conditions. d Median drift magnitude (|∆PO|) increases over time, and is comparable under baseline (black) and orientation-deprived (green) conditions. e Same as d but for mean drift rate (|PO_day–PO_day–1|). f Median convergence (∆|rPO|) over time for baseline (black) and orientation-deprived (green) conditions. g Initial distance from experienced orientation (|rPO|) shows little correlation with drift magnitude after 28-day deprivation. Two-sided Spearman’s correlation r = 0.118 (p = 0.008, n = 500). h Spearman’s correlation between drift magnitude and initial distance from experienced orientation (as shown in g) in networks with different ratios of Hebbian plasticity to synaptic volatility ($H/\xi$; see b). Synaptic volatility is constant, while the Hebbian component is scaled by k. Correlation increases with longer deprivation time. Large Hebbian plasticity component also leads to increased correlation. Dashed lines indicate $H/\xi$ ratio in the other panels and 28-day deprivation length. i Shuffling drift direction but not magnitude abolishes the median convergence effect of the model during orientation deprivation. Green: model data and shuffles (n = 500). Gray: experimental data from Fig. 2g, i, median convergence and shuffles. Error bars are bootstrapped 95% CIs; n = 414 neurons from seven mice. j Effect of omitting either the Hebbian or the synaptic volatility contribution from the plasticity rule on PO convergence, during 28 days of orientation-deprived conditions. k Effect of omitting either the Hebbian or the synaptic volatility contribution from the plasticity rule on PO drift magnitude, during 28 days of baseline stimulation conditions. l The model displays limited recovery after input statistics return to baseline conditions. During recovery, median convergence is negative and slowly increases in magnitude over time (blue), but is incomplete even after 1000 days (~3 years). Mean (solid line/data point) and standard deviation (shaded region/error bars) over 50 model iterations. Source data are provided as a Source Data file.