Sensory experience steers representational drift in mouse visual cortex

Bauer, Joel; Lewin, Uwe; Herbert, Elizabeth; Gjorgjieva, Julijana; Schoonover, Carl E.; Fink, Andrew J. P.; Rose, Tobias; Bonhoeffer, Tobias; Hübener, Mark

doi:10.1038/s41467-024-53326-x

Download PDF

Article
Open access
Published: 23 October 2024

Sensory experience steers representational drift in mouse visual cortex

Nature Communications volume 15, Article number: 9153 (2024) Cite this article

7360 Accesses
12 Citations
4 Altmetric
Metrics details

Subjects

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.

A model for the development of binocular congruence in primary visual cortex

Article Open access 25 July 2022

Coordinated drift of receptive fields in Hebbian/anti-Hebbian network models during noisy representation learning

Article 12 January 2023

Large-scale synaptic dynamics drive the reconstruction of binocular circuits in mouse visual cortex

Article Open access 01 July 2025

Introduction

A growing body of evidence indicates that stimulus-evoked neuronal responses change gradually over days to weeks, even in the absence of any experimental manipulation^{1,2,3,4,5,6,7}. This phenomenon, now referred to as representational drift^8,9, varies widely in magnitude across sensory and associative cortices, as well as stimulus types^{2,3,4,5,6,7,10}.

Little is known about the causes of representational drift. Is it the consequence of activity-independent, and therefore experience-independent, synaptic volatility^11,12,13, or does it reflect the effect of experience-dependent synaptic changes^14,15 ? If drift is caused by synaptic volatility alone, its dynamics should resemble a random walk in stimulus representation space^8,16. Conversely, if drift is a function of experience, its dynamics may be directed to reflect that experience. Several independent observations in different brain regions have indicated that drift can be sensitive to an animal’s experience between recordings. For example, in the piriform cortex, the daily experience of an odorant halves the drift rate to that particular stimulus⁵. Conversely, recent studies of factors contributing to drift in spatial tuning of hippocampal place cells have reported a destabilizing effect of experience^17,18.

Historically, neuronal responses in primary sensory areas have been considered relatively stable in adult animals. In the primary visual cortex (V1), the stability and perturbation resistance of large-scale functional maps for retinotopy, orientation, and ocular dominance^19,20,21,22 suggest that these basic tuning features are stably encoded. More recent chronic recordings of single-neuron activity in the mouse have confirmed this high level of stability and perturbation-resistance for certain aspects of V1 responsiveness, including preferred orientation (PO), ocular dominance, and spatial frequency preference^23,24,25,26. Other measures, however, such as tuning curves^3,27, especially for responses to complex stimuli and natural movies^10,27, have been found to drift to varying degrees.

The PO of a mouse V1 neuron—the contour orientation that elicits the strongest response—has previously been reported to be relatively stable^23,24,25,26. This is puzzling in light of continuous synaptic changes in V1 (observed as dendritic spine turnover^28,29,30), and reports that orientation tuning curves measured on separate days become progressively more dissimilar with time^3,10,27. A columnar organization for orientation preference, which exists in the visual cortex of many mammals, could place anatomical constraints on PO drift by ensuring that the set of synapses forming onto and retracting from a cortical neuron prefer the same orientation⁷. However, rodents lack clear orientation columns³¹, raising the question of how PO could be stable. This makes the PO of neurons in mouse V1 an ideal model for studying the effect of experience on drift.

Here, we use PO in mouse V1 to determine whether the dynamics of representational drift depend on the animal’s ongoing experience of its visual environment. We hypothesize that PO stability emerges from animals’ frequent experience of oriented contours during the normal visual experience and that the consequent coactivation of similarly tuned neurons maintains the existing connectivity in the network against a background of continuous synaptic volatility^5,12. Using chronic two-photon calcium imaging in adult mice, we first show that PO undergoes modest drift in V1 neurons. We then restrict the range of orientations that the mouse experiences^20,32,33 for several weeks by applying cylinder lens goggles that limit visual input to contours of one orientation for several weeks^34,35,36. We find that this manipulation biases drift towards the remaining experienced orientation, but does not affect drift magnitude. Finally, we build a network model of PO drift and show that in this model, experience steers drift to match the input statistics via Hebbian plasticity. This mechanism stabilizes PO under normal visual conditions by counteracting the destabilizing effect of synaptic volatility.

Results

PO of V1 neurons drifts over time

To characterize drift in the orientation tuning of V1 neurons, we first quantified how the tuning of single cells changes over time. To this end, we presented gratings moving in 12 directions (six orientations) to awake, adult mice on 12–15 days over the course of a month while measuring neuronal activity using two-photon imaging of layer 2/3 cells expressing GCaMP6s²⁴ (Fig. 1a; Supplementary Fig. 1a). All mice were exposed to the full stimulus set on at least two separate days before the start of the experiment to avoid confounding effects from stimulus-response attenuation³⁷. We found that the similarity of neuronal responses between experimental sessions decreased as a function of time (Fig. 1b; Supplementary Fig. 1b), similar to previous reports of representational drift in the visual cortex^3,10,27.

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

We hypothesized that this decay in tuning curve similarity is at least partly caused by shifts in the PO of neurons. Indeed, we found neurons whose tuning curves shifted over days to weeks (Fig. 1c; Supplementary Fig. 1c). To quantify these changes, we calculated the PO for each repeatedly identified and responsive neuron (see Methods; Supplementary Fig. 2) on each day as the vector sum of responses to the different orientations. Using a trial resampling approach (bootstrapping, see Methods), we determined the 95% confidence interval of this PO measure for each day. If the confidence interval of a neuron was above 45°, it was considered untuned (i.e., was considered to have no PO) and excluded from further analysis for that day. We determined the PO change of a neuron across two days to be significant if the PO of each respective day was outside of the confidence interval of the other day. On the population level, the POs of neurons appeared more similar on two consecutive days compared to 20 days apart (Pearson’s correlation: r = 0.968 and 0.80 respectively; Fig. 1d). Quantifying this observation across sessions revealed that both the relative number of significant PO changes (Fig. 1e) and the magnitude of these PO changes (Fig. 1f, g) increased over time. Nevertheless, the overall PO drift rate was low, with a median drift rate of ~0.3°/day (calculated from all 19–20 day intervals), in line with findings that basic visual tuning features are rather stable and show only limited representational drift^10,25,26.

Previous studies have raised the question of whether representational drift may be due, at least in part, to changes in neuronal firing caused by differences in animal behavior or state between sessions rather than to actual changes in neural tuning^38,39. In V1, behavioral variables such as running have been shown to modulate the activity of neurons^40,41. However, running or behavioral state have negligible effects on PO, specifically^{40,42,43,44,45}. Thus, PO drift should be largely robust to this confound. Indeed, while we observed slight variability in PO between running and non-running trials within a single session, it was too small to explain PO drift over several days (Fig. 1h; in line with ref. ⁴⁵). Additionally, we quantified running and arousal modulation indices for each neuron and scaled these by the change in running or arousal across session, in order to obtain an estimate of the effective modulation by behavioral state. Excluding the 50% of PO changes that were most affected by behavioral state changes had little effect on the overall PO drift (Supplementary Fig. 1d–g). The PO drift we observed is, therefore, unlikely to be explained by behavioral changes across days.

Visual experience steers PO drift

Representational drift of PO is moderate compared to drift in other brain areas^2,4,5. A potential factor that could stabilize responses to a stimulus is how often it is experienced by the animal⁵. Therefore, we hypothesized that the PO of neurons in V1 is stabilized by the frequent occurrence of oriented contours during normal vision. To test this, we used cylinder lens goggles^34,35,36 to deprive mice of experiencing all but one orientation (Fig. 2a). In the past, changes in neural representations observed during baseline conditions have been referred to as drift, while changes during or after experimental manipulations (e.g., monocular deprivation) were referred to as experience-dependent plasticity or learning⁸. In this study, we will refer to all time-dependent PO changes as drift, including the ones occurring during orientation deprivation. This is because we consider representational drift to encompass such experience-dependent changes in representations and hypothesize similar synaptic mechanisms to underlie both phenomena.

**Fig. 2: Orientation deprivation leads to PO convergence.**

We found that applying the goggles continuously for four weeks biased the distribution of POs towards the experienced orientation (Fig. 2b; Supplementary Fig. 3a–c; two-sample T test, t(6)2.85, p = 0.029), consistent with single time point recordings in juvenile mice^35,36. Using longitudinal imaging, we characterized the change of individual neurons’ PO over the course of the experiment. In our analyses, we excluded neurons that gained or lost orientation tuning (252 and 171 neurons, respectively, from a total of 835), focusing on changes in PO. However, we found that the PO of neurons that gained orientation selectivity over the course of the experiment tended to be near the experienced orientation, though this effect was not significant (Supplementary Fig. 3d, e; one-sample T test on change in cell number with relative PO between 0° and 45° t(6)2.1, p = 0.080).

Defining the PO relative to the experienced orientation as rPO, we calculated the overall PO convergence of the population towards the experienced orientation (convergence: Δ|rPO|). We found that, on average, the PO of neurons drifted towards the experienced orientation (Fig. 2c; Supplementary Fig. 4a). Notably, the size of this PO convergence was dependent on the duration of orientation deprivation (Fig. 2c).

This convergence towards the experienced orientation could be explained by two possible factors: an experience-dependent change in the magnitude of PO drift (Fig. 2d), and/or a bias in the direction of PO drift towards the experienced orientation (Fig. 2e). Note, that drift magnitude does not take the direction of the PO change into account, while convergence does. We found that the initial PO of neurons and their drift magnitude were not correlated (Fig. 2f), and the overall PO drift magnitude of the population was largely unaffected by the altered visual experience (Supplementary Fig. 4b). Furthermore, shuffling drift magnitude across the population while leaving cell-wise drift directions intact had little effect on the overall convergence (Fig. 2g). On the other hand, most neurons’ PO drifted towards the experienced orientation rather than away from it (Fig. 2h) and shuffling the drift directions, while keeping cell-wise drift magnitudes intact, abolished the population convergence entirely (Fig. 2i; Supplementary Fig. 4c). Together, this demonstrates that depriving neurons of their PO does not lead to larger drift magnitudes, but rather steers their drift direction towards the experienced orientation.

If the absence of experience of most orientations during deprivation is the cause of PO drift towards the remaining experienced orientation, restoration of normal vision should recover the initial PO distribution. Further, intermittent exposure to all orientations should reduce the convergence effect. Indeed, we found that exposing the awake mice to drifting gratings during imaging for a few hours every seven days resulted in substantially lower convergence compared to uninterrupted orientation deprivation (Supplementary Fig. 4d). We also found that after ending deprivation, the initial distribution of POs was largely recovered, as the population drifted in the opposite direction than during the deprivation phase (Fig. 2j–l; Supplementary Fig. 4e).

Taken together, these data show that while the magnitude of PO drift of neurons in adult mouse V1 is unaffected by experience, the direction of PO drift is determined by the distribution of experienced orientations. Thus, PO drift is experience-dependent.

Hebbian plasticity and synaptic volatility can explain experience-dependent PO drift

To understand the nature of synaptic plasticity, which could lead to experience-dependent PO drift, we built a computational model of a neural network subject to synaptic plasticity (Fig. 3a). Previous studies have shown that synaptic changes can occur both in the presence and the absence of neuronal activity or NMDA-receptor mediated plasticity^13,46,47,48. These activity-independent synaptic fluctuations are referred to as intrinsic synaptic volatility^12,49, here, synaptic volatility. We, therefore, reasoned that the observed PO drift might be driven by a plasticity rule combining experience-dependent Hebbian ($H$) changes and experience-independent synaptic volatility ($\xi$). We modeled the change in feedforward synaptic weights from a cortical layer of presynaptic neurons to a cortical layer of postsynaptic neurons as the sum of $H$ and $\xi$ scaled by a synaptic weight-dependent propensity function ($\rho \left(w\right)$): $\Delta w=\rho \left(w\right)\left({kH}+\xi \right)$ (Fig. 3b). This propensity function was inspired by experimental results showing that the magnitudes of changes in spine size—commonly considered a proxy for synaptic strength—is proportional to the initial size of the spines^{47,50,51,52,53}. We initialized the feedforward weights from orientation-tuned presynaptic neurons as circularly-symmetric Gaussian distributions with varying widths, resulting in orientation-tuned postsynaptic neurons with varying tuning widths (Fig. 3a; Supplementary Fig. 5a–d). We modeled the experience-driven component $H$ as Hebbian changes—i.e., the product of presynaptic and postsynaptic activities—in response to a series of orientation stimuli, and $\xi$ as a baseline random synaptic change. The weights were subject to homeostatic normalization on a slow timescale^54,55,56 (four orders of magnitude slower than Hebbian changes; see Methods) to preserve the total sum of input weights per neuron.

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

To model the experience of mice under baseline conditions, reflecting normal vision, we stimulated the network with orientations sampled uniformly from −90° to 90°. A ‘day’ in our model consisted of one orientation stimulus per second for 12 hours. The model produced PO drift similar to our experimental data, with the POs of neurons drifting progressively over time (Fig. 3c, d; Supplementary Fig. 5e). To mimic the effect of orientation deprivation using cylinder lens goggles, we stimulated the network with a single orientation, instead of all orientations. Here, POs slowly drifted towards the experienced orientation (Fig. 3f). Neurons that are tuned far from the experienced orientation have weak connections with presynaptic neurons close to the experienced orientation (Supplementary Fig 5a), and these are not strengthened much due to the propensity function, raising the question of how these neurons are able to drift towards the experienced orientation. We find that these neurons shift because the synapses that are already strong and close to the neurons’ own PO are strengthened and that among these potentiated synapses, there is an asymmetry favoring synapses that are closer to the experienced orientation (Supplementary Fig 5f–h). However, day to day drift rate—not measurable from experimental data directly—remained constant (Fig. 3e). As in the experimental data described above (Fig. 2f), drift magnitude was largely independent of the initial PO within 28 days of deprivation (Fig. 3g; Supplementary Fig. 5i, j). However, the correlation between drift magnitude and initial relative PO gradually increased with time (Fig. 3h). This suggests that the lack of correlation between drift magnitude and initial PO, within the timeframe tested experimentally (Fig. 2f), was caused by synaptic volatility masking the effect of Hebbian plasticity, with the associated correlation only becoming evident after much longer time intervals. The rate at which this correlation increased was also strongly dependent on the relative strength of the Hebbian component (Fig. 3h). Shuffling the magnitude or direction of PO changes produced the same effect as in our data (Fig. 3i), demonstrating that, both in experimental data and model, convergence of the population was due to a bias in drift direction. In the model, Hebbian changes drove directional PO changes and thus convergence towards the experienced orientation (Fig. 3j; Supplementary Fig. 6b), while synaptic volatility introduced non-directional drift (Fig. 3k; Supplementary Fig. 6a). Importantly, baseline drift was highest in the absence of Hebbian changes (Fig. 3k; Supplementary Fig. 6a). Hence, Hebbian plasticity compensates for synaptic volatility under baseline conditions, and steers drift when input statistics change.

Our model also aligned with the experimental effect of interruptions during the deprivation period. Brief exposure to the full range of orientation stimuli every seven days slowed down PO convergence by transiently removing the directional bias, while leaving drift magnitude unaffected (Supplementary Fig. 6c). However, the model exhibited much slower recovery on a population level than observed experimentally. After deprivation had ended, the population displayed a negative convergence (divergence) that grew with time (Fig. 3l), but this never reached the magnitude observed in the experimental data (Fig. 2k). Likewise, the population distribution only partially recovered from the effect of deprivation, even after almost 3 years (Supplementary Fig. 6d).

With this, our model yields two specific predictions about PO drift beyond what we have found experimentally. First, the slower recovery of the original PO distribution in our model suggests that the plasticity rule employed is insufficient to fully explain the observed recovery in V1. Additional mechanisms may be present in the mouse visual cortex that accelerate recovery, such as backbone spines (synapses with little or no plasticity)⁸. The second prediction is that removing the Hebbian contribution experimentally would result in progressively larger PO drift magnitudes, both during baseline and deprivation conditions (Fig. 3k; Supplementary Fig. 6a, b). Thus, our model strongly suggests that under conditions of normal visual experience, stimulus-driven activity acts via Hebbian plasticity mechanisms to limit the destabilizing effect of synaptic volatility.

Discussion

PO of V1 neurons undergoes representational drift

Previous studies have shown that representational drift occurs in the visual cortex, mainly by quantifying the decay in response correlations over time^3,10,27. Some observations of drift have been explained by time-varying changes in behavior^38,39. We establish drift in the PO of individual neurons, a canonical visual response property, and demonstrate that these changes accumulate over time irrespective of locomotion and arousal. Representational drift is thought to originate from an accumulation of synaptic weight changes and synaptic turnover, while further response variability is caused by behavioral variability, internal state fluctuations, or measurement noise⁹. Time-varying changes in behavior and signal-to-noise ratio can result in time-dependent tuning curve changes that can be mistaken for representational drift when measurements and state changes occur on similar timescales³⁹. PO is well-suited to distinguish representational drift from such variability for two reasons. First, in contrast to tuning to other visual features, PO is state-independent, i.e. it is unaffected by running or behavioral state^{40,42,43,44,45}. Second, a decrease in signal-to-noise ratio reduces the accuracy of PO estimation but does not cause PO to trend towards a specific value. This is in contrast to direction selectivity, for instance, where a decrease in signal-to-noise would cause a tendency towards lower estimates of selectivity. The confounding effect of decreases in signal-to-noise ratio can, therefore, be mitigated by applying a constant confidence threshold on PO changes.

What controls PO drift?

Isolating and quantifying drift in a fundamental visual response property, PO, allowed us to explore the mechanisms that underlie drift. The cause of drift in neuronal tunings, such as the PO drift we observe, may be the accumulation of synaptic strength changes and turnover over time. Dendritic spine volume changes and turnover have been used in the past to estimate the extent of these changes and clearly demonstrate that some degree of change in connectivity does occur in mouse V1. However, the majority of spines persist for many weeks, and there is no consensus on the degree of baseline synaptic changes or its effect on neuronal response properties to date^28,29,30,57. These synaptic changes, while often associated with activity-dependent synaptic plasticity^46,58,59,60, also arise from activity-independent synaptic volatility^{11,12,47,48,49,61}. Accumulation of the apparently undirected activity-independent component should, in principle, lead to progressive changes in neuronal tuning. However, tuning curve similarity remains relatively high, i.e., the signal correlation asymptotes far above zero even after long time intervals^{3,10,23,24,25,26}, raising the question of what might limit the effect of synaptic volatility. Two potential mechanisms are anatomical constraints and compensatory plasticity^16,62,63.

Stability in neuronal tuning could arise from the anatomical constraints imposed by the spatial organization of axons and dendrites. The tuning properties of neurons depend on the availability of appropriate axons in the vicinity of their dendrites⁶⁴, which is largely determined during development, with little change observed during adulthood^65,66,67. The position of a neuron relative to a cortical map is, therefore, expected to correlate with its tuning stability. In support of this, a recent study in mouse somatosensory cortex found that layer 2/3 neurons tuned to the topographically aligned whisker are more stable than neurons tuned to a surrounding whisker⁷. We speculate that in mammals that have orientation columns in V1, neurons at the pinwheel center, where more diverse inputs are available, should drift more than those within an iso-orientation column⁶⁸. However, as there is little spatial organization of PO in rodent visual cortex³¹, PO drift of V1 neurons in mice should not be constrained by the availability of inputs.

Hebbian plasticity rules suggest that the repeated experience of sensory stimuli can stabilize the connectivity between, and the tuning of, co-tuned neurons through their coincident activity, ultimately leading to slower drift^8,16. This has been proposed for olfactory cortex, where responses to odor stimuli are more stable for odorants that are experienced more frequently⁵. We, therefore, hypothesized that orientation tuning in V1 could be stabilized by the frequent experience of all orientations. While we found that orientation deprivation had no observable effect on the magnitude of PO drift within the time frame tested, the direction of PO drift was biased towards the experienced orientation, resulting in an overall convergence of PO towards the remaining experienced orientation. This shows that visual experience steers PO drift rather than limiting its magnitude.

Hebbian plasticity and synaptic volatility control PO drift

Using a computational network model, we found that a combination of synaptic volatility and Hebbian plasticity driven by visual experience recapitulated the experimental dynamics of PO drift both during normal vision and orientation deprivation. POs changed as a function of time, and PO convergence increased as a function of deprivation duration. There was also next-to-no correlation between the initial PO of modeled neurons and the drift magnitude during deprivation, within a timeframe corresponding to that of our experiments. Our experimental findings, both during normal experience and deprivation, are therefore compatible with the hypothesis that the dynamics of PO drift in the mouse visual cortex are driven by a combination of two synaptic processes, synaptic volatility and experience-dependent Hebbian plasticity. However, this was only possible because of the weight-dependent propensity function, which favored plasticity changes at stronger synapses.

Beyond demonstrating plausibility, our theoretical model makes several clear predictions. Completely removing the experience-dependent Hebbian component increases drift under baseline conditions but prevents convergence during orientation deprivation. Conversely, removing synaptic volatility reduces drift and allows faster convergence. The model fits our data best when the contribution of synaptic volatility exceeds Hebbian learning. These predictions could be tested in future experiments using our paradigm while manipulating the relative ratio of the two synaptic plasticity components, e.g., by transiently blocking NMDA receptor-dependent plasticity.

In the absence of the experience-dependent Hebbian component, our model predicts that PO drift would increase. A recent study found that PO changes were higher during a baseline ~8-day interval of normal vision compared to a subsequent ~8-day interval of dark exposure⁶⁹. The authors concluded that experience has a destabilizing effect on tuning stability. This contrasts with both our experimental and modeling findings. It may, however, be partially explained by the sequential design of the dark exposure study. Over the first few successive visual stimulation sessions, neural responses tend to decrease and become sparser³⁷. This could lead to larger tuning differences between the control and dark exposure imaging sessions in the study by Jeon et al.⁶⁹. However, uniform dark exposure and partial visual deprivation very differently affect input statistics, and the resulting correlation structure of activity generates different outcomes⁷⁰. We, therefore, cannot exclude that there are additional mechanisms, not contained within our network model, which are engaged during dark exposure that cause lower drift magnitudes.

We suggest that experience-dependent synaptic changes exert a net stabilizing force on visual representations in the presence of synaptic volatility. This is similar to the stabilizing effect of the repeated experience of the same olfactory stimulus in piriform cortex⁵. In contrast, recent experiments in the CA1 region of mouse hippocampus have shown that repeated experience of the same environment leads to drifting spatial tuning^17,18, whereas the passage of time alone largely affects response magnitude¹⁸. A higher rate of Hebbian plasticity at hippocampal synapses may explain these results, as ongoing experience could rapidly induce random exploration of a solution space, as it occurs for networks trained using stochastic gradient decent^71,72. Alternatively, the difference in the rate and experience-dependence of drift between areas may be due to differences in the dimensionality, or degrees of freedom, of the representations⁷². Whatever the reason, these recent data, together with our own results, add to the growing body of work showing that the phenomenon of representational drift, as well as its mechanistic underpinnings, need to be interpreted in the context of the specific computation, network architecture, processing hierarchy, molecular cell type, and the ethological relevance of a circuit^72,73.

Methods

Animals

All experimental procedures related to animal handling were carried out in compliance with institutional guidelines of the Max Planck Society and using protocols approved by the Regierung von Oberbayern (Beratende Ethikkommission nach §15 Tierschutzgesetz). Female wild-type C57BL/6NRj mice (Janvier) were used for all experiments in this study. Mice were housed at 55 ± 5 % humidity and 22 ± 1.5 °C under a 12-hour inversed light-dark cycle with food and water available ad libitum. Mice were housed in large cages (GR900, Tecniplast) with access to a running wheel and other enrichment material such as a tunnel and a house. Animals were usually group-housed, except for short time periods during recovery from surgery.

Cranial window implant and intrinsic optical signal imaging targeted virus injections

In order to repeatedly measure visual tuning properties of neurons in mouse V1, neurons were virally transduced with the genetically encoded calcium indicator GCaMP6s, and a cranial window implanted over the visual cortex. Mice between P35 and P45 were anesthetized using a mixture of Fentanyl (0.05 mg/kg), Midazolam (5 mg/kg), and Medetomidine (0.5 mg/kg), injected intraperitoneally. Anaesthesia was maintained by injecting 25% of the original dose after the first two hours and then every hour. All surgical equipment was heat sterilized using a bead sterilizer and rinsed with ethanol. After checking for anaesthesia depth using the toe pinch reflex, the mice were placed on a thermostatically controlled heating blanket (set to 37 °C), and the head was fixed in a stereotaxic frame. Carprofen (0.5 mg/kg) was administered subcutaneously as an analgesic. Throughout the procedure, the eyes were kept moist and protected from debris using eye cream (Isopto-Max). The scalp was disinfected using iodine and ethanol, and lidocaine was applied as a local anaesthetic. The scalp was then removed along with the periosteum, and the hair along the rim of the wound was fixed using Histoacryl. The skull was roughened using a scalpel, and a custom aluminum head bar was fixed onto the skull using superglue (Pattex Ultra Gel). Dental cement (Paladur) was used to further secure the head-bar to the skull, cover the edges of the scalp wound, and cover the exposed skull (except for where the craniotomy would be placed). A rim of cement was built up around the front of the craniotomy in order to form a well. Diluted ultrasonic gel (diluted with cortex buffer in a ratio of 3:1; cortex buffer: in mM: 125 NaCl, 5 KCl, 10 glucose, 10 HEPES, 2 CaCl₂, and 2 MgSO₄) was placed in this well, and a 10 mm cover glass place over it, making sure to prevent bubbles. This increases the optical transparency of the skull for subsequent intrinsic optical signal (IOS) imaging.

The mice were then moved to an IOS imaging setup. An image of the blood vessel pattern over the visual cortex was taken with 530 nm illumination light (using a tandem 135 mm f/2.0 and 50 mm f/1.2 objective system, pco.edge 4.2 LT sCMOS camera or a Thorlabs B-Scope equipped with a ×4 objective and a 1500 MonoChrome camera). The skull was then illuminated with 735 nm light, and the focal plane lowered to ~400 nm below the surface of the brain. A 700–740 nm bandpass filter was placed in front of the camera sensor. The contralateral eye was covered with a cone made of tape and drifting gratings were presented 7–14 times on a screen 16 cm in front of the mouse (8 directions at 2 cycles/s and 0.04 cycles/° changing direction every 0.6 s, presented for 7 seconds each time, with a horizontal retinotopic position spanning 0 to −30° azimuth). The resulting intrinsic signal changes were analyzed using custom-written MATLAB code. This allowed the localization of V1, to aid targeting of virus injections.

The mice were then returned to the stereotaxic frame, and a 4 mm craniotomy was performed using a high-speed micro drill with the intended virus injection and two-photon imaging locations at the center. The skull was kept moist using cortex buffer during drilling. A 4 mm bone patch was removed, and the exposed cortex (still covered by dura) was washed repeatedly with cortex buffer until any bleeding stopped. The brain was covered with Gelfoam pieces soaked in cortex buffer to keep it moist and protected from debris.

Before the surgery, borosilicate glass injection pipettes were pulled (using a PC-97 pipette puller). Volumetric markings were drawn onto the borosilicate capillaries before pulling (marks were drawn every 1 mm using a black permanent marker, corresponding to 45 nl). The pipette tips were broken off so that the tip diameter was between 25 and 35 µm and beveled to a sharp point using a modified computer hard disk (Canfield 2006). The pipettes were front-loaded with the virus by placing a drop of virus on parafilm, lowering the pipette tip into the virus drop, and applying negative pressure. The pipettes were either loaded with a mixture of AAV2/1.1CamKII0.4.Cre.SV40 (titer 1-5 E8), AAV2/1.Syn.Flex.mRuby2.GSG.P2A.GCaMP6s.WPRE.SV40 (titer 1.28 E13) and AAV2/1.Syn.GCaMP6s.WPRE.SV40 (titer 6 E12), or AAV2/1.Syn.mRuby2.GSG.P2A.GCaMP6s.WPRE.SV40 (titer 1.2 E13). The loaded pipette tips were kept on ice until use.

The pipette was placed into a patch pipette holder and inserted 300–400 µm into the cortex using a micromanipulator. After waiting 4–5 min, virus was injected at ~50 nl/min with a total volume of 150–250 nl. This was done using 30–40 psi with 20–40 msec pressure pulses at 0.8 Hz controlled by a pulse generator (Master-8) and a pressure micro-injection system (Toohey Company). The pipette was left in the cortex for another 4-5 min before retracting it. The process was repeated for 2–5 injection sites, with a spacing of 50–100 µm. Throughout the injection procedure, the cortex was kept moist using cortex buffer. After all injections were completed, the cortex was covered with a 4 mm cranial window (glass coverslip), secured in place using super glue (Pattex Ultra Gel) around the rim of the craniotomy. After allowing the glue to dry, dental cement was used to cover any remaining bone and further secure the glass cover slip.

The mice were then administered 0.5–1 ml Sterofundin subcutaneously, and anesthesia was antagonized using Naloxone (1.2 mg/kg), Flumazenil (0.5 mg/kg), and Atipamezole (2.5 mg/kg) also administered subcutaneously. The animals were kept in a warm environment and observed for several hours before returning to their home cage. Wet food was placed into the cages, and 0.5 mg/kg of Carprofen was administered for the first 3 days after surgery. Fluorophores were allowed to express for 2–3 weeks before checking expression. If excessive bone regrowth prevented imaging, a second short surgery was performed to replace the cranial window and remove any bone patches.

After surgeries, the mice were placed in a 14/10 h light dark-cycle reversed room, so that the animals could be imaged during their dark cycle. Mice were co-housed with conspecifics (littermates whenever possible) in a large 1500 cm² cage containing nesting material, dark retreats and a running wheel. These conditions were upheld during orientation deprivation.

IOS imaging and Cylinder lens goggle mounting

2–4 weeks after the virus injection, IOS imaging was repeated (as described above) but without the use of ultrasound gel and using a drifting and inverting checkerboard bar (Fourier stimulus) to map out the retinotopic gradient across V1⁷⁴. This stimulus allowed for better outlining of the V1 border. In these experiments, mice were anaesthetized only lightly, with Fentanyl (0.035 mg/kg), Midazolam (3.5 mg/kg) and Medetomidine (0.35 mg/kg), injected intraperitonially. Anesthesia was maintained by injecting 25% of the original dose every hour. For animals that would later undergo orientation deprivation, after IOS imaging custom aluminum goggle frames (without cylinder lenses) were fitted to the animal’s head and adjusted using tongs. The space between the frames and the eyes was such that the animals could still clean their eyes with their forepaws. The goggle frames were then attached to the head bars with small screws.

Chronic in vivo two-photon calcium imaging in awake mice

Prior to awake head fixation, mice were handled for five days to accustom them to the experimenter, walking on a Styrofoam ball with a fixed axis and brief head restraint. For imaging, mice were head-fixed on an air-suspended Styrofoam ball, allowing them to run, under a two-photon imaging system (Thorlabs Bergamo II). The two-photon system was equipped with a pulsed femtosecond Ti:Sapphire laser (Spectra Physics MaiTai DeepSee laser; tuned to 940 nm for calcium imaging or 1050 nm for structural imaging), resonant and galvo scanning mirrors, a ×16 NA 0.8 immersion objective (Nikon) and a piezoelectric stepper for multiplane imaging. The photon collection pathway had a 720/25 nm short-pass filter followed by a dichroic beam-splitter (FF560) that allowed simultaneous detection of green and red light using two GaAsP photomultiplier tubes (PMTs; Hamamatsu) with either a 500–550 nm or a 572–642 nm bandpass filter. The system was controlled by ScanImage 4.2⁷⁵. Ultrasound gel (diluted 3:1 with water and centrifuged to remove air bubbles) was applied over the cranial window implant, and the ×16 objective was immersed in the gel. Strips of tape were used to shield the imaging window from external light.

During all imaging sessions, videos of both eyes were recorded (Imaging Source infra-red camera, 30 Hz), and running was tracked via an infra-red sensor attached to the air-suspended Styrofoam ball. The pupil diameters were extracted using DeepLabCut⁷⁶ and averaged across both eyes. The running/arousal modulation for each PO change (i.e., for each neuron across each pair of sessions) was defined as the correlation coefficient between the stimulus-response amplitude of the neuron and the running speed/pupil diameter of the mouse, multiplied by the average change in running speed/pupil diameter across sessions, i.e.:

$${running\; modulation}= {corr}\left({running\; speed},{stimulus\; response\; amplitude}\right) \\ * \overline{\Delta {running\; speed}}$$

(1)

$${arousal\; modulation}= {corr}\left({pupil\; diameter},{stimulus\; response\; amplitude}\right) \\ * \overline{\Delta {pupil\; diameter}}$$

(2)

Mice were presented with visual stimuli on a gamma-corrected LCD monitor (60 Hz, 2560 × 1440 pixels, 27 inches) 16 cm from the eyes. Light contamination during two-photon imaging was minimized by shuttering the LCD monitor⁷⁷. The screen was either placed directly in front of the mouse if imaging binocular V1 or offset when imaging monocular V1, and the screen tilt was aligned to the angle of the mouse’s head. The same screen position and screen angle were used for repeated sessions of each mouse. Visual stimuli were generated in MATLAB using the Psychophysics Toolbox^78,79. The visual stimulus consisted of a 25° radius (+12° fading-edge) sine wave grating drifting in 12 directions (0.04 cycles/° and 3 cycles/s) on a gray background with a stimulus duration of 5 s and an interstimulus interval of 6 s (gray screen). The mice were presented with these stimuli on two separate days before data acquisition began as part of habituation. On data acquisition days, the stimuli were presented 32 times. During these days, appropriate fields of view were identified and imaged during stimulus presentation allowing for re-finding on subsequent days. We re-found this imaging field of view on subsequent days by searching for the matching blood vessel pattern under widefield illumination. After switching to two-photon imaging, we used custom-written MATLAB code to overlay the live field of view over a template (average fluorescence image) acquired during the first imaging session. This alignment was significantly aided by the tdTomato structural marker. We then adjusted the live position in X, Y, and Z until the two images were perfectly aligned. During the experiment, if there was any slow drift in depth this was manually corrected by the experimenter. Again, this process was made significantly easier by the tdTomato structural marker.

Orientation deprivation

Mice that underwent orientation deprivation were habituated to the goggle frames for a minimum of four days before the baseline imaging sessions. The goggle frames were removed when presenting visual stimuli and recording neuronal calcium signals and were again fixed to the animal’s head bar after the end of the imaging session. When imaging was followed by orientation deprivation, the cylinder lenses were pressed into the goggles (sometimes secured with super glue) before attaching them to the animal’s head bar again. Mice were subsequently returned to their home cage (which had stripes of black tape oriented along the experienced orientation through the cylinder lenses). Alternatively, mice were housed in a cage with monitors around two sides of the cage that presented slowly drifting full field gratings moving in eight directions during the light cycle. The experienced orientation through the cylinder lenses was between −22° to −45°.

Processing of in vivo two-photon calcium imaging data and calculation of POs

Fluorescence traces from in vivo data were extracted either using custom MATLAB software or Suite2P⁸⁰. Both methods produced average fluorescence FOV images, single neuron ROI average fluorescence over time, with ROIs independently calculated or drawn for each day. To match ROIs from different sessions, we used a custom-written MATLAB program, which registered the templates from different sessions using affine transformation. If ROIs had more than 50% overlap in their ROI masks, they were defined as putatively matched. We then inspected the ROIs of each neuron manually. If any of the ROIs of a neuron were not in the same location relative to local landmarks (other neurons, blood vessels, axons, dendrites) or the neuron was no longer clearly visible in one of the imaging sessions, the neuron was excluded from the dataset. Only neurons that could be reidentified on every imaging session were included in further analysis. We verified this approach by employing an additional quantitative method. To this end, we calculated pixel-wise correlations of ROIs for a subset of our data. We plotted the distribution across all comparisons as-well-as the minimum correlation for each neuron across all time points (Supplementary Fig. 2a). Excluding neurons with a minimum ROI correlation below 0.1, 0.25 or 0.5 did not substantially change the overall drift magnitude (Supplementary Fig. 2b–e).

The calcium activity of a neuron was defined to be ∆F/F and was calculated as:

$$\Delta {{\rm{F}}}/{{\rm{F}}}=\frac{({{\rm{F}}}({{\rm{t}}})-0.7*({{{\rm{F}}}}_{{{\rm{neuropil}}}}({{\rm{t}}})-{\widetilde{{{\rm{F}}}}}_{{{\rm{neuropil}}}}))-{\widetilde{{{\rm{F}}}}}_{{{\rm{baseline}}}}}{{\widetilde{{{\rm{F}}}}}_{{{\rm{baseline}}}}}$$

(3)

Where F(t) is the green fluorescence trace of a neuron, ${{{\rm{F}}}}_{{{\rm{neuropil}}}}\left({{\rm{t}}}\right)$ is the neuropil green fluorescence trace, ${\widetilde{{{\rm{F}}}}}_{{{\rm{neuropil}}}}$ is the median of the neuropil fluorescence trace and ${\widetilde{{{\rm{F}}}}}_{{{\rm{baseline}}}}$ is the median of all 1 s pre-stimulus periods of the ROI fluorescence²⁴.

The stimulus tuning curve of a neuron was defined as the mean ∆F/F during the stimulus presentation across all trials for a given grating direction (resulting in a vector length 12). To calculate the pairwise signal correlation (PSC) matrix, the Pearson’s correlation between the stimulus tuning curves of all pairs of neurons were calculated (resulting in a square matrix size neurons x neurons). The pairwise stimulus correlation matrix of each day was vectorized and correlated with the vectorized PSC matrix of every other day (resulting in a square matrix size days x days).

For further analysis, the PO was calculated as follows. First, neurons were tested for visual responsiveness. For each stimulus, a rank sum test between the stimulus period mean ∆F/F and the pre-stimulus period mean ∆F/F was applied (with Bonferroni correction significance threshold of p < 0.05). If a neuron’s ∆F/F was significantly above the ∆F/F during the pre-stimulus period for any stimulus, it was considered visually responsive. Second, the PO of each neuron was calculated using the vector sum across all responses:

$${{\rm{X}}}=\frac{1}{{{{\rm{n}}}}_{{{\rm{rep}}}}*{{{\rm{n}}}}_{{{\rm{stim}}}}}{\sum }_{{{\rm{rep}}}=1}^{{{{\rm{n}}}}_{{{\rm{rep}}}}}{\sum }_{{{\rm{stim}}}=1}^{{{{\rm{n}}}}_{{{\rm{stim}}}}}{{{\rm{R}}}}_{({{\rm{stim}}},{{\rm{rep}}})}\cos ({2 \, {{\rm{\theta }}}}_{({{\rm{stim}}},{{\rm{rep}}})})$$

(4)

$${{\rm{Y}}}=\frac{1}{{{{\rm{n}}}}_{{{\rm{rep}}}}*{{{\rm{n}}}}_{{{\rm{stim}}}}}{\sum }_{{{\rm{rep}}}=1}^{{{{\rm{n}}}}_{{{\rm{rep}}}}}{\sum }_{{{\rm{stim}}}=1}^{{{{\rm{n}}}}_{{{\rm{stim}}}}}{{{\rm{R}}}}_{({{\rm{stim}}},{{\rm{rep}}})}\sin ({2\, {{\rm{\theta }}}}_{({{\rm{stim}}},{{\rm{rep}}})})$$

(5)

$${{{\rm{\theta }}}}_{{{\rm{PO}}}}=\frac{1}{2}\arctan \left(\frac{{{\rm{Y}}}}{{{\rm{X}}}}\right)$$

(6)

Where ${{{\rm{R}}}}_{({{\rm{stim}}},{{\rm{rep}}})}$ is the stimulus period mean ∆F/F per stimulus per trial, ${{{\rm{\theta }}}}_{({{\rm{stim}}},{{\rm{rep}}})}$ are the corresponding visual stimulus angles, ${{{\rm{n}}}}_{{{\rm{stim}}}}$ is the number of different stimulus angles and ${{{\rm{n}}}}_{{{\rm{rep}}}}$ is the number of stimulus repetitions (usually 32). Bootstrap resampling across all stimulus repetitions was applied in order to obtain a distribution of ${{{\rm{\theta }}}}_{{{\rm{PO}}}}$. PO is defined as the $\bar{{{{\rm{\theta }}}}_{{{\rm{PO}}}}}$ across all bootstrap samples. The 95% confidence interval (upper PO_CI and lower PO_CI) was also calculated from the distribution of ${{{\rm{\theta }}}}_{{{\rm{PO}}}}$. If the difference between the upper PO_CI and lower PO_CI was greater than 90°, then the neuron was considered not orientation-tuned. A neuron was considered concurrently tuned across a time interval, if it was both visually responsive and tuned on both days. For these neurons the $\Delta {{\rm{PO}}}$ was calculated. In order to determine if the PO of a cell was significantly different on two separate days, the PO of the second day had to be outside the PO 95% confidence interval of the first day, and the PO of the first day had to be outside the PO 95% confidence interval of the second day.

The experienced orientation during orientation deprivation was calculated by taking an image of the mouse, wearing the goggle frames, from the center of the stimulus presentation screen. The angle of the goggles in the image was measured using ImageJ. This angle was adjusted based on the angle of the image presentation screen to give the experienced orientation relative to the mouse. To calculate PO convergence, the change in a neuron’s PO relative to the experienced orientation (i.e., if it moved towards or away from the experienced orientation, ${experienced}$), was calculated as follows:

$${{\rm{for}}}\; {{\rm{mice}}}\; {{\rm{without}}}\; {{\rm{goggles}}}\!:\Delta \left|{{\rm{rPO}}}\right|={|{{\rm{PO}}}}_{{{\rm{t}}}1}-\overline{{{\rm{experienced}}}}{{|}}\\ -{|{{\rm{PO}}}}_{{{\rm{t}}}2}-\overline{{{\rm{experienced}}}}{{|}}$$

(7)

$$ {{\rm{for}}}\; {{\rm{orientation}}}\; {{\rm{deprivation}}}\; {{\rm{data}}}\!:\Delta \left|{{\rm{rPO}}}\right| \\ ={|{{\rm{PO}}}}_{{pre}}-{{{\rm{experienced}}}}_{{{\rm{mouse}}}}{{|}}-{|{{\rm{PO}}}}_{{{\rm{post}}}}-{{{\rm{experienced}}}}_{{{\rm{mouse}}}}{{|}}$$

(8)

Where $\overline{{{\rm{experienced}}}}$ is the mean experienced orientation through the cylinder lens goggles of all orientation deprived mice, and ${{{\rm{experienced}}}}_{{{\rm{mouse}}}}$ is the experienced orientation for the mouse from which the neuron came.

Statistics

When normality could not be ruled out using the Kolmogorov–Smirnov Goodness-of-Fit test, two-tailed two-sample unequal-variance, Pearson’s correlation, one-way ANOVA, paired or unpaired T tests were used. When normality was ruled out or could not be assumed, Spearman’s correlation, the Wilcoxon signed-rank test, the Mann–Whitney U test, or the Kruskal–Wallis test were used. In the case of periodic variables, a von Mises distribution was assumed, and circular statistics were used (circ_stats toolbox⁸¹). If ANOVA or Kruskal–Wallis tests were significant, we used a Dunnett’s posthoc test when data were parametric, or Mann–Whitney U tests with Bonferroni corrected alpha thresholds when data were not parametric. Animal number and cell or PO change numbers are indicated by capital and lower case “n”, respectively, in the figure legends. Asterisks indicate p values under significance threshold α = 0.05 (*). Alpha values were adjusted using the Bonferroni method for multiple comparison correction where necessary.

Network model

We considered a feedforward network in which $N$ = 500 presynaptic excitatory neurons with firing rates ${{\bf{u}}}$ drive $N$ postsynaptic excitatory neurons with firing rates ${{\bf{v}}}$, via a matrix of synaptic connections ${{\bf{W}}}$ and a linear response function such that ${{\bf{v}}}={{{\bf{W}}}}^{T}{{\bf{u}}}$. Presynaptic neurons were assumed to be tuned to a given orientation from 0° to 180°, and their firing rates were modeled with Gaussian tuning curves centered at the orientation to which they were tuned. In the model, we assumed that the drift in orientation tuning seen in our data can be mainly explained by changes to the feedforward connectivity between excitatory neurons in two cortical layers of the primary visual cortex.

Learning rule

We found that the representational drift observed in our data, both under baseline and deprivation conditions, can be captured by three main assumptions:

1.
Synaptic changes consist of a combination of both activity-dependent synaptic plasticity in response to external stimuli, represented by a Hebbian component $H$, and activity-independent fluctuations in synaptic strength (synaptic volatility), represented by a random component $\xi$.
2.
Synaptic changes are scaled by a weight-dependent propensity function, $\rho \left(w\right)$.
3.
Synaptic weights are normalized by homeostatic mechanisms acting on a slower timescale than the timescale of stimulus-driven Hebbian plasticity.

We combined these assumptions into a learning rule which describes the change in the synaptic strength ${w}_{{ij}}$ connecting pre-synaptic neuron ${u}_{j}$ to post-synaptic neuron ${v}_{i}$ as a weighted sum of Hebbian $H$ and random $\xi$ changes, scaled by a weight-dependent propensity function $\rho \left(w\right)$:

$$\Delta {w}_{{ij}}=\epsilon \cdot \rho \big({w}_{{ij}}\big)\cdot \left[k{H}_{{ij}}+\xi \right]$$

(9)

The learning rate $\epsilon$ (1 × 10⁻⁴) scales the overall magnitude of the weight update, and the scaling factor $k$ determines the relative impact of the Hebbian and random contribution to the total weight changes in the model. We modeled the propensity function to be weight-dependent $\rho \left(w\right)=\tanh \left(10w\right)$ based on experimental observations^{47,50,51,52,53}. The Hebbian contribution ${H}_{{ij}}$ for each weight ${w}_{{ij}}$ was given by the product of presynaptic and postsynaptic activities in response to each stimulus presentation:

$${H}_{{ij}}={u}_{j}{v}_{i}$$

(10)

The random component $\xi$ represents activity-independent synaptic strength fluctuations, and was sampled independently for each weight ${w}_{{ij}}$ from a standard normal distribution at each weight update:

$$\xi \sim N(0,1)$$

(11)

The weights were normalized on a timescale slower than Hebbian activity-dependent plasticity, motivated by experimental findings of slower homeostatic mechanisms such as synaptic scaling^54,55,56. Following the presentation of ${N}_{\theta }$ stimuli, we implemented a divisive normalization which preserves the total sum of incoming weights ${w}_{{ij}}$ onto each postsynaptic neuron $i$:

$${w}_{{ij}}\leftarrow \frac{{w}_{{ij}}}{{\sum }_{j}{w}_{{ij}}}$$

(12)

Given that homeostatic plasticity is typically reported to occur on a timescale of hours to days⁵⁵, we assumed that this normalization occurs once per day. Assuming each pyramidal neuron in the mouse visual cortex experiences approximately one orientation stimulus per second, we took ${N}_{\theta }$ to be (60 seconds) × (60 minutes) × (12 waking hours) = 43,200 stimuli per day. We then scaled the overall learning rate $\epsilon$ to match the mean drift magnitude to that observed over the experimental time frame.

Weight initialization

The population of $N$ orientation-selective presynaptic neurons had Gaussian-shaped tuning curves uniformly distributed across the space of oriented stimuli from −90° to 90°. The input weights to the $N$ postsynaptic neurons were initialized as Gaussians across the space of orientations, such that each postsynaptic neuron inherited an initial tuning curve from its presynaptic partners (Supplementary Fig. 5a). The drift rate of a model neuron was correlated with the width of its initial tuning curve (Supplementary Fig. 5d). In the experimental data, the drift magnitudes were approximately log-normally distributed across neurons. To match this variability seen in drift magnitude, we, therefore, initialized the postsynaptic tuning curves with widths sampled from a log-normal distribution (Supplementary Fig. 5a, b).

PO and drift metrics

The PO of neurons in the network model was calculated as follows: a series of orientation stimuli from −90° to 90° were presented to the network, and the PO of each neuron was defined as the orientation that elicits the highest firing rate in the postsynaptic neurons. Weights were frozen during this test protocol and were not subject to synaptic plasticity. Drift magnitude and convergence were defined as in the experimental analysis, and drift rate was defined as the absolute circular distance between POs on consecutive days.

Visual experience conditions

To simulate the experience of a mouse during baseline visual conditions, we repeatedly stimulated the network with stimuli drawn uniformly between −90° to 90°. To mimic orientation deprivation conditions, we stimulated the network instead with a constant experienced orientation $\hat{\theta }$ (value selected randomly). For the deprivation condition, as in the experiment, the network was either stimulated constantly with the same orientation for 28 days, or interrupted every 7 days by a test protocol in which a repeated series of stimuli across the entire range of −90° to 90° were randomly presented. Weights were plastic and continuously updated by the exact same learning rule during both baseline and deprivation conditions, unless otherwise stated (Fig. 3h, j, k; Supplementary Fig. 6a, b). For all experiments, a short warm-up period (2–3 days) of baseline conditions was run to allow the weights to settle to a constant distribution before measurement began.

Ratio of Hebbian plasticity to synaptic volatility contribution

Network parameters (amplitude and offset of presynaptic firing rates, overall weight scaling) were set such that when $k=1$, the mean values of ${H}_{{ij}}$ and $\xi$ across neurons were equivalent under baseline conditions. To modulate the relative strength of the Hebbian component with respect to synaptic volatility (Fig. 3h), we therefore scaled $k$ from 0 to 1. To determine the individual contributions of the Hebbian and random component to the drift (Fig. 3j, k), we removed each component respectively.

Recovery from deprivation

To test the recovery of orientation tuning in the network after deprivation (Fig. 3l), we simulated ${N}_{d}$ days of deprivation conditions, followed by ${N}_{b}$ days of baseline conditions, maintaining the same parameters throughout. The total directional PO changes undergone during the deprivation period were then compared to those during the following baseline period.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The processed data are available at https://gin.g-node.org/Joel-Bauer/orientation_tuning_drift. Source data are provided with this paper.

Code availability

The code is available at https://github.com/Joel-Bauer/orientation_tuning_drift and https://github.com/betsyherbert/bauer-lewin-drift-2024/tree/main.

References

Kentros, C. G., Agnihotri, N. T., Streater, S., Hawkins, R. D. & Kandel, E. R. Increased attention to spatial context increases both place field stability and spatial memory. Neuron 42, 283–295 (2004).
Article CAS PubMed Google Scholar
Ziv, Y. et al. Long-term dynamics of CA1 hippocampal place codes. Nat. Neurosci. 16, 264–266 (2013).
Article CAS PubMed PubMed Central Google Scholar
Montijn, J. S., Meijer, G. T., Lansink, C. S. & Pennartz, C. M. A. Population-level neural codes are robust to single-neuron variability from a multidimensional coding perspective. Cell Rep. 16, 2486–2498 (2016).
Article CAS PubMed Google Scholar
Driscoll, L. N., Pettit, N. L., Minderer, M., Chettih, S. N. & Harvey, C. D. Dynamic reorganization of neuronal activity patterns in parietal cortex. Cell 170, 986–999 (2017).
Article CAS PubMed PubMed Central Google Scholar
Schoonover, C. E., Ohashi, S. N., Axel, R. & Fink, A. J. P. Representational drift in primary olfactory cortex. Nature 594, 541–546 (2021).
Article ADS CAS PubMed Google Scholar
Aschauer, D. F. et al. Learning-induced biases in the ongoing dynamics of sensory representations predict stimulus generalization. Cell Rep. 38, 110340 (2022).
Article CAS PubMed Google Scholar
Wang, H. C., LeMessurier, A. M. & Feldman, D. E. Tuning instability of non-columnar neurons in the salt-and-pepper whisker map in somatosensory cortex. Nat. Commun. 13, 6611 (2022).
Article ADS CAS PubMed PubMed Central Google Scholar
Clopath, C., Bonhoeffer, T., Hübener, M. & Rose, T. Variance and invariance of neuronal long-term representations. Philos. Trans. R. Soc. B 372, 20160161 (2017).
Article Google Scholar
Rule, M. E., O’Leary, T. & Harvey, C. D. Causes and consequences of representational drift. Curr. Opin. Neurobiol. 58, 141–147 (2019).
Article CAS PubMed PubMed Central Google Scholar
Marks, T. D. & Goard, M. J. Stimulus-dependent representational drift in primary visual cortex. Nat. Commun. 12, 5169 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Nagaoka, A. et al. Abnormal intrinsic dynamics of dendritic spines in a fragile X syndrome mouse model in vivo. Sci. Rep. 6, 26651 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar
Mongillo, G., Rumpel, S. & Loewenstein, Y. Intrinsic volatility of synaptic connections—a challenge to the synaptic trace theory of memory. Curr. Opin. Neurobiol. 46, 7–13 (2017).
Article CAS PubMed Google Scholar
Hazan, L. & Ziv, N. E. Activity dependent and independent determinants of synaptic size diversity. J. Neurosci. 40, 2828–2848 (2020).
Article CAS PubMed PubMed Central Google Scholar
Hofer, S. B., Mrsic-Flogel, T. D., Bonhoeffer, T. & Hübener, M. Experience leaves a lasting structural trace in cortical circuits. Nature 457, 313–317 (2009).
Article ADS CAS PubMed Google Scholar
Keck, T. et al. Massive restructuring of neuronal circuits during functional reorganization of adult visual cortex. Nat. Neurosci. 11, 1162–1167 (2008).
Article CAS PubMed Google Scholar
Rule, M. E. & O’Leary, T. Self-healing codes: How stable neural populations can track continually reconfiguring neural representations. PNAS 119, e2106692119 (2022).
Article CAS PubMed PubMed Central Google Scholar
Khatib, D. et al. Active experience, not time, determines within-day representational drift in dorsal CA1. Neuron 111, 2348–2356 (2023).
Article CAS PubMed Google Scholar
Geva, N., Deitch, D., Rubin, A. & Ziv, Y. Time and experience differentially affect distinct aspects of hippocampal representational drift. Neuron 111, 2357–2366 (2023).
Article CAS PubMed Google Scholar
Chapman, B., Stryker, M. P. & Bonhoeffer, T. Development of orientation preference maps in ferret primary visual cortex. J. Neurosci. 16, 6443–6453 (1996).
Article CAS PubMed PubMed Central Google Scholar
Sengpiel, F., Stawinski, P. & Bonhoeffer, T. Influence of experience on orientation maps in cat visual cortex. Nat. Neurosci. 2, 727–732 (1999).
Article CAS PubMed Google Scholar
Shtoyerman, E., Arieli, A., Slovin, H., Vanzetta, I. & Grinvald, A. Long-term optical imaging and spectroscopy reveal mechanisms underlying the intrinsic signal and stability of cortical maps in V1 of behaving monkeys. J. Neurosci. 20, 8111–8121 (2000).
Article CAS PubMed PubMed Central Google Scholar
Kind, P. C. et al. Correlated binocular activity guides recovery from monocular deprivation. Nature 416, 430–433 (2002).
Article ADS CAS PubMed Google Scholar
Mank, M. et al. A genetically encoded calcium indicator for chronic in vivo two-photon imaging. Nat. Methods 5, 805–811 (2008).
Article CAS PubMed Google Scholar
Chen, T.-W. et al. Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300 (2013).
Article ADS CAS PubMed PubMed Central Google Scholar
Rose, T., Jaepel, J., Hübener, M. & Bonhoeffer, T. Cell-specific restoration of stimulus preference after monocular deprivation in the visual cortex. Science 352, 1319–1322 (2016).
Article ADS CAS PubMed Google Scholar
Jeon, B. B., Swain, A. D., Good, J. T., Chase, S. M. & Kuhlman, S. J. Feature selectivity is stable in primary visual cortex across a range of spatial frequencies. Sci. Rep. 8, 15288 (2018).
Article ADS PubMed PubMed Central Google Scholar
Deitch, D., Rubin, A. & Ziv, Y. Representational drift in the mouse visual cortex. Curr. Biol. 31, 4327–4339 (2021).
Article CAS PubMed Google Scholar
Grutzendler, J., Kasthuri, N. & Gan, W.-B. Long-term dendritic spine stability in the adult cortex. Nature 420, 812–816 (2002).
Article ADS CAS PubMed Google Scholar
Holtmaat, A. J. G. D. et al. Transient and persistent dendritic spines in the neocortex in vivo. Neuron 45, 279–291 (2005).
Article CAS PubMed Google Scholar
Majewska, A. K. Remodeling of synaptic structure in sensory cortical areas in vivo. J. Neurosci. 26, 3021–3029 (2006).
Article CAS PubMed PubMed Central Google Scholar
Ohki, K., Chung, S., Ch’ng, Y. H., Kara, P. & Reid, R. C. Functional imaging with cellular resolution reveals precise micro-architecture in visual cortex. Nature 433, 597–603 (2005).
Article ADS CAS PubMed Google Scholar
Hirsch, H. V. B. & Spinelli, D. N. Visual experience modifies distribution of horizontally and vertically oriented receptive fields in cats. Science 168, 869–871 (1970).
Article ADS CAS PubMed Google Scholar
Blakemore, C. & Cooper, G. F. Development of the brain depends on the visual environment. Nature 228, 477–478 (1970).
Article ADS CAS PubMed Google Scholar
O’Hashi, K., Miyashita, M. & Tanaka, S. Experience-dependent orientation plasticity in the visual cortex of rats chronically exposed to a single orientation. Neurosci. Res 58, 86–90 (2007).
Article PubMed Google Scholar
Kreile, A. K., Bonhoeffer, T. & Hübener, M. Altered visual experience induces instructive changes of orientation preference in mouse visual cortex. J. Neurosci. 31, 13911–13920 (2011).
Article CAS PubMed PubMed Central Google Scholar
Yoshida, T., Ozawa, K. & Tanaka, S. Sensitivity profile for orientation selectivity in the visual cortex of goggle-reared Mice. PLoS One 7, e40630 (2012).
Article ADS CAS PubMed PubMed Central Google Scholar
Montgomery, D. P., Hayden, D. J., Chaloner, F. A., Cooke, S. F. & Bear, M. F. Stimulus-selective response plasticity in primary visual cortex: progress and puzzles. Front. Neural Circuits 15, 815554 (2022).
Article PubMed PubMed Central Google Scholar
Sadeh, S. & Clopath, C. Contribution of behavioural variability to representational drift. Elife 11, e77907 (2022).
Article CAS PubMed PubMed Central Google Scholar
Jensen, K. T., Harpaz, N. K., Dhawale, A. K., Wolff, S. B. E. & Ölveczky, B. P. Long-term stability of single neuron activity in the motor system. Nat. Neurosci. 25, 1664–1674 (2022).
Article CAS PubMed PubMed Central Google Scholar
Niell, C. M. & Stryker, M. P. Modulation of visual responses by behavioral state in mouse visual cortex. Neuron 65, 472–479 (2010).
Article CAS PubMed PubMed Central Google Scholar
Keller, G. B., Bonhoeffer, T. & Hübener, M. Sensorimotor mismatch signals in primary visual cortex of the behaving mouse. Neuron 74, 809–815 (2012).
Article CAS PubMed Google Scholar
Andermann, M. L., Kerlin, A. M., Roumis, D. K., Glickfeld, L. L. & Reid, R. C. Functional specialization of mouse higher visual cortical areas. Neuron 72, 1025–1039 (2011).
Article CAS PubMed Google Scholar
Ayaz, A., Saleem, A. B., Schölvinck, M. L. & Carandini, M. Locomotion controls spatial integration in mouse visual cortex. Curr. Biol. 23, 890–894 (2013).
Article CAS PubMed PubMed Central Google Scholar
Reimer, J. et al. Pupil fluctuations track fast switching of cortical states during quiet wakefulness. Neuron 84, 355–362 (2014).
Article CAS PubMed PubMed Central Google Scholar
Mineault, P. J., Tring, E., Trachtenberg, J. T. & Ringach, D. L. Enhanced spatial resolution during locomotion and heightened attention in mouse primary visual cortex. J. Neurosci. 36, 6382–6392 (2016).
Article CAS PubMed PubMed Central Google Scholar
Matsuzaki, M., Honkura, N., Ellis-Davies, G. C. R. & Kasai, H. Structural basis of long-term potentiation in single dendritic spines. Nature 429, 761–766 (2004).
Article ADS CAS PubMed PubMed Central Google Scholar
Yasumatsu, N., Matsuzaki, M., Miyazaki, T., Noguchi, J. & Kasai, H. Principles of long-term dynamics of dendritic spines. J. Neurosci. 28, 13592–13608 (2008).
Article CAS PubMed PubMed Central Google Scholar
Minerbi, A. et al. Long-term relationships between synaptic tenacity, synaptic remodeling, and network activity. PLoS Biol. 7, e1000136 (2009).
Article PubMed PubMed Central Google Scholar
Ziv, N. E. & Brenner, N. Synaptic tenacity or lack thereof: spontaneous remodeling of synapses. Trends Neurosci. 41, 89–99 (2018).
Article CAS PubMed Google Scholar
Loewenstein, Y., Kuras, A. & Rumpel, S. Multiplicative dynamics underlie the emergence of the log-normal distribution of spine sizes in the neocortex in vivo. J. Neurosci. 31, 9481–9488 (2011).
Article CAS PubMed PubMed Central Google Scholar
Berry, K. P. & Nedivi, E. Spine dynamics: are they all the same? Neuron 96, 43–55 (2017).
Article CAS PubMed PubMed Central Google Scholar
Humble, J., Hiratsuka, K., Kasai, H. & Toyoizumi, T. Intrinsic spine dynamics are critical for recurrent network learning in models with and without autism spectrum disorder. Front. Comput. Neurosci. 13, 38 (2019).
Article PubMed PubMed Central Google Scholar
Statman, A., Kaufman, M., Minerbi, A., Ziv, N. E. & Brenner, N. Synaptic size dynamics as an effectively stochastic process. PLoS Comput. Biol. 10, e1003846 (2014).
Article ADS PubMed PubMed Central Google Scholar
Turrigiano, G. G. The self-tuning neuron: synaptic scaling of excitatory synapses. Cell 135, 422–435 (2008).
Article CAS PubMed PubMed Central Google Scholar
Zenke, F. & Gerstner, W. Hebbian plasticity requires compensatory processes on multiple timescales. Philos. Trans. R. Soc. B: Biol. Sci. 372, 20160259 (2017).
Article Google Scholar
Keck, T. et al. Integrating Hebbian and homeostatic plasticity: the current state of the field and future research directions. Philos. Trans. R. Soc. B: Biol. Sci. 372, 20160158 (2017).
Article Google Scholar
Runge, K., Cardoso, C. & de Chevigny, A. Dendritic spine plasticity: function and mechanisms. Front. Synaptic Neurosci. 12, 36 (2020).
Maletic-Savatic, M., Malinow, R. & Svoboda, K. Rapid dendritic morphogenesis in CA1 hippocampal dendrites induced by synaptic activity. Science 283, 1923–1927 (1999).
Article ADS CAS PubMed Google Scholar
Engert, F. & Bonhoeffer, T. Dendritic spine changes associated with hippocampal long-term synaptic plasticity. Nature 399, 66–70 (1999).
Article ADS CAS PubMed Google Scholar
Yuste, R. & Bonhoeffer, T. Morphological changes in dendritic spines associated with long-term synaptic plasticity. Annu. Rev. Neurosci. 24, 1071–1089 (2001).
Article CAS PubMed Google Scholar
Quinn, D. P. et al. The stability of glutamatergic synapses is independent of activity level, but predicted by synapse size. Front. Cell Neurosci. 13, 291 (2019).
Article CAS PubMed PubMed Central Google Scholar
Rule, M. E. et al. Stable task information from an unstable neural population. Elife 9, e51121 (2020).
Article CAS PubMed PubMed Central Google Scholar
Kossio, Y. F. K., Goedeke, S., Klos, C. & Memmesheimer, R.-M. Drifting assemblies for persistent memory: neuron transitions and unsupervised compensation. PNAS 118, e2023832118 (2021).
Article CAS PubMed Google Scholar
Bauer, J. et al. Limited functional convergence of eye-specific inputs in the retinogeniculate pathway of the mouse. Neuron 109, 2457–2468 (2021).
Article CAS PubMed Google Scholar
Lee, W.-C. A. et al. Dynamic remodeling of dendritic arbors in GABAergic interneurons of adult visual cortex. PLoS Biol. 4, e29 (2006).
Article PubMed Google Scholar
De Paola, V. et al. Cell type-specific structural plasticity of axonal branches and boutons in the adult neocortex. Neuron 49, 861–875 (2006).
Article PubMed Google Scholar
Stettler, D. D., Yamahachi, H., Li, W., Denk, W. & Gilbert, C. D. Axons and synaptic boutons are highly dynamic in adult visual cortex. Neuron 49, 877–887 (2006).
Article CAS PubMed Google Scholar
Bonhoeffer, T. & Grinvald, A. Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns. Nature 353, 429–431 (1991).
Article ADS CAS PubMed Google Scholar
Jeon, B. B., Fuchs, T., Chase, S. M. & Kuhlman, S. J. Visual experience has opposing influences on the quality of stimulus representation in adult primary visual cortex. Elife 11, e80361 (2022).
Article CAS PubMed PubMed Central Google Scholar
Mower, G. D., Berry, D., Burchfiel, J. L. & Duffy, F. H. Comparison of the effects of dark rearing and binocular suture on development and plasticity of cat visual cortex. Brain Res. 220, 255–267 (1981).
Article CAS PubMed Google Scholar
Chaudhari, P. & Soatto, S. Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks. 2018. Inf. Theory Appl. Workshop 00, 1–10 (2018).
Google Scholar
Micou, C. & O’Leary, T. Representational drift as a window into neural and behavioural plasticity. Curr. Opin. Neurobiol. 81, 102746 (2023).
Article CAS PubMed Google Scholar
Bauer, J. & Rose, T. Mouse vision: variability and stability across the visual processing hierarchy. Curr. Biol. 31, R1129–R1132 (2021).
Article CAS PubMed Google Scholar
Kalatsky, V. A. & Stryker, M. P. New paradigm for optical imaging temporally encoded maps of intrinsic signal. Neuron 38, 529–545 (2003).
Article CAS PubMed Google Scholar
Pologruto, T. A., Sabatini, B. L. & Svoboda, K. ScanImage: flexible software for operating laser scanning microscopes. Biomed. Eng. Online 2, 13 (2003).
Article PubMed PubMed Central Google Scholar
Mathis, A. et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21, 1281–1289 (2018).
Article CAS PubMed Google Scholar
Leinweber, M. et al. Two-photon calcium imaging in mice navigating a virtual reality environment. JoVE 84, e50885 (2014).
Google Scholar
Pelli, D. G. The VideoToolbox software for visual psychophysics: transforming numbers into movies. Spat. Vis. 10, 437–442 (1997).
Article CAS PubMed Google Scholar
Brainard, D. H. The psychophysics toolbox. Spat. Vis. 10, 433–436 (1997).
Article CAS PubMed Google Scholar
Pachitariu, M. et al. Suite2p: beyond 10,000 neurons with standard two-photon microscopy. bioRxiv https://www.biorxiv.org/content/10.1101/061507v2 (2017).
Berens, P. CircStat: a MATLAB toolbox for circular statistics. J. Stat. Softw. 31, 1–21 (2009).
Article Google Scholar

Download references

Acknowledgements

We thank Sandra Reinert for discussions and comments on the manuscript, Pieter Goltstein for help with the two-photon microscope and programming, Volker Staiger, Claudia Huber, Frank Voss, Dominik Lindner, and Max Sperling for technical assistance. We also thank Sasha Hay, Ezgi Kaya and Meike Hack for help with image annotation. This project was funded by the Max Planck Society, the German Research Foundation (Grant SFB870 A07, project number 1188 03580, to T.R., M.H., and T.B.), and the European Research Council under the European Union’s Horizon 2020 research and innovation program (grant agreement no. 804824 to J.G.).

Funding

Open Access funding enabled and organized by Projekt DEAL.

Author information

Joel Bauer
Present address: Sainsbury Wellcome Centre for Neural Circuits and Behaviour, University College London, London, UK
Carl E. Schoonover
Present address: Allen Institute for Neural Dynamics, Seattle, WA, USA
Andrew J. P. Fink
Present address: Department of Neurobiology, Northwestern University, Evanston, IL, USA
Tobias Rose
Present address: Institute for Experimental Epileptology and Cognition Research, University of Bonn, Medical Center, Bonn, Germany
These authors contributed equally: Joel Bauer, Uwe Lewin.

Authors and Affiliations

Max Planck Institute for Biological Intelligence, Martinsried, Germany
Joel Bauer, Uwe Lewin, Tobias Rose, Tobias Bonhoeffer & Mark Hübener
International Max Planck Research School for Molecular Life Sciences, Martinsried, Germany
Joel Bauer
Graduate School of Systemic Neurosciences, Ludwig-Maximilians-Universität München, Planegg, Germany
Uwe Lewin
School of Life Sciences, Technical University of Munich, Freising, Germany
Elizabeth Herbert & Julijana Gjorgjieva
Mortimer B. Zuckerman Mind Brain Behavior Institute, Department of Neuroscience, Columbia University, New York, NY, USA
Carl E. Schoonover & Andrew J. P. Fink

Authors

Joel Bauer
View author publications
Search author on:PubMed Google Scholar
Uwe Lewin
View author publications
Search author on:PubMed Google Scholar
Elizabeth Herbert
View author publications
Search author on:PubMed Google Scholar
Julijana Gjorgjieva
View author publications
Search author on:PubMed Google Scholar
Carl E. Schoonover
View author publications
Search author on:PubMed Google Scholar
Andrew J. P. Fink
View author publications
Search author on:PubMed Google Scholar
Tobias Rose
View author publications
Search author on:PubMed Google Scholar
Tobias Bonhoeffer
View author publications
Search author on:PubMed Google Scholar
Mark Hübener
View author publications
Search author on:PubMed Google Scholar

Contributions

All authors contributed to the conceptualization of the study. J.B. and U.L. conducted the experiments and analyzed the data. E.H. implemented the model with input from J.G., J.B., and U.L. All authors discussed the results and wrote the manuscript.

Corresponding authors

Correspondence to Joel Bauer or Mark Hübener.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks Frank Sengpiel, and the other, anonymous, reviewer for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Peer Review File

Reporting Summary

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Bauer, J., Lewin, U., Herbert, E. et al. Sensory experience steers representational drift in mouse visual cortex. Nat Commun 15, 9153 (2024). https://doi.org/10.1038/s41467-024-53326-x

Download citation

Received: 20 December 2023
Accepted: 08 October 2024
Published: 23 October 2024
DOI: https://doi.org/10.1038/s41467-024-53326-x

This article is cited by

Temporal coding carries more stable cortical visual representations than firing rate over time
- Hanlin Zhu
- Fei He
- Chong Xie
Nature Communications (2025)