Cerebral Cortex, August 2020;30: 4424–4437 doi: 10.1093/cercor/bhaa030 Advance Access Publication Date: 7 March 2020 Original Article ORIGINAL ARTICLE Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Anna Powell1 , William M. Connelly2 , Asta Vasalauskaite3 , Andrew J. D. Nelson1 , Seralynne D. Vann1 , John P. Aggleton1 , Frank Sengpiel3,4 and Adam Ranson4,5,6 1 School of Psychology, Cardiff University, CF10 3AS Cardiff, UK, 2 School of Medicine, University of Tasmania, Hobart, TAS, Australia, 3 School of Biosciences, Cardiff University, Cardiff, CF10 3AX, UK, 4 Neuroscience and Mental Health Research Institute, Cardiff University, Cardiff, CF24 4HQ, UK, 5 Faculty of Medicine and Health Sciences, Department of Basic Sciences, Universitat Internacional de Catalunya, Barcelona, 08195, Spain and 6 Institut de Neurociènces, Universitat Autònoma de Barcelona, Bellaterra, 08193, Spain Address correspondence to Adam Ranson. Email: adam@ransonlab.net Abstract The rodent retrosplenial cortex (RSC) functions as an integrative hub for sensory and motor signals, serving roles in both navigation and memory. While RSC is reciprocally connected with the sensory cortex, the form in which sensory information is represented in the RSC and how it interacts with motor feedback is unclear and likely to be critical to computations involved in navigation such as path integration. Here, we used 2-photon cellular imaging of neural activity of putative excitatory (CaMKII expressing) and inhibitory (parvalbumin expressing) neurons to measure visual and locomotion evoked activity in RSC and compare it to primary visual cortex (V1). We observed stimulus position and orientation tuning, and a retinotopic organization. Locomotion modulation of activity of single neurons, both in darkness and light, was more pronounced in RSC than V1, and while locomotion modulation was strongest in RSC parvalbumin-positive neurons, visual-locomotion integration was found to be more supralinear in CaMKII neurons. Longitudinal measurements showed that response properties were stably maintained over many weeks. These data provide evidence for stable representations of visual cues in RSC that are spatially selective. These may provide sensory data to contribute to the formation of memories of spatial information. Key words: behavior, navigation, retrosplenial cortex, V1, vision Introduction contribution to these processes. One proposal, which has been The rodent retrosplenial cortex (RSC), which comprises much of at the forefront of RSC research for the last decade, is that the medial part of the dorsal cortex, is reciprocally connected to this structure has an important role in integrating sensory and the hippocampal formation, anterior thalamic nuclei, and visual nonsensory information (Wolbers and Büchel 2005; Byrne et al. cortex. It is, therefore, perhaps unsurprising that this cortical 2007; Vann et al. 2009; Alexander and Nitz 2015); this proposal region has been repeatedly shown to be important for spatial reflects its anatomical connectivity, making it well placed to memory and navigation (Vann et al. 2009). However, there is combine cortical and subcortical sensory and motor-related still much uncertainty with respect to the nature of the RSC’s signals. Consistent with a role in integrating sensory signals, © The Author(s) 2020. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4425 experiments in anesthetized mice provide evidence of sensory surface of the skull. For RSC recordings, a custom head plate responses in the RSC (Murakami et al. 2015; Zhuang et al. 2017), was attached to the cranium using dental cement (Super Bond, while other work has described visually evoked responses in C&B), with an aperture approximately centered over the right feedback axons relaying signals between the RSC and visual hemisphere RSC. A 3-mm circular craniotomy was then made, cortex (Makino and Komiyama 2015). Meanwhile, unit record- centered approximately 2.5 mm caudal to bregma. The lateral ings in freely moving rats have showing that retrosplenial unit placement of the craniotomy was such that it encompassed a ensembles simultaneously map position in both the external small portion of the left hemisphere. This positioning meant and internal frames of reference (Alexander and Nitz 2015). that is was possible to visualize the central sinus through the This integrative model of retrosplenial function is further cranial window, which provided a useful reference point when supported by behavioral data, as RSC lesioned rats are particu- imaging. The craniotomy was closed with a glass insert made larly impaired on tasks where they are explicitly called upon to from three layers of circular glass (#1 thickness; 1 × 5 mm, integrate different types of information (Cooper and Mizumori 2 × 3 mm diameter) bonded together with optical adhesive 1999; Vann and Aggleton 2002; Pothuizen et al. 2008; Elduayen (Norland Products; catalogue no. 7106). The window was placed and Save 2014; Hindley et al. 2014). Despite the apparent impor- such that the smaller pieces of glass were in contact with the tance of the RSC for spatial cognition, there is surprisingly brain surface and the larger piece rested on the skull surround- little evidence from awake animals examining how sensory ing the craniotomy. The window was then sealed with dental information required for these functions are encoded in the cement. After surgery, all animals were allowed at least 1 week RSC. In particular, the extent to which simple visual stimuli are to recover before being imaged. For V1 recordings, the procedure encoded in an abstract or higher order versus veridical represen- was performed as described above except the craniotomy and tation (i.e., which encodes the physical properties of the stim- head plate position was centered −3.4 mm posterior and 2.8 mm ulus) remains unclear. More broadly, it remains uncertain how lateral from bregma of the right hemisphere. visual drive is integrated with motor input in the RSC. Finally, it remains unknown to which extent RSC visual and nonvisual Imaging and Locomotor Data Acquisition representations, in individual neurons, are longitudinally stable (as in early sensory cortex) or alternatively dynamic and context In vivo 2-photon imaging was performed using a resonant sensitive. scanning microscope (Thorlabs, B-Scope) with a 16× 0.8NA To address this gap in our knowledge, we used cellular res- objective (Nikon). GCaMP6 was excited at 980 nm using a olution 2-photon imaging, together with genetic labeling of cell Ti:sapphire laser (Coherent, Chameleon) with a maximum laser types, to assess how visual and motor-related signals are orga- power at sample of 50 mW. Data were acquired at approximately nized and processed in the RSC. A key step was to study head- 60 Hz and averaged, resulting in a frame rate of approximately fixed awake animals, allowing precise control and disentan- 10 Hz. For all imaging experiments, animals were head-fixed glement of visually driven and motor-driven neural activity. and free to run on a custom designed fixed-axis cylindrical This allowed an examination of whether overlapping or distinct treadmill. Movement was measured using a rotary encoder populations of RSC neurons process visual and motor signals, (Kübler, 05.2400.1122.0100). Imaging, behavioral, and visual and how stable or dynamic representations of these signals are stimulation timing data were acquired using custom written in single neurons over timescales of weeks. DAQ code (Matlab) and a DAQ card (NI PCIe-6323, National Instruments). Materials and Methods Visual Stimuli Animals Visual stimuli were generated in Matlab using the psychophysics All experimental procedures were carried out in accordance toolbox (32), and displayed on two calibrated LCD screens with the UK Animals (Scientific Procedures) Act 1986 and (Iiyama, B2080HS; width × height 26 × 47 cm) at right angles European Commission directive 2010/63/EU. Mice expressing to one another placed 20 cm from the eye. All visual stimuli GCaMP6f were generated by crossing the Ai95D line (Jax, 024105) were circular, unidirectionally drifting square wave gratings with either the CaMKII-alpha-cre T29-1 line (Jax, 005359), for of uniform size (40 × 40 ◦ ), spatial (0.08 cycles per degrees) expression in CaMKII positive cells, or the PV-cre line (Jax, and temporal frequency (1 Hz) presented at full contrast. 008069), for expression in parvalbumin (PV) positive cells. Stimulus position was corrected for viewing angle. Stimulus Experiments were carried out on adult male mice, housed under parameters were optimized in pilot experiments. Orientation normal light conditions (12:12 h light:dark). All recordings were and spatial location varied depending on the experiment. For all made during the light period. experiments, drifting gratings were presented for 2 s followed by a 2-s intertrial interval during which a gray screen was displayed. In all experiments, visual stimuli were presented passively, such Animal Surgical Preparation that the drifting of the grating was not yoked to the animal’s All surgical procedures were conducted under aseptic condi- movement. tions. Prior to cranial window surgery, animals were adminis- tered with the antibiotic Baytril (5 mg/kg, s.c.) and the anti- Experimental Design inflammatory drugs Rimadyl (5 mg/kg, s.c.) and dexametha- sone (0.15 mg/kg, i.m.). Anesthesia was induced at a concen- Low magnification imaging of the entire extent of RSC visible tration of 4% isoflurane in oxygen and maintained at 1.5–2% within the cranial window was used to localize the visually for the duration of the surgery. Once anesthetized, animals responsive region. In order to elicit visually driven activity, a were secured in a stereotaxic frame (David Kopf Instruments) vertically orientated drifting grating was presented in the center and the scalp and periosteum were removed from the dorsal of the binocular region of the visual field, centered at 20 ◦ 4426 Cerebral Cortex, 2020, Vol. 30, No. 8 elevation. Fro analysis of visual responsiveness as a function of Furthermore, let depth, Recordings were made at 11 different depths from pial surface to −300 μm, at two partially overlapping 900 × 900 μm fields of views within the RSC (one rostral and one caudal). M (θ) = e (κ cos (θ − θPref )) + ξ × e (κ cos (θ − θPref − π)) For experiments to determine spatial and orientation selec- tivity, 48 different stimuli were presented, with each stimulus where κ is the dispersion of the response and is related to the repeated 10 times in a pseudorandom order. The stimuli were tuning width of the neuron. θ is the orientation of the stimuli circular drifting gratings of one of eight different orientations, and θ Pref is the orientation preference of the neuron. Finally, ξ is presented in one of six locations in a 3 × 2 grid, centered at 0, the ratio of the response when the stimulus is displayed at θ Pref 45, and 90 ◦ in azimuth, and 0 and 20 ◦ in elevation. Recordings to the response at θ Pref –π radians. were made from a 400 × 400 μm field of view, aligned on The overall response of a neuron to stimuli of different the medial side with the midline, and centered approximately positions and orientations was fit with: 1.75 and 3.25 mm caudal of bregma for the rostral and caudal recording sites, respectively. Recordings were made at depths of 150–250 μm. Response(x,y.θ) = A(x,y) × M(θ) + E Calcium Imaging Data Analysis where E is the background firing rate. Calcium imaging data were registered and segregated into neu- Our experience was that Pref was difficult to fit due to ronal regions of interest (ROI) using Suite2P (Pachitariu et al. numerous local minima, hence the model was repeatedly fit 2016). Pixel-wise stimulus preference maps were constructed with Pref starting at all presented orientations, and the fit by first calculating the mean of the registered imaging frames with the lowest residual was chosen as the best fit. This model recorded during the drifting phase of each stimulus, and then was initially fit with the lsqcurvefit() function and subsequently, determining for each pixel, the stimulus which elicited the model performance was evaluated with fitnlm() and we excluded largest mean response. from further analysis fits where R2 < 0.5. Cross sections of the For experiments in which the stability of visual responses fits of the neurons included for further analysis were manually was tracked longitudinally, recordings were made from the inspected and compared with the raw data (i.e., the orienta- same field of view over multiple sessions. Cortical surface tion tuning curve at the fitted preferred position, and the 2D vascular landmarks were used to locate the same neurons Gaussian at the fitted preferred orientation). Supplementary between sessions. In order to match ROI masks detected with Figure S8 shows examples of raw data and cross sections of Suite2P between imaging sessions, the cpselect, fitgeotrans, and representative fits of varying goodness of fit. imwarp functions in Matlab were used to warp the ROI masks Orientation tuning curves were calculated using data pooled using manually identified control points selected from the mean over the two directions of grating drift. Orientation tuning fit image frame from each experiment. Masks that overlapped by curves were used to calculate orientation selectivity index (OSI) more than 60% were detected in all sessions and were flagged defined as (peak response−trough response)/peak response. as potential longitudinally recorded neurons. These masks were An integration index was calculated to quantify the linearity then manually verified by visual observation of the region of integration of visually and locomotion evoked activity in RSC of the mean each frame that they corresponded to in each neurons. The integration index was calculated as: (integrated − experiment. sum of isolated)/(integrated + sum of isolated). Here “integrated” is defined as the mean response of a single neuron during simul- taneous visual stimulation at its preferred orientation and any Quantification and Statistical Analysis of form of locomotion, and “sum of isolated” is defined as the sum Visual and Motor Responses of the response during locomotion without visual stimulation The visual responses of individual neurons were quantified as and visual stimulation without locomotion. the mean dF/F value between 0.5–1.5 s after stimulus onset, A neuron was classified as visually responsive if it showed with a baseline period quantified for each trial as the mean dF/F a statistically significant increase in activity between the base- between 1 and 0 s before stimulus onset. Stimulus response was line and visual stimulation periods for at least one stimulus, fit with the product of a 2D Gaussian (to represent the spatial with statistical significance calculated with a shuffle test (1000 extent of the receptive field) and the sum of two Von Mises shuffles, P < 0.05), corrected for multiple comparisons using the functions (to represent the orientation selectivity of the neuron). mafdr false discovery rate function in Matlab. Note that, using Specifically, let these criteria approximately 5% of neurons would be expected to be classified as visually responsive by chance. A one-way ANOVA of response amplitudes over stimulus position, orientation, or  2 −(x − μx )2 − y − μy direction was used to assess whether neurons discriminated A(x,y) = a × e + 2σx2 2σy2 these stimulus features. Statistical significance of correlations between run speed and neural activity was calculated with a shuffle test in which the entire trace of run speed was randomly where a is the maximum response, x is the position of the shifted relative to neural activity traces 1000 times (P < 0.05). stimuli, μx is the center of the receptive field, and σ x is the spa- Where appropriate similarity of variance and normality of distri- tial extent of the receptive field, respectively, in the horizontal bution were checked with the vartestn Matlab function, and the dimension. Similarly, y is the position of the stimuli, μy is the Kruskal–Wallis test was used as noted when the assumptions of center of the receptive field, and σ x is the spatial extent of the the one-way ANOVA or t-test were not met. T-tests used were receptive field, respectively, in the vertical dimension. two sided. Correction of P values for multiple comparisons was Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4427 Figure 1. Localization of visually responsive area of dysgranular RSC. (A) Setup schematic (left) and cranial window over RSC (right). Blue box indicates area imaged. Scale bar: 1 mm. (B) Low magnification 2-photon visual response maps showing visually evoked activity of CaMKII neurons in the cRSC (red box) but not rRSC (green box). Scale bar: 160 μm. (C,D) Higher magnification of rostral (C) and caudal (D) regions showing visually responsive neurons in caudal but not rostral areas. Scale bar: 30 μm. (E) Mean fraction of visually responsive neurons (error bars represent SEM) detected in caudal and rostral areas. calculated using the Matlab function multcompare using the the visual responsiveness of this caudal region (Supplementary Tukey–Kramer method. Fig. S1D). Response maps of the parvalbumin retrosplenial cortex (PV RSC) population showed a similar pattern of greater visual responsiveness in the cRSC versus rostral RSC (rRSC), Results but with significantly lower amplitude (Supplementary Fig. S1E). We next made higher resolution recordings from the Spatial Organization of Visually Responsive Cells in RSC layer 2/3 of the cRSC and rRSC in which individual CaMKII We first used low magnification 2-photon imaging of GCaMP6f neurons could be resolved (Fig. 1C,D). Consistent with our low expressed in CaMKII positive cells (i.e., primarily excitatory magnification observations, higher magnification pixel-wise neurons, Supplementary Fig. S1A) or parvalbumin expressing maps showed a large fraction of visually responsive cells in inhibitory neurons (PV neurons, Supplementary Fig. S1B) to cRSC but not rRSC regions (Fig. 1E; see also Supplementary localize visually responsive areas of RSC. Recordings were made Fig. S2A). A cell-wise analysis of visual responses of neurons in awake animals free to walk on a cylindrical treadmill (Fig. 1A). confirmed that a significantly larger fraction of caudal than We sampled the region between −2.4 and −4.2 mm AP and 0.0 rostral neurons were visually responsive (fraction of neurons and 0.9 mm ML relative to bregma, and depths between the visually responsive: 40.4 ± 6.4% caudal vs. 6.3 ± 7.4% rostral; pial surface and 300 μm. The area was tiled by two partially t-test: t(8) = 5.26, P = 0.0058; n = 5 and n = 5 mice, respectively; overlapping 900 × 900 μm fields of view (one anterior, one note that, 5% of neurons are expected to be classified as posterior), and data were acquired at depths spaced by 30 μm visually responsive by chance). These results show that visually in z (Supplementary Fig. S1C), resulting in 11 depths per field responsive CaMKII and PV neurons are largely confined to of view. We first generated mean pixel-wise response maps to caudal regions of the RSC and that within cRSC a larger fraction binocularly presented drifting grating visual stimulation and of neurons is visually responsive than previously reported in averaged across the 11 depths. Response maps of the CaMKII anesthetized animals (Murakami et al. 2015). RSC population showed a visually responsive area limited to the Spatial and Orientation Selectivity of Visually caudal RSC (cRSC) that was visible both in maps from individual Responsive Cells in cRSC animals and in the mean response maps averaged across five animals (Fig. 1B). Analysis of response maps across the 11 depths We next investigated the response properties of individual (spanning 300 μm) showed no systematic variation with depth in CaMKII neurons in the visually responsive cRSC. A recent 4428 Cerebral Cortex, 2020, Vol. 30, No. 8 Figure 2. Visual selectivity of CaMKII cRSC neurons. (A,B) Mean imaging frame from cRSC (A), and pixel-wise map from the same field of view of retinotopic preference in azimuth (B). Scale bar: 30 μm. (C) Azimuth position preference compared with distance of neurons from midline. Gray lines represent individual animals, and black line is average. (D) Mean raster plot of neural responses to stimuli at different retinotopic locations, with neurons (rows) sorted by positional preference. (E) Mean fractions of neurons (error bars indicate SEM) statistically significantly discriminating stimulus position or orientation in cRSC and V1. (F) Example traces of orientation selective and direction selective neurons. (G) Polar plots showing distribution of directional preference of RSC neurons (red) and V1 neurons (green) divided into those representing the monocular and binocular visual field. P values indicate statistical significance of Raleigh test of nonuniformity and show nonuniformity of RSC direction preference distribution. study using widefield imaging suggested the presence of a directions (four orientations). Pixel-wise maps were then retinotopic map in the RSC (Zhuang et al. 2017). We therefore calculated for azimuth stimulus preference (Fig. 2B). This first tested for the existence of retinotopy at single cell showed a broad range of retinotopic preferences within the resolution. Animals were visually stimulated with 40 × 40 ◦ 400 × 400 μm field of view, which included the full range of grating patches at one of six retinotopic locations and eight stimulus positions presented. Retinotopic organization was Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4429 Figure 3. Comparison of the activity of CaMKII RSC and V1 neurons in darkness. (A) Run speed, and below, raster representation of neural activity (rows are individual neurons) sorted by run speed correlation. (B,C) Histogram of run speed correlations of RSC neurons (B, black) and V1 neurons (C, white) with the nonsignificantly correlated fraction of neurons (determined with shuff le test) represented by gray bars. (D,E) Mean fraction of neurons which are significantly negatively locomotion correlated (D) or positively locomotion correlated (E). Error bars indicate SEM. apparent in the pixel maps of azimuth (which we sampled over responses (Fig. 2D; Supplementary Fig. S2B). In the azimuth map, a broader range than elevation due to restrictions in possible more medial and lateral parts of the map corresponded to more display positions, Fig. 2B,C; mean correlation between azimuth binocular (i.e., nasal) and monocular (i.e., temporal) visual space, preference and distance from midline = 0.85 ± 0.07; n = 5 mice), respectively. We next compared the spatial selectivity of cRSC and individual neurons typically exhibited spatially selective CaMKII neurons to that of CaMKII primary visual cortex (V1) 4430 Cerebral Cortex, 2020, Vol. 30, No. 8 neurons stimulated with the same stimulus. A similar fraction RSC vs. 10.5 ± 0.02% V1; t-test: t(9) = 2.43, P = 0.0338; n = 10 fields of visually responsive cRSC and V1 neurons discriminated of view [five mice] and n = 6 fields of view [three mice], Fig. 3D,E). stimulus position (fraction of neurons discriminating stimulus Consequently, some RSC neurons were found to be active in position: 73.7 ± 0.71% cRSC vs. 73.69 ± 7.1% V1; t-test: t(9) = 0.367, darkness and reliably suppressed by locomotion (Fig. 3A, upper P = 0.723; n = 5 and n = 6 mice respectively; Fig. 2E). In contrast, rows), while other neighboring neurons recorded simultane- we found that only 25.1 ± 2.6% of visually responsive cRSC ously exhibited the opposite behavior (Fig. 3A, lower rows). We neurons significantly discriminated stimulus orientation or next recorded from PV RSC neurons under the same dark condi- direction (at preferred position; Fig. 2E; F), which compared tions (Fig. 4A). A similar distribution was again observed in cRSC with 43.94 ± 0.06% of V1 neurons (t-test: t(9) = 2.64, P = 0.027; and rRSC and so neurons from the two areas were again pooled n = 5 and n = 6 mice respectively; Fig. 2E). This was reflected (see Supplementary Fig. S4A,B for data split into rostral and in a median orientation selectivity index of 0.63 in cRSC caudal regions). In contrast to the CaMKII population (Fig. 4B), as compared with 0.81 in V1 (Kruskal–Wallis test: P < 10−12 , PV neurons exhibited an overall higher degree of locomotion n = 436 and n = 307 neurons, respectively). When we binned correlation (Fig. 4C), both in neurons which were positively cor- neurons by their preferred direction of stimulus motion, we related (median R = 0.19 CaMKII vs. 0.34 PV; Kruskal–Wallis test additionally observed a significant bias in directional preference P < 10−8 ; Fig. 4B–D) and in negatively correlated neurons (median of orientation selective cRSC neurons (defined as neurons with R = −0.13 CaMKII vs. −0.26 PV; Kruskal–Wallis test: P < 10−14 ; OSI > 0.5) toward naso-temporal motion (Fig. 2G). This bias was Fig. 4B,C,E). not observed in orientation selective V1 neurons and was apparent both in neurons encoding monocular visual space Interaction of Visual and Locomotion Signals in (Raleigh test for nonuniformity; V1: n = 40 neurons; P = 0.1575; Responses in RSC RSC, n = 56 neurons, P = 0.0004) and binocular visual space (V1: n = 115 neurons, P = 0.7779; RSC: n = 92 neurons, P = 0.0325), We next sought to determine the interaction between visual although it appeared more pronounced in monocular neurons. input and locomotion-related input in individual cRSC Finally, we observed that the degree of habituation of responses neurons (Fig. 5A–D). We first quantified the extent to which over the course of the experiment of visually responsive neurons the visually responsive cRSC population overlaps with the to their preferred stimuli differed between V1 and RSC neurons, locomotion correlated population. To do this, we compared with RSC neurons exhibiting a greater reduction in F/F than the run speed correlation coefficients of visually responsive V1 neurons (reduction of 0.35 vs. 0.11; t-test: t(677) = 3.31, versus nonvisually responsive cells. We found no significant P = 0.001; Supplementary Fig. S2D). The apparent greater rate differences between these groups, either in CaMKII neurons of adaptation observed in RSC complicates the interpretation (Supplementary Fig. S5A) or PV neurons (Supplementary Fig. of the findings of fraction of cells responding significantly to S5B), suggesting no correlation between visual stimulus sensi- different stimulus features. To address this issue, we limited tivity and locomotion sensitivity in individual cRSC neurons. As analysis to the first 20% of the trials of the experiment where was the case in darkness, we found that PV neurons on average adaptation is similar in V1 and RSC. While this reduced the exhibited a higher degree of correlation with run speed than overall fractions of discriminating neurons due to the reduction CaMKII neurons during visual stimulation, both in neurons that in the number of trials, we observed the same overall pattern were positively correlated (median R = 0.11 CaMKII vs. 0.16 PV; described above of similar position discrimination between Kruskal–Wallis test: P < 10−4 ; Supplementary Fig. S5B, left panel) the RSC and V1, and greater orientation discrimination in V1 and in negatively correlated neurons (median R = −0.06 CaMKII (Supplementary Fig. S2E). Together these data show that a vs. −0.15 PV; Kruskal–Wallis test: P < 10−12 ; Supplementary Fig. significant fraction of cRSC neurons shows response selectivity S5B, right panel). We next studied how visual and locomotion- for retinotopy, orientation, and direction of visual flow and as a related activity interacts in RSC neurons, specifically whether whole, the population is organized retinotopically. the integration of these two types of activity is sublinear, linear, or supralinear. To examine this question, we calculated an integration index which varied between −1 and 1, and where Locomotion Sensitivity of RSC Neurons in Darkness a value of 0 indicates linear integration (i.e., activity when Sensory processing by V1 neurons is modulated by the loco- visual stimulation occurs at the same time as locomotion can be motor state of the animal (Niell and Stryker 2010; Keller et al. explained by the sum of isolated visually and locomotion evoked 2012; Saleem et al. 2013; Ranson 2017; Dipoppa et al. 2018). We activity). Negative and positive values of the integration index next examined the extent to which RSC sensory processing is indicate sublinear and supralinear integration, respectively (see subject to the same modulation by recording RSC and V1 CaMKII Materials and Methods section). Consistent with supralinear neurons in complete darkness. We found that, in both the RSC integration of visual and locomotion signals, 76% of CaMKII and V1, a fraction of neurons was significantly correlated with neurons had an integration index of greater than 0 (t-test run speed (either positively or negatively; shuffle test, P < 0.05; to test difference of index from zero: t(450) = 11.19, P < 1010 ), Fig. 3A–C; see Supplementary Fig. S3C for a similar plot from an and the CaMKII population overall had a mean integration example V1 recording). Both cRSC and rRSC exhibited a simi- index of 0.27 ± 0.024. In contrast, on average PV neurons lar pattern of run speed correlations in darkness and so were integrated near linearly with an integration index of 0.02 ± 0.058 combined in this analysis (see Supplementary Fig. S3A,B for data (t-test to test difference of index from zero: t(48) = 0.33, split into rostral and caudal regions). A larger fraction of RSC P = 0.75), and the mean integration indices differed signifi- than V1 CaMKII neurons was found to be significantly positively cantly between CaMKII and PV cell types (t-test: t(498) = 3.31, correlated with run speed (25.9 ± 0.02% RSC vs. 10.8 ± 0.02% V1; P = 0.001). t-test: t(9) = 4.32, P = 0.0003; n = 10 fields of view [five mice] In summary, in darkness, RSC neurons are significantly and n = 6 fields of view [three mice] respectively, Fig. 3D,E) or more locomotion modulated than V1 neurons, and within the significantly negatively correlated with run speed (17.1 ± 0.02% RSC population, PV neurons are more locomotion modulated Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4431 Figure 4. Activity of PV RSC neurons in darkness. (A) Run speed, and below, raster representation of neural activity (rows are individual neurons) sorted by run speed correlation. (B,C) Histogram of run speed correlations of CaMKII neurons (B, black) and PV neurons (C, white) with the nonsignificantly correlated fraction of neurons represented by gray bars. (D,E) Median fraction of neurons which is significantly positively locomotion correlated (D) or negatively locomotion correlated (E). Error bars indicate interquartile range. 4432 Cerebral Cortex, 2020, Vol. 30, No. 8 Figure 5. Interaction of visual and locomotion induced activity in CaMKII and PV neurons. (A,B) Run speed, and below, raster representation of neural activity during visual stimulation sorted by run speed correlation in CaMKII neurons (A) and PV neurons (B). (C,D) Example traces of running suppressed (upper) and running facilitated neurons (lower) during visual stimulation, together with running periods and stimulus onset times (black) in CaMKII neurons (C) and PV neurons (D). (E) Median run speed correlations during visual stimulation of significantly positively and negatively correlated neurons (left and right respectively). Error bars indicate interquartile range. (F,G) Histogram of indices of visual-locomotion integration of CaMKII (black, F) or PV (white, G) neurons. than CaMKII neurons, with a subset of PV neurons particu- locomotion-related activity differently: PV responses are largely larly strongly suppressed during locomotion. Comparison of explained by a summation of activity during visual stimulation responses during running and visual stimulation additionally and locomotion, while CaMKII neurons exhibit responses that suggests that RSC CaMKII and PV neurons integrate visual- and suggest supralinear integration. Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4433 Long-term Stability of RSC Visual and Locomotion in 52 longitudinally tracked cells from three animals (mean Signals neurons per animal 17.33 ± 1.8; see Supplementary Fig. S7B) as well as the 170 CaMKII neurons described above (Supplemen- The degree of long-term stability of sensory-evoked activity tary Fig. S7A). We generated pixel-wise maps of locomotion varies between brain regions (Clopath et al. 2017), with early modulation by calculating for each pixel (moving frames − sta- sensory areas tending to be on the whole more stable (Mank et al. tionary frames)/(moving frames + stationary frames) and then 2008; Margolis et al. 2012; Rose et al. 2016; Ranson 2017; Jeon et al. segregated this map into the ROIs which could be identified 2018), while regions containing more abstract representations, longitudinally (Fig. 7A; Supplementary Fig. S7A). Neurons of both such as regions of the hippocampus, exhibit more dynamic and cell types within cRSC exhibited a range of degrees of locomotion context sensitive representations (Mankin et al. 2012; Ziv et al. modulation, and a high degree of intersession stability could 2013). As the RSC occupies an intermediate position between be seen in many neurons (Fig. 7A, columns 1 and 2). We next early sensory cortex and the hippocampus, we next examined segregated the field of view into longitudinally tracked neurons the extent to which sensory-evoked activity and locomotion and calculated run speed correlations of each neuron in the modulation of RSC activity was stable over timescales of weeks. two sessions (Fig. 7B,C). This also revealed a striking degree of To measure the stability of visually evoked activity, we stability of run speed correlation between sessions that was focused on cRSC and the CaMKII population as this was the area most pronounced in PV neurons (R = 0.78; Pearson’s correlation and cell type in which we most frequently observed visually coefficient: P < 10−11 ), but also highly significant in CaMKII evoked activity. Over the course of 4 weeks, we made three neurons (Pearson’s correlation coefficient: R = 0.41, P < 10−7 ). evenly spaced recordings of the same CaMKII cRSC neurons, In summary, these results show that a large fraction of RSC relocated from session to session using vascular landmarks. neurons exhibits a high degree of stability of both sensory Pixel-wise maps of stimulus preference showed clear similarity responses and modulation by locomotion. PV neurons were of retinotopic preference in all animals measured over the observed to exhibit a particularly high correlation with locomo- 4-week period (Fig. 6A). To quantitatively assess similarity, we tion and also a high degree of stability of run speed correlation segregated the field of view of each experiment into neuronal over the timescale measured. ROIs and validated each ROI as present and valid in each session. This resulted in 170 unambiguously identified CaMKII neurons from three animals that were visible in all three sessions (mean neurons per animal 56.7 ± 9.9, see Supplementary Fig. S6A). We Discussion next measured the fraction of neurons with significant visual The RSC has been suggested to function as an integrative hub for responses over the three sessions and found that the largest sensory and motor signals. Here, we examined the organization fraction (36.5%) had significant responses in all three sessions of visually evoked activity within the RSC in awake mice, how with 19.4% responsive in two sessions, 27.7% responsive in one it integrates locomotion-related activity, and how stable these session, and 16.5% of neurons not responding in any session motor and visual representations are over periods of weeks. (Fig. 6B). For each neuron, we then calculated the difference We observed that a substantial fraction of dysgranular RSC between its fitted azimuth retinotopic preference for each pair excitatory neurons exhibit visually evoked activity, and that this of sessions and, consistent with the observed similarity of pixel subpopulation is largely limited to the more caudal RSC. The maps, found that the majority of neurons (90%) had differences dysgranular RSC has gradients of connections that are consis- in positional preference of less than 10 ◦ (mean difference tent with this observation. In particular, caudal dysgranular RSC 3.86 ± 0.89, compared with mean difference with shuffled is more interconnected (both afferent and efferent) with primary neuron identities 30.91 ± 0.358; Fig. 6C). We next compared the and higher cortical visual areas (Wyss and Van Groen 1992). similarity of orientation preference in longitudinally identified Consequently, these results suggest that visual responsiveness orientation selective neurons (OSI > 0.5) over all pairs of is concentrated in those RSC areas with direct visual inputs, that sessions. The above OSI threshold was used to ensure analysis is, the intrinsic connections linking different parts of dysgran- of stability of orientation preference was only undertaken in ular RSC (Jones et al. 2005) are not sufficient to convey visual neurons in which orientation preference could be accurately responsiveness as measured in this study. Likewise, it is the measured. We found that while the largest fraction of neurons caudal dysgranular RSC that shows the clearest difference in c- (38.9%) had orientation preferences which differed by less than fos activity when contrasting spatial behaviors in the light and 10 ◦ between sessions, many neurons had large differences in the dark (Pothuizen et al. 2009). preference (median difference 20.55 ◦ , Fig. 6D). This difference in We further found that many visually responsive cRSC tuning preference is considerably larger than previous reports CaMKII neurons exhibited retinotopic, orientation, and direction from awake mouse V1 of median differences in orientation selectivity. This was largely preserved over periods of several preference of approximately 3–4 ◦ over a 1-week period (Jeon weeks stably. Unexpectedly, over the short timescale of the daily et al. 2018). In summary, these results show longitudinal stability experimental sessions, we observed an apparent difference of visual response tuning over periods of weeks in many cRSC between V1 and RSC in the rate of adaptation of visually neurons, particularly with respect to retinotopic preference, and evoked responses, with RSC adapting more rapidly. This to a lesser degree orientation preference. has not previously been reported in other similar studies of We next sought to determine whether the degree of loco- visual processing in RSC, which may be due to differences in motion modulation of individual neurons was also stable over preparation such as anesthetized versus awake recordings. time. As both CaMKII and PV cRSC neurons exhibited clear In future studies of visual responses of RSC, it will be of locomotion modulation, we analyzed both populations with a interest to understand the mechanism of this adaptation, spacing of 1–2 weeks between recordings. We used the same cri- as well as more generally taking potential adaptation into teria outlined above to identify PV neurons, which were unam- account in experiment design. The observation of a high biguously detected in all recording sessions, and this resulted degree of retinotopic selectivity is perhaps surprising given 4434 Cerebral Cortex, 2020, Vol. 30, No. 8 Figure 6. Longitudinal stability of retinotopic positional preference in cRSC neurons over weeks. (A) Single field of view imaged over three sessions spread over 4 weeks, showing mean imaging frame (above) and pixel-wise azimuth retinotopic preference map (below). Scale bar: 30 μm. (B) Examples of longitudinally tracked neurons, which have stable visual responses, unstable responses or were stably unresponsive. (C,D) Between session difference in retinotopic preference (C) and orientation preferences (D) for each neuron on each pair of sessions on which it was measured. previous evidence of the importance of dysgranular RSC in bias in directional preference of cRSC neurons toward naso- tasks dependent on allocentric encoding (Vann and Aggleton temporal motion may indicate a role in the processing of visual 2005) as well as recent evidence of place field-like activity (Mao flow information. Computational models show that visual flow et al. 2017). Our observation of cellular resolution, retinotopic provides information about egocentric motion and influences organization of cRSC is consistent with the view that RSC firing patterns in spatially tuned cells during rodent navigation contains sensory representations of an egocentric nature, by (Raudies et al. 2012). Head direction cells are present in the which it is meant that stimuli in specific locations defined rodent RSC (Chen et al. 1994), therefore this area may play a role by their position relative to the observer can activate specific in updating head orientation during movement. RSC cells. Therefore, in future work, it will be important to A large number of studies in mice has reported locomo- reconcile how the retinotopically structured map of visual space tion modulation of both baseline and sensory-evoked cortical reported here, interacts with any allocentric maps of space. The activity in early sensory areas, including the primary visual and Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4435 Figure 7. Longitudinal stability of locomotion correlation of cRSC neurons over weeks. (A) Examples of longitudinally tracked CaMKII neurons (left) or PC neurons (right) which had stable locomotion modulation, unstable modulation or were stably unmodulated. (B,C) Inter-session comparison of locomotion modulation of CaMKII neurons (B) or PV neurons (C). auditory cortices, with the direction of effects varying between Interestingly, the visual responsiveness and locomotion cortical areas, stimulation conditions, and cell types (Niell and modulation of cRSC neurons were found to be largely unre- Stryker 2010; Ayaz et al. 2013; Polack et al. 2013; Keller et al. 2013; lated to one another, with both highly visual and visually Ranson 2017; Saleem et al. 2013; Benevento et al. 2016; Dipoppa nonresponsive neurons exhibiting similar levels of locomotion et al. 2018; Shimaoka et al. 2018). Our recordings of RSC in modulation on average. The parvalbumin-positive population the dark revealed that a significant fraction of dysgranular RSC of inhibitory RSC neurons was particularly strongly modulated neurons also exhibit baseline activity, which is either strongly by locomotion, both in the dark and during visual stimulation, positively or negatively correlated with locomotion. As rRSC is and this modulation was strikingly stably maintained over the more strongly connected than cRSC to both primary motor and interval of weeks examined. This could provide a mechanism of secondary motor cortex (Wyss and Van Groen 1992; Shibata and distinct modes of processing in RSC dependent upon behavioral Naito 2008; Yamawaki et al. 2016), we might have expected to see context, whereby functionally distinct subpopulations are either the strongest correlates of locomotion in rostral regions of RSC. facilitated or suppressed depending upon locomotion state. Instead, we observed comparable locomotion-related activity Interestingly, PV and CaMKII neurons were found to integrate across the rRSC and cRSC subregions, which may indicate that responses to visual stimulation and responses to locomotion locomotion correlations are driven by other motor- or arousal- differently. Specifically, PV neurons showed more linear and related signals such as neuromodulatory input (Fu et al. 2014). CaMKII neurons more supralinear response integration on These findings may also relate to previous studies showing average, suggesting responses of CaMKII neurons could amplify the importance of the RSC for path integration, which relies the salience of certain combinations of behavior state and visual on locomotion-derived cues to update internal representations input. It may be critical to explore this integrative process in a of space (Cooper and Mizumori 1999, 2001; Elduayen and Save more enriched context of active navigation to understand the 2014). significance of this operation. It is additionally important to 4436 Cerebral Cortex, 2020, Vol. 30, No. 8 note that, we do not know whether this integration is occurring Auger SD, Maguire EA. 2018. Dissociating landmark sta- in RSC neurons or elsewhere and being inherited by RSC. In fact, bility from orienting value using functional magnetic we know that the latter is almost certainly at least partly the resonance imaging. J Cogn Neurosci. 30(5):698–713. doi: case as it is now well established that visually evoked activity 10.1162/jocn_a_01231. in the visual cortex, which provides visual input to RSC, is itself Auger SD, Mullally SL, Maguire EA. 2012. Retrosplenial cor- locomotion modulated (Niell and Stryker 2010; Keller et al. 2012; tex codes for permanent landmarks. PLoS One. 7(8). doi: Ayaz et al. 2013; Polack et al. 2013; Saleem et al. 2013; Dipoppa 10.1371/journal.pone.0043620. et al. 2018). Auger SD, Zeidman P, Maguire EA. 2015. A central role for the Given the diverse roles ascribed to RSC, the similarities to retrosplenial cortex in de novo environmental learning. elife. primary visual cortex with respect to the organization of visually 4 e09031 doi: 10.7554/eLife.09031. responsive neurons and their response properties are surprising. Ayaz A, Saleem AB, Schölvinck ML, Carandini M. 2013. Locomo- In common with V1, many RSC neurons respond to particular tion controls spatial integration in mouse visual cortex. Curr positions, orientations, and directions of flow of visual stim- Biol. 23(10):890–894. doi: 10.1016/j.cub.2013.04.012. uli, are topographically organized into a retinotopic map, and Benevento M, Iacono G, Selten M, Ba W, Oudakker A, Frega M, exhibit stable visual preferences over periods of weeks. It will Keller J, Mancini R, Lewerissa E, Kleefstra T. et al. 2016. Histone be of interest to determine how other modalities of sensory methylation by the Kleefstra syndrome protein EHMT1 medi- information are functionally represented and topographically ates homeostatic synaptic scaling. Neuron. 91(2):341–355. doi: organized, and how these maps are integrated with those of 10.1016/j.neuron.2016.06.003. visually responsive cells. Byrne P, Becker S, Burgess N. 2007. Remembering the past In summary, our results show that a subpopulation of RSC and imagining the future: a neural model of spatial neurons surprisingly stably relay retinotopic, orientation, and memory and imagery. Psychol Rev. 114:340–375. doi: direction information about visual cues as well as locomotion 10.1037/0033-295X.114.2.340. signals. It will be important in future to determine how Chen LL, Lin LH, Green EJ, Barnes CA, McNaughton BL. 1994. these view-dependent visual feature representations interact Head-direction cells in the rat posterior cortex—I. Anatom- with and contribute to view-independent or allocentric RSC ical distribution and behavioral modulation. Exp Brain Res. representations. One emergent issue relates to evidence that the 101(1):8–23. doi: 10.1007/BF00243212. RSC may be involved in the identification of stable landmarks Clopath C, Bonhoeffer T, Hübener M, Rose T. 2017. Vari- in an environment (Auger et al. 2012, 2015; Auger and Maguire ance and invariance of neuronal long-term representa- 2018; Mitchell et al. 2018). While the separation between view- tions. Philosl Trans R Soc B Biol Sci. 372(1715):20160161. doi: dependent and view-invariant representations enables different 10.1098/rstb.2016.0161. forms of scene and object-based information (Wood 2009), it is Cooper BG, Mizumori SJ. 1999. Retrosplenial cortex inactiva- their co-operation that would help the formation of landmarks tion selectively impairs navigation in darkness. Neuroreport. to assist flexible navigation. 10:625–630. doi: 10.1097/00001756-199902250-00033. Cooper BG, Mizumori SJ. 2001. Temporary inactivation of the Supplementary Material retrosplenial cortex causes a transient reorganization of spa- tial coding in the hippocampus. J Neurosci. 21:3986–4001. doi: Supplementary material is available at Cerebral Cortex online. 21/11/3986 [pii]. Dipoppa M, Ranson A, Krumin M, Pachitariu M, Carandini M, Funding Harris KD. 2018. Vision and locomotion shape the interac- tions between neuron types in mouse visual cortex. Neuron. This work was supported by a Biotechnology and Biological Sci- 98(3):1–14. doi: 10.1016/j.neuron.2018.03.037. ences Research Council research grant awarded to JA, AN, SV and Elduayen C, Save E. 2014. The retrosplenial cortex is necessary FS (BB/L021005/1), a Sêr Cymru fellowship (80762-CU-080) to AR, for path integration in the dark. Behav Brain Res. 272:303–307. a Wellcome Trust Strategic Award (100202/z/12/z) to Michael J. doi: 10.1016/j.bbr.2014.07.009. Owen, J.H., Lawrence Wilkinson, Adrian Harwood, Meng Li, David Fu Y, Tucciarone JM, Espinosa JS, Sheng N, Darcy DP, Linden, John Aggleton, Vincenzo Crunelli, and Derek Jones, and Nicoll R a, Stryker MP. 2014. A cortical circuit for gain a Wellcome Trust ISSF Seedcorn Award (105613/Z/14/Z) to A.R. control by behavioral state. Cell. 156(6):1139–1152. doi: S.V. is funded by a Wellcome Trust Senior Research Fellowship 10.1016/j.cell.2014.01.050. (212273/Z/18/Z). Hindley EL, Nelson AJD, Aggleton JP, Vann SD. 2014. The rat retrosplenial cortex is required when visual cues are used Notes flexibly to determine location. Behav Brain Res. 263:98–107. For the use of GCaMP6S, we acknowledge Vivek Jayaraman, Rex doi: 10.1016/j.bbr.2014.01.028. A. Kerr, Douglas S. Kim, Loren L. Looger, Karel Svoboda from the Jeon BB, Swain AD, Good JT, Chase SM, Kuhlman SJ. 2018. GENIE Project, Janelia Farm Research Campus, Howard Hughes Feature selectivity is stable in primary visual cortex across Medical Institute. a range of spatial frequencies. Sci Rep. 8(1):1–14. doi: Conflict of Interest: None declared. 10.1038/s41598-018-33633-2. Jones BF, Groenewegen HJ, Witter MP. 2005. Intrinsic connec- tions of the cingulate cortex in the rat suggest the existence References of multiple functionally segregated networks. Neuroscience. Alexander AS, Nitz DA. 2015. Retrosplenial cortex maps the 133(1):193–207. doi: 10.1016/j.neuroscience.2005.01.063. conjunction of internal and external spaces. Nat Neurosci. Keller GBB, Bonhoeffer T, Hübener M. 2012. Sensorimotor 18(July):1–12. doi: 10.1038/nn.4058. mismatch signals in primary visual cortex of Stable Encoding of Visual Cues in the Mouse Retrosplenial Cortex Powell et al. 4437 the behaving mouse. Neuron. 74(5):809–815. doi: Raudies F, Mingolla E, Hasselmo ME. 2012. Modeling the influ- 10.1016/j.neuron.2012.03.040. ence of optic flow on grid cell firing in the absence Makino H, Komiyama T. 2015. Learning enhances the relative of other cues 1. J Comput Neurosc. 33(3):475–493. doi: impact of top-down processing in the visual cortex. Nat 10.1007/s10827-012-0396-6. Neurosci. 18(8):1116–1122. doi: 10.1038/nn.4061. Rose T, Jaepel J, Hubener M, Bonhoeffer T. 2016. Cell-specific Mank M, Santos AF, Direnberger S, Mrsic-Flogel TD, Hofer restoration of stimulus preference after monocular depri- SB, Stein V, Hendel T, Reiff DF, Levelt C, Borst A. et al. vation in the visual cortex. Science. 352(6291):1319–1322. doi: 2008. A genetically encoded calcium indicator for chronic 10.1126/science.aad3358. in vivo two-photon imaging. Nat Methods. 5(9):805–811. doi: Saleem AB, Ayaz A, Jeffery KJ, Harris KD, Carandini M. 2013. 10.1038/nmeth.1243. Integration of visual motion and locomotion in mouse Mankin EA, Sparks FT, Slayyeh B, Sutherland RJ, Leutgeb S, visual cortex. Nat Neurosci. 16(12):1864–1869. doi: Leutgeb JK. 2012. Neuronal code for extended time in the 10.1038/nn.3567. hippocampus. Proc Natl Acad Sci. 109(47):19462–19467. doi: Shibata H, Naito J. 2008. Organization of anterior cingulate and 10.1073/pnas.1214107109. frontal cortical projections to the retrosplenial cortex in the Mao D, Kandler S, McNaughton BL, Bonin V. 2017. Sparse rat. J Comp Neurol. 506(1):30–45. doi: 10.1002/cne.21523. orthogonal population representation of spatial context Shimaoka D, Harris KD, Carandini M. 2018. Effects of arousal in the retrosplenial cortex. Nat Commun. 8(1):243. doi: on mouse sensory cortex depend on modality. Cell Rep. 10.1038/s41467-017-00180-9. 22(12):3160–3167. doi: 10.1016/j.celrep.2018.02.092. Margolis DJ, Lütcke H, Schulz K, Haiss F, Weber B, Kügler S, Hasan Vann SD, Aggleton JP. 2002. Extensive cytotoxic lesions of the rat MT, Helmchen F. 2012. Reorganization of cortical population retrosplenial cortex reveal consistent deficits on tasks that activity imaged throughout long-term sensory deprivation. tax allocentric spatial memory. Behav Neurosci. 116(1):85–94. Nat Neurosci. 15(11):1539–1546. doi: 10.1038/nn.3240. doi: 10.1037//0735-7044.116.1.85. Mitchell AS, Czajkowski R, Zhang N, Jeffery K, Nelson AJD. 2018. Vann SD, Aggleton JP. 2005. Selective dysgranular retrosplenial Retrosplenial cortex and its role in spatial cognition. Brain cortex lesions in rats disrupt allocentric performance of the Neurosci Adv. 2. doi: 10.1177/2398212818757098. radial-arm maze task. Behav Neurosci. 119(6):1682–1686. doi: Murakami T, Yoshida T, Matsui T, Ohki K. 2015. Wide-field Ca2+ 10.1037/0735-7044.119.6.1682. imaging reveals visually evoked activity in the retrosple- Vann SD, Aggleton JP, Maguire EA. 2009. What does the ret- nial area. Front Mol Neurosci. 08(June):1–12. doi: 10.3389/fn- rosplenial cortex do? Nat Rev Neurosci. 10(11):792–802. doi: mol.2015.00020. 10.1038/nrn2733. Niell CM, Stryker MP. 2010. Modulation of visual responses Wolbers T, Büchel C. 2005. Dissociable Retrosplenial and by Behavioral state in mouse visual cortex. Neuron. hippocampal contributions to successful formation of 65(4):472–479. doi: 10.1016/j.neuron.2010.01.033. survey representations. J Neurosci. 25(13):3333–3340. doi: Pachitariu M, Stringer C, Schröder S, Dipoppa M, Rossi LF, 10.1523/JNEUROSCI.4705-04.2005. Carandini M, Harris KD. 2016. Suite2p: beyond 10,000 neu- Wood JN. 2009. Distinct visual working memory systems for rons with standard two-photon microscopy. BioRxiv. doi: view-dependent and view-invariant representation. PLoS One. 10.1101/061507. 4(8):e6601. doi: 10.1371/journal.pone.0006601. Polack P-O, Friedman J, Golshani P. 2013. Cellular mechanisms of Wyss JM, Van Groen T. 1992. Connections between the brain state-dependent gain modulation in visual cortex. Nat retrosplenial cortex and the hippocampal formation Neurosci. 9:1331–1339. doi: 10.1038/nn.3464. in the rat: a review. Hippocampus. 2(1):1–11. doi: Pothuizen HHJ, Aggleton JP, Vann SD. 2008. Do rats with 10.1002/hipo.450020102. retrosplenial cortex lesions lack direction? Eur J Neurosci. Yamawaki N, Radulovic J, Shepherd GMG. 2016. A cortico- 28:2486–2498. doi: 10.1111/j.1460-9568.2008.06550.x. cortical circuit directly links retrosplenial cortex to M2 in Pothuizen HHJ, Davies M, Albasser MM, Aggleton JP, Vann the mouse. J Neurosci. 36(36):9365–9374. doi: 10.1523/JNEU- SD. 2009. Granular and dysgranular retrosplenial cortices ROSCI.1099-16.2016. provide qualitatively different contributions to spatial Zhuang J, Ng L, Williams D, Valley M, Li Y, Garrett M, Waters working memory: evidence from immediate-early J. 2017. An extended retinotopic map of mouse cortex. elife. gene imaging in rats. Eur J Neurosci. 30(5):877–888. doi: 6:1–29. doi: 10.7554/eLife.18372. 10.1111/j.1460-9568.2009.06881.x. Ziv Y, Burns LD, Cocker ED, Hamel EO, Ghosh KK, Kitch LJ, El Ranson A. 2017. Stability and plasticity of contextual modula- Gamal A, Schnitzer MJ. 2013. Long-term dynamics of CA1 tion in the mouse visual cortex. Cell Rep. 18(4):840–848. doi: hippocampal place codes. Nat Neurosci. 16(3):264–266. doi: 10.1016/j.celrep.2016.12.080. 10.1038/nn.3329.