The rapid developmental rise of somatic inhibition disengages hippocampal dynamics from self-motion

  1. Robin F Dard
  2. Erwan Leprince
  3. Julien Denis
  4. Shrisha Rao Balappa
  5. Dmitrii Suchkov
  6. Richard Boyce
  7. Catherine Lopez
  8. Marie Giorgi-Kurz
  9. Tom Szwagier
  10. Théo Dumont
  11. Hervé Rouault
  12. Marat Minlebaev
  13. Agnès Baude
  14. Rosa Cossart  Is a corresponding author
  15. Michel A Picardo  Is a corresponding author
  1. Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, France
  2. Turing Centre for Living systems, Aix-Marseille University, Université de Toulon, CNRS, CPT (UMR 7332), France
  3. Mines ParisTech, PSL Research University, France

Abstract

Early electrophysiological brain oscillations recorded in preterm babies and newborn rodents are initially mostly driven by bottom-up sensorimotor activity and only later can detach from external inputs. This is a hallmark of most developing brain areas, including the hippocampus, which, in the adult brain, functions in integrating external inputs onto internal dynamics. Such developmental disengagement from external inputs is likely a fundamental step for the proper development of cognitive internal models. Despite its importance, the developmental timeline and circuit basis for this disengagement remain unknown. To address this issue, we have investigated the daily evolution of CA1 dynamics and underlying circuits during the first two postnatal weeks of mouse development using two-photon calcium imaging in non-anesthetized pups. We show that the first postnatal week ends with an abrupt shift in the representation of self-motion in CA1. Indeed, most CA1 pyramidal cells switch from activated to inhibited by self-generated movements at the end of the first postnatal week, whereas the majority of GABAergic neurons remain positively modulated throughout this period. This rapid switch occurs within 2 days and follows the rapid anatomical and functional surge of local somatic GABAergic innervation. The observed change in dynamics is consistent with a two-population model undergoing a strengthening of inhibition. We propose that this abrupt developmental transition inaugurates the emergence of internal hippocampal dynamics.

Editor's evaluation

This study investigates hippocampal dynamics over the course of early postnatal development with respect to spontaneous movements. Pioneering in vivo imaging in the hippocampus of neonatal mice, the authors find evidence for an abrupt developmental transition in this neural activity at the end of the first postnatal week in rodents and contributes to understanding how cognitive functions could emerge from the immature brain.

https://doi.org/10.7554/eLife.78116.sa0

Introduction

The adult hippocampus serves multiple cognitive functions, including navigation and memory. These functions rely on the ability of hippocampal circuits to integrate external inputs conveying multisensory, proprioceptive, contextual, and emotional information onto internally generated dynamics. Therefore, the capacity to produce internally coordinated neuronal activity detached from environmental inputs is central to the cognitive functions of the hippocampus such as planning and memory (Buzsáki, 2015; Buzsáki and Moser, 2013). In contrast to the adult situation, the developing hippocampus, like many developing cortical structures, is mainly driven by bottom-up external environmental and body-derived signals, including motor twitches generated in the spinal cord and/or the brainstem (Dooley et al., 2020; Inácio et al., 2016; Karlsson et al., 2006; Mohns and Blumberg, 2010; Del Rio-Bermudez et al., 2020; Valeeva et al., 2019a). These produce early sharp waves (eSW) conveyed by inputs from the entorhinal cortex (Valeeva et al., 2019a). The emergence of self-organized sequences without reliance on external cues in the form of sharp wave ripples (SWRs) is only observed after the end of the second postnatal week and sequential reactivations even a week later (Farooq and Dragoi, 2019; Muessig et al., 2019). Therefore, early hippocampal activity as measured with electrophysiological recordings is first externally driven while the emergence of internal dynamics is protracted. The timing and the circuit mechanisms of the switch between motion-guided and internally produced hippocampal dynamics remain unknown. They have been proposed to rely on the maturation of CA3 and extrinsic hippocampal inputs; however, a possible role of local connectivity, in particular, recurrent somatic inhibition, cannot be excluded (Cossart and Khazipov, 2022).

Local GABAergic interneurons could be critically involved in this phenomenon for several reasons. First, both theoretical and experimental work suggest that self-organized internal neuronal network dynamics require feedback connections to produce an emergent state of activity independently from the incoming input (Hopfield and Tank, 2005; Hopfield, 1982; Yuste, 2015). Feedback circuits are mainly GABAergic in CA1 (but not necessarily inhibitory), given the scarcity of recurrent glutamatergic connections in that hippocampal subregion (Bezaire and Soltesz, 2013). Second, GABAergic interneurons, in particular, the perisomatic subtypes, are long known to shape the spatial and temporal organization of internal CA1 dynamics (Buzsáki, 2015; Lee et al., 2014; Soltesz and Losonczy, 2018; Valero et al., 2015). However, GABAergic perisomatic cells display a delayed maturation profile both at structural (Jiang et al., 2001; Marty et al., 2002; Morozov and Freund, 2003; Tyzio et al., 1999) and functional levels (Ben-Ari, 2002; Doischer et al., 2008; Jiang et al., 2001; Khazipov et al., 2004; Marty et al., 2002; Morozov and Freund, 2003; Murata and Colonnese, 2020; Tyzio et al., 1999), and the precise developmental timeline for their postnatal development remains unknown, partly due to the difficulty in labeling them (Donato et al., 2017).

Here, we investigate the evolution of CA1 dynamics during the first and second postnatal weeks of mouse development with an eye on the specific patterning of activity of CA1 GABAergic neurons. To this aim, we adapted two-photon calcium imaging of CA1 dynamics using virally expressed GCaMP6 through a cranial window in non-anesthetized pups. We show that the first postnatal week ends with an abrupt switch in the representation of self-motion in CA1: principal neurons were synchronized by spontaneous movement before P9, whereas self-motion decreased their activity after that time point. Consistent with a two-population neuronal model, this switch was locally paralleled by the rapid anatomical and functional surge of somatic GABAergic interneurons and no significant change in external inputs. Self-generated bottom-up inputs may thus directly contribute to the emergence of somatic GABAergic inhibition and in this way calibrate local circuits to the magnitude of external inputs prior to the opening of experience-dependent plasticity.

Results

Progressive evolution of CA1 neuronal dynamics

In order to induce stable and early expression of the calcium indicator protein GCaMP6s, pups were injected with the AAV1-hSyn-GCaMP6s.WPRE.SV40 virus in the brain lateral ventricle on the day of birth (P0, Figure 1A, Figure 1—figure supplement 1A). Five to twelve days after injection, the hippocampal CA1 region of non-anesthetized pups was imaged through a window implant placed on the same day (Figure 1A, see ‘Materials and methods’). We first quantified the amount of sleep/wake cycle in P5–6 mice after cranial window surgery and electromyogram (EMG) nuchal electrodes implantation. We found that mice spent 74% (±6) of their time in active sleep (Figure 1—figure supplement 1B), which is comparable to previous reports (Jouvet-Mounier et al., 1970). This indicates that the window implant did not alter this characteristic of behavior in early postnatal stages. In the same way, the acute window implant did not significantly alter electrophysiological network patterns. These were measured using in vivo bilateral silicon probes recordings of eSW (Figure 1—figure supplement 1C, see ‘Discussion’) in P6–8 (n = 4) and P11 (n = 2) pups expressing GCaMP6s with a frequency of 2.6 eSW/min (25% 1.15 eSW/min and 75% 4.16 eSW/min) for the ipsilateral side and 3.49 eSW/min (25% 1.96 eSW/min and 75% 5.1 eSW/min, p-value=0.39) for the contralateral side (Figure 1—figure supplement 1C). This slight but nonsignificant reduction in eSW frequency recorded from the ipsilateral hemisphere was similarly reported in a previous study using the same surgical approach (see ‘Discussion’, Graf et al., 2021). eSW synchronization between hemispheres was preserved (Valeeva et al., 2019b) but with a 12 ms delay between the two hemispheres (peak at 0.087 ± 0.027 s, Figure 1—figure supplement 1C), possibly explained by a slight drop in local temperature due to the chamber placement as described previously (Reig et al., 2010). Finally, we checked for the presence of other types of oscillations in both hemispheres and observed a peak in the theta range in the P11 mouse pups in both hemispheres (ipsi: peak at 4.3 Hz of amplitude 3.6 × 10–3 ± 1.2 × 10–3 mV2/Hz; contra: peak at 4.1 Hz of amplitude 2 × 10–3 ± 3 × 10–4 mV2/Hz (jackknife standard deviation), Figure 1—figure supplement 1C). In general, a slight increase in the peak power of most electrophysiological network oscillations (below 20 Hz) was observed (Figure 1—figure supplement 1C). Altogether, we can conclude that the presence of the window implant minimally disrupted the electrophysiological network patterns and sleep–wake cycle of developing rodents during that early postnatal period. Thus, we pursued the description of early multineuron CA1 dynamics using calcium imaging (62 imaging sessions, 35 mouse pups aged between 5 and 12 days, yielding a total of 33,412 cells, see Supplementary file 1 for details of each session and their inclusion in the figures).

Figure 1 with 1 supplement see all
Evolution of CA1 dynamics during the first two postnatal weeks.

(A) Schematic of the experimental timeline. On postnatal day 0 (P0), 2 µL of a nondiluted viral solution was injected into the left lateral ventricle of mouse pups. From 5 to 12 days after injection (P5–P12), acute surgery for window implantation above the corpus callosum was performed and followed by two-photon calcium imaging recordings. Top panel: four example recordings are shown to illustrate the imaging fields of view in the stratum pyramidale of the CA1 region of the hippocampus (scale bar: 100 µm). Middle panel: contour maps showing the cells detected using Suite2p in the corresponding fields of view. Bottom panel: raster plots inferred by DeepCINAC activity classifier, showing 300 randomly selected cells over the first 5 min of recording obtained for these imaging sessions (P5, P7, P10, and P12, see full raster plots for these imaging sessions in Figure 1—figure supplement 1D). In the raster plot from the P5 mouse, the blue rectangle illustrates one synchronous calcium event (SCE). Scale bar for time is 1 min. Calcium Imaging Complete Automated Data Analysis (CICADA) configuration files to reproduce example rater plots and cell contours are available in Figure 1—source data 1. (B) Evolution of the ratio of calcium transients within SCEs over the total number of transients across age. Each dot represents a mouse pup and is color coded from light gray (P5) to black (P12), the open blue circles represent the median of the age group. The red line represents the linear fit of the data with r2 = 0.78, p<0.0001 (N = 32 pups). Results to build the distribution, as well as CICADA configuration file to reproduce the analysis, are available in Figure 1—source data 1. (C) Evolution of the number of transients per minute across the first two postnatal weeks. Each dot represents the mean transient frequency from all cells imaged in one animal and is color coded from light gray (P5) to black (P12). The red line represents the nonlinear fit (fourth-order polynomial, least-squares method) of the data with r2 = 0.30 (N = 32 pups). The open blue circles represent the median of the age group. Results to build the distribution, as well as CICADA configuration file to reproduce the analysis, are available in Figure 1—source data 1.

Figure 1—source data 1

Analysis configuration files and numerical data used in Figure 1.

(A) Field of views and the Calcium Imaging Complete Automated Data Analysis (CICADA) configuration files necessary to plot the contours map and raster plots used for the illustration. (B) Numerical data used to plot the evolution of the transient in synchronous calcium event (SCE) and the CICADA configuration file necessary to reproduce the analysis. (C) Numerical data used to plot the evolution of the transient per minute and the CICADA configuration file necessary to reproduce the analysis.

https://cdn.elifesciences.org/articles/78116/elife-78116-fig1-data1-v3.zip

The contours of the imaged neurons and their calcium fluorescence events were extracted using Suite2P (Pachitariu et al., 2017) and DeepCINAC (Denis et al., 2020), respectively. Representative examples of fields of view, contour maps, and activity raster plots from recordings in P5, P7, P10, and P12 mouse pups are shown in Figure 1A. Neuronal activity was stable over the duration of the recording (Figure 1—figure supplement 1D and E – median change: 0.08 transients/minute, N = 31). Consistent with previous electrophysiological studies (Leinekugel et al., 2002; Mohns and Blumberg, 2008; Valeeva et al., 2019a), spontaneous neuronal activity in the CA1 region of P5–6 pups alternated between recurring population bursts (synchronous calcium events [SCEs]) and periods of low activity (Figure 1A). In P5–6 mouse pups, more than half of the detected calcium transients occurred within SCEs (P5: median value 55% N = 4, n = 8; P6: median value 55% N = 3, n = 5; N: mice, n: imaging sessions, Figure 1B). Activity then became progressively continuous as evidenced by the linear decrease in the proportion of calcium transients occurring during SCEs (r2 = 0.78, p<0.0001) to finally reach 33% in P12 mouse pups (P12: N = 5, n = 8, Figure 1B). Reminiscent of a transient period of ‘neural quiescence’ at the beginning of the second postnatal week (Domínguez et al., 2021), we observed a nonlinear evolution in the cell activation frequency with a local minimum around P10 (r2 = 0.30, Figure 1C). We conclude that CA1 dynamics progressively evolve from discontinuous to continuous during the first two postnatal weeks, in agreement with previous electrophysiological studies (Cossart and Khazipov, 2022; Mohns and Blumberg, 2008; Valeeva et al., 2019a).

Early SCEs correlate with spontaneous motor activity

Previous extracellular electrophysiological recordings indicated that, in developing rodents, CA1 dynamics followed spontaneous motor activity during the first postnatal week (Karlsson et al., 2006; Del Rio-Bermudez et al., 2020; Valeeva et al., 2019a). Hence, we next examined the relationship between population activity and movement as monitored using either piezo recordings or infrared cameras (see ‘Materials and methods’). Because our surgical procedure could potentially affect CA1 dynamics in response to contralateral movement, we computed peri-movement time histograms (PMTHs) by plotting the fraction of active neurons centered on the onset of all ipsi- or contralateral limbs movements. Both spontaneous limb movements were followed by an increase in CA1 activity (peak ipsi = 3.6%, peak contra = 2.8%, chance level 3.4%, Figure 1—figure supplement 1E), showing that the surgery was not preventing the hippocampal response to contralateral limb movements. Still, contralateral limb movements recruited a slightly lower fraction of active cells (see ‘Discussion’). In mouse pups younger than P9, movements were followed by a significant increase in the percentage of active cells exceeding the chance level (Figure 2A; P5–8 median above chance level, Videos 13) and an increase in the average DF/F fluorescence signal (Figure 2—figure supplement 1A). In contrast, after P9, movements were followed by a significant decrease in activity below chance level (Figure 2A, P10–12 median below chance level, Videos 46) and a decreased DF/F fluorescence signal (Figure 2—figure supplement 1A). Short myoclonic movements such as twitches, happening during periods of active sleep (Gramsbergen et al., 1970; Jouvet-Mounier and Astic, 1968; Karlsson et al., 2006) as opposed to longer movements, happening mostly during wakefulness, may induce different activity patterns in the hippocampus (Mohns and Blumberg, 2008). This difference between wake movements and active-sleep twitches during development is proposed to rely on a gating of sensory feedback associated with movement during wake (Dooley and Blumberg, 2018; Tiriac and Blumberg, 2016). Accordingly, when combining calcium imaging with nuchal EMG recordings in one P5 mouse pup, we observed an increase in the percentage of active cells and in the DF/F fluorescence signal following movements occurring both during REM sleep and wakefulness (Figure 2—figure supplement 1C). However, when combining all mouse pups, and considering separately twitches (occurring during REM/active sleep) and complex movements (occurring during wakefulness), based on video recordings, we found that the two movement types did not significantly differ in their impact on CA1 activity (Figure 2—figure supplement 1B). Given this lack of difference, all movement types were thus combined in the following analysis steps. Post-movement activity was next computed, as defined by the number of active cells in the 2 s following movement onset divided by the number of active cells within a 4-s-long time window centered on movement onset (see ‘Materials and methods,’ Figure 2B, and Figure 2—figure supplement 1D). The median post-movement activity progressively decreased from P5 to P9 (mean difference between consecutive age groups of 3.1 ± 0.7%) until it suddenly dropped at P10 (–13.5% between P9 and P10) and stabilized until P12. P9 marked the transition in the relationship between movement and CA1 activity. Indeed, the median post-movement activity exceeded 50% from P5 to P8 (P5: 71%; P6: 65%; P7: 60%; P8: 56%). This is consistent with the evolution of PMTHs (Figure 2A). After P9, the median post-movement activity was lower than 50% (P10: 39%; P11: 35%; P12: 40%), thus revealing the inhibitory action of movement on activity. We next defined as ‘inhibiting movements’ all the movements with a post-movement activity lower than 40% and computed their proportion in each mouse (Figure 2C). The proportion of ‘inhibiting’ movements was stable before P9 (P5: 11%; P6: 16%; P7: 10%, P8: 15%). Again, P9 marks a transition since we observed that approximately half of the movements were followed by an inhibition of CA1 activity in P10–12 mice (P10: 55%; P11: 58%; P12: 48%). The proportion of ‘inhibiting’ movements varies with age as a sigmoid function with P9 being the transition time point (V50 = 9.015, r2 = 0.75). In line with the emergence of movement-induced inhibition, the fraction of neurons significantly associated with immobility also increased with age, also following a sigmoidal function (Figure 2D, sigmoid fit V50 = 9.022, r2 = 0.55). Altogether, these results indicate that the end of the first postnatal week marks a transition in the evolution of CA1 dynamics, with both a decorrelation and a ‘detachment’ of neuronal activity from spontaneous motor activity. We next investigated the circuit mechanisms supporting these changes.

Figure 2 with 1 supplement see all
Linking CA1 dynamics to movement during the first two postnatal weeks.

(A) Peri-movement time histograms (PMTH) representing the percentage of active cells centered on the onset of the mouse movements. The dark line indicates the median value, and the two thick gray lines represent the 25th and 75th percentiles from the distribution made of all median PMTHs from the sessions included in the group. Overall are included: P5: N = 4, n = 8; P6: N = 3, n = 5; P7: N = 5, n = 12; P8: N = 5, n = 8; P9: N = 5, n = 11; P10: N = 1, n = 1; P11: N = 1, n = 1; P12: N = 5, n = 7 (N, number of mice; n, number of imaging sessions). In all panels, the thin straight gray lines represent the 5th percentile, the median, and the 95th percentile of the distribution made of all median PMTHs resulting from surrogate raster plots from the sessions included in the group. Black asterisk indicate that the median value is above the 95th percentile or below the 5th percentile from the surrogates. Results to build the PMTH, as well as Calcium Imaging Complete Automated Data Analysis (CICADA) configuration file to reproduce the analysis, are available in Figure 2—source data 1. (B) Distribution of post-movement activity across age. Each box plot is built from all detected movements for the given age group. Whiskers represent the 5th and 95th percentiles with post-movement activity falling above or below represented as small dots. The average post-movement activity observed for each mouse pup is represented by the large dots color coded from light gray (P5) to black (P12). The red area illustrates the movement falling in the category of ‘inhibiting’ movements. P5: four mice, 1519 movements; P6: three mice, 766 movements; P7: five mice, 2067 movements; P8: five mice, 1105 movements; P9: five mice, 1272 movements; P10: one mouse, 83 movements; P11: one mouse, 57 movements; P12: three mice, 493 movements. Global effect of age was found significant (ANOVA, eight groups, F = 107.7, p-value<0.0001). Comparison between age groups shows that except all three possible pairs made of P10–P11–P12 and the P8–P9 pair, all pairs were significantly different (p-value<0.005, post hoc Bonferroni’s multiple-comparison test). Results to build the distributions, as well as Calcium Imaging Complete Automated Data Analysis (CICADA) configuration file to reproduce the analysis, are available in Figure 2—source data 1. (C) Distribution of the proportion of ‘inhibiting’ movements across age. Each dot represents a mouse pup and is color coded from light gray (P5) to black (P12). The open blue circles represent the median of the age group. The red line shows a sigmoidal fit with V50 = 9.015, r2 = 0.75 (least-squares method). Results to build the distribution, as well as CICADA configuration file to reproduce the analysis, are available in Figure 2—source data 1. (D) Distribution of the proportion of significantly immobility-associated cells as a function of age. Each dot represents a mouse and is color coded from light gray (P5) to black (P12). The open blue circles represent the median of the age group. The red line shows a sigmoidal fit with V50 = 9.022, r2 = 0.55 (least-squares method). Results to build the distribution, as well as CICADA configuration file to reproduce the analysis, are available in Figure 2—source data 1.

Figure 2—source data 1

Analysis configuration files and numerical data used in Figure 2.

(A) Numerical data used to plot all the peri-movement time histograms (PMTHs) (in Figure 2A) and the Calcium Imaging Complete Automated Data Analysis (CICADA) configuration file necessary to reproduce the analysis. (B) Numerical data used to plot Figure 2B and the CICADA configuration file necessary to reproduce the analysis. (C) Numerical data used to plot Figure 2C and the CICADA configuration file necessary to reproduce the analysis. (D) Numerical data used to plot Figure 2D and the CICADA configuration file necessary to reproduce the analysis.

https://cdn.elifesciences.org/articles/78116/elife-78116-fig2-data1-v3.zip
Video 1
First example of calcium imaging movies from P5 mouse pups centered on the onset of a twitch.

The twitch is indicated by T in the upper-left corner of the movie. Imaging 2× speed up.

Video 2
Second example of calcium imaging movies from P5 mouse pups centered on the onset of a twitch.

The twitch is indicated by T in the upper-left corner of the movie. Imaging 2× speed up.

Video 3
Third example of calcium imaging movies from P5 mouse pups centered on the onset of a twitch.

The twitch is indicated by T in the upper-left corner of the movie. Imaging 2× speed up.

Video 4
First example of calcium imaging movies from P12 mouse pups centered on the onset of a complex movement.

The complex movement is indicated by M in the upper-left corner of the movie. Imaging 2× speed up.

Video 5
Second example of calcium imaging movies from P12 mouse pups centered on the onset of a complex movement.

The complex movement is indicated by M in the upper-left corner of the movie. Imaging 2× speed up.

Video 6
Third example of calcium imaging movies from P12 mouse pups centered on the onset of a complex movement.

The complex movement is indicated by M in the upper-left corner of the movie. Imaging 2× speed up.

GABAergic neurons remain activated by spontaneous movement throughout the first two postnatal weeks

As a first step to identifying the circuit mechanisms for this switch, we focused on local circuits and disentangled the respective contribution of local GABAergic neurons and principal cells to CA1 dynamics as well as their relation to movement. To this aim, we identified GABAergic neurons with the expression of a red reporter (tdTomato) in GAD1Cre/+ pups virally infected with AAV9-FLEX-CAG-tdTomato and AAV1.hSyn.GCaMP6s (Figure 3A and B, Figure 3—figure supplement 1A). In addition, we used these imaging experiments (Figure 3—figure supplement 1B, top row) to train a cell classifier inferring interneurons in the absence of any reporter (Figure 3—figure supplement 1B, bottom row). This classifier was able to infer interneurons with 91% precision (Figure 3—figure supplement 1C; Denis et al., 2020). ‘Labeled’ and ‘inferred’ GABAergic neurons were combined into a single group referred to as ‘interneurons’' in the following (Figure 3—figure supplement 1D and E). As illustrated in a representative raster plot from a P5 mouse, both pyramidal cells (black) and interneurons (red) were activated during movement (vertical gray lines, Figure 3A). This was confirmed when computing the PMTH for pups aged less than P9, with the activation of the two neuronal populations after movement exceeding chance level (P5–8: N = 17, n = 33, pyramidal cells: baseline value = 0.51%, peak value = 2.1%, interneurons: baseline value = 2.1%, peak value 7.9%, N: number of mice, n: number of imaging sessions, Figure 3A, Figure 3—figure supplement 2A). In line with the above results (Figure 2A), pups older than P9 showed a significant reduction (below chance level) in the proportion of active pyramidal cells following movement (Figure 3B P10–12: N = 7, n = 9, baseline value = 1.3%, trough value = 0.4%). In contrast, interneurons remained significantly activated following movement even past P9 (P10–12: N = 7, n = 9, baseline value = 3.9%, peak value = 10%, Figure 3B, Figure 3—figure supplement 2B). We conclude that the link between movement and activity evolves differentially toward the start of the second postnatal week when comparing pyramidal neurons and GABAergic interneurons, the former being inhibited or detached from movements while the latter remaining activated. This suggests that pyramidal neurons could be directly inhibited by local interneurons after the first postnatal week, following a functional maturation of GABAergic outputs onto principal cells. Alternatively, this could result from differential changes in the synaptic inputs driving both cell types. In the following, we have addressed both, nonmutually exclusive, hypotheses.

Figure 3 with 3 supplements see all
Differential recruitment of CA1 glutamatergic and GABAergic neurons.

(A). Top panel: imaged field of view and associated raster plot from an example imaging session in the stratum pyramidale from one P5 Gad1Cre/+ mouse pup (scale bar = 100 µm). Imaged neurons expressed GCaMP6s. Interneurons were identified by the Cre-dependent expression of the red reporter tdTomato. In the raster plot, neurons are sorted according to their identification as pyramidal cells (black) or interneurons (red), vertical gray lines indicate movements of the mouse. Scale bar: 60 s. Bottom panel: peri-movement time histograms (PMTHs) for pyramidal cells and interneurons combining all imaging sessions from mice aged between P5 and P8 (N = 17 mice, n = 33 imaging sessions). The dark line indicates the median value, and the thick gray lines represent the 25th and 75th percentiles from the distribution made of all median PMTH obtained from the sessions included in the group. Thin gray lines represent the 5th, median, and 95th from the distribution made of all median PMTH obtained from surrogate raster plots from the sessions included in the group. Black asterisk indicate that the median value is above the 95th percentile or below the 5th percentile from the surrogate dataset (B). Same as (A), but illustration is made with one P12 Gad1Cre/+ mouse pup and PMTHs are built with all imaging sessions from pups aged between P10 and P12 (N = 7 mice, n = 9 imaging sessions). Note the presence of red labeled processes in the neuropil of the stratum pyramidale of P12 in contrast to P5. Results to build the PMTHs, as well as Calcium Imaging Complete Automated Data Analysis (CICADA) configuration files to reproduce the analysis, are available in Figure 3—source data 1.

Figure 3—source data 1

Analysis configuration files and numerical data used in Figure 3.

‘example_FoVs’: images used for illustration in Figure 3. ‘example_raster_plots’: two Calcium Imaging Complete Automated Data Analysis (CICADA) configuration files necessary to reproduce the raster plots used for illustration in Figure 3. ‘psths’: numerical data used to plot Figure 3 peri-movement time histograms (PMTHs) and the CICADA configuration file necessary to reproduce the analysis.

https://cdn.elifesciences.org/articles/78116/elife-78116-fig3-data1-v3.zip

We first compared the developmental time course of extra-hippocampal synaptic afferences onto CA1 GABAergic neurons and pyramidal cells using a rabies retrograde tracing method (Wickersham et al., 2007). We focused on changes that may occur around the end of the first postnatal week. To do so, two groups were compared, an early (AAV1-hSyn-FLEX-nGToG-WPRE3 – helper virus – injected at P0; SAD-B19-RVdG-mCherry – pseudotyped-defective rabies virus – at P5; and immunohistochemistry [IHC] at P9, Figure 3—figure supplement 3A–C) and a late one (AAV1-hSyn-FLEX-nGToG-WPRE3 – helper virus – injected at P0; SAD-B19-RVdG-mCherry – pseudotyped defective rabies virus – at P9; IHC at P13, Figure 3—figure supplement 3D–F). Injections were performed in either GAD1Cre/+ or EmxCre/+ pups in order to specifically target GABAergic or glutamatergic cells, respectively. Four GAD1Cre/+ pups (two early and two late injections) and three EmxCre/+ pups (one early and two late) were analyzed with injection sites restricted to the hippocampus. Starter and retrogradely labeled cells were found all over the ipsilateral hippocampus. For both GAD1Cre/+ and EmxCre/+ pups, we found no striking difference in the retrogradely labeled extra-hippocampal regions between the early and late groups. In agreement with previous studies (Supèr and Soriano, 1994), we found that GABAergic and glutamatergic neurons in the dorsal hippocampus received mainly external inputs from the entorhinal cortex, medial septum, and contralateral CA3 area (retrogradely labeled cells in these regions were found in four out of four GAD1Cre/+ pups and three out of three EmxCre/+ pups, Figure 3—figure supplement 3). Thus, we could not reveal any major switch in the nature of the extra-hippocampal inputs impinging onto local CA1 neurons. Thus, we next explored the maturation of local somatic GABAergic innervation given its significant evolution throughout that period (Jiang et al., 2001; Marty et al., 2002; Morozov and Freund, 2003) as well as our observation of a dense tdTomato signal in the pyramidal layer from GAD1Cre/+ mouse pups at P12 (Figure 3B), not visible at P5 (Figure 3A).

Abrupt emergence of a functional somatic GABAergic innervation at the beginning of the second postnatal week

We first analyzed the anatomical development of somatic GABAergic innervation within the CA1 pyramidal layer from P3 to P11, focusing on the innervation from putative parvalbumin-expressing basket cells (PVBCs), its main contributor. To this aim, we performed immunohistochemistry against Synaptotagmin2 (Syt2), which has been described as a reliable marker for parvalbumin-positive inhibitory boutons in cortical areas (Figure 4A, Sommeijer and Levelt, 2012). Using a custom-made Fiji plugin (RINGO, see ‘Materials and methods’), we quantified the surface of the pyramidal cell layer covered by Syt2 labeling at different stages and found that between P3 and P7, PV innervation remained stable (median values: P3: 0.34%; P5: 0.57%; P7: 0.49%, three mice per group, Figure 4B). However, after P9, an increase in the density of positive labeling was observed (P9: 1.03%; P11: 1.48%, three mice per group, Figure 4B). These results are consistent with previous work (Jiang et al., 2001; Marty et al., 2002), as well as with our tdTomato labeling (Figure 3A and B) and GCaMP imaging (Figure 4—figure supplement 1, Video 7). They also match the transition observed in CA1 dynamics (Figure 3). We next tested whether GABAergic axons in the pyramidal layer were active during periods of movement. We restricted the analysis of these experiments to P9–10 as axonal arborization innervating the CA1 pyramidal layer was not present before (i.e., there was no fluorescent signal before P9 in the stratum pyramidale, Figure 4—figure supplement 1, Video 7). To do so, we restrained the expression of the calcium indicator GCaMP6s to the axon (Broussard et al., 2018) of interneurons using GAD1Cre/+ mouse pups and specifically imaged axonal arborization in the pyramidal cell layer (Figure 4C, left panel). Fluorescence signals were extracted from axonal branches using PyAmnesia (a method to segment axons, see ‘Materials and methods,’ Figure 4C, right panel), and then normalized using DF/F. As expected (see Figure 3B), an increase in the fluorescent signal from GABAergic axonal branches was observed following movement (P9–10: n = 3 mice, Figure 4D). As a result, we reasoned that the emergence of functional perisomatic GABAergic activity could contribute to the reduction in activity observed after movement during the second postnatal week in pyramidal neurons.

Figure 4 with 1 supplement see all
Emergence of perisomatic GABAergic innervation.

(A) Representative example confocal images of the CA1 region in a P7 (left) and P11 (right) mouse pup. DAPI staining was used to delineate the stratum pyramidale (sp) from the stratum radiatum (sr) and stratum oriens (so, top row). Synaptotagmin-2 labeling (Syt2) is shown in the top and bottom rows. Illustrated examples are indicated by red dots in the associated quantification in (B). (A) Scale bar = 50 µm. (B) Fraction of the pyramidal cell layer covered by Syt2-positive labeling as a function of age. Each gray dot represents the average percentage of coverage from two images taken in the CA1 region of a hippocampal slice. Open black dots are the average values across brain slices from one mouse pup. Blue arrows indicate the slices used for illustration in (A). A significant effect of age was detected (one-way ANOVA, F = 13.11, p=0.0005, three mice per age group). Multiple-comparison test shows a significant difference between age groups (Bonferroni’s test, *p<0.05, **p<0.01, ***p<0.001). (C) Averaged image of a field of view in the pyramidal cell layer of a P9 Gad1Cre/+ mouse pup injected with a Cre-dependent Axon-GCaMP6s indicator (left) and the segmented image resulting from PyAmnesia (right). (C) Scale bar = 50 µm. (D) Peri-movement time histogram (PMTH) showing the DF/F signal centered on the onsets of animal movement (N = 3 mice, n = 3 imaging sessions). The dark gray line indicates the median value from the surrogate. Results obtained from surrogates are represented by light gray lines. Black asterisk indicate that the median value is above the median from the surrogate.

Figure 4—source data 1

Analysis configuration files and numerical data used in Figure 4.

(A) Images used for illustration in Figure 4A. (B) Numerical data of the plot in Figure 4B. (C) Calcium Imaging Complete Automated Data Analysis (CICADA) configuration file necessary to plot the contour map shown in Figure 4C. (D) Numerical data used to plot Figure 4C peri-movement time histogram (PMTH) and the CICADA configuration file necessary to reproduce the analysis.

https://cdn.elifesciences.org/articles/78116/elife-78116-fig4-data1-v3.zip
Video 7
Calcium imaging movie from the field of view (FOV) shown in the middle panel of Figure 4—figure supplement 1B.

Increasing feedback inhibition in two-population models explains the developmental transition

To test whether an increase in perisomatic inhibition alone can explain the switch in network dynamics between the first and second postnatal weeks, we simulated a two-population network model mimicking the development of perisomatic innervation (Figure 5A, see ‘Materials and methods’). Using a rate model and a leaky integrate and fire (LIF) model, we show that increasing the number of perisomatic inhibitory connections can account for the experimentally observed decrease in responses to movement-like feedforward inputs (Figure 5B). Time constants of the rate model and synaptic time constants of the LIF model were chosen to match the slow kinetics of synaptic transmission that exist at early developmental stages (see Figure 5—source data 1). Faster excitatory and inhibitory timescales, on the order of a few milliseconds, generate network dynamics that could not be followed by our calcium sensor. We chose to model them by simply providing a noisy, normally distributed, input to all the cells. Durations of feedforward inputs were chosen similar to experimental movement durations (see Figure 5—figure supplement 1A for a log-normal fit of the movement durations). When inhibition is weak, the average activity of the pyramidal neurons increases at the onset of a given twitch. Then, it quickly relaxes to the baseline with a timescale that follows the synaptic time constant (Figure 5B, left panel). In the presence of strong inhibition, there is a reduction in response to movement inputs. In addition, due to strong feedback inhibition following the movement responses, network activity relaxes to the baseline with an undershoot (Figure 5B, right panel), recapitulating the experimental findings (see Figure 2A). Similarly, PMTHs for interneurons were obtained (see Figure 5—figure supplement 1B).

Figure 5 with 1 supplement see all
Modeling the effects of perisomatic inhibition on pyramidal cell response.

(A) The model consists of two populations, Excitatory (E) and Inhibitory (I) receiving feedforward input Iext. The interaction strengths, Jab, represent the effect of the activity of population b on a. We studied the effects of perisomatic inhibition on the activity of pyramidal cells by varying the parameter JEI (i.e., the number of I to E connections). For the rate model, the overall scale of the rates is arbitrary. For the leaky integrate and fire (LIF) model, the parameters were tuned to match the percentages of active cells per experimental bin (100 ms). (B) Peri-movement time histogram (PMTH) response of excitatory neurons to pulse input in the rate model and LIF network. (C) Cross-correlations during periods of immobility, in experimental data (left), rate model (middle), and LIF network (right). In the rate model, dotted lines are the predicted correlation from the analytic expressions and solid lines are the results from numerical integration. In all the simulations, the signals were convolved with an exponential kernel of characteristic time 2 s to account for GCamp6s decay time.

The inhibition of principal cells following movements after P9, revealed by the PMTHs, could result from the development of a direct external inhibitory input or from changes in local circuits. In order to further examine the nature of the interaction between local pyramidal cells and interneurons displayed a direct interaction in the absence of the external movement-related input, we next used our experimental data to compute the cross-correlograms between interneurons and pyramidal cells calcium transients (inferred by DeepCINAC [Denis et al., 2020]) in the periods without movement (P5–8 light gray curve, P9–12 dark curve, Figure 5C, left panel). For P9–12, we observed a rapid drop in the correlation at positive time points, suggesting a feedforward inhibition of principal cells’ activity. Such drop was absent for P5–8, where the cross-correlogram was symmetric and centered at zero. Both the rate and LIF models displayed similar activity correlograms as the experimental data, with an undershoot in the presence of strong inhibitory feedback for positive time points (Figure 5C, middle and right panels). Thus, increasing JEI has the effect of strengthening the cross-correlation undershoot. JEI for both models were chosen to match the amplitude of this undershoot while at the same time matching the inhibition observed in the PMTHs. Auto-correlograms of the excitatory and inhibitory activity were also measured and compared to our model predictions (Figure 5—figure supplement 1C).

The consistency of our models with the experimental cross-correlograms, which were computed from the activity recorded during periods of immobility, further shows that the observed network dynamics and, in particular, the correlation undershoots most likely result from recurrent perisomatic inhibition rather than a feedforward drive from upstream areas. Therefore, in our model, the maturation of perisomatic inhibition alone was sufficient to support a switch in network dynamics.

Discussion

Using for the first time in vivo two-photon calcium imaging in the hippocampus of non-anesthetized mouse pups and a deep-learning-based approach to infer the activity of principal cells and interneurons, we show that the end of the first postnatal week marks a salient step in the anatomical and functional development of the CA1 region. Indeed, within 2 days (P8–10), the link between CA1 principal cells’ activity and self-triggered movements is inverted and neurons are preferentially active during immobility periods. This is likely due to the time-locked anatomical and functional rise of somatic GABAergic activity, given that interneurons remain highly active throughout this period, including in response to spontaneous movements. In this way, CA1 circuits start detaching from external inputs. Given the importance of local dynamics for hippocampal function and cortical circuits operation in general, this is likely to be a critical general step in the proper maturation of cognitive circuits.

Early postnatal calcium activity in CA1 is driven by sensorimotor inputs

We found that, until P7–9, spontaneous movements are followed by a significant peak in calcium events in the CA1 principal cell layer and that most neuronal activity occurs during SCEs. This early link between sensorimotor inputs and early cortical dynamics has been previously reported using electrophysiological recordings in various areas and species, including humans (Milh et al., 2006). Here, we extend that observation to calcium transients, which not only indirectly report action-potential firing as well as other modes of cell activation during development but also critically regulate activity-dependent genetic processes. In addition, we could describe the response to these movements with single-cell resolution. Of note, it is important to keep in mind that part of the overlying cortex, including the primary sensory cortex, was removed to grant optical access to the hippocampus. This region may contribute to relaying the sensory feedback from the twitches to the hippocampus (Khazipov and Milh, 2018; Valeeva et al., 2019a). The surgical procedure may thus damage incoming axons from the temporoammonic track linking the entorhinal cortex to the hippocampus. Accordingly, we observed (1) that the CA1 response to movements from the contralateral limbs was slightly reduced, (2) the eSW frequency in the ipsilateral hemisphere was slightly diminished, and (3) there was a small increase in the power spectra of network oscillations below 20 Hz. In addition, it cannot be excluded that other movements that we have not detected, like whisker movements, could also contribute to the patterning of CA1 activity. It is also possible that self-generated activity from other sensory organs but independent from movement, like the retina or the olfactory bulb, also contributes to hippocampal dynamics. Interestingly, in contrast to previous reports (Tiriac and Blumberg, 2016), we could not observe any significant difference between twitches (occurring mainly during active sleep) and longer, more complex movements (associated with wakefulness). In one mouse pup, we directly combined two-photon imaging with EMG recordings to better define brain states and confirmed that both REM and wake-associated movements were followed by an activation of CA1 neurons, with the latter triggering a smaller response. This difference from previous reports (Tiriac and Blumberg, 2016) may reveal a difference between calcium imaging and electrophysiology, the former sampling from a larger population but at a lower temporal and spike signal resolution. The patterning of CA1 dynamics in the large imaged population did not reveal any obvious spatial distribution for movement-activated cells but we cannot exclude that these would vary along the radial and transverse directions, which are the two main axes of principal cell development (Caviness, 1973), and are differentially targeted by perisomatic PVBCs (Lee et al., 2014; Valero et al., 2015).

Passed the end of the first postnatal week, between P8 and P10, a significant decrease in the fraction of coactive principal cells following movement was observed (while interneurons remained mostly activated by movement). We cannot exclude that some spikes fell below the threshold for the detection of calcium events. In this case, rather than full inhibition, it may be that a strong shortening of the time window for neuronal integration occurred (due to feedback inhibition), which would limit the number of spikes produced by principal cells and thus keep them below detection levels. Yet, a novel machine learning-based algorithm (Denis et al., 2020) was used since it was especially designed to infer activity in the dense CA1 pyramidal cell layer. This change in the polarity of principal cells’ response to movements is quite abrupt as it happens within less than 2 days (between P8 and 10). This contrasts with the progressive evolution of single-cell-firing frequencies but matches the fast redistribution of neuronal firing towards immobility periods. In this way, hippocampal neuronal dynamics ‘internalize’ as they stop being driven by movements and preferentially occur within rest. Of note, a small increase in the fraction of active cells can be observed approximately 1 s before the onset of movement in P12 pups, indicating that activity would start building up in CA1 prior to movement. A corollary discharge would increase activity prior to movement on a much shorter timescale. There is therefore no obvious explanation for this interesting phenomenon. Anticipatory cell firing prior to locomotion has been previously reported in the adult cortex (e.g., see Vinck et al., 2015). Different mechanisms could support such anticipatory firing, including the influence of top-down inputs, changes in arousal states, or any complex neuromodulatory interactions possibly associated with changes in the sleep–wake cycle and that could involve, for example, the norepinephrine, serotonin, or acetylcholine systems.

This ‘internalization’ of hippocampal dynamics is reminiscent of similar phenomena observed in other cortical areas, such as the barrel cortex where whisker stimulation induces a reduction in the size of cell assemblies following P9 while the same stimulation widens cell assembly size a few days before (Modol et al., 2019). It is also reminiscent of the recently described transient quiescent period observed in the somatosensory cortex using extracellular electrophysiological recordings (Domínguez et al., 2021). Lastly, it goes in hand with a sparsification of activity, which is a general developmental process supported by the emergence of inhibition (Golshani et al., 2009; Rochefort et al., 2009; Wolfe et al., 2010).

Circuit basis for the movement-triggered inhibition of CA1 dynamics

Our results demonstrate that the change in CA1 dynamics occurring at the end of the first postnatal week most likely relies on structural changes in local CA1 circuits rather than rewiring of the long-range extra-hippocampal connectivity.

The long-range circuits mediating the bottom-up flow of self-triggered or externally generated sensory information to the hippocampus are starting to be elucidated. The two main structures directly transmitting sensorimotor information to the dorsal CA1 are the entorhinal cortex and septum. The former processes multisensory information from all sensory cortices (visual, auditory, olfactory, somatosensory), including movement-related sensory feedback (Rio-Bermudez and Blumberg, 2022), and was shown to be activated by spontaneous twitches prior to CA1 (Mohns and Blumberg, 2010; Del Rio-Bermudez et al., 2020; Valeeva et al., 2019a) while the latter is more likely to be involved in transmitting internal information (Fuhrmann et al., 2015; Wang et al., 2015), as well as unexpected environmental stimuli (Zhang et al., 2018). In addition to these two canonical pathways, one cannot exclude the involvement of a direct connection from the brainstem, given their existence in the adult and their role in promoting sleep as well as motor twitches (Liu et al., 2017; Szőnyi et al., 2019). However, our retrograde-tracing experiments did not reveal any direct connection between the CA1 cells and the brainstem at the early ages analyzed here. In addition, we found that both CA1 interneurons and principal cells receive inputs from the septum and entorhinal cortex before the time of the switch (i.e. P9) and that there was no major qualitative change of inputs after, as expected from previous work (Supèr and Soriano, 1994). Still, these experiments do not allow a quantitative assessment of the number of inputs nor the type of inputs (GABAergic, cholinergic, etc.), and we cannot fully exclude that a stronger or different source of excitatory drive would be impinging onto interneurons after the switch. Neither can we exclude a functional maturation of those extrinsic inputs. Therefore, future optogenetic and slice physiology work is needed to characterize the bottom-up information flow onto specific components of the local CA1 circuits. Similarly, one cannot exclude a change in the CA3 to CA1 connectivity. Indeed, Schaffer collaterals are known to reach CA1 roughly around the end of the first postnatal week (Durand et al., 1996). In addition, roughly at the time of the switch, we do see the emergence of SWRs (Buhl and Buzsáki, 2005), a pattern strongly relying on CA3 inputs and perisomatic GABAergic transmission. However, we could not restrict the pool of starter cells to the CA1 region in our retrograde viral-tracing experiments, which precluded analysis of the development of CA3–CA1 connectivity. Interestingly, among the external inputs onto CA1 described above, the entorhinal cortex and CA3 were both shown to exert a mild influence on the organization of intrinsic CA1 dynamics, possibly pointing at a critical role of local interneurons in this process (Zutshi et al., 2021).

As indicated by our computational model, the disengagement from movement of CA1 dynamics can be fully explained by the observed rise in anatomical (Syt2 labeling) and functional (axonal GCaMP imaging) connectivity from perisomatic GABAergic cells onto pyramidal cells at the onset of the second postnatal week. This increased connectivity could not be easily captured with our retrograde viral labeling since the absence of early PV expression precludes the identification of PVBCs, the most prominent subtype of perisomatic GABAergic cells, among retrogradely labeled cells in EmxCre/+ pups. Early anatomical studies had already indicated that an increase in somatic GABAergic inhibition, including from CCK-basket cells, occurred in CA1 during the first postnatal week (Gour et al., 2021; Danglot et al., 2006; Jiang et al., 2001; Marty et al., 2002; Morozov and Freund, 2003). However, this rise was expected to be more progressive and not as abrupt as observed here as it happened within 2 days. If the axonal coverage of the stratum pyramidale by PV-basket cells axons increases, we cannot exclude that this is a general phenomenon, concerning all perisomatic subtypes, including soma-targeting CCK-expressing basket cells that develop anatomically at around the same time (Morozov and Freund, 2003) or chandelier cells. In addition, our computational model indicates that the emergence of feedback inhibition is sufficient to reproduce the developmental shift observed here, which could also involve other types of CA1 interneurons, including dendrite-targeting ones.

Interestingly, a similar rise of somatic GABAergic axonal coverage occurs in the barrel cortex at the same time. Indeed, recent connectomic mapping using 3D electron microscopy in that region revealed that the preferential targeting of cell bodies by GABAergic synapses increased almost threefold between postnatal days 7 and 9 (Gour et al., 2021), whereas two-photon imaging of putative GABAergic somatic axons in the same region revealed broader domains of co-activation (Modol et al., 2019). This time period for the shift may be synchronous within brain regions involved in sensorimotor integration such as the hippocampus and somatosensory cortex. Otherwise, PV expression was shown to develop sequentially in a region-specific manner Reh et al., 2020 following their intrinsic developmental age (Donato et al., 2017).

We found that many principal cells are inhibited by movement while most imaged GABAergic cells remained activated during the second postnatal week. This therefore indirectly suggests a net inhibitory effect of GABAergic transmission after the first postnatal week. This is expected since the shift from excitatory to inhibitory synaptic transmission was reported to occur earlier in the hippocampus in vivo (Murata and Colonnese, 2020). On a side note, the lack of somatic GABAergic inputs before P7 indicates that the early excitatory GABAergic drive in CA1 circuits likely originates from nonsomatic GABAergic interneurons, which include long-range, dendrite-targeting or interneuron-specific interneurons. The circuit role of excitatory GABAergic transmission should be revisited taking into account this new finding.

The movement-associated inhibition could result equally from feedforward (direct activation of interneurons from movement-transmitting inputs such as the entorhinal cortex) or feedback (from local CA1 cells) inhibition. Our analysis of pairwise correlations in the absence of movement, as well as our computational simulations, indicates the latter. A similar strengthening of feedback inhibition has previously been observed in the developing somatosensory cortex (Anastasiades and Butt, 2012). The inhibition of activity following movement is likely to be occurring during a transient developmental period. Indeed, in the adult, both interneurons and principal cells usually increase their activity as the animal moves (Fuhrmann et al., 2015). Therefore, the switch observed here opens another developmental time window that probably closes with the emergence of perineuronal nets and cell activation sequences at the end of the third developmental week (Farooq and Dragoi, 2019; Horii-Hayashi et al., 2015; Muessig et al., 2019). We would like to propose this developmental window to be the critical period for CA1 development, a period during which experience-dependent plasticity can be observed.

Conclusion

Cognitive hippocampal maps rely on two forms of representation, one that is map-based or allocentric and the other that is self-referenced, or egocentric and requires body movement. We would like to propose that the early postnatal period described here, where the hippocampus learns the statistics of the body, and which terminates with the rise of a recurrent inhibitory network, is a key step for the emergence of self-referenced representations onto which exploration of the external world can be grafted. An imbalance between self-referenced and environmental hippocampal representations due to a miswiring of local somatic inhibition could have major outcomes. It could be on the basis of several neurodevelopmental disorders, including autism spectrum disorders (ASDs) and schizophrenia. Interestingly, both disorders have been associated with an aberrant maturation of PV-expressing interneurons (Gogolla et al., 2014; Jurgensen and Castillo, 2015; Lewis et al., 2005). In addition, the proper development of the peripheral sensory system, which is partly initiating the early CA1 dynamics reported in our study, is also critically involved in ASD (Orefice et al., 2016). The period described here corresponds to the third trimester of gestation and likely extends postnatally given the protracted integration of GABAergic interneurons into functional circuits in the human brain (Murphy et al., 2005; Paredes et al., 2016). Future work should determine when a similar rise in somatic inhibition occurs in human infants and test whether it could constitute a valuable biomarker for cognitive neurodevelopmental disorders.

Materials and methods

Mice

All experiments were performed under the guidelines of the French National Ethics Committee for Sciences and Health report on ‘Ethical Principles for Animal Experimentation’ in agreement with the European Community Directive 86/609/EEC (Apafis #18-185 and #30-959).

Experimental procedures and data acquisition

Viruses

In vivo calcium imaging experiments were performed using AAV1-hSyn-GCaMP6s.WPRE.SV40 (pAAV.Syn.GCaMP6s.WPRE.SV40 was a gift from Douglas Kim & GENIE Project [Addgene viral prep# 100843-AAV1; http://n2t.net/addgene:100843; RRID:Addgene_100843]), AAV9-FLEX-CAG-tdTomato (pAAV-FLEX-tdTomato was a gift from Edward Boyden [Addgene viral prep# 28306-AAV9; http://n2t.net/addgene:28306; RRID:Addgene_28306]), AAV9-hSyn-FLEX-axon-GCaMP6s (pAAV-hSynapsin1-FLEx-axon-GCaMP6s was a gift from Lin Tian [Addgene viral prep# 112010-AAV9; http://n2t.net/addgene:112010; RRID:Addgene_112010]). Retrograde-tracing experiments were performed using AAV1-hSyn-FLEX-nGToG-WPRE3 (Charité# BA-096) and SAD-B19-RVdG-mCherry (gift from the Conzelmann laboratory).

Intracerebroventricular injection

Request a detailed protocol

This injection protocol was adapted from already published methods (Rübel et al., 2021Kim et al., 2014). Mouse pups were anesthetized on ice for 3–4 min, and 2 µL of viral solution (titration at least 1 × 1013 vg/mL) was injected into the left lateral ventricle whose coordinates were estimated at the 2/5 of the imaginary line between the lambda and the eye at a depth of 0.4 mm. Correct injection was visualized by the spreading of the viral-dye mixture (1/20 of fast blue). In SWISS mouse pups. we injected 2 µL of AAV2.1-hSyn-GCAMP6s.WPRE.SV40; in GAD1Cre/+ mouse pups we injected either a mix of 1.3 µL of AAV2.1-hSyn-GCAMP6s.WPRE.SV40 with 0.7 µL of AAV9-FLEX-CAG-tdTomato or 2 µL of AAV9-hSyn-FLEX-axon-GCaMP6s.

Intra-hippocampal injection

Request a detailed protocol

When hippocampal viral injections were performed at P0 (AAV1-hSyn-FLEX-nGToG-WPRE3), mouse pups were anesthetized by inducing hypothermia on ice and maintained on a dry ice-cooled stereotaxic adaptor (Stoelting, #51615) with a digital display console (Kopf, model 940). Dorsal hippocampus was targeted by empirically determined coordinates, based on the Atlas of the Developing Mouse Brain (Paxinos and Watson, 2020), using transverse sinus and superior sagittal sinus as reference: 0.8 mm anterior from the sinus intersection; 1.5 mm lateral from the sagittal sinus; 1.1 mm depth from the skull surface. Under aseptic conditions, an incision was made in the skin, the skull was exposed, and gently drilled (Ball Mill, Carbide, #¼ 0.019″ –0.500 mm diameter, CircuitMedic). Then, 10 nL of undiluted viral solution was injected using an oil-based pressure injection system (Nanoject III, Drummond Scientific, rate of 5 nL/min). The tip of the pipette was broken to achieve an opening with an internal diameter of 30–40 μm. When hippocampal viral injections were performed at P5 or P9 (SAD-B19-RVdG-mCherry), pups were anesthetized using 3% isoflurane in a mix of 90% O2–10 % air and maintained during the whole surgery (~0:30 hr) between 1 and 2.5% isoflurane. Body temperature was monitored and maintained at 36°C. Analgesia was controlled using buprenorphine (0.05 mg/kg). Under aseptic conditions, an incision was made in the skin, the skull was exposed, and anteroposterior and mediolateral coordinates of the dorsal hippocampus were estimated by eye looking at the skull sutures. The skull was gently drilled and 10 nL of a viral solution was injected (Nanoject III, Drummond Scientific, rate of 5 nL/min) at a depth of 1.25 mm below the dura.

Window implant surgery

Request a detailed protocol

The surgery to implant a 3-mm-large cranial window above corpus callosum was adapted from previous methods (Villette et al., 2015). Anesthesia was induced using 3% isoflurane in a mix of 90% O2–10% air and maintained during the whole surgery (~1:30 hr) between 1 and 2.5% isoflurane. Body temperature was controlled and maintained at 36°C. Analgesia was controlled using buprenorphine (0.05 mg/kg). Coordinates of the window implant were visually estimated. Then a small custom-made headplate was affixed with cyanoacrylate and dental acrylic cement. The skull was removed and the cortex was gently aspirated until the appearance of the external capsule/alveus. At the end of the cortectomy, we sealed a 3 mm glass window diameter circular cover glass (#1 thickness, Warner Instrument) attached to a 3-mm-diameter large and 1.2-mm-height cannula (Microgroup INC) with Kwik-Sil adhesive (WPI) and fixed the edge of the glass with cyanoacrylate. We let the animal recover on a heated pad for at least 1 hr before the imaging experiment.

Imaging

Request a detailed protocol

Two-photon calcium imaging experiments were performed on the day of the window implant using a single-beam multiphoton pulsed laser scanning system coupled to a microscope (TriM Scope II, LaVision Biotech). The Ti:sapphire excitation laser (Chameleon Ultra II, Coherent) was operated at 920 nm. GCaMP fluorescence was isolated using a bandpass filter (510/25). Images were acquired through a GaAsP PMT (H7422-40, Hamamatsu) using a ×16 immersion objective (NIKON, NA 0.8). Using Imspector software (LaVision Biotech), the fluorescence activity from a 400 μm × 400 µm field of view was acquired at approximately 9 Hz with a 1.85 μs dwell time per pixel (2 μm/pixel). Imaging fields were selected to sample the dorsal CA1 area and maximize the number of imaged neurons in the stratum pyramidale. Piezo signal, camera exposure time, and image triggers were synchronously acquired and digitized using a 1440A Digidata (Axon Instrument, 50 kHz sampling) and the AxoScope 10 software (Axon Instrument). During the imaging session, body temperature is continuously controlled.

Behavioral recordings

Request a detailed protocol

Simultaneously with imaging experiments, mouse motor behavior was monitored. In a first group of animals, motor behavior was monitored using two or three piezos attached to the paws of the animal. The signal from the piezo was acquired and digitized using a 1440A Digidata and the AxoScope 10 software. In a second group of animals, pups were placed and secured on an elevated platform (with the limbs hanging down on each side without touching the ground nor the support, as described here; Blumberg et al., 2015). Motor behavior was monitored using two infrared cameras (Basler, acA1920-155um) positioned on each side of the animal. For each camera, a square signal corresponding to the exposure time of each frame from the camera was acquired and digitized using a 1440A Digidata and the AxoScope 10 software. If the number of behavior frames from the square signal was higher than the number of saved frames (meaning that some camera frames were dropped during the acquisition), the imaging session was excluded from any movement related analysis (see Supplementary file 1).

Recording of EMG activity in neonatal mice

Request a detailed protocol

The vigilance state of neonatal mice was assessed through analysis of EMG signals obtained from a single insulated tungsten wire (A-M Systems 795500) implanted in the nuchal muscle. A stainless steel wire (A-M Systems 786000) wire inserted on the skull surface above the cerebellum and secured in place with dental cement served as the reference electrode. Signals from the electrodes were first passed through a headstage pre-amplifier before being digitized at 16,000 Hz (Digital Lynx SX, Neuralynx [the pre-amplifier and digitizer were both from Neuralynx, as was the acquisition software, Cheetah]) and saved to a hard disk. TTL signals from the imaging and camera acquisition systems were simultaneously recorded as well to enable precise synchronization of EMG recordings with the camera and imaging data.

In vivo extracellular electrophysiological recordings

Request a detailed protocol

Multisite probes (16-channel silicon probes with 50 µm separation distance, NeuroNexus, USA) were used to record electrophysiological activity below the window implant and in the intact hippocampus. To do so, we positioned the mouse pup (that had previously undergone a window implant) on the experimental setup. To head-fix the animal, the skull surface was covered with a layer of dental acrylic except the area above the intact hippocampus. In the intact (contralateral) hippocampus, the electrodes were positioned using the stereotaxic coordinates of approximately 1.5 mm anterior to lambda and 1.5 mm lateral from the midline. Hippocampus under the window was recorded through the hole drilled in the window implant. Both multisite silicon probes were positioned at the depth to record strata oriens (SO), pyramidale (SP), radiatum (SR), and lacunosum moleculare (SLM). After the positioning of the electrodes, the animal was left in the setup for 1 hr to recover followed by 2 hr recordings of the neuronal activity in both hippocampi simultaneously.

Histological processing

Request a detailed protocol

Pups were deeply anesthetized with a mix of Domitor and Zoletil (0.9 and 60 mg/kg, respectively), then transcardially perfused with 4% paraformaldehyde (PFA) in 0.1 M phosphate-buffered saline (PBS) (PBS tablets, 18912-014, Life Technologies). For perisomatic innervation analysis, brains were post-fixed overnight at 4°C in 4% PFA in 0.1 M PBS, washed in PBS, cryo-protected in 30% sucrose in PBS, before liquid nitrogen freezing. Brains were then sectioned using a cryostat (CM 3050S, Leica) into 50-µm-thick slices collected on slides. Sections were stored at –20°C until further usage. For tracing experiments, brains from GAD1Cre/+ and EmxCre/+ pups (The Jackson Laboratory JAX:005628), were post-fixed overnight at 4°C in 4% PFA in 0.1 M PBS, washed in PBS, and sectioned using a vibratome (VT 1200s, Leica) into sagittal 70–80-μm-thick slices. Sections were stored in 0.1 M PBS containing 0.05% sodium azide until further usage. Immunocytochemistry was processed as described previously (Bocchio et al., 2020). Briefly, sections were blocked with PBS-Triton (PBST) 0.3 and 10% normal donkey serum (NDS), and incubated with a mix of up to three primary antibodies simultaneously diluted in PBST with 1% NDS overnight at room temperature with the following primary antibodies: rabbit anti-dsRed (1:1000; Clontech, AB_10013483), chicken anti-GFP (1:1000, Aves Labs, GFP-1020, AB_10000240), and mouse anti-synaptotagmin-2 (1:100; Developmental Studies Hybridoma Bank, AB_2315626). After several washes, according to the mixture of primary antibodies, the following secondary antibodies were used: donkey anti-chicken Alexa 488 (1:500, SA1-72000), donkey anti-rabbit Alexa 555 (1:500, Thermo Fisher, A31570), donkey anti-mouse Alexa 488 (1:500, Thermo Fisher, A21202), and donkey anti-mouse Alexa 647 (1:500, Thermo Fisher, A31571). After Hoechst counterstaining, slices were mounted in Fluoromount. Epifluorescence images were obtained with a Zeiss AxioImager Z2 microscope coupled to a camera (Zeiss AxioCam MR3) with an HBO lamp associated with 470/40, 525/50, 545/25, and 605/70 filter cubes. Confocal images were acquired with a Zeiss LSM-800 system equipped with a tunable laser providing excitation range from 405 to 670 nm. For quantifying synaptotagmin-2, 11-µm-thick stacks were taken (z = 1 µm, pixel size = 0,156 µm) with the confocal microscope using a Plan-Achromat ×40/1.4 oil DIC objective.

Data preprocessing

Motion correction

Request a detailed protocol

Image series were motion corrected either by finding the center of mass of the correlations across frames relative to a set of reference frames (Miri et al., 2011) or using the NoRMCorre algorithm available in the CaImAn toolbox (Pnevmatikakis and Giovannucci, 2017), or both.

Cell segmentation

Request a detailed protocol

Cell segmentation was achieved using Suite2p (Pachitariu et al., 2017). Neurons with pixel masks including processes (often the case for interneurons located in the stratum oriens) were replaced by soma ROI manually drawn in ImageJ and matched onto Suite2p contours map using Calcium Imaging Complete Automated Data Analysis (CICADA; source code available on Cossart lab GitLab group ID: 5948056). In experiments performed on GAD1Cre/+ animals, tdTomato-labeled interneurons were manually selected in ImageJ and either matched onto Suite2p contours map or added to the mask list using CICADA.

Axon segmentation

Request a detailed protocol

Axon segmentation was performed using pyAMNESIA (a Python pipeline for analyzing the Activity and Morphology of NEurons using Skeletonization and other Image Analysis techniques; source code available on Cossart lab GitLab group ID: 5948056). pyAMNESIA proposes a novel image processing method based on three consecutive steps: (1) extracting the axonal morphology of the image (skeletonization), (2) discarding the detected morphological entities that are not functional ones (branch validation), and (3) grouping together branches with highly correlated activity (branch clustering). To extract the skeleton, we first performed 3D Gaussian smoothing and averaging of the recording, producing an image that summarizes it; on this image are then successively applied a histogram equalization, a Gaussian smoothing, an adaptive thresholding, and finally a Lee skeletonization (Suen et al., 1994), allowing for the extraction of the skeleton mask and the morphological branches. To ensure the functional unity of the segmented branches, we only kept those that illuminated uniformly, where uniformity was quantified by the skewness of the pixel distribution of the branch during a calcium transient (branch validation). To cluster the valid branches based on their activity, we first extracted their average trace – being the average image intensity along the branch for each frame – and then clustered the branches traces using t-SNE and HDBSCAN algorithms with Spearman’s correlation metric (branch clustering).

Cell-type prediction

Request a detailed protocol

Cell-type prediction was done using the DeepCINAC cell-type classifier (Denis et al., 2020). We used imaging sessions from GAD1Cre/+ mouse line injected with a viral mixture of AAV2.1-hSyn-GCAMP6s.WPRE.SV40 and AAV9-FLEX-CAG-tdTomato allowing us to manually identify in our recordings genetically labeled interneurons to train and test a cell-type classifier. Overall, we used 643 cells (245 labeled interneurons, 245 putative pyramidal cells, and 153 noisy cells) to train the cell-type classifier and 100 cells (38 labeled interneurons, 51 putative pyramidal cells, and 11 noisy cells) to evaluate its performance. Briefly, a neuronal network composed of a convolutional neural network (CNN) and long short-term memory (LSTM) was trained using labeled interneurons, pyramidal cells, and noisy cells to predict the cell type using 100-frame-long movie patches centered on the cell of interest. Each cell was classified as interneuron, pyramidal cell, or noise. Cells classified as ‘noisy cells’ were removed from further analysis. ‘Labeled interneurons’ were first kept in a separate cell-type category and added to the interneurons list that were inferred with a 91% precision.

Activity inference

Request a detailed protocol

Activity inference was done using DeepCINAC classifiers (Denis et al., 2020). Briefly, a classifier composed of CNN and LSTM was trained using manually labeled movie patches to predict neuronal activation based on movie visual inspection. Depending on the inferred cell type, activity inference was done using either a general classifier or an interneuron-specific classifier. Activity inference resulted in a (cells × frames) matrix giving the probability for a cell to be active at any single frame. We used a 0.5 threshold in this probability matrix to obtain a binary activity matrix considering a neuron as active from the onset to the peak of a calcium transient.

Behavior

Request a detailed protocol

Piezo signals were manually analyzed in a custom-made graphical user interface (Python Tkinter) to label the onset and offset of ‘twitches,’ ‘complex movements,’ and ‘unclassified movements.’ Twitches were defined as brief movements (a few hundred milliseconds long) occurring within periods of rest and detected as rapid deflections of the piezo signal. ‘Complex movements’ were defined as periods of movement lasting at least 2 s. A few other detected movements could not be categorized based on their duration and occurrence as twitches or complex movements. These are referred to as ‘unclassified movements.’ These unclassified movements were excluded from the analysis when it is specified that complex movements or twitches only were used but included in analysis when all kinds of movements were combined. Analysis of video tracking was done using CICADA, and behavior was manually annotated in the BADASS (Behavioral Analysis Data And Some Surprises) GUI. If camera frames were dropped during the acquisition, the imaging session was excluded from any movement-related analysis.

Neurodata without border (NWB:N) embedding

Request a detailed protocol

For each imaging session, imaging data, behavioral data, cell contours, cell-type prediction, calcium traces, and neuronal activity inference were combined into a single NWB:N file (Rübel et al., 2021). Our NWB:N data set is accessible on DANDI archive (https://gui.dandiarchive.org/#/) – ref DANDI:000219. NWB offers a common format for sharing, among others, calcium imaging data and analyzing them. Subsequently, we developed an open-source Python toolbox to analyze imaging data in the NWB format.

Modeling

Network implementation

Request a detailed protocol

We constructed a simple rate model and subsequently a more realistic spiking network in order to test our hypothesis that an increase in perisomatic inhibition could explain the switch in network dynamics between the first and second postnatal weeks. Both models consisted of one excitatory and one inhibitory population with recurrent interactions (Figure 5A, Appendix 1). The development of perisomatic innervation was simulated by increasing the strength from inhibitory to excitatory cells (JEI). The external input to the model was composed of a constant and a white noise term. To estimate the responses to twitch-like inputs, an additional feedforward input composed of short pulses was fed to the network. In the rate model, the rates represented the population averaged activities. The spiking network was constructed with LIF neurons. The network connectivity was sparse and each neuron received inputs from randomly selected neurons. Presynaptic spikes resulted in exponentially decaying postsynaptic currents (see Figure 5—source data 1 for the model parameter values). All codes used for the modeling are available at https://gitlab.com/rouault-team-public/somatic-inhibition/ (copy archived at swh:1:rev:98b243e4bd38be6bc9addbe86bc750942cb89e21, Hervé, 2022) project ID: 33964849.

Data analysis

Sample size estimation

Request a detailed protocol

This study being mainly exploratory in the sense that the evolution of population activity in the CA1 region of the hippocampus during early development using large-scale imaging has not been described before, we have not been able to use explicit power calculation based on an expected size effect.

Histological quantifications

Request a detailed protocol

Confocal images of synaptotagmin-2 immunostaining were analyzed using RINGO (RINGs Observation), a custom-made macro in Fiji. We first performed a max-intensity projection of the z-stack images of the top 6 µm from the slice surface, then images were cropped to restrict the analysis to the pyramidal cells layer. Obtained images were denoised using Fiji ‘remove background’ option and then by subtracting the mean intensity of the pixels within a manually drawn ROI in the background area (typically the cell body of a pyramidal neuron). Denoised images were then binarized using a max-entropy thresholding (Fiji option). Finally, particles with size between 0.4 µm² and 4 µm² were automatically detected using the Fiji ‘Analyse particle’ option. We then computed the proportion of the pyramidal cell layer (i.e., surface of the cropped region) covered by positive synaptotagmin-2 labeling.

In vivo electrophysiology

Request a detailed protocol

The neuronal activity recorded from both hippocampi in vivo using a 64-channel amplifier (DIPSI, France) was analyzed post hoc. Firstly, data was downsampled to 1 kHz to save disc space. The local field potential (LFP) was band-passed (2–100 Hz) using the wavelet filter (Morlet, mother wavelet of order 6), and the common reference was subtracted to exclude the bias produced by volume conducted fluctuations of LFP. Sharp wave events (SWs) were detected using a threshold approach. Firstly, LFP was band-passed (2–45 Hz) and the difference between LFP recorded in the strati oriens, pyramidale, and radiatum was calculated. Events were considered as SWs if (1) LFP reversion was observed in the stratum pyramidale and (2) their peak amplitude in the resulting trace exceeded the threshold of 4 SDs calculated over the entire trace (the threshold corresponds to p-values<0.01). The occurrence rate of SW was calculated over the entire recording and normalized to 1 min. SW co-occurrence was also calculated by cross-correlating the SW timestamps from ipsilateral and contralateral hippocampi using a bin size of 10 ms. Spectral analysis was carried out using the Chronux toolbox. Spectral power was estimated using direct multi-taper estimators (three time-bandwidth products and five tapers).

Statistics for in vivo electrophysiology

Request a detailed protocol

Group comparisons were done using nonparametric Wilcoxon rank-sum test for equal medians, and p-value of 0.05 was considered significant. Variability of the estimates was visualized as shaded bands of standard deviation computed using jackknife.

Vigilance state determination in neonatal mice

Request a detailed protocol

All analyses of EMG data were completed using custom scripts in MATLAB (Cossart lab GitLab, group ID 5948056, project ID 36204799). For each experiment, the raw EMG data was first downsampled to 1000 Hz and subsequently high-pass-filtered at 300 Hz and rectified. The processed data was then plotted to allow for manual inspection. Consistent with prior reports (Mohns and Blumberg, 2010), the data was primarily composed of alternating periods of high EMG tone (referred to as wakefulness) associated with ‘complex’ movements as well as periods of low EMG tone associated with a general behavioral quiescence and the presence of periodic brief myoclonic twitches (referred to as ‘active sleep’ due to the frequent observation of muscle twitches; Mohns and Blumberg, 2010). For vigilance state determination, we therefore utilized a protocol similar to that described previously (Del Rio-Bermudez et al., 2015). For both the ‘high’ and ‘low’ EMG tone conditions, five periods, each 1 s in duration, were first sampled from locations spread out over the entire recording length. Data from the samples were then pooled for each condition and the average value of the rectified signal was determined. Next, the midpoint between the average rectified signal values calculated for the ‘high’ and ‘low’ EMG tone conditions was determined for subsequent use as a threshold to separate periods of non-wakefulness (below the midpoint threshold value) from periods of wakefulness (above the midpoint threshold value), while the quarter point between these two values was calculated to further separate periods of non-wakefulness into active sleep (below the quarter point threshold value) or a sleep–wake transitory state (above the quarter point threshold value but below the midpoint threshold value). Once these thresholds were determined, the entire length of data was divided into 1 s nonoverlapping bins and the average filtered rectified EMG signal was determined for each. A hypnogram was then created by automatically applying the threshold-derived criteria to the binned averaged data. Data bins scored as being active sleep were further analyzed to determine the presence of muscle twitches; this was accomplished by automatically identifying data points with values exceeding the mean +5× standard deviation value determined from the low EMG tone representative samples. As a final step, the hypnogram and filtered rectified EMG signal data were plotted and manually inspected to ensure the accuracy of results. The hypnogram was then incorporated in the final NWB:N file to serve in the definition of the epochs of wakefulness and active sleep.

Analysis of calcium imaging data in the NWB format using CICADA

Request a detailed protocol

Analysis was performed using CICADA (Cossart lab GitLab, group ID 5948056, project ID 14048984), a custom-made open-source Python toolbox allowing for the automatic analysis of calcium imaging data in the NWB:N format. CICADA offers a user-friendly graphical user interface allowing the user (1) to select the NWB files of the recording sessions to include in a given analysis, (2) select the analysis to run and set up the parameters, and (3) generate result tables and/or ready to use figures. In addition, each analysis run in CICADA generates a configuration file that can be loaded in CICADA with the option ‘Load a set of parameters’ allowing for the replication of the analysis. CICADA can be installed following the installation guidelines presented at https://gitlab.com/cossartlab/cicada (copy archived at swh:1:rev:2ef0c25d7da5b69849c663ed56a0033cfe8488ca; Dard, 2022).

Calcium transient frequency analysis

Request a detailed protocol

Analysis launched from CICADA ‘Transient’s frequency’ analysis. The transient frequency for each cell was computed using the count of calcium transient onsets divided by the duration of the recording and was then averaged across all cells imaged in one given mouse pup across one or more imaging sessions.

SCE detection

Request a detailed protocol

Analysis launched from CICADA ‘SCE description’ analysis. SCEs were defined as the imaging frames within which the number of co-active cells exceeded the chance level as estimated using a reshuffling method. Briefly, an independent circular shift was applied to each cell to obtain 300 surrogate raster plots. We computed the 99th percentile of the distribution of the number of co-active cells from these surrogates and used this value as a threshold to define the minimal number of co-active cells in an SCE. Peak of synchrony above this threshold separated by at least five imaging frames (500 ms) was defined as SCE frames. To compute the percentage of transients within SCEs, we counted, for each cell, the number of its calcium transients (from onset to peak) crossing SCE frames and divided it by its total number of calcium transients. We averaged the obtained values over all the cells imaged per animal.

Peri-movement time histograms (PMTH)

Request a detailed protocol

Analysis launched from CICADA ‘Population-level PSTH’ analysis. A 20-s-long time window centered on movement onset was used. For each movement within an imaging session, the number of cells activated or the sum of all cells’ fluorescence was calculated for each time bin in that 20-s-long window. We obtain as many values as movements per time bin; for each individual imaging session, the 25th, median, and 75th percentiles of the distributions of these values per time bin are computed and divided by the number of imaged cells. To display the percentage of active cells at a given time bin, these values were multiplied by 100. To combine imaging sessions in an age group (i.e., P5, 6, 7, 8, 9, 10, 11, 12), all the median PMTHs from individual imaging sessions belonging to the given group were stacked and we represented at each time bin the 25th percentile, median, and 75th percentile value of these median PMTHs. To evaluate chance level around movement onsets, 500 surrogate raster plots per imaging session were computed, and the above procedure was used to obtain chance level in each imaging session and then grouped. To obtain each surrogate raster plot, the activity of each imaged cell was translated by a randomly selected integer (between 1 and the total number of frames). We used the 95th percentile of the surrogate PMTH to conclude significant activation and the 5th percentile to conclude significant activity reduction. PMTHs obtained from fluorescence signals were built from DF/F calcium traces.

Movement-related inhibition

Request a detailed protocol

Analysis launched from CICADA ‘Activity ratio around epochs’ analysis. A 4-s-long window centered on the onset of movements was used. The total number of cells activated during this time period was calculated. If less than 40% of these cells were activated within 2 s following movement onset, the movement was classified as an ‘inhibiting’ movement. This procedure was applied to all detected movements to obtain for each mouse pup the proportion of ‘inhibiting’ movements.

Movement- and immobility-associated cells

Request a detailed protocol

Analysis launched from CICADA ‘Epoch-associated cells’ analysis. The number of transients per cell occurring during movement or immobility was calculated. These transient onsets were then circularly shifted 100 times and the same calculation was performed on each roll. We used the 99th percentile of this distribution as a threshold above which the cell was considered as associated with movement or immobility. Finally, the proportion of cells associated with rest or immobility was calculated for each imaged mouse.

Statistics

Statistical tests were performed using GraphPad (Prism).

Appendix 1

Perisomatic inhibition onto pyramidal neurons in CA1 is established during the early stages of development. Firstly, we investigated the effect of perisomatic inhibition on response to twitch-like inputs in a rate model. It allowed us to analytically characterize and compare the changes in the dynamics of the model under the conditions of weak and strong inhibition. Secondly, we show through numerical simulations that similar changes can be observed in a more realistic spiking network model. We show that increasing the strength of perisomatic inhibition in our models can account for the experimentally observed decrease in responses to twitch-like feedforward inputs.

Rate model

The rate model consists of two populations (excitatory and inhibitory) with interaction strengths Jab, (a,bE,I,Jab>0). They receive feedforward input of strengths Ja0>0 from an external excitatory population with average firing rate r0. The rates rE,rI represent the population-averaged activities. The timescale of their evolution is determined by tE and tI and follows

(1) (r˙Er˙I)=(-rE/tE-rI/tI)+f((JEE-JEIJIE-JII)(rErI)+(JE0JI0)r0)

where f(x) is the neuronal transfer function. We choose it to be threshold linear, that is, f(x)=[x]+. For positive x, the dynamics in matrix notation can be written as

(2) r˙=(J-I/t)r+r0=Mr+r0

The fixed point of this system is given by

(3) r*=-M-1r0

where

(4) M1=1det(M)(J111τ1JE1J1EJEE+1τE)
(5) det(M)=JEIJIEJEEJII+J11τEJEEτI+1τEτI

Linear stability

The fixed point r* is stable to small perturbations if the real parts of all the eigenvalues of the Jacobian matrix M(r*) are negative. The eigenvalues can be expressed as

(6) λ±=12(Tr(M)±Tr(M)2-4det(M))

where

(7) Tr(M)=JEE-JII-1tE-1tI

Equivalently, the system is always stable if det(M)>0 and Tr(M)<0.

Requiring re>0 and ri>0 gives the conditions

(8) (JII+1τI)JE0JEIJI0>0JE0JI0>JEIJII+1/τI
(9) JIEJE0(JEE1τE)JI0>0JE0JI0>JEE1/τEJIE

det(M)>0 gives,

(10) (JEE1τE)(JII+1τI)+JEIJIE>0JEIJII+1/τI>JEE1/τEJIE

Combining the inequalities above gives the constraints for stable nonzero rates

(11) JE0JI0>JEIJII+1/τI>JEE1/τEJIE

When the solutions are stable, small perturbations δr will decay to zero. Twitches can be considered as perturbations around the fixed point. The transient response to such short impulses can be expressed as r(t)=C1exp(λ+)+C2exp(λ2)

For a system with external white noise, ξ(t) as input we have

(12) r˙=Mr+r0+Σξ(t)
(13) ξ(t)t=0,ξ(t)ξ(s)t=δ(t-s)

We set the off-diagonal elements to zero, that is, Σij=0 if ij. Let δr(t)=r(t)-r*, so the linearized system is

(14) δr˙=Mδr+Σξ(t)dt

This can be seen as a 2D Ornstein–Ulenbeck process defined as

(15) x(t)=-Ax(t)dt+BdW(t)

which is well documented (Gardiner and Bennett, 1985, pp. 109–111), and we can immediately write down the expression for the covariance matrix C(t). With (A=-M), we have

(16) C(t)=δr(t)δrT(t+t)=exp(-At)σ
(17) σ=detAΣ+[ATr(A)1]Σ[ATr(A)1]T2Tr(A)det(A)

Spiking model: LIF network

Excitatory and inhibitory neurons are modeled as LIF neurons. The LIF network consists of NE excitatory neurons and NI inhibitory neurons with exponentially decaying postsynaptic currents. Each neuron receives exactly KE excitatory and KI inhibitory inputs from randomly selected neurons in the network. And we assume that the network is sparse, that is, KN. The evolution of the subthreshold membrane voltage Via of neuron i in population a{E,I} is given by

(18) dViadt=Viaτm+Isyn+Iext+η,Isyn=b,jJijabSjab
(19) dSjabdt=-Sjabtsyn+tjbδ(t-tjb)

When the membrane voltage reaches the threshold, Vthreshold, it is reset to Vreset. τsyn is the synaptic time constant, and tjb is the spike time of neuron (i,b). The coupling strengths Jijab=Jab, if there is a connection from neuron (j,b) to (i,a) and zero otherwise. The contribution of external inputs is represented by Iext. Isyn represents the total synaptic currents due to spikes. Spikes are modeled as delta functions.

If a neuron (j,b) emits a spike at time tjb and projects to a postsynaptic neuron (i,a), this will result in a change of the membrane voltage Via of the postsynaptic neuron by an amount Jab. The membrane voltage decays exponentially to its resting potential in a time tm. Each neuron receives an independent Gaussian white noise of amplitude η.

Simulations and data analysis

The network simulations were conducted using custom code written in Python and C++, and all the analyses was done in Python. We use the forward Euler method to solve the set of coupled ODEs with a time step of 0.1ms.

Covariance

Given a stationary stochastic process Xt with mean μX=E[Xt], the autocovarince is given by

(20) CXX(t)=E[(Xt-μX)(Xt-t-μX)]

The covariance of Xt with another process Yt is defined as

(21) CXY(t)=E[(Xt-μX)(Yt-t-μY)]

Data availability

NWB dataset is available at DANDI Archive (https://dandiarchive.org/dandiset/000219). All codes are on GITLAB (Cossart Lab - GitLab).

The following data sets were generated
    1. Robin FD
    (2022) DANDI
    ID 000219. Two photon calcium imaging in the CA1 region of the hippocampus in neonatal mice.

References

    1. Jouvet-Mounier D
    2. Astic L
    (1968)
    Study of the course of sleep in the young rat during the 1st postnatal month
    Comptes Rendus Des Seances de La Societe de Biologie et de Ses Filiales 162:119–123.
  1. Book
    1. Paxinos H
    2. Watson K
    (2020)
    Atlas of the Developing Mouse Brain
    Elsevier Science.
  2. Book
    1. Suen CY
    2. Wang PSP
    3. Lam L
    (1994) Thinning Methodologies for Pattern Recognition
    In: Suen CY, editors. S Mach Perc. World Scientific. pp. 239–261.
    https://doi.org/10.1142/2100
    1. Tyzio R
    2. Represa A
    3. Jorquera I
    4. Ben-Ari Y
    5. Gozlan H
    6. Aniksztejn L
    (1999)
    The establishment of GABAergic and glutamatergic synapses on CA1 pyramidal neurons is sequential and correlates with the development of the apical dendrite
    The Journal of Neuroscience 19:10372–10382.

Decision letter

  1. Adrien Peyrache
    Reviewing Editor; McGill University, Canada
  2. Laura L Colgin
    Senior Editor; University of Texas at Austin, United States
  3. Simon JB Butt
    Reviewer; University of Oxford, United Kingdom

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "The rapid developmental rise of somatic inhibition disengages hippocampal dynamics from self-motion" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Laura Colgin as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Simon J B Butt (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

While each reviewer has raised a number of specific concerns about the present study, there was an agreement that the following essential revisions needed to be addressed to warrant publication of the manuscript.

1. Statistical analysis needs to be improved throughout the manuscript and unified across figures. In their present form, some of the claims cannot be supported. Reviewer #1 (points #3, 4, 7) provided a detailed list of statistical tests to improve. Reviewer #2 (point #4) also raised concerns about the statistics demonstrating that PMTH differed between pyramidal and inhibitory neurons.

2. Although the method has been already published, the classification of inhibitory neurons with a deep neural network is interesting but requires more details. Specifically, there seems to be some discrepancy between ground truth and automatically labelled data (Reviewer #1, point #5; Reviewer #2, point #3) and it is unclear whether this classifier works correctly across developmental age.

3. The assessment of behavioural states must be improved especially since it is unclear how much the effects reported in the study depend or not on brain states (Reviewer #1 point #2). Furthermore, it seems that a whole class of "unclassified movements" was not used and it is unclear whether it is related to different brain states (Reviewer #3, point #3).

4. It is somehow still unclear whether the functional inhibitory inputs to pyramidal cells are changing over the ages studied as it is only really tested at P9-P10 (Reviewer #1 point #7; Reviewer #2 point #5; Reviewer #3 point #1). It would interesting to include any (even partial) data that were collected before P9. Note that this is not a request to collect more data. At any rate, some of the claims should be perhaps tempered, especially the discussion about a switch from external to internal models (Reviewer #1, point #9; Reviewer #3 point #5).

Reviewer #1 (Recommendations for the authors):

(1) From a methodological standpoint, sufficient data to support the conclusion that the invasive neurosurgical procedure used to perform hippocampal imaging does not disrupt network function is not presented. An estimated 7mm3 of cortical tissue of the ipsilateral somatosensory cortex is aspirated to allow insertion of the window implant. The authors do not present any data documenting the physiologic recovery of the pup from the surgery or the quantitative stability of active cells over the recording period. Such data would be important to ensure that pups of all ages had similarly recovered from the procedure and that age-dependent recovery processes do not contribute to the observed results. Furthermore, bilateral silicon probe recordings are used to suggest that ipsilateral and contralateral hippocampal network activity is comparable, but these data are difficult to interpret. Rates of eSW are presented and labeled as not statistically different between the bilateral hippocampi, but 5/6 pups show an increase in eSW rate for the contralateral hippocampus and it is concerning that the sample size is underpowered to detect what could be physiologically meaningful differences. The correlogram of eSW activity is also not centered around zero, suggesting that one hippocampus is consistently leading the other; which hippocampus (ipsilateral or contralateral) serves as the reference for this graph is not detailed in the figure or legend, but this could also indicate a difference in function of the surgically altered hippocampus. To my reading, it is also unclear what data are used for the calculation of the hippocampal power spectra, and no statistical comparison between ipsilateral and contralateral hippocampi is actually performed to support similarity. Furthermore, does aspiration of the ipsilateral somatosensory cortex affect spontaneous twitches/waking movements or the hippocampal-cortical response? For instance, is there a difference in response to body movements corresponding to the limbs contralateral to the window (cortex responding to the movements has been aspirated) vs. limbs ipsilateral to the window (cortex responding to the movements is intact)?

(2) How state-dependent dynamics are incorporated into the analysis requires further clarification. Given the state-specific activity patterns that occur in the more mature brain (e.g. hippocampal theta restricted to REM sleep and movement, sharp wave ripples in NREM and wakeful immobility) and the emergence of electrophysiologic differentiators of sleep states around P10, care should be taken to ensure that data from similar states are compared across ages. If there is a disproportionate representation of behavioral states across ages (for instance related to post-surgical comfort), sampling data across states would not control for state-dependent neural activity patterns. Specifically, the data comprising Figures 1B and 1C would be susceptible to such effects.

(3) The approach to statistical analysis of neural activity parameter changes across age is incompletely detailed and insufficient to support the authors' conclusions. The exact statistical tests, group numbers, and p-values used for data presented in Figure 1B-C, Figure 1-supplement 1A, Figure 2B-D should be detailed in the figure legends. It seems as though a linear correlation coefficient is used to quantify data in Figure 1B, but Kruskal-Wallis ANOVA between groups is used in Figure 1C. In Figure 2C, a sigmoidal fit is chosen, and in Figures 2B and 2D, no fitting is performed, though a linear relationship is mentioned for Figure 2B in the text. A unified statistical approach to modeling changes in neural properties over time should instead be used. Justification for different fits should be provided for each dataset. For example, examining the data in Figure 1C, there appears to be a decrease in the number of transients at P9-P10, which subsequently increases through P11-P12. The data appear to be underpowered for use of an 8 group ANOVA given the small number of data points per group, and potentially significant non-linearities in the number of transients over time are therefore ignored. Similarly, there appears to be a peak in the number of cells imaged at P9, but no details of which statistical testing was used are provided and again, more rigorous statistical approaches are required to understand whether significant parameter changes are occurring over time.

(4) PMTH plots require more rigorous and detailed quantitative analysis. In Figure 2A and Figure 3, these are portrayed as percentiles across the population. To properly evaluate such data, statistical methodology (for example, a shuffling procedure) should be employed to determine whether a significant modulation is present at each age and for different cell types. Modulation values could then be quantitatively compared (using statistical testing) between groups (e.g. P5-8 and P10-12 pyramidal cells and interneurons in Figure 3). Furthermore, examining the PMTH plots in Figure 2, it appears that there may be a leftward shift in the peak after P9 (i.e. increased activity occurs after movement in P5-9, but then seems more centered around the time of movement, or even before the detected movement in P10-12). If this observation were to be supported statistically, it could suggest an important change in the flow of hippocampal-cortical activity around the time of movement.

(5) The conclusions of the manuscript rely strongly on the differentiation of pyramidal cells and interneurons. Therefore, more details regarding the cell classifier (DeepCINAC) should be provided for this particular dataset. What features permit specific identification of interneurons, and are these features demonstrated to be constant across developmental age? In Figure 3-supplement 2, the average PMTHs for predicted vs. labelled interneurons are quite different; the peak percentage of active cells is ~10% for predicted interneurons, but ~25% for labeled interneurons. It also appears as though labeled interneurons have essentially no firing outside of movement-related activity in the surrounding 20-second window, in contrast to the predicted interneurons. These differences should be further explored and explained.

(6) The extra-hippocampal synaptic afferents appear to be evaluated using structural connectivity metrics, and no substantial changes were observed. However, no functional assessment of these synapses was performed. The conclusion that the mechanism of hippocampal activity changes is a result of local GABAergic innervation should be tempered given this, and more clearly explained in the discussion.

(7) In Figure 4, GABAergic axon activity was determined at P9-10 only, but this was interpreted as an increase. Without comparison to other ages, this conclusion would need to be tempered. Statistical evaluation of data in Figure 4B also does not directly support the authors' conclusions. They explain in the results text a sharp increase in labeling at P9, but their statistical testing does not show a significant difference between P7 and P9, which is the supposed inflection point. This point further illustrates the need for more rigorous and sensitive statistical testing for age-dependent changes.

(8) For the modeling data provided in Figure 5, why do the effects of inhibition persist over up to 5 seconds? Such effects would seem to extend beyond the direct effects of monosynaptic perisomatic inhibition. Most such models are presented over a span of milliseconds, not seconds. The parameters governing these timings, as well as the reasoning behind the choice of parameters used to generate the model, require justification and would improve clarity.

(9) The authors demonstrate evidence for detachment of pyramidal cell activity from movements and suggest a transition to internal representations. However, the data presented are strongly focused on peri-movement epochs, without any indication of the characteristics of neural activity during non-movement epochs. Unless the authors consider pursuing further analysis of their non-movement epoch data, implications regarding internal models and plasticity should be perhaps removed from the discussion, and this limitation acknowledged.

Reviewer #2 (Recommendations for the authors):

– Table 1: it is not clear why some mice were used in some figures and not in others. Some further detail would be useful here.

– Some further details on the CICADA analysis software would be very helpful, as this does not seem to have been published separately. How does this work, and why did the authors use it?

– It would be very helpful to add more details on those interneurons that were identified by the deep learning network. How do the authors justify using this method (as opposed to just looking for tdTomato labelled cells), and why was it necessary? What does 91% reliability refer to – false-positive or false-negative? The authors should provide a full breakdown of numbers of pyramidal cells misclassified as interneurons, and vice versa. Figure 3 Figure supp 2C shows a striking difference in the background activity levels of classified as opposed to labelled neurons – the authors should comment on why this is, and justify treating these cell types equally in spite of this.

– With respect to figure 3, the authors state that "the link between movement and activity evolves differentially towards the start of the second postnatal week when comparing pyramidal neurons and GABAergic interneurons, the former being inhibited or detached from movements while the latter remaining activated". Whilst this appears to be true from looking at average activity PMTH plots, there is no statistical quantification to demonstrate that this is the case. By quantifying either the peak and/or the trough of the PMTH, the authors should try to show a statistically significant interaction between how these develop with age and cell type.

– Figure 4D: in addition to the problems associated with having only one age point, this data also appears to show a striking dissociation between axonal and somatic interneuron activity: in the axonal trace (only) there is a significant dip below baseline following the movement peak. The authors should comment on the significance of this, and what it means for their conclusions.

Reviewer #3 (Recommendations for the authors):

1) On line 52 the authors build the argument that feedback connections are important for self-organisation of the internal state. Further, that is likely GABAergic in nature in CA1 due to the paucity of recurrent glutamatergic connections onto pyramidal projections neurons. The authors then show some evidence of the strengthening of the perisomatic connections (Figure 4), but – as they acknowledge in the discussion (lines 478-490) – could not strengthen the JEI connection as shown in their model (Figure 5A) be equally important? Is it possible to use their existing imaging data to test if more GABAergic interneurons respond to movement at P10-12 in the data shown in Figure 3? And assess the timing of the interneuron activity relative to the pyramidal cells?

(2) It would be interesting to see the primary imaging data for that shown in Figure 2A i.e. deltaF/F of cells following movement.

(3) Line 155, the authors define myoclonic twitches versus movements associated with wakefulness. In the methods (line 791), the authors further delineate "unclassified movements". Why are these not included in the results? Further, it is a shame that the authors only did one recording with nuchal EMG to better separate active sleep (AS is perhaps a better descriptor as opposed to REM) versus wakefulness. Again, it would be interesting to see the raw data from this animal.

(4) Do the authors think that the drop in peak %active cells from P8 to P9 (Figure 2A) is important? Could not a sigmoidal function be fitted to Figure 2D and if so, what would the R(2) value be?

(5) Line 368, it is not immediately clear to me why an increase in feed-forward inhibition leads to a detachment from external inputs. In primary sensory areas, an increase in feed-forward inhibition is observed in line with the emergence of fast sensory processing (see, for example, Chittajallu and Isaac, 2010).

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "The rapid developmental rise of somatic inhibition disengages hippocampal dynamics from self-motion" for further consideration by eLife. Your revised article has been evaluated by Laura Colgin (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

(1) The manuscript should be clearer regarding the potential consequences of the experiments on the monitored features, especially when it comes to ipsi-contralateral differences. Reviewer #1 has raised again several concerns in this regard, from differences in electrophysiological patterns to possible differences in left-right twitches and how they specifically affect the recorded hemisphere.

(2) Reviewer #2 still has concerns regarding the statistics of PMTHs and requires further explanation.

Reviewer #1 (Recommendations for the authors):

The authors have addressed many of the points raised. Points requiring additional modification/clarification are documented below.

Initial Review – From a methodological standpoint, sufficient data to support the conclusion that the invasive neurosurgical procedure used to perform hippocampal imaging does not disrupt network function is not presented.

Author Response – We gently disagree with this reviewer given that, as previously indicated in the text and illustrated in Figure1 —figure supplement 1C, we had performed electrophysiological recordings in ipsi- and contralateral hippocampi and observed that hippocampal oscillations in the form of eSWs were present in both hemispheres, at several ages.

Reviewer Response – Demonstrating the presence of one type of oscillation that actually shows a trend toward a decreased occurrence rate in the surgically manipulated hemisphere, in my opinion, is not sufficient to claim that hippocampal network function is unaffected. The work that the authors perform below (with caveats also noted below), however, helps to provide support for at least a minimal disruption of the network.

Initial Review – An estimated 7mm3 of cortical tissue of the ipsilateral somatosensory cortex is aspirated to allow insertion of the window implant. The authors do not present any data documenting the physiologic recovery of the pup from the surgery or the quantitative stability of active cells over the recording period. Such data would be important to ensure that pups of all ages had similarly recovered from the procedure and that age-dependent recovery processes do not contribute to the observed results.

Author Response – We thank the reviewer for pointing out this important issue. We have performed simultaneous CA1 dynamics and EMG recordings in two animals. We found that craniotomy did not alter the structure of the sleep-wake cycle, as revealed by the quantification performed in Figure 1—figure supplement 1B. As previously described in Jouvet et al. 1967, P5-6 mice spend 70-80% of their time in active sleep, which is in agreement with our experiments. In addition, to probe the stability of active cells over our recording periods, we have computed the difference in the frequency of calcium transients between the first and last quarters of the recordings. We found that single-cell activity was stable throughout our recordings, for all ages recorded (see Author response image 1, median difference = 0,08 transient/mins).

Reviewer Response – The quantification of the sleep-wake cycle as performed in figure supplement 1B is inconsistent with currently proposed methods to do so. Indeed, the authors reference a paper from 1967 to support their work. Previously, sleep was suggested to be initiated in an undifferentiated mixed state (1,2), but more recently, quiet and active sleep is thought to be present from birth (3, 4, 5). The current figure shows "twitches", "REM", "transition" and "awake", which are not consistent with either classification scheme. This analysis should be redone in a manner consistent with a classification scheme to enable comparisons with other current literature in the field.

The quantification of the calcium transients over the course of the recording does seem convincing, and I would suggest that the authors include this figure in their supplementary data.

(1) M.G. Frank, H.C. Heller Development of REM and slow wave sleep in the rat

Am J Physiol, 272 (6 Pt 2) (1997), pp. R1792-R1799

(2) M.G. Frank, H.C. Heller The ontogeny of mammalian sleep: A reappraisal of alternative hypotheses J Sleep Res, 12 (2003), pp. 25-34

(3) A.M. Seelke, M.S. Blumberg The microstructure of active and quiet sleep as cortical δ activity emerges in infant rats Sleep, 31 (2008), pp. 691-699

(4) A.M. Seelke, K.A. Karlsson, A.J. Gall, M.S. Blumberg Extraocular muscle activity, rapid eye movements and the development of active and quiet sleep Eur J Neurosci, 22 (2005), pp. 911-920

(5) M.S. Blumberg, A.J. Gall, W.D. Todd The development of sleep-wake rhythms and the search for elemental circuits in the infant brain Behav Neurosci, 128 (2014), pp. 250-263

Initial Review – Furthermore, bilateral silicon probe recordings are used to suggest that ipsilateral and contralateral hippocampal network activity is comparable, but these data are difficult to interpret. Rates of eSW are presented and labeled as not statistically different between the bilateral hippocampi, but 5/6 pups show an increase in eSW rate for the contralateral hippocampus and it is concerning that the sample size is underpowered to detect what could be physiologically meaningful differences.

Author Response – We thank the reviewer for bringing up this important issue. Indeed, our experimental access to early hippocampal activity with 2-photon calcium imaging relies on a quite invasive procedure. However, the many control experiments we have performed indicate that early hippocampal dynamics were not significantly altered by the surgery. First, our extracellular electrophysiological recordings from a sample of 6 mice (ranging from P6 to P11, Figure 1—figure supplement 1C) show that the frequency of early sharp waves (eSW) was slightly but not significantly reduced in the ipsilateral hemisphere compared to the contralateral one. Of note, a similar “non-significant” decrease had been previously reported by another group (Graf et al. 2021 Figure S6C). As suggested by the reviewer, we can speculate that this slight decrease may result from a reduction of the sensory feedback re-afference originating from the right limbs. Indeed, we observed that movements of the right limbs (contralateral to the window implant) elicited a slightly smaller response than those from the left limbs. This observation has been added to Figure 1 – Supplement 1E and described in the results (lines 128-134) and discussion (lines 314-320).

We have performed additional control experiments using EMG nuchal electrodes in two pups aged P5 and P6. We observed that, an hour following the surgery (corresponding to the recovery time in our experimental procedure), the composition of the sleep-wake cycle (with 70 to 80 % of active sleep) was comparable to previous reports (Jouvet-Mounier, 1969, Figure 4). This quantification was added to Figure 1—figure supplement 1B (lines 82-86).

Reviewer Response – The authors have decided to acknowledge a likely difference due to reduced sensory feedback, which is reasonable, but this is not reflected in the manuscript as the statement is made "In the same way, the acute window implant did not significantly alter electrophysiological network patterns." (lines 89-90) and no mention is made in this section of the point. The Results section referenced speaks to differences in movements rather than activity patterns.

Initial Review – The correlogram of eSW activity is also not centered around zero, suggesting that one hippocampus is consistently leading the other; which hippocampus (ipsilateral or contralateral) serves as the reference for this graph is not detailed in the figure or legend, but this could also indicate a difference in the function of the surgically altered hippocampus.

Author Response – We thank the referee for making this important observation and apologize for the lack of clarity. The onsets of eSWs recorded in the ipsilateral hippocampus served as reference. As the referee noted, the two hippocampi are not perfectly synchronous with eSWs from the contralateral hippocampus leading the ipsilateral one by 12 ms. Such delay could be explained by a slight drop in local temperature due to the chamber placement as described previously for cortical up-states (Reig et al., 2010).

Reviewer Response – In the response here, the authors explain the finding by citing Reig et al., 2010, but in the manuscript, they cite this difference as being expected as per Graf et al., 2021; Valeeva et al., 2019b (lines 97-98). This is confusing and should be clarified.

Initial Review – To my reading, it is also unclear what data are used for the calculation of the hippocampal power spectra, and no statistical comparison between ipsilateral and contralateral hippocampi is actually performed to support similarity.

Author Response – We thank the reviewer for bringing up this important point. To characterize the developmental changes in the theta band power we have performed spectral analysis using the Chronux toolbox on the entire recordings (Chronux toolbox J Neurosci Methods. 2010 Sep 30;192(1):146- 51). Spectral power was estimated using direct multi-taper estimators (3 time-bandwidth products and 5 tapers). The shaded area corresponds to the confidence interval (p>0.05). As requested by this referee, we have updated the figure and added comparison between ipsi- and contralateral hippocampi. Our results demonstrated the similarity in the spectral distribution (see peak in the theta frequency band in P11 animals, which is not present in younger animals). Surprisingly, we found that the spectral power was slightly higher for the ipsilateral hippocampus.

Reviewer Response – These statistics are helpful. What is the proposed reason for this difference in spectral power? Taken together, it seems as though there are several indicators that the ipsilateral hippocampus is different in its activity than the contralateral hippocampus (trend toward different eSW rate, consistent lead/lag in regard to eSW activity, and significant differences in power spectra). These points should be mentioned in the manuscript text for transparency even if the authors argue that they do not affect the primary conclusions of the work.

Initial Review – Furthermore, does aspiration of the ipsilateral somatosensory cortex affect spontaneous twitches/waking movements or the hippocampal-cortical response? For instance, is there a difference in response to body movements corresponding to the limbs contralateral to the window (cortex responding to the movements has been aspirated) vs. limbs ipsilateral to the window (cortex responding to the movements is intact)?

Author Response – We thank the reviewer for this important comment. To address it, we have quantified in 6 imaging sessions from 5 animals (aged between P5 and P8) the CA1 response to body movements from the limbs contralateral to the imaging window (right) vs. limbs ipsilateral to the window (left). We found that movements from the right limbs elicited a slightly reduced response than the left limb movements. Regarding the number of twitches and waking movement of the right limbs compared to the left, we found fewer twitches on the contralateral side (right) as compared to the ipsilateral one (See Author response image 2). This could result from a partial lesion of the ipsilateral motor cortex in our experimental conditions. Related to this point, we would like to mention that the motor cortex, before P12, is more passively driven by spontaneous twitches than involved in driving them (Dooley Blumberg, 2018), but that up to 24 % of the twitches in P3-5 rats could already be originating from that region (An, Luhmann 2014). Our data is somehow consistent with that observation. This is now included in the manuscript but not studied any further as it is beyond the scope of the present paper.

Reviewer Response – This quantification is also informative. Although the difference is now shown in the supplementary figure, the point is to ensure that this difference does not affect the conclusions of the present study, not to study the changes themselves. For instance, it would be important to know if only the ipsilateral limb twitches are used (routed through the intact contralateral somatomotor region), is the trend in cell recruitment over development different?

Initial Review – State-dependent parameters are not adequately described, controlled, and examined quantitatively to ensure that data from similar behavioral states is being used for analysis across ages. Network activity from wakefulness, REM/active sleep, and NREM/quiet sleep should not be presumed to be indistinguishable.

Author Response – We would like to point out that our analysis across ages focused on the population response following animal movements, and not across all behavioral states. That said, it is true that two types of movements can be distinguished, namely the twitches and the complex ones. To take this behavioral heterogeneity into account, we have now separately quantified the hippocampal activation following twitches (movement during active sleep) and complex movement (during wakefulness). We show in Figure 2 —figure supplement 1B that the hippocampal response to twitches and complex movements is similar across ages. Thus, even if the amount of time spent in each behavioral state is modified over the developmental period that we have studied, we are pretty confident that it does not impact the transition we have described in the relationship between animal movements and hippocampal activity. Additionally, we were able to combine one P5 mouse pup 2p-imaging with nuchal EMG recordings and separately computed the PMTH for movements observed during REM or wakefulness (Figure 2 —figure supplement 1C). We show that CA1 hippocampal neurons were activated and time-locked to movement in both behavioral states, with only the amplitude of the population response differing between wakefulness and during REM. This point is now included in the result section (lines 148-152) and discussed (lines 324-327).

Reviewer Response – I am a bit unclear as to the conclusion here. In figure supplement 1B, the hippocampal response to twitches and complex movements is not different, but in figure supplement 1C, the hippocampal response is different between "REM" and wakefulness. Is the conclusion that the movement semiology-based analysis doesn't show differences, but the state-based analysis does (2x difference in population response), raising the concern that the semiology-based analysis proposed is not actually capturing state-based differences? The relationship between the two figure supplements should be clarified. Also, "REM" is not typically ascribed to a P5 pup (would usually be called active sleep).

Initial Review – It seems as though a linear correlation coefficient is used to quantify data in Figure 1B, but Kruskal-Wallis ANOVA between groups is used in Figure 1C.

Author Response – To address this point, we now perform a nonlinear fit instead of the Kruskal-Wallis ANOVA between groups to test for age effects in Figure 1C. Our analysis indicates a non-linear increase of the mean transient frequency over development with a local minimum at around P10. This is now included in the figure legend and results. Thanks.

Reviewer Response – This is helpful. Typically, would be appropriate to mention somewhere (perhaps in the Methods) how the fit was selected (e.g. minimizing MSE or some other method). An r2 value is not typically valid statistically for a non-linear regression (mentioned as "r2=0,30" in the figure legend at present).

Initial Review – In Figure 2C, a sigmoidal fit is chosen, and in Figures 2B and 2D, no fitting is performed, though a linear relationship is mentioned for Figure 2B in the text.

Author Response – We thank the reviewer and have now used a sigmoidal fit in Figure 2D (as in 2C) to demonstrate the increase around P9 of hippocampal activity during the period of mouse immobility. In Figure 2B we used one-way ANOVA to show a significant effect of age on the proportion of cells activated after movement and post hoc corrected multiple comparison tests to compare between age groups. Here, we show that the post-movement activity was not different for all pairs of age groups from P10 to P12, nor between P8 and P9 while all other pairs of age groups were highly significantly different. In addition, we mentioned in the text that the difference in the median between 2 consecutive age groups was similar from P5 to P9 and that the decrease was 4 times stronger between P9 and P10 than between any other pair. We have modified the text accordingly (lines 156-157).

Reviewer Response – This is helpful. Typically, would be appropriate to mention somewhere (perhaps in the Methods) how the sigmoidal fit was selected (e.g. minimizing MSE or some other method). Again, an r2 value is not typically valid statistically for non-linear regression.

Initial Review – A unified statistical approach to modeling changes in neural properties over time should instead be used. Justification for different fits should be provided for each dataset. For example, examining the data in Figure 1C, there appears to be a decrease in the number of transients at P9-P10, which subsequently increases through P11-P12. The data appear to be underpowered for use of an 8-group ANOVA given the small number of data points per group, and potentially significant non-linearities in the number of transients over time are therefore ignored. Similarly, there appears to be a peak in the number of cells imaged at P9, but no details of which statistical testing was used are provided and again, more rigorous statistical approaches are required to understand whether significant parameter changes are occurring over time.

Author Response – See response 3-b above.

Reviewer Response – See reviewer response to 3b above.

Initial Review – In Figure 4, GABAergic axon activity was determined at P9-10 only, but this was interpreted as an increase. Without comparison to other ages, this conclusion would need to be tempered.

Author Response – We thank the reviewer for this comment that reveals some lack of clarity in the previous description of our experiments. Indeed, functional GABAergic activity was also assessed before P9, however, given that there are no GABAergic axons in the CA1 pyramidal layer at early stages (for both CCK cf. Morozov and Freund 2003, and prospective PV cells cf. Figure 4A,B), there is no signal to be measured either. We have now added a new figure (Figure 4 - figure supplement 1) to clarify this point. In agreement with our Syt2 longitudinal quantification, we show, using tdTomato expression in the Gad67cre driver mouse line, that GABAergic perisomatic innervation is only visible after p9. This matches as well our attempted imaging experiments using axon enriched GCaMP in mice before P9.

As explained above, there are few, if any, GABAergic axons in the CA1 pyramidal layer before P9 (Figure 4 —figure supplement 1 specifically, supplementary video 7). There is no signal, thus comparison is pointless. We have now clarified this point (lines 242-244), thanks.

Reviewer Response – Thanks for clarifying this issue. To ensure that no pointless comparisons are inferred by the reader, it may be of benefit to change the wording of lines 255-256 from "increase in functional perisomatic GABAergic activity" to "emergence of functional perisomatic GABAergic activity."

Initial Review – Statistical evaluation of data in Figure 4B also does not directly support the authors' conclusions. They explain in the results text a sharp increase in labeling at P9, but their statistical testing does not show a significant difference between P7 and P9, which is the supposed inflection point. This point further illustrates the need for more rigorous and sensitive statistical testing for age-dependent changes.

Author Response – We had observed a significant difference between the P7 and P11 age groups (2 age groups exactly centered around P9), which demonstrates an increase in PV coverage around P9. In fact, P9 being the inflection point it is not surprising that it is neither different from P7 nor from P11 since it is precisely when axonal arborization is changing the most.

Reviewer Response – The statistical interpretation is still problematic here. If the statistical comparison between the P7 and the P9 groups is not significant, it is not accurate to report in the results that "However, around P9, a sudden increase in the density of positive labeling was observed" (lines 240-241). The reasoning about P9 being an inflection point may be the case, but the currently employed statistics do not show it. If the authors wanted to show there is an inflection point, rather than just stating what their statistics show (a significant difference between P7 and P11), a modeling approach that demonstrates a change in the curvature of the modeled fit around P9 would be appropriate.

Reviewer #2 (Recommendations for the authors):

Overall the authors have done a good job of responding to many of the points raised. However, I feel they have not provided a complete answer to three key points, raised by both myself and reviewer 1.

1) Statistics for PMTHs. Thanks to the authors for clarifying that there was already shuffled significance criteria in place for the PMTHs and the addition of an asterisk to the plots helps to clarify this. However, I still do not feel that the explanation as it stands is fully satisfactory.

a) It is necessary to provide more details of the shuffling procedure. The methods state simply that '500 surrogate raster plots per 770 imaging session were computed', which is insufficient detail. Unless the authors can provide a good reason otherwise, I think that the surrogates should be 20sec sections of data, drawn at random from the entire dataset, from both movement and non-movement epochs. If surrogates are not defined as above, the authors should justify why. In any case, more detail should be provided in the methods.

b) Are the 95th and 5th percentiles calculated with respect to each bin, or the entire 20s PMTH? If the former, then the authors need to account for the problem of multiple comparisons across the multiple time bins of the PMTH.

2) Statistics in addition to the PMTHs. Both Reviewer 1 and myself requested additional statistical analyses of the %cell activation data, in particular comparisons of how activation peaks and inhibition troughs evolve over development. The current argument of the authors is that it is sufficient to rely on whether PMTH activity crosses a significance threshold, or not, as a form of cross-age comparison. I don't agree with this – to take an example, the movement-activated activity peaks at P8 and P10 look visually very similar (Figure 2A). The proper statistical approach to test whether they are different is to compare them directly. Using the author's current approach, two very similar samples, not significantly different in themselves, could be judged to be from different populations based on one crossing an arbitrary shuffling-based threshold, and the other not.

The authors already have a model for better analysis in the manuscript – the analysis of activity troughs in Figure 2B. I think that a proper analysis of the data requires this method to be applied to movement-related peaks and post-movement throughs, for all data in figures 2, 3, and 4.

3) Thanks to the clarification from the authors, I now understand and accept that there are no axon terminals to image before P9. However, some more temporal precision regarding the emergence of perisomatic axon terminal activity would be helpful. The key transition dates for pyramidal cell activity run from P8 (still immature), P9 (which is transitional – no activity peak but also no trough) to P10 (mature activity). How does the emergence of axonal activity relate to this timeline? Is there a difference between P9 and P10? Is the response already mature at its first emergence (at P9)? Or does it continue to gradually increase between P9 and P12? This information would help make a more specific link between increases in perisomatic inhibition and PC activity.

Unless the authors can show these data, then phrases such as 'functional surge' should be avoided ('surge' implies a rapid maturation, which cannot the demonstrated using one time point), the authors should restrict their conclusions to stating that 'functional perisomatic activity from inhibitory interneurons can be observed at P9-10 (or similar).

https://doi.org/10.7554/eLife.78116.sa1

Author response

Essential revisions:

While each reviewer has raised a number of specific concerns about the present study, there was an agreement that the following essential revisions needed to be addressed to warrant publication of the manuscript.

1. Statistical analysis needs to be improved throughout the manuscript and unified across figures. In their present form, some of the claims cannot be supported. Reviewer #1 (points #3, 4, 7) provided a detailed list of statistical tests to improve. Reviewer #2 (point #4) also raised concerns about the statistics demonstrating that PMTH differed between pyramidal and inhibitory neurons.

Statistical analysis is unified across figures when similar datasets (size, distribution) are being tested. We think this main point arises from the lack of clarity in our description of the statistical analysis being performed. We apologize for that and have now justified the choice for the various statistical tests employed. Statistical tests are now detailed for all figures as suggested by reviewers# 1 and 2.

2. Although the method has been already published, the classification of inhibitory neurons with a deep neural network is interesting but requires more details. Specifically, there seems to be some discrepancy between ground truth and automatically labelled data (Reviewer #1, point #5; Reviewer #2, point #3) and it is unclear whether this classifier works correctly across developmental age.

We thank the reviewers for pointing out this lack of detail. We have now clarified this point in the manuscript (see responses to Reviewer #1, point #5 and Reviewer #2, point #3).

3. The assessment of behavioural states must be improved especially since it is unclear how much the effects reported in the study depend or not on brain states (Reviewer #1 point #2). Furthermore, it seems that a whole class of "unclassified movements" was not used and it is unclear whether it is related to different brain states (Reviewer #3, point #3).

Behavioral states are now further examined in the revised manuscript (see response to Reviewer #1 point #2 and Reviewer #3, point #3). We thank the reviewers for this suggestion.

4. It is somehow still unclear whether the functional inhibitory inputs to pyramidal cells are changing over the ages studied as it is only really tested at P9-P10 (Reviewer #1 point #7; Reviewer #2 point #5; Reviewer #3 point #1). It would interesting to include any (even partial) data that were collected before P9. Note that this is not a request to collect more data. At any rate, some of the claims should be perhaps tempered, especially the discussion about a switch from external to internal models (Reviewer #1, point #9; Reviewer #3 point #5).

We directly addressed this concern. See responses to Reviewer #1 point #7, Reviewer #2 point #5; Reviewer #3 point #1 and Reviewer #1, point #9; Reviewer #3 point #5

Reviewer #1 (Recommendations for the authors):

1) From a methodological standpoint, sufficient data to support the conclusion that the invasive neurosurgical procedure used to perform hippocampal imaging does not disrupt network function is not presented.

We gently disagree with this reviewer given that, as previously indicated in the text and illustrated in Figure1 —figure supplement 1C, we had performed electrophysiological recordings in ipsi and contralateral hippocampi and observed that hippocampal oscillations in the form of eSWs were present in both hemispheres, at several ages.

(1b) An estimated 7mm3 of cortical tissue of the ipsilateral somatosensory cortex is aspirated to allow insertion of the window implant. The authors do not present any data documenting the physiologic recovery of the pup from the surgery or the quantitative stability of active cells over the recording period. Such data would be important to ensure that pups of all ages had similarly recovered from the procedure and that age-dependent recovery processes do not contribute to the observed results.

We thank the reviewer for pointing out this important issue. We have performed simultaneous CA1 dynamics and EMG recordings in two animals. We found that craniotomy did not alter the structure of the sleep wake cycle, as revealed by the quantification performed in Figure 1—figure supplement 1B. As previously described in Jouvet et al. 1967, P5-6 mice spend 70-80% of their time in active sleep, which is in agreement with our experiments. In addition, to probe the stability of active cells over our recording periods, we have computed the difference in the frequency of calcium transients between the first and last quarters of the recordings. We found that single-cell activity was stable throughout our recordings, for all ages recorded (see Author response image 1, median difference = 0,08 transient/mins).

Author response image 1

(1c) Furthermore, bilateral silicon probe recordings are used to suggest that ipsilateral and contralateral hippocampal network activity is comparable, but these data are difficult to interpret. Rates of eSW are presented and labeled as not statistically different between the bilateral hippocampi, but 5/6 pups show an increase in eSW rate for the contralateral hippocampus and it is concerning that the sample size is underpowered to detect what could be physiologically meaningful differences.

We thank the reviewer for bringing up this important issue. Indeed, our experimental access to early hippocampal activity with 2-photon calcium imaging relies on a quite invasive procedure. However, the many control experiments we have performed indicate that early hippocampal dynamics were not significantly altered by the surgery. First, our extracellular electrophysiological recordings from a sample of 6 mice (ranging from P6 to P11, Figure 1—figure supplement 1C) show that the frequency of early sharp waves (eSW) was slightly but not significantly reduced in the ipsilateral hemisphere compared to the contralateral one. Of note, a similar “non-significant” decrease had been previously reported by another group (Graf et al. 2021 Figure S6C). As suggested by the reviewer, we can speculate that this slight decrease may result from a reduction of the sensory feedback re-afference originating from the right limbs. Indeed, we observed that movements of the right limbs (contralateral to the window implant) elicited a slightly smaller response than those from the left limbs. This observation has been added to Figure 1 – Supplement 1E and described in the results (lines 128-134) and discussion (lines 314-320).

We have performed additional control experiments using EMG nuchal electrodes in two pups aged P5 and P6. We observed that, an hour following the surgery (corresponding to the recovery time in our experimental procedure), the composition of the sleep-wake cycle (with 70 to 80 % of active sleep) was comparable to previous reports (Jouvet-Mounier, 1969, Figure 4). This quantification was added to Figure 1—figure supplement 1B (lines 82-86).

The correlogram of eSW activity is also not centered around zero, suggesting that one hippocampus is consistently leading the other; which hippocampus (ipsilateral or contralateral) serves as the reference for this graph is not detailed in the figure or legend, but this could also indicate a difference in function of the surgically altered hippocampus.

We thank the referee for making this important observation and apologize for the lack of clarity. The onsets of eSWs recorded in the ipsilateral hippocampus served as reference. As the referee noted, the two hippocampi are not perfectly synchronous with eSWs from the contralateral hippocampus leading the ipsilateral one by 12 ms. Such delay could be explained by a slight drop in local temperature due to the chamber placement as described previously for cortical up-states (Reig et al., 2010).

To my reading, it is also unclear what data are used for the calculation of the hippocampal power spectra, and no statistical comparison between ipsilateral and contralateral hippocampi is actually performed to support similarity.

We thank the reviewer for bringing this important point. To characterize the developmental changes in the theta band power we have performed spectral analysis using the Chronux toolbox on the entire recordings (Chronux toolbox J Neurosci Methods. 2010 Sep 30;192(1):146-51). Spectral power was estimated using direct multi-taper estimators (3 time-bandwidth product and 5 tapers). Shaded area corresponds to confidence interval (p>0.05). As requested by this referee, we have updated the figure and added comparison between ipsi- and contralateral hippocampi. Our results demonstrated the similarity in the spectral distribution (see peak in theta frequency band in P11 animals, that is not present in younger animals). Surprisingly, we found that the spectral power was slightly higher for the ipsilateral hippocampus.

(1d) Furthermore, does aspiration of the ipsilateral somatosensory cortex affect spontaneous twitches/waking movements or the hippocampal-cortical response? For instance, is there a difference in response to body movements corresponding to the limbs contralateral to the window (cortex responding to the movements has been aspirated) vs. limbs ipsilateral to the window (cortex responding to the movements is intact)?

We thank the reviewer for this important comment. To address it, we have quantified in 6 imaging sessions from 5 animals (aged between P5 and P8) the CA1 response to body movements from the limbs contralateral to the imaging window (right) vs. limbs ipsilateral to the window (left). We found that movements from the right limbs elicited a slightly reduced response than the left limb movements.

Indeed, our experimental access to early hippocampal activity with 2-photon calcium imaging relies on a quite invasive procedure. However, the many control experiments we have performed indicate that early hippocampal dynamics were not significantly altered by the surgery. First, our extracellular electrophysiological recordings from a sample of 6 mice (ranging from P6 to P11, Figure 1—figure supplement 1C) show that the frequency of early sharp waves (eSW) was slightly but not significantly reduced in the ipsilateral hemisphere compared to the contralateral one. Of note, a similar “non-significant” decrease had been previously reported by another group (Graf et al. 2021 Figure S6C). As suggested by the reviewer, we can speculate that this slight decrease may result from a reduction of the sensory feedback re-afference originating from the right limbs. Indeed, we observed that movements of the right limbs (contralateral to the window implant) elicited a slightly smaller response than those from the left limbs. This observation has been added to Figure 1 – Supplement 1E and described in the results (lines 128-134) and discussion (lines 314-320).

We have performed additional control experiments using EMG nuchal electrodes in two pups aged P5 and P6. We observed that, an hour following the surgery (corresponding to the recovery time in our experimental procedure), the composition of the sleep-wake cycle (with 70 to 80 % of active sleep) was comparable to previous reports (Jouvet-Mounier, 1969, Figure 4). This quantification was added to Figure 1—figure supplement 1B (lines 82-86).

Regarding the number of twitches and waking movement of the right limbs compared to the left, we found less twitches on the contralateral side (right) as compared to the ipsilateral one (See Author response image 2 ). This could result from a partial lesion of the ipsilateral motor cortex in our experimental conditions. Related to this point, we would like to mention that the motor cortex, before P12, is more passively driven by spontaneous twitches than involved in driving them (Dooley Blumberg, 2018), but that up to 24 % of the twitches in P3-5 rats could already be originating from that region (An, Luhmann 2014). Our data is somehow consistent with that observation. This is now included in the manuscript but not studied any further as it is beyond the scope of the present paper.

Author response image 2

(2) How state-dependent dynamics are incorporated into the analysis requires further clarification. Given the state-specific activity patterns that occur in the more mature brain (e.g. hippocampal theta restricted to REM sleep and movement, sharp wave ripples in NREM and wakeful immobility) and the emergence of electrophysiologic differentiators of sleep states around P10, care should be taken to ensure that data from similar states are compared across ages. If there is a disproportionate representation of behavioral states across ages (for instance related to post-surgical comfort), sampling data across states would not control for state-dependent neural activity patterns. Specifically, the data comprising Figures 1B and 1C would be susceptible to such effects.

We thank the reviewer for this major point. To address the heterogeneity of the movements occurring in different behavioral states, we have examined the CA1 response to twitches and complex movements occurring during REM sleep and awake states across ages, respectively and found no significant difference (Figure 2 —figure supplement 1). As a result, we decided to combine all types of movements. This is now detailed in the manuscript (lines 143-148).

(3a) The approach to statistical analysis of neural activity parameter changes across age is incompletely detailed and insufficient to support the authors' conclusions. The exact statistical tests, group numbers, and p-values used for data presented in Figure 1B-C, Figure 1-supplement 1A, Figure 2B-D should be detailed in the figure legends.

We obviously agree with this reviewer that rigorous statistics should be employed and can certify that the data analyzed in the submitted manuscript was carefully examined following that principle. We feel that his/her strong criticism regarding that point was not fully justified. In particular, we do not understand why statistical tests should be “unified” across different figures of the paper. Rather, statistical tests should be adapted to the sample size and distribution. Of course, the same tests were used for similar datasets. This revised manuscript now contains further description and justification of all the tests included in every figure panels.

We have noticed that indeed details on statistical analysis could sometimes be missing in the legends. Figure 2B-D and Figure 1-supplement 1A now mention all tests. Thanks.

(3b) It seems as though a linear correlation coefficient is used to quantify data in Figure 1B, but Kruskal-Wallis ANOVA between groups is used in Figure 1C.

To address this point, we now perform a nonlinear fit instead of the Kruskal-Wallis ANOVA between groups to test for age effects in Figure 1C. Our analysis indicates a non-linear increase of the mean transient frequency over development with a local minimum at around P10. This is now included in the figure legend and results. Thanks.

(3c) In Figure 2C, a sigmoidal fit is chosen, and in Figures 2B and 2D, no fitting is performed, though a linear relationship is mentioned for Figure 2B in the text.

We thank the reviewer and have now used a sigmoidal fit in Figure 2D (as in 2C) to demonstrate the increase around P9 of hippocampal activity during the period of mouse immobility. In Figure 2B we used one-way ANOVA to show a significant effect of age on the proportion of cells activated after movement and post hoc corrected multiple comparison tests to compare between age groups. Here, we show that the post movement activity was not different for all pairs of age groups from P10 to P12, nor between P8 and P9 while all other pairs of age groups were highly significantly different. In addition, we mentioned in the text that the difference in median between 2 consecutive age groups was similar from P5 to P9 and that the decrease was 4 times stronger between P9 and P10 than between any other pair. We have modified the text accordingly (lines 156-157).

(3d) A unified statistical approach to modeling changes in neural properties over time should instead be used. Justification for different fits should be provided for each dataset. For example, examining the data in Figure 1C, there appears to be a decrease in the number of transients at P9-P10, which subsequently increases through P11-P12. The data appear to be underpowered for use of an 8 group ANOVA given the small number of data points per group, and potentially significant non-linearities in the number of transients over time are therefore ignored. Similarly, there appears to be a peak in the number of cells imaged at P9, but no details of which statistical testing was used are provided and again, more rigorous statistical approaches are required to understand whether significant parameter changes are occurring over time.

See response 3b above.

(4a) PMTH plots require more rigorous and detailed quantitative analysis. In Figure 2A and Figure 3, these are portrayed as percentiles across the population. To properly evaluate such data, statistical methodology (for example, a shuffling procedure) should be employed to determine whether a significant modulation is present at each age and for different cell types.

We apologize for this confusion. We had used a shuffling procedure to assess whether post movement activation exceeded chance level. To clarify this, we have placed a star on the PMTHs illustrated in figures 2, 3, 4 when statistical significance level is reached. This procedure is presented in detail in the methods part (section Peri-Movement-Time-Histograms (PMTH)).

(4b) Modulation values could then be quantitatively compared (using statistical testing) between groups (e.g. P5-8 and P10-12 pyramidal cells and interneurons in Figure 3).

Regarding pyramidal cells at P5–8 vs P10-12: since the response at P5-8 is significantly higher than the 95th percentile from surrogate data and the response at P10-12 is significantly lower than the 5th percentile we can already conclude that the response of pyramidal cells is different between the two ages. Regarding interneurons, the point here was to say that they are significantly activated after movement in the two age groups which is shown by the fact that the response amplitude exceeds the chance level. We did not aim at comparing the modulation levels for interneurons at P5-8 compared to P10-12.

(4c) Furthermore, examining the PMTH plots in Figure 2, it appears that there may be a leftward shift in the peak after P9 (i.e. increased activity occurs after movement in P5-9, but then seems more centered around the time of movement, or even before the detected movement in P10-12). If this observation were to be supported statistically, it could suggest an important change in the flow of hippocampal-cortical activity around the time of movement.

As noticed by this reviewer, a brief increase in the proportion of active cells following movement can be seen in Figure 2 on P10-11 mouse pups. This increase does not pass the chance level and is completely suppressed in the following second leading to a significantly reduced fraction of active cells. This may be the sign of feedback inhibition. At P12, this small increase in the fraction of active cells can be observed approximately one second before the onset of movement, indicating that activity would start building up in CA1 prior to movement. A corollary discharge would increase activity prior to movement on a much shorter time scale. There is therefore no obvious explanation for this interesting phenomenon. Anticipatory cell firing prior to locomotion has been previously reported in the adult cortex (see for example Vinck et al. Neuron 2015). Different mechanisms could support such anticipatory firing, including the influence of top-down inputs, changes in arousal states or any complex neuromodulatory interactions possibly associated with changes in the sleep-wake cycle and that could involve, for example, the norepinephrine, serotonin or acetylcholine systems. Thus, this intriguing observation remains to be further explored in future studies.

(5a) The conclusions of the manuscript rely strongly on the differentiation of pyramidal cells and interneurons. Therefore, more details regarding the cell classifier (DeepCINAC) should be provided for this particular dataset. What features permit specific identification of interneurons, and are these features demonstrated to be constant across developmental age?

We thank the reviewer for bringing up this important point. All the details regarding our cell type classifier can be found in the publication cited (Denis et al. 2020) as well as online on our gitlab account (https://gitlab.com/cossartlab/deepcinac). The specific dataset used is this paper is actually the one we used to train and validate the DeepCINAC cell type classifier across ages from P5 to P12; thus, the performance of the classifier in terms of sensitivity, precision and F1 score is exactly as described in (Denis et al). The features used by the classifier to perform are not easily accessible from the artificial network and we have not investigated this specific question since the performance at the classification was 91 % precision for interneurons. The input to the neuronal network is a 100 frames long movie of 25 × 25 pixels window centered on the cell body of interest. Thus, we hypothesize that the network would use directly from movie visualization several parameters such as: neuronal shape, localization and activity. Whether all, several or only one parameter has been used by the network remains an open question.

(5b) In Figure 3-supplement 2, the average PMTHs for predicted vs. labelled interneurons are quite different; the peak percentage of active cells is ~10% for predicted interneurons, but ~25% for labeled interneurons. It also appears as though labeled interneurons have essentially no firing outside of movement-related activity in the surrounding 20-second window, in contrast to the predicted interneurons. These differences should be further explored and explained.

The difference in the peak percentage of active cells is simply a direct consequence of the relatively low number of labeled interneurons in each imaging session. For example, in an imaging session with a low number of labeled interneurons the median percentage of active labeled interneurons can only take discrete values, sometimes up to 100 %. Because we then represented for each time bin the median from each session median it could bias the percentage of active labeled interneurons toward low baseline and high response. To circumvent this limitation, we represent the activation of labeled interneurons using the median DF/F signal that is more representative in case of a low number of cells. The point of Figure 3 —figure supplement 1E was to show that labeled interneurons and inferred interneurons have the same relationship with movement. This is confirmed by the PMTH using the DF/F signal from these 2 groups.

(6) The extra-hippocampal synaptic afferents appear to be evaluated using structural connectivity metrics, and no substantial changes were observed. However, no functional assessment of these synapses was performed. The conclusion that the mechanism of hippocampal activity changes is a result of local GABAergic innervation should be tempered given this, and more clearly explained in the discussion.

We agree with reviewer 1 that we did not functionally assess the extrinsic inputs, given the experimental challenge to perform such experiments. This was acknowledged in the first section of the discussion (line 301) “This [activity change] is likely due to the time-locked anatomical and functional rise of somatic GABAergic activity given that interneurons remain highly active throughout this period, including in response to spontaneous movements.” We therefore had already tempered the conclusion that the mechanism of hippocampal activity changes is a result of local GABAergic innervation. This conclusion is softened even further in the discussion of the revised manuscript. That said, our modeling approach further supports the possibility that a rise in local GABAergic inhibition alone can account for the changes in CA1 dynamics reported here.

(7a) In Figure 4, GABAergic axon activity was determined at P9-10 only, but this was interpreted as an increase. Without comparison to other ages, this conclusion would need to be tempered.

We thank the reviewer for this comment that reveals some lack of clarity in the previous description of our experiments. Indeed, functional GABAergic activity was also assessed before P9, however, given that there are no GABAergic axons in the CA1 pyramidal layer at early stages (for both CCK cf. Morozov and Freund 2003, and prospective PV cells cf. Figure 4A,B), there is no signal to be measured either. We have now added a new figure (Figure 4 - figure supplement 1) to clarify this point. In agreement with our Syt2 longitudinal quantification, we show, using tdTomato expression in the Gad67cre driver mouse line, that GABAergic perisomatic innervation is only visible after p9. This matches as well, our attempted imaging experiments using axon enriched GCaMP in mice before P9.

As explained above, there are few, if any, GABAergic axons in the CA1 pyramidal layer before P9 (Figure 4 —figure supplement 1 specifically, Video 7). There is no signal, thus comparison is pointless. We have now clarified this point (lines 242-244), thanks.

(7b) Statistical evaluation of data in Figure 4B also does not directly support the authors' conclusions. They explain in the results text a sharp increase in labeling at P9, but their statistical testing does not show a significant difference between P7 and P9, which is the supposed inflection point. This point further illustrates the need for more rigorous and sensitive statistical testing for age-dependent changes.

We had observed a significant difference between the P7 and P11 age groups (2 age groups exactly centered around P9), which demonstrates an increase of PV coverage around P9. In fact, P9 being the inflexion point it is not surprising that it is neither different from P7 nor from P11 since it is precisely when axonal arborization is changing the most.

(8) For the modeling data provided in Figure 5, why do the effects of inhibition persist over up to 5 seconds? Such effects would seem to extend beyond the direct effects of monosynaptic perisomatic inhibition. Most such models are presented over a span of milliseconds, not seconds. The parameters governing these timings, as well as the reasoning behind the choice of parameters used to generate the model, require justification and would improve clarity.

We agree that the time scales used in our simulations need clarification. The fast excitatory and inhibitory synaptic time scales fall beyond the time resolution offered by GCamp6s. It is believed that they tend to produce irregular spiking as demonstrated by the balanced state models (van Vreeswijk and Sompolinsky, 1996). In our simulation of the leaky integrate and fire network, we account for this irregular spiking by providing a noisy, normally distributed, input to all the cells.

We believe the longer observed time scales (in the range of a few hundreds of ms to seconds) originate from slow synaptic transmission. In order to make that more precise, we have updated the synaptic time scales employed in our simulations to match those receptors (Destexhe, Mainen, and Sejnowski, 1994) and generated new panels accordingly in figure 5. This is now better clarified in the manuscript (lines 260-264).

(9) The authors demonstrate evidence for detachment of pyramidal cell activity from movements and suggest a transition to internal representations. However, the data presented are strongly focused on peri-movement epochs, without any indication of the characteristics of neural activity during non-movement epochs. Unless the authors consider pursuing further analysis of their non-movement epoch data, implications regarding internal models and plasticity should be perhaps removed from the discussion, and this limitation acknowledged.

We thank the reviewer for his/her helpful comment. In the abstract and discussion parts we have removed the statements regarding ‘internal models’.

Reviewer #2 (Recommendations for the authors):

– Table 1: it is not clear why some mice were used in some figures and not in others. Some further detail would be useful here.

We apologize for the lack of details regarding the criteria for inclusion of mice from imaging experiments. Due to technical problems during the recording of the mouse behavior (behavioral frames dropped at unknown timestamps) we were not able in some cases to realign mouse behavior with calcium imaging. We decided to exclude these imaging sessions from all movement-related analysis. This explains why, for instance, some animals from figure 1 were not included in figure 2. To clarify this point, we have added a sentence in the Methods section (DATA PREPROCESSING, behavior lines 535-538).

– Some further details on the CICADA analysis software would be very helpful, as this does not seem to have been published separately. How does this work, and why did the authors use it?

We agree with reviewer #2 and added further details on the CICADA analysis pipeline in the methods section (Data Analysis, Analysis of calcium imaging data in the NWB format using CICADA lines 722-730)

– It would be very helpful to add more details on those interneurons that were identified by the deep learning network. How do the authors justify using this method (as opposed to just looking for tdTomato labelled cells), and why was it necessary?

This method allows us to infer neuronal cell types (Interneurons or Pyramidal Cells) in wild type animals. This is useful and necessary as not all imaged mice were from transgenic animals with tdTomato expressed in GABAergic neurons.

What does 91% reliability refer to – false-positive or false-negative? The authors should provide a full breakdown of numbers of pyramidal cells misclassified as interneurons, and vice versa. Figure 3 Figure supp 2C shows a striking difference in the background activity levels of classified as opposed to labelled neurons – the authors should comment on why this is, and justify treating these cell types equally in spite of this.

We apologize for the confusion. 91% refers to the specificity of the classifier meaning the probability of being a GABAergic neuron when the classifier predicts it to be the case. This reviewer is concerned by the fact that the PMTHs plotting the proportion of active interneurons differ when they are labeled vs. inferred. The difference in background activity levels likely originates from the low number of labeled interneurons per imaging session. The point of Figure 3 —figure supplement 1E was to show that labeled and inferred interneurons had the same relationship with movement (i.e. an increased activation after movement). This is now further supported by a PMTH using the raw DF/F signals from these 2 groups (which are less sensitive to the number of cells). More details on the cell type classifier and on the identified interneurons are provided in Figure 3 —figure supplement 1.

– With respect to figure 3, the authors state that "the link between movement and activity evolves differentially towards the start of the second postnatal week when comparing pyramidal neurons and GABAergic interneurons, the former being inhibited or detached from movements while the latter remaining activated". Whilst this appears to be true from looking at average activity PMTH plots, there is no statistical quantification to demonstrate that this is the case. By quantifying either the peak and/or the trough of the PMTH, the authors should try to show a statistically significant interaction between how these develop with age and cell type.

There was actually a statistical quantification included in the PMTH plots. We used surrogates of neuronal activity to estimate the chance level. We showed that PMTH peaks in P5-8 mice are above the chance level for both interneurons and pyramidal cells, indicating significant co-activation, whereas in P10-12 mice PMTH peaks are below the chance level for pyramidal cells and remain above chance for interneurons. We did not specifically quantify differences in peak or trough values but rather demonstrated that these peaks/troughs are significantly above/below chance levels. This shows that “the link between movement and activity evolves differentially towards the start of the second postnatal week when comparing pyramidal neurons and GABAergic interneurons, the former being inhibited or detached from movements while the latter remaining activated". To further clarify this point, we now indicate with a star on the PMTHs from figures 2, 3, 4 and 5, cases when statistical significance level is exceeded. This analysis procedure is presented in detail in the methods part (section Peri-Movement-Time-Histograms (PMTH)).

– Figure 4D: in addition to the problems associated with having only one age point, this data also appears to show a striking dissociation between axonal and somatic interneuron activity: in the axonal trace (only) there is a significant dip below baseline following the movement peak. The authors should comment on the significance of this, and what it means for their conclusions.

As explained above, we have imaged activity in GABAergic axons only after P9, not due to a technical problem but rather because there is no signal to measure given that perisomatic innervation is absent before P9. We have dedicated a new supplementary figure (Figure 4 —figure supplement 1) to explain why we could not image GABAergic axons in the pyramidal cell layer at earlier developmental stages.

The apparent dissociation between axonal and somatic interneuron activity is in fact due to the method used to build the PMTHs. PMTHs on somatic interneuron activation (Figure 2-3) are built on inferred activity from DeepCICNAC classifiers and represent the fraction of active cells across time. The PMTH in Figure 4 corresponds to the median DF/F signal centered on the onset of movement. It is not expected that these two PMTH should have the exact same shape.

We now also provide somatic PMTHs using DF/F signals. Figure 3 —figure supplement 2B illustrates the somatic response of interneurons to movement at P10-12. This time, the shape of the PMTH nicely matches the one observed for axonal imaging in Figure 4D. We thank the reviewer for pointing out this possible source of confusion.

Reviewer #3 (Recommendations for the authors):

(1) On line 52 the authors build the argument that feedback connections are important for self-organisation of the internal state. Further, that is likely GABAergic in nature in CA1 due to the paucity of recurrent glutamatergic connections onto pyramidal projections neurons. The authors then show some evidence of the strengthening of the perisomatic connections (Figure 4), but – as they acknowledge in the discussion (lines 478-490) – could not strengthen the JEI connection as shown in their model (Figure 5A) be equally important? Is it possible to use their existing imaging data to test if more GABAergic interneurons respond to movement at P10-12 in the data shown in Figure 3?

We thank the reviewer for an excellent suggestion. Using our dataset, we have quantified the differences between the peak and baseline values in the PMTHs of active interneurons at P5-8 and P10-12 and found no significant difference (P5-8=5.7%; P10-12=6.1%). This suggests that the same amount of interneurons relative to baseline were recruited when the animal was moving at both stages.

And assess the timing of the interneuron activity relative to the pyramidal cells?

The timing of interneurons activity relative to pyramidal cells is visible in the cross-correlogram plots (Figure 5C). It shows that at P9-12, pyramidal cells are inhibited following interneurons activation.

(2) It would be interesting to see the primary imaging data for that shown in Figure 2A i.e. deltaF/F of cells following movement.

We agree with the reviewer and now provide the PMTHs built using the DF/F signal in addition to the ones built using inferred activity, whenever possible (Figures 2,3). These results are now illustrated in Figure 2 Figure supplement 1, Figure 3 Figure supplement 1 and Figure 3 Figure supplement 2.

(3) Line 155, the authors define myoclonic twitches versus movements associated with wakefulness. In the methods (line 791), the authors further delineate "unclassified movements". Why are these not included in the results?

We apologize for this lack of clarity. Movements referred to as ‘unclassified’ were indeed included in the main analysis. They were only excluded when analysis was restricted to ‘twitches’ or to ‘complex movements’. We have now clarified the definition of movements in the methods section (DATA PREPROCESSING, behavior lines 636-643)

Further, it is a shame that the authors only did one recording with nuchal EMG to better separate active sleep (AS is perhaps a better descriptor as opposed to REM) versus wakefulness. Again, it would be interesting to see the raw data from this animal.

Combining nuchal EMG and 2p imaging in young mice is extremely difficult. As a result, we had to perform 8 experiments for one animal meeting our standard criteria for EMG and imaging experiments. To address reviewer #3 concern, we now present the result obtained on the raw DF/F for this mouse (Figure 2 Figure supplement 1C).

(4) Do the authors think that the drop in peak %active cells from P8 to P9 (Figure 2A) is important?

Yes, we think that this drop in the peak of percentage of active cells is important. CA1 dynamics switch from being mainly driven by bottom up inputs to be inhibited by these inputs. Interestingly, in adults it is known that again CA1 activity increases when the mouse moves. There is therefore a transient inhibitory relationship between movement and hippocampal activity that is likely to carry a developmental function.

(4b) Could not a sigmoidal function be fitted to Figure 2D and if so, what would the R(2) value be?

We would like to thank this reviewer for an excellent suggestion. We were indeed able to fit a sigmoidal function to the data presented in Figure 2D (r2 = 0.55).

(5) Line 368, it is not immediately clear to me why an increase in feed-forward inhibition leads to a detachment from external inputs. In primary sensory areas, an increase in feed-forward inhibition is observed in line with the emergence of fast sensory processing (see, for example, Chittajallu and Isaac, 2010).

We agree with this reviewer that “detachment from external inputs” might not be the only interpretation of our observations. We used this terminology, inspired by this review (Buzsaki et al. Emergence of cognition from action, 2014) and also because, as discussed above, we know that P9 opens a transient period of quiescence in CA1 with locomotion, in contrast to sensory processing in cortical areas, which remains under the tight control of inhibition.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

(1) The manuscript should be clearer regarding the potential consequences of the experiments on the monitored features, especially when it comes to ipsi-contralateral differences. Reviewer #1 has raised again several concerns in this regard, from differences in electrophysiological patterns to possible differences in left-right twitches and how they specifically affect the recorded hemisphere.

(2) Reviewer #2 still has concerns regarding the statistics of PMTHs and requires further explanation.

Reviewer #1 (Recommendations for the authors):

The authors have addressed many of the points raised. Points requiring additional modification/clarification are documented below.

1. Initial Review – From a methodological standpoint, sufficient data to support the conclusion that the invasive neurosurgical procedure used to perform hippocampal imaging does not disrupt network function is not presented.

Author Response – We gently disagree with this reviewer given that, as previously indicated in the text and illustrated in Figure1 —figure supplement 1C, we had performed electrophysiological recordings in ipsi- and contralateral hippocampi and observed that hippocampal oscillations in the form of eSWs were present in both hemispheres, at several ages.

Reviewer Response – Demonstrating the presence of one type of oscillation that actually shows a trend toward a decreased occurrence rate in the surgically manipulated hemisphere, in my opinion, is not sufficient to claim that hippocampal network function is unaffected. The work that the authors perform below (with caveats also noted below), however, helps to provide support for at least a minimal disruption of the network.

We agree with reviewer #1 and modified the manuscript accordingly. We thus reworded line 104 “the window implant preserved the electrophysiological network patterns” by “the window implant minimally disrupted the electrophysiological network patterns”. We also added in the Results section line 92-94: “This slight and non-significant reduction in eSW frequency in the ipsilateral hemisphere was comparable with previous study using the same surgical approach (see Discussion, Graf et al., 2021).”

2. Initial Review – An estimated 7mm3 of cortical tissue of the ipsilateral somatosensory cortex is aspirated to allow insertion of the window implant. The authors do not present any data documenting the physiologic recovery of the pup from the surgery or the quantitative stability of active cells over the recording period. Such data would be important to ensure that pups of all ages had similarly recovered from the procedure and that age-dependent recovery processes do not contribute to the observed results.

Author Response – We thank the reviewer for pointing out this important issue. We have performed simultaneous CA1 dynamics and EMG recordings in two animals. We found that craniotomy did not alter the structure of the sleep-wake cycle, as revealed by the quantification performed in Figure 1—figure supplement 1B. As previously described in Jouvet et al. 1967, P5-6 mice spend 70-80% of their time in active sleep, which is in agreement with our experiments. In addition, to probe the stability of active cells over our recording periods, we have computed the difference in the frequency of calcium transients between the first and last quarters of the recordings. We found that single-cell activity was stable throughout our recordings, for all ages recorded (see Author response image 1, median difference = 0,08 transient/mins).

Reviewer Response – The quantification of the sleep-wake cycle as performed in figure supplement 1B is inconsistent with currently proposed methods to do so. Indeed, the authors reference a paper from 1967 to support their work. Previously, sleep was suggested to be initiated in an undifferentiated mixed state (1,2), but more recently, quiet and active sleep is thought to be present from birth (3, 4, 5). The current figure shows "twitches", "REM", "transition" and "awake", which are not consistent with either classification scheme. This analysis should be redone in a manner consistent with a classification scheme to enable comparisons with other current literature in the field.

We thank the reviewer for his/her comment. As presented in the method section (lines 708-732) the classification of animal states was similar to the one used in Rio-

Bermudez et al. 2015. To perfectly match this nomenclature, we renamed the “transition” state to “quiet sleep” (Figure 1 —figure supplement 1B). Of note, twitches were not defined as a stage but just detected events during REM/active sleep (lines 729-730).

The quantification of the calcium transients over the course of the recording does seem convincing, and I would suggest that the authors include this figure in their supplementary data.

We have now included this figure in Figure 1 —figure supplement 1E and mention it in the Results section lines 115-116 : “Neuronal activity was stable over the duration of the recording (Figure 1 —figure supplement 1D, E – median change 0.08 transients/minute, N=31 pups).”

3. Initial Review – Furthermore, bilateral silicon probe recordings are used to suggest that ipsilateral and contralateral hippocampal network activity is comparable, but these data are difficult to interpret. Rates of eSW are presented and labeled as not statistically different between the bilateral hippocampi, but 5/6 pups show an increase in eSW rate for the contralateral hippocampus and it is concerning that the sample size is underpowered to detect what could be physiologically meaningful differences.

Author Response – This question on the rate of eSW raised here by the reviewer was addressed in the public review #1.

Reviewer Response – The authors have decided to acknowledge a likely difference due to reduced sensory feedback, which is reasonable, but this is not reflected in the manuscript as the statement is made "In the same way, the acute window implant did not significantly alter electrophysiological network patterns." (lines 89-90) and no mention is made in this section of the point. The Results section referenced speaks to differences in movements rather than activity patterns.

We thank reviewer #1 for his/her comment (see first Author re-Response). We now clearly acknowledge these differences in the discussion part (lines 323-330): “Of note, it is important to keep in mind that part of the overlying cortex, including the primary sensory cortex, was removed to grant optical access to the hippocampus. This region may contribute in relaying the sensory feedback from the twitches to the hippocampus (Khazipov and Milh, 2018; Valeeva et al., 2019a). The surgical procedure may thus damage incoming axons from the temporoammonic track linking the entorhinal cortex to the hippocampus. Accordingly, we observed: (i) that the CA1 response to movements from the contralateral limbs was slightly reduced; (ii) that the eSW frequency in the ipsilateral hemisphere was slightly diminished; (iii) a small increase in the power spectra of network oscillations below 20Hz”

4. Initial Review – The correlogram of eSW activity is also not centered around zero, suggesting that one hippocampus is consistently leading the other; which hippocampus (ipsilateral or contralateral) serves as the reference for this graph is not detailed in the figure or legend, but this could also indicate a difference in the function of the surgically altered hippocampus.

Author Response – We thank the referee for making this important observation and apologize for the lack of clarity. The onsets of eSWs recorded in the ipsilateral hippocampus served as reference. As the referee noted, the two hippocampi are not perfectly synchronous with eSWs from the contralateral hippocampus leading the ipsilateral one by 12 ms. Such delay could be explained by a slight drop in local temperature due to the chamber placement as described previously for cortical up-states (Reig et al., 2010).

Reviewer Response – In the response here, the authors explain the finding by citing Reig et al., 2010, but in the manuscript, they cite this difference as being expected as per Graf et al., 2021; Valeeva et al., 2019b (lines 97-98). This is confusing and should be clarified.

We clarified this point (lines 92-97), thank you. (i) Reference Graf et al., refers to the expected slight decrease in the eSW frequency in the ipsilateral hemisphere. (ii) Reference Valeeva et al., refers to the expected synchrony in the eSW in the two hemispheres. (iii) Reference Reig et al., may provide an explanation for the delay that we observed in our experiments.

5. Initial Review – To my reading, it is also unclear what data are used for the calculation of the hippocampal power spectra, and no statistical comparison between ipsilateral and contralateral hippocampi is actually performed to support similarity.

Author Response – We thank the reviewer for bringing up this important point. To characterize the developmental changes in the theta band power we have performed spectral analysis using the Chronux toolbox on the entire recordings (Chronux toolbox J Neurosci Methods. 2010 Sep 30;192(1):146- 51). Spectral power was estimated using direct multi-taper estimators (3 time-bandwidth products and 5 tapers). The shaded area corresponds to the confidence interval (p>0.05). As requested by this referee, we have updated the figure and added comparison between ipsi- and contralateral hippocampi. Our results demonstrated the similarity in the spectral distribution (see peak in the theta frequency band in P11 animals, which is not present in younger animals). Surprisingly, we found that the spectral power was slightly higher for the ipsilateral hippocampus.

Reviewer Response – These statistics are helpful. What is the proposed reason for this difference in spectral power? Taken together, it seems as though there are several indicators that the ipsilateral hippocampus is different in its activity than the contralateral hippocampus (trend toward different eSW rate, consistent lead/lag in regard to eSW activity, and significant differences in power spectra). These points should be mentioned in the manuscript text for transparency even if the authors argue that they do not affect the primary conclusions of the work.

We agree with reviewer #1 and have now mentioned all these points in the manuscript (see discussion, lines 323-330, results line 102).

6. Initial Review – Furthermore, does aspiration of the ipsilateral somatosensory cortex affect spontaneous twitches/waking movements or the hippocampal-cortical response? For instance, is there a difference in response to body movements corresponding to the limbs contralateral to the window (cortex responding to the movements has been aspirated) vs. limbs ipsilateral to the window (cortex responding to the movements is intact)?

Author Response – We thank the reviewer for this important comment. To address it, we have quantified in 6 imaging sessions from 5 animals (aged between P5 and P8) the CA1 response to body movements from the limbs contralateral to the imaging window (right) vs. limbs ipsilateral to the window (left). We found that movements from the right limbs elicited a slightly reduced response than the left limb movements. Regarding the number of twitches and waking movement of the right limbs compared to the left, we found fewer twitches on the contralateral side (right) as compared to the ipsilateral one (See Author response image 2). This could result from a partial lesion of the ipsilateral motor cortex in our experimental conditions. Related to this point, we would like to mention that the motor cortex, before P12, is more passively driven by spontaneous twitches than involved in driving them (Dooley Blumberg, 2018), but that up to 24 % of the twitches in P3-5 rats could already be originating from that region (An, Luhmann 2014). Our data is somehow consistent with that observation. This is now included in the manuscript but not studied any further as it is beyond the scope of the present paper.

Reviewer Response – This quantification is also informative. Although the difference is now shown in the supplementary figure, the point is to ensure that this difference does not affect the conclusions of the present study, not to study the changes themselves. For instance, it would be important to know if only the ipsilateral limb twitches are used (routed through the intact contralateral somatomotor region), is the trend in cell recruitment over development different?

We have performed an additional set of analyses that confirms that the trend in the cell recruitment over development is similar when considering separately ipsi (left) and contra-lateral (right) body twitches, with the emergence of inhibition around P9 (Author response image 3) . For this, we used the already annotated twitches for two P5 and P7 pups and manually annotated randomly selected twitches in one P9 and one P12 mouse. Of note the manual annotation of all of the twitches at P9 and 12 would take several weeks. We hope that this reviewer will now be convinced that the trend reported in Figure 2A with combined twitches is also valid when these are treated separately.

Author response image 3
PMTHs for P5, 7, 9 and P12 pups built on twitches only.

N represents the number of animals. n represents the number of imaging sessions.

7. Initial Review – State-dependent parameters are not adequately described, controlled, and examined quantitatively to ensure that data from similar behavioral states is being used for analysis across ages. Network activity from wakefulness, REM/active sleep, and NREM/quiet sleep should not be presumed to be indistinguishable.

Author Response – We would like to point out that our analysis across ages focused on the population response following animal movements, and not across all behavioral states. That said, it is true that two types of movements can be distinguished, namely the twitches and the complex ones. To take this behavioral heterogeneity into account, we have now separately quantified the hippocampal activation following twitches (movement during active sleep) and complex movement (during wakefulness). We show in Figure 2 —figure supplement 1B that the hippocampal response to twitches and complex movements is similar across ages. Thus, even if the amount of time spent in each behavioral state is modified over the developmental period that we have studied, we are pretty confident that it does not impact the transition we have described in the relationship between animal movements and hippocampal activity. Additionally, we were able to combine one P5 mouse pup 2p-imaging with nuchal EMG recordings and separately computed the PMTH for movements observed during REM or wakefulness (Figure 2 —figure supplement 1C). We show that CA1 hippocampal neurons were activated and time-locked to movement in both behavioral states, with only the amplitude of the population response differing between wakefulness and during REM. This point is now included in the result section (lines 148-152) and discussed (lines 324-327).

Reviewer Response – I am a bit unclear as to the conclusion here. In figure supplement 1B, the hippocampal response to twitches and complex movements is not different, but in figure supplement 1C, the hippocampal response is different between "REM" and wakefulness. Is the conclusion that the movement semiology-based analysis doesn't show differences, but the state-based analysis does (2x difference in population response), raising the concern that the semiology-based analysis proposed is not actually capturing state-based differences? The relationship between the two figure supplements should be clarified. Also, "REM" is not typically ascribed to a P5 pup (would usually be called active sleep).

We apologize for the confusion in this figure. It is true that Figure 2 —figure supplement 1B and 1C display some differences in the cell recruitment during twitching and in active/REM sleep movements that are mainly twitches as well. This can be explained by the number of animals used to build each figure. For figure 2 —figure supplement 1B we had pooled multiple animals (e.g., P5 represents n=4 pups and 8 imaging sessions). However, Figure 2 —figure supplement 1C shows the PMTHs for REM and Awake only for one P5 mouse pup.

In Author response image 4 we confirm that comparing twitches and complex movements gives the same result as comparing REM/active sleep movements and awake movements in the same mouse pup compare Figure 2 —figure supplement 1C top panels with (Author response image 4). This confirms that the twitches and movements identified by manual annotation of the video recording captures a state-based difference between REM and AWAKE movement when there is one.

Author response image 4
PMTHs for P5 pup on twitches and complex movements (same pup as the one used in Figure 2 —figure supplement 1C to compare REM/ active sleep movements and awake movements).

Additionally, we have now clarified the manuscript (see lines 152-158): “Accordingly, when combining calcium imaging with nuchal EMG recordings in one P5 mouse pup, we observed an increase in the percentage of active cells and in the DF / F fluorescence signal following movements occurring both during REM sleep and wakefulness (Figure 2 —figure supplement 1C). However, when combining all mouse pups, and considering separately twitches (occurring during REM/active sleep) and complex movements (occurring during wakefulness), based on video recordings, we found that the two movement types did not significantly differ in their impact on CA1 activity (Figure 2 —figure supplement 1B)”.

8. Initial Review – It seems as though a linear correlation coefficient is used to quantify data in Figure 1B, but Kruskal-Wallis ANOVA between groups is used in Figure 1C.

Author Response – To address this point, we now perform a nonlinear fit instead of the Kruskal-Wallis ANOVA between groups to test for age effects in Figure 1C. Our analysis indicates a non-linear increase of the mean transient frequency over development with a local minimum at around P10. This is now included in the figure legend and results. Thanks.

Reviewer Response – This is helpful. Typically, would be appropriate to mention somewhere (perhaps in the Methods) how the fit was selected (e.g. minimizing MSE or some other method). An r2 value is not typically valid statistically for a non-linear regression (mentioned as "r2=0,30" in the figure legend at present).

The fourth order polynomial fit in Figure 1C uses least squares method. This is now mentioned in the figure legend (line 808).

9. Initial Review – In Figure 2C, a sigmoidal fit is chosen, and in Figures 2B and 2D, no fitting is performed, though a linear relationship is mentioned for Figure 2B in the text.

Author Response – We thank the reviewer and have now used a sigmoidal fit in Figure 2D (as in 2C) to demonstrate the increase around P9 of hippocampal activity during the period of mouse immobility. In Figure 2B we used one-way ANOVA to show a significant effect of age on the proportion of cells activated after movement and post hoc corrected multiple comparison tests to compare between age groups. Here, we show that the post-movement activity was not different for all pairs of age groups from P10 to P12, nor between P8 and P9 while all other pairs of age groups were highly significantly different. In addition, we mentioned in the text that the difference in the median between 2 consecutive age groups was similar from P5 to P9 and that the decrease was 4 times stronger between P9 and P10 than between any other pair. We have modified the text accordingly (lines 156-157).

Reviewer Response – This is helpful. Typically, would be appropriate to mention somewhere (perhaps in the Methods) how the sigmoidal fit was selected (e.g. minimizing MSE or some other method). Again, an r2 value is not typically valid statistically for non-linear regression.

The sigmoïdal fit in Figure 2C,D uses least squares method. This is now mentioned in the figure legend (lines 837 and 842).

10. Initial Review – A unified statistical approach to modeling changes in neural properties over time should instead be used. Justification for different fits should be provided for each dataset. For example, examining the data in Figure 1C, there appears to be a decrease in the number of transients at P9-P10, which subsequently increases through P11-P12. The data appear to be underpowered for use of an 8-group ANOVA given the small number of data points per group, and potentially significant non-linearities in the number of transients over time are therefore ignored. Similarly, there appears to be a peak in the number of cells imaged at P9, but no details of which statistical testing was used are provided and again, more rigorous statistical approaches are required to understand whether significant parameter changes are occurring over time.

Author Response – See response 3b above.

Reviewer Response – See reviewer response to 3b above.

See Re-response to point #8.

11. Initial Review – In Figure 4, GABAergic axon activity was determined at P9-10 only, but this was interpreted as an increase. Without comparison to other ages, this conclusion would need to be tempered.

Author Response – As explained above, there are few, if any, GABAergic axons in the CA1 pyramidal layer before P9 (Figure 4 —figure supplement 1 specifically, supplementary video 7). There is no signal, thus comparison is pointless. We have now clarified this point (lines 242-244), thanks.

Reviewer Response – Thanks for clarifying this issue. To ensure that no pointless comparisons are inferred by the reader, it may be of benefit to change the wording of lines 255-256 from "increase in functional perisomatic GABAergic activity" to "emergence of functional perisomatic GABAergic activity."

We changed the manuscript accordingly see lines 259-260.

12. Initial Review – Statistical evaluation of data in Figure 4B also does not directly support the authors' conclusions. They explain in the results text a sharp increase in labeling at P9, but their statistical testing does not show a significant difference between P7 and P9, which is the supposed inflection point. This point further illustrates the need for more rigorous and sensitive statistical testing for age-dependent changes.

Author Response – We had observed a significant difference between the P7 and P11 age groups (2 age groups exactly centered around P9), which demonstrates an increase in PV coverage around P9. In fact, P9 being the inflection point it is not surprising that it is neither different from P7 nor from P11 since it is precisely when axonal arborization is changing the most.

Reviewer Response – The statistical interpretation is still problematic here. If the statistical comparison between the P7 and the P9 groups is not significant, it is not accurate to report in the results that "However, around P9, a sudden increase in the density of positive labeling was observed" (lines 240-241). The reasoning about P9 being an inflection point may be the case, but the currently employed statistics do not show it. If the authors wanted to show there is an inflection point, rather than just stating what their statistics show (a significant difference between P7 and P11), a modeling approach that demonstrates a change in the curvature of the modeled fit around P9 would be appropriate.

We changed the manuscript accordingly. See lines 245-247.

Reviewer #2 (Recommendations for the authors):

Overall the authors have done a good job of responding to many of the points raised. However, I feel they have not provided a complete answer to three key points, raised by both myself and reviewer 1.

1) Statistics for PMTHs. Thanks to the authors for clarifying that there was already shuffled significance criteria in place for the PMTHs and the addition of an asterisk to the plots helps to clarify this. However, I still do not feel that the explanation as it stands is fully satisfactory.

a) It is necessary to provide more details of the shuffling procedure. The methods state simply that '500 surrogate raster plots per 770 imaging session were computed', which is insufficient detail. Unless the authors can provide a good reason otherwise, I think that the surrogates should be 20sec sections of data, drawn at random from the entire dataset, from both movement and non-movement epochs. If surrogates are not defined as above, the authors should justify why. In any case, more detail should be provided in the methods.

We would like to thank reviewer 2 for pointing out the lack of details present in the method section. To obtain a surrogate raster plot, the activity of each cell was translated by a randomly selected integer (between 1 and the total number of frames). This procedure has two consequences: it conserves the inter-transient time interval distribution y for each cell while disorganizing population activity.

Chance level PMTHs were built from these surrogate raster plots using the timestamps of movement onsets observed in the data. This allowed us to test the hypothesis that movement triggers a population response by synchronizing CA1 hippocampal cells. The reviewer proposes to use 20sec sections of data, drawn at random from the entire dataset. To our understanding this would be equivalent to selecting random movement onsets and conserving population activity. This would thus test a different hypothesis, namely that population synchrony could happen independently from movement and that, by chance, movement would precede SCEs. This was already tested at single-cell level in Figure 2 —figure supplement 1E where we found that 60% of the cells recorded at P5 are more active than expected by chance during movement epochs. We have now clarified the method section accordingly (see lines 769-771)

b) Are the 95th and 5th percentiles calculated with respect to each bin, or the entire 20s PMTH? If the former, then the authors need to account for the problem of multiple comparisons across the multiple time bins of the PMTH.

We would like to thank reviewer 2 for pointing out the lack of details present in the method section. To obtain a surrogate raster plot, the activity of each cell was translated by a randomly selected integer (between 1 and the total number of frames). This procedure has two consequences: it conserves the inter-transient time interval distribution y for each cell while disorganizing population activity.

Chance level PMTHs were built from these surrogate raster plots using the timestamps of movement onsets observed in the data. This allowed us to test the hypothesis that movement triggers a population response by synchronizing CA1 hippocampal cells. The reviewer proposes to use 20sec sections of data, drawn at random from the entire dataset. To our understanding this would be equivalent to selecting random movement onsets and conserving population activity. This would thus test a different hypothesis, namely that population synchrony could happen independently from movement and that, by chance, movement would precede SCEs. This was already tested at single-cell level in Figure 2 —figure supplement 1E where we found that 60% of the cells recorded at P5 are more active than expected by chance during movement epochs. We have now clarified the method section accordingly (see lines 769-771).

2) Statistics in addition to the PMTHs. Both Reviewer 1 and myself requested additional statistical analyses of the %cell activation data, in particular comparisons of how activation peaks and inhibition troughs evolve over development. The current argument of the authors is that it is sufficient to rely on whether PMTH activity crosses a significance threshold, or not, as a form of cross-age comparison. I don't agree with this – to take an example, the movement-activated activity peaks at P8 and P10 look visually very similar (Figure 2A). The proper statistical approach to test whether they are different is to compare them directly. Using the author's current approach, two very similar samples, not significantly different in themselves, could be judged to be from different populations based on one crossing an arbitrary shuffling-based threshold, and the other not.

We gently disagree with reviewer 2. As quantified in Figure 2B, the post movement activity in P8 mouse pups is significantly higher than the one in P10 mice. This is further confirmed by Figure 2C showing that most movements at P10 are followed by a reduction of activity whereas most movements at P8 are followed by an increase of activity. This allows us to directly compare the PMTHs presented in Figure 2A. Overall, even if PMTHs at P8 and P10 seem visually very similar to the reviewer, these two quantifications (Figure 2 B,C) show that they are not. In fact P10 PMTH is visually more similar to the one observed at P11 than to the P8, which is confirmed by Figures 2 B,C. Of note, reviewer #1 is now convinced by all of the changes made (see point 8 above).

The authors already have a model for better analysis in the manuscript – the analysis of activity troughs in Figure 2B. I think that a proper analysis of the data requires this method to be applied to movement-related peaks and post-movement throughs, for all data in figures 2, 3, and 4.

This method is already applied to all PMTHs included in Figure 2A and cannot be directly applied to Figure 4D that uses DF / F signal and not a percentage of active cells. Figure 3 shows the PMTHs for interneurons and pyramidal cells before P9 (P5-8) and after (P1012). The quantification presented in Figure 2B based on Figure 2A data was useful to quantify subtle differences in peaks and troughs between PMTHs (such as the P8 vs P10 example pointed out by the reviewer). In Figure 3, among the four PMTHs, three (interneurons P5-8 and P10-12 and pyramidal cells P5-8) show a significant increase and 1 (pyramidal cells P10-12) a significant decrease. Our only claim is that the activity of the two neuronal populations evolves differentially with respect to movement. Because these two very different PMTHs types cross the statistical threshold in opposite directions we do not think the data requires the Figure 2B method to be applied to movement-related peaks and post-movement throughs in this case (in agreement with reviewer #1 comment, see point #8). We agree with reviewer # 2 that quantification of peaks with this method could be useful to directly compare interneurons response to movement in P5-8 with P10-12 but this claim is outside the scope of the present study.

3) Thanks to the clarification from the authors, I now understand and accept that there are no axon terminals to image before P9. However, some more temporal precision regarding the emergence of perisomatic axon terminal activity would be helpful. The key transition dates for pyramidal cell activity run from P8 (still immature), P9 (which is transitional – no activity peak but also no trough) to P10 (mature activity). How does the emergence of axonal activity relate to this timeline? Is there a difference between P9 and P10? Is the response already mature at its first emergence (at P9)? Or does it continue to gradually increase between P9 and P12? This information would help make a more specific link between increases in perisomatic inhibition and PC activity.

Unless the authors can show these data, then phrases such as 'functional surge' should be avoided ('surge' implies a rapid maturation, which cannot the demonstrated using one time point), the authors should restrict their conclusions to stating that 'functional perisomatic activity from inhibitory interneurons can be observed at P9-10 (or similar).

We have changed the manuscript accordingly see lines 259-260.

https://doi.org/10.7554/eLife.78116.sa2

Article and author information

Author details

  1. Robin F Dard

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
  2. Erwan Leprince

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Julien Denis

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0537-6483
  4. Shrisha Rao Balappa

    Turing Centre for Living systems, Aix-Marseille University, Université de Toulon, CNRS, CPT (UMR 7332), Marseille, France
    Contribution
    Software, Formal analysis
    Competing interests
    No competing interests declared
  5. Dmitrii Suchkov

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Formal analysis, Methodology
    Competing interests
    No competing interests declared
  6. Richard Boyce

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Data curation, Formal analysis, Methodology
    Competing interests
    No competing interests declared
  7. Catherine Lopez

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Methodology
    Competing interests
    No competing interests declared
  8. Marie Giorgi-Kurz

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Methodology
    Competing interests
    No competing interests declared
  9. Tom Szwagier

    1. Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    2. Mines ParisTech, PSL Research University, Paris, France
    Contribution
    Software
    Competing interests
    No competing interests declared
  10. Théo Dumont

    1. Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    2. Mines ParisTech, PSL Research University, Paris, France
    Contribution
    Software
    Competing interests
    No competing interests declared
  11. Hervé Rouault

    Turing Centre for Living systems, Aix-Marseille University, Université de Toulon, CNRS, CPT (UMR 7332), Marseille, France
    Contribution
    Software, Formal analysis, Validation, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4997-2711
  12. Marat Minlebaev

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Formal analysis, Investigation, Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0722-7027
  13. Agnès Baude

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Data curation, Investigation, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7025-364X
  14. Rosa Cossart

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Validation, Investigation, Visualization, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    rosa.cossart@inserm.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2111-6638
  15. Michel A Picardo

    Turing Centre for Living systems, Aix-Marseille University, INSERM, INMED U1249, Marseille, France
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    michel.picardo@inserm.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1198-3930

Funding

European Resuscitation Council (646925)

  • Rosa Cossart

Fondation Bettencourt Schueller

  • Rosa Cossart

Neurodata Without Borders (R20046AA)

  • Michel A Picardo

Agence Nationale de la Recherche (ANR-16-CONV-0001)

  • Dmitrii Suchkov
  • Rosa Cossart

Ministère de l'Education Nationale, de l'Enseignement Superieur et de la Recherche (MESR)

  • Robin F Dard
  • Erwan Leprince

Fondation pour la Recherche Médicale (FDT202106012824)

  • Robin F Dard

Fondation pour la Recherche Médicale (FDM20170638339)

  • Julien Denis

Fondation pour la Recherche Médicale (ARF20160936186)

  • Michel A Picardo

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by the European Research Council under the European Union’s, Horizon 2020 research and innovation program grant# 646925, the Fondation Bettencourt Schueller, and an NWB seed grant# R20046AA. The project leading to this publication has received funding from the «Investissements d'Avenir» French government program managed by the French National Research Agency (ANR-16-CONV-0001) and the Excellence Initiative of Aix-Marseille University – A*MIDEX. RFD was funded by the “Ministere de l’Enseignement Supérieur, de la Recherche et de l’Innovation” and the Fondation pour la Recherche Médicale Grant FDT202106012824. EL was funded by the “Ministere de l’Enseignement Supérieur, de la Recherche et de l’Innovation” and A*Midex foundation and the French National Research Agency funded by the French government «Investissements d’Avenir» program (NeuroSchool, nEURo*AMU, ANR-17-EURE-0029 grant). JD was supported by the Fondation pour la Recherche Médicale Grant FDM20170638339. MP was supported by the Fondation pour la Recherche Médicale Grant ARF20160936186. TD and TS were funded by «Investissements d'Avenir» French government program managed by the French National Research Agency (ANR-16-CONV-0001) and from the Excellence Initiative of Aix-Marseille University – A*MIDEX. We would like to thank Dr. Pierre-Pascal Lenck-Santini for providing valuable feedback on our research project. We thank S Pellegrino-Corby, F Michel, and S Brustlein from the INMED animal and imaging facilities (InMagic). We would also like to thank Marion Sicre for her help with GAD1Cre/+ experiments. We are grateful to Pr. Hannah Monyer for providing the GAD1Cre/+ mouse lines. We thank the Centre de Calcul Intensif d’Aix-Marseille for granting access to its high-performance computing resources. The rabies virus was a gift from Conzelman laboratory.

Ethics

All experiments were performed under the guidelines of the French National Ethics Committee for Sciences and Health report on 'Ethical Principles for Animal Experimentation' in agreement with the European Community Directive 86/609/EEC (Apafis #18-185 and #30-959).

Senior Editor

  1. Laura L Colgin, University of Texas at Austin, United States

Reviewing Editor

  1. Adrien Peyrache, McGill University, Canada

Reviewer

  1. Simon JB Butt, University of Oxford, United Kingdom

Version history

  1. Preprint posted: June 9, 2021 (view preprint)
  2. Received: February 23, 2022
  3. Accepted: July 19, 2022
  4. Accepted Manuscript published: July 20, 2022 (version 1)
  5. Version of Record published: August 8, 2022 (version 2)
  6. Version of Record updated: August 10, 2022 (version 3)

Copyright

© 2022, Dard et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,596
    Page views
  • 299
    Downloads
  • 8
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Robin F Dard
  2. Erwan Leprince
  3. Julien Denis
  4. Shrisha Rao Balappa
  5. Dmitrii Suchkov
  6. Richard Boyce
  7. Catherine Lopez
  8. Marie Giorgi-Kurz
  9. Tom Szwagier
  10. Théo Dumont
  11. Hervé Rouault
  12. Marat Minlebaev
  13. Agnès Baude
  14. Rosa Cossart
  15. Michel A Picardo
(2022)
The rapid developmental rise of somatic inhibition disengages hippocampal dynamics from self-motion
eLife 11:e78116.
https://doi.org/10.7554/eLife.78116

Further reading

    1. Neuroscience
    Stijn A Nuiten, Jan Willem de Gee ... Simon van Gaal
    Research Article

    Perceptual decisions about sensory input are influenced by fluctuations in ongoing neural activity, most prominently driven by attention and neuromodulator systems. It is currently unknown if neuromodulator activity and attention differentially modulate perceptual decision-making and/or whether neuromodulatory systems in fact control attentional processes. To investigate the effects of two distinct neuromodulatory systems and spatial attention on perceptual decisions, we pharmacologically elevated cholinergic (through donepezil) and catecholaminergic (through atomoxetine) levels in humans performing a visuo-spatial attention task, while we measured electroencephalography (EEG). Both attention and catecholaminergic enhancement improved decision-making at the behavioral and algorithmic level, as reflected in increased perceptual sensitivity and the modulation of the drift rate parameter derived from drift diffusion modeling. Univariate analyses of EEG data time-locked to the attentional cue, the target stimulus, and the motor response further revealed that attention and catecholaminergic enhancement both modulated pre-stimulus cortical excitability, cue- and stimulus-evoked sensory activity, as well as parietal evidence accumulation signals. Interestingly, we observed both similar, unique, and interactive effects of attention and catecholaminergic neuromodulation on these behavioral, algorithmic, and neural markers of the decision-making process. Thereby, this study reveals an intricate relationship between attentional and catecholaminergic systems and advances our understanding about how these systems jointly shape various stages of perceptual decision-making.

    1. Neuroscience
    Manfred G Kitzbichler, Daniel Martins ... Neil A Harrison
    Research Article Updated

    The relationship between obesity and human brain structure is incompletely understood. Using diffusion-weighted MRI from ∼30,000 UK Biobank participants, we test the hypothesis that obesity (waist-to-hip ratio, WHR) is associated with regional differences in two micro-structural MRI metrics: isotropic volume fraction (ISOVF), an index of free water, and intra-cellular volume fraction (ICVF), an index of neurite density. We observed significant associations with obesity in two coupled but distinct brain systems: a prefrontal/temporal/striatal system associated with ISOVF and a medial temporal/occipital/striatal system associated with ICVF. The ISOVF~WHR system colocated with expression of genes enriched for innate immune functions, decreased glial density, and high mu opioid (MOR) and other neurotransmitter receptor density. Conversely, the ICVF~WHR system co-located with expression of genes enriched for G-protein coupled receptors and decreased density of MOR and other receptors. To test whether these distinct brain phenotypes might differ in terms of their underlying shared genetics or relationship to maps of the inflammatory marker C-reactive Protein (CRP), we estimated the genetic correlations between WHR and ISOVF (rg = 0.026, P = 0.36) and ICVF (rg = 0.112, P < 9×10−4) as well as comparing correlations between WHR maps and equivalent CRP maps for ISOVF and ICVF (P<0.05). These correlational results are consistent with a two-way mechanistic model whereby genetically determined differences in neurite density in the medial temporal system may contribute to obesity, whereas water content in the prefrontal system could reflect a consequence of obesity mediated by innate immune system activation.