Skip to main content
eLife logoLink to eLife
. 2022 Aug 17;11:e78811. doi: 10.7554/eLife.78811

An increase of inhibition drives the developmental decorrelation of neural activity

Mattia Chini 1,, Thomas Pfeffer 2, Ileana Hanganu-Opatz 1
Editors: Liset M de la Prida3, Laura L Colgin4
PMCID: PMC9448324  PMID: 35975980

Abstract

Throughout development, the brain transits from early highly synchronous activity patterns to a mature state with sparse and decorrelated neural activity, yet the mechanisms underlying this process are poorly understood. The developmental transition has important functional consequences, as the latter state is thought to allow for more efficient storage, retrieval, and processing of information. Here, we show that, in the mouse medial prefrontal cortex (mPFC), neural activity during the first two postnatal weeks decorrelates following specific spatial patterns. This process is accompanied by a concomitant tilting of excitation-inhibition (E-I) ratio toward inhibition. Using optogenetic manipulations and neural network modeling, we show that the two phenomena are mechanistically linked, and that a relative increase of inhibition drives the decorrelation of neural activity. Accordingly, in mice mimicking the etiology of neurodevelopmental disorders, subtle alterations in E-I ratio are associated with specific impairments in the correlational structure of spike trains. Finally, capitalizing on EEG data from newborn babies, we show that an analogous developmental transition takes place also in the human brain. Thus, changes in E-I ratio control the (de)correlation of neural activity and, by these means, its developmental imbalance might contribute to the pathogenesis of neurodevelopmental disorders.

Research organism: Human, Mouse

Introduction

Neural activity in the developing brain has several unique traits, such as discontinuity (Chini and Hanganu-Opatz, 2021), extremely low firing rates (Shen and Colonnese, 2016), a loose temporal coordination of excitation and inhibition (Dorrn et al., 2010), and weak modulation by behavioral state (Cirelli and Tononi, 2015; Chini et al., 2019). These patterns of early activity have been described in humans (Vanhatalo and Kaila, 2006) as well as in disparate model organisms ranging from fish (Avitan et al., 2017) to flies (Akin et al., 2019), and from rodents (Khazipov et al., 2004) to brain organoids (Trujillo et al., 2019). As brain networks mature, they gradually evolve into exhibiting motives with adult-like spatiotemporal properties. Oscillatory events become more rhythmic, increasing their amplitude and average frequency (Bitzenhofer et al., 2020), oscillation patterns become more complex (Trujillo et al., 2019), the ratio of excitatory and inhibitory conductances (E-I ratio) decreases (Zhang et al., 2011), excitation and inhibition tighten on a temporal scale (Dorrn et al., 2010; Moore et al., 2018), and brain activity decorrelates and sparsifies (Golshani et al., 2009; Rochefort et al., 2009). The relationship between decorrelation and E-I ratio has been the subject of extensive experimental and theoretical work in the adult brain (Zhou and Yu, 2014; Yu et al., 2014; Shew et al., 2014). Decorrelated and sparse activity is a hallmark of adult spike trains (Olshausen and Field, 2004; Vinje and Gallant, 2000) and artificial neural networks alike (Cun et al., 1990; Frankle and Carbin, 2019; Cogswell et al., 2016). This activity pattern bears important functional and behavioral relevance, as it allows for efficient storing and retrieval of information, while minimizing energy consumption (Olshausen and Field, 2004; Cun et al., 1990; Frankle and Carbin, 2019). However, it is still unknown whether changes in E-I ratio underlie the developmental decorrelation of brain activity. The (patho)physiological relevance of this process is underscored by the hypothesis that altered E-I ratio is the hallmark of later-emerging neurodevelopmental disorders, such as autism (Trakoshis et al., 2020; Antoine et al., 2019; Sohal and Rubenstein, 2019; Gao and Penzes, 2015; Medendorp et al., 2021) and schizophrenia (Gao and Penzes, 2015; Ferguson and Gao, 2018), diseases that have also been linked to disruption of correlated activity in animal models (Luongo et al., 2016; Hamm et al., 2017; Zick et al., 2018).

E-I ratio is controlled by the interplay between pyramidal neurons (PYRs) and interneurons (INs). Throughout development, both populations of neurons migrate into the cortex following an ‘inside-out’ sequence that corresponds to their birthdate (Lim et al., 2018; Sidman and Rakic, 1973). This process is guided by cues provided by PYRs, which populate the cortical layers at an earlier time point than INs (Lim et al., 2018). The functional integration of INs into the cortical circuitry is a slow process that is initiated by the establishment of transient circuits. During the first postnatal week, mouse inhibitory circuits are dominated by somatostatin-positive (SST+) INs (Tuncdemir et al., 2016; Marques-Smith et al., 2016). At later time points, parvalbumin-positive (PV+) INs are also integrated into local networks (Tuncdemir et al., 2016; Marques-Smith et al., 2016; Guan et al., 2018). In rodents, the development of inhibitory synapses is not complete until postnatal day (P)30 (Gour et al., 2021). It is thus conceivable that the developmental strengthening of inhibitory synapses and the ensuing tilting of E-I ratio toward inhibition (Zhang et al., 2011) might underlie the decorrelation of neural activity. In favor of this hypothesis, chronic manipulation of IN activity in the murine barrel cortex results in altered temporal and spatial structure of brain activity (Che et al., 2018; Duan et al., 2018; Modol et al., 2020). However, direct evidence linking E-I ratio and the strengthening of inhibition with the developmental decorrelation of brain activity is still lacking. Further, it is still debated whether GABA, the main inhibitory neurotransmitter in the adult brain, actually exerts an inhibitory effect also during early development. It has long been thought that, in the rodent neocortex, GABA might act as an excitatory neurotransmitter for the first 2 postnatal weeks, and only subsequently ‘switch’ to an inhibitory effect (Ben-Ari, 2002; Kalemaki et al., 2021). These findings have been called into question (Che et al., 2018; Kirmse et al., 2015; Murata and Colonnese, 2020), but the argument remains open.

Here, we combined in vivo extracellular electrophysiological recordings and optogenetics with neural network modeling to systematically explore the relationship between E-I ratio and the decorrelation of neural activity in the developing rodent and human brain. In the murine medial prefrontal cortex (mPFC), a brain area where E-I ratio is of particular relevance in the context of neurodevelopmental disorders (Trakoshis et al., 2020), we show that GABA inhibits neural activity from the very first postnatal days, and that an increase in the strength of the exerted inhibition progressively decorrelates the prefrontal spike trains. Using neural network modeling and bidirectional optogenetic manipulation of IN activity, we further uncover how the inhibition increase drives the change in the correlation structure. Moreover, in a mouse model of impaired neurodevelopment, we report that excessively low E-I ratio causes impaired spike train correlations. Finally, we investigate two different EEG datasets and illustrate the translational relevance of these findings by providing first insights into analogous developmental processes that take place in newborn babies.

Results

The patterns of prefrontal activity dynamically evolve with age

To investigate the relationship between E-I ratio and the decorrelation of neural activity that occurs throughout development, we interrogated a large dataset (n=117 mice) of multi-site extracellular recordings of local field potential (LFP) and single unit activity (SUA) from the prelimbic subdivision of the mPFC of unanesthetized P2-12 mice (Figure 1A). Across this developmental phase, the LFP evolves from an almost complete lack of activity (silent periods) to uninterrupted (continuous) activity, passing through intermediate stages in which silent periods alternate with bouts of neuronal activity (active periods) (Figure 1A). To quantify this transition, we calculated the proportion of active periods over the recording and found that it monotonically increases over age (age slope = 0.88, 95% CI [0.84; 0.93], p<10–50, generalized linear model) (Figure 1B). The increase in the proportion of active periods resulted from the augmentation of both number and duration of individual active periods until continuous activity was detected (Figure 1—figure supplement 1A-C). Accompanying these high-level changes in activity dynamics, the maximum amplitude of active periods, the broadband LFP power, and the SUA firing rate exponentially increased over age (age slope = 0.24, 0.47, and 0.21, 95% CI [0.18; 0.30], [0.40; 0.53], and [0.15; 0.27], p<10–9, p<10–23, p<10–11, respectively, generalized linear model) (Figure 1C, Figure 1—figure supplement 1D-F).

Figure 1. Active periods and local field potential (LFP) properties of the mouse medial prefrontal cortex (mPFC) across the first 2 postnatal weeks.

(A) Schematic representation (Claudi et al., 2021) of extracellular recordings in the mPFC of P2-12 mice (left), and representative LFP traces from P3, P6, P9, and P12 mice (right). (B and C) Violin plots displaying the percentage of time spent in active periods (B) and the single unit activity (SUA) firing rate (C) of P2-12 mice (n=117 mice and 2269 single units, respectively). (D) Log-log plot displaying the normalized median power spectral density (PSD) power in the 1–40 Hz frequency range of P2-12 mice (n=117 mice). Color codes for age with 1 day increment. (E) Violin plot displaying the 1/f exponent of P2-12 mice (n=117 mice). In (B) and (E) white dots indicate individual data points. In (C) data are presented as median, 25th, 75th percentile, and interquartile range. In (B), (C), and (E) the shaded area represents the probability distribution density of the variable. In (D) data are presented as median. Asterisks in (B), (C), and (E) indicate significant effect of age. ***p<0.001. Generalized linear models (B–C) and linear model (E). For detailed statistical results, see Supplementary file 1.

Figure 1.

Figure 1—figure supplement 1. Local field potential (LFP) and single unit activity (SUA) properties of the mouse medial prefrontal cortex (mPFC) across the first 2 postnatal weeks.

Figure 1—figure supplement 1.

(A and B) Violin plots displaying the number (A) and duration (B) of active periods of P2-12 mice (n=117 mice). (C) Scatter plot displaying the time spent in active periods with respect to the number of active periods per minute of P2-12 mice (n=117 mice). Color codes for age. (D and E) Violin plot displaying the maximum amplitude of active periods (D) and the LFP power in the 1–40 Hz frequency range (E) of P2-12 mice (n=117 mice). (F) Log-log line plot displaying the median LFP power in the 0.1–4 Hz frequency range of P2-12 mice (n=117 mice). Color codes for age. In (A), (B), (D), and (E) white dots indicate individual data points and the shaded area represents the probability distribution density of the variable. In (F) data are presented as median. Asterisks in (B), (D), and (E) indicate significant effect of age. ***p<0.001. Generalized linear models. For detailed statistical results, see Supplementary file 1.

Changes in the log-log power spectral density (PSD) slope (reflected by the 1/f exponent) have been linked to E-I ratio by several experimental (Trakoshis et al., 2020; Lendner et al., 2020; Colombo et al., 2019) and theoretical studies (Trakoshis et al., 2020; Lombardi et al., 2017; Gao et al., 2017). In particular, a relative increase in inhibition is thought of leading to a steeper PSD slope (higher 1/f exponent), whereas the opposite occurs when E-I ratio shifts toward excitation. Given that INs have a more protracted integration into cortical circuits than PYRs, this process might be accompanied by a developmental shift of the E-I ratio toward inhibition. In line with this hypothesis, the PSD slope grew steeper over age, as readily observed when the PSD was normalized by the area under the curve (Figure 1D). To quantify this observation, we parameterized the PSDs using a recently published protocol (Donoghue et al., 2020), and confirmed that the 1/f exponent increases with age (age slope = 0.12, 95% CI [0.11; 0.14], p<10–27, linear model) (Figure 1E).

Thus, the monitoring of age-dependent dynamics of prefrontal LFP and SUA let us propose that, throughout development, E-I ratio tilts toward inhibition.

E-I ratio controls pairwise spike train correlations in a neural network model

To explore the relationship between the 1/f exponent, E-I ratio, and the (de)correlation of neuronal spike trains, we simulated a neural network of 400 interconnected conductance-based leaky integrate-and-fire (LIF) neurons (Figure 2A). In line with anatomical data (Markram et al., 2004; Hendry et al., 1987), 80% of those simulated neurons were excitatory (PYRs), whereas 20% were inhibitory (INs). PYRs were simulated with outgoing excitatory AMPA synapses, while INs were simulated with outgoing inhibitory GABAergic synapses, including recurrent connections for both PYRs and INs. In keeping with theoretical and experimental work (Buzsáki and Mizuseki, 2014; Hazan and Ziv, 2020), excitatory and inhibitory synaptic weights were simulated with a lognormal distribution. Both neuron types received input noise and PYRs received an additional external excitatory Poisson stimulus with a constant spike rate of 1.5 spikes/s (see Materials and methods for details on the model). We parametrically varied the AMPA and GABA conductances on both PYRs and INs and defined the network’s net inhibition strength as the ratio between the inhibitory and excitatory conductances. The network was simulated for 60 s for each parameter combination. The network’s LFP was defined as the sum of the absolute values of all synaptic currents on PYRs, which was shown to be a reliable proxy of experimental LFP recordings (Trakoshis et al., 2020; Mazzoni et al., 2008). Across all parameters combinations, INs exhibited higher average firing rates compared to PYRs (INs = 4.06 Hz, 95% CI [3.53; 5.42], PYRs = 1.51 Hz, 95% CI [1.02; 2.98]).

Figure 2. Increased inhibition leads to an increase in the 1/f exponent and decorrelates spike trains in a neural network model.

Figure 2.

(A) Schematic representation of the neural network model. (B) Scatter plot displaying the 1/f exponent as a function of net inhibition strength. (C) Scatter plot displaying average STTC as a function of net inhibition strength. For (B) and (C) color codes for inhibition strength with fixed excitation level.

In agreement with previous results (Trakoshis et al., 2020; Gao et al., 2017), the increase in the network’s net inhibition strength robustly tilted the PSD slope (in the range from 30 to 100 Hz). Increased levels of net inhibition were associated with a steeper decay of power as a function of frequency and a corresponding increase of the 1/f exponent, across a range of AMPA conductance levels (Figure 2B); Pearson correlation coefficient, averaged across AMPA levels (inhibition strength slope = 0.022, 95% CI [0.021; 0.024], linear model). We next examined the effect of increasing the network’s net inhibition strength on neural correlations. To this end, we computed the spike time tiling coefficient (STTC; at a lag of 1 s), a parameter that measures pairwise correlations between spike trains without being biased by firing rate (Cutts and Eglen, 2014), on the network’s spike matrices (both PYRs and INs) and across all levels of net inhibition strength. For all levels of AMPA conductance, we found that increased net inhibition strength results in a robust logarithmic decrease in STTC (Figure 2C; inhibition strength slope = −0.068, 95% CI [0.075; 0.061], generalized linear model).

Thus, simulations of a biologically plausible neural circuit reveal that increased net inhibition strength leads to an increase of the PSD 1/f exponent that is accompanied by decorrelation of neural spike trains.

Prefrontal spike trains decorrelate over development

Since neural network modeling predicts that a shift of E-I ratio toward inhibition leads to higher 1/f exponent and decorrelation of neural activity, we tested on the experimental data whether the developmental increase in the 1/f exponent in the mouse mPFC was accompanied by a decorrelation of neural activity. For this, we calculated the STTC between >40,000 pairs of spike trains over a large range of lags (2.5 ms to 10 s) (Figure 3A). For the analysis, we only considered SUA that was recorded for at least 60 min. To verify the robustness of STTC as an estimator, we compared the STTC obtained on the first and the second half of the recording. The STTCs computed on the two halves of the recording strongly correlated with each other across all the investigated lags (0.70, [0.50; 0.80] median and min-max Pearson correlation; 0.70 [0.52; 0.79] median and min-max Spearman correlation) (Figure 3—figure supplement 1A-B), thus corroborating its robustness as an estimator. Throughout the manuscript, we will consider STTC computed at 1 s, yet the summary plots and the supplementary statistical table include values calculated at all lags.

Figure 3. Spike time tiling coefficient (STTC) decreases throughout development with a specific spatial profile in the mouse medial prefrontal cortex (mPFC).

(A) Schematic representation of the STTC quantification. (B) Multivariate linear regression coefficients with respect to STTC lag (n=40,921 spike train pairs and 82 mice). (C) Average STTC at 1 s lag of P4, P6, P8, P10, and P12 mice over distance (n=40,921 spike train pairs and 82 mice). Color codes for age. (D) Weighted adjacency matrices displaying average STTC at 1 s lag of P4, P6, P8, P10, and P12 mice as a function of the recording sites in which the spike train pair has been recorded. Color codes for STTC value. In (B) regression coefficients are presented as mean and 95% CI. In (C) data are presented as mean ± SEM. Asterisks in (B) indicate significant regression coefficients of the respective (interaction between) variables for STTC at 1 s lag. Asterisks in (C) indicate significant effect of age*distance interaction. ***p<0.001. Linear mixed-effect models. For detailed statistical results, see Supplementary file 1.

Figure 3.

Figure 3—figure supplement 1. The spike time tiling coefficient (STTC) developmental decrease follows specific spatial patterns in the mouse medial prefrontal cortex (mPFC).

Figure 3—figure supplement 1.

(A) Scatter plot displaying the STTC computed in the first half of the recording with respect to STTC computed in the second half of the recording (n=40,921 spike train pairs and 82 mice). (B) Scatter plot displaying Pearson and Spearman correlation coefficient for STTC computed on the first and second half of the recording over lags (n=40,921 spike train pairs and 82 mice). (C) Weighted adjacency matrices displaying average STTC at 2.5 ms lag of P4, P6, P8, P10, and P12 mice as a function of the recording sites in which the spike train pair has been recorded. Color codes for STTC value. For detailed statistical results, see Supplementary file 1.

Using multivariate mixed hierarchical linear regression, we found that STTC negatively correlated with the distance between neurons (i.e. nearby neurons had higher STTC values than neurons that are far apart) over all the investigated lags (main distance effect, p<10–78 at 1 s lag, linear mixed-effect model) (Figure 3B–D). This is in line with previous studies conducted in the adult (Smith and Kohn, 2008; Goltstein et al., 2015; Greenberg et al., 2008) and developing brain (Golshani et al., 2009; Cutts and Eglen, 2014; Blankenship et al., 2011) of several mammalian species. Further, STTC values negatively correlated with age at lags ≤5 s (main age effect, p<10–4 at 1 s lag, linear mixed-effect model), an effect that was strongest in the 100–1000 ms range (Figure 3B and D, Figure 3—figure supplement 1C). This developmental STTC decrease did not occur uniformly across all neuron pairs. Rather, age and distance had a significant interaction, nearby pairs of neurons displaying a more severe decorrelation over age than neurons that were further apart (age*distance interaction, p<10–12 at 1 s lag, linear mixed-effect model) (Figure 3B–D).

Taken together, these data indicate that, throughout development, as E-I ratio tilts toward inhibition, there is a concomitant decorrelation of pairwise neuronal activity computed over lags that span more than three orders of magnitude. This result is in agreement with data from the rodent barrel cortex (Golshani et al., 2009; Rochefort et al., 2009). In addition, we report that this process follows a specific spatial pattern, with the activity of nearby neurons being the most affected.

Optogenetic IN manipulation confirms developmental increase of inhibition

To substantiate the experimental evidence supporting the developmental increase of inhibition in the mouse mPFC, we optogenetically manipulated IN activity at different stages of early development. To this aim, we selectively transfected Dlx5/6cre and Gad2cre INs with either an excitatory (ChR2, n=19 mice) or an inhibitory opsin (ArchT, n=40 mice) using a combination of mouse lines and viral approaches. Briefly, expression of an excitatory opsin in INs was achieved by injecting P0-1 Dlx5/6cre and Gad2cre mice with a virus encoding for ChR2 (AAV9-Ef1alpha-DIO-hChR2(ET/TC)-eYFP). Expression of an inhibitory opsin was instead achieved by crossing Dlx5/6cre and Gad2cre mice with a mouse line (Ai40(RCL-ArchT/EGFP)-D) expressing ArchT under a cre-dependent promoter. No significant differences between experiments targeting Dlx5/6cre and Gad2cre neurons were detected and therefore, the datasets were pooled. In line with previously developed protocols (Bitzenhofer et al., 2020; Chini et al., 2020; Bitzenhofer et al., 2017b), we applied a 3-s-long ‘ramp-like’ optogenetic stimulation of increasing intensity (Figure 4A, Figure 4—figure supplement 1A).

Figure 4. Optogenetic stimulation of interneuron (IN) activity leads to widespread inhibition in the developing mouse medial prefrontal cortex (mPFC).

(A) Schematic representation of the effects induced by optogenetic IN stimulation (left). Representative local field potential (LFP) trace (4–100 Hz band-pass filter) with a corresponding wavelet spectrum at an identical timescale during ramp light stimulation (473 nm, 3 s) of INs in the mPFC of a P10 mouse. (B) Z-scored single unit firing rates in response to optogenetic stimulation of INs (left) and volcano plot displaying the modulation index of pre vs. stim single unit firing rates (right) for P4-6 (top, n=268 units and 5 mice), P7-9 (middle, n=480 units and 7 mice), and P10-12 (bottom, n=475 units and 7 mice) mice. Color codes for firing rate. (C) Scatter plot displaying the percentage of inhibited units with respect to age (n=19 mice). (D) First (top, putative pyramidal neurons [PYRs]) and second (bottom, putative INs) principal component analysis (PCA) component of trial-averaged spike trains in response to optogenetic stimulation of INs. Color codes for age group. In (C) the regression is presented as mean and 95% CI. Asterisks in (C) indicate significant effect of age. **p<0.01. Individual dots in (C) indicate distinct optogenetic protocols (up to two per mouse, for details see Materials and methods). Generalized linear mixed-effect model (C). For detailed statistical results, see Supplementary file 1.

Figure 4.

Figure 4—figure supplement 1. Optogenetic manipulation of interneuron (IN) activity affects local field potential (LFP) activity in the mouse developing medial prefrontal cortex (mPFC).

Figure 4—figure supplement 1.

(A) Schematic representation of the experimental approach employed to target INs with an excitatory and an inhibitory opsin. (B) Proportion of activated units in response to ramp light stimulation of INs as a function of age (n=19 mice). (C) Number of virally transfected neurons as a function of age (n=87 images and 16 mice). (D) Scatter plot displaying the percentage of inhibited units (restricted to the top 50% units for firing rate) with respect to age (n=19 mice). (E) Schematic representation of the neural network model. (F) Modeled pyramidal neurons (PYRs) (top) and INs (bottom) trial-averaged normalized firing rate in response to optogenetic stimulation of INs. Color codes for excitation-inhibition (E-I) ratio. (G) Violin plot displaying the log-transformed firing rate in simulated optogenetic experiments with an excitatory current injected into INs as a function of the difference between the chloride reverse potential and the membrane resting potential. (H) Modeled PYRs (top) and INs (bottom) trial-averaged normalized firing rate in response to optogenetic stimulation of INs. Color codes for the difference between the chloride reverse potential and the membrane resting potential. (I, J) Same as (G, H) for simulated optogenetic experiments with an inhibitory current injected into INs. In (B–D) and (C) the regression is presented as mean and 95% CI. Asterisks in (D) indicate significant effect of age. ***p<0.001. Individual dots in (B) and (D) indicate distinct optogenetic protocols (up to two per mouse, for details, see Materials and methods). Generalized linear mixed-effect model (B) and (D), linear mixed-effect model (C). For detailed statistical results, see Supplementary file 1.
Figure 4—figure supplement 2. Optogenetic manipulation of interneuron (IN) activity in cre- mice does not affect the developing mouse medial prefrontal cortex (mPFC).

Figure 4—figure supplement 2.

(A) Schematic representation of the experimental approach employed to generate control (cre-) mice (left) and schematic representation of the lack of effects induced by optogenetic IN stimulation in cre- mice (right). (B) Z-scored single unit firing rates response to optogenetic stimulation in cre- mice (left) and volcano plot displaying the modulation index of pre vs. stim single unit firing rates (right) for P2-12 mice (n=380 units and 10 mice),.

IN activation led to conspicuous modulation of SUA across all investigated ages. Upon stimulus, a small number of neurons (putative INs) gradually increased their firing rate (Figure 4B). The proportion of stimulated INs was similar among mouse lines (main mouse line effect, p=0.14, generalized linear mixed-effect model) and across ages (main age effect = 0.14, 95% CI [–0.04; 0.34], p=0.12, generalized linear mixed-effect model) (Figure 4—figure supplement 1B; individual dots correspond to optogenetic stimulation protocols, see figure legend and Materials and methods). This is in line with the histological quantification of the number of virally transfected neurons that led to similar results for all mouse lines (main mouse line effect, p=0.45, linear mixed-effect model) and developmental stages (main age effect = −1.35, 95% CI [–3.22; 0.51], p=0.18, linear mixed-effect model) (Figure 4—figure supplement 1C). While putative INs increased their firing rate in response to optogenetic stimulation, a larger proportion of neurons (putative PYRs) significantly decreased their firing rate (Figure 4B). In line with the results above that indicated increasing inhibition throughout development, the proportion of inhibited neurons augmented with age (main age effect = 0.31, 95% CI [0.09; 0.55], p=0.005, generalized linear mixed-effect model) (Figure 4C; individual dots correspond to optogenetic stimulation protocols, see figure legend and Materials and methods). These results are unlikely to be biased by a ‘floor effect’ due to the low firing rate of neurons in the youngest mice, as limiting the analysis to neurons in the top 50% for spikes fired during the optogenetic protocol yielded an even stronger effect (main age effect = 0.47, 95% CI [0.17; 0.81], p=0.002, generalized linear mixed-effect model) (Figure 4—figure supplement 1D; individual dots correspond to optogenetic stimulation protocols, see figure legend and Materials and methods). Regardless of age, after terminating the optogenetic stimulus, PYRs responded with a prominent ‘rebound’ increase in firing rate, similar to the effects reported for the adult brain (Roux et al., 2014; Sessolo et al., 2015). Such widespread inhibition upon IN stimulation supports the hypothesis that GABA exerts an inhibitory population-level effect already during the first postnatal days.

To dissect the main ‘neuronal trajectories’ in response to light stimulation, we performed principal component analysis (PCA) on trial-averaged smoothed and normalized spike trains and projected the time-varying activity of neurons onto a low-dimensional space. The two principal components captured two populations of neurons that, during optogenetic stimulation, responded with a monotonic decrease (first component) and increase (second component) in firing rate (Figure 4D). While opto-tagging at such an early age is not possible (Weir et al., 2014), we hypothesize that the first component corresponds to inhibited PYRs and the second one to INs that were activated with the optogenetic stimulus. Interestingly, the two dynamics were strikingly similar across age groups (Figure 4D).

To corroborate this hypothesis, we carried out a series of simulations in a LIF neural network model analogous to the one previously described. To mimic the ‘ramp-like’ optogenetic stimulation, we injected INs with repeated sweeps of an excitatory current with the same temporal profile (3 s of stimulation of increasing intensity) (Figure 4—figure supplement 1E). The model confirmed that, similarly to the neuronal trajectory described by the first PCA component, PYRs decreased their firing rate in response to the stimulation, while INs responded with an increase in firing rate that is analogous to the neuronal trajectory described by the second PCA component (Figure 4—figure supplement 1F). To test the role played by inhibition in generating the population dynamics, we systematically varied the reversal potential (Vrev) driving the inhibitory post-synaptic currents. Vrevs that were higher than the action potential threshold (Vthr) (excitatory GABA) between the resting membrane potential (Vrest) and the Vthr (depolarizing GABA) or equal to Vrest resulted in runaway excitation and average firing rates within 100–1000 Hz range (Figure 4—figure supplement 1G). Vrev that were at least 5 mV lower than the Vrest resulted in stable networks, but only Vrevs that were 10 mV or lower than Vrest were able to fully recapitulate the experimentally observed population trajectories (Figure 4—figure supplement 1H).

Taken together, these data indicate that, while the inhibition strength exerted by INs increases throughout development, the dynamics with which the PYR-IN network responds to IN activation does not change across the first 2 postnatal weeks. These data also provides support to the notion that, on a network level, GABA exerts an inhibitory effect already during the first postnatal week (Kirmse et al., 2015; Murata and Colonnese, 2020).

To further corroborate these results, we carried out analogous experiments in mice expressing an inhibitory opsin in INs (Figure 5A, Figure 4—figure supplement 1A).

Figure 5. Optogenetic inhibition of interneuron (IN) activity leads to widespread excitation in the developing mouse medial prefrontal cortex (mPFC).

Figure 5.

(A) Schematic representation of the effects induced by optogenetic IN inhibition (left). Representative local field potential (LFP) trace (4–100 Hz band-pass filter) with a corresponding wavelet spectrum at an identical timescale during ramp light stimulation (594 nm, 3 s) of INs in the mPFC of a P10 mouse. (B) Z-scored single unit firing rates in response to optogenetic stimulation of INs (left) and volcano plot displaying the modulation index of pre vs. stim single unit firing rates (right) for P2-3 (top left, n=164 units and 5 mice), P4-6 (top right, n=286 units and 11 mice), P7-9 (bottom left, n=470 units and 13 mice), and P10-12 (bottom right, n=691 units and 11 mice) mice. Color codes for firing rate. (C) Scatter plot displaying the percentage of activated units with respect to age (n=40 mice). (D) first (top) and second (bottom) principal component analysis (PCA) component of trial-averaged spike trains in response to optogenetic inhibition of INs. Color codes for age. (E) Schematic representation of the neural network model. (F) Modeled pyramidal neurons (PYRs) (left) and INs (right) trial-averaged normalized firing rate in response to optogenetic inhibition of INs. Color codes for excitation-inhibition (E-I) ratio. In (C) the regression is presented as mean and 95% CI. Asterisks in (C) indicate significant effect of age. ***p<0.001. Individual dots in (C) indicate distinct optogenetic protocols (up to two per mouse, see Materials and methods). Generalized linear mixed-effect model (C). For detailed statistical results, see Supplementary file 1.

Light-induced IN silencing resulted in a widespread progressive increase of firing, with very few inhibited neurons (12/1611 neurons) (Figure 5B). Independent of age group, the increased firing abruptly returned to baseline levels upon terminating the optogenetic stimulus (Figure 5B). The proportion of neurons responding with a firing rate increase during IN inhibition augmented with age (main age effect = 0.54, 95% CI [0.37; 0.74], p<10–8, generalized linear model) (Figure 5C; individual dots correspond to optogenetic stimulation protocols, see figure legend and Materials and methods). To qualitatively compare the ‘neuronal trajectories’, we performed trial-averaged PCA. The first component, that captured the activity of putative PYRs, responded to the light stimulus with a monotonic rise of firing rate that quickly dropped as soon as the stimulus stopped (Figure 5D). On the contrary, and in line with previous experimental (Moore et al., 2018; Tsodyks et al., 1997; Sadeh and Clopath, 2020) and theoretical work (Kato et al., 2017; Sanzeni et al., 2020) in the adult brain, putative INs (the trajectory captured by the second component), had a biphasic response. They were initially inhibited but, halfway through the optogenetic stimulation, displayed a steep increase in firing rate, that persisted until end of the stimulus (Figure 5D). This response pattern is typical for inhibition-stabilized networks (ISNs), a regime at which adult cortical circuits operate (Sanzeni et al., 2020; Sadeh et al., 2017), and that has been suggested to not characterize the early developing brain (Rahmati et al., 2017).

To corroborate the experimental findings of putative PYRs’ and INs’ temporal dynamics, we carried out a series of simulations in a LIF neural network model. To mimic the ‘ramp-like’ optogenetic inhibition of INs, we injected this population of neurons with repeated sweeps of an inhibitory current with the same temporal profile (3 s of stimulation of increasing intensity) (Figure 5E). The model confirmed that, similarly to the neuronal trajectory described by the first PCA component, PYRs monotonically increased their firing rate in response to the IN inhibition (Figure 5F). INs had a biphasic response analogous to the one described by the second PCA component. They initially decreased their firing rate and then, halfway through the ramp stimulation, robustly increased their firing rate (Figure 5F). To test the role of inhibition in generating the population dynamics, we systematically varied the Vrev driving the inhibitory post-synaptic currents. Analogously to what has been previously shown, excitatory and depolarizing GABA both resulted in runaway excitation and average firing rates that were in the 100–1000 Hz range (Figure 4—figure supplement 1I). Vrev that were at least 5 mV lower than Vrest resulted in stable networks, but only Vrev that were 10 mV or lower than Vrest were able to fully recapitulate the population trajectories that we experimentally observed (Figure 4—figure supplement 1J).

Thus, while the network excitation derived from IN inhibition increased throughout development, the dynamics with which the PYR-IN network responds to IN inhibition did not change during the first 2 postnatal weeks. These data further support the conclusion that, on a network level, GABA exerts an inhibitory effect already in the first postnatal week (Kirmse et al., 2015; Murata and Colonnese, 2020).

We have previously shown that the optogenetic paradigm that we utilized does not lead to significant tissue heating (Bitzenhofer et al., 2017b), but to further rule out possible nonspecific effects, we applied the same stimulation paradigm to cre- mice (n=10 mice, 380 neurons, Figure 4—figure supplement 2A). Pooling together all investigated mice, only 6 out of 380 units were activated, whereas none was inhibited. These results are in line with the statistical threshold (0.01) that was used for this analysis (proportion of modulated units 0.016, CI [0.007; 0.034], Figure 4—figure supplement 2B). Thus, the used light stimulation leads to minimal non-significant, if any, unspecific modulation of neuronal firing.

Taken together, these data show that optogenetic manipulation of INs robustly affects the neonatal prefrontal network in an age-dependent manner. Stimulating INs induced widespread inhibition of putative PYRs, whereas the contrary was true after IN inhibition. Both effects augmented with age. However, the ability of INs to control the cortical inhibition did not qualitatively change during the first 2 postnatal weeks, resembling adult patterns. These data provide evidence against the long-standing hypothesis of network-level excitatory effects of GABA in the developing mouse cortex.

Optogenetic manipulation of IN activity impacts pairwise spike train correlations

To investigate the relationship between age-dependent dynamics of inhibition and decorrelation of spike trains, we compared STTC before IN optogenetic manipulation (STTCpre) to STTC during optogenetic manipulation (STTCstim). Considering that STTCpre and STTCstim could only be computed in 3 s epochs (times the number of trials), we first verified whether STTCpre was a good predictor of ‘baseline’ STTC. Pooling across mice and different IN manipulations, STTCpre robustly correlated with baseline STTC across every investigated lag, from 2.5 ms to 1 s (0.66, [0.48; 0.72] median and min-max Pearson correlation; 0.68 [0.40; 0.71] median and min-max Spearman correlation) (Figure 6—figure supplement 1A-B). Further, STTCstim exhibited lower correlation values with baseline STTC across all lags, a first hint that optogenetic IN manipulation affected STTC (Figure 6—figure supplement 1A-B).

As predicted by the experimental and modeling results, optogenetic modulation of IN activity affected the STTC values across all investigated timescales (Figure 6A–B, Figure 6—figure supplement 1C). IN stimulation resulted in decreased STTC values (main IN stimulation effect, p<10–71, 1 s lag, linear mixed-effect model) (Figure 6A). On the other hand, IN inhibition increased STTC (main IN inhibition effect, p<10–286, 1 s lag, linear mixed-effect model) (Figure 6B). Moreover, in line with the strongest decorrelation along development for nearby neurons (Figure 3D), IN modulation had a larger impact on STTC values of nearby neurons when compared to pairs that are further apart (IN stimulation*distance interaction, p=2*10–4; IN inhibition*distance interaction p<10–5, 1 s lag, linear mixed-effect model) (Figure 6C–D).

Figure 6. Bidirectional optogenetic manipulation of interneuron (IN) activity affects spike time tiling coefficient (STTC) in the developing mouse medial prefrontal cortex (mPFC).

(A and B) 2D kernel density plots displaying STTC before IN optogenetic manipulation (STTCpre) and STTC during optogenetic manipulation (STTCstim) during IN activation (A) and inhibition (B) (n=10,173 spike train pairs and 19 mice, n=9778 spike train pairs and 40 mice, respectively). (C and D) Average STTCpre and STTCstim during IN activation (C) and inhibition (D) over distance (n=10,173 spike train pairs and 19 mice, n=9778 spike train pairs and 40 mice, respectively). In (C and D) data are presented as mean ± SEM. Asterisks in (A and B) indicate significant effect of IN activation and inhibition, respectively. Asterisks in (C and D) indicate significant effect of IN activation*distance and IN inhibition*distance interaction, respectively. ***p<0.001. Linear mixed-effect models. For detailed statistical results, see Supplementary file 1.

Figure 6.

Figure 6—figure supplement 1. Bidirectional optogenetic manipulation of interneuron (IN) activity effects on spike time tiling coefficient (STTC) depends on spatial configurations and age.

Figure 6—figure supplement 1.

(A and B) Scatter plot displaying Pearson (A) and Spearman (B) correlation coefficients for STTC before IN optogenetic manipulation (STTCpre) and STTC during optogenetic manipulation (STTCstim) with STTC computed on baseline data over lags (n=19,951 spike train pairs and 59 mice). (C) Multivariate linear regression coefficients as a function of STTC lag (n=19,951 spike train pairs and 59 mice). In (C) regression coefficients are presented as mean and 95% CI. Linear mixed-effect models. For detailed statistical results, see Supplementary file 1.

Thus, these data indicate that IN manipulation causally impacts pairwise correlations between spike trains. The effect of IN manipulation increases with age, in agreement with the notion that inhibition strengthens throughout development.

Mice with altered developmental E-I ratio have excessively decorrelated activity

Developmental imbalances in E-I ratio have been linked to the pathophysiology of neurodevelopmental disorders (Antoine et al., 2019; Bitzenhofer et al., 2021). A corollary of the results above is that impaired developmental E-I ratio should result in altered correlation levels of neuronal activity. To test this hypothesis, we interrogated an open-source dataset that we recently published (Chini et al., 2020; Bitzenhofer et al., 2021). The dataset was obtained from extracellular recordings of SUA from the mPFC of P4-10 control and dual-hit genetic-environmental (GE) mice. GE mice mimic the etiology (combined disruption of Disc1 gene and maternal immune activation) and cognitive impairment of schizophrenia, showing already at neonatal age reduced excitatory activity in the superficial layers of the mPFC (Chini et al., 2020; Xu et al., 2019; Figure 7A). On the flipside, deep layers of the mPFC are not affected. Therefore, we hypothesized that GE mice have lower STTC values than controls (i.e. mice lacking the abnormal genetic background and influence of environmental stressor). Considering the layer specificity of the deficits identified in the mPFC of GE mice, we reasoned that this effect should be present in spike trains from neurons in the superficial layers. Overall, GE mice had lower spike train correlations when compared to controls (main condition effect, p=0.032, 1 s lag, linear mixed-effect model) (Figure 7—figure supplement 1A). In line with the proposed hypothesis, this deficit depended on whether the neuron pair was situated in the superficial or deep layers of the mPFC (condition*layer interaction, p<10–7, 1 s lag, linear mixed-effect model). While there was no significant difference between STTC of controls and GE spike train pairs situated in the deep layers (p=0.15, 1 s lag, linear mixed-effect model), spike train pairs of GE mice in which one of the two neurons was located in the superficial layers had reduced STTC values (p=0.016, 1 s lag). This difference was even more robust if both neurons were situated in the superficial layers (p=10–3, 1 s lag) (Figure 7B–D). Last, the effect did not depend on the age of the mouse (condition*age interaction, p=0.16, 1 s lag, linear mixed-effect model) (Figure 7—figure supplement 1A-B).

Figure 7. Genetic-environmental (GE) mice have reduced spike time tiling coefficient (STTC) values with specific spatial profiles.

(A) Schematic representation of the excitation-inhibition (E-I) ratio imbalance affecting GE mice. (B) STTC of control and GE mice (n=18,839 and 11,051 spike train pairs; 33 and 30 mice, respectively) with respect to the number of neurons in the superficial layers in the medial prefrontal cortex (mPFC). (C) Weighted adjacency matrices displaying average STTC at 1 s lag of P4, P6, P8, P10, and control mice as a function of the recording sites in which the spike train pair has been recorded (n=18,839 spike train pairs and 33 mice). White inset indicates STTC values between spike trains that are located in the superficial layers of the mPFC. Color codes for STTC value. (D) Same as (C) for GE mice (n=11,051 spike train pairs and 30 mice).

Figure 7.

Figure 7—figure supplement 1. Age and spatial profile of spike time tiling coefficient (STTC) values in genetic-environmental (GE) and ES mice.

Figure 7—figure supplement 1.

(A) Multivariate linear regression coefficients as a function of STTC lag (n=29,890 spike train pairs and 63 mice). (B) STTC of control and GE mice (n=18,839 and 11,051 spike train pairs; 33 and 30 mice, respectively) over age. Asterisks in (A) indicate significant regression coefficients of the respective variable for STTC at 1 s lag. *p<0.05. In (A) regression coefficients are presented as mean and 95% CI. In (B) data are presented as mean ± SEM. Linear mixed-effect models. For detailed statistical results, see Supplementary file 1.

Taken together, these data support the hypothesis that decreased developmental E-I ratio results in reduced spike train pairwise correlations. Further, we show that this effect is remarkably specific. In GE mice, a mouse model characterized by reduced excitatory drive in prefrontal PYRs of the superficial layers, the reduced correlation levels were largely limited to spike train pairs involving PYRs of the superficial layers.

E-I ratio decreases with age in newborn babies

Considering the role of E-I ratio for neurodevelopmental disorders, it is of critical relevance to assess whether a developmental strengthening of inhibition occurs also in humans. To this aim, we interrogated two EEG datasets recorded in newborn babies of an age between 35 and 46 post-conception weeks (PCW), a stage of brain development that is roughly equivalent to the one that we studied in mice (Chini and Hanganu-Opatz, 2021). While it is not straightforward to compare intracranial recordings from a deep structure like the mouse mPFC to human EEG data, to maximize the consistency between approaches, we limited our analysis to channels from the frontal derivations of the EEG (Figure 8A).

Figure 8. 1/f exponent of EEG recordings increases with age in newborn babies.

(A) Schematic representation of EEG recording from frontal derivations of 36-45 post-conception week (PCW) newborn babies (left) displayed together with representative EEG traces from 36 and 45 PCW newborn babies (right). (B) Log-log plot displaying the normalized mean power spectral density (PSD) power in the 1–20 Hz frequency range of 36-45 PCW newborn babies (n=1110 babies). Color codes for age. (C) Violin plots displaying the 1/f exponent of 36-45 PCW newborn babies (n=1110 babies). (D) Same as (C) for 40 and 43 PCW newborn babies (n=72 EEG recordings and 40 babies). (E) 1/f exponent over age for the two EEG datasets (n=1110 babies and n=72 EEG recordings and 40 babies, respectively). In (D) black dots indicate individual data points. In (C) and (D) data are presented as median, 25th, 75th percentile, and interquartile range. In (C) and (D) the shaded area represents the probability distribution density of the variable. In (B) and (E) data are presented as mean ± SEM. Asterisks in (C) and (D) indicate significant effect of age. ***p<0.001. Linear model (C) and linear mixed-effect models (D–E). For detailed statistical results, see Supplementary file 1.

Figure 8.

Figure 8—figure supplement 1. EEG power spectral densities (PSDs) of newborn babies.

Figure 8—figure supplement 1.

(A) Log-log plot displaying the mean PSD power in the 1–20 Hz frequency range of 36-45 post-conception week (PCW) newborn babies (n=1110 babies). Color codes for age. (B) Same as (A) for 40 and 43 PCW newborn babies (n=72 EEG recordings and 40 babies). Color codes for age. (C) Same as (B) for normalized mean PSD power. In (A–C) data are presented as mean ± SEM.

The first dataset (Schetinin and Jakaite, 2017) consisted of 1100 EEG recordings from sleeping babies with an age comprised between 36 and 45 PCW. Similar to the PSDs of recordings from the neonatal mice mPFC, the PSD slope grew steeper over age (Figure 8—figure supplement 1A), a phenomenon that was readily apparent after normalization of the PSD (Figure 8B). We quantified the 1/f exponent on the 1–20 Hz frequency range and confirmed that it increased with age (age coefficient = 0.26, 95% CI [0.24; 0.27], p<10–183, linear model) (Figure 8C). A second dataset (Wielek et al., 2019) consisted of EEG recordings from 42 sleeping babies, recorded at 40 and 43 PCW. The analyses revealed that also for these data the PSD slope grew steeper (Figure 8—figure supplement 1B-C) and the 1/f exponent increased with age (age coefficient = 0.30, 95% CI [0.17; 0.42], p<10–4, linear mixed-effect model) (Figure 8D). The increase in 1/f slope over age was very similar across the two different datasets (mean age coefficients = 0.26 and 0.30) and no statistical difference was found between them (main dataset effect, p=0.15; age*dataset interaction, p=0.21, linear mixed-effect model) (Figure 8E).

Thus, the E-I ratio decreases along development also in newborn humans. This feature might represent a fingerprint of cortical circuit maturation in mammalian species.

Discussion

Integration of INs into the cortical circuitry has been proposed to bear important structural, functional, and behavioral consequences (Lim et al., 2018). Here, we show that, even though INs inhibit neuronal activity already in the first postnatal week, the relative strength of the exerted inhibition increases with age, leading to a decrease of E-I ratio. This developmental process contributes to a transition in brain dynamics from early highly synchronous activity patterns to decorrelated neural activity later in life. We further show that an early imbalance in E-I ratio in the mPFC results in altered temporal neural dynamics. Last, leveraging two EEG datasets recorded in newborn babies, we provide evidence for analogous developmental processes taking place in the human cortex.

Inhibitory synaptogenesis is an exquisitely specific process that is orchestrated by distinct molecular programs (Favuzzi et al., 2019) and neural activity (Kepecs and Fishell, 2014). Its timeline is protracted and it extends to early adulthood in the mouse somatosensory cortex (Gour et al., 2021). PYRs are thought of playing an instructive role in this process, by providing molecular cues guiding IN migration and by regulating their survival in an activity-dependent manner (Lim et al., 2018; Wong et al., 2018). Inhibitory circuits during the first postnatal week have several peculiarities, including a predominance of inhibitory synapses by SST+ INs (Tuncdemir et al., 2016; Marques-Smith et al., 2016; Guan et al., 2018) and the presence of hub neurons that control the formation of functional assemblies (Cossart, 2014; Picardo et al., 2011). One of the unique traits of early inhibitory circuits that has garnered much attention is the hypothesis that GABA might act as an excitatory neurotransmitter. This excitatory action of GABA has been most intensively investigated at single-cell level (Ben-Ari, 2002; Garaschuk et al., 2000) and has been proposed to result from the low expression of KCC2, a potassium chloride cotransporter extruding chloride (Ben-Ari, 2002), which leads to high intracellular chloride concentration in immature neurons. The ‘GABA-switch’ from an excitatory to inhibitory neurotransmitter has been suggested to take place between the first and the second postnatal week (Ben-Ari, 2002). Refuting this hypothesis, recent studies have shown that, already during the first postnatal week, GABAergic transmission, while depolarizing on single neurons, exerts an inhibitory action on population activity (Che et al., 2018; Kirmse et al., 2015; Murata and Colonnese, 2020). In line with this latter interpretation, the present study provides SUA-level evidence of the inhibitory effect of GABA in vivo already in the first postnatal days. This is of particular interest considering that the PFC is a brain area whose development is thought of being more protracted than sensory areas (Chini and Hanganu-Opatz, 2021), where the early effects of GABA have been more intensively studied (Che et al., 2018; Kirmse et al., 2015; Murata and Colonnese, 2020). We show that optogenetic inhibition of cortical INs results in a widespread increase of SUA firing rates and, conversely, increasing their activity leads to a reduction of neuronal activity. While the strength of the inhibition exerted by INs increases throughout development, the network effects generated by INs do not qualitatively change with age. Already during the first postnatal week, optogenetic inhibition of INs leads to a paradoxical increase in their firing rate. This dynamics is characteristic of ISNs that have a high degree of recurrent inhibition (Sadeh and Clopath, 2020), a regime that has been proposed to emerge at later developmental phases (Rahmati et al., 2017). Here, we find experimental and computational evidence that the mPFC might operate in an ISN regime already shortly after birth. Neural network modeling suggests that this is only possible in the presence of a strong hyperpolarizing chloride drive at inhibitory synapses. Of note, we only observed a paradoxical effect, the signature of ISN networks, in response to optogenetic inhibition and not stimulation of INs. This apparent discrepancy is most likely the result of the approach that we employed to express an inhibitory opsin in INs. The combination of two mouse lines (see Materials and methods) targeted a larger proportion of neurons than the approach that combined a mouse line with viral injections (see Materials and methods) and has been used to express the inhibitory opsin. The proportion of activated/inhibited neurons is a critical factor in determining whether a paradoxical response is elicited (Sadeh and Clopath, 2020; Sanzeni et al., 2020). Similar discrepancies between the two different targeting approaches have been previously reported for the adult cortex (Sanzeni et al., 2020). Further, in neural network models, the paradoxical effect is more readily observed in response to IN inhibition than IN stimulation (Sadeh and Clopath, 2020). Thus, while the effects of GABA might differ within brain regions (Murata and Colonnese, 2020), our data supports the notion that, in the mPFC, GABA has an inhibitory role already at P2 and that even in the first postnatal days, the cortex operates in an ISN regime.

As the brain develops and inhibitory synaptogenesis progresses, the temporal coordination between excitatory and inhibitory transmission tightens (Dorrn et al., 2010; Moore et al., 2018) with a relative strengthening of inhibition with respect to excitation (i.e. a decrease in E-I ratio) (Zhang et al., 2011). While E-I ratio is thought of being a critical feature of healthy neural networks (Moore et al., 2018), the functional consequences of the developmental E-I ratio decrease are still poorly understood. Individually, E-I ratio imbalances (Trakoshis et al., 2020; Antoine et al., 2019; Sohal and Rubenstein, 2019; Gao and Penzes, 2015; Medendorp et al., 2021; Ferguson and Gao, 2018) and altered correlational structure of brain activity (Luongo et al., 2016; Hamm et al., 2017; Zick et al., 2018) have both been linked to mental disorders. Studying how these two processes are linked to each other is therefore likely to be insightful for understanding the pathophysiology of these disorders. To address this knowledge gap, we explored the impact of varying E-I ratio in a biologically plausible neural network model. In line with previous results (Trakoshis et al., 2020; Gao et al., 2017), we show that E-I ratio can be indirectly tracked by measuring the 1/f exponent, a notion that we leveraged to show that there is an E-I ratio decrease in the mouse and human mPFC. We further show that, in a neural network model, a relative increase in inhibition results in decreased pairwise correlations of spike trains. Confirming the modeling results, we report that, in the mouse mPFC, the inferred E-I ratio decrease taking place across the first two postnatal weeks is accompanied by a reduction in pairwise spike trains correlations, as previously described in sensory cortices (Golshani et al., 2009; Rochefort et al., 2009; Cutts and Eglen, 2014; Siegel et al., 2012). To further strengthen the link between the two processes, we bidirectionally manipulated the activity of prefrontal INs by light. In line with our hypothesis, optogenetic stimulation of INs (i.e. decreasing E-I ratio) results in reduced correlations among spike trains. Conversely, optogenetic inhibition of INs (i.e. increasing E-I ratio) increases correlations among spike trains. IN inhibition results in increased spike train correlations even though, in the last portion of the optogenetic protocol, IN displays a paradoxical increase in firing rate. This might indicate that even a transient reduction in inhibition strength might be sufficient to increase neural correlations. Both the age-dependent as well as IN manipulation-induced effects on activity correlations do not impact all spike train pairs in a uniform manner. Rather, neuron pairs that are close to each other are more severely affected than those that are farther apart. While an IN subtype-specific dissection of the mechanisms accounting for neural activity decorrelation is beyond the scope of this study, this effect might be explained by the fact that PV+ INs preferentially provide local inhibition (Tremblay et al., 2016) and have a particularly protracted integration into the prefrontal cortical circuitry (Bitzenhofer et al., 2020). PV+ INs generally provide inhibition at the soma and the axon initial segment (Tremblay et al., 2016), a position that is particularly suited to inhibit the spiking output of PYRs and thereby reduce pairwise spiking correlations. The progressive embedment of PV+ INs in the rodent prefrontal circuitry has also been suggested to be responsible for the increase in the average frequency of the LFP oscillations that are generated by layer 2/3 PYRs (Bitzenhofer et al., 2020). Future studies might shed light on whether this process is related to the decorrelation of brain activity.

In the adult brain and modern artificial neural networks, most neurons are only sparsely active and correlations between neurons are low (Olshausen and Field, 2004; Cun et al., 1990; Frankle and Carbin, 2019; Moreno-Bote et al., 2014). The relationship between correlations, network activity, and behavioral performance is complex and has been reviewed elsewhere (Kohn et al., 2016; Averbeck et al., 2006). Briefly, correlated activity has been proposed as limiting the amount of information that can be encoded and as being energy-inefficient (Olshausen and Field, 2004; Moreno-Bote et al., 2014; Lennie, 2003). Conversely, theoretical and experimental work has shown that correlated neural activity can also enhance the robustness of information transmission (Averbeck et al., 2006; Valente et al., 2021). In sensory cortices, the decorrelated adult activity patterns result from a developmental sparsification process (Golshani et al., 2009; Rochefort et al., 2009; Modol et al., 2020; Cutts and Eglen, 2014; Mizuno et al., 2021). Along the same lines, neural activity has also been shown to become less global and more local throughout development (Siegel et al., 2012; Wosniack et al., 2021). Here, we describe that a similar transition also occurs in a higher-order brain area such as the mPFC. Several non-mutually exclusive mechanisms underlying this phenomenon have been proposed, such as a transition in synaptic plasticity rules (Wosniack et al., 2021), changes in NMDA receptor composition (Mizuno et al., 2021), a decrease in the input resistance of neurons (Golshani et al., 2009), or a combination of these (Rahmati et al., 2017). Here, we propose that an increase in inhibition also plays a role in the decorrelation of neural activity.

Further, we show that the 1/f exponent derived from LFP recorded from the mouse mPFC increases along the first two postnatal weeks. Similarly, the 1/f exponent derived from the frontal derivations of EEG recorded from human babies increases between the 36th and the 45th PCW (from ~2 weeks preterm birth to ~7 weeks after birth, when considering 38 weeks as the average length of human pregnancy; Jukic et al., 2013). This is the opposite of processes taking place in aging (Voytek et al., 2015), between childhood and adulthood (He et al., 2019), and even from the 1st to the 7th month of life (Schaworonkow and Voytek, 2021). The decline in the 1/f exponent (indicative of increased E-I ratio) occurring between childhood and elderliness can be explained by the decline of brain GABA levels (Hermans et al., 2018) and cortical inhibition (Lissemore et al., 2018). This is however unlikely to explain the discrepancy between the current study and the effect reported for babies of 1–7 months of age (Schaworonkow and Voytek, 2021), an age range that borders the one that we investigated. In age-matched mice, a wave of interneuronal cell death takes place around this age (Wong et al., 2018) and might induce an early shift from E-I ratio decrease to E-I ratio increase. Whether a similar process occurs in humans too and might explain the discrepancy between the two studies is still unknown and, due to the ethical and technical limitations of invasive recordings in humans, difficult to address. Directly investigating pairwise spike train correlations is not feasible without resorting to invasive intracranial recordings. It is however noteworthy that, albeit at a different spatial and temporal resolution, the density of phase and amplitude EEG spatial correlations decreases between the last trimester of pregnancy and the first weeks of life (Omidvarnia et al., 2014; Tokariev et al., 2016). Importantly, an impaired structure of frontal correlations at birth, such as the one induced by birth prematurity, is predictive of impaired neurological performance (Tokariev et al., 2019). Future work should address whether changes in E-I ratio also underlie the developmental maturation of these high-level EEG spatial correlations.

Several studies have reported E-I imbalances in the mPFC of mouse models of mental disorders and in patients affected by these diseases (Trakoshis et al., 2020; Sohal and Rubenstein, 2019; Medendorp et al., 2021; Ferguson and Gao, 2018; Hamm et al., 2017; Zick et al., 2018). Following this stream of evidence, we investigated a dataset previously obtained from in vivo electrophysiological recordings from the mPFC of a mouse mimicking the etiology of schizophrenia (Chini et al., 2020; Bitzenhofer et al., 2021), generated by combining two mild stressors, a genetic and an environmental one (Chini et al., 2020; Xu et al., 2019). Their synergistic combination results in severe deficits affecting PYRs that reside in the superficial layers of the mPFC. These neurons display a simplified dendritic arborization, a severe reduction in spine density and firing rate (Chini et al., 2020; Xu et al., 2019). Thus, GE mice have a reduced E-I ratio in the mPFC superficial layers, which leads to diminished STTC values among spike trains. While in sensory areas correlation might limit information carrying capacity, in associative brain areas, like the mPFC, correlations are thought of improving signal readout, and increased correlations have been linked to improved behavioral performance (Valente et al., 2021). This data support the hypothesis that imbalances in E-I ratio might be a possible unifying framework for understanding the circuit dysfunction characterizing neuropsychiatric disorders (Sohal and Rubenstein, 2019; Yizhar et al., 2011; Nelson and Valakh, 2015). In this perspective, it is relevant that the development of E-I ratio can also be quantified, albeit indirectly, also from EEG recordings of newborn babies. The fact that two different EEG datasets yield similar estimations for the age-dependent changes in the 1/f exponent supports the notion that this parameter might be a robust biomarker of E-I ratio development with potential translational relevance.

Materials and methods

Data and code availability

LFP and SUA data that were newly generated for this study are available at the following open-access repository: https://gin.g-node.org/mchini/development_EI_decorrelation.

Code supporting the findings of this study is available at the following open-access repository: https://github.com/mchini/Chini_et_al_EI_decorrelation; Chini, 2021.

Experimental models and subject details

All experiments were performed in compliance with the German laws and following the European Community guidelines regarding the research animals use. All experiments were approved by the local ethical committee (G132/12, G17/015, N18/015). Experiments were carried out on C57BL/6J, Dlx5/6-Cre (Tg(dlx5a-cre)1Mekk/J, Jackson Laboratory), Gad2-IRES-Cre (Gad2tm2(cre)Zjh, Jackson Laboratory), and ArchT (Ai40(RCL-ArchT/EGFP)-D, Jackson Laboratory) mice of both sexes. Mice were housed in individual cages on a 12 hr light/12 hr dark cycle, and were given access to water and food ad libitum. The day of birth was considered P0. To inhibit IN activity, mice from the Dlx5/6-Cre and Gad2-IRES-Cre driver lines were crossed with mice from the ArchT reporter line. To stimulate IN activity, P0-P1 mice from the Dlx5/6-Cre and Gad2-IRES-Cre driver lines were injected in the mPFC with a virus encoding for ChR2 (AAV9-Ef1alpha-DIO-hChR2(ET/TC)-eYFP) as previously described (Xu et al., 2021). Details on the data acquisition and experimental setup of open-access datasets that were used in this project have been previously published (Chini et al., 2020; Bitzenhofer et al., 2021; Schetinin and Jakaite, 2017; Wielek et al., 2019).

In vivo electrophysiology and optogenetics

Surgery

In vivo extracellular recordings were performed from the prelimbic subdivision of the mPFC of non-anesthetized P2-12 mice. Before starting with the surgical procedure, a local anesthetic was applied on the mice neck muscles (0.5% bupivacain/1% lidocaine). The procedure was carried out under isoflurane anesthesia (induction: 5%; maintenance: 1–3%, lower for older pups, higher for younger pups). Neck muscles were cut to reduce muscle artifacts. A craniotomy over the mPFC (0.5 mm anterior to bregma, 0.1–0.5 mm lateral to the midline) was performed by first carefully thinning the skull and then removing it with the use of a motorized drill. Mice were head-fixed into a stereotactic frame and kept on a heated (37°C) surface throughout the entire recording. (Opto)Electrodes (four-shank, 4×4 recording sites, 100 µm between recording sites, 125 µm shank distance; NeuroNexus, Ann Arbor, MI) were slowly inserted into the prelimbic cortex, at a depth varying between 1.4 and 2 mm depending on the age of the mouse. A silver wire implanted into the cerebellum was used as ground and external reference. Before signal acquisition, mice were allowed to recover for 30–45 min, to maximize the quality and stability of the recording as well as single units yield.

Signal acquisition

Extracellular signals were acquired and digitized at a 32 kHz sampling rate after band-pass filtering (0.1–9000 Hz) using an extracellular amplifier (Digital Lynx SX; Neuralynx, Bozeman, MO, Cheetah, Neuralynx, Bozeman, MO).

Optogenetic stimulation

Optical stimuli were delivered by an Arduino Uno-controlled (Arduino, Italy) diode laser (Omicron, Austria). The delivered light stimuli varied in wavelength (472 or 594 nm) according to the experimental paradigm (IN stimulation and inhibition, respectively). Laser power was titrated before signal acquisition and adjusted to the minimum level that induced reliable spiking response. Typical light power at the fiber tip was measured in the range of 15–40 mW/mm2. Optogenetic stimulations consisted of ramp-like stimuli of 3 s duration as previously described (Chini et al., 2020; Bitzenhofer et al., 2017b; Bitzenhofer et al., 2021; Bitzenhofer et al., 2017a). Ramp stimulations were repeated 30–120 times and carried out on the two outmost lateral shanks of the 4-shank electrodes, corresponding to superficial and deep layers of the mPFC. In Figures 3C and 4C, Figure 4—figure supplement 1B and D, these two distinct optogenetic protocols are plotted separately (statistical analysis is however run on the individual mouse level, see ‘Statistical modeling’ section).

Histology

Epifluorescence images of coronal brain sections were acquired postmortem to reconstruct the position of the recording electrode and quantify the amount of eYFP expressing neurons. Only mice in which the electrodes were placed in the correct position were kept for further analysis. eYFP expression was manually quantified on a slide-by-slide basis.

Neural network modeling

The architecture of the network was set similarly to Trakoshis et al., 2020, and is schematically illustrated in Figure 2A. The network was composed of a total of 400 conductance-based LIF units, 80% of which were excitatory (E) (N=320) and 20% were inhibitory (I) (N=80). The units in the network were all connected with each other, with a synaptic weight that was log-normally distributed. Excitatory (E→E, E→I) and inhibitory (I→I and I→E) synapses were mediated by AMPA and GABA conductances, respectively. All baseline parameter values used in the simulations are listed in Table 1. All simulations were performed using Brain2 for Python3.7 (Stimberg et al., 2019).

Table 1. Parameters of the leaky integrate-and-fire network.

Neuron model
Parameter Description Excitatory cells Inhibitory cells
VL Leak membrane potential –70 mV –70 mV
VThr Spike threshold potential –52 mV –52 mV
VRes Reset potential –59 mV –59 mV
τRef Refractory period 2 ms 1 ms
Cm Membrane capacitance 500 pF 500 pF
gL Membrane leak conductance 25 nS 20 nS
τm Membrane time constant 20 ms 10 ms
Synapse model
Parameter Description Excitatory cells Inhibitory cells
EAMPA Reversal potential (AMPA) 0 mV 0 mV
EGABA Reversal potential (GABA) –80 mV –80 mV
gAMPA Conductance (AMPA) lognormal(0, 1)*/50 * nS lognormal(0, 1)*/50 * nS
gGABA Conductance (GABA) lognormal(0, 1)*/12 * nS lognormal(0, 1)*/60 * nS
gAMPA,ext Conductance external input (AMPA) 0.234 * 5 nS
τAMPA Time constant of AMPA decay 2 ms 1 ms
τGABA Time constant of GABA decay 8 ms 8 ms
Current
Parameter Description Excitatory cells Inhibitory cells
Amplitude Max (final) current amplitude / 0.05–0.2 nAmp
Duration Duration of ramp current / 3 s
Interval Interval between currents / 6 s
Sweeps Number of repetitions / 60
*

The two parameters of the lognormal distribution refer to, respectively, the mean and the standard deviation of the underlying normal distribution.

The dynamics of each excitatory and inhibitory cell were governed by the following stochastic differential equation:

CmdVmdt=-gLVm-VL-gAMPAVm-EAMPA-gGABAVm-EGABA+η (1)

with

dgAMPAdt=-gAMPAτAMPA (2)

and

dgGABAdt=-gGABAτGABA (3)

where Vm is the membrane potential, VL is the leak membrane potential, and EAMPA and EGABA denote the AMPA and GABA current reversal potentials, respectively. The synaptic conductance parameters and the corresponding decay time constants are denoted by gAMPA, gGABA and τAMPA, τGABA, respectively. η is a noise term that is generated by an Ornstein-Uhlenbeck process with zero mean. Due to the near-instantaneous rise times of AMPA- and GABA-mediated currents (both typically <0.5 ms), we opted to neglect these in the current simulations. Moreover, synaptic transmission was assumed to be instantaneous (i.e. with zero delay). The excitatory units of the network received an additional external input in the form of AMPA-mediated Poisson spike trains from an external pool of 100 units with a constant spike rate of 1.5 spikes/s.

In order to assess the effect of altered E-I ratio (gE/gI), we parametrically modulated all excitatory (through multiplication with 25 linearly spaced values from 0.1 to 0.7) and all inhibitory (26 linearly spaced values from 0.2 to 1.2) synaptic conductances. The network was simulated for a duration of 30 s for each of the 25×26 parameter combinations. For each parameter combination, the LFP of the network was computed by taking the sum of the absolute values of the AMPA and GABA currents on all excitatory cells (Trakoshis et al., 2020). Neuronal correlation was estimated by means of the STTC (see below), assessed at a lag of 1 s.

To mimic optogenetic modulation of IN activity, we added an external current to the stochastic differential equation regulating the dynamics of inhibitory neurons:

CmdVmdt=-gLVm-VL-gAMPAVm-EAMPA-gGABAVm-EGABA+IStim+η (4a)

where Istim is a 3-s-long ramp-like inhibitory or excitatory current, administered in repeated sweeps, with an interval of 6 s.

Electrophysiological analysis

Data were analyzed with custom-written algorithms in the MATLAB and Python environment that are available on the following github repository: https://github.com/mchini/Chini_et_al_EI_decorrelation (copy archived at swh:1:rev:9a07c56f36c80a60a44a6607a5a4061a37d96ae7; Chini, 2021).

Detection of active periods

During early development, brain activity is characterized by an alternation of periods of isoelectric traces (silent periods) and oscillatory bursts (active periods). To detect and quantify the properties of active periods, we developed a novel detection algorithm. For this, the extracellular signal was band-pass filtered (4–20 Hz) and downsampled to 100 Hz, before being averaged across recording electrodes. The average signal (raw and z-scored) was then passed through a boxcar square filter (500 ms) on which a hysteresis threshold was applied. Active periods were firstly detected as oscillatory peaks exceeding an absolute or relative threshold (100 µV or four standard deviations, respectively) and subsequently extended to all neighboring time points that exceed a lower threshold (50 µV or two standard deviations, respectively). The combination of absolute and relative thresholding makes this approach suitable to a wide range of signals, from the highly discontinuous brain activity of P2 mice, to the nearly continuous brain activity of P11-12 mice (Figure 1A–B). Neighboring active periods whose inter-active period interval was shorter than 1 s were merged. Active periods whose duration was smaller than 300 ms were discarded.

Power spectral density

PSDs for mouse and human data (see below for exception) were computed with the mtspecgramc function of the Chronux Toolbox (10-s-long windows, 5 s overlap). Median averaging was the preferred measure of central tendency (Izhikevich et al., 2018). To quantify the PSD modulation by IN optogenetic stimulation/inhibition, we computed the MI (see below) of the PSD computed on the last 1.5 s of the optogenetic stimulation with the PSD computed on the 1.5 s preceding stimulus delivery.

EEG preprocessing

EEG signal was extracted only from frontal electrodes (Fp1, F7, F3, Fp2, F8, F4, Fpz, when available) and re-referenced to a common average reference before further analysis. From the EEG dataset of 1100 sleeping babies (Schetinin and Jakaite, 2017), epochs whose average envelope amplitude exceeded two standard deviations from the mean were considered as possible artifacts and were removed from further analysis. No preprocessing was applied to the EEG dataset of sleeping babies recorded at 40 and 43 PCW, as PSDs were already included in the freely available data (Wielek et al., 2019).

1/f exponent

The 1/f exponent was extracted on the 5–20 and 5–45 Hz (human and mouse data, respectively) frequency range of PSDs using the FOOOF package (Donoghue et al., 2020) with the ‘fixed’ aperiodic mode. To quantify the 1/f exponent modulation by IN optogenetic stimulation/inhibition, we compared the exponent obtained by PSDs computed on the second half of the optogenetic stimulation with the baseline exponent.

Spike sorting

Spike sorting was performed using Klusta (Rossant et al., 2016). Automatically obtained clusters were then manually curated using phy (https://github.com/cortex-lab/phy, Rossant, 2022).

Spike time tiling coefficient

The STTC, a metric that tracks correlations between spike trains and is robust to changes in firing rate, was calculated as previously described (Cutts and Eglen, 2014; Yang et al., 2021; Figure 3A):

STTC=12PA-TB1-PATB+PB-TA1-PBTA (4b)

where PA is defined as the proportion of spikes in spike train A that falls within ±Δt of a spike from spike train B. TA is defined as the proportion of time that occurs within (is ‘tiled’ by) ±Δt from the spikes of spike train A. The same applies for PB and TB. The ‘lag’ parameter ±Δt controls the ‘timescale’ at which the STTC is computed, a parameter that we systematically varied across more than three orders of magnitude (from 2.5 ms to 10 s). Baseline STTC analysis was limited to spike trains pairs that were recorded for at least an hour and for which both spike trains had at least 50 spikes (40,921 of 56,613 spike train pairs). To quantify the STTC modulation by IN optogenetic stimulation/inhibition, we compared the STTC derived by spike matrices obtained during the 3 s optogenetic stimulation with the STTC derived by spike matrices obtained during the 3 s preceding optogenetic stimulation.

Modulation index

The modulation index (MI) is a normalization strategy that was used to compute optogenetically induced changes in firing rate and LFP power. The MI has the desirable property of bounding the normalized value between –1 and 1. MI was computed as:

MI=ValuePRE-ValueSTIMValuePRE+ValueSTIM

Optogenetic modulation of electrophysiological parameters

Modulation of firing rate by optogenetic manipulation was quantified using the MI and signed-rank testing that compared the firing rate during the last 1.5 s of optical stimulation with the firing rate during the 1.5 s preceding stimulus delivery.

PCA of spike matrices during optogenetic stimulations

The first two PCA components of the spike during optogenetic stimulations were computed on trial-averaged spike trains that were convolved with a Gaussian window (500 ms length, 50 ms standard deviation) and z-scored across the time dimension.

Statistical modeling

Statistical modeling was carried out in the R environment. All the scripts and the processed data on which the analysis is based are available from the following github repository: https://github.com/mchini/Chini_et_al_EI_decorrelation (Chini, 2021).

Nested data were analyzed with (generalized) linear mixed-effects models (lmer and glmer functions of the lme4 R package; Bates et al., 2014). Depending on the specific experimental design, we used ‘mouse’ or ‘subject’ as random effects. For statistical analysis of STTC, to maximize interpretability of the results (i.e. avoid multiple triple interactions), each STTC lag was investigated as its own independent variable, using identical statistical models. Regression on that data, upon visual inspection seemed to be better fit by an exponential curve, were fitted with generalized linear (mixed-effect) models (family = Gamma, α=1, link = inverse). Proportions (e.g. the proportion of activated/inhibited units) were also fitted with generalized linear (mixed-effect) models (family = Binomial, link = logit). Statistical significance for linear mixed-effects models were computed with the lmerTest R package (Kuznetsova et al., 2017), using the Satterthwaite’s degrees of freedom method. When possible, model selection was performed according to experimental design. When this was not possible, models were compared using the compare_performance function of the performance R package (Lüdecke et al., 2021b), and model choice was based on an holistic comparison of AIC, BIC, RMSE, and R2. Model output was plotted with the plot_model (type=‘pred’) function of the sjPlot R package (Lüdecke et al., 2021a); 95% confidence intervals were computed using the confint R function. Post hoc analysis with Tukey multiple comparison correction was carried out using the emmeans and emtrends functions of the emmeans R package (Lenth et al., 2020).

Acknowledgements

We thank Amit Marmelshtein, Stefano Panzeri, Giulio Bondanelli, Sebastian Bitzenhofer, Johanna Kostka, Jastyn Pöpplau, and Lingzhen Song for valuable discussions and feedback on the manuscript, and P Putthoff, A Marquardt, and A Dahlmann for excellent technical assistance. This work was funded by grants from the European Research Council (ERC-2015-CoG 681577 to ILH-O), Marie Curie Training Network euSNN (MSCA-ITN-H2020-860563 to ILH-O), Horizon2020 DEEPER 101016787, the German Research Foundation (437610067, 178316478, and 302153259 to ILH-O) and Landesforschungsförderung Hamburg (LFF76, LFF73 to ILH-O).

Funding Statement

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

Contributor Information

Mattia Chini, Email: mattia.chini@zmnh.uni-hamburg.de.

Liset M de la Prida, Instituto Cajal, Spain.

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

Funding Information

This paper was supported by the following grants:

  • European Research Council ERC-2015-CoG 681577 to Ileana Hanganu-Opatz.

  • H2020 Marie Skłodowska-Curie Actions Marie Curie Training Network euSNN MSCA-ITN-H2020-860563 to Ileana Hanganu-Opatz.

  • Horizon 2020 Framework Programme DEEPER 101016787 to Ileana Hanganu-Opatz.

  • Deutsche Forschungsgemeinschaft 437610067 to Ileana Hanganu-Opatz.

  • Landesforschungsfoerderung Hamburg LFF76 to Ileana Hanganu-Opatz.

  • Deutsche Forschungsgemeinschaft 178316478 to Ileana Hanganu-Opatz.

  • Deutsche Forschungsgemeinschaft 302153259 to Ileana Hanganu-Opatz.

  • Landesforschungsfoerderung Hamburg LFF73 to Ileana Hanganu-Opatz.

  • American Friends of the Alexander von Humboldt Foundation Feodor-Lynen Fellowship to Thomas Pfeffer.

Additional information

Competing interests

No competing interests declared.

Author contributions

Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing.

Software, Formal analysis, Visualization, Writing - review and editing.

Conceptualization, Supervision, Funding acquisition, Project administration, Writing - review and editing.

Ethics

Human subjects: No new human data was collected for this study, only open-access datasets were used.

All experiments were performed in compliance with the German laws and following the European Community guidelines regarding the research animals use. All experiments were approved by the local ethical committee (G132/12, G17/015, N18/015).

Additional files

Supplementary file 1. Detailed statistical results.
elife-78811-supp1.xlsx (36.2KB, xlsx)
MDAR checklist

Data availability

LFP and SUA data that were newly generated for this study are available at the following open-access repository: https://gin.g-node.org/mchini/development_EI_decorrelation. Code supporting the findings of this study is available at the following open-access repository: https://github.com/mchini/Chini_et_al_EI_decorrelation, (copy archived at swh:1:rev:9a07c56f36c80a60a44a6607a5a4061a37d96ae7).

References

  1. Akin O, Bajar BT, Keles MF, Frye MA, Zipursky SL. Cell-type-specific patterned stimulus-independent neuronal activity in the Drosophila visual system during synapse formation. Neuron. 2019;101:894–904. doi: 10.1016/j.neuron.2019.01.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Antoine MW, Langberg T, Schnepel P, Feldman DE. Increased excitation-inhibition ratio stabilizes synapse and circuit excitability in four autism mouse models. Neuron. 2019;101:648–661. doi: 10.1016/j.neuron.2018.12.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Averbeck BB, Latham PE, Pouget A. Neural correlations, population coding and computation. Nature Reviews. Neuroscience. 2006;7:358–366. doi: 10.1038/nrn1888. [DOI] [PubMed] [Google Scholar]
  4. Avitan L, Pujic Z, Mölter J, Van De Poll M, Sun B, Teng H, Amor R, Scott EK, Goodhill GJ. Spontaneous activity in the zebrafish tectum reorganizes over development and is influenced by visual experience. Current Biology. 2017;27:2407–2419. doi: 10.1016/j.cub.2017.06.056. [DOI] [PubMed] [Google Scholar]
  5. Bates D, Mächler M, Bolker B, Walker S. Fitting Linear Mixed-Effects Models Using Lme4. arXiv. 2014 doi: 10.18637/jss.v067.i01. http://arxiv.org/abs/1406.5823 [DOI]
  6. Ben-Ari Y. Excitatory actions of gaba during development: the nature of the nurture. Nature Reviews. Neuroscience. 2002;3:728–739. doi: 10.1038/nrn920. [DOI] [PubMed] [Google Scholar]
  7. Bitzenhofer SH, Ahlbeck J, Hanganu-Opatz IL. Methodological approach for optogenetic manipulation of neonatal neuronal networks. Frontiers in Cellular Neuroscience. 2017a;11:239. doi: 10.3389/fncel.2017.00239. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bitzenhofer SH, Ahlbeck J, Wolff A, Wiegert JS, Gee CE, Oertner TG, Hanganu-Opatz IL. Layer-specific optogenetic activation of pyramidal neurons causes beta-gamma entrainment of neonatal networks. Nature Communications. 2017b;8:14563. doi: 10.1038/ncomms14563. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bitzenhofer SH, Pöpplau JA, Hanganu-Opatz I. Gamma activity accelerates during prefrontal development. eLife. 2020;9:e56795. doi: 10.7554/eLife.56795. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Bitzenhofer SH, Pöpplau JA, Chini M, Marquardt A, Hanganu-Opatz IL. A transient developmental increase in prefrontal activity alters network maturation and causes cognitive dysfunction in adult mice. Neuron. 2021;109:1350–1364. doi: 10.1016/j.neuron.2021.02.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Blankenship AG, Hamby AM, Firl A, Vyas S, Maxeiner S, Willecke K, Feller MB. The role of neuronal connexins 36 and 45 in shaping spontaneous firing patterns in the developing retina. The Journal of Neuroscience. 2011;31:9998–10008. doi: 10.1523/JNEUROSCI.5640-10.2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Buzsáki G, Mizuseki K. The log-dynamic brain: how skewed distributions affect network operations. Nature Reviews. Neuroscience. 2014;15:264–278. doi: 10.1038/nrn3687. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Che A, Babij R, Iannone AF, Fetcho RN, Ferrer M, Liston C, Fishell G, De Marco García NV. Layer I interneurons sharpen sensory maps during neonatal development. Neuron. 2018;99:98–116. doi: 10.1016/j.neuron.2018.06.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Chini M, Gretenkord S, Kostka JK, Pöpplau JA, Cornelissen L, Berde CB, Hanganu-Opatz IL, Bitzenhofer SH. Neural correlates of anesthesia in newborn mice and humans. Frontiers in Neural Circuits. 2019;13:38. doi: 10.3389/fncir.2019.00038. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Chini M, Pöpplau JA, Lindemann C, Carol-Perdiguer L, Hnida M, Oberländer V, Xu X, Ahlbeck J, Bitzenhofer SH, Mulert C, Hanganu-Opatz IL. Resolving and rescuing developmental miswiring in a mouse model of cognitive impairment. Neuron. 2020;105:60–74. doi: 10.1016/j.neuron.2019.09.042. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Chini M. Decorrelation. swh:1:rev:9a07c56f36c80a60a44a6607a5a4061a37d96ae7Software Heritage. 2021 https://archive.softwareheritage.org/swh:1:dir:3dea3196c72ed865c568c0daae2a331716872133;origin=https://github.com/mchini/Chini_et_al_EI_decorrelation;visit=swh:1:snp:5d83d8d2553613276ac1f6bebf713a48fcee2b66;anchor=swh:1:rev:9a07c56f36c80a60a44a6607a5a4061a37d96ae7
  17. Chini M., Hanganu-Opatz IL. Prefrontal cortex development in health and disease: lessons from rodents and humans. Trends in Neurosciences. 2021;44:227–240. doi: 10.1016/j.tins.2020.10.017. [DOI] [PubMed] [Google Scholar]
  18. Cirelli C, Tononi G. Cortical development, electroencephalogram rhythms, and the sleep/wake cycle. Biological Psychiatry. 2015;77:1071–1078. doi: 10.1016/j.biopsych.2014.12.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Claudi F, Tyson AL, Petrucco L, Margrie TW, Portugues R, Branco T. Visualizing anatomically registered data with brainrender. eLife. 2021;10:e65751. doi: 10.7554/eLife.65751. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Cogswell M, Ahmed F, Girshick R, Zitnick L, Batra D. Reducing Overfitting in Deep Networks by Decorrelating Representations. arXiv. 2016 http://arxiv.org/abs/1511.06068
  21. Colombo MA, Napolitani M, Boly M, Gosseries O, Casarotto S, Rosanova M, Brichant JF, Boveroux P, Rex S, Laureys S, Massimini M, Chieregato A, Sarasso S. The spectral exponent of the resting EEG indexes the presence of consciousness during unresponsiveness induced by propofol, xenon, and ketamine. NeuroImage. 2019;189:631–644. doi: 10.1016/j.neuroimage.2019.01.024. [DOI] [PubMed] [Google Scholar]
  22. Cossart R. Operational hub cells: a morpho-physiologically diverse class of gabaergic neurons united by a common function. Current Opinion in Neurobiology. 2014;26:51–56. doi: 10.1016/j.conb.2013.12.002. [DOI] [PubMed] [Google Scholar]
  23. Cun YL, Denker JS, Solla SA. Advances in Neural Information Processing Systems. Morgan Kaufmann; 1990. Optimal brain damage; pp. 598–605. [Google Scholar]
  24. Cutts CS, Eglen SJ. Detecting pairwise correlations in spike trains: an objective comparison of methods and application to the study of retinal waves. The Journal of Neuroscience. 2014;34:14288–14303. doi: 10.1523/JNEUROSCI.2767-14.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Donoghue T, Haller M, Peterson EJ, Varma P, Sebastian P, Gao R, Noto T, Lara AH, Wallis JD, Knight RT, Shestyuk A, Voytek B. Parameterizing neural power spectra into periodic and aperiodic components. Nature Neuroscience. 2020;23:1655–1665. doi: 10.1038/s41593-020-00744-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Dorrn AL, Yuan K, Barker AJ, Schreiner CE, Froemke RC. Developmental sensory experience balances cortical excitation and inhibition. Nature. 2010;465:932–936. doi: 10.1038/nature09119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Duan ZRS, Che A, Chu P, Modol L, Bollmann Y, Babij R, Fetcho RN, Otsuka T, Fuccillo MV, Liston C, Pisapia DJ, Cossart R, De Marco García NV. GABAergic restriction of network dynamics regulates interneuron survival in the developing cortex. Neuron. 2018;105:75–92. doi: 10.1016/j.neuron.2019.10.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Favuzzi E, Deogracias R, Marques-Smith A, Maeso P, Jezequel J, Exposito-Alonso D, Balia M, Kroon T, Hinojosa AJ, F Maraver E, Rico B. Distinct molecular programs regulate synapse specificity in cortical inhibitory circuits. Science. 2019;363:413–417. doi: 10.1126/science.aau8977. [DOI] [PubMed] [Google Scholar]
  29. Ferguson BR, Gao WJ. PV interneurons: critical regulators of E/I balance for prefrontal cortex-dependent behavior and psychiatric disorders. Frontiers in Neural Circuits. 2018;12:37. doi: 10.3389/fncir.2018.00037. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Frankle J, Carbin M. The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks. arXiv. 2019 http://arxiv.org/abs/1803.03635
  31. Gao R, Penzes P. Common mechanisms of excitatory and inhibitory imbalance in schizophrenia and autism spectrum disorders. Current Molecular Medicine. 2015;15:146–167. doi: 10.2174/1566524015666150303003028. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Gao R, Peterson EJ, Voytek B. Inferring synaptic excitation/inhibition balance from field potentials. NeuroImage. 2017;158:70–78. doi: 10.1016/j.neuroimage.2017.06.078. [DOI] [PubMed] [Google Scholar]
  33. Garaschuk O, Linn J, Eilers J, Konnerth A. Large-scale oscillatory calcium waves in the immature cortex. Nature Neuroscience. 2000;3:452–459. doi: 10.1038/74823. [DOI] [PubMed] [Google Scholar]
  34. Golshani P, Gonçalves JT, Khoshkhoo S, Mostany R, Smirnakis S, Portera-Cailliau C. Internally mediated developmental desynchronization of neocortical network activity. The Journal of Neuroscience. 2009;29:10890–10899. doi: 10.1523/JNEUROSCI.2012-09.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Goltstein PM, Montijn JS, Pennartz CMA. Effects of isoflurane anesthesia on ensemble patterns of ca2+ activity in mouse v1: reduced direction selectivity independent of increased correlations in cellular activity. PLOS ONE. 2015;10:e0118277. doi: 10.1371/journal.pone.0118277. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Gour A, Boergens KM, Heike N, Hua Y, Laserstein P, Song K, Helmstaedter M. Postnatal connectomic development of inhibition in mouse barrel cortex. Science. 2021;371:eabb4534. doi: 10.1126/science.abb4534. [DOI] [PubMed] [Google Scholar]
  37. Greenberg DS, Houweling AR, Kerr JND. Population imaging of ongoing neuronal activity in the visual cortex of awake rats. Nature Neuroscience. 2008;11:749–751. doi: 10.1038/nn.2140. [DOI] [PubMed] [Google Scholar]
  38. Guan WQ, Cao JW, Liu LY, Zhao ZH, Fu YH, Yu YC. Eye opening differentially modulates inhibitory synaptic transmission in the developing visual cortex. eLife. 2018;6:e32337. doi: 10.7554/eLife.32337. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Hamm JP, Peterka DS, Gogos JA, Yuste R. Altered cortical ensembles in mouse models of schizophrenia. Neuron. 2017;94:153–167. doi: 10.1016/j.neuron.2017.03.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Hazan L, Ziv NE. Activity dependent and independent determinants of synaptic size diversity. The Journal of Neuroscience. 2020;40:2828–2848. doi: 10.1523/JNEUROSCI.2181-19.2020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. He W, Donoghue T, Sowman PF, Seymour RA, Brock J, Crain S, Voytek B, Hillebrand A. Co-Increasing Neuronal Noise and Beta Power in the Developing Brain. bioRxiv. 2019 doi: 10.1101/839258. [DOI]
  42. Hendry SH, Schwark HD, Jones EG, Yan J. Numbers and proportions of GABA-immunoreactive neurons in different areas of monkey cerebral cortex. The Journal of Neuroscience. 1987;7:1503–1519. doi: 10.1523/JNEUROSCI.07-05-01503.1987. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Hermans L, Leunissen I, Pauwels L, Cuypers K, Peeters R, Puts NAJ, Edden RAE, Swinnen SP. Brain GABA levels are associated with inhibitory control deficits in older adults. The Journal of Neuroscience. 2018;38:7844–7851. doi: 10.1523/JNEUROSCI.0760-18.2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Izhikevich L, Gao R, Peterson E, Voytek B. Measuring the Average Power of Neural Oscillations. bioRxiv. 2018 doi: 10.1101/441626. [DOI]
  45. Jukic AM, Baird DD, Weinberg CR, McConnaughey DR, Wilcox AJ. Length of human pregnancy and contributors to its natural variation. Human Reproduction. 2013;28:2848–2855. doi: 10.1093/humrep/det297. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Kalemaki K, Velli A, Christodoulou O, Denaxa M, Karagogeos D, Sidiropoulou K. The developmental changes in intrinsic and synaptic properties of prefrontal neurons enhance local network activity from the second to the third postnatal weeks in mice. Cerebral Cortex. 2021;bhab438:bhab438. doi: 10.1093/cercor/bhab438. [DOI] [PubMed] [Google Scholar]
  47. Kato HK, Asinof SK, Isaacson JS. Network-level control of frequency tuning in auditory cortex. Neuron. 2017;95:412–423. doi: 10.1016/j.neuron.2017.06.019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Kepecs A, Fishell G. Interneuron cell types are fit to function. Nature. 2014;505:318–326. doi: 10.1038/nature12983. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Khazipov R, Sirota A, Leinekugel X, Holmes GL, Ben-Ari Y, Buzsáki G. Early motor activity drives spindle bursts in the developing somatosensory cortex. Nature. 2004;432:758–761. doi: 10.1038/nature03132. [DOI] [PubMed] [Google Scholar]
  50. Kirmse K, Kummer M, Kovalchuk Y, Witte OW, Garaschuk O, Holthoff K. GABA depolarizes immature neurons and inhibits network activity in the neonatal neocortex in vivo. Nature Communications. 2015;6:7750. doi: 10.1038/ncomms8750. [DOI] [PubMed] [Google Scholar]
  51. Kohn A, Coen-Cagli R, Kanitscheider I, Pouget A. Correlations and neuronal population information. Annual Review of Neuroscience. 2016;39:237–256. doi: 10.1146/annurev-neuro-070815-013851. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Kuznetsova A, Brockhoff PB, Christensen RHB. Christensen, lmertest package: tests in linear mixed effects models. Journal of Statistical Software. 2017;82:1–26. doi: 10.18637/jss.v082.i13. [DOI] [Google Scholar]
  53. Lendner JD, Helfrich RF, Mander BA, Romundstad L, Lin JJ, Walker MP, Larsson PG, Knight RT. An electrophysiological marker of arousal level in humans. eLife. 2020;9:e55092. doi: 10.7554/eLife.55092. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Lennie P. The cost of cortical computation. Current Biology. 2003;13:493–497. doi: 10.1016/s0960-9822(03)00135-0. [DOI] [PubMed] [Google Scholar]
  55. Lenth RV, Buerkner P, Herve M, Love J, Riebl H, Singmann H. Emmeans: estimated marginal means, aka least-squares means. 1.8.0Emmeans. 2020 https://CRAN.R-project.org/package=emmeans
  56. Lim L, Mi D, Llorca A, Marín O. Development and functional diversification of cortical interneurons. Neuron. 2018;100:294–313. doi: 10.1016/j.neuron.2018.10.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Lissemore JI, Bhandari A, Mulsant BH, Lenze EJ, Reynolds CF, Karp JF, Rajji TK, Noda Y, Zomorrodi R, Sibille E, Daskalakis ZJ, Blumberger DM. Reduced gabaergic cortical inhibition in aging and depression. Neuropsychopharmacology. 2018;43:2277–2284. doi: 10.1038/s41386-018-0093-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Lombardi F, Herrmann HJ, de Arcangelis L. Balance of excitation and inhibition determines 1/f power spectrum in neuronal networks. Chaos. 2017;27:047402. doi: 10.1063/1.4979043. [DOI] [PubMed] [Google Scholar]
  59. Lüdecke D, Bartel A, Schwemmer C. Titz, sjplot: data visualization for statistics in social science. Zenoda. 2021a https://CRAN.R-project.org/package=sjPlot
  60. Lüdecke D, Ben-Shachar M, Patil I, Waggoner P, Makowski D. Performance: an R package for assessment, comparison and testing of statistical models. Journal of Open Source Software. 2021b;6:3139. doi: 10.21105/joss.03139. [DOI] [Google Scholar]
  61. Luongo FJ, Horn ME, Sohal VS. Putative microcircuit-level substrates for attention are disrupted in mouse models of autism. Biological Psychiatry. 2016;79:667–675. doi: 10.1016/j.biopsych.2015.04.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Markram H, Toledo-Rodriguez M, Wang Y, Gupta A, Silberberg G, Wu C. Interneurons of the neocortical inhibitory system. Nature Reviews. Neuroscience. 2004;5:793–807. doi: 10.1038/nrn1519. [DOI] [PubMed] [Google Scholar]
  63. Marques-Smith A, Lyngholm D, Kaufmann A-K, Stacey JA, Hoerder-Suabedissen A, Becker EBE, Wilson MC, Molnár Z, Butt SJB. A transient translaminar gabaergic interneuron circuit connects thalamocortical recipient layers in neonatal somatosensory cortex. Neuron. 2016;89:536–549. doi: 10.1016/j.neuron.2016.01.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Mazzoni A, Panzeri S, Logothetis NK, Brunel N. Encoding of naturalistic stimuli by local field potential spectra in networks of excitatory and inhibitory neurons. PLOS Computational Biology. 2008;4:e1000239. doi: 10.1371/journal.pcbi.1000239. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Medendorp WE, Bjorefeldt A, Crespo EL, Prakash M, Pal A, Waddell ML, Moore CI, Hochgeschwender U. Selective postnatal excitation of neocortical pyramidal neurons results in distinctive behavioral and circuit deficits in adulthood. IScience. 2021;24:102157. doi: 10.1016/j.isci.2021.102157. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Mizuno H, Rao MS, Mizuno H, Sato T, Nakazawa S, Iwasato T. NMDA receptor enhances correlation of spontaneous activity in neonatal barrel cortex. The Journal of Neuroscience. 2021;41:1207–1217. doi: 10.1523/JNEUROSCI.0527-20.2020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Modol L, Bollmann Y, Tressard T, Baude A, Che A, Duan ZRS, Babij R, De Marco García NV, Cossart R. Assemblies of perisomatic gabaergic neurons in the developing barrel cortex. Neuron. 2020;105:93–105. doi: 10.1016/j.neuron.2019.10.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Moore AK, Weible AP, Balmer TS, Trussell LO, Wehr M. Rapid rebalancing of excitation and inhibition by cortical circuitry. Neuron. 2018;97:1341–1355. doi: 10.1016/j.neuron.2018.01.045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Moreno-Bote R, Beck J, Kanitscheider I, Pitkow X, Latham P, Pouget A. Information-limiting correlations. Nature Neuroscience. 2014;17:1410–1417. doi: 10.1038/nn.3807. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Murata Y, Colonnese MT. GABAergic interneurons excite neonatal hippocampus in vivo. Science Advances. 2020;6:eaba1430. doi: 10.1126/sciadv.aba1430. [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Nelson SB, Valakh V. Excitatory/inhibitory balance and circuit homeostasis in autism spectrum disorders. Neuron. 2015;87:684–698. doi: 10.1016/j.neuron.2015.07.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Olshausen BA, Field DJ. Sparse coding of sensory inputs. Current Opinion in Neurobiology. 2004;14:481–487. doi: 10.1016/j.conb.2004.07.007. [DOI] [PubMed] [Google Scholar]
  73. Omidvarnia A, Fransson P, Metsäranta M, Vanhatalo S. Functional bimodality in the brain networks of preterm and term human newborns. Cerebral Cortex. 2014;24:2657–2668. doi: 10.1093/cercor/bht120. [DOI] [PubMed] [Google Scholar]
  74. Picardo MA, Guigue P, Bonifazi P, Batista-Brito R, Allene C, Ribas A, Fishell G, Baude A, Cossart R. Pioneer GABA cells comprise a subpopulation of hub neurons in the developing hippocampus. Neuron. 2011;71:695–709. doi: 10.1016/j.neuron.2011.06.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  75. Rahmati V, Kirmse K, Holthoff K, Schwabe L, Kiebel SJ. Developmental emergence of sparse coding: A dynamic systems approach. Scientific Reports. 2017;7:13015. doi: 10.1038/s41598-017-13468-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  76. Rochefort NL, Garaschuk O, Milos RI, Narushima M, Marandi N, Pichler B, Kovalchuk Y, Konnerth A. Sparsification of neuronal activity in the visual cortex at eye-opening. PNAS. 2009;106:15049–15054. doi: 10.1073/pnas.0907660106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  77. Rossant C, Kadir SN, Goodman DFM, Schulman J, Hunter MLD, Saleem AB, Grosmark A, Belluscio M, Denfield GH, Ecker AS, Tolias AS, Solomon S, Buzsaki G, Carandini M, Harris KD. Spike sorting for large, dense electrode arrays. Nature Neuroscience. 2016;19:634–641. doi: 10.1038/nn.4268. [DOI] [PMC free article] [PubMed] [Google Scholar]
  78. Rossant C. Phy: interactive visualization and manual spike sorting of large-scale ephys data. v2.0b1GitHub. 2022 https://github.com/cortex-lab/phy
  79. Roux L, Stark E, Sjulson L, Buzsáki G. In vivo optogenetic identification and manipulation of gabaergic interneuron subtypes. Current Opinion in Neurobiology. 2014;26:88–95. doi: 10.1016/j.conb.2013.12.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  80. Sadeh S, Silver RA, Mrsic-Flogel TD, Muir DR. Assessing the role of inhibition in stabilizing neocortical networks requires large-scale perturbation of the inhibitory population. The Journal of Neuroscience. 2017;37:12050–12067. doi: 10.1523/JNEUROSCI.0963-17.2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  81. Sadeh S, Clopath C. Patterned perturbation of inhibition can reveal the dynamical structure of neural processing. eLife. 2020;9:e52757. doi: 10.7554/eLife.52757. [DOI] [PMC free article] [PubMed] [Google Scholar]
  82. Sanzeni A, Akitake B, Goldbach HC, Leedy CE, Brunel N, Histed MH. Inhibition stabilization is a widespread property of cortical networks. eLife. 2020;9:e54875. doi: 10.7554/eLife.54875. [DOI] [PMC free article] [PubMed] [Google Scholar]
  83. Schaworonkow N, Voytek B. Longitudinal changes in aperiodic and periodic activity in electrophysiological recordings in the first seven months of life. Developmental Cognitive Neuroscience. 2021;47:100895. doi: 10.1016/j.dcn.2020.100895. [DOI] [PMC free article] [PubMed] [Google Scholar]
  84. Schetinin V, Jakaite L. Extraction of features from sleep EEG for bayesian assessment of brain development. PLOS ONE. 2017;12:e0174027. doi: 10.1371/journal.pone.0174027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  85. Sessolo M, Marcon I, Bovetti S, Losi G, Cammarota M, Ratto GM, Fellin T, Carmignoto G. Parvalbumin-positive inhibitory interneurons oppose propagation but favor generation of focal epileptiform activity. The Journal of Neuroscience. 2015;35:9544–9557. doi: 10.1523/JNEUROSCI.5117-14.2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  86. Shen J, Colonnese MT. Development of activity in the mouse visual cortex. The Journal of Neuroscience. 2016;36:12259–12275. doi: 10.1523/JNEUROSCI.1903-16.2016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  87. Shew WL, Yang H, Yu S, Roy R, Plenz D. Information capacity and transmission are maximized in balanced cortical networks with neuronal avalanches. Journal of Neuroscience. 2014;31:55–63. doi: 10.1523/JNEUROSCI.4637-10.2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  88. Sidman RL, Rakic P. Neuronal migration, with special reference to developing human brain: a review. Brain Research. 1973;62:1–35. doi: 10.1016/0006-8993(73)90617-3. [DOI] [PubMed] [Google Scholar]
  89. Siegel F, Heimel JA, Peters J, Lohmann C. Peripheral and central inputs shape network dynamics in the developing visual cortex in vivo. Current Biology. 2012;22:253–258. doi: 10.1016/j.cub.2011.12.026. [DOI] [PubMed] [Google Scholar]
  90. Smith MA, Kohn A. Spatial and temporal scales of neuronal correlation in primary visual cortex. The Journal of Neuroscience. 2008;28:12591–12603. doi: 10.1523/JNEUROSCI.2929-08.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  91. Sohal VS, Rubenstein JLR. Excitation-inhibition balance as a framework for investigating mechanisms in neuropsychiatric disorders. Molecular Psychiatry. 2019;24:1248–1257. doi: 10.1038/s41380-019-0426-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  92. Stimberg M, Brette R, Goodman DF. Brian 2, an intuitive and efficient neural simulator. eLife. 2019;8:e47314. doi: 10.7554/eLife.47314. [DOI] [PMC free article] [PubMed] [Google Scholar]
  93. Tokariev A, Videman M, Palva JM, Vanhatalo S. Functional brain connectivity develops rapidly around term age and changes between vigilance states in the human newborn. Cerebral Cortex. 2016;26:4540–4550. doi: 10.1093/cercor/bhv219. [DOI] [PubMed] [Google Scholar]
  94. Tokariev A, Stjerna S, Lano A, Metsäranta M, Palva JM, Vanhatalo S. Preterm birth changes networks of newborn cortical activity. Cerebral Cortex. 2019;29:814–826. doi: 10.1093/cercor/bhy012. [DOI] [PubMed] [Google Scholar]
  95. Trakoshis S, Martínez-Cañada P, Rocchi F, Canella C, You W, Chakrabarti B, Ruigrok AN, Bullmore ET, Suckling J, Markicevic M, Zerbi V, MRC AIMS Consortium. Baron-Cohen S, Gozzi A, Lai M-C, Panzeri S, Lombardo MV. Intrinsic excitation-inhibition imbalance affects medial prefrontal cortex differently in autistic men versus women. eLife. 2020;9:e55684. doi: 10.7554/eLife.55684. [DOI] [PMC free article] [PubMed] [Google Scholar]
  96. Tremblay R, Lee S, Rudy B. GABAergic interneurons in the neocortex: from cellular properties to circuits. Neuron. 2016;91:260–292. doi: 10.1016/j.neuron.2016.06.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  97. Trujillo CA, Gao R, Negraes PD, Gu J, Buchanan J, Preissl S, Wang A, Wu W, Haddad GG, Chaim IA, Domissy A, Vandenberghe M, Devor A, Yeo GW, Voytek B, Muotri AR. Complex oscillatory waves emerging from cortical organoids model early human brain network development. Cell Stem Cell. 2019;25:558–569. doi: 10.1016/j.stem.2019.08.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  98. Tsodyks MV, Skaggs WE, Sejnowski TJ, McNaughton BL. Paradoxical effects of external modulation of inhibitory interneurons. The Journal of Neuroscience. 1997;17:4382–4388. doi: 10.1523/JNEUROSCI.17-11-04382.1997. [DOI] [PMC free article] [PubMed] [Google Scholar]
  99. Tuncdemir SN, Wamsley B, Stam FJ, Osakada F, Goulding M, Callaway EM, Rudy B, Fishell G. Early somatostatin interneuron connectivity mediates the maturation of deep layer cortical circuits. Neuron. 2016;89:521–535. doi: 10.1016/j.neuron.2015.11.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  100. Valente M, Pica G, Bondanelli G, Moroni M, Runyan CA, Morcos AS, Harvey CD, Panzeri S. Correlations enhance the behavioral readout of neural population activity in association cortex. Nature Neuroscience. 2021;24:975–986. doi: 10.1038/s41593-021-00845-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  101. Vanhatalo S, Kaila K. Development of neonatal EEG activity: from phenomenology to physiology. Seminars in Fetal & Neonatal Medicine. 2006;11:471–478. doi: 10.1016/j.siny.2006.07.008. [DOI] [PubMed] [Google Scholar]
  102. Vinje WE, Gallant JL. Sparse coding and decorrelation in primary visual cortex during natural vision. Science. 2000;287:1273–1276. doi: 10.1126/science.287.5456.1273. [DOI] [PubMed] [Google Scholar]
  103. Voytek B, Kramer MA, Case J, Lepage KQ, Tempesta ZR, Knight RT, Gazzaley A. Age-related changes in 1/f neural electrophysiological noise. The Journal of Neuroscience. 2015;35:13257–13265. doi: 10.1523/JNEUROSCI.2332-14.2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  104. Weir K, Blanquie O, Kilb W, Luhmann HJ, Sinning A. Comparison of spike parameters from optically identified gabaergic and glutamatergic neurons in sparse cortical cultures. Frontiers in Cellular Neuroscience. 2014;8:460. doi: 10.3389/fncel.2014.00460. [DOI] [PMC free article] [PubMed] [Google Scholar]
  105. Wielek T, Del Giudice R, Lang A, Wislowska M, Ott P, Schabus M. On the development of sleep states in the first weeks of life. PLOS ONE. 2019;14:e0224521. doi: 10.1371/journal.pone.0224521. [DOI] [PMC free article] [PubMed] [Google Scholar]
  106. Wong FK, Bercsenyi K, Sreenivasan V, Portalés A, Fernández-Otero M, Marín O. Pyramidal cell regulation of interneuron survival sculpts cortical networks. Nature. 2018;557:668–673. doi: 10.1038/s41586-018-0139-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  107. Wosniack ME, Kirchner JH, Chao LY, Zabouri N, Lohmann C, Gjorgjieva J. Adaptation of spontaneous activity in the developing visual cortex. eLife. 2021;10:e61619. doi: 10.7554/eLife.61619. [DOI] [PMC free article] [PubMed] [Google Scholar]
  108. Xu XX, Chini M, Bitzenhofer SH, Hanganu-Opatz IL. Transient knock-down of prefrontal DISC1 in immune-challenged mice causes abnormal long-range coupling and cognitive dysfunction throughout development. The Journal of Neuroscience. 2019;39:1222–1235. doi: 10.1523/JNEUROSCI.2170-18.2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  109. Xu X, Song L, Kringel R, Hanganu-Opatz IL. Developmental Decrease of Entorhinal Gate Disrupts Prefrontal-Hippocampal Communication in Immune-Challenged DISC1 Knockdown Mice. Research Square. 2021 doi: 10.21203/rs.3.rs-290304/v1. [DOI] [PMC free article] [PubMed]
  110. Yang W, Chini M, Pöpplau JA, Formozov A, Dieter A, Piechocinski P, Rais C, Morellini F, Sporns O, Hanganu-Opatz IL, Wiegert JS. Anesthetics fragment hippocampal network activity, alter spine dynamics, and affect memory consolidation. PLOS Biology. 2021;19:e3001146. doi: 10.1371/journal.pbio.3001146. [DOI] [PMC free article] [PubMed] [Google Scholar]
  111. Yizhar O, Fenno LE, Prigge M, Schneider F, Davidson TJ, O’Shea DJ, Sohal VS, Goshen I, Finkelstein J, Paz JT, Stehfest K, Fudim R, Ramakrishnan C, Huguenard JR, Hegemann P, Deisseroth K. Neocortical excitation/inhibition balance in information processing and social dysfunction. Nature. 2011;477:171–178. doi: 10.1038/nature10360. [DOI] [PMC free article] [PubMed] [Google Scholar]
  112. Yu Y, Migliore M, Hines ML, Shepherd GM. Sparse coding and lateral inhibition arising from balanced and unbalanced dendrodendritic excitation and inhibition. The Journal of Neuroscience. 2014;34:13701–13713. doi: 10.1523/JNEUROSCI.1834-14.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  113. Zhang Z, Jiao YY, Sun QQ. Developmental maturation of excitation and inhibition balance in principal neurons across four layers of somatosensory cortex. Neuroscience. 2011;174:10–25. doi: 10.1016/j.neuroscience.2010.11.045. [DOI] [PMC free article] [PubMed] [Google Scholar]
  114. Zhou S, Yu Y. Synaptic E-I balance underlies efficient neural coding. Frontiers in Neuroscience. 2014;12:46. doi: 10.3389/fnins.2018.00046. [DOI] [PMC free article] [PubMed] [Google Scholar]
  115. Zick JL, Blackman RK, Crowe DA, Amirikian B, DeNicola AL, Netoff TI, Chafee MV. Blocking NMDAR disrupts spike timing and decouples monkey prefrontal circuits: implications for activity-dependent disconnection in schizophrenia. Neuron. 2018;98:1243–1255. doi: 10.1016/j.neuron.2018.05.010. [DOI] [PMC free article] [PubMed] [Google Scholar]

Editor's evaluation

Liset M de la Prida 1

This manuscript presents a combination of in vivo recording and optogenetic experiments that together with modeling bring findings with important significance: inhibition is functionally present in the newborn frontal cortex having major effects on EEG dynamics. These important findings challenge the view on the switch in GABAergic excitation to inhibition and extend phenomenological observations to human infant EEG data. The strength of evidence is solid, with appropriate methodology used and only minor weaknesses noted regarding the human infant data.

Decision letter

Editor: Liset M de la Prida1
Reviewed by: Sampsa Vanhatalo2

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 "Developmental increase of inhibition drives decorrelation of neural activity" 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 review of your submission have agreed to reveal their identity: Sampsa Vanhatalo (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:

The manuscript by Chini et al., investigates emergent dynamics of neural activity during brain development, focusing on differences in the relative strength of excitatory and inhibitory neurotransmission. Using a combination of in vivo recordings and optogenetic experiments with computer modelling, the authors show that inhibition is functionally present in newborn frontal areas. They also describe how this process is dysfunctional in a mouse model of a neurodevelopmental disorder. The work challenges the simplified view of the switch in GABAergic excitation to inhibition. By phenomenologically comparing rodent and human infant EEG data, the manuscript may provide translational bridges with significant impact for clinical studies.

While there was overall consensus on the value of the study, issues arise in particular with the points listed below. We all agree that you will need to address these points specifically to warrant publication. Please, note that the next review round should reach a consensus.

(1) An issue related with EEG human data and in particular regarding how are they generated and analyzed. First, we are unclear about the real N and the type of recordings obtained from these infants. Second, we need to understand the selection of the frequency band (1-40Hz), since most data suggest that neonates have most of their signal power at <1Hz and there is very little contribution for >20Hz. We feel you should adhere to the most commonly used standards. Finally, we are surprised about the increase in slope with early maturation, which is not in agreement with earlier publications. Importantly, addressing these points is critical for your manuscript to advance to the next step. While we will be open to discuss the option of leaving human data out of the revised version, we feel they represent an important addition that increases the impact.

(2) A second concern is with optogenetic data in Figure 4. We have difficulties in understanding the number of units per mice and per group, as they refer to age. We feel the ranges of ages used should provide a consistent number of samples. We are also unclear about the statistical design, whether it is running longitudinally or not. We would like to see these parts clarified and improved and an appropriate statistical contrast for nested design implemented.

(3) We are unclear about the paradoxical effects of optogenetic activations of INs. This point will require clarification and possibly additional analysis.

(4) Finally, regarding the model we are not completely clear about the assumptions made, in particular regarding lognormal distribution of synaptic connection strengths. We feel that testing the effect of other distributions may improve the conclusion and better support the model.

Please, also go over the specific points raised in the individual reviews and address them all in your revised version and rebuttal letter.

Reviewer #1 (Recommendations for the authors):

Overall this is a nice paper which does a good job of exploring an important question using a broad range of different approaches. The focus on PFC and cross-species analysis are particularly novel and important. There are a few points which I feel could be clearer and some issues surrounding data in Figure 4 where a better picture of how the experiments were performed and analysed would be beneficial. Code is made available via GitHub in keeping with open access policy of the journal.

Abstract:

Claim that mechanism behind decorrelation of activity unknown is not strictly accurate, numerous factors have been shown to contribute, including sensory input, synaptic changes and developmental alterations in EGABA

Introduction:

Claim that SOM integrate before PV is not accurate. See, Pangratz-Fuehrer and Hestrin, 2011; Anastasiades et al., 2016; Daw et al., 2007. They make a similar claim in the discussion "Early inhibitory circuits have several peculiarities, including a predominance of inhibitory synapses by SOM+ INs". Again this is not supported by the data. SOM interneurons certainly have an important and unique role in early development. But this is not to say that PV synapses are less numerous or weaker (in fact one of the papers they cite shows that they are much stronger at P12 and have a 20% higher connection probability than SOM cells).

Results:

In the model of the local network do they include changes in GABAergic driving force (i.e reversal potential) or just the conductance? Although they provide evidence that GABA does not appear excitatory during early development, it does not mean that it may not be depolarizing and that EGABA may change across this period. This change could influence their results. Others have shown developmental changes in EGABA within the developing PFC and so this should be taken into account.

Could differences in baseline firing across P4-12 make it harder to detect inhibition at early ages due to a floor effect? Could this contribute in part to their observation?

Were any recordings made at P4 in Figure 4? If not, why not state P5-6 rather than P4-6?

In terms of the mice recorded at different ages. Were mice recorded at all ages in each time window? For example, at P11 there are 2 mice who show a very low modulation index but at P12 all the units seem to be strongly modulated, but there are only 2 data points plotted vs 4 at P11 and 7 at P10. Were only 2 mice recorded at P12, or do the data points overlap at certain ages? Overall, how many mice were recorded at each age?

In the methods for Figure 4 they state that laser power was "adjusted until it gave the desired response" how was this defined? Was there a difference in laser power across the different ages, could this account for differences in inhibition?

Figure S3B How were positive neurons quantified? Is this per slice or per animal?

Discussion:

While the strength of the inhibition exerted by INs increases throughout development, the ability of INs to control cortical inhibition does not qualitatively change with age. Already during the first postnatal week, inhibition of INs leads to a paradoxical increase in their firing rate.

This sentence could be a little clearer.

IN inhibition results in increased spike-train correlations even though, in the last portion of the optogenetic stimulation, IN display a paradoxical increase in firing rate. As could this. Is there a way that you could rephrase? Stimulation typically means to activate cells, whereas you are suppressing them. Even though there is a paradoxical increase in firing this occurs at the network level, this is not due to the direct effect of light and so the term "optogenetic stimulation" is not accurate.

Reviewer #2 (Recommendations for the authors):

1. As mentioned in the weakness, the authors should go into more details about the paradoxical effect. Why is not seen for optogenetic activations of INs, only for the optogenetic inactivations? Also, it would be good to bring in some citations of experimental and theoretical work (Sanzeni et al. eLife, Sadeh et al. J Neuro 2017).

2. The authors should really put their work in the context of other studies who have measured and analyzed spontaneous activity and discussed how it evolves over time. For e.g. the Lohmann lab proposes the existence of L and H events (low and high participation rate events observed in the primary visual cortex), see Siegel et al. 2012. In a modeling study with the Gjorgjieva lab (Wosniack et al. eLife 2021), they proposed a different mechanism that can lead to the desynchronization (or sparsification) of activity during development, where L events increase in frequency while H events increase. This should be at least discussed. In Leighton et al. (Curr Biol 2021) the Lohmann lab also talked about the role of inhibition (from SOMs) in development. Finally, an interesting study that should be discussed is Rahmati et al. (Sci Rep 2017) which also presents results on sparsification of neural activity in development and the connection to inhibition stabilization.

3. Can the authors discuss the use of lognormal weights in the model? What happens if they are constant or taken from a lognormal distribution? I don't doubt they come from a lognormal distribution in the real circuit, but it would be important for the model to point out why this is important, as many other modeling papers ignore this fact.

4. The authors should present a more extensive discussion of why decorrelation is something that the network might strive to archive and how this relates to the onset of sensory experience and the efficient processing of sensory information.

Reviewer #3 (Recommendations for the authors):

My review at this phase will only focus on the few items that I hope would help the authors strengthen the work:

(1) In your introduction, you note that E-I ratio is important in your context because it "is the hallmark of neurodevelopmental disorders, such as autism or schizophrenia". Please note that you are mixing periods in the lifespan: your work is on neonatal brain development, while those disorders are about toddler age (autism) or much much later in life (schizophrenia).

(2) The work is very strong with the case on early inhibition. I find it a bit confusing how the work starts from making a case why the slope of PSD curve (1/f exponent) should be taken as a relevant measure of E-I.

Why not move this part towards the end, just before you introduce the human data? After all, this component appears to have value mainly because it allows you to link your findings on inhibition to the human dataset.

(3) I find it a bit perplexing that you show increase in slope with early maturation. This is opposite to what has been published earlier, and what is the general finding among clinicians. The early EEG (from prematurity to the end of neonatal period) is characterized by a rapid/robust decline in the lowest frequency power ->this translates directly to a decrease of slope.

So there is something unexpected here?

(4) I would also like to understand why you select to analyse 1-40Hz while recent papers have clearly indicated that (i) neonates have most of their signal power <1hz, and (ii) there is very little to be found >20hz.

(5) The human dataset is elusive: You tell that you had N=1100 and N=42 infants (Figure 8 N=1110?`). This would be the by far largest newborn dataset ever published. BUT the papers you cite only have 71+42+40=153 EEG recordings (assuming that they are from different infants). Also, there is no information about the kind of recordings done from these infants.

So, in brief, the information about human data is virtually missing; please elaborate.

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

Thank you for resubmitting your work entitled "An increase of inhibition drives the developmental decorrelation of neural activity" 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:

– Regarding human data: while we were overall positive, we still feel the human data may require some clarification in view of the previous concerns. We appreciated your argument that there is no change of slope caused by contribution at 20-40Hz, but they were based on mouse data only (Figure 1-FS1). If you could provide some sort of additional/control analysis on human data as supplementary material, we feel that could help. We would like to stress this is just advice that we leave to your consideration. Importantly, we feel that it would be useful to discuss your observations on changes in slope with early maturation in the context of earlier publications. Please, be sure you add text to the discussion addressing these and previously raised issues and caveats regarding human results.

– In terms of the optogenetic data in Figure 4, we feel some additional clarification is required specifically regarding the way data is represented (per trial not per mice), and potential issues of low N.

Reviewer #1 (Recommendations for the authors):

The authors seem to have addressed my previous comments on an earlier version of this manuscript. I still think their statements regarding the predominance and importance of SST cells are a little strong, and largely unnecessary given they don't study them in this paper, but I guess it is still a matter of debate within the field.

In terms of the optogenetic data in Figure 4, their explanation makes sense. It does however seem a little strange to plot 2 trials from the same animal as separate data points. Their analysis seems to account for this, but it does mean that the N is a little low for some time points. That said their new analysis accounting for only the top 50% of active units seems to show a very robust effect consistent with their observations and overall model of inhibition's role in the early network.

Reviewer #2 (Recommendations for the authors):

The authors have appropriately addressed all of my, and the other reviewers', comments. There are still a few typos which I'm sure will be fixed in the final version (e.g. line 707 in the version with tracked changes "such AS a transition in synaptic plasticity rules", line 653 the word In's should be INs (all capital)). Overall, this is a very nice paper that people in the field will enjoy reading.

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

Thank you for resubmitting your work entitled "An increase of inhibition drives the developmental decorrelation of neural activity" 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:

As indicated in the previous decision letter, the editorial consultation on the issue regarding human data agreed to request that this would be addressed directly in the discussion. The decision letter stated "Importantly, we feel that it would be useful to discuss your observations on changes in slope with early maturation in the context of earlier publications. Please, be sure you add text to the discussion addressing these and previously raised issues and caveats regarding human results.". Please, consider this point carefully when providing a revised version. We specifically ask for the issues raised by the non-responding reviewer to be explicitly addressed in the manuscript. Please also note that eLife publishes reviews and decision letters together with manuscripts, so we prefer not to leave important issues unaddressed that were previously raised during reviews and consultation.

eLife. 2022 Aug 17;11:e78811. doi: 10.7554/eLife.78811.sa2

Author response


Essential revisions:

The manuscript by Chini et al., investigates emergent dynamics of neural activity during brain development, focusing on differences in the relative strength of excitatory and inhibitory neurotransmission. Using a combination of in vivo recordings and optogenetic experiments with computer modelling, the authors show that inhibition is functionally present in newborn frontal areas. They also describe how this process is dysfunctional in a mouse model of a neurodevelopmental disorder. The work challenges the simplified view of the switch in GABAergic excitation to inhibition. By phenomenologically comparing rodent and human infant EEG data, the manuscript may provide translational bridges with significant impact for clinical studies.

While there was overall consensus on the value of the study, issues arise in particular with the points listed below. We all agree that you will need to address these points specifically to warrant publication. Please, note that the next review round should reach a consensus.

(1) An issue related with EEG human data and in particular regarding how are they generated and analyzed. First, we are unclear about the real N and the type of recordings obtained from these infants. Second, we need to understand the selection of the frequency band (1-40Hz), since most data suggest that neonates have most of their signal power at <1Hz and there is very little contribution for >20Hz. We feel you should adhere to the most commonly used standards. Finally, we are surprised about the increase in slope with early maturation, which is not in agreement with earlier publications. Importantly, addressing these points is critical for your manuscript to advance to the next step. While we will be open to discuss the option of leaving human data out of the revised version, we feel they represent an important addition that increases the impact.

We apologize for the confusion created by the wrong reference that references only part of the open access dataset. We corrected the reference.

We clarified that for the EEG analysis, we already only considered the 1-20 Hz band, whereas 1-40 Hz band was used only for LFP data. We agree with the observation that in the immature brain, most of the PSD power resides at ultra-slow frequencies. However, we do not quantify the PSD power, but rather estimate the slope of the PSD. For this measure, it is less relevant, where most PSD power resides. On the contrary, it has been suggested to choose frequency bands that are “uncorrupted by oscillatory peaks” (Gao et al., 2017). Further, as shown in Figure 1D and Figure 1—figure supplement 1F, there is no change in the slope between 20 and 40 Hz. Thus, the 1/f slope estimation is not biased by the specific frequency range that we chose.

(2) A second concern is with optogenetic data in Figure 4. We have difficulties in understanding the number of units per mice and per group, as they refer to age. We feel the ranges of ages used should provide a consistent number of samples. We are also unclear about the statistical design, whether it is running longitudinally or not. We would like to see these parts clarified and improved and an appropriate statistical contrast for nested design implemented.

In the revised version, we added the required information (see reply to reviewers). We have also provided additional analysis (Figure 4—figure supplement 1D) and a figure (Figure 1) for the reviewers that addresses some of the criticism that was raised with respect to the optogenetic experiments / analyses.

To analyze these experiments, we used a generalized linear mixed-effect model (logit link) with mouse as random effect, and estimated the overall effect that age has on the proportion of units that increase/decrease their firing rate in response to optogenetics. Compared to ANOVAs, generalized linear mixed-effect models estimate the regression coefficients while explicitly taking in consideration the nested nature of the data (i.e. multiple single units per mouse), while allowing more freedom with respect to missing datapoints and the distribution of the data (in this case, since it is not normally distributed as it is bounded between 0 and 1, we used a logit link).

(3) We are unclear about the paradoxical effects of optogenetic activations of INs. This point will require clarification and possibly additional analysis.

We have provided further modeling analysis (Figure 4—figure supplement 1G-J) that strengthens the conclusions of the manuscript with respect to this point. Further, we have expanded the discussion of the paradoxical effect.

(4) Finally, regarding the model we are not completely clear about the assumptions made, in particular regarding lognormal distribution of synaptic connection strengths. We feel that testing the effect of other distributions may improve the conclusion and better support the model.

We have clarified our reasoning behind some of the assumptions that went into the model and have provided a figure for the reviewers that highlights how the effects that we report do not depend on the lognormality of the synaptic weights.

Please, also go over the specific points raised in the individual reviews and address them all in your revised version and rebuttal letter.

Reviewer #1 (Recommendations for the authors):

Overall this is a nice paper which does a good job of exploring an important question using a broad range of different approaches. The focus on PFC and cross-species analysis are particularly novel and important. There are a few points which I feel could be clearer and some issues surrounding data in Figure 4 where a better picture of how the experiments were performed and analysed would be beneficial. Code is made available via GitHub in keeping with open access policy of the journal.

We thank the reviewer for the constructive feedback and most helpful comments and suggestions.

Abstract:

Claim that mechanism behind decorrelation of activity unknown is not strictly accurate, numerous factors have been shown to contribute, including sensory input, synaptic changes and developmental alterations in EGABA

We agree with the reviewer that the sentence was not entirely accurate. We have rephrased it to better reflect the fact that, while a lot of what explain the developmental decorrelation of neural activity still remains to be understood, the underlying mechanisms are not entirely unknown (lines 30-31).

Introduction:

Claim that SOM integrate before PV is not accurate. See, Pangratz-Fuehrer and Hestrin, 2011; Anastasiades et al., 2016; Daw et al., 2007. They make a similar claim in the discussion "Early inhibitory circuits have several peculiarities, including a predominance of inhibitory synapses by SOM+ INs". Again this is not supported by the data. SOM interneurons certainly have an important and unique role in early development. But this is not to say that PV synapses are less numerous or weaker (in fact one of the papers they cite shows that they are much stronger at P12 and have a 20% higher connection probability than SOM cells).

We rephrased the sentences in Introduction and Discussion (lines 79-81, 605-607) to highlight that the statement on the predominance of SOM+ INs holds true for the first postnatal days (the beginning of the developmental phase that we investigated in this study). At this early stage, SOM+ interneurons have much larger synapses and higher connectivity than PV+ INs (Gao et al., 2017). Pangratz-Fuehrer and Hestrin, 2011 reports no (FS) PV-PYR connectivity until P4, and little connectivity also at P5-6. Daw et al., 2007 reports no functional recruitment of (FS) PV+ INs until P6-7.

Results:

In the model of the local network do they include changes in GABAergic driving force (i.e reversal potential) or just the conductance? Although they provide evidence that GABA does not appear excitatory during early development, it does not mean that it may not be depolarizing and that EGABA may change across this period. This change could influence their results. Others have shown developmental changes in EGABA within the developing PFC and so this should be taken into account.

We thank the reviewer for the insightful comment that has spurned us to extend our modeling analysis. As depicted in the novel Figure 4—figure supplement 1G-J, we show that in our model both depolarizing (chloride reversal potential between the resting membrane potential and the action potential threshold) as well as excitatory (chloride reversal potential more positive than the action potential threshold) GABA result in runaway excitation. Further, we also show that even a chloride reversal potential that is equal to or only slightly more negative to the resting membrane potential fails to replicate the population trajectories that we empirically observed in the optogenetic experiments. Only a chloride reversal potential that is 10 mV or more negative than the resting membrane potential fully recapitulates the experimental data.

Could differences in baseline firing across P4-12 make it harder to detect inhibition at early ages due to a floor effect? Could this contribute in part to their observation?

In line with the reviewer’s suggestion, we tested whether a floor effect might have partially explained our results. When we limited the analysis to neurons that were in the top 50% for spikes fired during the optogenetic protocol, the proportion of inhibited units as a function of age was higher than when considered the entire dataset. Therefore, a floor effect as cause of the described results seems unlikely. We added the novel analysis to the manuscript (lines 285-288) and displayed the results in Figure 4—figure supplement 1D.

Were any recordings made at P4 in Figure 4? If not, why not state P5-6 rather than P4-6?

The reviewer is correct, no mouse was recorded at P4. We corrected the figure.

In terms of the mice recorded at different ages. Were mice recorded at all ages in each time window? For example, at P11 there are 2 mice who show a very low modulation index but at P12 all the units seem to be strongly modulated, but there are only 2 data points plotted vs 4 at P11 and 7 at P10. Were only 2 mice recorded at P12, or do the data points overlap at certain ages? Overall, how many mice were recorded at each age?

As specified in Materials and methods (lines 759-771), each mouse has been acutely recorded and only once.

Of the 19 mice that were recorded for the Chr2 experiments: 2 mice were recorded at P5, 3 mice were recorded at P6, 1 mouse was recorded at P7, 3 mice were recorded at P8, 3 mice were recorded at P9, 4 mice were recorded at P10, 2 mice were recorded at P11, 1 mouse was recorded at P12. The individual dots (some were indeed overlapping) do not refer to single mice but rather to individual optogenetic stimulations. The optogenetic protocol was applied twice in each mouse, on the two outmost shanks of the 4-shank electrode (corresponding to superficial and deep layers of the mPFC). We added the missing information to the legend of Figure 4, 5, Figure 4—figure supplement 1, and Materials and methods (lines 781-784).

Of note, the statistical modeling (generalized linear mixed-effect model with mouse as nesting factor) explicitly takes the nesting of the data into account and estimates the parameter of the model (and its significance) on an individual mouse level. We limited the statistical analysis to the estimation of the main effect of age, thereby rending unnecessary recording a large number of mice at every individual postnatal day.

In the methods for Figure 4 they state that laser power was "adjusted until it gave the desired response" how was this defined? Was there a difference in laser power across the different ages, could this account for differences in inhibition?

We rephrased the sentence and specified that the “desired response” was reliable light-induced spiking (lines 778-779). Two pieces of evidence let us assume that it unlikely that different laser power across ages accounts for the differences in inhibition: (i) across the two optogenetic conditions, laser power showed a weak and non-significant (main age effect=-0.23, 95% C.I. [-0.50; 0.04], p=0.10; linear mixed-effect model) (Author response image 1A), trend towards decreased intensity as a function of age. This weak trend, if anything, should work against the increase in inhibition that we report; (ii) laser power and optogenetic effect (% of neurons with increased firing rate) were entirely uncorrelated (main laser power effect=-0.03, 95% C.I. [-0.18; 0.18], p=0.74 for ChR2 experiments; main laser power effect=0.03, 95% C.I. [-0.24; 0.29], p=0.81 for ArchT experiments; generalized linear mixed-effect model) (Figure for reviewer 1B-C) in both optogenetic conditions.

Author response image 1. Laser power across age and its effect on neuronal activity.

Author response image 1.

(A) Scatter plot displaying laser power with respect to age (n=59 mice). (B) Scatter plot displaying the percentage of activated units in ChR2 experiments with respect to laser power (n=19 mice). (C) Same as (B) for ArchT experiments (n=40 mice).

Figure S3B How were positive neurons quantified? Is this per slice or per animal?

We specified that eYFP-expressing neurons were manually quantified on a slide-by-slide basis (lines 785-788) and edited the y-axis label on figure S3.

Discussion:

While the strength of the inhibition exerted by INs increases throughout development, the ability of INs to control cortical inhibition does not qualitatively change with age. Already during the first postnatal week, inhibition of INs leads to a paradoxical increase in their firing rate.

This sentence could be a little clearer.

We rephrased the sentence (lines 621-623).

IN inhibition results in increased spike-train correlations even though, in the last portion of the optogenetic stimulation, IN display a paradoxical increase in firing rate. As could this. Is there a way that you could rephrase? Stimulation typically means to activate cells, whereas you are suppressing them. Even though there is a paradoxical increase in firing this occurs at the network level, this is not due to the direct effect of light and so the term "optogenetic stimulation" is not accurate.

We rephrased the sentence (lines 658-660).

Reviewer #2 (Recommendations for the authors):

1. As mentioned in the weakness, the authors should go into more details about the paradoxical effect. Why is not seen for optogenetic activations of INs, only for the optogenetic inactivations? Also, it would be good to bring in some citations of experimental and theoretical work (Sanzeni et al. eLife, Sadeh et al. J Neuro 2017).

In line with the reviewer’s suggestion, we extended the available discussion of the paradoxical effect in response to IN inhibition referring to the work by Sanzeni and Sadeh (lines 367-373, 633-637).

2. The authors should really put their work in the context of other studies who have measured and analyzed spontaneous activity and discussed how it evolves over time. For e.g. the Lohmann lab proposes the existence of L and H events (low and high participation rate events observed in the primary visual cortex), see Siegel et al. 2012. In a modeling study with the Gjorgjieva lab (Wosniack et al. eLife 2021), they proposed a different mechanism that can lead to the desynchronization (or sparsification) of activity during development, where L events increase in frequency while H events increase. This should be at least discussed. In Leighton et al. (Curr Biol 2021) the Lohmann lab also talked about the role of inhibition (from SOMs) in development. Finally, an interesting study that should be discussed is Rahmati et al. (Sci Rep 2017) which also presents results on sparsification of neural activity in development and the connection to inhibition stabilization.

Already in the previous version of the manuscript, a substantial part of the Introduction and Discussion has been dedicated to put the new results into the context of previous studies done by Rochefort, Lohmann, Colonnese and Khazipov labs. As suggested, we extended the discussion of the mentioned studies, highlighting the developmental decorrelation of neural activity and other potential underlying mechanisms (lines 673-686).

3. Can the authors discuss the use of lognormal weights in the model? What happens if they are constant or taken from a lognormal distribution? I don't doubt they come from a lognormal distribution in the real circuit, but it would be important for the model to point out why this is important, as many other modeling papers ignore this fact.

Abundant literature data have shown that synaptic weights are log-normally distributed in the adult mammal brain (reviewed in Barbour et al., 2007; Buzsáki and Mizuseki, 2014; Scheler, 2017) and, while evidence is not yet conclusive, this log-normality does not seem to arise from activity-dependent processes (Hazan and Ziv 2020). Thus, it is reasonable to assume that synaptic weights are log-normally distributed also in early development. We therefore chose to use log-normally distributed synaptic weights to use biologically plausible parameter choices. We added this explanatory note to the manuscript (lines 173-174).

Further, to verify that the results do not depend on a specific parameter choice, we performed additional modeling simulations with normally distributed synaptic weights. As shown in Author response image 2, the results were similar to those included in the manuscript.

Author response image 2. Figure 2.Increased inhibition leads to an increase in the 1/f exponent and decorrelates spike trains in a neural network model.

Author response image 2.

(A) Schematic representation of the neural network model. (B) Log-log plot displaying the normalized median PSD power in the 30-100 Hz frequency range for varying level of inhibition. Color codes for inhibition strength. (C) Scatter plot displaying the 1/f exponent as a function of net inhibition strength. (D) Scatter plot displaying average STTC as a function of net inhibition strength. For (C) and (D) color codes for inhibition strength with fixed excitation level.

4. The authors should present a more extensive discussion of why decorrelation is something that the network might strive to archive and how this relates to the onset of sensory experience and the efficient processing of sensory information.

We added a new paragraph to the manuscript (lines 673-686).

Reviewer #3 (Recommendations for the authors):

My review at this phase will only focus on the few items that I hope would help the authors strengthen the work:

(1) In your introduction, you note that E-I ratio is important in your context because it "is the hallmark of neurodevelopmental disorders, such as autism or schizophrenia". Please note that you are mixing periods in the lifespan: your work is on neonatal brain development, while those disorders are about toddler age (autism) or much much later in life (schizophrenia).

We previously showed that, while the disease-related symptoms emerge during later development, the disturbance of neural circuits is initiated already at neonatal age (Hartung et al., 2016; Chini et al., 2020). As suggested, we rephrased the sentence to better highlight this aspect (lines 71-72).

(2) The work is very strong with the case on early inhibition. I find it a bit confusing how the work starts from making a case why the slope of PSD curve (1/f exponent) should be taken as a relevant measure of E-I.

Why not move this part towards the end, just before you introduce the human data? After all, this component appears to have value mainly because it allows you to link your findings on inhibition to the human dataset.

We agree with the reviewer that the optogenetic experiments (Figure 3 and 4) strengthen the hypothesis that E-I ratio shifts towards inhibition during development that was proposed in line with the 1/f results presented in Figure 1. However, taken in isolation, we do not think that they are sufficient to make a strong case for the developmental E-I ratio shift. The results in Figure 3-4 suggest that inhibition increases throughout development. This is not the same as a shift in E-I ratio (relative strengthening of inhibition with respect to excitation).

(3) I find it a bit perplexing that you show increase in slope with early maturation. This is opposite to what has been published earlier, and what is the general finding among clinicians. The early EEG (from prematurity to the end of neonatal period) is characterized by a rapid/robust decline in the lowest frequency power ->this translates directly to a decrease of slope.

So there is something unexpected here?

To the best of our knowledge, there is only one other paper (Schaworonkow and Voytek, 2021) that has been published on the development of the 1/f exponent at such early stages. Of note, even this paper investigates a later developmental phase (1 to 7 months) when compared to the time window of 1-month prematurity to 1-month of age investigated in the present study. The Discussion includes several hypotheses as to why the two studies might diverge (lines 694-698).

Further, the fact that the slope becomes steeper over this period does not contradict the observation that the early EEG is characterized by a robust decline at the lowest frequency range. Our data solely indicate that the steepness of this decline increases over the two months that are covered by the two datasets. The entire range (1-month prematurity to 1-month of age) can be categorized as (very) early EEG.

(4) I would also like to understand why you select to analyse 1-40Hz while recent papers have clearly indicated that (i) neonates have most of their signal power <1hz, and (ii) there is very little to be found >20hz.

We agree with the observation that in the immature brain, most of the PSD power resides at ultra-slow frequencies. Therefore, we do not quantify the PSD power, but rather estimate the slope of the PSD. For this measure, it is not important to include frequencies at which most power resides. On the contrary, it has been suggested to choose frequency bands that are “uncorrupted by oscillatory peaks” (Gao et al., 2017). As shown in Figure 1D and Figure 1—figure supplement 1F, there is no change in the slope between 20 and 40 Hz. Thus, the 1/f slope estimation is not biased by the specific frequency range that we chose.

Further, we extended the frequency range up to 40 Hz to compute the 1/f exponent only on mouse data (for newborn babies we the cutoff was 20 Hz, see Materials and methods) and, how it is observable from Figure 1D and Figure 1—figure supplement 1F, there is no discontinuity in the rate with which power decays between 20 and 40 Hz. Thus, the 1/f slope estimation is not affected by the specific frequency range that we chose.

(5) The human dataset is elusive: You tell that you had N=1100 and N=42 infants (Figure 8 N=1110?`). This would be the by far largest newborn dataset ever published. BUT the papers you cite only have 71+42+40=153 EEG recordings (assuming that they are from different infants). Also, there is no information about the kind of recordings done from these infants.

So, in brief, the information about human data is virtually missing; please elaborate.

We thank the reviewer for spotting the inconsistency with respect to the reference for the first of the two human datasets that we analyze in the manuscript. We apologize for the confusion created by the wrong reference that references only part of the open access dataset. While the two references that were previously given for the dataset are the two that are available on figshare (where the repository is located), they indeed reference only part of the dataset. We have updated the reference by providing one that pertains to a paper describing the entirety of the dataset (1100 recordings) (Schetinin and Jakaite, 2017).

[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:

– Regarding human data: while we were overall positive, we still feel the human data may require some clarification in view of the previous concerns. We appreciated your argument that there is no change of slope caused by contribution at 20-40Hz, but they were based on mouse data only (Figure 1-FS1). If you could provide some sort of additional/control analysis on human data as supplementary material, we feel that could help. We would like to stress this is just advice that we leave to your consideration. Importantly, we feel that it would be useful to discuss your observations on changes in slope with early maturation in the context of earlier publications. Please, be sure you add text to the discussion addressing these and previously raised issues and caveats regarding human results.

Our answer was focused on the mouse data because only on mouse data we used the 1-40 Hz frequency range to estimate the 1/f slope. For the human data, we used the 1-20 Hz frequency range already from the first iteration of the manuscript. To avoid any ambiguity, we have now explicitly stated this on lines 579-580.

– In terms of the optogenetic data in Figure 4, we feel some additional clarification is required specifically regarding the way data is represented (per trial not per mice), and potential issues of low N.

We modified the text (lines 278-279, 287-288, 292-293, 369-370) to explicitly mention that individual dots correspond to optogenetic stimulation protocols on Figure 4C, Figure 5C and Figure 4—figure supplement 1B-D.

Reviewer #1 (Recommendations for the authors):

The authors seem to have addressed my previous comments on an earlier version of this manuscript. I still think their statements regarding the predominance and importance of SST cells are a little strong, and largely unnecessary given they don't study them in this paper, but I guess it is still a matter of debate within the field.

We thank the reviewer for the most helpful feedback and support.

In the manuscript we mention twice the SST+ interneurons (in Introduction and Discussion), referring to published data from other labs. While we agree with the reviewer that our investigation does not address this cell type, we consider important to mention the literature findings that highlight the differences between developing and adult circuits.

In terms of the optogenetic data in Figure 4, their explanation makes sense. It does however seem a little strange to plot 2 trials from the same animal as separate data points. Their analysis seems to account for this, but it does mean that the N is a little low for some time points. That said their new analysis accounting for only the top 50% of active units seems to show a very robust effect consistent with their observations and overall model of inhibition's role in the early network.

To avoid any ambiguity, we modified the main text (lines 278-279, 287-288, 292-293, 369-370) to clearly specify that individual dots correspond to optogenetic stimulation protocols on Figure 4C, Figure 5C and Figure 4—figure supplement 1B-D.

Reviewer #2 (Recommendations for the authors):

The authors have appropriately addressed all of my, and the other reviewers', comments. There are still a few typos which I'm sure will be fixed in the final version (e.g. line 707 in the version with tracked changes "such AS a transition in synaptic plasticity rules", line 653 the word In's should be INs (all capital)). Overall, this is a very nice paper that people in the field will enjoy reading.

We thank the reviewer for the most helpful feedback and support. We have corrected the two typos.

[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:

As indicated in the previous decision letter, the editorial consultation on the issue regarding human data agreed to request that this would be addressed directly in the discussion. The decision letter stated "Importantly, we feel that it would be useful to discuss your observations on changes in slope with early maturation in the context of earlier publications. Please, be sure you add text to the discussion addressing these and previously raised issues and caveats regarding human results.". Please, consider this point carefully when providing a revised version. We specifically ask for the issues raised by the non-responding reviewer to be explicitly addressed in the manuscript. Please also note that eLife publishes reviews and decision letters together with manuscripts, so we prefer not to leave important issues unaddressed that were previously raised during reviews and consultation.

As recommended, we added to the manuscript a discussion of our results in the context of previous human studies on early patterns of EEG activity. To the best of our knowledge, there are no previous publications directly investigating developmental changes of the 1/f exponent, beyond the paper by Schaworonkow and Voytek (2021) that we already discuss. Correspondingly, the concern of Reviewer #3 (first revision) refers to the fact that the early EEG is characterized by a robust decline at the lowest frequency range. Our findings do not contradict this observation, but solely point in the direction that the steepness of this decline increases over the two months that are covered by the two datasets that we investigated.

With this, we deem to have addressed all the concerns that have been previously raised during the various rounds of review. As you suggested, we carefully considered the issues raised by the non-responding reviewer and addressed them in the manuscript

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    Supplementary file 1. Detailed statistical results.
    elife-78811-supp1.xlsx (36.2KB, xlsx)
    MDAR checklist

    Data Availability Statement

    LFP and SUA data that were newly generated for this study are available at the following open-access repository: https://gin.g-node.org/mchini/development_EI_decorrelation.

    Code supporting the findings of this study is available at the following open-access repository: https://github.com/mchini/Chini_et_al_EI_decorrelation; Chini, 2021.

    LFP and SUA data that were newly generated for this study are available at the following open-access repository: https://gin.g-node.org/mchini/development_EI_decorrelation. Code supporting the findings of this study is available at the following open-access repository: https://github.com/mchini/Chini_et_al_EI_decorrelation, (copy archived at swh:1:rev:9a07c56f36c80a60a44a6607a5a4061a37d96ae7).


    Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd

    RESOURCES