Enriched experience increases reciprocal synaptic connectivity and coding sparsity in higher-order cortex - PMC

Experimental Paradigm:

We ran n=18 (2-month-old, 10M/8F) pair-housed mice on either an enrichment track (ET; n=5M/4F) or an exercise control track (CT; n=5M/4F) for 10–12 weeks (5 × 1-hour sessions per week) (Fig. 1A-C). ET mice were exposed to multiple obstacles that were changed daily to allow a rich repertoire of visual, somatosensory, and motor experiences, while CT mice ran on a track with simple ramp hurdles (20 cm × 11 cm per hurdle) (Fig. S1A, B). CT mice ran more laps than ET mice (F=5.938, p=0.033, group-effect, repeated-measures ANOVA (rmANOVA)) (Fig. 1C). A separate group of n=15 mice (7ET/8CT, age-matched to main group) underwent the same ENR protocol and confirmed that ET group showed enhanced spatial recognition in a Y-maze task (t=−4.2, p=0.0013, unpaired t-test; Fig. S1C-F), replicating previously shown behavioral benefits of ENR using this paradigm (10).

After ENR, mice (~5.5 months old) were head-barred and injected bilaterally with pAAV2-hSyn-DIO-hM3D(Gq)-mCherry virus in the parafacial zone (PZ), a SWS promoting region, to induce essentially natural SWS at will using Clozapine-N-oxide (CNO) (18) (Fig. 1D). This allowed stable data collection for long-duration offline periods, which is required for reliable estimation of connections between excitatory neurons (19). After ~2 months of head-fixation habituation, craniotomies were performed over granular and agranular retrosplenial cortex (RSCg, RSCag), primary visual cortex (V1), dentate gyrus (DG), and dorsal and ventral CA1 (dCA1, vCA1), and a catheter was implanted in the upper back for subcutaneous CNO injection (Fig. 1E-F).

Acute electrophysiology recordings were conducted the day after craniotomy surgery using two dual-shank 256-channel silicon probes (512 total) (20), targeting RSC-dCA1 (probe 1) and V1-vCA1 (probe 2). Recordings included 2.5–3 hours of awake rest, followed by at least 13 hours of CNO-induced slow-wave sleep (SWS), with CNO boosters administered subcutaneously (0.5 mg/kg), every 2.75±0.12 hours. Average recording duration was 2.85±0.06 hours (awake rest) and 17.67±0.44 hours (SWS). Brains were then sectioned to verify recording locations (Fig. 1F; Fig. S2A; Table S1) and virus expression, which did not differ across groups (t=0.276, p>0.05, unpaired t-test; Fig. 1D; Fig. S2B). Recordings were randomly interleaved between groups as the experimenters were blind to the group identity.

Spike sorting (21) yielded single units (CT: 176.86±11.82, 124.0±38.28; ET: 206.0±20.58, 118.72±36.85 for probe1 and probe2, respectively; Fig. 2A, Table S1). Putative excitatory (mean firing rate: 2.523±0.049 Hz) and fast-spiking (FS) interneurons (mean firing rate: 10.038±0.359 Hz) were classified based on the waveform and auto-correlogram features (Fig. S2D). Four animals were excluded due to low cell yield or surgical issues, resulting in a final sample of n=14 animals (CT: 7 (4M/3F); ET: 7 (4M/3F)) for the RSC probe and n=10 animals (CT: 5 (3M/2F); ET: 5 (3M/2F)) for the V1 probe.

Brain state (wake, SWS/NREM, REM) was scored using running speed, theta (5–10 Hz), and delta (1–4 Hz) oscillation power (Fig. 2B-C) (22). ET group showed slightly reduced delta power in the last 2–3 hours (F=1.741, p=0.023, group*time effect, rmANOVA) (Fig. S3B). Therefore, we analyzed only the first 15 hours (~12 hours SWS). No group differences were found in total awake: CT: 2.56±0.12, ET: 2.6±0.11 hours (t=−0.185, p=0.85, unpaired t-test), SWS duration: CT: 9.9±0.22, ET: 10.1±0.2 hours (t=0.688, p=0.504; Table S1), SWS proportion (F=0.007, p=0.935; Fig. S3C), SWR rate (F=0.045, p=0.835, group-effect rmANOVA; Fig. S3C1), and other SWR-related properties (p>0.05, unpaired t-test with Bonferroni correction; Fig. S3C2–4).

Changes in functional monosynaptic connectivity:

We first analyzed functional synaptic connectivity in NC to test whether enriched experiences remodel NC circuitry to support encoding of large number of experiences. We computed cross-correlogram (CCG) for all simultaneously recorded neuron pairs and identified short-latency peaks or troughs (0.8–4.8 ms) to identify putative monosynaptic excitatory or inhibitory connections (19, 23) (Fig. 2D). While we refer to these connections as functional, their validity is supported by juxtacellular recordings, which have demonstrated that such CCG peaks and troughs reflect true monosynaptic connections (24). Each significant connection was further classified as unidirectional excitatory-excitatory (EE1), bidirectional excitatory-excitatory (EE2), excitatory-inhibitory (EI), and inhibitory-excitatory (IE), based on short-latency peak/trough and cell classification (Fig. 2E). Common input connections with zero-lag synchrony, typically observed between interneuron pairs (II), were excluded (24).

We analyzed a total of 7,611.43±955.98 excitatory (EE1+EE2) and 3,818.43±525.23 inhibitory (EI+IE) cell pairs in RSC (n=7CT/7ET), and 3,376.5±710.25 excitatory and 1,880.5±344.81 inhibitory pairs in V1 (n=5CT/5ET). The total number of excitatory-to-excitatory (EE1+EE2) connections were comparable between groups in both RSC and V1. However, ET mice showed a significant increase in EE2 connections (t=−3.563, p=0.009; unpaired t-test; Table S2) and IE connections (t=−2.933, p=0.0158; Table S2) specifically in RSC, but not in V1 (Fig. 3A,B). This increase was accompanied by a significant decrease in EE1 connections in RSC (t=3.361, p=0.006; Table S2). No difference was observed in connection probability for EI connections.

Next, we assessed synaptic connection strength using: 1) sum of observed CCG values exceeding (or below) chance, normalized by the reference cell’s spike count, 2) z-score peak (or trough) relative to the chance-level CCG. For EE2 connections, strength was averaged across both sides of the CCG. No significant group differences in connection strength were observed for any connection type in either region (Fig. 3C,D; Table S2; Fig. S4A). Consistent with previous studies, connection probability declined with distance (Fig. S4B). EI connections were generally stronger than EE1 and EE2 connections (Fig. S4B), and all connection types showed lognormal strength distributions (Fig. S4C) (23, 25, 26). Unlike a previous study (23), we did not observe stronger EE2 connections relative to EE1 (Table S2). No sex differences were observed (p>0.05, ANOVA with Bonferroni correction, Fig. S7A).

The expected number of unconnected, EE1, and EE2 pairs was estimated using the following equations: N(1-p)², 2Np(1-p), and Np², respectively, where N is the total EE pairs and p is the overall EE connection probability (19, 23). Using p=0.1 (p_EE1 + p_EE2; Fig. 3A,B) and the observed N, we found that the number of EE1 connections were lower than expected, whereas EE2 were significantly higher (~3x) than expected (p<0.05). Similar increase in EE2 connections have been reported previously (19, 27). This overrepresentation of bidirectional EE2 (symmetric) connections is consistent with attractor dynamics and may reflect circuit-level changes supporting the encoding of multiple items experienced during ENR (5).

In V1, we analyzed intralaminar deep (L5/6) connections in Fig. 3B,D, as insufficient number of superficial (L2/3) neurons were available across animals. In contrast, RSC data included both L2/3 and L5/6 neurons from granular (RSCg) and agranular (RSCag) subdivisions. Therefore, we decided to analyze connections within and between L2/3 and L5/6of RSCg and RSCag. Interareal connections were rare, but we detected some interlaminar connections within the same brain region for EE1 (probability in RSCg: CT: 0.011±0.003, ET: 0.012±0.004; RSCag: CT: 0.025±0.004, ET: 0.022±0.005), EI (RSCg: CT: 0.015±0.004, ET: 0.016±0.006; RSCag: CT: 0.020±0.005, ET: 0.020±0.007), and IE (RSCg: CT: 0.019±0.004, ET: 0.016±0.004; RSCag: CT: 0.008±0.004, ET: 0.010±0.005) connection types, but not for EE2, with no significant group differences in their probability or strength. Thus, we focused on intralaminar connections within each RSC subregion.

The total number of intralaminar EE pairs were: L2/3 RSCg: 91.08±17.89; L5/6 RSCg: 1194.0±211.01; L2/3 RSCag: 82.64±10.56; L5/6 RSCag: 1417.21±176.36 and EI/IE pairs were: L2/3 RSCg: 71.63±11.5; L5/6 RSCg: 620.93±118.3; L2/3 RSCag: 70.69±13.44; L5/6 RSCag: 454.21±70.48. In ET mice, EE2 connection probability was significantly higher in deep (L5/6) layers of both RSCg (t=−3.149, p=0.013) and RSCag (t=−3.177, p=0.008) compared to CT mice (Fig. 3E). In contrast, the EE1 connection probability was significantly lower in ET mice in L5/6 of both RSCg (t=2.36, p=0.037) and RSCag (t=4.259, p=0.003), consistent with Fig. 3A results. EI connection rates were unchanged across groups, but IE connection probability was significantly increased in ET mice in L5/6 of both RSCg (t=−4.337, p=0.0013) and RSCag (t=−3.894, p=0.002) (Fig. 3E). Across animals, L2/3 connection probabilities were often zero, suggesting that the results in Fig. 3A are largely driven by L5/6 neurons. Finally, no group differences were observed in intralaminar synaptic strength across connection types in either RSCg or RSCag (Fig. S4D). Subsampling L5/6 neurons to match the number of L2/3 neurons yielded similar results, ruling out sampling bias.

We computed CCG over the entire 15-hour recording to obtain a reliable estimate of connection probability (19). However, awake rest and SWS differ in several aspects, including activity levels, neuromodulatory tone, memory replay dynamics, etc. (28–31). To assess state-dependent changes, we recalculated connection strength for connections identified in the full recording, separately for awake rest (~2.5–3 hours) and the first 3 hours of SWS. Previous studies reported that intralaminar L5/6 EE connection strength increases from awake rest to SWS (25). We also observed the largest differences in intralaminar L5/6 EE and IE connection probabilities (Fig. 3A,B). We therefore focused our analysis on intralaminar L5/6 connections to determine whether similar changes in connection strength occur. No significant group differences were found; although EI, IE, and EE2 connections showed a slight upward trend from awake rest to SWS in both groups (Fig. S4E), while EE1 remained stable. Together, these results highlight selective reorganization of deep-layer RSC connectivity following EE, characterized by a shift toward more reciprocal (bidirectional) excitatory connections and enhanced inhibitory feedback, consistent with attractor-like dynamics.

Changes in population coding dynamics:

Having established the changes in synaptic connectivity, we next asked whether ET mice would show sparser and more orthogonal population activity. We computed population and lifetime sparseness using 50 ms binned population vectors (PV) of excitatory units during awake rest and SWS. Lifetime sparseness reflects how selectively a single neuron fires over time, while population sparseness quantifies the proportion of active neurons in a population at any given time (32, 33). Neuron counts were balanced across animals and brain regions by sub-sampling 1000 times and averaging. We compared units from CA1, L2/3 RSC, and L5/6 RSC separately due to their distinct coding properties (34–36). For V1, we focused on L5/6 neurons due to insufficient L2/3 neurons.

During both awake rest and SWS, ET mice showed significantly increased population and lifetime sparseness in L5/6 RSCg (lifetime sparsity: Awk Rest: t=−3.147, p=0.012; SWS: t=−6.628, p<0.001; population sparsity: Awk Rest: t=−3.23, p=0.008; SWS: t=−4.897, p<0.001) and L5/6 RSCag (lifetime sparsity: Awk Rest: t=−3.812, p=0.004; SWS: t=−4.8, p<0.001; population sparsity: Awk Rest: t=−2.36, p=0.038; SWS: t=−3.15, p=0.009) (Fig. 4A,B; Table S3). A modest but significant increase in population sparseness was also observed in L2/3 RSCag in ET mice during SWS (t=−2.445, p=0.043; unpaired t-test). No group differences in sparseness were found in other regions (Fig. 4B, Table S3). Dividing the 12-hour SWS into 3-hour blocks showed comparable results, with ET mice exhibiting increased sparsity in RSC L5/6 neurons (Fig. S5A,B). As no systematic differences were observed across SWS blocks, we report data pooled across the entire SWS period (Fig. 4).

We then looked at orthogonality across brain regions using average cosine distance between pairs of PVs. ET mice exhibited significantly larger distances in L5/6 RSCg (Awk Rest: t=−2.816, p=0.017; SWS: t=−3.15, p=0.009) and L5/6 RSCag neurons (Awk Rest: t=−2.725, p=0.023; SWS: t=−3.464, p=0.005) (Fig. 4C, S5C; Table S3), indicating a more diverse neural state space. Other regions did not show significant group differences. Increase in sparsity and orthogonality were generally more pronounced during SWS, potentially reflecting differences in activity levels and memory replay dynamics between the two states (28–31). Perhaps a more diverse activity space is being explored during SWS than awake rest, which might be more prominent in ET mice. Differences in sparsity and orthogonality could not be fully explained by differences in mean firing rate or bursting index (proportion of spikes with inter-spike interval < 6 ms), as no significant group effects were detected for either measure (Fig. 4E,F, Fig. S5, Table S3). That said, a trend toward more bursting and lower firing rates in ET mice was visible. Consistent with previous studies, mean firing rates were generally higher in NC L5/6 compared to L2/3 (34–36) and lower during SWS than awake rest (37, 38). No significant sex differences were found (p>0.05, ANOVA with Bonferroni correction, Fig. S7b).

Changes in population and lifetime sparsity were primarily observed in L5/6 RSC. The lack of significant changes in CA1 and L2/3 RSC may be attributed to these regions already operating at near-maximal (or near-optimal) sparsity levels, whereas L5/6 neurons may have greater potential for improving coding efficiency (34, 35). No significant difference in lifetime sparsity was observed in DG neurons after pooling data across animals (Fig. S6A). This finding, combined with the lack of difference in CA1 sparsity, contrasts with previous studies examining hippocampal subregions after ENR (39–41). As CT mice ran more laps (Fig. 1C), more physical exercise may have promoted neurogenesis and hippocampal sparsity, potentially reducing differences between the groups (42, 43). L2/3 RSC likely follows a similar pattern, as it receives direct inputs from dCA1 (44, 45). However, a more likely explanation is that L5/6 neurons have been shown to code for categorical knowledge (46, 47). Consequently, larger effects are expected in L5/6 of ET mice to accommodate a greater diversity of sensory-motor experiences. Selective increase in L5/6 RSC, but not in L5/6 V1, may result from V1 being an early sensory area where feature coding is relatively less experience-dependent. Alternatively, RSC integrates multimodal inputs from multiple areas beyond V1, such as somatosensory and motor, which may exhibit more pronounced changes and drive the observed results (48). Supporting this, a recent study using the same ENR paradigm demonstrated rapid stabilization of spatially selective firing in secondary motor cortex, a primary input region to RSC (11). Further exploration of PV correlations using graph-based analysis trends towards lower clustering coefficient (not significant) for L5/6 units in ET mice, indicative of weaker coactivation, in alignment with higher orthogonality observed (Fig. S6C).