Adult female Wistar rats (>P60) were anesthetized with urethane (1.2 g/kg, i.p.) and placed in a stereotaxic device, and the body temperature was maintained at 37°C with a heating pad and feedback control. Glucose-saline was supplemented (10 ml/kg) every hour to maintain the animal hydrated, and experiments never lasted >4 h in total. The long-lasting anesthetic bupivacaine (0.75%) was applied at surgical wounds. In different experiments, concentric stimulating electrodes were placed in the soma layer of the CA3b region of the left hemisphere (AP 2.9; L ± 2.6; V 3.4 mm from Bregma and cortical surface) to activate the ipsilateral Schaffer input to CA1 and the commissural input to the contralateral CA1; in the angular bundle (AP 2.9; L ± 2.6; V 3.4) or the MEC (AP 2.9; L ± 2.6; V 3.4) to activate the medial perforant path (MPP); and in the lateral olfactory tract (LOT) (AP 2.9; L ± 2.6; V 3.4) to activate the lateral perforant path (LPP). Up to 64 simultaneous recordings were obtained with two linear silicon probes (32 sites, 65 μm intersite distance) from Atlas Neuroengineering (Leuven, Belgium). The probes were stereotaxically located at homotopic sites of the dorsal hippocampus across the CA1 region and that also spanned the DG/CA3 (AP 4-4.5; L ± 2.6 mm) (Figure 1A). Typically, the recording arrays also spanned the overlying V2 cortex and sections of the underlying thalamus. Mineral oil was used to cover the site of implantation after electrode penetration. Probes were soaked in DiI before insertion (Molecular Probes, Invitrogen, Carlsbad, CA) to assess their location post-mortem in histological sections. The built-in reference site was not used to avoid influence from nearby sources that severely distort the time course and spatial landmarks of local FPs. Instead, a silver chloride wire implanted under the skin of the neck served as a reference for recordings (note that volume-conducted contributions are readily separated by the ICA; see next section; Torres et al., 2019). Signals were amplified and acquired using MultiChannel System (Reutlingen, Germany), or Open Ephys hardware and software at a 50 kHz sampling rate. We used histological and electrophysiological criteria to identify cortical and hippocampal strata, such as the maximum of the population spike in evoked potentials, and the spontaneous cell firing to determine the position of cell layers in the CA1 and DG. At the end of the recording session the animals were sacrificed by anesthetic overdose, and their brain was removed and maintained in 4% paraformaldehyde in saline. Sagittal brain sections (100 μm) were stained with bis-benzimide and the electrode positions assessed by fluorescence microscopy.

Wide-band FPs (0.1 Hz–5 kHz) were recorded in 3 min periods separated by 15 min, and additional high-pass digital filter was set at 0.5 Hz to remove slow transient artifacts. Signals were down-sampled at 4 kHz to speed up analysis. Occasionally, the activity in a single recording site with transient artifacts was interpolated from the surrounding electrodes. We note that site interpolation in faulty electrodes does not affect the time-course of FP-generators extracted from a collection of recordings by the ICA (Herreras et al., 2015). Faulty recordings in outer sites were however rejected.

Coarse synchronization between different recording sites and time courses of FP-generators was estimated using the cross-correlation coefficient (CC) and spectral coherence. The CC was obtained as:

where (C_ij) is the covariance matrix of two random variables. Spectral coherence was calculated by:

where (P_ij(f)) is the matrix of cross-power spectral density. To determine the level of significance we used the surrogate data test. Randomizing phase relations and keeping other first order characteristics intact, we obtained surrogate time series from the original signals. For each experiment we generated 1000 surrogates and we evaluated pairwise spectral coherences. The level of significance (at α = 0.05) was then calculated for each frequency value and coherence above this level was considered statistically significant.

We used two additional tests aiming at different features. Firstly, we use the Pearson correlation coefficient, a statistical measure used to quantify the linear correlation between two variables:

which, applied to a sample {(x_i,y_i) : i = 1, …, n}, results in the following formula:

The Pearson coefficient, when applied to two signals, can be used to quantify their degree of synchrony. This measure fluctuates in the interval [−1, 1], where absolute values close to 1 indicate a good correlation while values near 0 mean that the variables have no correlation at all. In our particular case we intend to study the variability in the coherence between right and left pairs of FP generators containing highly irregular fluctuations with patterns varying in a scale of seconds. It can thus be anticipated that the coefficient will vary along the time. Hence, we chopped the signals into equidistant time windows, and we repeated the analysis for three different values, Δt = 1, 0.1, and 0.01 s. The longer time windows will focus on the envelope of the signal in the low frequency range, whereas shorter ones will emphasize the correlation of faster fluctuations in the high frequency band. Choosing a single time window is troublesome as there are competing influences when signals as these are structured as evolving mixtures of long and short waves. Therefore, to help their interpretation we distributed the Pearson values into density histogram for each pair of signals compared and time windows, in order to better visualize if these correlation coefficients assemble around certain values of interest. The shorter the time window the more symmetrical and flat become the density histograms. Note these coefficients do not directly reflect frequency components or waves of a specific duration, albeit their presence in the signal has an influence.

Then, we used Granger-causality (Granger-c), a statistical test that is widely used to infer directional connectivity among brain areas (Chen et al., 2006). It assumes that, if there is a link from region X to region Y, then the past activity of X may predict the dynamics of Y. It is based on autoregressive models, where the time-course of the region Y is modeled by its own past and a residual (Bressler and Seth, 2011):

Briefly, if there is a link from X to Y, the inclusion of the previous values of X(t) in the model of Y(t) would reduce the error of the prediction. Thus, the Granger-c from X(t) to Y(t) is determined by the F-statistics:

where Σ_YY = cov(ε_Y) and ΣY⁢Y′=c⁢o⁢v⁢(εY′) are the residual covariance matrices of the model in (7) and the model including the past values of X(t), respectively.

We employed the MVGC toolbox for Matlab to compute both temporal and spectral Granger-c (Barnett and Seth, 2014). As input time series, we used the activation signals of the right and left FP-generators (Dhamala et al., 2008). We downsampled the signals to 250 Hz and calculated Granger-c in a pairwise manner, using a sliding window of 5 s, with an overlap of 80% and a model order p = 12. The Granger-c associated to each link was the average value across time-windows.

To estimate the statistical significance associated to the Granger-c, we used a surrogate data analysis (N = 1000 in this work) by block-resampling. Each signal was cut at a single random time point and the blocks were permuted. By breaking the temporal relationship between signals, any connectivity between them would be by chance. For each surrogate, we kept the Granger-c values for the different links and approximated them to a normal distribution. We set the significance threshold at the value where the previous cumulative distribution was 0.95 (p = 0.05).

In order to compare the features of individual gamma waves and evaluate their synchronization and relative amplitude at left and right sites, accurate determination of the waveforms is required (start time, amplitude and duration). Here we used a formerly implemented method using deconvolution of FPs. The method employs the modeling of the time course of a FP generator, s(t), as a weighted sum of K single FP events of the pulse-like form:

where w_k, τ_k, δ_k are the relative weight, starting time, and time scale of the k-th event, respectively, and H(t) is the Heaviside step function. We aim at estimating the parameter set (w, τ, δ) from the observation of s(t). To accomplish this task, we use the method of maximization of the loglikelihood (for details see Benito et al., 2016). Events were considered paired when they overlap for at least 70 % of the duration. This procedure was also essayed on generators other than the Schaffer, but the results were unsatisfactory. This was partly due to fluctuating baselines and partly due to the imprecision of separated generators from FPs where multiple gamma oscillations are extensively overlapped in space (Makarova et al., 2011).

Layer-specific FP motifs and oscillations spanning several contiguous electrodes in linear arrays are observed in all hippocampal subfields with multiple frequencies and durations (Figure 1C). These two spatial and temporal features reflected, respectively, the stratified input of various synaptic pathways to major populations of pyramidal or GCs, and the wide variety of dynamics. According to previous research (Herreras et al., 2022, 2023), such variety of waveforms derived in part from the spatiotemporal mixtures of potentials contributed by various co-active sources that contaminate each other and distort the original time courses. Temporal distortion arose either from the overlapping of the spatial domains hosting different FP waves/motifs (Figure 1C, purple ovals) or due to their contamination by others generated at remote sites (waves outlined in brown). Therefore, an ICA was employed to retrieve clean time courses of activities in separated sources in each hemisphere (Figures 1D, E). This approach exhibits a notable improvement over the conventional CSD analysis (Figure 1F), which removes remote contributions but does not separate waves generated by overlapping pathways (Martín-Vázquez et al., 2013). Thus, single-site CSD plots (Figure 1G, purple traces) still showed distorted waves when sources and sinks from different pathways cohabited the same strata. As described in Methods, only a few FP generators contributed significantly to the total variance of the epoch and were highly stable and reproducible across animals. The six main FP generators explored were the CA3→CA1 Schaffer input (Figure 1D, cyan), input to st. lacunosum-moleculare (st. L-M; green), MEC→DG input (red), the LEC→DG input (blue), input to GC soma layer (GCsom; magenta), and a remote FP generator (brown). The identification of these components was accomplished through the recognition of their characteristic voltage profiles (V_prof) in conjunction with functional assessments, as reported in previous studies using selective stimulation of principal extrinsic and associational pathways (see below).

Left-right asymmetry was visually observed in raw FPs in certain spatial domains during epochs of irregular activity, which were the focus of our study. These epochs were characterized by fast activity of moderate amplitude (0.1–1 mV) which included interspersed slow waves in the st. L-M and occasional sharp waves (SPWs) within the st. radiatum of the CA1 area, commonly referred to as non-theta state. Given the known association between irregular FP states and diverse behaviors (Buzsáki et al., 1983), we conducted a preliminary evaluation to determine the relevance of sub-states with or without visually recognizable electrographic features (e.g., Figure 2A, rats #1 and #2), for bilateral comparisons. Specifically, we compared 10-second epochs (three per animal) at left and right homotypic sites in two demarcations: the st. L-M and the hilus. We calculated the mean linear correlation (Pearson coefficient) across the sample, yielding r = 0.55 ± 0.01 and r = 0.25 ± 0.01, respectively (mean ± s.e.m; n = 6 × 3; Δt = 0.1 s). Although there were different r values at different recording sites, there was very low variability across animals, regardless of the general pattern. Hence, no additional classification into raw FP electrographic sub-states was attempted.

However, due to the high short-term variability observed in the linear correlation (Figure 2A) it was deemed necessary to establish a minimum epoch duration during which the overall CC remained stable for all FP generators. To determine this, we gradually increased the epoch length until the CC between left and right homonymous FP generators did not deviate by more than 3% of the mean value observed in the shorter epochs (Figure 2B). Using this criterion, an epoch duration of 50 seconds was determined to be optimal under the present recording conditions. Thus, we analyzed three non-consecutive epochs per animal within a 60-minute period, and these epochs were bound together for subsequent analysis, resulting in a total epoch duration ranging from 50 to 90 seconds per animal.

Field potential generators in left and right hemispheres were matched by their V_prof. Hierarchical clustering analysis of all FP generator profiles obtained from the animal population (Figure 2C) revealed that homonymous left (blue dots) and right (black dots) FP generators clustered together within distances below 0.2, with some exceptions observed in the profiles of the MPP→DG generator (red), which contributed the smallest variance but still retained the main spatial landmarks. The tight spatial match of the left and right homonymous profiles was evident from the extensive overlap of the respective spatial bands outlining the standard error at each recording site (Figure 2D). This indicated a precise left-right symmetry of recordings, as displacements along the AP or LM axes rapidly modified the shape of the V_prof due to strong anatomical curvatures (Benito et al., 2014). Additionally, previous experiments have demonstrated that the spatial coherence of the MPP→DG generator drops the fastest of all FP generators, decaying rapidly for distances separated by more than 0.2 mm (Benito et al., 2014), which was used to estimate the interhemispheric symmetry of recordings and select the group of animals for study.

As an initial global approach, we computed the overall CC for left-right pairs of homonymous generators (Figure 2E). The highest correlation was found for a remote FP generator originating in the overlying V2 cortex (as shown by the flat profile in brown in Figure 2D). The remaining generators were ranked in decreasing order of correlation as follows: L-M > Schaffer > MPP > LPP = GCsom. The first four generators had a CCmax greater than 0.6, whereas the last two in the DG have very low CCmax (<0.3). All correlations were found to be statistically significant at CCmax lag using a surrogate test (n = 1000 replications, α = 0.05) and none of the pairwise comparisons showed significant delays. The mean τ was (in ms, n = 6): 0.05 ± 0.15 (Sch), 0.1 ± 0.36 (L-M), −1.1 ± 1.6 (LPP), −0.6 ± 0.59 (MPP), 0.4 ± 0.48 (GCsom), and −0.7 ± 0.52 (Rem). Our analysis thus revealed no significant anticipation of global irregular dynamics in one hemisphere with respect to the other. It is important to note that these CC analyses were performed on wideband activities, meaning that no selection of frequency bands or individual waves was made at this stage. Further differences were observed through the fine-grained analyses detailed in the following sections.

The Schaffer generator, which reflects spontaneous CA3→CA1 input (Figure 3A), exhibited a V_prof that closely resembled that of CA3-evoked potentials. It had a maximum at the CA1 st. radiatum and decayed toward both sides, the soma layer and distal apical dendrites. Additionally, a second, smaller peak appeared in the CA3 region, caused by coherent activation of CA3 axon collaterals that comprise the excitatory recurrent network (Martín-Vázquez et al., 2016). CSD contour plots during bouts of gamma waves closely matched subthreshold CA3→CA1 fEPSPs (Figure 3B). Notably, this generator is characterized by a conspicuous baseline on which gamma waves, SPWs, and other less stereotyped fluctuations grow in the same (negative) direction, denoting multiple firing regimes of the afferent CA3 population. The responsible synaptic pathway was functionally identified through selective capture of CA3-evoked potentials (Korovaichuk et al., 2010; Fernández-Ruiz et al., 2012). For estimating some quantifiers and statistics of activity, we manually removed the SPWs, as the large variance they contribute to the total epoch may bias comparisons.

In a representative experiment (Figure 3C) the tight left-right overlap of Schaffer activities was evidenced at different time scales and regardless of the amplitude of events. Therefore, all types of temporal motifs (gamma waves, SPWs or irregular waves) were highly coherent, despite the presence of some unpaired individual waves (asterisks). The left-right Schaffer CCs from all animals are presented in Figure 3D, all of which reached a significant level at CCmax lag (surrogate analysis, n = 1000; p < 0.05). This wide-ranging synchrony is best appreciated by the stable ceiling-like value of the Pearson’s linear correlation. However, we observed frequent short-lived drops of correlation, which we explored further by building density histograms using different time windows (Figure 3E). As Δt reduced from 1 s to 10 ms, we observed a gradual transition from a strong leptokurtic distribution with a large positive mean value to another markedly platykurtic but still with a positive mean value. The same trend was consistently repeated in all animals (Figure 3E, 3), which indicates tight bilateral correlation in both slow and fast fluctuations. These fluctuations belonged to temporally structured waves of stereotyped frequency, as denoted by the significant bilateral coherence found in spectral coherence analysis for delta/theta and gamma bands in all animals (Figure 3F, surrogate test, n = 400).

In former study we showed that the temporal characteristics of the Schaffer generator are optimal for extraction and quantification of individual gamma waves (Benito et al., 2016). Hence, we conducted a similar bilateral exploration in the present set of experiments (see Methods). The results confirmed tight matching of gamma waves but poor bilateral amplitude covariation of paired left and right gamma waves (Figure 4A). Bilateral SPWs showed, however, strong covariation (Figures 4B, C). Even composing wavelets were found to have extraordinary left-right covariation, distinguishing from baseline gamma waves. Thus, the larger and longer SPW events were bilateral in contrast to baseline gamma waves even though these were paired. Associated to SPWs, fast ripple oscillations were found to maintain global left-right coherence but the wavelet series were lateralized, could even fail on one side (Figure 4D). Note that these events do not belong to the Schaffer generator.

The L-M FP generator (Figure 5A) presented V_prof with a maximum in the distal apical dendrites of the CA1, which reversed polarity in the st. radiatum, and decayed without reversal into the DG (Figure 2D). However, in CSD analysis (Figure 5B), the FP waves captured in this generator were found to be associated with both, source/sink and sink/source pairs in the same domain, which is consistent with former reports that found cohabitation of excitatory and inhibitory pathways in the st. L-M (Benito et al., 2014). Notably, the ICA was not able to discriminate waves with mirroring spatial profiles. As previously reported, the L-M generator captured theta rhythm when it is present. In such state, additional L-M generators were uncovered (Benito et al., 2014; López-Madrona et al., 2020), indicating that multiple inputs with similar profiles may remain unsegregated in the L-M generator. We essayed a formerly deployed strategy involving the modification of the spatial content of the data matrix (Benito et al., 2014) in order to enable the ICA algorithm for different spatial feature selection. We thus applied the ICA algorithm on a complete dataset of bilateral recordings and used less restrictive reduction of dimensions (see Methods). This procedure yielded three different components with similar (but not identical) spatial profiles in all animals (Figure 5C). All of these components had maxima at the st. L-M of the CA1 and exhibited different dynamics. Since they appeared twice in the spatial profiles of ICA components, one in each hippocampus (LM1-LM3), their respective bilateral activities must be highly coherent, otherwise each side would segregate into two different components, one per hemisphere, as it was the case in other generators (e.g., the LPP). However, we maintained the use of the compounded L-M generator for bilateral analysis due to the algorithm limitation for retrieving optimal dynamics in weak generators when using a single shank per hemisphere (Benito et al., 2014). Internal inconsistencies (over time) can now be explained by the composite nature, as for native FPs.

Upon examination of the generic L-M activity, it was observed that the baseline was less conspicuous compared to that of Schaffer FPs. This made it difficult to determine whether the positive and negative waves belonged to different pathways that were mixed in the L-M generator or if it was due to a zeroing effect caused by AC-filtered recordings. Previous studies (see Martín-Vázquez et al., 2013) have also noted this limitation. Nonetheless, at a coarse grain, the positive and negative waves were highly symmetrical (Figure 5D, 1a) in concordance with the high and sustained levels of linear correlation at any time windows (Figure 5D, 2). It is worth noting that drastic drops of linear correlation lasting from tens to a few hundred milliseconds (as evidenced by the red and purple ovals) corresponded to low-voltage epochs containing gamma oscillations. This was further confirmed at a fine grain scale by the poor and shifting phase matching of bilateral gamma waves in gamma-filtered traces (Figure 5D, 1b). Quantitative analysis showed the absence of significant bilateral coherence in the gamma band in spectral coherence analyses in all animals (Figure 5E; surrogate test, n = 400). In contrast, significant levels of coherence were observed for low frequency bands (<10 Hz).

Given that the characteristics of wide-band bilateral correlation of the Schaffer and L-M activities were similar, and both reflect population synaptic inputs to the same CA1 pyramidal cell population, we sought to investigate possible relationships between their respective time courses in the same hemisphere. In sample epochs (Figure 6A, 1), the dynamics of these two generators displayed high levels of divergence. While the covariance was close to zero at all the times (Figure 6A, 2), the linear correlation showed strongly fluctuating positive and negative values (Figure 6A, 3), which is also indicative of a lack of correlation. This was further confirmed by the highly symmetrical r values around zero in the density histograms. However, a notable exception was observed for Δt = 0.1, which demonstrated a right-shift of the mean (positive correlation). This later result was stable across the animal population in both hemispheres (r = 0.06 ± 0.01, 0.24 ± 0.03, and 0.15 ± 0.03 for Δt = 1, 0.1, and 0.01 s, respectively; n = 10, data pooled from 5 animals and 2 sides). Moreover, we found no indication of lagged correlation between the Sch and L-M generators. The pairwise CC between Sch and L-M yielded a mean of 0.24 ± 0.03 and a mean T_max of −73.3 ± 40.4 ms, which did not reach significant level in any animal and hemisphere (surrogate test, n = 1000). Additionally, spectral coherence did not show significant values at any frequency band, animal, or hemisphere (Figure 6A, 4; surrogate test, n = 400 each).

We conducted further investigation into the possible relationship between Schaffer and L-M inputs, with a focus on gamma oscillations that may have been overlooked in wideband analysis. To achieve this, we bandpass-filtered FP generators in the 30–100 Hz range. The mean frequency of gamma-enriched time-courses (30–100 Hz bandpass) estimated on autocorrelation functions was 35.7 ± 1.2 Hz for the Schaffer generator and 37.3 ± 1.3 for the L-M generator. High-gamma (≈ 71 Hz) can also be disclosed in both by further narrowing of the bandpass filter (60–100 Hz). Visual inspection of sample epochs revealed that Sch and L-M gamma oscillations with amplitude above the baseline were not correlated, whether in long episodes or short bouts of gamma waves (Figure 6B, 1, 2). Close examination revealed that individual gamma waves were variably out-phased between the two generators (Figure 6B, 3; r = 0.2 ± 0.03, n = 5; Δt = 0.01 s), consistent with previous findings in bilateral comparisons of the L-M generator (r = 0.26 ± 0.03). Spectral coherence analyses between the two generators did not show any significant results for any frequency band. These results contrasted with the tightly matched bilateral gamma waves observed in the Schaffer pathway. Figure 6C provides a summary of the inter- and intra-hemispheric relationships between the two CA1 generators. Both generators showed strong bilateral coherence in the delta and theta bands but were uncorrelated between them, while only the Schaffer generator maintained strong bilateral gamma coupling. However, we cannot fully guarantee bilateral gamma decoupling in the L-M generator despite the current analysis, as we found it to be composed by a mixture of several pathways.

The ICA of FPs in the DG subfield yielded three distinct components with variance greater than 5%, indicating their stability (Figure 7A). These components displayed voltage profiles with higher values and positive-going fluctuations in the hilus between DG cell layers, but differed in the location of polarity reversal (Figure 1D). Raw FP waves with other spatial configurations were observed, but were not included in the study due to insufficient cumulative variance. The stronger and the weaker components were identified as the LPP and MPP generators, respectively, as they captured subthreshold evoked potentials elicited by stimuli in the LOT di-synaptic activation input to the DG through the LPP and the MPP pathways, respectively. Figure 7B displays FPs and CSD of a sample epoch, as well as the evoked potentials for each generator, revealing source/sink pairs that correspond tightly in space with those of spontaneous waves, with sinks in the outer or the middle molecular layers of the DG, respectively. They thus reflect the ongoing separate input from the LEC and MEC to the GC population. The third generator, GCsom, corresponds to inputs at the granule-cell layer and proximal dendrites, and has been identified as a local generator whose dynamics closely follow that of the LPP generator (Benito et al., 2014).

In the present experiments, the LPP and GCsom generators exhibited alpha activity dominating over bouts of shorter (gamma) waves and irregular activity (Figure 7C). The MPP generator, on the other hand, displayed sparse low-amplitude activity and few large slow waves (Figure 7D). In both generators, a consistent baseline was observed, from which fluctuations departed in a unidirectional manner. However, gamma waves typically rode on slower potentials, which difficulted their separate extraction for precise wave-to-wave characterization.

The superposition of left and right activities in the LPP generator revealed strong mismatch (Figure 7C, 1), as evidenced by fluctuations up and down the zero value in left-right linear correlation, (Figure 7C, 2), even during long epochs of simultaneous presence of gamma waves, which indicated hazardous bilateral phasing of individual waves. The corresponding density histograms displayed small right-ward shifts at Δt ≥ 0.1 s, but were symmetrical for Δt = 0.01 s (Figure 7C, 3), indicating moderate synchronization of slow fluctuations but little in faster ones. These values were highly reproducible across the population (Figure 7C, 4). In contrast, the MPP generator showed tight left-right matching of their activities (Figure 7D, 1), resulting in higher values of correlation (Figure 7D, 2), and clear right-ward drift in density histograms for any Δt (Figure 7D, 3). These values were also highly reproducible across the population (Figure 7D, 4), and significantly larger than those for the bilateral LPP (p = 0.02 or smaller, n = 5; Student-t test). Spectral coherence for the bilateral LPP generators showed minimum significant bars limited to the delta band in 3 out of 5 experiments (Figure 7E) (surrogate analysis, n = 400). Similar findings were obtained for the GCsom generator. Conversely, the MPP generator showed significant coherence in all animals spanning the delta and theta bands, but not in the gamma band.

To check for possible leadership of the synaptic pathways represented in FP generators in one side respect to the other we essayed the Granger-causality (Granger-c) between pairs of right and left homonymous generators. Figure 8A (1) shows two instances of temporal Granger-c in a representative experiment (upper and middle rows) and along the frequency spectrum in the two directions (2). The significant frequencies (over the gray profiles) in such spectra were largely but not completely coincident between directions (e.g., upper row). Nearly all pairwise comparisons (57 out of 60 comparisons: 5 animals x 6 generators x 2 directions: Supplementary Table 1) yielded statistically significant value (surrogate test, n = 1000), and we found no clear leading of one side respect to the other. The net value was, however, generator-specific (mean population value is color coded in Figure 8B along a diagram representation of directionality), and it followed the general trend found in precedent tests, that is, high between all homonymous pairs except the LPP and GCsom generators that showed weaker value (compare significant bars in Granger-c and spectral coherence plots: Figure 8A, 3).

We also essayed the Granger-c test on pairs of generators in the same hemisphere. The results were more varied than interhemispheric correlations (Figure 8B, 1), and numerous directional relations were found as well as non-significant pairs in either direction. Except in the case of the LPP generator that showed the strongest directionality over all other generators, the Granger-c indices were weak and reached statistical level in only a fraction of the animals tested (plot in gray squares). The results in Figure 8C correspond to ipsilateral comparisons in the left side, and were largely replicated in the right side.

Comparisons of population activity in neuron populations are traditionally performed by FPs in the frequency domain. The uncontaminated temporal fluctuations of FP generators allow for more secure comparisons in the time domain. Prior studies using realistic feed-forward models of FPs have demonstrated that while FPs are formed from elementary synaptic currents of relatively stable duration, their scaling up to mesoscopic FPs give raise to waves of variable length when combined in different temporal patterns, which enters energy in different frequency bands (Makarova et al., 2011). As such, their interpretation should be approached with caution as they do not directly correspond to true elementary waves or their intervals. To minimize over-interpretation by relying on a single aspect (Hillebrand et al., 2012; López-Madrona et al., 2017), we have utilized several tests exploring different temporal, frequency, and leadership characteristics.

The ICA has been successfully used to explore different aspects of FP mixtures, either in the spatial or the temporal and frequency domains (Makarov et al., 2010; Schomburg et al., 2014; Lopes-Dos-Santos et al., 2018; Rogers et al., 2019). Demixing of irregular FP activity is well suited to ICA, but that of gamma oscillations is difficult in certain regions, such as the DG, especially during periods of high co-activation due to excessive spatial and temporal overlap. In previous computational studies, we showed that co-activation of more than three spatially distinct generators with a similar frequency results in a significant reduction in temporal accuracy by the ICA (Makarova et al., 2011). Methods to increase the temporal precision of separation have been advanced, but they may not achieve sufficient precision in some epochs (Benito et al., 2014, 2016). Without optimization measures, accurate extraction of gamma waves in strongly overlapping DG generators (e.g., Fernández-Ruiz et al., 2021) is highly susceptible to error, while being more reliable when applied to CA1 gamma sources (see Herreras et al., 2023 for a repertoire of gamma-related phenomena). In this study, we have opted for cautious use of correlation or spectral coherence tests, which are suitable for studying global flow and coherence between paired generators of bilateral circuits. It is possible to extract individual gamma waves for bilateral wave-to-wave comparison in exceptionally favorable generators such as the Schaffer generator (Figure 4; Benito et al., 2016). We have attempted to perform the same process on gamma waves from other generators in the DG, but the results have not been satisfactory, mainly due to the lack of an adequate baseline that allows precise estimation of time and amplitude parameters of individual waves.

The FP generators analyzed show very different dynamics, even though some of them are part of the canonical excitatory trisynaptic circuit (Andersen et al., 1971; Herreras et al., 1987), where some impact from one node or population to another might have been expected. The failure to do so implies that the dynamics of each segment results from the convergence of several synaptic afferents in each population, and would explain the small number of ipsilateral generator pairs showing directionality with the Granger-c test, confirming previous findings in awake rats (López-Madrona and Canals, 2021). Only the GCsom generator showed dynamics inherited from the LPP, as can be expected from an interneuron subsystem that receives few inputs, preferentially from its upstream excitatory population.

Theoretical studies have found that effective connectivity in the hippocampus can be strongly modulated by multiple factors, including the interaction between common afferents, average firing frequency, or intrinsic excitability (Battaglia et al., 2012; Han et al., 2015; López-Madrona et al., 2017; Pariz et al., 2018). On the one hand, experimental studies conducted on single subfields will be highly dependent on the study conditions, and on the other hand, theoretical studies show that unless structural and functional constraints are incorporated, almost any parameter can be decisive in one range or another (Migliore et al., 1995; López-Aguado et al., 2000; Ibarz et al., 2006). This is the reason why simultaneous experimental observation of several nodes is a necessary approach to first delineate the physiological ranges of activity in specific synaptic pathways.

Changes in effective connectivity have already been observed within the hippocampus. A most remarkable case is the opposite modulation of excitability and neuronal firing rate in CA1 and DG upon changes from irregular to theta related behaviors, by which GCs augment firing rate whereas there is a drastic reduction of CA1 pyramidal cell firing (Leung, 1980; Buzsáki et al., 1983; Herreras et al., 1988; Vinogradova, 2001). The flow reversal occurs in the CA3 region that is capable to perform strong nonlinear transformations (Guzowski et al., 2004). This region is an important node where cortical and subcortical information converge, and is also the main hippocampal node of interhemispheric connection. The opposite behavior of two consecutive nodes in a circuit not only demonstrates it does not operate as synfire chains (feed-forward assembly-based transmission: Abeles, 1991), but also evidences that the output of a population may have little temporal correspondence with its inputs, mostly when it is engaged in performing specific logical operations, such as feature extraction, coincidence detection, or others. Despite the fact that entorhinal afferents contact all three hippocampal nodes (DG, CA3, and CA1) and bypass the canonical trisynaptic chain (Hjort-Simonsen and Jeune, 1972; Steward and Scoville, 1976; Ruth et al., 1982, 1988), each of these shortcuts can transmit different dynamics (Traub and Whittington, 2022). In this line, the different FP generator dynamics before and after CA3 reflects logical processing within specific nodes of the circuit.

Experimental studies with repetitive stimulation have shown that transmission through hippocampal circuits to the cortex is frequency-dependent (Andersen and Lomo, 1967; Herreras et al., 1987; Moreno et al., 2016). Optimal frequencies depend on the node used for stimulation, which indicate that different frequency filters could be sequentially imposed on successive nodes. Intuitively, this should end flow of activity after a few nodes. However, exogenous stimulation imposes frequency-dependent cumulative effects on single cells that do not take place during the sparse and scattered transmission of spontaneous activity. According to present data, spontaneous transmission appears to operate quite differently, as neither long nor short waves show coherence across ipsilateral FP generators.

Bilateral coordination may arise from external nuclei with bilateral projection to both hippocampi (e.g., raphe, medial septum, nucleus reuniens), or it may be established internally through direct interhemispheric (CA3-CA3, CA1-CA1), crossed (CA3-CA1), or indirect connections (DG-DG through hilar cells: Amaral, 1978). Interestingly, while the MEC and various retrohippocampal areas have reciprocal and crossed connections, the LEC lacks direct reciprocal connections (Steward and Scoville, 1976; Köhler et al., 1978; Amaral et al., 1984; van Groen and Wyss, 1990; Watson et al., 2017), which explains the independent activity of the left and right LEC generators. Given its position at the beginning of the cortico-hippocampal chain, this seems like a logical result given the considerable functional lateralization of the cortex in humans, as well as in rodents (Cohen and Wilson, 2017; Dieterich and Brandt, 2018; Bartolomeo and Seidel Malkinson, 2019).

A candidate for interhemispheric synchronization might come by the highly symmetric input from the MEC, which has strong reciprocal bilateral connections and may impact all nodes of the trisynaptic circuit (Blackstad, 1956; Köhler et al., 1978). Yet, its low level of activity under present conditions does not make it appropriate for acting as a coordinator of bilateral subfields. Instead, we consider node-by-node synchronization via direct or indirect (disynaptic) connections to be more likely. In this respect, it is notable that individual generators show different regimes of activity, displaying slow waves of varying duration interspersed with shorter highly stereotyped gamma waves. In a former study we showed that in the Schaffer segment a gamma wave corresponds to the synaptic envelope caused by synchronous firing of a CA3 assembly (Fernández-Ruiz et al., 2012), as also confirmed by realistic feed-forward modeling (Martín-Vázquez et al., 2013). Although the firing synchrony of the afferent neurons that elicit a gamma wave is stronger than that of the slower waves, these are much larger and therefore require the activation of a much larger pool of afferent neurons. Therefore, the synaptic envelope built on individual postsynaptic cells is more likely to reach threshold (Combe et al., 2018) and thus have a stronger impact on the contralateral side. For instance, SPWs are associated to strong recruitment of pyramidal cell firing in CA1 (Csicsvari et al., 1999) whereas this is smaller and very specific during gamma waves (Senior et al., 2008; Fernández-Ruiz et al., 2012). In fact, the bilateral synchrony of gamma waves in this segment is produced by coordination of autonomous interneuronal oscillators on each side (Bartos et al., 2002; Börgers et al., 2012; Benito et al., 2016), since it persists, although without bilateral pairing, after blockade of the commissure. In contrast, the longer and larger SPWs disappear on both sides although they resume later, indicating their dependence on interhemispheric reciprocal excitation (Martín Vázquez and Herreras, 2015).

The different dynamics of FP generators in the same hemisphere indicates that FP timing does not reflect simple forward transmission. Rather it is compatible with specific processing at each node, and the lack of bilateral coherence for fast waves indicates different streams of information in each side. However, we only count with a few FP generators. We lack the GC output to CA3 population and the CA1 output to the subiculum. The first of these two has been found to be lateralized during processing of contextual navigational cues using calcium imaging of GC populations (Cholvin and Bartos, 2022). This result is compatible with the LEC inputs differing in left and right sides we found here. Experiments to check if lateralization of the LPP maintains in behaving animals are on the way in our lab.

In turn, the wave-to-wave bilateral gamma coupling of the CA1-CA3 Schaffer generator appears to be an exception to the lateralized dynamics of gamma waves in the other FP generators. It seems to support that CA1 output at the end of the hippocampal chain has become bilateral on its way to the cortex. However, as previously shown and it is further confirmed in this study, the left and right Schaffer gamma waves have very low amplitude covariation, indicating that the neural assemblies responsible for each gamma wave in each hemisphere (Fernández-Ruiz et al., 2012) transmit essentially different information (Benito et al., 2016). Based on this, we infer that long (large) waves in the delta/theta band are likely to transfer information both ipsi and contralaterally while gamma waves would remain largely ipsilateral. Congruent with this view is that SPWs match bilaterally much more precisely than gamma waves in the same generator. Even, other faster waves that could be expected to match bilaterally as the ripples, are moderately coherent but poorly matching wave-to-wave (Villalobos et al., 2017; this work).