Summary
Theta-gamma cross-frequency coupling (CFC) is thought to route information flow in the brain. How this idea copes with the co-existence of multiple theta rhythm generators is not well understood. We have analysed multiple theta and gamma activities in the hippocampus to unveil the dynamic synchronization of theta oscillations across hippocampal layers, and its differential coupling to layer-specific gamma frequency bands. We found that theta-gamma CFC is stronger between oscillations originated in the same hippocampal layer. Interestingly, strong CFC was linked to theta phase locking across layers in a behaviourally related manner, being higher during memory retrieval and encoding. Systematic analysis of cross-frequency directionality indicated that the amplitude of gamma oscillations sets the phase of theta in all layer-specific theta-gamma pairs. These results suggest, contrary to an extended assumption, that layer- and band-specific gamma-oscillations coordinate theta rhythms. This mechanism may explain how anatomically distributed computations, organized in theta waves, can be bound together.
Introduction
Brain oscillations of different frequencies are thought to reflect a multi-scale organization in which information can be bound or segregated in oscillatory cycles1–3. Interactions between different oscillations, known as cross-frequency coupling (CFC)4–6, have been measured in multiple brain regions during perception7–9, attention10,11 and memory formation7,12–15. It is believed that these interactions play a role in the coordination of local computations and large-scale network communication1–3,16–18. In the hippocampus, theta and gamma oscillations are the most prominent rhythms recorded in freely moving animals19,20, and it has been proposed that information transmission between the CA1 region and its afferent regions in CA3 and the entorhinal cortex (EC) is organized in separated gamma frequency channels that are synchronized by the phase of the slower CA1 theta rhythm18. The output activity from CA1 has been shown to organize in yet another separated gamma band specifically overlapping the pyramidal cell layer21,22 and the coupling of the three layer-specific gamma oscillations to the underlying theta rhythm was shown to depend on the behavioural state21–24. However, theta oscillations originating in different anatomical layers are also known to coexist in the hippocampus19,20,25–31, and therefore theta-gamma interactions need to be understood in the context of multiple rhythm generators.
In addition to the classical medial septum/diagonal band of Broca input imposing a global rhythmicity to the hippocampus and EC, important rhythm generators are located in EC layers II (EC2) and III (EC3), whose activity reach the DG and hippocampus proper though the perforant and temporoammonic pathways respectively, and from CA3 activity reaching CA1 stratum radiatum through the Schafer collaterals19. Importantly, although theta oscillation in the hippocampus are most commonly studied as a unique coherent oscillation across hippocampal layers, exhibiting a characteristic amplitude/phase vs. depth variation19, the frequency and phase of the CA3 theta rhythm generator was shown to change relatively independently from the EC theta inputs32. How these multiple theta rhythm generators and layer-specific gamma oscillations interact in the hippocampus is not well understood. One intriguing possibility is that theta and gamma activities of different laminar origin may represent independent communication channels with possibility to coordinate distributed processes.
Here we investigated the function of layer-specific synchronization of oscillatory activity in the hippocampus of rats freely exploring known and novel environments and resolving a T-maze. Using high density electrophysiological recordings aided by source separation techniques we characterized the dynamic properties of three different theta and three different gamma dipoles in the hippocampus with origins in the CA3 Schaffer collateral layer, the EC3 projection to the stratum laconusum-moleculare and EC2 projection to the mid-molecular layer of the DG, respectively, and found strong support for the existence of relatively independent theta-gamma frameworks. We further characterized theta-gamma interactions between the different layers and established an association with the synchronization state in the hippocampal network. Finally, we investigated the functional role of the characterized theta-gamma coordinated time frames for contextual learning.
Results
Layer-specific theta and gamma oscillations
We performed electrophysiological recordings using linear array electrodes across the dorsal hippocampus in five rats (see Supplementary Methods, Figure 1). Recordings were carried out while the animal explored an open field or a T-maze. Using spatial discrimination techniques to separate LFP sources contributed by different synaptic pathways, based on independent component analysis (ICA)33–37, we dissected 3 robust components in all subjects (Figure 1; Supplementary Methods). The maximum voltages (Figure 1b) and dipoles in the current source density (CSD) depth profiles (Figure 1c) of the three components matched the stratified distribution of known terminal fields in the hippocampus, and the currents resulting from stimulation of the corresponding pathways, as previously shown21,22,35,36. The first component was located in the stratum radiatum, where the CA3 Schaffer collateral/commissural pathway targets the CA1 region (labelled as Schaffer component or Sch-IC). The second matched the EC3 projection in the stratum lacunosum-moleculare (lm-IC), and the third one the perforant pathway from EC2 to the mid-molecular layer of the DG (PP-IC). These three components, referred to as layer-specific LFPs or IC-LFPs, represent the synaptic contributions with distinct anatomical origins recorded in the LFP37.
The power spectra of these signals exhibited a clear peak at theta frequency (6-8 Hz) and prominent broadband gamma activity (Figure 1e). The dominant theta current sinks and sources calculated from the three layer-specific LFPs (Figure 1f) showed phase differences consistent with the firing properties of principal neurons in their respective upstream afferent layers38. Entorhinal principal cells in EC2 and EC3 have been shown to fire in anti-phase, relative to the theta oscillation and, accordingly, theta current sinks (Figure 1f) and large amplitude gamma oscillations (Figure 1g) in PP-IC and lm-IC were shifted 180°. CA3 and EC3 neurons have been previously shown to fire phase locked to discrete gamma band oscillations in the downstream Sch-IC and lm-IC, respectively21,39, with gamma oscillations segregated in the phases of the slower theta wave recorded in CA118,21,22. In good agreement, layer-specific gamma oscillations were segregated in the theta cycle, with lm-IC close to the theta peak and followed by Sch-ICin the descending phase to trough of the cycle and PP-IC in the transition from the trough to the ascending phase of the theta cycle (Figure 1g). Overall, these results support the use of multichannel recordings and source separation tools to investigate interactions between theta and gamma current generators in multiple layers of the hippocampal formation.
Different theta frameworks coexist in the dorsal hippocampus
The frequency and phase of the CA3 theta rhythm generator has been shown to change relatively independently from the EC theta inputs32. However, theta oscillations in the hippocampus are commonly viewed as a unique coherent oscillation. Taking advantage of the separation of theta activity in layer-specific theta oscillations (Figure 1), we now investigated their interaction across layers (Figure 2a). To this end we computed a synchronization index (SI) measuring the delay difference between two consecutive theta cycles, normalized to the maximal phase difference found in the complete time series (Supplementary Methods). We computed the SI for all pairs of IC-LFPs and for the three components simultaneously. Phase delay and phase difference distributions between layer-specific theta oscillations are shown in Figure 2b and Figure S1, respectively. Synchronization between theta current generators varied dynamically within recording sessions, with periods of low SI alternating with periods of high SI (Figure 2c). Measured over all recording sessions and animals, high synchronization between pairs of theta generators dominated (Figure 2d) with the distribution of SI values per theta cycle showing a peak close to perfect phase locking (Figure 2d). The highest synchronization was found between PP-IC and lm-IC (ANOVA with degrees of freedom corrected by Greenhouse-Geisser, F(1.165,4.660)=531.8; p<0.001 for mean SI values), likely reflecting the tight coordination between the afferences received from layers II and III of the EC. Interestingly, synchronization computed for the three theta oscillations simultaneously, showed a broader distribution with peaks at low and high SI values, revealing the coexistence of both, desynchronized and highly synchronized states, respectively (Figure 2d). Measuring the average SI in sliding time windows of increasing duration, we found that high synchronization epochs preferentially last three consecutive theta cycles (Figure S2)18.
We further found that theta power in lm-IC and PP-IC correlated with the synchronization state, with larger theta power associated with higher synchronization (Figure 3a and 3b). Theta frequency was higher in average in the Sch-IC (ANOVA, F(1.359,5.436)=54.94, p<0.05), and correlated with theta synchronization in all IC-LFPs, but most notably in lm-IC and PP-IC (Figure 3d). Larger SI values were associated with higher theta frequency (Figure 3d), with maximal frequency values undistinguishable between IC-LFPs (ANOVA, F(1.163,4.651)=4.443, p>0.05). In this way, high synchronization states between the three generators are reached by increasing the theta frequencies in PP-IC and lm-IC to match the Sch-IC frequency. Overall, these results indicate that although a single coherent theta rhythm can be found across hippocampal layers, layer-specific theta oscillations also appear desynchronized, supporting the coexistence of different theta frameworks. In contrast to theta power, broadband (30-150 Hz) gamma power did not correlate with theta synchronization in any IC-LFP (Figure 3a and 3c), indicating that broadband gamma power is not driven by theta synchronization.
Theta-gamma CFC reflects layer-specific interactions
The existence of different theta frameworks opens the possibility to multiple theta-gamma interactions (Figure 4a). We therefore computed the theta-gamma CFC within and between hippocampal layers. We first measured the amplitude of gamma oscillations in the three IC-LFPs referenced to the phase of theta recorded in the CA1 pyramidal layer, as is usually done14,15,18,21–24,40. The results (Figure 4b) confirmed previous findings showing coupling between CA1 theta and a slow gamma band of CA3 origin (Sch-IC; maximal modulation at 37.5 ± 5 Hz, CA1γS)18,21,22 and a medium gamma band of EC3 origin (lm-IC; 82.5 ± 4 Hz medium gamma, CA1γM)18,21,22. They also revealed an additional theta-nested fast gamma band (130 ± 10 Hz,) in the mid-molecular layer of the DG (DGγF) overlapping the terminal field of EC2 inputs, compatible with the previously found theta-gamma CFC in the DG4. DGγF and CA1γM in PP-IC and lm-IC, respectively, were locked close to 180° phase (Figure 4d) as shown before38. CA1γM activity was closely followed by CA1γS in the theta cycle (Figure 4d)21. The key new finding in our analysis was the existence of theta-gamma coupling dominant within each layer, this is, the CFC between the theta and gamma oscillations recorded in the same IC-LFP was stronger than any between-layer combination (Figure 4c and 4d and Figure S3). Theta-gamma CFC reflects layer-specific interactions and further supports the existence of independent theta-gamma synchronization frameworks in the hippocampus32, rather than a single theta framework that implies a unique carrier theta wave to which the gamma activity is multiplexed in segregated theta-gamma channels.
Synchronization between theta frameworks is associated to local theta-gamma CFC
Knowing that theta-gamma CFC is layer-specific (Figure 4) and that different layers can oscillate relatively independently (Figure 2 and 3)32, we explored theta-gamma CFC accounting for the different synchronization states. This analysis unveiled a striking correlation between the CFC and theta synchronization (Figure 5a). Strong theta-gamma modulation was associated to high SI values, while weak or nearly absent CFC was found in periods of low SI (Figure 5b). This result indicated that within-layer CFC was associated to the synchronization between layers and allowed us to hypothesize that CFC could represent a mechanism to synchronize theta frameworks.
Theta and gamma oscillations reflect the extracellularly added synaptic and active dendritic currents of two processes occurring at different timescales, the second being tightly paced by inhibitory neurons37. We asked which of these processes was driving the CFC between the two frequencies. We computed the cross-frequency directionality index41 (CFD) (Supplementary Methods), based on the phase-slope index, which computes the phase difference between two signals as an indication of directed interactions between both frequencies. An increase of the phase difference between the theta phase and the gamma amplitude with frequency gives rise to a positive slope of the phase spectrum (i.e. a positive CFD value) when the phase of the slow oscillation sets the amplitude of the fast, and negative otherwise. As shown in Figure 5c for the averaged data, and Figure S4 for individual animals, in all considered cases CFD resulted in negative values for the specific gamma bands nested to the theta oscillations in each IC-LFP (CA1γS, CA1γM and DGγF, respectively). This was the case whether we analysed the complete time series, or split them into theta-synchronized (SI > 0.8) or desynchronized (SI < 0.4) epochs (Figure 5c). The results suggested that the amplitude of the gamma oscillation effectively modulates the phase of the theta wave.
Behavioural modulation of theta synchronization and CFC
The strength of CFC in the hippocampus has been shown to correlate with learning15 and we found here that it correlates with theta coherence across hippocampal layers (Figure 5a). We hypothesized that the CFC represents a mechanism to synchronize theta oscillations and coordinate information transmission across hippocampal layers. This hypothesis allows us to predict that CFC and theta synchronization would predominate during learning, maximally when the animal updates an existing memory with novel information, a condition likely requiring coordination between incoming sensory inputs, the retrieval of the stored contextual information and updating the memory with new information42–44.
Therefore, in our final set of experiments we tested this prediction using two behavioural paradigms. In the first one, novel information was added on top of a stored contextual memory. In the second, a hippocampus-dependent delayed spatial alternation task was used in which the animal needed to remember the arm visited in the previous trial and update the memory with the choice made in the current trial45–47. In the first task, after habituation to an open field (8 min session 1 per day during 8-10 days), we introduced a novel tactile stimulus in the floor of the otherwise unchanged field (novelty session, see Supplementary Methods). We computed and compared theta synchrony and CFC between the novelty session and the habituation session the day before. When the animal entered the arena, the theta SI was high and comparable in both conditions during the first two minutes of exploration (Figure 6a, t1). As the animal explored the context, synchronization remained high during novelty, but rapidly decayed in the known environment (Figure 6a, t2). Consistent with the notion of information transmission to update an existing memory, by the end of the exploration time both conditions decreased to the same level of theta synchronization (Figure 6a, t3), when the introduced tactile stimulus lost its novelty. The simultaneously computed CFC MI correlated with theta synchronization during the complete session in both conditions, as shown in Figure 6a and 6b. Importantly, CFC strength was higher during the novelty sessions, when the theta synchronization was also higher, and decreased towards the end of the session in parallel with SI (Figure 6b).
In the second experiment, rats learned in an 8-shaped T-maze to alternate between the left or right arms on successive trials for water reward (Fig. 6d). In this task, the central arm is associated with memory recall, decision making and encoding of the current decision14,46–48. We first computed and compared theta-gamma CFC between the side and central arms and found significantly increased modulation in the central arm for the three IC-LFPs (Fig. 6e). This result is consistent with previously shown increased CFC in the same task in the CA1 radiatum and lacunosum-moleculare IC-LFPs21, although in that case the authors used a common theta reference recorded in CA1 pyramidal layer. Theta synchronization across hippocampal layers in the T-maze showed comparable average values as during open field explorations (Fig. 6f) and, again, differentiated between side and central arms (Fig. 6g). Importantly, the SI was higher in the central arm (Fig. 6g) consistent with the higher CFC MI found in the same behavioural epochs (Fig. 6e). Overall these results support the above prediction and the idea that CFC synchronizes theta frameworks in the hippocampus facilitating information transmission to update a contextual memory.
Discussion
Overall, our results provide functional evidence supporting independent theta oscillations in the hippocampus whose coordination can be seen as a mechanism to channel information between hippocampal layers and binds distributed computations. Less synchronized theta states may secure relatively independent processing in local circuits of the hippocampal formation. Theta coordination (phase locking) is achieved by increasing theta frequency in all layers, most notably in those receiving EC inputs, and strongly correlates with the strength of theta-gamma CFC, so that coherent theta oscillation across hippocampal layers is associated with stronger CFC. Furthermore, directionality analysis suggests that band- and layer-specific gamma activity contributes to the synchronization of theta oscillations across layers. We thus hypothesize that the CFC represents a mechanism operated by gamma activity to coordinate theta oscillations in separated regions. In a network with multiple connected nodes, theta-phase locking between specific nodes will further contribute to the directionality of the information flow, habilitating targets between which communication is permitted in defined time windows. We have provided evidence supporting this hypothesis by showing that CFC and the coordination between the theta generators recorded in the hippocampus increase in the mnemonic process.
Extensive previous research has demonstrated the existence of multiple theta rhythms and current generators in the hippocampus and EC (reviewed in 19. While septal activity is required for theta rhythmicity, and lesions targeting the medial septum eliminate theta oscillation in both structures, intrinsic hippocampal activity from CA3 and extrinsic EC inputs do also contribute to the recorded theta oscillations19. Surgical removal of the EC unveils a theta oscillation that depends on the integrity of CA3 and is highly coherent across hippocampal layers4. In the presence of an intact EC, however, the coherence between theta signals in the stratum radiatum and lacunosum-moleculare, receiving the inputs from CA3 and EC3 respectively, is reduced32. The relative independency of theta oscillations in these layers is supported by our findings, showing that frequency and phase can be regulated independently in these theta generators (Sch-IC and lm-IC, respectively, Figure 2 and 3). Furthermore, the key new finding of our experiments is that theta coherence between the generators is not fixed; it rather changes dynamically and is regulated by behaviour. Accordingly, periods of perfect phase locking (SI = 1) between Sch-IC and lm-IC were also frequent (Figure 2), and enhanced during particular behaviours (Figure 6). Brain oscillations can be seen as rhythmic changes in neuronal excitability that can define sequential information packages in the framework of assemblies16,49. Therefore, the dynamic variation in theta synchrony between hippocampal layers found in this study, likely reflects the existence of multiple theta-coordinated time frames with phase differences between oscillations having a large impact on the timing of principal cells firing in the respective layers. Synchronization of theta frameworks will, in turn, coordinate, though not necessarily synchronize50, firing sequences in consecutive hippocampal stations.
Interactions between the phase of the theta oscillation and the amplitude (power) of the gamma activity have been extensively documented and proposed as an effective mechanism to integrate activity across different spatial and temporal scales1–18. In the hippocampus and despite the evidence supporting the existence of multiple theta generators (see above), theta-gamma CFC has been almost exclusively investigated as the statistical dependence between the gamma activity in different frequency bands and a single theta reference, most commonly recorded in the pyramidal layer of the CA1 region. Our analysis demonstrates that phase-amplitude CFC between theta and gamma oscillations in the hippocampus varies for different theta generators, and is stronger when both oscillations are recorded in the same anatomical layer (Figure 4 and Figure S3). This observation, together with previous and important evidence demonstrating that firing of principal cells in CA3 and EC3 is phase-locked to downstream gamma oscillations recorded in the CA1 stratum radiatum (CA1γS) and lacusosum-moleculare (CA1γM), respectively18,21,22, suggest that CFC is a local phenomenon driven by upstream afferences organizing downstream gamma oscillations. Our CFD analysis further supports this interpretation, since it shows a predominant amplitude-to-phase coupling for the same layer-specific gamma frequency bands. This does not mean that 4-11 Hz oscillations in the hippocampus are generated by gamma activity, on the contrary, our data suggest that gamma activities, reflecting the interplay of inhibitory-excitatory networks51–56, impose phase shifts on the on-going theta oscillations in their corresponding layers. Therefore, we hypothesize that local gamma-generating circuits, driven by afferences from their respective upstream layers, might not be activated at a particular theta phase, as commonly interpreted, but rather, might be actually setting the phase of the local theta oscillation. This interpretation would also explain phase-phase coupling between CA1 theta and CA1γS and CA1γM57, as the consequence of theta phase driven by layer-specific gamma activity entrained by upstream inputs in CA3 and EC3, respectively.
The proposed new scenario provides a mechanism to coordinate distributed computations organized in theta waves, by synchronizing theta oscillations in connected regions through theta-gamma CFC. The highly significant positive correlation between CFC strength and theta synchronization found in our study (Figure 5) supports this view. In this way, our results link theta-gamma CFC6 and coherence-based communication49, the first being the mechanism to align higher excitability windows to open communication channels between defined network nodes. A number of different studies investigating learning and memory processes in the entorhinal-hippocampal network have indeed, but separately, reported increases in theta-gamma CFC or synchronization in different frequency bands associated with memory performance1–3,16. The association of both phenomena in our study has been possible by the separation of pathway-specific LFPs, identifying the CFC as a local phenomenon and separating theta activities, that otherwise would have render a mixed readout of theta-gamma interactions, highly dependent on electrode implantation coordinates and interindividual variability37. Here we have further demonstrated that the strong association between CFC and synchronization between the three theta oscillations predominantly occur when the animal updates a memory with novel information (Figure 6). The behavioural test used were selected to contrast behavioural epochs in which the animal explores a known environment, recalling its representation from memory, and simultaneously needs to encode new information (novel stimuli, last arm visited) to update the memory. In this condition, several processing streams likely coexist and need to coordinate in the entorhinal-hippocampal network42–44. It is in these conditions that we found an increased association between CFC strength and theta synchronization across layers, suggesting its role in coordinating distributed computation during memory formation.
We have suggested a causal link between CFC strength and theta synchronization through the gamma-driven adjustment of theta phase and frequency (Figure 3 and 5). While dissecting the precise circuit mechanisms supporting the transfer function from gamma activity to theta phase is out of the scope of the present work, several possibilities exist. Computational works have demonstrated that theta-gamma CFC emerges from the interactions between functionally distinct interneuron populations, as the basket and oriens-lacunosum-moleculare (OLM) cells, interconnected in a network of principal cells receiving an external theta rhythm generator, such as the septal input53–56. The on-going theta oscillation is thus modulated by the inhibitory network, whose activity is known to be reflected in the gamma activity in in vivo extracellular recordings40,51,52, and is associated to layer-specific inputs18,21,22. Subsets of interneurons can phase-lock to different hippocampal rhythms58,59 and, interestingly, recent findings showed in the CA1 region that some interneurons can specifically phase-lock to CA1γS and others to CA1γM, supporting the idea that different classes of interneurons drive slow and medium gamma oscillations22,24,60. Thus, an appealing mechanism for gamma-modulation of theta phase would be the control of different interneuron classes by layer specific inputs, which would entrain specific gamma networks modulating principal cell excitability and firing in response to on-going theta inputs, advancing or delaying theta phase. Spiking resonance in principal cells may contribute to this mechanism too, since optogenetic activation of basket interneurons (parvalbumin expressing cells) in the hippocampus and neocortex has been shown to pace pyramidal cell firing in the theta range, by virtue of postinhibitory rebound of Ih activity51. In that experiment, theta-band firing of excitatory neurons required rhythmic activation of basket cells, as white noise activation effectively modulated their activity but did not entrained pyramidal theta-band firing51, suggesting that feed-forward activation of interneurons from upstream layers or an external rhythmic input (i.e. cholinergic or GABAergic inputs form the septum), are required for resonance amplification. Thus, intrinsic cellular properties and network mechanism may interact to support gamma-dependent coordination of theta phases across hippocampal layers.
Interactions between slow and fast brain oscillations have been measured in multiple brain regions during perception, attention, learning and memory formation1–3. Despite its ubiquitous presence in fundamental cognitive processes, its function is largely unknown. Our results provide a mechanism for binding distributed computations packed on theta waves and routing the information flow based on theta-gamma cross-frequency coupling. Important questions remain to be answered. How theta synchronization in the hippocampus relates to hippocampal-neocortical interactions61,62 known to be favoured at theta and beta frequencies63,64 and modulated by synaptic plasticity in the hippocampus65,66? The conditions triggering the coordination between theta-gamma frameworks are not well understood, nor are the precise cognitive processes they subserve, but given that theta-gamma uncoupling seems to represent an early electrophysiological signature of hippocampal network dysfunction in Alzheimer’s disease67–70 as well as for schizophrenia and other psychiatric disorders71–73, further and detailed mechanistic investigations are granted.
Acknowledgments
We thank Begoña Fernández for excellent technical assistance and Laura Pérez-Cerveza for her input during data pre-processing with ICA. S.C. and D.M. were supported by the Spanish Ministerio de Economía y Competitividad (MINECO) and FEDER funds under Grant Nos. BFU2015-64380-C2-1-R and −2-R, respectively. S.C. was supported by the European Union Horizon 2020 research and innovation programme under Grant Agreement No. 668863 (SyBil-AA) and acknowledges financial support from the Spanish State Research Agency, through the Severo Ochoa Program for Centres of Excellence in R&D (SEV-2017-0723). C.R.M. and E.P. acknowledge support from MINECO trough project Nos. TEC2016-80063-C3-3-R and −2-R, respectively. C.R.M. also acknowledges financial support from the Spanish State Research Agency, through the María de Maeztu Program for Units of Excellence in R&D (MDM-2017-0711). O.H. was supported by MINECO under Grant No. SAF2016-80100-R. V.J.L. was supported by a predoctoral fellowship La Caixa-Severo Ochoa from Obra Social La Caixa.
References
Supplementary Bibliography
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.
- 16.
- 17.