Single neurons throughout human memory regions phase-lock to hippocampal theta

  1. Daniel R. Schonhaut
  2. Ashwin G. Ramayya
  3. Ethan A. Solomon
  4. Nora A. Herweg
  5. Itzhak Fried
  6. Michael J. Kahana

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Functional interactions between the hippocampus and cortex are critical for episodic memory. Neural oscillations are believed to coordinate these interactions, and in rodents, prefrontal neurons phase-lock to hippocampal theta oscillations during memory-guided behavior. We assessed inter-regional phase-locking to hippocampal oscillations in humans by recording 1,233 cortical and amygdala neurons and simultaneous hippocampal local field potentials in 18 neurosurgical patients. We identified 362 neurons (29.4%) from multiple regions that phase-locked to rhythmic hippocampal activity, predominantly at theta (2-8Hz) frequencies. Compared to baseline spiking, strong theta phase-locking coincided with regionally-specific increases in hippocampal theta power, local and hippocampal high frequency activity, and cross-frequency power correlations between the hippocampus and a phase-locked neuron’s local region. These results reveal that spike-time synchrony with hippocampal theta is a defining feature of cortico-hippocampal functional connections in humans. We propose that theta phase-locking could mediate flexible inter-regional communication to shape the content and quality of episodic memories.

Article activity feed

  1. eLife

    ###Reviewer #3:

    The manuscript by Schonhaut et al. presents novel analysis on an impressive dataset of more than 1200 neurons across diverse brain areas in the human brain to investigate their modulation by hippocampal theta oscillations. They found a substantive proportion of cells phase-locked to hippocampal activity, mainly in the theta frequency band, in several areas known to be functionally related to the hippocampus, some of them receiving monosynaptic hippocampal inputs but other only indirect ones. These results extend previous reports in humans showing hippocampal interactions with these structures but at the level of mesoscopic activity and highlight the ubiquity of spike-theta timing and the importance of single-unit studies in humans. Additional analysis, detailed below, will contribute to give a better description of the data, provide stronger support for some of the authors' claims and clarify some issues.

    1. I assume that the dataset also includes hippocampal units, why then excluding them from the analysis? Although the main novelty is in the coupling of cells in other structures with hippocampal LFPs, it would be useful to also compare it with the coupling of local hippocampal cells.

    2. Include average power spectrum of hippocampal LFPs. Additional examples of raw LFP traces overlaid to spectrograms (perhaps in Supplementary) will help to illustrate the nature of hippocampal oscillations.

    3. The authors compared fractions of significantly modulated units and their preferred frequencies across regions. While very informative, these analyses are not sufficient to capture the richness of spike-LFP interactions likely existing in the dataset. Were there differences in the strength of phase-locking across regions? (this analysis could be added to Figure 2). Studies in rodents have shown that theta phase-locked units in different structures have characteristics preferred firing phases (when hippocampal LFP is used as a reference). Authors can easily look if this is also the case in their data. They should include both pooled data statistics of mean phases across regions and single neuron examples of firing probability by LFP phase (such examples could be added to the single unit plots in Figure 1).

    4. Did phase-locked and non phase-locked units have different properties? The authors can compare if they differ in basic properties such as mean firing rate, waveform width, inter-spike intervals, burstiness, etc., as it has been reported in other studies in non-human primates and rodents. These analyses could be extended to show if units with different properties also differ in their preferred phase-locking frequency, or phase. It would be very interesting if these analyses reveal the existence of heterogeneous cellular populations with different relation to hippocampal theta, even if the single-unit isolation quality is limited due to the low density recordings. In relation to this, authors should also plot unit auto-correlograms. ACGs can be computed for all the spikes, but also only for the strongly phase-locked spikes, to show if, at least during periods of strong oscillatory activity, some units show rhythmicity.

    5. To better interpret the results in Figure 4, it would be important to know if the recording sites in both hippocampi were from the same sub-region and similar location along the longitudinal hippocampal axis in each subject and if the degree of synchrony between the LFP in both hemispheres. Coherence or phase-locking between LFPs across hemispheres should be computed and also power spectrum for both of them shown.

    6. In Figure 4C-D it seems that phase-locking strength across hemispheres was not correlated but preferred frequency was. This should be quantified and mentioned in text before moving to the correlation in Figure 4E.

    7. The analysis in Figure 5D should be complemented by also checking the LFP-LFP phase-locking between the local region and the hippocampus. Were periods of high LFP power correlation also reflect enhanced phase-phase coupling? Were the structures also more phase-synchronous during periods of stronger spike-LFP coupling? These analyses could provide a more direct support for the interpretation of the authors in line with the CTC hypothesis.

    8. Was there any relation of the "strongly phase-locked" periods with global variables reflecting brain state (e.g. drowsiness versus attention to the task, etc.) or with the firing dynamics of the units (instantaneous firing rate or inter-spike intervals)?

  2. eLife

    ###Reviewer #2:

    In this study, Schonhaut et al., describe the phase locking statistics of cortical and subcortical neurons with respect to hippocampal local field potential (LFP) recorded in 18 epilepsy patients undergoing seizure monitoring. Nearly 30% of extrahippocampal neurons showed phase locking to some bandpassed hippocampal signal. Amygdalar and entorhinal neurons were more likely to be phase locked, as compared to neurons recorded in other neocortical sites. Most neurons showed the strongest phase locking to hippocampal theta (2-8 Hz), though neocortical and amygdalar neurons tended to phase lock to lower theta bands. Spikes that were phase locked to hippocampal rhythms occurred during local LFP-states that showed moderate correlations with the spectral patterns observed in the hippocampus. These data are interpreted within the broader "communication through coherence" hypothesis.

    Large N, multi-region, single unit studies from humans are rare and the kind of mesoscopic descriptive analyses provided here serve as an important bridge between the large rodent literature on hippocampal physiology and human physiology and cognition. That said, there are some weaknesses in the analyses that could be addressed. Also, a deeper discussion of the biological origin of human theta is merited in the discussion to address alternate explanations - beyond communication through coherence - of the data.

    A similar statistical mistake was made several times. The author's logic goes like this: find the argmax in one sample, take the argument that generated that max, and use that to sample in another condition, and report that the max is higher in the first condition than the second. For example, on pg. 6 "This is difficult to reconcile with our results, in which 248/362 neurons (68.5%) phase-locked more strongly to hippocampal LFPs than to locally-recorded LFPs at their preferred hippocampal phase-locking frequency." The same flaw can be seen in Figure 5, where the spikes are sub-sampled to occur during strong phase locking in one condition, thus almost guaranteeing high power in the frequency bands that generated that strong phase locking (which was observed). This is a case in which cross-validating the data may be useful. The authors could take a subset of the hippocampal data to define the preferred frequency, and then test phase locking on the held out data from the hippocampus and cortex.

    The relationship between power and phase locking is not fully controlled in this paper. The phase seems to be calculated irrespective of whether there is any instantaneous power at that frequency band, introducing noise. This will bias away from finding significant phase locking to frequency bands that occur transiently. Therefore, I recommend defining some threshold of the existence of the spectral signal prior to using that signal to calculate phase.

    A related point has to do with the nature of the theta rhythm in the human. There has been considerable controversy over the years as to whether this is a comparable signal to that studied in the rodent. Based on the citations in this manuscript, and the nomenclature of the spectral band, the authors seek to make explicit the commonality of the underlying physiology, or function. Rodent theta is a sustained rhythm, while primate theta seems to come in bouts, perhaps even related to sampling statistics, such as saccades, leading to the suggestion that the apparent theta may be better thought of as semi-rhythmic evoked responses. How long were the bouts of high theta power? Was eye movement tracked? If so it would be important to relate the signal to eye movements. If the low frequency signal is phase locked to eye movement and potentially reflects semi-rhythmic information arriving to (from?) the hippocampus, then a stronger case could be made that hidden "third parties" synchronize the apparent communication through coherence observed here, and in fact there may be no communication at all.

    The authors dedicate much of their discussion to relating the current result to the communication through coherence analyses. Oddly, LFP coherence was never addressed. A strong prediction of the current framing would be that: when coherence is high, phase locking should be high, and higher than other moments when power in either region is high but coherence is not observed. The authors should directly measure how phase locking is modulated by coherence.

    The authors also lump together biological entities that should have different phase locking behaviors. The amygdala is not a monolithic region, does phase locking differ by nucleus? Also, do fast spiking inhibitory cells differ from excitatory cells? The authors should relate their phase locking measure to mean firing rate to show that it is insensitive to lower level cell statistics. This is important since the conclusions of the study would be quite different if neurons in the entorhinal cortex had high rates which artifactually drove up phase locking values.

  3. eLife

    ###Reviewer #1:

    Hippocampal theta oscillations are among the most prominent rhythms in the mammalian brain. Extensive research in rodents has shown that neurons not only within the hippocampus but in widespread cortical areas can be phase-locked to hippocampal theta. Such cross-regional communication within theta frames has been postulated to be the foundation of many hippocampal operations. While previous studies in humans have documented the relationship between LFP theta and spiking in the hippocampus, coupling between hippocampal LFP and more remote cortical areas have not been demonstrated in human subjects. This is the topic of the present work. The authors show that spikes of single (and mostly multi-unit) neurons in multiple cortical regions both in the same and opposite hemispheres are phase locked to transient occurrence of hippocampal theta LFP in the 2-6 Hz range. However, phase-locking is stronger in structures known to be part of the 'limbic system', such as the amygdala and entorhinal cortex. Theta phase locking was stronger to hippocampal than to local LFP and the magnitude of spike phase locking increased when the power of theta increased, associated with increased high frequency power. The results are straightforward and the analysis methods are reliable. The novel information is limited but informative and documents a missing aspect of theta communication in the human brain.

    Comments:

    1. Given the simple message, the text is a bit long with many repetitions and loose ends. This applies to both Introduction and Discussion. Potential implications to learning, etc are interesting but the findings do not provide additional clues, thus those aspects of the discussion are mainly distractions. Instead, perhaps the authors would like to discuss potential mechanisms of remote unit entrainment. They are talking about multi-synaptic pathways but these are unlikely to be a valid conduit. Instead, the septum, entorhinal cortex or retrosplenial cortex, with their widespread projections, may be responsible for coordinating both hippocampal and neocortical areas.

    2. Arguably, the weakest part of the manuscript is the lack of hippocampal neurons. The authors refer to their own previous papers, but in a story which compares hippocampal theta oscillations with remote unit activity, it is strange that the magnitude of theta phase-locking to local hippocampal neurons is not available for comparison.

    3. How was the hippocampal LFP reference site chosen and did it vary substantially from subject to subject? Anterior or posterior locations?

    4. The authors list 1233 single neurons but in the discussion they make it clear that most of them were multiple neurons. This should be emphasized up front and may be used as an excuse why the authors did not attempt to separate pyramidal cells from interneurons (interneurons have a much higher propensity to be entrained by projected rhythms).

    5. Given that units were mixed, a logical extension would be to examine how hippocampal theta phase modulates high gamma in neocortical areas. This could potentially yield a much larger data base, targeting the same question.

    6. In the Discussion, the authors suggest that cross-regional theta phase coupling could be related to learning and other cognitive performance. However, spike-LFP coupling and coherence is confounded by LFP power increase and the authors cite Herweg et al., 2020 which did not find a relationship between theta power and memory performance. Is it then not logical to assume that cross-regional coupling may also not be related to memory?

    7. Line 36. "Long-term potentiation and long-term depression in the rodent hippocampus are also theta phase-dependent (Hyman et al., 2003)." Pavlides et al. (Brain Res 1988) or Huerta and Lisman (Neuron 1995) are perhaps more relevant references here.

    8. Line 82: "significant neocortical and contralateral phase-locking suggests". This is a strange phase. Perhaps significant phase locking of neurons in the neocortex in both hemispheres or similar would be a better formulation.

  4. eLife

    ##Preprint Review

    This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 2 of the manuscript.

    ###Summary:

    This is a very intriguing paper showing how hippocampal local field potentials couple with the activity of other cortical regions. This mechanism has been and continues to be extensively studied in other mammals, and thus its existence and relevance in humans is exciting.

  5. Version published to 10.1101/2020.06.30.180174 on bioRxiv