Properties and Hemispheric Differences of Theta Oscillations in the Human Hippocampus

Abstract

The left and right primate hippocampi (LH and RH) are thought to support distinct functions, but little is known about differences between the hemispheres at the neuronal level. We recorded single-neuron and local field potentials from the human hippocampus in epilepsy patients implanted with depth electrodes. We detected theta-frequency bouts of oscillatory activity while patients performed a visual recognition memory task. Theta appeared in bouts of 3.16 cycles, with sawtooth shaped oscillations that had a prolonged downswing period. Outside the seizure onset zone, the average frequency of theta bouts was higher in the RH compared to the LH (6.0 vs. 5.3 Hz) with only the RH exhibiting a bimodal distribution with centers at 3.7 Hz and 6.6 Hz. LH theta bouts had lower amplitudes and a higher prevalence compared to the RH (26% vs. 21% of total time). Additionally the RH contained a population of thin spiking visually tuned neurons that were not present in the LH. This data shows that human theta appears in short oscillatory bouts whose properties vary between hemispheres, thereby revealing neurophysiological properties of the hippocampus that differ between the hemispheres.

Introduction

In 1800, Marc Dax, a French physician, encountered an infantryman who had suffered a saber wound through his left Parietal bone. Dax noted that the soldier “suffered a major impairment in his memory for words, while his memory for things was preserved in all its integrity” (Joynt & Benton, 1964). Dax’s observation of material specific deficits associated with lateralized damage was the first word in what has been a fiery multi-century debate on the importance of left-right lateralization in the primate cerebrum. Though initial theories of lateralization centered on notions of a “Dominant” left and “Non- dominant” right cerebral hemisphere, lesion studies lateralized cognitive motifs to the two hemispheres of one brain region in particular, the Hippocampus (Joynt & Benton, 1964; Harrington, 1989).

Research from a host of modalities indicates a primary role for the LH in semantic memory, and for the RH in memory of non-semantic entities such as spatial coordinates, acoustic sequences, difficult to describe images, or according to some studies even low level perceptual processes ( Corsi, 1973; Coleshill et al 2004, Jones-Gotman & Milner, 1978, Watanabe et al., 2008). There is also significant anatomical evidence for hippocampal lateralization. Two white matter tracts connect the RH and LH: the dorsal and lateral Hippocampal Commissures (DHC and VHC). In rodents, the VHC originates along the entire long axis of the Hippocampus, providing a strong link between the two hippocampi (Gloor et al., 1993; Jordan 2020) However, in humans, tractography, stimulation, and dissection studies indicate that this direct pathway is non-existent ( Gloor et al., 1993; Wei et al., 2017). In contrast, the DHC is equally and fully preserved across all mammals. But rather than directly connecting the RH and LH, in primates this pathway originates in the parahippocampal gyrus and terminates in the contralateral hippocampal lobe ( Gloor et al., 1993). These dual lines of evidence indicate that the two hippocampi might function relatively independently in humans.

One of the most prominent electrophysiological patterns of activity in the rodent hippocampus is a ~3–8Hz frequency rhythm known as the theta oscillation (Buzsáki, 2002; Kahana et al., 2001). During periods of movement, “type 1” theta oscillations in the 7–12 Hz range appear continuously and consist of many cycles (Kramis et al., 1975). During periods of immobility, intermittent bouts of a lower frequency 4–7Hz “type II” theta predominate in rodents (Kramis et al., 1975). The properties of theta in humans are, in comparison, poorly characterized (Halgren et al., 1978; Jacobs, 2014). However, recently, recordings in subjects undergoing monitoring for intractable epilepsy have started to reveal a number of insights into the properties of a human equivalent of type 2 theta by evaluating recordings in stationary subjects. This work has revealed that power changes in the theta band correlate with a variety of cognitive variables, such as associative memory formation and accuracy of verbal memory encoding (Greenberg et al., 2015; Lega et al., 2012). Some of the properties and/or relationships of theta to behavior vary along the long axis of the hippocampus (Goyal et al., 2020) or between hemispheres (Miller et al., 2018). Additionally, the extent and phase of theta to which simultaneously recorded single neurons coordinate their activity is indicative of the success of memory formation (Rutishauser et al., 2010), working memory content (Kamiński et al., 2020), and success of memory retrieval (Minxha et al., 2020).

While much work has investigated functional correlates of aspects of human theta, little work has been done on assessing the properties of human theta and comparing properties of theta between hemispheres. In particular, most existing studies apply detection methods such as MODAL or p-Bosc, which assume the presence of theta if power in the theta band is elevated relative to the background (Lega et al., 2012; Lin et al., 2017; Miller et al., 2018), either in a specific period of time or across the entirety of a relevant recording epoch ( Lin et al., 2017; Miller et al., 2018, Watrous et al, 2013). Though these studies have offered critical insight into human theta processes, assessed in this way, there is no guarantee that what is detected is an autocoherent oscillation. This is because power increases can be the result of brief transient events, non-oscillatory waveform features, or simply broadband increases in power (Jones, 2016). These purely power based methods thus cannot answer the question of whether theta in humans is oscillatory, and if so what the shape of the oscillatory waveform is – questions that require answers in order to craft biophysical models of theta generation, and to systematically compare the properties and function of theta across species (S. R. Cole & Voytek, 2019; Jones, 2016). In primates, including humans, theta seems to appear predominantly in short intermittent bouts (M. Aghajan et al., 2017; Stewart & Fox, 1991, Watrous et al, 2013). However, it remains unclear whether the theta that appears in such bouts is oscillatory or can be explained by transient band-limited increases in power that is not oscillatory.

Here, we use a recently described approach (S. Cole & Voytek, 2019) to detect and characterize oscillatory bouts of theta and to compare the properties of the detected bouts between hemispheres. N=33 subjects (17 female; n=48 recording sessions) undergoing monitoring for the localization of intractable epilepsy participated in our study, during which subjects performed a visual recognition memory task (Rutishauser et al., 2010). Only electrodes located in the hippocampus as confirmed by post-operative imaging were analyzed. The relative position of the electrode along the anterior-posterior axis was determined manually (Konrad et al., 2009). Single neurons were recorded using hybrid depth electrodes and spike sorted as published previously (Rutishauser et al., 2010). All the LFP data analyzed in this paper was recorded from microwires. For a bout of theta to be classified as oscillatory, we used the method of Cole and Voytek (S. Cole & Voytek, 2019) with parameters of 60% similarity in the periods between cycles, a 60% similarity in the amplitude between cycles, a signal that was continuously at least 60% monotonic on a per cycle basis, and a minimum of 2 cycles for all detected bouts (see methods). We excluded channels or trials with interictal epileptiform discharges (IEDs) to assure that detected theta bouts were not caused by IEDs (see methods). Additionally, we only included a microwire if the power in the theta band exceeded that expected by the 1/f spectrum. If a putative oscillatory bout was detected, we quantified its auto-coherence using the circular variance of the phase portrait [CVPP] (Burns et al., 2010). We only included detected oscillations that fell below the 0.3 variance metric that prior work indicates as a threshold for a true oscillation (Burns et al., 2010).

We detected in total 42,899 theta bouts from 161 microwires (across 28 patients) that satisfied above criteria (Fig. 1b–d shows examples; Fig. S1 depicts the number of usable electrodes in each participant). Bouts contained on average 3.16±1.71 cycles and were on average 0.690±0.61 sec long. The majority (66.9%) of the isolated bouts were highly autocoherent (Fig. 1g, <0.3). Visual inspection indicated that the detected bouts were often noticeably different from the background both in terms of voltage (Fig. 1b–d, middle) and power (Fig. 1c–d, top; and Fig. S2). However, there were a number of exceptions, with some detected bouts not corresponding to a continuous increase in power at the same time of the bout (Fig. 1b shows an example). Note that, on average, all included channels had elevated theta power (Fig. S2).

Figure 1: Properties of Human Hippocampal Theta Bouts.

(a) Location of recording electrodes, shown in Montreal Neurological Institute’s MNI152 space. (b-d) Example detected theta bouts. For each, shown is from top to bottom: (i) the power spectra from one second before to one second after the on/offset of the detected bout. Black lines indicate the frequency range of the detected bout. Power is in units of z-score. (ii) Single detected bout (purple), pre/and post period (blue), and band-pass filtered version in the detected frequency (dashed). Blue line shows area of zoom-in shown in below. Amplitude is z-scored. (iii) left: Zoom-in, showing every detected bout, bandpass filtered within a 0.5 Hz window around the mean frequency of the detected bouts (green). The same example bout is shown in purple. Black trace is the mean. Amplitude is z-scored. Right: Phase portrait of the detected bouts. (b) is in the LH in the 2.2–3.2 Hz range. (c) is in the LH in the 2.8–4.8 Hz range. (d) is in the right hippocampus in the 5.1–6.1 Hz range. (e) Rise-decay asymmetry of detected theta Bouts (top). Bouts were significantly skewed towards the falling phase (<0.5). Bottom shows average theta bouts with frequencies 4.5–5Hz that fall within the ‘Decay Skewed’, ‘Symmetrical’, or ‘Rise Skewed’ categories. (f) Peak-trough asymmetry of detected theta bouts (top). Bouts were significantly skewed towards the peak period (>0.5). Bottom shows average theta bout waveform with frequencies 4.5–5Hz (z-scored) that falls within the ‘Peak Skewed’, ‘Symmetrical’, or ‘Trough Skewed’ categories (as indicated by color). (g) Circular variance of the phase portraits (CVPP) for every putative bout. Black line depicts the cutoff point below which detected theta bouts are considered true oscillations. The three colored stars represent the variance of the three exemplar theta rhythms depicted in b-d. (h-j) Comparison of bout properties between electrodes located within and outside the seizure onset zone (SOZ). All time is relative to onset of the detected bout.

We evaluated the morphology of the LFP waveforms during detected theta bouts using the measures of trough to peak asymmetry, the proportion of an oscillation’s period spent in the trough vs the peak, and the rise decay asymmetry (S. Cole & Voytek, 2019). A perfectly symmetrical oscillation (i.e. a sine wave) would have a rise-decay symmetry and peak-trough symmetry of 0.5. Lower values indicate a skew towards the downswing phase and higher values indicate a skew towards the upswing phase. 67.5% of oscillations had rise-decay symmetries that were at least five percent different from 0.5, with more oscillations skewed towards the downswing phase (Fig. 1e; 25004/ 42899; 58.3%; had values below 0.5 while 8837/21329; 41.6%; had values above 0.5 sig; chi square test of proportions, p<1e⁻⁵, ttest compared to 0.5, p<1e⁻⁵). Overall, the mean value was 0.479.

Trough to peak asymmetry, on the other hand, was far less apparent (Fig. 1f). Bouts with trough to peak asymmetry values lower than 0.5 were skewed towards the trough, while bouts with a value higher than 0.5 were skewed towards the peak. Though there was a statistically significant trend for bouts to have longer peak than trough phases, the overall effect size was small with only 4.7% of detected bouts having a trough to peak asymmetry that was at least 5% different from 0.5 (Fig. 1f; 19457/42899; 45.4%; had values <0.5, while 22550/42899; 52.6%; had values >0.5 sig; chi square test of proportions, p<1e⁻⁵, ttest compared to 0.5, p p<1e⁻⁵). Overall, the mean value was 0.502. Both of these results also held true across channels (Fig. S3). There were no differences in rise-decay symmetry between Hippocampal hemispheres (Fig. S4).

Theta bout properties differed between whether the electrode was in a hippocampus that was inside vs. outside the SOZ. Outside the SOZ, theta bout frequency was unimodally distributed with a clear peak at 4 Hz, whereas the distribution was uniform in the 2–9Hz range inside the SOZ (Fig. 1h; 2-sided t-test frequency inside vs outside the SOZ p<1e⁻⁵). This difference was not driven by individual sessions or electrodes (Fig. S5, two-sided t-test frequency inside vs. outside the SOZ averaged across electrodes p<1e⁻⁵, averaged across sessions p<.01). The cycle length and number of cycles within a bout, z scored on the basis of frequency, were not significantly different between inside vs. outside the SOZ (Fig. 1i–j; p=0.10, p=0.25, respectively; see methods).

We next investigated whether the properties of detected theta bouts differed between hemispheres. Given the frequency distribution of the detected bouts (Figure 2a), we restricted the remainder of our analyses to oscillations in the frequency range of 3.5–8 Hz. There were several notable differences. First, the oscillatory frequency was significantly higher in the right hippocampus (6.00 ± 2.193 Hz vs. 5.30 ± 1.97 Hz; two-tailed t-test, p < 1e−5). Aggregated across all subjects, this difference was due to a second higher frequency mode around 6.6 Hz that was only visible in the RH (Figure 2a). A few select participants did not drive this finding nor was it sensitive to the location of the SOZ (Fig. 2b). Also, this finding was not due to differences in anterior-posterior positioning of the electrodes as confirmed by a General Linear Mixed Effects Model (GLME) with fixed effects referencing scheme, electrode placement within/outside SOZ, anterior-posterior position, laterality, patient session ID as a random effect, laterality and anterior-posterior position as interaction effects, and frequency as the predicted variable (see methods). Significance was assessed by calculating the tstat for each coefficient (Estimate/Standard Error). Significant coefficients were laterality (p<1e⁻⁵), anterior-posterior position (p<1e⁻⁵), and the interaction between anterior-posterior position and laterality (p<1e⁻⁵). Moreover, using model comparisons, we find that the model as a whole was significant p<1e⁻⁵ compared to a null model with all fixed factors removed. Also, our full model explained significantly more variance compared to a model with the laterality term removed (p<1e⁻⁵).

Figure 2: Right-Left Lateralized Theta Oscillation Differences in the Human Hippocampus.

(a) The frequency of detected theta bouts was higher in the right compared to the left hippocampus. (b) Same as (a), but for every individual session with at least 50 detected bouts and usable bouts in both hemispheres. Sessions are grouped by location of Seizure Onset Zone (SOZ; Bonferroni corrected p <0.001). 7/9 sessions had higher frequency detected theta in the RH. Bi-MTL indicates bilateral medial temporal lobe. (c) The amplitude of detected theta bouts was higher in the right hippocampus. (d) Total Percentage of the LFP occupied by a detected theta bouts is higher in the left hippocampus (sig; two-tailed ttest p<5e⁻³). (e) Trough to peak width for visually responsive cells in the RH and LH. (f) Average waveform of visually responsive neurons in the LH and RH. (g) Proportion of visually responsive cells with thin-and wide waveforms. Across all panels. * p<0.05, ** p<0.005, *** p < 1e-5.

Second, we examined differences in the amplitude of the detected bouts. Because amplitude and frequency were, as expected due to 1/f, negatively correlated (r=−0.215, p<1e⁻⁵), we first z scored the amplitude of each bout in frequency steps of 1 Hz on a per participant basis (see supplementary methods). We also removed bouts that had a peak amplitude >4 standard deviations from the mean. We then fit the same GLME model we used to examine frequency above, but now with z scored amplitude as the predictor. We found a significant effect of laterality (p<1e⁻⁵) and SOZ (p<1e⁻⁵). The model was significant over all (p<1e⁻⁵) and a model that included laterality as a coefficient explained significantly more variance then one that did not (p<1e⁻⁵). Overall, this analysis reveals that the amplitude of theta bouts was significantly larger in the RH compared to the LH (Fig. 2c).

Third, a significantly larger proportion of time was occupied by detected theta bouts in the LH (Fig. 2d, 26.3 % vs. 21.4%; ttest p<5e⁻³). This shows that the prevalence of theta in the LH was higher.

The RH has been classically assumed to be more involved in visual memory than the LH, so we next evaluated whether the proportion of neurons responsive to the onset of visual stimuli (see supplementary methods) varied between hemispheres. We evaluated this hypothesis separately for neurons with narrow and wide extracellular waveforms. Visually response neurons with narrow waveforms were significantly more common in the RH vs. the LH (Fig. 2g-i; 22/37 vs. 5/20; 59.5% vs. 25.0%% sig; chi square test of proportions, p<.02). In contrast, there was no significant difference in the prevalence of wide-waveform visually responsive cells (Fig 2i; 18/64 vs. 15/59; 28.1% vs. 25.4%, p=.74). There was no significant hemispheric difference in the relative prevalence of thin spiking neurons overall (37/107 vs. vs 20/82; 34.6% vs. 24.9%; chi square test of proportions, p=0.13).

In this work we evaluate the qualities and interhemispheric differences of autocoherent theta rhythms in humans, which is different from the majority of earlier work that focuses on changes in theta power alone. Replicating earlier work (Goyal et al., 2020), we found that theta rhythms had higher frequencies closer to the Posterior tail (an effect that held for both the RH and LH; Fig. 2a). However, this effect did not explain our laterality differences (Fig. 2a). Moreover, in line with work showing selective deficits in visual processing following RH excision or damage, we find that a distinct population of visually tuned thin spiking cells existed in the RH but not the LH (Coleshill et al., 2004; Jones-Gotman & Milner, 1978). A possible explanation for this effect is that thin spiking (putative interneurons) were preferentially entrained by IEDs and that this was particularly likely to occur in the RH (Lee et al., 2021.; Reed et al., 2020).

The salient differences in the properties of theta bouts between the RH and LH suggest that, in humans, the mechanisms of theta generation may be different in the two hemispheres. In rodents, a striking inter-hemispheric Hippocampal distinction is that postsynaptic spines of CA1 pyramidal cells targeted by projections from the right CA3 have a larger “mushroom shaped “ morphology and a higher expression of GluA1 receptors with a lower relative expression of GluN2B receptors compared to the smaller postsynaptic spines in contacts with projections from the left CA3 ( Shinohara et al., 2008). These differences are true regardless of the laterality of the postsynaptic contact. The lack of a functional VHC in humans suggests that such postsynaptic receptor asymmetry could lead to radically different theta generators between the RH and LH (Buzsáki, 2002).

To the best of our knowledge our paper is the first quantification of theta waveform asymmetry in humans. We observed that, as in rodents, human theta was skewed significantly towards the falling phase. Though it has long been appreciated that Theta is non-sinusoidal the significance and biophysical origin of this phenomenon remain elusive (Cole and Voytek 2017). Theta appears to become more ‘sawtooth-like’ as rats increase their movement speed and increased skewing towards the falling phase predicts successful memory formation (Cole and Voytek 2017). Given the recent advent of invasive recording techniques that are compatible with naturalistic movement in primates it will be fascinating to observe whether theta asymmetry is similarly important in humans (M. Aghajan et al., 2017).

Our observation that the SOZ had no apparent or consistent theta peak frequency is in line with previous studies that have described dysfunctional theta dynamics in epileptogenic regions (Amiri et al, 2019, Young et al 2018). We emphasize that this finding indicates that studies of human cognition should take into account whether an electrode is located within the SOZ even if IEDs are not apparent (Parvizi & Kastner, 2018).

A key result of our study is that theta was relatively rare – despite the patients actively engaged in a memory task, only 26.3% of time in the LH and 21.4% in the RH was occupied by a detected theta bout. This is different from the kind of continuous long stretches of theta seen in rodents during movement, suggesting that this is not Type 1 theta. If, as hypothesized, theta is critical for learning, why does it appear so infrequently during our task? There are several possibilities. First, it is possible that in humans, the neural code is based on phase coding that does not rely on oscillations, similar to bats (Eliav et al., 2018), but rather on phase-locking to non-autocoherent theta-band signals alone. Second, given that theta is a travelling wave, it may be that the intermittent presence of theta is a result of the bipolar recording technique (Zhang & Jacobs, 2015). Third, it may be that intermittent short bursts of theta are sufficient to transiently synchronize assemblies of cells and that the long continuous stretches of theta are specific to navigation rather than being necessary for the kind of learning and memory our task tests. As with primate studies evaluating hippocampal theta, we find that the rhythm is noticeably present but only in intermittent bursts, it seems likely that there are major species differences in both the circuit level participants that generate theta and the role the rhythm plays in generating and maintaining neural ensembles ( Jacobs, 2014; Stewart & Fox, 1991).