Trends in

OPEN ACCESS Neurosciences

Review

Gamma oscillations and episodic memory

Benjamin J. Griffiths1,* and Ole Jensen1

Enhanced gamma oscillatory activity (30–80 Hz) accompanies the successful Highlights
formation and retrieval of episodic memories. While this co-occurrence is well Gamma oscillations may coordinate pre-
documented, the mechanistic contributions of gamma oscillatory activity to and postsynaptic neuronal firing to en-
hance plasticity within the hippocampus.
episodic memory remain unclear. Here, we review how gamma oscillatory activity
may facilitate spike timing-dependent plasticity, neural communication, and Cross-regional gamma synchronisation
sequence encoding/retrieval, thereby ensuring the successful formation and/or may communicate sensory information
retrieval of an episodic memory. Based on the evidence reviewed, we propose to the hippocampus during memory for-
mation, and hippocampal representa-
that multiple, distinct forms of gamma oscillation can be found within the canonical
tions to the cortex during retrieval.
gamma band, each of which has a complementary role in the neural processes
listed above. Further exploration of these theories using causal manipulations Gamma oscillations nested within ongo-
may be key to elucidating the relevance of gamma oscillatory activity to episodic ing theta oscillations may encode and re-
call sequences of stimuli.
memory.
Multiple, distinct oscillations may exist
within the canonical gamma band
An established correlation between gamma oscillations and episodic memory (30–80 Hz), each with complemen-
When we talk of episodic memories (see Glossary), we mean long-term memories relating to tary roles in episodic memory.
personally experienced events anchored to a specific moment in time and space [1]. Although
these memories are, by definition, rich in detail and can last for decades, patterns of neural activity
lasting mere seconds will dictate whether these memories are formed or recalled [2,3]. Many
studies suggest that gamma oscillations (~30–80 Hz, although definitions can vary slightly
among researchers) have a key role in both the formation and retrieval of an episodic memory.
Supporting evidence comes from a range of species (including rodents [4–7], nonhuman
primates [8,9], and humans [10–12]), and a variety of empirical techniques (ranging from studies
of cell cultures in vitro [13,14] to behavioural responses in humans [15]). In our view, the extent of
this evidence provides firm support for a link between gamma oscillations and episodic memory,
and calls for a focus on understanding why this link exists. To address this question, we review
three distinct neural mechanisms that may link gamma oscillations to fundamental aspects of
episodic memory: (i) spike timing-dependent plasticity; (ii) neural communication; and
(iii) sequence encoding/retrieval, with the aim of elucidating how gamma oscillatory activity
supports episodic memory.

Gamma oscillations and spike timing-dependent plasticity

Our ability to form an episodic memory hinges upon long-term potentiation (LTP), a process
through which synaptic connections between two neurons are strengthened [16,17]. Gamma
oscillations have been proposed to play an important role in a type of LTP known as spike
timing-dependent plasticity (STDP), which depends on a precise temporal delay between the
firing of a presynaptic and a postsynaptic neuron. While there are numerous examples of
1
Centre for Human Brain Health,
gamma oscillatory activity enhancing STDP in vitro [13,18], in silico [19,20], and in vivo [21,22],
University of Birmingham, Birmingham,
the mechanistic explanation of this link is open to debate. Here, we discuss: (i) how STDP occurs; UK
(ii) how gamma oscillations may facilitate this process; and (iii) how the interaction between STDP
and gamma oscillations might result in the formation of complex, episodic memories.

STDP is thought to depend upon: (i) a presynaptic spike leading to the release of presynaptic gluta- *Correspondence:
mate, which promotes the opening of postsynaptic NMDA receptors; and (ii) the backpropagation of b.griffiths.1@bham.ac.uk (B.J. Griffiths).

832 Trends in Neurosciences, October 2023, Vol. 46, No. 10 https://doi.org/10.1016/j.tins.2023.07.003

© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Trends in Neurosciences
OPEN ACCESS

a postsynaptic spike leading to the unblocking of the Mg2+ block from the same postsynaptic NMDA Glossary
receptors [16]. Some have suggested that the comparatively slow binding of glutamate to the NMDA Associative binding: cornerstone of
receptor relative to the rapid removal of the Mg2+ block means that the presynaptic action potential episodic memory formation that
involves linking two previously unrelated
must precede the postsynaptic action potential by ~10–20 ms for STDP to occur (e.g., [23]). Indeed, concepts together.
in slices of rat hippocampus, presynaptic spikes that lead postsynaptic spikes by ~15 ms result in Cell assembly: collection of
synaptic strengthening, whereas presynaptic spikes that follow postsynaptic spikes by ~6 ms interconnected neurons that are tuned
to a particular stimulus.
lead to synaptic weakening (known as long-term depression; LTD) [24]. STDP effects have been
Episodic memory: long-term memory
reported across a range of species, including rodents (in vitro [24–26] and in vivo [27]), nonhuman relating to a unique experience,
primates (in vitro [28] and in vivo [29]), and humans (in vitro [30]). anchored to a single point in time and
space, that can be explicitly recalled and
re-experienced in vivid detail.
Although STDP depends upon correlated pre- and postsynaptic spiking, a solitary presynaptic
Gamma oscillation: fast, rhythmic
spike is unlikely to induce postsynaptic spiking [16]. Instead, convergent input is required. change in the activity of a collection of
Gamma oscillations may provide this convergent input in two ways. First, gamma oscillations neurons, often defined to be in the range
can synchronise the firing of multiple presynaptic neurons so that they exert a stronger of 30–80 Hz although definitions vary
across the literature.
depolarising effect on the target postsynaptic neuron than if they were to fire in isolation
Long-term potentiation (LTP): neural
[13,14]. Indeed, computational models show that the oscillation-driven synchrony of presynaptic phenomenon in which the strength of a
activity increases the likelihood of a postsynaptic spike [31,32] (see also [33]). Moreover, in vitro synaptic connection between two
studies show that synchronising multiple inputs to a postsynaptic neuron enhances the likelihood neurons grows following repeated
co-firing of the two neurons. LTP is
of LTP [13]. While an oscillation of any frequency could, in theory, synchronise neuronal firing, thought to be key in binding distinct cell
gamma oscillations are perhaps ideal because they provide a comparatively short window of assemblies together to form an episodic
excitability that ensures all neurons fire in near-perfect unison [34], while having oscillatory cycles memory.
that are long enough to ensure that neurons return to their resting potential before the next excit- Neural communication: process of
routing information through the brain.
atory part of the oscillation. In support of the former claim, LTP has been demonstrated to be Here, we use the term ‘neural
most effective when pre- and postsynaptic firing is coupled to a 50-Hz rhythm (relative to slower communication’ to refer to both local
rhythms) [13], although it remains to be seen how LTP is affected when pre- and postsynaptic (i.e., between local cells/assemblies) and
interareal (i.e., between regions)
firing is coupled to a frequency greater than 50 Hz.
communication.
Reinstatement: phenomenon in which
In addition to facilitating synchronised neuronal firing, gamma oscillations may also aid the post- patterns of neural activity observed
synaptic depolarisation necessary for STDP by inducing subthreshold oscillatory fluctuations in during memory formation reoccur during
memory retrieval. This reoccurrence is
postsynaptic membrane potential. For example, in vitro work has shown that pairing presynaptic
thought to allow an individual to vividly
spikes to the peak of a 40-Hz oscillation led to greater LTP than when spikes were paired to the re-experience the memory.
trough of the oscillation [35], purportedly because the change in potential at the oscillatory peak Sequence encoding/retrieval: act of
adds an additional drive for depolarising the postsynaptic neuron. This may explain why, in encoding/retrieving multiple, distinct
stimuli in the order that they were
humans and nonhuman primates, successful memory formation occurs when neuronal firing is
presented. Sequence encoding/retrieval
coupled to particular phases of the ongoing gamma oscillation [9,36]. Considering all the is critical to forming and recalling the
above, it appears plausible that gamma oscillations can facilitate LTP by increasing the likelihood temporal narrative of an episodic
of postsynaptic spiking. memory.
Spike-timing-dependent plasticity
(STDP): form of LTP that relies on the
Gamma oscillations may also provide a spike-timing delay between the pre- and postsynaptic precise firing of two neurons. Typically,
neurons that is optimal for STDP (~10–20 ms) [37]. However, little work has been conducted at STDP is thought to occur when the
the cellular level to demonstrate that observed links between gamma oscillatory activity and presynaptic neuron fires shortly (<20 ms)
before the postsynaptic neuron.
STDP are specifically due to gamma oscillations matching the optimal timing constraints of Theta–gamma coupling: oscillatory
STDP. This may be because the spiking delays necessary for STDP can vary across brain phenomenon in which patterns of
regions, cell types, and even between individual cells (Box 1). Consequently, a gamma oscillation gamma oscillatory activity fluctuate as a
function of theta oscillatory activity.
of a precise frequency cannot match the timing delay of every cell in a network. However, it
Often, this involves the amplitude of
remains possible that gamma oscillations match the average preferred delay of the network, gamma oscillations fluctuating as a
meaning that STDP-like phenomena could be reliable on a macroscopic (e.g., behavioural) function of theta oscillatory phase.
level. In line with this idea, humans are better able to learn pairings between stimuli that rhythmi-
cally fluctuate in intensity (at 37.5 Hz) when, during the initial pairing, the cue preceded the target
by ~7 ms (matching traditional STDP delays) relative to when the cue and target are presented

Trends in Neurosciences, October 2023, Vol. 46, No. 10 833

 Trends in Neurosciences
OPEN ACCESS

Box 1. Variations on the traditional STDP curve

The textbook STDP curve depicts pre-to-postsynaptic firing inducing LTP, post-to-presynaptic firing inducing LTD, and
the strength of the effects diminishing as the delay between the firing of the two neurons increases. Depictions of this curve
are based on NMDA receptor-dependent STDP in slices of the rat hippocampus (e.g., [24]), with the assumption that these
curves generalise to other cells, brain regions, and even species. However, this may not be the case (Figure I).

For example, GABAergic neurons fail to display LTP [24] (and can even display LTD [134]) following pre-to-postsynaptic
firing, suggesting that STDP curves vary based on cell type. In spiny stellate cells of the barrel cortex of rodents, pre-to-
postsynaptic firing also induces LTD [135], hinting that STDP curves vary based on both brain regions and cell types. In
human hippocampal tissue, LTP can be observed regardless of whether the presynaptic spike precedes or follows the
postsynaptic spike, with LTD only being observed when the postsynaptic spike precedes the presynaptic spike by
~100 ms [30], suggesting variation across species. Even in cells of the same type and region, STDP curves can vary simply
because of variation in the distance between the soma and the location where input arrives on the dendrite: short dis-
tances produce the stereotypical curves, while greater distances invert the curve [136]. Lastly, STDP curves of a single
neuronal pair can vary based on the time difference between glutamate arriving at the NMDA receptor and the occurrence
of a depolarising potential [137]. In short, STDP curves can take many forms, and one should exercise caution when
generalising STDP curves to other cells, brain regions, and species.

This conclusion is also pertinent to the discussion of STDP and gamma oscillations: given that no two neurons are iden-
tical, it appears unlikely that there is a single oscillatory frequency that is optimal for enhancing STDP in all neurons. Indeed,
variations in gamma oscillatory frequency across regions/species may well align with the differences in STDP curves that
neurons in these regions/species exhibit.

(A) Rodent hippocampus (B) Human hippocampus (C) Dendrite position

Transition Proximal Distal
LTD window LTP LTD LTP
Δ Plasticity

+ +
Δ Synaptic strength

Δ Synaptic strength

Pre. leads post.

Post. leads pre.

0
(D) Rodent cell type
0 Spiny stellate Pyramidal
Δ Plasticity

- -
-50 0 +50 -120 0 +30
Post-synaptic spike delay (ms) Post-synaptic spike delay (ms) Pre. leads post.

Trends in Neurosciences

Figure I. Variations on the traditional spike timing-dependent plasticity (STDP) curve. (A) Depiction of perhaps
the most common STDP curve, in which long-term potentiation (LTP) occurs when the presynaptic spike precedes the
postsynaptic spike, and long-term depression (LTD) occurs when the presynaptic spike follows the postsynaptic spike.
This curve was derived from observations made on slices of rat hippocampus [24]. (B) Depiction of an alternative STDP
curve, observed in slices of human hippocampal tissue [30]. (C) Depiction of how STDP curves vary as a function of
dendrite input location. When input is proximal to the postsynaptic soma, typical STDP patterns are observed, but
when input is distal, the effect inverts [136]. (D) Depiction of how STDP effects differ as a function of cell type in the
rodent barrel cortex. Pyramidal cells show the typical LTP effect when presynaptic spikes precede postsynaptic spikes,
but LTD is observed in spiny stellate cells following the same pattern of firing [135].

simultaneously during encoding [15]. Based on this finding, perhaps it is not a matter of ensuring
that synapses of every neuron pair undergo STDP, but rather that sufficient pairs undergo STDP
to ensure that a memory can be reliably formed.

While the explanations explored in the preceding text suggest gamma oscillatory activity
enhances STDP, they introduce an issue of firing order ambiguity. Specifically, if two neurons
repeatedly and reliably fire as a function of gamma oscillatory phase, it becomes unclear which
neuron leads which and, consequently, whether LTP or LTD will occur. In these instances,
evidence suggests that LTP and LTD do not sum linearly [13,38,39], with LTP possibly

834 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

supplanting LTD [13]. This suggests that the relevant timing of pre- and postsynaptic spikes
becomes irrelevant so long as both neurons fire regularly and in quick succession. Consequently,
ambiguity in firing order does not undermine the idea that gamma oscillations can enhance STDP.

There is also the question of how gamma-facilitated plasticity scales up to the formation of fully
fledged episodic memories. Unfortunately, linking gamma oscillations to both STDP and
behavioural expressions of episodic memory in a single experiment is a troublesome endeavour:
STDP is most easily observed in small cultures of cells and slice preparations, whereas behav-
ioural expressions of episodic memory can only be observed in active individuals. That said,
progress has been made. For example, in rodents, an increase in gamma oscillatory activity
correlated with both an enhanced fear response to an auditory tone (i.e., a learned response)
and a change in A1 receptive fields (a proxy for plasticity) [7]. Indirect links also exist in humans.
For example, successful memory formation occurs when two conditions are met: (i) when the
latency of firing between two neurons is ~20 ms (approximating the delay required for STDP);
and (ii) when the co-firing neurons couple to an ongoing gamma oscillation [36]. This suggests
that episodic memory formation is most probable when STDP-like delays in neuronal firing are
coupled to an ongoing gamma oscillation. Together, these studies support the idea that gamma-
facilitated STDP can lead to the formation of complex and highly detailed episodic memories.

In sum, STDP is intimately tied to gamma oscillatory activity, although the mechanistic explanation
of this link is open to debate. Moreover, it remains an open question whether gamma oscillations
are necessary for STDP to occur. These questions may benefit from causal interventions that
quantify the relevance of gamma oscillatory activity to STDP.

Gamma oscillations and neural communication

Neural communication refers to the process of relaying information across the brain, be that
between local cell assemblies or across large portions of the cortex. Neural communication is
relevant to almost all aspects of cognition, from perception to action, but we focus here on its
relevance to episodic memory. Effective neural communication ensures that, during memory
formation, incoming information in sensory cortices activates the relevant cell assemblies in the
hippocampus to ensure associative binding. Similarly, during retrieval, neural communication
ensures that reactivated hippocampal cell assemblies induce neocortical activity to allow for the
reinstatement of an episode. Here, we review key theories that link gamma oscillations to neural
communication and assess whether they may, as a result, explain the link between gamma
oscillatory activity and episodic memory.

A prominent theory tying gamma activity to information exchange is ‘communication through

coherence’ [40,41]. Communication through coherence proposes that information from one
neural population can be relayed to another population when: (i) gamma oscillations in the two
regions synchronise; and (ii) input from the ‘sender’ cell assembly arrives at the ‘receiver’
assembly when the ‘receiver’ assembly is at its most excitable oscillatory phase (Figure 1A).
Similar to STDP, gamma oscillations are well suited for facilitating communication because they
ensure tighter synchrony between neurons of the ‘sender’ cell assembly compared with slower
oscillations [34], while also ensuring that neurons can return to their resting potential before the
next excitatory phase of the oscillation. This may explain numerous findings linking gamma-
band coherence to episodic memory. For example, when learning object–reward associations,
primates display enhanced gamma-band coherence between inferotemporal areas (which
represent the object) and prefrontal sites (which are thought to link the object to the reward) [8]
(for similar effects in humans, see [12,42]). Moreover, when humans successfully recall a memory,
gamma-band coherence between the hippocampus and lateral temporal cortex predicts the

Trends in Neurosciences, October 2023, Vol. 46, No. 10 835

 Trends in Neurosciences
OPEN ACCESS

(A) (B) (C)

A B
‘Fast’ gamma I Feedforward I
A CA1 60-100Hz
II
Gamma (60-100Hz)
II
III III
B MEC IV IV
CA3 V V
‘Slow’ gamma
C 25-50Hz VI VI
Feedback
Beta (13-30Hz)

Trends in Neurosciences

Figure 1. Theories of gamma-based neural communication. (A) Depiction of ‘communication through coherence’
[41]. Two brain regions can communicate when: (i) gamma oscillations in the two regions become coherent; and (ii) the
transmission delay from sender to receiver ensures that the sender spike arrives at the moment when the receiver is most
excitable. (B) Depiction of how two hippocampal gamma oscillations may coexist [45,46]. A ‘fast’ gamma oscillation,
generated by the medial entorhinal cortex, can entrain cells in CA1. This allows to-be-encoded information to flow into the
hippocampus. A ‘slow’ gamma oscillation, generated by CA3, can also entrain cells in CA1. This allows reactivated
memory traces to be reinstated in CA1 (C) Depiction of how distinct oscillations allow information to be fed forward and
backward [55]. Bottom-up, sensory information is passed forward by gamma oscillatory coherence present in superficial
cortical layers (i.e., layer 2/3). Top-down influence is exerted via beta oscillatory coherence present in deeper cortical layers
(i.e., layer 5/6).

degree of reinstatement observed in the lateral temporal cortex [43], with such retrieval-related
coherence being absent in those with autobiographical amnesia [44]. Taken together, these
empirical reports demonstrate a correlative link between gamma coherence and the ability to
relay to-be-encoded and to-be-recalled information across the brain.

If gamma-band connectivity is involved in relaying both to-be-encoded and recalled information,

how can interference between the two streams of information be avoided? One suggestion,
based on observations from the rodent hippocampus, is that distinct gamma oscillations support
the two processes: a fast gamma oscillation (~60–100 Hz; slightly above the canonical gamma
band) facilitates encoding by allowing information to flow from the entorhinal cortex to the
hippocampus, while a slow gamma oscillation (~25–50 Hz; slightly beneath the canonical
gamma band) supports retrieval by allowing reinstated traces to propagate from CA3 to CA1
[45,46] (Figure 1B). These ideas have received empirical support from studies of rodents
(e.g., [4,5,47,48]) and humans [10,49,50]. While contradictions exist in the rodent literature
(e.g., [51,52]), these are often seen in studies in which encoding and retrieval overlap (e.g., at a
decision point in a maze: a rat may either be retrieving a past trajectory or encoding the current
trajectory for future reference), making it difficult to isolate encoding-/retrieval-specific neural
processes. However, studies in humans sidestep this issue by explicitly directing participants
to either encode or retrieve and then contrasting the resulting neural correlates. In these
instances, the typical pattern of ‘fast and ‘slow’ gamma activity is observed, with the former
favouring encoding and the latter favouring retrieval [10]. That said, in both rodent and human
studies, measurements of ‘slow’ gamma oscillations may be susceptible to distortion by theta
harmonics [53], meaning open questions remain about what can be attributed to ‘slow’
gamma oscillations and what is attributable to theta. Taken together, ‘fast’ and ‘slow’ gamma
oscillations may have separable roles in episodic memory, with the former biased toward
encoding and the latter toward retrieval.

The spectral separation of encoding and retrieval may also arise in the cortex [54–57], with
gamma oscillations feeding information forward to associative hubs (e.g., the hippocampus) for
memory formation [58], while beta oscillations (15–30 Hz) exert top-down inhibitory influence
over the same pathways to restrict activity to regions relevant to the processing of reactivated
memory traces [59] (Figure 1C). While these frequency-specific cortical routes have been
observed in low-level perceptual processes (e.g., [55–57]), these concepts are more difficult to

836 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

reconcile with episodic memory. For example, as described in the preceding text, gamma-band
coherence between the hippocampus and lateral temporal cortex supports memory retrieval
[43], suggesting that gamma oscillations are not the exclusively feedforward phenomenon that
the cortical routing hypothesis would suggest (see also [60,61]). However, this concern could
be addressed by hypothesising that hippocampal slow gamma oscillations and top-down
cortical beta oscillations reflect the same rhythm. Oscillations tend to propagate from areas
with faster intrinsic rhythms to areas with slower intrinsic rhythms [62]; thus, speculatively, a prop-
agating hippocampal slow ‘gamma’ oscillation could slow to a beta-like rhythm in cortical areas
as the reactivated memory trace propagates across the brain. Exploring the interactions between
hippocampal ‘slow’ gamma and cortical beta oscillations using simultaneous recordings would
allow for empirical testing of this idea.

While the theories outlined in the preceding text propose that coherence causes communication,
some have questioned how general these causal mechanisms may be [63–65] (Box 2). For
example, in macaques, coherence can be a product of communication if a ‘sender’ region
projects to both itself and a ‘receiver’ region in a rhythmic fashion [64]. Moreover, computational
modelling shows that two networks can communicate with one another without coherence,
and such communication can, in fact, produce coherence [63]. These findings suggest that
coherence is not necessary for communication but, nonetheless, could reflect instances when
successful communication occurred.

Others have questioned how gamma-band coherence can support communication over
distances of >1 cm, given factors such as variable axonal conduction delays [66]. Indeed, com-
putational models suggest that this ‘long-range’ communication is better supported by slower

Box 2. The pros and cons of communication through coherence

Communication through coherence (CTC) has been an influential theory of neural communication over the past 20 years,
but the generality of the mechanism has recently been questioned. Here, we consider some key critiques and enduring
strengths of CTC.

Some critics of CTC propose that coherence may be a consequence, rather than the cause, of communication. Indeed,
rhythmic spiking in the sender can produce postsynaptic potentials in both the sender and receiver area, which in turn pro-
duce gamma-band coherence in the local field potential between the two regions [64]. Computational models support this
idea, showing that coherence follows, rather than causes, communication [63]. Others have demonstrated that excitatory
neurons in the receiving region do not couple to the phase of the gamma oscillation despite the two regions becoming co-
herent [138], suggesting that coherence is not essential for neuronal communication. Lastly, the frequency and strength of
gamma oscillations depend on stimulus properties (e.g., contrast) [139,140], resulting in multiple frequencies of gamma
oscillation which may struggle to become coherent with one another. Altogether, these findings suggest that gamma os-
cillatory coherence may not be a prerequisite for neural communication.

That said, several aspects of CTC appear robust. For example, gamma coherence relates to performance in a range of
cognitive tasks, including attention (e.g., [141]) memory (e.g., [8,42,43]), and navigation (e.g., [142]), with suboptimal phase
delays between gamma oscillations in the sender and receiver regions impairing behavioural responses [143], all of which
support the idea that coherence with optimised phase delays aids communication and its behavioural consequences.
Moreover, while many investigations of CTC focus on the visual system of nonhuman primates, congruent findings can
be found in the rodent hippocampus (e.g., [46,142]), suggesting that the principles of CTC generalise across brain regions
and species. Lastly, while stimulus properties modulate gamma frequency and this frequency mismatch may hinder com-
munication [139], this can be beneficial: following a phase reset, highly excited networks produce fast gamma oscillations
that can quickly excite a connected region, with the resulting wave of inhibition suppressing inputs from less excited sender
regions (which exhibit slower gamma oscillations). In other words, variation in oscillatory frequency allows selective com-
munication, a key feature of CTC [41].

Given the evidence presented in the preceding text, it may be suggested that gamma-band coherence can support com-
munication, although it may not be the only means of coherence within the brain. Consequently, investigating the specific
conditions under which gamma coherence supports communication may prove fruitful.

Trends in Neurosciences, October 2023, Vol. 46, No. 10 837

 Trends in Neurosciences
OPEN ACCESS

(e.g., beta) oscillations [67]. However, these concerns may be addressed by cross-frequency
coupling, a phenomenon in which gamma oscillations nest within a slower rhythm [68]. When
the low-frequency oscillations of two distant regions become coherent, and local gamma oscilla-
tions phase lock to the coherent low-frequency oscillations, gamma-band coupling becomes
more precise than what would occur based on long-range gamma-band interactions alone.
In line with this idea, computational models [69], rodent hippocampal recordings [46], and
nonhuman primate visual cortical/thalamic recordings [70] all suggest a link between neural
communication and cross-frequency interactions. Consequently, future explorations into
communication through cross-frequency coupling may prove prosperous.

In sum, distinct gamma oscillations may track the flow of to-be-encoded and to-be-retrieved
information across the cortex. However, it remains to be seen whether gamma oscillatory
activity has a causal role in such communication or is simply a by-product. Of course, even if
it becomes apparent that gamma oscillations do not have a causal role in communication,
this does not preclude the potential role of gamma oscillations in other subprocesses of
episodic memory.

Gamma oscillations and sequence encoding/retrieval

Thus far, we have discussed episodic memories as though all information relevant to the memory
was presented and processed simultaneously. This is not the case, however: episodic memories
have an inherent temporal narrative, a property that gamma oscillations, nested in ongoing
hippocampal theta rhythms (~3–7 Hz), may facilitate [46,71–75]. In this partnership, the theta
oscillation defines the overarching episode (e.g., a birthday party), while nested gamma cycles
represent and separate distinct elements of the sequence (e.g., the birthday cake, the
presents). Here, we review how a theta–gamma code encodes and retrieves episodic memory
sequences.

Hippocampal theta–gamma coupling may support episodic memory by iteratively representing

elements of an episode in a manner optimal for LTP [73,76]. This process begins with a hippo-
campal cell assembly (i.e., a collection of cells tuned to an element of an event) firing in response
to a combination of external and theta phase-dependent excitation [77]. This results in feedback
inhibition that suppresses network activity momentarily (~20 ms). Once inhibition dissipates, the
next assembly in the sequence fires. The back-and-forth between excitation and inhibition
produces a gamma oscillation that helps segment elements of an event from one another [71]
(Figure 2A) and facilitates STDP. STDP allows the temporal order of an event to be encoded in

(A) (B)
Post-synaptic
A A B A B C A B C D A B C D
Synaptic strength

A
Pre-synaptic

B
C
D
Gamma

Theta

Trends in Neurosciences

Figure 2. Depiction of the contribution of theta–gamma coupling to episodic memory. (A) Individual elements of an
event are represented on individual gamma cycles, which are nested within an ongoing theta oscillation. Given the inhibitory
nature of theta, individual elements of an event occur at the trough of the theta cycle. Given that early elements always
precede later elements, unidirectional synaptic links can form between elements via asymmetric long-term potentiation
(LTP). (B) Sequential learning establishes synaptic links between elements, such that early elements can induce firing in
later elements, but not vice versa. Adapted from [88].

838 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

the asymmetric synaptic links between the neuronal representations of the event [78] (Figure 2B).
The alternation between cell assembly excitation and feedback inhibition continues until theta
reaches its inhibitory phase. This results in a gamma oscillation nested within the ongoing theta
oscillation, which supports the formation of a temporally coherent memory.

Supporting these ideas, computational modelling demonstrates that elements represented in a

biologically plausible theta–gamma code do indeed produce asymmetric synaptic connections
that later predict successful retrieval [79,80]. While a direct demonstration is lacking in biological
organisms, pairwise links between theta–gamma coupling, asymmetric LTP, and sequence
encoding have been observed. First, as discussed earlier, gamma oscillatory activity can facilitate
LTP [13,14], and LTP could be further enhanced if gamma couples to the phase of theta optimal
for LTP [81–84], suggesting that theta gamma coupling can enhance LTP. Second, sequences of
place cells representing a familiar path have been shown to fire more readily with each traversal
[85], an effect attributed to enhanced asymmetric synaptic connectivity between sequentially fir-
ing place cells, linking sequence learning to LTP (for similar evidence in humans, see [86]). Lastly,
in humans, the magnitude of theta–gamma coupling during object sequence learning predicts the
accuracy of temporal judgements about the objects presented in the sequence [11], explicitly
linking theta–gamma coupling to sequence memory. Weaving the strands together, it appears
that theta gamma coupling helps encode temporally coherent episodic memories through asym-
metric LTP.

With the sequence encoded through asymmetric synaptic connections, sequential recall is simply
a matter of cuing one element in a sequence and allowing the asymmetric synaptic links to
reactivate each remaining element in turn [72]. Given that memory reactivation is thought to
preferentially arise at the peak of the hippocampal theta cycle [82], sequence recall will inherently
be tied to this phase of theta. When this phase is reached, the cued assembly fires and excites
connected assemblies. As with encoding, feedback inhibition prevents immediate activation of
the connected assemblies, but as inhibition subsides, the most excited assembly (i.e., the one
that shares the strongest synaptic links) can fire [87]. This process repeats for each element of
the memory, resulting in theta–gamma-coupled activity that reflects the readout of sequential
information from episodic memory.

Several empirical studies support this explanatory link between theta–gamma coupling and
retrieval. For example, retrieval occurs at a preferred phase of theta in both rats (e.g., [88,89])
and humans (e.g., [90,91]), suggesting that recall is theta-phase dependent. Critically, disrupting
theta by injecting muscimol impairs both memory retrieval and theta–gamma coupling [6], while
optogenetically enhancing gamma boosts both memory retrieval and theta–gamma coupling
[92], suggesting a causal role for both theta and gamma in sequence retrieval (although this
may not be the only route to sequence retrieval; Box 3).

With evidence to suggest that theta–gamma coupling can carry codes related to both the past
and the present, we must once again address how the brain separates to-be-encoded and to-
be-retrieved information. This may be achieved by distinct hippocampal gamma rhythms nested
in differing phases of theta [45,46,82,93,94]. In rats, indirect support for this idea has come
from the observation that ‘fast’ gamma coupled to theta carries spatial sequences about where
the rat currently is (which is thought to be pertinent to forming new memories based on current
experience), while ‘slow’ gamma, coupled to a different phase of theta, represents sequences
of where rats are planning to go (which is thought to reflect the reactivation of past sequences
to aid navigation) [4,5]. Similar evidence has been reported in humans, with the two frequencies
coupling to different phases of theta [49]. Taken together, one could speculate that sequential

Trends in Neurosciences, October 2023, Vol. 46, No. 10 839

 Trends in Neurosciences
OPEN ACCESS

Box 3. Sharp-wave ripples as an alternate route to sequence retrieval

Similar to theta-gamma coupling, sharp-wave ripples have also been linked to the retrieval of sequences [144,145], al-
though it is unclear whether these two mechanisms are complementary or adversarial. Here, we briefly summarise the link
between sharp-wave ripples and sequences, and then discuss how the nature of LTP and theta–gamma coupling may
dictate which sequences sharp-wave ripples can retrieve.

By replaying previously experienced sequences, sharp-wave ripples are thought to facilitate processes such as systems
consolidation (e.g., [144]), reward learning (e.g., [146]), and creative planning (e.g., [147]). Intriguingly, this replay can occur
in a forward, backward, or novel order [146,148,149]. However, the observation of replay in any other direction than
forward is difficult to explain with any neural mechanism that implements asymmetric LTP because the latter does not
provide the synaptic connections necessary for traversing a sequence in any order other than the original. A possible
resolution to this problem comes from recent work indicating that, within CA3, symmetric STDP can occur [150]. Symmetric
STDP is ambivalent about which cell fired first, meaning it forms reciprocal connections between two cells. When applied to
theta–gamma coupling, this would see a sequence build both forward and backward connections. Importantly, synaptic
connections would still be moderated by temporal distance; thus, element C will have stronger connections to B and D than
to A and E, meaning that some degree of temporal order is retained. While symmetric LTP, relative to asymmetric LTP, may
make it more difficult to recall the exact order in which a sequence unfolded, it would provide additional flexibility for mentally
navigating elements of a sequence to understand unexpected rewards or exploring unexperienced sequences of events
[146,147]. It remains an open question whether theta–gamma coupling makes preferential use of either asymmetric or
symmetric LTP and, consequently, whether theta–gamma coupling principally creates memory traces with a strict temporal
narrative or traces that can be flexibly replayed and rearranged by sharp-wave ripples. Addressing this question will help
elucidate whether theta–gamma coupling and sharp-wave ripples reflect two complementary learning mechanisms or a
single cooperative mnemonic phenomenon [151].

information relating to the to-be-encoded present and to-be-remembered past is segregated by

nesting distinct gamma oscillations at different phases of ongoing theta activity.

Notably, the links made between theta–gamma coupling and sequence encoding/retrieval and
those made to neural communication may be complementary. For example, following the
theta–gamma-coupled reactivation of a sequence in CA3, theta-band coupling between the
CA3 and CA1 could ensure that gamma oscillations within the two regions become coherent
and, consequently, help the reactivated sequence propagate toward the cortex for reinstatement.
This highlights how theta–gamma coupling need not be thought of as solely aiding either commu-
nication or sequence representation, but may indeed facilitate both.

In sum, theta–gamma coupling provides the temporal scaffolding necessary for episodic
memory. This may be achieved by distinct gamma oscillations, nested in differing phases of
theta, supporting the encoding and retrieval of sequences. That said, much of the supporting
evidence for this latter claim comes from studies of rodent navigation. Therefore, it will be of
interest to see how these findings generalise to humans and nonhuman primates, which depend
less on locomotion to build representations of space (relative to rodents) and instead explore
space using different means (e.g., saccades; [95,96]).

From fundamental mechanisms to fully fledged episodic memories

So far, we have explored how gamma oscillations support low-level neural phenomena
(i.e., plasticity, communication, and sequence representation), which, in turn, can facilitate epi-
sodic memory. However, because much of this work relies on the study of single cells in vitro
or anthropomorphising the behaviour of rodents navigating a maze, it is unclear how well these
concepts generalise to the behavioural expression of episodic memories that we, as humans,
experience. Here, we explore recent work conducted in humans that directly links gamma
oscillations to the behavioural expression of episodic memories.

Fortunately, there is no shortage of studies linking gamma-band activity to human episodic

memory formation (e.g., [10,11,15,36,42,97–108]) or retrieval (e.g., [10,43,49,58,61,106,109–117]).

840 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

Many of these studies involve studying a series of stimuli and later recalling them on cue. These
stimuli are then split based on whether they were recalled and the associated neural activity is
contrasted. In doing so, these approaches isolate memory-related activity by minimising contribu-
tions from other cognitive phenomena (e.g., stimulus perception, motor response) that arise
regardless of whether encoding/retrieval was successful. Indeed, lab-based studies utilising
these approaches have demonstrated that inter-regional gamma-band coherence aids both
memory formation [42] and retrieval [43,116], while theta–gamma coupling predicts episodic
sequence encoding [11,98,106] and retrieval [106]. Moreover, the retrieval of real-life autobio-
graphical memories has been linked to enhanced gamma coherence [44] and theta–gamma cou-
pling [114,115], suggesting that gamma oscillations and the mechanisms they support have critical
and specific roles in life-like episodic memories.

While these studies have an advantage over cell- and animal-based studies in quantifying the
behavioural expression of episodic memory, they are at a disadvantage when quantifying
gamma oscillations. Specifically, there is difficulty in delineating oscillations from arrhythmic sig-
nals [118] that also correlate with episodic memory (e.g., [119]). Therefore, an analysis of spectral
power that does not attempt to separate out the contributions of gamma oscillatory activity from
broadband aperiodic activity will struggle to provide informative insights into the oscillatory under-
pinnings of episodic memory. However, analytical advances have been developed to tackle this
issue [118,120] and studies implementing these methods continue to suggest that narrowband
gamma oscillations relate to episodic memory formation/retrieval (and also highlight separable
contributions from lower-frequency oscillations; Box 4) [10,98,121].

Taken together, numerous human studies demonstrate that gamma oscillatory activity specifi-
cally maps onto the formation and retrieval of life-like episodic memories, further strengthening
the link between the two phenomena.

Future directions
While there are numerous threads that relate gamma oscillatory activity to synaptic plasticity,
neural communication, and sequence representation, further research is required to fully
elucidate these links. Here, we discuss how investigating multiple oscillations, causality, and
neuropathology may help address such open questions.

A common theme in the research reviewed in the preceding text is that multiple distinct oscilla-
tions, with differing frequencies and distinct mechanistic functions, sit (more or less) within the
canonical gamma band. For better understanding of the role of gamma oscillations in episodic
memory, it would be beneficial to account for these different forms of oscillation. This could be
achieved analytically by distinguishing putative gamma oscillations from aperiodic components

Box 4. Synchronisation versus desynchronisation

While this review has principally focused on the relevance of gamma oscillatory synchrony in episodic memory formation
and retrieval, growing evidence suggests that a second oscillatory correlate has an equally important role: widespread
cortical power decreases, predominately in the lower frequencies (<30 Hz; e.g., [10,101–103,121,152–157]), but also
within the gamma-band itself (e.g., [116]). While one might be led to believe that gamma-band power increases and
low-frequency cortical power decreases are inversely correlated, several studies suggest that these two electrophysiolog-
ical phenomena have distinct roles in episodic memory formation and retrieval [10,98,121]. For example, these cortical
power decreases may be involved in information processing [158], decreasing inhibition [159], and/or moderating top-
down control (e.g., [55,57]). Based on these functions, one could differentiate low-frequency oscillations from gamma
oscillations by suggesting that the former support the allocation of resources (e.g., attention), while the latter reflect active
processing (e.g., plasticity, neural communication, or sequence representation) (e.g., [55,160]). Consequently, low-
frequency power decreases may complement gamma oscillations in the formation and retrieval of episodic memories.
For further discussion on this point, see [3].

Trends in Neurosciences, October 2023, Vol. 46, No. 10 841

 Trends in Neurosciences
OPEN ACCESS

of the power spectrum (e.g., [118,122]) or by using spatial separation methods (e.g., independent Outstanding questions
components analysis; ICA) to distinguish hippocampal and cortical gamma rhythms Do gamma oscillations provide any
(e.g., [43,55]). Complementing this, experimental paradigms could be finessed to better isolate additional benefit to plasticity above
and beyond what is provided by
distinct gamma oscillations. For example, matching sensory input/motor output between
arrhythmic but synchronous neural
encoding and retrieval would ensure that only the internal state differs between encoding and activity?
retrieval, allowing direct comparison of the distinct rhythms associated with each state (e.g., [10]).
Ultimately, these approaches may help separate the many forms of gamma oscillation that have How do NMDA antagonists
(e.g., ketamine) influence the inter-
critical roles in episodic memory.
action between gamma oscillatory
activity and STDP?
With numerous studies demonstrating a correlative link between gamma oscillations and
episodic memory, and mechanistic links proposing almost unanimously that gamma oscillatory Does gamma oscillatory coherence
support the flow of information from
activity results in memory formation and/or retrieval, causality also needs addressing. Fortunately, the cortex to the hippocampus during
numerous promising techniques can help establish causality, including optogenetic manipulation memory formation? Does the phase
of gamma oscillations in rodents (e.g., [92]), and sensory stimulation in humans (e.g., [15]). Of delay between the sender and reader
course, it is a challenge to determine absolute causality between gamma oscillations and episodic dictate whether a memory can be
formed?
memory because this requires total control over a system that can generate and recall episodic
memories. Therefore, cross-disciplinary approaches are essential. For example, combining Can separable ‘fast’ and ‘slow’
computational models that explore the impact of gamma oscillations in a controlled system gamma oscillations be observed
outside of the medial temporal lobe?
with behavioural experiments exploring how exogenous gamma stimulation impacts real episodic
How do they relate to feedforward
memories will provide deeper insights into the causal role of gamma oscillations in episodic gamma and feedback beta rhythms
memory compared with either approach in isolation. observed in the sensory cortices?

Can individual elements of an episodic

Lastly, it is worth considering how cognition changes in the face of pathological gamma oscil- memory be decoded from human
lations. While fundamental research linking gamma oscillations to episodic memories has been theta–gamma sequences during mem-
used to inform potential interventions for neurological disorders (e.g., Alzheimer’s disease; ory formation and/or retrieval?
[123–127]; however, see [128,129]), we propose that studies of clinical populations can also
Is there a trade-off between accurate
inform fundamental research. As discussed throughout this review, the brain does not execute temporal order memory and flexible
a singular ‘episodic memory’ process but rather executes a whole host of processes thinking? Do asymmetric synaptic links
(e.g., plasticity, neural communication, and sequence representation) from which episodic between elements favour the former
while symmetric synaptic links favour
memory emerges. Neurological disorders are similarly ambivalent to psychological constructs
the latter?
such as episodic memory: neurological disorders principally associated with memory-related
problems (e.g., dementia) often entail non-memory-related problems, while disorders that Does the magnitude of theta–gamma
are not typically thought of as memory-related disorders can, nonetheless, involve memory coupling during memory formation
predict the intensity and/or frequency
issues (e.g., schizophrenia). Considering the commonalities and idiosyncrasies between disor- of replay events during later offline
ders may provide an alternative view into the link between memory and gamma oscillations. periods (e.g., sleep)?
For example, gamma oscillatory dysrhythmia can be observed in Alzheimer’s disease [130]
Can gamma-band sensory stimulation
and autism [131], but memory function is markedly different between the two [132,133].
modulate memory? Given that ‘fast’
Understanding why gamma oscillatory dysrhythmia may relate to mnemonic impairment in gamma sits above the flicker fusion
some instances but not others may help elucidate the mechanistic role of gamma oscillations threshold, does this mean that gamma-
in episodic memory, complementing those provided by fundamental research conducted on band sensory stimulation offers an
imperceivable memory intervention?
healthy participants.

Concluding remarks
The well-documented link between gamma oscillations and episodic memory is likely to come
from not one but many underlying mechanisms, including synaptic plasticity, neural commu-
nication, and sequence representation. However, whether gamma oscillations have a causal
role in these mechanisms remains to be seen (see Outstanding questions). Consequently,
future research that causally manipulates gamma oscillatory activity may be the best step
forward to advance our understanding of the link between gamma oscillations and episodic
memory.

842 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

Acknowledgements
This work was supported by the Leverhulme Trust (www.leverhulme.ac.uk/, Early Career Fellowship ECF-2021-628
awarded to B.J.G.) and by the Templeton World Charity Foundation (TWCF0389). Thanks to Syanah Wynn for providing
insightful comments on a draft of this manuscript.

Declaration of interests
The authors declare no competing interest.

References
1. Tulving, E. (1972) Episodic and semantic memory. In Organization 22. Guerra, A. et al. (2020) Enhancing gamma oscillations restores
of Memory (Tulving, E. and Donaldson, W., eds), pp. 381–403, primary motor cortex plasticity in Parkinson’s disease.
Academic Press J. Neurosci. 40, 4788–4796
2. Staresina, B.P. and Wimber, M. (2019) A neural chronometry of 23. Nyhus, E. and Curran, T. (2010) Functional role of gamma and
memory recall. Trends Cogn. Sci. 23, 1071–1085 theta oscillations in episodic memory. Neurosci. Biobehav.
3. Hanslmayr, S. et al. (2016) Oscillations and episodic memory – Rev. 34, 1023–1035
addressing the synchronization/desynchronization conundrum. 24. Bi, G. and Poo, M. (1998) Synaptic modifications in cultured
Trends Neurosci. 39, 16–25 hippocampal neurons: dependence on spike timing, synaptic
4. Zheng, C. et al. (2016) Spatial sequence coding differs during strength, and postsynaptic cell type. J. Neurosci. 18, 1–9
slow and fast gamma rhythms in the hippocampus. Neuron 25. Wittenberg, G.M. and Wang, S.S.-H. (2006) Malleability of
89, 398–408 spike-timing-dependent plasticity at the CA3-CA1 synapse.
5. Bieri, K.W. et al. (2014) Slow and fast gamma rhythms coordi- J. Neurosci. 26, 6610–6617
nate different spatial coding modes in hippocampal place 26. Froemke, R.C. et al. (2005) Spike-timing-dependent synaptic
cells. Neuron 82, 670–681 plasticity depends on dendritic location. Nature 434, 221–225
6. Shirvalkar, P.R. et al. (2010) Bidirectional changes to hippocampal 27. Morera-Herreras, T. et al. (2019) Environmental enrichment
theta-gamma comodulation predict memory for recent spatial shapes striatal spike-timing-dependent plasticity in vivo. Sci.
episodes. Proc. Natl. Acad. Sci. U. S. A. 107, 7054–7059 Rep. 9, 19451
7. Headley, D.B. and Weinberger, N.M. (2011) Gamma-band acti- 28. Huang, S. et al. (2014) Associative Hebbian synaptic plasticity in
vation predicts both associative memory and cortical plasticity. primate visual cortex. J. Neurosci. 34, 7575–7579
J. Neurosci. 31, 12748–12758 29. Seeman, S.C. et al. (2017) Paired stimulation for spike-timing-
8. Csorba, B.A. et al. (2022) Long-range cortical synchronization dependent plasticity in primate sensorimotor cortex. J. Neurosci.
supports abrupt visual learning. Curr. Biol. 32, 2467–2479 37, 1935–1949
9. Jutras, M.J. et al. (2009) Gamma-band synchronization in the 30. Testa-Silva, G. (2010) Human synapses show a wide temporal
macaque hippocampus and memory formation. J. Neurosci. window for spike-timing-dependent plasticity. Front. Synaptic
29, 12521–12531 Neurosci. 2, 12
10. Griffiths, B.J. et al. (2019) Directional coupling of slow and fast 31. Salinas, E. and Sejnowski, T.J. (2000) Impact of correlated syn-
hippocampal gamma with neocortical alpha/beta oscillations aptic input on output firing rate and variability in simple neuronal
in human episodic memory. Proc. Natl. Acad. Sci. U.S.A. 166, models. J. Neurosci. 20, 6193–6209
21834–21842 32. Murthy, V.N. and Fetz, E.E. (1994) Effects of input synchrony on
11. Heusser, A.C. et al. (2016) Episodic sequence memory is supported the firing rate of a three-conductance cortical neuron model.
by a theta-gamma phase code. Nat. Neurosci. 19, 1374–1380 Neural Comput. 6, 1111–1126
12. Fell, J. et al. (2001) Human memory formation is accompanied 33. McNaughton, B.L. et al. (1978) Synaptic enhancement in fascia
by rhinal–hippocampal coupling and decoupling. Nat. Neurosci. dentata: cooperativity among coactive afferents. Brain Res.
4, 1259–1264 157, 277–293
13. Sjöström, P.J. et al. (2001) Rate, timing, and cooperativity jointly 34. Jensen, O. et al. (2007) Human gamma-frequency oscillations
determine cortical synaptic plasticity. Neuron 32, 1149–1164 associated with attention and memory. Trends Neurosci. 30,
14. Markram, H. et al. (1997) Regulation of synaptic efficacy by 317–324
coincidence of postsynaptic APs and EPSPs. Science 275, 35. Wespatat, V. et al. (2004) Phase sensitivity of synaptic modifications
213–215 in oscillating cells of rat visual cortex. J. Neurosci. 24, 9067–9075
15. Wang, D. et al. (2022) Altering stimulus timing via fast rhythmic 36. Roux, F. et al. (2022) Oscillations support short latency co-firing
sensory stimulation induces STDP-like recall performance in of neurons during human episodic memory formation. eLife 11,
human episodic memory. bioRxiv Published online November e78109
3, 2022. https://doi.org/10.1101/2022.11.02.514843 37. Traub, R.D. et al. (1998) Gamma-frequency oscillations: a
16. Bliss, T.V.P. and Collingridge, G.L. (1993) A synaptic model of neuronal population phenomenon, regulated by synaptic and
memory: long-term potentiation in the hippocampus. Nature intrinsic cellular processes, and inducing synaptic plasticity.
361, 31–39 Prog. Neurobiol. 55, 563–575
17. Hebb, D. (1949) The Organisation of Behavior, John Wiley & 38. Wang, H.-X. et al. (2005) Coactivation and timing-dependent in-
Sons tegration of synaptic potentiation and depression. Nat.
18. Whittington, M.A. et al. (1997) Recurrent excitatory postsynap- Neurosci. 8, 187–193
tic potentials induced by synchronized fast cortical oscillations. 39. Froemke, R.C. and Dan, Y. (2002) Spike-timing-dependent
Proc. Natl. Acad. Sci. U.S.A. 94, 12198–12203 synaptic modification induced by natural spike trains. Nature
19. Li, K.T. et al. (2021) Gamma oscillations facilitate effective learn- 416, 433–438
ing in excitatory-inhibitory balanced neural circuits. Neural Plast. 40. Fries, P. (2005) A mechanism for cognitive dynamics: neuronal
2021, 1–18 communication through neuronal coherence. Trends Cogn.
20. Park, K. et al. (2020) Optogenetic activation of parvalbumin and Sci. 9, 474–480
somatostatin interneurons selectively restores theta-nested 41. Fries, P. (2015) Rhythms for cognition: communication through
gamma oscillations and oscillation-induced spike timing- coherence. Neuron 88, 220–235
dependent long-term potentiation impaired by amyloid β oligomers. 42. Costa, M. et al. (2022) Aversive memory formation in humans
BMC Biol. 18, 7 involves an amygdala-hippocampus phase code. Nat. Commun.
21. Guerra, A. et al. (2018) Boosting the LTP-like plasticity effect of 13, 6403
intermittent theta-burst stimulation using gamma transcranial 43. Pacheco Estefan, D. et al. (2019) Coordinated representational
alternating current stimulation. Brain Stimulation 11, 734–742 reinstatement in the human hippocampus and lateral temporal

Trends in Neurosciences, October 2023, Vol. 46, No. 10 843

 Trends in Neurosciences
OPEN ACCESS

cortex during episodic memory retrieval. Nat. Commun. 10, 72. Lisman, J. and Buzsaki, G. (2008) A neural coding scheme formed
2255 by the combined function of gamma and theta oscillations.
44. Fuentemilla, L. et al. (2018) Gamma phase-synchrony in autobio- Schizophr. Bull. 34, 974–980
graphical memory: evidence from magnetoencephalography and 73. Lisman, J.E. and Idart, M.A.P. (1995) Storage of 7 +/- 2 short-
severely deficient autobiographical memory. Neuropsychologia term memories in oscillatory subcycles. Science 267, 1512–1515
110, 7–13 74. Bragin, A. et al. (1995) Gamma (40-100 Hz) oscillation in the
45. Colgin, L.L. (2015) Do slow and fast gamma rhythms corre- hippocampus of the behaving rat. J. Neurosci. 15, 47–60
spond to distinct functional states in the hippocampal network? 75. Belluscio, M.A. et al. (2012) Cross-frequency phase–phase cou-
Brain Res. 1621, 309–315 pling between theta and gamma oscillations in the hippocampus.
46. Colgin, L.L. et al. (2009) Frequency of gamma oscillations J. Neurosci. 32, 423–435
routes flow of information in the hippocampus. Nature 462, 76. Jensen, O. and Lisman, J.E. (2005) Hippocampal sequence-
353–357 encoding driven by a cortical multi-item working memory buffer.
47. Zheng, C. et al. (2016) Fast gamma rhythms in the hippocam- Trends Neurosci. 28, 67–72
pus promote encoding of novel object-place pairings. eNeuro 77. Csicsvari, J. et al. (1999) Oscillatory coupling of hippocampal
3, ENEURO.0001-16.2016 pyramidal cells and interneurons in the behaving rat. J. Neurosci.
48. Carr, M.F. et al. (2012) Transient slow gamma synchrony under- 19, 274–287
lies hippocampal memory replay. Neuron 75, 700–713 78. Muller, R.U. et al. (1996) The hippocampus as a cognitive
49. Vivekananda, U. et al. (2021) Theta power and theta- graph. J. Gen. Physiol. 107, 663–694
gamma coupling support long-term spatial memory retrieval. 79. Jensen, O. and Lisman, J.E. (1996) Theta/gamma networks
Hippocampus 31, 213–220 with slow NMDA channels learn sequences and encode epi-
50. Kucewicz, M.T. et al. (2017) Dissecting gamma frequency activity sodic memory: role of NMDA channels in recall. Learn. Mem.
during human memory processing. Brain 140, 1337–1350 3, 264–278
51. Yamamoto, J. et al. (2014) Successful execution of working 80. Jensen, O. and Lisman, J.E. (1996) Hippocampal CA3 region
memory linked to synchronized high-frequency gamma oscillations. predicts memory sequences: accounting for the phase preces-
Cell 157, 845–857 sion of place cells. Learn. Mem. 3, 279–287
52. Igarashi, K.M. et al. (2014) Coordination of entorhinal-hippocampal 81. Griffiths, B.J. and Fuentemilla, L. (2020) Event conjunction: how
ensemble activity during associative learning. Nature 510, 143–147 the hippocampus integrates episodic memories across event
53. Zhou, Y. et al. (2019) Methodological considerations on the use boundaries. Hippocampus 30, 162–171
of different spectral decomposition algorithms to study hippo- 82. Hasselmo, M.E. (2005) What is the function of hippocampal
campal rhythms. eNeuro 6, 1–30 theta rhythm? Linking behavioral data to phasic properties of
54. Miller, E.K. et al. (2018) Working memory 2.0. Neuron 100, field potential and unit recording data. Hippocampus 15,
463–475 936–949
55. Bastos, A.M. et al. (2020) Layer and rhythm specificity for pre- 83. Huerta, P.T. and Lisman, J.E. (1995) Bidirectional synaptic plasticity
dictive routing. Proc. Natl. Acad. Sci. U.S.A. 117, 31459–31469 induced by a single burst during cholinergic theta oscillation in CA1
56. Michalareas, G. et al. (2016) Alpha-beta and gamma rhythms in vitro. Neuron 15, 1053–1063
subserve feedback and feedforward influences among human 84. Hölscher, C. et al. (1997) Stimulation on the positive phase of
visual cortical areas. Neuron 89, 384–397 hippocampal theta rhythm induces long-term potentiation that
57. Bastos, A.M. et al. (2015) Visual areas exert feedforward and can be depotentiated by stimulation on the negative phase in
feedback influences through distinct frequency channels. Neuron area CA1 In Vivo. J. Neurosci. 17, 6470–6477
85, 390–401 85. Mehta, M.R. et al. (1997) Experience-dependent, asymmetric
58. Haque, R.U. et al. (2020) Feedforward prediction error signals expansion of hippocampal place fields. Proc. Natl. Acad. Sci.
during episodic memory retrieval. Nat. Commun. 11, 6075 U.S.A. 94, 8918–8921
59. Barron, H.C. et al. (2020) Prediction and memory: a predictive 86. Ekman, M. et al. (2023) Successor-like representation guides
coding account. Prog. Neurobiol. 192, 101821 the prediction of future events in human visual cortex and
60. Gregoriou, G.G. et al. (2009) High-frequency, long-range hippocampus. eLife 12, e78904
coupling between prefrontal and visual cortex during attention. 87. de Almeida, L. et al. (2009) A second function of gamma
Science 324, 1207–1210 frequency oscillations: an E%-Max Winner-Take-all mechanism
61. Treder, M.S. et al. (2021) The hippocampus as the switchboard selects which cells fire. J. Neurosci. 29, 7497–7503
between perception and memory. Proc. Natl. Acad. Sci. U.S.A. 88. Manns, J.R. and Eichenbaum, H. (2009) A cognitive map for
118, e2114171118 object memory in the hippocampus. Learn. Mem. 16, 616–624
62. Ermentrout, G.B. and Kopell, N. (1984) Frequency plateaus in a 89. Siegle, J.H. and Wilson, M.A. (2014) Enhancement of encoding
chain of weakly coupled oscillators, I. SIAM J. Math. Anal. 15, and retrieval functions through theta phase-specific manipula-
215–237 tion of hippocampus. eLife 3, e03061
63. Rolls, E.T. et al. (2012) Communication before coherence: com- 90. Kerrén, C. et al. (2018) An optimal oscillatory phase for pattern
munication before coherence. Eur. J. Neurosci. 36, 2689–2709 reactivation during memory retrieval. Curr. Biol. 28, 3383–3392
64. Schneider, M. et al. (2021) A mechanism for inter-areal coherence 91. ter Wal, M. et al. (2021) Theta rhythmicity governs human
through communication based on connectivity and oscillatory behavior and hippocampal signals during memory-dependent
power. Neuron 109, 4050–4067.e12 tasks. Nat. Commun. 12, 7048
65. Vinck, M. et al. (2023) Principles of large-scale neural interactions. 92. Etter, G. et al. (2019) Optogenetic gamma stimulation rescues
Neuron 111, 987–1002 memory impairments in an Alzheimer’s disease mouse model.
66. Ray, S. and Maunsell, J.H.R. (2015) Do gamma oscillations play Nat. Commun. 10, 5322
a role in cerebral cortex? Trends Cogn. Sci. 19, 78–85 93. Lopes-dos-Santos, V. et al. (2018) Parsing hippocampal
67. Kopell, N. et al. (2000) Gamma rhythms and beta rhythms have theta oscillations by nested spectral components during
different synchronization properties. Proc. Natl. Acad. Sci. spatial exploration and memory-guided behavior. Neuron
U.S.A. 97, 1867–1872 100, 940–952
68. Bonnefond, M. et al. (2017) Communication between brain areas 94. Aguilera, M. et al. (2022) How many gammas? Redefining
based on nested oscillations. eNeuro 4, ENEURO.0153-16.2017 hippocampal theta-gamma dynamic during spatial learning.
69. Quax, S. et al. (2017) Top-down control of cortical gamma- Front. Behav. Neurosci. 16, 811278
band communication via pulvinar induced phase shifts in the 95. Staudigl, T. et al. (2018) Hexadirectional modulation of high-
alpha rhythm. PLoS Comput. Biol. 13, e1005519 frequency electrophysiological activity in the human anterior
70. Saalmann, Y.B. et al. (2012) The pulvinar regulates information medial temporal lobe maps visual space. Curr. Biol. 28,
transmission between cortical areas based on attention 3325–3329
demands. Science 337, 753–756 96. Jutras, M.J. et al. (2013) Oscillatory activity in the monkey
71. Lisman, J.E. and Jensen, O. (2013) The theta-gamma neural hippocampus during visual exploration and memory formation.
code. Neuron 77, 1002–1016 Proc. Natl. Acad. Sci. U.S.A. 110, 13144–13149

844 Trends in Neurosciences, October 2023, Vol. 46, No. 10

Trends in Neurosciences
OPEN ACCESS

97. Karlsson, A.E. et al. (2022) Out of rhythm: compromised preci- 121. Fellner, M.-C. et al. (2019) Spectral fingerprints or spectral tilt?
sion of theta-gamma coupling impairs associative memory in Evidence for distinct oscillatory signatures of memory formation.
old age. J. Neurosci. 42, 1752–1764 PLoS Biol. 17, e3000403
98. Griffiths, B.J. et al. (2021) Disentangling neocortical alpha/beta 122. Huang, N.E. et al. (1998) The empirical mode decomposition
and hippocampal theta/gamma oscillations in human episodic and the Hilbert spectrum for nonlinear and non-stationary time
memory formation. NeuroImage 242, 118454 series analysis. Proc. R. Soc. Lond. A 454, 903–995
99. Lega, B. et al. (2016) Slow-theta-to-gamma phase-amplitude 123. Adaikkan, C. et al. (2019) Gamma entrainment binds higher-
coupling in human hippocampus supports the formation of order brain regions and offers neuroprotection. Neuron 102,
new episodic memories. Cereb. Cortex 26, 268–278 929–943
100. Rubinstein, D.Y. et al. (2021) Contribution of left supramarginal 124. Chan, D. et al. (2022) Gamma frequency sensory stimulation in
and angular gyri to episodic memory encoding: An intracranial mild probable Alzheimer’s dementia patients: results of feasibility
EEG study. NeuroImage 225, 117514 and pilot studies. PLoS ONE 17, e0278412
101. Greenberg, J.A. et al. (2015) Decreases in theta and increases in 125. Martorell, A.J. et al. (2019) Multi-sensory gamma stimulation
high frequency activity underlie associative memory encoding. ameliorates Alzheimer’s-associated pathology and improves
NeuroImage 114, 257–263 cognition. Cell 177, 256–271
102. Burke, J.F. et al. (2013) Synchronous and asynchronous theta and 126. Jones, K.T. et al. (2023) Gamma neuromodulation improves
gamma activity during episodic memory formation. J. Neurosci. 33, episodic memory and its associated network in amnestic mild
292–304 cognitive impairment: a pilot study. Neurobiol. Aging 129,
103. Hanslmayr, S. et al. (2009) Brain oscillations dissociate between 72–88
semantic and nonsemantic encoding of episodic memories. 127. Traikapi, A. et al. (2023) Episodic memory effects of gamma
Cereb. Cortex 19, 1631–1640 frequency precuneus transcranial magnetic stimulation in
104. Sederberg, P.B. et al. (2007) Hippocampal and neocortical Alzheimer’s disease: a randomized multiple baseline study.
gamma oscillations predict memory formation in humans. J. Neuropsychol. 17, 279–301
Cereb. Cortex 17, 1190–1196 128. Soula, M. et al. (2023) Forty-hertz light stimulation does not
105. Solomon, E.A. et al. (2019) Dynamic theta networks in the entrain native gamma oscillations in Alzheimer’s disease
human medial temporal lobe support episodic memory. Curr. model mice. Nat. Neurosci. 26, 570–578
Biol. 29, 1100–1111 129. Schneider, M. et al. (2023) Cell-type-specific propagation of
106. Saint Amour Di Chanaz, L. et al. (2023) Gamma amplitude is visual flicker. Cell Rep. 42, 112492
coupled to opposed hippocampal theta-phase states during 130. Mably, A.J. and Colgin, L.L. (2018) Gamma oscillations in
the encoding and retrieval of episodic memories in humans. cognitive disorders. Curr. Opin. Neurobiol. 52, 182–187
Curr. Biol. 33, 1836–1843 131. Wilson, T.W. et al. (2007) Children and adolescents with autism
107. Wang, D.X. et al. (2021) Cross-regional phase amplitude coupling exhibit reduced MEG steady-state gamma responses. Biol.
supports the encoding of episodic memories. Hippocampus 31, Psychiatry 62, 192–197
481–492 132. Renner, P. et al. (2000) Implicit and explicit memory in autism: is
108. Staudigl, T. and Hanslmayr, S. (2013) Theta oscillations at autism an amnesic disorder? J. Autism Dev. Disord. 30
encoding mediate the context-dependent nature of human 133. Carlesimo, G.A. and Oscar-Berman, M. (1992) Memory deficits
episodic memory. Curr. Biol. 23, 1101–1106 in Alzheimer’s patients: a comprehensive review. Neuropsychol.
109. Sederberg, P.B. et al. (2007) Gamma oscillations distinguish Rev. 3, 119–169
true from false memories. Psychol. Sci. 18, 927–932 134. Tzounopoulos, T. et al. (2004) Cell-specific, spike timing-
110. Yaffe, R.B. et al. (2017) Cued memory retrieval exhibits rein- dependent plasticities in the dorsal cochlear nucleus. Nat.
statement of high gamma power on a faster timescale in the Neurosci. 7, 719–725
left temporal lobe and prefrontal cortex. J. Neurosci. 37, 135. Egger, V. et al. (1999) Coincidence detection and changes of
4472–4480 synaptic efficacy in spiny stellate neurons in rat barrel cortex.
111. Staresina, B.P. et al. (2016) Hippocampal pattern completion is Nat. Neurosci. 2, 1098–1105
linked to gamma power increases and alpha power decreases 136. Letzkus, J.J. et al. (2006) Learning rules for spike timing-
during recollection. eLife 5, 1–18 dependent plasticity depend on dendritic synapse location.
112. Steinvorth, S. et al. (2010) Human entorhinal gamma and theta J. Neurosci. 26, 10420–10429
oscillations selective for remote autobiographical memory. 137. Kampa, B.M. et al. (2004) Kinetics of Mg2+ unblock of NMDA
Hippocampus 20, 166–173 receptors: implications for spike-timing dependent synaptic
113. Tan, R.J. et al. (2020) Direct brain recordings identify hippocampal plasticity. J. Physiol. 556, 337–345
and cortical networks that distinguish successful versus failed 138. Spyropoulos, G. et al. (2023) Distinct feedforward and feedback
episodic memory retrieval. Neuropsychologia 147, 107595 pathways for cell-type specific attention effects. bioRxiv
114. Hebscher, M. et al. (2019) A causal role for the precuneus Published online March 23, 2023. https://doi.org/10.1101/
in network-wide theta and gamma oscillatory activity during 2022.11.04.515185
complex memory retrieval. eLife 8, e43114 139. Ray, S. and Maunsell, J.H.R. (2010) Differences in gamma
115. Roehri, N. et al. (2022) Phase-amplitude coupling and phase frequencies across visual cortex restrict their possible use in
synchronization between medial temporal, frontal and posterior computation. Neuron 67, 885–896
brain regions support episodic autobiographical memory recall. 140. Hermes, D. et al. (2015) Stimulus dependence of gamma oscil-
Brain Topogr. 35, 191–206 lations in human visual cortex. Cereb. Cortex 25, 2951–2959
116. Solomon, E.A. et al. (2017) Widespread theta synchrony and 141. Womelsdorf, T. et al. (2006) Gamma-band synchronization in
high-frequency desynchronization underlies enhanced cognition. visual cortex predicts speed of change detection. Nature 439,
Nat. Commun. 8, 1704 733–736
117. Staudigl, T. et al. (2012) Memory signals from the thalamus: early 142. Montgomery, S. and Buzsáki, G. (2007) Gamma oscillations
thalamocortical phase synchronization entrains gamma oscilla- dynamically couple hippocampal CA3 and CA1 regions during
tions during long-term memory retrieval. Neuropsychologia 50, memory task performance. Proc. Natl. Acad. Sci. U.S.A. 104,
3519–3527 14495–14500
118. Donoghue, T. et al. (2020) Parameterizing neural power spectra 143. Rohenkohl, G. et al. (2018) Gamma synchronization between
into periodic and aperiodic components. Nat. Neurosci. 23, V1 and V4 improves behavioral performance. Neuron 100,
1655–1665 953–963.e3
119. Burke, J.F. et al. (2015) Human intracranial high-frequency activity 144. Carr, M.F. et al. (2011) Hippocampal replay in the awake state:
during memory processing: Neural oscillations or stochastic a potential substrate for memory consolidation and retrieval.
volatility? Curr. Opin. Neurobiol. 31, 104–110 Nat. Neurosci. 14, 147–153
120. Wen, H. and Liu, Z. (2016) Separating fractal and oscillatory 145. Buzsáki, G. (2015) Hippocampal sharp wave-ripple: a cognitive
components in the power spectrum of neurophysiological biomarker for episodic memory and planning. Hippocampus
signal. Brain Topogr. 29, 13–26 25, 1073–1188

Trends in Neurosciences, October 2023, Vol. 46, No. 10 845

 Trends in Neurosciences
OPEN ACCESS

146. Foster, D.J. and Wilson, M.A. (2006) Reverse replay of behavioural 154. Karlsson, A.E. et al. (2020) Item recognition and lure discrimina-
sequences in hippocampal place cells during the awake state. tion in younger and older adults are supported by alpha/beta
Nature 440, 680–683 desynchronization. Neuropsychologia 148, 107658
147. Kurth-Nelson, Z. et al. (2023) Replay and compositional 155. Long, N.M. and Kahana, M.J. (2015) Successful memory
computation. Neuron 111, 454–469 formation is driven by contextual encoding in the core memory
148. Diba, K. and Buzsáki, G. (2007) Forward and reverse hippo- network. NeuroImage 119, 332–337
campal place-cell sequences during ripples. Nat. Neurosci. 156. Griffiths, B.J. et al. (2021) Alpha/beta power decreases during
10, 1241–1242 episodic memory formation predict the magnitude of alpha/beta
149. Liu, Y. et al. (2019) Human replay spontaneously reorganizes power decreases during subsequent retrieval. Neuropsychologia
experience. Cell 178, 640–652 153, 107755
150. Mishra, R.K. et al. (2016) Symmetric spike timing-dependent 157. Griffiths, B.J. et al. (2016) Brain oscillations track the formation of
plasticity at CA3–CA3 synapses optimizes storage and recall episodic memories in the real world. NeuroImage 143, 256–266
in autoassociative networks. Nat. Commun. 7, 11552 158. Hanslmayr, S. et al. (2012) Oscillatory power decreases and
151. Geschwill, P. et al. (2020) Synchronicity of excitatory inputs long-term memory: the information via desynchronization
drives hippocampal networks to distinct oscillatory patterns. hypothesis. Front. Hum. Neurosci. 6, 1–12
Hippocampus 30, 1044–1057 159. Jensen, O. and Mazaheri, A. (2010) Shaping functional architec-
152. Griffiths, B.J. et al. (2019) Alpha/beta power decreases track ture by oscillatory alpha activity: gating by inhibition. Front.
the fidelity of stimulus-specific information. eLife 8, 1–22 Hum. Neurosci. 4, 1–8
153. Martín-Buro, M.C. et al. (2020) Alpha rhythms reveal when and 160. Bauer, M. et al. (2014) Attentional modulation of alpha/beta and
where item and associative memories are retrieved. J. Neurosci. gamma oscillations reflect functionally distinct processes.
40, 2510–2518 J. Neurosci. 34, 16117–16125

846 Trends in Neurosciences, October 2023, Vol. 46, No. 10