Imagination as a fundamental function of the hippocampus. Alison E. Comrie, Loren M. Frank and Kenneth Kay. Philosophical Transactions of the Royal Society B: Biological Sciences. October 31 2022. https://doi.org/10.1098/rstb.2021.0336
Abstract: Imagination is a biological function that is vital to human experience and advanced cognition. Despite this importance, it remains unknown how imagination is realized in the brain. Substantial research focusing on the hippocampus, a brain structure traditionally linked to memory, indicates that firing patterns in spatially tuned neurons can represent previous and upcoming paths in space. This work has generally been interpreted under standard views that the hippocampus implements cognitive abilities primarily related to actual experience, whether in the past (e.g. recollection, consolidation), present (e.g. spatial mapping) or future (e.g. planning). However, relatively recent findings in rodents identify robust patterns of hippocampal firing corresponding to a variety of alternatives to actual experience, in many cases without overt reference to the past, present or future. Given these findings, and others on hippocampal contributions to human imagination, we suggest that a fundamental function of the hippocampus is to generate a wealth of hypothetical experiences and thoughts. Under this view, traditional accounts of hippocampal function in episodic memory and spatial navigation can be understood as particular applications of a more general system for imagination. This view also suggests that the hippocampus contributes to a wider range of cognitive abilities than previously thought.
5. Organization and origin of generative activity in the brain
Having reviewed multiple types of generative neural activity in the hippocampus, we turn to our next question of how generative representations may be organized and ‘parsed’ from representations of actual, ongoing experience. One would expect that neural processes are in place to separate actual and generative activity to avoid their confusion, reminiscent of the subject-level ability to internally distinguish actual from imagined experience [4]. Multiple organizational schemes are possible; different sets of neurons could participate in actual versus generative representations, these representations could occur at different relative times, or some combination of these schemes could take place.
Findings in the rodent hippocampus indicate that neural firing corresponding to actual and generative representations occur at different relative times that are internally determined [105]. Generative representations tend to occur not only with temporal separation from representations of actuality, but also in alignment with underlying network-level activity patterns in the hippocampus that are internally generated: SWRs and the theta rhythm (figure 3a) [68,106]. This results in a serial alternation of neural firing corresponding to actuality and generativity, or a temporal ‘multiplexing’ of actual and generative representations in the brain.
This serial alternation is present across behavioural states. During immobility, neural firing corresponding to the animal's actual present location is maintained for prolonged periods, transiently suppressed during SWR events that typically contain generative replays (tens to hundreds of milliseconds), and then subsequently restored (figure 3) [106,107].
Similarly, during movement and exploratory behaviours, neural firing corresponding to actual present and non-actual alternative experience, or actual and generative representations, occurs serially and in alignment with characteristic phases of the theta rhythm [3,68]. More specifically, early phases characteristically contain representations of the animal's actual past and present experience, while late phases may contain firing corresponding to a variety of hypothetical experiences, resulting in alternating actual and generative representations (examples in figure 2, schematic in figure 3) [68]. Furthermore, there are multiple levels of alternation between actual and generative activity during movement—representations not only alternate within approximately 125 ms theta cycles (e.g. actual and upcoming position), but also across consecutive theta cycles (e.g. alternation of two possible paths ahead; figure 2) [68]. Additional findings are also consistent with the idea that multiple representations can be accommodated in the hippocampus via serial alternation at a sub-second timescale. For instance, studies in the rat hippocampus have reported theta-modulated ‘flickering’ between representations of two environmental contexts, as well as dynamic switching between two spatial reference frames, and separate reverse and forward-ordered location sequences within theta cycles [108–110].
The organization of actual and generative neural firing in the hippocampus also extends to other brain areas, consistent with the engagement of a distributed network in these representations [20,111,112]. Network-level neural activity patterns underlying generative representations can be coherent across the hippocampus and prefrontal cortex during replays and along the theta rhythm, with some reports of concurrent expression of actual versus alternative location representations across both regions [107,113–117]. Additionally, some generative firing events in the hippocampus are not only coordinated with but also predicted by the activity of cells in the medial prefrontal cortex [70]. Numerous other cortical and subcortical areas also share coordinated firing patterns with the hippocampus, during both replay events and the theta rhythm [67,118–125]. Recruitment of a large network of brain areas during activity related to actual and generative experience appears to reflect brain-wide organization, and the question of how firing patterns in other regions across the brain specifically contribute and respond to generative representations in the hippocampus remains an active area of research [113,122].
How might organized generative neural firing patterns in the hippocampus come about through hippocampal and extrahippocampal processes? This remains largely unknown, but some initial points can be made. First, one would expect generative firing patterns, which do not correspond to immediately ongoing circumstances, to arise primarily from internally driven activity patterns, as opposed neural activity driven directly by external stimuli. Consistent with this, generative events are observed during SWRs and in association with the theta rhythm—and both of these activity patterns are generated internally in the brain (spontaneously) rather than elicited by external stimuli [76,126]. More specifically, SWRs spontaneously occur during sleep in the absence of dynamic sensory stimuli and can be intrinsically generated in isolated hippocampal slices in vitro [76]. Hippocampal theta oscillations arise in vivo in coordination with a rhythm generator region, the medial septum, and can also be generated in isolated rodent hippocampus in vitro [127,128]. Furthermore, late phases of theta, during which generative representations tend to occur, are associated with increased recurrent network activity from within the hippocampus, and relatively weaker influence from cortical areas that are thought to provide multimodal information to the hippocampus [63,67,129,130].
While SWR and theta oscillations are understood to be internally generated and are associated with the occurrence of generative neural firing patterns in the hippocampus, the question of how specific groups of neurons (such as place cells with overlapping place fields) are recruited during generative events remains open [131]. In addition to mechanisms that support SWR and theta generation, it is likely the case that input from brain regions beyond the hippocampus have a role in this process [67]. One possibility is that the activation of particular sets of spatially tuned neurons during generative events is guided by extrahippocampal areas, such as the prefrontal cortex, that are also implicated in the default mode network [20]. This possibility is consistent with evidence that cortical activity can predict generative spiking during theta oscillations several cycles in advance, as well as SWR activity during sleep, and would argue against the idea that hippocampal ensembles are activated by exclusively unstructured input [70,125]. Studies focusing on the internal correlates of generative activity within the brain, over external behavioural correlates, may be especially important to understand what determines the generative neural firing patterns observed in the hippocampus.
The segregation of generative and actual representations in the hippocampus also raises the question of whether the hippocampus further differentiates subtypes of generative representations. For example, are events that reflect veridical experience from the past somehow distinguished from those that reflect constructed alternatives, or those that are predictive of future choices? At the level of neural firing, it remains unclear whether or how the hippocampus might separate these possible representations. However, two points of reference in the human literature offer clues that the relevant neural substrates may be outside the hippocampus. First, patients with hippocampal amnesia can entertain thoughts that distinguish the past or the future, despite impairments in episodic memory [132,133]. Additionally, hippocampal activation during mental simulations without temporal placement versus those specifically set in the future result in similar activation levels in the medial temporal lobe and default mode network [132,134]. These results are consistent with the idea that temporally differentiating representations related to the past or the future may not be hippocampally dependent. Second, healthy human subjects can subjectively discriminate internally and externally derived information, an ability known as reality monitoring [135]. Based on functional imaging studies in both healthy subjects and patients with schizophrenia who experience hallucinations, reality monitoring is thought to rely primarily on prefrontal cortical networks [112]. By contrast, another study reports that hippocampal activation was similar across cases of true and false recognition memory [136], further suggesting that this ability does not strictly rely on the hippocampus. Although probing reality monitoring in rodents is not straightforward, it would be notable if, for example, frontal cortical firing patterns systematically differed based on the representation of possibilities in the hippocampus that reflected veridical experience versus constructed alternatives. Such a result would be consistent with the idea that the hippocampus alone may not distinguish subcategories of generative events, but that the brain may do so via the engagement of prefrontal circuits.
Looking beyond rodents, it remains an open question as to which patterns of generative activity in the hippocampus are shared across species [137]. On the one hand, SWRs have been observed in a range of vertebrates, as have neural reactivation patterns suggestive of replay [138–144]. In humans, replay and replay-like patterns have also been reported, including activity patterns consistent with reactivating prior experience, as well as inferred sequential activity that is not simply recapitulative [145–149]. By contrast to the ubiquity of SWRs across vertebrates, the theta rhythm appears to be more prominent and continuous in the rodent hippocampus than in various other species [137]. A notable example is the bat hippocampus, which shows network-level activity fluctuations that are not generally rhythmic yet still organize place cell firing according to phase [140,150–153]. This may suggest that actual and generative representations can be organized via temporal multiplexing even in the absence of strong rhythmicity. In nonhuman primates and humans, the hippocampal theta rhythm appears to occur in intermittent bouts and at a lower frequency [140,150–153]. Recently, theta phase coding has also been shown in single cells in human subjects [154,155]. In all, these results indicate some conservation across species of the organization of neural firing with respect to network-level hippocampal activity. More generally, they leave open the possibility that the brains of many species temporally multiplex actual versus generative internal representations.