New Insights Into The Specific Roles Of Hippocampus In Navigation

For dwellers in a big city like London, it is not unusual to discover new favourite restaurants, shops or other places of interest in distant, unfamiliar regions. When considering going again, most people would face the problem of choosing the appropriate transportation, therefore a particular factor to think about is how far away the place is. But the challenge does not stop there. Even with the help of public transportation, one has to navigate to the exact whereabouts of the place once dropped into the 'foreign' territory. Initially he might have to rely on the GPS in his mobile phone; but after a few visits, this place would likely to have been marked on his internal 'map', thereby rendering extra aid unnecessary.

This scenario illustrates two important tasks the brain has to perform to allow successful navigation: encoding the location of the goal, and encoding the distance to the goal. The former helps the navigator to know where the goal is, while the latter is important for planning and executing routes to the goal. Together with knowing the current location of the navigator himself, they comprise the three challenges of wayfinding.

Starting with the uncovering of 'place cells'1, more than 3 decades of research on rodent spatial cognition has thrown much light on the potential mechanisms by which mammals know their self location2. There is a large population of cells ('place cells') in the hippocampus, each of which fires predominantly when the animal is within a small field of the local environment (its 'place field'). The currently active ensemble of place cells, analogous to the dot on the GPS, is thought to help determine self location.

However, less is known about how the brain meets the other two challenges. There have been other breakthroughs, including the discovery of 'head direction cells'3 in areas surrounding the hippocampus, which fire when the animal's head is facing a particular direction, and the discovery of 'grid cells'4 in the medial entorhinal cortex, whose firing field consists of multiple vertices covering the entire local environment in a grid-like pattern. The emergence of these cells have fuelled various efforts to model navigation, each assigning prominent roles to different cells. One class of place-cell-centred theory posits that a signal derived from place cell activities represents the navigator's proximity to the goal, and the animal navigates in order to match their currently active cell ensemble with the activity pattern previously encoded at the goal5,6. Such a goal proximity signal has been demonstrated in the hippocampus of rodents7, and two papers discussed in this essay8,9 showed that hippocampus is also a key structure for representing distances in humans.

The other two papers10,11 provide new insights into the representation of the environment. An important theory argues that the place fields essentially constitute a 'cognitive map' of the local environment12,13, which allows flexible navigation such as taking shortcuts or performing detours. Evidence for such a map comes from both rodent lesion studies and human patient studies2. But how could this map be built? It was found that, during sharp wave ripples (SWRs), which reflect transient but high-frequency field potential oscillations of hippocampal neuronal networks while the animal is asleep14 or awake but inactive15, place cells reactivate in a sequence that echoes their sequence of activation in a preceding time. This hippocampal 'replay', which can occur both forward and backward, is thought to be a candidate mechanism for consolidating memory of recent experiences16. However, passive consolidation of experienced trajectories may not be enough to establish a cognitive map, and the two studies discussed here10,11 suggest that reactivation may be a more active process that facilitates the construction of a meaningful map.

Hippocampal representation of distance

The fMRI study by Viard et al.8 used a task set in a virtual reality (VR) rectangular room, where the subject has a first-person view at one end, facing two parallel walls (Fig. 1). Each wall has two gaps and the target, a man, stands behind one of the gaps of the farther wall.

Each trial is composed of an encoding phase, when the subject takes the view, followed by a test phase, when the subject has to indicate which gap in the nearer wall to go through to minimise the distance to the target.

Two factors, memory of the encoding phase and requirement of detour, were manipulated to create four types of trials. Importantly, during the encoding phase, the subject was always viewing from the middle, but during the test phase, the start position was varied, whereby goal proximity is manipulated (Fig. 1). This allowed the authors to investigate which brain regions are sensitive to goal proximity. A control condition was also included, where the correct gate was highlighted for the subject to select, thus minimising the spatial planning involved.

Figure 1. Layout of the VR room and illustration of the starting positions and correct responses in the Viard et al.8 study. The lilac lines represent walls and the red crosses mark the gaps in them. The black figure represents the target man. The digits at the bottoms represent different start positions and the lines represent shortest, hence correct, routes to the target. From Ref. 8.

The key finding was that goal proximity did have an effect on hippocampal activation: bilateral anterior hippocampus displayed greater activation the closer the goal was. Further analysis revealed that neither detour nor memory interacted with goal proximity to produce this effect. Besides, hippocampal activity was enhanced for no-detour trial (shorter distance to goal) compared with detour trials (longer distance to goal), further supporting the finding.

However, as the distance to goal increases, the difference between the correct route and the alternative also decreases, making it more difficult to make a correct response. Hence the observed effects may be attributable to variation in difficulty. Additional analysis where task difficulty was covaried out confirmed the parametric response of hippocampus to goal proximity. The effects of response-switching or novelty were also ruled out.

Morgan and colleagues9 took a different approach. They prepared pictures of 10 landmarks on the University of Pennsylvania campus, and recruited students who are familiar with them for the experiment. During the fMRI scan, single pictures were shown successively to the subjects, who had to press a button to signal that they recognised the landmark. The rationale behind the design was that, for pairs of pictures of different landmarks, the closer the currently visible landmark is to the previous one, the more attenuation in fMRI signal would be observed in the brain area encoding distance. This is a phenomenon known as 'fMRI adaptation'17, which generally refers to the decrease in response induced by repeated presentation of the same stimulus or stimuli with common features, and the brain regions showing this decrease are deemed as where such stimulus or features are represented.

Data analysis revealed that, the smaller the distance between pairs of landmarks, the more pronounced the associated adaptation in the left anterior hippocampus is (Fig. 2, A). Besides, an subjective measure of distance, taken from subjects before the scan, also correlates well with the amount of signal change seen in the left anterior hippocampus (Fig. 2, B). Figure 2. Key results from the Morgan et al.9 study. A. The fMRI response in the left anterior hippocampus is plotted against the real-world distance between successively presented non-identical landmarks; B. The fMRI response in the left anterior hippocampus is plotted against the subjective distance, evaluated by the subjects as the number of minutes needed to walk between pairs of landmarks. From Ref. 9.

This effect was not seen in other hippocampal regions. Also curiously, 10 landmarks should yield 45 pair combinations, thus giving 45 distances, yet only 4 data points were plotted on the graph (Fig. 2). In addition, adaptation related to pairs of pictures showing identical landmarks was not seen in the hippocampus. The authors argued that the rarity of these landmark repetition trials may have triggered novelty processing mechanisms in the hippocampus, thereby reducing or masking any adaptation.

Hippocampal reactivation as an active process

Unlike previous studies of hippocampal replay which have generally used linear tracks, both papers discussed here10,11 deployed a more complex environment, enabling more sophisticated manipulations. Gupta and co-workers10 used a maze that contains two loops (Fig. 3). During training, the rat were rewarded for running through either the left loop, the right loop, or alternating between the two. The reward contingency stayed the same within each training session, but during testing, the reward contingency would be changed approximately halfway through the session, so that the rats could experience both loops within a single session, but the exact experience would vary with varying contingency.

Figure 3. Layout of the maze used in the Gupta et al.10 study. MS = mace start; T1 = turn 1; T2 = turn 2; F1 = feeder 1; F2 = feeder 2. From Ref. 10.

Neuronal activity was recorded from the CA1 field of hippocampus using tetrodes, and spikes during SWRs as the animals stopped at the reward locations were analysed. The results showed that, replays for the distal loop (n = 534) were not infrequent compared with replays for the proximal loop (n = 869), indicating that reactivation is not restricted to the immediate environment. Moreover, the number of backward replays (n = 646) is also comparable with that of forward replays (n = 757). This is interesting because the animals had not actually experienced the backward sequences. Further analysis was performed on the distal replays, and showed that a big proportion of them are 10 min or 15 laps away from the last experience of the corresponding trajectory. Hence replays can happen to temporally distant experiences. Furthermore, replays were not biased by the total preceding experience neither.

Most intriguingly, the authors have identified 19 shortcut sequences in 3 rats. Most of them were very rarely experienced, and 11 of them, all recorded from one rat, were never experienced by the animal, even when training sessions were taken into account. Other possible explanations, such as chance alignment of separate replays, were investigated and found to be very unlikely.

Singer and Frank11 used a maze composed of 6 longitudinal arms jutting out a transverse arm (Fig. 4) and recorded from the CA3 hippocampal subfield in three rats. There were two reward contingencies, and the experiment similarly included switches between them within the same session. This produced a considerable number of unrewarded trials without altering the behaviour too much. SWRs events and replays at the end of rewarded trials were compared with those after unrewarded ones, and there were three key findings: reactivation was significantly enhanced by receipt of reward relative to its absence; when comparing rewarded trials associated with familiar contingency with those during periods of active learning following contingency switch, the latter induced further enhancement of SWRs and reactivation relative to the former; at the location of reward, cell with place field on the track to or from the reward exhibited greater reactivation.

Figure 4. Layout of the maze used in the Singer & Frank11 study. Red dots at the end of longitudinal arms represent feeders. From Ref. 11.

Discussion

The findings presented in this essay concern the role of the hippocampus in navigation. The firs two papers8,9 demonstrated that there is signal in the hippocampus that is sensitive to distances between locations. Other brain regions were also identified to show such sensitivity, although there is not much overlap between the extrahippocampal regions reported in these two papers, probably due the different nature of the tasks used. Notably, medial prefrontal cortex (mPFC) was shown to respond to the parametric variation in goal proximity8, consistent with the findings of a previous study that recruited London taxi drivers to navigate in a VR London and found a positive correlation between goal proximity and mPFC activation18. This study18 did not find parametric effects of goal proximity on the hippocampus, although hippocampal activation was reported for route planning in a related paper19. Whether the hippocampus keeps track of goal proximity during active navigation remains to be elucidated.

With respect to theorizing, the fact that the hippocampus can encode distance lends support to the 'cognitive map theory'8,9, as do the other two papers10,11. If hippocampal reactivation is indeed responsible for consolidating memory, reward-enhanced reactivation shown by Singer and Frank11 would be ideal for encoding goals, while the replay of trajectories over a wide spatial and temporal range as well as the construction never experienced shortcuts could serve to build a cognitive map10. Admittedly, results from one rat10 are not convincing enough, so more evidence is called for. Interestingly, it seems that the data sets from these two papers10,11 could be used to examine each other's claims. Another outstanding question is exactly how spatial goals are represented. Are there 'goal cells'5? Are goals represented as a biased cluster of place fields6?

The Hippocampus is a structure implicated in many cognitive functions, while navigation is a complicated behaviour that demands multiple cognitive faculties. These studies8,9,10,11 help clarify the specific roles the hippocampus plays in navigation, and consequently advanced our understanding of both hippocampal function and navigation.