Execution of new trajectories toward a stable goal without a functional hippocampus

Abstract The hippocampus is a critical component of a mammalian spatial navigation system, with the firing sequences of hippocampal place cells during sleep or immobility constituting a “replay” of an animal's past trajectories. A novel spatial navigation task recently revealed that such “replay” sequences of place fields can also prospectively map onto imminent new paths to a goal that occupies a stable location during each session. It was hypothesized that such “prospective replay” sequences may play a causal role in goal‐directed navigation. In the present study, we query this putative causal role in finding only minimal effects of muscimol‐induced inactivation of the dorsal and intermediate hippocampus on the same spatial navigation task. The concentration of muscimol used demonstrably inhibited hippocampal cell firing in vivo and caused a severe deficit in a hippocampal‐dependent “episodic‐like” spatial memory task in a watermaze. These findings call into question whether “prospective replay” of an imminent and direct path is actually necessary for its execution in certain navigational tasks.

coupled to the computation of a return vector to any start point (McNaughton et al., 2006;Mittelstaedt & Mittelstaedt, 1980). The importance of the hippocampus as a biological substrate for a cognitive map was suggested by the discovery of place cells, and is supported by many causal lines of evidence including lesion, pharmacology and molecular genetic studies (McHugh et al., 1996;Morris et al., 1982Morris et al., , 1986Tsien et al., 1996). Theoretical models of path integration (McNaughton et al., 2006) are supported by data showing that the hippocampus can sometimes be involved in this process (Whishaw, 1998). Other navigational strategies include learning an egocentric orientation from a start point and/or approach to specific landmarks.
Place cell sequences are, however, also present when the animal is awake and performing a navigation task. They can occur concomitantly with theta oscillations (Colgin, 2013;Dragoi & Buzsáki, 2006;Foster & Wilson, 2007;O'Keefe & Recce, 1993;Skaggs et al., 1996;Wikenheiser & Redish, 2015), or be compressed in time during awake SWRs (Foster & Wilson, 2006;Jadhav et al., 2012). In a landmark study by Pfeiffer and Foster (2013), rats were trained to shuttle, each day, between a novel daily "Home" location and multiple "Random" locations in a large arena. Importantly, a minority of SWR-associated trajectory sequences were observed to contain information about future paths from Random to Home locations. Whether the Home locations were really encoded and then recalled in an explicit sense is unclear, but this finding, recently replicated in a similar task on a smaller but more complex arena (Widloski & Foster, 2022), raised the tantalizing possibility that the hippocampus may be involved in the active planning of spatial trajectories toward future goals.
These physiological observations are, however, correlational and observed in a task whose status with respect to involving allocentric spatial memory or involving path-integration is unclear. Notwithstanding these differing representational aspects, the question arises of whether such neural activity in the hippocampus is on the neural pathway causal to the execution of such movement trajectories or, alternatively, whether spatial trajectories computed elsewhere are merely being reported to the hippocampus. To compare these two possibilities, we trained rats on the same spatial navigation task in which future trajectory sequences had been observed (Pfeiffer & Foster, 2013). We first established comparable performance to that observed by them and then inactivated the dorsal and intermediate hippocampus pharmacologically. The aim was to determine whether hippocampal activity during ongoing navigation is indispensable. Various control experiments were conducted in parallel, notably to establish a drug concentration sufficient to block cell firing in the dorsal and intermediate hippocampus of anesthetized rats in vivo and to disrupt performance in an 'episodic-like' allocentric spatial memory task in a watermaze.

| Ethics and reproducibility statement
Growing interest in the replicability of biomedical studies has led us (and others) to be explicit about blinding and other procedures as advised by the CAMARADES consortium (CAMARADES, 2019). The behavioral studies were conducted by experimenters JIR and AJD who were "blind" to the drug infused in any animal. The main study was conducted in two separate cohorts to examine whether comparable data was obtained in each (it was, and the data was combined). The electrophysiological experiment could not be conducted "blind" by AM as the effects of muscimol were so dramatic, but it did include counterbalanced vehicle infusions and all monitoring and measurements were conducted automatically. The animals were handled carefully and all surgery conducted with suitable anes-

| Animals
The subjects were adult male Lister Hooded rats (200-300 g on arrival; n = 73; Charles River, UK) housed in groups of 3-4 for the duration of the study. The experimental cohorts were as follows: Cohort 1: Behavioral study of Home-Random navigation task, n = 10 Cohort 2: Hippocampal drug infusions and acute electrophysiology, n = 17 Cohort 3: Hippocampal drug infusions in Home-Random navigation task, and watermaze delayed matching-to-place task (after surgical implantation of bilateral hippocampal cannulae), n = 15 Cohort 4: Hippocampal or medial prefrontal cortex (mPFC) drug infusions in Home-Random navigation task (after surgical implantation of bilateral hippocampal and mPFC cannulae), n = 13, two animals excluded from mPFC dataset due to off-target implants Cohort 5: mPFC drug infusions in Home-Random navigation task (after surgical implantation of bilateral mPFC cannulae), (n = 14) (Supporting information) Cohort 6: Muscimol diffusion in the mPFC and hippocampus (using fluorescent muscimol bodipy, n = 4; Supporting information) The animals had access to food and water ad libitum and were kept on a 12-h light: 12-h dark schedule (lights on at 6 am; behavioral testing conducted during light phase). For the arena experiments, food was restricted (20-25 g per day) and the animals kept at between 85% and 90% of free-feeding weight. The animals were handled for at least 3 days before the start of all behavioral procedures.

| Behavioral apparatus and tasks: Home-Random navigation task
Arena experiments were conducted in an apparatus based upon that used by Pfeiffer and Foster (2013

| Main task
Daily training on the navigation task itself consisted of continuous running between reward locations. For analysis purposes, this was divided into "trials", each consisting of two "phases" called "Home" and "Random." In the Home phase, the rat would start at a previous Random location and approach the Home location at which liquid reward would be delivered upon arrival. This daily Home location was not marked by any local cue but was one in which the location of reward availability was stable within a session but changed between sessions. In the immediately following Random phase, the animal would leave the Home location to search all over the arena for wherever reward would be delivered, one location having been silently filled with reward while the animal was at the Home location; a Random location was defined as being one of 25 varying locations available across trials on that session, these also being changed between sessions. These two phases completed each successive trial giving two measures of latency and path-length (Random to Home, and Home to Random). Without interfering with the animal, the next trial began immediately, in continuous mode as in the original study of Pfeiffer and Foster (2013). Sessions began with a "Start Box to Random" search path, which was not included in the behavior analyses.

| Behavioral apparatus and tasks: Watermaze task
The watermaze, with associated extramaze cues and overhead video monitoring equipment was used as previously described (Morris, 1984).
Trial 1 (T1) of each session was given as a rewarded probe test, and this was done using an "on-demand" or "Atlantis" platform (Burešová et al., 1985;Spooner et al., 1994).

| Procedure
The procedure of the "episodic-like" delayed matching to place (DMP) protocol in the watermaze task is described in detail elsewhere (Rossato et al., 2018;Steele & Morris, 1999). In this task, the hidden escape platform moves from one location to another between sessions, directly analogous to the moving Home location in the Home-Random arena navigation task. This version of the watermaze is hippocampal-dependent, with effective learning of the new daily platform location requiring the integrity of sufficient tissue in the dorsal hippocampus (Hoz et al., 2005) and its functional integrity with respect to fast synaptic transmission, plasticity and dopaminergic neuromodulation (O'Carroll et al., 2006;Riedel et al., 1999;Steele & Morris, 1999). Performance is typically characterized by a long escape latency on T1 as the animal searches for the platform whose location on that session is still unknown, followed by rapid 1-trial location learning that occurs during the 30 s period on the escape platform which rose to near the water surface after 60 sec swimming. The allocentric memory of where escape was possible on the last session is followed by relatively direct paths to that location on T2-T4 from any start location in the pool, with a small but significant residual memory last through to the next day's session.

| Main task
Four trials per session needed to be used, rather than the 25 trials of the Home-Random navigation task, as asymptote is reached within 2-3 trials. The intervals between trials and escape platform locations were described in detail in Rossato et al. (2018). In this study, there were also 8 sessions of initial training.

| Drug-infusion sessions
As in the Home-Random navigation task, drug-associated sessions followed the 8 training sessions and consisted of 2 counterbalanced drug infusion sessions interleaved with 1 regular training session.
They involved bilateral intrahippocampal infusions of 2 μL of aCSF or muscimol (1.3 mM). The drugs were given 40 min prior to the start of the session.

| Focus of the analyses
The primary measure of performance computed was the time taken (in sec) until the animal first crossed the correct location (12 cm diameter) where the platform would become available (T1) or was available (T2-T4)-the "first crossing latency." We also computed path length which, given stable swimming speeds, is directly correlated with latency. The secondary measure was, during T1 only, the time spent swimming in a virtual zone of 40 cm diameter centered on the location of the platform during the previous session ("24-h memory retention"). This time is normalized and represented as a percentage relative to the 4% area of the pool that the zone occupies. This 4% level represents "chance" if the animal were to be swimming around the pool randomly. Having both measures provided a measure of daily learning within the domain of short-term memory (as in the Home-Random navigation task), and separately a measure of 24-h memory retention.

| Surgery
Guide cannulae for drug infusions were implanted into either the hippocampus or mPFC. Anesthesia was induced using isoflurane (induction, 5%, maintenance, 1%-2%; air-flow, 1 L/min were allowed to recover on a heating pad until normal behavior resumed, and then returned to the vivarium where they were closely monitored over the ensuing days. For all infusions, 2 μL of drug per cannula was infused at 0.5 μL/ min; this is a relatively large volume which risked infusions spreading beyond the target structure, but it was essential to maximize the possibility of intra-regional spread of effect. The infusion cannulas were left in place for a further 2 min to aid drug absorption before being replaced with stylets. The drug infusions were performed 40 min prior to the start of critical test sessions, this interval being based on data from electrophysiological recordings in vivo. The rats were habituated to the experimental procedure of injection for several days before the infusion sessions in order to minimize stress.

| In vivo hippocampal electrophysiology
The aim of the electrophysiology studies was to establish an effective location, dose and volume of muscimol that would successfully inhibit cell firing in the hippocampus in vivo.

| Apparatus
We used a stereotaxic apparatus (

| Extracellular field potential recording
Conventional field potential recordings were made, with stimulation every 20 sec, and these monitored and calculated online using custom software (EPS software; PS). In response to biphasic 200 μs stimulus pulses of circa 600-800 μA, we measured both the early-rising slope of the evoked potential by linear regression over several points, and the amplitude of the evoked population spike in the dentate gyrus.
The stimulus intensity was adjusted to secure initial population spike amplitudes of circa 3-6 mV. Once acquired using suitable electrode placements, potentials typically remained relatively stable over periods of up to 3-4 h, with a small upward drift of the population spike [but not the field excitatory postsynaptic potential (fEPSP)] that rarely exceeded 15% over this long period. Animals for which the potentials were unstable were discarded. The same long time-period stability was observed when aCSF was infused into the dorsal hippocampal formation at a depth targeting stratum lacunosum-moleculare of the CA1 area. A volume of 2 μL at 1.3 mM was infused that, on the basis of previous autoradiographic and electrophysiological data (Morris et al., 1989;Rossato et al., 2018) has been shown to diffuse throughout the entire CA1, CA3 and dentate gyrus regions of the dorsal and intermediate hippocampus.

| Histology and diffusion profile of fluorescently labeled muscimol
We conducted routine Nissl staining of brain sections from animals in the behavioral studies to determine the site at which the implanted cannulae were located in the hippocampus and mPFC. Animals with cannulae implanted off-target (n = 2, mPFC implants) were removed from the analysis.
To identify and confirm visually the site of infusion and spread of maximal concentration of muscimol in mPFC and hippocampus, we infused fluorophore-conjugated muscimol (FCM) from the stereotaxically defined sites of the guide cannulae. We hoped to use this drug to quantify the extent of diffusion. However, its much higher molecular weight session was still relatively new (F 1,27 = 7.42, p < 0.05; pairwise comparisons with Bonferroni correction: early, t 27 = 3.58, p < 0.01; middle, t 27 = 1.14, p > 0.05; late, t 27 = 1.12, p > 0.05). Thus, in the later trials starting after T4, there was no muscimol-associated impairment.
Hippocampal inactivation did not affect the ability of rats to find the reward on the Random phase of a trial (path length, F 1,27 = 2.87, p > 0.05, Figure 3d; latency, F 1,27 = 0.01, p > 0.05). A more detailed presentation of these data is in Figure S3.

| Muscimol blocks performance of an episodic-like spatial memory task in the watermaze
While the electrophysiological data indicates that spiking activity would have ceased in the dorsal and intermediate hippocampus during task performance, we found that it had only a transient effect on the Home-Random navigation task. Faced with this almost "null" result, we turned to a different behavioral task that is definitively hippocampal-dependent-the "episodic-like" delayed matching-toplace (DMP) task in the watermaze. This task requires animals to remember the most recent daily location of the escape platform which also changes from day to day; it is, however, a task which is learned in a definitively allocentric manner. A massive and highly significant deleterious impact of muscimol was observed in this task.
Using again the same animals of Cohort 3 (n = 15), the animals learned to search for the varying location of the hidden escape platform each session during an initial set of 5 sessions (Figure 4a). Thereafter, repeated "blocks" of 3 successive sessions (Session N, N + 1, N + 2; with muscimol only given on the N + 1 sessions; Figure 4b) were given as a counterbalanced within-subjects design to examine the impact of intrahippocampal muscimol or vehicle injections (d) Muscimol infusion impaired recall of the previous session's platform location both on the session of infusion (Session N + 1) and the following session (Session N + 2), without affecting new learning on session N + 2. The impairment on Session N + 2 in the absence of muscimol was due to muscimol having been present on Session N + 1 to block encoding of the Session N + 1's platform location. **p < 0.01, **p < 0.001; means ± S.E.M.
normally on T2-T4 (Figure 4c). Overall, these results demonstrate that the same pharmacological intervention used in the Home-Random task impairs allocentrically encoded recency memory with respect to both learning within the day and recall the next day.

| Lack of effect of medial prefrontal cortex (mPFC) inactivation on arena task
Since the successive trials in the Home-Random task are continuous, that is, occur without any experimenter interference or imposed delay, we hypothesized that, except for the early trials, the task does not engage long-term memory mechanisms and the performance may instead be supported by working memory. Human and animal studies established that the prefrontal cortex is an important node in the working memory circuitry (Curtis & D'Esposito, 2003;Funahashi, 2017). We thus examined the impact of muscimol in the prelimbic region of mPFC, the prefrontal cortical area involved in working memory in rodents (Baeg et al., 2003;Bolkan et al., 2017;Spellman et al., 2015;Yang et al., 2014).
We first trained two cohorts of rats [Cohorts 4 and 5, n = 25; data pooled for analysis] and implanted them with bilateral cannulae directed at mPFC (Figure 5a). Analysis of separate animals established that our muscimol infusions were on-target (Figures S4A-C and 4 infusions in two rats). As in the case of hippocampal inactivation, intra-mPFC muscimol caused no impairment in path length for Home-to-Random paths across trials (F 1,24 = 1.75, p > 0.05; Figure 5d). Importantly, in contrast to hippocampal inactivation, Random-to-Home trajectories were unaffected by muscimol infusion into the mPFC (path length, F 1,24 = 0.61, p > 0.05; latency, F 1,24 = 3.91, p > 0.05; Figure 5b), even if the trials were separated into early (T2-T4), intermediate (T5-T7) and late (T8-T10) (pairwise comparisons with Bonferroni correction: early, t 24 = 0.75; middle, t 24 = 1.06; late, t 24 = 0.81, p > 0.05 for all comparisons; Figure 5c). In the absence of either a working or long-term memory deficit following muscimol in the Home-Random task, it would seem that some other navigational process supports performance.

| DISCUSSION
There are two main findings of this study. The first is that the successful inhibition of hippocampal cell-firing has relatively minimal effect on the Home-Random navigation task reported by Pfeiffer and Foster F I G U R E 5 Inactivation of mPFC with muscimol did not affect the ability of rats to navigate to the Home location. (a) Schematic of bilateral mPFC cannula positions and timeline of the drug infusion and task window. (b and c) mPFC inactivation with muscimol had no effect on the path lengths to the home location. (d) mPFC inactivation had no effect on the path lengths in search of the Random location. n.s, not significant; means ± S.E.M.
(2013). Specifically, despite bilateral hippocampal infusion of muscimol, we observed similar acquisition of the short latency/short path-length asymptote during Home phases of successive trials as they had observed, and the same consistently longer latencies in the Random phases. The early trial adoption of this pattern was, however, slightly but significantly affected by muscimol compared to aCSF vehicle treated animals. The dose and volume of intra-hippocampal muscimol was titrated to yield an inhibition of cell-firing for >2 h, measured electrophysiologically in both dorsal and intermediate hippocampus, with the behavioral tests conducted during a 30 min task-window of minimal cell-firing. The second main finding is that this same dose of muscimol had a devastating effect on an allocentrically encoded delayed matching-to-place watermaze task (recency memory) which was used used as a positive behavioral control for the effectiveness of intra-hippocampal muscimol microinjection. The performance of muscimol-treated rats in the DMP task showed very little improvement over successive daily trials, closely mimicking the effect of complete hippocampal lesions in the same task (Bast et al., 2009).
What is the implication of this dissociation? We first discuss features of the experimental design that support these claims about the results, together with reference to additional findings, and then the key issue of the putative causal role that prospective replay events might be playing in this type of navigational behavior. That prospective replay might be part of a causal path is explicit in the abstract of the Pfeiffer and Foster (2013) paper where they write: "[these sequences are] … supporting a goal-directed, trajectory-finding mechanism, which identifies important places and relevant behavioral paths, at specific times when memory retrieval is required." The heart of our interpretation is to suggest a critical difference between a true allocentrically encoded episodic-like memory task and one involving path integration.

| Sufficient inhibition of spiking activity by muscimol, and its regional diffusion in hippocampus
It is unlikely that insufficient muscimol was infused to disrupt hippocampal spiking activity in the awake freely moving rat during the Home-Random navigation task. Local muscimol infusion into the dorsal hippocampus in anesthetized rats largely abolished evoked spiking activity in dorsal (99% decrease) as well as in intermediate hippocampus (85% decrease). We did not perform single-cell recording of place cells in freely moving rats because published unit recording data is available showing the deleterious impact of muscimol (Bonnevie et al., 2013). For example, even with a lower volume and concentration (0.5 mg/mL, 0.3 μL; 1.5 nanomoles compared to our use of 2.6 nanomoles), there was a complete loss of place cell firing in the dorsal hippocampus adjacent to the infusion as well as a loss of grid-like tuning by grid cells in the medial entorhinal cortex (MEC) (Bonnevie et al., 2013). The concentration used in our study was 2.5 times higher with a volume 7 times larger (in nanomolar terms, circa 17 times higher). Bonnevie et al. (2013) also mapped the spread of muscimol (FCM) revealing a spread largely limited to the hippocampus and the cannula track. Our FCM mapping data also revealed infusion into the targeted site, with the drug remaining largely in hippocampus (and, in separate animals, in mPFC). However, the spread of the high molecular weight FCM underestimates the spread of the normal "non-conjugated" muscimol used in the main experiments. The evidence for this claim is the robust blockade of evoked cell-spiking activity observed in the intermediate hippocampus where FCM fluorescence was not detected. The functional impact, rather than mere diffusion, of an antagonist of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionate)/kainate-type of glutamate receptors has also been studied using 2-deoxyglucose mapping to reveal widespread hypometabolism (23% reduction) throughout the dorsal and intermediate hippocampus (Riedel et al., 1999). Deleterious effects on watermaze learning and memory consolidation were observed in this latter study. However, in that case, while the site of intrahippocampal infusion was comparable While we have not investigated whether muscimol infusions successfully inactivated the ventral-most part of the hippocampus, we deem it unlikely that ventral hippocampus is capable of supporting navigation to the learned Home location to a higher degree than its dorsal counterpart. Ventral hippocampus has classically been associated with processing emotional, affective and/or contextual information (Bannerman et al., 2003;Fanselow & Dong, 2010). Its contribution to spatial navigation tasks is a matter of dispute (Bast et al., 2009;Fanselow & Dong, 2010;McDonald et al., 2018;Moser et al., 1995;Ruediger et al., 2012). Ventral hippocampal place cells have larger, less defined place fields (Jung et al., 1994;Royer et al., 2010) and show less theta rhythmicity (Royer et al., 2010) than their dorsal counterparts, which makes them less suited for accurate representation of spatial trajectories but perhaps sufficient to achieve context discrimination. Moreover, in awake rats, sharp-wave ripples (SWRs) in ventral hippocampus occur independently of those in dorsal hippocampus and are functionally coupled to separate less spatially selective downstream targets (Sosa et al., 2020).
Taken together, the lack of an effect on the Pfeiffer and Foster

| Superficial similarity of the Home-Random navigation and DMP watermaze tasks
In the Pfeiffer and Foster (2013) task, the animals shuttle continuously between Home and Random locations, with the fixed Home location only changed across sessions. In the DMP watermaze task, the animal swims from any of four separate start positions to a hidden escape platform whose location is also changed across sessions. On the face of it, the tasks have a conceptual similarity and both appear to involve spatial recency memory.
We shall argue, however, that this apparent conceptual similarity is only superficial. Numerous lines of data indicate that watermaze spatial learning is allocentric and involves some "explicit" memory of a location in space identified with reference to extramaze cues. The animals climb onto the platform and look around to see where they are.
Importantly, they do not swim back to the starting point of any trial (which changes across trials). Even in situations such as the present watermaze training protocol where the wider extramaze context of learning had been learned prior to any drug infusions, the encoding of that session's location of safety within this allocentric framework is known to involve the hippocampus and associated structures (Steele & Morris, 1999). Using the DMP task, we observed both a complete failure to learn the new session's escape location in the presence of muscimol or to recall the previous session's location the next day in its absence. From this, we infer that the dose was enough to prevent both new spatial memory encoding in the presence of the drug and subsequent spatial memory recall. In terms of the longitudinal axis of the hippocampus, the DMP task is exquisitely sensitive to both lesions and AMPA receptor blockade restricted to the dorsal and intermediate hippocampus (Hoz et al., 2005;Riedel et al., 1999), and the effect of muscimol in our DMP experiment was essentially the same as the effect of complete hippocampal lesions and much more severe than the effect of partial lesions (Bast et al., 2009).
Still, it is of note that we did observe a modest impairment in the initial learning of the new Home location in the early trials of the Home-Random task. The animals with an inactive hippocampus required a few extra trials to match the asymptotic performance of the control group. This is consistent with the observed failure to remember the new platform location in the DMP task on subsequent trials of the same day. The hippocampus thus does play some role in the initial rapid learning of the new goal location, but this function seems to be orthogonal to the subsequent ability to navigate back to the goal once its location has been encoded. Pfeiffer and Foster's (2013) beautiful data show unambiguously that prospective replay is playing out a representation of future trajectories in a brain structure implicated in spatial navigation. Their supposition was that this matters causally for accurate spatial navigation from Random to Home. But is successful navigation in their Home-Random navigation task really an allocentric task that is dependent on the hippocampus? Our data suggest not.

| The Home-Random navigation task is likely a path integration task
Understanding the difference between the Home-Random navigation task and the DMP watermaze task is the critical issue here, but first a word about what "hippocampal-dependence" means. This term in widespread use was first defined in terms of the sensitivity of a task to lesions, but other interventions such as drugs, regional-specific genetic manipulations, and both optogenetic and chemogenetic procedures are displacing the older lesion criteria in new ways. For example, some drugs such as N-methyl-D-aspartate (NMDA) receptor antagonists impair spatial memory encoding without effect on retrieval (Steele & Morris, 1999). This is still a hippocampal-dependent profile, but different from that obtained with lesions. Very low concentrations of muscimol can, under some circumstances, impair retrieval without any effect on encoding, a finding consistent with dendritic computation (Rossato et al., 2018), but such concentrations are an order of magnitude lower than those used here for which muscimol impairs both new learning and episodic-like recall. Other interventions such as intrahippocampal blockade of AMPA receptors impair both encoding and retrieval of other allocentric spatial tasks (Riedel et al., 1999).
Interestingly, the supposition that path integration is "hippocampaldependent" was made primarily on the basis of the impact of lesions of the fornix (Maaswinkel et al., 1999;Whishaw, 1998;Whishaw et al., 2001), but it is not supported by hippocampal lesion studies (Alyan & McNaughton, 1999). Hippocampal involvement in path integration has not, to our knowledge, yet been investigated in animals using more contemporary approaches that intervene directly with hippocampal physiology, but study of patients with definitive hippocampal damage has revealed no impairment of path integration (Shrager et al., 2008).
We doubt the Home-Random navigation task, at least in its plateau phase, is either episodic or allocentric, whereas the DMP watermaze task is an episodic-like task as usually defined (Clayton et al., 2003). This is because the animal remembers an event (escaping the water-what) at a specific location (where) in a familiar allocentric framework and, specifically, the most recent occasion that this happened (when). As discussed in detail by Steele and Morris (1999), recall over trials T2-T4 involves remembering where this "event" happened most recently, followed by a natural tendency to head to the place where it happened (because navigation to that location had consistently brought safety). Disrupting hippocampal cell firing with muscimol should, as shown here, be devastating for such recency-recall within the session. A modest but significant memory of the most recent learned location lasts 24 h (in vehicle treated controls), can be measured and is also impaired in rats treated with muscimol on the previous day. Another facet of the DMP watermaze task is that trials start and stop and have experimenter involvement in carrying the animals, whereas the successive phases and trials of the Home-Random navigation task are continuous. This discontinuous versus continuous issue is relevant to path integration, and likely also relevant to a new navigational honeycomb maze task in which trials are discontinuous but the animal must, in successive phases of choice opportunity, approach a learned goal appropriately . We predict that the honeycomb task also would be severely impaired by intrahippocampal muscimol injection.
In the Home-Random navigation task, in contrast, finding liquid reward at a specific location in this continuous task likely does not constitute an event in quite the same way. This assertion may seem self-serving for our interpretation, but bear in mind that with up to 25 trials per session, the reward will be found in every session in one place 25 times and in numerous other places one or more times each.
In short, reward is potentially available almost everywhere. Such a training protocol is ideally suited to path integration solution (McNaughton et al., 2006). The idea is that a brain system keeps track of and accumulates the animal's translational and rotational movements during the Random phase of each trial, and then uses the accumulated vectors to compute a direct path back to the starting position (in this case, the Home location). The accumulator is then reset.
Importantly, in such a system, the animal never learns or needs to learn where the starting position is located in an allocentric sense. It may know it on the basis of other aspects of experience, but accurate paths do not require explicit spatial knowledge. With respect to the Home-Random navigation task, the accumulator would have operated primarily during the Random search phases, with the return vector applied to get back to the start point of the random-search. There is a longstanding debate about whether such path integration involves the hippocampal formation , with some data suggesting that it does (Maaswinkel et al., 1999;Whishaw, 1998;Whishaw et al., 2001) and others implicating instead the MEC (Banino et al., 2018;Burak & Fiete, 2009;Campbell et al., 2018;Tennant et al., 2018), the retrosplenial cortex (Elduayen & Save, 2014;Mao et al., 2020) and/or the head-direction system (Butler et al., 2017;Valerio & Taube, 2012). Our muscimol infusions did not spread to the MEC, but similar muscimol infusions into the dorsal hippocampus have previously been shown to disrupt the periodic firing of grid cells (Bonnevie et al., 2013), indicating that entorhinal cortex is unlikely to support the Home-Random differences in latency and path-length (see also Shrager et al., 2008). The lack of effect of mPFC inactivation observed in this study indicates that prefrontal areas are also unlikely to serve as an anatomical substrate for path integration. Nevertheless, a path integration system, wherever located anatomically, might report the prospective trajectory to the hippocampus and this could be the basis of the SWR-associated prospective replay sequences representing future trajectories. In our view, there are two separate points: there need be no causal link between these place cell sequences recorded in the hippocampus and the executed trajectories to the goal; and there is also no need for explicit allocentrically coded declarative memory of the Home location in the Home-Random task.
It is worth noting, however, that the recent modification of the Home-Random task that includes intramaze barriers (Widloski & Foster, 2022) necessitates adoption of indirect trajectories to the goal location and may be more dependent on the SWR-associated sequences. Establishing whether the addition of barriers makes the Home-Random task hippocampal-dependent is of future interest.
An alternative approach to explain the animal performance in the task is via Pavlovian place conditioning. That one location in the arena has reward multiple times and others only once per session, could nonetheless have served to increase the reward value of a particular area of the arena, with this location varying across sessions. The animals switch repeatedly between (a) directed approach to the Home location where they get reward multiple times (high-value), and (b) a search strategy to diverse Random locations (low-value), where they also get reward but only once for each location. The learning of value would then take place gradually using place conditioning. On this account, and in addition to path integration, the gradual increase in relative value of the Home location over 3-5 trials at the start of each session might have helped develop home path directionality, but not critically. Interestingly, a temporal-difference conditioning algorithm for mediating directed approach to a valued location was developed by Foster et al. (2000) that does not appeal to declarative event memory but may nonetheless be relevant.
When we began this series of experiments, we had in mind to explore precisely timed optogenetic inactivation of the SWRs in the hippocampus that constituted the prospective replay. The notion was that halorhodopsin or ArchT disruption of the prospective path from Random to Home as represented by place-cell sequences would reduce or even block the Home-Random difference in latency and path length. The analysis just presented reveals that such a study is misconceived and, to the contrary, our analysis leads us to make the opposite prediction. Were it possible to direct exquisitely timed optogenetic inactivation throughout the Home to Random phase of a trial to a sufficient volume of the relevant brain structure(s) performing the path integration computations (wherever these are anatomically), only then would the faster latency to Home compared to Random, and lower path length, be reduced or abolished. In contrast, optogenetic disruption of SWRs at the start of the Home phases should be without effect. Such an experiment is, unfortunately, beyond the scope of our study.

| The place of prospective replay in goal-directed navigation
The analysis presented suggests, ironically, that animals in the Home-Random navigation task never learn explicitly where Home is located each session, they merely execute paths computed using path integration to get them there accurately, aided modestly by some Pavlovian place conditioning of a non-declarative nature. It is, therefore, worth pausing to reflect on Figure 1d; the differential pattern of the paths is striking; the radical claim being made here is that the animals need not to remember the Home location in any explicit sense. Newly learned goal locations can, nonetheless, be incorporated into the brain's cognitive map on multiple levels, including enriched representations of the goal in hippocampal place cell codes (Dupret et al., 2010;Hok et al., 2007;Hollup et al., 2001;Markus et al., 1995) and entorhinal grid cell codes (Boccara et al., 2019;Butler et al., 2019). This updating of the cognitive map with a new goal location, likely taking place in the early Random-Home trials in the present study, could perhaps be aided by plasticity-promoting neuromodulatory inputs to the hippocampus McNamara et al., 2014;Takeuchi et al., 2014Takeuchi et al., , 2016. In bats, hipocampal cells can also encode distance and direction toward the goal (Sarel et al., 2017). The presence of the animal at a goal location is associated with awake SWR-associated hippocampal replay events that include (Davidson et al., 2009;Diba & Buzsáki, 2007;Foster & Wilson, 2006;Pfeiffer & Foster, 2013), but are not limited to, representations of immediate past and future trajectories (Gupta et al., 2010;Karlsson & Frank, 2009). These findings highlight the possible importance of awake replay in ongoing navigation to and from goal locations. Interestingly, in a spatial memory task involving both a working memory component encoded in one trial and a long-term spatial reference memory component encoded over many trials, awake SWRs depict trajectories associated with both components (Karlsson & Frank, 2009), but their disruption (Jadhav et al., 2012) or prolongation (Fernández-Ruiz et al., 2019), respectively, impaired or enhanced only the working memory component, indicating that not all functions of awake SWRs are causally related to ongoing long-term spatial memory.
The veracity of Pfeiffer and Foster's (2013) exciting results remains unchallenged by our findings. We did not monitor the single cell activity of hippocampal neurons during our investigation of the Home-Random navigation task, and we recognize the importance of such prospective replay as indicating the likely output of a path integration system located outside the hippocampus. However, we did explore the dataset of Bonnevie et al. (2013) only to find that the electrode placement used for the purposes of their study was inappropriate for examining SWRs. An intact and functioning hippocampus gets to know and may even "need to know," but we question the likely causal role of such reported sequences in the hippocampus in planning subsequent navigation. We are therefore obliged to end with an important qualification. Such a report to the hippocampus may not be essential for accurate navigation in this path integration task, but it may be very important for true allocentric tasks involving explicit memory (Jadhav et al., 2012;Kim & Frank, 2009) whereby it may query and/or consolidate the cognitive map via replay of a repertoire of possible trajectories within present and past environments (Gupta et al., 2010;Karlsson & Frank, 2009). To test these hypotheses, concomitant high-density place cell recording or calcium imaging to identify and possibly disrupt specific prospective replay sequences during a definitively allocentric task would constitute a valuable follow-up to the present work.