Rapid and Independent Memory Formation in the Parietal Cortex

Home , Hippocampal formation, Recognition memory

Rapid and independent memory formation in the parietal cortex

Svenja Brodta, Dorothee Pöhlchena, Virginia L. Flanaginb,c, Stefan Glasauerb,c,d, Steffen Gaisa,c,1, and Monika Schönauera,c

aMedical Psychology and Behavioral Neurobiology, Eberhard-Karls-Universität Tübingen, 72076 Tuebingen, Germany; bGerman Center for Vertigo and Balance Disorders, University Hospital Munich, 81377 Munich, Germany; cBernstein Center for Computational Neuroscience, Ludwig-Maximilians- Universität München, 82152 Planegg-Martinsried, Germany; and dCenter for Sensorimotor Research, Department of Neurology, University Hospital Munich, 81377 Munich, Germany

Edited by Larry R. Squire, Veterans Affairs San Diego Healthcare System, San Diego, CA, and approved September 30, 2016 (received for review April 22, 2016)

Previous evidence indicates that the brain stores memory in two and neocortex, as well as increased intracortical connectivity of complementary systems, allowing both rapid plasticity and stable representational areas (6). representations at different sites. For memory to be established in A region that is currently receiving increased attention in a long-lasting neocortical store, many learning repetitions are memory research is the posterior parietal cortex (PPC) (9). Al- considered necessary after initial encoding into hippocampal though many fMRI studies have shown activation of the PPC circuits. To elucidate the dynamics of hippocampal and neocortical during memory tasks (10), it has only recently been confirmed contributions to the early phases of memory formation, we closely that there is a genuine contribution of the PPC to episodic followed changes in human functional brain activity while volun- memory that goes beyond attentional processes (11). Studies on teers navigated through two different, initially unknown virtual functional connectivity identifying a hippocampal-parietal memory environments. In one condition, they were able to encode new network (12), together with findings that posterior parietal areas information continuously about the spatial layout of the maze. In can maintain material-specific working memory representations the control condition, no information could be learned because the (13), suggest that the PPC might be involved in hippocampal- layout changed constantly. Our results show that the posterior neocortical interactions during memory formation. parietal cortex (PPC) encodes memories for spatial locations The domain of spatial navigation offers great possibilities to investigate the gradual process of building a complex memory rapidly, beginning already with the first visit to a location and representation. Similar to other domains of memory, lesion and steadily increasing activity with each additional encounter. Hip- neuroimaging studies in animals and humans indicate that spatial pocampal activity and connectivity between the PPC and hippo- learning and memory primarily rely on the hippocampus (14, 15) campus, on the other hand, are strongest during initial encoding, but can, at some point, become independent (16). In a longitu- and both decline with additional encounters. Importantly, stron- dinal study, the hippocampus was only activated during mental ger PPC activity related to higher memory-based performance. navigation when subjects had just moved to a new city, but not Compared with the nonlearnable control condition, PPC activity in when they had already lived there for 1 y. In the latter case, the the learned environment remained elevated after a 24-h interval, subjects relied instead on neocortical areas like the retrosplenial indicating a stable change. Our findings reflect the rapid creation cortex (RSC) and the PPC (17). Indeed, in their metaanalysis, COGNITIVE SCIENCES of a memory representation in the PPC, which belongs to a Boccia et al. (18) found activation in the hippocampus only for PSYCHOLOGICAL AND recently proposed parietal memory network. The emerging pari- studies investigating recently learned environments but not for etal representation is specific for individual episodes of experi- highly familiar surroundings. Studies in rodents support these ence, predicts behavior, and remains stable over offline periods, findings by showing a decrease in metabolic activity and expression and must therefore hold a mnemonic function. of immediate early genes in the hippocampus following retrieval after 4 wk, whereas, at the same time, these measures increased in long-term memory | posterior parietal cortex | precuneus | memory systems several neocortical areas (5). Together, these findings indicate that consolidation | virtual reality Significance earning enables adaptive and effective interaction with the Lenvironment based on past experience. How this essential ca- When it comes to memory, our brain needs to deal with two op- pability of the brain to encode, store, and later retrieve new in- posing demands: remaining plastic to acquire new information formation is implemented on the systems level has been the focus rapidly and maintaining old information faithfully over long pe- of many studies. However, although there is consistent evidence riods. It is assumed that two distinct brain systems are in charge of that specific brain regions are involved in learning and memory, the these functions: the fast-learning hippocampus and the slow- interactions between these regions and their temporal dynamics learning neocortex. The interaction between these two systems has remain unclear. For declarative memory, one influential model traditionally been seen as very slow, requiring weeks or months to proposes complementary roles of the hippocampus and neocortex build a neocortical memory trace. In this study, brain activity during in supporting memory representations (1–3).Itassumesthatthe virtual-reality navigation shows that contributions of hippocampus highly plastic hippocampus serves as a fast learner, transiently and parietal neocortex to memory are changing substantially al- storing newly encountered information. Later on, this information ready at the time a spatial memory representation is built. Notably, is gradually integrated into more stable neocortical networks (4). we show that the posterior parietal cortex fulfills all criteria for a Many experiments in animals and humans have confirmed de- hippocampus-independent memory representation. creased hippocampal but increased neocortical contributions to memory retrieval with longer consolidation intervals (5–7). Con- Author contributions: S.B., V.L.F., S. Glasauer, S. Gais, and M.S. designed research; S.B., D.P., cerning the time frame during which hippocampal independence of and M.S. performed research; S.B., D.P., and M.S. analyzed data; and S.B., V.L.F., S. Glasauer, amemoryisestablished,accountsdivergewidely.Inthecaseof S. Gais, and M.S. wrote the paper. patients with medial temporal lobe (MTL) damage, retrograde The authors declare no conflict of interest. amnesia is reported to have a temporal gradient spanning years or This article is a PNAS Direct Submission. decades (8). However, other studies describe a decrease in hippo- 1To whom correspondence should be addressed. Email: [email protected]. campal activation as early as 1 d after learning (6, 7). These changes This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. were paralleled by decreasing connectivity between the hippocampus 1073/pnas.1605719113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1605719113 PNAS | November 15, 2016 | vol. 113 | no. 46 | 13251–13256 Downloaded by guest on October 1, 2021 show a gradual increase of activity over trials in the static environment that did not similarly occur in the random environment. The interaction between the factors environment (static/random) and trial should therefore represent those brain areas that develop a spatial representation, whereas it controls for changes related to moving through the maze and to the passage of time. We found significant clusters of activity predominantly in the bilateral precuneus, as well as a few smaller activations in other parietal and frontal areas (Fig. 2A and Table S1). The mean beta values for the precuneus clusters increase steadily over learning trials in the static condition. No such increase could be observed over repeated navigation trials in the random maze (Fig. 2B). This increase in precuneus activity is mirrored by the overall increase in performance over trials [Fig. 2B, orange line; r(58) = 0.55, P < 0.001]. The beginning of session 3, on the second day, reveals a drop in precuneus activity as well as in navigation performance. However, when comparing the first trials of sessions 1 and 3 in the static and random conditions, it becomes obvious that the precuneus response to the first trial of the static maze on the second experimental day is already higher than on the first day (t26 = −2.19, P = 0.037) and also higher than the precuneus response of the corresponding random condition (t26 = 2.33, P = 0.028). Learning-dependent changes in precuneus response therefore persist between experimental days (additional analysis is provided in SI Results and Discussion)(seealsoFig. S1 and Table S2). The time-based analyses above implicate the precuneus in the Fig. 1. Properties of the maze. (A) Layout of the virtual labyrinth in the static development of a memory representation in the static maze condition. Locations of the five objects are depicted as black dots, and junctions condition. Because junctions in a maze are the locations where are marked with red triangles. (B) View of a junction in the static condition from memory-based decisions have to be made, we compared brain the participant’s perspective. (C) Layout of the virtual labyrinth in the random condition. All pathways within a hexagonal structure were theoretically possi- activity at these decision points with activity while traveling along ble, but some were blocked at random. The experience for the participant in a corridor without junctions. Again, we found a very similar the random condition was identical to the experience in the static condition, network of brain areas as in the comparison between static and because wall locations and wall textures were changed constantly but only variable conditions, featuring the precuneus and some other outside the field of view of the subject (red circle). (D)Samplewalltextures. parietal and frontal areas (Table S3). To assess the dynamics underlying learning-related changes in more detail, we investigated the effect of repeated encounters independent neocortical memory representations may be formed with specific spatial locations. For this analysis, we compared more rapidly than previously expected. brain responses during initial and later encounters with a given Although many studies focus on the retrieval of an already location and modeled a linear change over repetitions. To avoid learned environment, in this study, we targeted memory processes bias, these preplanned analyses were based on the anatomically during the acquisition period. We aimed to identify areas involved defined regions (Fig. S2). We found higher activation in the during the establishment of a new memory representation and to precuneus and RSC during late compared with initial encounters examine the changes in brain activity over the course of the initial (Fig. 3A and Table S4). The mean beta values over all precuneus β = t = P < R2 = encounters with this information. In an MRI scanner, we let par- voxels increased linearly ( 0.96, 16 13.32, 0.001, ticipants navigate through a completely unknown, complex virtual environment and find specific objects over an extended amount of time. Because we were particularly interested in those brain regions specifically involved in the storage of spatial memory, and not in navigating through an environment per se, we designed two conditions. In the experimental condition, participants were able to encode new information continuously about the spatial layout of themaze(“static maze”; Fig. 1). In the control condition, no information could be learned because the maze layout changed constantly outside the field of view (“random maze”). In all other respects (e.g., construction of the maze, optic flow, parameters of movement), the conditions were identical (Movie S1). Our main interest in this study was to elucidate the dynamics of hippocampal and neocortical contributions during spatial learning, with a special Fig. 2. Effect of number of trials in the static condition vs. the random con- focus on the relation between both systems. We hypothesized that dition. (A) Precuneus shows a higher increase in activation with an increasing the hippocampus would be required mainly for the initial encoding number of trials for the static condition compared with the random condition. of new information, and that its influence should diminish over the Highlighted clusters exceed 10 voxels and exhibit significant peak-level effects – < course of learning. Instead, neocortical, probably parietal, regions at a family-wise error corrected probability threshold (PFWE) 0.05. No masking should gradually support the spatial representation. was applied (also Table S1). (B) Precuneus activation increases with the number of trials only in the static condition, but not in the random condition. Precuneus Results activation increase in the static condition parallels the percentage of target objects found in each trial (orange line). Trials 1–30 occurred on day 1, and trials Activation Changes over Repeated Encounters. The main goal of this 31–60 occurred on day 2. Mean beta values of all voxels in an anatomical study was to track brain activity during the development of a precuneus region of interest that were significantly activated in A are displayed. spatial representation. Subjects were navigating alternately through a Error bars indicate SEM. Linear regression for the static condition was significant 2 learnable virtual environment (static maze) and nonlearnable virtual (β = 0.76, t59 = 8.85, P < 0.001, R = 0.58), whereas it was not significant for the 2 environment (random maze). We first looked for brain regions that random condition (β = −0.24, t59 = −1.84, not significant, R = 0.06).

13252 | www.pnas.org/cgi/doi/10.1073/pnas.1605719113 Brodt et al. Downloaded by guest on October 1, 2021 memory performance during navigation. We used four different analyses to test for a relation between memory performance and brain activity. First, we considered the percentage of target objects found within a session. We find that participants form a memory of the static maze over time: performance in the static condition increased linearly from session 1 to session 4 (F1,26 = 10.39, P = 0.003; Fig. 5A and Table 1). Learning occurred continuously over all sessions because learning rates did not differ between the two experimental days (t26 = −3.96, P = 0.695). The random condition showed no such trend (F1,26 = 0.73, P = 0.400; interaction static/ random × session: F1,26 = 4.36, P = 0.047; Table 1). Although both good and bad learners show similar activity in the static maze compared with the random maze (Table S7), comparing brain activity between good and bad learners based on a median split revealed significantly higher activity in the precuneus in good learners in the interaction good/bad × static/random (Fig. 5C and Table S8). Analysis of corresponding beta values showed higher precuneus activation in the static condition in good performers compared with bad performers (t25 = 3.14, P = 0.004; Fig. S3A). For a second analysis, we calculated a route bias score in the static condition. This score shows whether the preference for certain routes in the maze changes over learning. Over repeated encounters, some routes should be preferred over others if a spatial rep- Fig. 3. Effect of repeated encounters with locations. (A) Precuneus shows resentation is built. This preference changes the distribution of path greater activation in the last encounter compared with the first encounter with a use frequencies. Behaviorally, route preference indicated significant location in the static condition. Highlighted clusters exhibit significant peak-level learning over sessions in the static condition in an ANOVA (F3,78 = < effects at PFWE 0.05 (also Table S4). (B) Precuneus activation increases with the 5.01, P = 0.003), with route bias increasing linearly over the sessions number of encounters. Mean beta values of all voxels in an anatomical pre- (F1,26 = 11.35, P = 0.002; Fig. 5B and Table 1). Performance was cuneus region of interest (ROI) over encounters are displayed. Error bars indicate related to increased blood oxygen level-dependent signal intensities β = = < 2 = SEM. Linear regression was significant ( 0.96, t16 13.32, P 0.001, R 0.92). in the good/bad × static/random interaction, again bilaterally in the (C) Hippocampus shows greater activation in the first encounter compared with same region of the precuneus (Fig. 5C, Fig. S3B,andTable S8). the last encounter with a location in the static condition. The highlighted cluster < There was no difference in connectivity between the precuneus and exhibits a significant peak-level effect at an uncorrected P value (Puncorr) 0.001 hippocampus over encounters between good and bad learners (also Table S4). (D) Hippocampus activation decreases with the number of en- (based on percentage of targets found: t25 = −0.14, P = 0.887; based counters. Mean beta values of all voxels in an anatomical ROI of the left hip- t = − P = pocampus over encounters are displayed. Error bars indicate SEM. Linear on route bias: 25 0.07, 0.995). 2 Above, we reported higher precuneus activity for junctions vs. regression was significant (β = −0.70, t16 = 3.78, P = 0.002, R = 0.49). All highlighted clusters exceed 10 voxels, and no masking was applied. corridors. This activity can be due to memory retrieval while making navigational decisions or to higher familiarity with the COGNITIVE SCIENCES junctions, which are encountered more often than the corridors. PSYCHOLOGICAL AND 0.92; Fig. 3B). Most strikingly, in contrast to this increase in Similarly, subjects with a higher route bias score, based on their activity with repeated encounters, the opposite pattern emerges recall of the maze, choose certain routes more and more often in the hippocampus. Comparing the initial with later encounters over sessions, but they also encounter certain corridors more yielded higher activity in the left hippocampus [uncorrected often, which increases familiarity. Therefore, as a third measure P value (Puncorr) < 0.001; Fig. 3C and Table S4]. Opposing the of performance, we looked at brain activity when participants trend of precuneus activity, the mean beta values over the reached a junction and made a correct choice compared with an left hippocampus decreased linearly with encounters (β = −0.70, 2 t16 = −3.78, P = 0.002, R = 0.49; Fig. 3D). Comparing the course of hippocampal and precuneus activity on a single-subject level yielded a significant negative average correlation over all subjects (r = −0.23, t26 = −2.66, P = 0.013), indicating that en- gagement of these two regions follows an opposing course. Finally, to assess the interaction between the precuneus and hippocampus over repeated encounters with a location, we com- puted a multilevel psychophysiological interaction (PPI) analysis, which represents the activity commontobothareasduringnavi- gation. Linear regression on the results of this PPI analysis revealed that functional connectivity declines linearly with repeated en- 2 counters (β = −0.65, t16 = −3.30, P = 0.005, R = 0.42; Fig. 4). Together, our data speak for the rapid build-up of an independent spatial representation in the precuneus over the first encounters with a novel location. The hippocampus, although initially active and functionally connected to the precuneus, shows a gradually dimin- ished contribution over encounters (additional analyses on the role of the hippocampus and other regions are provided in SI Results and Discussion and Tables S5 and S6). Overall, our analyses confirm that precuneus activity increases during the development of a spatial Fig. 4. Connectivity between precuneus and hippocampus with repeated representation, initially supported by the hippocampus. encounters. Coactivation of precuneus and hippocampus decreases with increasing number of encounters. The beta values represent the average Behavioral Results and Performance-Dependent Effects. Because the values over all anatomical hippocampal voxels in a PPI analysis with the present study aimed at finding correlates of spatial memory repre- precuneus as a seed region. Error bars indicate SEM. Linear regression was 2 sentations, we tested whether brain activity was related to spatial significant (β = −0.65, t16 = 3.30, P = 0.005, R = 0.42).

Brodt et al. PNAS | November 15, 2016 | vol. 113 | no. 46 | 13253 Downloaded by guest on October 1, 2021 Fig. 5. Performance and precuneus activation. (A) Mean percentage of correct trials in which the target object was found within the time limit increased linearly over the four experimental sessions

(F1,26 = 10.39, P = 0.003). Error bars indicate SEM (also Table 1). (B) Route bias based on the distribution of frequencies with which each of the 79 corridors was traveled in a session when sorted in descending order. Gray bars show the mean frequency for each corridor for session 1. Mean logarithmic fits for sessions 1 (solid line) and 4 (dashed line) and time constants for every session (vertical black bars) are displayed. Decreasing time constants

on the x axis indicate that route bias increases with every session (F1,26 = 11.35, P = 0.002). (C) Precuneus shows overlapping performance-dependent activation in four different analyses (white). Higher precuneus activation was found in good navigators compared with poor navigators based on the percentage of correct trials (red) as well as on the route bias score (blue). Similar precuneus activity was found at junctions in the static maze when the correct decision in terms of the optimal path leading to the target was made (green), as well as when comparing brain activity at the beginning of successful and unsuccessful

trials (yellow) (also Fig. S3 and Tables S8–S10). All four analyses revealed peak activity in the precuneus of PFWE < 0.05 full-volume corrected and are displayed at Puncorr < 0.001. Dorsolateral prefrontal cortex activity, in contrast to activity in the precuneus, did not reach family-wise error–corrected thresholds (also SI Results and Discussion). All highlighted clusters exceed 10 voxels. No masking was applied.

incorrect choice. We found higher bilateral precuneus activation Furthermore, activity in the PPC is able to represent category-spe- at junctions in the static maze when the optimal path was chosen cific information during a delay period after category information (Fig. 5C and Table S9). has been learned (21). It is also one of very few regions in the brain Finally, to investigate brain regions that predict the subsequent that represents memory accuracy in human fMRI when memory outcome of a trial, we analyzed brain activity only at the begin- confidence is held constant (22). Moreover, GABAergic and gluta- ning of each trial, when the subjects were presented with their matergic signatures of memory exist in the PPC, which speak to its starting location and target. Again, the interaction analysis correct/ role in episodic memory (23). Together, these and other recent incorrect × static/random yielded a significant effect bilaterally in findings led to the proposal of a parietal memory network, which is C the precuneus (Fig. 5 and Table S10). This effect stems from a anatomically distinct from the spatially adjacent default-mode net- greater activation of the precuneus during the initial 5 s of a trial work (9, 24). This network increases its activity with stimulus repe- for correct (target found) compared with incorrect (target not tition and represents familiarity. Even though it is therefore a good found) trials specifically in the static condition (t26 = 4.47, P < C candidate for long-term memory storage, the mnemonic nature of 0.001; Fig. S3 ). Therefore, precuneus activity predicts successful activity in the parietal cortex is heavily discussed. navigation already at the very beginning of a trial. In all four Although several other brain regions also elicited activity in analyses, overlapping areas of the precuneus were found in re- C some of the analyses, only posterior parietal regions, and par- lation to performance (Fig. 5 ). Together, these four measures of ticularly the precuneus, were consistently engaged in all memory- performance in conjunction with the finding of a gradual increase related analyses (other regions are discussed in SI Results and of precuneus activity over successive encounters, provide strong Discussion). Our findings strongly support previous suggestions evidence that the precuneus is involved in the encoding and retrieval of spatial representations, and thus has a clear memory- that these areas in the parietal memory network have a genuine related function that emerges already early during learning. role in memory representation (9). In line with previous evidence (11), a purely attentional account of PPC function (25) does not Discussion sufficiently explain our data, because posterior parietal regions In this virtual navigation fMRI study, we followed the development showed increased activation over time and with every additional of the memory representation of a previously unknown, complex encounter with a location. Regarding the question as to which environment. We investigated specifically memory-related processes memory processes are most likely supported by the PPC, the by comparing navigation in a static, and thus learnable, environment increase in activity with repeated encounters suggests that it is with navigation in a continuously changing environment. We were mainly interested in how the involvement of regions that have previously been implicated in spatial processing (i.e., PPC, including Table 1. Performance indices over the four sessions the precuneus and RSC, as well as the hippocampus) develops over Session (M ± SEM) multiple encounters with a specific location. Our analyses point toward a role of the hippocampus during early encoding, whereas Cond 1234 the relevance of the precuneus increases when the memory representation gets stronger. The dissociated activation gradients of the Percentage of targets found S32.6± 2.8 38.5 ± 3.6 41.0 ± 4.1 49.1 ± 4.5 hippocampus and precuneus, together with the decreasing con- ± ± ± ± nectivity between both areas, shed light on the possible location of R46.12.8 45.7 3.0 50.6 2.3 48.4 3.0 extrahippocampal spatial memory representations and suggest that Time to target neocortical representations can develop rapidly during memory S37.3± 1.5 38.2 ± 1.0 37.4 ± 1.5 39.1 ± 1.4 acquisition. Our findings are in good agreement with the standard R26.5± 0.6 26.0 ± 0.9 27.6 ± 0.8 26.2 ± 0.9 model of memory consolidation, which posits that the role of the Route bias hippocampus in memory diminishes with time, whereas the neo- S −0.63 ± 0.02 −0.70 ± 0.03 −0.72 ± 0.03 −0.76 ± 0.04 cortical trace gains in importance (19). We show that this process Session mean and SE values for the percentage of correct trials, for the can occur at a much faster time scale than previously thought. time needed to find the target for the static (S) and random (R) conditions Activation of the PPC, and especially the precuneus, has only (Cond), and for the route bias score in the static condition are displayed. In infrequently been discussed in terms of memory-related processing. the static condition, but not in the random condition, participants found an Only recently, more studies on memory have focused their attention increasing number of target objects within the trial time limit in each on the precuneus. Harvey et al. (20) have found choice-specific firing session. The static condition also shows a steady increase over the sessions in patterns of PPC neurons during memory-based navigation in the rat, the strength of preference for certain corridors, which is indicated by more which can be interpreted as the representation of a memory. negative values.

13254 | www.pnas.org/cgi/doi/10.1073/pnas.1605719113 Brodt et al. Downloaded by guest on October 1, 2021 involved in the retrieval of an existing memory trace. This view is plus offline reactivation, our experiment involved up to 17 en- also supported by studies on spatial (26) and episodic memory, counters with the learning material. Our findings suggest that both where precuneus activation has been shown during recognition can result in a similar extent of systems consolidation. as well as cued recall (10, 27) and parietal lesions disrupt recall The speed with which consolidation integrates the memory rep- performance (28, 29). resentation into neocortical networks is supposed to be very slow (2). Although the MTL, and especially the hippocampus, has been Can the neocortex support an independent memory representation strongly implicated in the representation of space, a number of already after only a few learning trials? Indeed, outside declarative previous studies show no hippocampal activity in spatial memory memory, a rapid establishment of a hippocampus-independent tasks (16, 30). In line with these studies, we did not observe differ- neocortical memory trace within a few trials has already been shown, ences in hippocampal activity during memory-related navigation for example, in the domains of prism adaptation and priming (38, over time on the level of sessions or individual trials (SI Results and 39). Additionally, patients with hippocampal damage can rapidly Discussion). However, we did find hippocampal activity when a acquire lasting relational memories via a fast-mapping strategy, corridor was encountered for the first time (i.e., during early memory which is supposed to be mainly reliant on a neocortical network (40). encoding). Similar to a study by Wolbers and Büchel (31) showing Also, when incorporating new information into preexisting knowl- that the hippocampus is most active when encoding is strongest, this edge networks in the spatial domain, hippocampus-independent activation decreased gradually with every additional encounter. First, neocortical memory traces can be successfully established within this pattern is consistent with the finding that the hippocampus is 3–48 h in rats (41). Thus, our data are in line with previous literature especially active when perceptually new stimuli with high entropy are and support the idea that under certain circumstances, the neocortex encoded (32). Second, our study shows that decreasing reliance on is able to store new information within a very short time frame. the hippocampus can already occur at very early stages of memory By tracking the initial development of a complex memory rep- formation when a complete representation has not yet been estab- resentation over an extended amount of training, we were able to lished. Further analyses (SI Results and Discussion) support the idea uncover early dynamical changes in the relative contribution of that the hippocampus acts as a novelty detector (33) or as an en- and interaction between the hippocampus and the PPC during coder that supports information storage in other areas. memory formation. Our data implicate the hippocampus mainly in It has been proposed that the primary function of parietal regions early encoding processes, whereas memory-related processing in lies in integrating (egocentric) sensory inputs, which cue recall of the PPC, especially in the precuneus, is gradually strengthened (allocentric) the long-term spatial representations residing in the with each encounter of a location, indicating the immediate es- MTL. These representations, in turn, elicit associated mental im- tablishment of a memory representation upon exposure to new agery in the PPC (34). Our results go beyond this view in several information. We show an early decline in connectivity between important aspects. Increasing activity in the PPC over the course of both regions, indicating a rapid independence of parietal memory- learning, concurrently decreasing hippocampal activation as well as related processing from hippocampal function. The observed ac- decreasing connectivity between both areas with every additional tivity in the precuneus was specific for individual episodes of the encounter with a location, speaks against stronger integration of learning experience, predicted behavior, and remained stable over hippocampal with parietal memory systems over learning. If the offline periods, which attests to its mnemonic function. Our study PPC only functions as a retrieval hub for long-term spatial in- provides a general lead for further investigation on the role of the formation, this spatial representation does not seem to be retrieved PPC in a parietal memory network and its putative function as an from within the hippocampus. Instead, we believe that it is more independent store for episodic memory.

likely that the increasing parietal response with growing familiarity COGNITIVE SCIENCES PSYCHOLOGICAL AND is due to the build-up of a memory representation in the PPC itself. Materials and Methods The PPC has often been implicated in the context of working Participants. Twenty-eight healthy, right-handed subjects [13 female and 15 memory (20, 35). Most recently, it has been shown that activity male; mean (M) age: 23.04 ± 2.37 y (M ± SD)] participated in this study. changes in subregions of the PPC during working memory reflect Experimental procedures were approved by the Ethics Committee of the not only task-general but also feature-specific information (13), Ludwig-Maximilians-Universität München, and participants gave written arguing for a transient memory representation in these areas. informed consent. Data from one participant was excluded because of ex- Our data support that the PPC stores a feature-specific repre- cessive movement (>10 mm in translational parameters), and another par- sentation. Moreover, evidence for short-term, map-like repre- ticipant felt nauseous and aborted the experiment, resulting in a smaller sentations in the PPC, especially involving the precuneus, has number of trials for one session. Self-rated motivation on a scale of 0–10 was already been reported in human navigation studies (36). Here, consistently high for all participants (day 1: 8.22 ± 1.60, day 2: 8.15 ± 1.79). we show that the enhanced response of the precuneus to the learned spatial representation is preserved between experimental Virtual Maze Task. A discussion of the virtual maze task is provided in SI Methods. days. This finding can be best explained by a representation of the static environment within the precuneus, which is stable for Procedure. Before the experimental scanning sessions, subjects attended a at least 1 d. Correspondingly, a recent study showed that working screening and pretraining session to even out interindividual differences in memory signatures can be traced in the PPC of monkeys during experience with moving in virtual realities and to assess proneness to cyber the delay period of a categorization task, but only after the sickness. Pretraining consisted of walking around in a virtual test maze, using corresponding task has been learned over a long-term period the same finger-to-movement direction mapping as in the MRI scanner later (21). Therefore, both short-term working memory and long-term on. Maze layout and wall textures differed from the actual maze, and no episodic representations seem to exist in the precuneus. hidden objects were placed in the training maze. Several female subjects Models of systems consolidation postulate that a neocortical dropped out during this stage because of nausea. Participants who had successfully completed the training were handed a questionnaire concerning memory representation may be strengthened by neuronal reac- their experience and preferences in computer games and a sleep diary to tivation between the hippocampus and neocortex, as observed, keep track of their sleeping behavior before and during scanning days. for example, during resting wakefulness and sleep (1, 2). It is Participants were scanned on two consecutive days at approximately the therefore surprising that the effect of additional learning repe- same time of day. On each day, the experiment consisted of two scanning titions on memory systems consolidation has so far received little sessions of 30 min each and a rest period of 10 min in-between. Each session attention. Previous studies assessing systems consolidation in had a blocked design and consisted of 30 blocks with one trial per block and humans have mainly investigated how the memory trace develops alternating conditions. The starting condition was counterbalanced between over offline periods, and their main focus was on the time between participants and sessions. Trials in the static maze lasted either until the target learning and recall (6, 7, 37). We suggest that it is mainly the object was found or maximally for 60 s. The length of the random trials was amount of neuronal activation, not time or level of performance, jittered between 20 and 40 s. On each experimental day, participants filled that determines the progress of systems consolidation. Whereas out pre- and postscan questionnaires assessing their fatigue, motivation, previous studies have mainly used one to three learning repetitions and nausea.

Brodt et al. PNAS | November 15, 2016 | vol. 113 | no. 46 | 13255 Downloaded by guest on October 1, 2021 Behavioral Data and Performance Measures. Three different parameters were maze independent of the particular objects participants were navigating obtained as performance indicators. The first was the percentage of trials in between. The third approach to analyze behavioral performance looked at which the target object was found within the time limit. The second was the decision taken at each junction. Here, we compared junctions where based on the assumption that the development of a spatial representation is subjects chose the corridor that led to the target on the optimal path with accompanied by a preference for those corridors that lead to target objects. the ones where a suboptimal path was chosen in an event-related fMRI Thus, the distribution of frequencies with which every corridor was visited analysis. Statistical testing relied on univariate ANOVAs. All tests were two- provides information about learning-induced changes in navigation behav- tailed with an α-level of 0.05. If not indicated otherwise, means are reported ior. For each session, we standardized the frequencies to the total number of together with their SEs (M ± SEM). corridor encounters and fitted a natural logarithmic curve to the frequency distribution sorted in descending order (Fig. 5B). The time constant of the MRI Data Acquisition and Analysis. A discussion of MRI data acquisition and fitted curve describes its steepness and basically represents how many cor- analysis is provided in SI Methods. ridors make up 63.2% of all encounters within a session. Thus, this measure reflects the strength of bias or memory for individual routes within the static ACKNOWLEDGMENTS. This study was supported by the Deutsche Forschungs- maze. A steeper distribution has a lower time constant and indicates that gemeinschaft (Grant GA730/3-1), the German Center for Vertigo and Balance fewer corridors were visited a higher number of times, and thus a higher Disorders [Bundesministerium für Bildung und Forschung (BMBF) Grant 01EO0901], route bias. In this way, we can assess overall memory for routes within the and the Bernstein Center Munich.

1. Frankland PW, Bontempi B (2005) The organization of recent and remote memories. 28. Ben-Zvi S, Soroker N, Levy DA (2015) Parietal lesion effects on cued recall following Nat Rev Neurosci 6(2):119–130. pair associate learning. Neuropsychologia 73:176–194. 2. McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are complementary 29. Haramati S, Soroker N, Dudai Y, Levy DA (2008) The posterior parietal cortex in learning systems in the hippocampus and neocortex: Insights from the successes and recognition memory: A neuropsychological study. Neuropsychologia 46(7): failures of connectionist models of learning and memory. Psychol Rev 102(3):419–457. 1756–1766. 3. Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory system. Science 30. Schinazi VR, Epstein RA (2010) Neural correlates of real-world route learning. 253(5026):1380–1386. Neuroimage 53(2):725–735. 4. Marr D (1970) A theory for cerebral neocortex. Proc R Soc Lond B Biol Sci 176(1043): 31. Wolbers T, Büchel C (2005) Dissociable retrosplenial and hippocampal contributions 161–234. to successful formation of survey representations. J Neurosci 25(13):3333–3340. 5. Maviel T, Durkin TP, Menzaghi F, Bontempi B (2004) Sites of neocortical re- 32. Fraedrich EM, Flanagin VL, Duann JR, Brandt T, Glasauer S (2012) Hippocampal in- organization critical for remote spatial memory. Science 305(5680):96–99. volvement in processing of indistinct visual motion stimuli. J Cogn Neurosci 24(6): 6. Takashima A, et al. (2009) Shift from hippocampal to neocortical centered retrieval 1344–1357. network with consolidation. J Neurosci 29(32):10087–10093. 33. Kumaran D, Maguire EA (2009) Novelty signals: A window into hippocampal in- 7. Takashima A, et al. (2006) Declarative memory consolidation in humans: A prospective formation processing. Trends Cogn Sci 13(2):47–54. functional magnetic resonance imaging study. Proc Natl Acad Sci USA 103(3):756–761. 34. Byrne P, Becker S, Burgess N (2007) Remembering the past and imagining the future: 8. Rempel-Clower NL, Zola SM, Squire LR, Amaral DG (1996) Three cases of enduring A neural model of spatial memory and imagery. Psychol Rev 114(2):340–375. memory impairment after bilateral damage limited to the hippocampal formation. 35. D’Esposito M, Postle BR (2015) The cognitive neuroscience of working memory. Annu J Neurosci 16(16):5233–5255. Rev Psychol 66:115–142. 9. Gilmore AW, Nelson SM, McDermott KB (2015) A parietal memory network revealed 36. Wolbers T, Hegarty M, Büchel C, Loomis JM (2008) Spatial updating: How the brain by multiple MRI methods. Trends Cogn Sci 19(9):534–543. keeps track of changing object locations during observer motion. Nat Neurosci 11(10): 10. Wagner AD, Shannon BJ, Kahn I, Buckner RL (2005) Parietal lobe contributions to 1223–1230. episodic memory retrieval. Trends Cogn Sci 9(9):445–453. 37. Gais S, et al. (2007) Sleep transforms the cerebral trace of declarative memories. Proc 11. Hutchinson JB, et al. (2014) Functional heterogeneity in posterior parietal cortex Natl Acad Sci USA 104(47):18778–18783. across attention and episodic memory retrieval. Cereb Cortex 24(1):49–66. 38. Henson RN (2003) Neuroimaging studies of priming. Prog Neurobiol 70(1):53–81. 12. Vincent JL, et al. (2006) Coherent spontaneous activity identifies a hippocampal-pa- 39. Luauté J, et al. (2009) Dynamic changes in brain activity during prism adaptation. rietal memory network. J Neurophysiol 96(6):3517–3531. J Neurosci 29(1):169–178. 13. Ester EF, Sprague TC, Serences JT (2015) Parietal and frontal cortex encode stimulus-specific 40. Sharon T, Moscovitch M, Gilboa A (2011) Rapid neocortical acquisition of long-term mnemonic representations during visual working memory. Neuron 87(4):893–905. arbitrary associations independent of the hippocampus. Proc Natl Acad Sci USA 14. Burgess N, Maguire EA, O’Keefe J (2002) The human hippocampus and spatial and 108(3):1146–1151. episodic memory. Neuron 35(4):625–641. 41. Tse D, et al. (2007) Schemas and memory consolidation. Science 316(5821):76–82. 15. Maguire EA, Burke T, Phillips J, Staunton H (1996) Topographical disorientation fol- 42. Kovács G, Schweinberger SR (2016) Repetition suppression - an integrative view. lowing unilateral temporal lobe lesions in humans. Neuropsychologia 34(10):993–1001. Cortex 80:1–4. 16. Winocur G, Moscovitch M, Fogel S, Rosenbaum RS, Sekeres M (2005) Preserved spatial 43. Vogels R (2016) Sources of adaptation of inferior temporal cortical responses. Cortex memory after hippocampal lesions: Effects of extensive experience in a complex en- 80:185–195. vironment. Nat Neurosci 8(3):273–275. 44. Gotts SJ, Chow CC, Martin A (2012) Repetition priming and repetition suppression: A 17. Hirshhorn M, Grady C, Rosenbaum RS, Winocur G, Moscovitch M (2012) The hippo- case for enhanced efficiency through neural synchronization. Cogn Neurosci 3(3-4): campus is involved in mental navigation for a recently learned, but not a highly fa- 227–237. miliar environment: A longitudinal fMRI study. Hippocampus 22(4):842–852. 45. Grill-Spector K, Henson R, Martin A (2006) Repetition and the brain: Neural models of 18. Boccia M, Nemmi F, Guariglia C (2014) Neuropsychology of environmental navigation stimulus-specific effects. Trends Cogn Sci 10(1):14–23. in humans: Review and meta-analysis of FMRI studies in healthy participants. 46. Farovik A, Dupont LM, Arce M, Eichenbaum H (2008) Medial prefrontal cortex sup- Neuropsychol Rev 24(2):236–251. ports recollection, but not familiarity, in the rat. J Neurosci 28(50):13428–13434. 19. Squire LR, Alvarez P (1995) Retrograde amnesia and memory consolidation: A neu- 47. Yonelinas AP, Otten LJ, Shaw KN, Rugg MD (2005) Separating the brain regions in- robiological perspective. Curr Opin Neurobiol 5(2):169–177. volved in recollection and familiarity in recognition memory. J Neurosci 25(11): 20. Harvey CD, Coen P, Tank DW (2012) Choice-specific sequences in parietal cortex 3002–3008. during a virtual-navigation decision task. Nature 484(7392):62–68. 48. Huijbers W, Pennartz CM, Daselaar SM (2010) Dissociating the “retrieval success” 21. Sarma A, Masse NY, Wang XJ, Freedman DJ (2016) Task-specific versus generalized mne- regions of the brain: Effects of retrieval delay. Neuropsychologia 48(2):491–497. monic representations in parietal and prefrontal cortices. Nat Neurosci 19(1):143–149. 49. Blumenfeld RS, Ranganath C (2006) Dorsolateral prefrontal cortex promotes long- 22. Urgolites ZJ, Smith CN, Squire LR (2015) True and false memories, parietal cortex, and term memory formation through its role in working memory organization. J Neurosci confidence judgments. Learn Mem 22(11):557–562. 26(3):916–925. 23. Myskiw JC, Izquierdo I (2012) Posterior parietal cortex and long-term memory: some 50. Blumenfeld RS, Ranganath C (2007) Prefrontal cortex and long-term memory en- data from laboratory animals. Front Integr Neurosci 6:8. coding: An integrative review of findings from neuropsychology and neuroimaging. 24. Levy DA (2012) Towards an understanding of parietal mnemonic processes: Some Neuroscientist 13(3):280–291. conceptual guideposts. Front Integr Neurosci 6:41. 51. Dolcos F, et al. (2013) Neural correlates of opposing effects of emotional distraction 25. Cabeza R, Ciaramelli E, Olson IR, Moscovitch M (2008) The parietal cortex and episodic on working memory and episodic memory: An event-related FMRI investigation. memory: An attentional account. Nat Rev Neurosci 9(8):613–625. Front Psychol 4:293. 26. Boccia M, Guariglia C, Sabatini U, Nemmi F (2016) Navigating toward a novel envi- 52. Javadi AH, Walsh V (2012) Transcranial direct current stimulation (tDCS) of the left ronment from a route or survey perspective: neural correlates and context-dependent dorsolateral prefrontal cortex modulates declarative memory. Brain Stimul 5(3): connectivity. Brain Struct Funct 221(4):2005–2021. 231–241. 27. Kragel JE, Polyn SM (2016) Decoding episodic retrieval processes: Frontoparietal and 53. Rossi S, et al. (2011) Temporal dynamics of memory trace formation in the human medial temporal lobe contributions to free recall. J Cogn Neurosci 28(1):125–139. prefrontal cortex. Cereb Cortex 21(2):368–373.

13256 | www.pnas.org/cgi/doi/10.1073/pnas.1605719113 Brodt et al. Downloaded by guest on October 1, 2021