(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Retrieval of contextual memory can be predicted by CA3 remapping and is differentially influenced by NMDAR activity in rat hippocampus subregions [1]

['Magdalena Miranda', 'Laboratorio De Memoria Y Cognición Molecular', 'Instituto De Neurociencia Cognitiva Y Traslacional', 'Conicet-Fundación Ineco-Universidad Favaloro', 'Ciudad Autónoma De Buenos Aires', 'Buenos Aires', 'Azul Silva', 'Laboratorio Bases Neuronales Del Comportamiento', 'Departamento De Ciencias Fisiológicas', 'Facultad De Ciencias Médicas']

Date: 2024-07

Episodic memory is essential to navigate in a changing environment by recalling past events, creating new memories, and updating stored information from experience. Although the mechanisms for acquisition and consolidation have been profoundly studied, much less is known about memory retrieval. Hippocampal spatial representations are key for retrieval of contextually guided episodic memories. Indeed, hippocampal place cells exhibit stable location-specific activity which is thought to support contextual memory, but can also undergo remapping in response to environmental changes. It is unclear if remapping is directly related to the expression of different episodic memories. Here, using an incidental memory recognition task in rats, we showed that retrieval of a contextually guided memory is reflected by the levels of CA3 remapping, demonstrating a clear link between external cues, hippocampal remapping, and episodic memory retrieval that guides behavior. Furthermore, we describe NMDARs as key players in regulating the balance between retrieval and memory differentiation processes by controlling the reactivation of specific memory traces. While an increase in CA3 NMDAR activity boosts memory retrieval, dentate gyrus NMDAR activity enhances memory differentiation. Our results contribute to understanding how the hippocampal circuit sustains a flexible balance between memory formation and retrieval depending on the environmental cues and the internal representations of the individual. They also provide new insights into the molecular mechanisms underlying the contributions of hippocampal subregions to generate this balance.

In this work, we study the interaction between pattern separation and pattern completion during object in context memory at the behavioral (i.e., memory retrieval), molecular, and electrophysiological (i.e., remapping) level. To model episodic memory retrieval in an incidental task without the bias of motivational influences over contextual representations, we developed a context-degraded task. Taking into account the fundamental role of the balance between discrimination and generalization for episodic memory function, we decided to test the role of molecular mechanisms as potential modulators of this balance in the DG-CA3 circuitry and analyze CA3 neuronal activity. We found that internal context representation, seen as the efficacy of memory retrieval to guide behavior, could be explained by CA3 neuronal context representation. Furthermore, we showed that the limits between retrieval of a previous experience and encoding of a new one can be shifted to favor one process or the other by modulating NMDARs activity in the DG-CA3 circuitry.

Experimental evidence and computational models suggest a role of the DG-CA3 in pattern separation and pattern completion, both at the electrophysiological and behavioral level [ 22 – 24 ]. In this regard, previous studies have implicated NMDA receptors (NMDARs) in DG and CA3 in behavioral memory discrimination and rate remapping [ 25 , 26 ]. In addition, it has been shown that NMDARs are particularly relevant to recover memory from a cue-degraded context [ 26 ] and to allow CA1 neurons to maintain place field characteristics between these conditions [ 27 , 28 ]. Since some theories suggested that the DG-CA3 circuit of the HP mediates the dynamic competition between memory differentiation (supported by pattern separation) and generalization (supported by pattern completion), understanding the nature of this interaction could provide a framework to explain how episodic memory is rooted on contextual representation.

But can remapping explain retrieval or encoding of spatial memories that sustain behavior? Despite the intuition of place cell importance for episodic memory, there is still no clear evidence that remapping is directly related to the expression of different contextual memories. One reason for this is that most studies have only related remapping to observable properties of the environment but not to behavioral outputs that could give some insight into the inferences the animal is making about a particular environment or experience (i.e., its internal context representation). In other words, the variability of circuit activation in the same physical environment has been overlooked, variability that could be influenced by motivation [ 18 , 19 ], attention [ 20 ], and experience [ 21 ]. Moreover, it is not clear how individual mnemonic variability (i.e., behaving as being in a familiar context or not) relates to internal context representation and place cell activity.

Learned behavior is the outcome of an interaction between the environment and representations stored in memory. Many contextual responses depend on whether experiences are novel or similar to previous ones. Since the environment is continuously changing, episodic memory retrieval usually occurs under contextually modified conditions and the ability to retrieve previous memories despite partial contextual change becomes crucial [ 1 ]. Changes in environmental cues force the brain to make a difficult choice: should we discard differences between 2 similar experiences (i.e., retrieve an already stored representation) or should we build a new memory? Under each particular context, the brain evaluates environmental and internal cues to determine whether a context is familiar or not. Based on the proposed dual role of the hippocampus (HP) in context recognition, computational studies suggested the need for 2 independent systems [ 2 ] that optimize information processing under small or large contextual changes. These studies proposed that unique features of hippocampal subregions could allow the development of computationally distinct and complementary functions needed for correct episodic memory formation and retrieval. These processes, based on the activity of the DG-CA3 circuit, are known as pattern separation and pattern completion [ 2 – 4 ].

Finally, we performed a dichotomic analysis to account for the putative effect of these drugs on memory retrieval, similar to that used in the electrophysiological experiments (i.e., putative internal context representation). We divided the data into 2 groups by pooling together all “memory boosting” (left) or “memory impairing” (right) drug infusion experiments (see Methods , Fig 6A ). We found that in the “memory boosting” group, in a two-cue context (where most animals show a behavior inconsistent with retrieval), the treatment increased the proportion of rats that behaved in a retrieval-like manner (similar internal context representation between test and training, Fig 6B , left). Conversely, in the “memory impairing” group, in a three-cue context where most animals behave in a retrieval-like fashion, the treatment decreased the number of rats behaving in a retrieval-like manner ( Fig 6B , right). These results show that, although there is a great variability in animal behavior, NMDAR activity in the hippocampal circuit can bias the internal spatial representation towards memory retrieval or away from it.

Changes in engram cell excitability [ 53 ] or plasticity [ 54 – 56 ] could be controlling the efficacy of memory retrieval under contextually degraded conditions in our task. In this regard, NMDA receptor activation can trigger AMPA-type trafficking to the dendritic surface in an L-type voltage gated calcium channel (L-VGCC)-dependent manner [ 57 ] and L-VGCC can also regulate neuronal excitability [ 58 , 59 ]. Therefore, we evaluated their role as potential effectors of NMDARs. Animals were infused in CA3 with the L-type calcium channel inhibitor Nimodipine (Nim) prior to the D-Cyc infusion and then tested in the PC2 context. While D-Cyc infusion decreased the percentage of object exploration, co-infusion of D-Cyc with Nimodipine prevented this effect ( Fig 5G ). This result suggests that NMDARs interact with L-VGCC in the CA3 region to favor memory generalization. Histological analysis did not reveal any lesion in the infusion sites (CA3 or DG) ( S2 Fig ). Results cannot be explained by drug-related changes in motivation to explore the object during the test phase, since infusion of D-Cyc in the DG or CA3 region 15 min before exposure to an object in the AC context did not alter exploration percentages compared with Vehicle-infused groups (paired t test: t DG = 0.67, p = 0.528, n = 7; W CA3 = −8, p = 0.578, n = 7).

However, we predicted that we could favor memory discrimination by activating NMDAR in the DG. We tested this prediction in animals trained in the 2-day version of the task that received an infusion of the NMDAR partial agonist D-Cycloserine (D-Cyc) or Vehicle in the DG 15 min before exposure to a PC context ( Fig 5D ). Animals infused with D-Cyc in the dentate gyrus before the test session showed significantly higher exploration percentages than the vehicle group, with object exploration times that did not differ from the training session. In line with this, the proportion of animals that show high exploration consistent with non-retrieval in test sessions (in red) goes from 12% in Veh injected sessions to 37% in the AP5 injected sessions ( Fig 5E ). It is important to note that animals infused with D-Cyc showed no increase in the distance traveled or linear velocity during the test session compared with Veh injected animals. This outcome rules out an effect of D-Cyc over general mobility levels ( S3 Fig ). All in all, these results indicate that NMDAR activity in the DG can decrease the ability to generalize from a degraded context. In contrast, we have previously shown that NMDARs in the CA3 region are important for contextual generalization. We reasoned that by increasing NMDAR activity in CA3, we could increase retrieval guided by incomplete cue stimuli. We infused D-Cyc or Vehicle in CA3 before the test phase under the PC2 context and found that Veh-infused animals showed exploration levels consistent with a lack of object in context memory retrieval ( Fig 5D and 5F ). However, the infusion of D-Cyc in the CA3 previous to this test phase significantly decreased the percentage of object exploration. Consistently, the percentage of animals that show high exploration (i.e., non-retrieval) goes from 56% to 22% ( Fig 5F ). This indicates that increasing NMDAR activity in CA3 can enhance memory retrieval under cue-degraded conditions, potentially through tipping the balance towards contextual generalization.

(A) Time of infusion of NMDAR agonist D-Cyc or Vehicle (Veh) in the DG 15 min before the evaluation session in the AC, PC, or PC2 condition of the task. (B) Percentage of exploration for objects presented during the test session under the AC or PC conditions of the task in animals infused with either Vehicle (light) or AP5 (dark) in the DG. Two-way ANOVA GD: n = 8–9, Finteraction = 0.42, p = 0.527, Fcondition = 1.93, p = 0.185, Fdrug = 0.65, p = 0.432. One sample t test against 100%, Veh AC t = 3.94, p = 0.043; AP5 AC t = 5.05, p = 0.001; Veh PC t = 1.467, p = 0.186; AP5 t = 2.30, p = 0.055. ACVeh: 66.83 ± 8.42, ACAP5: 65.37 ± 6.86, PCVeh: 85.76 ± 9.71, PCAP5: 72.51 ± 11.95. (C) Percentage of exploration in the PC2 context during test session in Vehicle infused (light) or AP5 infused animals (dark). Paired t test t = 3.26, p = 0.010, n = 10. One sample t test against 100%, Veh p = 0.669, t = 0.44; AP5 p = 0.005, t = 3.65. PC2Veh: 103.9 ± 8.89, PC2AP5: 74.09 ± 7.09. (D) Time of infusion of D-Cyc or Vehicle in the DG or CA3 15 min before test session in the PC or PC2 condition. (E) Percentage of object exploration during test session in PC condition of the task in animals infused with Veh (light) or D-Cyc (dark) in DG prior to the session. Paired t test: t = 4.198, p = 0.004, n = 8. One sample t test against 100%: Veh t = 3.90, p = 0.006; D-Cyc t = 2.54, p = 0.04. PC Veh : 68.36 ± 8.10, PC DCyc : 84.80 ± 5.98. (F) Percentage of exploration during test session in PC2 condition of the task after animals were infused with Veh (light) or D-Cyc (dark) in CA3. Paired t test: t = 3.03, p = 0.016, n = 9. One sample t test against 100%, Veh t = 0.31, p = 0.767; D-Cyc t = 4.15, p = 0.003. PC Veh : 103.00 ± 9.69, PC DCyc : 70.65 ± 7.08. (G) (Left) Time of Nimodipine (Nimo)/Veh infusion and D-Cyc/Vehicle in CA3 20 and 15 min, respectively, before test session. (Right) Percentage of exploration during test session in PC2 condition of the task after animals were infused with Veh/Veh (light) or Veh/D-Cyc (red) or Nim/D-Cyc (dark) in the CA3 region. RM one-way ANOVA: F = 5.44, p = 0.037, n = 8. * p < 0.05, ** p < 0.01. Veh/Veh t = 3.62, p = 0.0085, Veh/D-Cyc t = 6.717, p = 0.0003; Nimo/D-Cyc t = 0.61, p = 0.5668. PC2 Veh/Veh : 84.30 ± 4.34, PC2 Veh/DCyc : 56.64 ± 6.45, PC Nim/DCyc : 93.01 ± 11.63. Individual values used to calculate the mean and SEM are presented as dots. Red dots represent animals whose object exploration was above the memory retrieval threshold (i.e., internal context representation). The data that support these findings is available in OSF at https://osf.io/7pw23/ . AC, all cue; PC, partial cue; RM, repeated measures.

Many findings indicate the importance of NMDAR in the HP (especially the DG) for correct pattern separation function, both at the behavioral [ 25 ] and electrophysiological level [ 26 , 49 – 51 ]. Interfering with NMDARs could hinder memory differentiation thereby increasing memory generalization. To address this, we tested the role of NMDAR activity in the DG for retrieval under the AC, PC conditions, and an additional PC2 condition in which only 2 cues were left ( Fig 5A ). We found no effect of AP5 infusions in the DG 15 min previous to the test session in the AC or PC condition ( Fig 5B ). However, using a two-cue context (PC2) during test session, we observed that DG-Vehicle-infused animals showed percentages of exploration that did not differ significantly from training, while AP5 infused animals had a percentage of exploration significantly lower than 100% and significantly different from the Vehicle-infused group. Consistent with this, the proportion of animals showing non-retrieval behavior in test sessions in the PC2 (in red) goes from 60% in Veh injected sessions to 20% in the AP5 injected sessions ( Fig 5C ). The lack of effects in the less degraded AC and PC conditions rules out any nonspecific influence of AP5 on exploration levels. Our results suggest that NMDARs in the DG are necessary for the orthogonalization that allows a distinction to be formed between a cue-degraded context and the original full context, and that the absence of this distinction could lead to memory retrieval even in highly degraded contexts.

Then, we asked whether NMDARs were required for reactivation of the object-context memory under degraded cues. For this, after training on a context-object association as in previous experiments (day 1), animals infused with either AP5 or Veh in CA3 were subjected to an “all cues” contextual re-exposure in the absence of any object, and Emetine or Vehicle was infused in the Prh immediately after (TS1). On day 3, the original object memory of day 1 was tested on the triangular context against a novel object (TS2). Consistent with our previous result ( Fig 2 ), Emetine infusion in the Prh immediately after context exposure (TS1) significantly reduced the discrimination ratio of the previously presented object even under a cue-degraded context ( Fig 4D ). But when we combined pre-TS1 infusion of AP5 into the CA3 with post-TS1 infusion of Emetine into the Prh, Emetine only reduced the discrimination ratio of the object memory under AC but not under PC context exposure, suggesting that inhibiting NMDAR activity during PC exposure can interfere with the ability to activate the original memory representation. The PC specificity of AP5-effect during the exposure session rules out an effect due to changes in memory labilization induced by AP5. Furthermore, the absence of any significant difference in the number of rearings exclude possible AP5-induced changes in exploration levels during the exposure session (paired t test: AC Mean ± SEM Veh = 36.65 ± 5.84 AP5 = 42.53 ± 32.77, t = 0.78, p = 0.444, PC Mean ± SEM Veh = 56.9 ± 4.37 AP5 = 53.05 ± 4.89, t = 0.84, p = 0.409). In sum, these results showed that NMDARs in the CA3 region are not required for memory reactivation when guided by the original contextual information, but are crucial for the reactivation of object-in-context memories under partial contextual information when a memory completion process is required.

(A) Time of infusion of AP5 or Vehicle (Veh) in CA3 15 min before test session in the AC and PC conditions. (B) Data from PC and TC groups from Fig 1D was used to obtain the 2SD threshold criteria for discriminating sessions where putative internal context representations would be consistent with non-retrieval, as done in Fig 3 . (C) Percentage of exploration for the object presented in the presence of the AC or PC conditions during test session, in animals infused with Vehicle (light) or AP5 (dark) in the CA3 region of the HP. Two-way ANOVA: n = 10–11, F interaction = 4.50, p = 0.047, PC Veh-AP5 p = 0.042, AC Veh-AP5 p = 0.887. Wilcoxon test against 100%, AC-Veh W = −48, p = 0.032; PC-Veh W = −66, p = 0.001; AC-AP5 W = −55, p = 0.002; PC-AP5 W = −7, p = 0.769. AC Veh : 72.85 ± 8.64, AC AP5 : 67.02 ± 4.59, PC Veh : 71.98 ± 4.39, PC AP5 : 97.42 ± 5.42. (D) (Up) Trained animals were infused with AP5 or Veh in CA3 before test session in an AC (1) or PC (2) conditions and infused with Veh or Emetine (Eme) in the Prh immediately after test session (TS1); 24 h after, memory for the original object was tested in another familiar context against a novel object (TS2). (Down) (1) Discrimination ratio for the TS2 session of the task, 24 h after exposure to an empty original context followed by an Emetine or Vehicle infusion in animals that had previously received AP5 or Vehicle infusions. Two-way ANOVA: F interaction = 0.26 p = 0.618, F Veh-Eme = 35.93 p < 0.0001, F Veh-AP5 = 0.001 p = 0.970, n = 8–9. One sample t test against 0, Veh-Veh t = 3.81, p = 0.007; AP5-Veh t = 5.98, p = 0.0006; Eme-Veh t = 1.85, p = 0.101; Eme-AP5 t = 0.89, p = 0.399. AC Veh/Veh : 0.29 ± 0.08, AC AP5/Veh : 0.26 ± 0.04, AC Veh/Eme : −0.10 ± 0.05, AC AP5/Eme : −0.06 ± 0.07. (2) Discrimination ratio for TS2 session of the task, 24 h after exposure to an empty partial cue context followed by an Emetine or Vehicle infusion in animals that had previously received AP5 or Vehicle infusions. Two-way ANOVA: F interaction = 4.42 p = 0.049, n = 10–11; Veh/Veh vs. Veh/AP5 p > 0.99, Veh/Eme vs. Eme/Veh p = 0.046. One sample t test against 0: Veh-Veh t = 5.83, p = 0.0003; AP5-Veh t = 3.23, p = 0.010; Eme-Veh t = 1.44, p = 0.181; AP5-Eme t = 3.90, p = 0.003. PC Veh/Veh : 0.21 ± 0.04, PC AP5/Veh : 0.18 ± 0.06, PC Veh/Eme : 0.04 ± 0.03, PC AP5/Eme : 0.19 ± 0.05. Red * represents significance against 0, * p < 0.05, *** p < 0.001. Individual values used to calculate mean and SEM are presented as dots. Red dots represent animals whose object exploration was above the memory retrieval threshold (i.e., putative internal context representation). The data that support these findings is available in OSF at https://osf.io/7pw23/ . AC, all cue; PC, partial cue.

We have demonstrated a correlation between CA3 place cell remapping and retrieval memory in our behavioral paradigm. It is well-established that the generation and modification of place fields are related to NMDAR-activity [ 46 – 51 ]. Furthermore, previous studies have as well shown that the CA3 region [ 24 , 52 ], and NMDARs in particular, are required when retrieval occurs in the absence of some of the original contextual cues [ 28 ]. To assess the requirement of CA3-NMDARs for retrieval under cue-degraded conditions, we used a pharmacological approach. Animals trained in the 2-day version of the task received an injection of the NMDAR antagonist AP5 or Vehicle in the CA3 region 15 min before the test session in the AC or PC condition ( Fig 4A ). Vehicle-infused animals spent significantly less time exploring the original object, independently of the condition. However, animals that received an AP5 infusion prior to the test session in the PC condition showed significantly higher exploration percentages than Vehicle infused animals. In contrast, no effect of AP5 was observed in the AC condition. In accordance with this, the proportion of animals that show high exploration (consistent with non-retrieval in the test, in red) varies from 18% in “all cues” and 0% “partial cues” in vehicle injected sessions to 50% in the “partial cues” AP5 injected test sessions ( Fig 4C ). Our results suggest that the NMDAR activity in the CA3 region is involved in recovering memories in the presence of partial contextual information but not with all the original cues.

The differences in the spatial coding between phases of the task were also analyzed using a population vector analysis ( S7C Fig ). This analysis showed the same tendency as the one observed at the level of single units, where the spatial coding is more stable in non-retrieval sessions. We also fitted our neuronal data with a general linear model (GLM) with a binomial distribution, using the putative internal context representation as the response variable. Models including only one of the variables (spatial correlation or firing rate change) or both were significant (internal rep ~ spatial corr, p = 1.27e-08; internal rep ~ firing rate change, p = 1.63e-05 and internal rep ~ spatial Corr + firing rate change, p spatial corr = 1.97e-06, p firing rate change = 0.02). The best of the 3 models is the one that includes both variables (AIC 1-variable model = 294.30 spatial corr, 314.01 firing rate change, AIC 2-variables model = 291.16, and ANOVA comparing models with increasing variables, p 1-variable model versus p 2-variables model = 0.009). Although the firing rate change appears to be a less reliable estimator of the animal’s internal representation compared to spatial information, it still provides significant information, as demonstrated by the t-SNE analysis and the GLM model.

To better understand how the neuronal population might represent this dichotomic behavioral output, we examine whether our neuronal population can be classified according to its spatial correlation and firing rate change values. To do this, we computed t-SNE (t-distributed stochastic neighbor embedding) over the 2 variables and performed a k-mean clustering, finding 2 clear clusters (evaluated by the Silhouette method). Cluster A has a higher mean spatial correlation than Cluster B ( Fig 3I , n = 165–167, LMM, p = 1.9186e-63) and a lower firing rate change ( Fig 3I , n = 165–167, LMM, p = 1.0141e-14). Looking at the proportion of each cluster according to the internal representation, we found that Cluster B is more represented in the non-retrieval session ( Fig 3J , 80%, 4 out of 5) and Cluster A is more represented in the retrieval sessions ( Fig 3J , 75%, 12 out of 16, X2 (1, N = 21) = 4.887, p = 0.0271). In addition, most cells from the session where animals had a different internal representation (79%) are in Cluster B. This suggests that Cluster B grouped neurons represent a change in the animal’s internal representation. In summary, when animals are unable to recall the initial memory, there is a greater representation of cells from Cluster B. These cells exhibit a low spatial correlation and a high firing rate change.

According to dichotomic accounts of memory retrieval, animals’ behavioral output should be represented by only 2 distributions, i.e., retrieval of the prior experience/high place cell activity correlation or no retrieval/low place cell activity correlation [ 43 – 45 ]. To understand if our behavioral data fitted this dichotomic account of memory, we performed a bootstrap likelihood ratio test (LRT) for assessing the number of mixture Gaussian components that could model our behavioral data. We found that the addition of a second distribution significantly increased the ability of the mixture Gaussian component model to explain our data (LRT, p = 0.001), but adding a third Gaussian component did not lead to a significant improvement (LRT, p = 0.997). This suggests that the behavioral output of our animals can be modeled by 2 independent distributions (i.e., memory retrieval/no memory retrieval, S7A Fig ). We then classified the recording sessions in terms of the animal’s dichotomic behavioral output (i.e., putative internal context representation) instead of the experimental cue setup (AC, PC, or NC). We separated sessions in which animals discriminated between contexts (high exploration, consistent with non-retrieval) from sessions in which they did not (low exploration, consistent with retrieval), using an estimated 2-SD threshold for context discrimination (see Methods , Fig 3E ). With this new classification, we found significant differences in the spatial correlation ( Fig 3F , n = 65–267, LMM p = 1.822e-06) 3G, n = 65–267, LMM p = 0.0038). Similar results were observed when comparing only neurons from the NC condition with different putative internal context representations ( S7B and S7C Fig ). Overall, place cell activity seems to better represent animal’s internal representation than just the physical properties of a context. To rule out nonspecific effects, we repeated the same analysis but using the activity of neurons that were not classified as place cells and found no differences between groups ( S7D Fig ). In addition, we calculated the percentage of time the animals explore an area similar in size to the object and opposite to it, and sorted place cell’s activity according to this new behavioral output. We found no differences between groups for the 2 studied variables ( S7E Fig ).

One advantage of the present task is its variability. We could find sessions in which, despite having reduced the number of cues (NC or PC condition), animals showed low percentage of exploration and sessions in which animals exhibited high percentage of exploration under a degraded context. This behavioral characteristic was useful to look for a relationship between place cell activity and the memory output (represented as the object exploration percentage) independently of the experimental condition. There was a significant inverse correlation between spatial correlation and percentage of object exploration and a significant positive correlation between firing rate change and percentage of object exploration. Thus, animals tend to explore the object more as firing rate change in CA3 increases and spatial correlation decreases ( S6D and S6E Fig , spatial corr p = 1.47e-10 R = −0.34 and firing rate change p = 0.00015 R = 0.20).

(A) CA3 place cell responses. Place maps for training session (left) and test session (right). Color-coded firing maps normalized by the min and max firing rate of each neuron. Spatial correlation and firing rate change were calculated for each neuron. (B) Schematic of the calculation of spatial correlation and firing rate change. (C) CA3 spatial correlation sorted by condition. Only in NC sessions, place cells had a significantly lower spatial correlation. Moreover, there were no differences between the AC and PC conditions. n = 54–141, Kruskal–Wallis test NC-AC p = 0.0027, NC-CP p = 0.0074, and AC-CP p = 0.8503. AC: 0.53 ± 0.03, PC: 0.56 ± 0.04, NC: 0.39 ± 0.03. (D) CA3 firing rate change sorted by condition. There are no differences between conditions. n = 54–141, Kruskal–Wallis test NC-AC p = 0.3430, NC-CP p = 0.0720, and AC-CP p = 0.4788. AC: 0.22 ± 0.02, PC: 0.19 ± 0.03, NC: 0.29 ± 0.02. (E) Object exploration percentage for every condition and for the internal representation: sessions were divided by a context discrimination threshold (mean (OE% AC OE% PC) + 2 STD)) red dots represent sessions where animals discriminated between contexts and gray dots, sessions where animals did not discriminate. (F) CA3 spatial correlation sorted by putative internal representation. n = 65–267, LMM p = 2.311e-06, =: 0.53 ± 0.02, ≠:0.25 ± 0.04. (G) CA3 firing rate change sorted by internal representation. n = 65–267, LMM p = 0.006, =: 0.21 ± 0.01, ≠:0.36 ± 0.04. (H) t-SNE over the spatial correlation and firing rate change and K-means clustering, gives 2 clear clusters evaluated by the Silhouette method. (I) Spatial correlation and firing rate change sorted by cluster. Cluster A has higher mean spatial correlation and low firing rate change than cluster B. n = 165–167, LMM, p spatial corr. = 1.9186e-63 (A: 0.76 ± 0.01, B: 0.20 ± 0.02) and p firing rate change = 1.0141e-14 (A: 0.14 ± 0.01, B: 0.34 ± 0.02). (J) Percentage of sessions with majority of cluster A or B sorted by putative internal representation. X2 (1, N = 21) = 4.887, p = 0.0271. The data that support these findings is available in OSF at https://osf.io/7pw23/ . AC, all cue; NC, no cue; PC, partial cue; t-SNE, t-distributed stochastic neighbor embedding.

While several studies have shown that different experiences can be represented in the hippocampus [ 41 , 42 ], a direct link between the individual’s internal context representation and memory retrieval is still missing. Taking advantage of our new behavioral paradigm, which allows us to measure how well animals remember a given context, we aimed to address an important yet unresolved question: How is place cell coding (in particular remapping) related to spatial memory? To assess the relationship between contextual representation and object in context memory, we recorded CA3 activity as animals performed the 1-day version of the task (4 rats, 24 sessions, n = 332 place cells). We compared place cell firing properties between training and test phases using parameters that describe similarities and differences between their place fields in both phases of the task, like the spatial correlation, and firing rate change ( Fig 3 , see also Methods section). We found that a place cell can maintain the position where they fire (similar location of its place field, high spatial correlation) and its firing rate (low firing rate change) or change one or both variables at the same time (Figs 3A and S5 ) without changing its spatial information coding, the size of their place fields or the proportion of place cells coding the environment ( S6A–S6C Fig ). When we grouped place cells according to the test condition, we found that the spatial correlation differed significantly between the NC and the other 2 conditions ( Fig 3C , n = 54–141, Kruskal–Wallis test NC-AC p = 0.0048, NC-PC p = 0.0108, and AC-PC p = 0.8523), but not the firing rate change ( Fig 3D , n = 54–141, Kruskal–Wallis test NC-AC p = 0.6012, NC-PC p = 0.1467, and AC-PC p = 0.4754). Interestingly, even after removing half of the cues (PC), there were no significant differences in the spatial correlation or firing rate change of place cells when compared with the AC condition ( Fig 3C and 3D ). This phenomenon cannot be explained by differential context exploration, as the animals show no significant differences between conditions in the distance traveled, in the mean instantaneous velocity, or in the comparison of time spent in each spatial bin between phases ( S5 Fig ). These neuronal results are consistent with the behavioral output observed in the PC condition where behavior was not different between PC and AC ( Fig 1G ). Crucially, this suggests that retrieval of the original contextual memory in the PC context is directly associated with CA3 neuronal representation by place cells.

To rule out nonspecific effects of Emetine, we tested the contextual dependency of memory labilization. During the training session, rats were exposed to an object in the NC condition and after 2 h to another object in the AC condition in a counterbalanced manner, and 24 h later, animals were placed in either the empty NC context or the empty AC context and both groups received Emetine or Vehicle infusion immediately after context exposure. Memory for the original objects was tested 24 h later against a novel object in both cases in a counterbalanced manner ( Fig 2C , see Methods section). Time spent exploring each object during training did not differ between the AC and NC context ( Fig 2D ). Nevertheless, animals infused with Emetine after “no cues” exposure only showed a reduction of the discrimination ratio for the object originally associated with that condition. On the other side, animals infused with Emetine after “all cues” condition had selective reduction for the discrimination of the object associated with AC during the training session ( Fig 2E ). This indicates that the effect of Emetine on memory reconsolidation is specific for the object-context association because only the trace that was previously associated with the presented context was sensitive to Emetine infusion. Additionally, this suggests that all the contexts by themselves are capable of guiding reactivation of an object memory originally associated to that context, leaving the memories of objects linked to a different context unaffected by the action of protein synthesis inhibitors.

(A) Procedure and time points of infusions. Animals were trained in the task and 24 h after were exposed to an empty AC or NC context (TS1) and immediately after they received an Emetine (Eme) or Vehicle (Veh) infusion in the Prh. A test session was given to evaluate the original object against a novel one (TS2). (B) Discrimination ratio for the TS2 test session, 24 h after exposure to an empty context with AC or NC followed by an Emetine (dark) or Veh (light) infusion. RM two-way ANOVA: F interaction = 6.25, p = 0.029, F condition = 2.90, p = 0.117, F drug = 2.50, p = 0.142, n = 6–7, Veh vs. Eme AC p = 0.014. NC p = 0.479. One sample t test against 0: NC-Veh t = 4.77, p = 0.003; NC-Eme t = 4.21, p = 0.008; AC-Veh t = 4.38, p = 0.005; AC-Eme t = 1.30, p = 0.252. ACVeh: 0.22 ± 0.05, ACEme: 0.03 ± 0.02, NCVeh: 0.18 ± 0.04, NCEme: 0.24 ± 0.06. (C) Experimental protocol and time points of the Emetine or Vehicle infusion. Animals were trained to an object A (circle) in the NC context and an object C (star) in the NC context. The next day, they received an exposure session to an empty AC or NC context (no object) and immediately afterward they were infused with either Emetine (dark) or Vehicle (light) in the Prh. Lastly, object memories were tested against novel objects in a different context 24 h after (TS2). (D) Object exploration time during training in the AC or NC context. Mann–Whitney test U = 60 p = 0.089. AC: 94.08 ± 4.88, NC: 87.00 ± 10.11. (E) Discrimination ratio for the A or C objects against a novel object during TS2, 24 h after exposure to an empty AC or NC context followed by Emetine (gridded) or Vehicle (smooth) in the Prh. Two-way ANOVA: F interaction = 2.08, p = 0.132, F condition = 4.05, p = 0.020, F drug = 4.77, p = 0.039, n = 6–7. AC retrieval: CircleVeh: 0.17 ± 0.07, CircleEme: 0.03 ± 0.05, StarVeh: 0.26 ± 0.02, StarEme: 0.27 ± 0.05. NC retrieval: CircleVeh: 0.19 ± 0.06, CircleEme: 0.19 ± 0.06, StarVeh: 0.20 ± 0.07, StarEme: −0.01 ± 0.03. Red * represents significance against 0, # p < 0.1, * p < 0.05, ** p < 0.01. Individual values used to calculate mean and SEM are presented as dots. The data that support these findings is available in OSF at https://osf.io/7pw23/ . AC, all cue; NC, no cue; RM, repeated measures.

Under certain conditions, memory reactivation can increase susceptibility to post-retrieval protein synthesis inhibitors, specifically targeting the reactivated memory trace. Inactive memories during the test remain stable and are unaffected by protein synthesis inhibition [ 34 – 39 ]. If contextual information is enough to guide retrieval of the original object-in-context memory in our task, exposure to the context alone should destabilize object memory in regions where this memory is stored, like the perirhinal cortex (Prh) [ 34 , 37 , 40 ]. To address this question, animals trained on day 1 in the task were exposed to a retrieval session on day 2 in the same context but in the absence of an object in either AC or NC conditions (TS1). Immediately after context exposure, they were infused into the Prh with the protein synthesis inhibitor Emetine (Eme) or with Vehicle (Veh). Object memory was evaluated 24 h later (day 3) on a test session (TS2), by placing the original object next to a novel one on a novel triangular context lacking any contextual cues so that object memory could be estimated independently of any contextual contribution ( Fig 2A , see Methods section). Vehicle-infused animals had significantly higher discrimination ratios than zero during this last session, evidencing 48 h-object memory. On the other hand, Emetine infusion in the Prh significantly decreased the discrimination ratio in the “all cues” condition when compared to Vehicle, while no effect was seen in the “no cues” condition ( Fig 2B ) (one sample t test against 0: NC-Veh t = 4.77, p = 0.003; NC-Eme t = 4.22, p = 0.008; AC-Veh t = 4.38, p = 0.005; AC-Eme t = 1.30, p = 0.252). These results indicate that the context associated with the object is sufficient to guide the reactivation of the original “object-in-context” memory.

(A) Associative retrieval task. “All cues” (AC, blue), “partial cues” (PC, green), and “no cues” (NC, orange) conditions. Right panel: representative animal trajectory. (B) Schematic illustration of 2-day version of the task. (C) Total object exploration time during training session for AC, PC, and NC during the 2-day version of the task. Rats spent an equal amount of time exploring the object during the training phase under the AC, PC, and NC conditions. One-way RM ANOVA F = 0.74, p = 0.453, n = 9. AC: 78.91 ± 5.38, PC: 87.23 ± 7.54, NC: 79.73 ± 6.31. (D) Percentage of object exploration in the presence of a variable number of cues (AC, NC, and PC) in the test session of the 2-day version of the task with respect to training. One-way RM ANOVA, F = 8.27, p = 0.006; AC-NC p = 0.008, PC-NC p = 0.036; one sample t test against 100% AC t = 9.39, p < 0.0001; PC t = 5.18, p = 0.0008; NC t = 1.83, p = 0.105. AC: 61.84 ± 4.07, PC: 72.56 ± 5.29, NC: 89.91 ± 5.51. (E) Schematic illustration of the 1-day version of the task. (F) Total object exploration time during training session for AC, PC, and NC during the 1-day version of the task. Two-way RM ANOVA, interaction: F = 0.5755, p = 0.7484, n = 10, sessions = 75, each dot represents a session. AC: 47.89 ± 3.38, PC: 47.39 ± 3.07, NC: 49.74 ± 3.50. (G) Percentage of object exploration in presence of a variable number of cues (AC, NC, and PC) in the test session of the 1-day version of the task with respect to training. Two-way RM ANOVA, main effect condition F = 16.04, p <0.0001. Tukey’s post hoc test: p < 0.0001 AC vs. NC; p = 0.25 AC vs. PC; p = 0.001 NC vs. PC. One sample t test against 100% AC t = 14.45, p < 0.0001; PC t = 8.76, p < 0.0001; NC t = 0.37, p = 0.91. AC: 56.98 ± 3.31, PC: 63.93 ± 4.11, NC: 92.13 ± 7.34. TR, training session; TS, test session. * p < 0.05, ** p < 0.01, **** p < 0.0001. Individual values used to calculate mean and SEM are presented as dots. The data that support these findings is available in OSF at https://osf.io/7pw23/ . AC, all cue; NC, no cue; PC, partial cue; RM, repeated measures.

Rodent’s inherent preference for novelty requires engaging retrieval to determine an event as familiar or novel. To determine whether rats would be able to retrieve experiences under degraded environmental cues, we adapted the spontaneous object recognition task to an associative retrieval task [ 29 – 31 ]. During a training session, animals incidentally associated a novel object with a context with 6 distal cues. They were later tested with an identical copy of the object in the same location but under a variable number of the original contextual cues: “all cues” (AC), “partial cues” (PC, 3 distal cues), and “no cues” (NC, no distal cues) conditions ( Fig 1A, 1B and 1E ; see SI Appendix). For pharmacological experiments, animals were tested 24 h after the training session (2-day version) and for the electrophysiological experiments, animals were tested 4 h after the training session (1-day version). The time animals spent exploring the object during the test session was considered a measure of object memory in that context, with decreased object exploration when animals retrieve the original object-context memory. We found no difference in object exploration time between groups during training ( Fig 1C and 1F ). Since all groups were trained in AC conditions, this indicates all groups were, a priori, equal in terms of object exploration levels. Critically, there was a significant condition effect over the percentages of object exploration during test. Animals in the NC condition had significantly higher percentages of object exploration than the AC and PC conditions ( Fig 1D and 1G ). Percentages of object exploration were significantly lower than 100% for both the AC and PC, but not for the NC in both versions of the task. The absence of differences due to the cue removal in the PC condition suggests that rats are able to retrieve the original representation using degraded information. Rearing behavior, an alternative measure of environmental novelty less dependent on object memory than object exploration time [ 32 , 33 ], showed a similar tendency ( S1 Fig ).

Discussion

In this study, we investigated the molecular and neural mechanisms involved in the recovery of spatial representations stored as incidental memories of an object experienced in a particular context. Our findings indicate that memory differentiation and memory generalization functions compete for behavioral control. We showed that the amount of CA3 remapping, signaling a separation process, is related to the recall of object-in-context memory. When animals recognized an object in context as familiar, regardless of the available contextual cues, there was less remapping in CA3 than when animals recognized the relationship as novel. This finding is crucial for understanding how contextual representations can influence episodic memory recall and behavior. In the same line, we showed that NMDAR activity in the DG-CA3 circuit can influence this discrimination/generalization balance. We found that pharmacological treatments that favor pattern separation (CA3 NMDAR antagonist and DG NMDAR agonist) lead animals to behave as if the relationship between object and context is novel while treatments that promote pattern completion (CA3 NMDAR agonist and DG NMDAR antagonist) bias animals to retrieve a prior experience.

Holistic memory retrieval is considered a key element of episodic memory allowing all aspects of an event to be recovered in an integrated manner. Our results indicate that animals can retrieve a memory of an object in a particular context, guided by a limited amount of contextual information, aligned with prior studies [24,28]. Furthermore, the similar patterns of object exploration and rearings in the presence of a variable number of retrieval cues indicate that exploration levels of familiar objects in the present task can be directly related to contextual novelty.

Place cells in the hippocampus can remap in response to contextual changes [60], enabling different activity patterns to represent different environments [61,62]. However, none of the previous studies have shown whether remapping has any behavioral significance (i.e., is there any link between remapping and memory?). There are interesting studies showing that artificial activation of a selected group of hippocampal neurons can modify mice behavior in a contextual fear memory [63–66] or a reward-oriented paradigm [67]. Here, we go a step further and show that the degree of spontaneous incidental memory, not involving reward or punishment, correlates with the amount of remapping of CA3 place cells. This result is fundamental for a better understanding of contextual representations and their role in memory, without the influence of explicit motivational factors.

Our results contain a large variability between (and within) animals in terms of how these contextual representations, coded by place cell activity, change with contextual variation. This has been repeatedly shown in the literature [19,68,69]. Place cells in CA1 can remap within the same session even though there are no changes in the context, possibly due to modifications in the internal state of the animal [19]. This variability could be explained if remapping does not solely reflect the observable and controlled properties of the environment, but rather subjective perceptions and predictions the animal is making about the environment during retrieval [70]. In fact, inter- and intra-individual variability in behavior is the rule rather than the exception [71–73]. Interestingly, we showed that the amount of memory retrieved was not only reliant on the test condition itself (number of cues) but was specifically related to the animal’s internal representation of the context. Thus, sorting cell activity by animal’s internal representation more clearly reflects the different encoding of distinct contexts for the animal. This contrast between the effect of the contextual manipulation (AC, PC, NC) and the actual change in contextual representations reveals that controlled experimental contextual manipulations only reflect a subset of what the animal perceives as context. Accordingly, individual representation of the same space is variable, but behaves in a dichotomous manner as expected according to the dual-process model of memory [43–45]. This representation can be more or less subject to discrimination or generalization depending on uncontrolled experimental variables that increase in importance as the context becomes more degraded. Our results contrast with those of Kentros and colleagues [20], who described an increase in the number and tuning of place cells when the number of environmental cues is higher. Importantly, that particular study recorded place cells from CA1, not CA3 like in the present manuscript, so those results are not easily comparable with ours. Having said this, in our task the 6 distal visual cues are likely not the only cues animals use to orient themselves in the arena, this could make the AC and NC conditions more similar in terms of their contextual representation. This might be the reason why we were able to disentangle the internal representation from the external condition in the electrophysiology experiments. Despite the difference in the distal removable visual cues, there is a conserved ambiguity between the contexts that is reflected in place cell activity and also in behavior. In addition, the effect of the pharmacological manipulations over the generalization/discrimination was also variable and likely dependent on the retrieval state of the animal (Fig 6B). We propose that the pharmacological treatments biased individual pattern completion/pattern separation balance, leading to changes in the activation of the internal contextual representation during retrieval, and therefore modifying the distribution of exploration times towards discrimination or generalization. In the labilization/reconsolidation experiments, both contextual and object information are degraded. Thus, object memory retrieval demands a reconstruction of the original experience from degraded information to guide memory. Combined, our results show that incidental reactivation occurs for all the elements of the original event even in contextually degraded conditions, and that retrieval is related to a low degree of remapping in CA3 place cells. In addition, these results reveal a contextual specificity of the reactivation and subsequent labilization of object memory traces, since only the corresponding contextually associated object memory trace was reactivated under contextual exposure in the NC and AC retrieval sessions (Fig 2E). Consistent with this idea, human studies found that activity in the Prh correlates with recollection in response to partial spatial cues [74,75] and associative retrieval led to neocortical activity corresponding to incidental reactivation of all the items of a particular event [76].

Our results are consistent with other studies suggesting a requirement of NMDAR activity in the CA3 region for long-term spatial memory reactivation, but only when the amount of contextual information available is limited [28,77]. Since spatial place fields are generated and modified in an NMDAR-dependent manner [46–51], it is possible that memory reactivation under cue-degraded conditions (where presumably information input via the perforant pathway is reduced) relies more heavily on the activity of attractor states and the strength of recurrent collateral connections in CA3. Changes in NMDAR activity could affect the attractor circuit activation threshold since enhancement of CA3 recurrent collateral connections is NMDAR-dependent [55,56]. Therefore, NMDAR antagonists could be preventing reverberant activity in the CA3 region, leading to a retrieval deficit, and NMDAR agonists could favor attractor circuit stability for the corresponding memory trace, thereby promoting retrieval. Could memory retrieval under contextually degraded cues rely on LTP expression mechanisms that involve an NMDAR-dependent component in CA3? Although the role of NMDAR for LTP was mainly associated with induction [78,79], a recent publication showed that recall cues can drive transient increases in excitability of engram cells and this, in turn, can influence the efficacy of memory retrieval [53], pointing out a role of NMDAR in the expression of LTP. Interestingly, synaptic activation of NMDARs triggers internalization of Kir2.1 channels leading to increases in engram cell excitability. In our task, NMDAR-dependent increases in cell excitability in CA3, driven by the degraded context, could be controlling the efficacy of memory retrieval. Moreover, there is evidence that NMDARs could participate in both spatial memory and place field activity [28,50,80,81]. In addition, the fact that a voltage-gated calcium channel blocker (Nimodipine) prevented the increase in memory generalization due to CA3 D-Cyc infusion points toward a role of calcium channels and possible excitability changes regulating the strength of cue-driven retrieval under partial cue conditions.

Computational models suggest that the attractor circuit in CA3 could lead to pattern separation or pattern completion, as a function of the relative strength of the attractor circuit and the nature of the external inputs from the EC and DG [3,82,83]. Sensory information entering through the perforant path could be used to either recover the original engram in CA3 and cause retrieval of the entire memory, or to form a new engram guided by the new entrance of information. Thereby, these theories imply that recognition of an experience will ultimately depend on whether pattern completion-like or pattern separation-like processes rule over CA3 activation patterns. This is exactly what we have seen in the electrophysiological data, in which the degree of remapping balances between retrieval of a familiar memory with low remapping or recognizing a different situation with no retrieval and higher remapping. In this context, we speculate that, under AC and PC conditions, input from the EC through the perforant path could lead to attractor activity in CA3 corresponding to an object-in-context representation and, by the time DG inputs arrive, the strong attractor dynamics of the circuit dominate and prevent remapping favoring a pattern completion process. But if the sensory inputs induced by retrieval cues are weaker, like in the PC2 condition, CA3 attractor circuit activity will also be weaker, offering the opportunity for these attractor states to follow DG input and promote environmentally induced remapping.

In this scenario, we have shown that interfering with plasticity-related mechanisms in the DG can actually favor memory retrieval in conditions where spatial cues would normally be insufficient to guide retrieval. In this sense, NMDAR inactivation could affect the ability of the DG to remap in response to novel stimuli [25], leading to stable memory representation in CA3 and memory retrieval under incomplete contextual information. Electrophysiological studies that support this line of thought have shown that LTP decay is an active process that requires NMDAR activity [84]. Consistent with the putative DG role in pattern separation, NMDAR partial agonist D-Cyc prevented memory retrieval in the PC condition, suggesting a shift in the hippocampal balance towards an acquisition mode instead of a retrieval mode. This is in line with the fact that DG granular cells are depolarized after exposure to a novel environment, indicating that these cells could be involved in setting the hippocampal circuit into an acquisition mode [85]. Moreover, although most studies focused on the role of NMDARs on spatial memory acquisition [86–89], a few studies already showed a post-acquisition facilitation by administration of NMDAR antagonists [84,90]. Although inactivation of NMDARs in the DG can modulate memory retrieval, it is likely that inactivation of NMDARs in the DG is not sufficient to support memory retrieval by itself. Rather, a requirement for external contextual inputs with the potentiality to act as a cue for retrieval on the DG-CA3 system is also needed in order to sustain memory generalization (even under this modulation). In this regard, it is possible that an absence of contextual cues, like in the NC condition, could prevent a context from acting as a cue for retrieval. Only further experiments will be able to tell us whether even under a higher reduction in the number of cues, DG NMDAR inactivation will still lead to an increase in the level of generalization.

Even though in real life, episodic memory retrieval usually occurs in a degraded context, this process and the balance between pattern completion–pattern separation have not been studied under these conditions. Relatively little work has been done linking remapping and behavior (see Allegra and colleagues [91]). Here, using an incidental memory task, we showed that retrieval of an object in context memory is reflected by the levels of CA3 remapping, demonstrating a clear relationship between remapping and associative episodic-like memory processing. Furthermore, we describe NMDARs as a key player in regulating the balance between retrieval and memory differentiation processes. While an increase in CA3 NMDAR activity boosts memory retrieval, DG NMDAR activity enhances the memory differentiation process. Our results contribute to understanding the adaptive nature of memory, and how it can guide behavior in a way that is consistent with changes in the environmental cues and the internal state of the individual.

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002706

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.