Hippocampal cells multiplex positive and negative engrams

The hippocampus is involved in processing a variety of mnemonic computations specifically the spatiotemporal components and emotional dimensions of contextual memory.1–3 Recent studies have demonstrated vast structural and functional heterogeneity along the dorsal-ventral axis1, 5 of the hippocampus. The ventral hippocampus has been shown to be important in the processing of emotion and valence.6–9 Here, we combine transgenic and all-virus based activity-dependent tagging strategies to visualize multiple valence-specific engrams in the vHPC and demonstrate two partially segregated cell populations and projections that respond to appetitive and aversive experiences. Next, using RNA sequencing and DNA methylation sequencing approaches, we find that vHPC appetitive and aversive engram cells display distinct transcriptional programs and DNA methylation landscapes compared to a neutral engram population. Additionally, while optogenetic manipulation of tagged cell bodies in vHPC is not sufficient to drive appetitive or aversive behavior in real-time place preference, stimulation of tagged vHPC terminals projecting to the amygdala and nucleus accumbens (NAc), but not the prefrontal cortex (PFC), had the capacity drive preference and avoidance. These terminals can also undergo a “switch” or “reset” in their capacity to drive either, thereby demonstrating their adaptable contributions to behavior. We conclude that the vHPC contains genetically, cellularly, and behaviorally distinct populations of cells processing appetitive and aversive memory engrams. Together, our findings provide a novel means by which to visualize multiple engrams within the same brain and point to their unique genetic signatures as reference maps for the future development of new therapeutic strategies. One sentence summary The hippocampus contains neurons that correspond to positive and negative engrams, which are segregated by their molecular, cellular, and projection-specific features.

The hippocampus is involved in processing a variety of mnemonic computations specifically the spatiotemporal components and emotional dimensions of contextual memory. [1][2][3] Recent studies have demonstrated vast structural and functional heterogeneity along the dorsal-ventral axis 1,5 of the hippocampus. The ventral hippocampus has been shown to be important in the processing of emotion and valence. [6][7][8][9] Here, we combine transgenic and all-virus based activity-dependent tagging strategies to visualize multiple valence-specific engrams in the vHPC and demonstrate two partially segregated cell populations and projections that respond to appetitive and aversive experiences. Next, using RNA sequencing and DNA methylation sequencing approaches, we find that vHPC appetitive and aversive engram cells display distinct transcriptional programs and DNA methylation landscapes compared to a neutral engram population. Additionally, while optogenetic manipulation of tagged cell bodies in vHPC is not sufficient to drive appetitive or aversive behavior in real-time place preference, stimulation of tagged vHPC terminals projecting to the amygdala and nucleus accumbens (NAc), but not the prefrontal cortex (PFC), had the capacity drive preference and avoidance. These terminals can also undergo a "switch" or "reset" in their capacity to drive either, thereby demonstrating their adaptable contributions to behavior.
We conclude that the vHPC contains genetically, cellularly, and behaviorally distinct populations of cells processing appetitive and aversive memory engrams. Together, our findings provide a novel means by which to visualize multiple engrams within the same brain and point to their unique genetic signatures as reference maps for the future development of new therapeutic strategies.
Within the brain we find a rich repository of memories that can be imbued with positive and negative valenced information. These experiences leave enduring structural and functional changes that are parceled up into discrete sets of cells and circuits comprising the memory engram 10 . Recent studies have successfully visualized and manipulated defined sets of cells previously active during a single experience [10][11][12][13][14][15][16][17] . However, how multiple engrams of varying valences (hereafter defined as cells that differentially respond to appetitive or aversive events in a stimulus-independent manner 18 ) are represented within the same brain region remains poorly understood. Previous studies have suggested the ventral hippocampus (vHPC) selectively demarcates and relays emotional information to various downstream targets. We sought to characterize the molecular and cellular identities and the behaviorally-relevant functions of vHPC cells processing appetitive and aversive engrams, with a focus on ventral CA1 (vCA1). In order to address the question of how the vCA1 processes multiple emotional experiences, we first devised a strategy combining two cFos-based tagging methods with endogenous cfos immunohistochemistry to visualize engrams across three discrete timepoints. First, we anatomically charted the projection patterns of appetitively and aversively-tagged vCA1 cells to measure structural overlap and segregation across various downstream targets. Second, we performed genome-wide RNA sequencing (RNA-seq) and DNA methylation sequencing to investigate the genetic features of these sets of cells. Lastly, we behaviorally tested the causal role and functional flexibility vHPC cells in both a cell body and projection-specific manner.
First, to access cells across multiple timepoints and in an activity-dependent, within-subject manner, we used the Fos-based transgenic animal, TRAP2 18 , under the control of 4-pattern observed within medial sections (Fig. 1c-d). We found that the order of tagging, tagging shock and then female exposure, did not impact the anatomical recruitment of these cells (Extended Data Fig. 2) Next, using the same dual tagging strategy as described above, we asked whether or not these appetitive-and aversive-tagged populations were recruited during subsequent experiences of similar valences. To that end, we added a third time point by visualizing endogenous cFos expression 90 minutes after a final behavioral experience, i.e. exposure to sweetened condensed milk or restraint stress 19 (Fig. 1e-f). To control the order of experiences, we counterbalanced the four groups, each of which contained two tagged populations of cells (i.e. "aversive" for cells tagged by fear conditioning and "appetitive" for cells tagged by male-female interactions), followed by a third experience of varying valence, restraint stress as another aversive and sweetened condense milk as the other appetitive experience, which was captured by endogenous cFos expression. Our groups were as follows: Aversive-Appetitive-Restraint Stress (GROUP 1), Aversive-Aversive-Restraint Stress (GROUP 2), Appetitive-Aversive-Sweetened Condensed Milk (GROUP3), and Appetitive-Appetitive-Sweetened Condensed Milk (GROUP 4) as shown in Fig. 1f. In each group we measured the cellular co-localization of TdTomato, EYFP, and cFos to infer which cells were active in one or more of the three experiences (Fig. 1g). All mice exhibited a similar proportion of tagged cells across all three tagging approaches regardless of method of tagging or valence (Fig. 1h, and Extended Data Fig. 3).
Histological analyses revealed that there were significantly higher rates of cells showing colocalization of TdTomato and EYFP when labelled with the same experiences (i.e. appetitive and appetitive) and lower rates of co-localization when labeling differently valenced experiences (i.e. appetitive and aversive) in Fig. 1i. Further, we observed significant colocalization between TdTomato, EYFP, and cFos when mice were subjected to three aversive or three appetitive experiences (Fig. 1l), i.e. GROUP 2 and GROUP 4 respectively. Appetitive-tagged cells were preferentially reactivated when mice were exposed to sweetened condensed milk, and aversivetagged cells were preferentially reactivated when mice were exposed to restraint stress ( Fig. 1jk). Together, these findings raise the intriguing possibility that vCA1 designates emotionallyrelevant information to two partially non-overlapping sets of cells.
Moving forward, we chose two valenced engrams, shock and male-female exposure, as a proxy to better understand how the vHPC processes these aversive and appetitive experiences. Therefore, we used the same dual-tagging strategies to characterize the basic physiological properties of vHPC cells (Extended Data Fig. 4). TRAP2 mice were co-injected with the same activity dependent viruses and tagged in the same manner as mentioned previously (Extended Data Fig. 4a-b). Interestingly, we did not observe any differences in firing frequency, suprathreshold characterization, adaptation rate, input rate, or spike rates (Extended Data Fig.   4c-l) suggesting that, despite recruiting partially non-overlapping sets of cells for appetitive and aversive experiences, these tagged cells themselves share similar physiological characteristics.
Despite these physiological similarities, vHPC cells have been shown to project to a myriad of distinct brain regions involved in stress and approach-avoidance behaviors, thereby forming multi-regional networks involving emotion and memory 1 . We speculated that within these networks there exists structural heterogeneity partly defined by whether an experience is appetitively or aversely valenced, as has been observed in areas including the amygdala 10,25,27 , nucleus accumbens 29 , and medial prefrontal cortex 36 . Using our dual tagging strategy, we next traced vHPC outputs tagged by appetitive and shock experiences in a within-subject manner and measured the axonal fluorescence intensities in the following target areas, given their crucial role in emotional processing: BLA, NAc, PFC, dorsal hypothalamus, fornix, and the dentate gyrus ( Fig.   2a-b). Interestingly, we observed both red and green fluorescent signal between appetitive and aversive-tagged vCA1 terminals in the medial BLA (A.P. -1.8) and the PFC (including IL and PL), while also finding evidence of structural segregation in the following regions: we observed predominantly stronger EYFP (appetitive) projections, as measured by fluorescence intensities, to the basomedial amygdala Posterior part (BMP), anterior commissure anterior part (ACA), fornix, dorsal medial hypothalamic nucleus (DMD and DMV), and the lower layer of the dentate gyrus ( Fig. 2c-o). Further, we found predominantly stronger TdTomato (aversive) projections to the anterior BLA, posterior BLA, NAc core, and the upper layer of the DG (Fig. 2c-o). Previous studies have demonstrated that neurons in regions such as the PFC 36 , NAc 29 , and BLA 10.25 , collectively process experiences in a manner that can be anatomically segregated or heterogeneous. We posit that appetitive-and aversive-tagged vHPC terminals are embedded in a larger network of emotional memory processing that can be partly defined both by unique anatomical patterns and by the activity-dependent recruitment of ensembles involved in processing a specific valence.
Recent studies have identified unique molecular profiles of vCA1 cells containing distinct projection targets. These vCA1 projections transmit information to multiple areas involved in emotion and memory and form a network that can be organized by unique architectural features, including vCA1 cell inputs, outputs, and transcriptional signatures 1 . We wanted to ask the question of whether cell populations as a whole, for appetitive or aversive engram, have distinct genetic differences based on experience. Accordingly, we next examined whether or not the molecular composition of vHPC cells contained distinct genetic profiles. To catalogue the molecular landscape of vHPC cells in an activity-dependent manner, we tagged an appetitive experience (i.e. male-female interactions), a aversive experience (i.e. multiple shocks), or a neutral experience (i.e. exposure to the same conditioning cage without an appetitive or aversive stimuli; see methods; Fig. 3a). We then performed RNA-seq using tagged nuclei (EYFP+) isolated by Fluorescence-Activated Cell Sorting (FACS), which showed 0.5~0.6% tagged nuclei from each group ( Fig. 3b-c). Comparing aversive and appetitive vCA1 cells to the neutral group, we identified 474 differentially expressed genes (DEGs) in aversive vCA1 cells ( Fig. 3d and 3f), including 340 down regulated genes and 134 upregulated genes (Fig. 3g). We also identified 1,104 DEGs in appetitive vCA1 cells compared to the neutral group ( Fig. 3e and 3f), including 1,025 down regulated genes and 79 upregulated genes (Fig. 3g). There were 842 unique DEGs in appetitive engram cells and 212 unique DEGs in aversive engram cells (Fig. 3f), suggesting distinct transcriptional landscapes in these two populations. Furthermore, the top 20 downregulated DEGs identified for aversive and appetitive vCA1 cells   showed no overlap with each other, and 14 among the top 20 upregulated DEGs for aversive and appetitive vCA1 cells do not overlap with each other (Fig. 3d and 3e, Table S1 and S2), supporting the distinct transcriptomes in these neurons. Interestingly, among the 262 shared DEGs we identified one gene Nufip1 (Nuclear Fragile X Mental Retardation Protein Interacting Protein 1) that was upregulated in appetitive vCA1 cells and downregulated in aversive vCA1 cells (Fig. 3g).
This gene encodes a nuclear RNA binding protein that contains a C2H2 zinc finger motif and a nuclear localization signal 47 . Diseases associated with NUFIP1 mutations include Peho Syndrome (progressive encephalopathy with Edema, Hypsarrhythmia and Optic atrophy), an autosomal recessive and dominate, progressive neurodegenerative disorder that starts in the first few weeks or months of life. Its interacting protein FMRP1 is essential for protein synthesis in the synapse 48 and CGG trinucleotide expansion mutation of FMR1 gene coding FMRP1 cause Fragile X syndrome, the most common intellectual disability in males 49 . Further investigation of the functional significance of NUFIP1 and other DEGs could reveal mechanistic insights to the transcriptomic plasticity engaged by the valences of memory.
To gain deeper insight into the molecular signatures of transcriptomes associated with aversive and appetitive vCA1 cells, we performed Gene Ontology (GO) analysis of the up-and downregulated pathways in aversive and appetitive vCA1 cells (Fig. 3h-k). We found that the top upregulated pathway in aversive vCA1 cells involved neurotransmitter complexes such as ionotropic glutamate receptor activity (Fig. 3h). This finding is consistent with previous studies using brain tissues that identified 3,759 differentially methylated DNA regions in the hippocampus associated with 1,206 genes enriched in the categories of ion gated channel activity after contextual fear conditioning 48,50,51 . Interestingly, we found the top downregulated pathway in aversive vCA1 cells involved DNA mismatch repair (Fig. 3i). Although the top upregulated pathways in appetitive vCA1 cells also include ionotropic glutamate receptor activity (Fig. 3j), the top downregulated pathways in appetitive vCA1 cells enrich on axoneme assembly and microtubule bundle formation (Fig. 3k) different from the pathways downregulated in appetitive vCA1 cells (Fig. 3i). Next, we compared the RNA-seq data between aversive and appetitive vCA1 cells directly. We identified 494 DEGs, including 47 upregulated genes and 447 downregulated genes (Extended Fig. 5a, Table S3). Furthermore, we found that five pathways were upregulated in appetitive vCA1 cells such as nuclear exosome RNAse complex and 16 pathways were down regulated in appetitive engram compared to aversive vCA1 cells, such as axoneme assembly (Extended Fig. 5b and 5c). These differentially altered signaling pathways between aversive and appetitive vCA1 cells, which resulted from multiple comparisons, support our conclusion that these neurons indeed represent transcriptionally distinct subpopulations. One future direction is to explore the functions of these DEGs and altered signaling pathways. As a proof-of-concept, we applied GeneMANIA 54 to predict the functions of the top 20 DEGs in Fig. 3l-o. For instance, the brain-specific angiogenesis inhibitor 2 (BAI2) is uniquely identified for the DEGs upregulated in aversive vCA1 cells (Fig. 3l), and the Ionotropic glutamate receptor is only implicated for the DEGs upregulated in appetitive vCA1 cells (Fig. 3n). Similarly, the WNT signaling pathway component Dishevelled is uniquely identified for the DEGs downregulated in aversive vCA1 cells (Fig. 3m), and the TRPV1-4 channel is only implicated for the DEGs upregulated in appetitive vCA1 cells (Fig.   3o). Loss-and gain-function study of these predicted genes will provide mechanistic insight for the molecular signatures of engram neurons with different memory valences.
Besides the distinct transcriptomic profiles in engram neurons described above, recent studies reveal the transcriptional priming role of epigenetic regulation in engram 43. To explore whether the dynamic epigenomic landscape also contributes to the specificity of engram neurons with different valences, we performed reduced representation bisulfite sequencing (RRBS) to characterize the DNA methylation landscapes in aversive and appetitive vCA1 cells. As shown in  (Table S5). Based on these DMCs, we identified differentially methylated regions (DMRs) that contain at least two DMCs for each DMR ( Fig. 4c and 4d). These DMRs allow us to identify the differentially methylated genes (DMGs) that either contain or close to these DMRs. The top 20 DMGs with the change of methylation level larger than 20% and p value smaller than 0.05 show no overlapping between aversive and appetitive vCA1 cells, suggesting different memory valences trigger different changes of DNA methylations. Among the 266 DMGs in appetitive vCA1 cells and the 98 DMGs in aversive vCA1 cells, only 32 DMGs are commonly shared (Fig. 4e, Table S6), confirming the distinct DNA methylation landscape between these two populations of engram cells. Last, we performed Gene Ontology (GO) analysis of DMGs in aversive and appetitive vCA1 cells ( Fig. 4f and 4g). We found that the pathways in aversive vCA1 cells mainly enriched in the structure and function of synaptic connections (Fig. 4f). However, the enriched pathways in appetitive vCA1 cells are much more diverse including axon growth, synaptic connection, ion channels, and RNA polymerase II transcription regulator complex (Fig. 4g). These differentially enriched pathways between aversive and appetitive vCA1 cells suggest potentially distinguished functional outputs attributed by DNA methylation at the transcriptional level to confer the specificity of memory valences. One interesting future direction is to explore the maintenance and functions of these DNA methylation changes during the consolidation and recall of memory. Overall, our results in Fig. 3 and 4 showed that the distinct molecular signatures of aversive and appetitive vCA1 cells are reflected at the transcriptomic and epigenomic levels likely contributing to the different valences of memory.
vCA1 is known to have monosynaptic projections to the BLA, NAc, and the mPFC 24 . Previous studies have shown that these brain regions are important in the modulation of both appetitive and aversive experiences, especially in the NAc 29, 34 and the BLA 25,26 , in which molecularly and topographically distinct cellular populations have been identified for each behavior. Therefore, we tested for a causal role of tagged vHPC cell bodies and its selected terminals in driving behavior by first infusing a virus cocktail of AAV9-cFos-tTA and AAV9-TRE-ChR2-EYFP or AAV9-TRE-EYFP into vCA1. We then implanted an optical fiber bilaterally above either the vCA1 or its terminals, vCA1-BLA, vCA1-NAc, vCA1-PFC (Fig. 5a). We first found that all terminals were capable of activating their corresponding downstream targets by assessing increases in cFos levels following stimulation (Extended Data Fig. 6). Following 10 days of recovery, a separate group of mice were taken off DOX and tagged with aversive (i.e. shock) or appetitive (i.e. malefemale interactions) experiences. As illustrated in Fig. 5b, for the first set of experiments, on Day 1, the aversively tagged mice were placed in a real-time place preference/avoidance (RTPP/A) chamber on day 1 to assess baseline levels. On day 3, the animals were placed back into the PTPP/A chamber; this time they received optical stimulation at 20Hz bilaterally on one side and no stimulation on the other. We found that optical stimulation of vCA1-BLA or vCA1-NAc terminals drove aversion ( Fig. 5e & f), whereas the EYFP controls, vCA1 cell body stimulation, and vCA1-PFC terminals did not statistically deviate from baseline levels ( Fig. 5c, d, & g). On day 5 of the experiment, the mice were subjected to an induction protocol, as previously reported 13 , to test the capacity of the engram to "switch" the behavior it drives. The aversive-tagged male mice were placed in a new chamber with a female mouse for 10 minutes while receiving optical stimulation for the entire duration of exposure. Afterwards, the animals were placed back in their chambers and assessed for behavioral changes on day 7. In this post-induction test, we observed that optical stimulation of vCA1-BLA terminals was now sufficient to drive preference despite driving aversion in the pre-induction test earlier (Fig. 5e). The induction protocol also revealed that vCA1-NAc terminals, which previously were sufficient to drive aversion, now had reversed or "reset" their capacity to modulate behavior and returned to baseline levels (Fig. 5g). Lastly, we saw no changes in the EYFP controls, vCA1 cell body, or vCA1-PFC stimulation (Fig. 5c, d, & g).
Next, we asked whether appetitive-tagged vCA1 terminals to the BLA, NAc, or PFC were sufficient to modulate behavior (Fig. 5h). Similar to the above findings, optical stimulation of vCA1-BLA ( Fig. 5k) and vCA1-NAc ( Fig. 5l) terminals, but not in any of the other groups ( Fig. 5i, j, & m), was sufficient to drive place preference. During the induction phase of the experiment, we placed the mice into a fear conditioning chamber where they received multiple foot shocks and simultaneous optical stimulation of the appetitive-tagged terminals. This experiment assessed if these vCA1 terminals are able to "switch" or "reset" their capacity to drive preference in a manner mirroring the experiments above. Indeed, we found that in the post-induction tests, the vCA1-BLA group switches from driving preference to aversion (Fig. 5k) while the vCA1-NAc group resets from driving preference back to baseline levels ( Fig. 5l). Furthermore, we did not observe this effect in neutral-tagged animals (Extended Data Fig. 7), nor did we see any statistically significant behavioral changes for the RTPP/A task in the EYFP controls, vCA1 cell body stimulation group, or vCA1-PFC stimulation group (Fig. 5i, j, & m). Importantly, we tested whether optical stimulation may cause increases in motor behaviors. We found no significant difference in distance travelled across all groups during light on or off epochs in an open field (Extended Data Fig. 8). Moreover, the lack of preference or avoidance observed from vCA1 cell body stimulation raised the possibility that vCA1's role in driving behavior is determined in a terminal-specific manner 24 ; a notion that dovetails with recent studies suggesting that computations along the axons of a given cell body can differentially drive behavior in accordance with the downstream target 8 . Taken together, these results suggest that vHPC cell bodies relay their behaviorally-relevant and valence specific content to its downstream targets despite sharing partially non-overlapping molecular and neuronal signatures and distinct projection patterns. This is corroborated by evidence showing that vHPC axonal outputs preferentially route independent features of a given behavior 7 .
Here we have shown that the vHPC processes appetitive and aversive experiences in defined populations of cells that are partially distinct at the molecular, cellular, and projection-specific levels. We also demonstrated their capacity to drive behaviors through functionally plastic projection-specific terminals. Our immunohistochemical data suggest that vCA1 contains at least three populations of neurons: two subsets that can be demarcated based on their anteriorposterior locations and preferential response to appetitive or aversive valences, and a third population that responds to both, perhaps reflecting a biological predilection for salience. 5 While their exact brain-wide structural and functional outputs remain undetermined, we speculate that our observed population of vCA1 cells responding to aversion are a superset of recently observed anxiety cells that transmit information to the hypothalamus 27 and PFC 28 . Moreover, vCA1 cells processing aversive or appetitive perhaps route information to and innervate the BLA and NAc at differing anatomic, receptor-and cell-type specifically optimizing their capacity to integrate mnemonic information 27 .
Additionally, by combining transgenic activity-dependent tagging strategies with all-virus-based expression of fluorophores, our design permits the visualization of multiple ensembles in a within-subject manner, which coalesces with previous studies monitoring and manipulating a single ensemble active at two discrete time points as well 29 . By intersecting these approaches with genetic sequencing strategies, these tools enable the tagging, manipulation, and molecular documentation of cells processing aversive and appetitive behaviors opening the possibility of cataloguing topographical similarities and differences between the two in a brain-wide manner.
For instance, future studies may probe the molecular composition of appetitive-and aversivetagged vCA1 cells along its anterior-posterior axis, and test whether or not the transcriptomic profiles of these cells differ both across valence and their physical location. Moreover, it is intriguing that vCA1 appetitive-and aversive-tagged terminals showed evidence of segregation within the amygdala and hippocampus. Consequently, subsequent research may seek to functionally tease out their contributions to behavior, ands measure the types of information they are transmitting through a combination of imaging and terminal-specific perturbation approaches.
It is important to constrain interpretations of engrams given the vast array of genetic tools used to access cells in an activity-dependent manner. For instance, vCA1 ensembles vary drastically in size and in activity patterns across learning and memory. These numbers range from single percentages and can reach ~35% of cells depending on immediate early gene marker used, tagging strategy, rodent line, and viral tools employed [32][33][34][35] . Fittingly, we believe that our dualtagging strategy partly over-samples the number of tagged cells given the time-frame necessary for tagging to occur (e.g. 48 hours off Dox; 72 hours post-4OHT injection [37][38][39][40], and future experiments may aim to improve the temporal resolution of contemporary tagging approaches to enhance the signal-to-noise ratio of experience-related tagging to background tagging or leakiness. Additionally, cFos+ cells only reflect a subset of an engram that is otherwise distributed throughout the brain and recruits numerous immediate early genes, cell types, and complementary physiological activity that activity-dependent markers may fail to capture due to the relatively slow timescales of gene expression. We indeed note that while cFos+ cells indicate recent neural activity, a cell that was recently active does not necessarily have to express cFos especially across brain regions and cell types which presumably have their own thresholds for cFos expression. These points highlight the complex nature of a memory engram and underscore a caution that is warranted when interpreting data in light of inherent technical limitations. We believe that a multi-faceted approach combining genetic tagging strategies with real-time imaging during complex behaviors will help to disentangle the relationship between neural activity, genetic modifications, and systems-level changes in response to learning and memory. Our terminal manipulations are in line with recent studies demonstrating terminal-specific routing of information from vCA1 cell bodies through a variety of single, bi-, and trifurcating processes 8,41,42 . Our data provide a "gain-of-function" demonstration that activated vCA1 terminals can drive preference or aversion, which we believe was obfuscated by cell body stimulation given that the latter presumably activates the cell body and most, if not all, corresponding axonal outputs. The former would selectively modulate a set of terminals emerging from vCA1 that project to a distinct target area while only minimally affecting vCA1 terminals projecting to other target areas. Additionally, while the molecular basis underlying our valence "switch" experiments remain unknown, we posit that the plasticity of the transcriptome could confer the ability to the switch which aspect of behavior a terminal drives given the rapid and enduring responses to learning and memory present at the genetic level in tagged cells. 43 Indeed, the comprehensive transcriptional landscape in the mouse hippocampus is dynamic across the lifespan of memory formation and recall, and its experience-dependent modifications are largely present specifically in tagged cells within minutes of learning and lasting for several weeks 42 . Future experiments may perform RNA-seq on cells before and after the induction protocol to measure the ensuing transcriptional changes and identify putative loci mediating such functional plasticity. Further, our sequencing data enhances our understanding of how the ventral hippocampus genetically parses out both appetitive and aversive engrams, how those experiences can cause a change in the upregulation and down-regulation of discrete genes, and how these experiences can have lasting effects on the epigenetic genome through the methylation study (Fig. 4). Future studies, for instance, can build on recent work assessing the molecular and projection-defined connectivity between vCA1 and its downstream targets with MAPSeq 1 , while adding an activity-dependent component, such as our dual memory tagging approach, to measure organizing principles of the ventral hippocampus and its projections in a manner that takes a cell's history into account. This study in particular provides an influential platform for characterizing the organization of the ventral hippocampus and its non-random input-output patterns, and we posit that this logic includes an activity-dependent-defined dimension such that appetitive and aversive experiences engage unique input-output hippocampal circuitry as well.
Finally, while the physiological basis by which terminals can switch or reset their capacity to drive valence-specific behaviors remains unclear, future studies may consider candidate mechanisms including homeostatic plasticity, dendritic growth and retraction, and counter-conditioningfacilitated changes along the axon-dendrite interface between vCA1 and the BLA or NAc 25,26 . In line with this speculation, previous studies have demonstrated that the dorsal hippocampus contains defined sets of functionally plastic cell bodies sufficient to drive aversive or appetitive behavior, while the BLA contains fixed populations that are sufficient to drive each behavior contingent on the anatomical location stimulated along the anterior-posterior axis as well as on their projection-specific sites. 7,8,[25][26][27] Subsequent research can be aimed at utilizing our dualmemory tagging approach to genetically and physiologically map out which cell-types and circuits show such hardwired or experience-dependent responses to emotional memories as well. In our study, we speculate that vCA1 cells become appetitive or aversive in an experience-dependent manner as opposed to being hardwired for either valence per se, as has been observed in many other brain regions (e.g. BLA 25 ). However, though it remains possible that a subset of these vCA1 cells are preferentially tuned to process a given valence in an experience-independent manner.
Indeed, the notion that experience itself modifies these neurons to become appetitive or aversive does not preclude the possibility that subsequent experiences will modify their capacity in a flexible manner to drive a given behavior associated with a given valence.
Moreover, in our study, while the criteria for valence was met in Fig. 1 (e.g. hippocampus cells responded differentially to stimuli of positive and negative valence in a manner independent of stimulus features), our subsequent data sets honed in on a single appetitive (e.g. male-female social interactions) and a single aversive (e.g. foot shocks) experience for thorough molecular, anatomical, and behavioral profiling. Importantly, we highlight here that in order to make a direct claim about valence, the task structure needs to be held constant. For instance, an alternative interpretation of our current data is that our observed differences in gene enrichment profiles and anatomical segregation are due to the inherent differences in tasks used (e.g. male-female social interactions vs. contextual fear conditioning), and not valence per se. It remains possible therefore that male-female social interactions, which require multi-modal integration of social and contextual cues, recruit a unique set of genes and ventral hippocampal anatomy in comparison to contextual fear conditioning, in which a conditioned cue (e.g. context) is associated with an unconditioned cue (e.g. shock), and therefore recruits task-specific molecular and anatomical activity. As such, we propose that future studies may focus on tagging experiences in which most, if not all, environmental features are held constant except the valence associated with a unimodal stimulus itself, which we believe opens up an experimental platform for studying how multiple experiences of similar or varying valence differentially engage cellular populations. While we believe that this partially accounted for by utilizing multiple experiences of similar valence in Fig. 1, additional future experiments may implement an in vivo recording approach in which putative negative-and positive-tagged cells are imaged while task structure and valence are parametrically varied. Building from this notion, our RNA-Seq and DNA methylation data provides an additional means by which memories alter the functions of the genome in a valence-dependent manner, both in healthy and pathological states. For instance, our methylation data identified Pin1 60, 61 among 266 DMGs in the appetitive vCA1 cells, which is known for its neuroprotective qualities specifically in neurodegeneration through the regulation in the spread of cis p-tau. Follow-up studies may leverage these appetitive engrams by altering these DMGs as a means to alleviate psychiatric and neurodegenerative diseases.
Together, we propose that, in addition to processing spatio-temporal units of information, the hippocampus contains discrete sets of cells processing aversive and appetitive information that relay content-specific and behaviorally-relevant information to downstream areas in a molecularly-defined and projection-specific manner, thus collectively providing a multi-synaptic biological substrate for multiple memory engrams. Immunohistochemistry. Immunohistochemistry follows protocols previously reported 15,16,18 .