The pERKs of temporal order memory in mice

Temporal Order Memory (TOM), the ability to discriminate between events according to when they occurred, is a key component of episodic-like memory, although whether this reflects temporal tagging itself or interactions between competing memory traces remains unresolved. Understanding the molecular mechanisms that support temporal memory requires behavioral approaches capable of capturing the continuous dynamics of natural exploration. Despite extensive evidence implicating the prefrontal cortex (PFC) in temporal memory, the intracellular signaling mechanisms supporting temporal order discrimination remain poorly understood. Here, we combined high-resolution automated behavioral phenotyping with molecular analyses to characterize the behavioral and signaling dynamics underlying TOM in mice. Mice were trained in a spontaneous object-recognition TOM task and tested after short-term (3 h) or long-term (24 h) retention intervals. Exploration was quantified using an artificial intelligence–based behavioral analysis pipeline that enables continuous and unbiased assessment of object exploration. Phosphorylation of extracellular signal-regulated kinase 2 (ERK2) and expression of the ERK phosphatase MKP3/DUSP6 were analyzed in the PFC and hippocampus (HIP) following habituation, a single training session, or two sequential training sessions. Additionally, a Temporal Novel Object Recognition (TeNOR) protocol was used to evaluate the integrity of memory traces. Mice displayed robust TOM performance across sexes and retention intervals. Molecular analyses revealed no significant changes in hippocampal ERK signaling, whereas the cytosolic fraction of the PFC exhibited dynamic, experience-dependent modulation of ERK2 phosphorylation. A single 15-min training session induced a transient increase in ERK2 activation, while a second session 45 min later was associated with a reduction in this increase. This modulation was accompanied by increased MKP3 expression, consistent with the recruitment of an active regulatory feedback mechanism. Continuous behavioral tracking further revealed temporal features of memory expression that were not captured by conventional summary measures; it identified an early, rapid decay of discrimination for older object memories in the TeNOR task, suggesting that TOM performance relies on resolving competitive retrieval between co-existing memory traces. Together, these findings reveal early dynamic ERK2-MKP3 signaling in the PFC associated with temporal discrimination, and demonstrate how high-resolution phenotyping in naturalistic behavioral paradigms can reveal molecular dynamics associated with temporal organization of experience.

Understanding how the brain organizes experiences across time is a central challenge in memory research. Episodic memory requires not only recognizing past events, but also encoding their temporal relationships (Eichenbaum, 2017; Tulving, 2002). In rodents, Temporal Order Memory (TOM) paradigms have been widely used to study episodic-like temporal discrimination by evaluating the ability to distinguish between familiar objects encountered at different moments in time (Barker et al., 2007; Dere et al., 2018; Mitchell, 1998). Successful TOM performance is typically inferred from the preferential exploration of an object experienced earlier relative to one encountered more recently (Barker and Warburton, 2011). An unresolved question in TOM research is whether the preference for the “older” object during testing reflects not genuine temporal discrimination, but rather an emergent property of competitive retrieval or retroactive interference (Autore et al., 2023). Spontaneous object-recognition paradigms, such as TOM, provide a naturalistic and minimally stressful framework for studying memory, as they rely on the animal’s intrinsic exploratory drive and novelty preference rather than appetitive or aversive reinforcements (Antunes and Biala, 2012; Ennaceur and Delacour, 1988). Furthermore, Temporal Novel Object Recognition (TeNOR) provides an opportunity to dissociate these possibilities by independently probing the recognition of old and recent familiar objects relative to a new one.

While ethologically relevant, spontaneous object-recognition paradigms pose important analytical challenges because exploratory behavior is inherently dynamic and continuous. Recent advances in automated behavioral quantification, pose-estimation tools, and machine-learning approaches have transformed the study of animal behavior by enabling automated, high-resolution quantification of spontaneous exploration (Datta et al., 2019; Mathis et al., 2018). These methods move beyond traditional manual analysis, which is often prone to subjective bias and limited reproducibility, and capture the continuous dynamics of behavior with greater precision and scalability.

In this context, we recently developed RAINSTORM (Real and Artificial Intelligence for Neuroscience—Simple Tracker for Object Recognition Memory; D’hers et al., 2025), an automated analysis pipeline that combines pose-estimation data, geometric analysis, and artificial intelligence–based behavioral classification to quantify exploratory behavior in rodents. By learning from expert-defined scoring criteria, RAINSTORM minimizes evaluator bias while providing rapid and accurate assessment of object exploration and recognition memory performance. In the present study, we used RAINSTORM to assess TOM in mice, enabling high-resolution analysis of temporal memory expression and exploratory dynamics.

The hippocampus (HIP) and prefrontal cortex (PFC) have both been implicated in temporal memory and object recognition processes (Barker and Warburton, 2011). However, their functional contributions are distinct: while the HIP is crucial for the integrated ‘what-where-when’ components of episodic memory (Clayton et al., 2001), the PFC specifically supports the organization of sequential information and the representation of temporal relationships between events (DeVito and Eichenbaum, 2011). Damage to the PFC often results in a “disordered” memory, characterized by the ability to recognize an object (the “what”) but a fundamental failure to recollect its specific temporal (the “when”) or spatial (the “where”) context (Barker et al., 2007; Janowsky et al., 1989). Despite this circuit-level understanding, the intracellular signaling mechanisms supporting these computations remain poorly understood.

One candidate mechanism for TOM processing is the extracellular signal-regulated kinase (ERK1/2) pathway, a central regulator of synaptic plasticity, learning, and long-term memory formation (Sweatt, 2004; Thomas and Huganir, 2004). ERK signaling is dynamically modulated during learning and novelty exposure across multiple species and memory paradigms (Adams and Sweatt, 2002; Antoine et al., 2014). Importantly, previous work in Drosophila proposed that repeated learning events can reset ERK activation states, potentially preventing molecular overlap between sequential memories (Cattaneo et al., 2020). Similarly, studies in the crab Neohelice granulata demonstrated that ERK signaling is critically involved in the formation of a long-term memory induced by a two-trial training protocol with a 45-min interval, further supporting the idea that tightly regulated ERK dynamics may contribute to memory formation under sequential learning conditions (Ojea Ramos et al., 2021). Whether similar mechanisms contribute to TOM processing in mammals remains unknown.

The duration and magnitude of ERK activation are tightly regulated by phosphatases such as MKP3/DUSP6, a cytosolic dual-specificity phosphatase that negatively regulates ERK1/2 signaling (Caunt and Keyse, 2013; Farooq and Zhou, 2004). Although ERK signaling has been extensively associated with synaptic plasticity and memory processes, the contribution of its regulatory phosphatases (MKP3 in particular) remains largely unexplored. Importantly, previous work from our group has consistently highlighted the relevance of cytosolic ERK dynamics during long-term memory consolidation and reconsolidation across different experimental models (Feld et al., 2005, 2008; Krawczyk et al., 2015, 2016; Ojea Ramos et al., 2026). These findings suggest that precise regulation of cytosolic ERK activity is critical for memory processing. Dynamic interactions between ERK phosphorylation and MKP3-mediated dephosphorylation could therefore provide an active regulatory mechanism to limit molecular overlap and compartmentalize intracellular signaling during sequential learning. Nevertheless, the role of MKP3 in TOM has not yet been investigated.

In the present study, we combined automated behavioral quantification with spontaneous object-recognition paradigms to characterize the behavioral and molecular dynamics underlying TOM in mice. Specifically, we analyzed short- and long-term TOM performance, characterized ERK2 phosphorylation dynamics and MKP3 expression in the PFC and HIP during memory acquisition, and used the TeNOR paradigm to investigate the relative stability of old versus recent memory traces. Our results identify a dynamic ERK2 and MKP3 signaling profile in the PFC associated with sequential learning and suggest a potential molecular mechanism involved in the early processing of temporally close memory traces.

Animals

Male and female C57BL/6 elite mice (obtained from the Academia Nacional de Medicina, Buenos Aires, Argentina) were used for all experiments. A total of 107 mice, starting at 8–9 weeks of age, were employed across different experimental cohorts, as detailed in Table 1. To mitigate potential carryover effects from repeated testing, the assignment of objects and their spatial locations were fully counterbalanced across all training and testing sessions. Furthermore, subjects were exposed to a unique set of novel objects for each session, ensuring no object was repeated for any experimental subject.

Mice were group-housed in plastic cages (2–4 individuals per cage) under standard laboratory conditions: controlled temperature (24° ± 2°), a 12-h light-dark cycle (lights on at 07:00), and ad libitum access to food and water. All animal handling and experimental procedures were performed in strict accordance with the NIH Guide for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee (CICUAL–FCEN, UBA, protocols N° 165/2021 and 165b/2025).

Object recognition protocols

Behavioral tasks were conducted in a white polyurethane open-field arena (20 × 30 cm). The arena floor was removable and equipped with four hidden magnets situated 5 cm from each corner, designed to securely anchor the objects and prevent displacement during exploration. The objects included blue LEGO bricks, glass jars, cylindrical plastic objects, and ping-pong balls. All objects were previously validated in our laboratory to ensure the absence of any innate bias or intrinsic preference among mice.

Temporal order memory (TOM)

The TOM protocol was composed of the following phases:

Habituation (HAB): Three sessions on consecutive days (5, 10, and 15 min), during which each mouse was placed in the empty arena and allowed to freely explore it.

First training session (TR1): On the fourth day, the mouse was placed in the arena containing two identical objects and was allowed to explore for 15 min.

Second training session (TR2): After a 45-min interval, the mouse was put into the arena again to explore a new pair of identical objects for 15 min.

Testing session (TS): 3 h (short-term) or 24 h (long-term) after the end of the second training, the animal entered the arena to explore one object from TR1 (old familiar object) and one from TR2 (recent familiar object) for 5 min. TOM is defined by the preference for exploring the old familiar object, known from TR1, over the recent familiar one (presented at TR2).

Temporal novel object recognition (TeNOR)

The arena, objects, and protocol are identical to the TOM protocol, except for the TS session.

Testing session (TS): 24 h after the end of the TR2 session, mice were divided into two groups. The first group (Prior TR) entered the arena to explore one object from TR1 and a novel object, while the second group (Latter TR) explored one object from TR2 and a novel object. In both cases, recognition memory is defined by the preference for exploring the novel object over the familiar (old from TR1, or recent from TR2) one.

In both protocols, discrimination performance was expressed as a Discrimination Index (DI), calculated as the proportion of exploration time directed toward the target object (i.e., the older object in TOM and the novel object in TeNOR) relative to the total object exploration time during the first 4 min of TS.

Automated quantification of exploratory behavior

Exploratory behavior was recorded using a Genius ECAM-8000 webcam interfaced with Bonsai software (version 2.8.5; Lopes et al., 2015). Body-part tracking was performed using DeepLabCut (version 3.0.0; Mathis et al., 2018; Nath et al., 2019). Specifically, we manually labeled 300–350 frames extracted from 30 representative videos, allocating 95% of the frames for training and the remaining 5% for testing. We trained a ResNet-50-based neural network for 800 epochs (test error: 3.02 px, train error: 1.39 px; image resolution: 720 × 640 px). The trained network was subsequently applied to extract positional data from all experimental videos. The resulting coordinates (nose, head, neck, ears, body center, and tail base) were processed using the RAINSTORM pipeline (D’hers et al., 2025), which leverages artificial neural networks to generate precise frame-by-frame behavioral labeling based on the animal’s position relative to each object. The models were previously trained and validated against manually annotated datasets generated by multiple independent experimenters, achieving performance comparable to expert human labeling (D’hers et al., 2025). This high-resolution labeling enables the automatic assignment of exploration to each object, allowing both the quantification of cumulative exploration times and the computation of time-resolved discrimination indexes throughout the testing session.

Tissue preparation and western blotting

To evaluate molecular dynamics, mice were euthanized at specific time points (0, 1, and 2 h) following either HAB, TR1, or TR2 (the full two-session training protocol). Following euthanasia by cervical dislocation, brain tissue from the HIP and PFC regions was rapidly dissected, and extracts enriched in nuclear or cytosolic proteins were prepared as described in Salles et al. (2015). Briefly, HIP or PFC tissue was homogenized in 250 or 150 μL, respectively, of buffer A (Table 2), with 8 strokes using a glass-glass Dounce homogenizer with a type B (tight) pestle. The homogenate was centrifuged for 15 min at 1000 g, and the supernatant (cytosolic protein-enriched extract) was aliquoted and stored at -20 °C until use. HIP or PFC pellets were resuspended in 50 or 20 μL, respectively, of buffer B (Table 2) and incubated on ice for 20 min. They were then centrifuged for 15 min at 12,000 g. The supernatant (nuclear protein-enriched extract) was aliquoted and stored at -20 °C until use. The entire extraction protocol was performed at 4 °C.

Protein concentration was quantified using the Bradford method (Bradford, 1976), with bovine serum albumin (BSA) standards of known mass serving as references. Absorbance was measured at 595 nm in triplicate, and the concentration of each sample was calculated using the linear portion of the BSA calibration curve. Samples (20–30 μg of total protein) were denatured in loading buffer (see Table 2) by heating for 5 min at 95°C, and were resolved by SDS-denaturing polyacrylamide gel electrophoresis (SDS-PAGE) for 20 min at 80 V, and 1–1.5 h at 120 V. Proteins were then electrotransferred under 350 mA for 1 h to polyvinylidene fluoride (PVDF) membranes. Samples from all experimental groups were distributed across gels to minimize potential blot effects.

Membranes were left to dry out and reactivated in methanol for a few seconds before blocking for 1 h with 4% fat-free powdered milk in TTBS (TBS—0.1% Tween-20). After blocking, the membranes were incubated with primary antibodies overnight at 4 °C, and then with secondary antibodies at room temperature for 1 h. The following primary antibodies were used: phospho-ERK1/2 (1:1,000, Santa Cruz Biotechnology cat. sc-7383, mouse monoclonal), total ERK1/2 (1:1,000, Cell Signaling Technology cat. #9102, rabbit polyclonal), and total MKP3 (1:1,000, Santa Cruz Biotechnology cat. sc-377070, mouse monoclonal). Total MKP3 was detected by chemiluminescence (Anti-Mouse IgG, Cell Signaling cat. #7076, 1:5,000) using Luminol reagent (ECL, BioRad, cat. #170-5060) and an Amersham Imager 600. Total and phospho-ERK1/2 were detected using IRDye subclass-specific secondary antibodies (donkey anti-mouse, 1:10,000, Li-Cor cat. #926-32212; donkey anti-rabbit, 1:10,000, Li-Cor cat. #926-68073) on an Odyssey imaging system. The images obtained were analyzed using ImageJ 1.53j (Schneider et al., 2012) and GelAnalyzer (version 26.1; Istvan and Istvan, 2026), and the signal intensity was determined for the specific bands of either phosphorylated or total ERK1 and ERK2, and total MKP3. Total MKP3 immunoreactivity was normalized to the total protein loaded in each lane, as determined by Ponceau Red staining of the PVDF membrane. After transfer, membranes were stained with Ponceau Red for 5 min and destained for 30 min. Total protein loading was estimated by densitometric quantification of five representative Ponceau-stained bands per lane, and the mean value was used as the loading reference for normalization. The five bands were of identical width, equally spaced, and uniformly distributed across the 25–70 kDa molecular-weight range; the same molecular-weight regions were quantified for all lanes across all membranes.

Sample sizes ranged from 8 to 10 animals per experimental group (corresponding figures depict exact n values).

Statistical analysis

Data processing and statistical analyses were conducted using R (version 4.6.0; R Core Team, 2026) and RStudio (version 2026.05.0-218; Posit Team, 2026). To robustly model our experimental data, which included bounded proportions and non-normally distributed continuous variables, we employed Generalized Linear Mixed-Effects Models (GLMMs).

Behavioral discrimination data (DI) from the TOM and TeNOR tasks were modeled using a Beta regression family with a logit link function via the glmmTMB package (version 1.1.14; Brooks et al., 2017). This approach is mathematically appropriate for variables strictly bounded between 0 and 1; notably, all calculated DI values fell within this range, as every mouse explored each object at least once per session. The primary model formula for TOM incorporated fixed effects for group, sex, and their interaction. To account for repeated measures and control for environmental factors, a parsimonious random effects structure was applied using random intercepts for individual subjects nested within their homecage identity (1| Cage/Mouse). Finally, the performance package (version 0.12.2; Lüdecke et al., 2021) was used to calculate the intraclass correlation coefficient (ICC) to assess the variability associated with these random intercepts. If the random intercept for “Cage” resulted in a variance of zero (a singular fit), the cage effect was considered negligible.

Molecular data exhibited a continuous, right-skewed positive distribution and were therefore analyzed utilizing a Gamma regression family with a log link function. The model evaluated the interacting fixed effects of training condition and wait time. To account for non-experimental variance and the nested structure of the data, sex was included as an additive fixed effect, and a random intercept was applied solely for home-cage identity.

Following model specification, assumptions for all GLMMs (including residual uniformity, absence of overdispersion, and outlier detection) were evaluated and confirmed using simulated residuals via the DHARMa package (version 0.4.7; Hartig, 2026). Main effects and interactions were assessed using Analysis of Variance (ANOVA). Type II ANOVA was explicitly utilized for the molecular data to account for the asymmetric experimental design (specifically, the absence of the HAB 1h and TR2 0h groups), whereas Type III tests were applied to balanced models. Finally, post-hoc pairwise comparisons and targeted contrasts—against the chance level (DI = 0.5) for behavior, and against the global baseline (HAB 0h) for molecular assays—were conducted using estimated marginal means (emmeans package, version 1.10.3; Lenth and Piaskowski, 2026). All-possible-pairwise comparisons between experimental groups were adjusted using Tukey’s method. Contrasts evaluating the DI against chance were adjusted using the False Discovery Rate (FDR) procedure. Conversely, targeted contrasts against the control group (HAB) were adjusted using Dunnett’s method. Correlation analyses were performed to assess associations between behavioral and molecular variables and are presented to complement the primary statistical analyses as exploratory observations; therefore, to preserve sensitivity to underlying biological signals, correlation p -values are reported without adjustment for multiple comparisons. Statistical significance was defined at α = 0.05. For detailed statistical analysis, also see the “Supplemental Statistical Analysis” file.

Behavioral characterization of TOM formation

First, we characterized the formation of Temporal Order Memory (TOM) using a behavioral protocol consisting of two training sessions (TR1 and TR2) separated by a 45-min interval, during which mice were exposed to different pairs of identical objects. The temporal order memory was then tested (TS) at short-term (3 h) or long-term (24 h) intervals after TR2. During TS, mice were presented with one object from each training session (Figure 1A). Given that both objects were familiar, successful TOM was inferred when mice preferentially explored the object presented earlier in time (old familiar object) over the one presented more recently.

Object exploration was quantified using the automated behavioral analysis pipeline RAINSTORM (D’hers et al., 2025). Artificial neural networks trained on manual annotations provided by five expert human scorers were used to identify exploratory behavior and automatically estimate object exploration times during TR1, TR2, and TS. Exploration times during TR1 and TR2 revealed no apparent differences between object/side (Supplementary Figure S1), indicating the absence of intrinsic object bias.

A first cohort of mice (see Table 1) was used to evaluate TOM performance after short-term retention intervals at both 2 and 3 months of age, and after a long-term retention interval at 3 months of age. No significant effects of sex [χ 2 (1) = 0.21, p = 0.6451], group [e.g., age and TS interval; χ 2 (2) = 0.70, p = 0.7042], or sex-by-group interaction [χ 2 (2) = 0.42, p = 0.8094] were detected in a type III ANOVA, indicating that both males and females performed similarly in this task at both ages and retention intervals. Moreover, post-hoc testing with FDR correction confirmed that discrimination performance was significantly above chance level (DI = 0.5) in each of the three TS (Figure 1B and Table 3), indicating that mice preferentially explored the older object in all conditions.

However, regarding total exploration time during TS (Figure 1C), a type III ANOVA revealed significant effects of the interaction between sex and group [χ 2 (2) = 10.47, p = 0.0053]. Tukey post-hoc comparisons revealed that 3-month-old females exhibited significantly higher total exploration times during LTM testing compared to all other sex/group pairs (Figure 1C).

Having established that 3-month-old mice exhibited robust TOM retention over a 24 h interval, we next evaluated long-term TOM performance in 2-month-old mice using an independent cohort (Figure 2 and Table 1). No significant effects of sex [χ 2 (1) = 0.18, p = 0.6703] were detected in a type III ANOVA, and discrimination index was significantly above chance level (DI = 0.5) in both males (11.4% above chance, z = 2.62, p = 0.0089; Figure 2A) and females (13.6% above chance, z = 3.83, p = 0.0003).

A detailed assessment of DI dynamics throughout the test session revealed a progressive emergence of temporal order discrimination in both sexes (Figure 2B). Consistent with this pattern, cumulative exploration curves showed a preferential exploration of the older object during testing, while no clear object/location preference was observed during training (Figure 2C). Comparable temporal dynamics were consistently observed across all TOM test sessions analyzed, irrespective of age or retention interval (Supplementary Figure S1).

ERK2 phosphorylation dynamics during TOM

To investigate the molecular mechanisms underlying TOM, we characterized ERK1/2 phosphorylation dynamics in both the PFC and the HIP. Mice were either placed on the empty arena (HAB), trained once (TR1) or twice (TR2), and then sacrificed at 0, 1, or 2 h following the first session. Brain tissue was dissected and processed to obtain purified cytosolic- and nuclear-enriched protein fractions (Figure 3A). Analysis of pERK1/2 and total ERK1/2 immunoreactivity following SDS-PAGE (Figure 3B) revealed that, although the antibodies detected both ERK1 (44 kDa) and ERK2 (42 kDa), the ERK2 band exhibited the most consistent experience-dependent modulation. Therefore, subsequent analyses focused on ERK2 phosphorylation. No significant changes in ERK2 activation were detected in either the nuclear (Supplementary Figure S2) or cytosolic (Supplementary Figure S3) fractions of the HIP. Nuclear and cytosolic fractions from PFC evidenced similar activation patterns; since significant experience-dependent modulations of ERK2 phosphorylation were primarily identified in the cytosolic fraction of the PFC, the main focus was placed there.

Using a Gamma regression model, we found a significant interaction between training condition and time in a type II ANOVA [χ 2 (2) = 15.15, p = 0.0005]. Specifically, a single exposure to novel objects (TR1) induced a significant increase in ERK2 activation 1 h after training (TR1 1h vs. HAB 0h, z = 2.94, p = 0.0178), while no significant difference was found immediately after training (TR1 0h vs. HAB 0h, z = −0.51, p = 0.9638). Interestingly, when animals were subjected to the second training session (TR2), the initial phosphorylation event observed at 1 h was suppressed (TR2 1h vs. HAB 0h, z = 0.35, p = 0.9879; TR1 1h vs. TR2 1h, z = 2.61, p = 0.0090; Figure 3C). However, a delayed and significant increase in ERK2 activation emerged 1 h after TR2 relative to habituated controls at the same time point (TR2 2h vs. HAB 2h, z = 2.58, p = 0.0282; Figure 3C). A similar pattern was observed in the nuclear fraction of the PFC, where pERK2 levels also showed a significant Training × Time interaction [χ 2 (2) = 9.43, p = 0.0089]. Post-hoc analyses likewise revealed significant differences at the 1-h time point (TR1 1h vs. HAB 0h, z = 3.51, p = 0.0025; TR1 1h vs. TR2 1h, z = 3.09, p = 0.0019; Supplementary Figure S2). We also assessed total ERK2 protein as a measure of ERK2 expression; however, no changes were found in any of the groups (Supplementary Figures S2, S3). These results are consistent with an experience-dependent regulation of cytosolic as well as nuclear ERK activation in the PFC during sequential learning.

Training-induced modulation of MKP3 expression in the PFC

Given that cytosolic ERK1/2 activation can be negatively regulated by the dual-specificity phosphatase MKP3 among others (Caunt and Keyse, 2013), we then characterized its cytosolic expression levels in the PFC (Figures 3B,D). A Type II ANOVA revealed a significant interaction between Training and Time [χ 2 (2) = 7.0, p = 0.0302]. Post-hoc comparisons against the global baseline (HAB 0h) showed that MKP3 expression remained largely stable across groups, but significantly increased immediately after TR2 (TR2 1h vs. HAB 0h; z = 3.00, p = 0.0146). Furthermore, pairwise comparisons within the same post-training interval revealed higher MKP3 levels immediately after TR2 rather than 1 h after TR1 (TR2 1h vs. TR1 1h; z = 2.5, p = 0.0123; Figure 3D). These results suggest a potential link between ERK2 signaling and MKP3 function during TOM acquisition.

Experience-dependent temporal coupling between ERK phosphorylation and MKP3 expression

Experience-dependent dynamics of ERK activation and MKP3 expression appear to be inversely related. Thus, increased ERK2 activation is associated with lower MKP3 protein levels, and vice versa. Consequently, we investigated whether they exhibit coordination at the individual level. To this end, we performed a Pearson correlation analysis segmented by experimental group in the cytosolic fraction of the PFC.

Immediately following a single training session (TR1 0h), we observed a significant positive correlation between ERK2 activation and MKP3 expression (R = 0.689, p = 0.027; Figure 4A). This result suggests a parallel mobilization of both the kinase and the phosphatase in response to the novel stimulus. However, this relationship was drastically different 1 h after a second training session (TR2 2h), observing a significant negative correlation (R = -0.773, p = 0.025; Figure 4A). This finding indicates that, only after two training sessions, higher levels of MKP3 are associated with a decrease in ERK phosphorylation, consistent with the recruitment of a negative feedback loop induced by training repetition. In contrast, habituated control groups (HAB) showed no significant associations at any time point evaluated (p > 0.05).

Exploration time predicts MKP3 response magnitude but not ERK activation

To determine whether object exploration modulates the molecular response, we evaluated whether the total exploration time during TR1 and TR2 correlated with ERK2 activation and/or MKP3 protein levels. Interestingly, we found no significant correlations between exploration time and ERK2 activation at any time point (p > 0.05; Supplementary Figure S4). In contrast, MKP3 expression did show a graded relationship with exploratory behavior. Specifically, in the TR1 1h group, a significant negative correlation was found between exploration time and MKP3 expression (R = -0.808, p = 0.0047; Figure 4B).

TeNOR reveals distinct memory traces

Finally, to delve into the TOM mechanism, we developed the Temporal Novel Object Recognition (TeNOR) protocol. This protocol mirrors the TOM training but diverges at the TS phase: mice were split into two groups, each tested 24 h later by exploring a completely novel object alongside either the object explored during TR1 (Prior TR) or during TR2 (Latter TR; Figure 5A).

Continuous behavioral tracking using RAINSTORM enabled us to examine the temporal dynamics of the DI throughout the entire TS (Figure 5B). Our analysis revealed a significant main effect of the tested group [χ 2 (1) = 4.98, p = 0.0257; Figure 5D]. Mice in the Latter TR group robustly discriminated between the familiar object and the novel one, exhibiting a strong preference for novelty (14.2% above chance, z = 3.17, p = 0.0030). Mice in the Prior TR group showed no significant discrimination against the novel object (0.5% above chance, z = 0.11, p = 0.9099). Both groups exhibited similar exploration times [χ 2 (1) = 2.16, p = 0.1412].

Notably, time-resolved analysis of the DI revealed a brief initial period (approximately 1 min) during which mice from Prior TR preferentially explored the novel object over the familiar one. Because this time window was defined post-hoc based on inspection of the DI trajectory, we considered this analysis exploratory. During the first minute, both testing groups exhibited a strong preference for the novel object compared either to the object explored during TR1 or TR2 (23.1% above chance, z = 3.23, p = 0.0025 for Latter TR; 17.5% above chance, z = 2.43, p = 0.0150 for Prior TR; Figure 5C). These findings suggest that memory of the old familiar object was preserved and capable of guiding behavior during the initial phase of testing. However, this effect was transient, suggesting that the relative influence of both memory representations on exploratory behavior changes rapidly over the course of testing.

In this study, we characterized behavioral and molecular dynamics underlying Temporal Order Memory (TOM) in mice. Our findings indicate that TOM is associated with highly dynamic early molecular signaling within the Prefrontal Cortex (PFC). Specifically, we identified a training-induced increase in ERK2 activation that is selectively modulated by subsequent experience and active, experience-dependent changes in MKP3 expression accompanying ERK2 dynamics. Furthermore, our results suggest that the preference for the older object in TOM may arise either from interference between competing memory traces or from differences in their relative stability during retrieval. The precise nature of TOM, whether it reflects the temporal discrimination of sequential events or the behavioral readout of memory interference, remains an open question.

Interestingly, a related form of retroactive interference has been described in an object-in-context task, where a second experience occurring 1 h after the first disrupts long-term memory formation in a PFC-dependent manner (Martínez et al., 2014). The authors proposed that temporally proximal memories may compete for common cellular resources during consolidation, highlighting the PFC as a key structure in regulating interactions between memory traces. Furthermore, the idea that the PFC regulates competition between memory representations is consistent with previous evidence showing that this structure contributes to the control of interference during retrieval of recognition memories (Bekinschtein et al., 2013; Morici et al., 2015). In these studies, the PFC was proposed to bias behavioral output toward the most relevant memory trace among competing alternatives. Also, Autore et al. (2023) recently demonstrated that retroactive interference can promote adaptive forgetting through competition of newly formed engrams. In line with this framework, TOM performance may not necessarily reflect a timestamping mechanism. Instead, it may emerge from a competitive process in which consolidation of a first recognition memory is affected by a second, temporally close experience. Under this view, both object representations may be retained, but their behavioral expression is determined by the relative accessibility of the competing memory traces and by the mechanisms through which the PFC regulates their interaction. Whether this brain area actively allocates temporally proximal memories to competing neuronal or molecular ensembles, thereby shaping their subsequent behavioral expression, deserves further investigation.

ERK2 signaling in the prefrontal cortex

The Extracellular Signal-Regulated Kinase (ERK) pathway is a well-established requirement for long-term memory (LTM) across species (Atkins et al., 1998; Blum et al., 1999; Feld et al., 2005; Martin et al., 1997; Purcell et al., 2003; Ribeiro et al., 2005; Sweatt, 2004; among others), and object recognition memory stabilization has been specifically shown to require ERK activation (Kelly et al., 2003; Nagai et al., 2007). Our results show that a single object exploration session (TR1) induces a significant increase in ERK2 phosphorylation in the PFC 1 h later. Remarkably, a second training session (TR2) administered 45 min after TR1 effectively “reverses” or suppresses this peak at the 1-h time point, followed by a delayed increase in activation of ERK2 at the 2-h time point (Figure 6), indicating that ERK2 signaling is highly sensitive to exploratory experience and dynamically regulated by the interaction between temporally proximal memory-encoding events.

This rapid, experience-dependent reset of ERK2 activation is highly reminiscent of the findings by Cattaneo et al. (2020) in Drosophila, where a second learning trial erased the increase in ERK activation induced by the first one. The observed ERK2 dynamics may reduce “molecular overlap” between separate events, thereby promoting the formation of distinct memory representations. Together with the behavioral evidence discussed above, these findings raise the possibility that dynamic ERK2 regulation contributes to the interaction and competition between coexisting memory traces. However, in the absence of information regarding the allocation of these signaling events within the brain, we cannot rule out the possibility that separate neural circuits compete for behavioral control.

Interestingly, in the absence of TR2, ERK2 activation eventually drops to baseline levels by the 2-h time point. This late decline does not appear to correlate with an upregulation of MKP3, suggesting that the spontaneous termination of ERK2 signaling after a single training experience may rely on constitutive dephosphorylation mechanisms or alternative phosphatase activity.

MKP3 expression in TOM

Feedback inhibition of ERK1/2 signaling can be mediated either by rapid posttranslational modifications of pathway components or by the slower induction of genes encoding pathway inhibitors, which requires de novo gene expression and protein synthesis (Lake et al., 2016). In support of this latter mechanism, Zhang and colleagues demonstrated that overexpression of wild-type MKP3 reduced EGF-induced ERK1/2 phosphorylation only after 2 h of stimulation, whereas a phosphatase-dead mutant had no effect. A third possibility is that differential expression of scaffold proteins can also significantly modulate ERK1/2 dynamics. We observed a significant increase in MKP3 expression specifically following TR2 (TR2 1h), which correlates with the suppression of ERK2 activation (Figure 3D). These findings are consistent with a role for MKP3 in delayed feedback regulation of ERK2 signaling.

A particularly intriguing feature of this regulatory dynamic is the positive correlation between pERK2 and MKP3 immediately following the first training session (TR1 0h, Figure 4A). While MKP3 is a canonical negative regulator of ERK—typically induced by ERK signaling to establish a negative feedback loop (Lake et al., 2016)—the immediacy of this correlation is unexpected. Because this parallel increase happens immediately after exploration, it is too fast to rely on de novo gene transcription. One possible explanation is that this early positive correlation reflects rapid translational control. The ERK pathway is known to regulate translation by phosphorylating 4E-BP1, promoting its dissociation from the translation initiation factor eIF4E (Herbert et al., 2002). By releasing this translational block, the initial surge in pERK2 could immediately drive the rapid synthesis of MKP3 from pre-existing mRNA pools. Under this framework, the early positive correlation does not contradict MKP3’s role as a negative regulator, but rather is compatible with the initiation of a feedback process. These dynamics may reflect coordinated regulation of kinase and phosphatase activity.

Finally, the transition to a strong negative correlation (TR2 2h, Figure 4A) suggests that after repeated training, the relationship between MKP3 and ERK2 becomes consistent with effective negative regulation. Further supporting this regulated response is the negative correlation between exploration time during TR1 and MKP3 expression 1 h later (TR1 1h, Figure 4B). While it might seem counterintuitive that a longer exploration time is associated with lower levels of this phosphatase, this relationship could reflect a permissive window for memory consolidation. A prolonged sensory experience may require more sustained kinase activity to process the memory trace, potentially delaying the upregulation of MKP3. Conversely, brief exploratory events might trigger an early deployment of the phosphatase, truncating the signaling cascade earlier. Under this view, the MKP3 response could help scale the duration of the molecular trace to the intensity of the behavioral experience.

It is important to note that the present mechanism was characterized using two discrete novelty-encoding events separated by a fixed interval under controlled laboratory conditions. In natural environments, animals experience continuous streams of novel stimuli, and whether similar ERK2/MKP3 dynamics are repeatedly engaged across successive encoding events or are subject to additional regulatory mechanisms remains unknown. Thus, while our findings identify a molecular signature associated with the processing of sequential experiences under these experimental conditions, extending this framework to more naturalistic patterns of experience will require future investigation.

TeNOR: insights on competing memories

To address whether the preference for the earlier-explored (Prior TR) object in TOM reflects a failure to recognize it, we developed the TeNOR (Temporal Novel Object Recognition) protocol. We found that while the memory of the recent object (Latter TR) drove sustained and robust novelty-seeking behavior throughout the test, the trace for the earlier-explored object appeared highly vulnerable. Time-resolved analysis of the DI revealed a brief initial period (approximately 1 min) during which mice in the PriorTR group preferentially explored the novel object over the familiar one. Because this time window was defined post-hoc following inspection of the DI trajectory, this analysis was considered exploratory. The transient initial discrimination is consistent with the possibility that the memory representation of the first object remained temporarily accessible and capable of influencing behavior during the initial phase of testing.

This behavioral dynamic is consistent with the retroactive interference framework: TR2 interferes with the ongoing consolidation of the object memory from TR1, a process that may be accompanied by the molecular dynamics observed here. In this context, the temporal preference seen in TOM could be an emergent property of two co-existing traces with different relative competitive strengths, rather than a strict binary state of remembering versus forgetting.

Limitations and scope

While our results provide a clear molecular correlate for TOM, some limitations must be considered. First, our molecular assays were performed on whole-tissue extracts; while this identifies a global signaling state, it does not account for the potential heterogeneity of neuronal populations (e.g., pyramidal neurons vs. interneurons) that may play distinct roles in temporal processing. Furthermore, while the correlation between behavior, ERK2 activation, and MKP3 expression is strong, further pharmacological or genetic interventions—specifically targeting MKP3 or ERK2 in the PFC during the inter-trial interval—would be required to establish not only a definitive causal link between this molecular reset and the behavioral preference, but also the specific PFC circuitry involved in this type of memory.

In conclusion, our findings support the hypothesis that Temporal Order Memory in mice may rely, at least in part, on Prefrontal Cortex mechanisms involved in regulating interactions between object recognition memory traces. The orchestrated activation of ERK2 and its rapid experience-dependent downregulation, together with increased MKP3 expression, define a molecular signature associated with the resolution of competing memory traces. Future causal manipulations will be required to determine whether this signaling directly regulates temporal discrimination or related forms of memory interference.