Behavioural interference at event boundaries reduces long-term memory performance in the virtual water maze task without affecting working memory performance

Narrative episodic memory of movie clips can be retroactively impaired by presenting unrelated stimuli coinciding with event boundaries. This effect has been linked with rapid hippocampal processes triggered by the offset of the event, that are alternatively related either to memory consolidation or with working memory processes. Here we tested whether this effect extended to spatial memory, the temporal specificity and extent of the interference, and its effect on working-vs long-term memory. In three computerized adaptations of the Morris Water Maze, participants learned the location of an invisible target over three trials each. A second spatial navigation task was presented either immediately after finding the target, after a 10-s delay, or no second task was presented (control condition). A recall session, in which participants indicated the learned target location with 10 ‘pin-drop ’ trials for each condition, was performed after a 1-h or a 24-h break. Spatial memory was measured by the mean distance between pins and the true location. Results indicated that the immediate presentation of the second task led to worse memory performance, for both break durations, compared to the delayed condition. There was no difference in performance between the delayed presentation and the control condition. Despite this long-term memory effect, we found no difference in the rate of performance improvement during the learning session, indicating no effect of the second task on working memory. Our findings are in line with a rapid process, linked to the offset of an event, that is involved in the early stages of memory consolidation.

An alternative account has been advanced by Leszczynski (2011), who argued for the role of working memory (WM) maintenance in subsequent long-term memory (LTM).WM maintenance starts at the offset of to-be-remembered input (such as at an event boundary) with relevant information maintained in neuronal networks that include cortical regions and the hippocampus (Dimakopoulos, Mégevand, Stieglitz, Imbach, & Sarnthein, 2022;Fuentemilla, Penny, Cashdollar, Bunzeck, & Düzel, 2010;Peters & Reithler, 2022;Poch, Fuentemilla, Barnes, & Düzel, 2011).Successful WM maintenance enables later LTM performance (for review see Hartshorne & Makovski, 2019).However, the presentation of novel stimuli during maintenance may disrupt WM, and so lead to reduced LTM performance.Thus, retroactive memory interference by novel stimuli presented at event offsets (Ben-Yakov et al., 2013) may equally indicate the disruption of either WM processes, or of a rapid and early stage of memory consolidation.
Research to date has primarily focussed on recall of single exposures to narrative episodes or lists, limiting the scope for dissociating WM and LTM components.Whether boundary-related retroactive memory interference occurs in other memory domains also remains unknown.One such domain is spatial memory where, additionally, relevant tasks allow the separation of WM and LTM.Spatial memory and learning have been widely studied using the Morris Water Maze task (Morris, 1984).In the traditional set-up, rodents are placed in a pool in which they learn the location of a submerged escape platform.For comparative studies in humans, a variety of computerized 'virtual' versions of the MWM have been developed (Astur, Taylor, Mamelak, Philpott, & Sutherland, 2002;Schoenfeld, Schiffelholz, Beyer, Leplow, & Foreman, 2017;Thornberry, Cimadevilla, & Commins, 2021).Successful LTM learning is quantified by shortened travel distances to reach the platform when returned to the maze after an extended break.Given multiple testing trials in a single session, the location of the target platform must be actively maintained in memory from one trial to the next.Thus, performance improvements within a session can be taken as a measure of WM (Frielingsdorf, Thal, & Pizzo, 2006;Vorhees & Williams, 2006;Wisman, Sahin, Maingay, Leanza, & Kirik, 2008) in line with WM's role in planning and carrying out behaviour (Cowan, 2008), while performance after an extended break is a measure of LTM in line with the role of memory in storing information over longer periods for later retrieval.
Here, we used a virtual MWM (vMWM) task in which participants used a game controller to navigate to an invisible platform within a computerized environment.Across three consecutive trials, participants learned the location of the target platform.To test whether retroactive interference by the presentation of novel stimuli at the event boundary extended to this form of spatial episodic memory, we presented a secondary spatial task, with no extra memory load, immediately after the first task.To investigate the temporal specificity of the effect, we introduced conditions in which the secondary spatial task was presented either after a 10-s delay or not at all, with a different version of the maze used for each condition.LTM for the target location in each maze was tested in separate participants either after a 1-h or a 24-h break.In a replication of Ben-Yakov et al. (2013), we found that the secondary task significantly reduced LTM performance, additionally this effect was only present when the second task was presented immediately after the first task.Introducing a 10s delay abolished the LTM deficit, in line with a rapid post-event process during which interference was effective.To our central question, we found that while LTM was impaired by the secondary task, WM was not, as participants learned the platform location at equal rates in all three experimental conditions.These findings support the idea of a rapid consolidation process that is triggered by event offsets.

Participants
155 participants (aged 18 to 24 years) completed the experiment after providing informed consent, and received course credits in exchange for participation.We first recruited participants for the 1-h delay condition before extending the experiment to the 24-h delay group.The study was approved by the local Psychology and Neuroscience Ethics Review Committee.Raw data and analysis scripts are available via Dataverse doi:10.34894/94MIE3.

Procedure
A complete description of tasks and data analysis is provided in supplementary methods, see also supplementary video.In brief, the experiment consisted of two sessions: a learning session and a recall session (Fig. 1A).During the learning session, participants learned the location of the hidden target in three vMWMs (Fig. 1B), with three attempts to learn the location of the target for each vMWM.Each vMWM was distinguishable by four distal cues and the colour of the boundary wall, and was assigned randomly to the three experimental conditions per participant.Within a 40-s inter-trial interval, a secondary spatial task (Fig. 1C) was presented either immediately after the vMWM task ('immediate' condition 0-10 s), after a short delay ('delayed' condition 10-20 s), or no secondary task was presented ('control' condition).The remainder of the inter-trial interval was occupied with a colour-changedetection fixation task, intended to maintain the participants' engagement while limiting possible interference with the vMWM task.Experimental conditions were blocked in 3-trial blocks and condition order was counterbalanced across participants.The trials did not include an indication of which condition the participant was in, nor of when a condition change occurred.Thus, for each vMWM the participant had three opportunities to learn the location of the target.For each opportunity the moment right after discovery, corresponding to an event boundary (ending an 'episode' of vMWM task), was either occupied by the secondary spatial task (in the 'immediate' condition) or by the fixation task (in both the delayed and the control condition).There were then at least 30-s of the fixation task before the next opportunity.
The recall session was performed either 1 or 24 h after the initial learning session.Participants revisited the three vMWMs encountered in the first session and indicated the corresponding target location with 10 virtual pins (Fig. 1D) per maze, presented in a pseudorandom order.All tasks were presented via MATLAB R2022a on a 24″ 16:9 LED LCD screen (iiyama).Participants performed the tasks with an Xbox One controller (Microsoft), using the joysticks to navigate and the trigger buttons to respond to the fixation task or place pins in the recall session.Practice trials for each of the four tasks were performed before the start of the first session.

Working memory is equal across conditions in the vMWM task
We first examined performance in the learning session, calculated as distance travelled from the start point to the target (Fig. 2).When a participant revisits a vMWM after a short period they are typically better at locating the target, as the target location is maintained in WM.Disruptions to WM are indicated by reduced learning over trials (Frielingsdorf et al., 2006;Vorhees & Williams, 2006;Wisman et al., 2008).The majority of our participants showed reliable improvement in performance over three repeated trials per maze.Those 44 participants who did not show performance improvement in any, or in only one maze were classified as 'non-learners' (desaturated lines and symbols, Fig. 2A, see supplementary results for additional analysis of these participants) and were excluded from further analysis, leaving 111 learners for further analysis (1-h retention delay: 53; 24-h: 58, Fig. 2B).For these participants, we tested the effect of the three experimental conditions (immediate, delayed and no secondary task as control) on the performance in the vMWM task using a 2-factor (trial, condition) split-plot (1-h / 24-h groups, taking into consideration that the groups were recruited separately) ANOVA.This analysis revealed an effect of trial number (F (1,987) = 90.5,p< 0.001) but no effect of condition (F(2,987) = 0.34,p = 0.7) and no interaction between trial and condition (F(2,987) = 1.2,p = 0.3).Interestingly, there was also a group effect (F(1,987) = 9.5,p = 0.002) and interaction between group and trial (F(1,987) = 4.2,p = 0.04).However, there was no interaction between group and condition (F(2,987) = 0.33,p = 0.72) indicating that while the 24 h group showed somewhat worse performance, the general trends were the same in both groups.
To better represent differences between conditions, while removing individual and group differences in overall performance, we normalized each participant's performance by their overall mean performance (Fig. 2C).After removing individual and group differences in this way, differences between trials were highly significant (two-factor ANOVA, F (1,987) = 153.0,p< 0.001) but there was no significant condition effect (F(2,993) = 0.87,p = 0.42), or interaction between trial and condition (F (2,993) = 1.55,p = 0.21).To further test for any effect of the secondary spatial task, we calculated differences in normalized performance between trials 1 and 2 (Fig. 2D) and between trials 2 and 3 (Fig. 2E) and tested for condition effects using a 2-factor ANOVA (condition, trial pair).Here, there was a significant difference between trial pairs (F (1,660) = 11.35,p= 0.0008), indicating a greater improvement in performance between trial 1 and 2 (mean = 0.48, standard deviation = 1.2), than between trial 2 and 3 (mean = 0.22, standard deviation = 0.76).However, there was no condition effect (F(2,660) = 1.2,p = 0.3) and no interaction between trial pair and condition (F(2,660) = 0.9,p = 0.41).Since participants must actively maintain the location of the target platform between trials, an improvement in performance between trials has been taken as a measure of WM in the MWM task.That trial-totrial improvement in performance was equivalent between conditions indicates that participants could maintain the platform location in WM equally well with or without the presentation of the secondary spatial task.

Reduced long term memory performance associated with immediate secondary task
We next examined LTM performance.Participants returned to the three mazes for 10 trials per maze and in each trial placed a pin at the remembered target location.LTM performance was quantified as the mean distance between their pins and the true location (Fig. 3A).A splitplot ANOVA showed a significant effect of condition (F(2,327) = 5.1,p = 0.007, see means in Fig. 3B) and a significant effect of group (1-h vs 24-h break, F(1,327) = 5.3,p = 0.022, Fig. 3B asterisk between groups).There was no interaction between group and condition (F(2,327) = 0.68,p = 0.5).For post-hoc tests, we repeated the ANOVA for each pair of conditions separately.There was a significant difference between the immediate and delayed conditions (F(1,218) = 10.12,p= 0.002, Fig. 3B asterisk between conditions) and a trend between the immediate and control conditions (F(1,218) = 3.357,p = 0.07).No significant difference was found between the delayed and control conditions (F(1,218) = 1.747,p = 0.19).An effect of group was found for two comparisons (immediate vs delayed F(1,218) = 5.3,p = 0.02; delayed vs control F (1,218) = 6.1,p = 0.01) but not for immediate vs control (F(1,218) = 2.39, p = 0.12).No comparison showed an interaction (immediate vs delayed F (1,218) = 1.16,p = 0.28; immediate vs control F(1,218) = 0.02,p = 0.89; delayed vs control F(1,218) = 0.94,p = 0.33).To better represent LTM performance, the pin-to-target distances were compared to chance performance, with chance performance estimated by comparing the participant's pin locations with randomly placed alternative target locations.While only a minority of participants performed significantly better than chance individually (Fig. 3C, black points), the group trend was better than chance for all conditions (1 sample t-tests, all p < 0.001 see T-values in figure).Notably, we found the largest proportion of participants with worse than chance performance per group in the immediate condition for both groups (percentages in figure).Differences between conditions were tested with a 2factor split-plot ANOVA as before.The results revealed an effect of condition (F(2,327) = 6.6,p = 0.0016) and of group (F(1,327) = 6.38,p = 0.012,asterisk between groups Fig. 3D), but no interaction (F(2,327) = 0.68,p = 0.51).Post-hoc tests indicated significant differences between the immediate and delayed conditions (F(1,218) = 13,p = 0.0004, asterisk between conditions Fig. 3D) and between immediate and control conditions (F(1,218) = 4.4,p = 0.037, Fig. 3D) but no significant differences between the delayed and control conditions (F(1,218) = 2.2,p = 0.14).Group differences were significant or trending for all comparisons (immediate vs delayed F(1,218) = 6.6,p = 0.01; immediate vs control F (1,218) = 3.3,p = 0.07; delayed vs control F(1,218) = 6.76,p = 0.001).No comparison showed an interaction (immediate vs delayed F(1,218) = 1,p = 0.32; immediate vs control F(1,218) = 0.0002,p = 0.98; delayed vs control F(1,218) = 1.12,p = 0.29).
Taken together, these findings suggest that immediately performing the secondary task after completing the vMWM task resulted in reduced LTM performance relative to both the delayed and control conditions for both a 1-h and 24-h retention delay.In contrast, a 10-s break between the vMWM and the secondary task resulted in no LTM deficit relative to the control condition.Thus, an uninterrupted pause after an event appears necessary for successful LTM retrieval.While overall LTM performance was reduced after 24-h, the length of the retention interval neither reduced nor enhanced the interfering effect of the secondary task.

Discussion
It is currently debated whether event-boundary related retroactive interference indicates interference with an early stage in LTM consolidation (Ben-Yakov et al., 2013;Ben-Yakov & Dudai, 2011;Wu et al., 2022), or interference of WM maintenance (Leszczynski, 2011).Here, we separately tested WM and LTM using a spatial memory task.We also investigated the temporal specificity and extent of the LTM deficit.Results indicated that LTM but not WM was affected by engaging in a secondary task at the event boundary; participants demonstrated similar rates of learning across the three experimental conditions, while recall of the platform location after 1-h or 24-h was significantly worse in the immediate second task condition compared to delayed or control conditions.Strikingly, we found no difference in LTM performance between the delayed-and control conditions.This indicates that those processes that were sensitive to interference by the second task were completed, and no longer sensitive to interference, within the first 10-s after solving the vMWM task.Trial-to-trial learning in the (v)MWM task relies on maintaining the target location in WM (Frielingsdorf et al., 2006;Vorhees & Williams, 2006;Wisman et al., 2008).That learning was not influenced by the secondary task, while LTM performance was, indicates a dissociation between LTM and WM processes for episodic memory formation.
Previous studies of post-boundary processes have relied on single encoding episodes, making it difficult to distinguish whether the presentation of distracting information interferes with learning the target information or with offline consolidation processes.Indeed, both the behavioural and neuroimaging results appear equally consistent with a consolidation or WM account (Leszczynski, 2011).WM maintenance of sensory stimuli has been proposed to facilitate the integration of information into LTM (Hartshorne & Makovski, 2019) and has been linked to continued hippocampal involvement throughout the maintenance period (Dimakopoulos et al., 2022;Fuentemilla et al., 2010;Peters & Reithler, 2022;Poch et al., 2011).Under this interpretation, previous results show the LTM consequences of interference with WM maintenance, rather than interference with memory consolidation per se.Making use of the vMWM in our experiment enabled us to isolate WM and LTM performance.Under the WM maintenance interpretation of boundary processes, we would expect retrograde interference induced by the secondary spatial task to impact both learning rate and recall.Our results were not consistent with this expectation, rather our findings suggest that consolidation, but not WM maintenance, of spatial memories is vulnerable to new information delivered within the first 10-s after the offset of an event.The dissociation between learning and LTM performance in our findings argues that WM and early stages of LTM correspond to distinct and relatively independent memory systems, whereby the WM system is robust to interference while early LTM consolidation is vulnerable.A more nuanced question is the degree to which LTM consolidation and WM maintenance are independent during the immediate post-boundary period.Our secondary spatial task was explicitly designed as a non-mnemonic task, with no requirement for participants to maintain the target locations in either the short or long term.Thus, WM may not have been subject to interference in our experiment, since we ensured that the amount of information that had to be maintained during the inter-trial interval did not exceed WM capacity.The degree of independence between the two systems could be tested by parametrically varying the WM load engaged by a secondary task.
Our findings are in line with previous investigations demonstrating impaired recall performance after post-stimulus interference for narrative memory (Ben-Yakov et al., 2013).We anticipated that the vMWM task would be a good spatial memory equivalent of narrative episodic memory since both rely on similar neuronal mechanisms (Burgess, Maguire, & O'Keefe, 2002;Deuker, Bellmund, Navarro Schröder, & Doeller, 2016).Nevertheless, to our knowledge, this is the first demonstration of boundary-related retroactive memory interference outside of the context of narrative memory.The similarity in our findings points towards the vulnerability of the post-event period as a general feature of episodic memory formation.Whether this vulnerability is specific to stimulus features of the interfering input is unclear.In Ben-Yakov et al. (2013) weaker retroactive interference occurred for the presentation of a scrambled movie clip compared to an intact clip, suggesting some level of specificity.In our study the secondary spatial task was designed to be maximally similar to the vMWM task, with a similar navigation task (but without a memory component), matching movement speed (Goyal et al., 2020) and a same-sized arena (Latuske, Kornienko, Kohler, & Allen, 2018).Such similarity may be analogous to viewing similar movie clips in previous studies, and contrasts with the control condition tasks which share relatively little similarity with the main task.However, whether the retroactive interference effect depends on such similarity, or would be observed also for the presentation of engaging but otherwise dissimilar stimuli, requires further investigation.The present experiment extended on previous findings also by including a 10s delay condition to measure the temporal specificity of the interference effect.We found no effect of presenting the secondary spatial task after 10-s, demonstrating a limit to the length of the period vulnerable to interference.Furthermore, we extended the retention duration over which the LTM deficit was found, from 20-min in the original study, up to 24-h in the current study.LTM formation is a multistage process that continues for many hours after the initial event (Dudai, Karni, & Born, 2015), with sleep being a critical component of long term consolidation (Klinzing, Niethard, & Born, 2019).We found no interaction between the retention delay and the effect of the secondary task.Episodic memories that were affected by boundary interference were neither repaired nor preferentially forgotten over a longer retention interval that included sleep, suggesting that the process influenced by the secondary task is specific to the post-boundary interval, leaving later consolidation processes unaffected.
It is worth considering what elements of our task might represent event boundaries.We have argued that the completion of each vMWM trial represents an important event boundary, since shifting to the secondary navigation task, or the color-change detection task, had a new environment and different task goals, both of which have been related to boundary processes (DuBrow & Davachi, 2013;Kurby & Zacks, 2008;Pettijohn & Radvansky, 2018a, 2018b;Radvansky, 2012;Schapiro et al., 2013;Zacks et al., 2007).An alternative definition of a boundary relies on when the prediction of the nature of upcoming events, based on past events, is no longer valid (Richmond & Zacks, 2017).Under this interpretation, the entire navigation task encompassing three vMWM trials in the same maze, the intervening secondary spatial tasks and the change detection tasks could be perceived as one event; since the general outline of the task stays the same, the prediction of the nature of the upcoming events remains unchanged.Salient boundaries may only occur when entering a new vMWM, or after completion of all three mazes.However, that our analysis revealed an LTM effect of the second spatial task presented at the offset of each vMWM trial, suggests that there was indeed some rapid process, related to LTM and vulnerable to interference, that occurs immediately after finding the platform.It is conceivable that there are several levels of event boundaries present in the experiment, representing coarser and finer chunking of the overall experience (Baldassano et al., 2017;Geerligs et al., 2022;Zacks et al., 2001).Future investigations could probe the interconnection between local and largerscale boundaries, and their individual contribution to LTM processes.
In conclusion, we present evidence for the critical importance of the moment of event offsets in LTM processes.Our findings refute the account that the relationship between event offsets and LTM is mediated by WM maintenance, since we found no effect of presenting a secondary task on WM performance.Additionally, we have extended on previous findings by showing that boundary processes are restricted to the first 10-s after an event offset, by showing that the effect persists over as much as 24-h after the initial encoding event, and by extending the field to include spatial memory in addition to narrative memory.
Supplementary methods, results and the supplementary video to this article can be found online at https://doi.org/10.1016/j.cognition.20 24.105859.

Fig. 1 .
Fig. 1.A) Schematic of experimental procedure.Top: learning session.Purple, orange, green indicate maze configurations associated with each experimental condition, defined by their trial structure shown below.Middle: blue, red, yellow indicate tasks performed in each trial.Bottom: recall session colours indicate pseudorandom presentation of mazes corresponding to each experimental condition.B -D) Participants' point of view (POV, left), and top-down maps (right) of the three spatial navigation tasks.B) vMWM, in the POV with a green border wall and distal cues consisting of line icons of scenes.Top-down map shows the path taken to the target platform (grey disk) over three trials.C) The secondary spatial task with (more distant) red and (closer) green columns as non-targets/targets in the POV.The score indicates 2 successive green hits.The top-down map shows the path taken on one trial including red/green locations.The locations change after each touch, numbers indicate each location pair.Here the participant successfully touched green targets 1 and 2 but failed to avoid red target 3.The trial ended before they reached target 4. D) The pin-drop task including the pin in the POV.The top-down map shows the location of 10 pin-drops (stars) relative to the unseen target location (grey disk).The fixation task (not shown) display only contained a central colour-changing dot.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3 .
Fig. 3. A) Violin plots (same format as Fig. 2) of recall performance (mean error distance) per condition for each group.Larger values indicate poorer performance.B) Overall mean performance per condition (colours) and per group (black/grey).Brackets indicate significant differences (* p < 0.05).C) Violin plots show recall performance relative to chance per condition for each group.Negative values indicate better-than-chance performance.Participants with significantly better-thanchance performance (bootstrapping) are marked in black, other participants in white.Numbers at the horizontal dotted line indicate the percentage of participants with worse than, or better than chance performance (respectively above and below the 0 line).Numbers below each distribution indicate 1 sample t-tests, comparing the distribution to 0. All P-values approached 0. D) Overall mean performance relative to chance per condition (colours) and per group (black/grey).Brackets indicate significant differences (* p < 0.05, ** p < 0.01).