Probing the neural dynamics of mnemonic representations in humans

Memories are not stored as static engrams, but as dynamic representations affected by processes occurring after initial encoding or even consolidation. How the modulation of memory traces after their formation is reflected in the neural activity during subsequent retrieval is currently not well understood. Using fMRI in 27 healthy human participants, we probed how neural representations of associative memories are dynamically modulated by two behavioral techniques that can either strengthen or weaken memories after encoding. Behaviorally, we demonstrated that, after an initial delay of 24 hours, associative memories can still be strengthened or weakened by repeated retrieval or suppression, respectively. Neurally, we show that repeated retrieval dynamically reduced activity amplitude in ventral visual cortex and hippocampus, but enhanced the distinctiveness of activity patterns in the ventral visual cortex. Critically, a larger reduction of activity amplitude in the ventral visual cortex associated with larger enhancement of distinctiveness of activity patterns in the same region. In contrast, repeated memory suppression was associated with reduced lateral prefrontal activity, but relative intact activity patterns. These results reveal dynamic adaptations of mnemonic representations in the human brains and how retrieval-related activity amplitude and distinctiveness of activation patterns change as a function of strengthening or weakening.

We aimed to assess differences in activity amplitude and activity patterns that are associated with the 2 4 9 reported behavioral effects of repeated retrieval and memory suppression. For the analysis of activity 2 5 0 amplitude, we used the standard whole-brain univariate General Linear Model (GLM) analysis,  For the analysis of activity patterns, we used a multivariate activity pattern-based neural index to measure 2 5 4 the fidelity of pattern reactivation during retrieval. We assumed that the ventral visual cortex (VVC) and 2 5 5 the hippocampus would demonstrate neural reactivation of picture-specific activity patterns during 2 5 6 retrieval. First, using the data from the familiarization task, we identified areas ( Figure 3A) within the 2 5 7 VVC that are sensitive to picture-specific information during perception and demonstrated the neural were regarded as regions-of-interest (ROIs). Second, trail-by-trail voxel-wise activity patterns of ROIs 2 6 1 during modulation and the final memory test were extracted. To quantify distinctiveness of retrieved 2 6 2 "mental images", we calculated the activation pattern variability using Pearson correlation ( Figure 3C).

6 3
We hypothesized that high activity pattern variability (low pattern similarity measured by R-values) 2 6 4 reflects neural reactivations of distinctive "mental images". Here, we first report the effects of repeated 2 6 5 retrieval on activity amplitude and activity patterns, and then the counterparts related to suppression. within the ventral visual cortex (VVC). We identified voxels whose activation patterns can be used to differentiate pictures that 2 6 9 were processed during the familiarization phase and were reactivated during successful memory retrieval during the final test images" were retrieved based on highly similar memory cues (different locations within maps were cued). We derived activation 2 7 2 patterns for each memory retrieval trials based on fMRI data, and then quantify the pattern variability across trials using Person's 2 7 3 r. Lower the similarity measure (r-value), higher the pattern variability.   Figure 4A; Figure S5; Table S1).

8 1
Most of these regions are located within the anatomically-defined VVC. The whole-brain analysis did not 2 8 2 show an effect of retrieval on the activity amplitude of hippocampal voxels under the same threshold. However, ROI-based analysis of hippocampal signal found reduced activity when retrieving RETRIEVAL 2 8 4 ASSOCIATIONS compared to CONTROL ASSOCIATIONS (left hippocampus: t=-2.43, p=0.022; Figure   2 8 5 4E; right hippocampus: t=-2.18, p=0.038; Figure 4G). Next, we confirmed that the observed activity reduction in VVC and bilateral hippocampus is related to a 2 8 7 linear decrease in activity with repeated retrieval using the data from the modulation phase. Specifically, 2 8 8 we extracted the beta coefficient from the VVC cluster defined by the RETRIEVAL vs CONTROL contrast 2 8 9 (as shown in Figure 3A) and the hippocampus for each run of the modulation phase and tested for the 2 9 0 change in activity amplitude across runs. We found reduced VVC over repeated retrieval attempts (F [4, We next examined whether the reduced activity amplitude was associated with reduced or enhanced 2 9 7 distinctiveness of activity patterns during the final memory test. Focusing on the identified VVC areas and 2 9 8 hippocampus ( Figure 3A and 3B), we calculated the trial-by-trial activity pattern variability for 2 9 9 RETRIEVAL ASSOCIATIONS and CONTROL ASSOCIATIONS separately. Results show that retrieval- CONTROL ASSOCIATIONS (t=2.3, df=26, p=0.029; Figure 4C). However, we did not observe a similar To characterize the dynamic modulation of activity pattern variability in the VVC, we further applied the 3 0 9 same variability analysis to each run of the modulation phase and analyzed these pattern variability values 3 1 0 using a 2×5 ANOVA (modulation; run; Figure 4D). We saw a significant main effect of the run, reflecting been presented ten times during the modulation, repeated retrieval more effectively enhanced pattern 3 1 8 distinctiveness compared to suppression. We did not perform the same dynamic analysis to the 3 1 9 hippocampal activity patterns because no effect was found in the final memory test. cluster showed decreased activity amplitude over repetitions of retrieval during the modulation phase. (C) Higher activation  Our activity amplitude and activity pattern variability analysis independently demonstrate that repeated 3 4 1 retrieval dynamically reduced activity amplitude, but enhances the distinctiveness of activity pattern in the 3 4 2 VVC. However, we performed these analyses at the whole-brain level and ROI level separately. To showed reduced activity amplitude during the final test and, at the same time carries picture-specific 3 4 6 information during perception and retrieval. Similar to our main analysis (as shown in Figure 4A and 4B), we found reduced activity amplitude (t=5.823, df=26, p<0.001; Figure 5B), but increased distinctiveness First, our behavioral results revealed that associative memories could still be strengthened or suppressed 4 0 0 by repeated retrieval or suppression separately after initial consolidation. The beneficial effect of retrieval and shorter reaction times during memory retrieval, suggesting strengthened associative memories after 4 0 6 this modulation. Corroborating the behavioral effect for the final memory test, we also found that repeat  For RETRIEVAL ASSOCIATIONS, fMRI revealed a dynamic process based on decreased retrieval-related how expectations shape brain responses. Expected stimuli reduce overall activity amplitude, a between changes in activity amplitude and distinctiveness of activity patterns across participants when we 4 3 7 restricted our analyses to the same cluster of visual processing voxels. One may argue that the observed 4 3 8 neural changes may just be associated with repeated visual processing of memory cues, but not 4 3 9 specifically with repeated retrieval. Our fMRI results from the modulation phase challenge this argument.

0
We did not only find reduced activity amplitude and more distinct activity patterns during the final 4 4 1 memory test, but also demonstrated that these neural changes gradually emerged across repeated retrieval 4 4 2 during the modulation phase. A similar effect was not found during repeated suppression, even though 4 4 3 memory cues were also presented repeatedly. Therefore, the observed neural changes cannot be simply  For SUPPRESSION ASSOCIATIONS, we observed lower LPFC activity amplitude, but relatively intact insufficient cognitive control resources to facilitate retrieval. It is noteworthy that our pattern analysis was between still-remembered associations and forgotten associations.

7 8
Taken together, our results provide behavioral and neural evidence for dynamically adapting 4 7 9 representations of episodic memories separately modulated by repeated retrieval and suppression. We 4 8 0 found that repeated retrieval reduced activity amplitude, but increased the distinctiveness of activation 4 8 1 patterns in visual areas. We propose that lower overall activity amplitude, but higher distinctiveness of 4 8 2 cortical reinstatement is evidence for more distinct neural representations of associative memories Thirty-two right-handed, healthy young participants aged 18-35 years were recruited from the Radboud excluded from further analyses due to memory performance lower than the chance level, three participants 5 0 3 were excluded from all of the analyses because of excessive head motion during scanning, and.

0 4
Neuroimaging data of one participant was partly used: he/she was excluded from the analysis of the final memory test ( Figure 1E). Among these phases, the familiarization, modulation, and final memory 5 4 1 test phase was performed in the scanner, while study phase and two typing tests were performed in the 5 4 2 behavioral lab. Familiarization phase 5 4 4 The first task in the scanner for our participants was the familiarization phase. This phase was 5 4 5 conducted to obtain the picture-specific brain responses to all of the 48 pictures, measured by the Blood-

4 6
Oxygen-Level Dependent (BOLD) activity patterns. The second purpose of the task is to let participants presentation was pseudorandom and pre-generated by self-programmed Python code. The dependence between the orders of different runs was minimized to prevent potential sequence-based memory encoding.

1
To keep participants focused during the task, we instructed them to categorize the presented picture via the 5 5 2 multiple-choice question with four options (animal, human, object, and location). We used an exponential 5 5 3 inter-trial intervals (ITI) model (mean=2s, minimum=1s, maximum=4s) to generate the ITIs between trials.

4
Participants' responses were recorded by an MRI-compatible response box.

5
Study phase 5 5 6 Each picture-location association was presented twice in two separate runs. During each study trial, 5 5 7 the entire map was first presented for 2.5s, then one of the 48 locations was highlighted with a BLUE 5 5 8 frame, for 3s, and finally, the picture and its associated location were presented for 6s. Participants were 5 5 9 encouraged to use both the relative position of the memory cues on the maps and the appearance of the 5 6 0 highlighted areas to facilitate association learning. We pre-generated a pseudorandom order of the trials to 5 6 1 minimize the similarity between the order used in familiarization and the order used in the study phase.

6 2
Typing test phase Immediately after the study phase, participants performed a typing test (day1) assessing picture- location association learning. Each location was presented again (4s) in an order which differs from the 5 6 5 study phase, and participants had maximally 60s to type on the standard keyboard to describe the 5 6 6 associated picture. Twenty-four hours later (day2), participants performed the typing test again at the same 5 6 7 2 7 behavioral lab. The procedure is identical to the immediate typing test, but with a different order of the 5 6 8 trials.

6 9
Modulation phase The modulation phase is the first task participants performed on the day2 MRI session. We used the 5 7 1 think/no-think (TNT) paradigm with the trial-by-trial self-report measures to modulate established 5 7 2 associative memories. The same paradigm has been used in previous neuroimaging studies, and the self- Anderson, 2012). Forty-eight picture-location associations were divided into three conditions. One-third 5 7 5 of the associations (16 associations) were assigned to retrieval ("Think") condition, one-third of the 5 7 6 associations were assigned to suppression ("No-Think") condition, and the remaining one-third of the 5 7 7 associations were assigned to control condition. The assignment process was counterbalanced between 5 7 8 participants. Therefore, at the group level, for each picture-location association, the possibility of 5 7 9 belonging to one of the three conditions of modulation is around 33%. Associations that belong to with the GREEN frame for 3s, and participants were instructed to recall the associated picture quickly and locations were highlighted with the RED frame for 3s, and our instruction for participants is to prevent the 5 8 5 potential memory retrieval and try to keep an empty mind. We also told the participants that they should cues. After each retrieval or suppression trial, participants have maximum 3s to report their experience (Never, Sometimes, Often, and Always) by pressing the button on the response box to indicate whether the 5 9 0 associated picture entered their mind during that particular trial. The modulation phase consisted of in total of five functional runs (64 trials per run). In each run, 32 5 9 2 locations (half retrieval trials, and half suppression trials) were presented twice. Therefore, for each 5 9 3 memory cue that not belongs to the control condition, it was presented ten times during the entire 5 9 4 modulation phase. Again, we pre-generated the orders of the presentation to prevent the similar order 5 9 5 sequences across five modulation runs. Between each trial, fixation was presented for 1-4s (mean=2s, 5 9 6 exponential model) as the ITI.

9 7
Final test phase 5 9 8 After the modulation phase, participants performed the final memory test within the scanner. All 48 5 9 9 locations (including both the retrieval/suppression associations as well as control associations) were category of the picture you were recalling?" They also had four options to choose from (Animal, Human, 6 0 8 Object, and Location). In this study, we did not focus on the accuracy of the category judgement because categorization could 6 1 2 be a subjective process. We mainly used the responses from participants to control for the effect of