Recruitment of a Long-term-memory Supporting System During Repeated Maintenance of an Abstract Visual Image in Working Memory: An EEG Study


 Humans can flexibly transfer information between different memory systems. Information in visual working memory (VWM) can for instance be stored in long-term memory (LTM) when a scene is repeatedly memorized. Conversely, information can be retrieved from LTM and temporarily held in WM when needed. It has previously been suggested that a neural transition from parietal- to midfrontal activity reflects transfer of information from WM to LTM during repeated visual search for the same single item. However, whether these observed neural changes truly represent consolidation and are also observed when memorizing and (explicitly) recollecting a rich visual scene (rather than a single target), remains unclear. The current EEG study investigates whether such a gradual shift in memory-correlates is also observed when an abstract colour-array is repeatedly memorized and explicitly recollected in a WM paradigm. Importantly, we tested the functional significance of a neural shift for longer-term consolidation in a subsequent recognition task. Our results provide supporting evidence for recruitment of an LTM-supporting storage system during WM, which facilitates visual WM maintenance and is indexed by (sustained) modulation of a midfrontal component. Enhanced explicit memory recollection during WM is associated with- and possibly facilitated by an emerging late contralateral parietal activity.


Introduction 34 35
When an artist paints a landscape from life, at first the chosen scene will need to be inspected 36 regularly, but gradually a detailed picture may develop in the painter's mind. How is the image 37 represented in the brain and how does this representation change, when the artist becomes more 38 familiar with the depicted landscape? The current EEG study investigated whether a shift in 39 mnemonic representation can be observed during learning of an abstract visual scene. Based on 40 an earlier study, which reported a transition in memory-sustaining neural representations when 41 subjects became gradually more efficient in spotting a visual search target 1-3 , we specifically 42 looked at parietal-and midfrontal components. 43 The ability to hold representations of several visual items online for direct access and 44 manipulation by higher cognitive functions, is known as visual working memory (WM). There 45 is a capacity limit to the number of items that can be contained in visual WM, which usually 46 sits around 4, but differs across individuals 4 . While traditionally visual WM is thought to rely 47 on a frontal-parietal network, recent evidence suggests that it can recruit MTL/hippocampal 48 structures as well, perhaps particularly when the number of memorized items exceeds the visual 49 WM capacity limit and/or when spatial location is an important dimension of the memory trace 50 (see 5 for a review). A proven EEG method to measure several aspects of visual WM, such as 51 capacity and filtering efficiency, is the contralateral delay activity (CDA). This is a sustained 52 negativity around parietal-occipital electrodes which scales with the number of memorized 53 items 4,6 (and see 7 for a review) and can also be observed while a search template is held online 54 during visual search 8 . 55 In the previously mentioned study 1 , it was shown that the CDA amplitude gradually decreases 56 in amplitude during repeated and increasingly efficient search for the same target in different 57 search contexts 2 . The decrease in CDA amplitude concurred with a negative modulation of a 58 midfrontal P170 component, which was interpreted as a transfer of information between a WM-59 to a long-term memory (LTM) store. Whether this shift in neural correlates is really linked to 60 an enhanced consolidation of the visual memory trace can however not be concluded from this 61 study, as recognition was not tested afterwards. In addition, a more efficient search performance 62 is likely to reflect (perhaps implicit) short-term strengthening of stimulus-response connections 63 (see 9 ), but not necessarily an enhanced explicit memory representation. 64 Negative modulation of the midfrontal P170 component has indeed been associated with long-65 term (implicit) memory processes such as perceptual priming 10 , involving perceptual 66 categorization and grouping processes 11,12 . Many other studies have implicated a later (positive) 67 midfrontal component in conscious recognition 13,14 . In these studies, a dissociation emerges 68 between recognition based on familiarity (knowing), signaled by the midfrontal component, 69 versus rich detailed recollection, reflected in a later parietal activity [30][31][32][33][34][35][36][37][38] . 70 The scalp location of the midfrontal signal (around the Fz electrode) may link this activity with 71 midfrontal theta power (4-8Hz), originating in the medial prefrontal cortex (mPFC). MPFC 72 theta power modulations have been associated with encoding and retrieval processes in LTM 73 and maintenance during WM 15 , possibly through functional connectivity with medial temporal 74 lobe (MTL) and hippocampal structures (and see 16 for a review). 75 In our paradigm, we aimed to investigate whether a parietal-to-midfrontal shift in neural 76 correlates can be observed when the repeatedly memorized stimulus is not a single search target, 77 but a complex colour-array, which needs to be recollected explicitly during the task. In our 78 design, the array consisted of more coloured items than can generally be held in working 79 memory, of which one was probed by spatial location during each trial. The array was therefore 80 more akin to an abstract visual scene, which had to be remembered as a whole, rather than a 81 collection of separate targets. However, we expected that initially, participants would encode 82 the coloured items as separate targets in visual working memory, with a gradual transition to a 83 coherent representation of the whole array, which would rely more on LTM sustaining 84 structures. 85 As in the visual search study 1 , we used the CDA component to trace the working memory 86 representation, while the modulation of the midfrontal P170 component was taken as a potential 87 neural signature of memory consolidation within an LTM-supporting system. Bilateral arrays 88 of 6 coloured squares each, were repeatedly presented during a testing block of 8 trials, with 89 one hemifield being cued for covert attention at the beginning of the testing block. In each trial, 90 the colour of one of the squares in the array was probed following a 2.5 second delay, with a 91 combined central colour-and a peripheral spatial-cue highlighting one of the array-squares 92 ('memory cue'). This way, participants had to construct an explicit mental representation of the 93 colour-array's spatial lay-out during each trial in order to perform the task. Importantly, the 94 item-set-size exceeded the visual WM capacity limit, which we expected to require stronger 95 recruitment of LTM (MTL/hippocampal) structures (see 5 ). 96 During the main experiment, parietal contralateral activity-and midfrontal-responses were 97 measured during the early learning-(trial 1-4) and late learning phase (trial 5-8) of the WM 98 blocks. These components were measured from array onset throughout the delay, relative to the 99 ERP in a control attention (ATT) condition, during which the cue was presented simultaneously 100 with the colour-array. The same components were also measured during recollection of the 101 memorized array after the delay, probed by the memory cue in the WM condition. With respect 102 to the parietal component, we refer to contralateral activity (CA) while the array or memory-103 cue is present on screen and contralateral delay activity (CDA) during the WM delay following 104 array-offset. 105 Memory consolidation was assessed in a subsequent post-hoc old/novel recognition task. The 106 midfrontal P170 response to old-and novel colour-arrays was measured, to assess lasting neural 107 modulation accompanying the longer-term perceptual memory consolidation 10 . 108 During the main experiment, we expected a gradual decline of a WM-delay-specific CDA 109 component (and no contralateral activity (CA) present in the ATT condition) when learning 110 progressed. Conversely, we expected an increasingly enhanced negative modulation of the 111 midfrontal component during the learning process 2,3 . We further predicted a more prominent 112 parietal component during gradually more successful (explicit) recollection of the colour-array 113 following the cue 30-38 . Furthermore, we hypothesized that, if the midfrontal response 114 modulation signaled transfer of the memorandum from a working-to another longer-term 115 memory store during the main experiment, this modulation should be predictive of accuracy in 116 a post-hoc recognition task. Finally, longer-term perceptual memory during the recognition task 117 should be visible as a lasting neural modulation of the midfrontal P170 response upon re-118 exposure to old colour-arrays 14 For the main experiment, all participants were included in the behavioural analysis and one 131 participant had to be excluded from EEG analysis, due to excessive artefacts (see for criteria 132 below). For the follow-up recognition task, the first 6 participants had to be excluded from 133 analysis due to a technical error. Both behavioural-and ERP analysis for this part of the 134 experiment were performed on the remaining 14 participants. 135 136

Behavioural Task 137
The experiment consisted out of 2 parts, the main experiment and a subsequent recognition 138 task. A schematic representation of the experimental paradigm during the main experiment 139 is shown in Fig. 1. Task (Working Memory (WM)/Attention (ATT)) was manipulated within 140 participants across 8-trial blocks. At the beginning of each testing block, a central arrow 141 indicated which hemifield to attend to during the subsequent 8 trials and would remain visible 142 throughout the duration of the testing block. After 750 ms following spatial cue onset, 2 143 different arrays, each consisting of 6 coloured squares at fixed locations, were bilaterally 144 presented (750 ms), followed by a scrambled colour mask (250ms). This was an initial 145 presentation of the colour-array-combination that was subsequently repeated in the ensuing 146 8 trials. A trial would start with a slight brightening of the central fixation square (750ms). 147 During each trial, participants had to match a central colour probe with the colour at one 148 square in the array, highlighted by a black rectangle around it. This combined cue (central 149 colour and peripheral rectangle) was presented for 750ms, either while the array was still 150 present on screen (ATT block) or following a 2.5 second delay (WM block). Locations of all 151 squares in the bilateral arrays and at central fixation remained visible as white placeholders 152 at all time when the colour-array was not on screen. ITI intervals were 2.5 seconds for both 153 conditions. Participants were encouraged to maintain a visual memory of the cued colour-154 array during the delay periods of the WM condition. 155 Participants responded with their right hand by using self-chosen keys on a keyboard for 156 "match" or "no-match". The central colour cue had to be the same as the colour at the 157 highlighted location in the array to be deemed a match (50% of trials) and was considered a 158 non-match when the colour was not present in the array at all (25% of trials) or present but not 159 at the highlighted location (25% of trials). The task was non-speeded and participants were 160 instructed to respond as accurately as possible. 161 Task (ATT or WM) and attended hemifield were randomized across testing blocks. Three The size of the screen was 50.8 × 33.9 cm. 172 The 6 different colours allocated to each square of an array were randomly drawn (separately 173 for each hemifield) from a pool of 8 highly discriminable but equi-luminant colours (red, gray, 174 brown, green, purple, turquoise, orange and pink (24,5 ± 2,5 cd/m 2 ). Arrays were presented on 175 a gray background (72 cd/m 2 ). Squares in the arrays subtended 0.75⁰ in visual angle each with 176 2 squares presented in 3 rows, with 1⁰ vertical distance between squares (center to center) and 177 3⁰ horizontal distance between squares in the upper and the lower row and 1⁰ horizontal distance 178 between squares in the middle row. Centre of the arrays were presented 4.5⁰ from fixation. 179 Preceding onset of the EEG experiment, participants completed 2 practice blocks of each task 180 condition. During the EEG experiment, lights were switched off, while participants were seated 181 in a comfortable chair, 66 cm from the monitor with their heads resting on a desk-mounted 182 chinrest. They were instructed to maintain fixation on a central square in the middle of the 183 monitor array, while covertly attending the cued hemifield throughout each testing block. 184 Adhered fixation was monitored with an Eyelink1000 eyetracker (SR Research). Participants 185 were instructed to refrain from blinking as much as possible until the 5 second breaks in 186 between testing-blocks. 187 Following a ca. 5 minute break after completion of the main experiment, during which 188 participants remained seated and attached to the EEG apparatus, the post-hoc recognition task 189 would commence. Participants were not informed about this task until after they had completed 190 the main experiment. For this task, all colour-arrays that had been presented during the main 191 experiment in the attended hemifield (either during ATT-or WM blocks) were randomly 192 intermixed with novel array configurations made up out of the same colour pool. In total there 193 were 72 trials, of which 36 contained novel-and 36 contained old arrays (18 presented during 194 either condition (ATT/WM)). Note that due to randomization of all trials, there was on average 195 an interval of ca. 35 minutes between initial array-exposure during the main experiment and the 196 test probe during the post-hoc recognition task. A trial would commence with a slight 197 brightening of the central fixation square (500 ms), followed by a central presentation of a 198 colour-array (2000 ms), with an equiprobable chance to be drawn from the 'old'-or 'novel' set. 199 Participants were asked to indicate whether they recognized the presented colour-array and 200 instructed to respond 'yes' when they felt more than 60% confident. Responses had to be 201 indicated with their right hand by pressing either of two self-chosen keys on the keyboard. 202 Participants were instructed to withhold their responses until offset of the colour-array and to 203 be as accurate, not as fast as possible. The following trial started after a 3.5 second inter-trial-204 interval, during which a white square would remain visible at fixation. During EEG recording, electrode impedances were kept below 20 kΩ. Each electrode was 245 measured on-line with respect to the CMS/DRL electrodes producing a monopolar channel. 246 The ongoing brain activity at each electrode site was sampled using the Actiview application at 247 2048 Hz and filtered with a low-pass filter of 100 Hz, and a high-pass filter of 0.16 Hz. Data quality was initially inspected using automated algorithms. Epochs containing excessive 263 noise or drift (±80 μV) at any electrode were marked and channels in which more than 20% of 264 the epochs were affected were excluded. One participant was excluded, of whom more than 265 20% of the channels was deemed bad according to these criteria. Blinks were identified as large 266 deflections (± 4 x SD) in the VEOG signal (subtraction of the signal from the EOG electrodes 267 above and below the left eye) and were marked. All marked artefacts were later accounted for 268 during robust averaging. Epochs were averaged according to task (ATT/WM), trial group (first 269 four trials/second four trials) and attended visual hemifield. an averaging procedure that preserves the electrode location relative to the attended hemifield 280 (contra-or ipsilateral) 4,6,20 . Previous studies investigating CDA generally studied relatively 281 short delay intervals (ca 300-1000ms following stimulus onset, see 1,2,4,6 ). We therefore limited 282 our analysis for these previously reported time-windows. Since stimulus presentation was 283 longer in our paradigm (compared to previous studies) only the first 300ms post colour array-284 offset was included (reflecting true delay-activity) into our first analysis. Mean amplitudes of 285 the difference waves between contra-and ipsilateral delay activity were computed during this 286 800-1100ms time-window. 287

288
The mid-frontal component was computed by averaging ERPs across 4 mid-frontal electrodes 289 (aFz, Fz, F1 and F2) and mean amplitudes were calculated during the peak response following 290 array onset (P170, see 2,10 ) and during the 300ms post array-offset. 291 292 During memory-cue presentation, parietal CA was measured from the moment it emerged (350 293 ms following cue onset 4,6,7 ) until 100 ms pre cue-offset. The P170 component was measured 294 during the peak 170ms following cue-onset. 295 296 Two-way repeated measures ANOVA were conducted for the mean amplitudes, testing the 297 effects of Task and Trial-group. To test whether individual differences in measured signals 298 predicted WM and/or recognition task performance, Robust Pearson correlations were 299 computed between mean amplitude and task performance (d'). We performed this correlation 300 analysis not only averaged across the 300ms time-window, but specifically also for a longer 301 time into the WM delay (1000ms following array-offset). 302 The percentage bend correlation is a robust method that protects against outliers among the The analysis of the post-hoc recognition task aimed to test longer-term perceptual priming and 314 measured a potential modulation of the midfrontal P170 response to repeated (old) arrays 315 compared to non-repeated (novel) arrays 10 . 316 317

Results 318
To test whether a learning-associated shift in visual WM correlates could be observed during 319 repeated encoding and recollection of a multi-item colour-array, we conducted an EEG 320 experiment using a visual WM paradigm as depicted in Fig. 1 (and see Methods). This was 321 followed by an array-recognition task to test whether observed modulation of neural correlates 322 during the WM task facilitated longer-term consolidation. In a previous pilot experiment we had shown that gradual learning of the colour-array 328 configuration reached a plateau performance after ca. 8 trials (see Supplementary Fig. S1b). 329 Moreover, the array-recognition task in this pilot experiment confirmed longer-term memory 330 consolidation, in particular for colour-arrays presented during the WM-as compared to the ATT 331 condition (see Supplementary Fig. S1c). In a second pilot experiment, we tested the hypothesis 332 that a set size exceeding the visual WM capacity-limit (6) would enhance the reliance on a 333 longer-term memory-supporting storage system for holding a mental representation online 334 during visual WM, which should be reflected in a stronger memory consolidation. We found 335 indeed support for this hypothesis, as 6-item arrays were better remembered in a post-hoc 336 recognition task than 4-item arrays (see Supplementary Fig. S2). Two different arrays, consisting out of six coloured spatially-fixed squares each, are bilaterally presented with one hemifield cued for covert attention at the beginning of (and throughout) a testing block. The same array-combination was repeated for 8 trials in one testing block and spatial locations of all squares would remain visible as white placeholders during stimulus intervals. Participants had to match a central colour cue with that at one highlighted square in the array, either while the colour-array was still present on screen (attention/ATT condition, left) or following a 2.5 second delay (WM condition, right).

Behavioural performance during the EEG experiment (WM task) 340 341
Guided by the pilot findings, 6-item colour-arrays were used in the EEG experiment, which 342 were repeated during 8-trial blocks. A similar gradual increase in accuracy-scores over trials 343 was observed in the EEG experiment, as had been shown for 6-item arrays in the second pilot 344 experiment (see Fig. 2A and Supplementary Fig. S2). For statistical analysis of task 345 performance, accuracy scores (d' ) were collapsed across the first-and the second four trials 346 separately and an ANOVA was applied with Trial-group (first four trials/second four trials) and 347 Task (ATT/WM) as factors. As shown in Fig. 2b, the gradual increase in performance during 348 the WM condition was reflected in a significant difference between trial-groups, while no 349

Behavioural performance during the EEG experiment (recognition task) 367 368
The main task was followed by a post-hoc recognition task, during which all colour-arrays that 369 had previously been presented in the attended hemifield during the main task (ATT-or WM 370 condition) were shown again, randomly intermixed with novel array configurations, made up 371 out of the same colour pool. Participants had to indicate whether they recognized the presented 372 arrays with a confidence higher than 60%. 373 374 As observed in the behavioural pilot experiments preceding the EEG study, the repeated 375 exposure to the colour-arrays resulted in a positive d' score during the post-hoc recognition task 376 as plotted in Fig. 3a (mean d'-score=0,32; T(13)=6,73 p<.001). Unlike during the first pilot 377 experiment (see Supplementary Fig. S1c), no significant difference was detected between d' 378 recognition scores for arrays shown during the ATT-versus the WM condition ( Fig. 3b; 379 pairwise comparison T(13)=-.53 p=.6 ns.). It should be noted however, that during the earlier 380 behavioural pilot, experimental conditions were slightly different and had more trials per block 381 (see discussion). However, also with 8 trials per test-block in the EEG session, we observe a 382 modest but significant above chance d' score across conditions in the recognition task, 383 indicating longer-term consolidation across conditions. 384 385 386

Fig. 3 Recognition task accuracy scores.
Accuracy scores on the post-hoc recognition task are plotted as d' scores across conditions (a) and separated for colour-arrays that had previously been shown during the ATT or the WM condition (b).

Parietal contralateral activity 389
First, the parietal contralateral response was measured from colour-array onset (Contralateral 390 Activity, CA) and throughout the memory delay or ITI (Contralateral Delay Activity, CDA) 391 per task condition (ATT/WM) and trial-group (first four trials/second four trials). This response 392 was plotted as difference waves between contra-and ipsilateral activity across occipital-parietal 393 electrodes (see Fig. 4a-b). As can be seen in Fig. 4a, a C(D)A component was visible for the 394 WM condition, starting around 300 ms post array-onset and persisting during a short time into 395 the delay period during early learning (first four trials). 396 As a CDA component was only detectable for a relatively short time during the delay, even 397 during the early learning trials, we calculated mean amplitudes during a selected time-window 398 of 300 ms following array offset (plotted in Fig. 4c; see also methods) and compared these 399 between both trial groups to test our prediction of a learning-associated decline in the CDA. An In summary, we do observe a learning-related decline in the WM-specific parietal C(D)A 411 response following array onset. However, even in the early learning phase (first four trials), this 412 component is only visible (averaged across participants) for a relatively brief time into the WM 413 delay period. Note that the signal was corrected for saccades and blinks up till 1 second post 414 array-offset, which may account for a noisier signal later during the delay. 415 To test whether the CDA may have become less prominent due to variable responses between 416 participants, but still sustains WM maintenance for a longer time into the delay, we performed 417 a second analysis in which we correlated responses with task performance per participant. CDA 418 mean amplitudes were averaged across a longer time into the WM delay (1 second) and 419 correlated with task accuracy (d') during the WM task. To test whether the CDA response 420 during the delay had any relevance for longer term memory consolidation we performed an 421 additional correlation analysis with task accuracy (d') during the subsequent recognition task. 422 No significant correlation was found between the CDA mean amplitudes and WM (d') 423 performance (see Fig. 4 h-i; first four trials: R=0.10; T=0.41; p=0.68 ns.; second four trials: 424 R=0.11 T=0.47 p=0,64 ns.). In contrast, a significant correlation was observed between the 425 CDA mean amplitudes during the first 4 trials of the WM task and subsequent performance (d') 426 during the later recognition task (see Fig. 4l; R=-0,57 T=-2,31 p=0.04), which was not observed 427 for the mean CDA amplitude during the second 4trials ( Fig. 4m; R=0,11 T=0,36 p=0.72 ns; 428 Williams's test between correlations T(12)=2,18 p=.05). Also for the short time-window post 429 array-offset (800-1100ms), no significant CDA-behaviour correlations were observed with 430 WM accuracy (p>.6 ns.), but a similar significant correlation was observed with recognition 431 accuracy for the CDA amplitude during the first 4 trials (R=-0.65; T=-2.87; p=0.015). No 432 significant activity-behaviour correlations were observed for the ATT condition (p>.35 ns.). 433 The CDA response therefore does not seem to reflect visual WM maintenance throughout the 434 delay, but may reflect a non-sufficient contribution to memory-maintenance, ie. keeping only a 435 subset of the coloured squares in mind, during early learning trials. Furthermore, our results 436 suggest that the C(D)A also supports a (longer-term) encoding mechanism. 437 438

Midfrontal response 439 440
We next tested whether learning of the colour-arrays in the WM condition concurred with a 441 negative modulation of the midfrontal component, which is associated with long-term 442 perceptual priming 1 . The midfrontal component was averaged across four electrodes (AFz, Fz, 443 F2 and F3) per task-condition (averaged across hemifield) and per trial-group, and plotted from 444 array-onset throughout the WM delay (see Fig. 4d), as well as for the first 350 ms following 445 array-onset separately to visualize the P170 peak (see Fig. 4e). Mean amplitudes were 446 calculated for all four conditions for the P170 peak (see Fig. 4f) and, as for the CDA analysis, 447 also for the 300 ms time-window post array-offset (see Fig. 4g).  Fig. 3g, but note that direct comparison between the task 457 conditions is problematic, since the ATT condition required a button-press immediately 458 following array-offset (this confound is canceled out in the difference potential of the CDA, but 459 not for the MF analysis). The effect of interest here is the pairwise comparison between the 460 first-and second 4 trials of the WM condition, which did not reach significance (T(18)=1,27 461 p=.22), possibly due to high inter-subject variability in the signal. 462 We subsequently performed a robust correlation analysis for the WM condition only, to 463 investigate whether inter-subject variability in the midfrontal response modulation during the 464 delay had any predictive value for WM task-performance (and hence any functional relevance). 465 As for the CDA component, the mean amplitude of the midfrontal signal was averaged during 466 the first second of the delay for each participant per trial group and correlated with individual 467 WM performance (see Fig. 4j-k). This analysis revealed a significant negative EEG-behaviour 468

EEG analysis: Explicit recollection during the visual WM task 499 500
Next, we assessed learning-associated changes in the parietal CA-response and the midfrontal 501 P170 component during explicit recollection of the learned colour-array, as probed by the 502 memory-cue in the WM condition following the 2500ms delay. 503 As can be seen in Fig. 5a, a memory-cue-induced CA component is not present during early 504 learning (first four trials), but emerges during late learning (pairwise comparison Trial-group: 505 T(18)= 2.2; p=0.04). For the midfrontal P170 peak (see Fig. 5b Plotted are the parietal CA difference waves (a) and midfrontal P170 responses (b) for both trial groups during explicit recollection of the memorized colour-array, as probed by the memory cue following the WM delay (t0=memory-cue onset). Mean CA amplitudes are calculated during a 300ms time-window while the cue was on screen (b; time-window indicated as filled gray box for CA in a) and for the midfrontal P170 peak after cue-onset (d; time-window indicated as as gray line for P170 peak in c ).

EEG analysis: Reflection of longer-term visual memory consolidation in the modulation of 513
the midfrontal response during the recognition task. 514

515
In line with our predictions based on previous studies 1,2 , we observed a WM-specific 516 modulation of the midfrontal P170 component, which became more pronounced with learning. 517 While we did not find a direct link with task-performance for modulation of the P170 peak, we 518 did observe a significant correlation between a sustained negative modulation of the midfrontal 519 component during the WM delay and WM accuracy, when learning progressed. These findings 520 imply a functional significance of this modulation for visual WM maintenance. Furthermore, 521 we provide supporting evidence indicating that the sustained midfrontal response modulation 522 reflects recruitment of an LTM-facilitating system, as significant correlations were also present 523 between this activity and performance accuracy during the subsequent recognition task. 524 To investigate whether perceptual memory consolidation was manifested as modulation of the 525 midfrontal P170 response upon re-exposure to the colour-arrays 10 , the EEG signal was 526 measured while participants were performing the post-hoc recognition task. Fig. 6(a) shows the 527 midfrontal response to the presented colour-arrays for correct trials, separated for repetition 528 (old/novel). Pairwise comparison revealed a greater P170 mean amplitude for repeated (old) 529 colour-arrays compared to novel colour-arrays, indicating a longer-lasting modulation (see

Discussion 541
The current study investigated whether the gradual strengthening of an explicit and rich visual 542 memory concurs with a shift of parietal-to-midfrontal neural activity, as was previously shown 543 for the memory of a single visual-search target (probed by the implicit memory measure of 544 response speed) 1,2 . In addition, we also investigated the hypothesis that this shift in neural 545 correlates could signify the recruitment of a longer-term-memory supporting storage system. 546 Our results reveal a learning-associated decline in parietal CDA during repeated memorizing 547 of a multi-item colour-array. However, even during the early learning phase, the CDA is, on 548 average, only briefly observed and not throughout the whole WM-delay period (of note here is 549 that in previous studies much shorter delay intervals were tested (ca 1000ms)) 1,2,4,6 . This could 550 indicate that maintenance of the memory trace is just briefly and not sufficiently sustained by 551 parietal activity even during these early learning trials. Alternatively, it is possible that the 552 memory representation becomes less lateralized throughout the delay and therefore hard to 553 detect by measuring the CDA. 554 In addition to this, our results provide clear support for an involvement of this contralateral 555 parietal activity in encoding of the colour-array, as a positive EEG-behaviour correlation with 556 performance on the post-hoc recognition task is observed. Qualitatively, the C(D)A narrows in 557 duration when learning progresses and is still (and only) present during array-presentation. This 558 may indicate involvement in a (gradually) more efficient encoding process. As is apparent from 559 the behavioural data, learning had not reached ceiling level after 8 trials and encoding-related 560 activity may therefore reflect ongoing learning. Results from our pilot experiment showed that 561 more trial repetitions (16) significantly enhanced post-hoc recognition and proper consolidation 562 may occur particularly while WM performance is at ceiling. Future study may clarify whether 563 encoding-related CA activity disappears completely during ceiling-performance. 564 While fronto-parietal regions have been implied in memory-encoding before 22,23 , our findings 565 suggest that parietal CA reflects a specific mechanism, which can be dissociated from selective attention (as no CA activity was observed during the attention condition). This mechanism may 567 involve processes that promote LTM by strengthening inter-item associations 24,25 , and/or 568 spatial binding, possibly through interaction with MTL-and hippocampal structures [26][27][28][29]  insufficient maintenance) of visual input and subsequently to visualize the detailed, spatial 579 context of the memorized information during recollection. The sustained parietal activity during 580 these processes may facilitate information transfer to (and from) MTL regions 5,41-43 . 581 582 Our experimental paradigm was deliberately designed to promote this interaction with MTL-583 hippocampal regions. We expected that the combination of a beyond-WM-capacity-set-size and 584 the spatial context of our task, would enhance reliance on MTL-hippocampal structures (see 8 ). 585 Due to the large set size, we expected that the memory trace could not be sustained by parietal 586 visual WM mechanisms alone and that an extended memory network would need to be 587 recruited. It seems indeed the case that the parietal CDA response did not sustain the visual 588 memory sufficiently and we observed a WM-specific modulation of the midfrontal P170 589 component (compared to the attention condition), which became more prominent as learning 590 progressed. This modulation is similar to what was described before in the repeated search 591 paradigm 1,2 which was interpreted as a transfer of the memorandum (search target) to an LTM 592 storage system. This interpretation was motivated by previous studies which demonstrated 593 modulation of this component during long-term perceptual memory priming processes 10 . In 594 our task we do not observe a significant correlation between modulation of the P170 component 595 and task-accuracy. However, we find that the level of negative modulation of the midfrontal 596 response during the delay period is predictive for WM accuracy when learning progresses (and 597 a transition from WM to short-term memory occurs). This is an indication that sustained 598 modulation of midfrontal activity is involved in maintaining the memory trace during the WM 599 delay. 600

601
In addition, the averaged midfrontal component during WM-delay correlated with performance 602 accuracy in the subsequent post-hoc recognition task. This suggests that the sustained 603 modulation of the midfrontal component reflects recruitment of an LTM-supporting system 1,2 604 which contributes to WM-maintenance during the delay phase. 605 Surprisingly, we observe a positive EEG-behaviour correlation during early learning trials, both 606 for task accuracy in the WM-and the post-hoc recognition task. Perhaps midfrontal activity 607 subserves different underlying mechanisms and memory is first actively consolidated during 608 early learning trials, while merely protected from interference during late learning trials. 609 The source of the midfrontal activity may be the ventromedial prefrontal cortex, which has been 610 implicated in memory encoding by multiple studies 44-46 (and see 47 for a review). The nature of 611 the signal remains unclear but further analysis could clarify a potential link with midfrontal 612 theta oscillations, which have been associated with maintenance during working memory 15 and 613 information-transfer to MTL/hippocampal regions during encoding-and retrieval processes in 614 long-term memory (and see 16 for a review). Modulation of the P170 peak might reflect a 615 memory-specific phase-reset mechanism of midfrontal theta-oscillations, which prepares the 616 system for memory encoding 15 . 617 618 Finally, we not only observed visual memory consolidation on the behavioural level, but also 619 observed lasting influences on the midfrontal response upon re-exposure to the previously 620 presented colour-arrays during the recognition task. A modulation in the P170 peak was 621 observed during trials in which colour-arrays were correctly recognized as old or novel. This 622 further supports the notion that the P170 modulation during the WM task is indicative of the 623 formation of a longer-term memory trace. 624 625 626

Conclusion 627
In conclusion, the present study provides evidence for the recruitment of an LTM-supporting 628 system when a visual image in repeatedly held in working memory. This is indexed by 629 modulation of a midfrontal component during encoding and maintenance and is accompanied 630 by the emergence of a parietal CA during gradually enhanced recollection. In contrast, there is 631 a gradual decline of parietal C(D)A following array presentation, which may partly reflect an 632 decreased (insufficient) contribution to WM maintenance, but seems more related to LTM 633 encoding. 634 635 636 Figure 1 Schematic representation of the task. Two different arrays, consisting out of six coloured spatially-xed squares each, are bilaterally presented with one hemi eld cued for covert attention at the beginning of (and throughout) a testing block. The same array-combination was repeated for 8 trials in one testing block and spatial locations of all squares would remain visible as white placeholders during stimulus intervals. Participants had to match a central colour cue with that at one highlighted square in the array, either while the colour-array was still present on screen (attention/ATT condition, left) or following a 2.5 second delay (WM condition, right).  Recognition task accuracy scores. Accuracy scores on the post-hoc recognition task are plotted as d' scores across conditions (a) and separated for colour-arrays that had previously been shown during the ATT or the WM condition (b).

Figure 4
Parietal contralateral activity and midfrontal response during encoding and WM maintenance. Upper panel: C(D)A components are plotted as difference waves between contra-and ipsilateral activity per task-condition (ATT/WM), from the moment of colour-array onset and throughout 2000ms of the WM delay period during the rst four trials (a) and the second four trials (b). The mean amplitudes of the C(D)A difference waves are plotted per task-condition and trial-group, as calculated during a 300 ms time-window immediately following array-offset (c) (time-window indicated as lled gray box in a and b). Second panel: Plotted are the midfrontal responses for all 4 task conditions from colour-array onset throughout the delay period (d.) and for the midfrontal P170 peak speci cally during 350 ms following colour-array onset s (e). Mean amplitudes per task-condition and trial-group were calculated and plotted for the P170 peak (f; time-point indicated as gray line in e) and also for the 300 ms time-window post array-offset (g). Lower panels: Correlations are shown between mean amplitudes (CDA and midfrontal responses), calculated across the rst 1000ms of the WM-delay following array-offset (indicated by open gray box in a, b and d) and d' accuracy scores during the WM task (CDA: h,i; midfrontal: j,k ) and the posthoc recognition task (CDA:l,m; midfrontal:n,o).

Figure 5
Parietal contralateral activity and midfrontal response during WM recollection. Plotted are the parietal CA difference waves (a) and midfrontal P170 responses (b) for both trial groups during explicit recollection of the memorized colour-array, as probed by the memory cue following the WM delay (t0=memory-cue onset). Mean CA amplitudes are calculated during a 300ms time-window while the cue was on screen (b; time-window indicated as lled gray box for CA in a) and for the midfrontal P170 peak after cue-onset (d; time-window indicated as as gray line for P170 peak in c ).

Figure 6
Midfrontal response during the recognition task. Plotted are the midfrontal responses during the post-hoc recognition task to previously shown (old)-and novel colour-arrays for correct trials (a). Mean amplitudes at 170 ms post array-onset (indicated by gray line in a) are plotted per array-type (old/novel) (b).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Heinenetal2021Supplements.pdf