Automatic and effortful control of interference in working memory can be distinguished by unique behavioral and functional brain representations

Goal-irrelevant information in working memory (WM) may enter the focus of attention (FOA) during a task and cause proactive interference (PI). In the current study we used fMRI to test several hypotheses concerning the boundary conditions of PI in WM using a modified verbal 2-back task. Temporal distance between item and lure presentation was manipulated to evaluate potential differences among hypothesized states of FOA, short-term memory and long-term memory. PI was present for the most proximal 3-back lures but dissipated with lure distance along with increased activation in brain regions critical for memory recollection, such as right prefrontal cortex, parietal cortex, and hippocampus. Reduced PI and less IFG activation were also observed after repeated item presentation, supporting the notion that a rehearsed encoding of item-context information reduces the need for interference control. Moreover, a trial-by-trial approach revealed activity in ACC, insula, IFG, and parietal cortex with increasing lure trial interference regardless of distance. The current results are first evidence for an observable transition of cognitive control, to include MTL regions involved in recalling task-relevant information from outside the FOA when resolving PI in WM.


Introduction
The ability to acquire new information requires a highly adaptive cognitive system, as it is unclear at the time of memory encoding whether certain information might prove useful or not. This inevitably places high demands on a system that is efficient in controlling unwanted memories and resolving interference from familiar, but currently goal-irrelevant representations. One such type of interference is proactive interference (PI). PI arises when past goal-relevant information interferes with current goal directed cognition and is commonly tested in experimental tasks by evoking familiarity from a task-irrelevant probe after recent exposure on a previous trial ( Hasher and Zacks, 1988 ;Jonides and Nee, 2006 ). While PI has primarily been studied in the context of long-term memory (LTM; Kliegl and Bäuml, 2021 ), there is substantial evidence for PI also in working memory (WM; Jonides and Nee, 2006 ). WM is typically characterized by two distinct component processes: on-line maintenance of information within the focus of attention (FOA) and retrieval of information from long term memory (LTM). Some propose that FOA supra-capacity items are actively maintained for use in ongoing cognition within a "direct access region ". This three-state model of WM that includes FOA, the direct access region, and activated LTM is supported by both behavioral and neural evidence ( Jonides et al., 2008 ;Nee et al., 2013 ;Nee and Jonides, 2011 ;Oberauer, 2002 ;Öztekin et al., 2009 a). To differentiate between these different memory states, studies have used a serial item-recognition task to manipulate temporal recency of goal-relevant (target) information. No study to date has examined how goal-irrelevant information in WM is controlled within different memory states.
Representations in WM that are no longer relevant for behavior can be quickly removed from the FOA, thereby reducing the load on the limited capacity of the FOA. It has been proposed that items outside the FOA persists in a state of heightened availability which can both facilitate target recognition ( Lewis-Peacock et al., 2012 ) and/or result in PI ( Lustig and Jantz, 2014 ). Little is known regarding the removal and control of no longer goal-relevant information in memory, and at what time such information no longer causes interference. Previous behavioral work has demonstrated that PI gradually dissipates when the temporal distance between the first presentation of an item and its reappearance as a lure item increased using a modified n-back task ( McCabe and Hartman, 2008 ;Samrani et al., 2017 ;Schmiedek et al., 2009 a;Shipstead et al., 2016 ). This suggests that when the distance between the item presentation and the reappearance of a lure item in-creases, the temporal offset aids correct decision-making while reducing the interference effect.
It has been suggested that familiarity-based PI can be counteracted by retrieval of item-context details (i.e., recollection; Badre and Wagner, 2005 ;Jonides and Nee, 2006 ;Oberauer, 2005 ;Szmalec et al., 2011 ). While familiarity is considered an automatic and implicit type of recognition that lacks contextual details, recollection is defined as a type of controlled retrieval of both memory content and contextual details associated with the encoding episode ( Yonelinas, 2002 ). In order for efficient control of PI to take place, recollection is required to override the misleading activation from the familiarity process by providing contextual evidence that the lure item is not in the correct temporal position. Possibly, items with weak memory representations or items that are unbound ( Oberauer, 2005 ) from their temporal context can be automatically inhibited without engaging recollection. Also, recollection can be affected by deep encoding, where efficient encoding can be seen to determine subsequent familiarity-and recollection-based recognition ( Rugg et al., 2012 ;Yonelinas et al., 2010 ).
Much evidence implicates the lateral prefrontal cortex (LPFC) as critical for interference control, including the inferior frontal gyrus (IFG; Badre and Wagner, 2007 ;Nee et al., 2013 ;Samrani et al., 2019 ;Thompson-Schill et al., 2002 ). The anterior cingulate cortex (ACC; Botvinick et al., 2001 ; and the insula ( Cieslik et al., 2015 ;Rottschy et al., 2012 ;Xu et al., 2016 ) have also been commonly associated with interference control. It has also been demonstrated that the medial temporal lobe (MTL) is involved in control of PI ( Öztekin et al., 2009b ;Öztekin and McElree, 2007 ). Co-activation of IFG and MTL has been implicated in strengthening of goal-relevant features of items during LTM encoding ( Addis and McAndrews, 2006 ;Blumenfeld and Ranganath, 2007 ), and is consistent with a controlled retrieval process that influences information from LTM (e.g Badre and Wagner, 2007 ). Possibly, LTM operations are needed for recollection of item-context details for distant, but not proximal, lure items that are outside the FOA.
Here, we set out to examine three novel aspects of working memory that may provide new insights into the neurocognitive underpinnings of cognitive inhibition in WM. Control of PI was investigated using the n-back task, which included familiar lure trials ( Gray et al., 2003 ;Marklund and Persson, 2012 ;Schmiedek et al., 2009 a). Participants were scanned with functional magnetic resonance imaging (fMRI) while performing a modified n-back task in which the temporal properties of PI was manipulated. Trial-level analyzes and a parametric design was used to estimate the impact of continuous predictors while accounting for RT-related effects in the data, and functional connectivity was used to examine task-related brain interactions. Also, a post-fMRI recognition test that included words from the n-back task was included to ensure that dissipation of interference did not occur as a function of forgetting.
First, we examined whether different temporal contexts (i.e., whether a lure was presented proximally or more distantly) affected the level of PI, and to what extent this manipulation resulted in changes in brain activation. We hypothesized that temporally distant lure items would result in reduced behavioral interference. Also, distant lure items that are presumably stored in LTM, needs to be reactivated into the FOA for recollection of item-context information critical for interference control. Thus, the MTL contribution to performance should become more apparent as the temporal delay increases. We also expected to find differential frontal contributions to proximal and distant lures, with more interference-related frontal activation for proximal lures and LTM-related frontal engagement for distant lures. We additionally utilized task-related functional connectivity to address the question of how hippocampal connectivity changes as a function of temporal distance. Increased hippocampal functional connectivity with increased target to lure distance in canonical LTM regions could indicate a higher reliance of LTM to control interference from distant task-irrelevant lures.
Second, we test the open question of whether repeated lure item presentation help reduce or leads to increased interference in WM as a result of rehearsed encoding. Based on previous findings that itemrepetition can facilitate recall when repetitions are relatively adjacent ( Crowder, 1968 ;Henson, 1998 ;Lee, 1976 ), resulting in increased performance, we hypothesized that repetition of lure trials would lead to increased distinctiveness and/or associated contextual information and thereby reduce interference. This will be tested by contrasting familiar 3-back trials that are presented three times (3B REPEAT ; extended encoding) with familiar 3B trials that are presented only twice (3B non-REPEAT ; limited encoding). Two alternative hypotheses are tested. (i) stronger WM representations may result also in a heightened familiarity signal and, therefore, result in increased interference. (ii) stronger WM representations may carry more diagnostic information in the form of itemcontext information. If interference is resolved by using item-context information, stronger representations would result in less PI, and less engagement of brain regions supporting interference control.
Third, we examine whether different neural processes underlie automatic compared to effortful inhibition of PI by separating lure trials based on the behavioral signature of PI (low vs. high interference) on a trial-by-trial basis. A similar approach of separating trials based on response times (RTs; Weissman et al., 2006 ;Yarkoni et al., 2009 ) and degree of conflict ( Cohen and Cavanagh, 2011 ), has successfully been used previously for identifying neurocognitive processes related to successful task performance, and at the same time avoid unwanted noise found in cross-trial averages ( Debener et al., 2007 ). It has been argued that a more faithful representation of brain-behavior dynamics is to parametrically modulate within-condition trial variability ( Cohen and Cavanagh, 2011 ;Pernet et al., 2011 ). In the current study we examined if brain activation was associated with trial-by-trial RT variability to explore qualitative differences, both behaviorally and functionally, between proximal and distant lure items.

Participants
Twenty-four healthy, right-handed participants between 18 and 35 years (M: 28.5; SD: 4.0; 12 women) took part in the study. Neurologic and psychiatric health status was assessed by a self-report questionnaire that included questions of whether participants were currently being treated for psychiatric or neurological conditions like depression, epilepsy, diabetes, trauma, and Parkinson's disease. Individuals with neurological or psychiatric conditions, or with implants and other surgery unfit for MRI scanning were excluded. One participant was excluded from the analysis due to excessive movement and technical problems during scanning, leaving 23 participants for all analyzes. Participants received financial compensation of 600 SEK for their participation. The study was approved by the Regional Ethical Review Board in Stockholm, and written consent was obtained from all participants.

Cognitive measures
PI was measured using a verbal 2-back (2B) WM task which included familiar lure items ( Gray et al., 2003 ;Marklund and Persson, 2012 ) occurring either at 3, 5, 6, 7, 8, 9 or 10 trial(s) after first item presentation (i.e. 3-to 10-back lures; Samrani et al., 2017 ). These are referred to as 3B, 5B etc., up to 10B. The task design is illustrated in Fig. 1 (see also Supplementary Table 1). The task was divided in to two equal blocks of 105 trials each, with a one-minute break between blocks. The proportion and number of trials are presented in Supplementary Table 1 and consisted of (1) non-familiar words presented for the first time ( new trials ), (2) words presented for the second time at the correct 2B position ( target trials ), (3) words presented for the second time at an incorrect position, one word after the target position (3B; proximal lures ), and (4) words presented a third time, three to ten trials from the target position ( Fig. 1 ; 3B to 10B; distant lures ). Lure trials consisted of stimuli already presented 3-to 10 trials earlier, and required a 'No' response, and new trials were non-familiar trials that had never been presented previously, which also required a 'No' response. Target trials were 2B trials and required a 'Yes' response. For each presented word, participants were instructed to press with the right index finger on a MRI compatible response box, which corresponds to 'Yes' ( "Yes, the word I now see has been shown two words ago ") and the button on the middle finger for 'No' ( "No, the word I now see has not been shown two words ago "). Distant lures were all recycled from either previous target trials or proximal lures with the aim to lower the total amount of new trials, consequently the proportion of 'No' answers.
Stimuli and trial conditions were presented in the same fixed order for all participants. Stimuli consisted of common English nouns with a maximum of two syllables, and were presented one at a time for 2.5 s, with an inter-trial jittering of either 2, 2.66, 3.33, or 3.99 s. There was an even distribution of the jitter-timings, and the timing from a new trial to the target position always added up to a total inter-trial time of 5.99 s to avoid any differences in encoding time between target items. Participants were instructed to answer as quickly and accurately as possible and were also asked to not overtly rehearse the words. Relative difference scores (PI scores) were calculated as the relative proportional difference in RT between non-familiar trials and familiar lure trials (see Samrani et al., 2017Samrani et al., , 2019. Interference can thus be observed as the difference in % between lure trials (high interference trials) and nonfamiliar trials (no interference trials). For example, an increase in RT from non-familiar-to lure trials, would reflect being affected by interference, as more time/effort was needed to resolve interference. A relative difference score should represent a more salient measure of executive control, as it considers baseline individual differences in the variables in question, such as processing speed ( Salthouse, 1996 ). Interference scores based on RT data are referred to as RT interference scores (RTIS). Median RTs were used to reduce the influence of extreme values.
To ensure that changes in PI was not related to forgetting, each participant also performed a recognition task immediately after the scanning session. The recognition task was presented on a 15.4-in. laptop computer screen (Compaq nx 7300), using E-Prime software (Psychology Software Tools, Inc., Pittsburgh, PA; version 1.1) for stimulus presentation. Stimuli were presented in white (Courier new font, font size 36 points) against a black background. Each word was presented one at a time for 5 s. A new word was shown either after the 5 s had passed or until the participant gave an answer. Participants made responses manually by either pressing the key "Ä", on a standard Swedish keyboard, with their right index finger for a yes response, or the key "' " with their right middle finger for a no response. All 105 words were either one-or two-syllable common English nouns. Thirty-five of these words (1/3 of total) had not been presented inside the scanner (i.e., lures), and seventy words (2/3 of total) had previously been presented in the fMRI 2-back task. Positive responses to previously presented items were scored as hits and positive responses to lure trials were scored as false alarms. Hit rate minus false alarm rate was used as a measure of recognition memory performance.

Behavioral statistical analyzes
RTs were calculated for correct trials only. The first two trials for each block were not included in the analysis due to their high predictability. Median RTs were extracted for target trials, new trials, and lure trials for each condition. Lure trial RTs were divided with new trials to get a relative difference score ( ( − 1 ) × 100 ) for each condition. A higher relative difference score indicates more interference. Moreover, each lure trial RT was characterized as either a "highinterference " (showing above-threshold interference) or "low interference " (showing below-threshold interference) trial, with the aim to separate proactive interference on a trial level. Single lure trials were compared against the median RT of non-familiar trials for each participant, creating a relative score (RTIS) for each trial to be used in the GLMs. This calculation included three steps. Firstly, we used a fixed median value to determine each participant's average RT for new trials and then, the relative difference between that mean RT value and each lure trial was calculated and thus obtaining a measure of PI for each lure trial. In the next step, a cutoff of 15% RTIS was used. Lure trials with an interference score above 15% was considered high-interference trials (PI + ) and lure trials with an interference score below 15% were considered low-interference trials (PI-). The cut-off was based on previous largescale individual level data ( Samrani et al., 2019 ). To test the validity of this approach, we pooled lure trials across all participants with an RTIS below the 15% threshold ( N = 1074) and this score was compared against 0 (as 0% RTIS would indicate no interference). The mean interference score for these trials was significantly below 0 (M RTIS = − 7.6%, SD RTIS = 14.2, t(1073) = − 17.6, P < .001), indicating that these trials were not eliciting interference. Behavioral analyzes on temporal distance (proximal vs. distant lure positions) included only repeated lure trials.
In addition to reporting classical frequentist p-values, we also calculated Bayes Factor (BF). The BF is a statistical technique that helps conclude whether the collected data favors the null-hypothesis (H0) or the alternative hypothesis (H1); thus, the BF could be considered as a weight of evidence provided by the data ( Wagenmakers et al., 2018 ). BFs were calculated using SPSS. Here we report BF01 values. BF01 values between 1 and 3 indicate anecdotal evidence for H0, while values between 3 and 10 indicate substantial evidence for H0. Conversely, while values between 0.33 and 1 indicate anecdotal evidence for H1, values between 0.1 and 0.33 indicate substantial evidence for H1. If the BF is below 0.1, 0.03, or 0.01, it indicates strong, very strong, or extreme ev-idence for H1, respectively. Values close to 1 do not support either H0 or H1.

MRI acquisition
Participants were scanned with an 8-channel phased array receiving head coil (Discovery MR750 3.0T scanner, General Electric). T1weighted 3D SPGR images were obtained with the following MRI scanner parameters: TR: 8.2 ms, TE: 3.2 ms, field of view: 25 cm, 176 axial slices, flip angle of 12°Task-related fMRI data were acquired using a gradient-echo-planar imaging sequence with the following MRI scanner parameters: repetition time = 2000 msec, echo time = 30 msec, flip angle = 70°, field of view = 25 cm. Forty-two transaxial slices with a thickness of 3 mm (0.35 mm gap) were acquired. Ten initial dummy scans were collected to allow for the fMRI signal to reach equilibrium. The stimuli were presented on a computer screen seen through a tilted mirror. E-Prime software (Psychology Software Tools, Inc., Pittsburgh, PA; version 2.0) was used for stimulus presentation and recordings. Headphones and earplugs were used to dampen scanner noise, and cushions inside the head-coil helped to minimize head movements.

Preprocessing
All fMRI data were preprocessed using the statistical parametric mapping software (SPM12; Wellcome Department of Cognitive Neurology, London, U.K.) implemented in MATLAB 9.3 (Mathworks, Inc., Natick, MA). Before analysis, the data were preprocessed in the following way: slice timing correction, movement correction by unwarping and realignment to the first image of each volume, co-registration, normalization to a sample specific template using DARTEL ( Ashburner, 2007 ), and affine alignment to Montreal Neurological Institute standard space and smoothing with an 8 mm FWHM Gaussian kernel. Following the co-registration step, the T1 image was segmented into gray matter and white matter. The final voxel size was 1.5 × 1.5 × 1.5 mm.

Selection of regions of interest
A-priori ROIs were identified from the automated anatomical labeling (AAL) atlas implemented in Matlab. The first analysis (GLM 1, see below) tested the hypothesis that activation in the HC would increase as a function of temporal distance to the lure item. Therefore, the HC was selected as an a-priori ROI. The second analysis (GLM 2, see below) tested the hypothesis that repeated item presentation would require less engagement in a network of regions previously identified as critical for interference control. These regions included the IFG, insula and ACC. The third analysis (GLM 3, see below) tested the hypothesis that only high-interference trials would engage regions with a known role in interference control. These regions included the IFG, insula and ACC. For all ROIs we used small-volume correction.

Statistical models
We estimated three first-level models in the current study to avoid co-linearity problems when answering the different aims of the current study. Only correct responses (comprising 95.8% of total trials) were included in the models. Regressors for each task condition were convolved with a canonical hemodynamic response function. Motion regressors from previous SPM preprocessing steps were added to each design matrix as regressors of no interest.
GLM1: The first model examined to what extent brain activation was modulated as a function of increased temporal distance time between the 1st presentation of an item to its reappearance as a lure item. We hypothesized that there should be an increased reliance on MTL regions when representations are reactivated into the FOA. Thus, any MTL contribution to performance will become more apparent as the temporal delay increases. High and low interference trials were modeled separately.
For each individual, the model consisted of target trials, non-familiar trials, all high interference lure trials (3-to 10-back), the parametric modulator of lure distance added to these lures, all low interference lure trials (3-to 10-back), and the parametric modulator of lure distance for these lures. A contrast image was made for each participant, weighting the parametric modulator of interest by 1, and then compared on a group level, using a one-sample t -test.
GLM2: The second model tests whether deeper encoding by repeated item presentation modulate brain activation associated with PI. This was tested by contrasting familiar 3B trials that are presented three times (3B REPEAT ; extended encoding) with familiar 3B trials that are presented only twice (3B non-REPEAT ; limited encoding). This model also enabled traditional analyzes of PI-related brain activation by contrasting familiar 3B lures with non-familiar trials. For each individual, the model consisted of target trials, non-familiar trials, non-repeated 3-back lures, and each repeated lure distance separately (from 3-to 10-back). Contrast images of 3B REPEAT versus 3B non-REPEAT , as well as for non-familiar trials versus 3B was made for each participant and then compared on a group level, using one-sample t-tests. GLM3: To examine whether different neural processes underlie automatic compared to effortful inhibition of PI we separated lure trials based on the behavioral signature of PI on a trial-by-trial basis. We therefore added RTIS as a parametric modulator in a third GLM. For each individual, the model consisted of target trials, non-familiar trials, all lure trials, and the parametric modulator for RTIS to all lure trials. A contrast image was made for each participant, weighting the parametric modulator of interest by 1, and then compared on a group level, using a one-sample t -test.
At the whole brain level, we set a family-wise error (FWE) corrected threshold of p < .05 at the cluster level for the SPM analyzes using a cluster-defining peak threshold of p < .001. Activation with a familywise error (FWE) corrected threshold of p < .05 at the peak level was also considered significant. For ROI analyzes we used an uncorrected threshold of p < .001, which were followed-up using small-volume correction (SVC) by applying a mask of the ROI with a radius of 8 mm centered on the peak voxel, assessing significance at a p-value < 0.05, FWE corrected.

Functional connectivity analyzes
Functional connectivity was examined using ROI-to-ROI connectivity maps implemented in the CONN v.18b ( www.nitrc.org/projects/conn ) toolbox ( Whitfield-Gabrieli and Nieto-Castanon, 2012 ). The hippocampus (x, y, z = -30 -14 -14) was used as a seed region in the analysis, and this ROI was defined from the GLM1 analysis (see above). A 4 mm volumetric sphere was created from those coordinates for subsequent analysis in the CONN connectivity toolbox. The idea was to investigate whether (a) there is a functional link between known frontal regions during interference control and the hippocampus with increasing lure distance, and (b) that this link is specific to effortful interference control in WM. Further, the calculations of functional connectivity followed recommended and standardized preprocessing pipeline and procedures in CONN, changing only the high-end band-pass filter [0.08 to infinite] to better fit a task fMRI paradigm. There were four conditions of interest in the analysis: (1) proximal low-interference trials (RTIS < 15%, see 2.5 below), (2) proximal high-interference trials (RTIS > 15%), (3) distant low-interference trials (lure distance 5 to 10) and (4) distant high-interference trials. The remaining task conditions (target trials and new trials) were used as regressors/covariates in the design matrix.

Post-fMRI recognition task performance demonstrates that low-interference trials were not forgotten
Forgotten items that are no longer familiar would not result in PI. Therefore, in order to rule out the possibility that low interference trials were not forgotten, we assessed memory performance after the scanning session. The offline recognition task showed that memory performance was high (M Hits-FA = 0.77, SD Hits-FA = 0.14). That is, a majority of the words used in the 2B task were well recognized up to 45 min after the fMRI 2B task. We also conducted a supplementary analysis to examine if any significant differences existed between prior high-interference and prior low-interference items in post-fMRI recognition performance. Using a paired-samples t -test we found that the difference in error rate between high-(M RTIS = 11.3%, SD RTIS = 11.3%) and low-interference trials (M RTIS = 7.5%, SD RTIS = 8.0%; was non-significant (t(22) = − 1.94, p = .069, d = 0.39, BF 01 = 1.21). Together, this indicates that the difference between high-and low-interference trials was not primarily a consequence of forgetting low-interference trials.

Reduced PI was observed when deeper target encoding was facilitated by repeated item presentation
Non-repeated 3B lure trials were presented twice, and repeated 3B lure trials were presented three times. A comparison between nonrepeated 3-back lure trials (M RTIS = 25.0%, SD RTIS = 14.1) and repeated 3-back lure trials (M RTIS = 17.4%, SD RTIS = 16.7) using a paired-samples t -test, showed a difference in RTIS ( Fig. 4 B; t (22) = 2.2, p = .04, d = 0.49 BF 01 = 0.74), although with limited support from the Bayesian analysis. However, there was a reduced proportion of high-interference versus low-interference lure trials between the two conditions (

Fig. 2.
Results showing the effect of lure distance in brain and behavior. Relative difference scores were calculated as the relative proportional difference between non-familiar trials and familiar lure trials (a) relative interference scores of RTs (RTIS) across temporal distance. A higher percentage implies being more negatively affected by interference. (b) Relative interference scores of accuracy across temporal lure distance. A higher percentage implies being more negatively affected by interference. (c) Proportion high-interference (RTIS > 15%) to low-interference trials across temporal lure distance. Error bars show standard error of the mean (SEM).

Distant lures resulted in greater hippocampal activation compared to proximal lures
The main idea for manipulating temporal lure distance was to capture the transition of WM representations to LTM. We hypothesized that for items outside the FOA, MTL-based recollection from LTM would be needed to resolve PI. Indeed, we found that activation in the left HC, together with other regions critical for LTM retrieval increased as a function of lure distance for high-interference trials only ( Fig. 3 B, Table 2 ). No significant brain activity was found for low-interference trials with increasing temporal distance. It should be noted that no significant activation was found as a function of distance when an additional control analysis was restricted to only high-interference lure trials with a distance of 5-10 back (i.e., not including 3-back trials). Even though this control analysis was somewhat underpowered with regards to the number of trials included, it might suggest that most of the observed main effect is related to a difference between proximal (3-back) and distant (5-10-back) trials.

Left anterior hippocampus show functional connectivity with prefrontal cortex for distant high-interference trials
Since the left anterior hippocampus showed greater activation as a function of increasing lure distance (GLM1), we further explored functional interactions between this region and the rest of the brain. To this end, we conducted a whole-brain functional connectivity analysis (see methods) in which we identified functional connectivity patterns between the left anterior hippocampus and the rest of the brain for highinterference distant lures (5B to 10B). We found significant positive functional connectivity between the HC and clusters in the right IFG (pars triangularis), basal ganglia, left orbitofrontal cortex, amygdala, ACC, insula and bilateral temporal regions ( Fig. 3 B; Supplementary Table 3) for distant high interference trials.

Repeated lure item presentation resulted in reduced IFG activation
The difference between repeated and non-repeated 3-back lures was examined in a whole brain analysis using the contrast of 3B non-REPEAT > 3B REPEAT (GLM2). This analysis showed significantly more activation in the left IFG (pars triangularis, x, y, z = − 52, 22, 15, p (FWE)SVC < 0.05, t = 4.50), for non-repeated 3-back lures ( Fig. 4 D; Table 3 ). No significant peak activation was found in the reverse contrast (3B REPEAT > 3B non-REPEAT ).

Trial-by-trial analyzes identifies brain regions associated with high-interference lure trials
Finally, we explored unique brain activity associated with high interference using RTIS as a parametric modulator (GLM3). These analyzes aimed to capture brain activation associated with within-condition triallevel variability in performance to dissociate brain activation directly related to level of interference. A whole brain analysis of all lure trials with RTIS as parametric modulator showed activity in the canonical interference control network, including IFG, insula, and ACC ( Fig. 5 B). Additional activation was also found in parietal regions ( Fig. 5 B; Table 4 ).  3. (a) Significant brain activity coupled with increasing lure distance, for all accurately resolved lure trials. peak activity extracted from uncorrected p = .001 for display purposes. (b) ROI-to-ROI connectivity analysis using anterior hippocampus as the seed ROI (MNI: 30 -14 -14; 4 mm sphere) with connectivity to the rest of the brain for high-interference distant lures (5-to 10-back) in the n-back task. Colors correspond to t-values of each connection to a specific ROI. L, left hemisphere; R, right hemisphere; SMG = supramarginal gyrus, IFG = inferior frontal gyrus, STG = superior temporal gyrus, TP = temporal pole, MTG = middle temporal gyrus, ICC = intracalcarine cortex, CG = cingulate gyrus, FOrb = fronto-orbital cortex, HC = hippocampus, LG = lingual gyrus, TFG = temporal fusiform gyrus, OFG = occipital fusiform gyrus, CO = central opercular cortex, PO = parietal operculum cortex, PP = planum polare, SSC = supracalcarine cortex.

Fig. 4.
Results showing the effect of repetition in brain and behavior. Relative difference scores were calculated as the relative proportional difference between non-familiar trials and familiar lure trials (a) relative interference scores of accuracy for repeated and non-repeated 3B lures. A higher percentage implies being more negatively affected by interference. (b) Relative interference scores of RTs for repeated and non-repeated 3B lures. A higher percentage implies being more negatively affected by interference. (c) Proportion high-interference (RTIS > 15%) to low-interference trials for repeated and non-repeated 3B lures. Error bars show standard error of the mean (SEM). (d) Significantly greater brain activity for 3B lures compared to repeated 3B lures. Peak activity is extracted from uncorrected p < .001 for display purposes. The coordinates (x, y, z) of each section in MNI space can be seen in brackets.

Discussion
The present experiment aimed to shed light on behavioral and neural properties of interference control during WM updating. Three main findings were obtained in the current study. First, increased temporal lure distance resulted in less behavioral PI and greater engagement of regions critical for LTM, including HC, parietal cortex and RIFG. In addition, HC -RIFG connectivity increased as a function of temporal target to lure distance supporting a view that effortful inhibition of lures implicates retrieval of LTM representations from outside the FOA. Second, repeated item presentation resulted in less PI along with reduced IFG and insula activation, indicating that recollection of item-context details may help counteract behavioral consequences of PI. Third, a within-condition, trial-by-trial analysis of familiar lure trials revealed a difference in engagement of interference-related brain regions, where low-interference items had reduced brain activation in these regions, and that neural engagement for these trials did not differ from non-familiar control trials. These results suggest that PI arising from familiar lure trials in WM can be resolved either by effortful engagement of regions including the IFG, ACC, and HC or being automatically inhibited by a process that does not implicate neural control. Importantly, automatic inhibition could not be attributed to forgetting, as participants retained high levels of recognition performance well up to 45 min after the main experiment. A single trial analytical model implemented in this study further extends these findings in several domains, showing novel evidence for the role of two separate modes of inhibition during PI.
Our first goal was to investigate the behavioral and neural boundary conditions for interference in WM. As expected, increased temporal lure distance was accompanied by reduced PI. This behavioral facilitation was primarily driven by a reduced number of above-threshold interference items. This is in line with previous behavioral evidence of reduced interference as a function of increased temporal lure distance   ( McCabe and Hartman, 2008 ;Schmiedek et al., 2009 b). Our findings extend these previous observations by using a longer temporal offset and trial-specific stimuli (i.e., no continuous recycling of items) which allows for assessment of long-term dissipation of PI for specific WM representations. On that note, increased temporal lure distance was associated with activation in regions with a known role in LTM retrieval, including the right PFC, hippocampus, and parietal cortex ( Addis et al., 2016 ;Eichenbaum, 2017 ;Guidotti et al., 2019 ;Rugg and King, 2018 ;Sestieri et al., 2017 ;Wagner et al., 2005 ;Xue, 2018 ). This finding supports the view that no longer goal-relevant items are removed from the focus of attention and stored in LTM. Re -exposure of distant lure items thus requires recollection of item-context information from LTM to successfully reject the item based on non-matching temporal context. It should be noted that only high-interference lure trials showed this pattern. A considerable amount of lure trials did not result in PI and also did not show any additional brain activity with increasing lure distance. Possibly, these low-interference lure trials are automatically inhibited, and do not involve LTM processes to be accurately resolved.
There is mounting evidence that the right PFC, hippocampus, and parietal cortex are critical for LTM. For example, the parietal cortex is central for episodic memory retrieval , and this region is particularly critical for source monitoring of LTM representations ( Mitchell and Johnson, 2009 ), source memory decisions ( Guidotti et al., 2019 ), and recovery of temporal order information ( Hsieh et al., 2014 ;Öztekin et al., 2009 a). These accounts are all well in line with a proposed role for recollection of contextual details in resolving familiaritybased conflict. Indeed, other two-alternative forced-choice studies have reported MTL activity for retrieval from LTM when participants distinguish between old and new items ( Öztekin et al., 2010 , 2009 a). Similarly, our finding that HC engagement tracks temporal lure distance also points to the involvement of LTM recollection processes, given that hippocampus shows limited engagement in familiarity-based recognition ( de Vanssay-Maigne et al., 2011 ;Montaldi and Mayes, 2010 ). Moreover, the HC has unequivocally been linked to memory retrieval and is expected by dual-store models to support LTM retrieval only. There is much evidence showing that amnesic patients can retain information over brief delays despite severe deficits in long-term memory ( Baddeley et al., 2010 ;Jeneson et al., 2010Jeneson et al., , 2011Jeneson et al., , 2012Shrager et al., 2008 ). While it has been demonstrated that these deficits become larger with increasing time between encoding and memory testing ( Aggleton et al., 1992 ;Buffalo et al., 1998 ;Holdstock et al., 1995Holdstock et al., , 2000Owen et al., 1995 ), damage to the MTL can produce memory deficits with retention intervals as short as 2-10 s ( Hannula et al., 2006 ). Our results are in line with the view that memory is best thought of as a continuum where MTL is increasingly critical as retention intervals increases.
Additional neural evidence that items are retrieved from LTM to resolve PI in the n-back task comes from our functional connectivity results. A longer temporal lure distance was associated with increased connectivity between the HC and other regions critical for LTM retrieval, such as IFG, insula and ACC. Moreover, since only high-interference items were included in the analysis, an increased activation as a function of temporal distance could not be explained by memory strength. This is in line with human brain imaging findings of a ventral polysynaptic pathway supporting PFC-MTL interactions during controlled episodic retrieval ( Barredo et al., 2015 ), and animal studies showing bidirectional hippocampus-prefrontal cortex interactions that support contextdependent memory retrieval ( Preston and Eichenbaum, 2013 ). While the MTL is not connected to VLPFC monosynaptically, between-region communication supporting interference control is likely propagated along polysynaptic pathways with multiple intermediate sites. Previous multimodal imaging studies that combined fMRI and diffusion tensor imaging to reconstruct tracts from ROIs in PFC and the HC showed that these connections are critical for episodic memory ( Schott et al., 2011 ;Takahashi et al., 2007 ). In addition to functional connectivity between the HC and IFG, the putamen, thalamus, and superior temporal regions also showed elevated functional connectivity with the HC as a function of increased lure temporal distance. This could indicate that when retrieval demands increase as a function of temporal distance, IFG signals for retrieval of information from multiple sources to support correct rejection of familiar items. The basal ganglia has been implicated in memory retrieval ( Han et al., 2010 ;Schwarze et al., 2013 ;Scimeca and Badre, 2012 ), and may interact with HC and IFG during recollection to support interference control.
The current findings run contrary to the view that when an item is no longer goal-relevant it is permanently removed from memory. Even at longer temporal distance of 9/10-back, some trials still evoked interference, which has interestingly been observed at a much higher rate with older adults ( Samrani et al., 2017 ). Rather, our results suggest that there is a gradual removal of irrelevant items from WM to LTM, but that these representations are not completely removed from memory. Current results also support a dual-store account in which WM and LTM representations are assumed to be stored and processed independently ( Norris, 2017 ). Such an account would predict that the HC primarily supports LTM operations ( Talmi et al., 2005 ), which corroborate current findings of stronger HC engagement for items with longer temporal target to lure distance. Importantly, control analyzes suggest that most of the HC engagement could be attributed to the transition from proximal-to distant lures. This make sense given that LTM operations may be relatively equally involved in all distances beyond the FOA. However, the number of high interference trials for each temporal distance was small in both analyzes, in particular when including only distant lures (5-to 10-back). The current study attempted to circumvent this limitation by either pooling all proximal (3-back) and distant (5-to 10-back) trials or treating distance as a continuous variable by including all trials with distance as a parametric modulator. Thus, while the current study was limited in estimating brain involvement for specific distances occurring distance-by-distance, it is a primary step in demonstrating the involvement of LTM operations in dealing with goal-irrelevant information in WM.
Our second goal was to examine if representational strength in WM influenced behavioral PI-measures and associated brain activation. The finding that item repetition resulted in reduced interference supports the view that recollection of item-context information can be used as a means for resolving PI. This view was further supported by the finding of less activation for 3B repeat compared to 3B non-repeat lures in regions previously associated with interference control, such as IFG and ACC. This observation could reflect that less resources for resolving interference are needed for items with well-established item-context representations. It should be noted that while the number of high-interference trials was lower for repeated 3B trials compared to non-repeated 3B trials, the amount of interference on average only approached significance. Clearly, more studies with larger samples are needed to test if and to what extent strength of memory representations affects the degree of proactive interference. In particular, studies designed explicitly to test the potential mechanisms resulting in PI by manipulating the degree of contextual encoding could provide important insight into the theories of proactive interference in working memory. Nevertheless, while the previously proposed idea which states that stronger WM representations may elicit more, and not less, interference is a sensible one ( Szmalec et al., 2011 ), it was not supported by the current results. Moreover, studies that include participants with memory impairments have commonly demonstrated more, not less, interference in these individuals. For example, increased sensitivity to both proactive and retroactive interference have been found in amnestic mild cognitive impairment ( Ebert and Anderson, 2009 ;Loewenstein et al., 2004 ), Alzheimer´s disease ( Crocco et al., 2014 ), and in patients with hippocampal lesions (e.g Winocur et al., 1996 .).
While the current version of the n-back task does not allow for a separation between encoding, maintenance, and retrieval phases of WM, previous findings using the recent-probes WM task have demonstrated that the activation in these regions originate primarily from the response/retrieval phase ( D'Esposito et al., 1999 ;Postle et al., 2004 ). This further supports the view that less engagement in these brain regions is needed during recollection, when item-context information is more easily available. Although this is in agreement with proposals that participants may try to rehearse during updating to reduce interference ( Bunting et al., 2006 ), it has also been suggested that an increased time between trials in updating tasks can lead to increased rehearsal of each item, thereby increasing familiarity ( Szmalec et al., 2011 ). Thus, the current results extend previous findings implicating IFG and ACC in the resolution of interference arising from stimulus familiarity.
Our final goal was to examine differences in within-condition neural signatures for high-and low-interference trials using a trial-by-trial analysis. A substantial proportion of lure trials, which are designed to induce PI, did not result in any behavioral consequence of this manipulation, as shown by RTs being on par with non-familiar new trials. Our results demonstrate remarkable differences in how the brain responded to lure trials associated with PI compared to lure trials where RTs were similar to RTs for non-familiar new trials. Presumably, a lack of PI for familiar lure trials could result from forgetting. However, participants retained high levels of recognition performance well up to 45 min after the main experiment, indicating that a lack of PI for these familiar lure trials could not be attributed to forgetting.
There are several ways by which PI can be resolved. For example, recollection of item-context information can aid the decision to reject a lure by retaining information that the item is not in its correct temporal position. However, such processing of lure items is presumably effortful and would result in increased RTs. Therefore, it is not clear why some lure trials, while not being forgotten, do not result in PI. One possibility is that these trials are automatically inhibited and/or enters a latent state in which the item is below the threshold for familiarity until it is activated by a goal-relevant cue. This view is in line with our observations that high-interference lure trials required more prefrontal engagement than low-interference lure trials, whereas the reversed contrast showed no additional brain involvement.
Current results indicate that active representations in working memory can be temporarily removed from WM by transitioning to a "neurally silent representation " that does not evoke PI ( LaRocque et al., 2013 ;Lewis-Peacock et al., 2012 ;Rose et al., 2016 ;Sprague et al., 2016 ;Stokes, 2015 ). Possibly, these items are easily rejected because they do not involve neither recollection nor conflict related processes. However, a sub-set of distant lure trials still induced interference, suggesting that such automatic inhibition does not occur for all lure trials outside FOA. Also, the fact that representations could subsequently be activated in long-term memory by a goal-relevant cue, suggests that items in a latent state can be reactivated. This finding corroborates previous observations that latent representations can be restored by goal-relevant cuing ( LaRocque et al., 2013 ;Lewis-Peacock et al., 2012 ;Sprague et al., 2016 ;Stokes, 2015 ) or even unspecific stimulation ( Rose et al., 2016 ). Similar to proposed mechanisms for priming, previously encountered representations might be characterized by an elevated baseline in the absence of continuous and sustained activation, and that such items could cross the threshold for retrieval more quickly when a goal-relevant cue is presented.
Our single-trial analytical approach showed that the presence of interference resulted in increased activation in the parietal cortex, IFG and ACC. These regions have consistently been specifically linked to PI ( Rottschy et al., 2012 ), but also to attentional control more generally ( Cazzoli et al., 2021 ;Milham et al., 2001 ). Thus, while these regions track the behavioral signature of PI, they may not all be related to processes specifically involved in resolving PI. Lure trials that elicited PI were associated with engagement of regions previously implicated in task difficulty ( Krebs et al., 2012 ;Shigemune et al., 2017 ) and effort ( Jansma et al., 2007 ;Samrani et al., 2018 ) more generally, and control of PI ( Badre and Wagner, 2007 ;Rottschy et al., 2012 ) and memory suppression ( Anderson et al., 2016 ;Benoit et al., 2015 ;Benoit and Anderson, 2012 ) more specifically. Thus, engagement of these brain regions may modulate resource allocation to be more effective. These results indicate that response to conflict varies dramatically from trial-to-trial, where only those trials that evoke a behavioral conflict seem to engage this brain network.
The current results demonstrate an average reduction of interference over even short durations, and that this process can be explained by a reduced number of lure trials eliciting interference as a function of increased temporal lure distance. Activation and functional connectivity in right PFC, hippocampus and parietal cortex tracked the temporal distance to the lure, supporting the notion that item-context information is transferred to LTM during the long delay intervals of our task and that this information can be recalled by goal-relevant cueing. Weaker representations resulted in more effort and greater engagement in the IFG, insula and ACC, which play an important role in effortful control of PI. Trial-level analyzes additionally suggest two modes of interference control, one is effortful and involves engagement of IFG and ACC, and the other is automatic, does not implicate extra neural control, and may be a result of representations residing in a latent state. Current results contribute to the theoretical understanding of how memory representations are controlled during goal-directed behavior. In general, understanding how information is maintained, controlled and accessed is fundamental for understanding human cognition, and may have important impli-cations for disorders that affect memory functions, such as dementia, schizophrenia and depression.

Data availability statement
Data will be publicly available via Open Neuro ( https://openneuro.org/ ) upon publication.
The MR and behavioral data that support the findings of this study are openly available in "OpenNeuro " at 10.18112/openneuro.ds004037.v1.0.0.