Varying demands for cognitive control reveals shared neural processes supporting semantic and episodic memory retrieval

The categorisation of long-term memory into semantic and episodic systems has been an influential catalyst for research on human memory organisation. However, the impact of variable cognitive control demands on this classical distinction remains to be elucidated. Across two independent experiments, here we directly compare neural processes for the controlled versus automatic retrieval of semantic and episodic memory. In a multi-session functional magnetic resonance imaging experiment, we first identify a common cluster of cortical activity centred on the left inferior frontal gyrus and anterior insular cortex for the retrieval of both weakly-associated semantic and weakly-encoded episodic memory traces. In an independent large-scale individual difference study, we further reveal a common neural circuitry in which reduced functional interaction between the identified cluster and ventromedial prefrontal cortex, a default mode network hub, is linked to better performance across both memory types. Our results provide evidence for shared neural processes supporting the controlled retrieval of information from functionally distinct long-term memory systems.


Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. Note that full information on the approval of the study protocol must also be provided in the manuscript.
In total, 47 undergraduate or postgraduate students were recruited for Experiment 1 and 169 students were recruited for Experiment 2. These sample sizes were chosen as relatively large samples for fMRI research based on prior literature on related topics (Lanzoni et al. 2019;Turnbull et al. 2019).
In Experiment 1, one participant had incomplete data which was excluded from further analysis. In Experiment 2, a total of 29 participants were excluded due to missing data and excessive motion inside the scanner based on the extensive head-motion correction procedures employed in this study that were pre-defined.
There were no direct replication attempts performed in this study, as the design and sample size are unique and relatively large.
There were no between group analyses conduced in this study.
There were no between group analyses conducted in this study.
All participants were right-handed, native English speakers with normal to corrected-to-normal vision. As per the exclusion criteria, none of the participants had any prior history of psychiatric or neurological disorders, incompatibility for MRI scanning, severe claustrophobia and anticipated pregnancy or drug use that could alter cognitive functioning. Consequently, the final group for Experiment 1 consisted of 46 participants (mean = 21.31 years old, SD = 2.17, range = 18-29, 29/17 female to male ratio), whereas the final number of participants included in Experiment 2 was 140 (mean = 20.70 years old, SD = 2.37, range = 18-31, 83/57 female to male ratio), who fully completed all the required neuroimaging-based and behavioural assessments.
All participants for both experiments were recruited from the undergraduate and postgraduate population at the University of York via online and poster advertisements. All volunteers received monetary compensation or course credit for their participation in line with the departmental policies. There may be a selection bias given that participants who agreed to take part in our experiments are generally young adults and are enrolled in higher education. Although, we do not expect this to have a significant influence on our results, future studies with more representative samples of the population will be required to assess the generalisability of our findings.
For both Experiment 1 and 2, ethical approval was obtained from the Department of Psychology and York Neuroimaging Centre, University of York ethics committees. All participants were briefed about the aims and objectives of the experiments before providing informed consent to take part in this study.

Statistical modeling & inference
Model type and settings Experiment 1 included two task-based, event-related fMRI studies that were repeated within same participants. Experiment 2 was an individual difference study that employed resting state fMRI.
In Experiment 1, semantic and episodic memory retrieval was assessed using 3-alternative forced-choice (3-AFC) paradigms. Both tasks consisted of 40 strong association, 40 weak association and 20 control trials. Following the presentation of the probe, target and 2 distractors, participants were given 4 seconds to respond, after which a fixation cross was displayed, jittered in duration between 1.5 to 3.5 seconds in 500 ms intervals.
For all task-based paradigms, both accuracy and response times were recorded. The mean inverse efficiency scores (i.e. reaction time divided by the percentage of incorrect responses) were calculated for each participant, the mean and standard deviation of which were used as indicators of compliant performance. The field of view employed in the MRI scans covered the whole brain.
FSL (Version 5.0.11) was used to pre-process the task-based MRI data. SPM (Version 12.0) and CONN Toolbox (Version 17.f) was used for the resting state MRI data.
Task-based (Experiment 1): After coregistration to the structural images, individual functional images were linearly registered to the MNI-152 template using FMRIB's Linear Image Registration Tool (FLIRT). Resting state (Experiment 2): Structural images were coregistered to the mean functional image via rigid-body transformation, segmented into grey/white matter and cerebrospinal fluid probability maps, and images were spatially normalized to the MNI-152 template using the unified segmentation-normalization procedure.

MNI-152.
Task-based (Experiment 1): The individual subject analysis first involved motion correction using MCFLIRT and slice-timing correction using Fourier space time-series phase-shifting. Functional images were spatially smoothed using a Gaussian kernel of FWHM 5mm, underwent grand-mean intensity normalization of the entire 4D dataset by a single multiplicative factor, and both high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=100s) and Gaussian low-pass temporal filtering, with sigma=2.8s. For all general linear models (GLMs), standard motion parameters (3 rotations and translations), their temporal derivatives and squared versions were added as potential motion confounds. No significant difference in head-motion was observed between the two task conditions. Resting state (Experiment 2): An extensive set of motion-correction and denoising procedures were employed, comparable to those reported in the literature (Ciric et al. 2017). In addition to the removal of six realignment parameters and their second-order derivatives using a GLM, a linear detrending term was applied as well as the CompCor method that removed five principal components of the signal from white matter (WM) and cerebrospinal fluid (CSF). The composite motion score (i.e. percentage of invalid scans) for each participant was also added as a covariate in group-level analyses to further account for the potential influence of head motion on functional connectivity estimates. No significant correlation was observed between in-scanner head motion and covariates of interest utilised in subsequent analyses.
Task-based (Experiment 1): Motion outliers were identified using the DVARS method and included in the model as a covariate of no-interest. Resting-state (Experiment 2): The functional volumes influenced by excessive head motion were identified and scrubbed based on the conservative settings of motion greater than 0.5 mm and global signal change larger than z = 3. Participants who had more than 15% of their data affected by motion were excluded from further analysis.
Task-based (Experiment 1): The pre-processed task fMRI data was modelled using mass univariate general linear models (GLM) with the onsets and durations (4s) of all trial types (i.e. strong, weak, control) included at the subject-level. Group-level statistical contrasts of strong > weak, weak > strong and task (all trials) > control was assessed using a mixed-effects approach