Age Differences in Brain Functional Connectivity Underlying Proactive Interference in Working Memory

ABSTRACT

Aging is typically accompanied by a decline in working memory (WM) capacity, even in the absence of pathology. Proficient WM requires cognitive control processes that can retain goal‐relevant information for easy retrieval and resolve interference from irrelevant information. Aging has been associated with a reduced ability to resolve proactive interference (PI) in WM, leading to impaired retrieval of goal‐relevant information. It remains unclear how age‐related differences in the ability to resolve PI in WM are related to patterns of resting‐state functional connectivity (rsFC) in the brain. Here, we investigated the association between PI in WM and rsFC cross‐sectionally (n = 237) and 5 years longitudinally (n = 134) across the adult life span by employing both seed‐based and data‐driven approaches. Results revealed that the ability to resolve PI was associated with differential patterns of inferior frontal gyrus (IFG) rsFC in younger/middle‐aged adults (25–60 years) and older adults (65–80 years) in two clusters centered in the vermis and caudate. Specifically, more PI was associated with stronger inferior frontal gyrus—vermis connectivity and weaker inferior frontal gyrus—caudate connectivity in older adults, while younger/middle‐aged adults showed associations in the opposite directions with the identified clusters. Longitudinal analyses revealed that a reduced ability to control PI was associated with reduced inferior frontal gyrus—insula and inferior frontal gyrus—anterior cingulate cortex connectivity in older adults, while younger/middle‐aged adults showed associations in the opposite direction with these clusters. Whole brain multivariate pattern analyses showed age‐differential patterns of rsFC indicative of age‐related structural decline and age‐related compensation. The current results show that rsFC is associated with the ability to control PI in WM and that these associations are modulated by age.

Results demonstrate converging evidence of age‐differential patterns of IFG and whole brain rsFC associated with PI in working memory, with several regions previously implicated in the ability to control PI in working memory showing differential rsFC patterns in older and younger adults, both cross‐sectionally and longitudinally.

1. Introduction

Getting older is commonly accompanied by increasing memory difficulties (Koen and Yonelinas 2014; Nyberg et al. 2020), even in the absence of any memory pathology such as dementia. An efficient memory system entails the ability to control the flow of information that we are almost constantly processing, to determine what information should be retained. Declining control processes might result in cognitive overload and impede goal‐directed behavior that relies on the ability to selectively engage relevant information. Crucially, irrelevant information might interfere with goal‐relevant information. Proactive interference (PI) occurs when old, outdated information interferes with memory for new information, leading to forgetting (Hasher and Zacks 1988; Jonides and Nee 2006).

While PI has traditionally been studied primarily in the context of long‐term episodic memory (Keppel 1968; Roediger and McDermott 1995; Kliegl and Bäuml 2021), evidence supports its presence also in working memory (WM; Bunting 2006; Emery et al. 2008). PI and the inability to appropriately control PI have been considered a major sources of forgetting, not only in long‐term memory (Kliegl and Bäuml 2021) but also in WM (Oberauer and Lewandowsky 2008). Resolving PI in WM has been consistently localized to a network of brain regions consisting of the left inferior frontal gyrus (IFG), the anterior cingulate cortex, and the striatum/insula (Badre and Wagner 2005; Burgess and Braver 2010; Jonides and Nee 2006; Marklund and Persson 2012; Nelson et al. 2009; Persson et al. 2013; Samrani et al. 2019).

An age‐related decrease in the ability to control PI has been found in both long‐term memory (Healey et al. 2013; Wahlheim 2014) and WM (Loosli et al. 2016; Samrani and Persson 2021). Furthermore, age‐related differences in cognition, such as episodic memory, are largely explained by PI in WM, over and above the effects of processing speed (Samrani and Persson 2021).

Increasing age is characterized by changes in both the structure and function of the brain. Specifically, aging is associated with a decrease in gray matter volume (Fjell et al. 2014) as well as reduced white‐matter integrity (Sexton et al. 2014). These structural changes have been postulated to underlie age‐related functional changes in the brain (Greenwood 2007; Park and Reuter‐Lorenz 2009). Functionally, increasing age is associated with neural dedifferentiation, which is identified by less selective neural processing and might explain reduced neural efficiency with age (Koen and Rugg 2019; for review: Goh 2011). Moreover, older adults tend to have decreased activity in posterior regions and increased activity in frontal regions, compared to younger/middle‐aged adults (Davis et al. 2008). It has been suggested that this pattern may be associated with less efficient functional communication, reflected by reduced resting‐state functional connectivity (rsFC) between networks (Goh 2011). Older adults also display less functional connectivity with posterior regions and more connectivity with frontal regions compared to younger/middle‐aged adults (Zhang et al. 2017). An age‐related increase in frontal activity together with the recruitment of additional regions or bilateral activation has been associated with maintained cognitive performance including WM (Goh 2011). This pattern is consistent with the Compensation‐Related Utilization of Neural Circuits (CRUNCH) model (Reuter‐Lorenz and Cappell 2008), which proposes that older adults compensate for deficient neural resources through additional functional recruitment. Moreover, weaker rsFC in older adults has been associated with poorer executive functioning and processing speed (Damoiseaux et al. 2008) and associative memory (Wang et al. 2010).

The structural and functional neurological changes that occur with aging may underlie age‐related decline in the ability to control PI. Specifically, more PI in older adults has been associated with smaller gray matter volume in the IFG (Samrani et al. 2019) and the hippocampus (HC; Andersson et al. 2023), lower white‐matter integrity (Andersson et al. 2022), altered task‐related BOLD activation (Loosli et al. 2016), and altered functional connectivity (Oren et al. 2017). Oren et al. (2017) examined functional connectivity synchronization using inter‐subject correlation in younger and older adults and its relationship with proactive interference. They found that whereas greater connectivity between the left IFG and the hippocampi predicted PI in all participants, PI was more related to inter‐subject correlation in the posterior cingulate cortex (PCC) in older adults. Importantly, though, this study did not employ a WM task to investigate PI but studied PI as context effects of successive movies on a linguistic task.

A few other studies have investigated functional connectivity in relation to control of PI, though without considering age‐related differences. For example, Samrani and Persson (2022) found that a longer interval between target and lure items in an n‐back task, which results in lower PI, was associated with increased connectivity between HC and IFG, insula, anterior cingulate cortex (ACC), thalamus, putamen, and superior temporal regions. Using the recent probes and directed‐forgetting tasks, Nee et al. (2007) found significant connectivity between IFG and the premotor cortex, medial temporal cortex, anterior cingulate cortex, inferior temporal pole, PCC, and caudate, and between the anterior prefrontal cortex and the anterior cingulate cortex in response to PI. Collectively, these results highlight the IFG, a region suggested to support PI resolution (Badre and Wagner 2005; Burgess and Braver 2010; Jonides and Nee 2006; Marklund and Persson 2012; Nee et al. 2007; Nelson et al. 2009; Persson et al. 2013; Samrani et al. 2019; Samrani and Persson 2024; Öztekin et al. 2008) as an important node in the functional connectivity network associated with controlling PI. Despite age‐related differences in rsFC, there are, to our knowledge, no previous studies investigating rsFC patterns associated with PI in WM in the context of aging. Such investigations can provide novel insights into the intrinsic organization of brain networks underlying the ability to control PI that may not be evident in task‐based or structural imaging and could thus improve our understanding of network alterations underlying age‐related decline in WM and, more specifically, PI control.

In this study, we investigated rsFC patterns associated with PI in a large (N_baseline = 237) population‐based, longitudinal (5 years) dataset using an interference version of the n‐back WM task. First, we investigated the rsFC between an established node of the brain network associated with controlling PI in WM, the IFG and the rest of the brain, and how this connectivity is related to PI, across the whole sample as well as dependent on age. Functional (Nee, Wager, et al., 2007) and structural (Samrani et al. 2019) brain imaging studies implicate that both right and left IFG might play a role in interference resolution tasks, and we therefore used bilateral IFG in the seed‐based analyses. Results using the left and right IFG as seed regions are reported in Supporting Information. Secondly, using data‐driven multivariate pattern analysis, we explored age‐related differences in whole brain connectivity patterns associated with PI, separately for individuals scoring low and high on PI. These approaches can provide unique and novel insights into the network dynamics associated with PI, as well as age‐related ability to appropriately control PI.

Based on previous findings, we expected to find (1) altered patterns of IFG rsFC in relation to increasing age; (2) alterations in whole brain rsFC patterns in relation to the ability to control PI in WM; and (3) weaker IFG connectivity with other regions within the PI network would be associated with a lower ability to control PI in older adults.

2. Materials and Methods

2.1. Participants

Participants were recruited from The Betula prospective cohort study: Memory, health, and aging (Nilsson et al. 1997; Nyberg et al. 2020), a deeply phenotyped longitudinal cohort. Participants were included from samples for which MRI measures were collected in 2008–2010 (timepoint 5, baseline) and 2013–2014 (timepoint 6, follow‐up). The Betula study was approved by the Regional Ethical Review Board in Umeå (dnr: 2008‐08‐132 and dnr: 2013‐92‐31M), and written consent was obtained from every participant.

Individuals with clinical dementia and other neurological disorders at baseline or follow‐up were excluded from the current analyses. Dementia status was assessed at baseline and reassessed every 5 years using a three‐step procedure. First, an overall evaluation was performed by an examining physician according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM–IV; American Psychiatric Association 1994). Second, using a composite measure based on scores from several cognitive tests (episodic memory, WM, processing speed, semantic memory, and fluid intelligence), each participant was compared to the mean cognitive score for their age cohort. If an individual scored more than two standard deviations below the mean of their age cohort, they would be flagged for further assessment of dementia by a clinical psychiatrist. Third, all participants scored at or above the cut‐off score for dementia of 24 using the MMSE (Mini Mental State Examination, Folstein et al. 1975). To retain the diversity of the sample, exclusions were not made for diabetes, hypertension, mild depressive symptoms, and other moderately severe medical conditions, which are common in older participants.

Participants with extremely low performance on the n‐back task (proportion hits minus proportion false alarms < 0.1), indicating a very low adherence to task instructions, were excluded (17 participants). The final total sample consisted of 237 participants for cross‐sectional analyses at baseline (25–80 years, M = 58.9, SD = 13.5, M education = 13 years (SD = 4 years); Supplementary Figure 1), and 134 participants for longitudinal analyses at follow‐up. For a drop‐out analysis, see Noroozian et al. (2023). Demographic information and cognitive performance can be found in Table 1.

TABLE 1.

Sample demographical and cognitive information.

	Baseline			Follow‐up
	Younger	Older	P	Younger	Older	P
N	131	106		82	52
Age (range)	49.7 (25–60)	70.2 (65–80)		54.5 (30–65)	74.1 (70–85)
Sex (F/M)	67/64	54/52	0.98	35/47	28/24	0.220
Education, in years	14.5 (3)	12.4 (4.4)	< 0.001	14.6 (3.2)	13.2 (4.5)	0.065
MMSE	28.6 (1.3)	28.3 (1.3)	0.126	28.4 (1.2)	28.2 (1.5)	0.44
Blood pressure, systolic	128.6 (16.1)	142.9 (14.6)	< 0.001	134 (14.2)	144.1 (20)	0.002
Blood pressure, diastolic	79.3 (10)	81.6 (8.4)	0.059	81.2 (7.7)	82.3 (8.3)	0.42
Combined PI score T5	−0.24 (0.76)	0.23 (0.77)	< 0.001	−0.22 (0.82)	0.26 (0.9)	0.002

Note: Mean (std); MMSE = mini mental state examination; PI = proactive interference.

2.2. Cognitive Measures

PI in WM was measured using a modified version of the verbal 2‐back task, which was designed to induce proactive interference (Gray et al. 2003; Marklund and Persson 2012; Nee et al. 2007). Participants were presented with a series of Swedish nouns, one after the other, and instructed to indicate for every noun whether it matches the one presented two trials prior. On target trials, the current stimulus matches the one presented two trials earlier (requiring a ‘Yes’ response); on new trials, the current stimulus has not been presented in previous trials, and so is non‐familiar (requiring a ‘No’ response); and on lure trials, the current stimulus matches a stimulus presented three or four trials prior, and so is familiar (requiring a ‘No’ response). The task consisted of 40 trials (9 target trials, 21 new trials, 8 3‐back lure trials, and 2 4‐back lure trials). All stimuli and trial conditions were presented in the same fixed random order to all participants. Stimuli were presented for 2500 ms (ITI = 2000 ms). Participants were instructed to respond as quickly and accurately as possible by pressing the “m” key for “yes” and the “x” key for “no,” on a standard Swedish qwerty keyboard, using their right and left index fingers, respectively.

RT calculations were based on correct responses (in milliseconds). To reduce the influence of extreme values, median RTs were used. Accuracy was estimated as a percentage of correct responses out of the total number of trials, excluding omissions.

PI scores were calculated by combining the relative proportional difference in RT and accuracy between new trials (requiring ‘No’ responses to non‐familiar stimuli) and lure trials (requiring ‘No’ responses to familiar stimuli). First, a relative difference score was calculated for RT and accuracy scores separately using the formula: ((/ −1) for RT scores and the formula: new trials (%) – lure trials (%) for accuracy scores. Using a relative difference score should provide a more salient and individual measure of executive control, taking into consideration the individual, age‐related differences in the variable. Relative difference scores based on RT and accuracy were positively correlated both at baseline, r(237) = 0.299, p = < 0.001, and follow‐up, r(132) = 0.330, p = < 0.001. Second, the two scores were normalized with z‐score transformation. Finally, the scores were combined into a PI score by calculating the average of the two difference scores. Higher PI scores indicate a lower ability to control PI.

2.3. Procedure

Scanning was performed at the MR‐center (Umeå Center for Functional Brain Imaging; UFBI) at the University hospital in Umeå, and the cognitive (2 h) and medical (2 h) test sessions were performed on two separate days at the Psychology department at Umeå University. The cognitive test session was performed before the scanning session for all participants. Informed consent was obtained before the cognitive test session from all participants.

2.4. MRI Data Acquisition

MRI data were collected using a 3 T GE scanner (32‐channel head coil). T1‐weighted images were acquired with a 3D fast spoiled gradient echo sequence (180 slices; thickness = 1 mm; TR = 8.2 ms; TE = 3.2 ms; 12° flip angle; FOV = 25 x 25 cm). Resting‐state data were acquired with a gradient echo‐planar imaging sequence (37 transaxial slices; thickness = 3.4 mm; gap = 0.5 mm; TR = 2 s; TE = 30 ms; 80° flip angle; FOV = 25 x 25 cm). The first 10 scans were excluded from the experimental image acquisition as dummy scans. The same scanner was used for baseline and follow‐up.

2.5. Scanner Stability

Scanner software changes were implemented from baseline to follow‐up. To check whether these changes affected image quality, a quality assurance program (based on Friedman and Glover 2006) was run weekly, confirming scanner stability.

2.6. Statistical Analyses

Based on previous demonstrations of age‐differential relationships between brain function and structure in younger and older adults (Burzynska et al. 2012; Koen and Rugg 2019; Rieckmann et al. 2018; Van Petten 2004), whole sample analyses were complemented with age‐stratified analyses. Participants were divided into two age groups based on their age at baseline, with 65 years as the cut‐off age: one younger/middle‐aged group (N = 131, 25–60 years) and one older group (N = 106, 65–80 years).

Based on previous demonstrations of differential relationships between brain activity and cognitive performance in older adults with preserved cognitive performance and those exhibiting cognitive decline (Damoiseaux et al. 2008; Goh 2011; Wang et al. 2010), PI groups were constructed based on a median split of the PI scores (median = −0.044) across age groups. The split resulted in a low PI group consisting of 119 individuals (39 older, 80 younger/middle‐aged) displaying relatively higher performance on the n‐back task indicated by less PI, and a high PI group consisting of 118 individuals (67 older, 51 younger/middle‐aged) displaying relatively lower performance on the n‐back task indicated by more PI (Supplementary Table 1). The PI groups were used in the multivariate pattern analyses (see Section 2.7.2) to investigate age‐group‐related differences in rsFC patterns within the low and high PI groups.

Whole sample and age‐stratified partial correlation analyses were conducted to investigate the relationship between age and PI at baseline and between age and change in PI from baseline to follow‐up, controlling for sex (F/M) and education level (in years). Change scores were calculated by dividing follow‐up PI scores by baseline PI scores.

A 2 (age group) × 2 (PI group) ANOVA was performed to evaluate the main effects of age group and PI group and the age group × PI group interaction effect on PI scores at baseline. Post hoc t‐tests were performed to compare PI scores at baseline between age groups, within as well as between PI groups.

2.7. MRI Preprocessing and Analysis

Data were preprocessed and analyzed using CONN (Whitfield‐Gabrieli and Nieto‐Castanon 2012; RRID:SCR_009550) release 22.a (Nieto‐Castanon and Whitfield‐Gabrieli 2022) and SPM (Penny et al. 2011) (RRID:SCR_007037) release 12.7771. All reported coordinates are in Montreal Neurological Institute (MNI) space.

Functional and anatomical data were preprocessed using CONN's default preprocessing pipeline (Nieto‐Castanon 2020a), including realignment with correction of susceptibility distortion interactions, slice‐timing correction, outlier detection, direct segmentation and MNI space normalization, and smoothing. Functional data were realigned using the SPM realign & unwarp procedure (Andersson et al. 2001), where all scans were co‐registered to a reference image (first scan of the first session) using a least squares approach and a 6‐parameter (rigid body) transformation (Friston et al. 1995), and resampled using b‐spline interpolation to correct for motion and magnetic susceptibility interactions. Temporal misalignment between different slices of the functional data (acquired in interleaved bottom‐up order) was corrected following the SPM slice‐timing correction (STC) procedure (Henson et al. 1999; Sladky et al. 2011), using sinc temporal interpolation to resample each slice BOLD time series to a common mid‐acquisition time. Potential outlier scans were identified using ART (Whitfield‐Gabrieli et al. 2011) as acquisitions with framewise displacement above 0.9 mm or global BOLD signal changes above five standard deviations (Nieto‐Castanon, submitted; Power et al. 2014). A reference BOLD image was computed for each subject by averaging all scans, excluding outliers. Functional and anatomical data were normalized into standard MNI space, segmented into gray matter, white matter, and CSF tissue classes, and resampled to 2 mm isotropic voxels following a direct normalization procedure (Calhoun et al. 2017; Nieto‐Castanon, submitted) using the SPM unified segmentation and normalization algorithm (Ashburner and Friston 2005; Ashburner 2007) with the default IXI‐549 tissue probability map template. Last, functional data were smoothed using spatial convolution with a Gaussian kernel of 8 mm full width half maximum (FWHM).

Functional data were denoized using a standard denoising pipeline (Nieto‐Castanon 2020a) including the regression of potential confounding effects characterized by white matter timeseries (5 CompCor noise components), CSF timeseries (5 CompCor noise components), motion parameters and their first‐order derivatives (12 factors; Friston et al. 1996), outlier scans (below 84 factors; Power et al. 2014), and linear trends (2 factors) within each functional run, followed by bandpass frequency filtering of the BOLD timeseries (Hallquist et al. 2013) between 0.008 Hz and 0.09 Hz. CompCor (Behzadi et al. 2007; Chai et al. 2012) noise components within white matter and CSF were estimated by computing the average BOLD signal as well as the largest principal components orthogonal to the BOLD average, motion parameters, and outlier scans within each subject's eroded segmentation masks. From the number of noise terms included in this denoising strategy, the effective degrees of freedom of the BOLD signal after denoising were estimated to range from 20.3 to 95.8 (average 77.1) across all participants (Nieto‐Castanon, submitted).

2.7.1. Seed‐Based Analyses

Seed‐based connectivity (SBC) maps were estimated characterizing the spatial pattern of functional connectivity with the IFG as the seed area. Functional connectivity strength was represented by Fisher‐transformed bivariate correlation coefficients from a weighted general linear model (weighted‐GLM; Nieto‐Castanon 2020b), estimated separately for the seed area and each target voxel, modeling the association between their BOLD signal timeseries.

The IFG was selected as a region of interest (ROI) in the seed‐based analyses based on previous evidence suggesting this region may support cognitive control processes involved in regulating PI (Badre and Wagner 2005; Burgess and Braver 2010; Jonides and Nee 2006; Marklund and Persson 2012; Nee, Wager, and Jonides, 2007; Nelson et al. 2009; Persson et al. 2013; Samrani et al. 2019; Samrani and Persson 2024; Öztekin et al. 2008). The IFG was anatomically defined by including AAL subregions pars opercularis, pars triangularis, and pars orbitalis. To reduce the number of comparisons, and because there were no hypotheses regarding laterality or subregional specificity, a binary mask of the IFG was created using the WFU Pickatlas toolbox (AAL atlas; RRID:SCR_007378), combining the left and right hemispheres.

Seed‐based connectivity correlational analyses were conducted to investigate the main effect of PI on IFG rsFC at baseline with the between‐subject contrast (all_participants, PI_baseline [0 1]).

Age‐group differences in PI‐related IFG rsFC at baseline were examined with a between‐subject contrast of (older, younger/middle‐aged, older × PI_baseline>younger/middle‐aged × PI_baseline [0 0 1–1]). Analyses of whether a change in PI scores over 5 years (baseline—follow‐up) was associated with a change in IFG rsFC were examined with a within‐subject contrast of (rsFC baseline<rsFC follow‐up [−1 1]). Age‐group differences in the relationship between change in PI and change in IFG rsFC over 5 years (from baseline to follow‐up) were examined with a between‐subject contrast of (older, younger/middle‐aged, older × PI_change>younger/middle‐aged × PI_change [0 0 1–1]. The main effect of age on IFG connectivity was examined cross‐sectionally (older, younger [1–1]) and longitudinally (baseline, follow‐up [−1 1]). A voxel‐wise threshold of p = 0.05 and a cluster‐level threshold (FDR‐corrected) of p = 0.05 were used to identify significant clusters. Participant‐specific mean cluster connectivity values from clusters identified in the seed‐based analyses were extracted and imported to SPSS (v.26; IBM, Armonk, NY, USA) for further visualization and analysis.

2.7.2. Multivariate Pattern Analyses (MVPA)

Functional connectivity multivariate pattern analyses (fc‐MVPA; Nieto‐Castanon 2022) were performed to estimate the first 12 eigenpatterns characterizing the principal axes of heterogeneity in functional connectivity across participants and time points. From these eigenpatterns, 12 associated eigenpattern score images were derived for each individual subject and condition characterizing their brain‐wide functional connectome state.

Eigenpatterns and eigenpattern‐scores were computed separately for each individual seed voxel as the left‐ and right singular vectors, respectively, from a singular value decomposition (group‐level SVD) of the matrix of functional connectivity values between this seed voxel and the rest of the brain (a matrix with one row per target voxel, and one column per subject and condition). Individual functional connectivity values were computed from the matrices of bivariate correlation coefficients between the BOLD timeseries from each pair of voxels, estimated using a singular value decomposition of the z‐score normalized BOLD signal (subject‐level SVD) with 64 components separately for each subject and condition (Whitfield‐Gabrieli and Nieto‐Castanon 2012).

Group‐level analyses were performed using a general linear model (GLM; Nieto‐Castanon 2020c). For each individual voxel, a separate GLM was estimated, with first‐level connectivity measures at this voxel as dependent variables (one independent sample per subject and one measurement per time point), and PI group (2: Low PI, High PI) × Age group (2: Older, Younger/middle‐aged) as independent variables. Voxel‐level hypotheses were evaluated using multivariate parametric statistics with random effects across participants and sample covariance estimation across multiple measurements. Inferences were performed at the level of individual clusters. Cluster‐level inferences were based on parametric statistics from Gaussian Random Field theory (Worsley et al. 1996; Nieto‐Castanon 2020d). Results were thresholded using a combination of a cluster‐forming p < 0.001 voxel‐level threshold and a familywise corrected p‐FDR < 0.001 cluster‐size threshold (Chumbley et al. 2010).

We used MVPA with 12 eigenpatterns to identify differences in brain‐wide functional connectivity patterns using the between‐participants contrast of the younger/middle‐aged vs. older group (−1 1) within the low PI and high PI groups, separately, with baseline PI score as a covariate. Additionally, an MVPA analysis with six eigenpatterns was conducted to identify differences in functional connectivity patterns using the between‐participant contrast of high PI vs. low PI older adults (−1 1). The eigenpattern maps were analyzed simultaneously using an omnibus/F‐test. Thus, three MVPA models were conducted in total. The number of eigenpatterns used in the respective analyses was chosen to correspond to an approximate 1:10 eigenpattern‐to‐participant ratio, which is considered a relatively conservative approach (Nieto‐Castanon 2022). A voxel‐wise threshold of p = 0.001 and a cluster‐level threshold (FDR‐corrected) of p = 0.001 were used to identify significant clusters.

2.7.2.1. Post Hoc Analyses

As MVPA is an omnibus test and, therefore, does not inform on the nature of identified differences in connectivity patterns, we conducted post hoc seed‐based analyses on any identified clusters from the MVPA analyses. These seed‐based analyses were conducted to investigate differences in connectivity patterns between age groups. Individual binary masks of each MVPA cluster were imported as ROIs and used in post hoc seed‐based analyses with between‐subject contrast PI High_Old, PI High_Young/Middle‐aged (1–1) and PI Low_Old, PI Low_Young/Middle‐aged (1–1), respectively (voxel p = 0.001 and cluster level p ^FDR = 0.001).

To summarize the results across analyses within the low PI and high PI groups, t‐statistics maps were exported from individual seed‐based analyses and processed in SPM's ImCalc to generate overlay maps where each voxel indicated the number of analyses showing significant connectivity at that location. Separate overlay maps were created for positive (older>younger) and negative (older<younger) effects and for the low PI and high PI analyses. First, each t‐statistic map was processed to extract binary maps of the positive (i1 > 0) and negative (i1 < 0) effects separately. Then, all positive and negative maps were combined within the respective PI group analysis (i1 + i2 + i3 … iX). The mean MNI coordinates for the cluster regions are reported in the results section.

3. Results

3.1. Age‐Related Differences and 5‐Year Change in PI

Age was positively correlated with PI scores across age groups (r(233) = 0.339 p < 0.001) at baseline, indicating that increasing age is associated with more PI. The younger/middle‐aged group showed a significant correlation between age and PI scores at baseline (r(127) = 0.297, p = < 0.001), but not the older group (r(102) = 0.086, p = 0.384). Age was not significantly correlated with change in PI from baseline to follow‐up (r(130) = 0.026, p = 0.77), in neither younger/middle‐aged (r(78) = 0.022, p = 0.85) nor older (r(48) = −0.133, p = 0.36) adults. PI did not change significantly between baseline and follow‐up in the whole sample (t(133) = 0.46, p = 0.65) or within either age group (younger/middle‐aged: t(81) = 0.33, p = 0.74; older: t(51) = 0.32, p = 0.76).

3.2. Age‐Related Differences in PI Between and Within the High and Low PI Groups

A 2 (age group) × 2 (PI group) ANOVA was performed to evaluate the effects of age group and PI group on PI scores at baseline. The results indicated a significant main effect of age group (F(1, 236) =6.38, p = 0.012, partial η² = 0.03) suggesting higher PI scores in the older group (M = 0.23, std. = 0.77) than in the younger/middle‐aged group (M = −0.24, std. = 0.78); a significant main effect for PI group (F(1,236) = 395.33, p < 0.001, partial η² = 0.63) suggesting higher PI scores in the high PI group (M = 0.62, std. =0.52) than in the low PI group (M = –0.67, std. = 0.42); but no interaction between age group and PI group (F(1, 236) = 0.06, p = 0.81, partial η² = 0.0). Note that since PI scores at baseline were used to define the PI groups, the main effect of PI group was expected and therefore negligible. The means and standard deviations for baseline PI scores are presented in Table 2.

TABLE 2.

Descriptive statistics for baseline PI scores.

Age group	PI group	M	SD	N
Younger/middle‐aged	Low PI	−0.73	0.43	80
	High PI	0.54	0.52	51
Older	Low PI	−0.56	0.4	39
	High PI	0.69	0.52	67

The age effect reached statistical significance in the low PI group (t(119) = −2.3, p = 0.023) but not in the high PI group (t(117) = −1.58, p = 0.117). The PI effect was significant both in younger to middle‐aged adults (t(131) = −15.48, p < 0.001) and older adults (t(105) = −13, p < 0.001).

3.3. Age‐Related Differences in IFG rsFC

Whole brain rsFC analyses with the IFG as a seed region and age group as an independent variable revealed seven clusters showing significant connectivity differences between the age groups (Supplementary table 2). Results from analyses on the left IFG are reported in supplementary table 3. Cluster 1 was centered in the left postcentral gyrus (x, y, z = −36 −20 + 40); cluster 2 was centered in the right parahippocampal gyrus (x, y, z = +18–26 –16); cluster 3 was centered in the left cuneal cortex (x, y, z = −8 −74 + 28); cluster 4 was centered in the right precuneus (x, y, z = +24–52 + 26); cluster 5 was centered in the right superior parietal lobule (x, y, z = +50–42 + 60); cluster 6 was centered in the left supramarginal gyrus (x, y, z = −62 –32 + 40); and cluster 7 was centered in the right angular gyrus (x, y, z = +44–58 + 40). Clusters 1, 3, and 7 showed stronger connectivity, whereas clusters 2, 4, 5, and 6 showed weaker connectivity in the older group compared to the younger/middle‐aged group. Analyses of age‐related differences in change in IFG connectivity between baseline and follow‐up did not reveal any significant differences between the age groups.

3.4. PI Is Differentially Associated With IFG rsFC in Younger/Middle‐Aged and Older Adults

Whole brain rsFC analyses across all participants revealed that higher PI scores were associated with stronger connectivity between the IFG and brain regions within a large cluster centered around the left inferior occipital cortex (x, y, z = −32 –80 –10). Full cluster details are reported in Supplementary table 4. Results from analyses on the left and right IFG are reported in Supplementary Tables 5 and 6, respectively.

PI was differentially associated with IFG connectivity in younger/middle‐aged and older adults in two large clusters (Supplementary table 7 and Figure 1). The first cluster was centered in vermis (x, y, z = +04–54 –24) and expanded across parts of the cerebellum and the brain stem. The second cluster was centered in the left caudate (x, y, z = −04 + 14 + 10) and expanded across the bilateral caudate and the subcallosal cortex. In the vermis cluster, more PI was associated with stronger connectivity in older adults but with weaker connectivity in younger/middle‐aged adults (Figure 1). In the left caudate cluster, the reverse pattern was observed, with more PI being related to weaker connectivity in older adults but stronger connectivity in younger/middle‐aged adults (Figure 1). Results from analyses on the left and right IFG are reported in supplementary Tables 8 and 9, respectively.

FIGURE 1.

Results from the analyses on age‐group differences in PI‐related IFG rsFC at baseline. (A) IFG seed region (to the left) and effect sizes for significant cluster (to the right), (B) Heat maps for the older>Younger/middle‐aged contrast (to the left) and scatter plots depicting PI score x connectivity value interaction (to the right) in the vermis (top) and left caudate (bottom) clusters. Younger/middle‐aged adults are represented in light gray and older adults represented in dark green. Note: Scatter plots depict the mean connectivity value for each cluster for each participant.

Analyses of whether the 5‐year change in PI was associated with the 5‐year change in IFG rsFC revealed a large (5833 voxels) cluster centered in the left inferior occipital cortex (MNI: –36–80–10) spanning several occipital regions (e.g., occipital pole, lingual gyrus, and lateral occipital cortex; Supplementary Table 10 and Figure 2). Results from analyses on the left and right IFG are reported in supplementary Tables 11 and 12, respectively. Increased PI was associated with decreased connectivity between the IFG and the inferior occipital cortex cluster across age groups. Age‐group differences in the association between change in PI and change in IFG rsFC were observed in two clusters centered in the right insula (2621 voxels; MNI: +30+22+12) and the right superior anterior cingulate cortex (ACC; 2008 voxels; MNI: +12+30+24) with older adults showing decreased rsFC between the IFG and these clusters with increased PI and younger/middle‐aged adults showing increased rsFC with these clusters with increased PI (Supplementary Table 13 and Figure 2). Results from analyses on the left and right IFG are reported in Supplementary Tables 14 and 15, respectively.

FIGURE 2.

(A) Results from Change (PI) – Change (IFG rsFC) analyses for the whole sample, including heat maps (left) and scatter plots (right) depicting PI scores x connectivity values. (B) Results from Change (PI) – Change (IFG rsFC) analyses of age‐group differences, including heat maps for the Older>Younger/middle‐aged contrast (left) and scatter plots depicting PI score x connectivity value interaction (right). Younger/middle‐aged adults are represented in light gray and older adults are represented in dark green. Note: Scatter plots depict the mean connectivity value for each cluster for each participant.

3.5. Age‐Related Differences in Whole Brain rsFC Patterns Associated With PI

The MVPA analysis examining age‐group differences in rsFC patterns within the high PI group revealed seven significant clusters (p = 0.001, cluster‐p ^FDR = 0.001) where older and younger/middle‐aged adults displayed significantly different rsFC patterns. These clusters were centered in the left superior temporal pole, right superior temporal pole, left cerebellum crus1, right parahippocampal gyrus, right middle temporal pole, right insula, and right posterior cingulate, respectively. Full cluster details are reported in Table 3.

TABLE 3.

Cluster seeds from the high PI MVPA.

	Coordinates	Voxels	p‐FDR	Peak region
Cluster 1	+40 + 8–20	289	< 0.001	Sup Temporal Pole R
Cluster 2	−42 + 4–16	269	< 0.001	Sup Temporal Pole L
Cluster 3	−44 –80 –36	156	< 0.001	Cerebelum Crus1 L
Cluster 4	+16–32 –12	142	< 0.001	ParaHC R
Cluster 5	+58 + 8–14	127	< 0.001	Mid‐Temporal Pole R
Cluster 6	+40–6 + 0	121	< 0.001	Insula R
Cluster 7	+4–44 + 28	112	< 0.001	Cingulum Post R

The MVPA analysis conducted in the low PI group revealed five significant clusters (p = 0.001, cluster‐FDRp = 0.001) where older and younger/middle‐aged adults displayed significantly different rsFC patterns. These clusters were centered in the left middle frontal gyrus, left postcentral gyrus, right thalamus, left posterior cingulate, and left precentral gyrus, respectively. Full cluster details are reported in Table 4.

TABLE 4.

Cluster seeds from the low PI MVPA.

	Coordinates	Voxels	p‐FDR	Peak region
Cluster 1	−26 + 12 + 42	284	< 0.001	Frontal_mid_L
Cluster 2	−36 − 20 + 44	237	< 0.001	Post‐CG L
Cluster 3	+14–8 + 14	198	< 0.001	Thalamus R
Cluster 4	–2 –46 + 14	172	< 0.001	Posterior cingulate L
Cluster 5	+44–14 + 62	152	< 0.001	Pre‐CG L

3.5.1. Post Hoc Seed‐Based Analyses

Post hoc seed‐based analyses were conducted with the MVPA‐defined clusters as seed regions. The results of these analyses revealed significant age‐group differences in rsFC patterns distributed across cortical and subcortical regions as well as parts of the cerebellum.

In the high PI group, older adults displayed patterns of weaker connectivity between six seed regions (the right superior temporal pole, left superior temporal pole, left cerebellum crus1, right parahippocampal gyrus, right middle temporal pole, and the right insula) with various frontal regions (e.g., precentral gyrus and postcentral gyrus), compared to younger/middle‐aged adults. Four seed regions (right superior temporal pole, left superior temporal pole, right middle temporal pole, and right insula) showed weaker connectivity with parietal regions (e.g., precuneus). However, the right posterior cingulate and left cerebellum crus1 seeds showed stronger connectivity with parietal regions (e.g., precuneus) in older adults, compared to younger/middle‐aged adults. Additionally, three seeds regions (right superior temporal pole, left superior temporal pole, and right insula) showed weaker connectivity with the insula and two seed regions (left superior temporal pole and right insula) showed patterns of weaker connectivity with temporal regions (e.g., temporal pole) in older adults, compared to younger/middle‐aged adults. Moreover, the left cerebellum crus1 seed showed weaker connectivity with other cerebellar regions (e.g., cerebellum crus2) while the right parahippocampal gyrus seed showed stronger connectivity with cerebellar areas (e.g., cerebellum crus2). Similarly, the right insula seed showed weaker connectivity with the right parahippocampal gyrus, while the left cerebellum crus1 seed showed stronger connectivity with the right parahippocampal gyrus and the right hippocampus in older adults. In addition, older adults showed patterns of stronger connectivity compared to younger/middle‐aged adults between the right superior temporal pole and thalamus, the left superior temporal pole and thalamus, the left superior temporal pole and caudate, and between the left cerebellum crus1 and posterior (e.g., lingual gyrus) and subcortical (e.g., posterior cingulate) regions.

The combined cluster seed map of negative (older<younger) connectivity, where younger adults in the high PI group showed patterns of stronger connectivity compared to older adults in the group, revealed some clusters of voxels that overlapped across post hoc seed‐based analyses. One left postcentral gyrus cluster (mean MNI coordinates: −43, −12, 46) was significant in five of the analyses; another postcentral gyrus cluster (mean MNI coordinates: −47, −15, 45) was significant in four of the analyses. Additionally, four of the analyses showed significant negative (older<younger) effects in clusters located in the right superior temporal gyrus (mean MNI coordinates: 55, −2, −2), the left middle cingulate cortex (mean MNI coordinates: −7, −32, 46), and the left superior temporal gyrus (mean MNI coordinates: −58, −12, 3). Three analyses overlapped in clusters centered in the right (mean MNI coordinates: 52, −7, 3 and 46, −32, 13) and left (−53, −6, 6 and − 63, −34, 12) superior temporal gyrus, left middle cingulate cortex (mean MNI coordinates: −2, −37, 50; −6, 11, 34; and − 5, −3, 37), left postcentral gyrus (mean MNI coordinates: −46, −14, 45), right putamen (mean MNI coordinates: 34, 11, 3), left (mean MNI coordinates: −59, −2, 34) and right precentral gyrus (mean MNI coordinates: 38, −13, 44), left insula (mean MNI coordinates: −35, −5, −8), left supplementary motor area (mean MNI coordinates: −4, 5, 49), and left Heschl's gyrus (mean MNI coordinates: −34, −28, 10). Combined negative effects across post hoc seed‐based analyses in the high PI group are depicted in Figure 3A.

FIGURE 3.

Combined results from the post hoc seed‐based analyses in the high PI group, depicting: (A) Regions exhibiting significant negative effects (older<younger/middle‐aged), and (B) Regions exhibiting significant positive effects (older>younger/middle‐aged. Voxel values reflect the number of analyses showing significance, with warmer colors reflecting higher consistency.

Less overlap was seen for the positive (older>younger) effects. The greatest overlap was seen in a cluster centered in the right caudate (mean MNI coordinates: 20, −15, 24) where three analyses showed positive effects. Additionally, some clusters showed overlaps across two analysis results. These were centered in the left lingual gyrus (mean MNI coordinates: −9, −55, 3), right caudate (mean MNI coordinates: 20, −16, 23), left hippocampus (mean MNI coordinates: −15, −37, 15), and the left parahippocampal gyrus (mean MNI coordinates: −24, −39, −9). One additional cluster was identified, but the mean coordinates did not correspond to a labeled region. Combined positive effects across post hoc seed‐based analyses in the high PI group are depicted in Figure 3B.

In the low PI group, older adults displayed patterns of weaker connectivity between three seeds (left postcentral gyrus, right thalamus, and left precentral gyrus) with frontal regions (e.g., frontal operculum), compared to younger/middle‐aged adults. Two seed regions (left postcentral gyrus and right thalamus) displayed weaker connectivity with posterior/cerebellar regions (e.g., occipital pole and cerebellum crus1). Additionally, weaker connectivity was observed between the right thalamus and subcortical areas (e.g., insula) and between the left middle frontal gyrus and temporal regions (e.g., inferior temporal gyrus). Finally, stronger connectivity was observed between the left postcentral gyrus and parietal regions (e.g., angular gyrus).

The combined cluster seed map of negative (older<younger) connectivity, where younger adults in the high PI group showed patterns of stronger connectivity compared to older adults in the group, revealed some clusters of voxels that overlapped across post hoc seed‐based analyses. The maximum overlap was across two analyses and was identified in clusters centered in the left inferior occipital cortex (mean MNI coordinates: −28, −78, −7 and − 40, −63, −6), right lingual gyrus (mean MNI coordinates: 17, −51, −7), right temporal fusiform gyrus (mean MNI coordinates: 24, −68, −5), and right superior temporal gyrus (mean MNI coordinates: 64, −26, 21). Combined negative effects across post hoc seed‐based analyses in the high PI group are depicted in Figure 4A.

FIGURE 4.

Combined results from the post hoc seed‐based analyses in the low PI group, depicting: (A) Regions exhibiting significant negative effects (older<younger/middle‐aged), and (B) Regions exhibiting significant positive effects (older>younger/middle‐aged. Voxel values reflect the number of analyses showing significance, with warmer colors reflecting higher consistency.

The combined map of the positive effects (older>younger) in the low PI group revealed no overlapping clusters/voxels. Combined positive effects across post hoc seed‐based analyses in the high PI group are depicted in Figure 4B.

4. Discussion

The present study investigated how rsFC is related to PI in WM and how this relationship differed between younger/middle‐aged and older adults. Both a hypothesis‐driven seed‐based approach was used to examine the role of IFG rsFC in the control of PI, alongside a data‐driven MVPA analysis to investigate how the ability to control PI in WM is associated with whole brain connectivity patterns. Of primary interest was to examine whether rsFC –PI associations differed between younger/middle‐aged and older adults.

The major findings from the study were that older individuals showed a pattern of both stronger and weaker IFG rsFC at baseline compared to younger adults, and that more PI was associated with stronger IFG rsFC with the left inferior occipital cortex across the whole sample. Additionally, reduced IFG—left inferior occipital cortex connectivity over 5 years was associated with a concurrent decline in the ability to control PI, and PI also showed age‐differential relationships in IFG rsFC with the left caudate and the vermis cross‐sectionally and with the insula and the anterior cingulate cortex longitudinally. MVPA analyses identified age‐differential patterns of connectivity contributing to the ability to control PI in WM. Taken together, these novel findings provide converging evidence for altered rsFC with increasing age and confirm that alterations in brain connectivity might have implications for the ability to control interference in WM.

Age‐stratified analyses revealed that weaker IFG connectivity with the left caudate and stronger IFG connectivity with the vermis were associated with more PI in older adults (Figure 1). The caudate is part of the dorsal striatum of the basal ganglia and has previously been implicated in WM (Lewis et al. 2004; McNab and Klingberg 2008; Owen et al. 1996; Postle and D'Esposito 1999). For instance, greater activation of the caudate/striatum has been linked to greater WM capacity and suggested to play a role in controlling attention and filtering distractions in WM (McNab and Klingberg 2008). Moreover, dopamine depletion in the caudate has been associated with impairments in updating and maintenance of information in WM, indicating the importance of the caudate in dopaminergic pathways involved in cognitive control processes (Cools et al. 2008). Furthermore, caudate dopamine has been associated with IFG activity during WM maintenance (Landau et al. 2009). Computational models have suggested that dopaminergic modulation of activity in the cortico‐striatal network mediates executive functioning (e.g., O'Reilly and Frank 2006). On a related note, Li, Bäckman, and Persson (2019) found that genetic markers associated with dopamine D2 receptor density were associated with frontostriatal activity and WM updating performance, especially in older age.

In a similar vein, previous studies have found that connections between the IFG and the caudate are important in WM updating, and that older adults show reduced coupling between these regions together with reduced performance (Podell et al. 2012). This functional coupling may reflect inhibitory control processing involved in successfully resolving PI during WM updating. The observed association between more PI and less IFG–caudate rsFC in older adults may, thus, indicate that aberrant IFG–caudate functional connectivity underlies the impaired ability to control PI in older individuals.

The possible contribution of the vermis to cognitive functioning, and to WM functioning more specifically, is less clear. Several previous studies have implicated the cerebellum in WM processing (e.g., Cabeza and Nyberg 2000; Chen and Desmond 2005; Desmond and Fiez 1998). However, others have found that while vermal gray matter volume is associated with cognitive functioning, these associations are no longer significant after controlling for prefrontal volume (Paul et al. 2009). It has also been suggested that the cerebellum contributes to WM by supporting articulatory rehearsal, postulated to help maintain information in WM (Ben‐Yehudah et al. 2007), which may be important in updating tasks such as the n‐back task.

Longitudinal analysis investigating age‐group differences in associations between change in IFG rsFC and change in PI across 5 years revealed that a reduced ability to control PI in older adults was associated with reduced connectivity between the IFG and two clusters centered in the insula and the anterior cingulate gyrus, respectively (Figure 2). These regions have previously been implicated in controlling PI in WM (Jonides and Nee 2006; Loosli et al. 2016; Nee et al. 2007; Samrani and Persson 2022). The anterior cingulate gyrus has previously been implicated in WM and postulated to play a role in monitoring and resolving conflict, thereby supporting the updating and maintenance of relevant information (Botvinick et al. 2004; Smith and Jonides 1999). Moreover, this region has been implicated in executive functioning more broadly by integrating information from cortical and subcortical regions in order to guide behavior. For instance, Kerns et al. (2004) found that the anterior cingulate cortex is active in tasks requiring conflict monitoring/resolution and this activity is associated with the ability to detect errors and adjust behavior. Additionally, Bush et al. (2000) found that the anterior cingulate cortex supports executive functioning through its role in attentional processes and interactions with other frontal regions. Furthermore, older adults have been found to display reduced anterior cingulate activity during WM tasks, indicating reduced efficiency of the region in older age (Madden et al. 2007), and may contribute to age‐related declines in WM and executive functioning.

To our knowledge, no previous study has employed fc‐MVPA to investigate age‐related differences in rsFC patterns related to cognitive function. Moreover, fc‐MVPA studies focusing on either age differences or cognition are also scarce. However, one recent study (Kim et al. 2024) employed this method to investigate connectivity patterns associated with cognitive functioning in a sample of older adults with suspected cognitive impairment. The authors used principal component analysis to identify cognitive performance patterns, resulting in two components: the first depicting general cognitive decline and the second depicting an association between lower memory and higher attention ability. Fc‐MVPA analyses, using the respective PCA scores, were conducted, resulting in two clusters localized to the cerebellum lobule VIII and the insula associated with the first component and two clusters localized to the temporal pole and the occipital gyrus associated with the second component. Given the methodological difference between this study and the present study, a direct comparison of the results is difficult. However, similar to Kim et al., the MVPA analyses in the high PI group identified seed regions located in the insula and the temporal pole(s). Speculatively, as these seeds were identified in the group with a lower ability to control PI, this may be related to general age‐related cognitive decline.

Moreover, several regions identified in the combined cluster connectivity maps have previously been implicated in working memory and executive functioning. For example, older adults in the high PI group showed greater connectivity with clusters in the hippocampus, insula, caudate, and cingulate across several of the MVPA seeds compared to younger/middle‐aged adults. As these regions have previously been implicated in proactive interference control in working memory (Andersson et al. 2023; Badre and Wagner 2005; Burgess and Braver 2010; Jonides and Nee 2006; Marklund and Persson 2012; Nelson et al. 2009; Persson et al. 2013; Samrani et al. 2019), these results may, speculatively, reflect alterations in how these regions are utilized/implicated in supporting this ability in older age. Importantly, however, while these findings align with previous research implicating these regions in PI control, they do not provide direct evidence of their functional role. Future studies using experimental manipulation or multimodal imaging approaches, such as combining rsFC with task‐based fMRI, could help to interpret how these connectivity patterns are related to PI control in aging.

The MVPA analysis within the group of individuals showing high PI demonstrated age‐differential patterns of both weaker and stronger functional connectivity. On the one hand, older adults with a lower ability to control PI displayed a pattern of weaker connectivity between most of the investigated clusters with frontal and parietal regions of the brain, as well as between some seeds and the insula, between one seed and the cerebellum, and between one seed and the parahippocampal gyrus, compared to younger/middle‐aged adults. Combined maps of negative effects (older>younger/middle‐aged) across seeds revealed several clusters with mean MNI coordinates located in frontal, temporal, and cingulate regions (Figure 3A). Additionally, low‐performing older adults showed patterns of stronger connectivity between individual seeds and thalamus, cerebellum, and limbic regions (including hippocampus and parahippocampal gyrus) compared to younger/middle‐aged adults. Combined maps of positive effects (older<younger/middle‐aged) across seeds revealed clusters with mean MNI coordinates located in subcortical (right caudate), limbic (left hippocampus, left parahippocampal gyrus), and occipital (left lingual gyrus) regions (Figure 3B).

The MVPA that included individuals showing low PI demonstrated age‐differential patterns of both weaker and stronger functional connectivity. Older adults with a greater ability to control PI displayed a pattern of weaker connectivity between the majority of the investigated seeds with frontal regions, as well as between some seeds and posterior/occipital regions, between one seed and the left inferior temporal gyrus, and between one seed and subcortical regions (insula and putamen), compared to young/middle‐aged adults. High‐performing older adults only showed patterns of stronger connectivity between one seed region (left postcentral gyrus) and the left angular gyrus. Combined maps of negative effects (older<younger/middle‐aged) across seeds revealed low overlap where two out of five seed‐based analysis results overlapped. These clusters had MNI coordinates centered in occipital and temporal regions (Figure 4A). The combined map of positive effects (older>younger/middle‐aged) across seeds did not reveal any overlapping significant voxels across the analyses (Figure 4B).

The patterns of weaker functional connectivity in older adults may be reflective of age‐related structural brain changes, including an anterior‐to‐posterior gradient of degradation in white matter (Sexton et al. 2014) and frontal gray matter shrinkage (Fjell et al. 2014). Previous investigations in the Betula sample, based on the same cohort, have found an association between reduced PI and reduced white‐matter integrity in aging (Andersson et al. 2022). Such structural alterations may result in increased activity and/or recruitment of additional regions to compensate for the effect of deterioration on cognitive function (Goh and Park 2009). For example, weaker rsFC with posterior regions may result in compensation by increasing rsFC with frontal regions, consistent with the posterior–anterior shift in aging (PASA; Davis et al. 2008). However, the patterns of weaker connectivity in low‐performing adults were largely centralized to frontal and parietal regions, and weaker connectivity in high‐performing older adults was largely centralized to frontal regions. Thus, the present results did not indicate a compensatory increase in connectivity with frontal regions in older adults with greater ability to control PI in working memory. Patterns of stronger connectivity in older adults, compared to younger/middle‐aged adults, are scarce, further indicating the lack of compensatory overactivation in older adults in the sample, both among the low‐ and high‐performing participants. Moreover, the MVPA analysis contrasting older adults in the high PI and low PI groups (low performing<high performing) did not reveal any clusters displaying significant differences in connectivity patterns at the specified threshold. Thus providing further support for a lack of compensatory connectivity in high‐performing older adults.

While previous studies investigating associations between the rsFC and the ability to control PI are rare, Samrani and Persson (2022) found that increased distance (5–10 back) between target and lure trials was associated with stronger HC connectivity with the left thalamus and bilateral temporal pole, suggesting the involvement of long‐term memory mechanisms in resolving PI. This is not consistent with the present results from the MVPA analyses, with the temporal pole seeds not exhibiting any significant positive or negative connectivity difference with the HC in relation to aging. Rather, we observed stronger connectivity between the superior temporal pole and the thalamus and caudate. However, low‐performing older adults did display stronger connectivity between the left cerebellum crus1 and the right hippocampus than younger/middle‐aged adults. The thalamus has been implicated in WM (e.g., Guo et al. 2017; Inagaki et al. 2019; Samrani et al. 2018) and has been suggested to be more active during low load compared to high load maintenance and to contribute to the inhibition of irrelevant information during high load conditions (Gomes et al. 2023). Furthermore, thalamus activity has previously been associated with WM load in the n‐back task (Chen et al. 2023), and thalamocortical interactions contribute to the modulation of distributed cortical activity during WM.

Moreover, several brain regions previously implicated in controlling PI (Nee et al. 2007; Samrani and Persson 2024) were demonstrated to be differentially involved in this ability in younger/middle‐aged and older adults in both the IFG‐based and whole‐brain analyses. Seed‐based analyses revealed that weaker IFG–caudate connectivity was associated with a lower ability to control PI and that reduced IFG–anterior cingulate gyrus connectivity was associated with a reduced ability to control PI in older adults.

4.1. Limitations

Important to note is that a relatively liberal voxel‐wise and cluster‐level threshold (0.05) was used in the seed‐based analyses on IFG rsFC due to the expected small effect sizes. Consequently, the identified clusters in both the whole sample and age‐stratified analyses of the PI effect were large and covered several regions located across different lobes. Thus, interpretations regarding the functional implications of these findings should be made with caution as the clusters cover regions also involved in cognitive functions and abilities other than PI in WM.

Longitudinal analyses of IFG rsFC associations with PI revealed a significant cluster centered in the inferior occipital cortex across the whole sample. Moreover, age‐differential associations were found between the declining ability to control PI and reduced rsFC between the IFG and two clusters centered in the insula and ACC, respectively. Importantly, however, the sample size for the longitudinal analyses was moderate (N = 134). Also, age did not significantly correlate with the change in PI scores in the whole sample or within the age groups. Thus, future studies should aim to replicate the present findings to strengthen conclusions regarding age‐related changes in the relationship between PI and IFG rsFC.

Finally, it should be noted that the working memory task used here included a limited number of trials, and all analyses relied on a difference score for measurement purity. This could have affected the reliability of the proactive interference measure and possibly led to an underestimation of true effects. Since the study was relatively well‐powered, we believe that this would, at least partially, compensate for low task reliability.

5. Conclusion

We found that older adults displayed stronger IFG rsFC with the postcentral gyrus, cuneal cortex, and angular gyrus but weaker connectivity with the parahippocampal gyrus, precuneus, superior parietal lobule, and supramarginal gyrus at baseline compared to younger to middle‐aged adults.

IFG‐based rsFC analyses revealed that weaker IFG connectivity with the left caudate and stronger connectivity with the vermis was associated with more PI in older adults. Moreover, a reduced ability to control PI over 5 years was associated with reduced IFG–inferior occipital cortex connectivity across the whole sample and reduced IFG–ACC and IFG–insula connectivity in older adults. Additionally, whole brain rsFC analyses revealed age‐differential patterns of rsFC associated with a lower and higher ability to control PI in WM.

Taken together, these novel results contribute to the understanding of decreased control of PI in aging by showing that both IFG and whole brain rsFC are differentially associated with the ability to control PI in older as compared with younger/middle‐aged adults. The present findings provide insights into the neural mechanisms underlying PI in working memory in aging, which may have practical implications for aging research. By identifying age‐related changes in rsFC associated with controlling PI, this study suggests potential targets that could inform future interventions targeting cognitive flexibility/cognitive control in older adults. Moreover, these results may enhance our understanding of age‐related working memory decline and could contribute to developing strategies to address interference‐related cognitive challenges in both clinical and everyday contexts.

Author Contributions

P. Andersson: formal Analysis, Writing – original draft, Writing – review and editing. M. G. S. Schrooten: supervision, Writing – review and editing. J. Persson: writing – original draft, Writing – review and editing, Supervision, Funding acquisition, Conceptualization.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1 Supporting Information.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. Due to data protection concerns, publicly sharing the entire data set underlying this study is not possible at the moment.