Age differences in the functional organization of the prefrontal cortex: analyses of competing hypotheses

Abstract

We employed a mixed design task for block and event-related functional magnetic resonance imaging with manipulations of levels of abstraction and duration in task-relevant cues and probes. Age-related differences between younger and older adults in task-related functional brain activity patterns of the prefrontal cortex (PFC) were reported. The results showed that (1) the low episodic condition evoked more activity in the more anterior PFC than the high episodic control condition for both age groups; (2) the low abstraction condition evoked more activity in the more anterior PFC than the high abstraction condition for both age groups; and (3) the signal change did not vary as a function of activity dynamics (transient and sustained responses) and maintenance duration (single-trial and multiple-trial). The findings showed that baseline conditions evoked more activity in the more anterior PFC for the older group than the younger group across most task contrasts and conditions, where these additional activities in the brain regions overlapped within the default mode network (DMN). We tentatively concluded that deficiency in the anterior DMN deactivation during externally driven tasks might be attributed to less efficiency in modulating local connectivity propagate to surrounding tissue, which may paradoxically increase brain activity.

Introduction

Literature has shown that the functional organization of the prefrontal cortex (PFC) exhibits a hierarchical manner (Sarafyazd and Jazayeri 2019; Kolk and Rakic 2022), in which its anterior–posterior organization (also known as rostro-caudal hierarchy) is either based on the timing with which cue stimuli reduce uncertainty in the action selection process (i.e. information cascade hypothesis; Koechlin et al. 2003; Koechlin and Summerfield 2007) or on the degree of abstraction of the task goals (i.e. levels of abstraction hypothesis; Badre and D’Esposito 2007, 2009; Badre 2008). A third hypothesis, namely the adaptive context-maintenance hypothesis, suggests that the organization of lateral PFC is based on the temporal dynamics of both anterior and posterior PFC to flexibly adjust the strategy used in different task-dependent context-maintenance demands (Reynolds et al. 2012). Despite the popularity of these hypotheses, there is no consensus on which hypothesis is more adequate to delineate the PFC functional organization. Hence, the primary aim of this study was to test these hypotheses on the organization of PFC. More critically, the existing literature evidence in support of either of these hypotheses was mainly derived from healthy young adults’ data. Rare research (yet see the one recently done by Yao and Hsieh 2022) has been reported to show whether an aged brain likewise exhibits similar functional organization of PFC as a younger brain. Therefore, whether there are age-related differences in the representational or processing demands that underlie this organization of frontal lobe function remains unclear. This study thus aimed to fill this research gap in the literature.

Particularly, we aimed to evaluate if an aged brain’s PFC functional organization would show a similar rostro-caudal hierarchy as a young brain’s, or instead show a dedifferentiation pattern that has often been reported in the literature regarding the posterior brain’s functional organization (e.g. visuoperceptual processing) for an aged brain. The dedifferentiation theory of cognitive aging suggests that task-evoked neural activity in the brain regions becomes less selective with increasing age. The interpretation of such neural dedifferentiation for older adults is contributed to a consequence of the age-related decline in functional specialization, which is originally seen as specialized for a single cognitive function for younger adults (Dennis and Cabeza 2011; Reuter-Lorenz and Park 2014; Koen and Rugg 2019; Seider et al. 2021). Alternatively, an aged brain’s PFC functional organization may show a frontal over-activation phenomenon. Literature has indicated another well-known aging brain theory, which suggests overactivation in brain activity of frontal regions or the activation shifting from posterior to anterior parts of the brain with aging (also known as a posterior–anterior shift in aging, PASA theory (Davis et al. 2008); additional readings refer to (Park and Reuter-Lorenz 2009; Grady 2012). For example, Yao and Hsieh (2022) applied Badre’s functional magnetic resonance imaging (fMRI) paradigm (Badre and D’Esposito 2007) by parametrically manipulating different levels of abstraction for the cue-response rule’s representations to investigate the age differences in the functional organization of PFC (Yao and Hsieh 2022). Their behavioral results showed that, compared with younger adults, older adults exhibited decreased efficiency in their responses when the choice hierarchy for cue-response pairings was more abstract. Their fMRI results showed a trend for the hierarchical organization along the rostro-caudal axis in both the younger and old groups. Yet, their brain-behavior correlations further showed that in experiments with the highest level of control demands, older adults displayed neural dedifferentiation (Goh 2011; Koen and Rugg 2019). Overall, Yao and Hsieh’s (2022) study indicated that older adults reflected maladaptive hyperactivation in task-unrelated areas that were detrimental to task performance with the highest control demands (Yao and Hsieh 2022). Since this previous study is one of the rare studies, it is therefore important to re-examine by using a different task paradigm to test whether the PFC functional organization for an aged brain exhibited a similar hierarchical manner as a younger brain, or instead, a dedifferentiation or frontal over-activation phenomenon to generalize the previous findings.

Materials and methods

Participants

We recruited 30 young and 30 older participants by means of advertisements on the Internet and bulletin boards from southern Taiwan. All participants were right-handed and with no evidence of neurological or psychiatric disorders based on their self-reports, and they were screened for probable dementia based on the Montreal Cognitive Assessment (MoCA; cut-off score < 26) before participating in the experiment. Of these participants, 3 of them (young n = 1, old n = 2) were not included in subsequent analyses as they failed to complete the experiment due to technical problems (e.g. temporary MRI scanner malfunction or cannot tolerate the environment inside the MRI). Further 11 of them (young n = 3, old n = 8) were excluded from the analyses due to their behavioral performance below the chance level. Total remaining 26 young participants were 16 males and 10 females with a mean age of 23.5 (age range: 20–27 years), a mean education year of 16.46 (range: 14–19 years), and a mean MoCA score of 29.24 (range: 27–30). The total remaining 20 old participants were 9 males and 11 females with a mean age of 65 (age range: 50–78 years), a mean education year of 15.58 (range: 12–18 years), and a mean MoCA score of 29 (range: 26–30). All participants provided written informed consent, and the study protocol was approved by the Research Ethics Committee of the National Cheng Kung University, Tainan, Taiwan, R.O.C. All participants were rewarded with 1,000 new Taiwan dollars (NTD) after completing the experiment, including neuropsychological tests, cognitive tests, and neuroimaging acquisition.

Behavioral tasks

Stimulus and experimental design

The visual stimuli used in the computerized task were programmed using the PsychoPy v2020.1.3 (Peirce et al. 2019) implemented in Matlab software. The task design was adopted from Reynolds et al’s. (2012) study with the modification of the stimuli. The stimuli used in this study consisted of characters retrieved from the Chinese Zodiac designed as a probe stimulus, such as    and Arabic digits designed as a task cue, such as “0,” “1,” “2,” “3,” “4,” “5,” “6,” “7.” Before the digit number cue, there was an additional color (such as yellow, green, blue, or red) cue, indicating which nesting task structure that the subsequent trials would belong to (see Fig. 1). Participants must use this additional color-dot cue to interpret the number cue that will then allow them to respond to the subsequent probe stimulus.

There were 5 conditions of a delayed-response task that each involved a continuous series of stimuli, presented one at a time. In the baseline (BS) condition (see Fig. 1a), participants were instructed to make a response to a particular Chinese character that was designed as a probe (e.g.) by using the index finger or the middle finger (the response rule was to respond to or with the index finger and to respond to or with the middle finger). This condition was considered to have 0 degrees of nesting task structure (Fig. 1a) because participants only need to consider determining the response solely based on the direct cue-response mapping rule irrespective of any prior number cue or color information (e.g. a yellow dot and “6” or “7” digit number acting as neutral cues denoting the baseline condition).

In addition to the BS condition, for the other 4 trial conditions, 2 factors were manipulated (in accord with the levels of abstraction hypothesis): (1) the number of trials (single- vs. multiple-trials) over which the cueing information needed to be maintained (aka maintenance duration); and (2) how abstract the relevant cue-response mappings were (low- vs. high-abstraction).

The definition of abstraction was the degree of nesting of the cue-response rules that the participants need to attend to select a correct response (see Fig. 1a–c for an illustration). The more levels of the cue-response rule contained; the higher abstraction was. Specifically, in the 2 low-abstraction (low-abstraction, single-trial; low-abstraction multiple-trial) conditions (Fig. 1b), determining the response to the probe depends on the prior number cue. For example, if the number was “2,” the participants had to respond to the probe of with the index finger, while if the number cue was “3,” they have to respond to with the middle finger instead. These conditions were considered to have 1 degree of nesting because the response to probes was nested within 1 factor (i.e. the number cue: “2” ~ “5”). In addition, for these 2 low-abstraction conditions, a green color cue was implemented. In the 2 high-abstraction conditions (high-abstraction, single-trial; high-abstraction, multiple trials), participants’ response to the probe depended on the prior number cue and color cue (i.e. if the color cue was red and the number cue was “0,” then made the response with the index finger to ; if the number cue was “1,” then made the response with the middle finger to ; if the color cue was blue, then the above cue-response mappings were reversed) (see Fig. 1c for an illustration). These conditions were considered to have 2 degrees of nesting because the choice of response to the probe was nested within 2 factors (i.e. the number cue nested within the color cue).

As for the manipulation of the maintenance duration, it was referred to as how often contextual cues were presented (see Fig. 2). For example, in single-trial conditions, all contextual cues were presented on every trial (trial-by-trial basis), whereas in the multiple-trial conditions (block-wise basis), contextual cues were only updated at the beginning of a 5-trial block. That is, in the multiple-trial conditions, the number cue was only presented on the first trial of a block in the low-abstraction multiple-trial condition, and the color cue was only presented on the first trial of a block in the high-abstraction multiple-trial condition; whereas the remaining trials’ contextual cues in the block were replaced with question marks to control for the equivalence of stimulus presentation and sequence effects as in the single-trial conditions.

The sequence parameters of task presentation and stimulation were identical to Reynolds et al’s. (2012) study (see Fig. 2 for an illustration). We counterbalanced the presentation order of all visual stimuli across participants and conditions. Before each block started, a resting session with eyes opened lasting for 15 s would be implemented. The composition of the presentation order for each trial within a block is as follows: (1) the start of a trial begins with presenting either a color cue or a question mark (both of which required a left-hand index finger response; the purpose of the question marks during trials’ presentation is to maintain the similar visual stimulation, number of manual responses, and trial timing for evoking same as other trials); (2) a question mark or number cue, the 2 of them required response with a left-hand index finger, and (3) a probe stimulus, requiring a response with either index or middle finger of right hand based on the instruction. All visual stimuli were presented on a black background, while all Chinese characters and Arabic letters were presented in white color in 24-point bold PMingLiU font for making the strongest contrast of visual stimuli clarity. All probe stimuli were underlined, to denote a different cue-response rule to the visual cue. The duration of visual stimuli was presented for 300 ms, while the duration of the stimulus-onset asynchrony (SOA) between cues and subsequent probes was presented for 2,500 ms. Participants must respond within 1,500 ms of cue and 2,500 ms of probe onset during each trial. After each probe or cue presented on screen, a cross sign (i.e. fixation stimuli) with the white color was displayed, and the duration of the intertrial interval (ITI) (Dale 1999) was varied between 2,200, 4,700, and 7,200 ms for estimating the transient the blood-oxygen-level-dependent (BOLD) signal response on each trial (Burock et al. 1998).

Behavioral data analysis: rationales of the design

Mean error rates and mean response times (RTs) (exclusion of RTs that were greater or smaller than mean RT ± 3SD, about 5% of trials) were calculated to determine whether they were different across conditions. We performed 2 sets of analysis of variance (ANOVA) on probe-RTs based on either the information cascade hypothesis, or the levels of abstraction hypothesis, and one set of analyses on cue-RTs to test the adaptive context maintenance hypothesis.

The design matrix for the first set of ANOVA to test the information cascade hypothesis is shown in Fig. 3A, which included four trial conditions: baseline (BS), low-abstraction single-trial (LS), low-abstraction multiple-trial (LM), and high-abstraction multiple-trial (HM) forming a 2 by 2 ANOVA, that is, episodic control (low: BS/LS vs. high: LM/HM), and contextual control (low: BS/LM vs. high: LS/HM). For the conceptual formula to derive these respective trial conditions for the design matrix, please see Reynolds et al’s. (2012) study for details (The rationales are as follows: BS and single-trial conditions are regarded as low episodic control conditions; whereas multiple-trial conditions are regarded as high episodic control conditions. This type of temporally distant, block-oriented control has been termed episodic control. Yet, low episodic control conditions do not mean that they require less cognitive activity because there is an orthogonal variable (i.e. contextual control), that needs to be considered, which is defined depending on the state of S-R mappings. If the S-R mappings are constant within blocks, then they are regarded as low contextual control conditions, such as the BS (S-R mappings always constant within and across blocks) and low-abstraction, multiple-trial (LM) conditions (S-R mappings remain the same within blocks, but can be varied across blocks; see Figs. 1 and 2); whereas if the S-R mappings are varied within blocks, then they are considered as high contextual control conditions, such as the low-abstraction, single-trial (LS) conditions, and the high-abstraction, multiple-trial (HM) conditions (see Fig. 2 for examples).). Whereas the design matrix for the second set of ANOVA to test the levels of abstraction hypothesis is shown in Fig. 3B, which included other 4 trial conditions: LS, high-abstraction single-trial (HS), LM, and HM, forming a 2 by 2 ANOVA, that is, maintenance duration (single: LS/HS vs. multiple: LM/HM) and levels of abstraction (low: LS/LM vs. high: HS/HM). To note, of the main interest of this study, we added a between-subject variable, i.e. group (young vs. old) into each of these 2 sets of ANOVAs.

The third set of the analysis targeting on the cue-RTs (sum RTs of two cues: a color cue and a number cue or question marks; see Fig. 2) to test the adaptive context-maintenance hypothesis by contrasting the cue-RTs for the single-trial conditions to those for the multiple-trial conditions. The hypothesis is that the contextual cues on each trial would be associated with additional encoding, updating, and maintenance processes (see Reynolds et al. 2012). Likewise, a between-subject variable of age was added to the analyses.

fMRI acquisition

The imaging data were acquired on a General Electric (GE) Discovery MR750 3 Tesla scanner (General Electric Medical Systems, Milwaukee, USA) in a 32-channel receive-only phased-array head coil in the Mind Research Imaging center at the National Cheng Kung University. High-resolution structural images were acquired with a fast-spoiled gradient recalled pulse sequence consisting of 166 axial slices (TR/TE/flip angle 7.6 ms/3.3 ms/12°; a field of view (FOV) 22.4 × 22.4 cm²; matrices 224 × 224; slice thickness 1 mm). Each functional scanning runs 8 alternating cycles of task and fixation blocks with an additional fixation block at the beginning. The EPI images were collected using an interleaved T2^* weighted gradient-echo planar imaging (EPI) pulse sequence (TR: 2,500 ms, TE: 30 ms, Flip angle:77°, matrix size: 64/64, FOV: 22 cm/1x, slice thickness: 4 mm, voxel size, 3.4375 × 3.4375 × 4 mm, number of slices:32). A total of 208 volumes and 6 dummy scans were acquired for each condition. The visual stimulus was displayed using PsychoPy v2020.1.3 on an MSI GS65 Stealth laptop and was projected onto a screen that was viewed through a mirror attached to the head coil.

fMRI imaging preprocessing

We used FMRIB Software Library (FSL) software (Jenkinson et al. 2012) to analyze the functional imaging data at 1st level. Specific steps are as follows: Firstly, preprocessing steps included head motion artifact corrections using the Motion Correction FMRIB’s Linear Image Registration Tool (MCFLIRT) (Jenkinson et al. 2002). We then used the brain extraction tool (BET; Jenkinson et al. 2002) to remove non-brain tissue from the preprocessed MR images. The FSL Motion Outliers tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLMotionOutliers; (Jenkinson et al. 2002) to identify outlier volumes based on frame displacement between volumes (above the 75th percentile +1.5 times the interquartile range). The output of this program was used to down weight those volumes in analyses. The functional image of an individual’s brain was registered to the high-resolution T1 structural image by a linear transformation, and then the individual structural image was registered to the standard MNI152 template by a linear transformation. Then, we set up the first-level GLM with a 9 mm full-width half-maximum (FWHM) Gaussian kernel for spatial smoothing.

fMRI blocked analyses: sustained brain responses

The blocked analysis procedure was designed to test the information cascade hypothesis (Koechlin et al. 2003; Koechlin and Summerfield 2007) and the levels of abstraction hypothesis (Badre and D’Esposito 2007, 2009; Badre 2008).

A general linear model (GLM) was used to estimate parameter values reflecting the mean difference between the task and fixation blocks for each experimental condition. A boxcar regressor starting from the stimulus and ending after a 50-s interval with a 15-s rest period was convolved with the double gamma. Please note that we followed the analysis procedures Reynolds et al. (2012) reported, in which we only implemented block regressors but not event (i.e. trial) regressors in the blocked analyses. The logic of this approach is that the transient effects should be decaying back to baseline during ITIs, whereas sustained effects should remain relatively constant and of increased amplitude relative to blocks of fixation.

Subsequently, the 6 single-column parameters from MCFLIRT, which describe the motion correction performed at each volume, and the FSL Motion Outliers files were added to the model as confounding variables to be excluded. When all the steps were completed, the functional imaging data were preprocessed for further analyses.

All the group-level analyses involved computing the activation level across the whole brain region for each participant and submitting each of those to a group-level ANOVA or t-test in which the participant was treated as a random effect. We identified clusters of activity that were sensitive at a cluster-level rate of 0.05, a 3.1 z-threshold was used to define contiguous clusters. Then each cluster’s estimated significance level (from Gaussian Random Field (GRF)-theory) was compared with the probability threshold. We then performed 2 different analyses in order to test the hypotheses regarding the organization of PFC.

The first set of analyses was to test the information cascade hypothesis

A 3-way with 2 (group: young vs. old) by 2 (episode control: low: BS/LS vs. high: LM/HM) by 2 (context control: low: BS/LM vs. high: LS/HM) repeated-measures ANOVA on sustained brain responses was performed. The main effects of group, episodic control, and contextual control, and the interaction of both variables were examined. If the interaction was significant, then we would further conduct simple interaction and main effect tests.

The second set of analyses was to test the levels of abstraction hypothesis

We performed a 3-way with 2 (group: young vs. old) by 2 (abstraction: low: LS/LM vs. high: HS/HM) by 2 (maintenance duration: single-trial: LS/HS vs. multiple-trial: LM/HM) repeated measure ANOVA on sustained brain responses. The main effects of group, abstraction, and duration, and the interactions of both variables were examined. If the interaction was significant, then we would further conduct simple interaction and main effect tests.

fMRI decomposition of temporal dynamics: transient (trial-based) vs. sustained (block-based) responses

The third analysis tested the adaptive context-maintenance theory proposed by Reynolds et al. (2012). An additional set of 1st level analyses was used to estimate transient activations associated with particular trial events and compare them with sustained activity associated with entire task blocks (i.e. those derived from the blocked analyses). To this end, we generated another GLM model to capture the transient RT-locked to the event (i.e. each trial of a block). Please note that only trials with correct responses and RTs of no more than 3 standard deviations were extracted from the different conditions’ behavioral data and programmed into an onset file. The onset file contains information on the start time (i.e. the first cue of a trial) and the duration (7.5 s) of each qualifying trial, which was edited as 1st regressor.

The third set of analyses was to test the adaptive context-maintenance hypothesis

This hypothesis makes 2 specific predictions. The first is that the more anterior areas of PFC should be more active for all conditions relative to baseline conditions since all conditions (i.e. LS, HS, LM, HM) involve an additional context-maintenance demand relative to the baseline condition. The second prediction is that activity is expected to be transient in the conditions in which contextual information is updated on each trial (i.e. LS, HS), but sustained in conditions requiring maintenance of contextual information across a block of trials (i.e. LM, HM). Therefore, we would expect to see more activity in more anterior PFC for the single-trial cueing condition on the transient responses, but more activity in more anterior PFC (i.e. DLPFC) for the multiple-trial condition on the sustained responses (i.e. a double-dissociation phenomenon for the type of trial condition (single vs. multiple) and type of responses (transient vs. sustained)). Furthermore, these temporal dynamics should be independent of levels of abstraction.

Results

Behavioral data

Young group

Overall, the behavioral performance (including the percentage of error rates, reaction time to cues (containing 2 different cues, see Fig. 2), and reaction time to probe stimuli) in both young and old groups were reported in Table 1. No condition contrasts revealed significant differences in error rates (all Bonferroni correction Ps > 0.05).

For the younger group, the ANOVA analysis on the probe-RT according to the information cascade hypothesis showed no significant main effects and interaction effects. Likewise, the ANOVA analysis on the probe-RT according to the levels of abstraction hypothesis showed no significant main effects and interaction effects. On the other hand, if we focus on the cue-RT data (i.e. by taking the summed RT for the 2 cues) according to the adaptive context-maintenance hypothesis, then we could expect to observe that participants were slower to respond to cues in the single-trial conditions (mean RT: 1,103 ms) relative to the baseline conditions [mean RT: 1,015 ms; F(2,50) = 14.50, P < 0.000; post-hoc Tukey test: q(50,3).01 = 4.325000], and to the multiple-trial conditions [mean RT: 1,081 ms; F(2,50) = 14.50, P < 0.000; q(50,3).01 = 4.325000]. Furthermore, the cue-RTs were also significantly slower in the single-trial condition than in the multiple-trial condition [F(2,50) = 14.50, q(38,3).05 = 3.450000]. Such results suggest that for single-trial conditions, task cues were associated with additional encoding, updating, and maintenance processes, resulting in slower cue responses, which accorded with the adaptive context-maintenance hypothesis.

Old group

For the old group, the ANOVA analysis on the probe-RT according to the information cascade hypothesis showed no significant main effects and interaction effects. The analysis of the probe-RT according to the levels of abstraction hypothesis showed a significant main effect of duration (single-trial vs. multiple trials), but no other significant main effect and interaction effects. On the other hand, if focusing on the cue-RT according to the adaptive context-maintenance hypothesis, the ANOVA results showed that participants were slower to respond to cues in the single-trial conditions (mean RT: 1,328 ms) relative to the baseline conditions [mean RT: 1,094 ms; F(2,50) = 27.86, P < 0.000; post-hoc Tukey test: q(38,3).01 = 3.450000], and to the multiple-trial conditions [mean RT: 1,244 ms; F(2,50) = 27.86, P < 0.000; q(38,3).01 = 3.450000].

Young vs. old groups

Of the main interest, we additionally contrasted younger and old groups in 3 sets of mixed-design ANOVAs separately based on the 3 hypotheses, respectively. The results of the ANOVA on the probe-RT according to the information cascade hypothesis showed no significant main effect of the group [F(1, 44) = 3.16, P = 0.083]. The results of the ANOVA on the probe-RT according to the levels of abstraction hypothesis, likewise showed no significant main effect of the group [F(1, 44) = 2.03, P = 0.161]. On the other hand, the results of ANOVA on the cue-RT according to the adaptive context-maintenance hypothesis now showed a significant main effect of the group [F(1, 44) = 10.84, P < 0.005)], with the older group yielding a larger context-maintenance effect than the younger group.

Neuroimaging data

The first set of analyses on the sustained responses to test the information cascade hypothesis

According to this hypothesis, we would expect to observe more activity occurring in the anterior regions of the PFC for the episodic control condition (i.e. the multiple-trial conditions: task cues that are presented at the beginning of a block of multiple trials, and are relevant over the entire block); whereas more activity in the posterior regions of the PFC for both the contextual (i.e. single-trial conditions: task cues that are presented on a trial-by-trial basis) and episodic control effects. Of the main interest, we further compared the activity patterns of older adults compared to young adults. For brevity, we reported only the results for the 3-way ANOVA in the main text and provided the results of subsequent simple interaction and main effects in Supplementary Material A.

A 3-way mixed-design (group x episodic control x contextual control) ANOVA was conducted. The results showed that for the contrast where activity for the high episodic control was greater than the low episodic control condition, younger adults exhibited more clusters of activity than older adults in the left middle frontal gyrus, left middle temporal gyrus, and left superior lateral occipital cortex; whereas older adults exhibited more clusters of activity than younger adults in the frontal pole region (see Fig. 4).

On the contrary, the BOLD activity for the high contextual control was greater than the low contextual control condition, and younger adults exhibited more clusters of activity than older adults in the inferior frontal gyrus, left orbital cortex, lingual gyrus, bilateral inferior temporal gyrus, and left superior parietal lobule, whereas older adults exhibited more clusters of activity than younger adults in the right frontal pole, right superior frontal gyrus, right precentral gyrus, and left middle frontal gyrus. Moreover, there was a significant 3-way interaction among age group, episodic control, and contextual control condition. For the subsequent simple interaction and main effects, please see the supplementary materials, where the effects of episodic control and contextual control were analyzed separately for the younger and older groups, respectively (Supplementary Material A).

Interim summary of the first set of analyses on the information cascade hypothesis

Some key findings regarding the information cascade model derived from the 3-way ANOVA and the subsequent simple effect tests (shown in Supplementary Materials A) are as follows: (1) Although there was a significant main effect of episodic control in several clusters of activity for both age groups, there was also a significant main effect of contextual control (but only for the younger group, see Supplementary Material A), and a significant interaction between the episodic control and contextual control in several other clusters of activity for both age groups—the results appeared to be inconsistent with the prediction derived from the information cascade hypothesis. (2) For the obtaining contrast where activity for the high episodic control was greater than the low episodic control condition, younger adults exhibited more clusters of activity in the left middle frontal gyrus, left middle temporal gyrus, and left superior lateral occipital cortex; whereas older adults exhibited more clusters of activity in the frontal pole region. The results showed an age-related frontal over-activation phenomenon. (3) More critically, regarding the main effect of episodic control, for the younger group, the high episodic control condition paradoxically evoked more activity in relatively more posterior regions than the low episodic control condition, such as in the left middle frontal gyrus and superior occipital lobe, whereas the low episodic control condition paradoxically evoked more activity than the high episodic control condition in a relatively more anterior region, such as the superior medial frontal cortex, postcentral gyrus, and the left frontal orbital cortex (see Supplementary Materials A, Fig. S1). Likewise, for the older group, there were more clusters of activity in the more anterior PFC (i.e. the right frontal pole) for the contrast where the activity for the low episodic was greater than the high episodic control condition, but paradoxically there were no significant clusters of activity for the contrast where the activity for the high episodic control greater than the low episodic control condition (see Supplementary Materials A, Fig. S2). The results again appeared to be inconsistent with the prediction derived from the information cascade hypothesis.

The second set of analyses on the sustained responses to test the levels of abstraction account

According to this hypothesis, we would expect to observe more activity occurring in the anterior regions of the PFC for the high abstraction condition (i.e. more degrees of nesting for the cue-response rules; see Fig. 1c) and to observe more activity in the posterior regions of PFC for the low abstraction conditions (see Fig. 1a and b). Of the main interest, we further compared the activity patterns of older adults compared to young adults. For brevity, we reported only the results for the 3-way ANOVA in the main text and provided the results of subsequent simple interaction and main effects in Supplementary Materials A.

A 3-way mixed-design (group x abstraction level x maintenance duration) ANOVA was conducted. The results showed that for the contrast where the activity for the higher abstraction was greater than the lower abstraction, the younger group evoked more activity in the bilateral middle temporal gyrus, bilateral occipital cortex, and the right temporal pole, whereas the older group evoked more activity in the frontal pole (see Fig. 5).

In the contrast where the activity for the multiple-trial condition was greater than in the single-trial condition, the younger group evoked more activity in the left inferior temporal gyrus, postcentral gyrus, bilateral middle frontal gyrus, and the left superior lateral occipital cortex, whereas the older group evoked more activity in the right frontal operculum cortex, insula, the right middle temporal cortex, and the right superior frontal cortex.

There was a significant 3-way interaction among three variables shown in the bilateral inferior lateral occipital cortex. Subsequently, we then examined the effects of abstraction and maintenance duration separately for the younger and older groups, respectively (see Supplementary Materials A for details).

Interim summary of the second set of analyses on the levels of abstraction hypothesis

Some key findings regarding the levels of abstraction model derived from the 3-way ANOVA and the subsequent simple effect tests (shown in Supplementary Materials A) are as follows: (1) Although there was a significant main effect of abstraction in several clusters of activity for both age groups, there was also a significant main effect of maintenance duration, and a significant interaction between the episodic control and contextual control in several other clusters of activity for both age groups (see Fig. 5 and Supplementary Materials A, Figs. S3 and S4)—the results appeared to be inconsistent with the predictions derived from levels of abstraction hypothesis. (2) The results of simple effect tests for the younger group showed the opposite direction from the model’s prediction, in which the low abstraction condition paradoxically evoked more activity in the more anterior PFC, such as the frontal pole, than the high abstraction condition (see Supplementary Materials A, Fig. S3). Likewise, for the older group, the lower abstraction condition paradoxically evoked more activity in the more anterior areas, such as the superior medial frontal lobe and the right middle frontal cortex than the higher abstraction condition (see Supplementary Materials A, Fig. S4). (3) More critically, the older group evoked more anterior PFC, such as the frontal pole, than the younger group for the contrast where the activity for the high abstraction is greater than the low abstraction (see Fig. 5). These results speak against the levels of abstraction hypothesis.

The third set of analyses on (1) the transient vs. sustained responses and (2) only the transient responses to test the adaptive context-maintenance theory

• (1) The rationale for the third theory is that the transient responses (the cueing information was transiently updated in a single trial, i.e. single-trial condition) were predicted to demonstrate increased activity in the single-trial conditions relative to the multiple-trial conditions, whereas the sustained responses (the cueing information needed to be sustained over several trials in a block, i.e. multiple-trial conditions) were predicted to demonstrate increased activity in the multiple-trial conditions relative to the single-trial conditions. Accordingly, we contrasted the conditions in which the image activity averaged across low- and high-level abstraction conditions for the single-trial conditions to compare with the image activity averaging across those for the multiple-trial conditions: [(LS + HS)/2] vs. [(LM + HM)/2], separately for different activity dynamics, i.e. sustained vs. transient responses. We analyzed the data separately for younger and old groups on these contrasts. We would expect to observe that on the transient responses, single-trial conditions would evoke more activity than multiple-trial conditions, whereas, on the sustained responses, the pattern would be reversed in which multiple-trial conditions would evoke more activity than single-trial conditions.

• (2) In addition, we contrasted the baseline conditions with all other conditions on the transient responses only: BS vs. LS; BS vs. HS; BS vs. LM; BS vs. HM; and BS vs. (LS + HS)/2; BS vs. (LM + HM)/2, and contrasted between younger versus old groups. We would expect for the transient responses the clusters of activity would be larger than the BS conditions.

Transient vs. sustained responses

Young group

For the younger group, out of the expectation, during the transient responses, multiple-trial conditions evoked more activity in the clusters of the middle temporal gyrus than single-trial conditions, whereas single-trial conditions paradoxically evoked no more activity than multiple-trial conditions. As for the sustained responses, there were no significant clusters of activity for the multiple-trial condition than in the single-trial condition—the results were inconsistent with the adaptive context-maintenance theory (see Fig. 6 upper panel).

Old group

For the old group, also out of the expectation, on the transient responses, multiple-trial conditions evoked more activity than the single-trial conditions in the clusters of the superior frontal cortex, anterior cingulate gyrus, temporal fusiform, and posterior gyrus, whereas single-trial conditions evoked more activity than the multiple-trial condition in the clusters of the left superior lateral occipital lobe. As for the sustained responses, likewise out of expectation, there were no more clusters of activity for the multiple-trial than the single-trial conditions, but there were significant clusters of activity in the left precentral region for the single-trial than the multiple-trial condition. These results appeared to be inconsistent with the adaptive context-maintenance theory (see Fig. 6 lower panel).

Transient responses only

Young group

Of the main interest, we examined if compared to the baseline conditions, all other conditions, such as LS, LM, HS, and HM, would evoke more clusters of activity in the more anterior regions of the PFC on the transient responses. The results showed the opposite patterns, in which it is the BS condition that evoked more anterior PFC, such as the frontal pole, than the reverse comparisons (see Fig. 7).

Old group

Likewise, for the old group, we examined if compared to the baseline conditions, all other conditions, such as LS, LM, HS, and HM, would evoke more clusters of activity in the more anterior PFC. The results showed the opposite pattern that the BS condition that evoked more anterior PFC, such as the frontal pole, than the reverse comparisons (see Fig. 8).

Young vs. old groups

Of the main interest, we contrasted younger and old groups and contrasted single-trial conditions vs. baseline conditions on the transient responses. The results showed that the younger group evoked more clusters of activity in the anterior cingulate gyrus, lingual gyrus, supramarginal gyrus, left middle frontal gyrus, and left superior parietal lobe, whereas the old group evoked more clusters of activity in the right frontal pole, angular gyrus, and superior frontal gyrus, for the contrast where single-trial condition greater than the baseline (see Fig. 9).

In addition, we contrasted younger and old groups and contrasted multiple-trial conditions vs. baseline conditions on the transient responses. The results showed that the younger group evoked more clusters of activity in the lingual gyrus, the left middle frontal gyrus, and the left superior parietal lobule, whereas the old group evoked more clusters of activity in the frontal pole and superior frontal gyrus, for the contrast where multiple-trial condition greater than the baseline (see Fig. 9).

Interim summary of the analyses on the adaptive context-maintenance hypothesis

We set out 2 sets of analyses to examine if the results would support this hypothesis. The key findings are as follows: (1) The results of the first analysis did not yield a significant interaction between the maintenance duration (single-trial vs. multiple-trial) and dynamics of activity (sustained vs. transient), which were inconsistent with the third hypothesis. (2) In addition, the second set of the analyses on the transient responses only, i.e. by comparing the activity in the BS condition to those in all other conditions, also did not support the third hypothesis, in which it was the BS condition that evoked more clusters of activity in the more anterior PFC, such as the frontal pole, than the inverse contrast for both age groups. (3) When directly comparing the 2 age groups, we observed more clusters of activity in the more anterior PFC (e.g. frontal pole) for the old group than in the younger group. The result seemed to accord with the well-known age-related frontal over-activation or the PASA theory proposed by prior research (Davis et al. 2008; Park and Reuter-Lorenz 2009; Grady 2012).

Discussion

The first aim of this study was to examine if the functional organization of the PFC coordinates with the information cascade hypothesis, the levels of abstraction hypothesis, or the adaptive context-maintenance hypothesis, by using a task-related fMRI paradigm that was adopted from the study by Reynolds et al. (2012). The second aim of the study was to further evaluate if an aged brain’s PFC functional organization would show similar hierarchical organization as seen in young adults, or conversely show either a dedifferentiation phenomenon, in which there was no such hierarchical activity of the PFC as a function of task-cue nesting structure or cueing information duration or a frontal over-activation phenomenon.

According to the information cascade hypothesis (Koechlin et al. 2003; Koechlin and Summerfield 2007), we would expect to observe a significant main effect of episodic control, but no other effects. Specifically, more anterior areas of PFC (such as mid-dorsal lateral PFC, mid-DLPFC) should be sensitive to only episodic control (i.e. multiple-trial cueing conditions relative to single-trial cueing or baseline conditions), whereas more posterior areas of PFC would be recruited in single-trial cueing conditions. This hypothesis predicts that the most anterior areas of PFC (i.e. mid-DLPFC) should not be sensitive to contextual control, while all posterior brain areas of PFC to it should be. However, the current study did not support the information cascade hypothesis. The current behavioral data on the probe-RT showed no significant main effects of episodic and contextual effects or interaction. Furthermore, based on this hypothesis, we would also expect to see only the high episodic control condition (i.e. multiple-trial cueing conditions) evoked more activity in the more anterior regions of PFC than the low episodic control conditions (i.e. single-trial cueing or baseline condition). Yet, the current results for the younger group did not support this prediction, in which it was the high episodic control condition that paradoxically evoked more activity in relatively more posterior regions, such as the middle frontal gyrus and superior occipital lobe, whereas the low episodic control condition evoked more activity in a relatively more anterior region, such as the medial PFC. Likewise, the older group also showed more clusters of activity in the more anterior PFC for the contrast where the activity for the low episodic was greater than the high episodic control condition.

As to the levels of abstraction hypothesis, the higher abstraction task condition (i.e. the higher degree of cue-response nesting: that is, more numbers of additional relevant task cues need to be attended to for determining the cue-response mapping rules) should evoke more activity in the anterior DLPFC compared to the low abstraction task condition, regardless of the cue duration (i.e. single-trial cueing, multiple-trial cueing). Specifically, mid-DLPFC regions should be engaged only in the high-abstraction conditions, whereas posterior PFC regions would be predicted to be active in both high and low-abstraction conditions relative to baseline. In addition, the maintenance duration factor (i.e. single-trial vs. multiple-trial cueing conditions) should be irrelevant based on the level of abstraction theory. However, these results did not support the levels of abstraction hypothesis. The current behavioral data on the probe-RT shows no significant main effects of abstraction and cueing maintenance duration, and no interaction effect did not in accord with the hypothesis’s prediction. Moreover, as predicted by the hypothesis, the high abstraction task condition should evoke more clusters of activity in the anterior regions of PFC as compared to the low abstraction task condition. Yet, the current imaging results for the younger group showed the opposite direction, in which it was the low abstraction condition that evoked more clusters of activity in the more anterior regions of PFC, including the right superior frontal gyrus, frontal pole, right inferior frontal region, and right angular gyrus than the high abstraction condition (see Supplementary Materials A, Fig. S5). For the older group, the low abstraction condition also paradoxically evoked more clusters of activity in the more anterior regions of PFC (e.g. the superior frontal region) than the higher abstraction condition.

Turning to the third hypothesis, differing from the previous 2 theories, emphasizes the recruitment and dynamics of the DLPFC activity depend on the demands for context processing along several dimensions, including the need for preparatory processing (i.e. single-trial vs. multiple-trial) and the temporal duration (sustained vs. transient) over which it extends (Reynolds et al. 2012). The current behavioral data on the cue-RTs showed that participants were slower to respond to cues in the single-trial conditions relative to both the BS conditions and the multiple-trial conditions. Such results suggest that for single-trial conditions, task cues might be associated with additional encoding, updating, and maintenance processes, resulting in slower cue responses, which behaviorally seem to accord with the adaptive context-maintenance hypothesis. However, the prediction from imaging results also did not accord with this third hypothesis. One critical prediction regarding the imaging data derived from this hypothesis is that we should expect to observe the transient responses showing more activity in the anterior regions of PFC for the single-trial cueing condition, whereas the sustained responses, on the contrary, show more activity in the anterior regions of PFC for the multiple-trial cueing conditions. Nonetheless, the current results turned out to be the opposite, in which it was the transient responses evoked more activity in the more anterior regions of PFC (such as the superior frontal cortex, anterior cingulate gyrus, temporal fusiform, and posterior gyrus for the older group) for the multiple-trial cueing conditions than the single-trial conditions. Furthermore, another prediction derived from this hypothesis is that we should expect to observe more clusters of activity for all other conditions than for the BS condition. Yet, when we contrasted the BS conditions with all other four conditions, we found the inverse pattern, in which it was the BS condition that evoked more clusters of activity in the more anterior PFC, such as the frontal pole, than the inverse contrast for both age groups. More surprisingly, these additional clusters of activity for the BS condition overlapped with the well-known default mode network (DMN), which we will discuss in the later paragraphs.

Regarding the second aim of the present study, we did not observe obvious different patterns between young and old groups in terms of no evidence to support any of the hierarchical functional organization hypotheses of the PFC. However, as we discussed previously, we consistently observed that the old group evoked more clusters of activity in the more anterior part of the PFC as compared to the younger group across many task condition contrasts for testing hypotheses. These results seemed to accord with the well-known age-related frontal over-activation or the PASA theory proposed by recent studies (Davis et al. 2008; Park and Reuter-Lorenz 2009; Grady 2012). The additional analysis with a finite impluse response (FIR) model (Please note, in the current analyses on the transient responses, we did not implement the FIR model as done in Reynolds et al’s. (2012) study. In order to generalize the current results, we have also incorporated an additional analysis to estimate values based on various time points within the hemodynamic response (i.e. using a FIR model). This approach to GLM coding of transient and sustained responses has been validated in both simulation and empirically based methodological studies (Visscher et al. 2003; see Supplementary Materials B, Figs. SB1–SB4).) also corroborate with our prior story on frontal overactivation, especially in the anteromedial PFC, which it was found to be activated in monitoring the external environment, contributing to faster reaction times (Gilbert et al. 2005, 2006). It also seems that the DMN is recruited in switching tasks, in the case of a demanding shift from a cognitive context to a different one (Crittenden et al. 2015). This finding, together with our previous findings may be aligned with a recent study (Malagurski et al. 2022) that suggested the compensatory mechanism of frontal overactivation as a result of strengthening in within-network connections with advancing age (Jones et al. 2011). These recruitments of additional cognitive resources/pathways may be seen as the habitual connectivity routes are damaged or not as efficient as before (Behfar et al. 2020) or as a maladaptive response that leads to short-term compensation but long-term damage (Damoiseaux 2012; Hillary and Grafman 2017). It highlights the role of frontal overactivation in aging cognition.

Although the current results could not be explained solely by either of the hierarchical-type hypotheses, we nevertheless observed a very interesting phenomenon. That is, we accidentally observed that the transient responses in the BS condition evoked more clusters of activity than those in all other conditions (i.e. BS > LS, BS > LM, BS > HS, BS > HM) in the brain regions where the DMN are consisted of (see Supplementary Materials A, Fig. S6). The DMN is a large-scale brain network primarily composed of parts of the medial frontal cortex, posterior cingulate, posterolateral parietal cortex, retrosplenial cortex, and hippocampal formation, which are commonly observed at rest (Greicius et al. 2003; Fransson and Marrelec 2008; Andrews-Hanna et al. 2010). The DMN was noticed to be deactivated in certain goal-oriented tasks and was referred to as the task-negative network (Fox et al. 2005; Vatansever et al. 2017) in contrast with the task-positive network. The DMN is unique in that it shows less activity during nearly all attention-demanding, externally directed cognitive tasks compared to rest. It is understood that failing to decrease activity within the DMN during executive function tasks may be related to task performance (Persson et al. 2007; Prakash et al. 2012; Burzynska et al. 2013). fMRI studies report that during cognitive testing, not only are some brain areas “activated” (i.e. BOLD signal is higher than that observed during baseline trials) but also are some brain areas “deactivated” (i.e. BOLD signal lower than during baseline trials). For instance, attention-demanding tasks are typically associated with the activation of various brain regions within the dorsal attentional network (DAN) and concurrent deactivation of various brain regions within the DMN (Shulman et al. 1997; Mazoyer et al. 2001; Raichle and Snyder 2007). Together with some other cortical areas, these are thought to constitute a DMN that is active at rest or when engaging in “stimulus-independent” thought, but which undergoes a reduction in activity when attentive demanding goal-directed cognition needs to be undertaken (Gusnard et al. 2001; Raichle et al. 2001; Greicius et al. 2003; Raichle and Snyder 2007).

Possible mechanistic explanations regarding the observed increased PFC activation for older adults might reflect a deficiency of DMN deactivation (Stevens et al. 2008; Anticevic et al. 2012). That is, the reduction of the DMN activity (i.e. task-induced deactivation of the DMN) for the 4 conditions relative to the BS condition might be larger for the younger group than for the old group. Possible explanations are that a deficiency of DMN deactivation can be seen as dysfunction in keeping neuronal excitability and vascular reactivity due to a decreased neuronal excitatory input to the cerebral cortex. Reduced nerve cell activity may be related to increased synaptic inhibition or deactivation caused by a decrease of excitatory synaptic input to a defined neuronal circuit (Rakic et al. 1994; Petanjek et al. 2011; Changeux et al. 2021). In line with these findings, we conjecture that the possible deactivation is the result of reduced efferent, excitatory activity in crossed projections from the cerebral cortex to brainstem nuclei and relay neurons in the PFC. This may partly explain why we would observe the older adults recruited more anterior regions of the PFC (e.g. frontal pole) relative to the young group across many condition contrasts. This speculation can be supported by a large body of research demonstrating that DMN resting-state connectivity is negatively impacted by both aging and Alzheimer’s disease (for reviews, see Lustig et al. 2003; Persson et al. 2007; Stevens et al. 2008; Anticevic et al. 2012).

Furthermore, among various cortical cortex regions, the PFC undergoes the largest overproduction and the slowest rate of elimination (Petanjek et al. 2011). Since the younger group’s age range in the current study was between 20 and 27 years old (covering most of the young adulthood age range, e.g. 18–25 years old, which was within the age range of final stage of cortical maturation), it is likely that the hierarchical functional representations in the PFC had not been well stabilized due to the protracted development of synaptic connections (Petanjek et al. 2011) (Researchers have provided evidence showing that dendritic spine density in childhood is about 2–3 times the adult values (overproduction period), and begins to decrease (elimination period) during puberty to reach the adult level at the onset of adolescence (Changeux and Danchin 1976; Rakic et al. 1986; Bourgeois et al. 1994). Yet, such a substantial elimination of synaptic spines continues beyond adolescence and throughout the third decade of life before stabilization (Petanjek et al. 2011, 2019).). The maturation of synaptic connections, especially layer III projection neurons of PFC has been demonstrated to be highly related to the hierarchical modular organization of the human brain connectome (Changeux et al. 2021; Kolk and Rakic 2022). The current old group’s age range was widely between 50 to 78 years old, which might be in the period of synaptic spines aging. Accordingly, it might not be surprising to observe non-hierarchal functional organization of PFC in the current study. Further research is needed to test this connectomic hypothesis (Changeux et al. 2021).

Conclusions and implications of the current study

To summarize, the current results showed no evidence of an abstraction hierarchy, an information cascade hierarchy, and adaptive contextual maintenance dynamics within the PFC. The current findings of no support of either of the theories regarding the PFC functional organization do not directly falsify any of them, but rather, simply warrant researchers to more carefully define what the high abstraction level, high episodic control, and high task demands should be. Nonetheless, the general observation from all tested competing hypotheses reveals shared frontal overactivation across many task condition contrasts, especially in the anteromedial PFC within anterior DMN. It highlights the role of frontal overactivation in aging cognition. We further suspected that it might be related to the deficient deactivation of the DMN activity for the old group. We believe our study has provided the initial evidence to link the relationship between frontal overactivation, and DMN in aged cognitive demanding performance.

Acknowledgments

We thank Mind Research and Imaging Center (MRIC) at National Cheng Kung University for consultation and instrument availability.

Funding

This work was supported by the National Science and Technology Council (NSTC), originally known as the Ministry of Science and Technology (MOST), Taiwan (ROC) of the Republic of China, Taiwan for financially supporting this research (grant number Nos. 108-2410-H-006-038-MY3, 109-2923-H-006-002-MY3, 110-2321-B-006-004, and 111-2321-B-006-008). In addition, this research was supported in part by Higher Education Sprout Project, Ministry of Education to the Headquarters of University Advancement at National Cheng Kung University (NCKU) (grant number: D111–F2903; D111-F2909; R111-B013).

Conflict of interest statement: None declared.

References