Impaired executive functioning mediates the association between aging and deterministic sequence learning

Abstract

Sensitivity to the fixed ordering of actions and events, or deterministic sequence learning, is an important skill throughout adulthood. Yet, it remains unclear whether age deficits in sequencing exist, and we lack a firm understanding of which factors might contribute to age-related impairments when they arise. Though debated, executive functioning, governed by the frontal lobe, may underlie age-related sequence learning deficits in older adults. The present study asked if age predicts errors in deterministic sequence learning across the older adult lifespan (ages 55–89), and whether executive functioning accounts for any age-related declines. Healthy older adults completed a comprehensive measure of frontal-based executive abilities as well as a deterministic sequence learning task that required the step-by-step acquisition of associations through trial-and-error feedback. Among those who met a performance-based criterion, increasing age was positively correlated with higher sequencing errors; however, this relationship was no longer significant after controlling for executive functioning. Moreover, frontal-based executive abilities mediated the relationship between age and sequence learning performance. These findings suggest that executive or frontal functioning may underlie age deficits in learning judgment-based, deterministic serial operations.

Learning about sequences involves sensitivity to the typical ordering of actions and events, or recognition of relationships between cues that predict targets in time or space. Although this type of learning supports thoughts and skills that are ubiquitous to daily life (Lashley, 1951), we still lack a coherent understanding of how aging influences the ability to learn serial operations. Existing literature seems to agree upon age-declines in learning probabilistic (irregular) sequences and age-equality in learning deterministic (fixed) sequences (for a review, see Howard & Howard, 2013). Yet our recent work using a deterministic serial learning task challenged this pattern (Dang et al., 2020).

In our “Kilroy” task, participants learned a fixed “chain” of stimulus-response associations in a step-by-step fashion, based on trial-and-error feedback. Starting with the correct final response, actions were added one-by-one, until the participant mastered a fully predictable sequence from start to finish to escape a computerized maze of rooms. Not only did younger adults outperform older adults, but we also observed steady age-related declines in learning sequential structure across the older adult lifespan. Dang et al. (2020) ruled out sequence complexity and manual motor demands as explanations for the observed age effects; however, higher-order executive functions may be required for action planning (e.g., monitoring of actions and their consequences or inhibiting extraneous information) or remembering temporal and sequential order (e.g., mnemonic maintenance of “chains” of associations) in order to govern goal-directed actions and processes (Salthouse et al., 2003) like those encountered in our task. Such executive processes are known to depend on the frontal lobes (for a meta-analysis, see Yuan & Raz, 2014), which show disproportionate age-related declines (e.g., volumetric declines, cortical thinning, pronounced white matter loss, and reductions in dopamine receptors and transporters). Indeed, these biological changes account for many cognitive declines seen in older people, known as the frontal lobe hypothesis of aging (Dempster, 1992; West, 2000; West, 1996). Accordingly, the present study asked whether such frontally-mediated executive functions may explain age-based losses in learning deterministic serial orders.

Though debated, ample evidence already suggests that sequence learning, broadly defined, requires intact frontal lobes and executive functioning (Curran, 1995; Janacsek & Nemeth, 2013, 2015; Wilson et al., 2017). Compared to controls, monkeys with dorsolateral prefrontal cortex lesions performed more poorly on a sequential task (Petrides, 1991). Relatedly, patients with prefrontal lesions are more impaired than healthy age-matched controls on both implicit and explicit visuomotor sequence learning tasks (Beldarrain et al., 2002; Beldarrain et al., 1999; Meier et al., 2013). Other work shows that performance on tests of executive functioning predict motor sequencing (Fama & Sullivan, 2002) and that behavioral impairments on serial ordering are amplified in healthy aging (Ma et al., 2018) due to weaker involvement of frontal brain networks and dedifferentiation of frontal regions (Ye et al., 2020). More recent work suggests that the role of frontal regions in sequence learning is task-dependent (Vekony et al., 2022), and specifically calls attention to frontal-involvement in situations where learning depends on making judgments (e.g., noticing pattern structures underlying stimuli or verbalizing an underlying regularity (1997)). The Kilroy task may be one such judgment-linked sequence learning task.

Still, research using the Kilroy task report no significant correlations between sequence learning and traditional measures of executive functioning, like digit span, verbal fluency or the Wisconsin Card Sorting Task (WCST) (Dang et al., 2020; Herzallah et al., 2013; Keri et al., 2008; Polgár et al., 2008; Shohamy et al., 2005). This may reflect the use of younger or middle-aged samples (c.f. Dang et al., 2020), as executive functions are more likely to relate to sequence learning among older than younger adults (Niermeyer et al., 2017). Varying results may also reflect differential effects of aging on disparate aspects of executive functioning (Rodríguez-Aranda & Sundet, 2006) or that the chosen frontally-mediated executive functioning tasks were suboptimal to compare with the Kilroy sequence learning task. Most existing tests of executive functioning test specific domains of executive ability (Iavarone et al., 2011); for example, the WCST (Berg, 1948) taps conceptualization and cognitive flexibility whereas the Controlled Oral Word Association Test (COWAT; Benton, 1969) assess only verbal fluency. Accordingly, research using the Kilroy task explicitly could not discount the possibility that sequencing errors may be attributable to executive abilities more generally (see Polgar et al., 2008).

To more suitably examine the association between age, sequencing ability, and executive functioning, we opted for a comprehensive measure of frontal functioning. Here, we administered the frontal assessment battery (FAB; Dubois et al., 2000), which tests multiple facets of executive functioning at once. In this 10-minute battery, participants are evaluated on six domains of executive ability including conceptualization, mental flexibility, motor programming, sensitivity to interference, inhibitory control, and environmental autonomy. Though other broad batteries of executive functioning exist (i.e., Behavioral Assessment of Dysexecutive Syndrome (BADS); Wilson et al., 1966) and Executive Interview (EXIT 25; Royall et al., 1992), they are time-consuming (Hobson & Leeds, 2001) or correlate with measures of general cognitive functioning rather than with tests of executive processes, respectively (Dubois et al., 2000; Royall et al., 1992). In contrast, the FAB is brief, generally accepted by participants, and encompasses multiple domains of executive processes (Hobson & Leeds, 2001; Iavarone et al., 2011). A rich body of research indicates that the FAB is a sensitive and specific indicator of frontal lobe structure and function (Dubois et al., 2000; Kopp et al., 2013). Lower FAB scores have been linked to below-average blood flow localized in frontal regions (Hurtado-Pomares et al., 2018), and lower gray matter volumes in frontal (Terada et al., 2017) versus parietal or temporal brain regions (Bezdicek et al., 2017). Importantly, the FAB can differentiate the frontal dysfunction of patients with cortical from subcortical lesions (Dubois et al., 2000), whereby frontal lobe lesion patients consistently score lower on the FAB relative to patients with other lesions (Han et al., 2020). Existing work also reveals poorer FAB scores with increasing age (Appollonio et al., 2005; Asaadi et al., 2016; Barbosa et al., 2012; Benke et al., 2013; Bezdicek et al., 2017; Dias et al., 2009; Iavarone et al., 2011; Miki et al., 2013; Paula et al., 2013), consistent with the frontal hypothesis of aging.

The current study, therefore, sought to replicate the finding that age predicts deterministic serial-based learning performance among older adults using the Kilroy task, and extend it by examining whether frontally-mediated executive functions mediate this association. Using a large sample of healthy older adults, we hypothesized that advancing age would predict deterministic sequencing errors as well as impaired executive functioning as measured by lower FAB scores, in line with previous work. We also hypothesized that lower FAB scores would predict greater deterministic sequencing errors. Finally, we tested whether age-related errors in deterministic sequence learning were mediated by FAB scores; that is, we hypothesized that age would no longer be a significant predictor of sequencing errors once FAB and age together were entered into a regression model.

Methods

Participants

Participants included 153 healthy older adult community volunteers (48 men, 105 women), aged 54 to 89 (M = 71.27 years, SD = 8.63) with at least high school education (M = 16.20 years, SD = 2.39). Recruited through posted flyers, this sample was predominantly non-Hispanic White (77.8%). People who identified as Black or African-American (13.7%), Asian (3.3%), Hispanic or Latino (2.6%), two or more races (.7%), or did not report their race (2.0%) comprised the rest. All participants were in good health, were not color blind, had normal or corrected-to-normal vision, had no existing neurological or psychological conditions, and did not use medications known to influence cognition. Participants were screened for dementia using the Mini Mental State Exam (MMSE; Folstein et al., 1975), and were excluded from our study if they scored a 27 or below. Before commencing experimental testing, participants provided written informed consent in accord with the St. Olaf College institutional review board. This study was in compliance with the ethical standards for research on human subjects delineated in the 1964 Declaration of Helsinki. Participants received monetary compensation for their participation.

General Procedure

Participants completed demographic and health screening forms, a neuropsychological battery, and the deterministic sequence learning task during a single 1- to 1.5-hour visit. Test ordering remained constant across participants; the neuropsychological test battery (see below) always came before the deterministic sequence learning task.

Neuropsychological Tests

To characterize our sample, we included the Wechsler Memory Scale – Third Edition (WMS-III) Logical Memory, immediate and delayed (Tulsky & Price, 2003), and the North American Adult Reading Test (NAART; Uttl, 2002). To test executive and frontal lobe functioning, we used the Frontal Assessment Battery (FAB), a quick test that is easy to administer and reportedly not frustrating to older participants or patients (Dubois et al., 2000; Moorhouse et al., 2009) despite encompassing six subtests, which assess conceptualization, mental flexibility, motor programming, sensitivity to interference, inhibitory control and environmental autonomy.

Deterministic Sequence Learning Task

Participants completed a computerized sequence learning task, programmed in the SuperCard language, that has distinct sequencing, probe, and retraining phases (Shohamy et al., 2005). The participant’s objective was to guide an avatar, “Kilroy,” through a sequence of rooms to reach a goal, the outside. Figure 1 describes the task structure and provides example stimuli.

Figure 1.

Upper image: A schematic presentation of the deterministic sequence learning task. The three phases are trained successively. For each phase, participants reach criterion before moving to the next phase. During sequencing, reward is presented only after participants complete the entire sequence successfully.

Lower image: a) The participant must choose a door to open in room 1 (or subphase A). b) The participant chose a locked door in room 1 and will need to try again. c) The participant chose the open door in room 1 and reaches the reward. d) In room 2 (or subphase B), the participant chose the open door, and enters into room 1.

Each room, regardless of phase, presents three different doors with highly discriminable colored cards. In each trial, participants use a mouse click to select the door they think is correct. The correct door in each room is unlocked and either leads to the outside or to the next room in the sequence. The two incorrect doors in each room are locked, and if chosen, the participant must keep trying to identify the unlocked door, without any limit on response times. The location of the correct door in each room is randomized so that the correct answer (left, right, center) varies across trials; it is the color of the card, not its location, that determines the correct response. The colored cards marking the correct doors in each of the rooms are randomized across participants.

On each trial, the computer records only the color cards selections and spatial ordering of the doors, the desired response, and the door(s) the participant chooses. Errors are counted as any instance when the incorrect door is selected by the participant. Notably, a participant may make multiple errors on a single trial by choosing an incorrect door one or more times before choosing the correct door, in each of one or more rooms in the sequence.

Sequencing.

Before the task begins, the participant is acquainted with the task in a practice phase, in which the participant must successfully guide Kilroy out of a single room to the outside several times. Next, the sequencing phase has four subphases, where Kilroy is eventually guided through a sequence of four rooms. The participant starts by learning the correct response in Room 1 (subphase A), which leads to the outside. Once the correct response is learned, the participant is taken into a second room (Room 2, subphase B) and is required to learn which door in the second room leads to Room 1. Once inside Room 1, the participant is prompted to choose the correct door that leads to the outside. After this two-step sequence is learned, additional rooms are added to the sequence one at a time. Participants are trained backwards with an additional room being added to the sequence in each subphase until the participant selects the correct door in each room of the full four-room sequence (Room 4 → Room 3 → Room 2 → Room 1 → reward).

A trial is complete when Kilroy reaches the outside, regardless of Kilroy’s starting position. To meet a minimum performance criterion, the participant must correctly complete four consecutive trials in each stage of the sequence. If participants fail to meet criterion within 15 trials starting in any given entry room (including practice), they skip the remaining sequencing training, as well as probe, and proceed directly to the retraining phase.

Probe.

Once the full sequence is learned, a probe phase verifies that participants understood the sequential structure acquired during sequencing. The transition to the probe phase is not signaled to the participant. Just as in the final stage of sequencing, Kilroy starts in Room 4 and the participant must guide him through the entire sequence of rooms until he reaches the outside; however, now the incorrect colored doors have changed in each room. One of the incorrect doors reflects a door that was correct in a different room elsewhere in the sequence, and the other door represents a door that was always incorrect elsewhere in the sequence. Though the sequence of correct stimuli from Room 4 to Room 1 has not changed, now the participant must accurately associate the correct colored door with the correct room in each subphase of the sequence. This tests whether the participant fully learned the chain of associations in the sequencing phase, as intended (i.e., the red door is correct in Room 2) versus specific stimulus-response relationships irrespective of the sequence (i.e., the red door is correct with no knowledge of the chained relationships between rooms). While either strategy could be used during the sequencing phase, only a participant who learned sequentially can perform well on the probe phase (making few errors, regardless of what other doors are presented); in contrast, the participant who learned specific stimulus-response relationships (non-sequentially) would make many errors whilst deciding between two doors that were both correct at some point during the sequencing phase). There are 6 trials in the probe phase.

Retraining.

In the final phase, the participant sees a new room with three new colored doors, and must choose the unlocked door that leads outside. This phase again evaluates the participant’s ability to make single stimulus-response associations, and because it occurs at the end of the task, checks for fatigue or other non-associative effects. The number of trials is variable; the phase does not end until the participant correctly completes four consecutive trials.

Results

Nine participants (M = 69.22 ± 8.67 years old, M = 16.00 ± 2.00 years of education) failed the practice trials before testing began and were not included in our analyses. As done in Polgar (2008), participants were also excluded if they could not acquire the full sequence, meaning they failed to meet the minimum performance criterion during any sequencing subphase (n = 32; M = 73.38 ± 9.39 years old, M = 15.97 ± 2.69 years of education). Participants identified as outliers, by performing more than three standard deviations from the norm in either the FAB or our Kilroy task, were also excluded (n = 2; one 82-year-old female with 16 years of education and one 63-year-old male with 16 years of education). Our final sample included 110 older adults (31 men, 79 women), aged 54 to 89 (M = 70.80 ± 8.36 years of age) with 16.29 ± 2.36 years of education.

Analyses comparing those who were excluded from the analysis (n = 34) versus those who successfully learned the full sequence (n = 110) indicated that the groups did not differ significantly in age, years of education, gender, or neuropsychological performance (all p’s > .05), with the exception of NAART errors, which indicates verbal ability (Excluded: M = 24.12, SD = 11.24; Included: M = 18.95, SD = 10.10, t(138)= 2.52, p = .01). Greater verbal ability among those who successfully completed the task relative to those who could not replicates a previous finding of a small but significant association between verbal ability and sequencing performance on a deterministic task (Cherry & Stadler, 1995).

Neuropsychological Outcomes.

All participants performed within expected age norms on our neuropsychological tests: WMS-III Logical Memory, immediate (M = 21.62, SD = 6.65) and delayed (M = 17.32, SD = 6.94), NAART errors (M = 18.95, SD = 10.10) and FAB (M = 16.23, SD = 1.74). As predicted, bivariate correlations showed a link between age and executive function as measured by the FAB, r(108) = −.25, p = .01, whereby higher age was associated with poorer FAB scores (see Figure 2a). Age was also related to NAART errors, r(108) = −.27, p = .005, mirroring past results that older age is associated with fewer errors (Uttl, 2002). There were no significant correlations between age and Logical Memory (immediate and delayed, p’s > .05).

Figure 2.

Scatterplots depicting correlation between (a) age and executive functioning, (b) age and sequencing errors, and (c) executive functioning and sequencing errors. Executive functioning was defined as performance on the Frontal Assessment Battery (FAB).

Relationship Between Age and Task Performance.

Linear regression examined whether age predicted the total number of errors during the sequencing phase of our task. As predicted, age is a significant predictor of sequencing errors (ß = 0.202, p = 0.034, 95% CI [.006, .166]; F(1,108) = 4.59, p = 0.040), accounting for 4.1% of the variance (see Figure 2b). Across all participants, the majority of sequencing errors (> 82%) occurred on the newest, least-practiced doors (the first room in a trial) versus practiced rooms that involved stimuli learned in previous subphases (i.e., selecting the incorrect door for a room on the third trial after selecting the correct door for the room on the two previous trials, consistent with forgetting). Age did not, however, predict errors at either post-learning phase: probe (ß = .107, p = .265, 95% CI [−.056, .200]) or retraining (ß = .092, p = .339, 95% CI [−.020, .057]). The average number of errors on probe trials was considerably low (M = 4.82, SD = 5.63), confirming that older adults displayed sequential knowledge as intended (i.e., learned the chain of stimuli during the sequencing phase in a sequential manner) as opposed to simply associating each stimulus with a reward regardless of its place in the sequence. Moreover, nearly 90% of participants completed retraining within 6 trials or less (the minimum is 4 trials), providing evidence that fatigue effects were not responsible for observed age deficits during sequencing and that all older adults could learn simple stimulus-response associations.

Of note, we also examined the number of trials to reach criterion during the sequencing subphases. If older adults chained each stimulus by simply memorizing or keeping increasingly longer sequences in mind, we would expect the trial length to grow for each additional subphase. Solving longer sequences may involve mnemonic retention of the correct responses from previously encountered rooms, and could tax working memory. A mixed-model analysis of variance (ANOVA) on trial length with age and subphase as factors revealed only a significant effect of subphase, F(3, 234) = 4.86, p <. 005 (i.e., no effect of age, p = .11, and no interaction, p = .36). Pairwise comparisons showed that trial length in chained rooms (i.e., starting in Room 2: 5.97 trials ± 2.0, Room 3: 5.77 trials ± 1.8, or Room 4: 5.80 trials ± 2.0) differed significantly from starting in Room 1: 5.09 trials ± 1.2 (p’s < .05), indicating that the first subphase, with only a single stimulus-response association, may be easiest to learn. Although significant, we contend there is no meaningful trial length effect because all chained subphases were learned in about 5 to 6 trials and this pattern did not interact with age. According to Shohamy et al., (2005), this result indicates that sequencing occurred in an associative manner for all adults as intended, as opposed to merely learning the task by rehearsing in working memory the correct sequence structure.

Relationship Between Neuropsychological Outcomes and Task Performance.

Linear regression examined whether executive functioning, as measured by FAB scores, could predict the total number of errors during sequencing. Figure 2c shows that FAB performance is a significant predictor of total sequencing errors (ß = −0.297, p = 0.002, 95% CI [−.981, −.236]; F(1,108) = 10.47, p = 0.002, R² = .088), though it is not related to the total number of errors at either probe (ß = −.183, p = .067, 95% CI [−1.11, .042]) or retraining (ß = .128, p = .182, 95% CI [−.059, .308]). The only other significant correlations were between total number of probe errors with both logical memory-immediate, r(108)=−.233, p = .016, and logical memory-delayed, r(108)=−.221, p = .023, similar to what is reported in Myers et al. (2008). No other associations between neuropsychological measures and task performance (i.e., errors at sequencing, probe and retraining) reached significance, all p’s > .05.

Mediation Analysis.

Above we showed that (1) increasing age predicted a higher number of sequencing errors, (2) increasing age predicted lower executive functioning, as measured by FAB scores, and (3) lower executive functioning (or FAB scores) predicted a higher number of sequencing errors. Given these associations, we used the hierarchical regression analysis procedures laid out by Baron and Kenny (1986) to examine whether individual differences in FAB mediated age differences in sequencing errors. In a new model, with both age and FAB scores as predictors of sequencing errors, FAB performance remained a significant predictor (ß = −2.78, p = 0.006, 95% CI [−.922, −.155), but not age (ß = 1.435, p > 0.05, 95% CI [−.022, .138). This regression model was significant, F(2,107) = 6.314, p = 0.003, R² = .106, suggesting that the FAB completely mediates the relationship between age and sequence learning. In fact, this model with FAB scores and age was a significantly better predictor of sequence learning than the model with age alone, accounting for an additional 6.6% of the variance (ΔF(1,107) = 7.75, p = 0.006). Moreover, a Sobel test (z = 1.972, p =.049) provided additional evidence for significant mediation between age and sequencing errors via FAB scores (see Figure 3).

Figure 3.

Mediation analysis revealing that executive functioning, as measured by the Frontal Assessment Battery (FAB), mediates the relationship between age and sequence learning. Path values are standardized regression coefficients.

To calculate the degree to which the FAB attenuated the variance in sequence learning that can be explained by age, the amount of variance uniquely associated with age (after partialling out the effect of FAB scores) was subtracted from the amount of variance associated with age as the sole predictor, and then divided by the amount of variance associated with age as the sole predictor (see Salthouse, 1991). The attenuation of the age effect was substantial, with the FAB attenuating age-related sequence learning variance by 58% (1- [0.017/0.041] × 100).

Inclusion of outliers.

All results were re-analyzed to include outliers who performed more than three standard deviations from the norm. The pattern of results held (all p’s < .05) with the exception of the final Sobel test which was only marginally significant (z = 1.79, p =.07).

Discussion

We studied whether frontal lobe and executive functioning declines often seen in healthy aging contribute to deterministic sequence learning, in which each event perfectly predicts a subsequent event. In a large sample of older adults, age significantly predicted sequencing errors, directly replicating our finding that advancing age correlates negatively with the ability to chain fixed sequences of events (Dang et al., 2020). We extend this result by revealing that low scores on the Frontal Assessment Battery (FAB) - a brief measurement of frontal lobe function – was associated with both increasing age and higher sequencing errors. Finally, FAB scores mediated the relationship between age and sequencing performance, supporting the view that cognitive processes involving the frontal lobes and executive functions are particularly vulnerable in aging. Specifically, age deficits in sequence learning may reflect declining executive functions that are recruited to process and produce deterministic chains of operations.

Older adults made more errors with advancing age during sequencing only. As is typical, most errors were related to learning new stimuli as opposed to older stimuli learned in previous subphases. This same pattern was identified in Dang et al. (2020) which compared younger, middle-aged and older adults, Shohamy et al. (2004) which compared Parkinson’s patients on and off levodopa to controls, and H. Nagy et al. (2007) which compared controls to patients with Parkinson’s or amnestic mild cognitive impairment. Taken all together, older adults’ deficits in sequencing do not simply reflect general slowing or forgetting (Salthouse, 2000), where earlier associations are lost before newer processing of the sequence is complete. Notably, age also did not correlate with errors on probe or retraining, indicating that age deficits are indeed selective to learning deterministic sequential information. Successful probe performance offers additional evidence that participants learned the chain of stimuli during the sequencing phase in a sequential manner, as intended, while low errors on retraining help rule out fatigue effects on performance.

The negative association we found between age and FAB scores is consistent with the literature (Appollonio et al., 2005; Asaadi et al., 2016; Barbosa et al., 2012; Benke et al., 2013; Bezdicek et al., 2017; Dias et al., 2009; Iavarone et al., 2011; Miki et al., 2013; Paula et al., 2013). While some report no relationship between age and the FAB (Beato et al., 2012; Dubois et al., 2000; Lima et al., 2008; Rodrigues et al., 2009), these studies relied on younger samples than ours. In fact, the age of the oldest participant in three of these four studies does not exceed 75.5 years, whereas the present study examined older adults from 54 to 89 years (71 years on average). Though there may be restriction of range in our FAB scores (i.e., a ceiling effect) amongst our well-educated and healthy older sample, our average scores align with those of both Parkinson’s patients on medication and older control participants without neurological or psychiatric history (Dubois et al., 2000). Furthermore, our finding of lower FAB scores with advancing age fits with the frontal hypothesis of aging, which postulates that executive functioning is progressively impaired in older adults. FAB scores robustly correlate with frontal lobe structure and function (Bezdicek et al., 2017; Kopp et al., 2013; Terada et al., 2017) as well as traditional tests of executive ability (Dubois et al., 2000; Han et al., 2020; Hurtado-Pomares et al., 2018), corresponding with losses in frontal brain volume and executive function that often accompany old age (e.g., Yuan & Raz, 2014).

Likewise, FAB scores significantly predicted deterministic sequencing errors, replicating data showing a role for frontal lobe integrity in sequence learning (Saint-Cyr, 2003). Importantly, the FAB showed no significant relationship with the probe or retraining phases in our task, indicating that frontal-based processes play a unique role in chaining sequences, and not in generalization of sequences or in maintenance of single stimulus-response associations, respectively. This finding complements earlier work on the Kilroy task claiming that the frontostriatal dopaminergic system underlies specific sequencing deficits in Parkinson’s Disease (PD) patients and young individuals with lower dopamine levels (H. Nagy et al., 2007; O. Nagy et al., 2007; Shohamy et al., 2005) despite no effects on post-learning probe or retraining. Our results extend these findings, and may even implicate a greater role of frontal lobe integrity in learning serial orders. Coupled with lower dopamine, frontal lobe declines may result in deficient executive abilities that can affect effortful monitoring, updating, or action planning of “chains” of associations (Braver & Barch, 2002). As such, executive function losses may partly explain weakening sequence learning abilities among those with suboptimal dopamine levels, including healthy older adults. That said, aforementioned work did not identify or study relationships between executive function and sequence learning in the Kilroy task, so more work is needed to support this view. To our knowledge, only one Kilroy study included the FAB in an examination of sequencing in younger adults with and without dyslexia, but did not examine its connection to sequencing ability directly (Wenande et al., 2019).

That the FAB predicted sequencing errors in older adults, in particular, supports existing neuroimaging work in which older adults depend on their frontal lobes to learn different types of sequences (e.g., Aizenstein et al., 2004). For instance, on a sequence learning task designed with a simple repeating structure whose presence participants rarely consciously detect, older adults’ white matter integrity scores along the caudate-DLPFC tract were correlated with sequence learning throughout the entire task (Bennett et al., 2011). In addition, using fMRI, older adults’ performance on a digit ordering sequence task related to weaker prefrontal activations than younger adults’ and further, older adults’ poorer performance was tied to connectivity deficits in networks linked to prefrontal regions (Ye et al., 2020). It is also possible that older adults increasingly use frontally-based strategies to learn chained sequences as they age, fitting with the rich literature of aging studies that report higher engagement of the frontal lobes even when this does not entail better performance (e.g., Reuter-Lorenz & Cappell, 2008). In line with this view, a recent study showed that application of anodal transcranial direct simulation to the prefrontal cortices of older adults worsened participants’ chunking of discrete motor sequences (Greeley et al., 2022).

One important consideration for future development may involve clarifying the role of specific facets of executive functioning in sequencing in order to reconcile our findings with past work (e.g., Herzallah et al., 2013; Keri et al., 2008). According to Shohamy et al. (2005), our finding that trial length remained relatively constant across chaining subphases is inconsistent with a working memory deficit. That is, longer sequences do not appear to have degraded from temporal decay or from keeping additional items active through online maintenance rehearsal during each subphase. This suggests that the FAB may be capturing broader frontal-mediated executive processes than simple memory load increases with growing chaining demands. Indeed, working memory (or active maintenance) is only one of several disparate executive functions that can contribute to associative processes (Braver & Barch, 2002; Wang et al., 2019). It also remains unclear if any particular items on the FAB are driving the effect between executive ability and deterministic sequence learning. The FAB includes six different factors, each of which may be uniquely informative; for example, the conceptualization/categories subtest of the FAB has been linked to the dorsolateral frontal lobe, the lexical fluency subtest to the frontal medial areas, and inhibitory control during the go-no-go test to gray matter volume in the orbitofrontal gyrus (Terada et al., 2017). Given the narrow response range for each item (0–3), we did not evaluate which one(s) may produce the observed effect of aging on deterministic sequence learning, trusting that future neuroimaging work could more directly and effectively pinpoint these contributions. Ultimately, we believe the FAB as a behavioral measure is most apt for holistically investigating frontally-based executive function declines in aging adults.

Finally, future work may wish to distinguish between motor-based vs. judgment-based elements to more completely and coherently characterize age-deficits in sequence learning. The critical role of frontal regions to sequencing is likely task-dependent, and more research is needed to understand the specific cognitive processes that underlie frontal involvement in sequential or associative learning (Vekony et al., 2022; Wang et al., 2019). Sequence learning linked to motor-processing may operate differently from sequence learning linked to judgments (Seger, 1997) and each may be differentially affected by aging (Dang et al., 2020). Most sequencing learning tasks rely on motor responses to gradually learn deterministic regularities, and generally report age equality (Cherry & Stadler, 1995; Dennis et al., 2006; Frensch & Miner, 1994; Gaillard et al., 2009; Howard & Howard, 1989; Howard et al., 1992; Salthouse et al., 1999). To our knowledge, little research has examined how aging affects deterministic judgment-based sequences, like those in the Kilroy task. Executive functions may uniquely subserve intentional efforts to improve learning of judgment-based sequences (Frank & Claus, 2006; Hay et al., 2002; Maddox & Ashby, 2004) by impairing the ability to monitor and link chain relationships to actions’ consequences, especially in our task where visual stimuli can be easily verbalized (i.e., participants can describe the sequence as “red door – green door”). Though older adults responded well to individual sequence elements in our task (i.e., Room 1 and retraining), they were less able to group three or four elements into one single chunk or sequential representation, perhaps because that would require intact executive functions (Verwey & Dronkert, 1996). Chaining judgment-based sequences may be impaired with age because of losses in executive control and context representations (Braver & Barch, 2002) and/or effortful reasoning processes about relationships between stimuli (Mitchell et al., 2009).

In sum, individual differences in age and frontal-based executive functioning can influence the existence and magnitude of age-related declines in a deterministic, judgement-based sequence learning task, at least among a highly educated sample. Still, the exact role of executive functions in the aging of sequence learning remains to be more fully elucidated through examination of various task-related factors under which age differences may emerge, including motor vs. judgment-based sequences, implicit vs. explicit processes, or probabilistic vs. deterministic regularities (Janacsek & Nemeth, 2013, 2015; Janacsek et al., 2020). Only then will we have a more complete understanding of how aging influences the ability to learn serial operations.

Data Availability Statement

Data available on request from the authors.