Age Differences in Episodic Associative Learning

Rachel Clark and Eliot Hazeltine Michael Freedberg University of Iowa The Henry M. Jackson Foundation for the Advancement of Military Medicine, Bethesda, Maryland

Michelle W. Voss University of Iowa

Compared with young adults, older adults demonstrate difficulty forming and retrieving episodic memories. One proposed mechanism is that older adults are impaired at binding information into nonoverlapping representations, which is a key function of the hippocampus. The current experiments evaluate age differences in acquiring new memories using a novel episodic associative learning (EAL) task designed to tap hippocampal-dependent binding. The task involved repeated exposure of stimuli pairs and required the formation of new representations of each stimulus pair, as each pair was mapped to a unique keypress response. Notably, individual stimuli appeared in multiple pairs, so pair retrieval was necessary for correct response production. Experiment 1 demonstrated that older adults learned more slowly, and less overall, than young adults on this task. We also found that older adults benefited less than young adults from correct responses and as the number of intervening pairs between repetitions of a pair increased, older adults showed larger decrements in accuracy than young adults. Experiment 2 replicated these findings while minimizing motor demands and providing more practice. We also measured processing speed and spatial reconstruction to determine the involvement of specific cognitive mechanisms in observed age effects. We found that young adults with better spatial reconstruction abilities performed better on the EAL task than young adults with lower abilities and older adults overall. These findings suggest that older adults’ lower performance on the task may be partly explained by a decline in hippocampal-supported binding processes and a greater reliance on extrahippocampal learning systems.

Keywords: cognitive aging, episodic memory, learning, associative binding, hippocampus

The ability to learn new information is essential to daily life and One core feature of episodic memory is the associative binding thus a critical component of healthy aging. There is substantial of relations among elements of an episode (Davachi & Wagner, research documenting aging-related decline in forming and retriev- 2002; Konkel & Cohen, 2009). The associative deficit hypothesis ing episodic memories (Addis, Roberts, & Schacter, 2011; Craik, (ADH; Naveh-Benjamin, 2000) posits that episodic memory de- 1994; Levine, Svoboda, Hay, Winocur, & Moscovitch, 2002; clines with aging due to impaired binding processes. The ADH Light, 1991; McIntyre & Craik, 1987). However, the precise emerges from findings that older adults are worse than young cognitive mechanisms that make episodic memories especially adults at creating and retrieving links between elements of a scene vulnerable to aging are not as clear. or episode, even though their memory for individual elements is not impaired (Chalfonte & Johnson, 1996; Naveh-Benjamin, 2000). Following these findings, the ADH predicts that older adults will perform more poorly on tasks that require the formation

This document is copyrighted by the American Psychological Association or one of its allied publishers. Rachel Clark, Interdisciplinary Graduate Program in Neuroscience, Uni- of associations between items within an episode or scene. Multiple

This article is intended solely for the personal use ofversity the individual user and is not to be disseminatedof broadly. Iowa; Eliot Hazeltine, Interdisciplinary Graduate Program in studies have shown that older adults demonstrate an associative Neuroscience and Department of Psychological and Brain Sciences, Uni- deficit for different types of pairs, such as word-word pairs versity of Iowa; Michael Freedberg, The Henry M. Jackson Foundation for the Advancement of Military Medicine, Bethesda, Maryland; Michelle W. (Naveh-Benjamin, 2000; Provyn, Sliwinski, & Howard, 2007), Voss, Interdisciplinary Graduate Program in Neuroscience, Department of picture-picture pairs (Guez & Lev, 2016; Naveh-Benjamin, Hus- Psychological and Brain Sciences, and Aging Mind and Brain Initiative sain, Guez, & Bar-On, 2003), and face-name pairs (Naveh- (AMBI), University of Iowa. Benjamin, Guez, Kilb, & Reedy, 2004; for a review, see Old & This research was funded by start-up funds provided to Michelle W. Naveh-Benjamin, 2008). Voss by the University of Iowa. We thank Joan Severson at Digital However, associative binding is not always impaired with aging Artefacts in Iowa City, IA for creation and use of the spatial reconstruction (Dennis, Howard, & Howard, 2006; Howard, Howard, Dennis, paradigm. Correspondence concerning this article should be addressed to Michelle Yankovich, & Vaidya, 2004), suggesting that we need a better W. Voss, Department of Psychological and Brain Sciences, University of understanding of the specific processes involved in binding that Iowa, 300 Iowa Avenue, Iowa City, IA 52245. E-mail: michelle-voss@ decline with aging. For instance, we also previously found that uiowa.edu older adults performed similarly to young adults at a configural

144 AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 145

learning task (Clark, Freedberg, Hazeltine, & Voss, 2015). In this Motivated by this prediction, we have created a novel Episodic task, each item was associated with a unique single keypress Associative Learning (EAL) task that capitalizes on characteristics response and the items were presented in pairs so that two-finger that place heavy demands on the relational and associative binding responses were required. Importantly, individual items could re- processes of the hippocampus. This EAL task requires that partic- peat across pairs, but certain pairs occurred more frequently than ipants intentionally learn, through a process of trial and error, others. Configural learning was assessed by comparing response configural keypress responses to specific stimulus pairs (i.e., pic- times (RTs) of frequently occurring pairs with RTs to infrequent tures of faces and buildings). This task follows our configural pairs. Relatively shorter RTs to the frequent pairs expresses mem- learning task (Clark et al., 2015) in that it tests the ability to make ory for distinct combinations of pairs, despite the overlapping distinct representations of pairs that have overlapping elements. items between pairs (Hazeltine, Aparicio, Weinstein, & Ivry, Moreover, it uses the same stimuli pairs and configural responses. 2007). Older adults demonstrated relatively shorter RTs to fre- However, to target relational and associative binding that should quently occurring pairs to the same extent as young adults, which rely on the hippocampus, rather than having a single key consis- indicates that older adults were as able as young adults to bind the tently associated with each individual stimulus, this task requires a distinct response components (Clark et al., 2015). These data distinct response associated with each item pair. As shown in suggest that some forms of associative binding are not affected by Figure 1A, each picture appears within three separate pairs. To aging. In turn, we view the configural learning task as an oppor- perform accurately, a representation of each unique pair must be tunity to examine specific components of the binding process that distinct from the representations of each element individually lead to impairment with aging. (Moses & Ryan, 2006). This places heavy demands on binding Clues as to why some forms of binding show impairment processes that differentiate among pairs sharing individual ele- whereas others appear preserved can be gleaned from cognitive ments. Further, because each item pair is associated with a unique neuroscience. Relational memory theory proposes that the hip- nonoverlapping response, the stimuli must be bound together be- pocampus rapidly encodes flexible relationships between elements fore the response can be produced. This places much greater in an episode (Konkel & Cohen, 2009), such that individual demands on binding processes than the previous task in which elements of an experience are linked in a flexible manner so that each individual stimulus was consistently mapped to a keypress. they can be recalled with respect to their relationship and over- Additionally, because the hippocampus has been implicated in lapping elements can be recombined during retrieval (for a review, rapid binding compared with slower learning that is more depen- see Eichenbaum, 2000). Because the hippocampus and hippocampal- dent on the caudate (Poldrack et al., 2001; Poldrack & Rodriguez, cortical systems implicated in relational memory typically experi- 2004), examining rate of learning adds a rich dimension of time ence decline with aging (Driscoll et al., 2009; Jack et al., 1998; that enables firmer links to cognitive processes that depend on the Kennedy et al., 2009; Raz et al., 2005; Raz, Rodrigue, Head, hippocampus. More specifically, early in learning the hippocam- Kennedy, & Acker, 2004), the relational memory theory provides pus is thought to be engaged to rapidly acquire new associations an avenue to further specify the associative binding processes that between stimuli (Poldrack & Packard, 2003; Simon, Vaidya, How- would be predicted to deteriorate with aging. ard, & Howard, 2012), whereas a skill that is solidified as an This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Figure 1. (A) Stimulus-response mapping grid for Experiment 1 with representative sample stimuli. The cells correspond to the unique response for each face-building stimuli pairing. Shaded boxes represent a key press, while nonshaded boxes represent no key press. Given an AB (face-building) pair, correct response is not indicated by A or B alone. (B) Example of correct and incorrect trials in the task. Photos of faces are from the Minear and Park (2004) research database. 146 CLARK, HAZELTINE, FREEDBERG, AND VOSS

invariant “habit” may be more influenced by the striatum (Myers Participants completed 14 blocks each comprised of 45 trials et al., 2003; Poldrack, Prabhakaran, Seger, & Gabrieli, 1999; during the first visit to the laboratory and six blocks approximately Rieckmann, Fischer, & Bäckman, 2010; but see also Sadeh, Sho- seven days later at a second visit. A research assistant explained hamy, Levy, Reggev, & Maril, 2011). The current task involves the task using illustrations as examples; however, there was no multiple exposures to each unique pair, as did our configural practice phase. During each trial, a 100 ϫ 100 pixel face learning task study (Clark et al., 2015), so we can measure how stimulus was presented on the left and a 100 ϫ 100 pixel building both learning magnitude and rate are affected by aging. We also stimulus was presented on the right. Left and right positioning examine learning across multiple days to distinguish between of stimuli was counterbalanced between subjects, but for each learning limits and rate and to ensure reduced learning is not due participant faces and buildings were always presented on consis- to fatigue or lack of familiarity with the task structure. In our tent sides. Face stimuli were chosen from young adult male neutral configural learning task, we did not observe age differences in faces in the Center for Vital Longevity Face Database (Minear & learning rate or magnitude. However, we predict the greater rela- Park, 2004) and building images were chosen from a database of tional binding demands and the necessity in the EAL to bind the neutral building photographs that was previously used in a study of representations rapidly before response production will particu- age differences in face and place processing (Voss et al., 2008; see larly impair early pair acquisition for older adults. In contrast, Figure 1A). Stimuli in this study were grayscale, presented on a learning at a slower rate might reflect intact binding through black background. Participants were encouraged to create stories extrahippocampal learning systems. in order to learn the correct responses. Each face/building pair remained on the screen for 2 s, during which the participant’s first response was recorded. If the partic- Experiment 1 ipant responded correctly upon stimulus presentation, the next trial would begin following a 1-s fixation screen. If the participant Method responded incorrectly, the pair would be presented again, after a 1-s blank screen, with the correct mapping appearing directly Participants. Twenty-seven healthy young (M ϭ 24.3 years, below the stimuli for 1 s. A 1-s fixation screen appeared before the SD ϭ 3.2; 14 female) and 24 healthy older (M ϭ 64.7 years, SD ϭ next trial (see Figure 1B). A correct trial lasted for 3 s, while an 3.3; 14 female) adults participated in this study. All study proce- incorrect trial lasted for 4 s. The nine possible face/building pairs dures were in accordance with the University of Iowa’s Institu- each appeared five times per block in a randomized order. Between tional Review Board’s (IRB) policies and procedures. Participants each block, participants received feedback regarding accuracy and were recruited from Iowa City and the surrounding area using speed of response of each hand. university email advertisement and local fliers. Participants had to Analysis. All statistical analyses were performed using R meet the following eligibility criteria: (a) strong right handedness, (Version 3.3.3) and SPSS (Version 23.0.0.2). with a score of 75% or above on the Edinburgh Handedness Learning magnitude and rate. To assess age differences in Inventory (Oldfield, 1971); (b) between the ages of 18 and 30 for learning magnitude and rate, block-wise accuracy was submitted to young adults and 60 and 80 for older adults; (c) score greater than nonlinear mixed effects modeling. We utilized mixed-effects mod- 24 on the Mini-Mental State Exam (MMSE), 2nd Edition, Stan- eling instead of repeated-measures analysis of variance to more dard Version (Tombaugh, 2005); (d) normal color vision; (e) accurately model both individual and group differences, as mixed- visual acuity of 20/40 or above (corrected, if needed); (f) no effects modeling is a more robust method for differentiating ran- self-reported psychiatric and/or neurological condition, including dom effects (such as intersubject variability) from fixed effects of stroke or clinical depression; (g) no self-reported regular use of interest (such as age differences). Nonlinear mixed-effects model- medication that could affect the central nervous system (e.g., ing was run using R’s “nlme” package. Day 1 (14 blocks) and Day recent or current chemotherapy, antihypertensive medication, psy- 2 (6 blocks) were modeled separately using different model tra- chotropics); and (h) sign a written consent. Young adults had jectories between the two days. Age group was effect coded (Ϫ1 similar years of education (M ϭ 16.8, SD ϭ 1.7) as older adults for younger, ϩ1 for older). (M ϭ 17.0, SD ϭ 2.8), p ϭ .4. Older adults were additionally Accuracy was modeled using a three-parameter exponential administered the MMSE (Folstein, Folstein, & McHugh, 1975)to function with block as the time variable. In this function, a indexes This document is copyrighted by the American Psychological Association or one of its allied publishers. screen for cognitive impairment. The mean score was 29.6 (SD ϭ the asymptote for learning (accuracy where performance would This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 0.7) out of a possible 30 points. eventually stop increasing), b indexes the change in accuracy from Episodic associative learning paradigm. Similar to the con- the initial block to the asymptote (magnitude of overall accuracy figural learning task (Clark et al., 2015), participants learned to change), and c indexes the rate of change. Note, we did not restrict respond to a pair of simple categorical stimuli. In contrast, how- the asymptote to 1.00 but rather allowed it to vary freely for each ever, individual stimuli were not consistently mapped to responses. individual. Magnitude is then calculated as the difference in accu- Instead, each pair of stimuli was mapped to a configural response racy from the initial accuracy to predicted asymptote. For this (see Figure 1A). Responses were to be made with some combina- reason, the magnitude effects do not always match the expectations tion of the left and right middle and index fingers; the full set of based on the figures. Our model-fitting approach was to start with responses included four bimanual responses (one finger from each the most complex model and compare simpler models using the hand), four unimanal responses (one finger only), and one re- Bayesian information criterion (Schwarz, 1978). Starting values sponse with no keypress. Therefore, seeing an individual face or for the nlme procedure were determined by using the nonlinear building was not predictive of either left- or right-hand response; least-square procedure (nls) to fit a three-parameter exponential only a pair could indicate the correct response. function to the data (Bates, DebRoy, & Gay, 2007). AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 147

Pair-based analysis. Additionally, mixed-effects modeling is effects indicate that (a) young adults performed better than older a robust method to examine performance based on individual pairs adults; (b) pairs which had previously accurate responses were and their histories, such as whether the previous instance of a pair more likely to be accurate, and (c) fewer intervening pairs pre- received a correct or incorrect response and how many trials dicted better accuracy. There was no significant difference in occurred between successive pair appearances. This analysis al- accuracy between unimanual and bimanual trials. Results also lows for finer-grained insight into the binding processes leading to revealed a two-way interaction between age group and accuracy on individual differences in performance. To do this, we fit a gener- the previous occurrence (z ϭϪ4.1, p Ͻ .001), as well as a alized linear mixed-effects model across all trials for Day 1 and 2 near-significant interaction between age group and number of separately and included factors of age group, whether the previous trials since last occurrence (z ϭϪ1.9, p ϭ .06). These two-way occurrence of the pair was correct or incorrect, whether the trial interactions indicate that (a) a previous correct response was more was unimanual or bimanual (trials where no response was correct likely to be followed by a correct future response for young adults were not considered in this analysis because accuracy was dispro- than older adults, and (b) there was a trend in which younger adults portionately higher for these trials), and how many trials occurred were more accurate than older adults even after a greater number since the last appearance of the pair. Generalized linear mixed- of intervening trials since the last occurrence (Figure 3A). Finally, effects modeling was run using R’s “glmer” package. results revealed a near-significant three-way interaction between age group, trial type, and previous accuracy (z ϭ 1.7, p ϭ .08). Results Day 2. Learning magnitude and rate. Day 2 included only six One younger adult participant was excluded from analyses for blocks. The maximum random effects model justified by the data inability to comply with experimenter instructions. One older adult included random effects for magnitude (b) and rate (c), as well as participant voluntarily dropped out after the first task visit due to fixed effects of intercept for asymptote (a), magnitude (b), and rate task difficulty. Therefore, final analyses included 26 young adult (c) and fixed effects of age group for magnitude (b) and rate (c). participants and 23 older adults participants (Table 1). A significant main effect of age group was observed for magni- Day 1. tude, t(241) ϭ 7.3, p Ͻ .001, and rate of change, t(241) ϭϪ3.6, Learning magnitude and rate. The maximum random effects p Ͻ .001, such that young adults demonstrated a faster learning model justified by the data for Day 1 included random effects for rate than older adults across the six blocks (see Figure 2B). Thus magnitude (b) and rate (c) of change. The best fitting model on both days young adults reached asymptote sooner than older included fixed effects of intercept for asymptote (a), magnitude adults. This may be partly explained by a ceiling effect, as young (b), and rate (c), as well as fixed effects of age group for magnitude adults performed near the asymptote at the end of Day 1 and Day (b) and rate (c), but not for asymptote. A significant main effect of 2, whereas this was not the case for older adults. ϭϪ Ͻ age group was observed for rate of change, t(633) 7.0, p Pair-based analysis. As with analysis of Day 1, the first ϭ ϭ .001, but not magnitude, t(633) 1.2, p .19. These results instance of each pair was not considered, and the same factors indicate that although both groups improved performance, young were included in the model. Results revealed main effects of age adults improved at a faster rate than older adults on Day 1 (Figure group (z ϭϪ7.4, p Ͻ .001), accuracy of previous response (z ϭ 2A). It should be noted that while the linear slopes appear similar 19.0, p Ͻ .001), and number of intervening trials since previous across young and older adults, in this three-parameter exponential occurrence (z ϭϪ13.8, p Ͻ .001). These main effects were the model, rate refers to how quickly the modeled trajectory reaches same as Day 1. Again, there was no significant difference in the asymptote. accuracy between unimanual and bimanual trials. Also consistent Pair-based analysis. To better understand age-related differ- with Day 1 were significant two-way interactions between age ences in learning we evaluated performance based on the history of group and accuracy on previous occurrence (z ϭϪ4.4, p Ͻ .001), each pair. For this analysis, the first instance of each pair was between age group and number of intervening trials since last excluded from the analysis, as there was no way for the participant occurrence (z ϭϪ3.4, p Ͻ .001; see Figure 3B), between trial type to know the correct response on its first presentation. Results and accuracy on previous occurrence (z ϭϪ2.3, p ϭ .02), and a ϭϪ Ͻ revealed main effects of age group (z 6.2, p .01), accuracy trending interaction between trial type and number of intervening ϭ Ͻ of previous response (z 16.8, p .001), and number of trials trials since last occurrence (z ϭϪ1.9, p ϭ .06). Finally, there was This document is copyrighted by the American Psychological Association or one of its allied publishers. ϭϪ Ͻ since previous appearance (z 12.1, p .001). The main a significant three-way interaction between age group, trial type, This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. and previous accuracy (z ϭ 2.3, p Ͻ .001). Whereas young adults had a significant interaction between trial type and accuracy Table 1 (z ϭϪ3.2, p ϭ .001), such that there was a greater effect of Participant Demographics previous accuracy for bimanual than unimanual trials, older adults Age, Years educ., MMSE, showed equivalent effects of previous accuracy for bimanual and Age group N Sex M (SD) M (SD) M (SD) unimanual trials (z ϭϪ.6, p ϭ ns).

Experiment 1 Young 26 13 F, 13 M 24.2 (3.3) 16.8 (1.7) NA Discussion: Experiment 1 Older 23 13 F, 10 M 64.5 (3.3) 17.2 (2.7) 29.6 (.7) Experiment 2 We designed the current task to place high demands on rela- Young 18 11 F, 7 M 22.4 (3.7) 15.0 (2.1) NA tional binding processes, and this resulted in significant learning Older 18 9 F, 9 M 65.0 (3.8) 17.9 (3.1) 29.4 (1.3) impairments for older adults. Notably, in Clark et al. (2015) Note. MMSE ϭ Mini-Mental State Exam; F ϭ female; M ϭ male. individual stimuli were mapped in a one-to-one fashion to a similar 148 CLARK, HAZELTINE, FREEDBERG, AND VOSS

Figure 2. Proportion correct by block for young and older participants on the two days of Experiment 1. The solid lines represent the raw accuracy data, the shaded area represents standard error of the mean (SEM), and the dotted lines represent the generalized linear model fit to the data for each age group. See the online article for the color version of this figure.

set of responses as the current task, and young and older adults detect three-way interactions. While this may partly account for acquired configural associations at a similar rate. Like that study, the near-significant interactions observed, it does not affect our the current task requires flexible binding and the resolution of main outcomes. interference between overlapping elements across pairs. However, the EAL places much greater demands on binding processes. As Experiment 2 predicted, we found that across both Day 1 and Day 2, older adults learned at a slower rate than young adults. Most notably, young The results of Experiment 1 show a robust age effect on EAL adults improved rapidly early in the task (between Blocks 2–8), magnitude and rate. However, there are potential alternative whereas older adults did not demonstrate this rapid early learning. explanations to the interpretation that age effects are due pri- Both human and animal studies have provided evidence that the marily to relational binding processes. First, the task is chal- hippocampus would support this early learning (for a review, see lenging and the stimuli are presented quickly. It is possible that Poldrack & Packard, 2003). Young adults also successfully older adults were more affected than young adults by the speed learned more pairs than older adults, reaching significantly higher and complexity of the task. It is also possible that some older accuracy by the end of Day 2 (see Figure 2B). adults did not understand the instructions and thus took a longer We did not observe strong effects of trial type, though uni- time to figure out how to use feedback to learn the stimulus- manual trials were consistently numerically more likely to be response mappings. A lack of understanding could have con- correct than bimanual trials. Note that trial type did not interact tributed to slower learning at the beginning of the task. How- directly with age, which suggests that age-related differences in ever, we note that older adults did not show signs of catching up bimanual dexterity are not a main reason for the age differences in throughout the task so it does not appear that the observed age performance. Nonetheless, we address this in more depth in Ex- effects are due primarily to lower initial understanding. Regard- This document is copyrighted by the American Psychological Association or one of its allied publishers. periment 2. Additionally, we found for both young and older adults less, to acquaint participants with the task prior to the test This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. that getting a pair correct meant a greater likelihood of responding blocks, we designed a second experiment that included a short correctly to the next occurrence of that pair. Interestingly, young practice session with slower stimulus presentation and a longer adults benefitted more from previously accurate responses, and response window prior to the actual task. Also, to match stim- this was particularly true for bimanual trials. The feedback ap- ulus type with our previous study of configural learning (Clark peared to be more effective for young adults than older adults, et al., 2015), we used color images of young and older male which indicates that older adults were slower at acquiring a robust faces for both stimuli in each pair, rather than black and white memory of individual pairs. Further, older adults were more neg- images of faces and buildings. atively affected by intervening trials between repeated occurrences Experiment 1 included responses that were either unimanual of particular pairs. In other words, the feedback seemed to be (only 1 finger) or bimanual (1 finger from each hand). While we effective only for a short time, particularly for older adults. Finally, did not observe an interaction between age group and trial type, we note that while our sample sizes for Experiment 1 were appro- some studies have shown that older adults have difficulty priate for detecting main effects and two-way interactions, it is coordinating simultaneous movements with both hands possible that our design may not have had sufficient power to (Bangert, Reuter-Lorenz, Walsh, Schachter, & Seidler, 2010). AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 149

Experiment 1 Day 1 Experiment 1 Day 2 1 A 1 B

0.8 0.8

Young Correct 0.6 0.6 Young Incorrect Older Correct 0.4 0.4 Older Incorrect

0.2 0.2 Accuracy (proportion correct)

0 0 0 1 2 3 4-6 7-11 12-20 21-30 0 1 2 3 4-6 7-11 12-20 21-30 Number of Intervening Trials Number of Intervening Trials C Experiment 2 Bimanual version D Experiment 2 Unimanual version 1 1

0.8 0.8

0.6 Young Correct 0.6 Young Incorrect

0.4 Older Correct 0.4

Older Incorrect 0.2 0.2 Accuracy (proportion correct)

0 0 0 1 2 3 4-6 7-11 12-20 21-30 0 1 2 3 4-6 7-11 12-20 21-30 Number of Intervening Trials Number of Intervening Trials

Figure 3. Accuracy as a function of the number of intervening trials since the last occurrence of the stimulus pair and whether the response was correct on the last occurrence for young and older participants. (A) Experiment 1, Day 1. (B) Experiment 1, Day 2. (C) Experiment 2, bimanual version. (D) Experiment 2, unimanual version.

Thus, the bimanual component of some responses may have Watson, Voss, Warren, Tranel, and Cohen (2013) found that contributed to age effects. Therefore, Experiment 2 measures individuals with hippocampal damage were more likely to performance on two versions of the task: one with only uni- commit a specific error on this task that involved “swapping” manual responses, and one with both unimanual and bimanual the relative positions of two items. Several other studies have responses (similar to Experiment 1). also shown that this relational memory ability is related to In addition, to better understand the specific cognitive pro- performance on other canonical neuropsychological tasks of cesses involved with acquisition and to enhance construct va- relational memory and to hippocampal volume (Monti et al., lidity of our task, we measured additional cognitive constructs 2015) and hippocampal tissue integrity (Schwarb, Johnson, including processing speed and relational memory. Processing McGarry, & Cohen, 2016; Schwarb et al., 2015) in healthy speed determines the capacity an individual has for processing adults. Together, these findings provide evidence that the hip- information at a given time, and much evidence demonstrates pocampus is critical for successful performance on the spatial slowing with age (Kail & Salthouse, 1994; Salthouse, 1996). This document is copyrighted by the American Psychological Association or one of its allied publishers. reconstruction paradigm used here, and that these hippocampal Some propose that this slowing can account for a broad range of This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. processes decline even with healthy aging. age-related changes in cognition (Salthouse, 1996). However, By examining relationships of performance on the processing there is also substantial evidence that age-related changes in speed and spatial reconstruction paradigms with performance cognition occur in addition to general slowing. As previously described, older adults tend to demonstrate difficulty creating on our EAL task, we can both further evaluate construct validity and retrieving arbitrary bindings, which may be attributed of our task and better specify the parameters of age-related more specifically to a deficit in binding (Naveh-Benjamin, differences in associative binding. That is, if we find that 2000). performance on our EAL task is strongly related to processing We administered a spatial reconstruction paradigm to tap into speed performance, it may be that age differences can be relational memory processing. In previous work in our lab, we explained by general age-related changes in processing speed. found that differences in performance between young and older However, if we find that performance on our task is related to adults were significantly larger for this spatial reconstruction spatial reconstruction ability, this would suggest that paradigm than for a pattern separation task that is also thought hippocampal-dependent relational binding is a critical factor of to involve the hippocampus (Clark et al., 2017). In addition, older adults’ poorer learning in this task. 150 CLARK, HAZELTINE, FREEDBERG, AND VOSS

Method The bimanual learning task was identical to Experiment 1 ex- cept that face stimuli were used for both stimuli, rather than face Participants. Participants were recruited from Iowa City and and building stimuli. Faces were chosen from young and older the surrounding area using email advertisement and fliers. Study adult male neutral faces in the Center for Vital Longevity Face procedures were in accordance with the University of Iowa’s IRB Database (Minear & Park, 2004). Full-color images were used policies and procedures. Participants met the same eligibility re- (Figure 4A). As in Experiment 1, stimulus presentation lasted 2 s. quirements as in Experiment 1. Twenty-one healthy young (M ϭ If the response was incorrect, the correct mappings were shown for 22.3 years, SD ϭ 3.5; 12 female) and 18 healthy older (M ϭ 65.0 1 s, followed by a 1-s fixation screen and the next trial. In contrast years, SD ϭ 3.7; 11 female) adults participated in this study. Two to Experiment 1, we shortened the delay between stimulus presen- of the 21 young adults enrolled in the study dropped out after the tation and feedback (10 ms) to reduce the working memory de- initial visit for unspecified reasons. Of the participants who com- mands and increase the saliency of the feedback (see Figure 4B). pleted all study visits, young adults (N ϭ 19; M ϭ 22.2, SD ϭ age In the unimanual task, while stimulus presentation was identical to 3.8, 11 female) had a mean of 14.8 years of education (SD ϭ 2.2), the bimanual task, all responses were limited to one finger such while older adults (N ϭ 18; M ϭ 65.0, SD ϭ 3.8; 9 female) had age that each of the nine face pairs were matched to a single keypress a mean of 17.9 years of education (SD ϭ 3.1). This was a statistically significant difference (p Ͻ .01); however, this is not with either the left- or right-hand fingers (excluding the thumb), necessarily reflective of sociodemographic differences because including a no-press response (see Figure 4C). As can be seen in young adults were still completing their education. Older adults Figure 4C, eight fingers (4 from each hand) were used in the were administered the MMSE (Folstein et al., 1975) as a screen for unimanual task. Unique faces were used in the bimanual and cognitive impairment. The mean score was 29.4, out of a possible unimanual tasks and were counterbalanced between participants. 30 points (SD ϭ 1.3). Processing speed. To measure processing speed, participants Participants completed four visits to the lab. The first visit completed a paper-and-pencil pattern and letter comparison task included consent, demographic surveys, self-report questionnaires, (Salthouse, 1996). The first section (pattern comparison) contained and MMSE (for older adults only). The second visit consisted of two trials, each consisting of one page with 30 pairs of patterns. practice sessions for bimanual and unimanual versions of the task. The participant was to compare the patterns and correctly write Participants were counterbalanced such that half the participants either “S” (same) or “D” (different) on a line between the items for completed the bimanual version during Visit 3 and the unimanual as many items as possible within 30 s. The second part of the task version during Visit 4 and vice versa for the other participants. (letter comparison) consisted of two trials, each consisting of one EAL paradigm. page with 15 pairs of letter strings. Again, the participant was to Practice phase. Novel animal and transportation stimuli were write S or D on a line between the letter strings for as many items used in the practice phase. The unimanual practice phase included as possible within 30 s. two blocks of 45 trials each and the bimanual practice phase Spatial reconstruction. To measure relational memory, we included four blocks of 45 trials each. During each practice trial, used a touchscreen (iPad) version of the spatial reconstruction the stimuli pairs were presented for 4 s and feedback mappings paradigm based on the original reconstruction task used by Hut- were shown for 2 s. tenlocher and Presson (1979) and a computerized version created Learning phase. As before, the critical computational de- by Monti et al. (2015). This task has been widely used to measure mands for our task included the binding of stimuli pairs with a spatial reconstruction (Lucas et al., 2016; Schwarb et al., 2016; unique, correct response. No individual stimulus was correlated Smith & Milner, 1981; Watson et al., 2013). The touchscreen with a particular response. As in Experiment 1, participants were version was created through a custom Digital Artifacts BrainBase- encouraged to mentally create stories to help learn the correct line application (https://www.brainbaseline.com/). The task con- responses. sisted of two practice trials followed by three blocks with 10 trials This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Figure 4. (A) Stimulus-response mapping grid for bimanual version of Experiment 2 with Stimuli Set 1. (B) Example of incorrect and correct trials. (C) Stimulus-response mapping grid for unimanual version of Experi- ment 2 with Stimuli Set 2. Photos are from the Minear and Park (2004) research database. See the online article for the color version of this figure. AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 151

each (an example trial is shown in Figure 5A). On each trial, three z-score normalization across age groups. We did the same for or five “squiggle” objects appeared on the screen. Participants the letter comparison: we averaged the number of items correct were instructed to touch each object with either their finger or from both trials for each individual and transformed each indi- stylus to confirm they studied the exact location. Participants had vidual’s average score to a z-score across age groups. We up to 1 min to study each layout. The objects then disappeared for computed an overall processing speed z-score value for each 5 s. After the delay, the original objects appeared at the bottom of individual by averaging across z-scores from the pattern and the screen and participants used their finger or stylus to drag each letter comparison tasks. The nlme package was run with the object to the location that best matched the original placement of same parameters as the rate and magnitude analysis, but with the object. For each trial, the objects were selected without re- factors of age group, processing speed, and the interaction of placement so that no trial included multiples of the same object. age group and processing speed. Objects could be repeated across trials, though each trial consisted of a unique layout. Spatial reconstruction. Error types were calculated based Analysis. All statistical analyses were performed using R on metrics developed by Watson et al. (2013) including mis- (Version 3.3.3) and SPSS (Version 23.0.0.2). placement (the distance in pixels between each item’s studied EAL task. location and the reconstructed location), edge resizing (the Learning magnitude and rate. Similar to Experiment 1, non- length of the vector in pixels between each pair of items in the linear mixed-effects modeling was used to estimate age differences reconstructed layout compared with the original configuration, in learning. The bimanual and unimanual versions of the task were and summed across all relationships on each trial), rearrange- analyzed independently, with the same nonlinear modeling proce- ment (the change in overall layout of the stimuli as defined by dures as Experiment 1. a sign change in either the x-or y-dimension at any vertex), and Pair-based analysis. Similar to Experiment 1, we evaluated proportion of swaps (misassignment when the correct location performance based on individual pairs and their histories in of an item was chosen but the identity of the object was order to probe more specific differences between how older and incorrect; Schwarb et al., 2016; see Figure 5B). For proportion young adults learned. We fit a generalized linear mixed-effects of swaps, different set sizes were accounted for by computing model across all trials for bimanual and unimanual versions the number of swaps per pairwise relation (3 pairwise relations separately and included factors of age group, trial type (only for for Set Size 3, 10 pairwise relations for Set Size 5). To reduce the bimanual version; “no response” trials were not considered multiple comparisons and obtain more stable measures of spa- in any analysis), whether the pair was previously correct or incorrect, and how many trials occurred since the last appear- tial reconstruction by canceling out weaknesses in individual ance of the pair. Generalized linear mixed-effects modeling was error metrics, we generated a composite error (SRerror). Be- run using R’s “glmer” package. cause each error type was measured in different units, we Processing speed. To obtain a processing speed score for z-score normalized across age groups for each error metric and each individual for the pattern comparison task, we first aver- the composite error variable was created by taking the average aged each person’s number of items correctly completed for of the variables. The nlme model included factors of age group, Trials 1 and 2. We then transformed that average score by SRerror, and the interaction of age group and SRerror. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Figure 5. (A) Spatial reconstruction paradigm example trial. (B) Example of reconfiguration with different error types demonstrated. See the online article for the color version of this figure. 152 CLARK, HAZELTINE, FREEDBERG, AND VOSS

Results group and number of intervening trials (z ϭϪ5.4, p Ͻ .001), trial type and number of intervening trials (z ϭ 6.1, p Ͻ .001), trial type Data were lost for one young adult due to a technical problem and accuracy on previous occurrence (z ϭ 2.9, p Ͻ .01), and (see Table 1 for final demographics). accuracy on previous occurrence and number of intervening trials EAL task. (z ϭ 6.4, p Ͻ .001). These results indicate that while both young Learning magnitude and rate. and older adults performed better on the unimanual trials than on Bimanual version. The maximum random effects model jus- the bimanual trials, this difference was especially pronounced in tified by the data included random effects for asymptote (a), older adults, and the previous inaccurate responses and number of magnitude (b), and rate (c) of accuracy change for participants. intervening trials resulted in lower performance, particularly on the The best fitting model for the data included fixed effects for bimanual trials. intercept for asymptote (a), magnitude (b), and rate (c), as well Finally, results revealed two significant three-way interactions. as fixed effects of age group for magnitude (b) and rate (c), but First, we found a significant interaction between age group, pre- not for asymptote. Significant main effects of age group were vious accuracy, and number of intervening trials (z ϭ 4.7, p Ͻ observed for model-predicted magnitude, t(477) ϭ 3.4, p Ͻ .001). To explain this interaction, young adults did not have an .001 and rate, t(477) ϭϪ5.8, p Ͻ .001 (Figure 6A). These results indicate that the difference in performance between interaction between previous accuracy and number of intervening ϭ ϭ starting position and model-predicted asymptote was greater for trials (z 1.3, p ns), whereas older adults did have a significant ϭ Ͻ older adults than young adults, but young adults improved interaction (z 6.4, p .001; see Figure 3C). Second, we found accuracy at a faster rate than older adults, replicating our a significant three-way interaction between trial type, previous ϭϪ Ͻ findings from Experiment 1. accuracy, and number of intervening trials (z 3.3, p .001). Unimanual version. The maximum random effects model This interaction seems to be driven by a stronger interaction justified by the data included random effects for rate. The best between previous accuracy and number of intervening trials for ϭ Ͻ ϭ fitting model for the data included fixed effects for intercept for bimanual trials (z 4.2, p .001) than for unimanual trials (z ϭ asymptote (a), magnitude (b) and rate (c), as well as fixed 1.7, p .08). effects of age group for magnitude (b) and rate (c), but not for Unimanual version. Results revealed main effects of age ϭϪ Ͻ ϭ Ͻ asymptote. Similar to the bimanual condition, significant main group (z 4.4, p .001), previous accuracy (z 14.9, p ϭϪ Ͻ effects of age group were observed for magnitude, t(464) ϭ 5.2, .001), and number of intervening trials (z 23.1, p .001). As p Ͻ .001, and rate, t(464) ϭϪ4.6, p Ͻ .001 (see Figure 6B), with the bimanual version, young adults outperformed older adults showing again that young adults learned episodic associative and previous accuracy predicted better accuracy on future trials. In information faster than older adults. addition, significant two-way interactions were found between age Pair-based analysis. As with analysis of Experiment 1, the group and previous accuracy (z ϭϪ2.1, p ϭ .03), age group and first instance of each pair was not considered. number of intervening trials since the last appearance (z ϭϪ6.1, Bimanual version. Results revealed main effects of all factors: p Ͻ .001), and between previous accuracy and number of inter- age group (z ϭϪ6.1, p Ͻ .001), accuracy of previous response vening trials (z ϭ 5.2, p Ͻ .001). Finally, a three-way interaction (z ϭ 12.7, p Ͻ .001), number of intervening trials (z ϭϪ19.7, p Ͻ was found between age group, previous accuracy, and number of .001), and trial type (z ϭ 3.2, p ϭ .001). As before, young adults intervening trials (z ϭ 3.8, p Ͻ .001), reflecting that older adults performed better than older adults, pairs that had previous accurate had a significant interaction between previous accuracy and num- responses were more likely to be accurate, accuracy decreased ber of intervening trials (z ϭ 5.3, p Ͻ .001), which young adults with more intervening trials, and unimanual responses were more did not have (z ϭ 1.5, p ϭ ns; see Figure 3D). This indicates that accurate than bimanual responses. All two-way interactions were for young adults, regardless of previous accuracy, performance significant: age group and trial type (z ϭ 2.9, p Ͻ .01), age group decreased as the number of intervening trials increased. Older and accuracy on previous occurrence (z ϭϪ2.1, p ϭ .04), age adults, on the contrary, had a larger decrease in performance with

This document is copyrighted by the American Psychological Association or one of its allied publishers. A B

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Bimanual Version of EAL task Unimanual Version of EAL Task 1.0 1.0 Young Model 0.8 Young 0.8

0.6 0.6 Older Model 0.4 Older 0.4

0.2

Proportion Correct Proportion 0.2 Proportion Correct

0.0 0.0 2468101214 2468101214 Block Block

Figure 6. Learning curve graphs for the (A) bimanual, and (B) unimanual versions of Experiment 2. See the online article for the color version of this figure. AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 153

greater intervening trials when the previous response was correct, strates that processing speed was not related to learning rate which was likely driven by the floor effect of performance on trials differently for young and older adults (see Figure 7B). which were previously incorrect. Spatial reconstruction. Processing speed. Bimanual version. Significant main effects were found for age Bimanual version. We found a main effect of age group on group, t(473) ϭϪ3.6, p Ͻ .001, and SRerror on learning rate, the magnitude (b) parameter, t(473) ϭ 2.7, p Ͻ .01, but no main t(473) ϭϪ3.6, p ϭ .002. These main effects indicate that young effect of processing speed for b (t(473) ϭϪ.2, p ϭ ns), suggesting adults outperformed older adults in terms of rate of learning and that that, across both groups, processing speed does not predict mag- across both age groups individuals with a lower error rate (better nitude of change in accuracy across the EAL task. Additionally, spatial reconstruction) had a faster learning rate on the EAL task. there was no significant interaction between age group and pro- Results also revealed a significant interaction between age group and SRerror, t(473) ϭ 2.7, p Ͻ .01, such that older adults performed cessing speed for magnitude (t(473) ϭ .23, p ϭ ns). For parameter equivalently regardless of SRerror rate, but young adults with less c (learning rate), we found significant main effects of age group, SRerror learned the EAL task significantly faster than both young t(473) ϭϪ2.99, p Ͻ .01 and processing speed, t(473) ϭ 2.3, p ϭ adults with more SRerror and older adults (Figure 8A). .02, but there was not a significant interaction between age group Unimanual version. For b, we found only a significant main and processing speed on learning rate (t(473) ϭϪ.66, p ϭ ns). effect of age group, t(460) ϭ 2.1, p ϭ .03. For c, we found a main This finding indicates that processing speed was related to rate of effect of age group, t(460) ϭϪ3.4, p Ͻ .001, as well as an learning but that this relationship was not significantly different interaction between age group and SRerror, t(460) ϭ 2.2, p ϭ .03 between young and older adults (Figure 7A). (see Figure 8B). These results are consistent with the findings from Unimanual version. As in the bimanual version, for the the bimanual version, and suggest that, in comparison to process- magnitude parameter we found a main effect of age group, ing speed, errors in spatial reconstruction, which may more heavily t(460) ϭ 2.7, p Ͻ .01 but no main effect of processing speed tap hippocampal-based binding processes, were more related to (t(460) ϭϪ1.3, p ϭ ns). We did not see a significant interac- age differences in EAL. tion between age group and processing speed for b, t(460) ϭ 1.8, p ϭ .07. For c, as with the bimanual version, we found significant main effects of age group, t(460) ϭϪ3.1, p ϭ .002 Discussion: Experiment 2 and processing speed, t(460) ϭ 3.1, p ϭ .002 and there was no In Experiment 2, we replicated the significant age differences in interaction between age group and processing speed learning on the EAL task that we observed in Experiment 1. When (t(460) ϭϪ.7, p ϭ ns). Consistent with the results of the learning associations between overlapping stimulus pairs and their bimanual task, these findings suggest that processing speed is unique configural responses, our results demonstrate that young related to the rate of learning. Importantly though, as with the adults learn quicker and are more resilient to intertrial interference bimanual version, the lack of a significant interaction demon- than older adults.

A B Processing Speed with Processing Speed with Bimanual Version of EAL Task Unimanual Version of EAL Task 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 This document is copyrighted by the American Psychological Association or one of its allied publishers. 0.2 This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 0.2 Proportion Correct Proportion Correct 0.0 0.0 2468101214 2468101214 Block Block Young, faster processing speed (1 SD up) Young, faster processing speed (1 SD up) Young, slower processing speed (1 SD down) Young, slower processing speed (1 SD down) Older, faster processing speed (1 SD up) Older, faster processing speed (1 SD up) Older, slower processing speed (1 SD down) Older, slower processing speed (1 SD down)

Figure 7. Nonsignificant interaction between age group and processing speed for accuracy on (A) bimanual, and (B) unimanual versions of the associative learning task. Values are based on model-estimated proportion correct for ϩ1 SD and Ϫ1 SD from the mean of processing speed. ϩ1 SD ϭ better processing speed. See the online article for the color version of this figure. 154 CLARK, HAZELTINE, FREEDBERG, AND VOSS

A B SRerror with SRerror with Bimanual Version of EAL Task Unimanual Version of EAL Task 1.0 1.0 0.8 0.8

0.6 0.6

0.4 0.4

Proportion Correct 0.2 0.2 Proportion Correct

0.0 0.0 2 4 6 8 10 12 14 2 4 6 8 101214 Block Block

Young, lower SRerror (1 SD down) Young, lower SRerror (1 SD down) Young, higher SRerror (1 SD up) Young, higher SRerror (1 SD up) Older, lower SRerror (1 SD down) Older, lower SRerror (1 SD down) Older, higher SRerror (1 SD up) Older, higher SRerror (1 SD up)

Figure 8. Interaction between age group and SRerror for the (A) bimanual, and (B) unimanual version of the episodic associative learning (EAL) task. Values are based on model-estimated proportion correct for ϩ1 SD and Ϫ1 SD from the mean of SRerror. Ϫ1 SD ϭ better spatial reconstruction. See the online article for the color version of this figure.

The results of Experiment 2 ruled out alternative explanations of Finally, in Experiment 2 we found that while processing speed the findings of Experiment 1. First, to acquaint participants with was positively related to better learning in both young and older the task prior to the test blocks, and to reduce the likelihood that adults, this relationship did not differ between age groups. Thus, older adults simply did not understand the task, we implemented a we can conclude that processing speed does not seem to be the key short practice session. Second, we also slowed down the stimulus factor underlying age-related differences in EAL performance. On presentation during the practice trials to help participants become the contrary, we did find an interaction between age group and accustomed to the trial and feedback structure. One of the most SRerror on accuracy in the EAL task, such that younger adults striking findings from Experiment 1 was that young adults im- with better SR performance learned more rapidly on the EAL task. proved their performance much earlier in the task than older adults. Interestingly, learning rate did not differ as much for older adults We predicted that if the age effects were due mainly to speed and based on SRerror. Overall, these findings suggest that our EAL lack of understanding of the task, adding a practice session would task may be tapping into similar hippocampal-specific processes increase older adults’ performance early in the task. We did not that are required for successful spatial reconstruction, particularly find evidence to support that prediction. After the practice session, for young adults. older adults still demonstrated slower early learning than young adults, and less learning overall than young adults. This strength- General Discussion ens the support for our hypothesis that rapid binding of overlapping pairs as here mapped to nonoverlapping responses is partic- Older adults tend to demonstrate poorer learning than young ularly impaired with aging. adults, particularly when acquiring various pieces of information To examine whether the use of bimanual responses was dispro- within an episodic experience (Chalfonte & Johnson, 1996; portionately affecting older adults, we designed Experiment 2 to Naveh-Benjamin, 2000). The primary goal here was to design a include a version of the task that consisted of responses from only task that taps binding processes necessary to differentiate events a single hand. These unimanual responses involved a single finger, with shared elements, whereby the correct response was deter- This document is copyrighted by the American Psychological Association or one of its allied publishers. so no bimanual or even unimanual multidigit coordination was mined by the pair of overlapping elements. Note that the elements This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. required. While both young and older adults performed better on individually were uninformative. We predicted that such a task unimanual trials than bimanual trials, the age effect on learning would be highly sensitive to age, and we found strong support for remained large even for the unimanual version of the task (see this prediction. Importantly, the EAL task was designed to share Figure 6B). stimuli and responses with our previous configural learning task, The pair-based analyses in Experiment 2 substantiate the find- which did not show age-related differences in learning (Clark et ings from Experiment 1: older adults do not receive as much al., 2015). Critically, in the previous task individuals learned direct benefit as young adults from a correct response, and older adults mappings between stimuli and responses, which may allow bind- lose the benefit of the feedback with fewer intervening trials ing to occur at the response level. In contrast, here each stimulus between repeated pairs. One way to extend this finding in the must be bound separately to unique responses, determined entirely future would be to develop a similar task that repeats pairs when by the combination of stimuli. Although this may have created the subject performs incorrectly. The present data suggests that additional binding processes, such as both item-item binding be- older adults would benefit more from this protocol than younger tween elements and a spatial location binding in the stimulus- adults. response association, our data suggest the pair-dependent response AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 155

required by the current task fundamentally altered the type of about the pairs may leave open the possibility that older adults binding rather than only the number of bindings. The number of performed worse due to a difficulty in implementing this strategy bindings associated with a single trial would be the same for a rather than basic binding processes. While we do not know what unimanual trial in the current EAL task and a bimanual trial in the specific strategies participants implemented, future work could configural response learning task used by Clark et al. (2015) and describe and evaluate the strategies utilized by participants. both tasks also share the involvement of spatial location bindings. Our results generally support the ADH (Naveh-Benjamin, 2000), Yet, age-related differences in learning were only present in the but our approach further specifies which aspects of binding may be EAL. Further, a single keypress response, such as in the unimanual most affected by aging. Binding that relies on the hippocampus is a task, involves fewer bindings than a bimanual response, but we process that is significantly affected by aging (Foster, Picklesimer, observed equivalent age differences for unimanual and bimanual Mulligan, & Giovanello, 2016; St. Jacques, Rubin, & Cabeza, 2012). responses. This pattern of results suggests age differences here are Although our results do not speak to the specific hippocampal pro- not driven by the number of bindings and binding processes related cesses involved, this binding deficit explanation is consistent with our to mapping a stimulus to a spatial response are likely not a strong data as the EAL task likely places high demands on the hippocampal component of the observed age difference. system by requiring the formation of associations between arbitrarily In all versions of the EAL task here, older adults had a slower paired stimuli, as well as the resolution of interfering representations learning rate than young adults, and learned fewer responses that overlap between different pairs. Real life tasks also place heavy throughout two days of task training. The age effect we observed demands on flexible binding processes. Consider meeting a new was robust and was not attenuated by the modifications we made coworker. You may form a representation of that individual in the to the task for Experiment 2 that allowed for practicing of the task context of work. Later you see the individual at the gym. Your and reduced the need for bimanual response coordination (Bangert behaviors toward this individual may differ based on the context, and et al., 2010). While many studies have evaluated age differences of you may need to reconfigure your relational memory to now include learning single items or item pairs, our task assesses the ability to an additional representation of this individual. Our task aimed to make distinct representations of pairs that have overlapping ele- simulate acquisition of these overlapping but unique memory repre- ments. This multiple-trial exposure to each unique pair in the sentations. current task allows for the measurement of both rate and magni- Our results also align with previous findings of larger age tude of learning. Learning that occurs continuously over many effects under intentional learning demands compared with inciden- presentations may be representative of encoding and reencoding tal learning demands (Naveh-Benjamin, 2000; Old & Naveh- episodes. Additionally, we have operationalized this process as Benjamin, 2008). In the current task, participants were aware that episodic memory because it requires rapid binding of elements into they were supposed to be learning the associations and responses. distinct, nonoverlapping representations, which is a characteristic Thus, this task involves intentional learning and we observed a feature of forming and maintaining distinct episodic memories large effect of age group. In contrast, in our previous configural (Kumaran, Hassabis, & McClelland, 2016). However, it is possible response-learning task (Clark et al., 2015), the learning measure that individuals engage multiple learning systems at different did not rely on participants’ awareness or effort, and we saw no stages in the task, particularly given the opportunities for addi- significant age effect. This distinction between intentional and tional study when errors were made. For instance, after many incidental learning demands may partially account for the marked presentations and successful encoding and retrieval, some individ- difference in age effects on our current task and our configural uals may retrieve the associations from procedural rather than response learning task. episodic memory. Additional experiments that incorporate neuro- Dealing with overlapping elements appropriately may require imaging will be needed to more specifically map binding processes hippocampal-based processes similar to resolving interference in pat- required in this task to learning systems in the brain and the extent tern separation (Kirwan & Stark, 2007). As with pattern separation, an to which their involvement differs between young and older adults. individual faced with stimuli that share components with a previous With regard to limitations of the present study, we did not pattern but require a distinct response must be able to access distinct measure the flexibility of the memory representations after training memory representations. However, the representations must also be sessions. One way to evaluate flexibility would be to test whether maintained and updated flexibly since items that overlap across pairs participants would respond correctly, given the same pair but with or episodes will provide rich contextual information for developing This document is copyrighted by the American Psychological Association or one of its allied publishers. each stimulus on the opposite side of the screen. This type of test abstract knowledge of interitem relationships. The hippocampus is This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. would be useful for elucidating how associations are encoded and well known for this balance of rapid and flexible encoding of repre- whether older adults encode associations more rigidly or “fused” sentations (Konkel & Cohen, 2009; Konkel, Warren, Duff, Tranel, & than young adults (Campbell, Hasher, & Thomas, 2010; Eichen- Cohen, 2008; Schapiro, Turk-Browne, Botvinick, & Norman, 2017). baum & Bunsey, 1995; Moses & Ryan, 2006). If older adults do Therefore, it will be critical to further test the flexibility of learned rely more than young adults on extrahippocampal systems, such as associations in this task, how such flexibility is affected by aging, and the basal ganglia, it could be that the representations older adults the extent to which aging impacts the brain systems involved in these make are less flexible, as the striatum encodes more slowly and binding processes. rigidly than the hippocampus (Henke, 2010; Poldrack & Packard, 2003). This could be tested by examining whether age moderates Conclusion the relationship between learning rate on the EAL task and learning rate on tasks thought to be more heavily influenced by the We found age-related differences in learning using a paradigm striatum (Knowlton, Mangels, & Squire, 1996; Seger, 2006). Ad- that requires flexible associative binding. Older adults learned ditionally, our encouragement for participants to create stories more slowly than young adults and were more susceptible to 156 CLARK, HAZELTINE, FREEDBERG, AND VOSS

interference in learning across trials. This pattern of age-related Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). “Mini-mental differences in learning is strikingly different from our previous state.” Journal of Psychiatric Research, 12, 189–198. http://dx.doi.org/ study (Clark et al., 2015), which was designed with similar stimuli 10.1016/0022-3956(75)90026-6 and responses to the current task. Importantly, the previous task Foster, C. M., Picklesimer, M. E., Mulligan, N. W., & Giovanello, K. S. comprised of direct one-to-one mappings between stimulus and (2016). The effect of age on relational encoding as revealed by hippocampal functional connectivity. Neurobiology of Learning and Mem- response, independent of pairing. On that task, older adults dem- ory, 134, 5–14. http://dx.doi.org/10.1016/j.nlm.2016.07.026 onstrated the same extent of configural learning as young adults. Guez, J., & Lev, D. (2016). A picture is worth a thousand words? Not when The current task, on the contrary, which consisted of mappings that it comes to associative memory of older adults. Psychology and Aging, were entirely dependent on the distinct pairing of overlapping 31, 37–41. http://dx.doi.org/10.1037/pag0000069 items, resulted in a significant age difference in learning. This Hazeltine, E., Aparicio, P., Weinstein, A., & Ivry, R. B. (2007). Configural comparison reveals a critical age-sensitive aspect of relational response learning: The acquisition of a nonpredictive motor skill. Jour- binding, which is a key function of the hippocampus. Although we nal of Experimental Psychology: Human Perception and Performance, did not directly relate performance to brain structure or function, 33, 1451–1467. http://dx.doi.org/10.1037/0096-1523.33.6.1451 the observed age differences in performance that may be a mani- Henke, K. (2010). A model for memory systems based on processing festation of decreased hippocampal structure or function with modes rather than consciousness. Nature Reviews Neuroscience, 11, aging provides a strong foundation for future studies to do so. 523–532. http://dx.doi.org/10.1038/nrn2850 Howard, J. H., Jr., Howard, D. V., Dennis, N. A., Yankovich, H., & Vaidya, C. J. (2004). Implicit spatial contextual learning in healthy References aging. Neuropsychology, 18, 124–134. http://dx.doi.org/10.1037/0894- 4105.18.1.124 Addis, D. R., Roberts, R. P., & Schacter, D. L. (2011). Age-related neural Huttenlocher, J., & Presson, C. C. (1979). The coding and transformation changes in autobiographical remembering and imagining. Neuropsycho- of spatial information. Cognitive Psychology, 11, 375–394. http://dx.doi logia, 49, 3656–3669. http://dx.doi.org/10.1016/j.neuropsychologia .org/10.1016/0010-0285(79)90017-3 .2011.09.021 Jack, C. R., Jr., Petersen, R. C., Xu, Y., O’Brien, P. C., Smith, G. E., Ivnik, Bangert, A. S., Reuter-Lorenz, P. A., Walsh, C. M., Schachter, A. B., & R.J.,...Kokmen, E. (1998). Rate of medial temporal lobe atrophy in Seidler, R. D. (2010). Bimanual coordination and aging: Neurobehav- typical aging and Alzheimer’s disease. Neurology, 51, 993–999. http:// ioral implications. Neuropsychologia, 48, 1165–1170. http://dx.doi.org/ dx.doi.org/10.1212/WNL.51.4.993 10.1016/j.neuropsychologia.2009.11.013 Kail, R., & Salthouse, T. A. (1994). Processing speed as a mental capacity. Bates, D., DebRoy, S., & Gay, D. M. (2007). nls: Nonlinear least square. Acta Psychologica, 86, 199–225. http://dx.doi.org/10.1016/0001-6918 Vienna, Austria: R Foundation for Statistical Computing. (94)90003-5 Campbell, K. L., Hasher, L., & Thomas, R. C. (2010). Hyper-binding. Kennedy, K. M., Erickson, K. I., Rodrigue, K. M., Voss, M. W., Colcombe, Psychological Science, 21, 399–405. http://dx.doi.org/10.1177/0956 S. J., Kramer, A. F.,...Raz, N. (2009). Age-related differences in 797609359910 regional brain volumes: A comparison of optimized voxel-based mor- Chalfonte, B. L., & Johnson, M. K. (1996). Feature memory and binding in young and older adults. Memory & Cognition, 24, 403–416. http:// phometry to manual volumetry. Neurobiology of Aging, 30, 1657–1676. dx.doi.org/10.3758/BF03200930 http://dx.doi.org/10.1016/j.neurobiolaging.2007.12.020 Clark, R., Freedberg, M., Hazeltine, E., & Voss, M. W. (2015). Are there Kirwan, C. B., & Stark, C. E. L. (2007). Overcoming interference: An age-related differences in the ability to learn configural responses? PLoS fMRI investigation of pattern separation in the medial temporal lobe. ONE, 10, e0137260. http://dx.doi.org/10.1371/journal.pone.0137260 Learning & Memory, 14, 625–633. http://dx.doi.org/10.1101/lm.663507 Clark, R., Tahan, A. C., Watson, P. D., Severson, J., Cohen, N. J., & Voss, Knowlton, B. J., Mangels, J. A., & Squire, L. R. (1996). A neostriatal habit M. (2017). Aging affects spatial reconstruction more than spatial pattern learning system in humans. Science, 273, 1399–1402. http://dx.doi.org/ separation performance even after extended practice. Hippocampus, 27, 10.1126/science.273.5280.1399 716–725. http://dx.doi.org/10.1002/hipo.22727 Konkel, A., & Cohen, N. J. (2009). Relational memory and the hippocam- Craik, F. I. M. (1994). Memory changes in normal aging. Current Direc- pus: Representations and methods. Frontiers in Neuroscience, 3, 166– tions in Psychological Science, 3, 155–158. http://dx.doi.org/10.1111/ 174. http://dx.doi.org/10.3389/neuro.01.023.2009 1467-8721.ep10770653 Konkel, A., Warren, D. E., Duff, M. C., Tranel, D. N., & Cohen, N. J. Davachi, L., & Wagner, A. D. (2002). Hippocampal contributions to (2008). Hippocampal amnesia impairs all manner of relational memory. episodic encoding: Insights from relational and item-based learning. Frontiers in Human Neuroscience, 2, 15. http://dx.doi.org/10.3389/ neuro.09.015.2008 This document is copyrighted by the American Psychological Association or one of its allied publishers. Journal of Neurophysiology, 88, 982–990. Kumaran, D., Hassabis, D., & McClelland, J. L. (2016). what learning This article is intended solely for the personal use ofDennis, the individual user and is not to be disseminated broadly. N. A., Howard, J. H., Jr., & Howard, D. V. (2006). Implicit sequence learning without motor sequencing in young and old adults. systems do intelligent agents need? Complementary learning systems Experimental Brain Research, 175, 153–164. http://dx.doi.org/10.1007/ theory updated. Trends in Cognitive Sciences, 20, 512–534. http://dx.doi s00221-006-0534-3 .org/10.1016/j.tics.2016.05.004 Driscoll, I., Davatzikos, C., An, Y., Wu, X., Shen, D., Kraut, M., & Levine, B., Svoboda, E., Hay, J. F., Winocur, G., & Moscovitch, M. Resnick, S. M. (2009). Longitudinal pattern of regional brain volume (2002). Aging and autobiographical memory: Dissociating episodic change differentiates normal aging from MCI. Neurology, 72, 1906– from semantic retrieval. Psychology and Aging, 17, 677–689. http://dx 1913. http://dx.doi.org/10.1212/WNL.0b013e3181a82634 .doi.org/10.1037/0882-7974.17.4.677 Eichenbaum, H. (2000). A cortical-hippocampal system for declarative Light, L. L. (1991). Memory and aging: Four hypotheses in search of data. memory. Nature Reviews Neuroscience, 1, 41–50. http://dx.doi.org/10 Annual Review of Psychology, 42, 333–376. http://dx.doi.org/10.1146/ .1038/35036213 annurev.ps.42.020191.002001 Eichenbaum, H., & Bunsey, M. (1995). On the binding of associations in Lucas, H. D., Monti, J. M., McAuley, E., Watson, P. D., Kramer, A. F., & memory: Clues from studies on the role of the hippocampal region in Cohen, N. J. (2016). Relational memory and self-efficacy measures paired-associate learning. Current Directions in Psychological Science, reveal distinct profiles of subjective memory concerns in older adults. 4, 19–23. http://dx.doi.org/10.1111/1467-8721.ep10770954 Neuropsychology, 30, 568–578. http://dx.doi.org/10.1037/neu0000275 AGE DIFFERENCES IN EPISODIC ASSOCIATIVE LEARNING 157

McIntyre, J. S., & Craik, F. I. M. (1987). Age differences in memory for Raz, N., Rodrigue, K. M., Head, D., Kennedy, K. M., & Acker, J. D. item and source information. Canadian Journal of Psychology/Revue (2004). Differential aging of the medial temporal lobe: A study of a canadienne de psychologie, 41, 175–192. http://dx.doi.org/10.1037/h00 five-year change. Neurology, 62, 433–438. http://dx.doi.org/10.1212/01 84154 .WNL.0000106466.09835.46 Minear, M., & Park, D. C. (2004). A lifespan database of adult facial Rieckmann, A., Fischer, H., & Bäckman, L. (2010). Activation in striatum stimuli. Behavior Research Methods, Instruments & Computers, 36, and medial temporal lobe during sequence learning in younger and older 630–633. http://dx.doi.org/10.3758/BF03206543 adults: Relations to performance. NeuroImage, 50, 1303–1312. http://dx Monti, J. M., Cooke, G. E., Watson, P. D., Voss, M. W., Kramer, A. F., & .doi.org/10.1016/j.neuroimage.2010.01.015 Cohen, N. J. (2015). Relating hippocampus to relational memory pro- Sadeh, T., Shohamy, D., Levy, D. R., Reggev, N., & Maril, A. (2011). cessing across domains and delays. Journal of Cognitive Neuroscience, Cooperation between the hippocampus and the striatum during episodic 27, 234–245. http://dx.doi.org/10.1162/jocn_a_00717 encoding. Journal of Cognitive Neuroscience, 23, 1597–1608. http://dx Moses, S. N., & Ryan, J. D. (2006). A comparison and evaluation of the .doi.org/10.1162/jocn.2010.21549 predictions of relational and conjunctive accounts of hippocampal func- Salthouse, T. A. (1996). The processing-speed theory of adult age differ- tion. Hippocampus, 16, 43–65. http://dx.doi.org/10.1002/hipo.20131 ences in cognition. Psychological Review, 103, 403–428. http://dx.doi Myers, C. E., Shohamy, D., Gluck, M. A., Grossman, S., Kluger, A., Ferris, .org/10.1037/0033-295X.103.3.403 S.,...Schwartz, R. (2003). Dissociating hippocampal versus basal Schapiro, A. C., Turk-Browne, N. B., Botvinick, M. M., & Norman, K. A. ganglia contributions to learning and transfer. Journal of Cognitive (2017). Complementary learning systems within the hippocampus: A Neuroscience, 15, 185–193. http://dx.doi.org/10.1162/0898929033212 neural network modelling approach to reconciling episodic memory with 08123 statistical learning. Philosophical Transactions of the Royal Society of Naveh-Benjamin, M. (2000). Adult age differences in memory perfor- London: Series B, Biological Sciences. Advance online publication. mance: Tests of an associative deficit hypothesis. Journal of Experimen- http://dx.doi.org/10.1098/rstb.2016.0049 tal Psychology: Learning, Memory, and Cognition, 26, 1170–1187. Schwarb, H., Johnson, C. L., McGarry, M. D. J., & Cohen, N. J. (2016). http://dx.doi.org/10.1037/0278-7393.26.5.1170 Medial temporal lobe viscoelasticity and relational memory perfor- Naveh-Benjamin, M., Guez, J., Kilb, A., & Reedy, S. (2004). The asso- mance. NeuroImage, 132, 534–541. http://dx.doi.org/10.1016/j.neuro ciative memory deficit of older adults: Further support using face-name image.2016.02.059 associations. Psychology and Aging, 19, 541–546. http://dx.doi.org/10 Schwarb, H., Watson, P. D., Campbell, K., Shander, C. L., Monti, J. M., .1037/0882-7974.19.3.541 Cooke, G. E., . . . Cohen, N. J. (2015). Competition and cooperation Naveh-Benjamin, M., Hussain, Z., Guez, J., & Bar-On, M. (2003). Adult among relational memory representations. PLoS ONE, 10, e0143832. age differences in episodic memory: Further support for an associative- http://dx.doi.org/10.1371/journal.pone.0143832 deficit hypothesis. Journal of Experimental Psychology: Learning, Schwarz, G. (1978). Estimating the dimension of a model. Annals of Memory, and Cognition, 29, 826–837. http://dx.doi.org/10.1037/0278- Statistics, 6, 461–464. http://dx.doi.org/10.1214/aos/1176344136 7393.29.5.826 Seger, C. A. (2006). The basal ganglia in human learning. The Neurosci- Old, S. R., & Naveh-Benjamin, M. (2008). Differential effects of age on entist, 12, 285–290. http://dx.doi.org/10.1177/1073858405285632 item and associative measures of memory: A meta-analysis. Psychology Simon, J. R., Vaidya, C. J., Howard, J. H., Jr., & Howard, D. V. (2012). and Aging, 23, 104–118. http://dx.doi.org/10.1037/0882-7974.23.1.104 The effects of aging on the neural basis of implicit associative learning Oldfield, R. C. (1971). The assessment and analysis of handedness: The in a probabilistic triplets learning task. Journal of Cognitive Neurosci- Edinburgh Inventory. Neuropsychologia, 9, 97–113. http://dx.doi.org/10 ence, 24, 451–463. http://dx.doi.org/10.1162/jocn_a_00116 .1016/0028-3932(71)90067-4 Smith, M. L., & Milner, B. (1981). The role of the right hippocampus in the Poldrack, R. A., Clark, J., Paré-Blagoev, E. J., Shohamy, D., Creso recall of spatial location. Neuropsychologia, 19, 781–793. Moyano, J., Myers, C., & Gluck, M. A. (2001). Interactive memory St. Jacques, P. L., Rubin, D. C., & Cabeza, R. (2012). Age-related effects systems in the human brain. Nature, 414, 546–550. http://dx.doi.org/10 on the neural correlates of autobiographical memory retrieval. Neurobi- .1038/35107080 ology of Aging, 33, 1298–1310. http://dx.doi.org/10.1016/j.neurobiolag- Poldrack, R. A., & Packard, M. G. (2003). Competition among multiple ing.2010.11.007 memory systems: Converging evidence from animal and human brain Tombaugh, T. N. (2005). Test-retest reliable coefficients and 5-year studies. Neuropsychologia, 41, 245–251. change scores for the MMSE and 3MS. Archives of Clinical Neuropsy- Poldrack, R. A., Prabhakaran, V., Seger, C. A., & Gabrieli, J. D. (1999). chology: The Official Journal of the National Academy of Neuropsy- Striatal activation during acquisition of a cognitive skill. Neuropsychol- chologists, 20, 485–503. http://dx.doi.org/10.1016/j.acn.2004.11.004 ogy, 13, 564–574. http://dx.doi.org/10.1037/0894-4105.13.4.564 Voss, M. W., Erickson, K. I., Chaddock, L., Prakash, R. S., Colcombe, Poldrack, R. A., & Rodriguez, P. (2004). How do memory systems inter- S. J., Morris, K. S.,...Kramer, A. F. (2008). Dedifferentiation in the

This document is copyrighted by the American Psychological Association or one of its allied publishers. act? Evidence from human classification learning, Neurobiology of visual cortex: An fMRI investigation of individual differences in older

This article is intended solely for the personal use of the individual userLearning and is not to be disseminated broadly. and Memory, 82, 324–332. http://dx.doi.org/10.1016/j.nlm adults. Brain Research, 1244, 121–131. http://dx.doi.org/10.1016/j .2004.05.003 .brainres.2008.09.051 Provyn, J. P., Sliwinski, M. J., & Howard, M. W. (2007). Effects of age on Watson, P. D., Voss, J. L., Warren, D. E., Tranel, D., & Cohen, N. J. contextually mediated associations in paired associate learning. Psychol- (2013). Spatial reconstruction by patients with hippocampal damage is ogy and Aging, 22, 846–857. http://dx.doi.org/10.1037/0882-7974.22.4 dominated by relational memory errors. Hippocampus, 23, 570–580. .846 http://dx.doi.org/10.1002/hipo.22115 Raz, N., Lindenberger, U., Rodrigue, K. M., Kennedy, K. M., Head, D., Williamson, A.,...Acker, J. D. (2005). Regional brain changes in aging healthy adults: General trends, individual differences and mod- Received August 23, 2017 ifiers. Cerebral Cortex, 15, 1676–1689. http://dx.doi.org/10.1093/ Revision received November 15, 2017 cercor/bhi044 Accepted December 5, 2017 Ⅲ