Multimodal cortical and hippocampal prediction of episodic‐memory plasticity in young and older adults

Abstract

Episodic memory can be trained in both early and late adulthood, but there is considerable variation in cognitive improvement across individuals. Which brain characteristics make some individuals benefit more than others? We used a multimodal approach to investigate whether volumetric magnetic resonance imaging (MRI) and resting‐state functional MRI characteristics of the cortex and hippocampus, brain regions involved in episodic‐memory function, were predictive of cognitive improvement after memory training. We hypothesized that these brain characteristics would differentially predict memory improvement in young and older adults, given the vulnerability of cortical regions as well as the hippocampus to healthy aging. Following structural and resting‐state activity magnetic resonance scans, 50 young and 76 older participants completed 10 weeks of strategic episodic‐memory training. Both age groups improved their memory performance, but the young adults more so than the older. Vertex‐wise analyses of cortical volume showed no significant relation to memory benefit. When analyzing the two age groups separately, hippocampal volume was predictive of memory improvement in the group of older participants only. In this age group, the lower resting‐state activity of the hippocampus was also predictive of memory improvement. Both volumetric and resting‐state characteristics of the hippocampus explained unique variance of the improvement in the older participants suggesting that a multimodal imaging approach is valuable for the understanding of mechanisms underlying memory plasticity in aging.

1. INTRODUCTION

Episodic‐memory function, or the ability to encode and retrieve previous events (Tulving & Thomson, 1973), is important for the successful execution of daily tasks, and vulnerable to decline in healthy aging (Buckner, 2004; Nyberg, Lövdén, Riklund, Lindenberger, & Bäckman, 2012; Salthouse, 2003). Hippocampal–cortical connections are known to be critical for episodic‐memory function (Buzsaki & Moser, 2013; Scoville & Milner, 1957), and numerous studies have shown cortical and hippocampal volume and activation to relate to episodic memory performance (Buckner, 2004; Buckner, Andrews‐Hanna, & Schacter, 2008; Buckner & Wheeler, 2001; Raichle et al., 2001; Rugg & Vilberg, 2013). Although to a varying extent, the brain shows volumetric changes and functional disruptions with increasing age (Addis, Roberts, & Schacter, 2011; Allen et al., 2011; Allen, Bruss, Brown, & Damasio, 2005; Andrews‐Hanna et al., 2007; Damoiseaux et al., 2008; Ferreira & Busatto, 2013; Fjell et al., 2009b; Greenberg et al., 2008; Koch et al., 2010; Lustig et al., 2003; Mevel, Chételat, Eustache, & Desgranges, 2011; Raz et al., 2005; Walhovd et al., 2005, 2011), and such changes have been associated with decline in episodic memory performance (Fjell et al., 2014; Mevel et al., 2011; Murphy et al., 2010; Persson et al., 2012). This suggests conserved cortical and hippocampal function to be pivotal for healthy aging and cognitive preservation.

Memory training has previously shown positive effects on episodic memory performance in both young and older adults (de Lange et al., 2016; Engvig et al., 2010; Lövdén et al., 2010; Nyberg et al., 2003). However, young participants tend to improve more than older with cognitive training (Baltes & Kliegl, 1992a; Burki, Ludwig, Chicherio, & de Ribaupierre, 2014; Carretti, Borella, & De Beni, 2007; Dahlin, Nyberg, Backman, & Neely, 2008; Lövdén et al., 2010; Lövdén, Brehmer, Li, & Lindenberger, 2012; Nyberg et al., 2003), and the individual variation in cognitive training benefits is considerable (de Lange et al., 2016; Engvig et al., 2012; Lövdén, Brehmer, et al., 2012). There is still uncertainty regarding why some individuals benefit more than others from training interventions. Reduced brain integrity may yield less learning potential with older age, and some studies have shown that individual differences in brain characteristics explain part of the variance in cognitive improvement after training (de Lange et al., 2016; Engvig, Fjell, Westlye, Moberget, et al., 2012). Given the known role of hippocampal–cortical dialog in episodic‐memory function (Buzsaki & Moser, 2013), it is interesting to test hippocampal–cortical brain characteristics as predictors of memory‐training improvement. Hippocampal volume has previously been found to predict cognitive improvement in a memory‐training study including patients with subjective memory impairment (Engvig et al., 2012). However, to the best of our knowledge, a multimodal imaging approach to test how hippocampal and cortical characteristics together account for cognitive plasticity has not been applied. Such approach could enhance our understanding of the neural basis for successful memory‐training outcomes in healthy young and older adults. Here, we tested whether volumetric and resting‐state (RS) functional magnetic resonance imaging (fMRI) brain characteristics of the hippocampus and the cortex were predictive of the cognitive benefit from memory training. We further tested whether these brain characteristics were differentially predictive in young and older adults. As the dependency of the hippocampus for memory performance is observed strengthened with older age, we expected that the correlates would be stronger in this age group. The term cognitive plasticity is used to refer to the extent of memory improvement in response to the memory‐training program.

Groups of healthy adults in their 20s and their 70s underwent 10 weeks of memory training using the mnemonic technique method of loci (MoL) (Bower, 1970), which has been shown to improve serial recall substantially in both young and older adults (de Lange et al., 2016; Engvig, Fjell, Westlye, Moberget, et al., 2012; Kliegl, Smith, & Baltes, 1990b; Nyberg et al., 2003). An active control group received popular scientific lectures and tasks, and a passive control group was scanned and tested with the same 10‐week interval as the two other groups. White matter (WM) integrity at baseline was predictive of cognitive plasticity in aging. We have previously reported that memory improvement was specific to the training group, that is, the memory‐training group improved significantly more than the control groups (de Lange et al., 2016). A group interaction was observed between the train, active, and passive control groups. The participants in the training group significantly improved their memory score, while no significant change after time points 1 and 2 was found in any of the control groups. Given the previously reported findings, we here proceeded to study volumetric and RS characteristics of the cortex and the hippocampus exclusively in the participants that underwent the training intervention. Given that the hippocampus is invariably involved in episodic memory, while the cortical substrates of episodic memory vary with specific content, an unbiased and properly corrected whole‐brain approach was used for cortical volume and RS analyses.

For RS, we measured the intensity of regional spontaneous brain activity, namely, the amplitude of low‐frequency fluctuation (ALFF) (Zou et al., 2013) and fractional amplitude of low‐frequency fluctuation (fALFF) of the RS‐fMRI signal (Zang et al., 2007). In contrast to the more commonly used measure of functional connectivity (FC)—temporal correlations in BOLD signal fluctuations—between regions of the brain, this approach enables the investigation of isolated brain regions and their states during rest. Here, mean fALFF of the hippocampus was calculated and used in the analyses. Given the multimodal approach of the study, this was done to include a modality that was more comparable to the volumetric values, as opposed to FC between the hippocampus and other regions of the brain (however, for exploratory purposes, comparisons with RS‐FC were undertaken, see “Section 2.3” and “Section 3”). As it has been indicated that ALFF is sensitive to physiological noise (Zou et al., 2008), fALFF was chosen for the main analyses, but ALFF results are also reported.

We hypothesized that (a) both cortical and hippocampal volumetric and RS characteristics would be predictive of training‐related memory improvement, and (b) this predictive power, given aging‐related neural changes, would be greater in older relative to young adults.

2. METHODS AND MATERIALS

2.1. Sample

The sample was drawn from the ongoing project Neurocognitive Plasticity at the Center for Lifespan Changes in Brain and Cognition (LCBC), Department of Psychology, University of Oslo. All procedures were approved by the regional ethical committee of Southern Norway, and written consent was obtained from all participants. Participants were recruited through newspaper and web page adverts and were screened with a health interview. Participants were required to be either young or older (in or around their 20s or 70s, respectively) healthy adults, right‐handed, fluent Norwegian speakers, and have normal or corrected to normal vision and hearing. Exclusion criteria were a history of injury or disease known to affect central nervous system function, including neurological or psychiatric illness or serious head trauma, being under psychiatric treatment, use of psychoactive drugs known to affect central nervous system functioning, and MRI contraindications. Moreover, for inclusion in the present study, participants were required to score ≥ 26 on the Mini‐Mental State Examination (MMSE) (Folstein, Folstein, & McHugh, 1975) and have scores within normal range (≥2 standard deviations below mean) for age and sex on the 5‐min delayed recall subtest of the California Verbal Learning Test II (Delis, Kramer, Kaplan, & Ober, 2000). All participants further had to achieve an IQ above 85 on the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999). Three participants in the older group were excluded based on these criteria. Participant scans were evaluated by a neuroradiologist and deemed free of significant injuries or conditions. The images were further manually quality checked for artifacts, and six older participants were excluded due to motion artifacts. Only participants who underwent a neuropsychological follow‐up assessment in addition to MRI scanning and neuropsychological tests at baseline were included in the current analyses. A total of 25 participants (11 young, 14 older) dropped out before the follow‐up session and were thus excluded from the analyses. The participants who dropped out reported that the participation was too time‐consuming or that the particular time frame for assessment was inconvenient. The group of participants who dropped out after the first scanning session performed lower than the rest of the sample in terms of IQ (mean ± SD dropouts = 114.4 ± 10.5; included = 120.5 ± 10.2; t[149] = 2.2, p = .03), MMSE score (mean ± SD dropouts = 27.7 ± 1.4; included = 28.7 ± 1.3; t[149] = 2.0, p = .06) and CVLT 5 min recall (mean ± SD dropouts = 8.3 ± 3.7; included = 10.2 ± 3.2; t[149] = 2.1, p = .04). Lower cognitive performance among dropouts is commonly observed in longitudinal studies, resulting in a selection bias effect toward higher functioning individuals (Salthouse, 2014). Refer de Lange et al. (2017) for further analysis addressing the selection bias in the current sample. A total of 50 participants in their 20s and 76 participants in their 70s were included in this study.

In the project from which the present subsample is drawn, the participants were assigned to one of three groups: memory training (young adults: N = 50, older adults: N = 76), active control (young adults: N = 13, older adults: N = 18) or passive control (older adults: N = 49, young adults: N = 28). Some of the passive control‐participants (16 young adults and 35 older adults) completed the memory training after 10 initial weeks as passive controls. Their participation in the training group was preceded by one MRI scan followed by 10 weeks as passive controls. After the 10 weeks as passive controls, they underwent a second MRI scan and entered directly into the training group. For analyses of the relationship between the brain measures and memory‐training outcome regressing out a number of scans (Supporting Information). We have previously reported that memory improvements were found in the training group only (de Lange et al., 2016). Since the focus of this article was memory improvement in response to the training intervention, only the participants who completed the memory‐training program (N = 126) were included in the analyses. For demographics, refer Table 1. For demographics of the active and passive control groups, in addition to a detailed description of the design and the control‐group conditions, refer de Lange et al. (2017).

Table 1.

Demographics of the memory‐training group

	Young adults		Older adults
	(34 females/16 males)		(47 females/29 males)
	M ± SD	Range	M ± SD	Range
Age	26.1 ± 3.2	20.5:30.9	73.4 ± 3.0	69:82.1
IQ	112.4 ± 9.3	88:130	120.7 ± 10.5	91:144
Education	15.7 ± 1.9	12:19	14.7 ± 2.9	7.5:21
MMSE	28.9 ± 1.1	26:30	28.9 ± 1.7	26:30
CVLT 5 min recall	14.3 ± 1.8	9:16	10.5 ± 3.3	3:16

2.2. Design and memory‐training program

All participants underwent a neuropsychological assessment and an MRI scan before the training intervention, and a follow‐up neuropsychological assessment at the completion of the intervention. The memory training involved learning and practicing the MoL technique specifically aiming to improve episodic memory performance. The training program included weekly in‐class course sessions during 10 weeks and eight weekly online home assignments. The assignments involved tasks of word lists to be memorized by utilizing the MoL. The first group session included a presentation of the project, an introduction to the MoL method with instructions, and an initial word list task consisting of 15 words. The research fellow leading the group session was available for questions and provided further explanations and repetition of instructions to ensure that all participants were able to utilize the technique. The following weekly group sessions included updating of the strategy, clarification of instructions and a new word list task, which was increased by five words each week to ensure a continuous challenge. However, the participants were encouraged to individually adjust the difficulty level of the tasks both in class and of the home assignments, with the aim of achieving a challenging but manageable training level across all the participants. The home assignments were completed online and all responses in addition to time spent on the tasks were registered to a database. Both age groups underwent the same program. The number of total tasks completed was on average 48.1% in the young training group (mean ± SD = 37.79, 16.87) and 72.6% (mean ± SD = 57.83, 14.1) in the older training group. The number of tasks completed did not affect the training outcome (de Lange et al., 2017). Memory improvement was measured by change in correct written recall of a word list consisting of 100 nouns administrated on the neuropsychological test sessions at baseline and on the follow‐up test. The total number of words recalled from the word list was included in the score, regardless of order. The participants were given 5 min to memorize the word list, followed by 10 min to recall as many words as possible. The words in the list differed between the two time points. Latency and response time of the memory tests were not registered, as all participants were given 10 min to recall as many words as possible. The test was deliberately designed to be so difficult that no participant would be able to reach a maximum score within this time frame. The test was performed using pen and paper. We have previously reported that a working memory task was assessed, namely, the backward digit span task (de Lange et al., 2016). As opposed to the word list task, there were no differences between the training groups and the controls. Thus, the working memory task did not show any training‐specific changes. The extensive length of the word lists was chosen to avoid ceiling effects. For more details regarding the memory‐training program and the individual adjustments, refer de Lange et al. (2017).

2.3. Image acquisition and pre‐processing

A Siemens Skyra 3T MRI scanner with a 24‐channel head‐coil was used (Siemens Medical Solutions, Erlangen, Germany). The volumetric analyses were based on a three‐dimensional (3D) T1‐weighted magnetization prepared rapid gradient echo (MP‐RAGE) pulse sequence (TR/TE/TI = 2,300/2.98/850 ms, FA = 8°, matrix size = 192 × 192, 176 sagittal slices, voxel size = 1.0 × 1.0 × 1.0 mm³, field of view = 240 mm) with a total duration of 9 min and 50 s. The morphometric data (volume) was processed with the FreeSurfer software recon‐all pipeline (version 5.3; http://surfer.nmr.mgh.harvard.edu/). The processing includes removal of nonbrain tissue, Talairach transformation, intensity correction, tissue segmentation, and cortical surface reconstruction, and is thoroughly described elsewhere (Dale, Fischl, & Sereno, 1999; Dale & Sereno, 1993; Fischl et al., 2002, Fischl, Salat, et al., 2004; Fischl, van der Kouwe, et al., 2004; Fischl & Dale, 2000; Fischl, Liu, & Dale, 2001; Fischl, Sereno, & Dale, 1999; Fischl, Sereno, Tootell, & Dale, 1999; Han et al., 2006; Jovicich et al., 2006; Reuter, Rosas, & Fischl, 2010; Reuter, Schmansky, Rosas, & Fischl, 2012; Segonne et al., 2004). All scan‐sets were manually checked for gross motion artifacts and quality of subcortical segmentation and cortical parcellation output. No manual edits were performed on the MRI images. The RS BOLD sequences included 43 transversally oriented slices (no gap), measured using a BOLD‐sensitive T2*‐weighted EPI sequence (TR/TE = 2,390/30 ms, FA = 90°, voxel size = 3.0 × 3.0 × 3.0 mm³, field of view = 224 mm, GRAPPA acceleration factor = 2), producing 150 volumes and lasting for 6 min and 6 s. The RS BOLD‐data were extracted from the middle point of the gray matter, as estimated by FreeSurfer. RS data were motion‐, slice timing‐, B0‐corrected, and smoothed (5 mm FWHM) in volume space using FSL's FMRI Expert Analysis Tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FEAT). Then, FSL's multivariate exploratory linear optimized decomposition into independent components (MELODIC) was used in combination with FMRIB's ICA‐based Xnoiseifier (FIX) to auto‐classify independent components into signal and noise components and remove noise components from the four‐dimensional (4D) fMRI data (Salimi‐Khorshidi et al., 2014). The FIX‐classification used a lab‐specific hand‐labeled training set of typical signal and noise components derived from 38 subjects within similar age range (18–80) to the sample in the current study and using the same fMRI acquisition parameters as the current study. Such ICA‐based procedure for denoising fMRI data has been shown to effectively reduce motion‐induced variability, outperforming methods based on removing motion spikes in the data set (Pruim et al., 2015). FreeSurfer‐defined individually estimated anatomical masks of cerebral WM and cerebrospinal fluid (CSF)/lateral ventricles were resampled to each individual's functional space. All anatomical voxels that constituted a functional voxel had to be labeled as WM or CSF for that functional voxel to be considered a functional representation of noncortical tissue. Average time series were then extracted from functional WM and CSF voxels and were regressed out of the FIX‐cleaned 4D volume together with a set of estimated motion parameters (rotation/translation) and their derivatives. To be defined as a functional voxel of the hippocampus, a voxel was required to be encompassed by at least 50% structural voxels labeled as hippocampus in the structural 1.0 mm resolution.

For the ALFF and fALFF analyses, each participant's preprocessed RS‐fMRI data were high‐pass filtered with FSL's fslmaths tool using a nonlinear method with a cut‐off frequency of 0.009 Hz (sigma value of 23.24). Then, the power spectral density function was calculated for the filtered RS‐fMRI data for all FreeSurfer‐labeled hippocampal voxels. ALFF was calculated as the mean power within the low‐frequency band (0.01–0.1 Hz), while fALFF was calculated by further dividing this value by the power of the entire frequency band (0–0.2092 Hz). The resulting voxel‐wise hippocampal ALFF and fALFF values were averaged across an individual's hippocampi to yield average hippocampal ALFF and fALFF per participant. For the FC analyses, mean RS‐fMRI BOLD time series were extracted from each FreeSurfer‐defined hippocampal voxel. Then, the average time series across the bilateral hippocampi were correlated with BOLD time series data at every gray matter voxel falling within the FreeSurfer‐defined cortical mask. The resulting map of hippocampal–cortical correlations (Pearson's r, one map per participant per RS‐fMRI run), was Fisher z‐transformed and transformed to FreeSurfer's average space (fsaverage) before statistical analysis. For exploratory purposes, change in volume and fALFF from time points 1 to 2 was investigated through repeated measures analyses in the old sample. No change was found (Supporting Information).

2.4. Statistical analyses

2.4.1. Memory improvement

A repeated measures general linear model (GLM) with sex and age as covariates was carried out to test the improvement in memory performance from baseline to time point 2 across the sample. To test the interaction between age group and memory‐training improvement, an additional repeated measures GLM was performed with time (memory score at time point 1, memory score at time point 2) × age group (young, older), using sex as a covariate. Greenhouse–Geisser corrections for violation of sphericity were used where applicable. Then, to control for influences of baseline differences in memory performance, standardized residuals were used as the measure of memory improvement from baseline to time point 2. The residuals were calculated from a linear regression analysis, using memory performance at time point 2 as the dependent variable and memory performance at baseline as the independent variable. The residuals were calculated across ages in the analyses of the whole sample, and from each age group in the analyses of the age groups separately. Effect sizes of the memory improvement in each group are reported on p. 5. Cohen's d was calculated by dividing the mean change of each group within the groups' standard deviation (SD).

2.4.2. Brain characteristics (volume and fALFF) as predictors of memory‐training improvement

Given the time (memory change) × age group interaction (“Section 3.1”), the young and older adults were investigated as separate groups for the subsequent analyses of volume and fALFF and their relationships to memory improvement from baseline to time point 2. Before the volumetric cortical analysis, all participants' cortical surface‐representations of vertex‐wise volume were brought to FreeSurfers' average space (fsaverage) and smoothed with a 15 mm FWHM Gaussian filter. GLMs were then performed separately for each hemisphere and age group to examine the relationship between volume across vertices at baseline and memory improvement. The analyses were corrected for individual differences in age, sex, and estimated intracranial volume (ICV). The significance of this relationship was assessed within the FreeSurfer‐framework (mri_glmfit), using cluster‐based inference to account for multiple comparisons. To verify the reliability of the findings, several cluster‐forming thresholds were tested, ranging from p < .05 to p < .001 (all tests were two‐sided). Although our main hypotheses pertained to hippocampal characteristics and their relationships to functional plasticity and memory‐training effects, for the sake of completeness we ran a set of exploratory whole‐brain analyses. To further investigate the relationships between fALFF and memory improvement, vertex‐wise analysis of fALFF on the cortical surface was conducted (cluster‐correction; initial threshold of p < .001). In addition, analysis of noncortical fALFF was carried out in MNI volume space, using Freesurfer's subcortical mask. Here, significance was assessed using threshold‐free cluster enhancement correction following 5,000 permutations of FSL's Randomize tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise). For additional exploratory analyses, we estimated hippocampal strength to all other nodes in the Desikan–Killian‐parcellation (Freesurfer aparc+aseg). Pearson's r was calculated between the hippocampal RS time series and all cortical/subcortical Desikan–Killian‐nodes and summed the resulting correlation weights after Fisher‐normalizing the r values (Supporting Information).

To test for age group differences in the two measures, univariate GLMs of hippocampal volume and fALFF values at baseline were conducted. Sex was used as covariate in both analyses in addition to ICV in the volumetric analyses, and hippocampal volume in the ALFF/fALFF analyses. Due to nonnormal distributions of the BOLD‐derived ALFF/fALFF scores (Shapiro–Wilk‐tests, p < .001), the relationships between baseline memory performance and hippocampal volume/ALFF/fALFF at baseline were investigated using partial Spearman correlations. Also here, age and sex were included as covariates in all analyses, in addition to controlling for ICV in the volumetric analyses, and hippocampal volume in the ALFF/fALFF analyses. This was repeated for partial correlations between memory improvement and baseline hippocampal volume/ALFF/fALFF. The statistical significance of individual relationships was assessed following Bonferroni–Holm correction for three tests in each group.

2.4.3. Additional FC analyses as a comparison modality to fALFF

fALFF represents the relative proportion of low‐frequency fluctuations in a measured BOLD signal, and functional coupling between brain regions is commonly considered to occur through such intrinsic slow fluctuations (Buckner et al., 2008). An exploratory FC analysis was carried out to test whether differences in hippocampal–neocortical FC were predictive of memory improvements in the older age group. First, for each hemisphere separately, analysis of RS‐FC between the hippocampus and the rest of the brain was carried out, vertex‐by‐vertex. A mask encompassing exclusively vertices across the cortex with statistical significant FC with the hippocampus (false discovery rate < .05) was made. Only uncorrected correlations with a p‐value of < .05 were included in the mask. Given that the analysis is based on 150 RS‐fMRI volumes (resulting in 149° of freedom; 150‐1), the p value requirement (p < .05), represent correlations of (r ≥ .17). Then, including exclusively vertices within the mask, a GLM was set up examining the relationship between participants' hippocampal–cortical connectivity, vertex‐by‐vertex, and their extent of memory improvement. Significance was assessed using a cluster‐based inference. Importantly, due to recent reports of higher than expected numbers of false positives when using lenient cluster‐forming thresholds in combination with fMRI data (Eklund, Nichols, & Knutsson, 2016), we used a stringent cluster‐forming threshold (p < .001; two‐sided test). Age, sex, and hippocampal volume were included as covariates of no interest in the analyses.

3. RESULTS

3.1. Memory improvement

Repeated measures analyses showed a significant improvement in memory performance from baseline to time point 2 (F[1,123] = 145.65, p < .001), and a time × age group (young, older) interaction (F[1,123] = 51.15, p < .001). As shown in Figure 1, both age groups improved, but the group of young adults improved more than the older adults (mean difference = 12.98, p < .001), with ample variation (change in performance young; range = −7–47, SD = 20.82; change in performance older; range = −10–29, SD = 7.84). Univariate GLM using the standardized residuals of memory improvement (accounting for baseline performance) also showed an effect of age group (F[1,123] = 48.85, p < .001). Both groups showed large effect sizes (young: d = 1.6; old: d = 0.99).

Figure 1.

Change in memory performance in the two age groups. Individual change is measured by the difference in a number of recalled words on the 100‐word test between baseline and time point 2 (time point 2 – Baseline). (a) the time × age group interaction (F[1,123] = 51.14, p < .001). The change is measured from raw scores on time point 1 (baseline) to time point 2 in the two age groups, revealing a significantly greater memory improvement in the group of younger adults relative to the group of older adults. (b) Individual change in memory scores from baseline to time point 2. Change in young participants are presented in lines of shades of blue, older participants are presented in shades of orange. Variations in color shades are for illustrational purposes to delineate the individual scores regardless of total memory change or slope of change

3.2. Brain characteristics as predictors of memory improvement

3.2.1. Hippocampal volume

Univariate GLM showed a significant effect of age group on hippocampal volume (F[122,1] = 69.43, p = .000), with the younger group having the greater volume at baseline (mean ± SD = 8,741.94 ± 759.31) than the older (mean ± SD = 7,623.76 ± 797.84). Baseline memory performance did not relate to the hippocampal volume at baseline in any of the age groups. Hippocampal volume at baseline was predictive of memory improvement in the group of older adults (Spearman's rho = .28, p = .035; Figure 2), but not in the group of young adults (Spearman's rho = .15, uncorrected p = .329; Figure 2). To assess the significance of the difference between the relationships between baseline hippocampal volume and memory in the two independent age groups, a Fisher r‐to‐z transformation (available at http://vassarstats.net/rdiff.html) was conducted. The two correlations did not differ (z = 0.73, p = .465). The focus of the article was to investigate the relationship between memory improvement and brain characteristics. A small subsample of the participants experienced a small decline in their memory performance (n = 11). These participants were removed in a subsequent analysis to test whether the significant correlation between baseline volume of the hippocampus and memory improvement remained. Again, a positive correlation was found (Spearman's rho =.32, p < .009).

Figure 2.

Volume prediction of memory‐training effects. Correlations between baseline hippocampal volume (Z‐scored) and change in memory score from baseline to time point 2 (standardized residuals of memory change, calculated on each age group separately). The plots and values are shown for descriptive purposes, p values are uncorrected. p values marked with an asterisk remain significant after correcting for age, sex, and ICV. The opaque gray lines indicate the least squares relationship; dashed lines indicate a confidence interval of the slope; black opaque lines indicate relationship following robust regression. Hippocampal volume correlated with memory improvement in the whole sample (Spearman's rho = .36, corrected p = .002) and in the older adults (Spearman's rho = .28, corrected p = .035), but not in the young adults (Spearman's rho = .15, uncorrected p = .329)

3.2.2. Hippocampal resting‐state activity

Univariate GLM showed a significant effect of age group on hippocampal fALFF (F[1,122] = 9.26, p = .003), The younger group had greater activity at baseline (mean ± SD = .513 ± .041) relative to the older group (mean ± SD = .497 ± .035). Baseline memory performance did not relate to hippocampal fALFF at baseline in any of the age groups. Individual differences in hippocampal fALFF at baseline were predictive of memory improvements in the older adults (Spearman's rho = −.30, p = .031; Figure 3). No significant relationship was observed in the young adults (Spearman's rho = .15, uncorrected p = .323; Figure 3) or in any of the age groups when investigating hippocampal ALFF (older adults: rho/p = −.20/.091; younger adults: rho/p = −.03/.828). Fisher r‐to‐z transformation showed that the correlations between memory gain and fALFF differed between the two age groups (z = − 2.46, p = .014). As with the volume analysis, the subsample of older adults that experienced a decline in memory performance (n = 11, see above) was removed from the total sample to repeat the correlation analysis between fALFF and memory gains. As with the analysis including the participants with a negative change, a significant negative correlation was found (Spearman's rho = −.34, p < .005).

Figure 3.

fALFF prediction of memory‐training effects. Correlations between baseline hippocampal fALFF (Z‐scored) and change in memory score from baseline to time point 2 (standardized residuals). The plots and values are shown for descriptive purposes, p values are uncorrected. p values marked with an asterisk remain significant after correcting for age, sex, and hippocampal volume. The opaque gray lines indicate the least squares relationship; dashed lines indicate a confidence interval of the slope; black opaque lines indicate relationship following robust regression. Hippocampal fALFF correlated with memory improvement in the older adults (Spearman's rho = −.30, corrected p = .03), but not in the young adults (Spearman's rho = .15, uncorrected p = .323) or in the whole sample (Spearman's rho = −.09, uncorrected p = .341)

3.2.3. Multimodal hippocampal analysis

Observing that both hippocampal volume and fALFF at baseline predicted memory improvements after training in the older adults, then we tested whether these measures explained the unique variance of the training effect. Multiple regression analysis with memory improvement as the dependent variable and hippocampal volume and RS fALFF as the independent variables showed that both hippocampal volume (β = 0.277, t[1,75] = 2.476, p = .016) and fALFF (β = −0.299, t[1,75] = −2.697, p = .009) uniquely contributed to the explained variance in memory improvement in the older adults. The combination of the two modalities yielded greater explained variance (R ² = .138, F[2,75] = 5.864, p = .004) than either modality alone, suggesting a value of a multimodal approach for the prediction of memory‐training improvement.

Given that the correlation between baseline volume and memory improvement was positive in the older adults, while the correlation between baseline fALFF and memory improvement was negative, then we explored the relationship between these two measures at baseline. Hippocampal volume and fALFF correlated negatively when correcting for age, sex, and ICV (Spearman's rho = −.244, p = .034). As each functional voxel is defined by 27 structural voxels, we proceeded to test whether the result could reflect a partial volume effect as a result of including voxels bordering CSF. The correlation analysis was repeated including fALFF from functional voxels encompassed by 27 structural voxels labeled exclusively as hippocampus by FreeSurfer. In accordance with the previous, less conservative analysis, a negative correlation was found between hippocampal volume and fALFF (Spearman's rho = −.240, p = .04).

3.2.4. Additional RS‐FC analysis as a comparison modality

In the RS‐FC analysis, a significant negative relationship was found between memory‐training improvement and RS‐FC between the hippocampus and a cluster in the left pericalcarine cortex (Spearman's rho = −.46, p < .001; Figure 4).

Figure 4.

The hippocampal–visual FC cluster. A significant correlation between the HC–VIS cluster and change in memory score from baseline to time point 2 (standardized residuals) in the group of older adults. The plots and values are shown for descriptive purposes, p values are uncorrected. p values marked with an asterisk remain significant after correcting for age, sex, and hippocampal volume. The opaque gray lines indicate the least squares relationship; dashed lines indicate a confidence interval of the slope; black opaque lines indicate relationship following robust regression. (a) Illustration of the hippocampal–visual cluster. (b) Correlations between change in memory score from baseline to time point 2 (standardized residuals) and the HC–VIS FC cluster. A significant negative relationship was found between the older adults (Spearman's rho = −.46, corrected p < .001) but not in the young adults (Spearman's rho = .03, uncorrected p = .866)

Hence, somewhat unexpected, lower levels of functional coupling between the hippocampus and the pericalcarine area of the visual cortex before training onset predicted larger memory improvement from the training intervention in the older adults. Thus, as an exploratory analysis, we investigated whether those who had the lower levels of hippocampal FC connectivity to the observed area in the visual cortex might show larger FC change relative to those with higher FC levels at baseline. We calculated the change in FC between the hippocampus and the visual cluster (FC HC–VIS) from baseline to after the 10‐week training period. Change in FC was defined as the difference between time point 2 and baseline connectivity values. A partial Spearman's correlation was then calculated, testing for a relationship between FC at baseline and FC change, controlling for age, sex, and hippocampal volume. A negative correlation was found (Spearman's rho = −.48, p < .001; Figure 5).

Figure 5.

Hippocampal–visual FC baseline and change. Correlation between change in the hippocampal–visual FC cluster (Z‐scored) from baseline to time point 2 (standardized residuals) in the group of older adults. The plot and values are shown for descriptive purposes. The opaque gray line indicates the least squares relationship; dashed lines indicate a confidence interval of the slope; black opaque line indicate relationship following robust regression. Values are residuals controlling for age, sex, and hippocampal volume, and are Z‐scored. A negative correlation was found (Spearman's rho = −.48, p < .001)

3.2.5. Additional whole‐brain analyses

No significant relationships were observed between memory improvement and surface‐level/vertex‐wise cortical volume or cortical fALFF. Similarly, no relationships were found at the MNI voxel‐level when investigating noncortical fALFF.

4. DISCUSSION

We aimed to investigate volume and RS characteristics of the hippocampus and cortex as predictors of memory‐training improvement in young and older adults. Both the young and older participants improved their memory performance after the memory training. This aligns with previously reported results of the study. We have further reported that the memory plasticity was specific to the participants who underwent the memory intervention (de Lange et al., 2016). This is in correspondence with results from other studies reporting change in memory performance after memory training (Baltes & Kliegl, 1992b; Brehmer, Kalpouzos, Wenger, & Lovden, 2014; Burki et al., 2014; Dahlin et al., 2008; de Lange et al., 2017; Engvig et al., 2010; Nyberg et al., 2003; Schmiedek, Lovden, & Lindenberger, 2010). Furthermore, the young improved more than the older participants did. Although older participants have been observed to improve proportionally similarly to young adults (Brehmer, Shing, Heekeren, Lindenberger, & Backman, 2016; Carretti et al., 2007; Lövdén et al., 2012) after cognitive training interventions, young tend to improve more than older adults (Baltes & Kliegl, 1992b; Burki et al., 2014; Dahlin et al., 2008; Lövdén, Brehmer, et al., 2012) with larger variation in the training outcome (de Lange et al., 2017; Kliegl, Smith, & Baltes, 1990a). However, some studies find less conclusive responses to cognitive training in older age. The reasons why some studies find more robust plastic responses than others in older participants is unclear. We have previously addressed this question (de Lange et al., 2017). It could be that known age‐related decline in general processing abilities (Jones et al., 2006) and mental image manipulation (Palermo, Piccardi, Nori, Giusberti, & Guariglia, 2016) constricts the successful utilization of the mnemonic strategies. In addition, older participants have shown compliance challenges with strategy training relative to young participants (Verhaeghen & Marcoen, 1996). Here, meticulous attention was given to each participant's challenges in using the method correctly. All class sessions included repetition of instructions of the utilization of the technique, and the participants were encouraged to share experiences regarding problems with the home assignments, both individually with the group leader, and in plenary discussions. These aspects of the present study may have contributed to the substantial training gains observed also in the older age group. On average, the older participants completed more tasks than the younger group. This could be motivated by a stronger concern among the older participants regarding their own cognitive performance, or that the effort required to experience a satisfying feeling of improvement was probably less in the younger group. The discrepancy in tasks completed between the age groups could also be a consequence of available time to invest in the training. Given that the younger participants on average augmented their memory performance more than the older group, it seems that the younger participants needed to complete fewer tasks to improve relative to the older participants.

We found that both volume and fALFF characteristics were differentially predictive of memory improvement in younger versus older adults. Hippocampal volume predicted memory‐training improvement both in the whole sample and in the older group separately, while prediction through RS measures of the hippocampus was restricted to the group of older adults. Importantly, in the sample of older adults, both volume and RS characteristics of the hippocampus contributed as unique predictors of memory improvement, showing the value of a multimodal imaging approach. In the older adults, a negative relationship was found between hippocampal volume and hippocampal fALFF at baseline. Baseline measures of cortical volume at the vertex level did not predict memory improvement. This suggests a somewhat specific importance of the hippocampus in episodic‐memory plasticity as a response to episodic‐memory training, rather than a more global implication of the whole brain.

4.1. Volumetric characteristics as predictors of memory improvement

The group of young adults showed larger training improvement relative to the group of older adults, which is in correspondence with previous studies and to some extent expected, based on the known decrements in brain and cognition in aging (Allen et al., 2005; Courchesne et al., 2000; Fjell et al., 2009a; Fjell, McEvoy, Holland, Dale, & Walhovd, 2013; Fotenos, Mintun, Snyder, Morris, & Buckner, 2008; Greenberg et al., 2008; Jernigan et al., 2001; Lebel, Walker, Leemans, Phillips, & Beaulieu, 2008; Raz et al., 2005; Salat et al., 2004; Scahill et al., 2003; Walhovd et al., 2005; Walhovd et al., 2011). No relationship was found between cortical volume at baseline and memory improvement, while hippocampal volume predicted memory improvement in the whole sample. After separating the two age groups, hippocampal volume predicted memory improvement in the group of older adults exclusively, but the difference between the correlations in the two age groups was not significant. Thus, we cannot conclude that the relationship between hippocampal volume and memory improvement is stronger in older relative to younger adults. However, older adults show considerably more changes in the hippocampus and in episodic memory over time relative to younger adults, given normal age changes in brain and cognition. Hence, it could be that in the older sample, variance in brain characteristics may reflect, albeit normal, neurodegenerative age changes to a greater extent than in the young adults. The hippocampus is observed to be strongly affected by age, while other areas of the brain are affected to various degrees (Fjell et al., 2009a). Given that cortical volume did not predict memory improvement, there may be a stronger dependency on the volumetric characteristics of the hippocampus for memory plasticity, specifically in older age.

4.2. Resting‐state fALFF characteristics as predictors of memory improvement

Given the multimodal approach of this study, fALFF was chosen for the RS analyses as the modality was considered to be, in terms of spatial confines, a more similar RS modality to the volumetric measures than the FC approach. In accordance with the results from the volumetric analysis, we found that hippocampal, but not cortical, fALFF at baseline predicted memory improvement, but only in the older group. In contrast to the positive relationship between hippocampal volume and memory improvement, hippocampal fALFF values at baseline correlated negatively with memory improvement in the older group. The correlations between fALFF and memory improvement were also significantly different between the two age groups. To make sure that partial volume, for example, the inclusion of more voxels possibly containing CSF with atrophy, in older adults did not drive the results, an analysis including only voxels in the middle of the hippocampus was performed. The results confirmed the pattern. Thus, lower intensity of spontaneous hippocampal activity during rest may mediate a higher extent of cognitive plasticity in response to memory training with older age.

The role of ALFF and fALFF in healthy aging and in relation to cognition and plasticity is unclear. Previous studies have found both positive and negative correlations in different brain areas between low‐frequency fluctuations and memory performance in healthy adults (Ren, Li, Zheng, & Li, 2015; Zou et al., 2013). A clinical study with patients with amnestic mild cognitive impairment (MCI) showed that compared to controls, patients had lower fALFF in several brain areas including the hippocampus, accompanied by higher fALFF values in occipital and temporal regions (Han et al., 2011). Another study found that MCI patients had lower ALFF in the right hippocampus and parahippocampal cortex, left lateral temporal cortex, and right ventral medial prefrontal cortex, and increased ALFF in the left temporal–parietal joint and inferior parietal lobule (Xi et al., 2012). In the same study, ALFF values in the right hippocampus and parahippocampal cortex were positively correlated with the scores of MMSE. Few studies have investigated the ALFF/fALFF in relation to cognitive training. One training study has shown that increases in ALFF in the superior frontal gyrus and decreases in ALFF in the default mode network correlated with performance improvement after language training in young adults (Deng, Chandrasekaran, Wang, & Wong, 2016). Another study found that ALFF in areas of the middle frontal gyrus correlated with behavioral change in older adults (Yin et al., 2014). Whether the negative relationship between hippocampal fALFF and plasticity is a deleterious effect of high fALFF, or if low fALFF facilitates plasticity through an unknown pathway is not clear. However, memory performance at baseline was not related to baseline fALFF. Thus, although the implication of ALFF/fALFF in different cognitive processes and plasticity across the lifespan is poorly understood, our results point to an impact of hippocampal fALFF on the potential of cognitive plasticity in healthy aging, rather than a reflection of memory abilities in general. The hippocampus is not the only assumed brain structure involved in episodic memory. One brain system, in particular, the default mode network (DMN) is known to be highly synchronized at rest and active during internally focused tasks, including episodic‐memory retrieval, envisioning the future, and conceiving the perspectives of others (Buckner et al., 2008). The DMN can be further divided into several subsystems. While the core midline is active during self‐relevant and effective decisions, the dorsal medial prefrontal cortex (dMPFC) subsystem is implicated in self‐referential judgments about present situations or mental states. Finally, the medial temporal lobe (MTL) subsystem has been shown to be of particular importance for episodic‐memory function (Andrews‐Hanna, Reidler, Sepulcre, Poulin, & Buckner, 2010). Here, both total fALFF of the DMN and in the MTL subsystem correlated negatively with memory improvement (Supporting Information). Thus, although we convey a certain specificity of hippocampal fALFF in memory plasticity, the impact of fALFF may also extend to adjacent, and possibly mainly episodic memory‐specific, areas.

There is still large uncertainty regarding the underlying mechanisms of fALFF and their reflections in terms of biology. Thus, we do not know through which pathways the different RS measures interact or differ as underpinnings of cognitive function or plasticity. Previous studies have observed area‐dependent disruptions of ALFF/fALFF in relation to cognitive performance in both healthy and clinical samples (Ren et al., 2015; Veldsman et al., 2017; Zou et al., 2013). Interestingly, correlations between ALFF/fALFF and functional connectivity have also been observed (Di et al., 2013; Mascali et al., 2015; Weiler et al., 2014). Related to this, one theory is that variances in ALFF are a reflection of different levels of neurotransmitters modulating variance in functional connectivity (Di et al., 2013). It has also been suggested that the proximity to large vessels of some areas causes noise sensitivity in both measures, which underlies the correlation between the two (Di et al., 2013). Here, although the relationships between memory improvement and each RS measure (fALFF, FC) were negative, fALFF of the hippocampus did not correlate with connectivity between the hippocampus and the episodic memory relevant DMN (Supporting Information). Thus, although the similarities in relationships with memory improvement in the two measures can suggest an interplay or similarity as premises of cognitive plasticity, they do not seem to reflect fully the same characteristics.

4.3. Additional FC analyses as a RS comparison modality

fALFF was chosen for the RS analyses. However, an additional RS‐FC analysis was carried out in the older adults for comparison reasons of the two RS measures, as the relationship between the two is not clear. The predictive power of the RS‐FC measure in the hippocampus resembled the fALFF results in that baseline RS‐FC between the hippocampus and parts of the visual cortex, namely, the pericalcarine cortex, correlated negatively with memory improvement. The pericalcarine cortex is critical in visual processes (Henschen, 1893). As the MoL requires the ability to visualize the objects that should be memorized, the observed results may reflect an implication of functional communication between the hippocampus and visual areas of the cortex for the visualization demands of the MoL (Kosslyn, Thompson, Kim, & Alpert, 1995). However, somewhat counter‐intuitively, the participants with the higher hippocampal–visual coupling at baseline showed less memory improvement. It could be that their low memory improvement was a result of high abilities to use imagery on cognitive tasks even prior to the training. Thus, these participants may have already been close to reaching their potential of successfully applying imagery and visualization as resources for this type of word recall. Hence, although the MoL facilitates memory recall, the participants with the lower coupling might have had more room for improvement in this episodic memory recall task as a result of the training. This may indicate that the communication observed between the hippocampus and the pericalcarine cortex was a reflection of fruitful usage of imagery for this specific tasks.

Activation of the hippocampus during rest has been linked to memory performance in healthy adults (Sormaz et al., 2017), although increased hippocampal activation during task‐fMRI has been associated with poor cognitive performance and cognitive decline in elderly with MCI or in phases of prodromal Alzheimer's disease (AD) (Dickerson et al., 2005; Huijbers et al., 2015; O'Brien et al., 2010). Thus, given the somewhat unexpected negative relationship between baseline hippocampal FC HC–VIS and memory‐training improvement in the older adults, a follow‐up correlation analysis between baseline FC HC–VIS values and FC HC–VIS change (TP2–TP1) was conducted. Here, the participants with the lowest FC HC–VIS at baseline were the ones with the greater positive FC HC–VIS change after the training. This could suggest that a positive change in FC is a mediator of cognitive plasticity in older adults with low FC HC–VIS by means of fulfilling a need for higher connectivity, beyond what is expected in terms of pure regression toward the mean.

The purpose of this study was to investigate comparable brain characteristics (volume and fALFF) as predictors of memory improvement, and the FC analyses were included merely for exploratory purposes and for possible future investigation. Both RS measures correlated negatively with memory‐training improvement across the older adults. It could be that the two measures are associated in their relationships with memory improvement after training and as underlying mechanisms for memory change in normal aging. Regional hippocampal fALFF could be a reflection of the ability for hippocampal communication with other parts of the brain. However, it is still unknown to what degree there is an interplay between the two RS measures. It may be that the relationship between different measures of RS activity and cognitive function and plasticity differs with health and disease as well as age. Thus, caution should be taken when interpreting similarities or differences between the two RS measures and their implication in cognitive plasticity.

4.4. The value of multimodal imaging: Volumetric and RS characteristics of the hippocampus as unique predictors

Both volume and RS characteristics of the hippocampus were predictive of memory improvement, and both modalities served as unique contributors to the explained variance of memory improvement. The present findings point to the value of multimodal imaging in the prediction of cognitive plasticity. This is in line with previous studies showing multiple modalities to be unique predictors of AD and MCI, as well as cognitive function in development and in different stages of cognitive decline in aging (Fjell et al., 2012; Walhovd et al., 2010; Walhovd et al., 2010). This means that both structural and functional characteristics of the hippocampus are of importance for cognitive plasticity in aging and that there may be more than one route to optimal cognitive function. The present results indicate that regional brain volume, reflecting a combination of reserve factors and accumulated tissue loss, as well as spontaneous brain activity, each explain unique variance in cognitive improvement after memory training.

To the best of our knowledge, the relationship between volume and fALFF of the hippocampus in memory training has not been previously investigated. Hippocampal volume and fALFF at baseline showed opposite directions in their relationships with memory‐training improvement in the older adults. Consequently, we performed a follow‐up correlation analysis to test how these measures were related to each other. Interestingly, a negative correlation was found. While the causes for this negative relationship is unknown, it could be that low fALFF is a reflection of less age‐related changes in brain tissue. The relationship between brain morphometry and fALFF in the posterior cingulate cortex has previously shown to differ between healthy adults and MCI patients (Zhao et al., 2015). In the patient group, gray matter volume and fALFF correlated positively, while the correlation was negative in the healthy controls. This is in accordance with our results in that low fALFF indicated less volumetric age‐effects in healthy aging in an area known to be engaged in episodic‐memory function (Buckner et al., 2008). However, more studies are necessary to understand how the interplay between volume and fALFF affects cognitive plasticity. Thus far, the interpretation of our results is limited to the conclusion that both measures add to the explained variance of the memory gains. The uncertainty regarding the differences and relationship between volume and fALFF should be taken into account when combining and interpreting the two measures.

5. CONCLUSIONS

The results demonstrate that the outcome of targeted memory training can to some degree be predicted by volume and RS characteristics of the hippocampus, especially in older adults. This may be related to the variation in hippocampal characteristics associated with degenerative processes that impact memory function, even in healthy older adults. Both volume and RS characteristics were unique predictors of memory improvement in the older sample, demonstrating the added value of combining the two modalities. In addition, a negative relationship was observed between volume and fALFF of the hippocampus in the old sample, suggesting a possible coaction or dependency between the two measures in healthy aging. Further cross‐sectional and longitudinal studies combining multiple brain modalities may contribute to a broader understanding of the underpinnings of cognitive plasticity in young and older adults.

Supporting information

Supporting Information

Bråthen ACS, de Lange A‐MG, Rohani DA, Sneve MH, Fjell AM, Walhovd KB. Multimodal cortical and hippocampal prediction of episodic‐memory plasticity in young and older adults. Hum Brain Mapp. 2018;39:4480–4492. 10.1002/hbm.24287