OUP user menu

Age effects on the neural correlates of successful memory encoding

Alexa M. Morcom, Catriona D. Good, Richard S. J. Frackowiak, Michael D. Rugg
DOI: http://dx.doi.org/10.1093/brain/awg020 213-229 First published online: 1 January 2003

Summary

Event‐related functional MRI (fMRI) was used to investigate the neural correlates of memory encoding as a function of age. While fMRI data were obtained, 14 younger (mean age 21 years) and 14 older subjects (mean age 68 years) made animacy decisions about words. Recognition memory for these words was tested at two delays such that older subjects’ performance at the short delay was comparable to that of the young subjects at the long delay. This allowed age‐associated changes in the neural correlates of encoding to be dissociated from the correlates of differential recognition performance. Activity in left inferior prefrontal cortex and the left hippocampal formation was greater for subsequently recognized words in both age groups, consistent with the findings of previous studies in young adults. In the prefrontal cortex, these ‘subsequent memory effects’ were, however, left‐lateralized in the younger group but bilateral in the older subjects. In addition, for the younger group only, greater activity for remembered words was observed in anterior inferior temporal cortex, as were reversed effects (‘subsequent forgetting’ effects) in anterior prefrontal regions. The data indicate that older subjects engage much of the same neural circuitry as younger subjects when encoding new memories. However, the findings also point to age‐related differences in both prefrontal and temporal activity during successful episodic encoding.

  • Keywords: ageing; episodic encoding; event‐related fMRI; prefrontal cortex; hippocampus
  • Abbreviations: AMIPB = Adult Memory and Information Processing Battery; BOLD = blood oxygenation level‐dependent; COWAT = controlled word association test; EPI = echo‐planar imaging; fMRI = functional MRI; FWHM = full width half maximum; HRF = haemodynamic response function; MMS = Mini Mental State Test; MNI = Montreal Neurological Institute; MTL = medial temporal lobe; NART = National Adult Reading Test; RT = reaction time; SPM = Statistical Parametric Map; WMS‐R = Wechsler Memory Scale—Revised

Introduction

Healthy people often report everyday memory difficulties with increasing age. This impression is confirmed by a wealth of experimental data, which suggest that these problems are most marked when tasks involve explicit memory for specific events (episodic memory; Tulving, 1983; Light, 1991; Craik and Jennings, 1992).

Episodic memory impairments in older adults appear to derive at least to some extent from difficulties in encoding the to‐be‐remembered material, although factors operating at retrieval are also thought to play a role (e.g. see Craik and Byrd, 1982; Craik and Rabinowitz, 1985; Light, 1991). Behavioural findings suggest that older adults may show encoding deficits because they are less likely spontaneously to engage effective encoding strategies (Craik 1983). Retrieval difficulties have been inferred from the consistent finding that age‐associated deficits in episodic memory are more marked for free recall than recognition tests (e.g. Schonfield and Robertson, 1966; Craik and McDowd, 1987). A related finding is that older adults are less likely to experience a detailed recollection of previously encountered events and, at retrieval, may rely to a greater extent on a general feeling of familiarity (Parkin and Walter, 1992; Jennings and Jacoby, 1997). There is also evidence that older adults’ memory is impaired more for details of the context in which items are presented than for the items themselves (for review see Spencer and Raz, 1995). It is difficult, however, to dissociate age effects on encoding and retrieval using behavioural measures alone. Neuroimaging provides an important complementary perspective, allowing the separate identification of brain regions that are active during memory encoding and retrieval (Fletcher et al., 1997).

As well as impairments in episodic memory, ageing brings with it a broader cognitive decline (for review see Light, 1991). Some theoretical accounts therefore place memory deficits in a wider framework, for example explaining age‐associated deficits at encoding and retrieval in terms of more general impairments in semantic and associative processing (Craik and Byrd, 1982), or in the ability to employ appropriate self‐initiated strategies (Craik, 1983). Other approaches are more general still, the most important being the notion that reductions with age in processing speed or in other generic processing resources can account for apparently selective deficits on various cognitive tasks (for review see Salthouse, 1996). Thus, if the same tasks are given to younger and older adults, age‐related differences in performance may merely be a function of the fact that the tasks are more difficult for older subjects, rather than a direct reflection of ageing itself. Exactly the same argument can be applied to studies of the neural correlates of ageing, so it is crucial to address the problem of task difficulty in neuroimaging studies of ageing.

Neural correlates of encoding in young subjects

Numerous functional neuroimaging studies of memory encoding in young adults have been reported (for review see Buckner et al., 1999). Most of these have employed a blocked experimental design, where neural correlates of encoding are defined using comparisons between activity associated with different study tasks (Kapur et al., 1994, 1996; Shallice et al., 1994; Demb et al., 1995; Kelley et al., 1998; Wagner et al., 1998a,b). Such studies have implicated a network of brain regions, most notably in the prefrontal cortex and, less consistently, in the medial temporal lobe (MTL) (Squire and Knowlton, 2000; see also Dolan and Fletcher, 1997). For verbal materials, study tasks that result in relatively good memory performance (generally, tasks requiring semantic analysis of the study items) are consistently associated with greater activity in a number of regions of left prefrontal cortex, particularly left inferolateral regions (Buckner et al., 1999). There is some evidence that this lateralization is reversed when non‐verbal material is studied (e.g. Kelley et al., 1998; Wagner et al., 1998b; McDermott et al., 1999).

An alternative approach, enabled by the advent of event‐related functional MRI (fMRI) (Dale and Buckner, 1997; Josephs et al., 1997; Zarahn et al., 1997), is to identify brain regions where activity during the performance of encoding tasks predicts successful item memory. Such ‘subsequent memory’ comparisons between activity associated with subsequently remembered and subsequently forgotten items have generally implicated a subset of the regions activated by comparisons between semantic and non‐semantic study tasks, including prefrontal and medial temporal areas (Brewer et al., 1998; Wagner et al., 1998a; Henson et al., 1999; Kirchhoff et al., 2000; Baker et al., 2001; Buckner et al., 2001; Davachi et al., 2001; Otten and Rugg, 2001a; Otten et al., 2001). Such item‐specific analyses have, however, also identified brain regions that, while insensitive to ‘depth of processing’ manipulations, are nonetheless associated with successful encoding. Most notable among such regions are the right prefrontal cortex and left posterior hippocampus (Otten and Rugg, 2001a; Otten et al., 2001).

Neural correlates of encoding and encoding deficits in older subjects

A number of studies have investigated age effects on memory encoding (for review see Grady and Craik, 2000). So far, all have employed a blocked design. Some of the studies that have used verbal material have reported reduced encoding‐related activity in older compared with younger subjects in left prefrontal regions (Grady et al., 1995; Cabeza et al., 1997; Anderson et al., 2000; Stebbins et al., 2002) and, less consistently, in medial temporal regions (Gradyet al., 1995, 1999). In other studies, however, older and younger subjects have shown equivalent levels of activity in the regions normally activated in the young (Schacter et al., 1996; Madden et al., 1999). Recent work by Logan and colleagues (Logan et al., 2002; but see Stebbins et al., 2002) has gone some way towards resolving these apparently discrepant findings by comparing ‘intentional’ encoding, where subjects determine their own study strategy, and ‘incidental’ encoding, where a task, such as semantic decision, is imposed by the experimenter. ‘Under‐recruitment’ of left prefrontal areas by older subjects was observed during the intentional but not the incidental task. These findings are consistent with evidence from behavioural studies that older subjects have difficulty initiating encoding strategies and benefit more than younger subjects from ‘environmental support’, such as explicit directions to process items for meaning (Craik and Simon, 1980; Backman, 1986). It is, however, unclear from such findings whether age effects on episodic encoding result only from a failure to engage effective encoding strategies. In particular, it is important to assess whether the processes and neural structures that subserve successful encoding change with age.

Age differences in brain activity during memory encoding do not, however, appear to consist simply of a reduction in activity in regions seemingly important for encoding in younger adults. Some brain regions appear to be more active during memory encoding in older subjects. For verbal material, these regions are found predominantly in the right prefrontal cortex (Cabeza et al., 1997; Madden et al., 1999; Logan et al., 2002; see Cabeza, 2002), and may include areas homotopic with those active in young subjects in the left prefrontal cortex (Madden et al., 1999; Logan et al., 2002).

This pattern of findings has been formalized as the ‘hemispheric asymmetry reduction in older adults’ (HAROLD) account of aging (Cabeza, 2002). This proposes that prefrontal activity in both episodic memory and other cognitive tasks, particularly those tapping working memory, becomes less lateralized with increasing age. As noted already, for any given encoding task, older subjects usually remember less of the study material than do younger subjects. Thus, age‐associated differences in brain activity may reflect differences in the efficiency of encoding, rather than age effects per se. This problem can be partially offset by the use of an event‐related design, which enables a specific comparison to be made of the neural correlates of successful memory encoding across age groups. However, a more subtle issue remains. If there is a reduced rate of subsequent recognition by the older compared with the younger age group, the items remembered by the older group may be comparable only to a subset of the items remembered by the younger group. For example, in a verbal memory task, older subjects might tend to remember only those items that are particularly imageable or that have many semantic associates. Such factors may influence the neural correlates of encoding (e.g. Schmidt et al., 2002), and thus be confounded with age effects on these correlates.

Using event‐related fMRI, we examined the effects of age on the neural correlates of successful memory encoding during an incidental semantic processing task. The issue of task difficulty was addressed by employing two different delays for the subsequent memory test, such that item recognition in both age groups was lower at the long than at the short delay. Thus, the level of subsequent memory performance could be equated between the age groups by examining subsequent memory effects for items tested in the older subjects at the short delay and in the younger subjects at the long delay. This allowed age differences in subsequent memory effects to be explored when recognition test difficulty was the same, as well as different, for the two groups. Because an incidental encoding task was used, it was predicted that similar subsequent memory effects would be seen in left inferolateral prefrontal regions in both older and younger subjects (Logan et al., 2002). In keeping with the existing data for verbal material, no age differences were expected in subsequent memory effects in the MTL (Cabeza et al., 1997; Madden et al., 1999; Anderson et al., 2000). The experiment also afforded the opportunity to investigate whether the more bilateral pattern of prefrontal activity reported previously in older subjects during episodic encoding (Cabeza, 2002) extends to subsequent memory effects and to circumstances where memory performance is equated.

Methods

Subjects

Subjects were 28 healthy, right‐handed adults, 14 from the age range 18–29 years and 14 from the age range 63–74 years. The younger age group contained three men and the older age group five men. Two additional younger subjects recruited originally were excluded on the basis of low IQ test scores. One additional older subject was excluded because of a recent medical event and two on the basis of neuropsychological test scores (see below). Two further younger subjects were excluded after fMRI scanning because they made too few confident responses in the recognition memory test (see below). Younger subjects were undergraduate and post‐graduate students, and older subjects were all high‐functioning community dwelling individuals. As Table 1 indicates, the two groups were matched for level of education. All subjects were right‐handed British English speakers, and had normal or corrected‐to‐normal vision. Volunteers were excluded from taking part if they had a history of significant neurological, cardiovascular or psychiatric illness, or any other serious systemic condition, or if they were taking any neurotropic or vasoactive medication. Informed consent was obtained prior to participation and the experimental procedures were approved by the Joint UCL and UCLH Committees on the Ethics of Human Research, and by the Institute of Neurology and National Hospital for Neurology and Neurosurgery Joint Research Ethics Committee.

View this table:
Table 1

Subject characteristics by age group, and performance on standardized neuropsychological tests

YoungerOlder P
Age21 (1.6)68 (3.3)
Years of education (from 16 years)5 (1.9)4 (2.6)n.s.
NART FSIQ estimate113 (6.6)119 (7.9)<0.05
Raven’s Advance Progressive Matrices II11 (0.9)9 (1.6)<0.01
Mini Mental State 29 (0.7)
Warrington–McKenna Graded Naming 23 (3.2)25 (4.1)n.s.
WAIS digit span10 (2.1)9 (2.3)n.s.
Verbal paired associates (WMS) – immediate23 (1.9)19 (4.1)<0.05
Verbal paired associates (WMS) – delayed8 (0.0)7 (1.3)<0.01
Short story recall (AMIPB) – immediate41 (8.0)35 (5.2)<0.01
Short story recall (AMIPB) – delayed39 (8.5)30 (7.0)<0.01
COWAT47 (9.7)51 (10.5)n.s.

Mean scores are shown (SD). P values reflect the results of t tests, except in the case of years of education, digit span and paired associates measures, where they reflect the results of Mann–Whitney U tests. n.s. = not significant. COWAT = controlled word association test. WAIS = Wechsler Adult Intelligence Scale. WMS = Wechsler Memory Scale.

Neuropsychological testing

All subjects were administered a battery of standardized neuropsychological tests to assess intelligence, and memory and language functioning. These tests were given in a separate 1½ h session prior to the MRI scanning session. The battery targeted cognitive functions that are impaired with age, along with some generally found to be spared. Older subjects were first given the Folstein Mini Mental State test (MMS) (Folstein et al., 1975) as a screening measure, with a minimum score of 26 out of 30 required for inclusion in the study (Lezak, 1995). The National Adult Reading Test (NART) was given as a measure of ‘crystallized’ verbal intelligence (Nelson, 1982), and Raven’s Advanced Progressive Matrices II were used as a measure of ‘fluid’ non‐verbal intelligence (untimed) (Spearman, 1927; Raven et al., 1994). Digit Span Forward from the Wechsler Memory Scale—Revised (WMS‐R) battery (Wechsler, 1987) was employed as a test of verbal short‐term memory expected to be unaffected by age. Two long‐term memory measures predicted to show impaired scores in the older group, particularly on delayed tests, were the Adult Memory and Information Processing Battery (AMIPB) Logical Memory short story recall test (immediate and delayed) (Coughlan and Hollows, 1985), and the WMS‐R Verbal Paired Associates Test (Wechsler, 1987). The Warrington‐McKenna Graded Naming Test was also included as a test of picture comprehension and naming ability, and as a further dementia screening test, using norms specifically generated for this age group (Clegg and Warrington, 2000). Finally, verbal fluency was assessed using the controlled word association test (COWAT) with the letters F, A and S (Lezak, 1995). As already noted, two older subjects with age‐inappropriate scores on more than one test (other than the MMS) were excluded from the study.

Stimulus materials

The stimulus lists were constructed from a pool of 557 words between four and nine letters in length, with a frequency of between 1–30 per million (Kucera and Francis, 1967). Three sets of 80 words each were selected at random from this pool, with the constraints firstly, that within each set, 40 words would be animate and 40 inanimate, and secondly, that the distribution of word lengths would be the same across the three sets and across animate and inanimate items. These sets were used to form three different study‐test list combinations by rotating the sets across the studied (160 items) and new (80 items) conditions. Each volunteer saw one of the three possible list combinations. For each study list, a pseudo‐random trial sequence was generated from the 160 words to be studied plus 80 fixation only trials on which no word was to be shown, such that no more than three events of a given type (animate, inanimate, fixation only) would occur in a row. This study sequence was divided into two blocks of 120 trials, and two filler words were added to the beginning of each block. A further 12 words were selected from the word pool to create two practice lists for the study task.

For the recognition memory test, two separate lists were constructed for each subject, one to be seen at the short and one at the long test delay. Each of the two lists comprised a random selection of half of the studied (old) words (80 items) and half of the new words (40 items). Items were ordered according to a pseudo‐random sequence, such that no more than three old or three new words would occur in a row. Each test list was divided into two blocks, and two filler words were added at the beginning of each block. Six additional unstudied items were selected from the word pool and combined with the 12 studied items from the second practice study list to create an 18‐item practice list for the test task.

Experimental tasks and procedures

The experimental procedure consisted of an incidental study task followed by a recognition memory test after a delay of 10–15 min (short test delay), and a further recognition memory test 30 min after the first test (long test delay).

Procedure at study

Scanning took place during the study task only. Subjects were told that they would see words on the screen, presented one at a time, and that their task was to decide whether or not the word referred to, pertained to, or was part of an animate object. Half of the subjects were asked to respond with their left forefinger if a word was animate, and their right forefinger if it was inanimate. For the other half, the opposite response assignment was employed. Instructions emphasized both speed and accuracy. There were two brief practice sessions, one before entering the scanner and one in the scanner, so that subjects could familiarize themselves with the study task and environment. They were not told that they would later perform a memory test. Subjects were positioned reclining in the scanner, with stimuli projected onto a mirror in direct view. Responses were made using a hand‐held response box.

Once in the scanner, subjects underwent a 15‐min structural scan. After this, the second practice session was carried out with the scanner operating. Subjects performed two blocks of the study task, each of 122 trials, during which the functional scans were acquired in a single session. During the study task, a series of 160 critical words were shown, one at a time, in pseudo‐random order, interspersed with 80 fixation only trials. Between trials, and on fixation only trials, subjects saw a white ‘+’ sign in the centre of the screen. The presentation of each word was preceded for 650 ms by a warning signal; this was a change in the colour of the fixation symbol from white to red. The screen was then blanked for 150 ms, following which the word was shown for 300 ms. Stimuli were presented visually in white uppercase Arial 30 point font on a black background. Both word trials and fixation‐only trials were of 3.0 s duration; this was the minimum time between successive word onsets. The words subtended an approximate vertical visual angle of 1° and a horizontal visual angle of 3–10°. There was a further 150 ms blank period between the offset of the word and the onset of the fixation symbol, which remained on‐screen for 1750 ms. Between the two study blocks, there was a short break in the task of total duration of 10 trials (30 s). Subjects saw the message ‘Short break’ for 1200 ms, followed by a blank screen. The message ‘Get ready’ then preceded the first trial of the second block. Subjects were required to remain immobile during this rest period.

Procedure at test

After completing the study task, subjects were taken from the scanner to another room, where they were informed about the first recognition memory task. It was explained that they would again be shown a sequence of words, some of which they had seen during the task they had performed in the scanner. Subjects were not informed as to the nature of the final test (the long delay memory test) until just before they were asked to perform it.

For each word, subjects were asked to decide whether they had seen it before during the experiment, and to indicate whether or not they were confident about that decision. No specific instruction was given about how sure someone should be about their response before pressing the confident key. One of four keys was pressed according to whether the word was (i) confidently judged to be old, (ii) non‐confidently judged to be old, (iii) confidently judged to be new, or (iv) non‐confidently judged to be new. Responses were made with the middle and index fingers of the left and right hands, which rested on the a, z, m, and k keys of a keyboard placed on a table in front of the volunteer. The middle fingers were always used for confident responses (z and m keys). The assignment of old responses to the left or right hand was counterbalanced across subjects. Instructions were to respond as fast as possible without sacrificing accuracy. A brief practice was given to familiarize subjects with the task.

The tests at the short and the long delay were identical in format. At each test delay, subjects undertook two test blocks of 62 trials, each ∼5 min in duration. For each recognition memory test, half of the words seen at study were presented again (80 old items), along with half of the list of words not seen before in the experiment by that subject (40 new items). Half of both the old and the new items were animate, and half inanimate. On each trial, a plus sign was first presented for 650 ms, as a fixation point and warning signal, followed by a 150 ms blank period before the onset of the word. All stimuli were presented visually on a computer screen, using the same format and font as at study. The words subtended a visual angle of 2–4° and were displayed for 300 ms. The screen was then blanked until the arrival of the next fixation symbol, 1.0 s before the onset of the subsequent word. The time between the subject’s response and the arrival of the next stimulus was 3.0 s. Short rests were given between test blocks. After the short delay test, subjects read or conversed with the experimenter for 30 min. Following this, they were asked to perform the second, long delay test. When this was completed, subjects were debriefed about the nature of the experiment.

fMRI procedure

Data acquisition

Both T1‐weighted anatomical volume images (1 × 1 × 1 mm voxels, MPRAGE sequence) and T2*‐weighted echoplanar (EPI) images [64 × 64, 3 × 3 mm pixels, TE (echo time) = 40 ms] with blood oxygenation level dependent (BOLD) contrast were acquired using a 2T Siemens VISION system (Siemens, Erlangen, Germany). Each EPI volume comprised 32 2.5 mm thick axial slices separated by 1.5 mm, positioned for full coverage of the cerebrum but not of the cerebellum. Data were acquired continuously during a single session and comprised 337 volumes, with an effective repetition time (TR) of 2.43 s/volume. The first five volumes were discarded to allow for T1 equilibration effects. The constant inter‐stimulus interval of 3 s allowed an effective sampling rate of the haemodynamic response of 1.67 Hz.

Preprocessing

Preprocessing and data analysis were carried out using Statistical Parametric Mapping (SPM99, Wellcome Department of Cognitive Neurology, London, UK; Friston et al., 1995). For each volunteer, all volumes in a session were realigned spatially to the first volume and re‐sliced using a sinc interpolation in space. To correct for their different acquisition times, the signal measured in each slice was then shifted relative to the acquisition of the middle slice using a sinc interpolation in time. Inspection of movement parameters generated during spatial realignment revealed that no subject moved more than ±2 mm in any direction during the session. No systematic differences were apparent between age groups in this respect. Each volume was normalized to a standard EPI template volume based on the Montreal Neurological Institute (MNI) reference brain (Cocosco et al., 1997) of 3 × 3 × 3 mm voxels in the space of Talairach and Tournoux (1988) using non‐linear basis functions (Ashburner and Friston, 1999). The T1 structural volume was co‐registered with the mean realigned EPI volume and normalized with the same deformation parameters. The EPI volumes were then smoothed with a 10 mm full width half maximum (FWHM) isotropic Gaussian kernel. This relatively large smoothing kernel was chosen (compared with similar studies in younger subjects, e.g. Otten et al., 2001) to compensate for the greater degree of anatomical variation anticipated across the two age groups (e.g. Good et al., 2001). The data were high‐pass filtered to a maximum of 1/120 Hz, and no low‐pass filter was used.

Data analysis

Population inferences were made using a two‐stage procedure. In the first stage, the volumes acquired for each subject were modelled as a continuous time series. Stimuli at study attracting a correct response were classified into two main event types, subsequently ‘remembered’ and subsequently ‘forgotten’ (see Results). In subsidiary analyses employed as masks, they were further classified according to whether they were presented at test after a long vs. a short memory test delay (see Results). The haemodynamic response to the onset of each event type of interest was modelled with two basis functions: a canonical haemodynamic response function (HRF) (Friston et al., 1998), and a delayed HRF (Henson et al., 2000), shifted 2.43 s later in time (i.e. by one TR) than the canonical HRF. Both an ‘early’ and a ‘late’ response function were employed for several reasons. First, in healthy young adults there is evidence that for some brain regions (e.g. the anterior prefrontal cortex), maximal activation occurs later post‐stimulus than in the sensory regions on which the canonical HRF is based (Schacter et al., 1997; Buckner et al., 1998; Henson et al., 2000). Secondly, as older adults’ cognitive performance is generally slower (e.g. Salthouse, 1996), event‐related responses might be expected to peak later than in younger subjects. Finally, the timing of the haemodynamic response could be affected by age‐related changes in the responsiveness of the cerebral vasculature. (That said, investigations of age effects on the BOLD response in primary visual and motor cortices have not revealed any evidence of age‐related changes in the onset or duration of event‐related responses, although reduced amplitude, increased variability and subtle differences in shape are sometimes seen, e.g. D’Esposito et al., 1999; Huettel et al., 2001).

The early and late response functions, when convolved with a sequence of delta functions representing the onset of each event, comprised the covariates in a general linear model, together with a constant term for each session. The covariates for the late HRF were orthogonalized with respect to those for the early HRF using a Gram–Schmidt procedure, giving priority to the early covariate (Andrade et al., 1999). Thus, variance common to the early and late covariates was attributed to the early covariate, loadings on the orthogonalized late covariate accounting only for residual variance in the data unexplained by the early covariate. Parameter estimates for each condition and covariate were calculated from the least mean squares fit of the model to the data. The data were proportionally scaled to a global mean of 100 across each session. Also included for each subject were six covariates to capture residual movement‐related artefacts (the three rigid body translations and rotations determined from the realignment stage).

In the second stage of the analysis, parameter estimates for both early and late covariates were tested using planned contrasts (as specified in the Results section). The linear combinations of parameter estimates for each contrast were stored as separate images for each subject. These contrast images were entered into one‐ and two‐sample t tests, to permit inferences about condition effects across subjects, and about group differences, that generalize to the population (i.e. a ‘random effects’ analysis). These contrasts produced statistical parametric maps (SPMs) of the t statistics at each voxel, which were subsequently transformed to the unit normal Z distribution. Unless otherwise specified, contrasts were one‐ or two‐ sample t‐tests, and were thresholded at P < 0.001, uncorrected for multiple comparisons. Only activations involving contiguous clusters of at least five voxels were interpreted. For masked contrasts, both the mask and the final image were thresholded at P < 0.001. When reporting masked contrasts, the Z values refer to the outcome of the masked contrast only. The locations of maxima of suprathreshold regions were established by rendering them onto both the volunteers’ normalized structural and mean EPI images, and the MNI reference brain (Cocosco et al., 1997). They were labelled using the nomenclature of Talairach and Tournoux (Talairach and Tournoux, 1988) and anatomical designations of Brodmann (Brodmann, 1909). Results from the late covariate are reported here only when they add meaningfully to the findings from the early covariate. Plots of the BOLD response to a given stimulus type in peri‐stimulus time represent data derived using a Finite Impulse Response model adjusted for movement confounds and activity due to error trials (Henson et al., 2001).

Results

Neuropsychological test performance

The results of the neuropsychological test battery are summarized in Table 1. These indicate that although the older group showed slightly superior verbal IQ as indicated by performance on the National Adult Reading Test, their long‐term memory was, as expected, significantly poorer. Fluid IQ, as measured by the Raven’s Advanced Progressive Matrices, was also significantly lower in the older group relative to the younger group.

Behavioural performance

Study task

Animacy decisions were made with an accuracy of 96% by both the younger and the older subjects. Mean reaction time (RTs) for items correctly classified at study were analysed according to subsequent memory performance, age group and memory test delay, in parallel with the analysis of the fMRI data reported below. Following Otten et al. (2001), words were categorized as ‘remembered’ only when they attracted a confident correct response in the subsequent recognition memory test. Those attracting either a non‐confident hit or a miss were classified as ‘forgotten’. For the younger subjects, the mean RT for remembered items was 779 ms (SD = 119), compared with 750 ms (SD = 118) for forgotten items. For the older subjects, these mean RTs were 850 ms (SD = 118) and 824 ms (SD = 98), respectively. Although the older group seemingly responded more slowly than the younger group, the main effect of age was not significant [F(1,26) = 2.90, not significant]. Responses to subsequently remembered study items were, however, reliably slower than those to forgotten items [F(1,26) = 16.80, P < 0.001]. Crucially, there was no interaction of the effects of age and subsequent memory on RT (F < 1, not significant). There were also no significant effects of memory test delay on study RT.

Recognition memory

Performance of older and younger subjects on the recognition memory test is shown in Table 2. Memory accuracy was measured by the discrimination index Pr, the difference between the probability of a hit (Phit) and the probability of a false alarm (Pfalse alarm) for both confident and non‐confident responses (Snodgrass and Corwin, 1988). The effects of age group, response confidence and memory test delay on Pr were assessed using analysis of variance (ANOVA), after excluding data from one older subject who generated no low confidence responses at the short test delay.

View this table:
Table 2

Recognition memory performance by age group, for old and new words tested at the short and the long delay

YoungerOlder
Short delay
 Old itemsConfident hits0.73 (0.11)0.70 (0.14)
Non‐confident hits0.09 (0.06)0.09 (0.06)
Confident misses0.10 (0.06)0.16 (0.10)
Non‐confident misses0.08 (0.05)0.06 (0.05)
 New itemsConfident correct rejections0.70 (0.15)0.67 (0.22)
Non‐confident correct rejections0.18 (0.12)0.13 (0.11)
Confident false alarms0.07 (0.05)0.13 (0.09)
Non‐confident false alarms0.06 (0.05)0.08 (0.07)
Long delay
 Old itemsConfident hits0.67 (0.13)0.65 (0.15)
Non‐confident hits0.12 (0.07)0.12 (0.09)
Confident misses0.12 (0.08)0.15 (0.11)
Non‐confident misses0.10 (0.07)0.08 (0.06)
 New itemsConfident correct rejections0.68 (0.17)0.60 (0.26)
Non‐confident correct rejections0.19 (0.15)0.15 (0.10)
Confident false alarms0.06 (0.07)0.16 (0.16)
Non‐confident false alarms0.06 (0.05)0.09 (0.09)

Values are across‐subject mean proportion of responses (SD).

The two age groups differed in that recognition memory was significantly more accurate for the younger than for the older group [F(1,25) = 7.88, P < 0.01], but this effect of age was unmodified by confidence or by test delay. A reliable interaction of response confidence with test delay [F(1,25) = 12.42, P < 0.005] and main effect of confidence [F(1,25) = 301.82, P < 0.001] reflected the fact that for confident responses, memory was more accurate at the short than at the long test delay [F(1,25) = 14.12, P < 0.001], whilst for non‐confident responses there was a non‐significant trend in the opposite direction; again, these effects were not influenced by age (for all interactions of age group with test delay and with confidence, F <1, not significant).

A separate ANOVA was conducted to assess the extent to which low confidence responses reflected veridical memory. This compared the rate of hits with the rate of false alarms among low confidence responses for young and old subjects at the two memory test delays. Discrimination of old from new items was slightly better than chance [for main effect, F(1,25) = 4.62, P < 0.05, mean hits = 0.11, mean false alarms = 0.08], but there was no interaction with age.

Subjects’ willingness to classify an item as ‘old’ was measured by the response bias index Br [Pfalse alarm/1–(PhitPfalse alarm)]. Values of Br >0.5 indicate a liberal bias (Snodgrass and Corwin, 1988). Response bias was not reliably affected by age group or by memory test delay (mean Br for younger group = 0.14, SD = 0.06; for older group mean = 0.21, SD = 0.12).

The above findings indicate that there was markedly better discrimination between old and new words for confident than for non‐confident responses. This is in keeping with previous data (Wagner et al., 1998a; Brewer et al., 1998; Otten et al., 2001) and supports the categorization as ‘remembered’ of only those words attracting a correct confident ‘hit’. The response times for older and younger subjects in the recognition memory test were analysed according to age, subsequent memory and test delay. Mean RTs at the short test delay were 1145 ms (SD = 216) for remembered and 1845 ms (SD = 377) for forgotten items; the corresponding means for the long delay were 1334 ms (SD = 272) and 1370 ms (SD = 296), respectively. There was an interaction of subsequent memory with test delay [F(1,26) = 89.86, P < 0.001], because RTs to ‘forgotten’ items were slower than those to ‘remembered’ items only at the short memory test delay [t(27) = 13.30, P < 0.001]. Age did not reliably influence test RT [for main effect, F(1,26) = 1.78, not significant; for interactions, F(1,26) <1, not significant].

Planned comparisons were carried out on the recognition performance of younger subjects at the long memory test delay, and of older subjects at the short delay. This comparison was expected to yield no differences between the two age groups, thus validating the equivalent imaging analyses as contrasts unconfounded by differences in the level of subsequent memory. As expected, there was no significant age difference in discrimination for confident responses [t(26) <1].

fMRI findings

All analyses of fMRI data were confined to study trials associated with a correct animacy decision. There were three principal comparisons. First, to explore age‐invariant subsequent memory effects, the data for both age groups were collapsed across the short and the long memory test delay, and exclusive masking was employed to discount any voxels where subsequent memory effects interacted with either age or test delay. A second analysis searched for differences between younger and older subjects in subsequent memory effects, again excluding voxels where these effects were influenced by the time at which memory was tested. Thirdly, planned comparisons were conducted both to examine subsequent memory effects in a priori defined regions of interest, and to test the hypothesis that the lateralization of subsequent memory effects is reduced in older compared with younger subjects.

In all of the above analyses, age‐associated differences in subsequent memory effects were correlated with the variation in memory performance with age. A final set of contrasts was therefore carried out using data for which performance was equated between the two groups on the subsequent memory test. Specifically, a comparison between the two groups was made that included fMRI data for study items tested after a short delay for the older group and data for items tested after a long delay for the younger group.

Subsequent memory effects common to both age groups

To identify regions where subsequent memory effects were common to both age groups and test delays, contrasts were computed comparing the signal for remembered and forgotten items for all subjects together across both delays. These contrast images were then exclusively masked with the images derived from tests of the interactions between subsequent memory and age, subsequent memory and test delay, and all of these factors. The results of this analysis are summarized in Fig. 1 and Table 3.

Fig. 1 Main effect of subsequent memory common to both age groups, exclusively masked with interactions of age with subsequent memory and test delay, age with subsequent memory, and subsequent memory with test delay. Prefrontal activations are shown rendered onto the MNI reference brain. The hippocampal activation is displayed on the smoothed averaged T1 image for all subjects in the analysis. Plots show event‐related data in terms of the across‐subject mean percentage signal change (relative to grand mean over voxels and scans) against peri‐stimulus time, adjusted for confounds and binned every 2 s. BA = Brodmann area.

View this table:
Table 3

Principal regions showing significant (P < 0.001, cluster size >5) signal increases, common to both age groups, on the early covariate for words that were subsequently remembered versus forgotten

Location (x, y, z)Peak ZRegion Brodmann area
–48, –54, –185.33 L fusiform gyrus37
–45, –57, –155.19L fusiform gyrus37
–42, –45, –184.88L fusiform gyrus37
–57, –45, –35.14L middle temporal gyrus21
–42, 30, –64.64L inferior frontal gyrus 47
–48, 6, 304.85L inferior frontal gyrus 44
–39, 33, 124.55L inferior frontal gyrus45/46
–42, 21, 214.72L inferior frontal gyrus45
–27, 15, 514.81L middle frontal gyrus6/8
–30, 3, 544.56L middle frontal gyrus6
–48, 6, 424.38L middle frontal gyrus9/8
–6, 39, 454.26L medial frontal gyrus8
–6, 24, 484.54L medial frontal gyrus/ AC8/32
–30, –15, –154.25L hippocampus
–21, –12, –184.69L parahippocampal gyrus28
–39, –33, –184.55L parahippocampal gyrus36
–27, 6, –274.18L superior temporal gyrus38
–9, 48, 364.00L medial frontal gyrus8
–3, –3, 603.99L medial frontal gyrus10
–18, 0, –243.92L parahippocampal cortex/amygdala34
–48, –3, 453.92L middle frontal/ precentral gyrus9/6
3, –24, 603.73R medial frontal gyrus6
24, –12, –213.21R parahippocampal gyrus28/34
39, –63, –154.17R occipital
–54, 3, –213.82 L inferior temporal gyrus20
6, –90, 33.81R cuneus 17
21, –78, 153.69R cuneus 18
33, –66, 363.68 R precuneus 7
48, 6, 333.66 R inferior frontal gyrus44
57, 3, –183.66 R middle temporal gyrus21
57, –33, 03.63 R middle temporal gyrus21
–21, –45, 603.63L superior parietal gyrus11
15, –51, –63.60R cerebellum
27, –72, 543.56R precuneus/superior parietal gyrus7
33, –60, 603.56R superior parietal gyrus7
18, –87, 363.54R cuneus 19
0, –87, –93.53R occipital lingual gyrus17/18
42, –9, 153.48R insula
–18, –93, 63.45L cuneus17
–27, –24, 63.43L thalamus
0, 3, 363.42L cingulate gyrus24
27, –54, 303.31R cingulate gyrus31

Location is with respect to the system of Talairach and Tournoux (1988). Z values refer to the peak of the activated cluster, the size of which is indicated in brackets.

A wide network of brain regions manifested subsequent memory effects that did not differ significantly according to age group or test delay. Among these areas were the inferolateral prefrontal cortex and a more superior lateral prefrontal region. There were prominent bilateral medial temporal activations, with peaks in the left anterior hippocampus and more posteriorly in the right hippocampus, as well as in bilateral parahippocampal areas. All of these activations were more extensive on the left than on the right. Other regions showing subsequent memory effects included the left fusiform gyrus, right precuneus and bilateral cuneus. Areas loading on the late covariate represented a subset of the areas loading on the early covariate.

Age effects on subsequent memory

Age differences in subsequent memory effects were examined by computing contrasts for the interaction of remembered versus forgotten items for younger versus older subjects using data collapsed over the short and the long memory test delay. To remove any influence of effects pertaining only to items remembered at just one of the two test delays, these interaction images were exclusively masked with those for the interaction of subsequent memory with age group and test delay. Regions identified from contrasts carried out on parameter estimates for the early covariate are shown in Fig. 2 (see also Table 4).

Fig. 2 Interactions of age group with subsequent memory, showing (A) the region where younger subjects showed a greater subsequent memory effect and (B) principal regions where older subjects showed a greater subsequent memory effect. Activations are displayed on the smoothed averaged T1 image for all subjects in the analysis. Graphs show the difference at peak voxels averaged across subjects for subsequently remembered minus forgotten items (arbitrary units). Error bars represent the between‐subject standard error. BA = Brodmann area.

View this table:
Table 4

Regions showing significant (P < 0.001, cluster size >5) interactions of age group with subsequent memory, loading on the early covariate

Location (x, y, z)Peak ZRegionBrodmann area
Younger >older
 –33 –3 –363.43 L anterior inferior temporal cortex20
Older >younger
 –24, 42, 123.92 L anterior prefrontal cortex10
 –21, 54, 03.28 L anterior prefrontal cortex10
 27, 57, 0 3.47 R anterior prefrontal cortex10
 –45, –51, 243.87 L supramarginal gyrus40
 6, –60, 33.59 Lingual gyrus 18
 9, –99, 93.45 Occipital pole18/19

The only area manifesting a greater subsequent memory effect in the younger than in the older subjects was in left anterior inferior temporal cortex [Brodmann area (BA) 20]. As can be seen from Fig. 2A, there was a subsequent memory effect in this region for the young group only. (Note that clear responses were evident in the older subjects for both remembered and forgotten items, so the difference in subsequent memory effects between groups could not have been secondary to differences in signal dropout due to susceptibility artefact. This observation also applies to other regions showing interactions of age with subsequent memory.)

In other brain regions, significantly greater subsequent memory effects were found in the older than in the younger subjects. These regions included bilateral anterior prefrontal cortex and the left supramarginal gyrus. The interaction with age in these areas resulted, at least in part, from the existence of so‐called ‘subsequent forgetting’ effects in the young group only, i.e. greater signal for subsequently forgotten than for subsequently remembered items (Otten et al., 2001; see Fig. 2B). In a further two regions—the lingual cortex and near to the occipital pole—interactions of age and subsequent memory reflected subsequent memory effects for the older age group only (see Table 4).

Age effects on subsequent memory controlling for memory performance

To what extent was the pattern of findings in the foregoing analyses secondary to the differences in memory performance between the two age groups? To address this issue, a further set of planned contrasts were computed for subsequently remembered versus subsequently forgotten items tested in younger subjects at the long delay compared with older subjects at the short delay (see behavioural results). Data from 13 subjects in each group were entered into this analysis (one subject from each age group was excluded because they generated fewer than 12 forgotten trials).

This analysis revealed a similar pattern of results to the overall analysis. A region in the left anterior inferior temporal cortex again loaded significantly on the early covariate (BA 20, 12 voxels, x = –36, y = –3, z = –33, Z = 3.57), with a subsequent memory effect only for the younger age group. Of the anterior prefrontal regions manifesting a greater subsequent memory effect in the old than in the young in the overall analysis, the more medial area in left BA 10 was also identified in the present comparison (19 voxels, x = 24, y = 42, z = 15, Z = 4.45). An additional left prefrontal region was also identified in this contrast (BA 10, 5 voxels, x = –33, y = 36, z = 30, Z = 3.33). As in the overall analysis, both of these regions manifested subsequent forgetting effects in the younger group, and (smaller) subsequent memory effects in the older group.

Lateralization of subsequent memory effects

To investigate age‐related differences in subsequent memory effects, and specifically in their lateralization in the prefrontal cortex and medial temporal lobe, a hypothesis‐driven region of interest analysis was carried out. To avoid bias, the regions were selected on the basis of the findings of Otten and colleagues (see Table 2 of Otten et al., 2001) from a subsequent memory experiment employing the same animacy decision task as that employed here. Parameter estimates were extracted from three bilateral pairs of voxels of interest, in anterior and posterior regions of the inferior prefrontal gyrus (BA 47, x = ±36, y = 36, z = –9 and BA 45, x = ±51, y = 27, z = 18) and in dorsolateral prefrontal cortex (BA 9/44, x = ±51, y = 12, z = 21), and from a pair of voxels in the anterior hippocampal formation (x = ±27, y = –15, z = –12). Voxel values represented the weighted mean of activity of a region corresponding to the 10 mm FWHM Gaussian smoothing kernel applied to the data. Data from the prefrontal voxels were analysed by ANOVA with factors of group, subsequent memory (remembered versus forgotten), region and hemisphere. Only those effects involving either the factor of group or subsequent memory are reported. Main effects were found for group [F(1,26) = 15.89, P < 0.001] and subsequent memory [F(1,26) = 21.07, P < 0.001]; significant interactions were found for region with group [F(1,26) = 9.48, P < 0.001], subsequent memory with region and hemisphere [F(1,26) = 5.36, P < 0.01] and, crucially, group, subsequent memory and hemisphere [F(1, 26) = 9.00, P < 0.01]. Equally importantly (see Discussion), the interaction of group with hemisphere fell far short of significance (F <1). The group and group by region effects reflect a tendency for event‐related responses to be of greater magnitude in the older subjects, especially in the posterior two voxels. The three‐way interaction between region, subsequent memory and hemisphere arose from a tendency for subsequent memory effects to be less strongly lateralized in the more anterior of the three voxels (BA 47) than in the posterior two.

To elucidate the theoretically important interaction between group, subsequent memory and hemisphere, subsidiary ANOVAs were carried out on data from the two age groups separately. These analyses gave rise to an interaction of hemisphere with subsequent memory in the young group only [F(1, 13) = 32.00, P < 0.001; for old group, F <1]. Further analyses conducted on the data from each hemisphere separately demonstrated a significant interaction of age group with subsequent memory only in the right hemisphere [F(1, 26) = 5.30, P < 0.05]. In the left hemisphere, there was a main effect of subsequent memory only [F(1, 26) = 35.62, P < 0.001; for interaction with age, F <1]. This pattern of interactions suggests that, as illustrated in Fig. 3, prefrontal subsequent memory effects were less strongly lateralized in the older than the younger subjects. It is also important to note that subsequent memory effects were reliable and equivalent in magnitude for the two groups in the three left hemisphere voxels (as predicted). However, as evident in the figure and confirmed by the ANOVA described in the preceding paragraph, this pattern occurred in the context of the finding of generally greater responses in the older group to both remembered and forgotten items. To establish that the age‐related difference in the laterality of subsequent memory effects was independent of this global effect, hemispheric asymmetries in the responses elicited by each class of item were scaled, on a per‐subject basis, according to the formula [L–R)/√(L2 + R2)], thereby scaling the asymmetries according to the magnitude of the underlying responses. ANOVA gave rise to a significant effect for subsequent memory [F(1,26) = 7.27, P < 0.025], a significant interaction between age group and subsequent memory [F(1,26) = 8.41, P < 0.01], but no main effect of group (F <1). These effects—reflecting a generally bilateral pattern of responses in the older group, and a shift in the young from a bilateral pattern for forgotten items, to a left‐lateralized pattern for remembered words—confirm the findings from the ANOVAs conducted on the unscaled data.

Fig. 3 Lateralization of subsequent memory effects according to age, averaged across voxels in BA 47 (x = ±36, y = 36, z = –9), BA 45 (x = ±51, y = 27, z = 18) and BA 9/44 (x = ±51, y = 12, z = 21). (A) Subsequent memory effects, i.e. parameter estimates (see Fig. 2) for remembered minus forgotten items. (B) Parameter estimates for remembered and forgotten items plotted separately.

A separate analysis was conducted on parameter estimates from the medial temporal voxel of interest. There was no hint here of an interaction of age group with hemisphere and subsequent memory (F <1), and no interaction of group with hemisphere [F(1,26) = 4.10, 0.05 < P < 0.1], although there was a main effect of subsequent memory [F(1, 26) = 13.58, P < 0.005]. The absence of group differences in the laterality of these hippocampal subsequent memory effects was supported by an ANOVA on the data after rescaling as described above [interaction between group and subsequent memory: F(1,26) = 1.90, not significant]. A further analysis of laterality effects was carried out using data from older subjects at the short memory test delay and younger subjects at the long delay to assess these effects when subsequent recognition performance was equated; the pattern of findings was the same as in the overall analysis.

Discussion

This study examined the extent to which the neural correlates of successful memory encoding vary with age. In both age groups, subsequently remembered and subsequently forgotten words encoded during a semantic decision task elicited differential activity in a network of brain regions that encompassed prefrontal, medial temporal and posterior cortical regions. Age‐related differences in subsequent memory effects were observed in several regions. Consistent with previous findings (Cabeza, 2002), these differences included more asymmetric effects in lateral prefrontal cortex in the younger subjects. When recognition memory performance in the two groups was equated by comparing data obtained at different memory test delays, the foregoing pattern of findings was essentially unchanged. The findings are discussed below in the light of the behavioural data and their implications for accounts of the effects of ageing on episodic memory and its neural correlates considered.

Behavioural performance

Despite their high levels of general health and intelligence, older subjects exhibited a pattern of cognitive performance typical of that reported in behavioural studies of ageing (Light, 1991). In particular, performance was significantly below that of the younger subjects on all tests of long‐term memory.

Turning to the performance measures obtained in the experiment proper, no differences between the groups were found on the study task for either accuracy or RT. RTs were, however, longer for items subsequently remembered relative to those that were subsequently forgotten. This pattern of findings is similar to that reported in one previous subsequent memory study employing a ‘semantic’ study task (see Experiment 2 reported by Wagner et al., 1998a). The pattern is, however, by no means typical (Kirchhoff et al., 2000; Baker et al., 2001; Otten and Rugg, 2001a; Otten et al., 2001) and is thus unlikely to be the cause of the subsequent memory effects observed in the present or previous fMRI data. Moreover, the absence of an interaction between this RT effect and age means that it cannot account for the interactions between age and subsequent memory effects discussed below.

The older subjects performed less accurately on the recognition memory test than the younger subjects and responded more slowly, although neither of these effects interacted with delay. Importantly, recognition accuracy at the long delay in the young subjects and the short delay in the old subjects was equivalent. This permitted a direct test of interactions between age and subsequent memory without the potentially confounding influence of differences in memory performance. That said, it must be recognized that the validity of this approach to unconfounding age and difficulty effects rests on the assumption that the same cognitive operations were engaged during retrieval at each test delay. While we know of no good reason to question this assumption, there is no evidence that directly supports it. Indirect evidence comes from the fact that the results obtained for the age by subsequent memory interaction from the contrasts that matched for performance converged well with, and indeed were a subset of, those obtained from the data collapsed across delay.

fMRI results

Common effects across age groups

Regions showing subsequent memory effects common to both younger and older subject groups were broadly consistent with the findings of previous studies of subsequent memory effects for words in young adults (Wagner et al., 1998a; Kirchhoff et al., 2000; Baker et al., 2001; Otten and Rugg, 2001a; Otten et al., 2001). These regions included much of the left inferior prefrontal gyrus, as well as more dorsolateral prefrontal cortex. As in some previous studies (Brewer et al., 1998; Wagner et al., 1998a; Kirchhoff et al., 2000; Otten et al., 2001), effects were also observed in the medial temporal lobe, including the hippocampus.

As noted in the Introduction, previous studies of aging effects on memory encoding, all of which have employed blocked designs and task‐wise comparisons, have yielded mixed findings as to whether left prefrontal activity is reduced in older subjects. The present results go beyond those obtained previously, by demonstrating equivalent activity in this region in a within‐task, trial‐based measure of encoding‐related neural activity. The results are consistent with those of Logan and colleagues (Logan et al., 2002; but see Stebbins et al., 2002), who argued that reduced recruitment of left prefrontal cortex during encoding is not found for incidental study tasks. An interesting question for future research is whether left prefrontal subsequent memory effects are attenuated in older subjects when an intentional rather than an incidental study task is employed.

Age differences in subsequent memory effects

There were clear age differences in the degree of lateralization of subsequent memory effects in lateral prefrontal cortex when assessed using a region of interest approach (Cabeza, 2002). Across three independently defined voxels of interest, prefrontal subsequent memory effects were markedly less left lateralized in the older subjects, consistent with previous findings (e.g. Backman et al., 1997; Cabeza et al., 1997; Grady et al., 1998; Madden et al., 1999; Logan et al., 2002; for review see Cabeza, 2002). There was no such effect in a voxel of interest in the anterior hippocampus, suggesting that age‐related differences in lateralization are not a general property of regions exhibiting robust subsequent memory effects (see also Reuter‐Lorenz et al., 2000). Since the difference in lateralization of subsequent memory effects in the two age groups was unaltered when their recognition performance was equated, the present findings indicate that this difference was not secondary to the older subjects remembering a smaller proportion of the studied items. Thus, the difference cannot be reduced to an item effect, reflecting variation in the activity elicited by more versus less memorable words. Nor was this age by hemisphere effect merely a confound of global differences between the groups in response magnitude; a qualitatively identical pattern of findings emerged from an analysis of data rescaled to remove such differences.

While consistent in general terms with findings of reduced prefrontal lateralization in older subjects (Cabeza, 2002), the present findings seemingly differ from those reported previously in one respect: whereas greater lateralization in younger than older subjects was observed in the activity elicited by remembered items, this effect was absent for forgotten items (see Fig. 3). Thus, whereas earlier studies demonstrated reduced lateralization in older subjects in task‐related activity unconditionalized on memory performance (e.g. Logan et al., 2002), here this effect was found for successfully encoded items only. There are at least three possible reasons for this apparent discrepancy. First, the effects observed in previous studies may have been carried by only a subset of the experimental items—for example, those subjected to ‘successful’ encoding. Secondly, the discrepancy may point to a difference between blocked and event‐related designs, in that while the former designs confound ‘item‐’ and ‘state‐related’ activity (e.g. Donaldson et al., 2001), the latter focus exclusively on item‐related effects. Finally, the present findings may be a consequence of the negligible responses elicited by forgotten items in the younger subjects (see Fig. 3), giving rise to ‘floor’ effects and hence providing no opportunity for the responses to exhibit hemispheric asymmetry.

What might be the functional significance of age‐associated reductions in lateralization of task‐ or performance‐related neural activity? As discussed by Cabeza (2002) and Logan et al., (2002), these effects could be in some sense ‘compensatory’ and reflect age‐related changes in the neuroanatomical organization of cognitive processes, or differences in the cognitive operations engaged to perform a particular task. Alternatively, the effects may reflect a failure in older subjects to selectively recruit specialized cortical regions in response to task demands (Logan et al., 2002), perhaps as a result of impairment in transcallosal inhibitory mechanisms. Such ‘dedifferentiation’ would be consistent with the idea that age‐related changes in functional specialization represent the reversal of the developmental trend towards increased regional specificity (see Cabeza, 2002). The present findings do not distinguish between these alternatives. It is perhaps worth noting in passing that we were unable to detect any relationship between the magnitude of either left or right prefrontal subsequent memory effects and measures of memory performance in either age group.

The analysis of age by subsequent memory interactions across the whole brain revealed several additional age‐related differences. First, so‐called ‘subsequent forgetting effects’ (Otten and Rugg, 2001b) were observed in anterior prefrontal cortex in the younger subjects only. Subsequent forgetting effects in both prefrontal cortex and elsewhere have been described both by Otten and Rugg (2001b), and Wagner and Davachi (2001); in the latter study, effects were reported in a left anterior prefrontal region close to the more superior of the effects identified in the present study. Otten and Rugg (2001b) suggested that subsequent forgetting effects might be a consequence of diverting cognitive resources to processing unhelpful to memory encoding. The implication of this suggestion is that, in the present study, older subjects had more resources at their disposal or were able more effectively to focus resources on those components of the study task that benefited subsequent memory. These possibilities seem highly improbable, however, given the generally held view that both cognitive resources, and the control processes through which they are allocated, show an age‐related decline (Light, 1991). Alternative accounts are likely to depend upon a better understanding of the functional significance of subsequent forgetting effects than exists at present.

A second age‐related difference was the finding of a subsequent memory effect in left anterior temporal cortex for younger subjects only. This region has been implicated in complex semantic processing in a number of neuroimaging studies of language and its degeneration is linked to word comprehension deficits in semantic dementia (Mummery et al., 1998; for review see Price, 2000). It has been suggested that activity in this area is greater when semantic processing is more ‘specific’, for example when stimuli are grammatical sentences compared with scrambled or unrelated words (Mazoyer et al., 1993; Bottini et al., 1994; Price 2000), or stories compared with unrelated sentences (Fletcher et al., 1995). The better verbal memory typically exhibited by young than older adults may therefore be related partly to their capacity to engage in more differentiated or elaborated semantic processing. This is consistent with behavioural evidence that greater elaboration at encoding leads to better subsequent memory performance (Craik and Tulving, 1975), as well as the finding that spontaneous semantic elaboration is reduced in older adults (e.g. Craik and Simon, 1980; Hashtroudi et al., 1989). As noted earlier, a parallel explanation has been offered for the finding of a reduction in left inferior prefrontal activation at encoding in older adults in blocked design neuroimaging studies. This latter result appears, however, to be associated predominantly with the use of intentional memory study tasks, in which elaborative processing must be engaged spontaneously (see Introduction). In this respect, the parallel between the two regions breaks down, since the interaction of age and subsequent memory observed here in temporal cortex was obtained in the context of an incidental study task. However, the present findings are consistent with the suggestion that, even when incidental tasks are employed, older adults encode information in a more general fashion (e.g. Cohen, 2000; see also Stebbins et al., 2002).

There were two occipital regions in the present study where subsequent memory effects were found in older subjects only, i.e. in the lingual gyrus and medial occipital pole. In earlier studies examining age effects on a variety of non‐memory tasks (Grady et al., 1994, 1995; Madden et al., 1996, 1997), older adults manifested reduced occipital activity compared with younger adults. Although the present finding appears at first sight to be discrepant with this pattern, this discrepancy could merely reflect the difference between blocked and item‐based comparisons. For example, older adults may allocate visual attention less effectively than younger adults overall, but when they do engage such resources effectively, this might have a greater impact on subsequent memory. Another, and perhaps more interesting possibility, is that posterior regions may be recruited to a greater extent by older subjects in remembering words because they rely more than younger subjects on ‘surface encoding’ strategies, involving, for example, visual or phonological processing. This suggestion is consistent with the findings of Otten and Rugg (2001a) (see also Davachi et al., 2001), who described subsequent memory effects in bilateral parietal and left occipital cortex in a non‐semantic (syllable count) encoding task.

Conclusions

In summary, this study demonstrated that a number of previously reported neural correlates of successful memory encoding are unaffected by age. In particular, there were no significant differences in the activity associated with subsequently remembered compared with forgotten items in left inferolateral prefrontal areas, or in the hippocampus and adjacent medial temporal cortex. These findings are consistent with previous claims that when sufficiently elaborate semantic processing is required by the study task, older adults are able to recruit many brain regions in support of memory encoding to the same extent as younger subjects (Logan et al., 2002). However, the finding of greater subsequent memory effects in the young group in an anterior temporal region suggests that, even under these circumstances, less differentiated semantic processing of study items may contribute to age‐related memory impairment. Older subjects also showed subsequent memory effects in posterior regions, possibly indicative of a greater reliance on the encoding of non‐semantic features. Finally, consistent with previous observations (Cabeza, 2002), the engagement of prefrontal regions in support of successful encoding was more bilateral in the older than in the younger subjects. This result was unaltered when subsequent memory performance was equated between the groups, suggesting that it does not reflect the influence of age‐associated differences in performance.

Acknowledgements

We gratefully acknowledge the advice and assistance of R. Henson, L. Otten and the radiography staff of the Functional Imaging Laboratory, and helpful suggestions from an anonymous referee. The authors are supported by the Wellcome Trust. This research was supported by the Wellcome Trust and an MRC Co‐operative Award to M.D.R.

References

View Abstract