Age-related memory deficits linked to circuit-specific disruptions in the hippocampus

Abstract

Converging data from rodents and humans have demonstrated an age-related decline in pattern separation abilities (the ability to discriminate among similar experiences). Several studies have proposed the dentate and CA3 subfields of the hippocampus as the potential locus of this change. Specifically, these studies identified rigidity in place cell remapping in similar environments in the CA3. We used high-resolution fMRI to examine activity profiles in the dentate gyrus and CA3 in young and older adults as stimulus similarity was incrementally varied. We report evidence for “representational rigidity” in older adults’ dentate/CA3 that is linked to behavioral discrimination deficits. Using ultrahigh-resolution diffusion imaging, we quantified both the integrity of the perforant path as well as dentate/CA3 dendritic changes and found that both were correlated with dentate/CA3 functional rigidity. These results highlight structural and functional alterations in the hippocampal network that predict age-related changes in memory function and present potential targets for intervention.

Long-term memory function is commonly known to deteriorate with increasing age. One of the sites that undergo the earliest changes is the hippocampus (1, 2), which has a well-known role in learning new facts and remembering events (3). Recently, electrophysiological recording studies in aged rodents have shed light on some of the possible neural mechanisms in the hippocampus that underlie this decline (1). These studies have demonstrated “rigidity” in aged CA3 place cell firing patterns in similar environments. In contrast to young CA3 place cells, which readily remap and shift their representations in these environments, aged CA3 place cells retain their original fields despite the changes in the environment. These data strongly suggest that aging is associated with a diminished capacity for pattern separation (learning new information by decorrelating similar inputs to avoid interference) and an increased propensity for pattern completion (retrieval of previously stored information from a partial cue), and further suggest that this shift could be the result of a functional imbalance in the hippocampal dentate gyrus (DG) and CA3 network.

The role of hippocampal subfields in these key processes has long been hypothesized in computational models (4–7). The models suggest that the DG granule cells are capable of performing especially strong pattern separation on the distributed representations arriving from layer II entorhinal neurons, projecting this signal onto the CA3 subfield of the hippocampus via the strong mossy fiber pathway. Empirical evidence for the involvement of the DG and CA3 in pattern separation has been demonstrated by using electrophysiological recordings (8–10), immediate-early genes (11), and high-resolution functional MRI in humans (12, 13). Ablation studies using DG-specific ibotenic acid lesions (14), as well as genetic NMDA receptor knockouts (15), have additionally shown that the DG is critical for, and the likely locus of pattern separation. Recent evidence also indicates that neurogenesis in the DG may be particularly important for pattern separation (16–19).

The shift from pattern separation to pattern completion with age is hypothesized to be one of the reasons why older adults have more difficulties learning new information and may only encode the “gist” and not as many event details as young adults. We (20) as well as others (21) have tested this hypothesis more explicitly in older adults by using an object discrimination task. Consistent with the predictions from the rodent work, both studies found that older adults tended to be biased more toward completion at the expense of separation. We also demonstrated that these behavioral impairments were correlated with hippocampal DG/CA3 network hyperactivity (20), consistent with the finding in the rodent that aged CA3 place cells exhibit generally elevated firing rates across novel and familiar environments (22).

In this investigation, we hypothesized that the pattern separation signals exhibited by the DG/CA3 would be diminished with age for highly similar, but not dissimilar inputs (our techniques do not allow us to isolate DG and CA3 activity). Consistent with our predictions, we found no differences in DG/CA3 fMRI activity between young and older adults when stimuli were made very different. However, as stimuli were made more similar, the DG/CA3 response was weakened in older adults (consistent with pattern completion) but remained high in young adults, indicating that older adults’ DG/CA3 had an attenuated separation response to the lures. We refer to this change as a representational rigidity, which is operationally defined as the requirement for increased dissimilarity before stimuli can be orthogonalized, thus showing greater resistance to change. Critically, the extent of this rigidity predicted performance deficits in a behavioral discrimination task similar to the one used in our previous work (20).

Further, to evaluate the structural correlates of the rigidity we observed in the DG/CA3 functional network, we used ultrahigh resolution microstructural diffusion tensor imaging (msDTI) techniques developed in our laboratory (23) to assess potential changes in diffusion properties within hippocampal subfield gray matter. We found a correlation between left DG/CA3 functional rigidity and the same region's fractional anisotropy (a measure of directional diffusion). Directional diffusion in gray matter is thought to be an index of dendritic integrity (23, 24). Thus, these results suggest that structural dendritic changes in this region may contribute to the functional impairments observed, consistent with the idea that the dentate and CA3 in particular are selectively vulnerable to the aging process, because no such relationship was found in other subfields.

Next, we tested another key prediction of the model proposed by Wilson et al. (1), namely that the degraded perforant path input to the DG and CA3 is linked to pattern separation deficits. The perforant path provides the primary afferent input to the dentate and CA3 from layer II entorhinal cortical neurons (25). Studies in the rodent have shown that this pathway is essential for normal hippocampal function (26) and documented its degradation with advancing age (23, 27–31). Here, we tested the hypothesis that the network imbalance we observed in the fMRI data may be the result of an age-related degradation in the perforant path. To accomplish this goal, we used msDTI to visualize and quantify the perforant path diffusion signal as an in vivo correlate of this pathway's integrity. We found that perforant path integrity was tightly correlated with the extent of left DG/CA3 rigidity. In addition, perforant path integrity itself was predictive of performance of older adults on a behavioral discrimination task. A potential functional consequence of a degraded perforant path is a decrement in the functional correlations between the entorhinal cortex (EC) and the DG/CA3. To evaluate this possibility, we performed a seed-style functional connectivity analysis at the subfield level. The functional coupling of the DG/CA3 with the entorhinal cortex was highly correlated with the rigidity in the CA3, suggesting that the level of rigidity in the DG/CA3 may be related to degradation in the signaling between the entorhinal cortex and the hippocampus.

These data support the notion that age-related changes in the perforant path and the DG/CA3 network bias this network against separation and contribute to the mnemonic deficits observed in older adults. These findings provide converging evidence in support of a well-characterized model of memory loss in aged rodents (1) and contribute to our understanding of the effect of aging on the hippocampal network.

Results

Twenty young adults and 20 older adults underwent a high-resolution (1.5-mm isotropic) BOLD fMRI scanning session while viewing pictures of everyday objects (Fig. 1) and making an indoor/outdoor judgment. In our previous work (12, 13), we identified voxels in the hippocampus that signaled a repetition (i.e., changed their response between a first presentation and a repetition) and then examined activity for similar lure items in these voxels. Lure activity similar to that of a first presentation is consistent with pattern separation. However, if activity for a lure is similar to that of a repetition, this activity is more consistent with pattern completion. To increase our power for assessing activity for lure items and avoid potential issues regarding circularity and double dipping (32), we used data from two previously collected samples to define repetition-sensitive regions of interest (ROIs) in the hippocampus and then applied these ROI masks to our new, independent sample to investigate the effect of varying the degree of stimulus similarity on activity for lures.

Fig. 1.

Task design. Pictures were shown one at a time and could either be novel, repeated, or similar lures (participants are not informed of the item status). Different trials have been color-coded with blue (novel), green (repeated), and red (lure) for the reader's benefit; however, participants only saw items on a white background. The only task instruction given was to press a button for indoor items and another button for outdoor items. The two lures shown (pineapple and piano) provide examples of the lure stimuli used.

Based on extensive behavioral investigations (20), lures were binned according to their mnemonic similarity, allowing us to conduct a parametric investigation of DG/CA3 input/output transfer functions. In our recent work (13), we found that in a population of young adults, activity in a repetition-sensitive portion of DG/CA3 increased markedly with even small amounts of change in the input (highly similar lures) and remained constantly elevated as lure similarity increased. We observed this constant activity consistent with strong separation again here in young adults in the left DG/CA3 (Fig. 2B). Older adults, however, had an attenuated left DG/CA3 separation response to the lures such that separation signals were only noted on the highly dissimilar lures (Fig. 2C). This representational rigidity was consistent with the theoretical predictions shown in Fig. 2A.

Fig. 2.

Representational rigidity in older adults DG/CA3. (A) Hypothetical predictions for young and aged DG/CA3 transfer functions. Older adults are expected to exhibit rigidity where more dissimilarity is required for separation to occur (thus a positive slope across lures is expected for older adults and a near-zero slope across lures is expected for young adults). Inset shows actual data from our previous work (13) showing no change in young adults between “high similarity” lures and “low similarity” lures, i.e., DG/CA3 responds equally high to both types of lures, consistent with the above predictions. (B) A sample ROI of repetition-sensitive voxels (novel minus repeat) isolated in the left DG/CA3. (C) A comparison of young and aged left DG/CA3 fMRI activity across lures showing a pattern remarkably consistent with the predictions shown in A. Percent change in signal is change from the novel foil baseline. The contrast of activity for novel versus repeat items based on an orthogonal dataset was used to select the voxels averaged here.

We conducted a 2 × 5 ANOVA with group (old vs. young) and condition (L1, L2, L3, L4, L5) as fixed factors and subject as a random factor nested within group. A number of studies (33, 34) have shown that increases or decreases in BOLD response between young and old individuals could easily arise as artifacts of basal state factors such as metabolic rate and neurovascular coupling. With this caveat in mind, testing a main effect of group would have been difficult to interpret (this difference was significant nevertheless (F_1,38 = 5.31, P < 0.05). Thus, we opted to look for a significant interaction, which we observed in the data (F_4,152 = 2.13, P < 0.05). Although the young adults showed no difference in left DG/CA3 activity across lure bins [consistent with our previous data (13) and with the idea that separation had occurred even with small amounts of change in the input], the older adults showed a clear positive slope across lure bins (Fig. 2C). This finding suggests that compared with young adults, the older adults’ DG/CA3 required more dissimilarity before patterns of activity shifted to reflect a transition from completion to separation.

One important question is whether the extent of this rigidity reliably predicted behavioral discrimination abilities. Because the incidental task used inside the scanner could not be used to assess behavior, we tested a number of older participants (n = 15) who underwent this fMRI task in a separate session outside the scanner in an explicit recognition memory task designed to tax pattern separation (with a different set of stimuli). In this task, participants saw novel, repeated and similar lure items and were asked to indicate whether the item was “old,” “similar,” or “new.” Our previous work (20) has shown that older adults have a diminished “separation bias” on this task compared with young adults (i.e., they were less likely to call lure items similar and more likely to call them old). We replicated this effect here (Fig. S1), and we also observed an inverse correlation between the behavioral separation bias in older adults and the slope of left DG/CA3 activity (Pearson's r = −0.53, P < 0.05; Fig. 3A). In addition, there was also a marginally significant correlation (Pearson's r = −0.46, P = 0.09; Fig. 3B) between another behavioral measure, delayed recall performance by participants on the Rey Auditory Verbal Learning Test (RAVLT) (35), and the slope of left DG/CA3 activity. Both of these correlations suggest that the degree of representational rigidity in DG/CA3 predicts mnemonic deficits.

Fig. 3.

Relationships between representational rigidity and behavior. (A) We found a negative relationship between the left DG/CA3 slope of activity in older adults and their separation bias on an explicit recognition task designed to tax their pattern separation abilities, suggesting that processing in this region plays a key role in discrimination and that its dysfunction with age may underlie mnemonic deficits. (B) We also found a marginally significant relationship between the left DG/CA3 slope of activity in older adults and their performance on a delayed recall task (verbal list learning), which is a general index of their hippocampal function.

In addition, we performed ultrahigh-resolution msDTI scans (23) on 15 of the 20 older participants who underwent the fMRI scan. We used these scans to quantify fractional anisotropy (FA) throughout hippocampal subfields after removing all white matter signals. These gray matter signals are thought to be a reflection of the integrity of pyramidal dendrites (23, 24). We found a significant correlation between FA in the left DG/CA3 and the slope of fMRI signal in the same region taken to indicate the extent of representational rigidity (Pearson's r = −0.55, P < 0.05; Fig. 4A). No age-related changes in FA or correlations with any other measure were observed in any of the other hippocampal subregions (CA1, Sub, EC).

Fig. 4.

Relationship between functional and structural measures of integrity. (A) We found a negative relationship between the left DG/CA3 slope of activity in older adults and fractional anisotropy in this region (an indication of dendritic integrity), providing evidence that intact synaptic function in this region is necessary for successful pattern separation. (B) We found a negative relationship between the left DG/CA3 slope of activity in older adults and their perforant path integrity, suggesting that degradation in this pathway may underlie the network's representational rigidity. (C) We also found a positive relationship between perforant path integrity in older adults and their separation bias on the explicit recognition task, providing further evidence that the perforant path is necessary for successful discrimination. (D) We found a negative relationship between the left DG/CA3 slope of activity in older adults and degree of functional coupling between the left entorhinal cortex and the left DG/CA3 suggesting that communication failure in this network is linked to the representational rigidity observed.

We then sought to determine whether perforant path integrity was related to the ability of the DG/CA3 to engage in pattern separation. Perforant path degradation with age is hypothesized to be a catalyst for the shift in the information processing balance in the hippocampal DG/CA3 network that biases this network away from pattern separation and toward pattern completion. To test this prediction, we examined the msDTI we collected. Using this method in the same participants, we have demonstrated our ability to visualize and quantify diffusion signals specific to the perforant path and showed that these signals were degraded in the course of aging and predicted delayed recall performance on the RAVLT (23). In this investigation, we found a significant negative correlation between perforant path integrity and the slope of left DG/CA3 fMRI activity (Pearson's r = −0.59, P < 0.05, Fig. 4B). Additionally we found a positive relationship between perforant path integrity and the separation bias of participants on the explicit memory task administered outside the scanner in a subset of participants (n = 11) who received both behavioral and DTI sessions (Pearson's r = 0.73, P < 0.05; Fig. 4C).

Finally, to assess the functional impact of structural perforant path degradation, we used functional connectivity analyses to investigate the correlation in hemodynamic activity between the EC and DG/CA3 regions. Here, we found that the degree of functional connectivity between these regions (thought to reflect the function of the perforant path) also predicted the extent of functional rigidity in the DG/CA3 in older adults (Pearson's r = −0.40, P < 0.05; Fig. 4D).

Discussion

Converging evidence from animals (1) and humans (20, 21, 36) demonstrates that, on a behavioral level, pattern separation abilities are reduced with aging. This deficit is hypothesized to underlie many of the memory problems reported with increasing age. Aging is associated with a decline in episodic memory formation (37), spatial memory and navigation (38), contextual source memory (39), and recollection (40, 41). These functions require intact pattern separation circuitry (7, 42). Here, we examined the potential neural locus for pattern separation deficits and uncovered both a functional change in the DG/CA3 network as well as a related structural change in the perforant path, both of which predicted pattern separation abilities.

Previous work in our laboratory isolated an elevated pattern of BOLD activity in the region during an explicit recognition task (20); however, we could not make inferences regarding the ability of this region to perform the separation computation per se. The explicit approach we previously used, although informative in terms of behavior, was quite limiting in terms of interpretation. Activity during many conditions could be contaminated by a “recall to reject” strategy, which would prevent us from making strong claims about the relationship between activity levels and pattern separation/completion (20, 43). However, by using the incidental encoding approach used here, and in our previous work in young healthy volunteers (12, 13), we were able to parametrically assess changes in activity in the DG/CA3 region as a function of mnemonic similarity of stimuli and avoid this pitfall.

We found that relative to the young, the aged DG/CA3 requires more dissimilarity (a greater change in the input) to shift its representation to reflect a novel learning experience. Instead of place cell firing patterns (which we could not assess with fMRI) we used the slope of lure bin activity in the DG/CA3 as our in vivo proxy to measure representational rigidity (resistance to change). In this scheme, a more positive slope suggests a requirement for greater dissimilarity to “remap,” thus showing greater resistance to change. This pattern was characteristic of the slopes identified in older adults, but not in the young adults whose activity was consistent with remapping even for very small changes in the input. Critically, the extent of this representational rigidity predicted the performance of older adults on an explicit recognition task that directly tested their pattern separation abilities. This task was used to demonstrate age-related impairments in pattern separation (20, 21). Further, there was evidence for a relationship between the rigidity in DG/CA3 fMRI activity and a simple standardized test of delayed recall performance, a memory measure that is known to be sensitive to hippocampal damage (44, 45). We have also shown that it is correlated with performance on discrimination tasks sensitive to pattern separation deficits (20). Overall, these brain-behavior correlations indicate that representational rigidity in the aged DG/CA3 underlies the behavioral deficit in discriminating among very similar experiences.

Next, we turned to the question of whether structural changes in perforant path integrity would contribute to the DG/CA3 network changes observed in aging. We used ultrahigh-resolution msDTI to examine the potential relationship between the integrity of the perforant path and DG/CA3 function in older adults. We found that the extent of DG/CA3 rigidity in older adults was inversely correlated with perforant path integrity. Furthermore, we also observed a direct correlation between perforant path integrity and performance on the explicit recognition task taxing the discrimination abilities of participants. We also used a functional assay to investigate the connectivity between the EC and DG/CA3 and found that the extent of this functional coupling also predicted the degree of DG/CA3 rigidity. Using the same msDTI techniques we used for perforant path quantification, we also found a negative correlation between fractional anisotropy in the DG/CA3 region and its functional rigidity. These results suggest that microstructural changes in the DG/CA3 region, possibly reflecting a reduction in dendritic integrity, may contribute to the functional deficits observed.

Although we hesitate to overinterpret these dendritic diffusion data, we should note that this finding is the first of its kind in humans to our knowledge. Thus, subfield-specific dendritic integrity is a critical future avenue of investigation. Assessing dendritic integrity could offer important clues as to the neural basis of changes in network function. Reductions in dendritic integrity specific to the DG/CA3 region may occur before or subsequent to the loss in afferent fibers observed with age and may be an important feature of the neurocognitive aging process. However, at this time, this relationship remains unclear. We should emphasize, however, that high-resolution DTI allows us to examine gray matter anisotropy in a way that traditional DTI methods do not. This finding extends the utility of DTI to investigating gray matter microstructure and not just white matter.

The results presented herein provide compelling evidence for an age-related reduction in pattern separation in human DG/CA3 and point to structural and functional deficits in the perforant path and the DG/CA3 as potential contributors to this shift. Our results are largely consistent with data showing that the perforant path is particularly vulnerable to age-related changes. For example, studies in rodents have shown that the perforant path input to the hippocampus is reduced in aged rats compared with young rats (29, 30). It should come as no surprise also that one of the primary targets of perforant path input, the dentate gyrus, also appears vulnerable to the aging process (46–50). Electrophysiological data in aged rats show reductions in field EPSPs recorded in the dentate (27, 51) as well as presynaptic fiber potential at the perforant path-dentate synapse (28, 31). Recent evidence in aged rodents also suggests that molecular and synaptic changes occur in the entorhinal cortex (EC) where the perforant path originates (52).

One interesting possibility is that the some of the impairments observed in the dentate gyrus could be the result of diminished neurogenesis with increasing age (53, 54). Although there is still a debate about the exact role of newborn versus immature neurons (18, 19), it is possible that the reduced capacity for neurogenesis with increasing age interferes with dentate gyrus function, although evidence for this possibility is not straightforward (e.g., ref. 55). Although studying neurogenesis directly in humans may not be possible, recent work by Small and colleagues (56) using neuroimaging of cerebral blood volume (CBV) has isolated an in vivo correlate of neurogenesis, which can be coupled with high-resolution fMRI methods in the future to study the link between DG neurogenesis and its pattern separation capabilities.

Taken together, the data presented here provide converging evidence from multiple imaging modalities that age-related changes, manifest in the hippocampus, establish a permissive environment for potentially pathological changes that affect memory function. In this work, we uncover several of these conditions, namely perforant path degradation, loss of functional connectivity between the entorhinal cortex and DG/CA3, and functional rigidity as well as decreased anisotropy in DG/CA3. These conditions may collectively contribute to the age-related shift in mnemonic function from encoding new memories to retrieval of preexisting memories.

In summary, we combined two advanced high-resolution imaging modalities to investigate structural and functional changes in the hippocampal system with age. These results provide unique empirical demonstration of reduced pattern separation-related activity in human DG/CA3 with age as well as evidence linking structural changes in the perforant path and the DG/CA3 to information processing in the same network. The observed changes may contribute to the mnemonic deficits observed with age by weakening the processing of novel information and strengthening the processing of previously stored information. These results bring us one step closer to a more complete understanding of age-related changes in hippocampal structure and function. Future studies should continue to make attempts to link across techniques and species to enhance our understanding of this complex system and how it changes in the course of aging and disease.

Methods

Participants.

We collected data from 20 young adults (8M:12F, mean age 21 y, SD 3) and 20 older adults (8M:12F, mean age 71 y, SD 4). Written informed consent was obtained from all participants. Participants were screened for any health conditions that may interact with their neurological status. Young adults were recruited from the University of California, Irvine community by using IRB-approved flyers. Older adults were referred directly through a large longitudinal study of healthy aging people at the University of California, Irvine. Exclusion criteria included any major medical conditions, (e.g., diabetes, heart disease), any neurologically active medication use, any history of mental or psychiatric disorder and any contraindications for MRI. All older adults scored ≥27 on the Mini-Mental State Examination (MMSE: mean score 28, SD 1). All participants also received a battery of neuropsychological tests including measures of IQ, memory, attention, and general cognitive function (Table 1). Details of the administration of the RAVLT are included in SI Methods.

Table 1.

Older adults demographic and neuropsychological variables

Measurement	Value (SD)
n	20
M:F	8:12
Age	71 (4)
Years of education	17 (2)
MMSE	28 (1)
RAVLT 5-trial total	51 (9)
RAVLT immediate recall	11 (2)
RAVLT delayed recall	10 (3)
Digit span	18 (5)
WAIS full-scale IQ	118 (7)
Trails A	28 (10)
Trails B	61 (15)
Verbal fluency	48 (11)
Category fluency	21 (5)
Letter-number sequencing	10 (3)

Scores are shown as mean (SD). MMSE, Mini-Mental State Examination; RAVLT, Rey Auditory Verbal Learning Test; WAIS, Wechsler Adult Intelligence Scale. All scores were within norms.

Behavioral Tasks.

To assess behavior, we used an out-of-scanner explicit three-alternative recognition task in which participants viewed novel, repeated, and similar stimuli (i.e., “lures”) and were asked to indicate whether items were old, similar, or new. Lures were used to test the ability of participants to successfully separate. Details of the task are described in SI Methods. Behavioral data are shown in Fig. S1. For the in-scanner version of the task, we used a different stimulus set of novel, repeated, and lure items and participants were only asked to indicate whether items were “indoor” or outdoor” objects and their memory for these objects was never explicitly tested as in ref. 12. In our previous work (12, 13), we used half as many lures as we used in this study. This limitation hindered our ability to detect changes across lures in more than two bins (i.e., high similarity vs. low similarity). However, in the current design, we were able to double the number of lure items by removing the repetitions, thereby increasing the number of items in each lure bin from 20 to 40 stimuli. This change was done to ensure that we had enough power to detect effects across different lure bins. This procedure was essential because we hypothesized that the DG/CA3 of older adults will shift to a pattern consistent with a first presentation (i.e., pattern separate) at greater levels of dissimilarity compared with the young adults. Thus, whereas young adults should readily shift DG/CA3 representation to reflect a new item as early as the Lure bin 1 (most similar), older adults may not do so until Lure bin 3 or 4 (less similar). Because we used an orthogonal mask from independent data sets to define the required repetition-sensitive voxels, including repeated items in this design was not necessary. Detailed parameters of the task are described in SI Methods.

High-Resolution fMRI Methods.

Functional MRI data were collected by using a 3-Tesla Philips scanner equipped with a SENSE head coil using both higher-order shims and SENSE imaging techniques. Functional EPI images were collected by using a high-speed echoplanar single-shot pulse sequence with a field of view of 230 × 230 mm, echo time of 25 ms, flip angle of 70°, SENSE factor of 2, a repetition time (TR) of 1,500 ms, a 4 TR initial skip, and a resolution of 1.5 mm, isotropic. In each run, 26 triple-oblique axial slices were acquired aligned to the principal axis of the hippocampus bilaterally. Additional structural scans were also collected (details in SI Methods).

Data analysis was performed by using Analysis of Functional NeuroImages (AFNI). Images were motion corrected and normalized to standard Talairach space to provide an initial rough alignment. They were then aligned by using a ROI-based method we developed in our laboratory that maximizes our sensitivity to local changes in the medial temporal lobes (20). Behavioral vectors based on trial type and condition were used to model the data by using a deconvolution approach based on multiple linear regression. The resultant fit coefficients (betas) estimated activity versus baseline (novel foils) for a given time point and trial type in a voxel. The sum of the fit coefficients over the expected hemodynamic response (3–12 s after trial onset) was taken as the model's estimate of the response to each trial type (relative to baseline). Beta estimates were generated for first presentations and each of the lure bins (L1, L2, L3, L4, L5; L1 being the most similar lures and L5 being the least similar). Sorting the lure trials into different similarity bins was conducted based on similarity ratings derived from a large normative study conducted in our laboratory (20). Mean beta coefficients were converted to relative signal maps (percent signal change) by using the mean of each run from the betas. We created an a priori mask (Fig. S2) for repetition-sensitive voxels based on data from two studies (12, 13) and applied this mask to the lure bin data. The mask included several clusters in the DG/CA3, CA1, and subiculum. Voxels within each of these clusters were collapsed for further analysis. Clusters within a specific subfield were also collapsed to increase signal-to-noise ratio. Collapsed subfields maps are shown in Fig. S3. Additional details on fMRI preprocessing and analysis as well as ROI mask creation are provided in SI Methods.

The functional connectivity analysis was based on a seed style correlation analysis. First, we removed all task-related activity and nuisance vector activity from our raw data, leaving a residual time series. We extracted a mean residual time series from each of our hand-drawn ROIs used for alignment. We then correlated the isolated mean residual time series with the residual activity of all voxels in the brain. The resulting correlation coefficients were Fisher's z transformed. We then assessed the degree of correlation with the seed time series for each region by using an anatomical ROI analysis. See SI Methods for a more detailed description of this analysis.

Ultrahigh-Resolution Microstructural DTI Methods.

Diffusion weighted scans were acquired on the same scanner by following the methods we described (23). Twelve 15-slice coronal single shot echo-planar imaging (EPI) scans were acquired with a 256 × 256 matrix, a field of view (FOV) of 170, voxel size of 0.664 mm × 0.664 mm in the plane of acquisition, a slice thickness of 3 mm (1 mm gap), a repetition time of 2,717 ms and an echo time (TE) of 67 ms, a flip angle of 90° and a SENSE reduction factor of 2.5. Diffusion weighting was applied with b = 1,200 s/mm² along 32 independent, noncollinear orientations. One additional image with no diffusion weighting (b = 0) was also acquired. In addition to the diffusion-weighted scans, we also collected an ultrahigh-resolution Fast Spin Echo scan centered on the medial temporal lobe with identical geometry and resolution (FOV = 230, TR/TE = 3,000/80, flip angle = 90°). These scans were used because they provide detailed anatomical information inside hippocampal subfields for each diffusion-weighted slice.

Diffusion weighted images were corrected for head motion and eddy current distortions by using FMRIB's Diffusion Toolbox—FDT v. 2.0 (57). Motion and distortion-corrected volumes were merged into a single volume, and tensor solving was accomplished by calculating the six elements at each pixel by using multivariate linear fitting, allowing us to construct fractional anisotropy (FA) maps as well as images for each of the three eigenvalues and eigenvectors. In each participant, we identified the location of the perforant path on the in-plane inverted T2-weighted scan (SI Methods) by using the boundary between gray matter and white matter in the entorhinal cortex as a guide. We calculated the amount of diffusion signal that is parallel to the canonical orientation of the perforant path by using a method we have described and validated in young and older adults (23). Area under the curve (Fig. S4) was calculated for each participant's perforant path diffusion profile and used as the key measure of perforant path integrity in this study. Additional details on these procedures as well as details on subfield-level quantification of FA are discussed in SI Methods.