Measurement of the spatial distribution of Tau pathology is critical for early diagnosis and disease monitoring. We sought to investigate a novel Tau PET ligand, 18F-PI2620, in aging and throughout the spectrum of Alzheimer's disease (AD). Seventeen participants within known Amyloid status (via CSF or Amyloid PET) underwent Tau PET scanning with 18F-PI2620 on a GE PET-MRI scanner: ten older clinically normal (CN) individuals (five Amyloid- CN, mean age=72.6±4.0; and five Amyloid+ CN, mean age=72.2±6.8), six clinically impaired patients on the AD trajectory (mean age=65.0±8.2; three Amyloid+ Mild Cognitive Impairment and three Amyloid+ AD dementia), as well as one Amyloid- patient with Dementia with Lewy Bodies (DLB). Standardized uptake value ratios were computed 60-90 minutes post-injection and normalized to the inferior cerebellum. We examined target regions known to show high Tau uptake in AD (entorhinal cortex, hippocampus, amygdala, inferior temporal cortex, precuneus, and lateral parietal cortex). Group differences (Amyloid- CN vs. Amyloid+ CN vs. Amyloid+ Impaired) in regional Tau were assessed with Wilcoxon signed-rank tests, whereas associations between continuous CSF measures (Aβ42 and pTau) with regional Tau within the CN group were assessed with Spearman's Rank correlation coefficients. Compared to Amyloid- CN, Amyloid+ CN showed greater PI2620 uptake in entorhinal cortex, hippocampus, and amygdala (p-values<0.032). The Amyloid+ Impaired group showed elevated Tau in all regions compared to Amyloid- CN (p-values<0.008), as well as elevated Tau in inferior temporal cortex (p= 0.016), precuneus (p<0.001), and lateral parietal cortex (p<0.001) compared to the Amyloid+ CN group (Figure 1). Within the CN group, continuous levels of CSF Aβ42 were negatively associated with elevated Tau PET in entorhinal cortext (p=0.026), hippocampus (p=0.004), and amygdala (p=0.007). CSF pTau was not related to any regional Tau PET value (Figure 2). The Amyloid- DLB case did not show evidence of uptake in any Tau PET region.
Primary pathology of Alzheimer disease (AD) includes both β-amyloid and neurofibrillary tangles (NFT) of misfolded tau. Diagnosing tau pathology in vivo and correlating with clinical status is crucial in diagnosis and development of anti-tau therapy. The goal of this study was to 1) create a clinical protocol for evaluating a novel Tau PET ligand, 18F-PI-2620, 2) evaluate concordance between radiologists, and 3) compare to clinical status and quantified SUVR. Per Braak staging, NFTs start in the entorhinal cortex (Braak 1–2), progress to the hippocampus (Braak 3–4) and lastly impact neocortex (Braak 5–6). Criteria for tau positivity was based on modified Braak staging, where Braak stages 1&2, 3&4, and 5&6 were collapsed into stage A, B, and C, respectively, with no uptake staged as 0. Positivity was determined based on visual uptake greater than off-target regions. Two physicians were blinded to patient status, and independently evaluated fused PET/MRI for 16 healthy older controls and 12 patients with MCI or dementia. Using non-parametric Wilcoxin signed rank tests, visual ratings were then compared to SUVR from medial and lateral temporal ROIs. Of the 16 healthy controls, 12, 1, 2, and 1 participants were classified as 0, A, B, and C, respectively. All 3 MCI patients were classified C. One AD patient was classified as A, while 4 others were C. Of the other tauopathies, one patient with semantic dementia was classified as stage C. There was 89.2% agreement rate between radiologists (Cohen's kappa coefficient of 0.78, standard error=0.12). Significantly elevated SUVRs were present in individuals with some binding (stage A-C) compared to individuals with no binding (stage 0). The greatest disagreement was for intermediate levels corresponding to Braak 3-4. Off target uptake included the substantia nigra, venous and nasal sinuses, and choroid plexuses.
Abstract Background Prior work has demonstrated that elevated CSF biomarkers in older adults are associated with diminished performance across several hippocampally‐dependent memory tasks. It is unknown whether early structural changes in medial temporal lobe subfields (1) may be impacted by elevated p‐tau 181 , and (2) mediate the effect of p‐tau 181 on memory performance in older adults. Method The current study draws on a cohort of 147 cognitively unimpaired older adults (Stanford Aging and Memory Study) that completed both 3T MRI and lumbar punctures. To examine memory performance, participants completed two hippocampally dependent tasks: a paired associate cued recall task and a mnemonic discrimination task (MST) that examined discrimination between previously studied “old” objects, novel “foil” objects and perceptually similar “lure” objects. Lure trials were binned in five difficulty levels, ranging from low to high similarity compared to previously presented stimuli. Hippocampal subregions (CA1, CA3/DG, subiculum) were manually delineated on T2 * weighted images following Carr et al., (2017). CSF was processed with Lumipulse to measure Ab 42 :Ab 40 ratio and p‐tau 181. Multiple regression was used to assess the associative memory d’ from the paired associate task. Given the repeated measures design of the MST task, linear mixed models were conducted to understand the association between hippocampal subfields and p‐tau 181 on performance as a function of similarity strength (Trelle et al., 2021). Age, sex, and education were controlled in all models. Result P‐tau 181 was correlated with reduced left CA1 volume (partial r 2 =0.043, p = 0.01), but not with volume in other hippocampal subfields. Increased CA1 volume was positively associated with better memory performance, including higher associative memory recall and old‐lure discrimination. After controlling for p‐tau 181, CA1 volume was no longer associated with performance on the associative memory task but remained significantly correlated with old/lure discrimination. Notably, increased CA1 volume improved overall discrimination for old/lure trials, independent of the difficulty level, while the association between p‐tau 181 and performance was strongest at the lower similarity level. Conclusion Reduced CA1 volume and p‐tau explain unique variance in memory performance in older adults. This combination of biofluid biomarkers and high‐resolution MRI has the potential to understand mechanisms underlying age‐related memory decline.
Article Figures and data Abstract Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Age-related episodic memory decline is characterized by striking heterogeneity across individuals. Hippocampal pattern completion is a fundamental process supporting episodic memory. Yet, the degree to which this mechanism is impaired with age, and contributes to variability in episodic memory, remains unclear. We combine univariate and multivariate analyses of fMRI data from a large cohort of cognitively normal older adults (N=100) to measure hippocampal activity and cortical reinstatement during retrieval of trial-unique associations. Trial-wise analyses revealed that (a) hippocampal activity scaled with reinstatement strength, (b) cortical reinstatement partially mediated the relationship between hippocampal activity and associative retrieval, (c) older age weakened cortical reinstatement and its relationship to memory behaviour. Moreover, individual differences in the strength of hippocampal activity and cortical reinstatement explained unique variance in performance across multiple assays of episodic memory. These results indicate that fMRI indices of hippocampal pattern completion explain within- and across-individual memory variability in older adults. Introduction Episodic memory – in particular the ability to form and retrieve associations between multiple event elements that comprise past experiences – declines with age (Spencer and Raz, 1995; Rönnlund et al., 2005; Old and Naveh-Benjamin, 2008). Retrieval of an episodic memory relies critically on hippocampal-dependent pattern completion, which entails reactivation of a stored memory trace by the hippocampus in response to a partial cue, leading to replay of cortical activity patterns that were present at the time of memory encoding (Marr, 1971; McClelland et al., 1995; Tanaka et al., 2014; Staresina et al., 2019). Given observed links between in vivo measures of pattern completion and episodic remembering (Nakazawa et al., 2002; Gelbard-Sagiv et al., 2008; Gordon et al., 2014), and evidence of altered hippocampal function with age (Lister and Barnes, 2009; Leal and Yassa, 2013), changes in hippocampal pattern completion may play an important role in explaining age-related impairments in episodic memory. While a leading hypothesis, the degree to which the integrity of pattern completion can explain (a) trial-to-trial differences in episodic remembering within older adults and (b) differences in memory performance between older individuals remain underspecified. Functional MRI (fMRI) studies in younger adults suggest that hippocampal pattern completion is associated with at least two key neural markers: (a) an increase in hippocampal univariate activity (Eldridge et al., 2000; Dobbins et al., 2003; Yonelinas et al., 2005) and (b) cortical reinstatement of content-specific activity patterns present during encoding (Nyberg et al., 2000; Wheeler et al., 2000; Kahn et al., 2004). Multivariate pattern analyses — machine learning classification (Norman et al., 2006) and pattern similarity (Kriegeskorte et al., 2008) — reveal evidence for cortical reinstatement of categorical event features (Polyn et al., 2005; Johnson and Rugg, 2007; Gordon et al., 2014) and event-specific details (Staresina et al., 2012; Ritchey et al., 2013; Kuhl and Chun, 2014) during successful recollection. Moreover, hippocampal and cortical metrics of pattern completion covary, such that trial-wise fluctuations in hippocampal univariate retrieval activity are related to the strength of cortical reinstatement (Staresina et al., 2012; Ritchey et al., 2013; Gordon et al., 2014), and both hippocampal activity and reinstatement strength are related to associative retrieval performance (Gordon et al., 2014; Gagnon et al., 2019). These findings support models (Marr, 1971; McClelland et al., 1995; Tanaka et al., 2014) positing that cortical reinstatement depends, in part, on hippocampal processes, and contributes to remembering. Initial data bearing on age-related changes in hippocampal pattern completion are mixed. Studies comparing hippocampal activity during episodic retrieval in older and younger adults have revealed age-related reductions in activity (Cabeza et al., 2004; Dennis et al., 2008) and age-invariant effects (Wang et al., 2016; Trelle et al., 2019). Similarly, while some have identified reduced category-level (McDonough et al., 2014; Abdulrahman et al., 2017) and event-level (St-Laurent et al., 2014; Folville et al., 2020) cortical reinstatement in older relative to younger adults, others observed age-invariant category-level reinstatement (Wang et al., 2016) or that age-related differences in reinstatement strength are eliminated after accounting for the strength of category representations during encoding (Johnson et al., 2015). Although extant studies have yielded important initial insights, the absence of trial-wise analyses relating hippocampal activity to cortical reinstatement, or relating each of these neural measures to memory behaviour, prevents clear conclusions regarding the degree to which hippocampal pattern completion processes are impacted with age. Aging may affect one or both of these neural measures, and/or may disrupt the predicted relationships between these neural variables and behaviour (e.g., Gordon et al., 2014). The first aim of the present study is to quantify trial-wise fluctuations in hippocampal activity and cortical reinstatement in older adults, and examine how these measures relate to one another, as well as how these measures relate to episodic remembering of trial-unique associative content. Critically, in addition to varying within individuals, the degree to which pattern completion processes are disrupted among older adults may vary across individuals. Indeed, age-related memory decline is characterized by striking heterogeneity, with some individuals performing as well as younger adults and others demonstrating marked impairment (de Chastelaine et al., 2016; Henson et al., 2016; see Nyberg et al., 2012 for review). Identifying the neural factors driving this variability is a clear emerging aim of cognitive aging research (Nyberg et al., 2012; Cabeza et al., 2018). However, due to modest sample sizes, extant studies typically lack sufficient power to examine individual differences in retrieval mechanisms among older adults (Dennis et al., 2008; McDonough et al., 2014; St-Laurent et al., 2014; Johnson et al., 2015; Wang et al., 2016; Abdulrahman et al., 2017; Trelle et al., 2019; Folville et al., 2020). Moreover, while recent work examining variability in hippocampal function has demonstrated relationships between hippocampal retrieval activity and associative memory performance in older adults (de Chastelaine et al., 2016; Carr et al., 2017), the direction of this relationship differed across studies; to date, the relationship between individual differences in cortical reinstatement and memory performance remains unexplored. As such, the second aim of the present study is to examine whether hippocampal and cortical indices of pattern completion vary with age, and to assess the degree to which these measures explain individual differences in episodic memory performance — both as a function of age and independent of age. To address these two aims, a large sample (N = 100) of cognitively normal older participants (60–82 years) from the Stanford Aging and Memory Study (SAMS; Table 1; Materials and methods) performed an associative memory task (Figure 1) concurrent with high-resolution fMRI. Participants intentionally studied trial-unique word-picture pairs (concrete nouns paired with famous faces and famous places), and then had their memory for the word-picture associations probed. During retrieval scans, participants viewed a studied or novel word on each trial and indicated whether they (a) recollected the associate paired with the word, responding 'face' or 'place' accordingly (providing an index of associative memory), (b) recognized the word as 'old' but were unable to recall the associate (providing an index of item memory — putatively reflecting familiarity, non-criterial recollection, or a mix of the two), or (c) thought the word was 'new'. Following scanning, participants were shown the studied words again and asked to recall the specific associate paired with each word, this time explicitly providing details of the specific image (providing an index of exemplar-specific recall). Figure 1 Download asset Open asset Experimental paradigm. Concurrent with fMRI, participants intentionally encoded word-picture pairs and completed an associative cued recall test. At test, they were presented with studied words intermixed with novel words, and instructed to recall the associate paired with each word, if old. Participants responded 'Face' or 'Place' if they could recollect the associated image; 'Old' if they recognized the word but could not recollect the associate; 'New' if they believed the word was novel. A post-scan cued recall test (not shown, visually identical to the 'Test Phase') further probed memory for the specific associate paired with each studied word (see Materials and methods). Table 1 Demographics and neuropsychological test performance. MeasureMean (SD)RangeGender61 F; 39 M --Age (yrs)67.96 (5.47)60–82Education (yrs)16.84 (1.94)12–20MMSE29.10 (.90)26–30CDR0 --Logical Memory Delayed Recall (/50)32.04 (6.16)18–44HVLT-R Delayed Recall (/12)10.49 (1.68)5–12BVMT-R Delayed Recall (/12)9.80 (2.16)5–12Old/New d'2.26 (0.68)0.86–4.78Associative d'1.64 (0.73)−0.27–3.92Exemplar-Specific Recall (proportion correct, post-scan)0.29 (0.19)0.00–0.84 BVMT-R = Brief Visuospatial Memory Test-Revised; CDR = Clinical Dementia Rating; HVLT-R = Hopkins Verbal Learning Test-Revised; MMSE = Mini Mental State Examination. See Supplementary file 1 for summary of full neuropsychological test battery scores, and Supplementary file 1 for a summary of retrieval reaction time data and trial counts by memory outcome. Table 1—source data 1 Demographic information and behavioural data presented in Table 1. https://cdn.elifesciences.org/articles/55335/elife-55335-table1-data1-v1.csv Download elife-55335-table1-data1-v1.csv To measure pattern completion during retrieval, we used univariate and multivariate analyses focused on a priori regions of interest (ROIs; Figure 2). To measure hippocampal function, our primary analyses examined univariate activity in the whole hippocampus bilaterally. In addition, we measured activity in three subfields within the body of the hippocampus — dentate gyrus/CA3 (DG/CA3), CA1, and subiculum (SUB) — given prior work suggesting that aging may differentially affect individual hippocampal subfields (Yassa et al., 2011; Carr et al., 2017; Reagh et al., 2018) and models predicting differential subfield involvement in pattern completion, including a key role for subfield CA3 (Nakazawa et al., 2002; Grande et al., 2019). To measure cortical reinstatement, we focused on two cortical regions — ventral temporal cortex (VTC) and angular gyrus (ANG) — which we predicted would support content-rich representations during memory retrieval based on prior evidence in healthy younger adults. In particular, while VTC has traditionally been associated with content coding during memory encoding and retrieval (Nyberg et al., 2000; Wheeler et al., 2000; Polyn et al., 2005; Johnson and Rugg, 2007; Staresina et al., 2012; Ritchey et al., 2013; Kuhl and Chun, 2014; Gordon et al., 2014; Gagnon et al., 2019), more recent studies have also demonstrated evidence for cortical reinstatement of both category and stimulus/event-specific features in ANG during episodic retrieval, and suggest that these representations may be differentially related to memory-guided behaviour (Kuhl et al., 2013; Kuhl and Chun, 2014; Favila et al., 2018; Lee et al., 2019). Category-level reinstatement (i.e., face/place) was quantified via pattern classification and event-specific reinstatement (e.g., Queen Elizabeth, Golden Gate Bridge) was quantified using encoding-retrieval pattern similarity. Figure 2 Download asset Open asset Regions of interest. (a) Sample MTL subfield demarcations. The whole hippocampus ROI reflects the summation of all subfields (delineated only in the hippocampal body, shown), as well as the hippocampal head and tail (not pictured). (b) Parahippocampal cortex combined with fusiform gyrus and inferior temporal cortex forms the ventral temporal cortex ROI. Ventral temporal cortex (blue) and angular gyrus (gold) masks projected on the fsaverage surface. Results Behavioural results We assessed performance on the associative cued recall task using three measures: 1) old/new d' — discrimination between studied and novel words during the in-scan memory test, irrespective of memory for the associate; 2) associative d' — correctly remembering the category of associated images encoded with studied words, relative to falsely indicating an associative category to novel words; and 3) post-scan exemplar-specific associative recall — proportion correct recall of the specific exemplars associated with studied words. Performance on all three measures declined with age (old/new d': β = −0.35, p < 0.001; associative d': β = −0.30, p < 0.005, Figure 3a; post-scan exemplar-specific recall: β = −0.34, p < 0.001, Figure 3b), but did not vary by sex (βs = −0.10,–0.33, −0.23; ps ≥0.10) or years of education (β = −0.03,–0.02, −0.07; ps >0.47). Associative d' was higher for word-face pairs than word-place pairs (t(99) = 5.37, p < 10−7). Critically, despite this decline in performance with age, we also observed considerable variability in performance across individuals in each measure (Figure 3 and Table 1). Figure 3 Download asset Open asset Associative memory behavioural results. (a) In-scanner associative d' and (b) post-scan exemplar-specific associative recall decline with age. (c) Associative d' is strongly correlated with post-scan exemplar-specific associative recall, controlling for the effect of age. Each data point represents a participant; plots show linear model predictions (black line) and 95% confidence intervals (shaded area). Figure 3—source data 1 Demographic information and behavioural data depicted in Figure 3a–c. https://cdn.elifesciences.org/articles/55335/elife-55335-fig3-data1-v1.csv Download elife-55335-fig3-data1-v1.csv Individual-differences and trial-wise analyses revealed that post-scan associative recall tracked in-scanner associative memory. First, individuals who demonstrated higher associative memory during scanning showed superior recall of the specific exemplars on the post-scan test (controlling for age; β = 0.62, p < 10−12; Figure 3c). Second, trial-wise analysis revealed that making an in-scan associative hit was a significant predictor of successful post-scan exemplar recall (χ2(1) = 159.68, p < 10−36). These findings suggest that post-scan exemplar-specific retrieval — while quantitatively lower due to the longer retention interval, change of context, and interference effects — is a good approximation of recall of the specific exemplar during scanning (relative to simply recalling more general category information). fMRI encoding classifier accuracy Following prior work (e.g., Kuhl et al., 2013; Kuhl and Chun, 2014; Favila et al., 2018; Lee et al., 2019), cortical reinstatement analyses focused on two a priori ROIs: VTC and ANG. To confirm that activity patterns during word-face and word-place encoding trials were discriminable for each participant in each ROI, we trained and tested a classifier on the encoding data using leave-one-run-out-n-fold cross validation. On average, encoding classifier accuracy was well above chance (50%) using patterns in VTC (M = 98.4%, p < 0.001) and ANG (90.0%, p < 0.001), with classifier accuracy significantly greater in VTC than ANG (t(99) = 12.86, p < 10−16). Classification was above chance in all 100 participants (minimum accuracy of 82.5% (p < 0.001) in VTC and 68.0% (p < 0.005) in ANG), and did not vary significantly as a function of age (VTC: β = −0.13, p = 0.133; ANG: β = −0.06, p = 0.544). To account for variance in encoding classifier strength (quantified using log odds of the classifier's probability estimate) on estimates of category-level reinstatement strength during memory retrieval (trial-wise: VTC: χ2(1) = 13.96, p < 0.001; ANG: χ2(1) = 30.16, p < 10−8; individual differences: VTC: β = 0.45, p < 10−5; ANG: β = 0.62, p < 10−11; see Figure 5—figure supplement 3), we controlled for encoding classifier strength in all subsequent models in which category-level reinstatement strength was related to behavioural variables (memory accuracy, RT), as well as in models in which reinstatement strength was the dependent variable (see Materials and methods – Statistical Analysis and Supplementary file 1 for details). Memory behaviour scales with trial-wise category-level reinstatement We quantified reinstatement of relevant face or scene features (i.e., category-level reinstatement) in VTC and ANG using subject-specific classifiers trained on all encoding phase runs for an individual (training set was balanced for category), and tested for cortical reinstatement in the independent retrieval phase data; significance was assessed using permutation testing (see Materials and methods – MVPA for further details). Classifier accuracy (Figure 4a) was above chance (50%) during associative hits in VTC (M = 68.3%, p < 0.005) and ANG (M = 72.3%, p < 0.001), but did not exceed chance when associative retrieval failed, including on associative miss trials (VTC: 49.8%, p = 0.57; ANG: 50.4%, p = 0.49), item hit trials (VTC: 53.5%, p = 0.29; ANG: 53.3%, p = 0.31), and item miss trials (VTC: 47.1%, p = 0.68; ANG: 51.6%, p = 0.41; see Materials and methods for trial type definitions). Classifier accuracy during associative hits was greater in ANG relative to VTC (t(99) = 3.96, p < 0.001). In VTC, classifier accuracy during associative hits was stronger on place trials (M = 71.5%) relative to face trials (M = 65.1%, t(99) = 5.25, p < 10−7), whereas in ANG the strength of reinstatement did not significantly vary by stimulus category (place: M = 73.3%; face: M = 71.3%, t(99) = 1.69, p = 0.094). To control for possible effects of stimulus category on the results, category is included as a regressor in all linear and logistic mixed effects models, and interactions between category and primary variables of interest are examined and reported in Supplementary file 1). Analyses of the time course of cortical reinstatement during associative hits revealed significant category-level reinstatement effects emerging ~4–6 s post-stimulus onset (Figure 4—figure supplement 1). Analogous category-level reinstatement effects were observed using a pattern similarity approach (i.e., encoding-retrieval similarity (ERS); see Figure 4—figure supplement 2). Figure 4 with 6 supplements see all Download asset Open asset Cortical and hippocampal metrics of pattern completion during retrieval. (a) Classifier accuracy is above chance in VTC and ANG during successful, but not unsuccessful, associative retrieval. (b) Trial-wise category-level reinstatement strength (logits) in VTC and ANG is related to an increased probability of an associative hit and (c) faster decision RT on associative hit trials. (d) Event-level reinstatement (within-event ERS > within-category ERS) is observed during associative hits in VTC and ANG. (e) Trial-wise event-level reinstatement (within-event ERS) significantly varies with the probability of an associative hit and (f) exemplar-specific hit. (g) Hippocampal activity shows a graded response across retrieval conditions. (h) Trial-wise hippocampal activity is related to an increased probability of an associative hit and (i) greater category-level reinstatement strength (logits) in VTC and ANG. For visualization, data for each participant are binned into quintiles based on category-level reinstatement strength (b,c), event-level reinstatement strength (e,f) and hippocampal activity (h,i). Statistics were conducted on trial-wise data, z-scored within participant. Error bars represent standard error of the mean. VTC = ventral temporal cortex; ANG = angular gyrus; RT = reaction time; ERS = Encoding Retrieval Similarity. Figure 4—source data 1 Classifier accuracy in VTC and ANG by trial type, depicted in Figure 4a. https://cdn.elifesciences.org/articles/55335/elife-55335-fig4-data1-v1.csv Download elife-55335-fig4-data1-v1.csv Figure 4—source data 2 Trial-wise cortical reinstatement (logits), hippocampal activity, and behavioural data used to generate Figure 4b–c,h–i and Figure 4—figure supplements 3, 5 and 6. https://cdn.elifesciences.org/articles/55335/elife-55335-fig4-data2-v1.csv Download elife-55335-fig4-data2-v1.csv Figure 4—source data 3 Encoding-retrieval similarity in VTC and ANG by trial type, depicted in Figure 4d and Figure 4—figure supplement 2a. https://cdn.elifesciences.org/articles/55335/elife-55335-fig4-data3-v1.csv Download elife-55335-fig4-data3-v1.csv Figure 4—source data 4 Trial-wise encoding-retrieval similarity, hippocampal activity, and behavioural data used to generate Figure 4e–f and Figure 4—figure supplements 2b–c and 4. https://cdn.elifesciences.org/articles/55335/elife-55335-fig4-data4-v1.csv Download elife-55335-fig4-data4-v1.csv Figure 4—source data 5 Hippocampal activity by trial type, depicted in Figure 4g. https://cdn.elifesciences.org/articles/55335/elife-55335-fig4-data5-v1.csv Download elife-55335-fig4-data5-v1.csv Evidence for reinstatement during successful, but not unsuccessful, associative retrieval is consistent with theories that posit that reinstatement of event features (here, face or scene features) supports accurate memory-based decisions (here, associate category judgments). More directly supporting this hypothesis, generalized logistic and linear mixed effects models (see Supplementary file 1 for full list of model parameters) revealed that greater trial-wise category-level cortical reinstatement in VTC and ANG — quantified using log odds of the classifier's probability estimate — was related to (a) an increased probability of an associative hit (VTC: χ2(1) = 102.18, p < 10−24; ANG: χ2(1) = 133.25, p < 10−31; Figure 4b), (b) an increased probability of post-scan exemplar-specific recall (VTC: χ2(1) = 62.85, p < 10−15; ANG: χ2(1) = 89.02, p < 10−21; Figure 4—figure supplement 3), and (c) faster retrieval decision RTs on associative hit trials (VTC: χ2(1) = 30.08, p < 10−8; ANG: χ2(1) = 21.73, p < 10−6; Figure 4c). We also found that age moderated the relationship between category-level reinstatement strength in VTC and behaviour, such that older individuals exhibited a weaker relationship between reinstatement strength in VTC and (a) associative retrieval success (χ2(1) = 7.12, p < 0.01) and (b) retrieval decision RT on associative hit trials (χ2(1) = 3.91, p < 0.05). This interaction was marginally significant in ANG with respect to associative retrieval (χ2(1) = 3.57, p = 0.059), but not decision RT (χ2(1) = 0.16, p = 0.685). Together, these data provide novel evidence that the strength of category-level reinstatement in VTC and ANG is linked to memory behaviour in cognitively normal older adults (see Figure 4—figure supplement 2 for analogous ERS findings), and also suggest that older age negatively impacts the translation of cortical evidence to memory behaviour. Memory behaviour scales with trial-wise event-level reinstatement We next used encoding-retrieval similarity (ERS) to quantify trial-unique, event-specific reinstatement of encoding patterns, comparing the similarity of an event's encoding and retrieval patterns (within-event ERS) to similarity of encoding patterns from other events from the same category (within-category ERS). Evidence for event-level reinstatement was present in both VTC (t(99) = 2.26, p < 0.05) and ANG (t(99) = 3.54, p < 0.001) during associative hits (Figure 4d). Moreover, the strength of trial-wise event-level reinstatement — controlling for within-category ERS (see Supplementary file 1 for full list of model parameters)— was related to (a) an increased probability of an associative hit (VTC: χ2(1) = 1.78, p = 0.183; ANG: χ2(1) = 7.50, p = 0.006; Figure 4e) and (b) an increased probability of post-scan exemplar-specific recall (VTC: χ2(1) = 5.35, p < 0.05; ANG: χ2(1) = 7.27, p = 0.006; Figure 4f), but not with decision RT on associative hit trials (VTC: p = 0.845; ANG: p = 0.231). These relationships were not significantly moderated by age (all p > 0.254). These results demonstrate a relationship between trial-unique, event-specific cortical reinstatement and associative retrieval in older adults. Behaviour and reinstatement scale with trial-wise hippocampal retrieval activity Successful associative retrieval, ostensibly driven by pattern completion, was accompanied by greater hippocampal activity (Figure 4g) relative to associative misses (t(75) = 4.90, p < 10−6), item only hits (t(59) = 3.87, p < 0.001), item misses (t(83) = 8.86, p < 10−13), and correct rejections (t(99) = 11.28, p < 10−16). Relative to item misses, hippocampal activity was greater during associative misses (t(68) = 4.0, p < 0.001) and item only hits (t(51) = 5.37, p < 10−6); activity did not differ between associative misses and item hits (t < 1) or between item misses and correct rejections (t < 1). Moreover, generalized logistic and linear mixed effects models revealed that greater trial-wise hippocampal activity was related to (a) an increased probability of an associative hit (χ2(1) = 63.45, p < 10−15; Figure 4h), (b) an increased probability of post-scan exemplar-specific recall (χ2(1) = 59.02, p < 10−14; Figure 4—figure supplement 3), but (c) not faster associative hit RTs (χ2(1) = 2.08, p = 0.149). These relationships were not moderated by age (associative hit: p = 0.616; exemplar-specific recall: p = 0.713). Thus, the probability of successful pattern-completion-dependent associative retrieval increased with hippocampal activity. This relationship was significant across hippocampal subfields, but greatest in DG/CA3 (see Figure 4—figure supplements 5–6 for subfield findings). Cortical reinstatement is thought to depend on hippocampal pattern completion triggered by retrieval cues (Marr, 1971; McClelland et al., 1995; Tanaka et al., 2014; Staresina et al., 2019). Consistent with this possibility, the magnitude of trial-wise hippocampal retrieval activity significantly varied with the strength of category-level cortical reinstatement across all retrieval attempts (VTC: χ2(1) = 43.36, p < 10−11; ANG: χ2(1) = 35.31, p < 10−9; Figure 4i) and when restricting analyses only to associative hit trials (VTC: χ2(1) = 5.77, p < 0.05; ANG: χ2(1) = 9.48, p < 0.005). Similarly, hippocampal activity significantly varied with within-event ERS (controlling for within-category ERS) in VTC (all trials: χ2(1) = 4.55, p < 0.05; associative hit only: χ2(1) = 3.73, p = 0.054; see Figure 4—figure supplement 4); this relationship did not reach significance in ANG (all trials: p = 0.328; associative hit only: p = 0.289). The relationship between hippocampal activity and reinstatement strength was not moderated by age (category-level reinstatement VTC: p = 0.777; ANG: p = 0.773; event-level reinstatement VTC: p = 0.493). Collectively, these results constitute novel evidence for a relationship between trial-wise hippocampal activity and cortical reinstatement in older adults. This relationship was also observed in select hippocampal subfields (see Figure 4—figure supplements 5–6). Cortical reinstatement partially mediates the effect of hippocampal activity on retrieval Having established a relationship between associative retrieval success and (a) hippocampal activity, (b) cortical reinstatement strength in VTC, and (c) ANG, we next sought to determine whether each of these putative indices of pattern completion explain common or unique variance in associative retrieval success. Using nested comparison of logistic mixed effects models, we found that compared to a model with image category and hippocampal activity, addition of VTC category-level reinstatement strength significantly improved model fit (χ2(1) = 103.48, p < 10−24). Addition of ANG category-level reinstatement to this model further improved model fit (χ2(1) = 115.42, p < 10−27), and all three variables remained significant predictors in the full model (hippocampus: b = 0.31, z = 8.36, p < 10−16; VTC: b = 0.32, z = 9.36, p < 10−16; ANG: b = 0.52, z = 14.36, p < 10−16). These results indicate that reinstatement strength and hippocampal activity, though related indices of pattern completion, nevertheless explain unique variance in the probability of a successful associative retrieval decision. Moreover, they indicate that measures of category-level reinstatement strength in different cortical regions are not redundant, and perhaps carry complementary information relevant for memory behaviour. Given our prediction that the present measures of cortical reinstatement are, at least in part, a read out of hippocampal pattern completion processes, we next sought to more directly test the hypothesis that cortical reinstatement mediates the relationship between hippocampal activity and associative retrieval success. We conducted a mediation analysis separately for each cortical ROI, in which the coefficient of the indirect path was computed as the product of the direct effects, a x b, and the significance of the indirect effect was calculat
Age-related episodic memory decline is characterized by striking heterogeneity across individuals. Hippocampal pattern completion is a fundamental process supporting episodic memory. Yet, the degree to which this mechanism is impaired with age, and contributes to variability in episodic memory, remains unclear. We combine univariate and multivariate analyses of fMRI data from a large cohort of cognitively normal older adults (N=100; 60-82 yrs) to measure hippocampal activity and cortical reinstatement during retrieval of trial-unique associations. Trial-wise analyses revealed that hippocampal activity predicted cortical reinstatement strength, and these two metrics of pattern completion independently predicted retrieval success. However, increased age weakened cortical reinstatement and its relationship to memory behaviour. Critically, individual differences in the strength of hippocampal activity and cortical reinstatement explained unique variance in performance across multiple assays of episodic memory. These results indicate that fMRI indices of hippocampal pattern completion explain within- and across-individual memory variability in older adults.
Age-related episodic memory decline is characterized by striking heterogeneity across individuals. Hippocampal pattern completion is a fundamental process supporting episodic memory. Yet, the degree to which this mechanism is impaired with age, and contributes to variability in episodic memory, remains unclear. We combine univariate and multivariate analyses of fMRI data from a large cohort of cognitively normal older adults (N=100) to measure hippocampal activity and cortical reinstatement during retrieval of trial-unique associations. Trial-wise analyses revealed that (a) hippocampal activity scaled with reinstatement strength, (b) cortical reinstatement partially mediated the relationship between hippocampal activity and associative retrieval, (c) older age weakened cortical reinstatement and its relationship to memory behaviour. Moreover, individual differences in the strength of hippocampal activity and cortical reinstatement explained unique variance in performance across multiple assays of episodic memory. These results indicate that fMRI indices of hippocampal pattern completion explain within- and across-individual memory variability in older adults.
Memory decline is a key feature of cognitive aging, even among putatively healthy individuals who do not meet clinical criteria for cognitive impairment. However, the magnitude of this decline varies considerably across individuals. Emerging evidence suggests that two hallmark AD pathologies, including the aberrant accumulation of the beta-amyloid (AB) and tau proteins, are present decades before clinical diagnosis of dementia. The Stanford Aging and Memory Study (SAMS) examines the contribution of abnormal AD biomarkers, together with changes in brain structure and function, to individual differences in episodic memory among healthy older adults. Cognitively normal older adults aged 60-88 years (CDR=0) provided molecular (CSF phospho-tau, AB42) biomarkers of Alzheimer's disease (AD) and underwent ultra-high resolution 7T structural MRI and whole-brain high-resolution functional MRI (fMRI). The structural metrics obtained here include CA1-SRLM thickness, ERC thickness, whole hippocampal volume, and DG/CA3 subfield volume. During high-resolution fMRI, participants engaged in encoding and retrieval phases of an associative memory task, yielding univariate measures of regional BOLD activity and multivariate measures of cortical reinstatement during memory retrieval. Behavioural measures of item memory and associative recollection werealso obtained. Functional measures of (a) univariate BOLD activity in the hippocampus and (b) hippocampal-mediated cortical reinstatement in ventral temporal and parietal cortex during memory retrieval explained significant variance in associative memory performance across participants. CSF AB42 exhibited a marginally significant positive relationship with associative memory and neuropsychological tests of delayed recall, and a significant positive relationship with CA1-SRLM thickness, DG/CA3 volume, and total hippocampal volume; by contrast, p-tau showed a significant negative relationship with reduced CA1-SRLM thickness. These initial results suggest that the presence of abnormal AD biomarkers exert early effects on both hippocampal microstructure and individual differences in memory among cognitively normal older adults. The results also reveal that fMRI assays of hippocampal-mediated retrieval processes partially explain individual differences in memory performance. Ongoing analyses will further examine the relationships between age-related changes in the functional and structural integrity of the hippocampus and MTL cortex, continuous levels of CSF AB42 and p-tau, and their unique or combined contributions to individual differences in episodic memory in cognitively normal older adults.
To determine whether memory tasks with demonstrated sensitivity to hippocampal function can detect variance related to preclinical Alzheimer disease (AD) biomarkers, we examined associations between performance in 3 memory tasks and CSF β-amyloid (Aβ)42/Aβ40 and phosopho-tau181 (p-tau181) in cognitively unimpaired older adults (CU).
Methods
CU enrolled in the Stanford Aging and Memory Study (n = 153; age 68.78 ± 5.81 years; 94 female) completed a lumbar puncture and memory assessments. CSF Aβ42, Aβ40, and p-tau181 were measured with the automated Lumipulse G system in a single-batch analysis. Episodic memory was assayed using a standardized delayed recall composite, paired associate (word–picture) cued recall, and a mnemonic discrimination task that involves discrimination between studied "target" objects, novel "foil" objects, and perceptually similar "lure" objects. Analyses examined cross-sectional relationships among memory performance, age, and CSF measures, controlling for sex and education.
Results
Age and lower Aβ42/Aβ40 were independently associated with elevated p-tau181. Age, Aβ42/Aβ40, and p-tau181 were each associated with (1) poorer associative memory and (2) diminished improvement in mnemonic discrimination performance across levels of decreased task difficulty (i.e., target–lure similarity). P-tau mediated the effect of Aβ42/Aβ40 on memory. Relationships between CSF proteins and delayed recall were similar but nonsignificant. CSF Aβ42 was not significantly associated with p-tau181 or memory.
Conclusions
Tests designed to tax hippocampal function are sensitive to subtle individual differences in memory among CU and correlate with early AD-associated biomarker changes in CSF. These tests may offer utility for identifying CU with preclinical AD pathology.
As the research focus shifts toward the earlier identification and treatment of Alzheimer's disease (AD), a more sophisticated approach for characterizing cognitive function is needed. Using cognitive processing models combined with hierarchical Bayesian methods (HBCP) is one approach for estimating underlying cognitive processes and variables that cannot be measured using traditional scorning methods. HBCP models were applied to a wordlist memory task in normal aging subjects to characterize changes in underlying processes. 681,428 normal subjects, aged 15-100 years old (Figure 1), were assessed with free recall and recognition memory tasks. The raw score data were the proportion of words correctly recalled overall, of first (primacy) and last (recency) list words recalled, and of words correctly recognized overall. Hits and false alarms were also tracked. One HBCP applied a primacy-recency model of serial position effects to the free recall data (Figure 2a). The other HBCP applied an equal-variance signal detection theory model of discriminability and response bias to the recognition data (Figure 2b). The cognitive processing parameters of each individual were drawn from an assumed Gaussian distribution for their age. For each age, the mean value and the 90% credible intervals were plotted, within which a new individual's parameter values would fall. Mean raw recall scores overall and for primacy (first word) decline after 70 (Figure 3) while the HBCP model of the primacy parameter more sharply declines after 70 years old than the raw primacy scores (Figure 4). Mean raw hit and overall recognition rates slightly decline, and false alarm rates slightly increase after 70 years old (Figure 5) while the HBCP model shows that discriminability gradually declines, and response bias gradually increases over the entire lifespan (Figure 6).
