In vivo and neuropathology data support locus coeruleus integrity as indicator of Alzheimer’s disease pathology and cognitive decline

Abstract

Several autopsy studies recognize the locus coeruleus (LC) as the initial site of hyperphosphorylated TAU aggregation, and as the number of LC neurons harboring TAU increases, TAU pathology emerges throughout the cortex. By conjointly using dedicated MRI measures of LC integrity and TAU and amyloid PET-imaging, we aimed to address the question whether in vivo LC measures relate to initial cortical patterns of Alzheimer’s disease (AD) fibrillar proteinopathies or cognitive dysfunction in 174 cognitively unimpaired and impaired older individuals with longitudinal cognitive measures. To guide our interpretations, we verified these associations in autopsy data from 1524 Religious Orders Study and Rush Memory and Aging Project and 2145 National Alzheimer’s Coordinating Center cases providing three different LC measures (pigmentation, tangle density and neuronal density), Braak staging, beta-amyloid and longitudinal cognitive measures. Lower LC integrity was associated with elevated TAU deposition in the entorhinal cortex among unimpaired individuals, consistent with postmortem correlations between LC tangle density and successive Braak staging. LC pigmentation ratings correlated with LC neuronal density but not with LC tangle density, and were particularly worse at advanced Braak stages. In the context of elevated beta-amyloid, lower LC integrity and greater cortical tangle density were associated with greater TAU burden beyond the medial temporal lobe and retrospective memory decline. These findings support neuropathologic data in which early LC TAU accumulation relates to disease progression, and identify LC integrity as a promising indicator of initial AD-related processes and subtle changes in cognitive trajectories of preclinical AD.

One Sentence Summary:

In vivo and autopsy evidence support locus coeruleus MRI measures as proxy of initial TAU deposition and amyloid-related memory decline

INTRODUCTION

Alzheimer’s disease (AD), the most common form of dementia, is neuropathologically characterized by the accumulation of beta-amyloid (Aβ) plaques and hyperphosphorylated TAU forming pretangle material and, when aggregating, neurofibrillary tangles (1, 2). Plaques and tangles progress in a predictable topographical pattern in the cortex described in the seminal Thal and Braak stages, respectively (3). However, several autopsy studies reported that the locus coeruleus (LC) accumulates hyperphosphorylated TAU in approximately 50% of the 30–40-year-old individuals, a time in life when no cortical pathology is detectable (2, 4, 5). The number of LC neurons harboring TAU increases as TAU pathology emerges throughout the cortex (5). Greater LC tangle burden correlates with steeper cognitive decline antemortem, separate from the number of neurons in the LC (6). While LC volume is inversely related to Braak stages, LC neurons are sturdy and appear to succumb only later in the disease process (7). Given its early predilection for TAU, integrity of the LC may represent a putative early marker of AD, potentially preceding the accumulation of cortical fibrillar pathology. Here, we examined whether in vivo LC MRI-measures are associated with the initial anatomic patterns of cortical AD pathology, and whether they relate to a pattern of cognitive dysfunction typical of preclinical AD.

Direct in vivo detection of TAU pathology in the LC is limited by its small size compared to the typical resolution and sensitivity of positron emission tomography (PET) scanners. In addition, current TAU radiotracers bind off-target to neuromelanin (8) which is abundant in the norepinephrine (NE) neurons of the LC (9). However, recent advances in high-resolution magnetic resonance imaging (MRI) of the LC permit its detection as a focal hyperintense signal. The biological correlate underlying this signal is likely related to processes contributing to pigmentation - and later depigmentation - of the neuromelanin cells in LC NE neurons (10, 11). The number of neuromelanin cells increase throughout life as they chelate environmental toxins. These cells also harbor TAU inclusions, and as the disease progresses, lose pigmentation and degenerate (12). This cumulative sequestration and buildup of neuromelanin-associated lipids, toxins and proteins may thus indirectly contribute to the MRI signal (13).

The LC signal intensity (expressed as relative intensity (intensity_r)) evaluated in imaging studies reflects a measure of LC cellular integrity (10). One might expect LC integrity to deteriorate with age, however, this has not been consistently observed: instead, reports suggest an inverted U-pattern (14, 15) or no relationship with age (16). Similar inconsistencies can be observed in immunohistochemistry, demonstrating an inverted U-shape of neuromelanin neuron numbers with age (17, 18), a linear decline with age (19) or even no age-related reduction in neuromelanin neuron numbers in older adults (20, 21). The common feature among these reports is the substantial amount of variability of LC integrity in later life. Given that the number of LC neurons harboring TAU correlates positively with increasing cortical TAU pathology, this variability may reflect varying amounts of LC TAU burden and concomitant cortical fibrillar Aβ and TAU pathologies. Thus, we hypothesize that older individuals exhibiting a downward age-relationship with LC integrity represent individuals with elevated cortical Aβ or TAU pathology and are at risk of AD-related cognitive decline.

To substantiate the hypothesis that in vivo LC MRI-measures convey AD risk-related information, we associated LC intensity_r with anatomic patterns of TAU and Aβ pathology in individuals at risk for preclinical AD as well as cognitively impaired individuals and examined whether LC intensity_r predicts AD-related cognitive decline. To that end, we combined dedicated LC-scans with Aβ and TAU PET imaging in the well-characterized participants of the longitudinal Harvard Aging Brain Study (HABS), who also had up to 8 years of retrospective cognitive assessments. In addition, we verified our associations in two supporting postmortem datasets - the Religious Orders Study and Rush Memory and Aging Project (ROSMAP)(22) and the National Alzheimer’s Coordinating Center (NACC)(23) – each providing study-specific neuropathologic or cognitive measures, and ratings of LC hypopigmentation, a measure comparable to our in vivo MRI-measure. In addition, we also included a subset of the Rush Memory and Aging Project (MAP)(22), with available LC tangle and neuronal density measures. Juxtaposing the patterns of associations between LC MRI-intensity_r and PET-measures of AD-pathology, and the patterns observed between autopsy LC measures (LC tangle or neuronal density or pigmentation) and AD-pathology may further guide our interpretation of pathologic processes contributing to the in vivo LC MRI-signal.

RESULTS

Study samples

Participants (n=221; “in vivo dataset”) came from the longitudinal HABS cohort, a HABS-related study recruiting young and middle-aged individuals, and the Massachusetts Alzheimer’s Disease Research Center. All participants (n=221) underwent 3T-MRI imaging, including our dedicated LC-sequence (age range: 22 to 92 years, 102 females (55.43%)). Participants were grouped according to age: young (n=15; median age 26.0 [IQR,23.00–29.00]), middle-aged (n=32; median age 50.50 years [IQR,45.75–55.25]), and the older individuals were subsequently grouped according to their Clinical Dementia Rating (CDR)-status: cognitively unimpaired (CU) with CDR=0 (n=138; median age 75.00 [IQR,70.00–80.00]; 78 (56.52%) females) and cognitively impaired (CI) with CDR=0.5 or 1 (n=36; median age 76.38 years [IQR,66.69–83.63]; 18 (50%) females). Within the CI group, Aβ-PET was available for 34 individuals and 22 (64.71%) exhibited elevated Aβ-PET. From this in vivo dataset, n=174 individuals (“PET-sample”, 14 middle-aged CU, 138 older CU and 22 CI; Table S1) underwent both Aβ and TAU-PET imaging and 165 individuals also received repeated neuropsychological assessments (up to 8 years prior to MRI).

Both autopsy datasets consisted of older participants with quantitative neuropathology data on Aβ, TAU and LC integrity using study-specific methods. The ROSMAP-dataset included 1524 cases who received a summary diagnosis at death of cognitively unimpaired (CU, n=495), mild cognitive impairment (MCI, n=366) or AD (n=663), made blinded to all neuropathologic data. Available autopsy measures include cortical Aβ, Braak staging and cortical tangle density (Table S1). Complete cognitive data was available for 1065 cases (median time difference to autopsy:0.67 years [IQR,0.36–0.95]) and longitudinal antemortem cognitive data was collected up to maximum 25 years prior to death. In a subset of 160 MAP cases (66 CU, 53 MCI, 41 AD), LC tangle and neuronal density were also available (Results S1). The NACC-dataset included 2145 cases (CDR=0 n=295; CDR=0.5 n=300; CDR=1 n=323; CDR=2 n=441; CDR=3 n=786) who received a normal cognition, MCI or AD diagnosis at their last visit prior to autopsy (median time to death:11 months [IQR,6.00–28.00]), had AD neuropathologic evaluation with LC pigmentation scoring and Braak/Thal staging (Table S1;Table S20).

The relationship between age and LC intensity_r is modulated by pathology

We assessed differences in LC intensity_r across the in vivo age groups using ANCOVAs. We examined different measures of LC intensity_r, selecting either the voxel with highest intensity, or 5, or 8, or 12 contiguous voxels with highest intensities and normalized each slice to the 2D reference region (pontine tegmentum) (Fig.S1, S2). LC intensity_r values (5 voxels) were lower in CI compared to CU (p<0.001) and middle-aged individuals (p=0.001; Omnibus F_(216,3)=6.02, p<0.001; Fig.1A). Similar group differences were observed for the other LC measures (Fig.S3). Consistent with the existing literature, we continued all analyses with the LC intensity_r measure using 5 contiguous voxels.

Fig. 1:

Relationships between LC measures and age, CDR score / diagnosis or APOE-status.

A) Group differences in LC intensity_r when selecting 5 contiguous voxels with highest intensities (n=221, ANOVA, p<0.001, results for 1, 8 or 12 voxels are in Figure S3). B) Group differences in LC tangle density in MAP (n=160, ANOVA, p=0.001). Group comparisons for neuronal density and the adjusted tangle density are in Figure S3. C) Proportion of cases with LC hypopigmentation across the diagnostic groups (n=1524, χ², p<0.001) in ROSMAP. D) Proportion of cases with LC hypopigmentation at different CDR global scores in NACC (n=2145, χ², p<0.001). E) LC intensity_r in individuals carrying at least one ε4 allele and in those not carrying an ε4-allele (n=169, Wilcoxon Test, p=0.067). F) LC tangle density in cases with at least one ε4 allele and in those with no ε4 allele in MAP (n=160, Wilcoxon Test, p=0.078). G) LC hypopigmentation among ROSMAP individuals with at least one ε4 allele and in those not carrying an ε4-allele (n=1524, ordinal regression, p=0.054). H) LC hypopigmentation in individuals with 1 or 2 ε4 alleles or no ε4 allele in NACC (n=1853, ordinal regression, p<0.001). I) LC neuronal density across LC hypopigmentation rating scores in MAP (n=160, ANOVA, p<0.01; Fig S3 for tangle density results). J) Association between age and LC intensity_r (n=185, only CU individuals, robust regression, p<0.001). The intensities of the CI group (n=36, boxplot) were included as comparison. K,L) Effect of PiB or entorhinal TAU on the association between age and LC intensity_r in CU older individuals (n=152, robust regression, PiB: p=0.031 and entorhinal TAU: p=0.027). Individuals with elevated pathology are indicated in orange, those with lower pathology in blue (continuous interactions are reported in the text). Box plots show median and interquartile range overlaid with individual data points.

In MAP, patients with MCI or AD exhibited greater LC tangle density compared to CU (Omnibus F_(155,4)=9.54, p=0.001; p=0.04 and p<0.001, respectively, Fig.1B), and no overall group difference was observed for LC-neuronal density (F_(155,4)=2.62, p=0.08; Fig.S3). In the ROSMAP and NACC dataset, we observed a positive gradual relationship between the proportion of worse hypopigmentation and clinical diagnosis for ROSMAP or CDR-scores for NACC (Fig.1C, ROSMAP: χ²:p<0.001 (CN<AD, MCI<AD for no-maybe and no-yes), Linear-by-Linear (LbL):p<0.001; NACC: r_s=0.27, p<0.001; χ²:p<0.001, LbL:p<0.001, Fig.1D, Fig.S3).

LC measures in the in vivo dataset and (ROS)MAP-datasets were not different between APOE-ε4 carriers and non-carriers (Wilcoxon Test: in vivo:p=0.067; MAP:p=0.078; ROSMAP:ordinal regression model OR=1.25, 95%CI[0.99,1.57], p=0.054; Fig.1E–G), but we observed differences between ε4-carriers and non-carriers in the NACC dataset (OR=1.37, 95%CI[1.20,1.57], p<0.001; Fig.1H). No sex differences in LC measures were observed in the in vivo or (ROS)MAP datasets (Wilcoxon: in vivo: p=0.64; MAP:p=0.65; ROSMAP:OR:1.13, 95%CI[0.92,1.41], p=0.24), but in the NACC-dataset, males exhibited a greater proportion of worse LC hypopigmentation than females (OR=0.77, 95%CI[0.66,0.90], p=0.001); however, this difference did not survive age-adjustment (OR=0.93, 95%CI[0.79,1.09], p=0.37). When relating the postmortem LC measures to each other in MAP, we found that LC hypopigmentation rating was worse at higher LC neuronal density (F_(156,3)=7.40, p<0.001, Fig. 1I), and not related to LC tangle density (F_(156,3)=1.41, p=0.25, Fig.S3).

Given the discrepancies regarding age-associations in the autopsy and neuroimaging literature, we examined associations between age and LC intensity_r in the in vivo dataset. We observed a quadratic age-relationship (β=−0.21, t(218)=−3.57, p<0.001, 95%CI[−0.322,−0.094], Fig.1J) with the peak at 53.82 years. The shape of this age-relationship is congruent with immunohistochemistry reports of an inverse U-shaped function of density of neuromelanin cells in the LC across the lifespan(18).

To evaluate whether variability in LC intensity_r in clinically normal older individuals (n=152), and hence the direction of the age-related slope, may be driven by individuals with elevated pathology, we classified individuals as either having elevated Aβ or entorhinal TAU, the initial cortical region vulnerable to TAU, using Gaussian Mixture Models (GMM). Individuals with elevated Aβ (n=43) exhibited a negative relationship between age and LC intensity_r, whereas the relationship was flat in those with lower Aβ (β=−1.25, t(148)=−2.13, p=0.031, 95%CI[−2.403, −0.101], Fig.1K). When considering neocortical Aβ as a continuous measure, the interaction with age became not significant (β=−0.82, t(148)=−1.39, p=0.167). We observed similar differences between individuals with elevated (n=51) and lower entorhinal TAU (β=−1.14, t(148)=−2.21, p=0.027, 95%CI[−2.157,−0.130], Fig.1L), including when entorhinal TAU was entered as continuous predictor (β=−1.16, t(148)=−2.17, p=0.031). These observations suggest that LC intensity_r tracks with disease severity and that individual variability in AD-pathology may explain the previously reported different age-relationships with LC intensity_r.

LC measures are associated with initial AD-pathology

We then examined whether LC intensity_r relates to specific anatomic patterns of TAU (entorhinal and other cortical regions) or Aβ deposition in the PET-sample (n=174). Vertex-wise surface analyses on TAU (age and sex-adjusted; robust regression) with LC intensity_r as predictor demonstrated negative associations predominantly in medial and lateral temporal lobe and smaller clusters in medial prefrontal and parietal regions (Fig.2A, Table S2). The peak p-value (also the cluster with the largest surface area) was observed in the entorhinal cortices (Fig.2E). These associations, particularly in the left entorhinal cortex, did not change when adjusting for vertex-wise cortical thickness values (Fig.2B) or neocortical Aβ (Fig.2C) or when only considering CU individuals (Fig.2D, Table S2). The medial temporal lobe (MTL) plays an important role in memory and (preclinical) AD, but hippocampal TAU-PET signal is confounded by off-target choroid plexus binding (24). Therefore, we restricted MTL-analyses to amygdala TAU binding and observed a negative association with LC intensity_r in the entire PET-sample (p<0.001) and the CU individuals (p=0.002, Fig.S4A).

Fig.2:

Associations between cortical TAU and LC measures

A) Cortical vertex-wise associations between FTP-PET binding and LC intensity_r (adjusted for age and sex, n=174, robust regressions). B) For each vertex, the association between TAU and LC intensity_r was adjusted for the cortical thickness value at that vertex, C) or was adjusted for neocortical PiB. D) Vertex-wise associations between FTP-PET binding and LC intensity_r in CU individuals (n=152),. Threshold in surface maps was set at p<0.01 cluster-wise corrected (expressed in −log(10), min. cluster size of 88.9mm²). PET surface data was adjusted for partial volume effects using the extended Müller-Gartner method. The scale bars reflect the magnitude of the negative associations (yellow: greater; blue: smaller). Abbreviations: L=left; R=right. The plot in (E) shows the relationship between LC intensity_r and left entorhinal FTP (SUVr, PVC) indicated in A (arrow). F) Correlations between LC tangle density and cortical tangle density (partial Spearman correlation, p<0.001) and G) between LC tangle density and Braak stages in MAP (n=160, partial Spearman correlation, p<0.001). Correlations between LC neuron density and cortical tangle density or Braak stage are shown in Fig. S4B–C. H) Comparison of cortical tangle density across LC hypopigmentation ratings in ROSMAP (n=1524, partial Spearman correlation, p<0.001) I) The proportion of LC hypopigmentation scores across different Braak stages in ROSMAP (n=1524, χ^2, p=0.003). J) The proportion of LC hypopigmentation across different Braak stages in NACC (n=2145, χ², p<0.001). Box plots show median and interquartile range overlaid with individual data points.

In the postmortem datasets, we aggregated early Braak stages or Thal phases to balance the groups and have sufficient cases per stage/phase, also for CDR=0 or CU groups. In doing so, we aimed to be consistent with the estimated detection ability of PET-tracers facilitating comparisons with the PET-sample. In the MAP-dataset, LC tangle density correlated positively with Braak stages and cortical tangle density in the entire sample (partial Spearman (age-adjusted): r_s=0.55, F_(157,2)=81.3, p<0.001 (all pairwise comparisons significant) and tangles: r=0.64, p<0.001, Fig.2F–G) and the CU cases (n=66, Braak: r_s=0.54, F_(63,2)=49.7, p<0.001; tangles: r=0.53, p<0.001, Fig.S4D; LC neuronal density results: Fig.S4B–C). In ROSMAP, the effect size was smaller, but cortical tangle density was associated with LC hypopigmentation rating (r_s =0.09, p<0.001, Fig.2H). In both the entire ROSMAP and NACC-datasets, LC hypopigmentation scores were worse at higher Braak stages (ROSMAP: r_s=0.09, p<0.001; χ²:p=0.003 (no-maybe: stage 0-II vs V/VI; no-yes: stage 0-II/III/IV vs V/VI, Fig.2I, Table S3); LbL:p=0.006; NACC: r_s=0.34, p<0.001; χ²:p<0.001 (most comparisons to stage VI are significant; Table S3); LbL:p<0.001; Fig.2J), and also in CDR=0 NACC-cases (r_s=0.23, p<0.001; χ²:p=0.005 (none-severe: stage 0-II vs IV); LbL:p<0.001; Fig.S4F), but not in the CU of ROSMAP (r_s=−0.01, p=0.82; χ²:p=0.22; LbL:p=0.62, Fig.S4E).

When performing vertex-wise Aβ analyses with LC intensity_r, we observed widespread negative associations in the entire PET-sample and the CU individuals (Fig.3A–B, Table S4). No cluster survived when regressing out entorhinal TAU signal possibly due to the high amount of shared variance between LC intensity_r and entorhinal TAU (Fig.S5). These results suggest, consistent with autopsy data, that initial cortical TAU pathology is closely linked to LC intensity_r.

Fig. 3:

Vertex-wise associations between Aβ and LC measures

A) Cortical vertex-wise PiB associations with LC intensity_r (adjusted for age and sex, n=174, robust regressions). The scale bar reflects the magnitude of the positive associations (yellow: greater; blue: smaller). Threshold was set at a cluster-corrected threshold of p<0.01 (expressed as −log10, min. cluster size of 88.9mm²). Abbreviations: L=left; R=right. The scatter plot in B) shows the association between LC intensity_r and left precuneus PiB (DVR) (left arrow and circle in A). C) Association between LC tangle density and cortical Aβ density in MAP (n=160, partial Spearman correlation, p<0.001). Association between LC neuron density and cortical Aβ density is shown in Fig. S6A. D) Comparison of cortical Aβ density across different LC hypopigmentation ratings in ROSMAP, (n=1524, ANOVA, p=0.003). Box plot shows median and interquartile range overlaid with individual data points. E) Proportion of LC hypopigmentation across Thal phases in NACC (n = 2145, partial Spearman correlation, p<0.001).

In both autopsy datasets, LC tangle density or hypopigmentation were positively associated with cortical Aβ density or Thal phase, respectively (MAP: r_s=0.35, p<0.001; ROSMAP: F_(1520,3)=5.97, p=0.003; NACC:r_s=0.31, p<0.001; LbL:p<0.001; Fig.3C–E). Similar observations were made for neuritic but less consistently for diffuse plaque density (Fig.S6B–G).

We then examined whether the association between LC intensity_r and regional TAU was dependent on neocortical Aβ. At lower LC intensity_r, the positive association between neocortical Aβ and TAU binding extended beyond the entorhinal cortex, into medial and lateral parietal and lateral temporal regions for both the entire PET-sample and CU individuals (Fig.4A–B, Table S5).

Fig. 4:

Interactive effects among Aβ, TAU burden and LC measures

A) Vertex-wise FTP analyses examining the interaction between neocortical PiB and LC intensity_r (adjusted for age and sex, n=174, robust regressions). B) The line plot depicts the interaction between LC intensity_r and neocortical PiB (DVR, PVC) on left entorhinal-inferior temporal FTP (SUVr, PVC). For visualization, the estimated marginal means of LC intensity_r is shown at mean, + and −1 standard deviation, but analyses were done continuously. Shaded regions show 95% confidence interval. The scale bar reflects the magnitude of the positive associations (yellow: greater; red: smaller). Threshold was set at a cluster-corrected threshold of p<0.01 (expressed as −log10). Abbreviations: L=left; R=right. C) Comparison of LC tangle density across groups of increasing likelihood of a NIA-Reagan AD-diagnosis (for LC neuron density: Fig. S6F) in MAP (n=160, ANOVA, p<0.001). Box plot shows median and interquartile range overlaid with individual data points. D) Proportion of LC hypopigmentation across NIA-Reagan AD likelihood diagnostic groups in ROSMAP (n=1524, partial Spearman correlation, p=0.004). E) Proportion of LC hypopigmentation across groups based on the likelihood of AD neuropathologic change in NACC (n = 2145, partial Spearman correlation, p<0.001).

To examine the relationship between LC measures and the convergence of Aβ and TAU in the postmortem cohorts, we used the NIA-Reagan AD diagnostic criteria in MAP and observed greater LC tangle density with increasing likelihood of AD-pathology (r_s=0.51, F_(155,4)=19.57, p<0.001; Fig.4C). In the ROSMAP and NACC-datasets, we observed a higher proportion of worse LC hypopigmentation at greater – in particular from higher - NIA-Reagan AD or AD Neuropathologic Change likelihood scores (ROSMAP:r_s=0.07, p=0.004; NACC:r_s=0.32, p<0.001; ordinal regression results: Table S6, Fig.4D–E).

Together, we showed that in vivo LC intensity measures are associated with the initial anatomic pathology patterns of preclinical AD. Our results also reveal similar patterns for the in vivo LC measure and postmortem LC tangle density in their relationship to cortical pathology. On the other hand, LC pigmentation ratings correlate with LC neuronal density but not LC tangle density, and seem to reflect more advanced cortical pathology (above Braak stage IV or V), when Aβ is elevated.

Initial cortical TAU mediates the relationship between LC measures and memory

For the PET-dataset, we evaluated associations between LC intensity_r and composite measures for memory, executive functioning and the PACC5 composite, designed to detect Aβ-associated cognitive deficits, as well as their subtests (n=165; missing data for cognition=9). Robust regression analyses demonstrated that LC intensity_r (adjusted for age, sex, education and the number of previous test exposures) was associated with memory and PACC5 composites, but not the executive composite (Table S7). Similarly, only subtests capturing memory processes, in particular retrieval, were positively related to LC intensity_r (Fig.5A–B, Table S7). Controlling for Aβ (Table S8) resulted in similar associations with predominantly memory processes. At Aβ>1.189 DVR (~9 CL), lower LC intensity_r was associated with lower memory performance (Fig.5C, Table S9). Probing these associations in CU individuals revealed similar associations (n=149, Table S10).

Fig. 5:

Associations between cross-sectional cognitive domains and LC measures

A) Radar chart showing the magnitude of the relationship between LC intensity_r and cognitive measures in the PET-sample (n=165, Table S7). The green line in the radar chart indicates the effect sizes of composite measures and the blue line the effect sizes of subtests. Effect size are expressed in t-values (robust regression); red dotted line indicates t-value=1.96 (p<0.05), the outer line indicates t-value=3.50 (p<0.0005). TMT scores were inversed to facilitate comparison. B) The scatterplot illustrates the association between LC intensity_r and memory composite (z-score) for the PET-sample (n=165, robust regression, p=0.003). C) Effect of LC intensity_r on the association between PiB and memory performance (n=165, robust regression, p<0.001). D) Radar chart showing the magnitude of the relationship between LC tangle density and cognitive measures in MAP (n=160, Table S11). The outer line indicates t-value=−8.00 (p< 2.43e⁻¹³). E) The association between LC tangle density and memory in MAP (n=160, robust regression, p<0.001). F) Effect modification of LC tangle density on the association between Aβ density values and memory performance in MAP (n=160, robust regression, p=0.016). The grey area represents the range of PiB values (C) or Aβ density values (F) where the interaction was significant. For visualization purposes, LC intensity_r or LC tangle density is shown at mean, + and −1 standard deviation, but analyses were done continuously. Abbreviations: PACC=Preclinical Alzheimer’s disease Cognitive Composite; CAT= Categorical Fluency; FAS=phonological fluency; FcSRT= free and cued selective reminding test; SRT=selective reminding test (DR=delayed recall and TR=total recall); LM=Logical memory (IR=immediate recall); TMT=Trail Making Test; DSST=Digit Symbol Substitution Test.

For the postmortem data, we only analyzed the (ROS)MAP data to avoid potential biases (11% of the NACC-participants had complete UDSv2 neuropsychological data available). In MAP (n=160) LC tangle density was negatively associated with established composite scores of episodic memory, language, processing speed and global cognitive functioning, but not working memory (adjusted for age, sex, education and time difference between the last evaluation and autopsy; Fig.5D–E, Table S11). These findings, except processing speed, remained significant when adjusting for cortical Aβ density (Table S12). Zooming in on the CU cases (n=66), greater LC tangle density was only associated with worse episodic memory performance (also after Aβ-adjustment; Table S13). The negative association between LC tangle density and episodic memory was modified by Aβ density in the entire MAP-dataset (Aβ<13, Fig.4F, β=0.58, t(152)=2.51, p=0.016, 95%CI[0.127,1.034]). In contrast, LC hypopigmentation was not associated with any cognitive domain in all ROSMAP (Table S14) or the CU individuals (Table S15). In sum, these findings indicate that lower LC integrity – measured with in vivo MRI-signal or LC tangle density- signals initial AD-related memory dysfunction.

Given that these cognitive patterns are reminiscent to previously reported associations between TAU and cognition(25), we examined in the PET-sample which cortical vertices of TAU-FTP mediated this relationship. We found that the relationship between LC intensity_r and memory was predominantly mediated by entorhinal TAU (Table S16) in the entire sample and CU individuals. In the left hemisphere, this mediation effect extended into the fusiform gyrus, inferior temporal cortex and parahippocampus (Fig.6A), reflecting the verbal nature of the tasks. We confirmed this partial mediation at the bilateral ROI-level using a similar approach for the entire group (10,000 simulations; Fig.6B; mediation effect:β=2.17, p=0.003, 95%CI[0.71,3.98]; proportion mediated:β=0.39, p=0.003, 95%CI[0.15,0.77], n=165) and the CU individuals (mediation effect:β=1.23, p=0.017, 95%CI[0.22,2.55]; proportion mediated:β=0.311, p=0.017, 95%CI[0.06,0.77], n=149).

Fig. 6:

Vertex-wise mediation by TAU on the relationship between LC measures and memory

A) vertex-wise Quasi-Bayesian Monte Carlo Mediation analyses showing in which regions, TAU (FTP SUVr, PVC) mediated the relationship between LC intensity_r and memory (n=174). For orientation purposes, the entorhinal cortex is marked in the black outline (see magnification). The blue spot indicates the location of the mediation effect in CU individuals (n=152). Analyses were corrected for multiple comparisons using a cluster-wise correction at p<0.01. The scale bar reflects the magnitude of the probability of the indirect (mediation) effect (expressed in −log10(p-value); yellow: greater; red: smaller). B) ROI-based analyses for entorhinal FTP (SUVr, PVC) as mediator of the relationship between LC intensity_r and memory (z-score, n=174, p=0.003). C) Analyses for likelihood of Braak stage 2 or higher as mediator of the relationship between LC tangle density and memory (n=160, p=0.033). Abbreviations: EC= entorhinal, L=left, R=right

Consistent with the PET-sample, we dichotomized Braak staging (stage 0-II versus higher) to avoid running several mediation models and contrast comparisons due to the ordinal nature of the variable. We observed that a higher odds of having tauopathy beyond Braak stage II partially mediated the negative relationship between LC tangle density and episodic memory in MAP (mediation effect:β=−0.095, p=0.033, 95%CI[−0.21,−0.003]; proportion mediated:β=0.39, p=0.032, 95%CI[0.02,0.64], n=160; Fig.6C). Given that there were no associations between LC hypopigmentation and cognition, we did not perform mediation models for the ROSMAP-dataset.

LC measures are associated with AD-related memory decline

As LC-imaging was only recently introduced in the PET-sample we examined memory change retrospectively (n=165, number of observations=867) with linear mixed effects models using time from the LC-scan, age, sex, education, LC intensity_r and interactions by time as fixed effects. Random effects included individual subjects’ intercepts and slope (time). We hypothesized that participants with lower LC intensity_r (measured at time zero) would show faster rates of retrospective memory decline, and that this decline would be stronger in individuals with elevated Aβ(26). As control cognitive measure, we included the executive composite, as there is no overlap in the subtests of these composites.

Lower LC intensity_r was associated with a faster decline in memory and executive functions (Fig.7A). When probing interactions, we observed that at a given value of Aβ, lower LC intensity_r were associated with faster decline on the memory composite, not on the executive function composite in the entire PET-sample and the CU individuals (Table S17). We then used floodlight analyses and quantified that the relationship between LC intensity_r and retrospective memory decline became significant (p_fdr<0.05) at Aβ>1.19 DVR (9 CL) for the entire sample or at Aβ>1.24 DVR (13 CL) for CU individuals. (Fig.7B). This indicates that LC intensity_r detects AD-related memory at subthreshold values, earlier than the established GMM-derived Aβ cut-off (18 CL).

Fig. 7.

Associations between LC measures and retrospective cognitive decline

A) Linear mixed effect analyses examining the relationship between LC intensity_r and retrospective memory decline (n=165, 867 observations, p<0.001). B) Effect of LC intensity_r on the association between neocortical PiB binding and retrospective memory decline (z-score) in the PET-sample (n=165, linear mixed effect analyses, p<0.001). The vertical solid line represents the range of PiB values at which the association is significant in the entire sample, and the dashed line for the CU individuals (n=152, floodlight analyses, estimated at p_fdr<0.05). C) Linear mixed effects model in MAP showing the associations between LC tangle density and memory decline. (n=160, 857 observations, p<0.001). D) Effect of LC tangle density on the association between Aβ density and retrospective memory decline (n=160, 857 observations, linear mixed effect analyses, p=0.009). For visualization, LC intensity_r or LC tangle density are shown at simple slopes estimated at mean, + and −1 standard deviation, but analyses were done continuously. Shaded regions show the 95% confidence interval.

For the MAP-dataset we examined antemortem cognitive change (n=160, number of observations=857, median time=3.97 years [IQR, 2.55–5.82]). LC tangle density was associated with declines in episodic memory (Fig.7C), language and processing speed. However, beta-amyloid only moderated the relationship between LC tangle density and decline in memory, not language or processing speed (Fig.7D, Table S18). In the ROSMAP-dataset, we observed no associations between LC hypopigmentation and retrospective cognitive changes (Table S19). These findings demonstrate that LC intensity_r and LC tangle density are associated with Aβ-related memory decline, starting at lower Aβ values.

DISCUSSION

Accumulation of beta-amyloid and hyperphosphorylated TAU are the defining neuropathologic hallmarks of AD (1, 2). As the first pathological changes emerge decades prior to the first clinical symptoms (27), being able to detect and measure the initial site of pathology will be critical to improve early detection and identify individuals eligible for clinical trials aimed at delaying the disease process. Autopsy studies have identified the LC as the origin location of TAU pathology (2, 4), before cortical Aβ or cognitive deficits are detectable. However, besides the difficulties in examining this region in vivo, the question remained whether in vivo LC measures relate to initial cortical patterns of AD-related fibrillar proteinopathies and cognitive dysfunction.

In the present work, we provide in vivo evidence that structural properties of the LC captured by MRI-intensity_r track with the pathophysiologic and cognitive features of preclinical AD. By combining dedicated LC-imaging with Aβ and TAU-PET data and longitudinal cognitive assessments in older individuals, we showed that the relationship between LC intensity_r and age is contingent on AD-pathology burden. In the entire sample, worse LC intensity_r related to greater TAU deposition in medial and lateral temporal regions, extending into smaller clusters of medial parietal-frontal regions. Of note, when restricting these correlations to the cognitively unimpaired individuals, the spatial relationships between LC intensity_r and TAU were predominantly restricted to the entorhinal cortex suggesting a temporal ordering consistent with the Braak staging. Upon verification of these associations in the autopsy cohorts, we similarly observed that LC tangle density measures correlated with Braak staging, even in CDR=0 cases, and a proportion of the individuals exhibited worse LC integrity when no cortical pathology was detectable confirming that LC measures may reflect initial TAU accumulation, possibly preceding substantial Aβ-pathology and cognitive decline. At elevated Aβ, LC intensity_r was associated with greater extra-medial temporal lobe TAU pathology, including the typical AD-pattern of lateral temporal and medial-lateral parietal and frontal regions. These patterns are consistent with current disease models, autopsy data as well as human in vivo studies demonstrating that TAU deposition outside the medial temporal lobe typically occurs when Aβ is elevated. Finally, our cognitive data demonstrated that LC intensity_r was associated with retrospective decline in the context of elevated Aβ burden. These results as well as the corresponding results within the autopsy-datasets support LC intensity_r as a potential promising marker to detect initial AD-related processes in clinically normal individuals.

Measuring TAU pathology in the LC in vivo is currently not feasible because of its diminutive size and off-target PET-binding to neuromelanin (8, 24). Our results are consistent with reports showing that the presence of neuromelanin cells may indirectly contribute to the MRI-contrast of the LC-signal (13, 28), but highlight that changes in neuromelanin pigmentation affects the LC predominantly in later disease stages. More importantly, the consistency in patterns between the PET-data and MAP-data indicates that the magnitude of MRI signal intensity_r may be a proxy of initial TAU, possibly including TAU inclusions in neuromelanin cells.

Critically, our associations between LC intensity_r and AD-pathology, as measured with PET-imaging, are not random, but followed topographical patterns reported in seminal autopsy staging studies (1, 2, 27). Our vertex-wise analyses revealed a specific anatomic pattern between LC intensity_r and TAU pathology in the vicinity of the transentorhinal / entorhinal cortex, the first cortical site of TAU, prior to the accumulation of Aβ (27). These results were confirmed after covarying for local cortical thickness or neocortical Aβ. This is further substantiated by analogues between the in vivo and autopsy correlations. The dichotomy between LC tangle density on the one hand and neuronal density or hypopigmentation on the other are consistent with earlier autopsy reports of LC neurons being sturdy against death, despite accrual of TAU (5). The in vivo LC intensity_r and autopsy LC tangle density measures also covaried with cognitive dysfunction, indicating that LC intensity_r represents a biologically and clinically relevant measure and a potential proxy of initial AD-related processes.

The protracted time between the initial occurrence of TAU in the LC and the first AD-related cognitive symptoms stirred a debate on whether the LC reflects age-related or AD-related changes. The role of the LC in cellular processes of learning, such as long-term potentiation, was first identified by Kety in 1972, and its powerful modulatory role in cognition is now well-established (29). Associations between LC intensity_r and memory performance were strongest for retrieval-related processes, echoing animal observations (30–32). Our data reveals that the association between lower LC intensity_r and memory dysfunction is mediated by entorhinal TAU and can be detected at Aβ values below our established GMM cut-off. In addition, lower LC intensity_r was associated with steeper AD-related retrospective memory decline, indicating that LC intensity_r conveys risk and is not merely an age-related phenomenon. These observations demonstrate AD-related associations and extend previous autopsy studies reporting that individuals with fewer norepinephrine neurons in the LC exhibited steeper antemortem cognitive decline, even after adjustment for common neuropathologies elsewhere in the brain (6, 33). Interestingly, in agreement with previous data (6), higher LC intensity_r in the context of elevated Aβ may signal resiliency to cognitive decline (Fig.S7), though the number of observations were limited in this range.

Our findings are in agreement with the seminal Braak staging, but they do not provide insight in how TAU may progress from the LC to cortical regions. Autopsy studies observed TAU pathology in the LC in 50% of the 30-year old cases who had no cortical pathology (4). By age 50, 50% of the cases exhibited TAU pathology in the entorhinal cortex, suggesting propagation of TAU from the LC to entorhinal cortex (34–36). Animal studies demonstrated that infusions of human hyperphosphorylated TAU in the LC increases local TAU and induced spread of TAU to cortical regions (37), including the allocortex (36). This local buildup of TAU affects the density of the ascending fiber projections from the LC (36, 38). Propagation may be associated with other physiological processes, such as dysregulation of adrenoreceptors leading to hyperexcitable neurons (39) and increases in microglial activation (40, 41), which in turn curtails Aβ clearance and perturbs synaptic plasticity in target zones(42). Together these processes facilitate cognitive decline, even at lower Aβ burden (40, 43–45).

On the other hand, seeding studies in postmortem tissue provided conflicting results on whether the LC is the origin of seeding to the entorhinal cortex (46, 47), and suggested that spreading of TAU from the LC may occur through other mechanisms, such as volume transmission (48). The fact that the limbic system, especially the entorhinal-hippocampal pathway, has a high density of adrenergic receptors (49), a greater propensity for long-term potentiation and other neuroplasticity processes (50), which are also under norepinephrine control, does not preclude the possibility of volume transmission, but provides a plausible hypothesis for the preferential spatial link between the LC and entorhinal cortex in the initial stages of the disease.

Our study has limitations. First, because of the recent introduction of LC-imaging and TAU-PET imaging in HABS, we were not yet able to assess the relationship between LC intensity_r and regional increases in TAU, or prospective cognitive decline. Another limitation is that our PET-data was only available in the older individuals. Future studies will be needed to relate LC intensity_r to TAU or Aβ in younger individuals. Finally, the LC hypopigmentation rating is inherently subjective and developing a semi-quantitative approach would enable more detailed analyses. Nonetheless, the patterns with pathology observed for ROSMAP and NACC were largely consistent, even though each cohort has its own intricacies.

In summary, MRI-measures of LC intensity_r are associated with initial TAU accumulation in the entorhinal cortex, consistent with successive Braak staging. Lower LC intensity_r was also associated with cognitive decline in the context of elevated Aβ in clinically normally older individuals. These findings suggest that LC intensity_r tracks with the initial stages of AD pathology and that it might be a promising marker to predict AD-related cognitive decline. These results bolster the idea that structural LC changes are not merely age-related but reflect initial AD-related processes. The directionality or mechanisms enabling TAU spread from the LC could not be tested with our data and remains an open question. Nonetheless, our findings provide critical empirical foundations for future work regarding the relevance of the LC in AD and illustrate LC intensity_r as a valuable marker of initial AD-related cortical fibrillar proteinopathy.

MATERIALS AND METHODS

Study design

This was an observational study design, with the objective to examine whether in vivo LC measures relate to initial cortical patterns of AD fibrillar proteinopathies or cognitive dysfunction, and to verify these patterns in two postmortem datasets. To that end, we included dedicated LC-MRI scans, ¹¹C Pittsburgh Compound B (PiB)-PET, ¹⁸F Flortaucipir (FTP) PET imaging and longitudinal cognitive observations from n=174 individuals from HABS (and its affiliated studies) and the MADRC. Experimenters collecting MRI or behavioral data were blind to Aβ-status or TAU binding and experimenters collecting PET-data were blind to the behavioral data and MRI results. For the postmortem data, we obtained data from the ROSMAP and the NACC, each providing data on Aβ, TAU and the LC, determined at autopsy using immunohistochemistry or visual ratings, and cognitive function measures (see below for more details). Clinicians determining the diagnosis were blinded to postmortem data. Each participant underwent identical procedures, so no randomization occurred. The inclusion and exclusion criteria are provided separately for each cohort. Sample sizes were not calculated or determined in advance; all available cross-sectional and longitudinal data from the studies were used.

Participants

In vivo dataset

Participants (n=221) were recruited from the HABS (51) (and its related studies) and a Tau-study recruiting from the MADRC. All participants underwent 3T MRI imaging, including the dedicated LC sequence. The age range of the entire sample was 22 to 92 years. Participants were grouped according to age: young (n=15), middle-aged (n=32) and older (n=174), and the older individuals were subsequently grouped according to their CDR-status: cognitively unimpaired (CU) with CDR=0 (n=138) and cognitively impaired (CI) with CDR=0.5 or 1 (n=36). From this sample, n=174 individuals (14 middle-aged CU, 138 older CU and 22 CI) underwent PiB-PET, FTP-PET-imaging, and 165 individuals (all from HABS) received extensive repeated neuropsychological assessments (up to 8 years prior to the MRI).

Younger participants (n=15) were recruited from the community for pilot-testing of new fMRI-designs for HABS. They were included if they had no cognitive complaints and no major psychiatric or neurological disorders. Middle-aged participants came from the HABS-related studies (n=16) or from HABS (n=13 and n=3 from its affiliated study). Older individuals were recruited from HABS (n=154) or from the MADRC (n=20). The CI group consisted of the MADCR participants and 16 individuals from HABS who received a CDR=0.5 or 1 during a clinical consensus meeting.

For HABS, participants were included at baseline if they had a CDR global score of 0 (52), a MMSE score ≥ 25 and performed within education-adjusted norms on the Logical Memory delayed recall (>10 for ≥ 16 years of education, > 6 for 8–15 years of education and >4 for <8 years of education). All participants underwent at least one comprehensive medical and neurological evaluation and had no major psychiatric or neurological disorders. Presence of clinical depression (Geriatric Depression Scale below 11/20 (53)) or other psychiatric illnesses, history of alcoholism, drug abuse or head trauma were exclusion criteria. Participants undergo annual extensive neuropsychological assessments, including CDR-rating. CDR rating is performed independently from the other cognitive test results by trained neuropsychologists and psychiatrists, who are blinded to biomarker status. As study participation is ongoing, participants had at least one test session within one year of the MRI-scan and additionally up to 8 assessments prior to this data (median: 6 assessments).

In addition, 3T MRI, including the dedicated LC-imaging method was also collected in 20 participants with clinical impairment (global CDR-score of 0.5 or 1) recruited at the memory disorders clinics at Massachusetts General Hospital (MGH) and BWH for a PET-study. PiB-PET was available for 18 individuals and 12 (66.67%) exhibited elevated PiB-PET, and for six individuals FTP-PET was available.

Study protocols were approved by the Partners Human Research Committee of MGH and all participants provided written informed consent.

ROSMAP-dataset

The dataset included 1524 participants from the Religious Orders Study (ROS) and the Rush Memory and Aging Project (MAP), two ongoing longitudinal clinical-pathologic studies known as ROSMAP that started in 1994 and 1997 (22, 54, 55). The two studies are run by the same team and share a larger common core of data at the item level including clinical and pathologic data. Eligibility criteria included age>55 years, absence of a previous dementia diagnosis and consent to annual clinical evaluation and brain autopsy at death. Participants were recruited from Convents and Monasteries, or from retirement communities, social service agencies and subsidized housing facilities, and individual homes in the Chicago metropolitan region. Participants included in the current analyses had a clinical diagnosis of cognitively unimpaired (n=495), MCI (n=366) or AD (n=663) at the time of death. Diagnosis was done each year by a neuropsychologist and clinician, and final diagnosis was by a neurologist blinded to postmortem data, based on the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS/ADRDA) criteria as previously reported (56, 57). In addition, neuropathologic assessments, including LC hypopigmentation, were available. Median time interval between last visit and death for these participants was 0.67 years [IQR, 0.36–0.95]. A subset of the 160 MAP participants was analyzed separately as these individuals also had LC tangle and neuronal density measures available. The studies were each approved by an institutional review board of Rush University Medical Center. All participants signed an informed consent, an Anatomical Gift Act, and a repository consent which allowed their data to be shared.

NACC-dataset

Data included visits between September 2005 and September 2019 from the NACC, a repository for data of the Alzheimer’s Disease Research Centers (ADRCs) located across the USA and funded by the National Institute on Aging (NIA). The ADRCs contribute standardized clinical data and neuropathological evaluations obtained at autopsy. Each local ADRC obtained written informed consent, including consent for autopsy from participants, and study protocols were approved by the ADRCs’ institutional review boards. Research using the NACC database was approved by the University of Washington Institutional Review Board. The clinical and neuropathology datasets have been described in detail previously (23, 58–60). The dataset included 2145 participants from 30 different ADRC’s who had neuropathology data available within the v10 neuropathology form and in particular the rating of the LC hypopigmentation. Participants were selected if they had either normal cognition (n=308), MCI (n=215) or AD (n=1622) at their last clinical visit prior to autopsy. Normal cognition was based on the UDS diagnosis form, including expert or consensus diagnosis, and also required a global CDR (CDR@ Dementia Staging Instrument) of 0 at the last visit prior to death. Median time interval between last visit and death for these participants was 11 months [IQR, 6.00–28.00]. All data was de-identified.

Structural Magnetic Resonance Imaging (for the in vivo dataset)

Magnetic resonance imaging was performed at the MGH, Athinoula A. Martinos Center for Biomedical Imaging on a 3T-imaging system (TRIM Trio; Siemens). Structural 3D T1-weighted volumetric magnetization–prepared rapid-acquisition gradient-echo (MPRAGE) images were collected (repetition time=2300 ms, echo time=2.95 ms, and inversion time=900 ms, flip angle=9°; and 1.05 × 1.05 × 1.20 mm resolution). To visualize the LC, we applied a 2D-T1-weighted Turbo-Spin-Echo (TSE) sequence with additional MT contrast (repetition time=743ms, echo time=16ms, flip angle=180°, 6 slices, 4 online averages, 0.4 × 0.4 × 3.00 mm resolution, acquisition time: 3:22 min). The short acquisition time minimizes the impact of motion, which is critical for such a tiny structure close to the fourth ventricle. Participants were repeatedly reminded to stay still and a piece of tape connecting their forehead with the coil provided proprioceptive feedback to minimize motion.

T1-weighted images were processed using FreeSurfer (FS) version 6.0.0 using the software package’s default, automated reconstruction protocol as described previously (61). Briefly, each T1-weighted image was subjected to an automated segmentation process involving intensity normalization, skull stripping, segregating left and right hemispheres, removing brainstem and cerebellum, correcting topology defects, defining the borders between grey/white matter and grey/cerebrospinal fluid and parcellating cortical and subcortical areas. Using FS’s native visualization toolbox, we visually inspected and, if necessary, edited each image.

Positron Emission Tomography (for the in vivo dataset)

The PiB-PET and FTP-PET data were acquired at MGH on a Siemens/CTI ECAT HR+ scanner as previously reported(26): PiB-PET was acquired with bolus injection (8.5 – 15 mCi) followed immediately by a 60-minute dynamic acquisition in 69 frames (12×15 seconds, 57×60 seconds) and FTP was acquired from 75–105 minutes after bolus injection (9.0 – 11.0 mCi) in 4 × 5-minute frames. PET data were reconstructed applying standard data corrections(26). Each frame was evaluated to verify adequate count statistics and motion correction was applied using an automated frame-to-frame realignment algorithm and visually checked.

To evaluate the anatomy of cortical FTP binding, each individual PET data set was rigidly coregistered to the subject’s MPRAGE data. The FS’s region-of-interest (ROIs) were transformed into the PET native space. PiB PET data were expressed as the distribution volume ratio (DVR) with cerebellar grey as reference tissue by using the Logan graphical method applied to data over the 40 to 60 minute post-injection integration intervals (62). Regional PET data were partial volume corrected using the Geometrical Transfer Matrix method as implemented in FS (63), assuming an isotropic 6mm point spread function. For surface-based FTP-PET analyses, we implemented the extended Muller-Gartner correction for partial volume effects in volume space prior to the surface sampling. We applied surface-smoothing equivalent to 8mm FWHM Gaussian kernel (Fig.S8 provides an overview of the PET processing steps). Regional FTP binding was expressed in FS-defined regions (entorhinal and amygdala) as the standardized uptake value ratio (SUVr) using cerebellar grey as reference.

Neocortical PiB retention was also assessed in a large cortical ROI aggregate that included frontal, lateral temporal and retrosplenial cortices (FLR). Aβ-status was ascertained in this FLR region using a previously determined cutoff value based on Gaussian mixture modeling approach cutoff value=1.324 DVR, PVC (64). Based on this cut-off, 56 individuals were classified as having elevated Aβ. To facilitate comparisons to other tracers or studies, we also calculated the centiloid (CL) values for important cut-points using the conversion previously reported (65) (the cutoff value of 1.324 DVR is equivalent to 18.5 CL).

For entorhinal TAU, we used the same approach and multiple Gaussian distribution were fit to the data (1 to 5 distributions or mixtures, allowing for either equal or unequal variances) and the optimal model was selected by evaluating the Bayesian information criterion, which penalizes models with more parameters) and a bootstrapped (n=10,000) sequential likelihood ratio test. The optimal model consisted of two distributions of unequal variance (BIC= −71.05, LRTS for 2 mixture components=54.24, p=0.001; LRTS for 3 mixture components=3.16, p=0.456). The probability threshold for belonging to either the low or high entorhinal TAU group (PVC) was set at 50%, which corresponded at a value of 1.36 SUVr.

Median time between the MRI scan and PiB-PET scan was 0.62 months [IQR, 0.03–2.20] and between the MRI scan and FTP-PET scan 1.08 months [IQR, 0.13–3.61].

Identification and quantification of the locus coeruleus intensity (for the in vivo dataset)

To calculate the signal intensity_r in the LC, we defined four equidistant boxes on the MNI 0.5mm template (Fig. S1) covering the bilateral LC region (~349mm² each) and the bilateral rostral pontine tegmentum (~2,715mm² each, reference region). These boxes were used to guide the search for intensities related to the structure of interest and removed any possible experimenter bias in identifying the LC (adapted from (66)). To ensure anatomical accuracy, we ensured that the LC boxes overlapped with another available validated template (11). The LC box spanned the rostral to caudal length from the inferior boundary of the inferior colliculus to the lateral recess of the fourth ventricle.

All these boundary regions were warped to each individual LC-scan by first registering the MNI template to each individual native T1-weighted scan using the high-dimensional symmetric diffeomorphic normalization transformation of the Advanced Normalization Tool (ANTs, Philadelphia, USA) warping first the MNI template to the each native T1 and consequently using the transformation model and the nearest neighbour interpolation to warp the four boxes to each T1 image (67). Each LC scan was rigid-body registered to its T1-weighted image and the inverse matrix was applied to linearly register the boxes in T1-space to the LC-scan space. Each scan was checked visually to check the registration and ensure that the boxes at the left and right LC or reference regions were not connected and that the reference regions did not contain hyperintense signal from the substantia nigra or the cerebrospinal fluid.

The reference region was used to remove the interslice variability that is inherent to the TSE sequence, by normalizing each slice to the average intensity of the reference region. This normalization ensures that intensity values can be compared across participants and removes the age-confounding previously reported in the reference region. The LC intensity_r was determined by the maximum of the average of iteratively (n=30) searching the i-number of connecting voxels with the highest intensity values (sensitivity analyses were done by varying i between 1 and 12; Fig. 1). Autopsy studies do not report asymmetry in LC TAU or neuronal changes and to maintain consistency with the other datasets, we averaged left and right.

Neuropathological measures (for the ROSMAP and the NACC-dataset)

ROSMAP includes a wide range of neuropathologic measures, obtained by quantitative histochemistry and immunohistochemistry, as well as Braak stage and the NIA-Reagan pathologic criteria for AD. Postmortem interval, the time between death and brain removal, was median: 6.80 hours [IQR, 5.00–10.25]. Upon participants’ death, brains were extracted, weighed, and the brainstem and cerebellar hemispheres removed. One hemisphere and the brainstem were sectioned into 1 cm-thick coronal slabs and stored (the other hemisphere was frozen). Neuropathological indices were examined to quantify neuropathologies. Modified Bielschowsky silver stain was used for Braak staging and to identify AD pathology based on NIA-Reagan and modified CERAD criteria. Amyloid-β load was quantified as percent area occupied by amyloid-β, labeled with a N-terminal directed monoclonal antibody, which identifies both the 1–40 and 1–42 length Aβ fragments, while PHF-TAU tangles, was quantified as the density of paired helical filament TAU tangles with an antibody specific for phosphorylated TAU, AT8. Quantification of tangle density per square millimeter was performed using stereologic methods across eight regions, and percent area amyloid-β was done using in-house automatic thresholding methods on the images of eight regions. The average for all regions was calculated for each pathology (68, 69). Within ROSMAP loss of pigmentation in the LC was macroscopically rated as yes, possible or not present. In a subset of 160 MAP participants, neuronal density (per mm²) and tangle density of the LC were examined using immunohistochemistry with a monoclonal anti-tyrosine hydroxylase antibody and an anti-paired helical filaments TAU antibody AT8, respectively, each bilaterally at two levels of the LC (rostral and main body) (6, 70).

The NACC Neuropathology Data set includes Braak staging of neurofibrillary tangles (0 or none; I-II: entorhinal; III-IV or limbic, and V-VI or isocortical)(1), Thal phases of amyloid deposition following NIA-AA guidelines (0 or no amyloid, 1/2 or isocortical/allocortical, 3 or basal ganglia, and 4/5 or brainstem/cerebellum) (71), CERAD scoring of neuritic neocortical or diffuse plaques (none, sparse, moderate, or frequent) (72) and AD neuropathologic change measures using the ABC criteria (A=Thal, B=Braak, C=CERAD) resulting in a likelihood estimation of AD neuropathologic change (no, low, intermediate and high) (73). Severity of LC hypopigmentation was rated from 0 (none) to 3 (severe) following the NACC guidebook (23, 58–60).

Genetics

Within the in vivo dataset, APOE-genotyping was done through a targeted Taqman assay for rs7412 and rs429358, and people with at least one APOE ε4 haplotype were grouped as APOE ε4 carriers. Within the ROSMAP-dataset, DNA was extracted from peripheral blood or frozen postmortem brain tissue and genotyping was performed at Polymorphic DNA Technologies by investigators blind to the clinical and pathologic data (74). For the NACC-dataset, genotyping was performed independently by the ADRC, Alzheimer’s Disease Genetics Consortium (ADGC) or the National Centralized Repository for Alzheimer’s disease (NCRAD) and provided to the NACC.

Neuropsychological assessments

In vivo dataset

We examined several composite measures that were created within HABS or that were created to detect early cognitive decline in preclinical AD. Memory and executive function composite scores were created based on a factor analysis from the entire HABS cohort (n = 284) (35). The memory composite (with factor loading weights between parentheses) included the z-score transformations of the delayed recall scores of the 6-Trial Selective Reminding test (0.739), the free recall of the Free and Cued Selective Reminding Test (0.605) and the delayed recall of the Logical Memory Test (0.534). Alternate forms were administered for the FCSRT (A-B-C-A-B-C), whereas same forms were administered for all other measures. The executive function composite included the z-score transformations of the Trail Making Test form B minus A (0.666), the Letter Number Sequencing test (0.533) and the phonemic fluency FAS test (0.622).

The Preclinical Alzheimer’s disease Cognitive Composite – 5 subtests (PACC5 (75)), is based on the PACC96 (76). In addition to averaging the z-scores across the MMSE, Logical Memory Delayed Recall, Digit-Symbol Substitution Test, Free and Cued Selective Reminding test (free and total recall), the PACC5 also includes a standard semantic memory test (Category Fluency (CAT). This configuration of the PACC5 has shown to add unique information about early cognitive decline, that is not captured by the PACC96.

As LC-imaging was added recently to the study, the number of prior exposures to these tests varied from 1 to 9. To account for differences in practice effects, we entered in all cross-sectional analyses the number of prior exposures to the analyses. Complete neuropsychological data within 1 year of the MRI and PET scan sessions (median:1.15 months [IQR, 0.10–2.92] was available for 165 participants (number of prior exposures: 44 individuals were tested only at baseline, 2 participants till year 2, 13 participants till year 4, 13 participants till year 5, 40 participants till year 6, 6 participants till year 7, 3 participants till year 8 and 44 participants till year 9). For longitudinal analyses, we used all available retrospective data (n=165, number of observations=867). The median retrospective follow-up duration was 5.02 years [IQR, 4.48–8.01]. The intra-class correlation coefficient (ICC) adapted for longitudinal data, demonstrated adequate measurement reliability over time for all composite measures (memory: 0.683; executive: 0.850; PACC5: 0.708).

ROSMAP-dataset

For each participant, comprehensive cognitive assessments were administered at baseline and each annual follow-up visit. The testing battery contains a total of 21 cognitive tests; the studies have 19 tests in common and 17 were used to summarize cognitive domains, including episodic memory (seven subtests), language (three subtests), working memory (three subtests), and processing or perceptual speed (four subtests). Composite measures were established based on factor analyses within the entire ROSMAP database using z-scores based on baseline mean and standard deviation. Global cognitive functioning was assessed using the global composite, the average of these composites, and the MMSE (77, 78).

For the longitudinal analyses of ROSMAP, n=1065 individuals had all cognitive data available at their closest time point to autopsy (median time difference: 0.67 years [IQR, 0.36–0.95]). These individuals had been followed for max. 25 years prior to death n=8,216 observations (see the distribution of number of observations in Fig.S8–9).

For longitudinal analyses of MAP we included 857 observations of n=160 participants during a 13 year period (see the distribution of number of observations in Fig.S9–10). For the language composite, there were in total 848 observations of 158 subjects. The median time of retrospective follow-up was 3.97 years [IQR, 2.55–5.82].

Statistical analyses

Statistical analyses were performed using statistical software (R version 3.5.1, http://www.r-project.org/). All analyses were done during March 2020 –March 2021. Group characteristics are represented in median and interquartile range. The threshold for statistical significance was set at p<0.05 (or lower, as indicated at the respective analyses).

Group differences in LC measures were tested with AN(C)OVA’s and followed up with post-hoc pairwise comparisons with Tukey correction for continuous LC variables. Proportional LC variables were tested using χ² tests with false discovery rate (FDR)-correction for pairwise comparisons. APOE-ε4 genotype or sex differences in LC measures were tested with the Wilcoxon Rank Sum Test or proportional odds logistic regression (cumulative link) for ordinal data. Associations between LC intensity_r and age and interactions between age and biomarker data (Aβ or entorhinal TAU; adjusted for age and sex) as well as associations between cross-sectional cognition and LC measures (covariates: age, sex, education and number of previous test exposures (or time between visit and death for autopsy data); and in a second step: Aβ) were examined using linear robust regression methods with the Huber-M estimator. Robust regression is a more conservative test compared to linear least-square regression methods, as the resulting models are stout against outliers. If the predictor was ordinal (LC hypopigmentation), post-hoc contrasts were estimated using the marginal means including FDR-adjustment for multiple comparisons.

Vertex-wise associations between either Aβ or TAU and LC intensity_r, and interactions between LC intensity_r and Aβ on TAU (adjusted for age and sex) also employed robust linear regressions. To mitigate inflated type 1 errors and at the same time take into account the point-spread function of PET data on surface areas, we chose to apply a cluster-wise threshold of p<0.01 with a minimal estimated surface-area size of an 88.9mm², taking the point-spread-function of 6mm FWHM into account (65). In order to understand the contributions of LC intensity_r and entorhinal TAU on the multivariate Aβ maps, we vertex-wise partitioned the unique effect of LC intensity_r as well as its shared variance with entorhinal TAU.

Given that the χ² test does not hold strong assumptions on the ordering of the data, we used the χ² test (and FDR-adjusted pairwise comparisons) to assess frequencies differences for ordinal data in which we changed the categorization to ensure sufficient observations per cell (concatenating Braak stage 0 to II). Data in which categories were not altered (ADNC likelihood) we used the proportional odds logistic regression. In addition, we also report the linear-by-linear (LbL) (79) association test for ordered data or the a partial Spearman rank correlation test (adjusted for age at death) where appropriate.

Mediation analyses were performed using the quasi-Bayesian Monte Carlo approximation (1,000 simulations per vertex for surface-based analyses and 10,000 simulations for ROI-analyses) in which the posterior distribution of quantities of interest is approximated by their sampling distribution. The proportion mediated is expressed as the causal-mediated effect divided by the total effect. A similar analytic strategy was created for the autopsy-datasets, by binarizing the mediator (Braak stages:0–2 versus 3–6) to reduce the number of contrasts and models. Mediation analyses were performed by implementing a logistic and linear regression into the quasi-Bayesian Monte Carlo approximation framework (10,000 simulations). In all mediation analyses we included age (age of death), sex and education and time between the last assessment and death as covariates.

Longitudinal analyses were performed with a hierarchical set of hypotheses-driven linear mixed-effects (LME) models using the restricted maximum likelihood estimation, containing a fixed effect for the predictor(s) of interest, a random intercept for each subject and random slope for retrospective time (time before the LC measurement). In all LME models, age, sex and education and their interaction with time were included as covariate if p < 0.10 (using the Wald t-statistic). Floodlight analyses were applied to determine the region of significance of moderation effects by calculating the value where Aβ produces the α-value of 0.05. Residual plots and Q-Q-plots were examined for all models. All reported p-values were two-sided.

Supplementary Material

Accepted supplemental data

Fig. S1: Overview of the methodology to quantify locus coeruleus intensity_r

Fig. S2: Examples of LC scans

Fig. S3: Sensitivity analyses of group comparisons of other LC measures

Fig. S4: Additional results of associations between TAU pathology and LC measures

Fig. S5: variance decomposition of LC intensity_r and entorhinal TAU on Aβ

Fig. S6: Additional results of associations between Aβ and LC measures

Fig. S7: Model relating various LC measures to pathology and cognition across the lifespan

Fig. S8: Processing pipeline of the PET-data

Fig. S9: Spaghetti plots of individual retrospective slopes of memory in the PET-sample and the MAP-dataset

Fig. S10: Distribution of the number of observations in ROSMAP and MAP

Table S1: Demographics of the PET-sample, the ROSMAP-dataset and the NACC-dataset

Table S2: Clusters of the TAU vertex-wise association with LC intensity_r

Table S3: Contrast comparisons between LC hypopigmentation and Braak stages in ROSMAP and NACC

Table S4: Clusters of the Aβ vertex-wise association with LC intensity_r

Table S5: Clusters of the TAU vertex-wise association with LC intensity_r, interacting with Aβ

Table S6: LC pigmentation in relationship to AD likelihood in ROSMAP and NACC

Table S7: Cross-sectional associations between LC intensity_r and cognitive performance in the entire PET-sample (n=165)

Table S8: Cross-sectional associations between LC intensity_r and cognitive performance, adjusted for Aβ in the entire PET-sample (n=165)

Table S9: Cross-sectional interactions between LC intensity_r and Aβ on cognitive performance in the PET-sample (n=165)

Table S10: Cross-sectional associations between LC intensity_r and cognitive performance in the CU-individuals of the PET-sample (n=149)

Table S11: Cross-sectional associations between LC tangle density and cognitive performance in the MAP-dataset (n=160)

Table S12: Cross-sectional associations between LC tangle density and cognitive performance adjusted for Aβ in the MAP-dataset (n=160)

Table S13: Cross-sectional associations between LC tangle density and cognitive performance in the CU cases of the MAP-dataset (n=66)

Table S14: Cross-sectional associations between LC hypopigmentation and cognitive performance, in the ROSMAP-dataset (n=1065)

Table S15: Cross-sectional associations between LC hypopigmentation and cognitive performance, in the CU of the ROSMAP-dataset (n=421)

Table S16: Clusters of the vertex-wise TAU mediation of the relationship between LC intensity_r and memory

Table S17: Associations between LC intensity_r and retrospective cognitive decline in the PET-sample

Table S18: Associations between LC tangle density and retrospective cognitive decline in the MAP-dataset

Table S19: Associations between LC hypopigmentation and retrospective cognitive decline in the ROSMAP-dataset

Table S20: Frequency of Thal and Braak stages across global CDR scores in the selected NACC-dataset

Data and materials availability

The Harvard Aging Brain Study project is committed to publicly releasing its data. Baseline data is already available online at http://nmr.mgh.harvard.edu/lab/harvardagingbrain/data. Follow-up data of the Harvard Aging Brain Study data, including the data used this manuscript, will be publicly to the research community, and data till year 5 is currently available by request, pending approval of a data request and agreement to abide by the Harvard Aging Brain Study online data use agreement. Data from ROSMAP is available upon request at https://www.radc.rush.edu and data from the NACC is available on request at https://www.naccdata.org/.