Atrophy links lower novelty‐related locus coeruleus connectivity to cognitive decline in preclinical AD

Abstract

INTRODUCTION

Animal research has shown that tau pathology in the locus coeruleus (LC) is associated with reduced norepinephrine signaling, lower projection density to the medial temporal lobe (MTL), atrophy, and cognitive impairment. We investigated the contribution of LC‐MTL functional connectivity (FC_LC‐MTL) on cortical atrophy across Braak stage regions and its impact on cognition.

METHODS

We analyzed functional magnetic resonance imaging and amyloid beta (Aβ) positron emission tomography data from 128 cognitively normal participants, associating novelty‐related FC_LC‐MTL with longitudinal atrophy and cognition with and without Aβ moderation.

RESULTS

Cross‐sectionally, lower FC_LC‐MTL was associated with atrophy in Braak stage II regions. Longitudinally, atrophy in Braak stage 2 to 4 regions related to lower baseline FC_LC‐MTL at elevated levels of Aβ, but not to other regions. Atrophy in Braak stage 2 regions mediated the relation between FC_LC‐MTL and subsequent cognitive decline.

DISCUSSION

FC_LC‐MTL is implicated in Aβ‐related cortical atrophy, suggesting that LC‐MTL connectivity could confer neuroprotective effects in preclinical AD.

Highlights

• Novelty‐related functional magnetic resonance imaging (fMRI) LC‐medial temporal lobe (MTL) connectivity links to longitudinal Aβ‐dependent atrophy.

• This relationship extended to higher Braak stage regions with increasing Aβ burden.

• Longitudinal MTL atrophy mediated the LC‐MTL connectivity–cognition relationship.

• Our findings mirror the animal data on MTL atrophy following NE signal dysfunction.

1. BACKGROUND

The neuropathology of Alzheimer's disease (AD) is characterized by aggregated misfolded proteins of amyloid beta (Aβ) and tau, contributing to neurodegeneration and cognitive impairment. The topography of these pathologies follows an almost predictable pattern throughout disease progression, starting two to three decades before the emergence of cognitive impairment. Autopsy studies of the locus coeruleus (LC), a small nucleus in the pontine tegmentum providing norepinephrine (NE) to the brain, reported that hyperphosphorylated tau starts to accumulate in the LC before tau pathology can be detected in allo‐ and neocortical areas and before deposition of fibrillar Aβ in the neocortex. ¹ , ²

The noradrenergic fibers of the LC project densely to all cortical regions, modulating memory, ³ , ⁴ novelty detection, ⁵ and attention reorientation. ⁶ , ⁷ Importantly, recent post mortem work and animal data showed that tau pathology in the LC was associated with axonal degeneration of LC neurons before loss of cell bodies occurs. ⁸ , ⁹ These animal studies reported that early reductions in hippocampal NE levels following either LC tau accumulation or LC lesions were associated with later reductions in fiber density projecting to the medial temporal lobe (MTL), hippocampal neurodegeneration, and impaired learning. ⁸ , ¹⁰ These degenerative processes in rodents were most pronounced at age 16 months, when detectable Aβ was substantially present in the MTL, suggesting that the emergence of Aβ can aggravate degeneration in the setting of lower NE release. Given the early reduction in NE levels in these animal AD models, we posit that a putative mechanism for these degenerative processes may include tau‐induced alterations in LC firing patterns.

Typically, LC neurons fire in two modes; tonic activity reflects arousal and behavioral flexibility, while phasic activity has been associated with goal‐directed behavior, novel or salient stimuli, and better cognitive performance. ¹¹ While NE is released in LC target regions during both modes, phasic bursts have been linked to an increased release. ¹² While chemogenetic applications like designer receptors exclusively activated by designer drugs (DREADDs) lack precise temporal control of neuronal activity, DREADD‐induced activation of LC neurons salvaged spatial learning in TgF344‐AD rats with tau and Aβ pathology. ⁸ Recent optogenetics work, providing better temporal specificity, stimulated LC neurons in rats infused with human pretangle tau at either tonic or phasic frequency and provided evidence that phasic activity protected LC fiber density as well as spatial learning performance. ¹³

To date, neuroimaging methods have not been able to directly measure tonic or phasic LC activity. However, given that novelty tasks consistently evoke phasic LC activity in animal studies, ⁵ , ¹⁴ , ¹⁵ such paradigms can be used cautiously to infer information on phasic activity. Similar to the aforementioned animal studies, we previously showed in asymptomatic (Clinical Dementia Rating = 0) older individuals that lower novelty‐related LC connectivity with the MTL was associated with Aβ‐related cognitive decline. ¹⁶ Guided by AD rat models showing that reduced projections from the LC to the MTL regions were associated with greater MTL atrophy and impaired learning, and the fact that the LC widely projects to cortical regions, we aimed to leverage and extend our previous work and hypothesized that lower functional connectivity (FC) between the LC and MTL would be associated with cortical atrophy. Consistent with animal findings, we expect these associations to strengthen as a function of Aβ‐positron emission tomography (PET) levels.

To evaluate this hypothesis, we used data from a subset of 128 well‐characterized participants of the Harvard Aging Brain Study (HABS) and specifically tested the relationship between LC‐MTL FC during a novel and repeated face–name matching task and longitudinal atrophy in regions grouped by Braak stage. Furthermore, we examined whether atrophy mediated the association between novelty‐related LC‐MTL FC and consecutive cognitive decline. Given the early involvement of the LC in AD's pathophysiology, mapping LC‐related cortical neurodegeneration can help identify the earliest stages of AD and facilitate the development of disease‐modifying interventions at a stage where irreversible brain changes are not yet too widespread.

2. METHODS

2.1. Demographics

The 128 people included in this analysis were selected from the HABS, which is a longitudinal observational study investigating aging and preclinical Alzheimer's dementia at the Massachusetts General Hospital (MGH) in Boston, MA, USA. ¹⁷ Participants underwent Pittsburgh compound B (PiB) PET and task‐based functional magnetic resonance imaging (fMRI) at baseline, structural MRI on average every 2.35 ± 1.07 years, and cognitive assessment on average every 0.92 ± 0.23 years. Our analysis sample subset was chosen based on having completed the task‐based fMRI paradigm and a PiB PET scan within 1 year of each other. All participants had to be cognitively unimpaired at enrollment, as determined by the Logical Memory Delayed Recall test, ¹⁸ Mini‐Mental State Examination (MMSE), ¹⁹ and Clinical Dementia Rating (CDR) scale ²⁰ administered by study‐associated physicians. Exclusion criteria were a history of alcoholism, drug abuse, head trauma, current serious medical/psychiatric illness, or a score of more than 10 on the Geriatric Depression Scale. ²¹ A description of the demographic information can be found in Sample characteristics and Table 1. The study was approved by the MGH Partners Human Research Committee and complied with the Declaration of Helsinki. ²² All participants gave their informed consent in writing and received monetary compensation for their participation.

TABLE 1.

Sample demographics at baseline. The full sample was used for cross‐sectional and longitudinal analyses.

Characteristic	Full sample, N = 128 a	Mediation sample, N = 71 a	p value b
Age	69.75 (63.75, 76.81)	71.00 (66.75, 77.00)	0.2
Years of education	16 (14, 18)	16 (12, 18)	0.7
PACC5 z‐score	0.27 (−0.19, 0.56)	0.27 (−0.20, 0.55)	>0.9
MMSE score	29.00 (29.00, 30.00)	29.00 (29.00, 30.00)	0.9
Sex			0.8
F	71 (55%)	41 (58%)
M	57 (45%)	30 (42%)
PiB‐PET group			0.5
Positive	36 (28%)	23 (32%)
Negative	92 (72%)	48 (68%)
CDR
0	128 (100%)	71 (100%)
Follow‐up time MRI (years)	5.0 (3.0, 8.0)	7.6 (5.0, 10.8)	<0.001
Follow‐up time NP (years)	6.83 (4.65, 10.82)	10.49 (8.11, 11.47)	<0.001

Note: The reduced sample was used for the mediation analysis. Participants were grouped based on their PiB‐PET measurements at a threshold of 1.324 DVR PVC (FLR).

Abbreviations: CDR, clinical dementia rating; F, female; M, male, MMSE, Mini‐Mental State Examination; MRI, magnetic resonance imaging; NP, neuropsychological testing; PACC5, preclinical Alzheimer's cognitive composite 5; PET, positron emission tomography; PiB, Pittsburgh compound B.

^a Median (IQR); n (%).

^b Wilcoxon rank sum test; Pearson's chi‐squared test.

RESEARCH IN CONTEXT

1. Systematic review: Literature database searches revealed a small body of animal literature linking early Alzheimer's disease (AD) pathology to dysregulated norepinephrine (NE) signaling from the locus coeruleus (LC), hippocampal atrophy, and cognitive impairment. In humans, the LC is one of the earliest sites accumulating AD pathology. We examined the relationship between LC functional connectivity with mediotemporal structures, cortical atrophy, and cognition in human preclinical AD.

2. Interpretation: Our findings corroborated previous animal reports by showing associations between novelty‐related LC‐medial temporal lobe (MTL) functional connectivity and longitudinal amyloid beta (Aβ)‐dependent atrophy. Further, we found that Braak stage 2 atrophy may be part of the underlying mechanisms relating lower LC‐MTL functional connectivity to cognitive decline.

3. Future directions: These findings expand our knowledge of the importance of LC function in early AD‐related atrophy. Future studies should explore the role of tau pathology in the observed relations and investigate whether NE‐targeted interventions can slow down AD‐related atrophy.

2.2. Longitudinal cognitive evaluation

The Preclinical Alzheimer's Cognitive Composite score (PACC5), a composite of cognitive tests, was designed to detect early, Aβ‐related cognitive decline. ²³ , ²⁴ The test score is an average aggregate of the Digit Symbol Substitution Test (DSST), ²⁵ free and total recall elements of the Free and Cued Selective Reminding Test (FCSRT), ²⁶ the Wechsler Memory Scale‐Revised (WMS‐R) Logical Memory Delayed Recall (LMDR), ¹⁸ the Category Fluency Test (CAT) for animals, fruits, and vegetables, ²⁷ and the MMSE, ¹⁹ all z‐transformed using the baseline mean and standard deviation. PACC5 was not computed if more than one subtest was missing. A median of seven PACC5 evaluations per participant (interquartile range [IQR] = [5, 10]) were available in total (see Table S1 for a detailed list).

2.3. Face–name paradigm

For task‐based fMRI, participants were asked to do a face–name association task by memorizing a total of 86 face–name pairs on a screen in the scanner bore. ²⁸ Faces, which varied in age, sex, and ethnicity, were shown in color, and names were displayed in white against a black background. The total task was split up into six functional runs, each consisting of two 39.8‐s blocks of novelty and repetition, respectively (Figure 1). Between face–name blocks a fixation cross was shown for 25 s. Each block consisted of seven face–name pairs to memorize with jittered fixation in between. Participants familiarized themselves with the task (two faces/names, one male and one female) outside of the scanner. These two faces/names were also shown in alternation during any repetition block inside the scanner. The novelty blocks consisted entirely of previously unseen faces and names. To increase task engagement, participants were asked to indicate via a button press whether they thought the face matched the name, while emphasizing the full subjectivity of this task. ²⁹

FIGURE 1.

Overview of fMRI task. Participants performed a total of six fMRI runs, each consisting of four blocks of seven faces: two blocks of novel faces and two blocks of repeated faces. Each face and name in any novelty block across the six runs only appeared once. During all repeated blocks across the six runs, the same two faces were shown. Participants encountered both faces of the repeated block before and during the familiarization with the task outside the scanner. Blocks of faces were separated by 25 s of a blank screen with a fixation cross.

2.4. PET acquisition and processing

Aβ PET imaging was performed after the injection of 8.5 to 15 mCi (11)C‐labeled PiB ³⁰ on a Siemens ECAT EXACT HR+ (Siemens Healthineers, Erlangen, Germany) in 69 frames over the span of 60 min (12 frames × 15 s, 57 frames × 60 s; 3D mode; 63 image planes; 15.2 cm axial field of view; 5.6 mm transaxial resolution; 2.4 mm slice interval). The median time span between the PiB PET imaging session and the baseline neuropsychological evaluation was 97 days (IQR = [70, 128]) and 70 days (IQR = [22, 119]) between PiB PET imaging and the baseline MRI. The PiB PET images were converted to distribution volume ratio (DVR) via the Logan graphical method ³¹ and using cerebellar gray as reference region. Further, images were aligned and corrected for subject motion using co‐registration to the high‐resolution anatomical T1 MRI scans. Geometric transfer matrix partial volume correction (PVC) was applied in FreeSurfer, ³² and cortical PiB retention was calculated in the FLR regions (frontal, lateral parietal and lateral temporal, retrosplenial cortices). ³³ , ³⁴ The FLR PIB DVR PVC was also converted to the Centiloid (CL) scale ³⁵ using a previously published linear transformation. ³⁶ For dichotomization into PiB+ and PIB− groups, we used a cut‐off value of 1.324 DVR PVC (18.49 CL), previously validated on the entire cohort data. ¹⁷

2.5. MRI acquisition and processing

MRI data were collected on a Siemens MAGNETOM Trio, A Tim System, software version syngo MR B17 (Siemens Healthineers, Erlangen, Germany) with a 12‐channel phased‐array head coil. T1‐weighted Magnetization Prepared Rapid Acquisition Gradient Echo (MPRAGE) sequences provided high‐resolution anatomical images for thickness and volume estimates. The sequence parameters were as follows: repetition/echo time (TR/TE) = 2300/2.95 ms; voxel size = 1.1 × 1.1 × 1.2 mm; inversion time = 900 ms; flip angle = 9°; acquisition matrix = 270 × 254 × 212 mm; 176 sagittal slices; 2× acceleration using GeneRalized Autocalibrating Partial Parallel Acquisition (GRAPPA). Functional BOLD data were acquired with a T2*‐weighted echo planar imaging (EPI) sequence: TR/TE = 2000/30 ms; voxel size = 3.1 × 3.1 × 5.0 mm; flip angle = 90°; acquisition matrix = 200 × 200 × 179 mm; 30 slices; 1 mm slice spacing; number of volumes = 127. Slice orientation was chosen to be coronal; perpendicular to the anterior commissure–posterior commissure (AC–PC) line to maximize in‐plane resolution in the hippocampus and brainstem.

2.5.1. Structural MRI processing

The anatomical T1 structural scans were loaded into FreeSurfer version 6.0 ³⁷ for segmentation and parcellation. The work flow consisted of conversion to isotropic 1‐mm voxel size, motion correction, transformation into Talairach space, intensity normalization, skull stripping, segregation of left and right hemispheres, removal of brainstem and cerebellum, correction of topology defects, segmentation of white matter, gray matter, and cerebrospinal fluid (CSF), and parcellation of cortical and subcortical gray matter regions using Desikan–Killiany atlas‐based automated heuristics. ³⁸ Cortical thickness values as well as hippocampus and amygdala volume were then extracted and bilaterally averaged across Desikan–Killiany regions representative of the Braak stages. ³⁸ , ³⁹ Hippocampus and amygdala volumes were residualized for intracranial volume. All thickness and volume measurements were z‐scored using their respective baseline distributions.

We formed thickness–volume (THV) aggregates based on Braak staging, ¹ , ⁴⁰ denoted by THV₂ to THV₆ (Table S2), by averaging the bilateral, z‐scored volume, and cortical thickness values. Braak stage 1 was omitted since FreeSurfer does not include a transentorhinal cortex segmentation.

2.5.2. Functional MRI processing

EPI BOLD images were preprocessed with the FMRIB Software Library (FSL, version 5.0.7, ⁴¹ ) performing brain extraction, slice‐timing correction, motion realignment of volumes, intensity normalization, independent component analysis (ICA) for automatic removal of motion artifacts (ICA‐AROMA), ⁴² and scrubbing of outlier volumes (4.67 ± 3.58%). ⁴³ , ⁴⁴ To minimize the influence of motion and physiological processes (CSF flow, respiration, cardiac pulse), we regressed out the signal from the lateral and fourth ventricles, white matter, previously extracted motion parameters, and their derivatives, as well as the squares of all previously mentioned time‐series. The resulting images were smoothed with a z‐direction ellipsoid Gaussian kernel (full width at half maximum = 4.2 × 4.2 × 13.2 mm) to preserve the signal in the elongated structure of the LC. ¹⁶

Localization of the LC in the fMRI space was performed by registering a pre‐existing post mortem‐evaluated template in Montreal Neurological Institude space to the native functional spaces of each run via the subject‐specific T1 images and a functional population‐specific T2* template space generated by the ANTs multivariate template construction tool (version 2.1.0) using the default number of iterations (30 × 50 × 20), default iteration limit of 4, symmetric normalization (Greedy‐SyN; ‐t GR), and cross‐correlation (‐c CC) metrics to guide the mapping. ⁴⁵ , ⁴⁶ The intermediary LC mask in the functional template space was chosen to be the area with the highest probabilistic overlap across the subject‐specific registrations from the T1 images (for a detailed flowchart of the methodology see Prokopiou et al., ¹⁶ Figure S1). Registration to the 2‐mm³ MNI‐152 space was performed with affine registration to the subject‐specific T1 images and subsequent non‐linear registration to the MNI space with increased weights for the brainstem (see Prokopiou et al. ¹⁶ for details). The robustness of this approach was demonstrated previously by replicating the functional connectivity analysis with unsmoothed data and an eroded LC mask (see the supplementary material of Prokopiou et al. ¹⁶ ).

2.6. Functional connectivity analysis

FC differences between novel and familiar faces were computed with generalized psychophysiological interaction (gPPI) analysis. We determined the novelty versus repeated faces FC between the LC and multiple brain areas involved in arousal and novelty: amygdala, hippocampus, insular cortex, parahippocampal gyrus, entorhinal cortex, and fusiform cortex. ¹⁶ , ⁴⁷ To account for the regional variability of the hemodynamic response function (HRF) in the LC, two parallel block‐cascade linear, finite impulse response models along with a spherical Laguerre basis of orthonormal basis functions was used to approximate the BOLD response to the novelty and repetition condition. ¹⁶ Five regressors were fitted for the gPPI analysis using a generalized linear model (GLM) with ordinary least‐squares optimization: two task blocks for novelty and repetition convolved with the subject‐specific LC hemodynamic response function (psychological terms), the BOLD time series in the LC seed region (physiological term), and the two first regressors element‐wise multiplied by the seed time series (interaction terms). After running gPPI for each subject individually, a group‐level analysis with linear mixed‐effect models (fixed effects: baseline age, sex; random intercept: subject; random slope: fMRI run 1 to 6) was performed to identify brain regions with significant task‐dependent FC with the LC. Cluster‐based multiple comparison correction was performed using the Gaussian random field theory option to determine significance levels at a z‐threshold of 4. Consistent with our previous work, we found clusters restricted to the bilateral MTL. FC between the LC and the MTL cluster (FC_LC‐MTL) was computed by calculating the mean contrast value (novelty minus repetition) across all voxels in the significant clusters.

To establish the specificity of our LC analyses, we also performed a voxel‐wise gPPI analysis targeting the whole brain with the MTL cluster as the seed region, using a cluster correction z‐threshold of 4. The clusters were grouped in a positive network (Net+) and a negative network (Net−), based on their mean contrast values. The average FC value in each network was then used for the specificity analysis.

2.7. Statistical modeling

All statistical analyses were conducted in R version 4.2.2. Baseline, cross‐sectional associations between FC_LC‐MTL, THV, and PACC5 were evaluated using linear models, correcting for age, sex, and education. We report regression coefficients and type II ANOVA derived p values.

Linear mixed‐effect models (LMMs) were used to examine relationships between baseline FC_LC‐MTL and longitudinal THV (outcome) with time as random slope and individual as random intercept, adjusting for baseline age, sex, and education, using maximum likelihood estimation. All predictor variables were interacted with time. Further, we expanded these LMMs to include three‐way interactions between baseline FC_LC‐MTL, PiB, and time. For all LMMs we report regression coefficients and p values derived from type II ANOVAs using the Kenward–Roger method of estimating degrees of freedom.

Additionally, we performed Johnson–Neyman analyses with false discovery rate (FDR)‐adjusted significance intervals to determine the range of PiB values where the relationship between FC_LC‐MTL and THV‐changes over time significantly differed from zero.

We also checked whether the effect of FC_LC‐MTL on THV differed across Braak stage regions for our baseline and longitudinal models by treating region as a repeated, within‐subject variable in our LMMs. Comparisons between Braak stage regions were tested using marginal means. We further used mediation analysis to examine whether changes in THV mediated the association between FC_LC‐MTL and cognitive decline. To ensure temporal separation of the rates of change of THV and PACC5 while retaining the largest number of participants, THV slopes were computed using data from year 0 to 5.5 using LMMs with random slope (time) and intercept (subject), while PACC5 slopes were computed analogously using data from year 5.5 to 13, with no overlap between them – all having baseline age, sex, and education as covariates. We then performed a bootstrapped mediation analysis (N = 10,000) determining the indirect effect of THV slopes on the relation between FC_LC‐MTL and PACC5 slopes. Finally, to test whether the mediation was contingent on PiB, we ran a bootstrapped moderated mediation analysis (N = 10,000) including interactions with baseline PiB. In a control analysis, we reversed the mediator and the outcome variable, testing whether PACC5 slopes would mediate the relationship between LC‐MTL FC and THV atrophy. Due to limited MRI data availability beyond year 5.5, PACC slopes were calculated up to year 3 for the control analysis, allowing for 51 participants to be included.

Unless stated otherwise, tests were two‐sided and significance thresholds set to p < 0.05. We additionally report FDR‐adjusted p values (P_adj ) to correct for multiple comparisons across Braak stage aggregates. Number of participants is abbreviated N and the number of observations N_obs. Effect sizes are reported as partial Cohen's f² scores, including their 95% confidence intervals.

3. RESULTS

3.1. Sample characteristics

The 128 participants included in this study had a median of two MRI follow‐ups (IQR = [1, 3]) and a median of six follow‐up sessions of neuropsychological (NP) testing (IQR = [4, 9]). Participants’ median age was 69.75 years (IQR = [63.70, 76.81]), 55.5% were female (71 F, 57 M), and they were predominantly White (White 88.3%, Black 9.4%, Native American 1.5%, Asian 0.8%) and well educated (16 years, IQR = [14, 18]). All participants were cognitively unimpaired at baseline: MMSE = median 29 (IQR = [29, 30]), CDR = 0; however, 24% progressed to CDR = 0.5 and 3% to CDR = 2 over the span of the study. Baseline PiB was elevated (PiB+) in 28.2%. As we required separation of time windows for the atrophy and cognition slopes in the mediation analyses, the mediation subset (N = 71) consisted of participants with baseline median age 71 years (IQR = [66.75, 77.00]), 57.7% female (41F, 30 M), predominantly White (White 91.5%, Black 7%, Native American 1.5%), well educated (median 16 years, IQR = [12, 18]), 32.4% PiB+, and all cognitively unimpaired at baseline (median MMSE = 29, IQR = [29, 30], CDR = 0). No significant differences in population characteristics existed between the full and mediation sample. Follow‐up times for MRI and NP differed between the full and mediation sample since we had to remove participants without enough longitudinal data points to pull slopes for the mediation analysis. All demographics are summarized in Table 1.

3.2. Cross‐sectional FC_LC‐MTL is associated with Braak stage 2 region atrophy and cognition

Lower baseline FC_LC‐MTL levels were related to smaller THV₂ values (N_obs = 128; β = 0.15, 95% CI = [0.05, 0.26]; p = 0.005, P_adj  = 0.026, f² = 0.07, 95% CI_f2 = [0.01, 0.19]; Figure 2A), but not for any of the other Braak stage aggregates (Table S3; Figure S1). Baseline PACC5 scores correlated positively with baseline FC_LC‐MTL levels (N_obs = 128; β = 0.1, 95% CI = [0.002, 0.19]; p = 0.045; f² = 0.03, 95% CI_f2 = [0, 0.13]; Figure 2B). Baseline PACC5 scores were not associated with any THV aggregate values (P_adj  > 0.05 exemplary THV₂ in Figure 2C; Table S3; Figure S2) or with baseline PiB (p = 0.32). Including interactions between FC_LC‐MTL and baseline PiB did not yield significant associations. (Table S4). At baseline, the association between FC_LC‐MTL and THV was significantly different between the Braak stage regions 2 to 4, 2 to 5, and 2 to 6 (Table S5).

FIGURE 2.

Cross‐sectional relations. (A) Higher FC_LC‐MTL was related to larger THV at baseline only for Braak stage 2 regions, but not others. (B) Higher FC_LC‐MTL was also correlated with higher PACC5 scores at baseline. (C) No association between larger PACC5 scores and larger THV could be established for any Braak stage region. Exemplarily THV₂ is shown here. Shaded areas indicate 95% confidence intervals of the predictions. All models included age, sex, and education as covariates. a.u., arbitrary units; FC_LC‐MTL, novelty‐related functional connectivity between the locus coeruleus and the medial temporal lobe; PACC5, preclinical Alzheimer's cognitive composite 5; THV, thickness–volume aggregate by Braak stage.

3.3. Baseline FC_LC‐MTL is associated with MTL atrophy in the context of elevated PiB

The longitudinal changes in THV values for each Braak stage with respect to baseline FC_LC‐MTL are shown in Figure S3. We found no relationship between baseline FC_LC‐MTL and longitudinal changes in THV for any of the Braak stage regions (Table S6, Figure S4). Controlling for longitudinal CDR did not change the results (Table S7).

Including baseline PiB in a three‐way interaction showed that particularly at higher PiB values, lower FC_LC‐MTL was associated with faster atrophy in THV₂ (N_obs = 394; β = 0.046, 95% CI = [0.008, 0.084]; p = 0.025, P_adj  = 0.041; f² = 0.07, 95% CI_f2 = [0, 0.25]; Figure 3A), THV₃ (N_obs = 394; β = 0.058, 95% CI = [0.027, 0.091]; p = 0.0009, P_adj  = 0.005; f² = 0.15, 95% CI_f2 = [0.03, 0.39]; Figure 3B), and THV₄ (N_obs = 394; β = 0.049, 95% CI = [0.016, 0.083]; p = 0.007, P_adj  = 0.017; f² = 0.1, 95% CI_f2 = [0.01, 0.29]; Figure 3C). This was not the case for THV₅ and THV₆ (Table S8, Figure S5). Johnson–Neyman analysis revealed that the association between FC_LC‐MTL and THV₂ was significant starting in year 0 (= baseline) from PiB = 1.2 DVR PVC (9.7 CL). For THV₃, the association was significant from year 3 at a PiB value of 1.39 DVR PVC (23.19 CL) and for THV₄ from year 8 at a PiB value of 1.65 DVR PVC (41.69 CL) (Figure 4). Including longitudinal CDR as covariate in the model did not change these results (Table S9). The moderating effect of PiB on the relationship between FC_LC‐MTL and THV over time was significantly different between the Braak stage regions 2 to 4 and 2 to 5 (Table S10). Higher baseline PiB without the interaction with FC_LC‐MTL was also associated with faster PACC5 decline (Nobs = 931; β = −0.21, 95% CI = [−0.28, −0.14]; p < 0.001, f2 = 0.35, 95% CIf2 = [0.15, 0.63]).

FIGURE 3.

(A–C) PiB‐dependent longitudinal associations between THV₂ to THV₄and baseline FC_LC‐MTL. The analysis was done with continuous PiB values, but for plotting the three‐way interaction between baseline FC_LC‐MTL, baseline PiB, and time, marginalized effects were plotted at three different PiB levels. The PiB DVR PVC value of 1.324 is the cut‐off between PiB− and PiB+. The PiB DVR PVC values 1.10 and 1.54 were the mean values of the PiB− and PiB+ group, respectively. Shaded areas indicate 95% confidence interval around predictions. a.u., arbitrary units; DVR, distribution volume ratio; FC_LC‐MTL, novelty‐related functional connectivity between the locus coeruleus and the medial temporal lobe; PiB, Pittsburgh compound B; PVC, partial volume corrected; THV, thickness–volume aggregate by Braak stage.

FIGURE 4.

Johnson–Neyman plots for FC_LC‐MTL:PiB interaction for THV₂ to THV₄ at three time points: 0, 3, and 8 years. All other model covariates are centered to their mean. Green vertical dashed lines and green sections indicate areas of PiB where the association between baseline FC_LC‐MTL and atrophy was significant. Shaded areas depict the 95% confidence interval around the model fit. From baseline (year 0) on, FC_LC‐MTL was associated with longitudinal atrophy in THV₂ for PiB values between 1.2 and 2.15 DVR PVC. From year 3 on, FC_LC‐MTL was also associated with longitudinal atrophy in THV₃ for PiB values of 1.39 DVR PVC and above. Lastly, from year 8 on, FC_LC‐MTL was related to longitudinal atrophy in THV₄ from PiB values higher than 1.65 DVR PVC. The effect of baseline FC_LC‐MTL on atrophy becomes significant later for increasing Braak stage regions and at the same time requires higher PiB values. a.u., arbitrary units; DVR, distribution volume ratio; FC_LC‐MTL, novelty‐related functional connectivity between the locus coeruleus and the medial temporal lobe; n.s., not significant; PiB, Pittsburgh compound B; PVC, partial volume corrected; SD, standard deviation; THV, thickness–volume aggregate by Braak stage.

3.4. Rates of Braak stage 2 atrophy mediate the association between baseline FC_LC‐MTL and rates of cognitive decline

Mediation analysis, using THV slopes (year 0 to 5.5) as mediator between baseline FC_LC‐MTL and later PACC5 slopes (year 5.5 to 13) revealed that the relationship between lower baseline FC_LC‐MTL and faster PACC5 decline was mediated by faster THV₂ atrophy (Figure 5A): mediation effect: β = 0.038, p = 0.026, 95% CI = [0.003, 0.09]; direct effect: β = 0.059, p = 0.032, 95% CI = [0.04, 0.12]. This effect was moderated by baseline PiB continuously, indicating that mediation by atrophy on the FC_LC‐MTL‐PACC5 relationship was more pronounced in individuals with higher PiB relative to those with lower PiB.

FIGURE 5 (.

A) Partial mediation effect of THV₂ slopes (year 0 to 5.5) on relation between baseline FC_LC‐MT and PACC5 slopes (year 5.5 to 13). (B) The mediation by steeper THV₂ atrophy on the relationship between lower FC_LC‐MTL and faster cognitive decline is stronger in individuals with PiB > 1.20 DVR PVC (PIB↑) compared to those with PiB < 1.20 DVR PVC (PIB↓). Asterisks behind numbers denote ranges of p values: *p < 0.05, **p < 0.01, ***p < 0.001. B, model coefficient; DVR, distribution volume ratio; CI, confidence interval; DE, direct effect; FC_LC‐MTL, novelty‐related functional connectivity between locus coeruleus and medial temporal lobe; ME, mediation effect; PACC5, preclinical Alzheimer's cognitive composite 5; PiB, Pittsburgh compound B; PVC, partial volume corrected; THV, thickness–volume aggregate.

Splitting our sample at the previously found Johnson–Neyman PiB cut‐off value 1.2 DVR PVC yielded 35 people (49.3%) above this cut‐off and 36 people (50.7%) below the cut‐off. Similarly, mediation by THV₂‐atrophy on the FC‐PACC5 relationship was stronger in individuals above the 1.2 DVR PVC cut‐off compared to those below the cut‐off (Figure 5B): mediation effects: β(PiB↑) − β(PiB↓) = 0.032, p = 0.036, 95% CI = [0.002, 0.07]; direct effects: β(PiB↑) − β(PiB↓) = 0.079, p = 0.23, 95% CI = [−0.06, 0.18]. Lastly, THV₃ to THV₆ did not mediate the relationship between baseline FC_LC‐MTL and PACC5 decline (Table S11). Reversing THV₂ atrophy and PACC5 in the mediation model did not yield any significant mediation effect, with and without interaction by baseline PiB (Figure S6).

3.5. Longitudinal FC‐atrophy relationship is specific to LC

In the voxel‐wise whole‐brain‐specificity analysis of novelty‐related FC with the MTL, two networks of significant clusters emerged: a positive network (FC_MTL‐Net+), including bilateral clusters in the anterior cingulate cortex, supramarginal gyrus, the parietal operculum, the cuneus, the precuneus, the posterior cingulate gyrus, and the right postcentral gyrus, and a negative network (FC_MTL‐Net−), including clusters in the bilateral thalamus, middle frontal and inferior frontal gyrus, the paracingulate gyrus, the superior frontal gyrus, and the right cerebellum (Figure S7).

FC_LC‐MTL was correlated with neither FC_MTL‐Net+ (N_obs = 128, β = 0.006, p = 0.95) nor FC_MTL‐Net− (N_obs = 128, β = 0.057, p = 0.52). At baseline, in a linear model controlling for baseline age, sex, and years of education, neither FC_MTL‐Net+ nor FC_MTL‐Net− was correlated with THV values of any Braak stage (Table S12). There were also no longitudinal associations between FC_MTL‐Net+ or FC_MTL‐Net− and THV of any stage, with or without interaction with baseline PiB (Table S13 & S14).

4. DISCUSSION

The LC is one of the earliest regions in the brain to accumulate hyperphosphorylated tau, which affects its projections to the cortex through which the LC exerts its modulatory effect on behavior. Animal studies in LC‐tau models reported reduced NE signaling and reductions in LC projections associated with hippocampal neurodegeneration and learning impairment. Combining a novelty‐related fMRI paradigm with longitudinal cortical thickness measures and cognitive data, we established for the first time in humans that lower novelty‐related LC‐MTL connectivity was associated with faster Aβ‐related atrophy. Interestingly, the association between lower LC‐MTL connectivity and greater atrophy followed a gradual pattern, where atrophy in higher Braak stage‐related regions was dependent on gradually higher Aβ levels. Finally, we demonstrated that Braak stage 2‐related atrophy mediated the relationship between lower novelty‐related LC‐MTL connectivity and subsequent cognitive decline, starting at subthreshold Aβ values. Neurodegeneration, measured as atrophy with MRI, is an important topographic but non‐specific marker of dementia. Our findings highlight that functional measures of the LC can enhance our understanding of early AD‐related atrophy, subsequently facilitating early identification, disease progression tracking, and stratification of patients with limited neurodegeneration into clinical trials.

Even though interest in the LC in AD has significantly increased in the last decade, there is little in vivo work relating LC properties to cortical atrophy. Two recent studies reported different relationships between LC MRI intensity and cortical thickness, ranging from no relationship to predominantly frontoparietal associations. ⁴⁸ , ⁴⁹ While a direct comparison is not possible, as these studies did not include information on possible underlying AD pathology, it is likely that LC MRI intensity‐ and novelty‐related LC FC reflect different aspects of LC alterations in older individuals. Notably, animal and in vivo human CSF‐based studies have associated dysfunction in NE signaling with increases in neuroinflammation and less clearance of Aβ and cortical atrophy. ⁵⁰ , ⁵¹ Such a dysfunction could stem from reduced NE release from LC neurons as well as impaired NE (re)uptake in the cortex. ⁵² In fact, AD rat models expressing LC tau accumulation or with LC lesions exhibited loss of efferent LC fibers, lower NE levels in the hippocampus, hippocampal atrophy, and memory impairment. ⁸ , ¹⁰ Our data corroborate these findings in humans by showing that lower novelty‐related LC‐MTL connectivity was associated with lower gray matter volume in Braak stage 2 regions (entorhinal cortex and hippocampus) and, in addition, that atrophy in these Braak stage 2 regions over 5.5 years was also associated with subsequent cognitive decline. Since baseline LC‐MTL FC was associated with both baseline and longitudinal cognitive performance but atrophy measures correlated only with longitudinal cognitive performance, it is likely that LC‐MTL FC is more closely linked to initial cognitive changes than atrophy, which occurs as a downstream process. Lower novelty‐related LC‐MTL connectivity has also been linked to faster Aβ‐related cognitive decline. ¹⁶ Here we expanded on this finding by showing that this effect was indeed mediated by the rate of MTL atrophy.

There are at least two, likely related, important mechanisms that can link the LC FC measures through MTL atrophy to cognitive performance. First, activity‐dependent NE release in the hippocampus and activation of its beta‐adrenergic receptors play critical roles in long‐term potentiation, the cellular process underlying learning. ⁵³ Atrophy in the earliest stages of the disease rarely represents neuronal death but rather declining synaptic density and changes in the extracellular matrix, affecting optimal neuronal signaling. ⁵⁴ Second, it is important to distinguish between different firing patterns of LC neuronal populations and their relationship to cognitive performance. In animal work, phasic firing of LC neurons has been associated with processing of novel stimuli, optimized task performance, and preserved LC axonal density. ¹³ As focusing on the novelty aspect of the task could be a proxy for phasic‐related LC connectivity, our findings reiterate the importance of phasic activity. The mechanisms underlying changes in these firing patterns of the LC are not yet clear, although accumulation of tau pathology and, consequently, loss of projections and communication with the MTL are assumed to have a negative impact. ⁵⁵

To determine whether the neurodegeneration patterns were LC‐ or MTL‐specific, we performed a sensitivity analysis with the MTL as the seed region and found FC clusters that overlap with previously described novelty‐related activation maps. ²⁸ , ⁵⁶ , ⁵⁷ , ⁵⁸ Many of the areas in our connectivity maps share known structural and functional connections with the LC. ⁵⁹ , ⁶⁰ The positive FC regions are often co‐activated with the MTL in response to novel stimuli. ⁵⁷ , ⁵⁸ The negative FC network possibly reflected familiarity (repetition condition in our task). ⁶¹ , ⁶² Critically, MTL‐Net+ and MTL‐Net− connectivity was not associated with longitudinal atrophy, nor with cognitive decline.

Cross‐sectionally, LC‐MTL connectivity was only associated with Braak stage 2 atrophy, but in our longitudinal analyses, LC‐MTL connectivity was associated with atrophy in the context of Aβ, but not independently of Aβ. Thus, LC‐MTL connectivity seems to be a necessary, but not sufficient, contributor to initial AD‐related neurodegenerative processes. We also found a dose–response relationship between LC‐related longitudinal atrophy across the Braak stage regions and levels of Aβ burden. The association between novelty‐related LC‐MTL FC and Braak stage 2 atrophy was detected starting at subthreshold Aβ levels, whereas Braak stage 3 and 4 atrophy was only observed at above‐threshold Aβ levels. These findings fit current disease models, in which pathology‐related processes in the MTL can occur without above‐threshold amyloidosis, but progression outside of the MTL requires the presence of Aβ. ³⁶ , ⁶³ Similar associations have been observed when associating LC integrity to cortical tau patterns, suggesting that tau is interacting with Aβ and possibly contributing to these LC‐related atrophy patterns. ⁶⁴ , ⁶⁵ However, it is important to acknowledge that other unmeasured co‐pathologies may also contribute to atrophy, including Lewy bodies, vascular lesions, hippocampal sclerosis, TDP‐43 inclusions, and argyrophillic grain disease. Similarly, we did not examine and therefore cannot exclude potential contributions from other neuromodulatory subcortical systems, including the ventral tegmental area, the basal forebrain, or the raphe nuclei.

Since we find LC‐related Braak stage 2 atrophy occurring at subthreshold Aβ levels, this could imply that non‐fibrillar forms of Aβ play a role in LC‐mediated neurodegeneration. While the role of Aβ in LC functioning in the early stages of AD remains understudied, animal research provides insights into potential mechanisms. These animal studies indicated that oligomeric forms of Aβ in the LC were able to dysregulate LC activity. ⁶⁶ Cortical Aβ, on the other hand, can damage unmyelinated axons projecting from the LC, ⁸ , ⁶⁷ resulting in lower cortical NE release from the LC, removing the neuroprotective effect of NE against Aβ. ⁶⁸ , ⁶⁹ Animal studies also provided evidence for potential interactions between NE signaling, Aβ deposition, and tau hyperphosphorylation by showing that Aβ oligomers can redirect NE signaling to set off a pathogenic GSK3β tau pathway. ⁷⁰

Our results can have important implications for the development of new interventions, indicating that strengthening the connections between the LC and MTL – possibly in combination with antiamyloid therapeutics – may slow down the cascade of AD‐related cognitive decline. In rodent AD models, both chemogenetic upregulation of LC neurons and novelty‐like optogenetic LC stimulation increased performance in a Morris water maze and a spatial and olfactory discrimination task, respectively. ⁸ , ¹³ While modulating LC function in humans is challenging, recent advances have been made with less invasive approaches. Auricular vagus nerve stimulation has been shown to increase salivary alpha amylase (a proxy of peripheral NE release), ⁷¹ and a pharmacological clinical trial with NE reuptake inhibitors demonstrated neuroprotective effects. ⁷²

This study has several limitations. The LC is a tiny structure in the pons, adjacent to the fourth ventricle, and fMRI data are therefore susceptible to partial volume effects and physiological artefacts (respiration, cardiac pulse). As described in detail in Prokopiou et al., ¹⁶ we mitigated these effects to the best of our ability. In the absence of heart rate and respiration monitoring, physiological noise was countered with nuisance regression using white matter, lateral ventricle, and fourth ventricle CSF signals, as well as ICA‐based removal of global motion and physiology components. Second, our focus on the LC, driven by its early accumulation of AD pathology, might overlook contributions of other neuromodulatory systems, for example, the dopaminergic system, which are implicated in novelty processing as well. ⁷³ , ⁷⁴ Third, we found no evidence that atrophy in Braak regions above stage 2 would mediate the relationship between LC‐MTL FC and cognitive decline. While the relationship between hippocampal volume and cognition is generally stronger than that observed with other cortical regions, ⁷⁵ it is likely that capping the window for calculating rates of atrophy at 5.5 years did not leave enough follow‐up time to measure atrophy in later stage regions, as was also suggested by our floodlight analyses. Further, voluntary enrollment in this study resulted in a mostly White and highly educated study population. ⁷⁶ We strongly encourage the replication of these findings in racially different or more diverse cohorts. ⁷⁷ , ⁷⁸ Finally, even though our mediation analysis was set up with non‐overlapping timelines for all variables, causality inferences need to be made cautiously, as it is conceivable that Braak stage 2 atrophy impacts LC‐MTL FC. Lacking longitudinal FC data prevented us from investigating this model in our mediation analyses. This shows the need for cohorts with longitudinal functional and structural LC imaging to better understand the temporal dynamics of different LC measures in the context of AD pathophysiology. Nonetheless, our data provide compelling evidence for a temporal chain of events that mimics animal data, indicating that loss of connectivity between the LC and MTL contributes to morphological and physiological changes in the MTL that are part of AD phenotypes.

To conclude, lower novelty‐related functional LC‐MTL connectivity in preclinical AD was associated with Aβ‐dependent longitudinal atrophy, which in turn mediated cognitive decline. Our findings, which are consistent with animal observations, demonstrate the importance of NE signaling in the neuronal health of limbic regions in preclinical AD. Future studies are required to investigate whether NE‐targeted interventions can slow down AD‐related atrophy.

CONFLICT OF INTEREST STATEMENT

KVP is funded by National Institute on Aging (NIA) grant K23 AG053422‐01 and the Alzheimer's Association and has served as a paid consultant for Biogen. APS has been a paid consultant for Janssen, Biogen, Qynapse, and NervGen. KAJ has served as a paid consultant for Janssen, Genzyme, Novartis, Biogen, Roche, and AC Immune. He is a site coinvestigator for Lilly/Avid and Janssen and receives research support for clinical trials from Eisai, Lilly, and Cerveau. He also received funding from NIH grants R01 EB014894, R21 AG038994, R01 AG026484, R01 AG034556, P50 AG00513421, U19 AG10483, P01 AG036694, R13 AG042201174210, R01 AG027435, and R01 AG037497 and the Alzheimer's Association grant ZEN‐10‐174210. RAS has served as a paid consultant for AC Immune, Alector, Acumen, Bristol Myers Squibb, Genentech, NervGen, Oligomerix, Prothena, Renew, Vigil Neuroscience, Ionis, and Vaxxinity. She receives research support from Eisai and Eli Lilly as part of public‐private partnership clinical trials and also receives research support from the following grants: P01 AG036694, U24 AG057437, R01 AG063689, R01 AG054029, R01 AG053798, GHR Foundation, NIA, Fidelity Biosciences, and the Alzheimer's Association. These relationships have not influenced the content of this manuscript. All other authors report no relevant conflicts. Author disclosures are available in the supporting information.

CONSENT STATEMENT

The study was approved by the MGH Partners Human Research Committee and complied with the Declaration of Helsinki. All participants gave their informed consent in writing and received monetary compensation for their participation.

Supporting information

Supporting Information

Supporting Information

Schneider C, Prokopiou PC, Papp KV, et al. Atrophy links lower novelty‐related locus coeruleus connectivity to cognitive decline in preclinical AD. Alzheimer's Dement. 2024;20:3958–3971. 10.1002/alz.13839

DATA AVAILABILITY STATEMENT

The Harvard Aging Brain Study project is committed to publicly releasing its data. Baseline structural MRI, PiB‐PET, and cognitive follow‐up data until year 5 are publicly available to the research community at http://nmr.mgh.harvard.edu/lab/harvardagingbrain/data. Task‐fMRI data are currently not yet publicly available but will be made available in future releases. Requests for material, currently available raw and processed data for all the datasets used in the study, and correspondence can be addressed to Dr. Sperling. Qualified investigators must abide by the Harvard Aging Brain Study online data use agreement, designed to protect the privacy of our participants.