Introduction

The locus coeruleus (LC), the primary site of norepinephrine neurons in the human brain, is an important site of neurodegeneration in Alzheimer’s disease (AD) [1, 2]. Neuropathological studies have found that the LC is the first brain region to accumulate hyperphosphorylated tau proteins years prior to the onset of cognitive impairment and clinical diagnosis [2,3,4], and postmortem data suggest that degeneration of the LC in AD may be a slow and gradual process that is delayed relative to the early accumulation of LC tau [1]. Although the LC is clearly implicated in AD, challenges studying LC physiology in humans in vivo have limited our understanding of the timing of LC changes and their association with characteristic aspects of AD pathophysiology and clinical features.

Neuromelanin-sensitive MRI (NM-MRI) [5] provides a practical means to overcome this obstacle by using neuroimaging to investigate the integrity of the LC in living human brain. This brief and non-invasive scan yields a high signal contrast in the LC, presumably due to its high concentration of neuromelanin (NM), a paramagnetic pigment [5, 6], although the relative contributions of various components of the signal are still under debate and investigation [7,8,9]. Reduced LC NM-MRI signal is associated with smaller LC volume postmortem [6], loss of norepinephrine terminals in the brain [10], AD diagnosis [7, 11,12,13,14], and CSF amyloid-β levels [15], strongly suggesting low LC NM-MRI signal is indicative of degeneration of norepinephrine LC neurons. While previous studies have demonstrated the utility of NM-MRI in AD [7, 11,12,13,14] and shown correlation to tau burden [8], further multimodal imaging work is needed to determine the contribution of LC degeneration to key features of the illness independent of amyloid-β and tau burden and gray matter atrophy. To support this, the validated [16, 17] radiotracers [18F]AZD4694 [18] (for amyloid-β) and [18F]MK6240 [19] (for tau) allow in vivo AD diagnosis and Braak staging [20,21,22].

Consistent with the known functions of the norepinephrine system, changes in the LC-norepinephrine system have been implicated in cognitive deficits and neuropsychiatric symptoms (NPS) in patients with AD [2, 3] and animal models [2, 23]. NPS are a common and burdensome aspect of AD [24, 25] that often emerge early in the course of the illness [26,27,28], render patients more likely to require residential care [29, 30], and are not easily treatable [29, 31,32,33]. Norepinephrine disturbances correlate to NPS in AD [24, 34,35,36,37] and may have a causal role because symptoms of agitation/aggression [38,39,40] and depression [41] respond to treatment with drugs targeting the norepinephrine system. The nature of LC dysfunction in AD may be complex, and compensatory changes may occur in response to LC degeneration, possibly even leading to hyperactivity in remaining LC neurons [2, 3, 36, 42, 43]. Indeed, cerebrospinal fluid levels of norepinephrine and biosynthetic capacity of norepinephrine (indexed as tyrosine hydroxylase expression) are elevated in AD despite LC degeneration [42,43,44]. These changes may have negative consequences in AD as some types of NPS, including agitated, aggressive, and psychotic symptoms and prescription of neuroleptic agents, have been linked to high or preserved norepinephrine function [34, 36,37,38,39,40, 45] and can respond to norepinephrine system blocking medication [38, 39]. Although no prior NM-MRI studies have investigated NPS in AD, in other populations the NM-MRI signal correlates to behaviors resembling aspects of NPS including depression [46], sleep disturbance [47], and autonomic nervous system function [48].

Similarly to norepinephrine function, cortical pathology, including aggregation of amyloid-β [19, 49] and phosphorylated tau [27, 50, 51], is also linked to NPS severity in AD. Thus, disentangling the pathophysiological correlates of NPS may require simultaneous examination of these different insults to determine their independent contributions to the emergence of NPS. We postulate that NPS reflect an imbalance in specific aspects of AD pathophysiology: integrity of the LC on the one hand and amyloid-β and tau accumulation in the cortex on the other hand. The combined effects of these processes may lead to a disruption in cortical and subcortical regulation of behavior, promoting emergence of NPS. Identifying neuroimaging measures that strongly predict NPS would not only help understand the mechanism of their pathogenesis but would also support the effort to find biomarkers to assess NPS risk, guide prescription of existing treatments or advance trials of novel treatments.

Here we combine these advanced neuroimaging methods with assessment of NPS using the validated Mild Behavioral Impairment Checklist (MBI-C) [52] an instrument that is sensitive and specific in capturing a broad spectrum of NPS in older adults across the cognitive spectrum from cognitively normal older adults through to moderate AD [53, 54]. We hypothesize that LC signal will be reduced in AD but correlate positively to NPS severity.

Materials and methods

Participants and clinical measures

Study participants from the community or outpatients at the McGill University Research Centre for Studies in Aging were enrolled in the Translational Biomarkers of Aging and Dementia (TRIAD) cohort [55], McGill University, Canada. The cohort participants had a detailed clinical assessment, including the Clinical Dementia Rating Scale (CDR) and Mini-Mental State Examination (MMSE; for clinical and demographic description of the sample see Table 1). Cognitively unimpaired participants had no objective cognitive impairment and a CDR score of 0. Mild cognitive impairment (MCI) individuals had subjective and objective cognitive impairment, preserved activities of daily living, and a CDR score of 0.5. Patients with mild-to-moderate sporadic AD dementia had a CDR score between 0.5 and 2, and met the National Institute on Aging and the Alzheimer’s Association criteria for probable AD determined by a physician [56]. Participants were excluded if they had other inadequately treated conditions, active substance abuse, recent head trauma, or major surgery, or if they had MRI/PET safety contraindication. AD patients did not discontinue medications for this study.

Table 1 Clinical and demographic measures.
Full size table

NPS severity was assessed using the MBI-C, http://www.MBItest.org [52]. The participant’s primary informant, most frequently their spouse, completed the MBI‐C. The MBI‐C is composed of 34 questions and subdivided into five domains: decreased drive and motivation, affective dysregulation, impulse dyscontrol (agitation, impulsivity, and abnormal reward salience), social inappropriateness, and abnormal perception/thought content. Each question answered “Yes” is accorded a severity rating (1 = mild, 2 = moderate, or 3 = severe). To be endorsed, symptoms must have emerged later in life and persisted for minimum 6 months continuously or intermittently. The Douglas Institute Research Ethics Board approved this study; all participants provided written informed consent.

MRI acquisition

All neuroimaging data were acquired at the Montreal Neurological Institute (MNI). Magnetic resonance (MR) images were acquired on a 3T Prisma scanner. NM-MRI images were collected via a turbo spin echo sequence with the following parameters: repetition time (TR) = 600 ms; echo time (TE) = 10 ms; flip angle = 120°; turbo factor = 4; in-plane resolution = 0.6875 × 0.6875 mm2; partial brain coverage overlaying the pons and midbrain with field of view = 165 × 220; number of slices = 20; slice thickness = 1.8 mm; number of averages = 7; acquisition time = 8.45 min. Whole-brain, T1-weighted MR images (resolution = 1 mm, isotropic) were acquired using an MPRAGE sequence for preprocessing of the NM-MRI and PET data. Quality of MRI images was visually inspected for artifacts immediately upon acquisition, and scans were repeated when necessary, time permitting. Cortical gray matter volume and estimated total intracranial volume were obtained using FreeSurfer version 6.0 (Martinos Center for Biomedical Imaging) standard segmentation pipeline.

Preprocessing of NM-MRI images

LC signal was measured for the whole LC and LC subregions (rostrocaudal sections) using a semi-automated algorithm incorporating steps similar to those described in previous studies [47, 57]. This method performs an intensity-threshold-free cluster search within an overinclusive mask of the LC in native space. For simplicity we refer to it as a “funnel tip” method, see Fig. 1 for summary of the steps in the algorithm. Although LC signal is measured on native space NM-MRI images, it is necessary to spatially normalize the NM-MRI images in order to register an overinclusive LC mask (referred to as the LC search mask) from MNI space to native space for each participant. Initial preprocessing steps were performed as in our prior work examining NM-MRI signal from the substantia nigra [58, 59] using ANTs software. To bring the NM-MRI image of each participant into standardized space, T1-weighted images were normalized to MNI space, then NM-MRI images were coregistered to the T1-weighted images, and finally these two transforms were applied to the NM-MRI images. A visualization template (Fig. 1) was created by averaging the spatially normalized NM-MRI images from all participants.

Fig. 1: Measurement of LC signal.
figure 1

Left: visualization template in MNI space created by averaging the spatially normalized NM-MRI images from all participants. Middle: magnified views of the visualization template with the LC search mask overlaid. This mask was manually traced on the visualization template over the hyperintense region surrounding the LC and divided into five sections (displayed in different colors), each spanning 3 mm in the z-axis. Top-right: unprocessed NM-MRI image showing the pons of a representative individual; the central pons reference region is encircled in white. Contrast-to-noise ratio for all voxels was calculated relative to signal extracted from this region. Bottom-right: segmentation of the LC in native space. The LC search mask (yellow, signifying the middle section) was deformed from MNI space to native space to provide a search space wherein the LC was identified on left and right sides as the four adjacent voxels with highest signal contrast. To minimize partial volume effects, of these 4, only the peak-contrast voxel was retained for each side and slice. LC signal was calculated per section by averaging CNR values from all such voxels within the section.

Full size image

Subsequent steps used custom Matlab scripts. An LC search mask was drawn over the MNI space visualization template to cover the LC, defined as the hyperintense voxels at the anterior-lateral edge of the 4th ventricle spanning 15 mm in the rostrocaudal axis (from MNI space coordinates z = −16 to −31, see Fig. 1). The rostrocaudal limits were set based on the position of the LC from a brainstem atlas [60] and cell counting work [61], spanning from the inferior collicus to the posterior recess of 4th ventricle, while excluding the extreme rostral and caudal ends to minimize edge effects. The mask was divided into five rostrocaudal sections of equal length (3 mm). The full LC search mask and the five mask sections were then warped to native space using the inverse transformation generated in the spatial normalization step and resampled to NM-MRI image space. This warped LC search mask defined a search space wherein to find the LC for each participant. A cluster-forming algorithm was used to segment the LC within this space, defined as the four contiguous voxels (total area = 1.96 mm2) on each side and axial slice with the highest mean signal. To minimize partial volume effects, only the peak intensity voxel of these four was retained for calculation of LC signal (on the assumption that this voxel had the highest fraction of LC tissue). The automated segmentation was visually inspected and was found to perform 2.2% of operations suboptimally (e.g. by locating the LC within a bright artifact occasionally present within the 4th ventricle), requiring manual correction. Contrast-to-noise ratio (CNR) for each voxel v in a given axial slice was then calculated as the relative difference in NM-MRI signal intensity I from a reference region RR in the same slice as: CNRv = (Iv − mode(IRR))⁄mode(IRR). We used a reference region with low NM concentration, the central pons (Fig. 1, similar to previous work) [62], defined by a circle of radius 11.6 mm, centered on the midline, 32.6 mm anterior to the LC. Finally, every LC-containing slice was linked to one of the five rostrocaudal LC sections based on which of the five sectioned LC masks was present on the same axial slice (if two sectioned masks were present on the same slice, the LC section was defined for each side by the sectioned mask covering the most LC voxels). LC signal was calculated for each of the five sections by averaging CNR values from all LC voxels within the section.

PET acquisition and analysis

All individuals had [18F]AZD4694 and [18F]MK6240 PET scans acquired with a brain-dedicated Siemens High Resolution Research Tomograph. See previous studies for more detailed PET methods [18, 19]. Tau [18F]MK6240 images were acquired at 90–110 min after the intravenous bolus injection of the tracer and were reconstructed using an OSEM algorithm on a 4D volume with four frames (300 s each) [63]. Amyloid-β [18F]AZD4694 images were acquired at 40–70 min after the intravenous bolus injection of the tracer, and scans were reconstructed with the same OSEM algorithm on a 4D volume with three frames (600 s each) [64]. At the end of each PET acquisition, a 6-min transmission scan was conducted with a rotating 137Cs point source for attenuation correction. The images were corrected for dead time, decay, random and scattered coincidences, and for motion. In order to normalize the PET data, T1-weighted MRIs were non-uniformity and field distortions corrected. PET images were then automatically registered to T1-weighted image space, and the T1-weighted images were linearly and non-linearly registered to the ADNI standardized space [65]. PET images were meninges- and skull-stripped and non-linearly registered to the ADNI template using the transformations from the T1-weighted image to ADNI template and from the PET image to T1-weighted image space. [18F]MK6240 standardized uptake value ratio (SUVR) and [18F]AZD4694 SUVR maps were calculated using the inferior cerebellum and whole cerebellum gray matter as the reference region, respectively [63, 64]. PET images were spatially smoothed to achieve a final 8-mm full-width at half-maximum resolution. [18F]MK6240 values were extracted from a temporal ROI used previously to define tau positivity [66] (we refer to this as the “temporal ROI”; see Fig. 2D). Tau-positive cases were defined as those with SUVR > 1.24 in the temporal ROI, as in our prior work [21]. A continuous measure of SUVR from this ROI was also included in several analyses. Subjects were divided into Braak stage groups [67,68,69,70] according to [18F]MK6240 SUVR values in Braak stage ROIs using methods previously employed by our group [22]. Discordant cases (where regional tau burden did not follow the anatomical progression proposed by Braak) were excluded from analyses of Braak stage. Global [18F]AZD4694 SUVR values were estimated to generate a measure of cortical amyloid-β burden based on a composite set of regions including the precuneus, prefrontal, orbitofrontal, parietal, temporal, anterior, and posterior cingulate cortices [66].

Fig. 2: LC signal and AD severity.
figure 2

A–C Schematic representations of the LC. A LC schematic overlaid on anatomical template in coronal view to illustrate position in the brain. B, C LC schematic showing signal loss in each of the rostrocaudal sections based on tau burden in the temporal ROI (left hemisphere shown in D, right hemisphere is similar). In B tau burden was dichotomized ([18F]MK6240 SUVR > 1.24 to define tau-positive cases) and in C it was left as a continuous measure. The strongest relationship was seen in the mid-caudal LC sections (MNI space z coordinate = −28 to −22; encircled in chartreuse green and matching the yellow and green LC sections shown in Fig. 1). Bilateral LC signal from these sections was retained as the metric used for subsequent analyses of LC signal. Scatterplots showing relationship of mid-caudal LC signal to tau status (E) and measures of AD severity including Braak stage as determined with [18F]MK6240 PET imaging (F), cognitive impairment (G), and dementia severity (H). L left, R right, MMSE mini-mental state exam, CDR clinical dementia rating scale.

Full size image

Statistical analysis

Statistical tests relating final imaging measures to clinical measures were performed on Matlab software. LC signal was related to clinical group using ANCOVAs with Tukey’s post hoc tests. LC signal was related to tau burden in the temporal ROI by linear regression with the model:

$${{{{{{{\mathrm{LC}}}}}}}}\;{{{{{{{\mathrm{signal}}}}}}}} = \beta _{{{{{{{\mathrm{o}}}}}}}} + \beta _1\left( {\left[ {{\,}^{18}{{{{{\rm{F}}}}}}} \right]{{{{{{{\mathrm{MK}}}}}}}}6240\;{{{{{{{\mathrm{SUVR}}}}}}}}\;{{{{{{{\mathrm{in}}}}}}}}\;{{{{{{{\mathrm{temporal}}}}}}}}\;{{{{{{{\mathrm{ROI}}}}}}}}} \right) + \beta _2\left( {{{{{{{{\mathrm{age}}}}}}}}} \right) + \beta _3\left( {{{{{{{{\mathrm{sex}}}}}}}}} \right) + \varepsilon$$

where LC signal was the average signal in whole LC in some models and the signal for each LC section in others; [18F]MK6240 SUVR was continuous in some models and binary (cutoff = 1.24) in others. Partial Spearman correlations were used to relate LC signal to measures of AD severity. Linear regressions and partial Spearman correlations were used to relate NPS severity to LC signal and other neuroimaging measures. The general form for the linear regressions was:

$${{{{{{{\mathrm{NPS}}}}}}}}\;{{{{{{{\mathrm{severity}}}}}}}} = \; \beta _{{{{{{{\mathrm{o}}}}}}}} + \beta _1\left( {{{{{{{{\mathrm{LC}}}}}}}}\;{{{{{{{\mathrm{signal}}}}}}}}} \right) + \beta _{{{{{{{{\mathrm{i}}}}}}}} + {{{{{{{\mathrm{1}}}}}}}}}\left( {{{{{{{{\mathrm{neuroimaging}}}}}}}}\;{{{{{{{\mathrm{measure}}}}}}}}_{{{{{{{\mathrm{i}}}}}}}}} \right) \ldots \beta _{{{{{{{{\mathrm{n}}}}}}}} + {{{{{{{\mathrm{1}}}}}}}}}\left( {{{{{{{{\mathrm{neuroimaging}}}}}}}}\;{{{{{{{\mathrm{measure}}}}}}}}_{{{{{{{\mathrm{n}}}}}}}}} \right)\\ + \;\beta _{{{{{{{{\mathrm{n}}}}}}}} + {{{{{{{\mathrm{2}}}}}}}}}\left( {{{{{{{{\mathrm{CDR}}}}}}}}\;{{{{{{{\mathrm{score}}}}}}}}} \right) + \beta _{{{{{{{{\mathrm{n}}}}}}}} + {{{{{{{\mathrm{3}}}}}}}}}\left( {{{{{{{{\mathrm{age}}}}}}}}} \right) + \beta _{{{{{{{{\mathrm{n + }}}}}}}}4}\left( {{{{{{{{\mathrm{sex}}}}}}}}} \right) + \varepsilon $$

Non-parametric analyses were favored where possible because many measures were ordinal (e.g. CDR score, Braak stage) or not normally distributed according to Lilliefors test (e.g. MMSE score, NPS severity). All analyses controlled for age and sex. See Results for details of the specific models used.

Results

Loss of locus coeruleus signal in AD

First, we confirmed that our novel method of LC signal measurement replicated past reports [7, 11,12,13,14] of reduced LC signal in clinically-diagnosed AD (clinical group effect on whole LC signal: F2,185 = 4.23, p = 0.016, one-way ANCOVA controlling for age and sex). Post hoc testing found a significant difference between CN and AD (p = 0.021) but not between CN and MCI or MCI and AD (p = 0.19 and p = 0.92 respectively; Tukey’s HSD). AD defined biologically by tau positivity ([18F]MK6240 SUVR >1.24 in the temporal ROI [21, 66] shown in Fig. 2D) was also associated with reduced whole LC signal (t186 = −3.26, p = 0.0013, Cohen’s d = 0.48, linear regression controlling for age and sex.

Next, we examined the anatomical topography of tau-associated signal loss within the LC, which we divided into five rostrocaudal sections on the left and right side. The middle section and the section below it (encircled sections in Fig. 2B, C) showed the greatest signal loss in relation to tau burden in the temporal ROI (linear regression controlling for age and sex; Fig. 2B, C). This was true whether tau burden was calculated as a dichotomous or a continuous measure of SUVR in this ROI. LC signal averaged from these sections was markedly reduced in tau-positive individuals (t186 = −4.00, p = 0.0001, Cohen’s d = 0.59, linear regression controlling for age and sex, Fig. 2E). Therefore, we retained this as the LC signal measure to be used for all subsequent analyses, referred to as the “mid-caudal LC” (Fig. 2B, C).

Locus coeruleus signal and AD stage and severity

We found that mid-caudal LC signal loss was significantly correlated to more advanced Braak stage of AD (Spearman ρ = −0.31, p = 0.00006, n = 160; partial correlations controlling for age and sex; see Fig. 2F). Analysis across all stages found that LC signal was lost at the rate of 0.67% CNR per stage, although the rate of loss was higher from stage 3–6, equal to 1.10% CNR/stage (linear regression controlling for age and sex). Consistent with this, general clinical severity, measured as cognitive impairment and dementia severity, was also negatively correlated to LC signal (MMSE errors: ρ = −0.14, p = 0.058, n = 187; CDR score: ρ = −0.28, p = 0.0001, n = 188; partial correlations controlling for age and sex; see Fig. 2G, H). Taken together, these findings suggest that degeneration of LC may be progressive throughout the early course of AD.

Locus coeruleus signal and symptoms of Alzheimer’s disease

We next investigated the clinical correlates of mid-caudal LC signal controlling for key pathophysiological measures to assess the independent contribution of the LC to these measures. First, we tested the relationship of the LC signal to NPS severity (MBI-C total score) in all participants. We found a significant interaction between tau positivity and LC signal on NPS severity (βInt = 0.75, t171 = 2.76, p = 0.006) due to a significant relationship between LC signal and NPS severity in tau-positive participants (β1 = 0.81, t171 = 3.36, p = 0.0009) but no such relationship in tau negative participants (β1 = 0.06, t171 = 0.42, p = 0.67, linear regression controlling for CDR score, age, and sex). We further investigated the LC signal’s association with NPS in the tau-positive group (n = 51) and found that it was present using parametric or non-parametric statistics and that the association was slightly strengthened when additional pathophysiological measures were included in the model (Table 2; in a full model the correlation of LC signal to MBI total score was ρ = 0.35, p = 0.019; partial Spearman correlation controlling for tau burden in temporal ROI, cortical amyloid-β burden, cortical gray matter volume, total intracranial volume, CDR score, age, and sex; see Table 2 and Fig. 3A, B). This positive correlation is consistent with our hypothesis that preserved and/or elevated LC function is associated with worse NPS. While in most models tau burden in the temporal ROI also significantly predicted higher NPS severity, LC signal was consistently the more influential predictor. Furthermore, the full model including all pathophysiological measures explained a substantial amount of variability in NPS severity (R2 = 0.50, adjusted R2 = 0.41; Table 2 and Fig. 3A). Post hoc analyses examining MBI-C domains and including the covariates from the full model found the correlation to LC signal was significant only for the impulse dyscontrol domain (ρ = 0.44, p = 0.0027, Fig. 3C). Subsequent examination of all LC sections (Fig. 3D) showed the relationship of LC signal to impulse dyscontrol was strongest in the second caudal-most section.

Table 2 Prediction of neuropsychiatric symptom severity (MBI-C total score) in tau-positive individuals.
Full size table
Fig. 3: Relationship between mid-caudal LC signal and neuropsychiatric symptom severity in n = 51 tau-positive older adults.
figure 3

A Neuropsychiatric symptom severity was strongly predicted in a linear regression model combining several multimodal neuroimaging measures of pathophysiology (“full model” in Table 2), including LC signal, tau burden in the temporal ROI, cortical amyloid-β burden, and cortical gray matter volume (adj. R2 = 0.41). The most influential predictor in this model was LC signal, which was positively correlated to NPS severity (B). Of the five domains of NPS, LC signal was most strongly correlated to the Impulse Dyscontrol domain (C). NPS severity score was adjusted in B and C based on other covariates in the model. D LC schematic showing correlation of LC signal to impulse control deficits for all rostrocaudal LC sections (controlling for covariates as in “full model”).

Full size image

Lastly, we examined the relationship of LC signal to cognitive impairment, measured as errors on the MMSE. Unlike the analysis in the section on AD stage and severity (Fig. 2G) we now tested this relationship while controlling for other measures of pathophysiology. We found that this relationship was not significant (ρ = −0.20, p = 0.18, n = 52, Spearman partial correlation on tau-positive participants controlling for covariates as in the full model). This would suggest that LC signal may not have a strong and direct association to general cognitive impairment in AD.

Discussion

Here we report several findings regarding the clinical and pathophysiological correlates of LC signal, a proxy measure of norepinephrine neuron loss, in AD. Loss of LC signal appears to be a progressive process that correlates with AD stage as indexed both by Braak stages of cortical tau proliferation and by severity of clinical symptoms. Despite these detrimental correlates of LC signal loss in AD, preservation of LC signal can also be detrimental as it is associated with worse NPS in AD patients. The relationship between LC signal and NPS was not confounded by the presence of cortical pathology; indeed, both LC signal and cortical tau burden independently predicted severity of NPS.

Critically, we found a significant interaction between tau status and LC signal on NPS severity suggesting that tau may dysregulate LC function in a disease-specific way that is distinct from alterations in LC function during normal aging. Specifically, we found no clear relationship between LC signal and NPS severity in healthy individuals. This finding is unsurprising since the level of endorsement of NPS was low in such individuals (Table 1) and furthermore, the assumption that variability in LC signal can be used as a proxy of the extent of LC degeneration may only apply in the AD/MCI groups, not the healthy group where LC degeneration is minimal and other factors may predominate in determining variability in the LC signal. On the other hand, we found that in tau-positive individuals, a preserved level of LC signal was associated with NPS risk. This is consistent with evidence of a positive relationship between norepinephrine function and NPS in AD [34, 36,37,38,39, 45] (but see [35, 71]), suggesting NPS are associated with LC preservation and enhanced norepinephrine function from compensatory changes in norepinephrine production, receptor expression, and number of axon terminals [3, 4, 42, 43]. We propose a model whereby variability in the progression of different disease processes may leave some patients with cortical tau pathology but spared LC integrity, possibly leading to dysregulation in the cortical and subcortical regulation of behavior and the expression of impulse control problems and other NPS. Perhaps the cortical tau insult interferes with top-down regulation of behavioral responses to stressful or arousing situations when the LC-norepinephrine system is intact or hyperactive, leading to agitated or aggressive behavior. Furthermore, NPS may be promoted not only by interaction of LC signal with cortical tau, but also with tau in the LC itself, which cannot be measured with PET imaging due the size of the LC but would be expected to be present in those with cortical tau [2,3,4] and could dysregulate LC function.

These findings are consistent with prior reports showing efficacy of norepinephrine blocking agents against aggressive and agitated behaviors in AD [38,39,40]. NM-MRI could have promise in this regard as a biomarker to indicate patients with high LC signal whose NPS may respond to such treatment, as opposed to patients with low LC signal whose NPS may have an origin unrelated to the norepinephrine system and who could be harmed by treatments exacerbating their already low norepinephrine system function. Furthermore, while NPS are often recognizable by clinical observation alone, a biological measure such as LC signal could show promise as a biomarker of NPS risk prior to their overt manifestation. Such a risk marker would be important for clinical decision-making, supporting vigilance of emergent NPS, and allowing administration of NPS treatments at the earliest stages, even during the prodrome.

Our findings provide insight regarding the anatomical topography and timing of LC signal loss in AD. Consistent with reports of a highly variable extent of LC cell loss in AD [61], we observed a large variability in LC signal loss (Fig. 2E). LC signal loss was most pronounced in central LC, consistent with this region having the greatest density of norepinephrine cells and the greatest number of cells lost in AD [61]. This anatomical variability underscores the strengths of our “funnel tip” LC signal measurement approach, allowing automated determination of approximate rostrocaudal position within the LC, while still extracting the signal from unprocessed images to avoid distortion of this small structure. This approach has the promise to target specific effects that may be anatomically segregated from other, potentially confounding, effects. For instance, we saw the tau effect was strongest in the middle LC section, whereas the NPS effect was strongest in the section below the middle (Figs. 2B, C and 3D). Such a subdivision of the LC could help probe specific circuits that may be subserved by distinct LC regions, consistent with recent work demonstrating a modular organization of LC circuitry [72]. Yet, despite this organization, distinct modules tend to be intermingled and more research is needed to determine the extent of any topographical pattern to LC projections in primates [72,73,74,75]. Thus, it may be premature to provide an explanation for why caudal LC would be specifically linked to impulse control symptoms and whether this is due to the connectivity of this subregion [73] or perhaps the degree to which it is vulnerable to degeneration [61].

Regarding the timing of LC signal loss in AD, mirroring postmortem evidence [1], we found LC signal loss to be a gradual process across Braak stages. This suggests that there is substantial delay between the accumulation of tau in the LC at the earliest stage of AD [3, 4] and the degeneration of LC neurons. Nevertheless, close examination of the relationship between LC signal and Braak stage (Fig. 2F) shows a curious phenomenon where LC signal is low at Braak stage 1 and appears to increase from Braak stage 1-3. Although this may be simply due to a low number of observations (e.g. at Braak stage 3), such a rebound in LC signal could reflect a biological process linked to a hyperactive LC-norepinephrine system after the onset of degeneration [2, 3, 36]. For instance this could lead to increased cell size, or accelerated NM accumulation, changes that could increase the LC signal [5, 76]. Indeed, a correlation of NM signal to catecholamine function has been observed in the dopamine system [58]. In this speculative scenario, it could be that loss of NM-MRI signal is apparent even at Braak stage 1 (when LC signal was significantly reduced relative to Braak stage 0, t88 = 2.36, p = 0.020) but that reductions in LC NM-MRI signal due to degeneration become somewhat confounded by increases in NM-MRI signal due to hyperactivity during intermediate Braak stages. While this would add noise when NM-MRI is used as a marker of early LC degeneration, it may enhance its sensitivity as a marker of NPS, which may be exacerbated not only by preservation but also hyperactivity of the LC [34, 36,37,38,39,40, 45].

Our study had many strengths including a relatively large sample and inclusion of multimodal neuroimaging measures of pathophysiology. However, certain methodological aspects may limit interpretation of the data. Our study supported a role for the norepinephrine system in NPS but function of other neurotransmitter systems (e.g. acetylcholine, serotonin) may also play an important role in NPS [36, 77] but was not measured. We found that combining measures of cortical and LC pathology explained a substantial amount of variance in NPS severity; however, a detailed examination of the role of the cortical measures in promoting NPS was beyond the scope of this work. We do not interpret, for instance, the absence of a relationship between our measure of cortical amyloid-β burden and NPS as being in conflict with prior studies targeted to specific brain regions and types of NPS [78]. Indeed, NPS are a heterogeneous combination of symptoms and it may be overly simplistic to expect LC signal or any measure to predict all NPS domains, some of which could even be associated with low LC signal [46]. One domain, psychotic symptoms, are of interest but could not be investigated in our sample due to very low level of endorsement. We focused on early stages of AD (CDR <3, with CDR for most AD cases <2) and cannot draw conclusions regarding the role of the LC in moderate to severe dementia. At these later stages, perhaps the influence of compensatory increases in norepinephrine function on NPS could be overwhelmed by a more advanced degeneration of the system. Indeed, some preclinical work suggests that with more advanced LC damage, NPS-like behavior begins to diminish [79]. Arguing against this, some postmortem studies (where cases are highly advanced) have found a positive relationship between antemortem NPS and norepinephrine function [24, 35,36,37]. A detailed assessment of cognition in AD was beyond the scope of this paper but future work could test whether the association seen between LC signal and specific cognitive domains in healthy aging [80] also applies in AD. Finally, our MRI data, like most published LC NM-MRI studies [7, 11,12,13,14, 47] was collected on a 3 Tesla scanner. Ideally, a 7 Tesla scanner [81] would be used due the increased spatial resolution afforded at ultra-high field strength. At the resolution employed here, measurement of LC signal may have been subject to partial volume effects in which LC voxels contain non-LC tissue. Nonetheless, the in-plane resolution was much smaller than the area of the LC (in-plane area of one voxel = 0.47 mm2, cross-sectional area of the LC ~1.8 mm2 [61]), the LC could be clearly identified on visual inspection, was segmented with 98% accuracy by our algorithm, and our LC signal measure revealed highly significant effects, in line with a priori hypotheses.

In summary, the LC signal tracks Braak stage of AD and is positively correlated to the severity of NPS, independently of other aspects of pathophysiology. These results demonstrate the utility of NM-MRI to interrogate the role of the norepinephrine system in human studies of AD pathophysiology. They also provide early evidence in favor of NM-MRI as a practical and non-invasive biomarker that could have potential to indicate NPS risk or likelihood of response to specific treatments.