Using neuroimaging to investigate Alzheimer’s disease heterogeneity

The heterogeneity of tau in people who are amyloid-β positive

In people who are Aβ-positive, low tau burden often correlates with low tau accumulation rates.¹ Dr Landau’s work showed that between 20−40% of people who are Aβ-positive and cognitively impaired have normal levels of tau. Using a composite of a number of cortical regions, she found that for patients where tau levels were low (n=96) compared to high (n=133), AD diagnosis was lower (28% vs 45%), they were older (mean 78 vs 74 years) and fewer were female (32% vs 53%).

Additionally, those with low tau burden were at an earlier stage of disease progression, were less cognitively impaired, had less AD-related hypometabolism and atrophy, and had lower Aβ levels. They did, however, have increased cerebrovascular risks, though lower scores on the Preclinical Alzheimer Cognitive Composite measure correlated with greatest cerebrovascular risk, and increased cardiovascular health conditions.

Tau burden should be considered in clinical trials

Dr Landau also examined regional tau distribution and found that relative distribution was the same regardless of tau level with highest levels in both high and low tau patients in the medial temporal region. In conclusion, Dr Landau’s work indicated that as up to 40% of people who are Aβ-positive have low tau, this may mean tau is a poor therapeutic target in these people and that in clinical trials tau burden should be considered as a variable that may affect outcome.

Can tau PET predict domain-specific cognitive decline in Alzheimer’s disease?

Cognitive impairments in episodic and semantic memory, as well as language, correlate with tau distribution, as shown with positron emission tomography (PET) scanning.² Dr Malpetti investigated whether the pattern of tau deposition in early, symptomatic AD can predict future cognitive performance declines. She used both tau-PET and 3T-Magnetic Resonance Imaging (MRI) and imaged patients at baseline and every 9 months, with cognitive testing at all time points. There were 79 Aβ-positive and 52 Aβ-negative patients.

Cognitive decline can potentially be predicted by tau burden

Over two follow up visits, there were little changes in episodic memory and executive function for all, with predominantly negatively but more individual changes in the other domains (semantic memory, language, visuospatial). Combining tau-PET load and grey matter MRI atrophy, they found that in the domains of episodic memory, semantic memory and language, higher tau was correlated with lower grey matter volumes in the relevant brain regions for each. They also found that baseline grey matter volumes mediated any predictive effect of tau-PET patterns. Dr Malpetti concluded that “Tau-PET may be a useful precision medicine tool for predicting patient-specific cognitive decline in symptomatic AD.”

The use of tau subtypes to map clinical progression in Alzheimer’s disease

Heterogeneity in AD can include time of onset, clinical variation, predominant brain location and rate of progression.³ There are also subtypes according to pattern of atrophy, which impacts resting-state networks.⁴ Dr Rauchmann’s study included 166 people with Aβ-positive AD and 155 Aβ-negative controls to identify a tau-PET subtype pattern. Four distinct subtypes were observed: occipito-temporo-parietal subtype (OTCS), limbic subtype (LS), temporo-parieto-frontal (medial temporal lobe sparing) subtype (TPFS) and lateral temporal subtype (LTS). Correlating this with neuropsychological testing, the OTCS subtype showed marginal cognitive sub-domain reductions, the LS showed reductions in memory subscores, TPFS showed more severe reductions in cognitive domain scores and the LTS showed global reduction in all subscores, especially language.

Assessing tau subtypes may help personalise Alzheimer’s disease treatment

Over approximately 3.3 years, in the TPFS and LTS, there was a steep increase in tau progression overall. Looking at individual brain regions, the OTPS, LS and TPFS showed increases in the occipital and temporal lobes, with the TPFS also showing frontal lobe increases. The LTS showed global increases and in amyloid-PET uptake. There were only marginal cognitive changes in this time in the OTPS and LS, with higher decreases in the visual cognitive domain subscore in the OTPS and language subscore in the LS. The LTS showed steep decreases in all but memory domains. Dr Rauchmenn concluded that by understanding a person’s subtype, treatments may be developed that are more personalised.

Subtype detection in early Alzheimer’s disease

Dr Venkatraghavan’s work aimed to develop a method to identify subtypes of AD progression, severity-based staging, subtype variations and early subtype detection using artificial intelligence. This was named SNOWPHLAKE: Staging NeurOdengeneration With Phenotype informed progression timeLine of biomarKErs. The retrospective cohort included people with young onset Aβ-positive AD at either the pre-clinical (n=184, 50% female, mean age 63.7 years), prodromal (n=322 44% female, mean age 66.5 years) or symptomatic (n=1054, 53% female, mean age 64.8 years) stage. Aβ-negative controls had subjective cognitive decline but did not progress to any form of dementia (n=184, 50% female, mean age 64 years).

Using MRI, they found four atrophy-based subtypes: frontal (FS) (26.8%), parietal (PS) (23.7%), ‘typical’ (TS) (early hippocampus atrophy) (19.8%) and subcortical (SS) (17.9%), as well as a number of outliers (11.8%). SNOWPHLAKE was able to produce estimated progression patterns and timelines for each subtype and showed that cognitive function differed significantly between subtypes. For example, executive functioning differed between FS/TS, PS/FS, PS/TS, SS/TS and SS/PS; visual memory differed between FS/PS and TS/PS; episodic memory between PS/FS and language between SS/PS.

Artificial intelligence can be used to identify Alzheimer’s disease subtypes

Dr Venkatraghavan then estimated the status of patients along an estimated timeline and found these mapped to the current disease stage, validating the estimated timelines. Using another dataset than included longitudinal follow-up data on later onset patients with AD, the SNOWPHLAKE model was validated. Dr Venkatraghavan concluded that SNOWPHLAKE can help identify a more homogenous patient population from a heterogenous cohort.

Patterns of pathology in people with comorbid Dementia with Lewy Bodies and Alzheimer’s disease

Distinct neuropathology patterns are found between AD and DLB.⁵ However, approximately 50% of people with DLB have AD co-pathology.⁶ In this study, Dr Grothe investigated the differential effects of DLB and AD pathology on fluorodeoxyglucose (FDG)-PET pattens in people with AD (n=51) or amnesiac mild cognitive impairment (aMCI; n=8) compared to cognitively healthy controls (n=179). On autopsy, 21 patients were classified as having AD only (mean age 82 years, 59% male, 57% Aβ-positive), seven as DLB only (mean age 89 years, 86% male, 0% Aβ-positive) and 24 as AD+DLB (mean age 81 years, 79% male, 71% Aβ-positive). Differences were also shown in substantia nigra neuronal loss and mini-mental state examination (MMSE) scores (highest in the DLB only cases) and memory composite scores (highest in AD-DLB cases).

Alzheimer’s disease predominates neurodegeneration patterns in people with comorbid Dementia with Lewy Bodies

The voxel-wise FDG-PET pattern was almost indistinguishable between AD ±DLB with typical temporo-parietal hypometabolism. The DLB only group showed typical posterior occipital FDG-PET with relative medial temporal lobe sparing. The cingulate island sign ratio (relative preservation of posterior cingulate metabolism compared to pronounced occipital hypometabolism in DLB⁷) was significantly elevated in the DLB only cases, but there was no difference between the AD only and AD+DLB cases. Cases with elevated substantia nigra neuronal loss showed elevated cingulate island sign, including three cases in the AD+DLB group. Braak tau stage was better correlated with a typical AD hypometabolic pattern with substantia nigra neurodegeneration correlating more with a typical DLB pattern. Dr Grothe concluded that regional neurodegeneration phenotype may be dominated by AD pathology in patients with AD+DLB.

Clustering of people with Alzheimer’s disease using MRI and self-organising maps

Dr Petersen’s work investigated if a SOM clustering algorithm utilising data from volumetric MRI measures of people with mild dementia could identify clusters of patients. The cohort included 1041 people with mild AD (mean age 73.3 years, 50% female, 66% Aβ-positive).

Self-organising maps using imaging data can help identify Alzheimer’s disease subtypes

They found all cases could be bucketed into one of three clusters with significant differences between age, Aβ-positivity, neuropsychological test scores (for instance on the MMSE or Alzheimer's Disease Assessment Scale subscales) and MRI volumes in a number of brain regions including the hippocampus, prefrontal cortex and entorhinal cortex. Cluster 1 showed the largest atrophy in the frontal and parietal lobes, Cluster 2 showed the least atrophy across all regions and the least change in the majority of neuropsychological test scores between the first and last visit, and Cluster 3 showed the largest atrophy in the temporal lobe. Dr Peterson concluded that clustering using SOM can potentially be useful to identify homogenous subsets of people with Aβ-positive AD.