The Spatial-Temporal Ordering of Amyloid Pathology and Opportunities for PET Imaging

Group 1: Postmortem Histology Studies

Histopathology is considered the gold standard method for amyloid plaque detection (19), with 2 landmark studies. Braak and Braak (20) evaluated the progressive topology and density of amyloid deposits in an analysis of 83 postmortem brains from demented and nondemented elderly patients. Based on both frequency and density of plaques, 3 stages of amyloid deposition were described, starting with deposits in basal portions of the frontal, temporal, and occipital lobes, followed by medium-density deposits throughout cortical association areas and white matter, as well as mild involvement of the hippocampus, and ending with inclusion of the sensorimotor cortex and expansion to subcortical structures, particularly striatum. Later, Thal et al. (14) followed a similar methodology but statistically tested for the presence of a constant sequence of regional amyloid deposition, adding strength to their proposed regional sequence compared with Braak albeit disregarding plaque density. In Thal et al., the first phase describes amyloid deposition in the whole neocortex with no mention of within-region differences, in contrast to the relatively more detailed regional assessment in Braak. In turn, 5 phases provide complementary information to Braak’s description, focusing on the subsequent spreading of amyloid through subcortical regions.

Although postmortem studies are considered the gold standard, limitations include the modest subject numbers and brain regions evaluated (14,20) and the focus on older end-of-life populations for which other end-stage medical illnesses could further compromise the analysis (8).

Group 2: Studies Comparing Amyloid PET Signal with Histology-Based Regional Thal Staging

These studies used cohorts with both PET (visual or semiquantitative evaluation) and autopsy data, which are highly informative but generally limited in availability.

Murray et al. (9) showed in a ¹¹C-PIB end-of-life cohort of 35 subjects that incremental Thal phasing significantly predicts global SUV ratio (SUVR) signal increases, albeit in a modest population. A follow-up study (21) in 179 end-of-life subjects demonstrated that increases in a global PET measurement scale (centiloid units (22)) are nonlinearly related to Thal phases. Also, global ¹¹C-PIB PET detected changes in only Thal phase 2, missing phase 1 entirely. However, the authors suggested that regional analysis could provide more sensitive estimates of early focal deposits than a global neocortical measure (9).

Similarly, in a 2015 Thal study (23), concordance between visual ¹⁸F-flutemetamol findings and amyloid phase (Thal phases 0–5) was examined in a dual PET/autopsy 68-subject cohort. PET was visually negative for amyloid in cases that were amyloid phases 1–2, whereas advanced-disease cases of phases 4–5 concurred with a positive scan. In the interim cases, phase 3, only 33% were positive, indicating that this phase may represent an inflection point in the progression of amyloid that results in the ability to visually assign positivity to a scan.

Thal et al. additionally demonstrated in an extended cohort of 97 patients that ¹⁸F-flutemetamol PET SUVR signal can be streamlined into 0–3 amyloid PET phase estimates. Using signal from a combination of cortical regions and the caudate nucleus, this PET phasing strategy can act as surrogate for Thal phasing between phases 3 and 5 with 97% accuracy when allowing ±1 Thal phase (24).

Group 3: Ranking of Regional Amyloid Signal by PET-Only Studies

Although limited by the lack of pathologic confirmation, in vivo studies can implement a wide range of strategies to disentangle amyloid deposition patterns and have investigated the regional cortical involvement in more detail than previous postmortem work. To date, most amyloid PET studies have included subjects across the clinical AD spectrum (25–27), and only recent work has studied cognitively unimpaired subjects alone (28,29).

Huang et al. (25) (n = 36) used a simple ANOVA between volumes of interest to assess regional ¹⁸F-florbetapir PET signal increments across clinical stages. They concluded that amyloid initially accumulates in precuneal, frontal, occipital, and posterior cingulate areas. In a larger dataset (n = 222), Villeneuve et al. (26) supported these findings by inferring a sequence of amyloid deposition from statistically significant signals of elevated ¹¹C-PIB distribution volume ratio among 66 permutations of 22-subject groups of interest with increasing mean ¹¹C-PIB distribution volume ratio. With this analytic approach, the authors found an initial deposition of amyloid in medial cortical regions, followed by more lateral regions, finally involving the lateral temporal lobes. Finally, Cho et al. (27) also used cognitively healthy subjects as a reference group to obtain ¹⁸F-florbetaben PET Z-transformed SUVRs from 195 subjects across the AD spectrum. The difference in frequency of regional involvement was assessed and then verified with bootstrapping statistics, allowing determination of rank confidence intervals and standard errors. In this work, the precuneus was reported to show early amyloid deposition, whereas the cingulate showed later involvement together with inferior temporal and the insular cortex, somewhat in contrast to previous studies (27).

More recently, studies have focused on amyloid PET signal from cognitively unimpaired subjects, aiming to capture amyloid accumulation, generally more apparent before cognitive impairment (30,31). One key study is by Grothe et al. (28), which in addition to assessing a regional ordering of amyloid deposition was first to create a corresponding staging system based solely on PET imaging. The authors ranked the frequency of regional involvement in amyloid deposition across 179 Alzheimer’s Disease Neuroimaging Initiative cognitively unimpaired cases imaged with ¹⁸F-florbetapir, essentially translating the traditional methodology from postmortem studies to PET imaging. With this method, 4 stages of amyloid deposition were described, for which early involvement of temporal regions was more pronounced than in previous work (Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

General mapping of amyloid deposition order. aMCI = amnestic mild cognitive impairment; CSF = cerebrospinal fluid; HC = healthy control; SPM = statistical parametric mapping; VOI = volume of interest.

Sakr et al., in 2019 (32), further demonstrated that Grothe’s 2017 model could be applied to a different cohort of 318 subjective cognitive impairment subjects (33) scanned with ¹⁸F-florbetapir PET, with only few subjects deviating from the model. However, when recomputing the model in this subjective cognitive impairment cohort, important discrepancies such as a more frequent involvement of posterior and lateral temporal regions were noted. The authors suggested that age differences between cohorts could explain altered regional deposition patterns due to, for example, late-age comorbidities (32).

Although results from Sakr (32) suggest between-tracer translatability of staging models, this could not be reproduced in work by Lopes Alves et al. (29). In this work, an event-based ordering of regional amyloid deposition was investigated in 3 cohorts (1,267 subjects) across the full AD spectrum. Based on a probabilistic approach, a deterministic event-based model (34) learned the temporal ordering of biomarker changes from normal to abnormal in large cross-sectional datasets. Ranking of regions was based on their probability of abnormality, and a populationwide ordering was determined by minimizing cross-subject differences. The 3 radiotracer datasets presented variations in regional probability of accumulation, especially in temporal and frontal regions. In turn, the striatum appeared in later stages for all tracers.

Considering the above, recent work by Collij et al. focused on overcoming the potential lack of cross-cohort and cross-tracer validity for previous staging models (35). Here, a 5-stage model was constructed from PET imaging data from 989 cognitively unimpaired subjects originating from 6 cohorts and scanned with 1 of 4 different PET tracers. The methodology expanded on the earlier work from Grothe et al. (28), with additional tests for abnormality cutoffs and alternative stage definitions. The first stage of deposition started in cingulate regions, followed by precuneus, inferior parietal, paracentral gyrus, lateral orbital, and middle frontal. The insular cortex was involved in stage 2. Then, basal temporal and associative parietal and frontal depositions were present in stage 3, and finally, other temporal and occipital regions were involved in stage 4 (Fig. 1). These results suggest that by incorporating the intrinsic heterogeneity in populations and available tracers, it is possible to construct a more generalizable staging system, applicable to increasingly common multitracer studies.

Studies in this section describe a range of strategies to understand the regional ordering of amyloid deposition and potentially transform this knowledge into staging systems. Despite differences, different studies were able to translate a temporal pattern of amyloid accumulation into a system for staging amyloid burden. Nonetheless, despite an overall agreement in a broad ordering of amyloid deposition, there is not a complete match in the final staging systems. This indicates that a universal staging system would either have to account for heterogeneity or allow a degree of variability to be accounted for, and further investigation is needed to determine whether the proposed staging models have distinct applications in a clinical setting or to support clinical trials.

Group 4: Studies Based on Longitudinal PET Imaging

These final studies used longitudinal PET data to derive the sequential ordering of amyloid deposition. These studies are suited to describing the dynamic amyloid spatial-temporal deposition pattern and can effectively account for population differences in Aβ burden, age-related nonspecific Aβ accumulation rates, and disease stage heterogeneities.

Hanseeuw et al. (36) based their methods on a priori knowledge from postmortem studies that striatal amyloid deposition is generally preceded by cortical involvement. They hypothesized a higher likelihood of transition of amyloid deposition from cortical to striatal regions than vice versa (14). Longitudinal PET data obtained from ¹⁸F-florbetapir and ¹¹C-PIB scans from 1,433 patients along the full dementia spectrum were used to assess the hypothesis and define 3 PET stages of amyloidosis: low overall PET signal, high PET signal in cortex only, and high PET signal in both cortex and striatum. With the available serial PET measures, individual trajectories advanced successfully through these 3 stages, providing longitudinal support to previous reports showing that the striatum is affected by amyloid deposition only after cortical involvement.

More recently, Palmqvist et al. (37) also assessed the longitudinal patterns in amyloid deposition from PET images, but their analyses were grounded on an independent amyloid biomarker, namely cerebrospinal fluid (CSF) Aβ₄₂. The authors hypothesized that early regions of amyloid deposition are those that display increasing accumulation rates (in SUVR/y) when comparing CSF-negative/PET-negative to CSF-positive/PET-negative subjects (38). Conversely, regions where accumulation rates are higher in CSF-positive/PET-positive than in CSF-positive/PET-negative individuals would be those that are affected at a later stage in the disease process. Their results described the cingulate, orbitofrontal, precuneus, and insula as early regions. Central, occipital, temporal, and some limited frontal regions were identified as later regions. Results were then successfully replicated in a separate population from the BioFINDER cohort (39).

Recently, Palmqvist’s work has been expanded by Mattsson et al. to construct a longitudinally valid staging model again grounded on CSF Aβ₄₂. In addition to the identification of the early and late regions as per Palmqvist’s methodology, this work defined an intermediate period by identifying remaining regions with longitudinal accumulation that did not distinguish early or late processes (40). From this replication and extended work, Mattsson et al. were limited to identifying 3 stages of amyloid accumulation. The first stage was defined by the precuneus, the posterior cingulate, the isthmus cingulate, the insula, and the medial and lateral orbitofrontal cortices. The second stage involved the medial and lateral frontal, lateral parietal, temporal, and additional occipital regions, and the final stage described amyloid accumulation in the lingual, pericalcarine, paracentral, precentral, and postcentral cortices (Fig. 1). This model was then applied to 2 cohorts and successfully staged 2,039 PET scans from 741 participants.

Cortical Generally Precedes Striatal Amyloid Accumulation

Cortical generally precedes striatal accumulation, as is apparent from histologic (14,20) and PET studies that included subcortical regions (24,28,29,32,35,36). Thal et al. (14) justified this order by postulating a sequential amyloid accumulation along neutrally connected areas, with the striatal region receiving inputs from the neocortex (48). Palmqvist (37) and Cho (27) carry this argument a step further, by suggesting a default-mode-network–driven mechanism for the regional accumulation process. In fact, several studies discussed in this review showed that striatal amyloid correlates to worsening cognitive scores and may indicate the potential for transition from preclinical to clinical AD (14,28,36). However, the region is characterized by a predominance of amyloid plaques of diffuse form, to which amyloid tracers are less sensitive than they are to higher-density neuritic plaques (41,49). Yet, the specificity of the thioflavin tracers for detecting varying densities of amyloid plaques in this region is high (13,50,51), and the striatum may therefore have a unique potential to stage patients and assess the risk of disease progression.

In PET Studies, Medial Neocortical Regions Are Seen to Accumulate Amyloid Earlier Than Lateral Regions

In subjects with moderate amyloid load, medial cortical regions are often more visually intense than lateral regions in amyloid PET scans (8,52,53), despite autopsy evaluation indicating that lateral aspects of temporal, frontal, and parietal lobes are involved earlier in the disease than medial regions (14,20,47). To examine this trend, Smith and Buckley (54) performed a post hoc analysis on a substantial subcohort of subjects from the ¹⁸F-flutemetamol autopsy validation cohort (n = 106) (12) and examined medial and lateral regions with similar levels of pathology. They showed that medial regions were read positively in over 20% more cases than lateral regions despite comparable pathology measures.

One possible explanation for these differences is PET signal distortion, derived from the partial-volume overlap effect between regions. The proximity of medial regions to white matter and additional gray matter signal from the contralateral hemisphere could raise a region’s relative intensity, adding to true tracer signal. Conversely, in lateral regions, the lack of intensity in juxtaposed regions results in decreased signal due to spill-out into interstitial cranial regions.

As a side effect of this signal distortion, medial regions (easily accessible from a parasagittal plane) can be helpful to visually detect early amyloid deposits with high sensitivity. However, when discussing the true regional amyloid deposition sequence, it is important to consider that the PET signal might represent a heterogeneous distortion of an otherwise uniform tracer distribution. Thus, uncorrected regional averages of quantitative data could influence composite scores and, possibly, the order of staging when early amyloid deposition is examined.

The Placement of Temporal Regions in the Pathologic Ordering Is Remarkably Variable Across Studies

Early deposition in temporal regions is particularly observed in studies using ¹⁸F-florbetapir PET (25,28,32), whereas elsewhere the temporal regions have been described as rather late-accumulating regions (26,35,37), perhaps paralleling the distinct processes of cognitive degeneration and aging as has been suggested by Chételat’s group (55,56). From the studies using ¹⁸F-florbetapir PET covered here, only Palmqvist (37) and Mattsson (40) do not report temporal regions as among the earliest to display amyloid accumulation. Compared with other ¹⁸F-florbetapir studies, the main difference in these studies can be attributed to the use of longitudinal PET imaging to derive the ordering of regional accumulation. In view of Chételat’s findings of a distinct pattern between individuals 20–50 y old and elderly subjects, it is possible that temporal regions have an increased signal in midlife and therefore display an elevated PET signal at baseline. This pattern would then be captured when assessing cross-sectional studies of elderly subjects with overall low amyloid PET uptake but would be missed when accounting for longitudinal rates of accumulation starting at 50 y of age or older. However, the reason for this observation’s being restricted to ¹⁸F-florbetapir scans remains unknown.