We used idiopathic PD patients to elucidate the role Aβ accumulation has on PD cognitive decline using both ‘top-down’ and ‘bottom-up’ analysis methods. We found that Aβ accumulation has a moderate association with future cognitive decline in PD, but only when specific brain regions are the target of this increased Aβ burden. Conversely, Aβ burden in HC only moderately predicted present cognitive scores and had no prospective utility. We also found that Aβ deposition in PD differed from HC in its clustering. In PD, Aβ deposition formed 3 major cortical clusters, with Aβ deposition in the more posterior located cluster 2 being more strongly related with cognitive decline than the more anterior-ventral cluster 1. In HC, 2 of the clusters were made up of the bilateral ACC and bilateral PCC respectively, with the other 16 ROIs forming one large, more heterogenous cluster.
Cluster analysis is a tool that helps identify variables (i.e., brain regions) sharing similarity in some fashion (Bridges, 1966; Anderberg, 2014). In PD cluster 1, the bilateral gyrus recti surprisingly had a significant positive Pearson correlation with MoCA score one year after scan; this means that in these frontal regions, cognitive performance was not negatively impacted by Aβ burden. The role of the rectus gyri in human cognition remains unclear, but it is hypothesized to be involved in emotional regulation, compulsive behaviour, and social cognition (Couto et al., 2013; Knutson et al., 2015; Wang et al., 2018). These cognitive abilities are not measured by the MoCA which could also explain why Aβ burden in this brain region was not associated with a cognitive decline. Alternatively, the positive relationship between Aβ in the left gyrus rectus and cognitive ability may suggest that this brain region may not significantly impact cognitive performance in the moderate stages of PD pathology.
We speculate that cluster 2 ROIs in the PD group are those most vulnerable to Aβ burden leading to cognitive decline in PD. The right parietal cortex and left occipital cortex had negative Pearson correlations with MoCA score at time of scan and two years after scan, respectively. Uribe and colleagues (2018) performed a hierarchical cluster analysis on MRI derived cortical thickness data in 77 PD patients and found two patterns of cortical atrophy. PD patients grouped under the more posterior atrophy cluster had more atrophy in the bilateral occipital and superior parietal lobes as well as more pronounced cognitive decline as measured by a neurocognitive test battery, compared to PD patients with the more anterior atrophy cluster.
The bilateral PCC was unusual in our analysis, forming its own cluster in both PD and HC. The PCC region is a central node in the default mode network (DMN) and is densely connected with numerous other brain regions (Leech et al., 2012). Previous studies have noted the association between Aβ deposition overlapping with nodes in the DMN (Buckner, 2012; Adriaanse et al., 2014; Palmqvist et al., 2017). In Alzheimer’s disease, it was found that Aβ deposition begins in the precuneus, medial orbitofrontal cortex, and PCC, which are all part of the DMN (Palmqvist et al., 2017). One hypothesis to this observation is that synaptic activity may drive Aβ deposition, and thus a lifetime of DMN activation would result in Aβ being deposited along the highly active DMN brain regions (Jagust & Mormino, 2011). Sepulcre and colleagues (2018) performed a longitudinal study examining how Aβ spread in HCs using another radiotracer that can measure cortical Aβ plaques in the brain, Pittsburgh compound B (PiB). Using graph theory analysis methods, they found that the PCC region acted as seed region for the spread of Aβ to neighbouring posterior and lateral parietal brain regions, namely the lateral fronto-parietal, midline frontal, and precuneus brain regions. In our study, these brain regions were all part of cluster 2 in PD. Greater Aβ burden in these regions may be working synergistically with PD-mediated dysfunctional α-synuclein and tau networks to create the conditions for PD cognitive decline (Irwin et al., 2013).
The stepwise linear regression analysis found models which explain between 14.8–49.5% of the variance in PD MoCA score across different years. Interestingly, the strongest model with the highest adjusted R2 was found for MoCA score one year after the [18F]FBB scan in the PD group, suggesting in the potential prognostic strength in using this model to assess future cognitive decline in PD patients. These findings are in line with previous studies which found that measuring Aβ levels in the CSF were a good predictor of future cognitive decline in PD (Siderowf et al., 2010). Gomperts and colleagues (2013) found that high Aβ deposition in the precuneus was not able to distinguish PD-MCI from PD-CU at time of the scan. However, patients with higher baseline Aβ deposition and PD-MCI categorization experienced a more severe cognitive decline 2.5 years after scan compared to baseline PD-CU or low Aβ burden. These findings have since been replicated with numerous other longitudinal studies showing higher baseline CSF Aβ is predictive of future PD cognitive decline and dementia risk (Compta et al., 2013; Parnetti et al., 2014; Alves et al., 2014; Bäckström et al., 2015; Terrelongue et al., 2016; Compta et al., 2016; Mondreanu et al., 2017; Caspell-Garcia et al., 2017; Buongiorno et al., 2017).
The right parietal cortex was found in the models predicting cognitive decline in PD both in the one-year and two-year follow-ups as well in the model predicting cognitive decline in HC at time of scan. The right parietal cortex is involved in attentional integration of sensory information for both halves of the body (Purves et al., 2001) and has been implicated in cognitive decline in PD using various imaging modalities. Structural MRI imaging studies have found increased cortical atrophy of the right parietal cortex in the brains of PD-MCI patients (Song et al., 2011; Melzer et al., 2013) while functional FDG-PET found reduced metabolism in the bilateral parietal cortex correlated with reduced cognitive abilities (Firbank et al., 2017; Wu et al., 2018). The correlation between the right parietal cortex in both the PD and HC groups’ with MoCA scores suggests it is a key brain region involved in Aβ mediated cognitive decline and may be particularly vulnerable to Aβ deposition.
In the HC group, the three clusters had notable differences from PD: while cluster three encompassed also the bilateral PCC, cluster two consisted of the bilateral ACC, and cluster one included the remaining 16 cortical ROIs together. Aβ deposition in neither the ACC nor the PCC had a significant correlation with MoCA score in our study. Instead, it was cluster one, specifically the right temporal cortices and right parietal cortex, which correlated with MoCA score at time of scan. This larger 16 ROI cluster found in HCs may suggest that an even distribution of Aβ deposition in the brain is likely healthier than localized deposits. Aβ plays a vital role in healthy brain function but is neurotoxic in both too high and too low amounts. If reaching a high threshold of Aβ in a brain region is pathological, a “subthreshold” distribution of the Aβ may not be so deleterious.
A linear regression model of the HC group was only found at time of scan with an R2 of 0.337 which explains just over a third of the variance in MoCA score. This model consisted of the right parietal cortex, right lateral temporal cortex and right mesial temporal cortex. The temporal cortices are involved in language processing, semantics, and memory encoding (Hickok & Poeppel, 2007; Jackson et al., 2018) which are all cognitive aspects measured by MoCA. Failing to find a linear regression model one-year and two-year post-scan suggests Aβ deposition mediated cognitive decline may work differently in HC than in the presence of PD pathology. The lack of accompanying brain pathologies in HC in the form of α-synuclein and neuroinflammation dysregulation present in PD pathology may offer a protective effect in which high Aβ levels alone may be insufficient for triggering cognitive decline.
We included only idiopathic PD patients in our sample to minimize the confounding effects genetic variants would have on our data. We did not find APOE status as a relevant variable in our linear regression analysis, however this is likely due to the small sample size of each allele variant.
When examining Aβ pathology, it is common for studies to examine an Aβ metric that measures global or composite Aβ burden in the brain as a binary marker of having too much Aβ in the brain (Aβ+) or not (Aβ-) (Seibyl et al., 2017). Many different studies have proposed different cut-off numbers of SUVR to classify Aβ+ (Villemagne et al., 2017; Fiorenzato et al., 2018; Kim et al., 2018; Melzer et al., 2019), which vary by radiotracer type, scanner resolution, disease pathology, method of comparing radiotracer uptake, and reference region (Bullich et al., 2017; Rowe et al., 2019; Doré et al., 2019). However, we propose that having a high composite level of Aβ in the brain is not telling the complete picture regarding amyloid burden in the brain. We found that having a high level of Aβ in some parts of the brain, such as the bilateral rectus gyrus, may be less detrimental to cognitive function. These brain regions are often included in a cortical composite score quantifying Aβ positivity (Mintun et al., 2006; Buddhala et al., 2015; Heeman et al., 2020), thus contributing towards Aβ + designation while also being less pathological. This may explain also why some studies have failed to find a difference in PD cognitive status between Aβ + and Aβ- groups.
The present study has some limitations that ought to be addressed. First, we only used MoCA score to quantify cognitive ability instead of a more comprehensive neuropsychological test battery that would measure cognitive ability in different domains. We used the MoCA due to its popularity as a screening tool for PD-MCI (Dalrymple-Alford et al., 2010) allowing for an easier time replicating these results in future studies. However, analyzing how specific cognitive domains are affected by cortical Aβ deposition may be an interesting follow-up. Second, there are also limitations due to the nature of the PPMI study protocol that we could not control for. The PET imaging data we used was collected from several different sites with different PET cameras which can affect the outcome measure of SUVR. We also had far more males than females in the PD group, as well as more PD-CU patients than PD-MCI especially in the follow-up years due to fact that the most pathological patients tend to drop out first, and this may have also impacted our results. Furthermore, although PPMI enrollment began with de novo PD patients, most patients were already on treatment by the time they were scanned with [18F]FBB radiotracer. Finally, we only used the Aβ values at baseline, but it would also be useful to measure Aβ in the yearly follow-ups parallel to the cognitive measurements to get a more comprehensive understanding of Aβ changes in concert with the cognitive decline.