In this study we employed two approaches to classify oNC subjects based on PIB index values. An iterative outlier approach using index values from the oNC group yielded a cutoff value of 1.16 (IO-cutoff). An approach using data from a sample of yNC subjects (age range 20-30, who are likely to lack Aβ deposition and hence provide a good estimate of PIB negativity), revealed a lower cutoff value of 1.08 (yNC-cutoff). oNC subjects falling above the IO-cutoff (PIB+ oNC; 11/75) showed substantial overlap with AD subjects across all examined measures of PIB uptake. Subjects falling between the 2 cutoffs were less obvious and labeled as ambiguous (Ambig oNC; 15/75), and the remaining subjects were considered PIB- oNCs (49/75). Further examination of these oNC groups across different PIB quantification measures consistently confirmed the intermediate status of Ambig oNC subjects, suggesting that this categorization was not merely an artifact of the employed classification approach. Furthermore, ambiguously elevated values were consistently more prevalent than analogous decreases and present in regions known to show amyloid deposition, suggesting that slightly elevated values do not solely reflect noise. Interestingly, a number of PIB- subjects showed evidence of elevation across the different measures, suggesting that a subset of PIB- subjects may also have elevated Aβ burden. Overall, the analyses presented in this manuscript, in conjunction with recently published pathological data, suggest a biological relevance of slight PIB elevations in oNCs.
4.1 An approach that combines magnitude and extent may be most sensitive
Qualitative comparison across the measures used to quantify PIB reveals that the number of high ROIs has the clearest stepwise progression from young NC to PIB-/Ambig/PIB+ oNC groups and AD, perhaps because the combination of magnitude and extent makes this measure less inherently noisy than the other measures examined. For instance, the maximum ROI approach identified a yNC subject whose value was much higher than all yNC and PIB- oNC subjects, as well as the majority of Ambig oNC subjects (1.44 in the left caudal anterior cingulate). However, the variance in the yNC group was substantially decreased using the total high ROI approach, and this subject's tally fell below all Ambig oNC as well as many PIB- oNC subjects. Therefore, it is possible that multiple ROIs should be considered to reduce noise in PIB classification. On the other hand, a global approach—such as the PIB index approach—may lose sensitivity by averaging too many regions. Overall, it will be necessary to apply these methods to independent cohorts to determine the generalizability of these approaches.
4.2 Slight elevations outnumber slight decreases
An obvious possibility is that slight elevations simply reflect noise amongst subjects lacking Aβ deposition. Noise within PIB-PET scans may be caused by bleed-in effects of nonspecific white matter binding, wash-out effects of neighboring cerebrospinal fluid, co-registration errors between MRI-defined ROIs and PIB PET scans, inaccurate labeling of gray matter ROIs, subject motion and errors during PET acquisition/reconstruction. It is difficult to predict how these factors will interact to affect signal, but one possibility is that these factors would simply produce noise without any bias towards higher or lower values. Thus, a higher prevalence of slight increases compared to slight decreases would support the claim that slightly elevated values reflect a biologically relevant signal. To this end, we contrasted the number of instances of slight elevations to the instances of analogous decreases. We excluded PIB+ oNC subjects to focus on the ambiguous signals present in Ambig and PIB- oNC. Across all 3 measures examined (high versus low PIB index values, right versus leftward shift in ROI distributions, and the total number of high versus low ROIs), there were greater numbers of ambiguous elevations compared to analogous reductions. This asymmetry argues that these slight elevations contain a meaningful signal rather than merely reflecting noise in the data.
4.3 Spatial distribution of elevated regions follows known pattern of amyloid deposition
It is also possible, however, that the noise in the PIB signal is positively biased towards higher cortical uptake because high white matter binding of the tracer is the predominant factor. Thus, we investigated the spatial pattern of elevated regions amongst oNCs by computing the percentage of subjects with high ROIs for the 68 FreeSurfer regions investigated. These percentages were plotted for the entire oNC group, and also separately for PIB+, Ambig and PIB- oNC subjects. This analysis revealed a pattern consistent with known patterns of amyloid deposition as measured with PIB-PET imaging (Fripp et al., 2008
; Mintun et al., 2006
) and postmortem staining (Braak and Braak, 1991
; Thal et al., 2002
). This regional specificity is difficult to explain by elevated white matter binding. Specifically, elevated PIB across subjects was seen across multiple association cortices, with the highest percentages in medial orbital, dorsolateral prefrontal, and temporoparietal cortices. Within-group examination revealed diffuse elevation in PIB+ oNC, with a more restricted pattern in Ambig oNC subjects. Elevated uptake was even present for a subset of PIB- oNC subjects (dorsolateral prefrontal, medial orbital and temporoparietal cortices), suggesting the presence of amyloid deposition in PIB- oNC subjects. Importantly, the pattern of regional elevation across these 3 levels of PIB uptake resembles the stages of amyloid deposition described by postmortem staining studies (Braak and Braak, 1991
; Thal et al., 2002
). Specifically, our analysis revealed the greatest vulnerability to Aβ burden in neocortical heteromodal regions (prefrontal and temporoparietal cortices), followed by the cingulate gyrus and medial temporal cortex whereas the least vulnerable cortical regions were unimodal sensory cortices. Overall, concordance between the spatial distribution of elevated PIB in this study and established patterns of Aβ deposition further strengthens the claim that these slight elevations are indicative of Aβ deposition.
4.4 Slightly elevated PIB values and postmortem data
Recent research from the Baltimore Longitudinal Study of Aging suggests that slightly elevated PIB indices may correspond to Aβ deposition measured at postmortem examination (Sojkova et al., 2011
). The authors describe a series of 6 oNC subjects that underwent both PIB imaging and postmortem examination. Although the PIB values are not directly comparable to the values reported in our study (due to inter-scanner differences, data processing methods, etc), the pattern presented in their research parallels the findings presented in our manuscript. Three cases in the Sojkova series had low to slightly elevated PIB index DVR values (1.01, 1.06 and 1.09; cases B, C and D respectively), as well as postmortem evidence of plaque deposition (all had a CERAD rating of moderate). In contrast, case A had a PIB index of 0.96 and no evidence of plaques at postmortem examination. In addition to reporting global PIB index values, PIB values for 9 ROIs were also examined. Interestingly, the maximum ROI value for the “Aβ -negative” case A was 1.03, whereas the maximum values for “Aβ-ambiguous” cases B-D were 1.16, 1.24, and 1.21 respectively. The PIB profile seen in cases B-D is reminiscent of the Ambig oNC reported in our study. The observation that these 3 ambiguous cases have postmortem confirmation for the presence of amyloid deposition greatly strengthens the argument that slightly elevated PIB values in oNC may reflect a biologically relevant signal.
4.5 Future approaches are needed to establish the relevance of ambiguous cases
The results presented in this manuscript are largely descriptive—it is clear that follow up studies are needed to determine whether ambiguously elevated PIB values represent a biologically meaningful signal. Future studies that directly compare PIB to post-mortem measurements of Aβ in slightly elevated cases are of utmost importance in determining the sensitivity of PIB imaging. It is likely that these studies will primarily draw from patient populations with low levels of Aβ since this type of data is extremely difficult to obtain in cognitively normal individuals. Longitudinal PIB-PET imaging in oNCs will also offer insight into the relevance of Ambig cases. For instance, longitudinal examination of regions showing high uptake will reconcile whether these regional elevations reflect noise or meaningful signal. It is also likely that Ambig oNCs are at risk for transitioning to PIB+, and longitudinal PIB imaging will be able to confirm or refute this speculation. More immediately, it is possible that interpretation of Ambig oNC subjects as PIB- may lead researchers to falsely accept a lack of difference between PIB+ and PIB- oNC. Overall, the potential relevance of these slightly elevated oNC subjects may help guide future studies examining PIB uptake in oNC populations.
4.6 Atrophy correction
A limitation in this study is the lack of atrophy correction to PIB-PET data. In previous publications our group has employed a 2-compartmental atrophy correction procedure to PIB data (Mormino et al., 2009
; Rabinovici et al., 2010
), however we choose to refrain from this procedure for suspicion that some older subjects may have high levels of atrophy without any amyloid deposition (for instance, atrophy may be due to vascular etiologies (Raz et al., 2007
)). During an initial analysis of this data we applied a 2-compartmental atrophy correction to yNC and oNC PIB scans, and found that the majority of oNC subjects were above the yNC PIB index cutoff (data not shown), and have assumed that this result represents overcorrection. Consequently, the lack of atrophy correction in this manuscript may in fact underestimate the prevalence of slightly elevated levels of amyloid in oNC (ie. the signal from amyloid must first overcome the washout effects of atrophy). Interestingly, this point may be applicable to the PIB- AD case presented in this manuscript (PIB index=1.03). Although this PIB index value falls amongst yNC and PIB- oNC values, this subject has cortical atrophy comparable to the 4 oNC subjects with the lowest PIB index values (0.88, 0.93, 0.96 and 0.98). Thus, it is unclear whether this PIB- AD case is truly Aβ negative, or has evidence for elevated uptake compared to “atrophy-matched” PIB- oNC subjects. The strategy of “atrophy matching” cases to assess slight elevations amongst oNC and patients against subjects with high atrophy that are confirmed to be amyloid-free (via postmortem examination, cerebrospinal fluid, etc) may be a promising alternative to traditional atrophy correction techniques.
A further complication is the possibility that increased signal amongst Aβ-negative subjects may be due to bleed-in from nonspecific white matter binding. A 3-compartmental model that accounts for PIB binding in white matter may address this issue, however, given the increased susceptibility to segmentation and registration errors associated with this approach (Meltzer et al., 1999
), we opted to refrain altogether. It is important to note that the influence of nonspecific white matter binding amongst Aβ negative cases would result in a higher young derived cut off value. Although bleed-in effects of white matter may likewise elevate the PIB signal in Aβ-free oNCs, we suspect that the washout effects of CSF in oNCs due to age-related atrophy will have a larger impact than bleed-in effects of white matter (ultimately resulting in underestimation rather than overestimation amongst oNCs). Therefore, it is likely that the lack of atrophy correction in this manuscript underestimates the quantity of elevated PIB values amongst oNCs.
4.7 Arbitrary nature of derived cut offs
In this study, we identified a cut off of 1.08 using data from a cohort of presumably Aβ-free young control subjects. This value was not affected by employing a different image processing pipeline, suggesting stability of this estimate within our young cohort (a situation of low variability due to young age, structural homogeneity, and negative scans of the subjects). This cut off value is below the cut offs typically used in PIB-PET imaging studies, which tend to isolate oNCs with AD-levels of PIB uptake. However, it is likely that the same methods applied to an independent cohort may yield a slightly different cutoff value. Therefore, it is important to keep in mind that the exact number derived from the methods herein represent an arbitrary value that is likely influenced by scanning parameters, image resolution, subject sample, etc, and may not be applicable to data from other laboratories. Thus, we do not posit that our methods circumvent the arbitrariness that accompanies demarcation of a continuous PIB index variable, but rather provides support for a biological relevance in values that fall beneath the levels that are typical in the context of AD.
It is possible that incorporation of test-retest reliability values may minimize the arbitrary nature inherent within PIB positivity cut off values. For instance, we established test-retest reliability in a subset of 14 oNCs scanned an average of 2.77 years apart and found that the average percent difference between PIB index values was 2.63%, consistent with previous test-retest reports (Lopresti et al., 2005
). Given this test-retest variability, a more conservative approach would have been to eliminate subjects falling within 2.63% of our young defined cut off (ie: 1.08+/-2.63%=1.05 to 1.11). However, the lack of a gold standard for determining whether these slightly elevated values are indicative of Aβ makes it difficult to determine whether this sort of approach is optimal, and should be addressed in follow up studies.