The distributions of both fibrillar amyloid plaques and tau were skewed, as would be expected if these two biomarkers reflect underlying pathology in some of these cognitively healthy participants. Only a small number of people are in the beginning stage of the disease and therefore have higher values of these two indices. In contrast, the distribution of CSF Aβ42 was not skewed. Instead, it was not significantly different from a normal distribution. Normal distributions are more often seen in individual difference variables such as height. Perhaps there are two influences on CSF Aβ42 level. One may reflect normal human biological variability, whereas the other indicates onset of pathology. The median value of CSF Aβ42 was slightly less than the mean (594 vs. 626; index of skewness = 0.40, see ). Whether this is due to sampling variability or indicates that CSF Aβ42 levels have dropped from prior levels for a small portion of the participants can only be determined by future studies, particularly longitudinal ones.
CSF Aβ42 and tau were not correlated linearly. There was a very small nonlinear relation between these two biomarkers indicating only 4% shared variation. Although we cannot find any report in the literature referring to this correlation in cognitively normal people, we expected it to be negative (low CSF Aβ42 associated with high CSF tau) if both CSF markers occurred as described by the amyloid cascade hypothesis. As indicated in the introduction, that hypothesis suggests that Aβ42 peptides aggregate to form amyloid plaques which, in turn, lead to synaptic loss and cell death, reflected in elevated CSF tau, thereby causing cognitive impairment.
The model observed in the analyses reported here indicates, as suggested previously (Holtzman et al., 2011
; Hyman, 2011
; Pimplikar, 2009
; Small and Duff, 2008
), that there are at least two independent processes, one represented by CSFAβ42
and the other by CSF tau, related to fibrillar amyloid plaque burden. These two biomarkers, and the processes they represent, were associated with an impressive amount of the variance in PIB binding (R2
= .60) in this cognitively normal sample, 17% of whom had MCBP values ≥ .18. This value was used previously to describe individuals as PIB positive (Fagan et al., 2009
; Mintun et al, 2006
) or having fibrillar amyloid plaque burden similar to those with AD. Similarly, Shaw et al. (2009)
reported that these two biomarkers made independent contributions to the differentiation of autopsy-verified AD cases from normal controls. There may be, however, still more processes contributing to plaque formation or, perhaps more important, to brain atrophy and dementia (Rowe et al., 2010
). In this cross-sectional sample fibrillar amyloid plaque formation had only a spurious relation with brain atrophy via their mutual association with age.
In addition to their main effects on PIB, there was also an interaction between CSF tau and Aβ42
. The correlation between CSF tau and PIB was much stronger when Aβ42
was low than when Aβ42
was moderate or high. This is analogous to the text book example of an interaction provided by Cohen et al. (2002)
showing a strong negative correlation between age and endurance on a treadmill that is much weaker in people with a history of aerobic exercise.
Multiple previous studies of these two CSF biomarkers in cognitively normal samples with a wide age range have alluded to the possibility of their interaction when they examined how the ratio of tau to Aβ42
was related to other variables. Long ignored by researchers in many fields, Pearson (1897)
pointed out over a century ago that the ratio of two variables can produce spurious results. The ratio actually represents a combination of the main effects of each of the variables in the ratio as well as their interaction. The analysis reported here used the appropriate procedure by first entering the main effects of Aβ42
and tau followed by their product, representing the interaction, in a subsequent step of a hierarchical regression analysis (Kronmal, 1993
). Each CSF biomarker was associated with unique, unrelated portions of the variance in PIB, and their interaction was associated with an additional, unrelated portion. This result, along with the minimal association between the two CSF biomarkers, again suggests the operation of at least two independent processes, one associated with lower levels of CSF Aβ42
and one associated with increased levels of CSF tau. One does not cause the other, but when both are present the effect is enhanced. A recent study (Desikan et al., 2011
) identified an interaction between Aβ42
and ptau in the prediction of longitudinal changes in the volume of the entorhinal cortext in healthy controls and individuals with amnestic mild cognitive impairment.
The effects of APOE and age on plaque burden were indirect through CSF Aβ42 and tau. It should be noted, however, that APOE and age together accounted for only 13% of the variance in Aβ42 and 14% of tau. There clearly are other important variables to be identified in explaining the processes associated with CSF levels of Aβ42 and tau, and their inclusion may change the model.
Brain volumes in the eight ROIs were uncorrelated with CSF measures or with PIB after controlling for age. As noted in the introduction, previous studies have reported mixed results in efforts to relate PIB to brain structure. There are likely numerous reasons for the differences. One involves the adjustment for age, which has not always been done. Others probably relate to brain regions examined, the measure of structural integrity (e.g., cortical thickness vs. volume), age range of the sample, and inclusion criteria. One study (Chetelat et al., 2010
), for example, found no correlation when people with subjective memory complaints were excluded. Similarly, in a study based on neuropathological data (Price et al., 2001
) little atrophy was observed in those without a diagnosis of at least very mild symptomatic AD.
The strengths of the current study include the wide age range and large sample size of carefully characterized cognitively normal individuals for whom a number of the biomarkers shown previously to be present prior to clinical diagnosis of AD were obtained in relatively close temporal proximity. A primary limitation is its cross-sectional nature, which allows only modeling of potential causative relations. These relations may be affected if individual biomarkers become abnormal at different times in the AD process. Further, reliability of some of the measures used in the analyses has not been well studied; failure to detect relations may reflect measurement error. A similar limitation is that PIB measures only fibrillar Aβ deposits; the model does not address earlier potential processes such as Aβ oligomerization and aggregation into diffuse plaques. Because the biology of AD undoubtedly is complex and may not be well-represented by the assumptions of our model, any potential causative relations must be confirmed by longitudinal studies. Our findings do, however, highlight the importance of considering multiple contributors to AD pathogenesis as that longitudinal research is pursued.