Transforming cerebrospinal fluid Aβ42 measures into calculated Pittsburgh Compound B units of brain Aβ amyloid

Abstract

Background: Positron-emission tomography (PET) imaging of amyloid with Pittsburgh Compound B (PIB) and Aβ42 levels in the cerebrospinal fluid (CSF Aβ42) demonstrate a highly significant inverse correlation. Both these techniques are presumed to measure brain Aβ amyloid load. The objectives of this study were to develop a method to transform CSF Aβ42 measures into calculated PIB measures (PIBcalc) of Aβ amyloid load, and to partially validate the method in an independent sample of subjects.

Methods: In all, 41 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) underwent PIB PET imaging and lumbar puncture (LP) at the same time. This sample, referred to as the "training" sample (nine cognitively normal subjects, 22 subjects with mild cognitive impairment, and 10 subjects with Alzheimer's disease), was used to develop a regression model by which CSF Aβ42 (with apolipoprotein E ɛ4 carrier status as a covariate) was transformed into units of PIB PET (PIBcalc). An independent "supporting" sample of 362 ADNI subjects (105 cognitively normal subjects, 164 subjects with mild cognitive impairment, and 93 subjects with Alzheimer's disease) who underwent LP but not PIB PET imaging had their CSF Aβ42 values converted to PIBcalc. These values were compared with the overall PIB PET distribution found in the ADNI subjects (n=102).

Results: A linear regression model demonstrates good prediction of actual PIB PET from CSF Aβ42 measures obtained in the training sample (R(2)=0.77, P<.001). PIBcalc data (derived from CSF Aβ42) in the supporting sample of 362 ADNI subjects who underwent LP but not PIB PET imaging demonstrate group-wise distributions that are highly consistent with the larger ADNI PIB PET distribution and with published PIB PET imaging studies.

Conclusion: Although the precise parameters of this model are specific for the ADNI sample, we conclude that CSF Aβ42 can be transformed into PIBcalc measures of Aβ amyloid load. Brain Aβ amyloid load can be ascertained at baseline in therapeutic or observational studies by either CSF or amyloid PET imaging and the data can be pooled using well-established multiple imputation techniques that account for the uncertainty in a CSF-based PIBcalc value.

Figure 1. Flow chart of subjects

Subsets of subjects that contributed to the analysis. The royal blue and gold notes at the bottom represent the APOE ε4 carriers and non-carriers in the training set (see Figure 2). The light blue node represents the 362 subjects without PIB PET whose CSF-based PIBcalc values constitute the independent supporting sample. The distribution of PIBcalc values from the supporting sample are compared to the PIB PET distribution of the 102 subjects in ADNI who had usable PIB PET (see Figure 3).

Figure 2. PIB PET and CSF Aβ42 in Training Data Set

Figure 2a: scatterplot of global cortical PIB ratio from PIB PET images on the y-axis vs. Aβ42 from CSF on the x-axis for the 41 subjects in the training data set. Figure 2b: same data as in Figure 2a but after a log₂ transformation and shown with least squares regression lines representing the fitted conversion model. Figure 2c: Scatterplot of calculated global cortical PIB ratio (PIBcalc) derived from CSF Aβ42 vs. global cortical PIB ratio from actual PIB PET imaging on the log₂ scale. The vertical lines indicate 95% prediction intervals illustrating the uncertainty in the estimated PIB values in 8 selected individuals. Figure 2d: Bland-Altman plot of the difference between PIBcalc vs. PIB PET over the range of PIB PET values. Solid line represents the mean difference while dotted lines represent Bland-Altman limits of agreement.

Figure 3. Comparing ADNI PIB PET distribution to PIBcalc distribution in the independent supporting sample

Figure 3a: Box plots by diagnosis comparing ADNI PIB PET distribution (n = 102) values to PIBcalc values based on 362 subjects in the supporting sample. Figure 3b: Kernel density estimates of the ADNI PIB PET distribution and PIBcalc distribution from the supporting sample. The two distributions were not found to be systematically different with an AUROC of 0.52 (P = 0.49).

Appendix Figure. Conversion Model Illustration

Illustration of the idea that there is not one conversion model with fixed parameters but a distribution of models. In practice, when a CSF Aβ value is converted to PIBcalc, a number of PIBcalc values will be generated based on this distribution of models. Each distribution shown above is centered at the least squares estimates with a standard deviation approximately equal to the standard error from the least squares fit. The intercept which we denote by β₀ has a distribution centered at the least squares estimate of 5.326. Similarly, the CSF Aβ (β₁) and the APOE (β₂) coefficients will the centered at -0.615 and 0.184. The differences in the standard errors among the coefficients are reflected in the differences in the heights and widths of the distribution curves. The error term (e) will be centered at 0 with a standard deviation of approximately 0.180, the residual standard error from the least squares fit.