Validation of a continuous model of amyloid‐β CSF/PET discordance across cohorts and PET tracers

Abstract Background Recently, we provided evidence that discordance in CSF vs. PET amyloid‐β biomarkers represents a natural phase in the progression of amyloid‐β pathology, with either CSF or PET biomarkers becoming abnormal first, based on well‐validated cut‐offs. We found that CSF‐/PET+ individuals had lower prevalence of APOE ‐ε4 genotype and slower rate of amyloid‐β PET accumulation compared with CSF+/PET‐ individuals [1]. Given the methodological limitations of binary cut‐offs, here we aimed to validate a continuous model of imbalance in CSF vs. PET amyloid‐β biomarkers across cohorts and tracers, and to investigate biological factors underlying discordance in CSF and PET results. Method The study included n=944 participants from two independent cohorts: ADNI and AMYPAD. We selected n=822 individuals (73.0±7.4 yr‐old, 53% female, 45% APOE ‐ε4 carriers, 16.3±2.6 yr‐education) from ADNI, ranging from cognitively‐normal to demented, with global cortical 18 F‐florbetapir‐PET and CSF‐Aβ42 measurements available within 3‐months; all participants also had CSF‐Aβ38 and CSF‐Aβ40 available. We also selected n=112 individuals (65.6±7.0 years, 59.8% female, 30.6% APOE ‐ε4 carriers, 15±3.5 yr‐education) ranging from cognitively‐normal to MCI from the AMYPAD Prospective and Natural History Study (PNHS) cohort with CSF‐Aβ42 and global cortical 18 F‐flutemetamol (n=75) or 18 F‐florbetaben (n=37) PET available within one‐year, expressed in Centiloids. In each cohort, we iteratively fitted the CSF/PET cross‐sectional data to a hyperbolic regression model with three parameters, by minimizing the Euclidean distance of the experimental points to the fitted line (Fig. 1). For each individual, we derived a measure of imbalance based on standardized residuals, with z imbalance >0 reflecting PET‐over‐CSF imbalance (z imbalance <0 reflecting CSF‐over‐PET imbalance). Results The hyperbolic model adequately fitted the CSF/PET data across cohorts and tracers (pseudo‐R 2 >0.90, Fig. 1). In ADNI, greater z imbalance (PET‐over‐CSF imbalance) was associated with lower APOE ‐e4 prevalence, and higher levels of CSF‐Aβ42, CSF‐Aβ40 and CSF‐Aβ38, corrected for demographics and disease stage. Studies in the AMYPAD cohort are ongoing. Conclusion We tested a continuous model of amyloid‐β CSF vs. PET imbalance across cohorts and PET tracers. Imbalance in CSF vs. PET amyloid‐β values carries meaningful biological and genetic information, which allows to identify heterogeneous pathological trajectories in the AD continuum , with implications for clinical trials. [1] Sala et al. (2021) Mol. Psychiatry 26(10):5864‐5874