1 BACKGROUND

The World Health Organization (WHO) estimates that more than 55 million people are living with dementia, and this number is predicted to rise to 139 million by 2050. In 2019, the estimated global cost of dementia was US$ 1.3 trillion.¹ Alzheimer’s disease (AD) is a continuum, which can be categorized by biomarker status according to the ATN classification system.² This system rates individuals upon the presence of amyloid β (Aβ) alterations in cerebrospinal fluid [CSF] or positron emission tomography [PET] as “A”, hyperphosphorylated tau (CSF or PET as “T”), and neurodegeneration (atrophy on structural magnetic resonance imaging [MRI], fluorodeoxyglucose [FDG]–PET, or CSF total tau as “N”). In addition, blood biomarkers have emerged and include candidates such as Aβ_1-42/Aβ_1-40 ratio, phosphorylated tau (P-tau), neurofilament light (NfL) chain, and glial fibrillary acidic protein (GFAP). Elevated levels of plasma P-tau181, P-tau217, and P-tau231 were indicative of prodromal and mild cognitive impairment (MCI) stages and predicted amyloid and tau pathology, whereas NfL has shown a high correlation with neurodegeneration in general, lacking AD specificity.^3–11 GFAP exhibited the ability to predict dementia and AD.^12–15 Moreover, combinations of Aβ_1-42/Aβ_1-40 ratio, GFAP, and NfL revealed the potential to identify cerebral amyloidosis and/or disease severity.¹⁶ These biomarkers have been quantified by immunoassays or mass spectrometry–based methods, which both detect low protein concentrations in blood plasma.^15–19 The single molecule array (Simoa) technology is an established immunoassay that we used in our analyses to determine P-tau181, GFAP, and NfL levels.

Complementary to widely used concentration-based protein biomarkers, the misfolding of Aβ in the initial phases of the disease in peripheral fluids has been established as a structure-based biomarker.^20–22 Misfolding and aggregation of native soluble forms into oligomeric and fibrillar, beta-sheet enriched structures are thought to be crucial in the development and progression of AD, where beta-sheet enriched species form amyloid plaques.^23, 24 The immuno-infrared sensor assay provides the only technology to directly measure Aβ misfolding in blood plasma and has been validated previously.^22, 25, 26 Plasma Aβ misfolding was able to predict the risk of clinical AD diagnosis up to 14 years in advance in a population-based cohort^25, 27 and up to 6 years in a cohort comprising participants with subjective cognitive decline.²⁶ In addition, Aβ misfolding in plasma in combination with plasma Aβ ratios led to higher disease prediction accuracy.²⁶ Because recent evidence has shown that plasma biomarkers identify pathological changes more than a decade before clinical manifestation, early risk prediction is crucial for successful therapy.^28–30

In this study, we compared the Aβ structure–based biomarker performance with P-tau181, GFAP, and NfL levels in plasma to predict clinical AD diagnosis within 17 years in a population-based cohort. In addition, the genetic risk marker apolipoprotein E gene (APOE) was considered when combining biomarkers.

2 METHODS

3 RESULTS

In this study, plasma AD biomarkers were assessed at baseline in a subset of participants from the ESTHER cohort (n = 308; 68 incident AD cases and 240 controls) (Figure 1). Table 1 summarizes the study population baseline characteristics. Table S1 shows in addition the participant characteristics compared to the overall ESTHER study. AD participants were on average 69 years of age, whereas controls were on average 66 years of age at baseline. Furthermore, 63% of AD subjects and 53% of controls were female. APOE ε4 genotype was positive in 49% of the AD group and 28% of the control group.

Aβ misfolding was determined by immuno-infrared measurements. AD cases had a mean amide I maximum frequency of 1641 cm^–1 (SD ± 4 cm^–1), whereas controls had a mean amide I maximum frequency of 1646 cm^–1 (SD ± 4 cm^–1). AD cases had significantly lower immuno-infrared sensor readout values as compared to controls, which means participants with an AD diagnosis within 17 years had a significantly higher degree of Aβ misfolding in plasma at baseline (P < .001) (Figure 2). Of interest, 65% of the AD cases diagnosed 9 to 17 years after study entry showed increased pathological misfolding of Aβ, compared to 71% of those diagnosed 0 to 9 years after study entry.

AD cases showed significantly increased mean levels of P-tau181, with 2.3 pg/mL (SD ± 1.4 pg/mL) compared to controls with 1.9 pg/mL (SD ± 1.0 pg/mL) (P < .01) (Figure 3A). Significant differences in mean concentrations of GFAP between AD cases with 159.0 pg/mL (SD ± 111.1) and controls with 99. 6 pg/mL (SD ± 46.7 pg/mL) were also evident (P < .001) (Figure 3B). In addition, NfL concentrations were increased in AD subjects (23.9 pg/mL SD ± 11.8 pg/mL) as compared to controls (18.9 pg/mL SD ± 9.4 pg/mL) at baseline (P < .001) (Figure 3C). Logistic regression results revealed that Aβ misfolding per SD decrease had the highest OR for incident AD, with 4.24 (95% CI 2.68–6.67), followed by APOE (≥1 ε4 allele vs none) with 2.36 (95% CI 1.32–4.24) and GFAP per SD increase with 2.08 (95% CI 1.44–3.01) (Table 1). In Contrast, NfL (1.37 95% CI 0.98–1.91) and P-tau181 per SD (1.25 95% CI 0.93–1.68) showed lower ORs.

The predefined discriminative threshold of ≤1642 cm^–1 for Aβ misfolding indicating a pathological biomarker transition toward AD. ROC analysis revealed the highest AUC of of all biomarkers measured: 0.78 (95% CI 0.71–0.85) (Figure 4A). Among the concentration markers, GFAP showed the highest AUC value (AUC 0.74, 95% CI 0.67–0.82), followed by NfL (AUC 0.68, 95% CI 0.61–0.75) and P-tau181 (AUC 0.61, 95% CI 0.53–0.70) (Figure 4A). Differences in concentration markers when categorized in misfolding positive and negative for AD cases and controls are provided in Figure S1. The distribution of concentration biomarker quartiles and Aβ misfolding status at baseline in incident AD cases and controls can be found in supplementary Table S2.

The combination of APOE and Aβ misfolding increased the AUC slightly to 0.80 (95% CI 0.73–0.86) (Figure 4B). AUC was further increased to 0.83 (95% CI 0.76–0.89) when GFAP and Aβ misfolding were combined. However, combining all three markers did not further improve the AUC (AUC 0.83, 95% CI 0.77–0.90) (Figure 4B). In addition, DeLong analyses showed that neither concentration markers nor APOE combined with Aβ misfolding led to a statistically significant difference between ROC curves of biomarker combinations and Aβ misfolding alone. GFAP + Aβ misfolding had the highest prediction accuracy of incident clinical AD conversion within 17 years. Adding APOE did not further improve the AUC.

4 DISCUSSION

Studies on predictive biomarkers for AD utilizing population-based cohorts, including symptom-free individuals, are still lacking. Here we presented results from a sub-study of the ESTHER cohort, a large population-based cohort study of older adults from Germany. The presence of Aβ misfolding, which had recently been verified and validated as an AD-specific biomarker,^{20–22, 25, 26} showed higher accuracy than concentration markers—P-tau181, GFAP, or NfL—to predict AD diagnosis within 17 years of follow-up. The combination of Aβ misfolding and GFAP exhibited the highest AD-prediction accuracy.

71% of the study participants who were diagnosed after 0 to 9 years showed Aβ misfolding, whereas 65% of participants who were diagnosed with AD between 9 and 17 years after baseline collection had increased misfolding. This suggests that Aβ misfolding could be a pre-screening biomarker for risk of clinical AD conversion up to 17 years before diagnosis. Due to the approval of a disease-modifying therapy, many more people will need to be screened periodically starting at about 60 years old. This helps to identify time points of biomarker changes and start of interventions based on blood tests.  

Most recently, elevated levels of P-tau181 showed great potential for predicting amyloidosis and tau pathology.^3, 4, 7 However, the performance of this biomarker in our study was at 17 years inferior to GFAP and NfL, with an AUC of 0.61, although levels were elevated in manifested AD cases compared to controls, as shown in other studies.^{3, 4, 7, 9, 10} We have shown previously that P-tau181 levels were associated with risk of clinical AD incidence only within 9 years of diagnosis.⁴³ Here, we identified 22 AD cases without Aβ misfolding, but with already elevated P-tau181 levels on average compared to controls (Figure S1A). Because Aβ misfolding was negative, these individuals might have non-AD pathological changes instead of being in the Alzheimer’s continuum, as suggested by the ATN classification system. Considering, that Aβ misfolding is currently not included in the ATN system in contrast to PET or CSF ratios, this biomarker could be added in the future to amyloid “A” section, not only but also because of the AD specificity. Further analyses revealed that the largest group of AD cases with positive misfolding status had rather low P-tau181 values (Table S2). This might be an indication that Aβ misfolding occurs before P-tau181 rises in plasma. However, a negative correlation of P-tau levels with Aβ misfolding was significant for the individuals with incident AD within 9 years (P < .01; r = −.53). Therefore, the question of which biomarker alteration occurs first in the disease progression needs further investigation.

NfL is a general neurodegenerative marker and not specific for AD. It did not improve the disease-prediction accuracy of AD diagnosis. Nineteen of 23 individuals (83%) with positive misfolding status and NfL levels in the highest quartile developed AD (Table S2). However, 40% of AD cases were Aβ misfolding positive but had no elevated NfL values. This also supports the idea that Aβ misfolding may be an early and NfL a later risk prediction marker. Our study with biomarker measurements at baseline provides insight into the risk of developing AD with 17 years of follow-up, whereas other studies had shorter follow-up time, only up to 11 years,^{3, 5, 7, 11, 12} or other types of AD such as familial cases,^44, 45 which may not be representative of the general population.

As a marker of astrogliosis, GFAP is currently of research interest for analyzing AD-specific associations.⁴⁶ In our study, GFAP levels were significantly higher in participants who were diagnosed with AD within 17 years. However, whether an increased GFAP level is AD specific needs further investigation. Recent study reporting has indicated that this is not necessarily the case.^46, 47 But the performance of GFAP alone is in concordance with recent results.^43, 46, 48 Of interest, there were 24 participants who had GFAP levels within the highest quartile and who were also Aβ misfolding positive at baseline, of whom 22 (92%) went on to receive an AD diagnosis during follow-up (Table S2).

In our study, a combination of Aβ misfolding and GFAP levels showed especially good predictive potential. Aβ misfolding alone showed the best discriminative performance and is on the same level as recently published plasma Aβ assays of prodromal and MCI stages.¹⁷ Furthermore, the combination of Aβ misfolding and GFAP brings an added value to the disease-prediction accuracy of incident AD within 17 years. This might be because elevated GFAP levels indicate abnormal activation of astrocytes that often surround amyloid plaques and may therefore be associated with Aβ misfolding.

The addition of APOE as a genetic risk factor did not improve disease-prediction accuracy. This could be because genetic predisposition, calculated from APOE genotype, indicates only increased AD risk. Blood-based biomarkers, however, directly indicate pathological changes that lead to neuronal loss and the development of the disease. In our study, Aβ misfolding had the strongest prognostic ability, as it directly detects the postulated first pathological process, Aβ misfolding and oligomerization, the base for plaque formation.²⁸ In addition, the measurement provides an advantage compared to quantitative assays with enabling analyses of heparin plasma samples. Furthermore, the misfolding measurement is a relative measure and the readout of the amide I maximum position is not dependent on concentration.

There are some limitations to this study. First, the dementia diagnoses may be inaccurate and not all participants visited a neurologist or other specialist. However, this characteristic of our community-based cohort study may better reflect real-life practice of AD diagnosis and care in community settings than cohorts conducted in specialized academic settings with highly selective study populations. Furthermore, another weakness is that Aβ_1-40 and Aβ_1-42 values were not measurable in Simoa analyses, thereby limiting comparability between structural and concentration changes of Aβ. One possible cause could be that heparin inside the collection tubes became unstable during long storage and precipitated during thawing. Because Aβ has a consensus sequence for binding heparin and the samples were centrifuged before Simoa measurements, most of the Aβ species may have been pelleted during processing and were therefore no longer in the solution.⁴⁹ Here, the immuno-infrared sensor revealed an advantage, since preprocessing was unnecessary and structural changes of the whole Aβ fraction could be analyzed regardless of the type of blood- collection tube. Other studies have recommended using EDTA tubes instead of heparin because of the analyte recovery.⁴⁰ Unfortunately, the ESTHER study has insufficient EDTA.

5 CONCLUSION

In summary, we presented results from a community-based cohort with AD diagnoses after 17 years of follow-up. A combination of Aβ misfolding and GFAP showed the best discrimination between AD and controls and the greatest potential for risk stratification. By the correct identification, these participants then avoid invasive and costly diagnostic tests like lumbar puncture or PET imaging. Our approach could lead to a non-invasive and cost-effective multifactorial diagnostic tool for prescreening older adults regarding the risk of developing AD. In further studies in which EDTA plasma samples were collected, it is also conceivable to expand the biomarker panel to include Aβ_1-40 and Aβ_1-42, and the Aβ_1-40/Aβ_1-42 ratio to improve the assay and biomarker panel. By identifying those individuals at high risk of AD development, disease-modifying therapies could be administered early in the disease’s progression, thereby preventing symptomatic clinical AD. These findings must be validated in additional longitudinal studies like ESTHER, preferably with PET and/or CSF data available, to confirm and enclose the time point when biomarker alterations appear.