Evaluation of Blood-Based Plasma Biomarkers as Potential Markers of Amyloid Burden in Preclinical Alzheimer’s Disease

INTRODUCTION

Alzheimer’s disease (AD) is a common neurodegenerative disease associated with aging. In the United States, more than 6.2 million individuals over the age of 65 are affected by AD dementia, a number that is expected to increase to 13.8 million by the year 2050 [1]. AD is thought to begin with a predementia stage that slowly progresses to mild cognitive impairment (MCI) then to symptomatic AD [2]. Older clinically normal (CN) individuals with evidence of amyloid pathology (Aβ+) demonstrate more rapid declines on longitudinal neurophysiological testing results; more rapid decline in cognitive function; and at a greater risk of progression to mild cognitive impairment (MCI) and dementia [2].

The current state of AD diagnostics allows for asymptomatic or CN individuals to be diagnosed with preclinical AD by examining the presence of Aβ, now detectable with neuroimaging modalities such amyloid positron emission tomography (PET) and cerebrospinal fluid (CSF) analysis. These methods are traditionally employed to indicate the presence of brain amyloid plaques in CN individuals. Numerous studies have correlated elevated brain Aβ levels and lower CSF Aβ₄₂/Aβ₄₀ levels with cognitive decline [3–5]. However, these methods are expensive, not readily accessible in primary care settings, time consuming, and invasive [6–8]. These limitations have sparked researchers to examine the feasibility of using blood biomarker measures (e.g., Aβ₄₀, Aβ₄₂, Aβ₄₂/Aβ₄₀, tau, phospho-tau) to detect amyloid pathology as blood sample collection is procedurally simple and readily accessible in primary care settings to minority populations in rural areas, the elderly, and those of diverse socioeconomic groups [9, 10]. Moreover, blood-based biomarkers may serve as a more cost-effective screening tool for patient enrollment into preclinical AD trials. Early detection is considered critical, in that AD-modifying interventions designed to stop or slow disease progression during asymptomatic stages are likely to be most effective [11]. Lastly, identifying and treating a cognitively impaired patient at earlier and more treatable time points may alleviate the financial hardships and caregiver burden imposed on close relatives and on the healthcare system [12].

In 2014, Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease (A4) Study, a first-of-its-kind secondary prevention trial was initiated to assess the efficacy of a preclinical drug intervention in CN older individuals who were “at risk” for progression to AD [13, 14]. Participant eligibility into the A4 Study was determined by amyloid PET imaging (elevated Aβ+versus non elevated Aβ-). Given the advantages of using blood-based biomarkers over PET imaging [9, 15–17], we investigated whether blood-based biomarker screening tests can predict brain amyloid pathology in the pre-randomization A4 participants. Quantifying plasma amyloid biomarkers has the potential to improve the efficiency of screening and enrolling participants into clinical trials similar to the A4 Study [11, 18, 19]. Over the past few years, accumulating evidence has demonstrated that lower plasma Aβ₄₂/Aβ₄₀ ratios, as measured using mass spectrometry (MS)-based methods, are associated with brain amyloid pathology, accelerated cognitive decline, and increased risk of developing AD dementia in CN, older adults [20, 21].

In the current study, we examined whether plasma Aβ₄₀, Aβ₄₂, and Aβ₄₂/Aβ₄₀ levels, as measured by two different MS platforms (MALDI-TOF-MS and LC-MS/MS), can predict brain amyloid PET positivity in cognitively normal, A4 Study participants. Secondly, we compared the impact of two different plasma processing protocols (2 h versus 24 h) on the plasma Aβ₄₀, Aβ₄₂, and Aβ₄₂/Aβ₄₀ levels to predict amyloid PET positivity.

METHODS

RESULTS

Two hundred and twenty-four EDTA plasma samples processed within 2 h (Protocol 2) were shipped frozen and analyzed for Aβ₄₀ and Aβ₄₂ concentrations (pg/mL) and Aβ₄₂/Aβ₄₀ concentration ratio at the Shimadzu and C₂ N labs. Participants’ baseline demographics and amyloid biomarker protein concentrations determined using both analytical platforms are summarized in Tables 1 and 2, overall and by amyloid PET status.

The ability to identify brain amyloid status using plasma Aβ₄₂/Aβ₄₀ quantified using MALDI-TOF-MS (Shimadzu) or LC-MS/MS (C₂ N) was evaluated using ROC curve analysis (Fig. 1). Plasma Aβ₄₂/Aβ₄₀ quantified using MALDI-TOF-MS identified brain amyloid PET status with AUC-ROC=0.80 (95%CI = 0.75 –0.86), and LC-MS/MS analysis identified brain amyloid PET status with AUC-ROC=0.76 (95%CI = 0.69 –0.82) (p = 0.13 for ROC comparison by DeLong). Individual concentrations of plasma Aβ₄₂ and Aβ₄₀ were quantified using both analytical platforms. MALDI-TOF-MS analysis identified brain amyloid PET status with AUC-ROC=0.70 (Aβ₄₂) and 0.56 (Aβ₄₀) (95%CI = 0.64 –0.77). LC-MS/MS analysis of Aβ₄₂ and Aβ₄₀ identified brain amyloid PET status with AUC-ROC=0.64 (Aβ₄₂) and 0.51 (Aβ₄₀) (95%CI = 0.57 –0.71). Shimadzu lab also provided an additional marker; the Composite blood plasma marker score (APP669–711/Aβ₄₂ and Aβ₄₀/Aβ₄₂) which returned an AUC of 0.82 (95%CI 0.76 - 0.87).

Fig. 1

Receiver operating characteristic (ROC) curves for the blood plasma samples (n = 224) Processed using Protocol 2 (2 h post phlebotomy) and analyzed at C₂ N and Shimadzu laboratories (left). The corresponding area under curve (AUC) estimates and 95%confidence intervals (right).

Aβ cut points identified from the Youden Index (see Table 3) were used to generate scatter plots of Aβ against PET SUVr for C2 N and Shimadzu under Protocol 2 (2 h processing) (Fig. 2). Shaded regions represent the areas where the plasma biomarker correctly identifies patients as amyloid negative (green) and amyloid positive (orange). The lighter color shaded between 1.11 and 1.15 SUVr is the region where a participant was considered amyloid positive only when the corresponding visual read was also considered positive by a two-reader consensus determination.

Fig. 2

Scatter plots of Aβ against PET SUVr for C2 N and Shimadzu under Protocol 2 (2 h post phlebotomy). Shaded regions represent the areas where the plasma biomarker correctly identifies patients as amyloid negative (green) and amyloid positive (orange).

Plasma Aβ₄₂/Aβ₄₀ was also quantified by LC-MS/MS in a subset (n = 100) of the 224 participants. These randomly selected blood samples were processed using both protocols (Protocol 1: blood processed to plasma 24 h after phlebotomy versus Protocol 2: blood processed to plasma within 2 h after phlebotomy; Table 2) and results were compared to examine the effects of blood processing and handling, prior to biomarker analysis. Samples processed within 2 h after phlebotomy returned a significantly higher AUC-ROC (0.80 (95%CI = 0.71 –0.89) as compared to samples processed within 24 h after phlebotomy (AUC-ROC=0.64 (95%CI = 0.52 –0.75) (Fig. 3, p < 0.001).

Fig. 3

ROC curves for the blood plasma samples (n = 100) collected using Protocol 1 (24 h post phlebotomy) and Protocol 2 (2 h post phlebotomy) and analyzed at C₂ N laboratory (left). The corresponding AUC estimates and 95%confidence intervals (right).

DISCUSSION

Plasma Aβ biomarkers were quantified by two independent laboratories in samples collected from screened, cognitively normal, A4 study participants with brain amyloid PET status determined by central readers. Our findings show that plasma Aβ₄₂/Aβ₄₀ quantified using either of the available MALDI-TOF-MS or LC-MS/MS analytical platforms can predict amyloid PET positivity among asymptomatic participants enrolled in the A4 trial. Blood samples were collected and processed to plasma within 2 h after phlebotomy (Protocol 2) provided a better prediction of amyloid PET status than blood samples that were collected and shipped on cold packs to the laboratory, where they were processed to plasma within 24 h after phlebotomy (Protocol 1). The findings imply that these plasma Aβ₄₂/Aβ₄₀ measures and analytical platforms can be used to pre-screen cognitively normal volunteers for eligibility into AD prevention trials; reduce traditional screen-failure rates; and save time and funds when used to screen for enrollment into prevention trials.

In 224 participant samples that were processed within 24 h, we determined that plasma Aβ₄₂/Aβ₄₀ quantified using the Shimadzu MALDI-TOF-MS platform provided a numerically better AUC = 0.80 than the C₂ N LC-MS/MS platform (AUC = 0.76), however these AUC values were not statistically different (p = 0.13). Conversely, the C₂ N LC-MS/MS platform provided an AUC = 0.80 for plasma Aβ₄₂/Aβ₄₀ in a subset of 100 participant plasma samples that were processed within 2 h. This was an improvement in AUC that was significantly better than what was achieved in the same 100 participants’ samples that were processed within 24 h (AUC = 0.64). Finally, Shimadzu calculated an additional Composite plasma biomarker that was generated by averaging the normalized scores of APP669–711/Aβ₄₂ and Aβ₄₀/Aβ₄₂ score [19, 23]. With an AUC of 0.82, we determined that measuring the Composite plasma marker could serve as a potential biomarker for predicting elevated brain amyloid, however further investigation is required. Previous studies have reported the combination of APP669-711/Aβ₄₂ ratio and Aβ₄₀/Aβ₄₂ ratio can serve as highly predictive surrogates for brain amyloid pathology [27]. We were not able to compare the biomarker performance of the Composite score between the MALDI-TOF-MS and LC-MS/MS methods as this marker is not measured by the C₂ N LC-MS/MS assay.

Plasma Aβ₄₂/Aβ₄₀ performed similarly between the two MS–based methods; however, we observed that the AUCs for plasma Aβ₄₂/Aβ₄₀ ratios were significantly lower than previously published reports (AUC 0.88 –0.97) [21, 27–30]. Moreover, the sensitivity (SN) and specificity (SP) varied significantly across individual plasma amyloid measures and across both analytical platforms. Plasma Aβ₄₂/Aβ₄₀ measures generated by C₂ N were highly sensitive (0. 74 –0.92) yet poorly specific (0.56 –0.70) in its ability to identify a participant who was not amyloid PET positive. While plasma Aβ₄₂/Aβ₄₀ measures generated by Shimadzu was highly specific (0.84) yet lack sensitivity (0.67) to identify an amyloid PET positive participant. Given the variance in sensitivity and specificity across both MS-based planforms, its plausible that our data may better interpreted as a measure to screen out those who are amyloid negative versus predicting those who are actually amyloid positive. Rabe and colleagues [31] determined that pre-analytical errors of plasma Aβ₄₂/Aβ₄₀ measures may lead to patient misclassifications (i.e., incorrectly ruling out amyloid pathology or vice versa) thus impacting sensitivity or specificity. As researchers continue to interpret their findings based on AUCs, additional clinical performance measures including the number of false positive and false negative [21] robustness [31–33] and population performance [34] must also need to be considered to support the use of plasma amyloid measures as reliable and accurate biomarkers for AD.

In general, MS-based assays are thought to have superior performance to detect amyloid pathology in brain as compared to immunoassays [35]. Recently, it has been reported that their level of performance as measured by AUC differed substantially among several MS-based and immunoassay-based analytical platforms that quantify plasma Aβ₄₂/Aβ₄₀ concentration ratio and identify brain amyloid status [35]. In particular, the immunoprecipitation LC-MS/MS assay (as performed by Bateman et al.) outperformed seven other analytical assays for detecting plasma Aβ₄₂/Aβ₄₀ [35]. The C₂ N LC-MS/MS analytical used herein is based on the Bateman laboratory assay. In addition, some disparities among analytical platform performance observed by Janelidze et al. [35] were attributed to differences in patient cohort characteristics and variations in sample processing. The current study supports prior reports [36] that sample handling procedures prior to analysis can significantly impact the sensitivity and specificity of plasma Aβ₄₂/Aβ₄₀ as biomarker for predicting amyloid PET positivity and this topic requires further investigation.

The interpretation of our data may be further complicated by the imaging protocols used to establish amyloid PET positivity. The existence of multiple imaging tracers with varying amyloid positivity (SUVr) cutoffs could have a direct impact on the ability of plasma Aβ₄₂/Aβ₄₀ to predict amyloid burden in the brain [37, 38]. Clinical trials, including the A4 Study, are opting to use newer generation, FDA approved, ¹⁸F-based radioligand tracers, such as florbetapir over the commonly used Pittsburgh compound B (PiB) amyloid tracer for imaging. Florbetapir enables a wider application due it its longer half-life (110 min) as compared to PiB (20 min) [38]. However, studies report that florbetapir-based amyloid measurements had higher variability due to nonspecific white matter deposition in the brain [37]. Because of this, tracers like florbetapir are likely to be less suitable for detecting subtle amyloid deposition in the brain [37, 38]. Future studies are needed to assess how amyloid tracer sensitivity influences the blood amyloid measurements across the spectrum of disease states. Understanding this will be vital for AD prevention trials where subtle amyloid deposition is more likely to occur in CN adults.

Despite age and APOE genotype being considered as risk factors for AD, we determined that these covariates included sex did not improve the ability of plasma Aβ₄₂/Aβ₄₀ to predict amyloid PET positivity. While previous studies have shown significant improvements in the sensitivity of plasma amyloid biomarkers when age, sex, and APOE ɛ4 carrier status are accounted for [19, 21, 29], several others have demonstrated similar results to ours. Keshavan and colleagues demonstrated that sex and APOE ɛ4 carrier status did not improve plasma Aβ₄₂/Aβ₄₀ ability to predict PET amyloid positivity in CN older adults [39] while Yamashita and colleagues [30] observed a slight numerical increase in the performance of the plasma Aβ₄₂/Aβ₄₀ ratio when incorporating the APOE ɛ4 status. Further investigation is required to reconcile these conflicting reports.

One major limitation of the current study is our relatively small sample number and the lack of racial and ethnic diversity among the trial participants. Blacks/African Americans and Hispanics/Latino(s) have a higher prevalence of AD as compared to non-Hispanic whites; yet they are consistently underrepresented in clinical trials for AD and other dementias [40]. A recent cross-sectional study found that blacks, Hispanics, and Asian participants were more likely to be ineligible for the A4 Study after the first screening visit [40]. In our sample, out of the 224 trial participants we analyzed, only 14 (9%) screened participants belonged to a minoritized ethnic group. Given that our data was generated from a relatively homogenous population, this may explain why our covariate analysis did not show any significant differences in AUC. Recent observational studies have revealed notable differences in the performance of AD biomarkers among different racial groups [41–43], therefore it is imperative that efforts are made to increase the diversity among trial participants. We anticipate that blood testing will increase accessibility to diverse minority and socioeconomic groups as well as improve screening and retention of trial participants [44].

In summary, our work and work of several others [21, 27–30] determined that plasma Aβ₄₂/Aβ₄₀ may serve as a potential biomarker for predicting elevated brain amyloid in cognitively normal adults. Additionally, the combination of APP669-711/Aβ₄₂ ratio and Aβ₄₂/Aβ₄₀ ratio may also serve as predictive surrogate for brain amyloid pathology as more studies are investigating a multi-biomarker approach for predicting cognitive decline and amyloid PET positivity [45]. Plasma p-tau181/Aβ_1 - 42 ratio [46] and t p-tau 217 [47] has recently emerged as alternative biomarker for predicting cognitive decline and amyloid PET positivity. A head-to-head comparison is necessary to determine which biomarker would be the most effective at predicting cognitive decline and amyloid PET positivity in older, CN adults.

Lastly, our exploratory analysis reveals the potential savings associated with using a blood test as a screening tool for enrollment into secondary prevention clinical trials like the A4 Study. Previous analysis has suggested that a blood test can reduce the use of amyloid PET scans by 50%[39]; however, future studies are still needed to understand how sample APOE status, race and ethnicity, and amyloid tracer sensitivity impacts plasma Aβ₄₂/Aβ₄₀ ability to predict amyloid brain pathology in normal controls.