A Review on MS-Based Blood Biomarkers for Alzheimer’s Disease

This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.

The accumulation of Aβ in amyloid plaques in the brain is a major hallmark of AD. Alterations of the proteolytic APP (amyloid precursor protein) processing and release of the Aβ peptides is a key event in AD pathophysiology [1]. The measurement of Aβ42 in CSF is an established AD biomarker of amyloid pathology and included in the diagnostic criteria [2]. MRM is used as a reference method for Aβ measurement in CSF [4, 5]. Pannee and colleagues used their antibody-free CSF reference MRM also to investigate blood levels of Aβ38, Aβ40 and Aβ42. In their small cohort of 9 AD and 10 control patients, they did not observe significant differences between the two groups [6]. There was also no correlation of Aβ42 levels in plasma and CSF but CSF was analyzed by ELISA. They also characterized additional Aβ variants in plasma by immunoprecipitation (IP) and MALDI-ToF MS (Aβ15, Aβ17, Aβ19, Aβ20, Aβ33, Aβ34, Aβ5-40, Aβ37, Aβ3-40 and Aβ39) but observed no differences between AD and control samples (n = 2).

Another group also reported an antibody-free MRM method with the aim to measure total Aβ in plasma [7]. However, they performed tryptic digestion of all plasma proteins as an initial step of sample preparation. The tryptic peptide (Aβ17-28) selected for Aβ quantification in this study is not specific for Aβ and is also released from full-length APP during tryptic digestion. Thus, measurement results did not represent Aβ in plasma. This is also reflected in the reported concentration which is more than tenfold higher compared with the more reliable method from Pannee et al. [6].

A Japanese group published a first study in 2014 using IP-MALDI ToF MS to measure Aβ peptide levels in plasma. They investigated subjects with and without cerebral amyloid deposition which was determined by positron emission tomography (PET) with Pittsburgh Compound B (PiB) [8]. The IP-MALDI ToF MS method used was already validated in terms of precision and dilution linearity previously [9]. The authors chose this pathology-driven approach because AD-like amyloid deposition is also found in a substantial number of non-demented aged individuals [10]. In this context, a syndromal comparison (i.e. AD patients vs controls) might underestimate the diagnostic performance of amyloid-related plasma biomarkers. Thereby, they also followed the recent suggestion of a biological definition of AD by the ATN classification of biomarkers [11]. Kaneko et al. measured plasma levels of Aβ40, Aβ42 and APP669-711 in 22 PiB-negative and 40 PiB-positive (including AD, MCI and healthy controls) subjects. They observed significantly reduced Aβ42 and Aβ42/Aβ40 and increased APP669-711/Aβ42 in PiB-positive vs PiB negative subjects [8]. Subgroup comparison showed, that the APP669-711/Aβ42 ratio was also increased in PiB-positive AD compared with PiB-positive healthy controls. The ratio APP669-711/Aβ42 yielded a high sensitivity and specificity (93% and 96%) for the detection of cerebral amyloid pathology. The authors confirmed their results recently with a slightly modified method in two additional patient cohorts from the Japanese National Center for Geriatrics and Gerontology (NCGG, n = 121) and Australian Imaging, Biomarker and Lifestyle Study of Ageing (AIBL, n = 252) [12]. APP669-711/Aβ42 and Aβ40/Aβ42 showed good discriminatory power for amyloid positive and negative subjects in the NCGG cohort (sensitivity 68% and 96%, specificity 92% and 87%) and AIBL cohort (sensitivity 83% and 85%, specificity 70% and 68%). In addition, they created a composite biomarker from both ratios with even better performance in the NCGG and AIBL cohort (sensitivity 86% and 86%, specificity 89% and 78%). They used stable isotope-labeled (SIL)-Aβ38 as internal standard (IS) for all Aβ peptides, which has the advantage that the Aβ ratios are independent from the spiked amount of IS. Therefore, the established cut-off values are transferable. A drawback could be that the different properties of the Aβ peptides are not optimally reflected by Aβ38. This might contribute to the center differences of the Aβ ratios between the NCGG and AIBL cohort due to different matrix effects.

Most recently, the Bateman group published a study confirming the good predictive power of Aβ40/Aβ42 in plasma for amyloid PET status [13]. They used tryptic digestion and MRM instead of MALDI-ToF but observed a similar sensitivity (88%) and specificity (76%) for the discrimination of amyloid positive and negative subjects compared with Nakamura et al. [12].The high performance for detection of cerebral amyloid pathology and the consistent findings with IP-MS methods in multiple cohorts and by different research groups makes it a very promising MS-based biomarker.

Phospholipids are an integral part of the plasma membrane and their composition substantially influences the function of membrane proteins, receptors, enzymes and ion channels. The reduced phospholipid amount in AD brains gives a first link of phospholipids to AD pathophysiology. Furthermore, the cleavage of APP, releasing the Aβ peptides, takes place at the cell membrane and thus in direct contact with phospholipids [14].

In 2014, Mapstone and colleagues [15] reported a panel of phospholipids (Table 1) with significant changes in plasma already in the preclinical phase of AD. The phospholipids and other metabolites were measured using a commercially available MRM kit. They compared AD/amnestic MCI (mild cognitive impairment) patients with cognitively normal controls from a longitudinal observational study. They especially looked at subjects converting from cognitively normal to AD/aMCI during the observational period. Here, they included plasma samples from the time before and after conversion. Combining the panel of eight phospholipids with two acylcarnitines (C3 and C16:1-OH, Table 1), they were able to identify AD/aMCI patients in their cohort of 106 patients at a time, when they were cognitively normal (on average 2.1 years before conversion). The sensitivity and specificity was 90%. They successfully validated their results in a different subset of 40 patients from the same cohort. Significant changes in AD, MCI or dementia of four lipids from the Mapstone panel (PC aa C36:6, PC aa C38:6, LysoPC 18:2 and C3) were also reported in other studies [16‐21] (see Table 1) but sometimes in the opposite direction [16].

Three studies from two other research groups evaluated the Mapstone panel of blood markers in different longitudinal cohorts. These included subjects converting to AD or MCI during the follow-up and the use of the same MRM kit [16, 17, 22]. Casanova et al. investigated samples from two large cohorts of patients, the Baltimore Longitudinal Study of Aging (BLSA) and the Age, Gene/Environment Susceptibility-Reykjavik Study (AGES-RS) comprising 392 patients in total [22]. They could not reproduce the high predictive performance of the Mapstone panel in blood for preclinical AD. In their much larger cohort of preclinical AD converters (n = 93 in BLSA cohort, n = 100 in AGES-RS cohort vs. n = 28 in Mapstone cohort), they observed sensitivities and specificities of 52% and 66% (BLSA cohort) and 47% and 36% (AGES-RS cohort). The use of different sample matrices (plasma vs serum) has been discussed as a possible reason for the discrepancy between these two studies. However, Casanova et al. also provided data that plasma and serum levels of the Mapstone panel are highly correlated [22].

Li and colleagues investigated plasma samples from the Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). A large majority of these samples were from African Americans (> 96%) and their aim was to evaluate the predictive performance of the Mapstone panel (established in a white population) in African American [16]. Their cohort included 95 MCI/dementia converters with a mean follow-up time of 7.3 years, which is considerably longer than in the Mapstone study (2.1 years). In agreement with the study by Casanova et al., they observed neither a large added value of the Mapstone panel for prediction of MCI/dementia [16] nor for detection of MCI/dementia in a cross-sectional comparison (n = 441) [17]. However, the MCI/dementia patients in the studies by Li et al. [16, 17] are not solely AD patients which might weaken the discriminatory performance of the (potentially AD-specific) biomarker panel. Changes of several other phospholipids in serum/plasma were reported in single studies and are listed in Table 2.

Sphingolipids are another class of lipids which are part of the plasma membrane and involved in a myriad of different biological processes such as signal transduction and cell–cell interaction [23]. Changes of different sphingolipids in AD blood using MS techniques have been reported in single studies and are listed in Table 2.

ApoE is the most prevalent lipoprotein in the brain and it exists as three major isoforms, ApoE2, ApoE3 and ApoE4 [24]. ApoE4 is the most important genetic risk factor for AD [1] and might be an interesting biomarker candidate. Han and colleagues used two orthogonal methods to measure total ApoE in serum including MRM and observed lower ApoE levels in AD with both methods [25], Two other studies did not find a difference [26, 27]. Inconsistent results of ApoE levels in AD blood were also reported in studies using immunassays [28‐31]. In addition, total ApoE levels in blood are different between carriers of different ApoE genotypes [26, 27, 32] which renders ApoE quite unsuitable as a disease biomarker. However, it turned out that MRM seems to be a valuable method for ApoE genotyping. Several groups reported the successful isoform identification of ApoE in blood by quantifying allele-specific peptides with up to 100% concordance to classical genotyping [32, 33].

L-Carnitine and acylcarnitines are essential players in energy metabolism but also other functions e.g. in the brain are reported. This includes protein modulation, neuromodulation, protection from excitotoxicity, antioxidant and anti-apoptotic functions all of which can be linked to neurodegenerative diseases [34]. Consistent changes in AD in two independent studies have been reported for five acylcarnitines. Acetyl-carnitine (C2), dodecanoyl-carnitine (C12), dodecenoyl-carnitine (C12:1) and tetradecenyl-carnitine (C14:1) were reduced in plasma [35] and serum [20] of AD patients compared with controls and MCI patients. C2, C12:1 and C14:1 were already reduced in MCI [20]. Both studies used MRM for acylcarnitine measurement. In contrast, octanoyl-carnitine (C8) was increased in AD in two studies analyzing plasma [35] or serum [36] samples. Reports on serum C8 levels in MCI were inconsistent [20, 36]. Changes of other acylcarnitines in blood of AD patients were reported only by single studies and are listed in Table 2. Free fatty acids and diacyl- and triglycerides measured with MS-based methods are also depicted in Table 2 because changes in AD were described in single studies only.

A link between cholesterol and AD comes from epidemiological studies indicating that high cholesterol levels in blood are associated with a higher risk for the development of AD. In addition, cholesterol can influence APP processing and Aβ aggregation [37]. In the brain, cholesterol is converted to the more hydrophilic 24S-hydroxycholesterol to eliminate it from the brain because cholesterol itself cannot cross the blood–brain-barrier. 24S-hydroxycholesterol is an important metabolite in the regulation of cholesterol homeostasis in the brain [38]. Several studies investigated 24S-hydroxycholesterol levels in blood of AD patients using MS but with conflicting results. Lütjohann et al. [39] and Zuliani et al. [40] observed significantly increased 24S-hydroxycholesterol levels in plasma of AD patients compared with controls although with a substantial overlap between groups. In contrast, Solomon and colleagues found reduced levels in AD [41] and another study observed no difference between AD and controls [42]. Thus, 24S-hydroxycholesterol might not be an optimal biomarker candidate for AD. Two precursors of cholesterol synthesis, lanosterol and lathosterol, have consistently been reported to be lower in AD patients in two studies [41, 43] but also with a high variation within groups. This might limit their use as a biomarker. Several other cholesterol-related metabolites such as bile acids, steroids and vitamin D were measured in blood of AD patients in single studies or with conflicting results and they are listed in Table 2.

Several other molecules were investigated in blood of AD patients using MS methods and were suggested as potential biomarker candidates. For a majority of them, significant changes in AD were reported in single studies only or with conflicting results and are therefore summarized in this section and listed in Table 2. This includes the group of metal ions. Here, consistent changes in AD in more than one study were described only for cobalt (down-regulated) and mercury (up-regulated) [44, 45]. Furthermore, markers of oxidative stress, amino acids and their derivatives and other metabolites or proteins have been described to be affected in blood of AD patients. However, their suitability as AD biomarkers is difficult to assess due to the single reports and also the question of specificity for AD.