Introduction

Alzheimer’s disease (AD) is a global health problem for western countries, representing more than 60% of dementia cases in the world. The pathological findings of AD include the progressive increase of Aβ peptides in the brain conforming extracellular amyloid plaques together with intracellular deposits of hyper-phosphorylated tau that form characteristic neurofibrillary tangles. Both pathology hallmarks accompany progressive neuronal loss that ultimately provokes memory loss and severe cognitive dysfunction [1].

AD is an intractable condition to date. The identification of early (preferably pre-symptomatic) biomarkers and true etiologic factors for this condition are the first steps to establish effective primary prevention programs for AD. Consequently, the search for a relatively inexpensive and harmless biomarker for AD continues. Beyond the neuropsychological assessment which still represents the most essential tool for AD and mild cognitive impairment (MCI) screening in humans[2], the most reputed biomarkers for AD are cerebral-spinal fluid (CSF) Aβ42 and phosphorylated-tau protein levels, hippocampal volume measured by magnetic resonance imaging (MRI) techniques and positron electronic tomography (PET) scan with brain Aβ radiotracers. These techniques represent the most studied methods for the detection of prodromal AD. However, there are also drawbacks for each one. For example, although MRI sensitivity showed high sensitivity at baseline, MRI specificity is not unexpectedly limited for MCI conversion to dementia[3] . Furthermore, MRI is restricted to patients with pacemakers. On the other hand, CSF measurements are sometimes variable and imprecise, since measurements vary between studies and laboratories, standardization of analytical as well as pre-analytical procedures will be essential[4]. Some subjects may be unwilling to undergo a lumbar puncture or may have contraindications, such as use of anticoagulants. Finally, something that must be considered is the expense associated with performing amyloid PET scans in large numbers of subjects. Furthermore, its sensitivity and specificity for MCI conversion would require further evaluation, due to a need for improved clinical diagnosis of those subjects with major risk of conversion to dementia, and standardized protocols of data acquisition and imaging analyses. Therefore, opportunities for diagnosis improvement in prodromal or even pre-symptomatic AD still remain.

In spite of an intensive worldwide research, there is not a definitive plasma or blood biomarker indicating high/low risk of conversion to Alzheimer’s disease to date[5]. Because of their involvement in the generation of amyloid deposits in the brain, the Aβ levels in blood have been widely investigated as potential markers for AD. However, regarding plasma measurements, contradicting results have been reached by using different molecular detection methods or research designs[6]. Beyond the role of plasma Aβ levels as potential AD biomarkers, there are a number of publications suggesting an association between plasma Aβ levels and blood pressure[7-9], body mass index (BMI)[10-12], and different biochemical blood parameters such as insulin levels[13], creatinine [14-18], cystatin C[18,19] or homocystein levels [15,16]. These last observations are of importance because they might help understand the true physiological meaning of APP derived peptides in different tissues.

Our group is actively involved in the development of novel ELISA sandwich colorimetric tests for detection of Aβ using whole blood instead of plasma alone [20,21]. In fact, using novel technology, we conducted a trial, called the AB128 project, which studied the efficacy of Aβ blood levels as an AD biomarker. Specifically, we found statistical significant differences of some measurements in different blood compartments when comparing healthy controls (HC) and MCI subjects [20]. In spite of these achievements, a definitive conclusion will require independent and extensive replication. Following this idea, a novel trial doubling sample size, the AB255 project, and independent replication efforts using samples obtained from reputed cross-sectional studies are underway to corroborate previous findings.

The aim of the present work is to identify the medical relevant variables correlated with blood Aβ measurements obtained using these novel ELISA techniques. So, we have conducted an unsupervised search of such correlates in the available dataset obtained from the AB128 project. Our study confirmed several variables previously associated with plasma levels of Aβ 40 and pointed to others that seem specific for the amyloid retained in the cell pellet. The information provided herewith can help to define the critical set of covariates for future studies and might provide novel insights in the peripheral function of Aβ.

Results

First, we analyzed common demographics and clinical differences between the three phenotypic groups (HC, MCI and AD) included in this study (for details see Methods section and table 1). Demographic analysis revealed important differences in age, education, plasma homocysteine levels and APOE allele ԑ4 carrier status (dominant model) (Bonferroni corrected p<0.002). However, no significant differences on gender, body mass index, creatinine plasma levels, heart rate, systolic blood pressure, diastolic blood pressure antihypertensive or statin treatment usage was observed. Strong differences on DA and RP Aβ levels were also detected among groups (p<0.001) as observed in a previous report[20].

Having these findings, we decided to start the covariate search for blood Aβ by applying unsupervised searches in all phenotypic groups together (table S1a,b). Our study indicated that age, creatinine plasma levels and homocysteine were significantly correlated with DA and RP Aβ 40 levels but not with CP Aβ 40 or Aβ 42 in any blood compartment (table 2). The study also suggested a significant, although not high, correlation between hematocrit, uric acid and diastolic blood pressure to different blood Aβ measurements in blood. Interestingly, only serum levels of immunoglobulin A can be nominally correlated to blood beta- Aβ measures in cell pellet (CP) (table 2).

Previous studies have suggested a strong correlation between age and Aβ levels, as well as between creatinine levels and Aβ levels[14,17,18]. Other studies also indicated sexual dimorphism in amyloid measurements between men and women[17]. The demographic analysis suggested differences in age, APOE genotype and education among phenotypic groups. Moreover, the co-linearity between many parameters during unsupervised analysis was also observed, especially true for creatinine, homocysteine, uric acid and urea plasma levels (data not shown). Considering these observations, we decided to conduct a partial correlation analysis of candidate covariates, using an adjusted model including age, gender, phenotypic group, and creatinine levels (table 3). By applying this relatively simple model, we found that most of the correlates between blood Aβ determinations and the other analyzed parameters disappeared, with the exception of DBP which maintained its correlation with RP Aβ 40 (r=-0.18; p=0.046; table 3). Remarkably, an improved correlation between DBP and DA Aβ 40 levels was observed using partial correlations analysis. This specific observation reached a nominally significant association (r=-0.19; p=0.034; table 3). Also, hematocrit correlations with RP Aβ 40 (r=-0.29; p=0.001) and DA Aβ 40 (r=-0.23; p=0.01) remains.

These results pointed out that in addition to age, gender, creatinine, phenotype, the hematocrit and DBP could be interesting covariates for DA or RP blood Aβ 40 levels. In contrast, neither of them seemed important for blood Aβ 42 determinations (DA, RP or CP fractions) nor CP Aβ 40 (table 3).

Because age, anti-hypertensive drug usage and individual renal function differences among groups may severely distort DBP/DA Aβ 40, or DBP/RP Aβ 40 relationships, we decided to conduct backward linear regression analyses in the three separate phenotypic groups (AD/MCI/HC). The main objective of this supplemental study was to ascertain whether there is a homogeneous correlation between DBP and DA Aβ in plasma (in terms of effect size and direction) in separate groups. Figure 1 shows linear regression between diastolic blood pressure (DBP) volume and levels of DA Aß 40 in the three groups separately. The analysis also permitted to check whether or not DA Aβ 40 can be retained in best regression models after extensive co-variation with well recognized parameters affecting DBP in elderly subjects (i.e. anti-hypertensive drug usage, BMI, age, gender, renal function variables, APOE genotype, etc.). Importantly, our results suggested a consistent correlation effect of DA Aβ 40 levels in three phenotypic groups (table 4). Most importantly, in spite of including an extensive number of covariates, DA Aβ 40 was retained in the best backward regression model of the three phenotypic groups. This last result supports the notion that plasma Aβ 40 levels/DBP correlation was independent of well-established physiological and pathological parameters affecting blood pressure in elderly subjects. Furthermore, the association between both variables was important enough to be selected as a critical one irrespective of cognitive status of studied individuals (table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Linear regression between diastolic blood pressure (DBP) volume and levels of DA Aß1-40 in Alzheimer’s disease patients (AD, red triangles), mild cognitive impairment subjects (MCI, blue circles) and healthy individuals (HC, green squares).

https://doi.org/10.1371/journal.pone.0081334.g001

Backward linear regression analyses also suggested that variables retained in the best model (and significantly affecting DBP) in the three phenotypic groups are different (table 4). In fact, cardiovascular factors such as total cholesterol levels and hematocrit seem much more important for DBP in the healthy (and younger) control group than in the MCI or AD groups. On the contrary, APOE genotype and creatinine levels seemed to only correlate well with the AD status. Remarkably, DA Aβ40 remains as one of the selected variables in the MCI group (table 4).

Discussion

The discovery of easily accessible, cost-effective and pre-symptomatic biomarkers for AD is mandatory to fuel secondary prevention strategies. There are numerous efforts underway to reach this objective[5]. Technologies under investigation are multiple and not only related to amyloid levels. In fact, promising results are emerging by applying novel technologies such as analyses of differences in gene expression patterns[22,23], micro-RNA analyses [24] or proteomic-based approaches among others[25].

The efficacy of Aβ blood levels on AD prediction has also been recursively investigated. Original investigations suggested an association between Aβ levels and AD, although results were not uniform among studies[6,8,9,19-21,26-28]. Thus, preliminary findings have remained controversial and would need independent validations [6,19]. Notably, the direct comparison between most studies is almost impossible due to the existence of many different immunodetection methods, including different polyclonal antibodies, differences on study designs or the lack of standardization or consensus in the analysis model. Therefore, the methods’ differences in previously published studies prevent a direct comparison among currently available results or meta-analysis.

In the present study, the Araclon’s blood Aβ immunodetection system permitted the identification of previously observed well-established correlations, including blood pressure, hematocrit, creatinine and homocysteine levels. These observations are of importance mainly for two reasons: ( i) on the one hand, the identification of similar correlations to those previously reported by independent research groups using our technology directly reassured the validity of our methods; (ii) on the other hand, the identification of such covariates may serve to a better design of future larger studies by controlling critical covariates affecting Aβ levels.

Based on previous investigations and on the findings presented here, variables such as age, gender and renal function biomarkers (i.e. serum creatinine levels) appeared as the most important covariates correlated to the levels of plasma (DA or RP) Aβ 40. The effect of these covariates in the blood levels of Aβ 42 appear to be smaller in our study. However, taking into account previous results, the incorporation of such covariates to future analyses also seems pretty reasonable for studies with Aβ 42. This could seem like an apparently obvious recommendation. In fact, it has been previously suggested by others[14]. However, only a minority of studies such as the cardiovascular health study (CHS) has incorporated renal function biomarkers in their models [19].

In contrast, the rest of correlates initially detected seem not to be independent of creatinine levels. Thence, we suggest that urea, uric acid and homocysteine associations have appeared only as a consequence of strong creatinine/ Aβ relationship as previously mentioned. Consequently, it makes no sense to incorporate them to regression models due to high co-linearity among them.

An exception to this rule might be the hematocrit, which retained its correlation with Aβ 40 levels even after the adjustments of age, gender, phenotype and plasmatic creatinine levels. It is possible that the hematocrit fraction is somehow affecting the equilibrium between cell-bound and cell-free in plasma amyloid. However, this association was not uniform enough among the phenotypic groups (HC, MCI and AD) to delineate a definitive conclusion. A replication increasing sample size is necessary to corroborate hematocrit findings. Interestingly, these results are consistent with previous findings in the Japanese population suggesting a genuine correlation between both physiological parameters[7]. Although some authors have speculated on hematocrit implication on cognition performance[29], its role on AD etiology is not well understood and deserves further investigation.

On the contrary, blood pressure (BP) is a well-recognized risk factor for dementia. Indeed, AD has been linked to cardiovascular risk factors, such as diabetes mellitus, hyperlipidemia, and, in particular, elevated BP[9]. DBP/Aβ 40 correlations appear consistent in our study irrespective of the individual’s phenotype. This observation corroborated, in terms of effect size and direction, previous reports [8,30]. The largest study reported to date on blood pressure and plasma Aβ levels was held by Lambert et al. in their “Three-Cities Study”[8]. The authors suggested associations between the Aβ 40/ Aβ 42 ratio and the systolic blood pressure, the DBP or the hypertension. However, they also pointed out that the observed effect could be mainly driven by plasma Aβ 40. Furthermore, the mechanism underlying this preferential association could be related to the Aβ 40 peptide’s properties on vascular vessels (for details, see Lambert et al. and references therein). Concretely, previous studies have indicated that Aβ 40 peptides could modify cerebral blood vessels in vitro and induce a decrease of cerebral flow and cerebral blood volume in vivo[31,32] and that Aβ 40 has important effects on vasoconstriction[32]. Our results independently confirm the importance of Aβ 40 in blood pressure in humans. Most importantly, the stratified analysis suggests that this effect can be detected in elderly patients affected or unaffected by MCI and AD.

The relationships between blood Aβ levels, blood pressure and Alzheimer’s disease have also been explored in the Honolulu Heart Program/Honolulu Asia Aging Study[9]. The authors of that study suggested that the risk for AD significantly increased with lower levels of plasma Aβ 40; hazard ratio: 2.1 [95% CI: 1.4 to 3.1]; and detected evidence of interaction between DBP and plasma Aβ 40. Importantly, low plasma Aβ 40 or 42 was associated with the presence of cerebral amyloid angiopathy but not with the other neuropathologies. Therefore, the disruption of blood pressure homeostasis in midlife could contribute to future risk of dementia. Therefore, reduced levels of Aβ 40 in midlife could be directly or indirectly involved in the early pathogenic process of AD.

A major limitation of this study is the observed differences in demographic characteristics among groups. In general, healthy controls are younger and with higher educational scores than MCI or AD subjects. However, despite this limitation the study was able to replicate previous findings as already pointed out, although the new insights shown here would deserve a further replication.

Another limitation would be the small sample size. This might explain the discordance observed in the relationships between amyloid measurements in different compartments and physiological parameters studied (Table 2 and 3). For example, backward regression analyses using Aβ40 RP fraction versus DBP were not fully consistent with the results obtained in Aβ40 DA compartment (data not shown). This divergence might be attributable to having a small sample size which in turn may provoke random chance oscillations during effect size estimation. Alternatively, unknown factors could be affecting peptide measurements in different compartments. For this reason, it would be advisable to independently replicate these results to corroborate these findings.

Finally, the biological meaning of these findings cannot be ignored because it could provide essential information about the real physiology of APP derived peptides in different human systems. For instance, the creatine/creatinine energy cycle in brain and muscle could have a closer relation with amyloid physiology than previously anticipated [33]. Therefore, observed correlations would need a more deep interpretation beyond a mere correlation with the renal function as proposed in previous studies. In contrast, the covariates related to Aβ CP fraction remained almost unknown. In fact, only serum immunoglobulin A levels displayed a weak correlation with the amyloid measured in the cell pellet fraction of the blood. Importantly, the presence of immunoglobulin fragments in the amyloid plaques of the AD brain had been observed[34] and this observation could be easily linked with AD neuroinflammatory hypothesis[35] . However, we considered that this last observation would require independent confirmation in future studies. Nevertheless, the novelty and potential of CP amyloid warrant further investigation.

Materials and Methods

Supporting Information

Acknowledgments

We would like to thank patients and controls who participated in this project. We are indebted to Trinitat Port-Carbó and her family who are supporting Fundació ACE research programs. Fundació ACE is an associated center to CIBERNED. Anonymised data table containing this dataset is available on request to corresponding author.