Molecular Markers of Amnestic Mild Cognitive Impairment among Mexican Americans

Abstract

Background:

Mexican Americans face a significant health disparity when it comes to Alzheimer’s disease (AD) as they present with higher rates of the disease and develop AD at an earlier age compared to other ethnic groups. Recent work identified a proteomic profile of AD among this population; however, no work to date has sought to examine the biological profile of pre-AD among Mexican Americans.

Objective:

This study aims to identify an amnestic mild cognitive impairment (aMCI) proteomic profile among Mexican Americans.

Methods:

Data were analyzed from 284 Mexican American participants (aMCI, n = 73; normal controls, n = 211) from the Health & Aging Brain among Latino Elders study. Fasting serum samples were analyzed using a multiplex biomarker assay platform. A biomarker profile was generated using random forest analyses.

Results:

Among aMCI cases, the biomarker profile was found to be largely inflammatory with the top three markers shown to include TNFα, IL10, and TARC. The overall diagnostic accuracy of the biomarkers in detecting aMCI was 96% (sensitivity = 0.82; specificity = 0.97). Inclusion of clinical variables with the selected biomarkers did not impact the overall detection accuracy (area under the curve = 0.96) but led to a slight improvement in specificity (specificity = 0.99) and decrease in sensitivity (sensitivity = 0.74).

Conclusion:

The biomarker profile of aMCI was shown to be different from our previously generated AD profile among Mexican Americans, which was largely metabolic in nature. The findings implicate a possible interplay between inflammatory and metabolic processes and additional work is needed to further examine this.

INTRODUCTION

Mexican Americans are among the fastest growing segment of the aging population [1], and it is estimated that Alzheimer’s disease (AD) will grow six-fold among the Hispanic population, which places them at a disproportionate risk when compared to other ethnic groups [2]. Hispanics, specifically Mexican Americans are 1) at increased risk for AD and mild cognitive impairment (MCI) at significantly younger ages, 2) are diagnosed at a more advanced stage of the disease progression, and 3) experience a higher rate of modifiable risk factors for MCI and AD (e.g., depression, diabetes, hypertension) [3–5]. Given these risks, the development of methods for the early detection of cognitive impairment is particularly salient for this population.

The growing demand for fast, reliable, efficient, and cost-effective methods for the detection of AD and MCI has caused a significant growth in the search for blood-based strategies [6, 7]. Blood-based biomarkers are more cost and time-effective than neuroimaging and cerebrospinal fluid (CSF) modalities and can serve as the first step in the multi-stage diagnostic process [6]. While neuroimaging and CSF approaches have been investigated as methods for detecting MCI and predicting risk of progression from MCI to AD [7–16]; the search for blood-based biomarkers for the detection of amnestic MCI (aMCI) has received less attention. Oh et al. [17] recently found that plasminogen activator inhibitor-1 (PAI-1) is altered in MCI as compared to normal controls and was related to cognitive abilities. Examining plasma metabolite profiles from 58 patients diagnosed as aMCI and 57 controls, Wang et al. [18] found a high diagnostic accuracy using a multi-marker approach. Clusterin has also been found significantly altered in MCI as compared to normal controls [19, 20].

On the other hand, outside of the current group, little work has been conducted examining blood-biomarkers of MCI and AD among Mexican Americans (the largest segment of the U.S. Hispanic population) despite the fact that prior work demonstrates that these biomarkers vary by ethnicity [5, 21]. A meta-analysis of the genome-wide allelic association studies supported the presence of ethnic specific genetic markers for AD [21]. This study identified CLU (SNP rs11136000), CR1 (SNP rs3818361) and PICALM (SNP rs3851179) as being associated specifically with non-Hispanic whites, while other markers such as ApoEε4 were found to be correlated with AD across a range of races/ethnic groups including Hispanics [21]. Other genetic markers for late-onset AD have been identified among Hispanics who are of Caribbean origin; however, such markers have been found to be overlapping but yet distinct [22]. Recently, Royall and Palmer [23] demonstrated that ethnicity is an important variable when considering blood biomarkers of AD when examining a Mexican American cohort.

Recent work from our group has sought to identify specific blood-based biomarker profiles of AD across ethnic groups including non-Hispanic white [24–26] and Mexican Americans [5]. These findings supported significant differences in the proteomic profiles of AD across groups with non-Hispanic whites exhibiting an inflammatory based profile as compared to a metabolic based profile shown among Mexican Americans [5, 24–26]. To date, no study has expanded this work to explore biomarkers of pre-AD stages of aMCI, specifically among Mexican Americans. Therefore, this study sought to address this dearth in the literature by generating a serum-based biomarker profile of aMCI among community-dwelling Mexican Americans.

MATERIALS AND METHODS

Participants

Data were analyzed from 284 Mexican American participants (aMCI n = 73, normal controls n = 211) from the Health & Aging Brain among Latino Elders (HABLE) study [27, 28]. The HABLE study is designed to explore factors related to aging among Hispanics, primarily Mexican Americans. Recruitment for HABLE is based on a community-based participatory research (CBPR) approach, which involves partnering communities with scientific groups in an effort to conduct studies of human disease. The current team has extensive CBPR experience working with underserved Hispanic populations from both rural and urban settings. The study team established and maintains community ties through local advisory boards, presentations and hiring of local workers into the research infrastructure. Our community recruiters, community individuals, and research staff present information about the study at community events as well as through door-to-door solicitation. This research was conducted under an IRB approved protocol with each participant (and/or informant for cognitively impaired persons) providing written informed consent.

Methods

Each participant underwent a standardized evaluation, which includes an interview (i.e., medical history, health behaviors), neuropsychological assessment, clinical labs, and medical examination. Interviews and testing were completed in English or Spanish depending on the participant’s specified preference. For each participant, an informant interview was conducted to obtain information regarding activities of daily living. Participant in formation was reviewed weekly by a consensus committee and diagnosis of aMCI (single and multi-domain) was established based on Mayo Clinic criteria for MCI [29]. Diagnosis of normal cognition was based on participants performing within normal limits on psychometric assessments [30].

Neuropsychological testing

The neuropsychological battery for HABLE consisted of the following tests of executive functioning (Trail Making Test Form B, Clock Drawing [CLOX1]) [31, 32], language (FAS, Animal Naming) [33], visuospatial functioning (Clock Drawing [CLOX2]) [32], memory (Wechsler Memory Scale- 3rd ed. [WMS=3] Logical Memory, Consortium for the Establishment of Registry for Alzheimer’s Disease List Learning) [33], and attention (WMS-3 Digit Span, Trail Making Test Form A) [34]. Global cognitive functioning was assessed using the Mini Mental State Examination (MMSE) [35] and disease severity was rated according to the Clinical Dementia-Rating Scale sum of boxes (CDR SB) [36, 37].

Human serum sample collection

Samples were collected based on a protocol, in line with the recently published pre-analytic guidelines [38]: 1) fasting serum samples were collected in 10 mL tiger-top tubes using 21 G needles, 2) allowed to clot for 30 minutes at room temperature in a vertical position, 3) centrifuged for 10 minutes at 1,300 × g within 1 hour of collection, 4) 1.0 ml aliquots of serum were transferred into cryovial tubes, 5) Freezer works TM barcode labels were firmly affixed to each aliquot, and 6) samples were then place into −80°C freezer for storage until use in an assay. Total processing time (stick to freezer) was kept under 2 hours.

Assays

All samples were assayed in duplicate via a multi-plex biomarker assay platform using electrochemiluminescence (ECL) on the SECTOR Imager 2400A from Meso Scale Discovery (MSD; http://www.mesoscale.com) per our previously published methods [5]. ECL measures have well-established properties of being more sensitive and requiring less volume than conventional ELISAs, which is the current gold standard for most assays [39]. Twenty-one biomarkers were selected to be assayed based on their identification through a series of generated and cross-validated AD algorithms [24–26]. The initial AD algorithm consisted of 108 proteins, which was later refined to 30 and then later further refined to the top 21 proteins associated with the disease state [24–26]. The later AD algorithm of 21 proteins consisted of the following and were included in our subsequent analyses due to their link with AD: CRP, SSA, ICAM, VCAM, A2M, B2M, FVII, TNC, CA125, Eotaxin3, IL5, IL6, Il7, IL10, IL18, TARC, TNFα, Fatty Acid Binding Protein (FABP), I309, PPY, and THPO.

Statistical analyses

Analyses were performed using R (V 2.10) and SPSS 19 (IBM) statistical software [40]. Chi square and t-tests were used to compare case versus controls for categorical variables (gender) and continuous variables (age and education). As with our prior methods [24–26], the biomarker data was transformed using the Box-Cox transformation and then randomly split into a training (aMCI n = 50; normal controls n = 148) and a test set (aMCI n = 23; normal controls n = 63). Within the training set, a random forest (RF) biomarker profile was generated using R Package randomForest (V 4.5), with all software default settings [41]. The RF biomarker risk score was then applied to the test sample to calculate classification accuracy estimates. The ROC (receiver operation characteristic) curves were analyzed using R package. ROC curves reflect sensitivity and 1-specificity for all possible scores on an item with AUC reflective of the overall discriminability based on the ROC figure. An adjusted model was then run utilizing the same methodology described above to examine the inclusion of select clinical variables (age, gender, education). Significance was set at p < 0.05. In order to place the current findings into the context of clinical diagnostic predictions, positive (PPV) and negative (NPV) predictive power were calculated. PPV and NPV were calculated based on Baye’s theorem as previously described [42] utilizing estimated base rates of 10% and 20% within a community-based population of MCI.

RESULTS

Table 1 presents the demographic characteristics of the sample. When compared to normal controls, aMCI cases were significantly older (t = 8.01, p < 0.001), had fewer years of education (t = −2.61, p = 0.001), obtained lower MMSE scores (t = −9.21, p < 0.001), and were rated higher on the CDR SB scores (t = 21.26, p < 0.001). Several of the biomarkers were shown to be significantly different between diagnostic groups (see Table 2) including TNFα (p < 0.001), IL10 (p = 0.001), TARC (p = 0.035), FABP (p = 0.002), THPO (p = 0.006), A2M (p = 0.006), IL18 (p < 0.001), IL6 (p = 0.003), sVCAM1 (p = 0.024), FVII (p = 0.035), sICAM1 (p = 0.015), CA125 (p = 0.017), and B2M (p = 0.028).

Table 1.

Demographic characteristics

	aMCI Mean (SD) n = 73	Normal Control Mean (SD) n = 211	p value
Gender (% male)	21%	36%
Age	66.30 (8.45)	58.75 (6.29)	<0.001*
Education	6.96 (4.79)	8.94 (4.41)	0.001*
MMSE	23.64 (3.69)	27.13 (2.40)	<0.001*
CDR SB	1.21 (0.82)	0.00 (0.00)	<0.001*

^*p < 0.05.

Table 2.

Descriptives for twenty-one proteins examined across amnestic MCI and cognitively normal cases

	aMCI Mean (SD) n = 73	Normal Control Mean (SD) n = 211	p value
A2M, μg/L	2,368,148,245 (936,101,608.0)	2,087,517,920 (666,234,588.9)	0.006*
B2M, μg/L	2,618,737.02(949,306.1)	2,235,257.61 (1,339,209.6)	0.028*
CA125, μg/L	37.18 (30.13)	30.19 (16.83)	0.017*
CRP, μg/L	4,157.44 (4,819.75)	4,195.8 (5,151.98)	0.956
Eotaxin3, μg/L	3.87 (23.15)	1.61 (12.53)	0.333
FABP, μg/L	88,853.4 (50,038.39)	67,331.63 (51,830.85)	0.002*
FVII, μg/L	1,145,559.99 (276,851.39)	1,220,950.29 (256,906.34)	0.035*
I309, μg/L	5.48 (5.68)	6.40 (7.71)	0.356
IL10, μg/L	5.48 (5.68)	3.64 (6.15)	0.001*
IL18, μg/L	326.96(178.91)	256.47 (125.28)	<0.001*
IL5, μg/L	1.08 (1.76)	0.76(1.41)	0.149
IL6, μg/L	5.08 (10.68)	2.52(1.97)	0.003*
IL7, μg/L	9.54 (3.38)	10.10(4.19)	0.356
PPY, μg/L	1,000.86 (268.38)	945.84 (293.77)	0.161
SAA, μg/L	7,625.27 (8,083.61)	8,233.98 (21,965.34)	0.817
sICAM1, μg/L	366.11 (107.82)	334.09 (92.29)	0.015*
sVCAM1, μg/L	513.61 (138.72)	468.78 (147.42)	0.024*
TARC, μg/L	237.01 (132.89)	179.83 (128.41)	0.035*
THPO, μg/L	744.51 (232.78)	652.57 (246.51)	0.006*
TNC, μg/L	35,944.79 (13,899.35)	37,604.07 (10,747.58)	0.294
TNFα, μg/L	5.36 (1.65)	3.08 (1.82)	<0.001*

^*p <0.05.

First, the biomarker profile for aMCI was generated using RF within the training set. Among the top 21 markers, 18 were found to be overexpressed among aMCI cases whereas 3 were under expressed. The overall biomarker profile for aMCI was then compared to the profile of the top 30 markers previously identified among Mexican Americans with AD [43]. When examining the Gini plot, the top markers for aMCI were shown to be inflammatory in nature whereas in our prior work looking at markers of AD among Mexican Americans the profile was weighted towards metabolic proteins (see Fig. 1).

Fig. 1.

Gini Plot of top 21 serum biomarkers of aMCI among Mexican Americans. Note: The top markers for aMCI provided by the Gini Plot supported an inflammatory profile as an increase in inflammatory based biomarkers were shown.

The overall accuracy of the proteomic profile in determining aMCI was 0.96 (95% CI: 0.916–1) with a sensitivity of 0.82 (95% CI: 0.685–0.914) and a specificity of 0.97 (95% CI: 0.941–0.995)(see Fig. 2). The clinical variables alone (age, gender, and education) yielded an AUC of 0.68 (95% CI = 0.549–0.757), sensitivity of 0.24 (95%CI = 0.130–0.381), and specificity of 0.96 (95% CI = 0.922–0.988). When the demographic variables were combined with the proteomic profile, the overall accuracy remained at 0.96 (95% CI: 0.915–1) while specificity increased to 0.99 (95% CI: 0.962–0.999) and sensitivity decreased to 0.74 (95% CI: 0.596–0.853) (see Fig. 3).

Fig. 2.

Area under the curve for proteomic profile of aMCI. Note: The overall accuracy (area under the curve) of the 21 proteins in determining aMCI was 0.96 (95% CI: 0.916–1).

Fig. 3.

Area under the curve for proteomic profile, age, gender, and education for aMCI. Note: The overall accuracy (area under the curve) of the 21 proteins in combination with demographic variables in determining aMCI was 0.96 (95% CI: 0.915–1).

In order to put the findings into the context of clinical prediction, PPV and NPV predictive values were calculated as described above. This reflects the likelihood that a clinician would be correct (PPV) or incorrect (NPV) in classifying a Mexican American elder as aMCI based on the current findings. Assuming a base rate of 10%, PPV = 0.89 and NPV = 0.97 based on the sensitivity and specificity estimates outlined above. Assuming a 20% base rate, PPV = 0.95 and NPV = 0.94. Presented another way, the Likelihood ratio of a positive test = 15.29 (95% CI: 14.06–16.62) and likelihood ratio of a negative test = 18.29 (95% CI: 14.77–22.64).

DISCUSSION

Hispanics are among the fastest growing segment of the aging population and suffer a disproportionate risk for the development of neurodegenerative diseases. Cost and time-efficient tools for early diagnostics are imperative to aid with treatment and overall healthcare. Such methods are also needed for primary care providers to meet the Centers for Medicare and Medicaid Services (CMS) cognitive exam as part of annual wellness visits. Our group was among the first to examine risk factors for MCI and AD among both a rural as well as community based sample of Mexican Americans. Additionally, work from our group was the first to identify proteomics of AD among this underserved population. The current work was an effort to 1) expand toward examining pre-AD stages (aMCI) through the utility of proteomics and 2) extend prior efforts towards examining MCI among a community based sample.

The utilization of an algorithmic approach towards identifying aMCI among a sample of Mexican Americans was shown to be effective. Overall, the algorithm, irrespective of inclusion of demographics factors, was able to maintain an excellent diagnostic accuracy of 96%. This work expands upon our prior algorithmic approach toward identifying AD among Mexican Americans, which demonstrated an overall accuracy level of 88% (SN=0.83, SP=0.78) [40]. The top markers among AD were shown to be metabolic (FABP, GLP-1, CD40); however, within the current study, the top markers of aMCI among Mexican Americans were inflammatory in nature (IL10, TARC, TNF-alpha). Interestingly, in our preliminary work (unpublished data), metabolic markers continue to be among the top within the profile that predicts AD among Mexican Americans. It is possible that the underlying causes of aMCI and AD among this ethnic group change with the progression of the disease such that initial inflammation triggers later metabolic processes, which is reflected by the change in inflammatory to metabolic markers found with disease progression. This work suggests that additional work is needed to understand the interplay between metabolic and inflammatory mechanisms during the transition from MCI to AD. Furthermore, replication work is needed to better clarify if the results found were due to biological changes related to ethnicity or if the results are specific to the target population studied.

The molecular markers of AD among non-Hispanic whites displayed a comparable inflammatory based profile with several of the top proteins (TNF-alpha and IL10) overlapping with the top proteins found among Mexican Americans with aMCI [43]. As Mexicans Americans have been shown to present with cognitive decline earlier than their non-Hispanic white counterparts, the overlap in inflammatory based proteins suggest a comparable underlying inflammatory process, which may precede metabolic changes later seen among this subgroup. Prior work has demonstrated the impact of inflammatory changes on the development of AD [44–47] and therefore, detection of specific inflammatory changes identified through blood-based biomarkers may serve to identify a possible therapeutic window for treatment opportunities.

The vast majority of prior work seeking to identify blood-based biomarkers of MCI and AD have been conducted through dementia specialty clinic settings (e.g., Alzheimer’s Disease Neuroimaging Initiative, Australian Imaging, Biomarkers and Lifestyle study, Texas Alzheimer’s Research & Care Consortium) with few studies examining community-based cohorts. The primary utility of blood-based biomarkers in the multi-stage detection process is within primary care settings in order to enhance referrals for patients to receive advanced neuroimaging, CSF, and clinical examinations, such is the case with cardiology and oncology. Additionally, the base rate (i.e. underlying prevalence of the condition within the population) of AD and MCI within primary care clinics and community-based settings is drastically lower than that observed in specialty clinics, which has a tremendous impact on the predictive accuracy of a test [42]. Therefore, as with our prior work, we applied the appropriate estimated base rate for the calculation of PPV and NPV [43] and demonstrated that the current approach will yield clinically meaningful outcomes if implemented within community-based and primary care settings. Lastly, given that Mexican Americans rarely seek specialty clinic examinations [7], the current work is novel as it was conducted within a community-based setting.

Due to sample size limitations, this study was unable to examine non-amnestic MCI (naMCI) cases. Future work should look towards examining the proteomic profile of Mexican American among naMCI cases, as it is likely that the proteomic profile varies. Furthermore, the proteomics selected for this study were previously identified to be significant predictors of AD. Additional research is needed to explore other potential blood biomarkers among Mexican Americans. Also, ApoE was not examined within this study; however, future work should examine the potential effect of ApoE on the biomarkers associated with MCI as it has been shown to impact inflammatory markers [48, 49]. Additionally, future research should focus on the biological properties underlying the distinctive profile of AD and MCI among Mexican Americans to better adjudicate mechanisms for targeted treatments. Future work should also examine the proteomic profile of those who progress from MCI to AD, a component this study was unable to examine due to limited sample size of those who progressed.

Another limitation posed is related to the diagnosis of MCI as it serves to capture a heterogeneous group of individuals. Our study sought to examine specifically those with memory-related impairment (aMCI); however, despite this effort, aMCI is still not a homogenous group and reflects some degree of heterogeneity. Some of the heterogeneity within MCI may be due to the myriad of etiologies that contribute to the disease state and associated with the various comorbid chronic disease states. Within the Mexican American population, elevated rates of comorbid disorders such as chronic kidney disease (CKD), diabetes, and hypertension have been found [4, 5, 27, 28]. Related to this, recent work has utilized biomarkers in an effort to detect proteomic profiles of cognitive impairment related to specific comorbid disease states. An example of this is Szerlip and colleagues [28], who identified proteomic profiles specific to CKD related MCI. This work was the first of its kind to examine comorbid disease specific risk pathology among those within the MCI diagnostic category. Their group was able to identify a unique profile for those with CKD who were at risk for MCI [28]. This work lends to future research examining the utility of proteomics in identifying specific types of MCI outside of its amnestic and non-amnestic forms.

This is the first-ever study of a proteomic profile of aMCI among Mexican Americans within a community-based setting. Additional work is needed in order to move the work from a laboratory test to an in vitro diagnostic; however, the current work sets the stage for that work. Despite the National Alzheimer’s Plan Act call for targeted recruitment of minority populations into clinical trials and research of neurodegenerative disease, less than 1% of all patients enrolled into AD clinical trials have been Hispanic [50]. The establishment of a blood-based screening tool would greatly increase access to clinical trials through primary care settings where Mexican Americans receive the vast majority of care.