Abstract
Using 1H NMR metabolomics, we biochemically profiled saliva samples collected from healthy-controls (n = 12), mild cognitive impairment (MCI) sufferers (n = 8), and Alzheimer’s disease (AD) patients (n = 9). We accurately identified significant concentration changes in 22 metabolites in the saliva of MCI and AD patients compared to controls. This pilot study demonstrates the potential for using metabolomics and saliva for the early diagnosis of AD. Given the ease and convenience of collecting saliva, the development of accurate and sensitive salivary biomarkers would be ideal for screening those at greatest risk of developing AD.
INTRODUCTION
Alzheimer’s disease (AD) is a neurodegenerative disorder for which there is no cure and few reliable diagnostic biomarkers [1]. It is characterized by the accumulation of amyloid-β plaques and tau tangles [2, 3]. Mild cognitive impairment (MCI) is progressive degree of impairment that is greater than might be attributed to normal age-related cognitive decline, but is not so severe as to merit a diagnosis of dementia. MCI is thought to be a transitional state between normal aging and AD [4]. The conversion rate from MCI to AD is estimated to be approximately 10% per year [5] and that number increases every year [6]. Current therapies for AD are initiated only after diagnosis. The development of valid and reliable biomarkers for AD will not only aid clinicians in recognizing the disease in its earliest symptomatic stages, but will be especially important should effective disease prevention or modifying therapies be developed. Sensitive biomarkers would permit intervention before substantial neuropathological damage has occurred and before the manifestation of dementia. Researchers are currently developing novel treatments for AD that could potentially prevent neurodegeneration [7]. The development of early biomarkers is a necessary first step to the design of such prevention and early- intervention trials.
Metabolomics is a discipline dedicated to the global study of small molecules in cells, tissues, or biofluids [8]. It involves the comprehensive, simultaneous, and systematic profiling of multiple metabolite concentrations and can measure fluctuations in these in response to disease, drugs, diet, and lifestyle. The value of metabolic profiling techniques for successfully distinguishing neurodegenerative diseases from healthy controls have been demonstrated [9–13]. This technique has recently been shown to be very successful in identifying biomarker panels for diagnosing AD using plasma [14]; however, this study has yet to be validated.
Saliva is a clear, watery biofluid produced by the salivary glands to protect and lubricate the oral cavity. The chemical composition of saliva changes quite dramatically in response to a variety of different physiological states, stimuli, insults, and stressors making it a good candidate for monitoring biological responses of the body to any directed case [15, 16]. This is because many of the components in serum and cerebrospinal fluid pass into the saliva via the blood by transcellular, intracellular, paracellular, or extracellular routes involving passive diffusion or active transport within the salivary glands and the gingival sulcus [17]. Unlike blood and cerebrospinal fluid, saliva is easily available through non-invasive means and represents a constant source of potential material for diagnostics.
A recent study by Figueira et al. [18] demonstrated the use of metabolomics and 1H NMR for the identification of biomarkers for those suffering from dementia, to include AD and vascular dementia [18]. Further, Kim et al. [19] reported using saliva as a biomarker medium with diagnostic potential for AD. In this study they employed antibody-conjugated magnetic particles to measure the levels of the Aβ peptides in normal subjects and in patients with AD or mild cognitive impairment [20]. This study builds on previous reports [18] and represents a potential opportunity for the surveillance and identification early of individuals at risk of conversion to MCI and in turn AD, and therefore delineate a population for early intervention. Using a 1H NMR based metabolomics approach, saliva from patients suffering from MCI and AD was biochemically profiled and compared with healthy controls. Our goal for this study is to determine whether salivary biomarkers can distinguish MCI and AD from healthy control patients.
MATERIALS AND METHODS
Patient characteristics
Study subjects were recruited from an academic geriatric practice that is heavily focused on memory disorder (Supplementary Table 1). All subjects underwent for following cognitive assessments: a) Clinical dementia rating scale (CDR), b) Mini-Mental Status Examination (MMSE), c) logical memory test, d) digit span forward and backward, e) category fluency test, f) ordering test, g) trails A&B, and finally Geriatric Depression Scale.
Samples collection
Human saliva samples were collected from adult volunteers (12 Controls, 8 MCI, and 9 AD). The subjects were instructed to refrain from eating, drinking, smoking, or using any oral hygiene products for at least 1 h prior to saliva collection. Study participants were asked to rinse their mouths with water for 5 min and were subsequently instructed to spit into 50-cc Falcon tubes. The subjects were reminded not to cough throughout the collection process. The saliva samples were centrifuged at 2600×g for 35 min at 4°C to remove any sedimentary material [21]. Subsequently, aliquots of the supernatant were stored in Eppendorf tubes at –80°C until analyzed.
Sample preparation
Samples thawed at room temperature were filtered through 3-kDa cut-off centrifuge filter units (Amicon Micoron YM-3; Sigma-Aldrich, St. Louis, MO) to remove any proteins. Sample preparation was completed following the protocol presented by Dame et al. [22]. Samples were analyzed in a randomized order and maintained at 4°C prior to analysis using the state-of-the-art SampleJetTM (Bruker) automated sample changer.
NMR analysis
All 1H-NMR experiments were recorded at 300.0 (±0.05) K on a Bruker Avance III HD 600 MHz spectrometer (Bruker-Biospin, USA) operating at 600.13 MHz equipped with a 5 mm TCI cryoprobe. Data collection was conducted as previously described by Ravanbakhsh et al. [23]. The singlet produced by the DSS methyl groups was used as an internal standard for chemical shift referencing (set to 0 ppm, concentration 500μM) and for quantification. All 1H-NMR spectra were processed and analyzed using the Chenomx NMR Suite Professional Software package version 8.1 (Chenomx Inc,Edmonton, AB).
Statistical analysis
Normalized and Pareto scaled data were analyzed using logistic regression to generate optimal predictive models for MCI and AD. Prior to logistic regression analysis unsupervised PCA was performed to make sure that no outlier was incorporated into the analysis. Significant metabolites subsets were generated using Random Forest analysis. Subsequently, stepwise variable selection was utilized to optimize prediction model components via 10-fold cross-validation. The area under the receiver operating characteristics curve (AUROC or AUC) was calculated using previously described techniques [24]. Sensitivity and specificity values were also calculated for each model.
RESULTS
Logistic regression models were used for the detection of MCI and AD. The following metabolites: i) galactose, imidazole and acetone; ii) creatine and 5-aminopentanoate; and iii) propionate and acetone were used for distinguishing controls versus MCI, MCI versus AD, and controls versus AD, respectively. Some of the observed variances in said biomarkers were also relatively large (Supplementary Figure 1). Using logistic regression modelling, statistically significant prediction of MCI and AD from controls and MCI from AD were achieved.
The logistic regression models are represented below;
The descriptive characteristic for the ROC and the logistic regression analysis based on metabolites of interest are available in Fig. 1 and Table 1, respectively.
Receiver operating characteristics (ROC) curve analysis of metabolite data. a) The logistic regression ROC analysis for controls versus MCI sufferers (AUC: 0.826 (0.634–1.000), Sensitivity: 0.909 (0.909–1.000), Specificity: 0.897 (0.707–1.000). b) The logistic regression ROC analysis of MCI sufferers versus AD (AUC: 0.871 (0.689–1.000), Sensitivity: 0.909 (0.909–1.000), Specificity: 0.842 (0.678–1.000). c) The logistic regression ROC analysis for controls versus AD (AUC: 0.897 (0.707–1.000), Sensitivity: 0.900 (0.900–1.000), Specificity: 0.944 (0.839–1.000)). All models were validated following a 10-fold cross validation.
Results for the pairwise logistic regression modeling for healthy control versus MCI sufferers, MCI sufferers versus AD patients and healthy controls versus AD patients
DISCUSSION
Here we present the first NMR based metabolomics study discriminating MCI sufferers, AD patients, and healthy controls from each other. Previous studies have used similar techniques as presented herein but differed in that they attempted to identify a panel of diagnostic salivary biomarkers of dementia in general (AD and vascular dementia combined) [18]. In this study, we analyzed specimens collected from AD patients, MCI sufferers, and corresponding age- and gender-matched controls. Using the 1H NMR acquired data we positively identified (Supplementary Figure 2) and accurately quantified 57 metabolites (Supplementary Table 2) in saliva.
Our results demonstrate that there are significant differences in the concentrations of a large number of salivary metabolites in AD and MCI versus unaffected controls. Similarly, differences were found when the AD and the MCI groups were compared. Based on the metabolite concentration data we were able to generate regression models that significantly differentiated both MCI and AD cases from controls. The regression model with the greatest predictive ability was created when separating controls from MCI using the concentrations of galactose, imidazole, and acetone with an AUC = 0.826 (95% CI: 0.634–1.00) and with a sensitivity and specificity of 0.909 and 0.889, respectively. When the concentration values of creatinine and 5-aminopentanoate from MCI sufferers AD patients and were analyzed using logistic regression an AUC value of 0.897 (0.707∼1.000) with a sensitivity and specificity of 0.900 and 0.944 was achieved.
Furthermore, the logistic regression model based on the concentration values of propionate and acetone for separating controls from AD patients produced an AUC = 0.871 (0.689–1.000) with 0.909 and 0.842 sensitivity and specificity values, respectively. Of all the logistic regression models created, controls versus MCI was found to be the weakest. This could be the result of the heterogeneous nature of MCI; 10% of the MCI sufferers progress to AD while a small percentage regress to being considered healthy controls.
While this study is novel, there are several obvious limitations associated with it. The main limitation being the small sample size, which limits what we can take from the results. However, as metabolomics is a hypothesis generating discipline, we first needed to complete a small study to determine how many samples would be required to achieve significance. Having completed the study a power analysis using the acquired data shows that we require a minimum of n = 100 samples per group to develop significant models (p < 0.05) with 90% power. Further, we only use one analytical platform for our analyses. We understand that by including other techniques such as mass spectrometry coupled with either liquid or gas chromatography, we will significantly increase our coverage of the salivary metabolome and identify potentially stronger diagnostic biomarkers of AD.
Conclusions
In this pilot study, we provided preliminary evidence that salivary metabolites may be useful for AD biomarker development. Given the convenience and the frequency with which saliva can be obtained, larger studies are justified.
DISCLOSURE STATEMENT
Authors’ disclosures available online (http://j-alz.com/manuscript-disclosures/16-1226r2).
Footnotes
Appendix
ACKNOWLEDGMENT
This work was partly funded by the generous contribution made by the Fred A. & Barbara M. Erb Foundation.