Hearing Loss in Alzheimer’s Disease Is Associated with Altered Serum Lipidomic Biomarker Profiles

Recent data have found that aging-related hearing loss (ARHL) is associated with the development of Alzheimer’s Disease (AD). However, the nature of the relationship between these two disorders is not clear. There are multiple potential factors that link ARHL and AD, and previous investigators have speculated that shared metabolic dysregulation may underlie the propensity to develop both disorders. Here, we investigate the distribution of serum lipidomic biomarkers in AD subjects with or without hearing loss in a publicly available dataset. Serum levels of 349 known lipids from 16 lipid classes were measured in 185 AD patients. Using previously defined co-regulated sets of lipids, both age- and sex-adjusted, we found that lipid sets enriched in phosphatidylcholine and phosphatidylethanolamine showed a strong inverse association with hearing loss. Examination of biochemical classes confirmed these relationships and revealed that serum phosphatidylcholine levels were significantly lower in AD subjects with hearing loss. A similar relationship was not found in normal subjects. These data suggest that a synergistic relationship may exist between AD, hearing loss and metabolic biomarkers, such that in the context of a pathological state such as AD, alterations in serum metabolic profiles are associated with hearing loss. These data also point to a potential role for phosphatidylcholine, a molecule with antioxidant properties, in the underlying pathophysiology of ARHL in the context of AD, which has implications for our understanding and potential treatment of both disorders.


Introduction
Aging-related hearing loss (ARHL) and Alzheimer's Disease (AD) are common disabling disorders in the elderly. Over the age of 65, approximately 10% of individuals develop AD, while approximately 40% develop ARHL [1,2]. Both disorders are rising in prevalence as the population ages, and an

Clinical Diagnosis and Hearing Loss Assessment
AD was diagnosed using NINCDS/ADRDA criteria for probable AD [28]. MCI patients had a memory complaint, an abnormal score on the Logical Memory II subscale from the Wechsler Memory Scale, a Mini-Mental Status Exam score between 24-30 and a Clinical Dementia Rating scale score of 0.5. Normal subjects did not have a memory complaint, had a normal score on the Logical Memory II subscale and had a Clinical Dementia Rating scale score of zero. Hearing was not systematically measured in the ADNI database. Similar to a previous report [21], we used subjective hearing loss complaints found in the following datasheets: ADSXLIST.csv, BLSCHECK.csv, INITHEALTH.csv, MEDHIST.csv, NEUROEXM.csv, PHYSICAL.csv, RECBLLOG.csv, RECMHIST.csv. We used the search terms "hear", "auditory", "ear", "deaf", "presbycusis" and "HOH (hard of hearing)" and eliminated those reports that were clearly not related to aging-related hearing loss (e.g., skin cancer on ear, earwax, etc.), as well as entries that referred to tinnitus without mention of hearing loss and eliminated duplicates. These search terms are identical to those used by Xu et al. (2019) and were selected prior to the data being seen. Subjects with a hearing complaint are labeled in this study as "hearing loss" or HL. Other subjects are listed as "non-hearing loss" or NHL, notwithstanding the fact that hearing was not objectively measured (see below).

Statistical Methods
The effect of each individual lipid species on hearing loss in AD subjects was assessed via analysis of covariance (ANCOVA) after adjusting for gender and age as covariates, and log transforming the lipid expression values. Samples with an absolute value of studentized residuals from this model exceeding 3 were identified as outliers and excluded from further analysis. The summary measures reported from this analysis include the area under the receiver operating characteristic curve (ROC AUC), covariate-adjusted significance (p-value) and false discovery rate [29].
The effect of each of the 16 known lipid classes and 28 empirically derived lipid sets (Barupal et al., 2019) on hearing loss in AD subjects was assessed via "lipid set analysis" (LSA). See Supplementary  Table S1 for a list of the lipids in each of the 28 sets. This LSA of the lipid classes and lipid sets was based on the maxmean statistic of the gene set analysis algorithm [30], which was applied on the residuals from the above ANCOVA model on the individual lipid species to adjust for the effects of age and gender. Individual subject-level standardized composite scores were determined for each lipid class and each lipid set from this algorithm. These scores were then used to assess the effect of each of the lipid classes and lipid sets on hearing loss in AD subjects. The results were summarized in terms of ROC AUC, covariate-adjusted significance (p-value) and false discovery rate (q-value). Lipid sets with q-value < 0.05 were considered as statistically significant. The corresponding lipid classes and individual lipid species with Bonferroni-adjusted p-value < 0.05 were highlighted and studied further in terms of their potential connections to hearing loss in AD subjects.

Lipidomic Biomarker Sets That Separate HL from NHL Subjects
Levels of 349 lipids were measured across 16 classes. Because the levels of many of the lipids are strongly correlated due to co-regulation, and because of the high potential for false discovery when comparing the levels of all 349 lipids, we attempted to reduce the data by grouping the lipids. A previous report measured correlations between all of the serum lipid biomarkers, and using a dynamic clustering algorithm known as dynamicTreeCut, determined that 28 co-regulated sets of lipids were present [23]. They also found that many of these lipid sets were associated with either AD diagnosis or AD biomarkers. Although most of the sets were homogeneous (or near-homogeneous) clusters of single lipid types, others comprised a mixture of lipids (see supplementary Table S1 for a list of lipids in each class).
Given the robust performance of these clusters to signal changes in AD biomarkers, we asked whether these same clusters were also associated with the presence of HL. The pand q-values for the 28 groups of lipids are shown in Table 3. We found that two sets of lipids correlated with the presence of hearing loss: set 23 and set 4, both with pand q-values below 0.05, with set 23 producing the best performance. We therefore focused on the lipids found in these two sets for subsequent analyses of lipid class and individual lipids.

Lipid Classes and Individual Lipids That Separate HL from NHL Subjects
Using the biomarker sets to narrow our hypotheses about which lipids exhibit signal changes in hearing, we attempted to determine which lipid classes were most significantly associated with HL. Within the two significant sets identified above (q < 0.05), 25 lipids in seven classes were identified, with only the phosphatidylcholine class surviving correction for multiple comparisons (uncorrected p-value = 0.0057, Bonferroni corrected to 0.04). See Figure 1 for boxplots of the seven biomarker classes comparing HL and NHL subjects. See Table 4 for a list of lipid classes found in sets 4 and 23 and their associated capacity to separate HL from NHL subjects.  Among the 25 lipids in the two significant lipid sets identified above, the most commonly appearing lipid class was phosphatidylcholine (14/25 lipids or 56%), which is significantly greater than the proportion of all tested lipids that were in the phosphatidylcholine class (82/349 lipids or 23.4%, p < 0.05, chi-squared test). See Table 5 for a list of individual lipids in sets 4 and 23 and their associated capacity to separate HL from NHL subjects. Both of these analyses point to phosphatidylcholine levels as the main factor distinguishing between HL and NHL subjects.  Among the 25 lipids in the two significant lipid sets identified above, the most commonly appearing lipid class was phosphatidylcholine (14/25 lipids or 56%), which is significantly greater than the proportion of all tested lipids that were in the phosphatidylcholine class (82/349 lipids or 23.4%, p < 0.05, chi-squared test). See Table 5 for a list of individual lipids in sets 4 and 23 and their associated capacity to separate HL from NHL subjects. Both of these analyses point to phosphatidylcholine levels as the main factor distinguishing between HL and NHL subjects.

Analysis of Non-AD Subjects and Apolipoprotein E (APOE)
Similar analyses were done in subjects with MCI (n = 225, 64 with HL) and control subjects without memory loss (n = 373, 104 with HL). None of the lipid sets were found to differentiate HL from NHL subjects in either control or MCI cohorts (see Table 6). Interaction of disease diagnosis (AD, MCI, control subjects) and hearing loss status (HL, NHL) with respect to specific lipid classes was formally assessed within the framework of a two-way ANOVA. Post hoc evaluation of this interaction effect from this model revealed that phosphatidylcholine was significantly differentiated between HL vs. NHL only in the AD subjects (p < 0.05), but not in the MCI and control subjects. Subjects across all groups (control, MCI and AD) were also separated based on genotype (having at least one copy of APOE4 or none), and no association was found between genotype and likelihood of HL.

Discussion
In the current study, 349 serum biomarkers were measured in 185 subjects with AD. Using previously identified co-regulated sets of biomarkers [23], we found two sets of lipids that were strongly associated with the presence of HL. Within these sets, the most common class of lipids was phosphatidylcholine, and as a class and as individual biomarkers, phosphatidylcholines were found to be significantly diminished in individuals with HL. Similar analyses in non-AD subjects (control and MCI) did not reveal significant associations between lipidomic biomarkers and HL

Weaknesses in the Study
Hearing loss in this study was assessed in a non-systematic way-via subjective reports obtained from the subjects. Using the National Health and Nutrition Examination Survey (NHANES), which captured both objective hearing loss (using pure tone audiograms) and subjective hearing loss, previous data have established concordance values between subjective and objective hearing loss ranging from 65-77% depending on demographic factors [31]. Older subjects, such as the ones in this study, tended to underestimate their degree of hearing loss. These data suggest that some subjects with HL may have inappropriately been placed in the NHL category, and vice versa, but with a greater likelihood of missing HL subjects. Although there are several publicly available databases that have measured hearing loss objectively (e.g., the Baltimore Longitudinal Study of Aging or National Health and Nutrition Examination Survey), these did not systematically measure an extensive panel of lipid biomarkers. Conversely, despite the richness of biomarker data available in the ADNI, hearing was not systematically measured. Thus, additional future work in subjects with objectively-measured hearing loss will be required to confirm the associations reported here.
In addition, it is not possible to extrapolate the current findings to a therapeutic intervention. As an observational study, the current work cannot be used to support the idea that supplementation of phosphatidylcholine can protect against ARHL in subjects with AD. It is possible that phosphatidylcholine levels and ARHL are related by a third, unmeasured, factor. Only a prospective, randomized and blinded trial can determine whether phosphatidylcholine can improve ARHL.

Phosphatidylcholine, Alzheimer's Disease and Hearing Loss
Phosphatidylcholine is one of the major phospholipids and a fundamental constituent of cell membranes and may activate enzymatic antioxidants situated in the cell membrane. There is also evidence for disrupted phosphatidylcholine metabolism in AD. For example, the enzymes that break down phosphatidylcholine (phospholipase D and phospholipase A2) are altered in AD [32,33]. In addition, low plasma levels of phosphatidylcholine docosahexaenoic acid have been associated with the development of AD [34] as well as thinning of the prefrontal cortex [35]. With respect to ARHL, phosphatidylcholine's protective role in hearing loss was suggested by work from Seidman et al., who observed that lecithin (a polyunsaturated phosphatidylcholine) can protect against aging-related hearing loss in rats [36]. In this study, the investigators observed higher mitochondrial membrane potentials in the lecithin-treated group, suggesting preserved mitochondrial function. Lecithin treatment also diminished the occurrence mtDNA4834 deletion (common aging-related mitochondrial deletion) in the brain and cochlear tissue of the treated group. These data point to a role of phosphatidylcholine in protecting cochlear mitochondrial function. In addition, the antioxidants activated by phosphatidylcholine may protect the cell membrane from damage by reactive oxygen species [37] that arise during aging-related cochlear hypoperfusion, which can lead to cochlear degeneration [38,39]. These data all suggest that phosphatidylcholine levels may be depleted in AD and ARHL.

Origins of Measured Lipids
The lipids measured in this study were extracted from blood samples, which brings about the question of the origins of these lipids. Dietary fats are absorbed into the portal system to the liver. In the liver, fatty acids are incorporated into lipoprotein particles which are then released into the bloodstream. Additionally, adipocytes can release stored fatty acids into the blood as lipid levels in the blood decrease. Evidence also suggests that some fatty acids can be synthesized in the brain, but essential fatty acids still have to be transported across the blood-brain barrier [40]. Additional studies done on adult rats to study the rate of polyunsaturated fatty acid incorporation from plasma into the brain further suggests that this is a dynamic process with active daily turnover [41]. The exact mechanism behind how fats enter the brain is still unclear. One study performed on cholesterol homeostasis and hearing loss indicates that since the blood-brain barrier prevents the uptake of this lipoprotein from circulation, brain cholesterol is synthesized in astrocytes; further, excess cholesterol is metabolized into 24 (S)-hydroxycholesterol before secretion from the blood-brain barrier to the liver [42]. Thus, measured lipids in this study are likely derived from a variety of sources.

Conclusions
In the current study, we observed that in the context of AD, lower serum levels of phosphatidylcholine were associated with ARHL. The fact that this association was found in AD subjects, but not in non-AD subjects, suggests that there is an interaction between the presence of AD and the relationship between phosphatidylcholine and ARHL. Given that AD is associated with diminished brain mitochondrial function and increased levels of lipid peroxidation, it is possible that individuals with AD may not have the metabolic reserve to withstand additional metabolic stressors, such as declining levels of antioxidant molecules such as phosphatidylcholine. These data also suggest that normalizing phosphatidylcholine levels in AD subjects, but not in non-AD subjects, may have a role in the treatment or prevention of ARHL. Future studies will need to be done to investigate the potential therapeutic role of phosphatidylcholine in this context.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2073-4409/9/12/2556/s1. Author Contributions: D.A.L.: conceived of the study, curated data, wrote paper, L.K.I.: curated data, wrote paper, P.D.: wrote paper, V.D.: conceived of the study, curated data, wrote paper, analyzed data. All authors have read and agreed to the published version of the manuscript.
Funding: Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). The named authors did not receive specific funding for this work.