Ethics approval

The BLSA study protocol has ongoing approval from the Institutional Review Board (IRB) of the National Institute of Environmental Health Science, National Institutes of Health (“Early Markers of Alzheimer’s Disease (BLSA)”, IRB No. 2009–074). Informed written consent was obtained at each visit from all participants.

The ADNI study protocol was approved by the IRBs of all the participating institutions/study sites [5]. Informed written consent was obtained from all participants at each site. All ADNI studies are conducted according to Good Clinical Practice guidelines, the Declaration of Helsinki, and United States of America 21 CFR Part 50 (Protection of Human Subjects) and Part 56 (IRBs). Additional details can be found at adni.loni.usc.edu.

The ROSMAP study, including the parent study and substudies, was approved by the IRB of Rush University Medical Center. Informed written consent was obtained for all participants as well as an Anatomical Gift Act and a repository consent to share data and biospecimens.

CPRD data are anonymized, general practitioners do not need to seek patient consent when sharing data with CPRD, and patients have the option of opting out. Additional details can be found at https://www.cprd.com/public. This study was approved by the CPRD Independent Scientific Advisory Committee (ISAC; Protocol # Protocol 18_173) and exempted from full IRB review by the National Institutes of Health Office of Human Subject Research.

Step 1: Test associations between cholesterol catabolism (i.e., BA synthesis) and neuroimaging markers of dementia

We performed targeted metabolomics assays measuring the principal cholesterol breakdown products (i.e., CA and CDCA) as well as their principal biosynthetic precursor, (7α-hydroxycholesterol; 7α-OHC) in serum samples from participants in the Baltimore Longitudinal Study of Aging Neuroimaging (BLSA-NI) substudy who also underwent in vivo brain amyloid positron emission tomography (PET) and longitudinal structural magnetic resonance imaging (MRI).

In order to validate index results from BLSA, we used the ADNI sample to test associations between CA and CDCA and neuroimaging outcomes (note: 7α-OHC was not assayed in the ADNI serum samples).

Study participants

The BLSA is a prospective cohort study that began in 1958 and is administered by the National Institute on Aging (NIA) [6]. BLSA-NI substudy imaging and visit schedules have varied over time and have been described in detail previously [7] and included in S1 Text.

ADNI is an ongoing longitudinal study launched in 2003. The primary goal of ADNI has been to test whether longitudinal MRI, PET, and other biological markers can measure the progression of mild cognitive impairment (MCI) and early AD. ADMC performed serum BA assays in participants enrolled in ADNI, and data are publicly available. Study design details have been published previously [5] and are available at www.adni-info.org.

Quantitative serum metabolomics assays

Blood serum samples were collected from BLSA participants at each visit; details on collection and processing have been published previously [8] and included in S1 Text.

Blood serum samples were collected from ADNI participants at baseline; details on collection and processing have been published previously [9,10] and are described in detail in the Biospecimen Results section (link: AD Metabolomics Consortium Bile Acids Methods (PDF)–Version: January 21, 2016) at http://adni.loni.usc.edu. Data used in this study are available in the Biospecimen Results section (link: AD Metabolomics Consortium Bile Acids–Post Processed Data [ADNI1, GO, 2]–Version: June 28, 2018).

In vivo brain amyloid imaging, WMLs, and brain volumes

BLSA-NI participants underwent ¹¹C-Pittsburgh compound-B (PiB) PET scans to assess brain amyloid-β burden. A detailed description of acquisition and preprocessing procedures has been published previously [11]. Individuals were characterized as amyloid +ve or amyloid −ve based on a mean cortical distribution volume ratio (cDVR) threshold of 1.066 [11]. Among amyloid +ve individuals, we examined mean cDVR, a weighted global average of brain amyloid deposition, and regional DVR in the precuneus, a region vulnerable to early amyloid deposition in AD [12]. The total sample included 141 individuals (66 male; 75 female) of whom 36 were amyloid +ve (21 male; 15 female).

BLSA brain MRI was performed on a 3T Philips Achieva scanner (Philips Healthcare, Netherlands) to quantify both global and regional brain volumes and WMLs. A detailed description is included in S1 Text.

We a priori defined a set of brain regions to examine brain atrophy over time based on prior work using BLSA-NI data suggesting that these regions were sensitive to age-related change [7]. These regions included global brain volumes: total brain, ventricular cerebrospinal fluid (CSF), total gray matter, and white matter; lobar volumes: temporal, parietal, and occipital white matter and gray matter; and additional regions sensitive to early neurodegeneration: hippocampus, entorhinal cortex, amygdala, parahippocampal gyrus, fusiform gyrus, and precuneus.

All BLSA MRI data, including brain volumes and WMLs, after onset of clinical symptoms among individuals who developed MCI or AD were excluded (21 visits). The total sample included 134 individuals (62 male; 72 female) with an average of 2.5 longitudinal MRI visits (male: 2.3; female: 2.6).

ADNI brain MRI was used to quantify both global and regional brain volumes. A detailed description of acquisition and preprocessing is included in S1 Text. As our analyses in the ADNI sample were performed to confirm index results from BLSA, we restricted these analyses to gray matter and subcortical brain regions described above, excluding all white matter regions based on the lack of associations in BLSA analyses.

ADNI is enriched for individuals with MCI and AD at baseline, and all data across baseline diagnoses (control (CON), MCI, and AD) were included in analyses. Similar to our primary analyses in BLSA, MRI data from all CON individuals in ADNI, after onset of clinical symptoms among individuals who subsequently developed MCI or AD, were excluded (200 visits). The total sample included 1,666 individuals (918 male; 748 female) with an average of 5.2 longitudinal MRI visits (male: 5.3; female: 5.1).

Statistical methods

Step 1 of the analytic plan using BLSA data were developed in January 2018 prior to starting analyses in June 2019. The inclusion of ADNI BA and neuroimaging data to validate significant BLSA findings and sensitivity analyses (i.e., adding statin as a covariate) were performed in June 2020 in response to reviewer recommendations.

Similar to prior work in BLSA [8], metabolite concentrations above the upper limit of quantification (ULOQ) were excluded, concentrations below the limit of detection (LOD) were imputed as the threshold LOD/2, and resulting concentrations were natural log transformed. Outliers ± 3 × interquartile range (IQR) were excluded.

For ADNI, data processing steps for serum BA concentrations have been described in detail previously [9,10]. Metabolite concentrations below the LOD were imputed as LOD/2, and all samples were log₂ transformed. Outliers ± 3 × IQR were excluded.

To test for group differences between amyloid +ve and amyoid −ve individuals, we examined associations between serum concentrations of metabolites (i.e., 7α-OHC, CA, and CDCA) and brain amyloid deposition, in overall and sex-stratified linear regression models with metabolites as the dependent variable and the binary brain amyloid variable (i.e., amyloid +ve/amyloid −ve) as the main predictor. Covariates included mean-centered age and sex in the overall model and mean-centered age only in the sex-stratified model. We next tested the association between metabolite concentrations and mean cDVR (BLSA) and precuneus DVR (BLSA) in amyloid +ve individuals only. We used similar linear regression models for the continuous DVR predictors. The significance threshold was uncorrected and set at p = 0.05 to accommodate the limited sample size.

To test the association between serum concentrations of metabolites (i.e., 7α-OHC, CA, and CDCA) and longitudinal changes in (1) regional brain volumes and (2) WMLs, we used total and sex-stratified linear mixed models with brain regions of interest (ROIs) volumes and WMLs as the dependent variable (i.e., outcome) and metabolite concentration as the predictor. We first performed analyses in BLSA and then validated results in ADNI. The statistical significance threshold for both BLSA and ADNI was set at a false discovery rate (FDR)-corrected p = 0.05. Additional details on model specifications are included in S1 Text.

In sensitivity analyses, we explored the effect on associations of adding statin drug use as a covariate in BLSA models.

Step 2: Test whether pharmacological modulation of BAs alters dementia risk in a large, real-world clinical dataset

In order to extend findings from our Step 1 analysis relating BA levels with brain amyloid accumulation, rates of brain atrophy, and progression of WMLs, we next tested whether pharmacological modulation of de novo BA synthesis influences dementia risk. As BAS are lipid-modifying treatments that are known to decrease the circulating pool of BAs [13–15] and promote the breakdown of cholesterol, we hypothesized that exposure to these drugs would alter the risk of dementia. We therefore tested associations between exposure to BAS and dementia risk using data from the UK’s CPRD, an anonymized electronic health record (EHR) covering more than 4 million active registrants from the UK general practice clinics. Our results from Step 1 suggested a plausible sex difference in the effect of BAS on dementia risk, a hypothesis that we tested using the CPRD.

Data source and study population

The CPRD is a primary care database covering >four million active registrants from >650 general practice clinics and is representative of the broader UK population in terms of age and sex [16].

From the August 2018 CPRD data release, we identified patients ≥18 years old who had a first prescription record (i.e., new users) for BAS (colestipol, colesevelam, and cholestyramine) or non-statin lipid-modifying therapies (LMT; fibrate, cholesterol absorption inhibitor, nicotinic acid derivative, and probucol) between January 1, 1995 and August 1, 2018. BAS are often used as a second-line treatment independently or in combination with statins, and therefore, we selected non-statin LMT users as an active comparator group. In both groups (BAS or non-statin LMTs), we allowed for prior statin use in combination with either BAS or LMTs. Individuals who only had a prescription record of statin use were excluded from this study.

The index date was defined as the date of the first BAS or LMT prescription. For each BAS user, we selected up to five LMT users matched on sex, year of birth (±5 year), region, year of clinic registration (±2 year), and year of first prescription (±2 years). Analysis was restricted to those with at least 12 months of clinical registration prior to the index date (to allow for covariate evaluation). We restricted BAS users to those aged ≥50 years and to those with two or more BAS/LMT prescriptions. The final analysis included 3,208 (1,083 male; 2,125 female) new BAS users and 23,483 (8,977 male; 14,506 female) new LMT users (Fig 1).

Download:

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

Fig 1. Flowchart of study participants included in the CPRD analyses.

BAS, bile acid sequestrants including colestipol, colesevelam, and cholestyramine; CPRD, Clinical Practice Research Datalink; LMT, lipid-modifying therapies including fibrate, cholesterol absorption inhibitor, nicotinic acid derivative, and probucol.

https://doi.org/10.1371/journal.pmed.1003615.g001

The outcomes of interest were all-cause dementia, and its subtypes: AD, VaD, and other dementia not otherwise specified (NOS). We used the last reported dementia diagnosis to identify the disease subtype (read codes are available upon request). The significance threshold was set at p = 0.05 considering that each outcome of interest was a priori specified.

Statistical methods

Step 2 of the analytic plan using CPRD data was developed in January 2018 prior to starting data analyses in June 2019. We added a comparison of patient characteristics across outcome categories (i.e., dementia subtypes) based on reviewer recommendations.

We compared patient characteristics and their comorbidity profiles across dementia subtypes (AD, VaD, and NOS) as well as drug use (BAS and LMT) using the chi-squared test for categorical variables and Wilcoxon rank-sum tests for continuous variables.

For multivariable analyses, we used Cox proportional hazard models to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) comparing dementia risk (all-cause and subtypes) in BAS versus LMT users in the overall and sex-stratified samples. Our sex-specific analyses were a priori specified and based on findings from Step 1. We also tested the dose–effect relationship between dementia risk and drugs of interest using the number of prescriptions. Models were adjusted for factors that were significantly different between BAS and LMT groups to account for potential confounding by indication (since patient comorbidity profiles can lead to a BAS versus LMT prescription decision). Models were also adjusted for statin use during follow-up (until one year prior to exit date) using a time-varying covariate to account for its impact on dementia (26% versus 80% of BAS and LMT users, respectively, were prescribed statins in the 12 months before the index date and the majority (63%) continued its use for all or part of the follow-up). See S1 Text for details on model specifications.

Step 3: Test plausible molecular mechanisms relating BA signaling in the brain to dementia pathogenesis using targeted metabolomics and transcriptomics

Given our findings that peripheral levels of the primary BAs, CA, and CDCA are associated with neuroimaging markers of dementia (Step 1) and that their pharmacological manipulation influences dementia risk (Step 2), we hypothesized that alterations in brain BA-mediated signaling may be a plausible biological mechanism underlying these findings. We first tested whether concentrations of CA and CDCA were detectable in the brain and whether they were altered in AD in participants from the BLSA autopsy program. We then tested whether gene expression of BA receptors was altered in AD using single-cell RNA sequencing (scRNA-Seq) data from the ROSMAP autopsy program.

Study participants

The autopsy program of the BLSA was initiated in 1986 and has been described previously [17]. See S1 Text for additional details. Tissue samples from AD (n = 16) and CON (n = 13) from the inferior temporal gyrus (ITG) and middle frontal gyrus (MFG), regions representing areas of early neurofibrillary (i.e., tau) and neuritic plaque (i.e., amyloid) accumulation, respectively [18,19], as well as the cerebellum (CB) were included in these analyses.

scRNA-Seq gene expression data [20] from ROSMAP [21] were downloaded from Synapse (https://www.synapse.org/#!Synapse:syn18485175) under the doi 10.7303/syn18485175; code used to run analyses presented in Mathys and colleagues [20] was requested from coauthors. Data came from postmortem participants in ROSMAP including 46 individuals: 32 individuals (18 male and 14 female) in the AD category and 14 individuals (5 male and 9 female) in the CON category. The AD category included individuals with a clinical diagnosis of AD, including individuals with AD and no other condition contributing to cognitive impairment and AD and another condition contributing to cognitive impairment, as well as individuals with a clinical diagnosis of MCI and no other condition contributing to cognitive impairment. The CON category included individuals with a clinical diagnosis of no cognitive impairment.

Tissue was profiled from the prefrontal cortex (Brodmann area 10) across eight major cell types in the aged dorsolateral prefrontal cortex including inhibitory neurons, excitatory neurons, astrocytes, oligodendrocytes, microglia, oligodendrocyte progenitor cells, endothelial cells, and pericytes. Additional details are provided in the index paper [20]. We identified BA receptor genes (including receptors involved in BA homeostasis) using a literature search [22–24] and include the full list in S2 Table. There were 21 BA receptor genes that had available data in the ROSMAP dataset: Nuclear Receptor Subfamily 1 Group I Member 3 (NR1I3); Retinoid X Receptor Gamma (RXRG); Nuclear Receptor Subfamily 5 Group A Member 2 (NR5A2); Cholinergic Receptor Muscarinic 3 (CHRM3); G Protein-Coupled Bile Acid Receptor 1 (GPBAR1); Peroxisome Proliferator Activated Receptor Gamma (PPARG); Nuclear Receptor Subfamily 1 Group I Member 2 (NR1I2); Kinase Insert Domain Receptor (KDR); Nuclear Receptor Subfamily 3 Group C Member 1 (NR3C1); Retinoid X Receptor Beta (RXRB); Peroxisome Proliferator Activated Receptor Delta (PPARD); Cholinergic Receptor Muscarinic 2 (CHRM2); Retinoid X Receptor Alpha (RXRA); Nuclear Receptor Subfamily 1 Group H Member 3 (NR1H3); Vitamin D Receptor (VDR); Nuclear Receptor Subfamily 1 Group H Member 4 (NR1H4); Retinoic Acid Receptor Alpha (RARA); Hepatocyte Nuclear Factor 4 Alpha (HNF4A); Nuclear Receptor Subfamily 1 Group H Member 2 (NR1H2); Formyl Peptide Receptor 1 (FPR1); Peroxisome Proliferator Activated Receptor Alpha (PPARA).

Quantitative brain metabolomics assays

Quantitative metabolomics assays were performed on brain tissue samples to measure concentrations of the primary BAs, including CA and CDCA, using the Biocrates Bile Acids kit (Biocrates Life Sciences AG, Austria). Details on both assay kits, as well as calibration steps, have been published previously [25]. Additional details regarding the use of internal standards are included in S1 Text.

Statistical methods

Step 3 of the analytic plan using BLSA data was developed in January 2020 in order to address a plausible molecular mechanism explaining findings from Step 1 and Step 2. The inclusion of scRNA-Seq data from ROSMAP occurred in June 2020 in response to reviewer recommendations to use non-array, non-bulk tissue-based gene expression data.

In order to assess whether primary BAs were present in the brain, we visualized CA and CDCA concentrations in AD and CON samples in the ITG, MFG, and CB using dot plots. Concentrations above and below LOD were indicated. We used tobit regression models to determine whether mean metabolite concentrations were significantly different between AD and CON samples. We set the lower limit as the metabolite-specific LOD threshold and included covariates age and sex (mean centered). In brain regions where metabolite concentrations were all below LOD (i.e., CB), we used chi-squared tests to determine whether percentage of samples below LOD was significantly different between AD and CON samples. Due to a small number of individuals with BA metabolite values above LOD, we were not able to sex-stratify these analyses. Additionally, the statistical significance threshold was set at p = 0.05 to accommodate the limited sample size.

For gene expression data, we scaled each sample to have the same total read count. To test differences between AD and CON, we used the Wilcoxon rank-sum test in the total and sex-stratified samples. Similar to the index paper [20], each single-cell–specific sample from a participant was treated as an independent sample. We summarized age- and sex-corrected fold changes (total sample) as well as sex-specific fold changes indicating whether genes were differentially expressed in AD versus CON samples. We additionally visualized results for significant associations using a heatmap. The significance threshold was set at an FDR-corrected p = 0.05.

Step 1: Test associations between cholesterol catabolism (i.e., BA synthesis) and neuroimaging markers of dementia

Participant demographic details are included in S3A and S3B Table. Results of cross-sectional analyses testing associations between serum metabolite concentrations and amyloid status in BLSA are included in S4 Table, and associations between serum metabolite concentrations and brain amyloid-β burden among amyloid +ve individuals are included in Table 1. Brain amyloid +/−ve status was not significantly associated with serum concentrations of 7α-OHC, CDCA, or CA in the total or sex-stratified samples. In the total sample and in males only, serum 7α-OHC concentration, representing the rate-limiting biosynthetic precursor of the primary BAs [26,27], was significantly, negatively associated with mean cDVR (p = 0.034 and p = 0.041, respectively) (Fig 2) and precuneus DVR (p = 0.033 and p = 0.022, respectively), indicating that lower serum concentration of 7α-OHC was associated with higher levels of global and precuneus brain amyloid-β deposition. We observed no significant associations in the female-only sample.

Download:

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

Fig 2. Association between serum 7α-OHC concentration and brain amyloid-β burden in amyloid +ve individuals.

Weighted global average of brain amyloid-β burden (mean cDVR); analyses restricted to amyloid +ve individuals (n = 36). 7α-OHC, 7α-hydroxycholesterol; cDVR, cortical distribution volume ratio.

https://doi.org/10.1371/journal.pmed.1003615.g002

Download:

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

Table 1. Associations between serum metabolite concentrations and brain amyloid-β deposition, rates of brain atrophy, and longitudinal changes in global brain WML burden–BLSA.

https://doi.org/10.1371/journal.pmed.1003615.t001

Download:

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

Table 2. Associations between serum metabolite concentrations and rates of brain atrophy and longitudinal changes in global brain WML burden–ADNI.

https://doi.org/10.1371/journal.pmed.1003615.t002

Results of longitudinal analyses in BLSA testing associations between serum metabolite concentrations and brain atrophy are shown in Table 1. In males, lower serum CDCA and CA concentrations were associated with faster rates of atrophy in the parietal gray matter (CDCA: FDR p = 0.003; CA: FDR p = 0.013) and precuneus (CDCA: FDR p < 0.001; CA: FDR p < 0.001). In females, lower serum CA concentration was associated with slower total gray matter atrophy (FDR p = 0.038).

Sensitivity analyses including statin drug use as a covariate are included in S5 Table; results were not substantially altered.

Results of longitudinal analyses in ADNI testing associations between serum metabolite concentrations and brain trophy are shown in Table 2. In the total sample, lower serum CDCA and CA concentrations were associated with faster rates of atrophy in the entorhinal cortex (CDCA: FDR p = 0.032; CA: FDR p = 0.009), frontal gray matter (CDCA: FDR p = 0.045; CA: FDR p = 0.005), fusiform gyrus (CDCA: FDR p = 0.012; CA: FDR p = 0.001), total gray matter (CDCA: FDR p = 0.030; CA: FDR p = 0.003), hippocampus (CDCA: FDR p = 0.030; CA: FDR p = 0.012), parahippocampal gyrus (CDCA: FDR p = 0.012; CA: FDR p = 0.009), temporal gray matter (CDCA: FDR p = 0.016; CA: FDR p = 0.002), and ventricles (CDCA: FDR p = 0.030; CA: FDR p = 0.008). Lower CA was also associated with faster rates of atrophy in the amygdala (FDR p = 0.030), occipital gray matter (FDR p = 0.012), parietal gray matter (FDR p = 0.016), and precuneus (FDR p = 0.030). In males, lower serum CDCA and CA concentrations were associated with faster rates of atrophy in the parahippocampal gyrus (CDCA: FDR p = 0.049; CA: FDR p = 0.049). Lower serum CDCA was associated with faster rates of atrophy in the ventricles (CDCA: FDR p = 0.049), and lower serum CA was associated with faster rates of atrophy in the entorhinal cortex, frontal gray matter, fusiform gyrus, total gray matter, hippocampus, and temporal gray matter (FDR p = 0.049). In females, CDCA was not significantly associated with rates of brain atrophy; lower CA was associated with faster rates of atrophy in only the fusiform gyrus and temporal gray matter (FDR p = 0.039).

Results of longitudinal analyses in BLSA testing associations between serum metabolite concentrations and WML are shown in Table 1. In males, lower serum CDCA concentration was associated with faster accumulation of WML (p = 0.050), and in females, lower serum 7α-OHC was associated with slower accumulation of WML (p = 0.010).

Results of longitudinal analyses in ADNI testing associations between serum metabolite concentrations and WML are shown in Table 2. In ADNI, we did not observe significant associations between serum metabolite concentrations of BAs (i.e., CA and CDCA) and WML in the total male or female samples.

Step 2: Test whether pharmacological modulation of BAs alters dementia risk in a large, real-world clinical dataset

S6 Table summarizes characteristics of BAS and LMT users. LMT users were more likely to be overweight or obese compared with BAS users (73% versus 57%, respectively). In the 12 months prior to index date, LMT users were more likely to have used statins (80% versus 26%, respectively) or metformin (15% versus 7%) and had a record of coronary artery disease (7% versus 3%), type 2 diabetes (7% versus 3%), or dyslipidemia (25% versus 5%). BAS users were more likely to have a prior record of cancer (16% versus 8%).

Table 3 summarizes results from Cox proportional hazard models. During the median follow-up of 4.9 years, 809 incident dementia cases occurred (N = 72 for BAS versus 737 for LMT) corresponding to crude incidence rates of 4.8 (95% CI = 3.8 to 6.1) and 5.5 per 1,000 person-years (95% CI = 5.1 to 5.9) among BAS and LMT users, respectively. In multivariable adjusted models including all patients and compared to LMT use of > = two prescriptions, BAS use was not statistically significantly associated with risk of all-cause dementia (Table 3) or with its subtypes (any dementia: HR = 1.03, 95% CI = 0.72 to 1.46, p = 0.88; AD: HR = 1.24, 95% CI = 0.72 to 2.14, p = 0.43; VaD: HR = 1.27, 95% CI = 0.70 to 2.31, p = 0.43; other dementia: HR = 0.50, 95% CI = 0·22 to 1.15, p = 0.10).

Download:

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

Table 3. Association between BAS use and the risk of dementia in individuals who received at least two BAS or LMT prescriptions with at least one year of follow-up after second prescription.

https://doi.org/10.1371/journal.pmed.1003615.t003

In analyses stratified by sex, we observed a significant (p = 0.040) difference between the HR of VaD in males compared to females, indicating a sex difference in the relationship between BAS and risk of VaD. BAS use was associated with nonsignificantly elevated risk of VaD in males (HR = 2.89, 95% CI = 0.96 to 8.68, p = 0.06). We identified a statistically significant dose–response relationship between BAS and risk of VaD in males. Specifically, risk of VaD was higher with the increased number of BAS prescriptions (p-trend = 0.045) (Table 3). There was no statistically significant association with VaD in females (overall or by number of prescriptions). Differences in patient characteristics across outcome categories are included in S7 Table.

Step 3: Test plausible molecular mechanisms relating BA signaling in the brain to dementia pathogenesis using targeted metabolomics and transcriptomics

Participant demographic details for the BLSA autopsy study are included in S8 Table. Demographic details of ROSMAP participants included in scRNA-Seq analyses have been published previously [20].

The primary BAs, CDCA, and CA were detectable in postmortem brain tissue samples, although the majority were below the LOD (i.e., <LOD) (Fig 3, S9 Table). Tobit regression models indicated marginally higher (nonsignificant) concentrations of the primary BAs in AD samples compared to CON samples in the ITG and MFG. Chi-squared models indicated significantly more participants with metabolite concentrations above LOD in AD compared to CON in the CB. Due to a small number of individuals with BA metabolite values above LOD, we were not able to sex-stratify these analyses.

Download:

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

Fig 3. Differences in brain primary BA concentrations between AD and CON.

Dots in gray indicate concentrations below the LOD; dots in blue and red indicate AD and CON brain tissue sample metabolite concentrations, respectively. p-Values indicate differences between AD and CON from a tobit regression model; the lower limit in the model is set at the metabolite-specific LOD (indicated by the dashed line). The p-value for CDCA in the CB is derived from a chi-squared test because there were no detectable concentrations above the LOD in the CON sample. AD, Alzheimer disease; BA, bile acid; CB, cerebellum; CDCA, chenodeoxycholic acid; CON, control; ITG, inferior temporal gyrus; LOD, limit of detection; MFG, middle frontal gyrus.

https://doi.org/10.1371/journal.pmed.1003615.g003

We observed that gene expression of several BA receptors was different in AD versus CON in the total and male samples, mainly within neurons (i.e., both inhibitory and excitatory neurons). The majority of differentially expressed genes showed lower expression in AD relative to CON. Results across all 8 major brain cell types are included in S10 Table. As indicated below in a heatmap visualizing sex-stratified differences in AD versus CON samples (Fig 4), for inhibitory neurons, in males, 10 out of 21 genes were significantly altered (FDR pval < 0.05); six had lower gene expression in AD compared to CON (AD<CON); and four had higher gene expression in AD compared to CON (AD>CON). In females, there were no differentially expressed BA receptor genes within inhibitory neurons. Within excitatory neurons, in males, 16 out of 21 genes were significantly altered (FDR pval < 0.05); 10 had lower gene expression in AD compared to CON (AD<CON); and 6 had higher gene expression in AD compared to CON (AD>CON). For females, four genes showed lower gene expression in AD compared to CON (AD<CON).

Download:

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

Fig 4. Differences in brain BA receptor gene expression between AD and CON.

Summary of differentially expressed BA receptor genes (including receptors involved in BA homeostasis) in neurons in AD compared to CON. Statistically significant (FDR-corrected p-value < 0.05) fold change differences are indicated in green or red shading. Green shading indicates that gene expression was significantly reduced in AD compared to CON. Red shading indicates that gene expression was significantly increased in AD compared to CON. Gray shading indicates gene expression was not significantly different between AD and CON. AD, Alzheimer disease; BA, bile acid; CHRM2, Cholinergic Receptor Muscarinic 2; CHRM3, Cholinergic Receptor Muscarinic 3; CON, control; FDR, false discovery rate; FPR1, Formyl Peptide Receptor 1; NR1H2, Nuclear Receptor Subfamily 1 Group H Member 2; NR1H3, Nuclear Receptor Subfamily 1 Group H Member 3; NR1H4, Nuclear Receptor Subfamily 1 Group H Member 4; NR1I2, Nuclear Receptor Subfamily 1 Group I Member 2; NR1I3, Nuclear Receptor Subfamily 1 Group I Member 3; NR3C1, Nuclear Receptor Subfamily 3 Group C Member 1; NR5A2, Nuclear Receptor Subfamily 5 Group A Member 2; PPARA, Peroxisome Proliferator Activated Receptor Alpha; PPARD, Peroxisome Proliferator Activated Receptor Delta; PPARG, Peroxisome Proliferator Activated Receptor Gamma; RARA, Retinoic Acid Receptor Alpha; RXRA, Retinoid X Receptor Alpha; RXRB, Retinoid X Receptor Beta; VDR, Vitamin D Receptor.

https://doi.org/10.1371/journal.pmed.1003615.g004