Bile acids are associated with baseline and longitudinal amyloid and tau pathology in patients with Alzheimer's disease

Abstract

INTRODUCTION

Alzheimer's disease (AD), a neurodegenerative disease, involves early alterations in the gut microbiota. Bile acids (BAs), which are metabolites produced by the microbiota, may impact brain function through the gut‐brain axis.

METHODS

By reference to multimodal datasets from the Chinese Preclinical Alzheimer's Disease Study (n = 1397) and the Alzheimer's Disease Neuroimaging Initiative (n = 1275), we analyzed differences in BA levels and their associations with AD biomarkers.

RESULTS

Lithocholic acid (LCA) –family BAs are associated with the amyloid‐β status. Longitudinal changes in BA levels correlated with amyloid and tau pathologies. LCA and the deoxycholic acid (DCA) family exhibited predictive value with respect to AD pathology. Imaging transcriptomic analyses suggested that BAs modulated amyloid pathology through multiple mechanisms.

DISCUSSION

DCA‐ and LCA‐family BAs were proposed as molecular bridges that connect age signatures with AD pathology. They represent a new avenue for the development of biomarkers and therapeutic interventions.

Highlights

• Metabolomics data from two large‐scale cohorts: the Chinese Preclinical Alzheimer's Study (CPAS; n = 1397) and the Alzheimer's Disease Neuroimaging Initiative (ADNI; n = 1275) with cerebrospinal fluid (CSF) or positron emission tomography (PET) –defined Alzheimer's disease (AD) were analyzed, confirming bile acid alterations by amyloid‐β status.

• Spatial transcriptomic analyses revealed previously unexplored mechanisms by which bile acids influence amyloid pathology initiation and progression, involving energy metabolism, immune modulation, signal transduction, and ion channel/receptor regulation.

• Lithocholic acid (LCA) and deoxycholic acid (DCA) demonstrated strong predictive value for AD pathology, with high performance in detecting amyloid and tau pathology (area under the curve [AUC] = 0.73–0.78). Longitudinal analyses showed that bile acid predicts future amyloid and tau pathology progression, providing the first evidence for their prognostic value over time.

1. BACKGROUND

Alzheimer's disease (AD) is the most common cause of dementia in older people and is characterized by the presence of extracellular amyloid‐β (Aβ) aggregates and the deposition of intracellular neurofibrillary tangles consisting of hyperphosphorylated microtubule‐associated tau protein. ¹ , ² The current diagnosis of this disease relies on the collection and testing of cerebrospinal fluid (CSF) samples and positron emission tomography (PET) imaging. These characteristic pathological changes usually begin to appear years before cognitive impairment occurs. However, indicators that can predict the progression of these pathologies at an early stage are lacking. Beyond these classical features, metabolic dysfunction, including perturbations in lipid homeostasis, mitochondrial energetics, and neuroinflammation, is increasingly implicated in AD pathogenesis. ³ , ⁴ , ⁵ , ⁶ Critically, the gut–brain axis serves as a key regulator of these metabolic processes, with dysbiosis in AD patients altering the production of microbial metabolites such as bile acids (BAs). ⁷

Primary bile acids are predominantly synthesized in hepatocytes via classical and alternative pathways and are subsequently converted into secondary bile acids through deconjugation and dehydroxylation reactions. ⁸ , ⁹ Previous studies have indicated that BAs are directly or indirectly associated with AD. ⁴ , ¹⁰ Importantly, elevated taurolithocholic acid (TLCA) levels correlates with disease severity, ¹¹ , ¹² whereas unconjugated primary bile acids are also associated with cognitive decline. ¹³ These peripheral BAs can cross the blood–brain barrier, where they activate nuclear receptors such as farnesoid X receptor (FXR) and G protein‐coupled bile acid receptor (TGR5) to modulate AD pathology. ¹¹ , ¹⁴ , ¹⁵ , ¹⁶ Specifically, deoxycholic acid (DCA) promotes Aβ generation by binding to TGR5 to regulate amyloid precursor protein processing. ¹⁴ In contrast, tauroursodeoxycholic acid (TUDCA) exerts anti‐inflammatory effects on microglia. ¹⁷

Despite these advances, the existing evidence relies primarily on cross‐sectional designs, and comprehensive longitudinal relationships between BA profiles and evolving AD pathologies remain uncharacterized. Moreover, the potential mechanisms underlying these associations remain undefined. We addressed these gaps by analyzing the associations of blood‐based metabolites with CSF biomarkers and AD pathologies measured by PET imaging from the Chinese Preclinical Alzheimer's Disease Study (CPAS) and the Alzheimer's Disease Neuroimaging Initiative (ADNI) (Figure 1). We explored the relationships of BAs with baseline and longitudinal amyloid and tau pathologies to validate the potential of the BA signature as an early predictive marker of AD progression and to identify BA‐associated genes and pathways driving AD pathologies by performing an imaging transcriptomic analysis.

FIGURE 1.

Graphic overview of this study. The design of this study focused on sample collection and measurements, analyses of baseline and longitudinal data, and imaging transcriptomics studies. (A) This study included the CPAS and ADNI cohorts. Neuropsychological tests, [¹⁸F]‐florbetapir PET scans, and plasma biomarker and metabolite measurements were completed in the CPAS cohort. (B) Baseline data obtained from these two cohorts and sequential analyses: profiles of identified bile acids across patients stratified by cognitive status and amyloid pathology, correlations with AD pathologies and predictive value. (C) Imaging transcriptomics research consisted of two steps: spatial BA–amyloid mapping and gene set enrichment analysis to elucidate the function of genes regulating BAs associated with amyloid pathology. (D) Longitudinal data obtained from the ADNI cohort were used to determine the associations between longitudinal changes in bile acid levels and pathological and cognitive decline. ADNI, Alzheimer's Disease Neuroimaging Initiative; BA, bile acid; CPAS, Chinese Preclinic Alzheimer's Study; PET, positron emission tomography.

2. METHODS

2.1. Study cohort and samples

The CPAS followed the principles of the Declaration of Helsinki and was approved by the Institutional Review Boards of Fudan University Affiliated Huashan Hospital and Shanghai Jiao Tong University Affiliated Sixth People's Hospital. The inclusion and exclusion criteria and other information about CPAS have been reported previously. ¹⁸ The participants were included from the memory clinic and communities in Shanghai, and written informed consent was also obtained. The ADNI is an ongoing and longitudinal study designed to determine the progression of AD by assessing multiple dimensions, including clinical assessments, PET, and other biological markers. Details on participant recruitment and study approval have been published previously. ¹⁹

RESEARCH IN CONTEXT

1. Systematic review: Patients with Alzheimer's disease (AD) exhibit disturbances in bile acid (BA) metabolism. A systematic review of the literature using PubMed revealed that evidence linking bile acids to AD pathological progression and elucidating their underlying mechanisms remains limited and inconclusive.

2. Interpretation: Deoxycholic acid (DCA) and lithocholic acid (LCA) were associated with cognitive decline and exacerbation of amyloid and/or tau pathology and demonstrated good predictive performance. Available evidence suggests that bile acids may influence amyloid pathology through multiple biological pathways.

3. Future directions: This study underscores the potential role of bile acids in the pathological progression of AD and supports their use as early predictive biomarkers. Future studies incorporating longitudinal data from independent cohorts are warranted to validate these findings and clarify causal mechanisms.

2.1.1. CPAS

A total of 1397 participants in the CPAS whose plasma metabolites were available were included; 421 had undergone [¹⁸F]‐florbetapir (FBP) PET scans, and 416 of them had plasma biomarkers, including phosphorylated tau (p‐tau) 181 and Aβ42/40. The diagnosis of AD was based on the 2018 National Institute on Aging and Alzheimer's Association (NIA‐AA) criteria. In this study, Aβ deposition was determined by a visual reading according to the guidelines for interpreting amyloid PET scans. Three board‐certified nuclear medicine physicians, comprising two mid‐career specialists and one senior expert, independently assessed all scans while being blinded to the clinical, demographic, and diagnostic information. ²⁰ , ²¹ , ²² Scans were classified as positive for Aβ deposition only when at least two raters concurred on the tracer retention patterns. This consensus‐driven approach ensured diagnostic reliability while mitigating individual rater bias. The metabolomics data from the CPAS cohort comprise 22 bile acids, with 38 concentration ratios calculated based on these and employed for analysis. 

2.1.2. ADNI

A total of 1275 participants in the ADNI with data on serum metabolites were included; 830 of them had CSF biomarker data, and 519 had [¹⁸F]‐FBP PET scans. Amyloid and tau positive signals were determined by measuring CSF Aβ42 and CSF p‐tau levels according to their protocol. Specifically, the cutoff value for the CSF Aβ42 concentration was <1098 pg/mL, and the cutoff value for the CSF p‐tau concentration was >19.2 pg/mL. ²³ The metabolomics data used in this study were analyzed by the University of Hawaii Cancer Center. This laboratory measured a total of 104 metabolites, including 33 BAs, in serum samples from participants in ADNI 1, GO, and 2. Based on the measured concentrations of 33 BAs, 40 concentration ratios were additionally calculated and incorporated into the analysis. Detailed information on the collection and processing of samples, demographic information, and clinical data available in previous publications can be downloaded from https://adni.loni.usc.edu/.

2.2. Neuropsychological assessment

All participants in the CPAS underwent comprehensive neuropsychological assessments. Global cognitive function was evaluated using the Chinese version of the Mini‐Mental State Examination (MMSE) and Montreal Cognitive Assessment‐Basic (MoCA‐B). Domain‐specific cognitive functions were assessed on the basis of standardized tests targeting three core domains: memory was evaluated with the Auditory Verbal Learning Test (AVLT), including the total score, 30‐minute delayed free recall (AVLT‐LDR, 12 items), and recognition subtest (24 items); language abilities were measured using the Animal Fluency Test (AFT) total score and Boston Naming Test (BNT, 30 items); and executive function was examined using completion times for Parts A and B of the Shape Trails Test (STT). Diagnoses for CPAS participants were made based on the NIA‐AA 2018 diagnostic criteria for probable AD dementia. Individuals with mild cognitive impairment (MCI) were identified if they had an impaired score on both measures within at least one cognitive domain or one impaired score in each of the three cognitive domains, as we previously reported. ²⁴ In the present study, participants were grouped into two groups, namely, those with impaired cognition (CU) and those with impaired cognition (CI), two groups as determined by their results on comprehensive cognitive tests.

2.3. Plasma biomarker measurements

Measurements of plasma biomarkers were conducted with the assistance of the Quanterix Simoa HD‐1 platform. More specifically, the Neurology 3‐Plex A Assay Kit was used to determine Aβ42, Aβ40, and total‐tau (t‐tau) levels, the P‐tau 181 Assay Kit V2 was used to measure plasma p‐tau levels, and the NF‐light Assay Kit was used to measure plasma neurofilament light chain (NfL) levels. The pretreatment of the reagents and loading of the samples were performed according to the manufacturer's instructions. The concentration (pg/mL) of each plasma biomarker was determined using single‐molecule array (Simoa) HD‐1 Analyzer software.

All samples were measured in duplicate on the Quanterix HD‐X analyzer using the following reagent kits: the Neurology 3‐Plex A Kit (for Aβ42, Aβ40, and t‐tau), the P‐tau 181 V2 Kit (for p‐tau181), and the NF‐light Kit (for NfL). The instrument automatically acquired and calculated the average enzyme per bead (AEB) for each replicate well. The AEB values were then converted to analyte concentrations via fitting with a four‐parameter logistic (4PL) curve. Finally, the system compiled the concentration values from the duplicate measurements and calculated the mean, standard deviation, and coefficient of variation (CV), which served as the final concentration and repeatability metric for each sample. The results demonstrated that all biomarkers showed good agreement between duplicate measurements. The detection precision was highest for p‐tau181 and t‐tau, with the vast majority of samples exhibiting a CV below 5%. The duplicate results for Aβ42, Aβ40, and NfL were also stable, with CVs predominantly below 10%.

2.4. Measurement of metabolite levels

Among the numerous metabolites, the levels of 189 metabolites from 12 metabolite classes were stably measured using ultra performance liquid chromatography tandem mass spectrometry (UPLC‐MS/MS43). The raw data files were processed using TMBQ software (V1.0, HMI, Shenzhen, China), encompassing peak integration, calibration, quantification, quality control, and adjustment for batch effects each metabolite, in accordance with the manufacturer's guidelines. The outliers were identified through a Cauchy distribution robust fit (K sigma = 7). Any outliers (<0.2%) and missing values (<0.1%) were substituted using multivariate normal imputation. The data were subjected to logarithmic transformation (base = 2) to normalize their distribution for statistical analyze.

For quality control (QC), rigorous procedures were applied to ensure data reliability. Specifically, isotope‐labeled internal standards (IS) were added to each sample to correct for peak intensity variations. Within each batch, small pooled QC samples‐prepared by mixing a fraction of all samples in the batch‐were inserted at regular intervals (every 12 samples) to monitor and correct for intra‐ and inter‐batch variation. Additionally, three to five large QC samples, generated by pooling multiple normal serum samples, and two NIST SRM 909c reference sera were analyzed per batch to allow inter‐batch calibration and cross‐project comparability. Instrument stability was monitored by analyzing QC samples, maintaining a CV below 15% and retention time shifts within ±0.2 min. System suitability tests (SST) were performed prior to each analytical run to ensure consistency and reproducibility across batches.

2.5. PET and magnetic resonance imaging

Participants in the CPAS underwent PET/computed tomography (CT) imaging using [¹⁸F]‐FBP with the parameters described previously. ²⁵ , ²⁶ In brief, each subject received an intravenous bolus injection of [¹⁸F]‐FBP at ∼370 MBq (±10%). Twenty‐minute static PET acquisitions were performed at 50 min after the injection using a Biograph mCT Flow PET/CT system (Siemens, Germany). PET images were reconstructed using a filtered back‐projection algorithm incorporating corrections for decay, normalization, dead time, photon attenuation, scatter and random coincidences. The reconstructed images featured 168 × 168 × 148 voxel matrices with a 2.04 × 2.04 × 1.50 mm³ resolution.

Complementary structural magnetic resonance (MR) images were acquired on Siemens PRISMA 3.0T system via magnetization prepared rapid gradient echo (MPRAGE) sequence and sagittal scans at different locations. The imaging parameters included a repetition time = 3000 ms, an echo time = 2.56 ms, a flip angle = 7°, an acquisition matrix = 320 × 320, an in‐plane resolution = 0.8 mm × 0.8 mm, and a slice thickness = 0.8 mm, for a total of 208 sagittal slices.

2.6. Data preprocessing

Image processing was performed using an established pipeline with SPM12 (Welcome Trust Centre for Neuroimaging, London, UK; https://www.fil.ion.ucl.ac.uk/spm) ²⁰ : T1‐weighted MR images were coregistered to the PET data, followed by spatial normalization to the Montreal Neurological Institute (MNI) template space, and segmented into gray matter, white matter, and CSF components. Utilizing the transformation parameters derived from this MR normalization, all the PET images were then spatially normalized to the same MNI space for precise anatomical alignment. Finally, the PET data were smoothed using a Gaussian kernel with an 8‐mm full width at half maximum (FWHM). The standardized uptake value ratio (SUVR) for each region of interest (ROI) was calculated by normalizing the regional FBP uptakes to the uptake in the whole cerebellum, which was utilized as the reference region. Moreover, the centiloid units were transformed to estimate the standardized global amyloid burden more accurately. ²⁷

2.7. Imaging transcriptomics analysis of regional gene expression and enrichment analysis

We performed imaging transcriptomics analysis using previously described method to explore the functions of genes related to BA. ²⁸ Regional gene expression profiles were derived from microarray data of six post mortem human brains available in the Allen Human Brain Atlas (AHBA; https://netneurolab.github.io/neuromaps/). The raw expression data were processed using abagen, a standardized Python toolkit designed for preprocessing and analyzing AHBA microarray datasets. Generally, the steps included generating a gene expression map and a BA–amyloid correlation map. We projected the gene expression data into Schaefer 200 parcels to construct a gene expression map, and Pearson's correlation coefficient was calculated between the SUVR of each region and the level of each BA. The r values were subsequently used as the dependent factors to compute a map of the spatial correlation with gene expression using the spatial null module from NeuroMaps (https://netneurolab.github.io/neuromaps/). ²⁹ , ³⁰ Genes were considered to have expression profiles significantly associated with BA levels when they had a spatial correlation of p < 0.05. These genes were subsequently grouped into positive and negative correlation sets based on their r values and ranked accordingly. The top 200 genes positively and negatively correlated with each BA were then pooled to form a combined dataset. Enrichment analyses of the genes were subsequently performed using Gene Ontology (GO) and KEGG pathway enrichment analyses.

2.8. Statistical analysis

All the statistical analyses were implemented in R version 4.4.2. The analyses in this study were strictly confined to the subsets of participants with corresponding valid detection data. We applied z score normalization to the data within each cohort to better compare the data across cohorts, and the statistical significance was set at p < 0.05 after the Benjamini–Hochberg correction. Although FBP PET and CSF Aβ42 levels represent two distinct detection methods, both fundamentally observe amyloid. Therefore, based on the existing data types for the two cohorts and adhering to the principle of maximizing the sample size, we used the results of the FBP PET visual interpretation and CSF Aβ42 thresholds to distinguish amyloid‐negative/amyloid‐positive (A–/A+) subjects in the CPAS and ADNI cohorts, respectively. We assessed group differences in demographic characteristics and neuropsychological scores using the chi‐square test for categorical variables and the Mann–Whitney U test for nonnormally distributed continuous data across cognitive status and amyloid pathology categories. For the BA data across two cohorts, all statistical analyses and multiple corrections were applied independently. Specifically, inter‐group difference analyses, intra‐group longitudinal comparisons, and population‐wide correlation analyses were each adjusted according to the phenotypes, total number of bile acids and ratios covered by CPAS and ADNI, respectively. Multiple correction for group‐level correlation analyses was performed based on the bile acid families of interest identified in prior results, and according to the number of shared members within these families across both cohorts. After subjects with impaired cognition or positive Aβ deposition at baseline but unimpaired cognition or negative Aβ deposition in the longitudinal were excluded, the Wilcoxon signed‐rank test was used to compare differences in bile acid levels before and after the longitudinal visit within each group. Longitudinal trajectories of target biomarkers (BAs, CSF Aβ42, CSF p‐tau, and FBP‐PET centiloid values) and cognitive assessments were modelled using subject‐specific linear mixed‐effects regression analyses with time as the primary predictor. The fixed‐effect slopes derived in this context quantified linear rates of change in the level of each biomarker and were carried forward as dependent variables in association analyses. Multivariable linear regression models adjusted for age, sex, and apolipoprotein E (APOE) Ɛ4 status were used to examine cross‐sectional and longitudinal associations between BA levels and AD hallmarks.

Moreover, we performed a voxel‐wise analysis of amyloid PET scans with BAs levels as independent variables using SPM12 with a threshold of P_FDR < 0.05, unless specified otherwise, and adjusted for age and sex. Finally, BA levels were used as dependent variables to predict the hallmark classification by applying a logistic regression model, adjusted for age, sex, and APOE Ɛ4 status. For the classification of the pathological status, amyloid positivity was defined as a plasma Aβ42 concentration less than the median and tau positivity as a p‐tau concentration greater than the median in the CPAS cohort; the ADNI cohort, amyloid positivity was defined as a CSF Aβ42 concentration <1098 pg/mL, and tau positivity as a p‐tau concentration >19.2 pg/mL. FBP SUVR ≥ 1.11 was used as an indicator of Aβ PET positivity in ADNI. ³¹

3. RESULTS

3.1. Demographic characteristics and clinical assessments

A total of 1397 individuals (CU: n = 726; CI: n = 671) from the CPAS were included in this study. CI participants were older (68.7 vs. 63.8, P_FDR  < 0.0001) and had fewer years of education (10.6 vs. 12.2, P_FDR < 0.0001) than CU individuals. The prevalence of APOE Ɛ4 (39.9% vs. 18.7%, P_FDR  < 0.0001) was higher in CI individuals than in CU individuals. CI patients also had worse cognition than CU patients (MMSE score: 20.9 vs. 28.0, P_FDR  < 0.0001; MoCA‐B score: 17.4 vs. 25.5, P_FDR  < 0.0001) (Table 1). Four hundred twenty‐one participants with Aβ‐PET scans were divided into A– (n = 317) and A+ (n = 104) subgroups. A+ individuals have fewer years of education and poor cognitive assessment scores compared with A– individuals. More details can be found in the Table S1.

TABLE 1.

Sample characteristic.

	CPAS			ADNI
Parameter	All (n = 1397)	CU (n = 726)	CI (n = 671)	All (n = 1275)	CU (n = 377)	CI (n = 898)
Age	66.1 ± 8.6 b	63.8 ± 8.0	68.7 ± 8.5	73.6 ± 7.1 b	74.6 ± 5.8	73.2 ± 7.6
Male/Female	481/916 c	233/493	248/423	719/555 c	195/182	524/374
Education (year)	11.4 ± 3.2 b	12.2 ± 3.1	10.6 ± 3.2	15.9 ± 2.8 b	16.4 ± 2.7	15.8 ± 2.9
APOE Ɛ4 carrier%	28.8% c	18.7%	39.9%	46.2% c	27.9%	53.8%
MMSE	24.6 ± 5.6 b	28.0 ± 1.8	20.9 ± 6.1	27.3 ± 2.7 b	29.1 ± 1.1	26.5 ± 2.8
MoCA	23.3 ± 4.8 b	25.5 ± 3.2	17.4 ± 7.7	23.7 ± 3.7 b	25.7 ± 2.3	22.9 ± 3.8
Plasma Aβ42	9.5 ± 3.1	9.6 ± 3.0	9.4 ± 3.3	N/A	N/A	N/A
Plasma p‐tau181	2.3 ± 1.2 b	2.0 ± 0.9	2.8 ± 1.5	N/A	N/A	N/A
CSF Aβ42	N/A	N/A	N/A	1065 ± 625 b	1348 ± 687	952 ± 559
CSF p‐tau	N/A	N/A	N/A	27.6 ± 13.5 b	22.3 ± 9.0	29.7 ± 14.4
PET acceptance%	30.1% (421) c	36.4% (264)	23.4% (157)	40.7% (519)	41.6% (157)	40.3% (362)
Centiloid	3.1 ± 25.6 b	−3.3 ± 15.9	14 ± 34.1	39.0 ± 43.7 b	24.1 ± 37.0	45.4 ± 44.8
Aβ(+)% a	24.7% c	14.0%	42.7%	63.3% c	42.0%	71.8%
M12	N/A	N/A	N/A	1038	321	717
M24	N/A	N/A	N/A	886	292	594

Note: Data are presented as mean ± SD unless specified otherwise.

Abbreviations: ADNI, Alzheimer's Disease Neuroimaging Initiative; APOE Ɛ4 carrier%, the percentage of participants with APOE Ɛ4 genotype in all participants; CI, cognitive impaired; CPAS, Chinese Preclinical Alzheimer's Disease Study; CU, cognitive unimpaired; M12, the first 12‐month visitation after assessment at baseline; M24, the second 12‐month visitation after assessment at baseline; MMSE, Mini‐Mental State Examination. MoCA, Montreal Cognitive Assessment; PET, positron emission tomography; PET acceptance%, the percentage of participants who underwent the brain [¹⁸F]‐flobetapir PET scan.

^a The percentage of participants with positive amyloid (individuals from CPAS were defined through visual assessment following the guidance for interpreting amyloid PET, and those from ADNI cohort were defined based on their amyloid levels in CSF) in those who underwent the brain [¹⁸F]‐flobetapir PET scan or volunteered and consented to have a lumbar puncture.

^b Mann‐Whitney U test P_FDR  < 0.05 compared CU with CI.

^c Chi‐squared test P_FDR < 0.05 compared CU with CI.

A similar situation was observed in the ADNI cohort. A total of 1275 individuals (CU: n = 377; CI: n = 898) from the ADNI were included in this study. Compared with CU individuals, CI individuals were younger (73.2 vs. 74.6, P_FDR  = 0.08) and had lower levels of education (15.8 vs. 16.4, P_FDR  = 0.001). The prevalence of APOE Ɛ4 in the CI group was much higher (53.8% vs.27.9%, P_FDR < 0.0001). CI patients also had worse cognition than CU patients (MMSE score: 26.5 vs. 29.1, P_FDR  < 0.0001; MoCA score: 22.9 vs. 25.7, P_FDR  < 0.0001) (Table 1). Eight hundred thirty participants underwent CSF sample collection, 305 with negative Aβ deposition and 525 with positive Aβ deposition. A+ individuals have performed worse with cognition. Detailed results were listed in the Table S1.

3.2. Differences in the level of secondary BAs in AD patients

Previous studies have demonstrated that patients with AD exhibit changes in the levels of primary bile acids, secondary bile acids, and their ratios in blood compared with healthy controls. ¹⁰ In this study, we observed that the levels of several secondary bile acids, such as DCA, and the glycodeoxycholic acid/taurodeoxycholic acid (GDCA/TDCA) ratio, increased in the CI individuals compared with those in the CU individuals in the CPAS cohort (Figure 2A). CI subjects from the ADNI cohort showed increased GDCA and ursocholic acid (UCA) levels and decreased cholic acid (CA) levels compared with CU individuals (Figure 2B).

FIGURE 2.

The concentrations of DCA‐family and LCA‐family BAs differed across the AD subgroups. (A–D) Volcano plots depicting the concentrations of BAs in different groups: CI versus CU (A, B) and A+ versus A– (C, D). The x‐axis represents the z score normalized U values. The red line represents the threshold of P_FDR < 0.05, and BAs with significant differences are colored blue. (E) Upset plot displaying group‐specific changes in the levels of BAs and their isomeric forms, and ratios. Top bar graph: Quantification of the forms of BAs with between‐group differences within each bile acid family separately. The left five bars represent five common BA families, and the right bar contains less common BAs. The remaining bars represent the ratio of primary bile acids, the ratio of secondary bile acids and the ratio between secondary and primary bile acids. We categorized BA families based on the free form, conjugated form, isomeric forms, and the ratio between the forms of a given BA. Left bar graph: Summary of the number of subgroup‐specific differentially abundant BAs. Central matrix: Visualization of overlaps of differentially abundant BAs between subgroups. AD, Alzheimer's disease; BA, bile acid; CI, cognitively impaired; CU, cognitively unimpaired; DCA, deoxycholic acid; LCA, lithocholic acid.

When we classified these participants by their amyloid status based on PET scans (CPAS) and CSF Aβ42 levels (ADNI), we observed that glycochenodeoxycholic acid (GCDCA) levels, taurochenodeoxycholic acid (TCDCA) levels, lithocholic acid (LCA) levels, the LCA/chenodeoxycholic acid (CDCA) ratio, (ursodeoxycholic acid [UDCA] + LCA)/CDCA ratio and the LCA/UDCA ratio were associated with the amyloid status in participants in the CPAS. The levels of these BAs or their ratios were significantly reduced in the A+ subjects compared with the A– subjects, while the A+ subjects did not present significantly increased levels of any BAs (Figure 2C). We also identified that several bile acids were significantly associated with the Aβ deposition status in the ADNI cohort, but the differences were distinct. Compared with those in the amyloid‐negative (A–) group, the allolithocholic acid (alloLCA) levels, isolithocholic acid (isoLCA) levels, GLCA levels, DCA/(UDCA + LCA) ratio, GLCA/CDCA ratio, GLCA/UDCA ratio, and glycolithocholic acid (GLCA)/LCA ratio were increased in the amyloid‐positive (A+) group compared with those in the ADNI (Figure 2D). Here, we defined unconjugated bile acid and its conjugated forms, isomers, and ratios between them as a bile acid family. We identified two secondary bile acid families, namely, the DCA family, whose levels differed between CI individuals and CU individuals across both cohorts. In contrast, levels of the LCA family were found to be associated with amyloid deposition in these two cohorts (Figure 2E).

3.3. BAs were significantly associated with amyloid and tau pathology

We further explored the relationship between the levels of BAs and amyloid and tau pathologies, especially the levels of the LCA family and the DCA family. First, in the CPAS cohort, we observed that the plasma p‐tau 181 level was correlated with decreased sulfolithocholylglycine (LCA‐S) levels (β = –0.10, P_FDR  = 0.02), but increased CA levels (β = 0.16, P_FDR  = 0.04), as well as glycholic acid/taurocholic acid (GCA/TCA) ratio (β = 0.12, P_FDR  = 0.04). Plasma Aβ42 levels were significantly correlated with increased GCDCA levels (β = 0.10, P_FDR  = 0.04). In the ADNI cohort, we observed that CSF Aβ42 levels correlated with an increased (TUDCA+ [glycoursodeoxycholic acid] GUDCA)/UDCA ratio (β = –0.13, P_FDR  = 0.04). CSF p‐tau levels were associated with GDCA (β = 0.09, P_FDR  = 0.02) and GLCA (β = 0.07, P_FDR  = 0.04) levels (Figure 3A). Here, we identified that LCA‐ and DCA‐family BAs were significantly associated with amyloid and tau pathologies.

FIGURE 3.

LCA‐family BAs were associated with amyloid and tau pathologies. Heatmaps depicting the relationships between BAs levels and amyloid or tau pathology. Plasma Aβ42, Aβ42/40, p‐tau181, and amyloid PET centiloid were obtained from the CPAS cohort, and CSF Aβ42, CSF p‐tau, and amyloid PET centiloid levels were measured in the ADNI cohort. The colors of the dots/square indicate β coefficients (blue, negative; red, positive), and the symbol * indicates P_FDR  < 0.05. (A) BA–pathology associations at the population level, with the dot size reflecting ‐log₁₀(P_FDR ). (B–E) Subgroup‐specific (B, CU group; C, CI group; D, A‐ group; E, A+ group) correlations between the levels of the LCA‐family and DA‐family BAs and pathological markers. (F–M) Logistic regression models assessing the ability of the LCA‐family and DCA‐family BAs to predict amyloid/tau pathology (adjusted for sex, age, and APOE genotype). ADNI, Alzheimer's Disease Neuroimaging Initiative; APOE, apolipoprotein E; BA, bile acid; CI, cognitively impaired; CPAS, Chinese Preclinical Alzheimer's Study; CSF, cerebrospinal fluid; CU, cognitively unimpaired; DCA, deoxycholic acid; LCA, lithocholic acid; PET, positron emission tomography.

Next, when we stratified participants by cognitive status, the DCA levels (β = –0.10, P_FDR  = 0.001) (Figure 3B) and the GCA/TCA ratio (β = 0.11, P_FDR  = 0.03) (Figure S1A) were associated with plasma p‐tau181 levels in the CU group from the CPAS. Increased GDCA levels were positively correlated with CSF p‐tau levels (β = 0.11, P_FDR  = 0.004) in CU individuals from the ADNI (Figure 3B). In the CI group, the (TCDCA+GCDCA)/CDCA ratio was negatively associated with CSF Aβ42 levels (β = –0.11, P_FDR  = 0.01) (Figure S1B) and the GLCA levels (β = 0.08, P_FDR  = 0.04) and the GLCA/LCA ratio (β = 0.11, P_FDR  = 0.03) were positively associated with CSF p‐tau levels (Figure 3C). When patients were stratified by amyloid status, more results were obtained. GLCA/LCA levels were associated Centiloid levels (β = 0.03, P_FDR  = 0.04) and DCA levels (β = –0.08, P_FDR  = 0.02) and the GCA/TCA ratio (β = 0.11, P_FDR = 0.02) (Figure S1C) were associated with plasma p‐tau in the A– population from CPAS. In addition, DCA levels (β = 0.10, P_FDR  = 0.003) (Figure 3D), were also associated with CSF p‐tau levels in the A– population. Increased GLCA (β = 0.28, P_FDR  = 0.02) (Figure 3E) and GCDCA levels (β = 0.24, P_FDR  = 0.04) (Figure S1D) were positively associated with plasma Aβ42 levels, while increased GLCA/LCA levels were associated with plasma Aβ42 levels (β = 0.34, P_FDR  = 0.004) and CSF Aβ42 levels (β = 0.02, P_FDR  = 0.03)in the A+ population (Figure 3E).

3.4. Accuracy in the context of detecting Aβ or tau positivity

We compared the area under the curve (AUC) between the LCA family and the DCA family for classifying Aβ or tau positivity. When LCA or DCA metabolites were used as the sole predictors of plasma Aβ42 and p‐tau181 levels, both models displayed limited predictive performance, with AUC values of 0.581–0.584 for Aβ42 and 0.632 for p‐tau. However, the predictive capacity of DCA and LCA for Aβ PET positivity was similar (AUC = 0.68 vs. AUC = 0.68) (Figure 3F,J). The inclusion of the entire LCA family and DCA family in the CPAS cohort did not substantially improve the predictive performance (Figure 3G,K). In contrast, in the ADNI cohort, DCA levels alone exhibited good performance in the context of detecting CSF Aβ42 (AUC = 0.759), CSF p‐tau (AUC = 0.738), and Aβ PET (AUC = 0.784). The DCA family also showed significantly greater predictive power for CSF Aβ42 (AUC = 0.766), CSF p‐tau (AUC = 0.742), and Aβ PET (AUC = 0.788) (Figure 3H,I). In addition, the use of only the levels of LCA and the LCA family exhibited similar performance in the context of detecting amyloid and tau pathologies.

3.5. The enrichment analysis revealed gene that influence metabolic activity

The voxel‐wise analysis revealed spatial relationships between the levels of DCA and LCA family BAs and Aβ deposition. For P_FDR  < 0.05, a negative correlation between GLCA levels and Aβ deposits in the parietal lobe was observed only in the CPAS cohort (Figure 4D). No results were observed for the levels of the remaining bile acids in the CPAS cohort or in the ADNI cohort. Therefore, we report the results based on p < 0.001 below. In the CPAS cohort, no significant correlations were observed between Aβ deposition in any brain region and DCA levels (Figure 4A). The GDCA level was negatively correlated with Aβ deposition in the temporal and parietal lobes (Figure 4B), the LCA level was negatively correlated with Aβ deposition in the parietal lobe (Figure 4C), and the GLCA level was positively associated with Aβ deposition in the temporal cortex (Figure 4D). In the ADNI cohort, DCA and GDCA levels, as well as LCA and GLCA levels, displayed similar relationships and patterns. DCA and GDCA levels were positively associated with Aβ deposition in the temporal and occipital lobes (Figure 4E,F), and LCA and GLCA levels were positively associated with Aβ deposition in the parietal lobe (Figure 4G,H). CA, GCA/TCA, GCDCA, and LCA‐S levels, which exhibited significant associations in the previous assessment, were also analyzed for their relationships with amyloid deposition in the CPAS cohort. Plasma CA levels were associated with increased Aβ deposition in the bilateral occipital cortices (Figure S2A), whereas an increased GCA/TCA ratio correlated with Aβ deposition in parts of the parietal and temporal lobes (Figure S2B). Additionally, lower GCDCA levels were associated with Aβ deposition in the left temporal region (Figure S2C). The association between LCA‐S levels and amyloid deposition was limited (Figure S2D)

FIGURE 4.

Distinct BAs exhibit differential spatial associations with amyloid deposition, and gene set enrichment analysis reveals their multifaceted regulatory mechanisms in AD pathology. (A–H) Voxel‐wise analyses were performed to assess the spatial correlations of the four BAs with amyloid plaques in the CPAS cohort (A–D) and ADNI cohort (E–H) after adjustment for sex, age and APOE genotype. The color bar represents the T value with a statistical threshold of p < 0.001. (I–N) GO enrichment analysis of biological process, cellular component and molecular function terms. (I, J) The BP, CC, and MF terms for the negative genes. (K–N) Results of the KEGG pathway enrichment analysis of the negative (K, L) and positive (M, N) genes. All p values displayed were adjusted for the FDR. AD, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; APOE, apolipoprotein E; BA, bile acid; BP, biological process; CC, cellular component; GO, Gene Ontology; MF, molecular function.

We subsequently performed an imaging transcriptomics analysis to identify BA‐associated genes and their functional implications. Genes with negative β coefficients for the DCA and LCA families were enriched primarily in biological processes (BPs), including the transmission of nerve impulses, proton transmembrane transport, the assembly of mitochondrial respiratory chain complex І, and the NAD metabolic process. Similarly, these genes were localized to the synaptic membrane, transmembrane transport complex, and respiratory chain complex І and were functionally involved in electron transfer activity, proton transmembrane transport activity, and NADH dehydrogenase activity (Figure 4I,J). The KEGG pathway analysis revealed enrichment in oxidative phosphorylation, neuroactive ligand signaling and Alzheimer's disease (Figure 4K,L). Genes with positive β coefficients yielded few significant results in the Go enrichment analysis but were enriched in pathways such as neuroactive ligand signaling, the MAPK signaling pathway and calcium signaling (Figure 4M,N).

Furthermore, an analysis of genes associated with CA, GCA/TCA, GCDCA, and LCA‐S revealed that genes with negative β coefficients were enriched in the regulation of transmembrane transport, nervous system processes, and the Toll‐like receptor 2 signaling pathway. These genes exhibited spatial enrichment in the monoatomic ion channel complex, postsynaptic density membrane, gamma‐aminobutyric acid‐ergic (GABAergic) synapse, and synaptic membrane (Figure S2E). Conversely, genes showing positive β coefficients were involved in molecular functions (MFs) related to potassium channel activity and voltage‐gated channel activity. These proteins were predominantly localized to the synaptic membrane, the neuronal cell body membrane, and various channel complexes (Figure S2F). Notably, the pathway enrichment analysis revealed significant associations with Type І diabetes mellitus, muscle cell cytoskeletal organization (Figure S2G), oxytocin signaling, MAPK signaling, and dopaminergic synapse pathways (Figure S2H).

3.6. BAs can predict longitudinal amyloid and tau pathologies

Furthermore, we attempted to identify whether BAs levels could predict longitudinal amyloid and tau pathologies in the ADNI cohort. We included 1038 individuals who had 12‐month (M12) visits and 886 individuals who had 24‐month (M24) visits. Following the exclusion of subjects whose cognitive status or amyloid deposition status is inconsistent, we compared differences in BA levels before and after M12 or M24 visits within each cognitive subgroup and amyloid subgroup. No significant differences were observed between the CU and A+ subjects. Compared with the baseline values, CI subjects exhibited significantly elevated CA levels at M12 (Figure 5A,E). In addition, dehydrolithocholic acid (dehydroLCA), LCA, and UCA levels were significantly increased from baseline to M24 (Figure 5C,E). In A– subjects, norcholic acid (NorCA) levels, TDCA levels and the glycohyocholic acid/hyocholic acid (GHCA/HCA) ratio were significantly increased at M12 (Figure 5B,F); DCA levels were increased at M24 compared with the baseline value (Figure 5D,F).

FIGURE 5.

Longitudinal alterations in bile acids and their associations with AD pathology. (A–D) Volcano plots depicting BAs levels at different time points. (E, F) Changes in the concentration of various bile acids at three time points in the CI group (E) and A– group (F) are shown as different colored lines. (G, H) Heatmaps depicting the correlations of baseline and longitudinal changes in BAs levels with AD pathologies and the cognitive status. The x‐axis represents the z score normalized results for the W values. The colors of the dots indicate β coefficients (blue, negative; red, positive) and the sizes reflect ‐log₁₀(P_FDR) values. The symbol * indicates P _FDR < 0.05. AD, Alzheimer's disease; BA, bile acid; CI, cognitively impaired.

In the longitudinal analysis, we observed that the slopes of the levels of the CA family (GCA [β = 0.10, P_FDR  = 0.02], TCA [β = 0.44, P_FDR  = 0.007]), CDCA family (GCDCA [β = 0.04, P_FDR  = 0.02], TCDCA [β = 0.35, P_FDR  < 0.001]), and (GCA+GCDCA)/(CA+CDCA) (β = 2.29, P_FDR  = 0.02), (TCA+TCDCA)/(CA+CDCA) (β = 13.64, P_FDR  < 0.001), (GCA+GCDCA+TCA+TCDCA)/(CA+CDCA) (β = 2.12, P_FDR  = 0.007) ratios were associated with the slope of CSF Aβ42 levels. Additionally, the slopes of the levels of the secondary BAs TDCA (β = 0.41, P_FDR  = 0.003) and TUDCA (β = 8.63, P_FDR  = 0.007) were also linked to the slope of the CSF Aβ42 level. The slope of the CSF p‐tau level was correlated only with hyodeoxycholic acid (HDCA) levels (β = –0.08, P_FDR  = 0.01) (Figure 5G). Baseline levels of TCDCA (β = –0.015, P_FDR  = 0.01), TDCA (β = –0.02, P_FDR  = 0.04), and the (TCA+TCDCA)/(CA+CDCA) ratio (β = –0.54, P_FDR  = 0.05) were associated with the slope of the CSF Aβ42 level. Baseline levels of GUDCA/CDCA (β = –0.008, P_FDR  = 0.004) and (TDCA+GDCA)/(TCA+GCA) (β = –0.006, P_FDR  = 0.02) were associated with changes in the MMSE scores (Figure 5H).

4. DISCUSSION

In this study, we analyzed 2672 blood samples from 1397 participants in the CPAS cohort and 1275 participants in the ADNI cohort. Our results revealed that differences in BA concentrations between CI and CU patients were primarily attributable to variations in the secondary bile acid DCA family. Furthermore, concentrations of the LCA family differed significantly between amyloid‐negative and amyloid‐positive individuals. The longitudinal analysis revealed that the levels of DCA‐family BAs in A– subjects continuously changed from baseline levels, while the levels of LCA‐family BAs in CI subjects also varied. Although we only observed an association between the slopes of TDCA and CSF Aβ42 levels and did not directly observe significant associations between the levels of other DCA and LCA‐family BAs and Aβ or p‐tau pathology, the ratios between these secondary bile acids and their primary counterparts were associated with these biomarkers. These findings suggest interactions between enzymes involved in secondary bile acid metabolism and pathological proteins. Collectively, these findings underscore the importance of LCA and DCA in AD pathogenesis. The potential underlying mechanisms may involve transmembrane transport, signal transduction, immune regulation, ion channel/receptor activity, electron transfer, and energy metabolic processes.

DCA and LCA are secondary bile acids metabolized by the gut microbiota from CA and CDCA, respectively, and typically exist in glycine‐ or taurine‐conjugated forms. ³² Previous studies focusing on variations in the cognitive stage reported elevated levels of DCA, GDCA, LCA, and GLCA in AD patients and model mice, with higher concentrations correlated with more severe cognitive decline. ⁴ , ³³ , ³⁴ Moreover, plasma levels of LCA‐family BAs exhibit moderate specificity but limited sensitivity in distinguishing AD patients from controls. ³⁵ In our study, we observed increased DCA levels in the CI groups, which is consistent with previous findings.

Quantitative analyses have shown an average 3.2‐fold increase in the plasma levels of LCA and only a slight increase in GLCA levels in healthy individuals who converted to AD over an 8‐ to 9‐year follow‐up period. ³⁵ Furthermore, increased levels of specific LCA isomers were detected in centenarians. ³⁶ This evidence strongly suggests that LCA‐family bile acids play a role in ageing‐related conditions. ³⁷ , ³⁸ , ³⁹

We next systematically investigated the relationships of these bile acids with amyloid and tau pathologies by measuring CSF, plasma and PET imaging to elucidate their roles in AD pathogenesis in further detail. We observed that the glycine‐conjugated primary bile acid GCDCA was consistently associated with amyloid pathology, whereas DCA‐family BAs (GDCA) and LCA‐family BAs (GLCA and LCA‐S) were strongly associated with tau pathology. Previous studies showed that higher levels of GLCA and TLCA were significantly associated with elevated CSF p‐tau levels, suggesting a link between gut‐derived BA metabolism and tau‐related neurodegeneration. Additionally, higher levels of GLCA/CDCA were associated with lower CSF Aβ42 levels. ⁴⁰ , ⁴¹ Moreover, our previous findings demonstrated that conjugated BAs, particularly GDCA, were associated with advanced clinical stages and disease severity in patients with AD. ⁴²

Based on the above findings, we tested whether LCA‐family and DCA‐family BAs can be used as predictors to detect Aβ or tau positivity. The results suggested that they showed moderate performance in the context of detecting the levels of the Aβ42 and p‐tau markers in plasma, but high performance for detecting markers in CSF. This discrepancy may be caused by the different blood components used to acquire the metabolite data. Additionally, compared with direct measurement of plasma and CSF levels of markers, the predictive capacity for Aβ PET was stronger. The relationships between longitudinal AD pathologies and these bile acids have not been extensively explored. A large‐scale study reported that the levels of GDCA and 7‐ketolithocholic acid were associated with changes in the Aβ PET, whereas the level of GDCA was also associated with changes in the Tau PET rate. ⁴¹ In our study, we observed that the levels of conjugated bile acids and the ratio of conjugated to unconjugated primary bile acids were associated with the progression of CSF Aβ42 levels. This phenomenon may reflect functional alterations in enzymes or other substances that mediate the synthesis of conjugated bile acids, which occurs during the progression of amyloid pathology. The accumulated taurine‐conjugated bile acids could exert protective effects through anti‐inflammatory and antiapoptotic pathways. ⁴³ , ⁴⁴

Although accumulating evidence from animal and clinical studies supports the impact of the gut microbiota on central nervous system function via the gut–brain axis, ⁴⁵ the specific mechanisms by which BAs influence amyloid and tau pathology remain to be elucidated. Nevertheless, in addition to directly binding to Aβ40 to promote the aggregation of Aβ42, ⁴⁶ accumulating evidence suggests that the regulatory effects of BAs on AD pathogenesis are mediated by multiple mechanisms. ⁴⁷ , ⁴⁸ In support of the involvement of these diverse mechanisms, the results of the gene set enrichment analysis revealed functional associations between BA‐related pathways and mitochondrial dysfunction, energy metabolism, signal transduction, ion channel/receptor activity, immune regulation, and neurotransmitter or synaptic functions within the nervous system.

The initial evidence suggests receptor‐dependent mechanisms. BAs influence the central nervous system primarily via the nuclear receptor FXR and the membrane receptor TGR5. ⁴⁹ , ⁵⁰ , ⁵¹ Notably, LCA acts as the most potent activator of the TGR5 receptor to protect against cognitive deficits, ⁵² , ⁵³ but excessive FXR signaling induces neuronal apoptosis. ¹⁵ In parallel to signaling transduction, microbe‐derived BA metabolites (e.g., isoLCA and lithocholic acid 3‐sulfate) regulate systemic inflammation by modulating the Treg/Th17 balance and through interactions with the NF‐κB/MAPK pathway. ⁵⁴ , ⁵⁵ , ⁵⁶ Severe neuro inflammation and oxidative stress, which are largely caused by electrons escaping from the respiratory chain, can induce mitochondrial dysfunction. As research progresses, the role of mitochondrial dysfunction in AD is garnering increasing attention, and the mitochondrial cascade hypothesis offers a new perspective. ⁵⁷ Finally, BAs can also directly modulate neuronal excitability through interactions with ion channels, including epithelial sodium channels and calcium‐activated potassium channels, influencing membrane potential and synaptic plasticity. ⁵² , ⁵⁸

Given that ageing drives progressive shifts toward cytotoxic BA profiles (including LCA accumulation) and considering the dual roles of LCA in metabolic regulation via derivatives such as isoLCA, ³⁶ we observed that LCA‐family BAs serve as molecular bridges connecting ageing signatures with AD pathology, particularly through the modulation of the cytoskeleton in muscle cells. These findings are consistent with those of previous studies reporting that the effects of LCA on ageing‐related phenotypes mainly involve muscle regeneration ³⁸ and that clinical evidence of elevated levels of LCA‐family BAs is correlated with AD pathologies, suggesting that these BAs have potential as biomarkers reflecting accelerated brain aging.

Although our study employed two distinct cohorts, potential inconsistencies in the specific findings between these cohorts may arise because of inherent differences in genetic backgrounds and environmental exposures across ethnic groups. More importantly, the two cohorts included in this study exhibit distinct dietary habits. The Asian population in the CPAS cohort typically consumes a higher proportion of dietary fiber in their diet, with some individuals potentially maintaining a habit of drinking green tea. Conversely, the Western population in the ADNI cohort tends to consume more meat and coffee, which may lead to elevated levels of secondary bile acids, while long‐term consumption of green tea may also alter the composition of bile acids. ⁵⁹ , ⁶⁰ , ⁶¹ Bile acids are one of the important modulators of AD progression. However, the lifestyle of AD patients may also influence the bile acid pool. For example, a high‐fat diet can increase the concentration of DCA and 12‐keto LCA. ⁶² The ketogenic diet has not only been correlated with lower Aβ, ⁶³ , ⁶⁴ but also alters bile acid metabolism; elevated serum levels of TDCA and TUDCA may induce weight loss by reducing energy absorption through the inhibition of intestinal carbonic anhydrase 1 expression. ⁶⁵ , ⁶⁶ Additionally, older adults with dementia show significantly lower levels of 25‐hydroxy‐vitamin D, and vitamin D appears to modulate the bile acid metabolites by shaping gut microbiota composition. ⁶⁷ , ⁶⁸ Patients with AD frequently showed weight loss, and the percent of weight loss was significantly associated with elevated Aβ, declined memory, and worse cognitive function. ⁶⁹ , ⁷⁰ Whatever, weight loss caused by a low‐calorie diet, diabetes, or other factors may also induce an increased level of bile acids. ⁷¹ , ⁷² , ⁷³ To conclude, those factors that may disturb the level of bile acids should be controlled in future research.

The metabolic data obtained from the two cohorts were derived from different blood components, one from plasma and one from serum, which to some extent, affects the accuracy and consistency of the measurements of BA concentrations. The ADNI cohort underwent a strict overnight fast before blood sampling, whereas the CPAS cohort was only required to fast for 6 hours. The difference in bile acid measurements between these two cohorts was likely influenced by the fasting status, which could cause these inconsistencies.

In the Section 2, we clarified that different approaches were used to define A–/A+ subjects in the two cohorts based on the shared substances detected using PET and in CSF, as well as considerations for maximizing the sample size. The fact that these differing methods yielded similar results indicates that our findings are not coincidental. However, for consistency in the study design, a more appropriate approach for subsequent research would be to adopt unified criteria across both cohorts. Additionally, longitudinal analyses were conducted exclusively within the ADNI cohort. Thus, while important longitudinal observations are provided, these findings regarding temporal dynamics are derived from a single cohort, and their broader applicability should be considered alongside the aforementioned methodological constraints and await further validation.

In conclusion, we observed that the dysregulation of gut microbiota‐derived LCA‐ and DCA‐family BAs was intricately linked to AD pathology and cognitive decline. BAs are correlated with the baseline and longitudinal amyloid and tau burden, and contribute to disease progression through diverse mechanisms. We proposed that the LCA and DCA‐family BA profiles reflect shared biological processes underlying both brain ageing and AD pathogenesis, positioning them as promising targets for monitoring disease progression and developing therapeutic agents targeting the gut–brain axis.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Written informed consent was obtained from all participants.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Fu W, Chao X, Wang Y, et al. Bile acids are associated with baseline and longitudinal amyloid and tau pathology in patients with Alzheimer's disease. Alzheimer's Dement. 2026;22:e71307. 10.1002/alz.71307