This study was a preplanned secondary analysis of data derived from two prospective cohort studies conducted at The Second Hospital of Hebei Medical University. The original studies focused on perioperative neurocognitive disorders and related biomarkers. The comprehensive data collection, including neuropsychological assessments, DCE-MRI for BBB evaluation, and fecal sample collection for microbiota analysis, was an integral part of the parent study protocols, designed prospectively to investigate the multifactorial nature of cognitive decline in this patient population. The rationale for utilizing these cohorts for the current analysis was the comprehensive collection of cognitive, imaging, and biospecimen data relevant to the gut-brain axis. Ethical approval for this secondary analysis and the parent studies was obtained from the Research Ethics Committee of The Second Hospital of Hebei Medical University (approval no. 2023-R077). Both parent studies were registered on the Chinese Clinical Trial Registry (ChiCTR2200063774 and ChiCTR2300070477). All participants provided written informed consent prior to enrollment.

Participants were recruited between April 2023 and August 2023. Inclusion criteria were: (1) age between 60 and 80 years; (2) a minimum of 6 years of education; (3) American Society of Anesthesiologists (ASA) physical status grade II–III; and (4) a diagnosis of coronary heart disease (CHD), confirmed by coronary angiography. Exclusion criteria included: (1) history of stroke, brain tumor, brain surgery, Parkinson’s disease, or epilepsy; (2) severe auditory or visual impairments, or communication difficulties hindering neuropsychological testing; (3) pre-existing major psychiatric disorders (e.g., schizophrenia, major depressive disorder requiring hospitalization); (4) contraindications for magnetic resonance imaging (MRI) scanning (e.g., metallic implants, claustrophobia); (5) active lower gastrointestinal tract diseases (e.g., inflammatory bowel disease), or acute/chronic infectious diseases; (6) use of antibiotics for more than 3 consecutive days within one month prior to enrollment, or regular use of probiotics/prebiotics. Each participant underwent a comprehensive medical history evaluation, detailed neuropsychological assessment, fecal sample collection, and brain MRI.

Demographic characteristics (age, sex, body mass index [BMI], education level), lifestyle factors (smoking, alcohol consumption), history of COVID-19 infection, and comorbidities (hypertension, diabetes, hypercholesterolemia, constipation, anxiety, New York Heart Association [NYHA] class, Pittsburgh Sleep Quality Index [PSQI]) were collected by trained researchers using standardized questionnaires.

A comprehensive neuropsychological test battery was administered by trained psychometrists in a quiet, controlled environment. Global cognitive function was assessed using the Montreal Cognitive Assessment-Basic (MoCA-B). Specific cognitive domains evaluated included: memory (Auditory Verbal Learning Test-Hua Shan version [AVLT-H], long-delayed free recall); executive function (Trail Making Test A [TMT-A]); language (Verbal Fluency Test [VFT], animal and fruit categories); and visuoconstructive skills (Clock Drawing Test-30 [CDT-30]). Activities of Daily Living (ADL) were assessed to evaluate functional independence, and the Clinical Dementia Rating Scale (CDR) was used to stage dementia severity (though primarily to exclude dementia). All scales were validated Chinese versions.

Participants were classified into two groups: normal cognition (NC) and mild cognitive impairment (MCI). The diagnosis of MCI was based on the criteria proposed by Albert et al. (2011), (1) subjective cognitive complaint by the patient or an informant; (2) objective evidence of impairment in one or more cognitive domains, defined as a score ≥1.5 SD below age-and education-adjusted norms on any of the domain-specific tests (AVLT-H, TMT-A, VFT, CDT-30), with norms derived from published Chinese population data; (3) largely preserved independence in daily living activities; and (4) absence of dementia. While the MoCA-B was used to assess global function, a specific cut-off score was not used in isolation for diagnosis due to its use as a screening tool.

Brain imaging was conducted using a 3.0 Tesla GE SIGNA Architect MRI system (GE Healthcare). For anatomical localization, 3D T1-weighted BRAVO and 3D T2-weighted FLAIR CUBE sequences were acquired. The integrity of the BBB was evaluated through dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). The DCE-MRI acquisition parameters were as follows: axial plane, repetition time (TR) of 4.4 ms, echo time (TE) of 1.6 ms, slice thickness of 2.0 mm with a 0.6 mm interslice gap, field of view (FOV) measuring 26.9 cm × 24 cm, and a flip angle of 12°. A gadolinium-based contrast agent (Gadobutrol, 0.1 mmol/kg of body weight) was administered intravenously using a power injector at a flow rate of 4.5 mL/s, succeeded by a 20 mL saline flush. This injection commenced at the start of the third dynamic acquisition phase. In total, 40 dynamic scans were obtained over a period of approximately 7 min.

The acquired DCE-MRI datasets were subsequently processed on a GE Advantage Workstation (AW4.7) utilizing the GEN IQ advanced analysis module, which includes functionality for 3D motion correction. To quantify BBB permeability, the extended Tofts two-compartment pharmacokinetic model was employed, generating the volume transfer constant (Ktrans, units: min⁻¹). An individual arterial input function (AIF) was determined for each participant from the M2 segment of the middle cerebral artery, initially identified automatically by the software and then meticulously reviewed and confirmed by an experienced neuroradiologist for accurate placement and characteristic curve shape. T1 mapping was carried out to correct for T1 relaxation times. For this study’s focus, regions of interest (ROIs) were specifically delineated in the bilateral hippocampi. This manual delineation was performed on co-registered T2-FLAIR images by two skilled neuroradiologists, who were kept unaware of the participants’ cognitive diagnoses and group assignments; any discrepancies in delineation were resolved through mutual agreement. The mean Ktrans values derived from these bilateral hippocampal ROIs served as the primary measure for subsequent statistical analysis.

Fecal specimen collection was conducted within the hospital setting. Participants were provided with standardized collection kits and clear instructions. Samples were typically obtained in the morning prior to major study-related procedures or assessments, and subsequently as defined by the study protocol. Upon collection, specimens were immediately refrigerated at 4°C. They were then transported to the laboratory facilities on dry ice, generally within a 2 h timeframe, and subsequently archived at −80°C pending DNA extraction. Efforts were made to coordinate fecal sample collection to occur within 48 h of the cognitive and imaging assessments.

Total microbial genomic DNA was extracted from approximately 200 mg of fecal material using an Omega Soil DNA Kit according to the manufacturer’s instructions. DNA concentration and purity were assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States). The V3-V4 hypervariable regions of the 16S rRNA gene were amplified by PCR using specific primers (F: ACTCCTACGGGAGGCAGCA, R: GGACTACHVGGGTWTCTAAT) via PCR. PCR products were purified and pooled. Sequencing was performed on an Illumina NovaSeq 6,000 platform (Illumina, San Diego, CA, USA) by Personalbio Technology Co., Ltd. (Shanghai, China), generating 2×250 bp paired-end reads.

Raw sequencing reads were processed using QIIME2 (version 2022.2) (Bolyen et al., 2019). Briefly, reads were demultiplexed, followed by quality filtering, denoising, merging of paired-end reads, and chimera removal using the DADA2 plugin to generate amplicon sequence variants (ASVs). To ascertain the taxonomic composition, the representative sequences derived from the ASV generation step were subjected to a classification process within the QIIME 2 (version 2022.2) environment. This procedure involved querying these sequences against the Greengenes 16S rRNA gene reference database (release 13.8), enabling the assignment of taxonomic labels from phylum to genus levels based on sequence similarity.

Statistical analyses were performed using SPSS (version 27.0; IBM Corp, Armonk, NY, United States) and R software (version 4.3.1; R Core Team, Vienna, Austria). Data distribution was assessed using Shapiro–Wilk tests. Continuous variables were expressed as mean ± standard deviation (SD) for normally distributed data or median [interquartile range (IQR)] for non-normally distributed data. Categorical data were presented as numbers (proportions, %). Group comparisons for continuous variables were made using independent samples t-tests or Mann–Whitney U tests, as appropriate. Categorical data were compared using Chi-squared (χ²) tests or Fisher’s exact test. A p-value < 0.05 (two-sided) was considered statistically significant.

For gut microbiota analysis: Alpha diversity indices (Chao1 for richness, Shannon for richness and evenness, Pielou’s e for evenness) were calculated based on ASV counts and compared between groups using Mann–Whitney U tests. Beta diversity was assessed using Bray–Curtis dissimilarity and visualized with Principal Coordinates Analysis (PCoA). Permutational multivariate analysis of variance (PERMANOVA) with 999 permutations was used to test for significant differences in overall microbiota composition between groups. Linear discriminant analysis Effect Size (LEfSe) was employed to identify differentially abundant taxa (from phylum to genus level) between the NC and MCI groups, using a logarithmic LDA score threshold of >2.0 and a Kruskal-Wallis test p-value < 0.05. Correlations between microbial taxa (relative abundances of genera identified as significant by LEfSe), hippocampal Ktrans values, and MoCA-B scores were assessed using Spearman’s rank correlation coefficients.

Sample size was initially estimated based on expected differences in hippocampal Ktrans values between NC and MCI patients from pilot data or literature review. The pilot trial indicated a mean Ktrans of 6.0 × 10⁻³ min⁻¹ in MCI patients and 3.5 × 10⁻³ min⁻¹ in NC patients. With a significance level (α) of 0.05 and a power (1-β) of 0.80, the required sample size was calculated to be 20 participants per group.

The final analysis included 40 participants: 20 in the NC group and 20 in the MCI group. Demographic and clinical characteristics are presented in Table 1. There were no significant differences between the NC and MCI groups in terms of age, sex distribution, BMI, smoking or drinking habits, education level, history of COVID-19 infection, or prevalence of hypertension, diabetes, hypercholesterolemia, NYHA class, constipation, or anxiety symptoms (PSQI scores). Additionally, key cardiometabolic markers and medication use were well-balanced between the groups (Table 1). This similarity in baseline characteristics helps to minimize their confounding effects on the primary outcomes.

As expected, patients in the MCI group exhibited significantly lower scores on the MoCA-B compared to the NC group (19.05 ± 1.19 vs. 25.25 ± 1.65, p < 0.001). Significant differences were also observed in AVLT-H long-delayed recall, TMT-A completion time, and CDT-30 scores, indicating poorer performance in memory, executive function, and visuoconstructive skills in the MCI group (Table 2). There was a trend towards lower VFT scores in the MCI group, though it did not reach statistical significance (p = 0.059).

Hippocampal BBB permeability, assessed by Ktrans values, was significantly higher in the MCI group (median 6.04 ± 3.02 × 10⁻³ min⁻¹) compared to the NC group (median 3.90 ± 1.03 × 10⁻³ min⁻¹; p = 0.006).

Rarefaction curve analysis indicated sufficient sequencing depth to capture the majority of bacterial diversity in all samples (Supplementary Figure 1). The Venn diagram of shared and unique ASVs revealed a lower number of observed ASVs in the MCI group compared to the NC group (Figure 1: 3927 unique/shared ASVs in MCI vs. 6,128 in NC).

Alpha diversity analysis showed significantly lower microbial richness and diversity in the MCI group compared to the NC group. Specifically, the Chao1 index (343.9 ± 162.2 vs. 524.5 ± 199.3; p = 0.002), Shannon index (4.94 ± 0.84 vs. 5.64 ± 1.06; p = 0.009), and Pielou’s evenness index (0.60 ± 0.07 vs. 0.63 ± 0.10; p = 0.030) were all significantly reduced in MCI patients (Figure 2).

Beta diversity analysis using PCoA based on Bray–Curtis dissimilarity revealed a distinct clustering of gut microbiota composition between the NC and MCI groups (Figure 3). PERMANOVA confirmed this significant difference in overall microbial structure (p = 0.003).

At the phylum level, Firmicutes and Bacteroidetes were the most abundant phyla across all samples, accounting for over 65% of the total sequences. We compared the Firmicutes-to-Bacteroidetes (F/B) ratio between the groups but found no statistically significant difference (p = 0.609), suggesting that broad phylum-level shifts were not a primary feature of dysbiosis in this cohort. The LEfSe analysis also did not identify any phyla as being differentially abundant between the groups.

LEfSe analysis identified several taxa that were differentially abundant between the NC and MCI groups (Figure 4). At the family level, the relative abundances of Ruminococcaceae (LDA score = 4.661, p = 0.016), Rikenellaceae (LDA score = 3.752, p = 0.042), and Barnesiellaceae (LDA score = 3.698, p = 0.038) were significantly higher in the NC group compared to the MCI group.

At the genus level, the NC group exhibited significantly higher relative abundances of Faecalibacterium (LDA score = 4.456, p = 0.003) and Oscillospira (LDA score = 4.096, p = 0.002) compared to the MCI group. These genera are well-known for their SCFA-producing capabilities, particularly butyrate.

Spearman correlation analysis revealed significant associations between the relative abundances of the depleted genera in the MCI group, hippocampal Ktrans values, and MoCA-B scores across all participants (n = 40). The relative abundance of Faecalibacterium was negatively correlated with hippocampal Ktrans (r = −0.466, p = 0.002) and positively correlated with MoCA-B scores (r = 0.596, p < 0.001). Similarly, the relative abundance of Oscillospira was negatively correlated with hippocampal Ktrans (r = −0.322, p = 0.043) and positively correlated with MoCA-B scores (r = 0.369, p = 0.019).

To investigate potential confounding by cardiometabolic factors, a new correlation analysis was performed between key cardiometabolic markers (SBP, DBP, Fasting Blood Glucose, LDL-C) and the relative abundances of Faecalibacterium and Oscillospira. No significant correlations were found (all p > 0.10), suggesting that the observed associations between these genera and cognitive/BBB measures are independent of these specific metabolic parameters within our cohort.