The trial was conducted in a double-blind, randomized, placebo-controlled manner. HIV-infected patients were recruited through the Comparative HIV and Aging Research in Taizhou (CHART). The detail of the prospective cohort has been described previously (21). In brief, we screened 126 HIV-infected adults. The inclusion criteria for this trial require male and female patients diagnosed with HIV-1 infection and were under stable cART (6 months prior and within the study period) and diagnosed with NCI. The presence of NCI was diagnosed according to the global diagnosis score (GDS) from a set of neuropsychological tests (NP tests) (22). Exclusion criteria included patients taking drugs with psychic effects, patients with a value of BMI >30 kg/m², women during lactation or pregnancy, patients with diabetes mellitus history, cardiovascular and cerebrovascular diseases, or any other diseases in the liver, kidney, and hematopoietic system.

Finally, 88 eligible participants were randomized in 1:1 to receive daily algal oil DHA supplement (Supplementary Table 1) or a matching placebo of soy oil capsules. Each capsule provided 450 mg DHA. Doses were 3.15 g/day. The treatment duration was 24 weeks. Randomization was done according to a computer-generated random schedule. Both investigators and participants were blind to the intervention condition while assessing outcome measures before the completion of data analysis. All participants received instructions on maintaining their current diets and physical activities throughout the intervention period. All participants were contacted at weeks 4, 8, 12, 16, and 20 to evaluate any problems that might occur due to the drugs studied. After 24 weeks of intervention, 68 participants (77.3%) completed the whole trial.

Study visits were undertaken on day 1 (baseline) and week 24 (final follow-up). At each visit, blood and fecal samples were collected. All participants were administered with a questionnaire interview to collect dietary intake, basic demographics, and lifestyle. We used a valid food frequency questionnaire (FFQ) for dietary assessment (23). Current smoking and alcohol use were defined as having smoked and drink in the last 30 days (22). The main outcome of this study was mental status which was assessed by two questionnaires. Face-to-face neurocognitive testing was conducted on all participants by trained health staff according to both the MMSE in mandarin (1) and the NP tests. Additionally, they received comprehensive physical and biochemical examinations and B-mode ultrasound examinations. Carotid intima-media thickness (CIMT) was imaged using a high-resolution B-mode ultrasound machine with a 10 MHz multi-frequency linear transducer (LOGIQ P5 pro, General Electric Medical Company, WI, USA). A standardized protocol was used for images and procedures. This study was approved by the Institutional Review Board (IRB) of the School of Public Health at Fudan University (no.IRB#2019-06-0759), Shanghai, China. All individuals gave informed consent at enrollment. The registration of this study can be found in ClinicalTrials.gov (no. NCT04242004).

Participants were asked to record acceptability and adverse events. The research Staff inquired the participants about whether they consumed the assigned capsules and how many capsules they consumed each day during routine follow-up communications (once a month). Compliance with the supplementation was monitored by counting the capsules returned by the participants. The adherence to the intervention was defined as a percentage; (number of capsules supplied a minus number of capsules not taken)/number of capsules supplied × 100. Participants consumed a mean of 94.6% of the packets (mean of 94.6% in the DHA group and mean of 94.7% in the placebo group).

Multiple cognitive-motor ability domains that were contently found in HIV-associated brain disease in patients from the United States were tapped in the test battery. The battery was carefully reviewed and approved to be culturally appropriate for the study populations in China by mental health professionals (22). Ten NP tests were used to evaluate seven domains: semantic verbal fluency (animals); Hopkins Verbal Learning Test; Brief Visuospatial Memory Test; Stroop Color and Word Test; Trail Making Test (parts A and B); Wisconsin Card Sorting Test 64 Card Version (WCST-64); Grooved Pegboard Test; Paced Auditory Serial Addition Test (PASAT); the Digit Symbol, Symbol Search, and Letter-Number Sequencing tests from the Wechsler Adult Intelligence Scale (WAIS-III).

Scaled scores were converted from raw test scores from these tests. For each test, standardized T scores were generated with the mean of 50 (SD, 10) and adjusted for age, sex, and education. Based on T scores, deficit scores (DSs) ranging from 0 (no deficit) to 5 (severe deficit) were created: T score ≥ 40 = DS score of 0; T score 35–39 = DS score of 1; T score 30–34 = DS score of 2; T score 25–29 = DS score of 3; T score 20–24 = DS of 4; T score <20 = DS score of 5. To generate global T and GDS, domain T and DS took the average scores of individual tests within each domain and across all tests, respectively. The global cognitive score was defined by the GDS dichotomized as impaired (GDS ≥ 0.5) or unimpaired (GDS < 0.5).

Serum samples were taken from the supernatant of blood after centrifugation and were immediately stored at −80°C for further analysis. As described previously, the clinical biomarkers, including glucose (Glu) and lipid profiles [total cholesterol (TC), TG, low-density lipoprotein (LDL), and high-density lipoprotein (HDL)], were measured (24). Concentrations of plasma DHA, proinflammatory factors, including tumor necrosis factor-α (TNF-α), interleukin 6 (IL-6), soluble CD14 (sCD14), and high-sensitive C-reactive protein (hs-CRP) and oxidative stress factor, including 8-F2c-isoprostane and malondialdehyde (MDA) were measured using ELISAs kits (Shanghai Enzyme-linked Biotechnology Co., Ltd., Shanghai, China) according to the instructions of the manufacturer (Supplementary Table 2). The precision of the assay was confirmed for a low coefficient of variations, <10% for intra-assay and inter-assay.

The fecal samples were collected in a feces container and stored at −80°C immediately upon collection. Total genomic DNA was isolated using DNA Extraction Kit (Qiagen, Düsseldorf, Germany) as per instructions from the manufacturer. Details of the quantitative measurement of fecal metabolomics using 16S rRNA sequencing as well as relevant data analysis can be found in Supplementary Materials.

The sample size was calculated on the basis of the MMSE. With a power of 80% and a type I error of 5%, a sample size of 36 per group would be required to detect an effect size of the change in 0.5 SD between groups (1). Allowing for a 10% loss during over 6 months of intervention, a sample size of 40 per group was considered adequate. In the end, 34 participants per group completed the whole trial.

Comparisons of clinical parameters, demographic factors, and dietary intake between the two groups at baseline were included in the descriptive statistics analysis. According to the distribution of data, median (interquartile range [IQR]) or mean (SD) were used to describe continuous variables. Categorical variables were summarized as numbers and percentages (%). Differences for 24 weeks changes from baseline of clinical parameters, proinflammatory and oxidative stress markers were analyzed using Student's t-test or Mann–Whitney U-test according to the distribution of data.

To explore the role of gut microbiota and the underlying metabolic mechanisms, we compare the gut microbiota and profiles of fecal metabolomics before and after DHA intervention. The α-diversity indices, including Chao, Shannon, and Simpson, were calculated. Beta diversity was measured by unweighted UniFrac distance. Differences in characteristics and diversity indices between groups were analyzed using the Mann–Whitney U-test (continuous variables and skewed distribution), the Student's t-test (continuous variables and normal distribution), or the chi-square test (categorical variables). Wilcoxon rank-sum test was employed to identify the differential bacterial genera abundance between the two groups. Correlations between changes in clinical parameters, proinflammatory and oxidative stress markers, and alterations in genus relative abundance, were evaluated with Spearman's rank.

To investigate the overall microbial metabolites distribution between the two groups, orthogonal projections to latent structure-discriminant analysis (OPLS-DA) were performed. The fit and predictability of the models obtained were determined by the R² Y and Q² values, respectively. In the OPLS-DA model, we generated the variable importance plot (VIP) to select potential biomarkers. In biochemical pathways of predicted Kyoto Encyclopedia of Genes and Genomes (KEGG), a Wilcoxon matched-pairs test was adopted. To analyze the association between fecal metabolites and genus relative abundance, as well as clinical parameters, proinflammatory cytokines, and oxidative stress markers were analyzed using Spearman's rank test. A p < 0.05 was defined as statistically significant. To control the false discovery rate (FDR), the values of p were adjusted. All statistical analyses were conducted using R version 3.4.

We selected 88 subjects and grouped them in a randomized manner and 68 (77.3%) completed the evaluation for this 24-week study (Figure 1). Patient characteristics at baseline are summarized in Table 1. We found no significant differences in sex, age, and BMI between the two groups of participants. DHA group had a lower LDL level compared with the placebo group, while other clinical parameters and the intakes of dietary components had no difference between the two groups from the beginning of this study.

Anthropometric data, GDS, MMSE score, and clinical parameters are summarized in Table 2. There were no significant changes in NCI according to GDS and MMSE score within the two groups after the 6-month intervention. After the 24 weeks intervention, levels of TG and WBC in the DHA group were significantly reduced than those observed in the placebo group. Moreover, the level of LDL was significantly increased compared with that observed in the soy group. There were no significant changes in other clinical parameters between the two groups. Additionally, compared with the placebo group, levels of TNF-α and hs-CRP were significantly reduced in the DHA group (Table 2). There were no significant changes in dietary intake (Supplementary Table 3).

Similarly, in the DHA supplement group, the change in plasma DHA was negatively correlated with the level of TG (r = −0.346, p = 0.005), CRP (r = −0.696, p < 0.001), MDA (r = −0.765, p < 0.001), C8-sio-PGF2i (r = −0.362, p = 0.042), sCD14 (r = −0.885, p < 0.001), CR (r = −0.295, p = 0.021), and WBC (r = −0.288, p = 0.024). The increased level of DHA was positively correlated with increased level of LDL (r = 0.449, p < 0.001) (Figure 2).

Based on the Shannon, Simpson, and Chao indices, the community diversity was significantly increased in the DHA supplement group compared with the baseline group [Chao index (p = 0.002), Shannon index (p = 0.037), and Simpson index (p = 0.015), Figures 3A–C]. The β-diversity of gut microbiota was evaluated based on the unweighted (p = 0.001) UniFrac distance matrix of the differences between the two groups in the fecal microbial communities (Figure 3D).

One hundred sixty-two genera showed significant differences between the two groups. The top 10 most abundant genera are presented in Supplementary Table 4. We conducted the following correlation analyses using the top 10 most abundant difference genera. The DHA supplement led to a significant increase in the relative abundance of the bacterial genus including Blautia, Bifidobacterium, Dorea, Lactobacillus, Faecalibacterium, Fusobacterium, and Agathobacter after the intervention, but led to a reduction in Bacteroides and Prevotella (Figure 4 and Supplementary Table 4).

The relative abundance of Blautia was negatively correlated with the level of TG (r = −0.494, p = 0.004), TC (r = −0.423, p = 0.016), sCD14 (r = −0.358, p = 0.044), and C8-sio-PGF2i (r = −0.379, p = 0.032), whereas the change in Bacteroides was positively correlated with the changes in TG (r = 0.701, p = 0.002), Glu (r = 0.505, p = 0.039) and negatively correlated with the changes in LDL (r = −0.407, p = 0.021). The change in relative abundance of Faecalibacterium was negatively associated with the changes in IL-6 (r = −0.386, p = 0.029), CR (r = −0.402, p = 0.034), and Glu (r = −0.485, p = 0.048) (Figure 5). Additionally, the relative abundance of Agathobacter (r = −0.488, p = 0.047) and Dorea (r = −0.517, p = 0.034) were negatively associated with the level of Glu (data not shown).

We observed that the composition of fecal metabolites shifted significantly in response to the DHA supplement (p = 0.005) (Supplementary Figure 2). Multivariate and univariate statistical significance criteria (VIP > 1 and FDR < 0.05) were used to filter the metabolites that are different between the groups. In total, we identified 15 fecal metabolites that have been significantly changed. These metabolites include eight lipids, four amino acid metabolites, two nucleosides, and one organic oxygen compound after FDR correction (Supplementary Table 5). Among them, 12 were decreased [Cer(d18:0/14:0), Cer(d18:0/16:0), glycocholic acid, glycodeoxycholic acid, 5,8-tetradecadienoic acid, 2-nonenal, deoxyinosine, inosine, L-gamma-glutamyl-L-valine, L-gamma-glutamyl-L-isoleucine, N-docosahexaenoyl GABA, and DL-histidinol] and three were increased [PI (20:0/22:4(7Z,10Z,13Z,16Z), PS (17:2(9Z,12Z)/22:1(11Z), and pregnenolone] (Figure 6A).

The significantly perturbed pathways are shown in Figure 6B. After FDR correction, there were significant differences in both the cholesterol metabolism (Rich factor = 0.200, p = 0.003) and purine metabolism (Rich factor = 0.042, p = 0.003) pathways. The changes in fecal concentrations of inosine and deoxyinosine were found to be positively correlated with the level of CD4 count (r = 0.35, p < 0.001) and WBC count (r = 0.28, p = 0.002), while no other significant associations were observed (data not shown). Multiple correlations were found between fecal metabolites and specific gut bacteria (Figure 7). The changes in metabolites that decreased in the DHA supplement group were consistently negatively correlated with the beneficial bacterial genera increased in the DHA group, such as Blautia, Faecalibacterium, and Lactobacillus.