Gut microbiome in people living with HIV is associated with impaired thiamine and folate syntheses

Abstract

People living with HIV have a high incidence of cardiovascular and neurological diseases as comorbid disorders that are commonly linked to inflammation. While microbial translocation can augment inflammation during HIV infection, functional microbiome shifts that may increase pro-inflammatory responses have not been fully characterized. In addition, defining HIV-induced microbiome changes has been complicated by high variability among individuals. Here we conducted functional annotation of previously-published 16S ribosomal RNA gene sequences of 305 HIV positive and 249 negative individuals, with adjustment for geographic region, sex, sexual behavior, and age. Metagenome profiles were inferred from these individuals’16S data. HIV infection was associated with impaired microbial vitamin B synthesis; around half of the gene families in thiamine and folate biosynthesis pathways were significantly less abundant in the HIV positive group than the negative control. These results are consistent with the high prevalence of thiamine and folate deficiencies in HIV infections. These HIV-induced microbiota shifts have the potential to influence cardiovascular and neurocognitive diseases, given the documented associations between B-vitamin deficiencies, inflammation, and these diseases. We also observed that most essential amino acid biosynthesis pathways were downregulated in the microbiome of HIV-infected individuals. Microbial vitamin B and amino acid synthesis pathways were not significantly recovered by antiretroviral treatment when we compared 262 ART positive and 184 ART negative individuals. Our meta-analysis provides a new outlook for understanding vitamin B and amino acid deficiencies in HIV patients, suggesting that interventions for reversing HIV-induced microbiome shifts may aid in lessening the burdens of HIV comorbidities.

Introduction

Microbiome profiling may improve our understanding of comorbid disorders in people living with HIV due to the unique interplay between HIV infection, the gut microbiome, and inflammation [1–10]. HIV infection is associated with increased inflammatory responses that are closely linked to comorbid cardiovascular diseases and neurocognitive disorders [11–13]. While microbial translocation is reported to promote inflammatory responses in HIV infections [2], which functional shifts in the bacterial community may increase inflammation has not been fully addressed. Our goal is to investigate HIV-related microbiome shifts which may further contribute to inflammation and subsequent comorbid diseases.

Considerable efforts have been directed towards characterization of gut microbiome changes in HIV infection [1–10]. However, the identification of an HIV-infection-specific microbiome can be confounded by multiple host factors, resulting in heterogeneous microbiome compositions between HIV infected individuals [1–9]. Antiretroviral therapy (ART) generally shifts the HIV-infection-specific microbiome towards an HIV-negative phenotype; however, the degree of this restoration is unclear [1, 4, 5, 8, 9]. In addition, sexual behavior was reported to drive major changes in microbiota composition that may obscure more subtle HIV-induced microbiome changes [1, 4, 14]. Therefore, it is of importance to define HIV-infection-specific signatures by adjusting for confounding factors, as these signatures have the potential to guide the development of therapeutics for HIV-induced gut dysbiosis.

Metagenome data provides a comprehensive functional catalog of gut microbiota and thus functional metagenomics has shown great promise in the annotation of disease-specific microbial pathways [15, 16]. Predicted metagenomes have been widely used to investigate disease-associated microbial functional markers, including inflammatory bowel disease [17], obesity [18] and cardiovascular disease [19]. Accordingly, a recent study has reported an increase in pro-inflammatory pathways within the HIV-infected individuals’ microbiome [20]. In the present study, we conduct functional annotation on over 800 individuals’ 16S ribosomal RNA (rRNA) gene sequences to identify functional microbiota signatures of HIV infection. Metagenome profiles predicted by the 16S data are compared between HIV negative and positive groups after adjusting for host factors. We demonstrate that microbial vitamin B synthesis and amino acids synthesis are significantly downregulated by HIV infection. We then hypothesize how these functional shifts are potentially associated with inflammation and HIV comorbidities. We also investigate the degree of ART restoration on HIV-induced functional microbiome shifts.

RESULTS

Microbiome specimen characteristics

Figure 1A showed the geographic distribution of the 821 specimens analyzed in the current meta-analysis, which were collected from eight cities of three continents: Africa, America, and Europe [1–9]. These specimens were obtained from Abuja (Nigeria) (130, 15.8%), Barcelona (Spain) (156, 19.0%), Denver (US) (247, 30.1%), Isingiro (Uganda) (9, 1.1%), Kiruhura (Uganda) (4, 0.5%), Logrono (Spain) (53, 6.5%), Madrid (Spain) (33, 4.0%), Mbarara (Uganda) (43, 5.2%), and Stockholm (Sweden) (146, 17.8%) [1–9]. This global population consisted of 249 HIV negative, 310 HIV positive ART negative (not receiving ART at the time of specimen collection), and 262 HIV positive ART positive specimens, as listed in Tables S1–S3. More than half of the specimens were collected from MSM (52%). Females were 32% and non-MSM males were 16%, as shown in Figure. 1B. This meta-analysis cohort’s age distribution was relatively homogeneous within the range of 16 to 78 (Figure. 1B). Within the subset of individuals whose BMI was available, 12% were obese, 36% were overweight, 49% were normal, and 2% were underweight (Figure. 1B). The ethnicities of the cohort included Asian (0.4%), Black (25%), Hispanic (9%), Non-Hispanic (26%), and White (39%), where information was available (Figure. 1B).

Figure 1.

Meta-analysis cohort characteristics. (A) Geographic distribution of 249 HIV negative, 310 HIV positive ART negative and 262 HIV positive ART positive specimens listed in Tables S1–S3. Pie charts show the proportion of HIV negative, HIV positive-ART negative, and HIV positive-ART positive groups in America, Europe and Africa and the diameter denotes the proportional representation of each continent to the total specimen. (B) The 821 specimens’ host factor distributions. MSM designation and sex are combined into one variable, sex:sexual behavior, with three categories: female, male and non-MSM (M:nonMSM), and MSM.

Multivariable linear regression

To identify microbial functional signatures associated with HIV infection, we compared gene family profiles of 249 HIV negative individuals with those of 305 region-matched HIV positive ART negative individuals. Geographic location, sex, age and MSM designation were available from all the specimens examined. Univariable linear regression was first used to select host factors that were associated with each gene family. Our multivariable model then estimated the effect of HIV on each gene family level by adjusting for the selected host factors. A total of 7,847 gene families were assessed using 249 HIV negative and 305 HIV positive (ART negative) individuals; 3,089 gene families showed a significant difference (adjusted p value (P_adj) < 0.05) between the HIV negative and positive groups after controlling for the selected host factors that were associated with each gene family. Downstream functional pathway enrichment analysis then revealed 33 significant differentially expressed pathways (adjusted pathway p < 0.05 and > 40% coverage).

Downregulation of microbial vitamin B synthesis by HIV infection

We found that HIV infection was associated with the downregulation of B-vitamin metabolism pathways: Thiamine (vitamin B1) metabolism (map00730), One carbon pool by folate (vitamin B9) (map00670), and Pantothenate (vitamin B5) and CoA biosynthesis (map00770), as presented in Table 1.

Table 1.

Significant differential pathways between 249 HIV negative and 305 HIV positive individuals. Significantly differentially expressed pathways related to B-vitamin and amino acid synthesis with coverage (proportion of differential gene families in each pathway), adjusted pathway p value (Fisher’s exact test and Benjamini-Hochberg adjustment), and number of differentially up or down regulated gene families.

Pathway	Coverage	Adjusted pathway p	Differential gene families
B-Vitamin Synthesis
Thiamine metabolism (map00730)	50%	0.008	1 ↑ 17 ↓
One carbon pool by folate (map00670)	47%	0.02	1 ↑ 16 ↓
Pantothenate and CoA biosynthesis (map00770)	55%	0.001	2 ↑ 19 ↓
Amino Acid Synthesis
Biosynthesis of amino acids (map01230)	56%	<0.001	12 ↑ 117 ↓
Glycine, serine and threonine metabolism (map00260)	44%	<0.001	13 ↑ 31 ↓
Phenylalanine, tyrosine and tryptophan biosynthesis (map00400)	46%	0.001	4 ↑ 29 ↓
Valine, leucine and isoleucine biosynthesis (map00290)	63%	0.004	1 ↑ 11 ↓
Arginine and proline metabolism (map00330)	40%	0.004	15 ↑ 27 ↓
Alanine, aspartate and glutamate metabolism (map00250)	43%	0.006	4 ↑ 26 ↓
Histidine metabolism (map00340)	44%	0.027	5 ↑ 15 ↓
Lysine biosynthesis (map00300)	42%	0.045	3 ↑ 16 ↓

Thiamine metabolism (map00730) is a thiamine biosynthesis pathway which covers the production of thiamine and its conversion into its active form, thiamine pyrophosphate [21]. Bacterial-derived thiamine is absorbed by thiamine transporters in the gut and distributed throughout the body, while thiamine pyrophosphate is directly absorbed by the colon [22]. We observed a significant reduction in Hydroxymethylpyrimidine kinase (K14153) abundance associated with HIV infection, which is one of the essential enzymes for thiamine biosynthesis [23]. Figure 2A compared the level of Hydroxymethylpyrimidine kinase (K14153) of 249 HIV negative and 305 HIV positive subjects in eight different cities. In all regions except Abuja (Nigeria), K14153 levels were lower in the HIV positive group compared to the HIV negative group, and K14153 levels significantly associated with geographic region (p < 0.001). As presented in Figure 2B, sex:sexual behavior (female, male:nonMSM, and MSM) was associated with K14153 level (p < 0.001) and thus this variable along with geographic region were included in the multivariable model. Age was not associated with K14153 level as shown in Figure. 2C (p=0.99). After adjusting for geographic region and sex:sexual behavior, there was a significant difference in the level of K14153 between the HIV positive and negative groups (effect size = −1.36 ± 0.24 and P_adj < 0.001).

Figure 2.

Hydroxymethylpyrimidine kinase (K14153) abundance reduction by HIV infection. (A) Hydroxymethylpyrimidine kinase (K14153) level of 249 HIV negative (light blue) and 305 HIV positive (dark blue) individuals in different geographic regions. (B) HIV negative (light blue) and HIV positive (dark blue) groups’ K14153 level of 184 female (F), 88 non-MSM male (M:nonMSM), and 282 MSM. (C) Scatter plot of age and K14153 level of 249 HIV negative (light blue) and 305 HIV positive (dark blue) individuals. (D) K14153 level of 36 Hispanic (his), 129 non-Hispanic (nonhis), 171 White, and 142 Black subjects. Each racial/ethnic group is stratified by HIV negative (light blue) and positive (dark blue). (E) Scatter plot of BMI and K14153 level among 168 HIV negative (light blue) and 153 HIV positive (dark blue) individuals.

While racial/ethnic information was not available for all the individuals examined above, K14153 levels were significantly different across racial/ethnic groups in the subset of 224 HIV negative and 254 HIV positive individuals whose race/ethnicities were reported (Figure 2D, p<0.001). Within the subset of 168 HIV negative and 153 positive subjects whose BMIs were reported, BMI was not related to K14153 level (r = 0.029 and p = 0.61), as shown in Figure 2E.

In addition to Hydroxymethylpyrimidine kinase (K14153), 16 more gene families in thiamine metabolism (map00730) were downregulated by HIV infection while only one gene family was upregulated. Figure 3A shows the effect size of differential gene families between the HIV negative and positive groups adjusted for host factors.

Figure 3.

Significant differential gene families of microbial thiamine and folate synthesis pathways. (A) HIV infection effect size (β_HIV) and standard error of 18 significant differential gene families within the Thiamine metabolism (map00730) pathway. (B) HIV infection effect size of 17 significant differential gene families within the One carbon pool by folate (map00670) pathway.

HIV infection significantly downregulated One carbon pool by folate (map00670) where 16 out of 36 gene families were downregulated (Figure. 3B). This pathway is a tetrahydrofolate biosynthesis pathway within the larger Folate Biosynthesis (map00790). Folate is not metabolically active until converted to tetrahydrofolate by dihydrofolate reductase [24]. Dihydrofolate reductase (K00287) was one of the genes significantly downregulated in HIV infection with the effect size of −0.32 (± 0.09) (P_adj = 0.004), as shown in Figure 3B. We also observed a decrease in the abundance of Formyltetrahydrofolate dehydrogenase (K00289), another enzyme that catalyzes tetrahydrofolate production [25], in HIV infected persons, with the effect size of −0.65 (± 0.25) (P_adj = 0.031). Bacterial-derived tetrahydrofolate is directly absorbed by a proton-coupled folate transporter in the colon and distributed to the body [22].

Our analysis revealed that 19 out of the total 38 gene families in the Pantothenate and CoA biosynthesis (map00770) pathway were downregulated by HIV infection (Table 1). Pantothenate and CoA biosynthesis is a pantothenate biosynthesis pathway which covers the production of pantothenate and metabolism of pantothenate to CoA. Pantothenate is a key precursor to CoA, which is an essential cofactor for cell growth, phospholipid synthesis, fatty acid synthesis and degradation, and the function of the TCA cycle [26]. Both dietary vitamin B5 and bacterial pantothenate are absorbed in the large intestine, converted to CoA, and distributed to the body [22].

Signatures of impaired microbial amino acid synthesis in HIV infections

Our meta-analysis on the 249 HIV negative and 305 HIV positive individuals in eight different cities showed that HIV infection was related to the downregulation of amino acid synthesis pathways. Arogenate (pretyrosine) dehydrogenase (K15226) within Biosynthesis of amino acids (map01230) was downregulated in seven different cities, yielding an HIV infection effect size of −0.92 (±0.38) with P_adj=0.04 (Supplementary Figure 1S). Arogenate (pretyrosine) dehydrogenase (K15226) catalyzes the synthesis of tyrosine [27].

In addition, Glycine, serine and threonine metabolism (map00260), Phenylalanine, tyrosine and tryptophan biosynthesis (map00400), Valine, leucine and isoleucine biosynthesis (map00290), Arginine and proline metabolism (map00330), Alanine, aspartate and glutamate metabolism (map00250), Histidine metabolism (map00340) and Lysine biosynthesis (map00300) were significantly downregulated in the HIV positive group, compared to HIV negative (Table 1). We also found other differential pathways including fatty acid biosynthesis (map00061) and steroid (cholesterol) degradation (map00984), as presented in Table S4.

ART-associated microbial functional shifts

We investigated the impact of ART in the HIV infection setting. When we compared 262 ART positive individuals with 184 ART negative HIV positive individuals, there were no differential pathways with greater than 40% coverage. As shown in Figure. 4A, Hydroxymethylpyrimidine kinase (K14153) level significantly differed across geographic regions (p < 0.001), and was greater in the ART group than in the ART negative group across six different cities, all except Isingiro (Uganda). After adjusting for geographic region and sex:sexual behavior, the effect size between the ART negative and positive groups in the abundance of Hydroxymethylpyrimidine kinase (K14153) was 1.1 (± 0.26) with P_adj < 0.001. As in Figure 4B, Thiamine metabolism’s 16 other gene families downregulated by HIV infection were upregulated by ART, however those differences were not statistically significant (P_adj > 0.05). Likewise, 15 out of the 16 gene families downregulated in One carbon pool by folate were upregulated by ART, but those changes were not statistically significant (Figure 4C). We did not find any differential gene families in the Pantothenate and CoA biosynthesis (map00770) pathway between the ART negative and positive specimens. Therefore, we concluded that the microbial vitamin B synthesis impaired by HIV infection was not significantly recovered by ART.

Figure 4.

Hydroxymethylpyrimidine kinase abundance changes associated with ART. (A) Hydroxymethylpyrimidine kinase (K14153) level of 184 HIV positive ART negative (dark blue) and 262 HIV positive ART positive (light green) individuals in different geographic regions. (B) HIV infection effect size (grey) and ART effect size (green) of 18 gene families within the Thiamine metabolism (map00730) pathway. (C) HIV infection effect size (grey) and ART effect size (green) of 17 gene families within the One carbon pool by folate (map00670) pathway.

Lysine biosynthesis (map00300) was marginally enriched in the ART group, compared to ART negative HIV patients (P_adj = 0.063). Four differential gene families out of a total of 45 gene families in this pathway were upregulated in the ART group (coverage=8.9%), suggesting that HIV-infection-specific depletion of lysine biosynthesis was moderately recovered. In addition, Phenylalanine metabolism (map00360) was downregulated by ART (coverage=11% and P_adj = 0.001) and Steroid degradation (map00984) was downregulated by ART (coverage=27% and P_adj = 0.001).

DISCUSSION

We conducted functional annotation of the 16S rRNA gene data of the 249 HIV negative and 305 HIV positive individuals. Around half of the gene families among several B-vitamin synthesis pathways were significantly downregulated in the HIV positive group compared to the negative. The observed downregulation suggests a decreased capacity for the microbiome to synthesize and activate thiamine, folate, and pantothenate, and potentially decreased bioavailability of these B-vitamins to the host. Further studies, including metabolomic profiling and in vitro experiments, would be required to corroborate our observations. Our observations are concurrent with the high prevalence of thiamine deficiency [28–30] and folate deficiency [31–33] among people living with HIV, suggesting the potential role of gut microbiota for HIV-associated B-vitamin deficiencies.

B-vitamins are essential micronutrients, and their deficiencies are associated with increased pro-inflammatory responses [34, 35]. While B-vitamins are mainly contributed by diet, human gut bacteria can supply B-vitamins via de novo synthesis [36, 37]. Gut bacteria are estimated to contribute a maximum of 2.3% to the host’s daily intake of thiamine and a more substantial 37% to folate [38]. The significance of microbial-derived B-vitamins has been demonstrated in animal models; germ-free rats fed vitamin B-deficient diets suffered a higher degree of deficiency than conventional rats on the same diet [39, 40]. Likewise, germ-free drosophila fed B-vitamin free diets did not survive to pupation, while conventional animals did [41]. Germ-free animals also previously required substantial supplementation to overcome inherent vitamin deficiencies caused by a lack of commensal bacteria [42], indicating the importance of bacterial vitamin synthesis. Gut microbiome changes that cause downregulation of microbial vitamin B synthesis may therefore contribute to these deficiencies.

Microbial dysbiosis resulting in insufficient B-vitamin biosynthesis may contribute to HIV-associated neurocognitive disorders. Thiamine, folate, and pantothenate deficiencies have been commonly associated with neurodegenerative disorders with increased inflammation in the central nervous system [34, 43–45]. The impaired microbial vitamin B synthesis observed in this study may contribute to HIV-associated B-vitamin deficiencies, suggesting a potential link to neurocognitive impairment in HIV-infected individuals [46].

The altered microbiome’s impaired thiamine and folate biosynthesis may also contribute to HIV-associated cardiovascular disease. Thiamine deficiency has been reported to cause abnormal heart and circulatory system function [43], and its deficiency has been commonly reported among patients with cardiovascular disease [47]. Likewise, folate deficiency has been considered as a risk factor for cardiovascular disease [48, 49]. Both thiamine and folate deficiencies are associated with an increase in pro-inflammatory signaling [34, 35] that may play a causative role in the progression of cardiovascular diseases [50]. Thiamine supplementation was reported to improve endothelial function [51] and attenuate hypertension [52] and heart failure [53–55]. Similarly, supplementation with folic acid has shown to reduce the risk of stroke in cardiovascular disease patients [56]. The reported associations between B-vitamin deficiencies and cardiovascular disorders are consistent with the elevated thiamine and folate deficiencies observed in HIV patients, who have a higher risk of hypertension and cardiovascular disease than uninfected individuals [57, 58]. The observed downregulation of microbial vitamin B synthesis suggests that the gut microbiome may contribute to cardiovascular dysfunction in HIV-infected individuals.

Vitamin supplementation has been explored as a low-cost HIV intervention, which generally slowed HIV disease progression [59–61]. Improvement of survival in HIV patients by thiamine and vitamin B6 supplements has been reported [62]. Additionally, probiotic supplementation may ensure constant bioavailability of essential vitamins to the host, as demonstrated by folate-producing probiotics in animal models [63]. More importantly, the capacity for folate-producing bacteria to colonize the human intestine has been shown; these probiotics significantly increased fecal folate levels [64].

We also observed that most essential amino acid biosynthesis pathways are downregulated in the microbiome of HIV-infected individuals. Our results are in concordance with a recent metabolite screening study of gut bacteria where Proline, Phenylalanine, and Lysine were below the detection limit in HIV infected individuals [65]. Furthermore, it has been demonstrated that the plasma concentrations of most essential amino acids were significantly reduced in HIV-infected individuals [66–68] with some conflicting reports [69, 70].

Amino acid deficiencies may also be involved in the progression of HIV-associated cardiovascular disease and neurocognitive disorders. Amino acids, including glycine and leucine, were reported to have anti-atherogenic properties, while other amino acids to have pro-atherogenic effects [71]. Glycine was also reported to ameliorate inflammation and improve neurological function [72]. Furthermore, HIV-infected individuals with cognitive deficits were reported to have decreased levels of glutamate in the brain [73]. Our findings of impaired microbial amino acid syntheses indicate potential associations of the gut microbiome with HIV comorbidities.

A number of studies have explored amino acid supplementation as an HIV treatment option. Supplementation with glutamine has been demonstrated to improve several symptoms of HIV, including decreases in intestinal permeability in AIDS patients [74], increases in body weight in AIDS patients with wasting syndrome [75], and increased killer cell activity in HIV infected individuals [76]. Supplementation with a formula containing arginine also increased body weight [77] and supplementation with sulfur-containing amino acids increased natural-killer and T-cell function in patients with HIV [78]. However, we must keep in mind the upper limit of tolerance to supplemented amino acids, after which there was a risk of adverse effects and toxicity [79]. Our findings suggest the potential therapeutic use of probiotics. While the net microbial contribution of amino acids to host metabolism remains uncertain, several studies have reported the presence of bacterial-derived amino acids in human blood plasma [80–82]. Therefore, the reversal of HIV-induced microbiome shifts via probiotics may increase the bioavailability of endogenous essential amino acids and thereby reduce their deficiencies in HIV.

There were several limitations to our meta-analysis. First, ART regimen and duration can impact microbiota differently [8, 9], however, these data were limited and thus not included in our analysis. Second, metagenome profiles were inferred from 16S data. Metagenomic prediction is limited in that i) it utilizes reference genomes and thus rare functions not present in reference genomes may not be identified and ii) strain-level resolution cannot be inferred from 16S data and therefore strain-specific functionality cannot be annotated [83]. Nonetheless, metagenome profiles inferred from 16S rRNA gene data have shown a high level of consistency with corresponding metagenomics; correlation coefficients between the two profiles ranged from 0.83 to 0.87 [83]. Finally, inferred metagenomics is less powerful than metatranscriptomics or metabolomics as they can provide more direct measures of functional microbiome shifts [84].

Our meta-analysis provides a new perspective for understanding vitamin B and amino acid deficiencies in HIV infected individuals. Positive correlations among HIV infection and microbial vitamin B and amino acid deficiencies suggest a potential relationship between the microbiome and HIV-associated disorders. Microbiome restoration may therefore have important implications for therapeutics in HIV comorbidities.

MATERIALS AND METHODS

Meta-analysis cohorts

We collected previously published 16S ribosomal RNA (rRNA) gene sequences from 821 individuals [1–9]. These studies documented HIV status, MSM designation, sex, age, geographic location, and ART status. 16S data were generated from fecal samples [1, 3, 4, 6–9], fecal aspirates [2], rectal swabs [2, 5], and colon biopsies [2] via Miseq sequencing [1–5, 7–9] or pyrosequencing [6]. The current meta-analysis consisted of 249 HIV negative, 310 HIV positive ART negative (not receiving ART at the time of specimen collection), and 262 HIV positive ART positive specimens. Individual-level ART regimen and duration data were limited and were therefore not included in the meta-analysis. Tables S1–S3 presented each specimen’s host factors along with their publication source [1–9].

Functional annotation of microbiota from 16S rRNA data

Each downloaded fastq file was preprocessed using the software-as-a-service, CCMP (Cloud Computing for Microbiome Profiling (CCMP, https://ccmp.usc.edu) [85], which processed raw 16S rRNA gene sequence data by assembling paired-end reads to single reads, quality filtering, and removing chimeric sequences using MOTHUR [86]. CCMP used this preprocessed fasta file as an input for QIIME [87] operational taxonomic unit (OTU) picking. The representative sequences for the OTUs and OTU table created by QIIME were used as inputs for PICRUSt2 to produce metagenome predictions of the 16S rRNA gene data [83].

Using KEGG Orthology (KO) metagenomes, the gene family abundance table (pred_metagenome_unstrat.tsv) was obtained using picrust2_pipeline.py with default parameters. These metagenomic profiles were then normalized using the ALDEx2 R package (https://github.com/ggloor/ALDEx2) and subjected to multivariable regression for the identification of differential gene families.

Multivariable regression

We fit a multivariable linear model to gene family abundance profiles of 249 HIV negative and 305 HIV positive (ART negative) individuals. In order to geographically match the 249 HIV negative individuals, 305 of the 310 total HIV positive (ART negative) individuals were chosen. Available host factors from these individuals included HIV status, MSM designation, sex, age, and geographic location. MSM designation and sex were combined into one variable (sex:sexual behavior) with three categories: female, male:non-MSM, and male:MSM.

We first selected host factors that were associated with each gene family’s abundance. When host factors were associated with gene family abundance (p < 0.2), they were included in the multivariable regression. For example, a multivariable model with significant predictors of age (Ai), sex:sexual behavior (Si), and geographic location (Li) has the form,

$$F_i={\beta }_0+{\beta }_{HIV}HI{V}_i+{\beta }_A{A}_i+{\sum }_{j\ne 1}{\beta }_{Sj}I\left(S_i=j\right)+{\sum }_{k\ne 1}{\beta }_{Lk}I\left(L_i=k\right)+{\epsilon }_i,$$ Eq. (1)

where F_i is individual i’s normalized abundance of a given gene family, β₀ is the intercept (assuming reference levels of sex:sexual behavior=female and geographic location=Barcelona), β_HIV is the conditional effect of HIV infection (HIV_i = 0 for HIV negative and HIV_i = 1 for HIV positive), β_A is the conditional effect of age, β_Sj is the conditional difference in effect of the jth sex:sexual behavior compared to the first (female, male:MSM or male:non-MSM), β_Lk is the conditional difference in effect of the kth geographic location compared to the first and ε_i is the residual. Here we selected eight cities (Barcelona, Madrid, Logrono, Stockholm, Abuja, Mbarara, Kiruhura, and Denver) where both HIV negative and positive specimens were available. We compared 184 ART negative patients with 262 ART positive HIV patients. Again, 184 of the 310 total ART negative individuals were subsetted to geographically match the 262 ART positive individuals. Seven cities (Barcelona, Madrid, Logrono, Abuja, Mbarara, Isingiro, and Denver) were selected, from which both ART negative and positive specimens were available.

The multivariable model estimated β_HIV, the effect size of HIV infection on gene family abundance along with its standard error with adjustment for these host factors. Each of the total 7,847 gene families were individually modeled and these gene families’ p values were multiplicity-adjusted by the Benjamini and Hochberg (BH) method. Differential gene families with an adjusted p value (P_adj) for the HIV effect were designated for the identification of differential pathways.

Differential pathway identification

The list of differential gene families along with the total gene family list were used as inputs for FMAP (Functional Mapping and Analysis Pipeline) [88]. Using FMAP_pathway.pl, each pathway’s p value was obtained from Fisher’s exact test. The Benjamini and Hochberg (BH) method was then used to adjust each pathway’s p value. Differential pathways were designated when their adjusted p value was less than 0.05 and coverage (the proportion of differential families) was greater than 40%. This coverage cut-off was chosen as to designate differential pathways with a significant portion of differential gene families.

Highlights.

• HIV infection was associated with impaired microbial vitamin B synthesis

• Microbial thiamine and folate biosynthesis pathways were estimated to be down regulated by HIV infection

• Microbial amino acid biosynthesis pathways were estimated to be down regulated by HIV infection

• Microbial vitamin B and amino acid synthesis pathways were not significantly recovered by ART