High Abundance of genus Prevotella in the gut of perinatally HIV-infected children is associated with IP-10 levels despite therapy

Perinatal HIV infection is characterized by faster HIV disease progression and higher initial rate of HIV replication compared to adults. While antiretroviral therapy (ART) has greatly reduced HIV replication to undetectable levels, there is persistent elevated inflammation associated with HIV disease progression. Alteration of gut microbiota is associated with increased inflammation in chronic adult HIV infection. Here, we aim to study the gut microbiome and its role in inflammation in treated and untreated HIV-infected children. Examination of fecal microbiota revealed that perinatally infected children living with HIV had significantly higher levels of genus Prevotella that persisted despite ART. These children also had higher levels of soluble CD14 (sCD14), a marker of microbial translocation, and IP-10 despite therapy. The Prevotella positively correlated with IP-10 levels in both treated and untreated HIV-infected children, while genus Prevotella and species Prevotella copri was inversely associated with CD4 count. Relative abundance of genus Prevotella and species Prevotella copri showed positive correlation with sCD14 in ART-suppressed perinatally HIV-infected children. Our study suggests that gut microbiota may serve as one of the driving forces behind the persistent inflammation in children despite ART. Reshaping of microbiota using probiotics may be recommended as an adjunctive therapy along with ART.


Results
Perinatally HIV-infected children show a distinct pattern of gut microbiota compared to uninfected controls. To determine the composition of gut microbiota in the setting of perinatal HIV infection, we assessed fecal microbiota in perinatally HIV-infected children. We determined the relative abundance of fecal microbiota at the phylum, family, and genus level and compared their levels between perinatally HIV-infected children and uninfected controls. Perinatally HIV-infected children showed distinct pattern of fecal microbiota compared to uninfected controls ( Fig. 1A-R). The three most predominant phyla accounting for >97% of the overall bacterial abundance in fecal samples of both HIV-infected children and uninfected controls were Bacteroidetes, Firmicutes and Proteobacteria. No significant difference was observed in the overall relative abundances at the phylum level between perinatally HIV-infected children and uninfected controls (p = 0.08) (Fig. 1A). However, when relative abundance of individual phylum was compared between HIV-infected children and controls, Bacteroidetes and Spirochaetes were found to be significantly elevated (p = 0.002 and p = 0.0007, respectively) whereas Firmicutes, Actinobacteria and Cyanobacteria were significantly lower (p = 0.002, p = 0.0007 and p = 0.03, respectively) in HIV-infected children compared to uninfected controls (Figs 1B-E and S1B). The relative abundance of the remaining phyla did not show any significant difference between perinatally HIV-infected children and uninfected controls (Fig. S1A,C-E). On comparing the bacteria at the family and genus level, we observed overall relative abundances to be significantly different between HIV-infected children and uninfected controls (p = 0.0001 and p = 0.002, respectively) ( Fig. 1F,L). Then, we individually compared the relative abundance of top abundant bacterial families between perinatally HIV-infected children and uninfected controls and observed that the relative abundance of Prevotellaceae was significantly higher (p = 0.0007), while Bacteroidaceae (p = 0.0008), Ruminococcaceae (p = 0.005), Lachnospiraceae (p = 0.01), and Rikenellaceae (p < 0.0001) were significantly lower in perinatally HIV-infected children compared to uninfected controls (Figs 1G-K, S2A-G). Among the abundant genera, the relative abundance of Prevotella and Megasphaera were significantly increased (p = 0.0007, and p = 0.02, respectively), whereas that of Bacteroides (p = 0.0008), Faecalibacterium (p = 0.01), Ruminococcus (p = 0.006) and Haemophilus (p = 0.04) was significantly lower in HIV-infected children compared to uninfected controls ( Fig. 1M-R). The relative abundance of remaining abundant families and genera did not show difference between perinatally HIV-infected children and uninfected controls (Figs S2 and S3). The ratio of abundance of Bacteroides to Prevotella was significantly lower in perinatally HIV-infected children compared to uninfected controls (p = 0.0008, Fig. S5A). Our data shows that perinatally HIV-infected children have distinct pattern of gut microbiota compared to uninfected controls. The median values of relative abundance of significantly distinct phyla, families, and genera between perinatally HIV-infected children and uninfected controls are shown in Table S1.

Impact of ART on the fecal microbiota in perinatally HIV-infected children.
In order to determine the impact of antiretroviral therapy on fecal microbiota of perinatally HIV-infected children, we segregated perinatally HIV-infected children into treatment naïve and ART-experienced group and compared fecal microbiota across three groups. At the level of phylum, we observed overall relative abundance to be significantly different between treatment naïve HIV-infected children and uninfected controls (p = 0.01) but not between ART-suppressed HIV-infected children and uninfected controls (p = 0.27) (Fig. 2A). The relative abundance of Bacteroidetes (p = 0.002) and Spirochaetes (p = 0.002) were found to be significantly higher whereas Firmicutes The horizontal line and deviation in the dot plot represents median and interquartile range (IQR), respectively. Mann-Whitney-U-test was performed for statistical analysis. p < 0.05 was considered to be significant. Significant differences in the overall relative abundance of fecal microbiota at phylum, family and genus level between HIV negative and HIV positive children were estimated using adonis function of vegan R package.   (p = 0.0007) and Actinobacteria (p = 0.0002) were significantly lower in treatment naïve HIV-infected children compared to uninfected controls ( Fig. 2B-E). In ART-suppressed HIV-infected children, the relative abundance of bacterial phyla were normalized to that of uninfected controls except Spirochaetes which remain elevated despite therapy (p = 0.002) (Fig. 2E). Comparisons of bacterial families showed significant difference in the overall relative abundances between uninfected controls and treatment naïve (p = 0.001) or ART-suppressed HIV-infected children (p = 0.02) (Fig. 2F). The relative abundance of Prevotellaceae were significantly increased (p = 0.0009 and p = 0.02 respectively) whereas Bacteroidaceae (p = 0.01 and p = 0.002) and Rikenellaceae (p = 0.0002) were significantly lower in both treatment naïve and ART-suppressed HIV-infected groups compared to uninfected controls (Fig. 2G,H). The relative abundance of other abundant families Ruminococcaceae (p = 0.002) and Lachnospiraceae (p = 0.009) were significantly lower in treatment naïve HIV-infected children whereas no significant difference was found in these families in ART-suppressed HIV-infected children compared to controls. (Fig. 2G-K). We also found overall relative abundance to be significantly different at the genus level between uninfected controls and treatment naïve (p = 0.005) or ART-suppressed HIV-infected children (p = 0.01) (Fig. 2L). ART did not show any effect on the relative abundance of Prevotella and Bacteroides. Compared to uninfected controls, treatment naïve and ART-suppressed HIV-infected children had significantly higher relative abundance of Prevotella (p = 0.0009 and p = 0.02 respectively) and lower abundance of Bacteroides (p = 0.01 and p = 0.002 respectively) ( Fig. 2M,N). The relative abundance of Megasphaera was significantly higher (p = 0.02) and Ruminococcus (p = 0.0007) was significantly lower only in treatment naïve HIV-infected children compared to controls (Fig. 2P,Q). However, compared to treatment naïve HIV-infected children, the relative abundance of Ruminococcus was significantly higher in ART-suppressed children, suggesting the complete restoration of this genus by ART (Fig. 2Q). We did not observe difference in the relative abundance of Faecalibacterium (p = 0.06) and Haemophilus (p = 0.1) in treatment naïve HIV-infected children compared to controls (Fig. 2O,R). However, ART-suppressed HIV-infected children had decreased Faecalibacterium compared to controls (Fig. 2O). Our results suggest that the fecal microbiota of HIV-infected children did not completely get restored after ART treatment. The median values of relative abundance of significantly distinct phyla, families, and genera among treatment naïve, ART-suppressed HIV-infected children and uninfected controls are shown in Table S2.

Microbial taxa associated with HIV and ART usage.
In order to gain insight into the bacterial taxa mostly associated with HIV, we performed linear discriminant analysis effect size analysis (LEfSe) that revealed significant difference in the abundance of 27 taxa between perinatally HIV-infected children and uninfected controls (LDA score >2.5, Fig. 3A,B). In addition to the taxa shown in Fig. 1, LEfSe revealed decreased Odoribacteraceae, Barnesiellaceae and Oscillospira in HIV-infected children compared to controls (Fig. 3A,B). Further, we carried out LEfSe analysis to determine the taxa most associated with ART usage. We found 32 significant differentially abundant taxa between treatment naïve and ART-suppressed children. Of the families, Clostridiaceae and Dehalobacteriaceae were enriched whereas Enterobacteriaceae and Leptotrichiaceae were decreased in ART-suppressed HIV-infected children. Additionally, genera Selenomonas, Clostridium, Leptotrichia and Pantoea were enriched in treatment naïve HIV-infected children whereas Longilinea, Dehalobacterium, PD-UASB-13 were more prevalent in ART-suppressed children (Fig. 3C,D).

Analysis of diversity of microbial community.
Alpha diversity is the assessment of diversity within a habitat. We analysed the alpha diversity of fecal microbiota using the Shannon and Simpson index. The fecal microbiota diversity of both treatment-naïve and ART-suppressed HIV-infected children was significantly decreased when compared to uninfected controls (Fig. 4A,B). Beta diversity with principal component analysis (PCA) showed that the fecal microbiota composition of most of the treatment-naïve and ART-suppressed HIV-infected children were grouped into cluster and different from uninfected controls (Fig. 4C).

Prevotella is indirectly associated with CD4 T cells in perinatally HIV-infected children. As CD4
depletion is a hallmark of HIV disease progression, we sought to determine association of genera that were significantly differentially abundant in HIV-infected children with CD4 count. The genus Prevotella was only found to be indirectly associated with CD4 count in all HIV-infected children (p = 0.007; r = −0.6) (Figs 4D and 6C). The indirect association between Prevotella and CD4 T cell count remained significant in treatment naïve and ART-suppressed perinatally HIV-infected children (Fig. 4E,F). Prevotella did not show any correlation with CD4 T cell count in uninfected controls (data not shown).

Increased abundance of Prevotella copri in the gut of perinatally HIV-infected children despite ART.
We were intrigued by the increased relative abundance of Prevotella in perinatally HIV-infected children despite suppressive therapy and wanted to further characterize the genus Prevotella to species level. At species level, we identified Prevotella copri and Prevotella stercorea. However, only the relative abundance of Prevotella copri was found to be elevated in HIV-infected children compared to controls (p = 0.001) (Fig. 5A). The relative abundance of Prevotella copri was also significantly elevated in treatment naïve (p = 0.003) and ART-suppressed HIV-infected children (p = 0.02) compared to uninfected controls, suggesting that ART does not normalize the the dot plot represents median and interquartile range (IQR), respectively. Kruskal-Wallis one-way ANOVA with Dunn's multiple comparison was performed for statistical analysis. p < 0.05 was considered to be significant. Significant differences in the overall relative abundance of fecal microbiota at phylum, family and genus level between HIV negative controls, On ART and treatment naïve HIV-infected children were estimated using adonis function of vegan R package. *p < 0.05, **p < 0.005, p < 0.0005. levels of Prevotella copri in HIV-infected children (Fig. 5B). Additionally, the relative abundance of other abundant species Bacteroides plebieus, Bacteroides uniformis and Faecalibacterium prausnitzii in the fecal sample of HIV-infected children was significantly decreased compared to controls (Fig. S4) and remained lowered regardless of treatment. Like genus Prevotella, the relative abundance of P. copri also showed indirect association with CD4 T cells in overall and treatment naïve perinatally HIV-infected children (Fig. 5C,D) however it failed to achieve level of significance in ART-suppressed HIV-infected children (p = 0.08; Fig. 5E). The median values for the relative abundance of Prevotella copri in three groups are shown in Table S2.
Soluble IP-10 is elevated in HIV-infected children and are associated with Prevotella despite ART. Multiplex analysis revealed significantly high levels of IL-1β, TNF-α, IP-10 (Interferon-gamma inducible protein 10), MDC (Macrophage-derived chemokine), GRO (Growth regulated oncogene chemokine), and sCD40L (Soluble CD40-ligand) in treatment naïve HIV-infected children compared to uninfected controls. Treatment appeared to restore the levels of these cytokines except IP-10 ( Table 1). The IP-10 level was significantly elevated in perinatally HIV-infected children (p = 0.002) and it remained elevated in treated group (p = 0.03) in comparison to uninfected controls (Fig. 6A,B). This suggests a potential role of IP-10 in maintaining    Given that gut microbiota maintains gut epithelial barrier, we then evaluated whether significantly differentially abundant genera in HIV-infected children was associated with plasma level of sCD14. We found Haemophilus to be negatively associated with sCD14 (r = −0.56, p = 0.01) (Fig. 6C). Relative abundance of Prevotella and also Prevotella copri directly correlated with plasma sCD14 level in treated HIV-infected children (Fig. 7F,G).

Discussion
In this paper, we studied the fecal microbiota in treatment-naïve and ART-suppressed perinatally HIV-infected children, and compared them with that of uninfected controls. Our study showed a unique pattern of fecal microbiota with elevated relative abundance of the genus Prevotella and species Prevotella copri in HIV-infected children that persisted during suppressive ART. The genus Prevotella was directly associated with soluble IP-10 and inversely correlated with CD4 count despite ART. The genus Prevotella and species Prevotella copri positively correlated with sCD14, a marker of microbial translocation, in ART-suppressed perinatally HIV-infected children implying that Prevotella may potentially be involved in persistent inflammation in perinatally HIV-infected children during therapy. Alteration in the overall gut microbial diversity during HIV infection in adults currently remains controversial. Some studies reported gut microbiota to be less diverse in untreated HIV-infected adults 28,29,40,41 or the diversity to be decreased after ART 28,36 , while others found no alteration in the diversity of gut microbiota 26,27,31,33,34 . In our study, we found decreased α-diversity in the fecal microbiota of HIV-infected children despite therapy. This discrepancy among previous studies could be due to different ethnicity, age, different stage of HIV infection of enrolled HIV-infected subjects, selection of sample type, use of antibiotics by study participants or different technologies used to study gut microbiota composition. Perinatally HIV-infected children and uninfected controls participated in our study had preserved CD4 count and shared similar geographical region and dietary habits. As antibiotics alter gut microbiota composition profoundly 42 , we only included children who were not on antibiotics within last one month of sample collection. Similar to studies on HIV-infected adults, beta diversity indices such as PCA showed composition of fecal microbiota in most of the perinatally HIV-infected children to be different from uninfected controls. Moreover, most of the treated HIV-infected children were clustered with treatment naïve HIV-infected children suggesting similar pattern of gut microbiota in treated and untreated perinatally HIV-infected children. Thus, despite successful viral suppression, ART could not completely restore fecal microbiota in children with HIV.
Untreated HIV-infected children had increased relative abundance of Bacteroidetes and decreased Firmicutes whereas studies on HIV-infected adults reported decreased Firmicutes with no difference in Bacteroidetes   The horizontal line and deviation in the dot plot represents median and interquartile range (IQR), respectively. Mann-Whitney U-test and Kruskal-Wallis one-way ANOVA with Dunn's multiple comparison were performed for statistical analysis. p < 0.05 was considered to be significant. Association between (F) relative abundance of Prevotella and plasma sCD14 level (ng/ml) (G) relative abundance of Prevotella copri and plasma sCD14 level (ng/ml) in ART-suppressed perinatally HIV infected children. Statistical analysis was performed using Spearman t-test. A linear regression line is included in each plot. The p < 0.05 was considered to be significant. abundance compared to controls 25,29,30,43 . We did not observe any difference in the relative abundance of phylum Proteobacteria in perinatally HIV-infected children compared to uninfected controls, which is in contrast to a recent study by Dinh et al. who showed significantly enriched relative abundance of Proteobacteria in the fecal samples of adults living with HIV 31 . The variations in the relative abundance of Proteobacteria in HIV-infected subjects have been observed across studies. Some previous studies that evaluated intestinal mucosal biopsies showed increased relative abundance of Proteobacteria in HIV-positive patients 25,28 , whereas other failed to observe any difference in the fecal and rectal sponge samples of HIV-positive patients compared to healthy controls 26,29 .
Our observations of increased relative abundance of Prevotellaceae and decreased relative abundance of Bacteroidaceae and Rikenellaceae in untreated HIV-infected children are in agreement with the previous findings in HIV-infected adults 25,27,36 . Similar to HIV-infected adults [25][26][27][28]32 , we also observed significantly increased relative abundance of Prevotella with decreased abundance of Bacteroides in perinatally HIV-infected children. Increase in gut Prevotella has been reported in chronic inflammatory diseases such as Rheumatoid arthritis 44,45 and metabolic syndrome 46 . Recent studies have shown that the high abundance of Prevotella in HIV-infected individuals is attributed to their sexual behavior rather than to the disease itself 32,47 . However, in our study prior sexual behavior does not have any relevance because the study is focussed on perinatally HIV-infected children who have either not reached or just reached an early stage of puberty. A study by Dillon et al. showed positive correlation of relative abundance of Prevotella with mucosal CD4+ and CD8+ T cell activation, and pro-inflammatory cytokine-producing mucosal CD4+ and CD8+ T cells 25 however, Ling et al. did not find any significant correlation between Prevotella abundance and systemic inflammatory cytokines 27 in HIV-infected adults. This suggests that the potential inflammatory role of Prevotella is only confined to the local gut immune system. In contrast to the previous study in adults 25,27 , we found that relative abundance of Prevotella was indirectly associated with peripheral CD4 count and directly associated with MDC in perinatally HIV-infected children. MDC has emerged as novel modulator of lung inflammation in mice 48 and an inflammatory marker to assess the severity of atopic dermatitis in children and infants 49 . More recently, the increased levels of MDC in HIV-infected adults have also been reported 50 . This suggests the potential inflammatory role of MDC in HIV-infected children, however the level of MDC normalizes after initiation of ART. Interestingly, absolute abundance of Prevotella was found to be directly associated with soluble IP-10 in untreated and ART-experienced perinatally HIV-infected children. Our results indicate that Prevotella outgrowth (represented by increase in the absolute number of Prevotella), rather than Prevotella evenness may drive the production of IP-10.
Level of IP-10 has been shown to be elevated in many viral infections, including acute West Nile virus 51 , acute and chronic HCV 52,53 and severe influenza infection 54,55 suggesting the role of this chemokine in immune response to viral infection. Elevated IP-10 has been reported during acute and chronic HIV infection 52,[56][57][58][59] . Consistent with previous studies on HIV-infected adults 60,61 , ART could not normalise the plasma level of IP-10 in perinatally HIV-infected children to that of uninfected controls. Previous studies on acute and chronic HIV infection in adults showed positive correlation of IP-10 with viral load 57,59 and negatively with CD4 count 56 , thus proposed IP-10 to be the marker of disease progression. However, we failed to find direct correlation between IP-10 and HIV viral load in perinatally HIV-infected children, possibly because of limited sample size.
Infection of monocyte-derived macrophages (MDM) with HIV-1 has been shown to stimulate the production of GRO-α chemokine in-vitro. This chemokine in turn prompt HIV replication in macrophages and lymphocytes, thereby contributing to HIV pathogenesis 62 . Our observation of elevated level of GRO in untreated HIV-infected children might be due to viral replication.
Elevated sCD14, a marker of microbial translocation is associated with increased mortality among HIV-infected individuals 63 and also sCD14 level in HIV-infected mothers have reported to be directly associated with risk of mother to child transmission 64 . Consistent with prior findings [65][66][67][68] , perinatally HIV-infected children had increased plasma level of sCD14 compared to uninfected children, showing gut epithelial barrier disruption and microbial translocation that persists even after suppressive therapy. Our study also showed positive correlation of sCD14 with abundance of genus Prevotella and Prevotella copri in treated HIV-infected children. However the presence of virus appears to disrupt this correlation as seen in untreated HIV-infected children. This suggests that either HIV or other unknown factors driven by HIV viral load contribute to microbial translocation in untreated subjects. Additional contributory factors to increased microbial translocation in untreated children are the decreased relative abundance of Lachnospiracea and Ruminococcacea families that include many butyrate-producing bacterial species that are essential for preserving the gut mucosal integrity 69 .
Ours is the first study to investigate the fecal microbiota composition of treatment naïve and ART-suppressed perinatally HIV-infected children that showed association of genus Prevotella with IP-10 and sCD14, a microbial translocation marker. These results suggest that Prevotella is proinflammatory in nature that may augment the inflammation in perinatally HIV-infected children despite ART, and may potentially promote chronic complications. Our study has few limitations. First, in our cross sectional study fecal microbiota has been used as a proxy for gut microbiota. Fecal microbiota only represents gut lumen microbiota and does not reflect mucosal-associated microbiota, which is important to consider as microbes on the gut mucosa can potentially interact with gut associated lymphoid tissue (GALT). However fecal sample was only the realistic sample for non-invasive study. Second, though we showed impact of ART on the fecal microbiota of HIV-infected children, the effect of individual antiretroviral drugs on fecal microbiota is not investigated in our study.
In summary, our study found distinct gut microbiota composition in untreated and ART-suppressed perinatally HIV-infected children compared to uninfected controls. They had higher abundance of Prevotella and decreased Bacteroides, and elevated levels of IP-10 that persists despite ART. The higher abundance of Prevotella may promote HIV pathogenesis as suggested by the negative association of Prevotella with CD4 count. Interestingly, the direct correlation between Prevotella and IP-10 suggests that IP-10 may be regulated by gut microbiota and could potentially contribute to ongoing inflammation and T cell depletion. Future studies to understand the mechanism of Amplicon Library preparations and Sequencing. Identification of bacteria was carried out by amplifying and sequencing bacterial 16S rRNA gene at Eurofins Genomics India Pvt Ltd (Bangalore, India). Oligonucleotide primers (Forward -5′ GCCTACGGGNGGCWGCAG 3′ and Reverse-5′ ACTACHVGGG TATCTAATCC 3′) that target V3-V4 region of 16S rRNA gene were used to generate amplicons. The amplicon libraries were prepared using Nextera XT Index Kit (Illumina inc.) as per the 16S Metagenomic Sequencing Library preparation protocol and also include multiplexing index sequences and illumina adapters. The amplicon libraries were purified by 1X AMpureXP beads, quantified using Qubit fluorometer and analyzed in 4200 Tape Station system (Agilent Technologies) using D1000 Screen tape as per manufacturer instructions. Sequencing of the amplicons was performed on Miseq Platform using 2 × 300 v3 Miseq Kit.  sCD14 detection. Level of sCD14 in plasma was quantified by commercially available Human sCD14 Elisa kit according to the manufacturers' protocols (R&D Systems, Minneapolis, MN). Each sample was analysed in duplicate and sCD14 level was expressed in ng/ml.

Sequence data and Statistical analysis.
Sequencing data was analysed using QIIME 1.9.1 70 . Raw reads were quality filtered using Trimmomatic v0. 38. Quality control involved the removal of reads with errors in barcode or primer sequences, reads containing ambiguous bases and low quality sequences (reads with more than 10% quality threshold (QV) <20 phred score) along with a sliding window of 20 bp and a minimum length of 100 bp. According to barcodes, the high quality reads obtained were then assigned to samples. Using UCLUST in QIIME the high quality sequences were clustered into operational taxonomic units (OTUs) with 97% sequence similarity. OTUs was used to analyze alpha diversity indices (Shannon and Simpson) using QIIME. For PCA, distance matrices were calculated using R. Using Greengenes reference databases, the taxonomies are assigned to the OTUs by UCLUST with a threshold of 90% sequence similarity. Explicet software v2.8.5 was used to analyze OTUs and create figures 71 . Significant differences in the overall relative abundance of fecal microbiota at phylum, family and genus level between treatment naïve, ART-suppressed HIV-infected children and uninfected controls were estimated using adonis function of vegan R package 72 . Statistical analysis of alpha diversity indices, the relative abundance of taxa, plasma cytokines, and soluble sCD14 levels between two and three groups was performed using non-parametric Mann Whitney U test and Kruskal-Wallis 1-way ANOVA with Dunn's multiple comparison respectively in GraphPad Prism v. 6. LEfSe analysis (http://huttenhower.sph.harvard.edu/lefse/) was used to identify differentially abundant bacterial taxa across three groups with alpha value of 0.05 and LDA score of >2.4 73 . Association between two variables was carried out by Spearman rank correlation test. Correlation coefficient r = 1, describes a positive association and r = −1, describes a negative correlation. A p value below 0.05 was considered significant. Heatmap was generated to visualize association between two variables using heatmap.2 function of gplots R package (https://CRAN.R-project.org/package=gplots).