The Relationship Between the Cervical Microbiome, HIV Status, and Pre-Cancerous Lesions

Nearly all cervical cancers are causally associated with Human Papillomavirus (HPV). The burden of HPV-associated dysplasias in Sub-Saharan Africa is influenced by HIV. To investigate the role of the bacterial microbiome in cervical dysplasia, cytobrush samples were collected directly from cervical lesions of 144 Tanzanian women. The V4 hypervariable region of the 16S rRNA gene was amplified and deep-sequenced. Alpha diversity metrics; Chao1, PD whole tree, and operational taxonomic Unit (OTU) estimates, displayed significantly higher bacterial richness in HIV positive patients (P = 0.01) than in HIV negative patients. Within HIV positive patients, there was higher bacterial richness in patients with high grade squamous intraepithelial lesions (HSIL; P = 0.13) than those without lesions. The most abundant OTUs associated with high-grade squamous intraepitheilal lesions (HSIL) were Mycoplasmatales, Pseudomonadales, and Staphylococcus. We suggest that a chronic mycoplasma infection of the cervix can contribute to HPV-dependent dysplasia by sustained inflammatory signals.

K l e i n e t a l A b s t r a c t K l e i n e t a l behind this approach was that the sight of the lesion is where tumors form, thus it bacteria associated at this sight were more likely to be relevant to disease status.

Participants and Ethical Precautions
This study reports findings derived from a larger cross-sectional cohort study analyzing demographics of HPV and cervical cancer in HIV positive and negative women from rural and urban Tanzania. All human subjects protocols were approved by safety committees at Ocean Road Cancer Institute (ORCI) and the University of Nebraska-Lincoln in accordance with the Helsinki Declaration. Participation by patients was entirely voluntary and written patient consent was required for inclusion in the study.
Tanzania is part of an ongoing study site to follow HIV and associated secondary viral infections in women at ORCI, the only cancer treatment hospital in Tanzania. Between March 2015 and February 2016, female patients undergoing cervical cancer screening were approached for enrollment in the study. Those who were pregnant, menstruating, under 18, reported being sick in the past 30 days, or had a preexisting, non-HIV, immunologic defect were excluded from the study. Disease histories as well as physical examinations were carried out to rule out any clinical symptoms or visible signs for these conditions. Samples were collected at three sites: ORCI in Dar es Salaam, and rural clinics in Chalinze and Bagamoyo. A total of 144 cervical cytobrush samples obtained from these women were sequenced, of which 138 produced at least 1000 reads, and 132 included successful measurements of both HIV status and cervical cytology.
K l e i n e t a l assigned to the OTUs using the assign_taxonomy.py command available in QIIME using the latest version of the Greengenes database (May 2013).

Statistical Analyses
The OTU table was rarefied across samples to the lowest sample depth (1000 reads) using QIIME based on the Mersenne Twister pseudorandom number generator. All statistical analyses were performed with samples at an even depth. Bar charts summarizing average taxonomic makeup of samples by HIV status and cervical cytology were constructed from the rarefied OTU table in QIIME. Heatmaps showing the relative abundance of bacterial taxonomic families were constructed using the 'plot_ts_heatmap' command using the mctoolsR package for R. Differences in bacterial families by HIV status or cervical cytology were evaluated using the 'taxa_summary_by_sample_type' command in mctoolsR using Kruskal Wallace. Families with less than 1% abundance were excluded in this analysis. Alpha diversity estimators Chao1, observed OTUs, and PD whole tree and rarefaction curves were calculated for the overall bacterial community using QIIME. Good's coverage test was performed to evaluate if adequate sampling depth was achieved. Mean alpha diversity estimates for HIV positive, HIV negative, NILM, LSIL, and HSIL groups were compared using nonparametric two-sample t-tests using Monte Carlo permutations in QIIME. The weighted and unweighted UniFrac distance matrix for bacterial communities were calculated using QIIME. Even depth across samples avoided biases that could be encountered when using the Unifrac metric [26]. Bacterial community composition differences were evaluated using the unweighted UniFrac distance matrix as an input for a distance-based redundancy analysis (db-RDA) in Qiime, where HIV status, cervical cytology, or HPV status were used as main effects. A heatmap was generated using the heatmap.2 command in the "ggplots" package for "R" using the Bray-Curtis distance matrix to visualize relationships between samples. Significance was declared at P ≤ 0.1 throughout this K l e i n e t a l study. The linear discriminant analysis effect size (LEfSe) was used to identify specific OTUs that differed HIV status and cervical cytology [27]. LEfSe uses a non-parametric factorial Kruskal-Wallis sum-rank test followed by a linear discriminate analysis to identify both statistically significant and biological relevant features. The OTU relative abundances were used as an input for LEfSe. Demographic data was examined using odds ratio and an associated p value to test for factors associated with HIV status and/or a positive VIA status. All p values are reported as FDR corrected p values.

Ethics Statement
All human subjects protocols were approved by safety committees at Ocean Road Cancer Institute (ORCI) and UNL in accordance with the Helsinki Declaration. Participation by patients was entirely voluntary and written patient consent was required for inclusion in the study.

Demographics
Of the (Figure 5c).
An abundance of non-Lactobacillus Bacilli was the most significant differentiating taxonomy between HIV positive and negative samples. Mycoplasma was also associated with HIV+ individuals, supporting the significant difference in relative abundance between HIV positive and negative groups shown previously using a direct Kruskal Wallace comparison. Interestingly, Ureaplasma (a member of Mycoplasmatales) and Lactobacillus reuteri were associated with HIV-patients, while other members of their respective families were associated with HIV+ K l e i n e t a l patients. This suggests the existence of metabolic niches in the cervical microbiome which may be populated by pathogenic or non-pathogenic associating bacteria.

Discussion
It is well established that certain members of the cervicovaginal microbiome protect against infection and pathogenesis. The primary defense mechanisms of the cervicovaginal mucosa are antimicrobial peptides, a pH of less than 4.5, and a microbiome dominated by Lactobacilli.
An imbalance in these defenses can result in physiochemical changes which produce alterations of the vaginal mucosa and cervical epithelium [28]. In particular, an abundance of  Mycoplasma is a low-abundance microbe which has been shown to cause cervicitis. However, the lack of significant associations in previous metagenomic studies, is largely due to a lack of optimization of statistical analyses for the presence of low abundance microbes. In our study, Mycoplasma was a prominent result, likely due to the large HIV positive proportion of the cohort, wherein immunosuppression allowed higher abundance of the bacteria to accumulate.
Statistical analysis of just the HIV negative portion of the cohort did not identify Mycoplasmatales as a significant factor. However, a linear increase in the abundance of Mycoplasmatales from NILM to HSIL seen in both HIV positive and negative groups.
Analysis of HIV negative, NILM samples showed an abundance of anaerobic bacteria. This may be a result of the sampling method used, rather than a representation of the average HIV negative NILM cervical microbiome in Tanzania. Because samples were only obtained from women attending the women's clinic, it is possible that some of the women had slightly unusual microbiota due to bacterial infections, but were NILM by the pap smear. If this were the case, it would explain the high contribution of anaerobic and/or pathogenic bacteria in NILM women attending the clinic. Another possibility is that cervicovaginal infection with anaeroabic bacterial in the general population is higher than previoiusly thought.
K l e i n e t a l In this study, we took great effort to control for variation in the cervical microbiome so as to reduce confounding effects that might obsure the bacterial communities which were associated with HPV pathogenesis. The HIV positive population is of particular interest, since they appear to show enhanced cervical microbiota associated with HPV pathegenesis. In future studies, recruiting a cohort of all HIV positive women with and without cervical lesions would be desirable in order to better characterize HIV-associated microbiota which promote HPV infection and progression to cervical cancer. Currently, no study has been conducted with such a focused and controlled group. It is clear that variables such as diet, genetic background, antibiotics or ART, can dramatically effect the microbiota, and thus should be carefuly controlled at the point of recruitment to the study.
Longitudinal studies of the cervical microbiome are needed to understand how microbe populations change over time, particularly in individuals with HSIL. Long-term longitudinal studies will be able to determine if changes in the cervical microbiota preempt and predict the development of lesions, or if the shift in microbiota happens after lesions have developed.
Because progression of HPV infection to cervical cancer is a process that takes decades, and in many individuals never reaches cancer at all, such a study would need to be large. Studies of the cervical microbiome can be further improved using metagenomic sequencing, rather than 16s or other targeted sequencing techniques which lack depth. 16s amplification ignores microbes which lack a gene to match the primers, for example; viruses, archaea, and eukaryotes are not accounted for. Because only a portion of one gene is being sequenced, the microbes present may only be estimated at the genus level or worse. Since the majority of medium or large scale cervicovaginal microbiome studies have used this method, the role of non-bacterial components of cervicovaginal microbiome in HPV infection and disease has not been characterized.
K l e i n e t a l As the world's HIV positive population grows, cervical cancer is expected to become an even more significant problem, despite increasing coverage of anti-retroviral treatment (ART).
Compared to the risk reduction after ART seen in other AIDS-defining cancers like Kaposi's sarcoma and non-Hodgkin's lymphoma, the risk of cervical cancer is not significantly affected, and recurrence rates remain high with or without treatment [37][38][39][40]. Studies in this area suggests that progression of HPV infection depends on immunological status of the host such that ART is only able to indirectly affect HPV pathogenesis, potentially through an effect on circulatory CD4+ cell count. Identifying which aspects of the local and systemic effects of HIV infection contribute to progression from latent HPV infection to cervical cancer is crucial to understanding and predicting HPV pathogenesis. Current knowledge suggests effects on the cervical immune microenvironment may be key in this process. Understanding microbes which influence this environment will help identify cervical microbiota associated with low and highgrade cervical lesions. This may allow certain cervical microbiota to be used as diagnostic markers for those at high risk of developing cervical cancer, and for the development of preventative probiotic or antibiotic treatments which could control the cervical microbiome by promoting bacterial colonization with a microbiota associated with healthy cervical cytology.
Our studies have identified a unique microbiota associated with HSIL. Data derived from of our precise sampling of cervical lesions leads us to propose that Mycoplasma contributes to a cervical microbiome status which promotes HPV-related cervical lesions. These results suggest a greater influence of the bacterial microbiota on the outcome of HPV infection than previously thought.      Figure S1: Bacterial 16s deep sequencing data was analyzed with rarefaction curves generated from the OTU data. These rarefactions were then compared with HIV status and cervical cytology. S1A. The red square represents HIV negative: "0," and the black square represents HIV positive: "1." The line indicated as "NA" is the unadjusted control. S1B. The red square represents HSIL , and the blue square represents LSIL. The green square represents NILM. The line indicated as "NA" is the unadjusted control.   M  u  n  g  o  C  ,  C  o  h  e  n  C  R  ,  M  a  l  o  b  a  M  ,  B  u  k  u  s  i  E  A  ,  H  u  c  h  k  o  M  J  .  P  r  e  v  a  l  e  n  c  e  ,  c  h  a  r  a  c  t  e  r  i  s  t  i  c  s  ,  a  n  d  o  u  t  c  o  m  e  s  o  f  H  I  V  -p  o  s  i  t  i  v  e  w  o  m  e  n  d  i  a  g  n  o  s  e  d  w  i  t  h  i  n  v  a  s  i  v  e  c  a  n  c  e  r  o  f  t  h  e  c  e  r  v  i  x  i  n  K  e  n  y  a  .  I  n  t  J  G  y  n  a  e  c  o  l  O  b  s  t  e  t  .  2  0  1  3  ;  1  2  3  (  3  )  :  2  3  1  -5  .  E  p  u  b  2  0  1