Shifts in gut and vaginal microbiomes are associated with cancer recurrence time in women with ovarian cancer

Study population

Patients, who carried a diagnosis of advanced (Stage III/IV) epithelial ovarian cancer and who were classified as primary platinum-resistant (platinum-free interval (PFI) from completion of primary platinum-based chemotherapy <6 months) or platinum super-sensitive (PFI > 24 months) and were being treated at the Stephenson Cancer Center at the University of Oklahoma Health Sciences Center, were approached to participate in this study. Patients were approached when they either developed primary platinum-resistant disease or when they were identified as platinum super-sensitive. Patients who had already been diagnosed with primary platinum-resistant disease or as platinum super-responders but had moved on to additional therapy, as well as patients on active anti-cancer therapy or in surveillance, were also included in this study. Patients were excluded if: (1) they were taking antibiotics at the time of sample collection or within 14 days prior to sample collection, or (2) they had active vaginal bleeding or known entero-vaginal fistulae. Samples were also collected from five individuals who were referred for ovarian cancer treatment at the Stephenson Cancer Center but ultimately had benign tumors; these served as a non-chemotherapy exposed control group. Sample collection periods and treatment procedures are outlined in Fig. S1. A brief summary of demographic and medical treatment history for participants in this study (n = 45, median age 62.2, age range 33–83) is provided in Table 1. This study was approved by the University of Oklahoma Health Sciences Center Institutional Review Board (February 22^nd, 2016, reference #6458) and all participants gave written informed consent for their participation in the study.

All patients were treated initially with platinum and taxane chemotherapy for a planned six to eight cycles. These regimens included paclitaxel and carboplatin given every 21 days, paclitaxel given weekly with every 21^st day carboplatin, or intraperitoneal administration of either cisplatin or carboplatin with intravenous and intraperitoneal paclitaxel. In patients with platinum resistant disease, standard of care options after recurrence included pegylated liposomal doxorubicin (PLD), weekly paclitaxel, gemcitabine, topotecan or bevacizumab given as monotherapy or in combination therapy. Patients were also screened for eligibility for clinical trials. Patients with a PFI > 24 months had not recurred at the time of study participation and were followed every 6 months with surveillance of Ca-125 values and exams. For those who had recurred beyond 24 months, treatment options included several platinum-based doublets including carboplatin and PLD given every 28 days, carboplatin and paclitaxel given every 21 days or carboplatin and gemcitabine given on a day one and day eight or day one and day 15 schedule. Each patient completed a minimum of six cycles of treatment and could undergo more cycles as long as the patient was responding and tolerating therapy.

Sample collection

Samples were collected during standard of care exams in the gynecologic oncology clinic at the Stephenson Cancer Center in Oklahoma City, OK. Catch-All Sample Collection Swabs (Epicentre) were used to collect vaginal and fecal samples. Vaginal swabs were collected from three sites per individual: vaginal introitus (VIT), mid-vagina (MDV), and posterior fornix (VPF), and then placed into a dry sample collection tube. Fecal samples were collected via a rectal digital exam, after which any stool collected was placed on a Catch-All swab and placed in a dry collection tube. Two swabs were collected from each site (bilaterally from the vaginal sites and sequentially for the rectal samples). Each participant completed a quality-of-life survey regarding their medical treatment history, antibiotic use within the past year, vitamin consumption, socioeconomic status, and other lifestyle metadata (Table S1).

Laboratory methods

DNA was extracted from the left-side vaginal swab and first fecal swab from each patient, using the MoBio PowerSoil DNA Isolation Kit (now Qiagen DNeasy PowerSoil Kit), following manufacturer’s protocols with the addition of a ten-minute incubation at 65 °C prior to the initial bead-beating step, as recommended in the Manual of Protocols for the Human Microbiome Project (McInnes & Cutting, 2010). A quantitative PCR (qPCR), using the SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, USA) and primers targeting the V4 region of the bacterial 16S rRNA gene (Caporaso et al., 2011), was conducted; dilutions of Escherichia coli DNA corresponding to known 16S rRNA gene copy numbers were used as quantification standards for the DNA extracts. DNA extracts were amplified in triplicate, using Phusion High-Fidelity DNA polymerase (ThermoFisher Scientific, Waltham, MA, USA) and Illumina-compatible primers 515F and 806R (targeting the V4 region of the16S rRNA gene) with error-correcting Golay barcodes incorporated into the 806R reverse primer (Caporaso et al., 2011). PCR products were pooled in equimolar concentrations, purified with the MinElute PCR purification kit (Qiagen, Valencia, CA, USA), then size-selected between 300 and 450 basepairs using a PippenPrep, quantified using KAPA Biosystems Illumina library quantification kit, and sequenced across multiple runs of an Illumina MiSeq (500 cycles paired-end sequencing, v2 reagent kit).

Bioinformatic methods

AdapterRemoval (v2) (Schubert, Lindgreen & Orlando, 2016) was used to filter out reads with uncalled bases, reads with Phred quality threshold <30, and reads less than 150 bp in length. Quality filtered paired-end reads were merged using AdapterRemoval (v2) (Schubert, Lindgreen & Orlando, 2016) and then demultiplexed with QIIME (v1.9), followed by removal of chimeric sequences and low-abundance (<5 total sequences) reads (Caporaso et al., 2010). The remaining sequences were used for de novo Operational Taxonomic Unit (OTU) clustering with USEARCH (v10) at 97% sequence similarity (Edgar, 2010). Taxonomy was assigned to each OTU representative using the EzBioCloud 16S rRNA gene database (Yoon et al., 2017). The resulting OTU table was rarefied to 9000 reads and downstream analysis was performed in QIIME (v1.9) (Caporaso et al., 2010). The post-rarefaction sample breakdown was: PFI > 24 (n = 23), PFI < 6 (n = 17), and benign (n = 5). Further details of the bioinformatic methods are given in the Supplemental Material.

Statistical methods

Phylogenetic diversity and weighted/unweighted UniFrac (Lozupone & Knight, 2005) metrics were generated in QIIME (v1.9) with FastTree2 (Price, Dehal & Arkin, 2010). Tests for significance between study groups for alpha and beta diversity were performed using Kruskal-Wallis and PERMANOVA tests, respectively, in R (Team RC, 2013). Vaginal samples were classified into clusters by the dominant bacterial taxon found in each sample, as determined by Ward hierarchical clustering (Ward, 1963), and visualized as a heatmap using the gplots package in R (Warnes et al., 2016). If there were no dominant bacteria, the sample was classified as diverse. Median-unbiased estimated odds ratios were calculated to determine whether study groups had significantly different odds of dominant bacteria; reported odds ratios and 95% confidence intervals were log-transformed. Kruskal–Wallis tests with a Benjamini and Hochberg false-discovery rate adjustment were used to evaluate differential abundance of individual taxa between study groups. Odds ratios were calculated using epitools (Aragon, 2012) in R. Plots were generated using the ggplot2 (Wickham, 2016) and ColorBrewer (Harrower & Brewer, 2003) packages in R.

Vaginal microbiome

Samples from the different vaginal sites (VIT, MDV, VIT) that originated from the same individual showed similar taxonomic beta-diversity profiles (Figs. S2A–S2C). Sequencing failed for at least one of the three vaginal sites of eleven individuals (Supplemental Material), which presented difficulty in analyzing each vaginal site individually as the failed specimens were not limited to only one vaginal site. Due to the similarity in composition between vaginal sites in each individual and the relatively high number of samples which failed sequencing, we concatenated sequencing reads from each individual’s three vaginal samples into a single representative vaginal microbiome sample per individual, and then performed downstream analysis with this single representative sample, unless otherwise noted. Combining vaginal sites for each individual allowed us to retain a sample size of 45. Firmicutes was the most dominant phylum in the vaginal microbiome and it was found at over 50% relative abundance in 40% of individuals (Fig. S3A), while Proteobacteria, Bacteroidetes, and Actinobacteria were the next most abundant phyla and found at >50% relative abundance in the vaginal microbiome of 13.3%, 13.3%, and 6.7% of individuals, respectively. At the genus level, Lactobacillus, Prevotella, Escherichia, Gardnerella, and Streptococcus were the most dominant bacteria and accounted for 57.2% of all reads (Fig. S3B).

Individuals were grouped into the following study groups as outlined in the methods section (Fig. S1): PFI < 6 Months (n = 17), PFI > 24 Months (n = 23), and benign (n = 5). Table 1 presents summary statistics for these study groups. Iterative cytoreductive surgery (iCRS) was more common in PFI < 6 months while primary cytoreductive surgery (pCRS) was more common in PFI > 24 months individuals (log Odds Ratio (OR) = 3.02, 95% CI [1.37–4.68], p-value = 0.0003). As expected, PFI < 6 months individuals underwent platinum chemotherapy treatment more recently than PFI > 24 months individuals (p-value = 4.32 × 10⁻⁷) but no other metadata variable was significantly associated with PFI status (Table S2).

Vaginal microbiome communities were clustered into five community-dominance groups using Ward hierarchical clustering: Lactobacillus cluster, Escherichia cluster, Gardnerella cluster, Prevotella cluster, and a high diversity cluster (Fig. 1). Results below are presented demonstrating how each vaginal community (excluding Gardnerella due to small cluster size) relate to clinical variables, p-values for each vaginal community and clinical variable are given in Table S3.

Dominance groups were evenly distributed between patients with PFI < 6 months, PFI > 24 months, and benign, with the exception of higher-than-expected dominance of Escherichia in patients with PFI < 6 months; five of the six Escherichia-dominated vaginal communities identified with hierarchical clustering belonged to patients with PFI < 6 months (Fig. 1). Vaginal microbiomes dominated by Escherichia had higher odds of occurring in PFI < 6 months compared to PFI > 24 months (log OR = 2.812, 95% CI [0.267–5.62], p-value = 0.024, Fig. 2A). Additionally, one of the patients with PFI < 6 months and one of the patients with benign pathology identified with a ‘diverse’ vaginal microbiome had Escherichia at greater than 20% relative abundance, while no other patients with PFI > 24 months had Escherichia relative abundance above 5% (Fig. 1). In total, 35.3% of the patients with PFI < 6 months had Escherichia at greater than 20% relative abundance in the vaginal microbiome, compared to 4.34% of PFI > 24 months and only one of five benign individuals. Even though iterative cytoreductive surgery and Escherichia are more common in PFI < 6 months individuals, we found no significant relationship between iCRS and Escherichia abundance (Table S3, Fig. 2A, p-value = 0.292). Moreover, outside of the relationship between Escherichia and platinum resistance, Escherichia showed no significant association with any of the other health or lifestyle factors we examined (Table S3, Fig. 2A). Although Escherichia is a common lab-grown bacterium and found in feces, our analysis demonstrates Escherichia abundance in the vaginal samples is biological and not a technical artifact (Supplemental Material).

Approximately 24% (11 of 45) of patients in this study had Lactobacillus-dominated communities, which is significantly lower as compared to studies of similarly aged women without ovarian cancer (p-value = 0. 037) (Brotman et al., 2014a; Nené et al., 2019). Other studies have found Lactobacillus to be less abundant in Black and Hispanic women (Ravel et al., 2011). and our study consisted of 39 women who self-reported ethnicity as white, two self-reported as Black, and four self-reported as Native American (Table 1). Each of the Black and Native American women had a non-Lactobacillus dominated vaginal microbiome (Table S1) but ethnicity was not a statistically significant determinant of Lactobacillus-dominance (Fig 2B, log OR = −1.63, 95% CI [−4.62 to 1.30], p-value = 0.27). High microbial cell density, as gauged through qPCR with a standard curve generated from controls with known cell density, was positively correlated with vaginal Lactobacillus-dominance, although somewhat weakly (R² = 0.278, Fig. S4). Consumption of antibiotics within the past month was associated with a lack of Lactobacillus-dominance (Fig. 2B); however, this relationship was not significant (log OR = −2.12, 95% CI [−5.54 to 0.84], p-value = 0.0515). Additionally, Lactobacillus abundance was uncommon in individuals likely to be estrogen negative (post-menopause and not on hormone replacement therapy) but this result was also not significant (log OR = −1.32, 95% CI [−2.83 to 0.33], p-value = 0.0744). Lactobacillus-dominance did not have a strong relationship with any of the other health or lifestyle factors we tested, including PFI length (Table S3, Fig. 2B). Only 20% (1 of 5) of patients with benign pathology had a Lactobacillus-dominated microbiome but small sample size prohibits statistical inference.

Previous studies have indicated that different Lactobacillus species in the vaginal microbiome may have different roles and differential influence on host biology (Lamont et al., 2011; Arumugam et al., 2011; Schloissnig et al., 2013). While we identified L. iners, L. brevis, L. mucosae, L. reuteri, L. zeae, and L. delbrueckii in the vaginal microbiome, 99% of the Lactobacillus reads in our study were either unclassified at the species level or mapped to L. iners (Table S4); ultimately there was not sufficient data to assess the relationship between most of these Lactobacillus species and clinical variables. Nevertheless, in individuals with high Lactobacillus abundance (n = 11), we found L. iners at significantly higher abundance in patients with either no gross residual disease or residual disease < 1 cm (n = 7), compared to patients with residual disease > 1 cm (n = 4, p-value = 0.0359, Fig. 3).

We did not observe significant associations between either Prevotella and clinical variables (Table S3, Fig. 2C) or the high diversity community and clinical variables (Table S3, Fig. 2D). However, Prevotella was less common in those with residual disease > 1 cm and those with a history of hormonal disease (Fig. 2C), while a highly diverse vaginal microbiome was more common in women over 60 years old and in patients with a history of hormonal disorders, such as thyroid disease (Fig. 2D); but once again, none of these associations were statistically significant (Table S3, p-value > 0.05).

Gut microbiome

The gut microbiome was colonized by typical members of the gut microbiome at the phylum (Bacteroidetes, Firmicutes, Proteobacteria) and genus (Bacteroides, Akkermansia, Faecalibacterium, Ruminococcus, and Prevotella) levels (Fig. S5A–S5B). The PFI > 24 months and PFI < 6 months groups were not significantly different with respect to fecal unweighted and weighted UniFrac beta diversity distances as tested through PERMANOVA (Fig. 4, Fig. S6A–S6C); however, similar to the vaginal microbiome samples, there was a small subset of individuals (n = 9) with a unique microbiome signature. Patients with PFI < 6 months individuals had higher odds (log OR = 1.85, 95% CI [−0.85 to 4.497], p-value = 0.12) of being in this unique/outlier fecal microbiome group. These fecal outliers have significantly lower phylogenetic diversity compared to the other fecal samples (p-value = 0.001, Fig. 5A) and have increased abundance of genera belonging to the order Clostridiales (Lachnospira (Kruskal–Wallis p-value = 0.00039), unidentified Ruminococceae genus (Kruskal–Wallis p-value = 0.001337), and Subdoligranulum (Kruskal–Wallis p-value = 0.01121)) (Fig. S7). Of the six platinum-resistant patients in this outlier subset, two also had Escherichia-dominated vaginal communities. Eight of the nine patients in this subgroup reported consuming antibiotics within the past 6 months but this was not a significantly greater proportion of individuals with recent antibiotic consumption compared to individuals outside this fecal outlier subgroup (log OR = 0.35, 95% CI [−1.44 to 3.05], p-value = 0.48).

Overall, patients with platinum-resistant disease had lower fecal phylogenetic diversity compared to patients with platinum-sensitive disease, but the difference was not significant (Fig. 5B, p-value = 0.18). Regardless of PFI, patients with ovarian cancer had significantly higher relative abundance of Prevotella in the gut microbiome compared to benign individuals (p-value = 0.028, Fig. S8). Outside of the above-mentioned associations, there were no other significant associations between the gut microbiome and platinum sensitivity or other health/lifestyle variables.