Results
Thirty-two women were enrolled in the study, nine of whom underwent
salpingectomy due to hydrosalpinx (“cases”), and 23 due to other
indications (“controls”). One patient in the “case” group was later
excluded from the analysis due to low material yield from all samples
taken, as described below, leaving 8 patients in the case group. Table 1
describes the baseline characteristics of the study participants. The
mean age of patients was 41, with women in the case group being slightly
younger than controls (40 vs. 41.5). Of cases, 88% were premenopausal,
vs. 91% of controls. Two ethnic groups were represented in this study,
Jewish and Arabs, with no differences in their distribution between
groups (25% Arabs in the cases and 26% in controls). Indications for
surgery in the case group were either infection (4/8) or infertility
(4/8), and in the control group indications varied, the most common
being sterilization (11/23) and leiomyoma (8/23). All of the women in
the control group, and 75% of women in the case group underwent
bilateral salpingectomy, and samples were taken from both fallopian
tubes.
Ninety-four samples from 32 participants were available for sequencing,
producing a total of 2,310,035 raw reads (Supplementary Table 1). After
QC and in silico decontamination 872,742 were available for analysis.
After decontamination as described in the methods, 31/32 vaginal (V)
samples (8/9 cases, and 23/23 controls) were available for further
analysis, while only 10/62 fallopian tube (FT) samples belonging to 8/32
women (1/9 cases and 7/23 controls) produced sufficient genetic material
for the final analysis (Supplementary Table 1). The total read count of
the samples included in the final analysis was 787,362, with an average
read count of 22,847 for V samples and an average read count of 7909 for
FT samples (p<0.001). The average read count for V samples of
controls was 24,016, while the average read count for V sample of cases
was 19,485 (p=0.064). Since read counts are a surrogate marker for
bacterial load, we further analyzed the average read count of
case-patients who received antibiotic treatment prior to sample
collection and those who didn’t (24,005 vs. 14964, p=0.205).To account
for the possible relation between vaginal read count and positive FT
samples we compared average vaginal read counts of patients with
positive and negative FT samples, and found no difference (25, 091 vs.
23,569 respectively p= 0.079).
For analysis of OTUs and diversity measures, rarefaction was performed
to a depth of 4,000 sequences/sample as described in the methods,
producing 145 total observed OTUs, ranging from 1 to 76 OTUs per sample,
with an average of 8.8 OTUs for V samples and 28.2 OTUs for FT samples
(p<0.001). Looking into the differences between vaginal
samples of cases and controls, we examined the alpha diversity at the
genus level, as measured using the Shannon index, and while the
difference nearly reached statistical significance (p=0.0641, Figure
1a), the difference was further accentuated by comparing only vaginal
case samples taken from patients who were operated due to infection, to
control samples (none of the latter were operated due to infection)
(p=0.0243). These results remaned statistically significant when
analysed at the phylum level. As most FT samples available after
decontamination and rarefaction were from controls, we analyzed the
differences in measures of diversity between matched vaginal samples of
control patients to samples taken only from FT of the same control
patients. Figure 1 illustrates the significant differences in both alpha
(measured using the Shannon index, p=0.003, Figure 1b) and beta
diversity (measured using Bray–Curtis dissimilarity index, p=0.004,
Figure 1c) between both groups.
At the phylum level, both V and FT samples were composed mostly of
Firmicutes, (83.1% in vaginal samples vs. 65.4% in FT samples
p=0.073), though the relative abundances of Proteobacteria and (18.3%
vs. 2.16% respectively p=0.011) and Cyanobacteria (0% in V, 2.73% in
FT p=0.011) in FT samples were significantly higher, as shown in Figure
2.
Vaginal samples of both cases and controls contained mostly Firmicutes.
While no significant difference was found in relative abundances of
different genera, using linear discriminant analysis effect size (LEfSe)
to account for differences in specific bacteria associated with vaginal
dysbiosis, we found bacteria such as Prevotella ,Gardnerella, and Atopobium , to be more prevalent in case
patients (p=0.015, p=0.013, p=0.024 respectivly), whereas Lactobacilli
were more dominant in controls (p=0.005), as shown in Figure 3.