Study design
The present randomised, double-blind, parallel-group, placebo-controlled trial (RCT) was conducted at three University-affiliated fertility clinics and one private fertility clinic in Denmark. The EudraCT clinical trial identifier is 2016-002385-31; first registration day 2016-07-11. The current version of the protocol is 11, 2021-04-29. The trial protocol was published in 202019. The primary centre from which the ethical approval was accepted was Skive Regional Hospital.
Patients
Inclusion criteria were, female aged 18-42 years old, BMI <35, negative chlamydia/gonorrhoea test within 6 months of IVF treatment, normal cervical smear within 3 years of IVF treatment, written informed consent, and abnormal vaginal microbiota (AVM), according to criteria stated below with the vaginal swab being obtained less than 90 days before the randomisation day. Exclusion criteria were Hepatitis/HIV positivity, intrauterine malformations and severe concomitant disease, including inflammatory bowel disease. Patients were not allowed to take vaginal probiotics, neuromuscular blocking drugs, immunosuppressive medication, or investigational drug preparations other than the study product. Each patient could only participate once. Patients were approached when attending their first, second or third IVF stimulation cycle or frozen embryo transfer (FET) therefrom. If eligible, a vaginal swab was collected by the treating physician (N=1238) or the patient herself (N=273) using the ESwab™ system (Copan, Brescia, Italy). Previous studies found that self-collection of vaginal swabs is as valid as physician obtained swabs both considering qPCR of e.g. Gardnerella29 and sequencing30. The ESwab™ was subsequently shipped for central testing at Statens Serum Institut, Copenhagen where it was analysed for AVM according to criteria stated below and as previously reported16. AVM is a qPCR-based diagnosis of a BV-type vaginal dysbiosis, targeting a high absolute abundance of Gardnerella spp. and Fannyhessea vaginae with 93% sensitivity and 93% specificity compared to BV diagnosed by Nugent score (Gold standard)16. Patients were randomised on the first day of ovarian stimulation or during the first days of elective FET allowing for at least 12 days of study medication.
Randomisation, masking and intervention
The present RCT randomised three parallel groups 1:1:1. The first active treatment arm (CLLA) consisted of oral clindamycin 300 mg two times daily for 7 days followed by vaginal LACTIN-V (Osel Inc.). LACTIN-V is an investigational drug that contains L. crispatus CTV-05 (2×109 CFU/dose, 200 mg, delivered with pre-filled, single-use vaginal applicators) which was applied vaginally once daily after the last day of clindamycin treatment for a total of 7 consecutive days; thereafter twice weekly up to a total usage of 21 applicators or until completion of the clinical pregnancy scan at week 7–9.
The second active treatment arm (CLPL) consisted of oral clindamycin 300 mg, twice daily for 7 days followed by a LACTIN-V placebo as in the regimen described above.
Finally, the inactive treatment arm (PLPL) consisted of an identically appearing clindamycin placebo and a LACTIN-V placebo. Placebo clindamycin consisted of encapsulated mannitol. The placebo LACTIN-V formulation contained the same ingredients as LACTIN-V, without L. crispatus CTV-05.
Randomisation code and allocation concealment was performed by the pharmacy providing the study medication, using a computer-generated code. The identical medication packs were labelled with the randomisation number and received at the IVF centres from the pharmacy in blocks of 15, five of each of the three treatments, to secure equal distribution of treatment arms at the centres. The randomisation number was continuous and unique for each patient, and it was prelabelled from the pharmacy before distribution to the clinics; thus, both patients and study personnel were blinded to the intervention. The pharmacy did not play any role in or had knowledge about the IVF treatment. The first person to investigate the unblinded dataset was an external statistician at Aarhus University, Denmark who analysed the reproductive outcome. After that, data was unblinded for the principal investigators.
Outcomes
The primary outcome was clinical pregnancy rate, defined as an ultrasound proven intrauterine fetal heartbeat at gestational week 7–9. Secondary outcomes were live birth rate, biochemical pregnancy rate (hCG positive 9–11 days after embryo transfer according to local laboratory standards), implantation rate, pregnancy loss, preterm birth rate, birth weight and adverse events. Considering adverse events, we recorded all adverse events reported to the clinics from the day of randomisation to the day of embryo transfer, including an adverse event questionnaire on the day of embryo transfer. Patients without embryo transfer were approached to also fill in the questionnaire.
Compliance to medication was defined as those patients reporting to take all study medicine notwithstanding those patients who took all study medicine, but inadvertently in a wrong way. Outcomes were analysed by intention to treat (ITT), modified intention to treat (mITT) and per protocol (PP). ITT included all randomised patients except those withdrawn from the study within 24 hours from randomisation. For the mITT analysis, additionally, patients needed to fulfil in/exclusion criteria and to have embryo transfer less than 63 days from the active treatment start to cycle day 1 in the same menstrual cycle in which the embryo transfer was performed. For the PP analysis, additionally, all patients should adhere strictly to the protocol. Two authors (TH and MBJ) stratified patients for the mITT and PP analysis independently prior to breaking the randomization code.
Sample size
We estimated a 40% chance for clinical pregnancy per embryo transfer in the active treatment arm as compared to the placebo arm which was estimated to have a maximum of 20% chance of clinical pregnancy/transfer based on a previous study16. By a two-sample proportion test with a power of 80% and an alpha at 5%, the aim was to randomise 92 patients in each group. A potential difference between the two active arms was considered exploratory and consequently, this was not part of the power calculation, but we decided to include the same number of patients in the CLLA arm to investigate a potential added benefit of LACTIN-V. An interim analysis was performed, and to adjust for this, we added 10% more patients to the 92 randomised patients as suggested in Wittes et al.31. We estimated that 10% of couples would have no embryos for transfer after randomisation in fresh cycles, and we adjusted for this by adding another 10% to each randomised group, that is, 92+19=111. Interim analysis was pre-planned and conducted at the time 167 patients were randomised. At this point and under the conditions described previously19, the study board decided to continue the trial on March 12th, 2020.
Statistics
For each treatment group CLLA, CLPL and PLPL the estimated proportions, risk ratios (RR) and their confidence intervals were calculated using uni- and multivariate logistic regression analyses by generalized linear models with log-link function. The significance level for the final analysis was set at 4.9% (95.1% confidence intervals) due to the preplanned interim analysis where an alpha of 0.1% was used. The outcomes were analysed with and without adjusting for the following confounders: quality of the embryo (blastocyst/cleavage state –preimplantation genetic testing for aneuploidy (PGT-A) was not performed in this study) and female age (continuous variable) which are well-described parameters affecting pregnancy rates. It was also pre-planned to adjust for double embryo transfer and for private/public clinics, however, only five patients received double embryo transfer without achieving clinical pregnancy. Furthermore only 10 patients were included from the one participating private IVF clinic of whom only one patient had a clinical pregnancy. These numbers were not sufficient to adjust for double embryo transfer and private/public centre in the statistical model. We pre-planned to adjust for the abovementioned confounders since the primary analysis (mITT) was not performed per randomised patient but per transferred patient. The linear relation between the log of odds and age was evaluated using splines. To examine the sensitivity of the estimates, all the outcomes were further analysed under PP and ITT conditions. In Tables 1 and 3, we used Fisher’s exact test for binary variables, whereas the ANOVA was used for the continuous variables. We decided to provide a statistical test in Table 1 because all patients randomised did not necessarily have an embryo transfer and as such were not eligible for mITT analysis. Safety analysis (Table 3) was done per ITT. All these analyses were performed in STATA version 18 (StataCorp LLC).
Laboratory methods
After arrival of the vaginal eSwab specimen at the central laboratory at Statens Serum Institute (SSI), DNA from 100 µL of the vaginal screening sample was released boiling in 300 µL Chelex resin slurry as previously described32. Quantitative (q)PCRs detecting Gardnerella spp. (previously described as G. vaginalis) and Fannyhessea (F.) vaginae (previously Atopobium vaginae) were performed as previously described33. A vaginal sample was considered positive if having more than 5.7 × 107 and/or 5.7 × 106 copies/ml for Gardnerella spp. and F. vaginae, respectively. DNA extraction of vaginal samples was performed on a MagNAPure instrument (Roche Molecular Systems Inc., Pleasanton, CA, USA), using off-board enzymatic lysis by mixing 100 µL of sample with 150 µL MagNAPure bacterial lysis buffer with final concentrations of lysozyme (20mg/mL), mutanolysin (250U/mL) and lysostaphin (22U/mL) (Merck Life Sciences, Søborg, Denmark) for 60 minutes at 37°C. A total of 200µl was extracted, using the Pathogen Universal 200 MagNA Pure protocol and eluted in 100µl. The quality of the DNA extraction process was documented by simultaneous extraction of a vaginal mock community (ATCC® MSA-2007™, LGC Standards, Teddington, UK). After dissolving the bacteria according to the manufacturer’s instructions, the pellet of the vaginal mock community was re-suspended in 1 mL PBS and 10 µL 100µL ESwab® (Copan) transport medium. Negative controls for H2O were analysed on the same plates as the samples. Primers for 16S qPCR and amplification for 16S Illumina sequencing were the same as used by Fredricks et al. (Table S3) and have previously been published34.
For the sequencing, adapters and spacers added to the primer sequences, see Table S4. Amplification of the 16S Illumina sequencing, Kapa HiFi HotStart polymerase ready mix (Sigma-Aldrich) was used with the primer sequences described above, but with adaptors with heterogeneity spacers for MiSeq indexing and sequencing as described in Table S4. The indexing PCR was carried out using 2x KAPA HiFi, Nextera XT DNA Library Prep Kit v2 for indexing (384 combinations) (Illumina) and with 2 μL amplicon from the amplification PCR in a final volume of 25 μL on a 2720 Thermal Cycler using the following conditions: 3 min at 95 °C, 15 cycles of 20s at 98 °C, 15 s at 55 °C and 45 s at 72 °C, and final 5 min elongation step at 72 °C. After indexing, post-PCR cleanup was performed using a 1:1 ratio of AMPure XP beads (Beckman Coulter) following manufacturer’s instructions before quantification using AccuClear Ultra High Sensitivity dsDNA Quantification Kit (Biotium) following manufacturer’s instructions. Subsequently, samples were pooled in equimolar concentrations and quantified using the Qubit dsDNA HS assay kit (Thermo Fisher Scientific) prior to sequencing on a MiSeq with a 600-cycle MiSeq Reagent Kit v3 (Illumina) and a pool of libraries loaded at 10 pM final concentrations.
Bioinformatics pipeline
The amplicon samples were multiplexed on multiple sequencing run (with 20% phiX spike in). Nextera XT barcode incorporation and sequencing using the Illumina MiSeq platform (Illumina, Inc., San Diego, United States), with 300 bp paired-end reads, this was performed according to the manufacturer’s directions. Raw reads were demultiplexed using bcl2fastq (RRID:SCR_015058). Primers and heterogeneity spacers were trimmed using cutadapt (v. 2.3) in paired-end mode at an 8% error rate35. Subsequent sequence analysis was performed in Rstudio (v. 2022.07.0; R Core Team (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.) with R version 4.2.1. Decontamination was done using the Decontam package (v. 1.18.0) using method “either” with frequency threshold = 0.1 and prevalence threshold = 0.5. Amplicon Sequence Variants were called using the dada2 pipeline (v. 1.26.0) using the function "filterAndTrim" with following parameters: truncLen = (270, 210), maxEE = 2, truncQ = 2, maxN = 036. Community state types were subsequently computed according to the publication by France et al5.
Role of the funding source
PH, TH and JSJ received—through their institutions—an unrestricted research grant from Osel, Inc., which produces LACTIN-V. A clinical trial agreement was made ensuring full data ownership and publication rights to PH. Osel Inc. had inputs to study design but no role in data collection, data analysis, data interpretation, or writing of the present manuscript.
Data handling
Study data was collected and managed, using REDCap electronic data capture tools37 hosted at Aarhus University and monitored by the University affiliated ICH-GCP unit. The randomization code was broken April 21, 2023, when the primary outcome was monitored, and patients had been stratified to ITT, mITT or PP analysis.
Data availability statement
In 2028, five years after study completion we are obliged to deliver the deidentified clinical trial data to the Danish National Archives upon which the data is accessible for all interested parties. The metadata and statistical analysis-log can be made available to reviewers upon submission. Researchers interested in the individual participant data prior to 2028 may contact the first author to access the data under a data sharing agreement as decided by the Danish Data Protection Agency. The full study protocol with statistical analysis plan is uploaded with this manuscript, albeit it has to a great extent already been published19. Sequencing data will be uploaded to a relevant data repository with accession codes given when this manuscript has been accepted for publication.
Methods-only references
28. Wittes, J. Sample size calculations for randomized controlled trials. Epidemiol Rev 24, 39–53 (2002).
29. Jensen, J. S., Björnelius, E., Dohn, B. & Lidbrink, P. Use of TaqMan 5’ nuclease real-time PCR for quantitative detection of Mycoplasma genitalium DNA in males with and without urethritis who were attendees at a sexually transmitted disease clinic. J Clin Microbiol 42, 683–692 (2004).
30. Datcu, R. et al. Vaginal microbiome in women from Greenland assessed by microscopy and quantitative PCR. BMC infectious diseases 13, 480-2334-13–480 (2013).
31. Golob, J. L. et al. Stool Microbiota at Neutrophil Recovery Is Predictive for Severe Acute Graft vs Host Disease After Hematopoietic Cell Transplantation. Clinical Infectious Diseases: An Official Publication of the Infectious Diseases Society of America 65, 1984–1991 (2017).
32. Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal 17, 10–12 (2011).
33. Callahan, B. J. et al. DADA2: High-resolution sample inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016).
34. Harris, P. A. et al. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform 42, 377–381 (2009)