INTERACTIONS BETWEEN THE CERVIX AND VAGINAL MICROBIOME IN PREGNANCY

Lindsay Kindinger

A thesis submitted to the University of London for the degree of Doctor of Philosophy July 2016

Institute of Reproductive and Developmental Biology Department of Surgery and Cancer Imperial College London

Declaration

Statement of originality All work presented in this thesis was performed by myself unless otherwise stated.

Copyright Declaration The copyright of this thesis rests with the author and is made available under a Creative Commons Attribution Non-Commercial No Derivatives licence. Researchers are free to copy, distribute or transmit the thesis on the condition that they attribute it, that they do not use it for commercial purposes and that they do not alter, transform or build upon it. For any reuse or redistribution, researchers must make clear to others the licence terms of this work

Interactions between the cervix and vaginal microbiome in pregnancy

Table of contents

INTERACTIONS BETWEEN THE CERVIX AND VAGINAL MICROBIOME IN PREGNANCY ...... 1

DECLARATION ...... 2

TABLE OF CONTENTS ...... 3

ACKNOWLEDGEMENTS ...... 8

PUBLICATIONS AND PRESENTATIONS ...... 9

ABBREVIATIONS ...... 13

THESIS ABSTRACT ...... 16

1 INTRODUCTION ...... 18

PRETERM BIRTH ...... 19 1.1 Incidence a global issue ...... 19 1.2 Neonatal morbidity and mortality ...... 19

PHYSIOLOGY OF PARTURITION ...... 21 1.3 Normal term parturition ...... 21 1.4 Oestrogen and progesterone ...... 22 1.5 Cervical remodelling ...... 23

ADAPTIVE AND INNATE IMMUNITY IN PREGNANCY ...... 25

PATHOPHYSIOLOGY OF PRETERM BIRTH ...... 26 1.6 Infection and inflammation ...... 27 1.7 Structural factors ...... 29 1.8 Disorders of placentation ...... 31

PREDICTION OF SPONTANEOUS PRETERM BIRTH ...... 32 1.9 Fetal fibronectin ...... 32 1.10 Transvaginal ultrasound of the cervix ...... 33

PREVENTION OF PRETERM BIRTH ...... 37 1.11 Cervical cerclage ...... 37 1.12 Progesterone supplementation ...... 40 1.13 Tocolytics ...... 41 1.14 Antibiotics ...... 41

THE VAGINAL MICROBIOME ...... 43 1.15 Vaginal microbiota and reproductive health ...... 43

3 Interactions between the cervix and vaginal microbiome in pregnancy

1.16 Pregnancy and vaginal microbiota ...... 45 1.17 Vaginal microbiota and the neonate ...... 46

JUSTIFICATION FOR PROJECT ...... 47 1.18 Project aims and hypotheses ...... 47

2 MATERIALS AND METHODS ...... 49

RETROSPECTIVE STUDY ...... 50 2.1 Study design ...... 50

PROSPECTIVE STUDIES ...... 51 2.2 Study design ...... 51 2.3 Eligibility criteria ...... 52 2.4 Research timeline ...... 52 2.5 Ethical approval ...... 52

PROCEDURES ...... 53 2.6 Preventative interventions for preterm birth ...... 54

CERVICAL ASSESSMENT ...... 56 2.7 Virtual Organ Computer-aided AnaLysis (VOCAL) technology ...... 57 2.8 Statistical analyses ...... 58

VAGINAL MICROBIOTA: 16S RIBOSOMAL RNA GENE SEQUENCING ...... 59 2.9 Materials ...... 59 2.10 Isolation of DNA from vaginal samples ...... 61 2.11 DNA sequencing ...... 62 2.12 Statistical analyses ...... 63

QUANTITATIVE PCR ...... 65 2.13 Materials ...... 65

ASSESSMENT OF CYTOKINES IN CERVICO-VAGINAL SECRETIONS ...... 66 2.14 Magnetic Luminex Screen Assay for Cytokines ...... 66 2.15 Analyte extraction ...... 69 2.16 Statistical analyses ...... 72

3 ULTRASOUND ASSESSMENT OF THE CERVIX IN PREGNANCY ...... 73

CHAPTER ABSTRACT ...... 74

INTRODUCTION ...... 75

4 Interactions between the cervix and vaginal microbiome in pregnancy

STUDY DESIGN...... 78

RESULTS ...... 81 3.1 Recruitment ...... 81 3.2 Participant demographics ...... 82 3.3 Correlation of cervical length, volume and vascularity ...... 83 3.4 Cervical length and volume ...... 86 3.5 Cervical vascularity...... 88 3.6 Parity and cervical morphology ...... 89 3.7 A comparison of high and low-risk pregnancy ...... 90 3.8 Gestation at birth ...... 92 3.9 Prediction of preterm birth ...... 94

DISCUSSION ...... 97

4 VAGINAL MICROBIOTA AND THE CERVIX IN PREGNANCY ...... 100

CHAPTER ABSTRACT ...... 101

INTRODUCTION ...... 102

STUDY DESIGN...... 104

RESULTS ...... 106 4.1 Recruitment ...... 106 4.2 Participant demographics ...... 107 4.3 Community state type classification ...... 110 4.4 Ethnicity...... 112 4.5 Advancing gestational age ...... 114 4.6 Correlation between cervical assessment and vaginal microbiota ...... 117 4.7 Vaginal microbiota and a short cervix ...... 119 4.8 Comparisons of term and preterm birth outcomes...... 125 4.9 Cervical length, the vaginal microbiota and gestation at birth ...... 134 4.10 The prediction of preterm birth ...... 136

DISCUSSION ...... 141

5 A COMPARISON OF PREVIOUS PRETERM BIRTH AND CERVICAL TREATMENT IN PREGNANCY .... 145

CHAPTER ABSTRACT ...... 146

INTRODUCTION ...... 147

5 Interactions between the cervix and vaginal microbiome in pregnancy

STUDY DESIGN...... 150

RESULTS ...... 153 5.1 Cervical assessment by transvaginal ultrasound ...... 153 5.2 The vaginal microbiota in cervical treatment and prior preterm birth groups ...... 165 5.3 Associations between vaginal microbiota and cervical parameters ...... 183 5.4 Preterm birth prediction by vaginal microbiota and cervical length ...... 201

DISCUSSION ...... 202

6 CERVICAL CERCLAGE, VAGINAL MICROBIOTA AND PRETERM BIRTH PREVENTION ...... 206

CHAPTER ABSTRACT ...... 207

INTRODUCTION ...... 208

STUDY DESIGN...... 211 6.1 Retrospective study ...... 211 6.2 Prospective study ...... 212

RETROSPECTIVE STUDY RESULTS ...... 215 6.3 Results of a ten year audit...... 215

PROSPECTIVE STUDY RESULTS ...... 221 6.4 Participant demographics ...... 221 6.5 Cervical assessment ...... 223 6.6 Impact of suture material on vaginal microbial structure ...... 225 6.7 Hierarchical clustering of bacterial class, genera and species ...... 226 6.8 Expression of inflammatory cytokines ...... 245 6.9 Correlation between cervical vascularity, vaginal microbiota and cytokine expression...... 254

DISCUSSION ...... 257

7 PROGESTERONE SUPPLEMENTATION AND VAGINAL MICROBIOTA ...... 260

CHAPTER ABSTRACT ...... 261

INTRODUCTION ...... 262

STUDY DESIGN...... 264

RESULTS ...... 266 7.1 Demographics ...... 266 7.2 Vaginal microbiota and progesterone ...... 268 7.3 Longitudinal microbial profiles ...... 273

6 Interactions between the cervix and vaginal microbiome in pregnancy

7.4 Vaginal microbiota and gestation at birth ...... 277 7.5 Progesterone and gestation at birth ...... 283

DISCUSSION ...... 286

8 SUMMARY DISCUSSION ...... 288

SUMMARY OF THE CLINICAL PROBLEM ...... 289

CLINICAL RELEVANCE AND FUTURE WORK ...... 290 8.1 Preterm birth surveillance in high-risk pregnancy ...... 290 8.2 Preterm birth risk following excisional cervical treatment ...... 292 8.3 Investigation and treatment of vaginal infection in pregnancy ...... 293 8.4 Progesterone versus cerclage for preterm birth prevention ...... 293

STRENGTHS AND LIMITATIONS ...... 295

FINAL CONCLUSIONS ...... 298

9 REFERENCES ...... 299

10 APPENDIX ...... 316 10.1 Twins individual patient meta-analysis ...... 316 10.2 Re-sequencing of 16S rRNA gene sequences ...... 320

11 PEER REVIEWED PUPLICATIONS...... 324

7 Interactions between the cervix and vaginal microbiome in pregnancy

Acknowledgements

Firstly I would like to extend thanks to my supervisors Dr David MacIntyre, Professor TG Teoh, Dr Julian Marchesi and Professor Phillip Bennett who have all supported and contributed to this work immensely over the 4 years of its research and writing. Professor Teoh, thank you for being a fantastic mentor, and providing the most incredible introduction into the world of research. Your support, guidance and generosity has been unparalleled through my career as a trainee in Obstetrics and Gynaecology. Professor Bennett, thank you for your clarity of ideas. You have offered invaluable advice on matters, both scientific and clinical and I have benefited hugely from your wide experience. It has been an honour to work with you. And to Dr David MacIntyre - my boss, friend and fellow migrant from the colonies. I am eternally grateful for all the countless hours you have spent patiently teaching me, tolerating my ignorance, painstakingly critiquing my work, making me laugh and being an outstanding supervisor. Thank you for your encouragement, guidance and demanding of high standards – you have continually pushed and motivated me to strive for better.

Thank you to Yooni for showing me the lab ropes - you have been an invaluable designer of protocols, teacher and friend. Kamsiah and Edna, you were so kind and accommodating to me in the Fetal Medicine Unit, making the juggling of hundreds of prem and research clinics infinitely easier. Rachael, Larissa and Ramona, your support with sample collection and storage helped enormously in my reaching of recruitment targets. Thank you to my third floor friends Jo, Anita, Kim, Roberta, Holly, Ana, Julia, Stefano and Richard – nothing quite like PhD student camaraderie.

Thank you to my wonderful in-laws Thelma and Roger, the O’Briens and Hamblings, who have been incredibly understanding of my infrequent visits and my seemingly endless need to work rather than play.

A big thank you is reserved for my parents Carl and Dorcas, and my brother Matt. You have been so supportive and patient, despite the added demands of organising two weddings, Mum’s golf captaincy (as stressful as a PhD) and having us to stay for a protracted 2 years of house purchasing/renovating. Mum and Dad you are so kind, supportive, enthusiastic and loving. I can honestly say thanks to you, the whole experience has been bearable! I’d also like to thank my Granny Jean, who through her demonstrations of strength and determination, has provided a huge inspiration to me.

Finally, to my extraordinary husband Martyn, who has been my rock and source of strength over the past 4 years. When I felt I was struggling, your positivity, encouraging words, delicious meals, great company (and incredible diagrams) kept me afloat. I am looking forward to spending the rest of our (post-PhD) lives together.

8 Interactions between the cervix and vaginal microbiome in pregnancy

Publications and presentations

PEER REVIEWED ARTICLES

Kindinger LM, Bennett PR, Lee YS, Marchesi JR, Smith A, Holmes E, Nicholson JK, Teoh TG and MacIntyre DA. The interaction between vaginal microbiota, cervical length and vaginal progesterone treatment for preterm birth risk. Microbiome. 2017 Jan 19;5 (1):6. doi: 10.1186/s40168-016-0223-9.

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Smith A, McDonald J, Holmes E, Nicholson JK, Teoh TG and Bennett PR. Relationship between vaginal microbial dysbiosis, inflammation, and pregnancy outcomes in cervical cerclage. Science Translational Medicine 8 (350), 2016 2016 Aug 3;8(350): doi: 10.1126/scitranslmed.aag1026.

Kindinger LM, Kyrgiou M, MacIntyre DA, Cook J, Yulia A, Terzidou V, Teoh TG and Bennett PR. Preterm birth prevention post-conisation: a model of cervical length screening with targeted cerclage. PLoS ONE 11(11): e0163793. doi:10.1371/journal. pone.0163793

Kindinger LM, Poon LC, Cacciatore S, MacIntyre DA, Fox NS, Schuit E, Mol BW, Liem S, Lim AC, Serra V, Perales A, Hermans F, Darzi A, Bennett PR, Nicolaides KH and Teoh TG. The effect of gestational age and cervical length measurements in the prediction of spontaneous preterm birth in twin pregnancies: an individual patient level meta-analysis. BJOG: An International Journal of Obstetrics and Gynaecology, May 2016; Vol.123 (6):877-84.

Mitra A, Kindinger LM, Kalliala I, Smith JR, Paraskevaidis E, Bennett PR and Kyrgiou M. Obstetric complications following conservative treatment of Cervical Intraepithelial Neoplasia. British Journal of Hospital Medicine, August 2016, Vol 77, No 8

Cook J, Terzidou V, Chandiramani M, Kindinger LM, Sykes, L Shennan A and Bennett PR. Cerclage position, cervical length and preterm delivery in women undergoing ultrasound indicated cervical cerclage: A retrospective cohort study. PlosOne, accepted June 2016.

MacIntyre DM, Chandiramani M, Lee YS, Kindinger LM, Smith A, Angelopoulos N, Lehne B, Arulkumaran S, Brown R, Teoh TG, Holmes E, Nicoholson JK, Marchesi JR and Bennett PR. The vaginal microbiome during pregnancy and the postpartum period in a European population. Scientific Reports, March 2015, 11:5: 8988.

Articles in preparation

Kindinger LM, MacIntyre DA, Lee YS, Mitra A, Kyrgiou M, Marchesi JR, Smith A, Holmes E, Nicholson JK, Teoh TG and Bennett PR. The vaginal microbiome, cervical length and specific etiologies of preterm birth. In preparation.

9 Interactions between the cervix and vaginal microbiome in pregnancy

ORAL PRESENTATIONS

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Holmes E, Nicholson JK, Teoh TG, and Bennett PR. The vaginal microbiome, cervical length and preterm birth. International Fetal Medicine Congress, Majorca, Spain, June 2016.

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Terzidou V, Cook JR, Lees C, Toozs-Hobson P, Slack M, Holmes E, Nicholson JK, Teoh TG and Bennett PR. Cervical cerclage using braided suture induces vaginal dysbiosis, inflammation and is associated with increased rates of preterm birth. British Maternal Fetal Medicine Conference, Birmingham, April 2016 – Awarded 1st prize for best oral presentation

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Smith A, Holmes E, Nicholson JK, Teoh TG and Bennett PR. Identification of vaginal microbial communities associated with specific etiologies of preterm birth. Society for Reproductive Investigation, Montreal, Canada, March 2016.

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Smith A, Holmes E, Nicholson JK, Teoh TG and Bennett PR. Understanding the interaction between the cervix and the vaginal microbiome underlying specific aetiologies of preterm birth. Blair Bell Annual Academic Meeting, London, March 2016 – Awarded 1st prize for best platform presentation

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Teoh TG and Bennett PR. Cervical cerclage using braided suture induces vaginal dysbiosis, inflammation and is associated with increased rates of preterm birth. Blair Bell Annual Academic Meeting, London, March 2016.

Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Teoh TG and Bennett PR. The suture material used at cervical cerclage impacts on the vaginal microbiome in pregnancies at high-risk of preterm birth. Harris- Wellbeing UK Preterm Birth Research Conference, Liverpool, September 2015.

Kindinger LM, Kyrgiou M, MacIntyre DA, Cook J, Yulia A, Terzidou V, Teoh TG and Bennett PR. Preterm birth prevention following excisional cervical treatment for CIN: a model of cervical length screening with targeted cervical cerclage. Harris-Wellbeing UK Preterm Birth Research Conference. Liverpool, September 2015.

Kindinger LM, Poon LC, Cacciatore S, MacIntyre DA, Fox NS, Schuit E, Mol B, Liem S, Lim AC, Serra V, Perales A, Hermans F, Darzi A, Bennett P, Nicolaides KH and Teoh TG. Prediction of preterm birth by cervical length in twins: an individual patient level metaanalysis. International Fetal Medicine Congress, Nice, France, June 2014

10 Interactions between the cervix and vaginal microbiome in pregnancy

POSTER PRESENTATIONS

Kindinger LM, Lee YS, Marchesi JR, Teoh TG, Bennett PR and MacIntyre DA. Progesterone does not adversely affect the vaginal microbiome in high-risk pregnancy with a short cervix. Arch. Dis. Child. Fetal Neonatal Ed. June 2016; 101 (Suppl 1): 1-172. British Maternal and Fetal Medicine Conference, Birmingham, April 2016.

Kindinger LM, MacIntyre DA, Lee YS, Teoh TG and Bennett PR. Identification of Vaginal Microbial Communities Associated with Specific Etiologies of Preterm Birth. BJOG. 2016 Vol. 123, Suppl 1. Blair Bell Annual Academic Meeting, London, March 2016.

Kindinger LM, Lee YS, Marchesi JR, Teoh TG, Bennett PR and MacIntyre DA. Progesterone does not adversely affect the vaginal microbiome in high-risk pregnancy with a short cervix. Reproductive Sciences, Mar 2016, Vol.23, Suppl 1, p. 51A-344A. Society for Reproductive Investigation, Montreal, March 2016.

Kindinger LM, MacIntyre DA, Yun YS, Marchesi JR, Teoh TG and Bennett PR. The impact of cervical cerclage material on the vaginal microbiome in pregnancies at high-risk of preterm birth. The European molecular biology Conference on the Human Microbiome, Heidelberg, June 2015

Kindinger LM, MacIntyre DA, Teoh TG and Bennett PR. The cerclage and cervical vascularity: a prospective longitudinal study in pregnancy. Arch. Dis. Child. Fetal Neonatal Ed, June 2015; 100. British Maternal and Fetal Medicine, April 2015.

Kindinger LM, MacIntyre DA, Terzidou V, Teoh TG and Bennett PR. Change in cervical length for the prediction of preterm birth with pre-pregnancy excisional cervical treatment. Arch. Dis. Child. Fetal Neonatal Ed. 2015 Jun; 100. British Maternal and Fetal Medicine, April 2015.

Kindinger LM, MacIntyre DA, Terzidou V, Teoh TG and Bennett PR. Prediction of preterm birth with pre- pregnancy excisional cervical treatment. Reproductive Sciences, March 2015, Vol.22, Suppl 1, p.152A- 152A. Society for Gynecologic Investigation, San Francisco, March 2015.

Kindinger LM, MacIntyre DA, Teoh TG and Bennett PR. Cervical volume: a longitudinal observational study in pregnancy. Society for Gynecologic Investigation, San Francisco, March 2015.

Kindinger LM, MacIntyre DA, Teoh TG and Bennett PR. Cervical volume: a longitudinal observational study in pregnancy. Annual Academic Meeting in Obstetrics and Gynaecology, January 2015.

Kindinger LM, MacIntyre DA, Teoh TG and Bennett PR. Change in cervical length for the prediction of preterm birth in pre-pregnancy excisional cervical treatment. Annual Academic Meeting in Obstetrics and Gynaecology, January 2015.

Kindinger LM, Ashrafian H, Poon L, Nathan F, Nicolaides K, Darzi A, Teoh TG and Bennett PR. Prediction of preterm delivery with cervical length in twin pregnancy: A meta-analysis and systematic review. Reproductive Sciences, March 2014, 21 (3 SUPPL.1) (pp 256A), 2014. Society for Gynecologic Investigation, Florence, March 2014.

11 Interactions between the cervix and vaginal microbiome in pregnancy

Lewis H, Kindinger LM, Brown R, Bennett PR and MacIntyre DA. Relationship between yeast infection and vaginal microbiome composition in pregnancy. Reproductive Sciences, March 2016, Vol.23, Suppl 1, p. 51A-344A. Society for Reproductive Investigation, Montreal, March 2016.

Brown R, Kindinger LM, Lee YS, Marchesi JR, Bennett PR and MacIntyre DA. Characterisation of the vaginal microbiome in patients subsequently experiencing preterm pre-labour rupture of membranes. Reproductive Sciences, Mar 2016, Vol.23, Suppl 1, p. 51A-344A. Society for Reproductive Investigation, Montreal, March 2016.

12 Interactions between the cervix and vaginal microbiome in pregnancy

Abbreviations

2-D Two dimensional 3-D Three dimensional 4-D Four dimensional ANOVA Analysis of variance B'ham Birmingham Women's Hospital BMI Body mass index BV Bacterial vaginosis C&W Chelsea and Westminster Hospital Cam Cambridge University Hospitals CCL5 Chemokine ligand 5 Cerc Cervical cerclage CIN Cervical intra-epithelial neoplasia CKC Cold knife conization CL Cervical length COX Cycle-oxygenase CST Community state type CT Cervical treatment CV Cervical volume CVF Cervicovaginal fluid DNA Deoxyribonucleic acid ECM Extracellular matrix EDTA Ethylenediaminetetraacetic acid ELISA Enzyme-linked immunosorbent assays ENA European Nucleotide Archive ERα Oestrogen receptor (alpha) FI Flow index GA Gestational age G-CSF Granulocyte colony-stimulating factor GM-CSF Granulocyte-macrophage colony-stimulating factor HCA Hierarchical clustering analysis HIV Human immunodeficiency virus HPV Human papilloma virus

13 Interactions between the cervix and vaginal microbiome in pregnancy

HVS High vaginal swab ICAM Intercellular adhesion molecule IFN Interferon IL Interleukin IQR Inter-quartile range K-W Kruskal-Wallis (test) LDA Linear discriminative analysis LEEP Loop electrosurgical excision procedure LEfSe Linear discriminative analysis effect size LLETZ Large loop excision of the transformation LPS Lipopolysaccharide LXSAHM Luminex Human Premised Multi-analyte Kit MCP Monocyte chemotactic protein MMP Matrix metalloproteinases MTL Mid-trimester loss NF-ⱪB Necrosis factor – kappa B NGS Next generation sequencing NK (cells) Natural killer (cells) NRES National Research Ethics Service ns Non-significant NSAIDS Non-steroidal anti-inflammatory drugs PBS Phosphate-buffered saline PCA Principal components analysis PPROM Preterm prelabour rupture of membranes PR Progesterone receptor QCCH Queen Charlotte's and Chelsea Hospital qPCR Quantitative polymerase chain reaction RANTES Regulated on Activation Normal T Expressed and Secreted RCOG Royal College of Obstetricians and Gynaecologists RCT Randomised control trial RDP Ribosomal Database Project rRNA ribosomal Ribonucleic acid SMH St Mary's Hospital Sobs Species observed

14 Interactions between the cervix and vaginal microbiome in pregnancy

SOP Standard operating procedure sPTB Spontaneous pre-term birth STAMP Statistical Analysis of Metagenomic Profiles STI Sexually transmitted infection TBE Tris/Borate/EDTA TLR Toll-like receptor TNF Tumour necrosis factor TVS Transvaginal scan UV Ultraviolet VEGF Vascular endothelial growth factor VI Vascularity index VOCALTM Virtual Organ Computer-aided AnaLysis

15 Interactions between the cervix and vaginal microbiome in pregnancy

Thesis abstract

During pregnancy, the cervix protects the growing fetus and uterine cavity from risk of ascending infection. Microbial–induced inflammation has been shown to disrupt the cervical epithelial barrier, cause premature cervical ripening, and ultimately result in preterm birth; the leading cause of death in children under five. Current prediction and prevention strategies have been ineffective at managing the global dilemma of preterm birth.

Prior to the work presented in this thesis, there had been no examination of the association between the cervix, as a mechanical and immunological barrier to ascending bacterial infection, and the co-existing vaginal microbiota during pregnancy. Furthermore the impact of preventative interventions, namely cervical cerclage or vaginal progesterone supplementation, had not been assessed.

16S rRNA gene sequencing techniques were employed for the longitudinal assessment of vaginal microbial profiles in pregnancy, to compare women at high- and low-risk of preterm birth. At each sampling time-point (12, 16, 22, 28 and 34 weeks gestation) a matched transvaginal ultrasound scan was performed for measurements of cervical length, volume and vascularity, using 2D and 3D/4D ultrasound technology. The interaction between cervical phenotypes and the vaginal microbiota in pregnancy were assessed with respect to subsequent gestation at birth.

In this thesis I demonstrate that high abundance of L. crispatus appears to be advantageous and is associated with subsequent term birth, while L. iners is a risk factor for preterm birth. Although assessment of cervical volume and vascularity did not provide improved prediction of preterm birth over current cervical length screening, second trimester vaginal microbial composition effectively differentiated subsequent early from late preterm birth (before and after 34 weeks gestation). Additionally, vaginal microbial profiles, in conjunction with ultrasound assessment cervix, may provide future potential stratification for preterm birth risk, although larger validation studies are required. I also demonstrated that the interaction between cervical length, vaginal microbiota, and gestation at birth varies according to underlying aetiology of preterm birth. In particular, women with pre-pregnancy excisional CIN treatment displayed substantially different cervical and microbial profiles to women with a prior preterm birth.

16 Interactions between the cervix and vaginal microbiome in pregnancy

Finally I assessed the impact of cervical cerclage and suture material on vaginal microbiota in high risk pregnancies. Multifilament braided suture, the material predominantly used by obstetricians without an evidence base, was shown to induce vaginal dysbiosis, local excretion of pro-inflammatory cytokines and prematurely increased cervical vascularity. The monofilament suture alternative had minimal impact on vaginal microenvironment. Similarly vaginal progesterone supplementation administered for a shortened cervical length had minimal impact on antenatal vaginal microbiota profiles. Overall, the studies presented in this thesis provide an improved understanding of microbial–host interactions in both low- and high-risk populations for preterm birth.

17 Introduction

1 INTRODUCTION

18 Introduction

Preterm birth

1.1 Incidence a global issue

Preterm birth, defined as delivery before 37 completed weeks of gestation, affects 7-15% of pregnancies worldwide, but accounts disproportionately for 80% of neonatal morbidity and mortality 1-4. The short and long term health sequelae of preterm birth have major social and economic implications. Despite a significant increase in prematurity focused research over the last 20 years, the incidence of preterm birth is largely unchanged; preterm birth remains a global health concern affecting resource-rich and poor countries alike. The challenge is in understanding and preventing this heterogeneous condition.

1.2 Neonatal morbidity and mortality

Prematurity is responsible for more than half of all neonatal deaths 5, 6, and has recently emerged as the largest single cause of childhood death under the age of five 7. Whilst advances in neonatal care over the past few decades have dramatically improved survival of extremely premature infants, there remains a significant risk of disability among surviving children6. The severity of neonatal morbidity relates to extremity of prematurity, where earlier gestation at birth correlates with poorer prognosis 8. Preterm birth is frequently classified into extremely preterm (birth <28+0 weeks), early preterm (birth between 28+0 and <34+0 weeks) and late preterm (birth between 34+0 and <37+0 weeks). Survival rates improve among babies born after 24 weeks, but most significantly after 28 weeks (Figure 1-1) 9.

Morbidity among surviving infants of preterm birth is high, particularly those born before 34 weeks (Figure 1-2) 10, 11. These babies are at risk of multi-organ short-term complications including respiratory distress, sepsis, intraventricular haemorrhage and necrotising enterocolitis 9. Longer-term complications of prematurity are predominantly neurodevelopmental, including cerebral palsy, global developmental delay, blindness and hearing impairments 8, 12, 13. In a 6 year follow up of children born before 26 weeks, the EPICure study group reported 41% of surviving infants had significant cognitive impairment, defined as an IQ greater than two standard deviations below than their school age counterparts, while 13% had disabling cerebral palsy and a further 13% had severe sensory impairment 8.

19 Introduction

100% 90% 80%

70% Survival

60% Mortality 50% 40% 30% 20% 10% 0% 22 24 26 28 30 32 34 36 38 40 42 Gestation at Birth (weeks)

Figure 1-1 Survival by gestation at birth among live-born infants. Reproduced from Mercer et al 9

100% 90% 80% 70% Respiratory Distress Sepsis 60% Intraventricular Hemorrhage 50% Necrotising Enterocolitis 40% 30% 20% 10% 0% 22 24 26 28 30 32 34 36 38 Gestation at birth (weeks) Figure 1-2 Acute morbidity by gestation among surviving infants. Reproduced from Mercer et al 9

These short and long term sequelae of prematurity carry significant emotional and financial burden on families, carers, and healthcare services 10, 11, 14. In addition to the immediate costs of caring for premature babies in neonatal intensive care units (estimated at £317,166 per night), the cost of hospital re-admissions in the first five years of life are 20 times greater for babies born before 28 weeks than those born at term 10. The projected annual cost for 18 years of subsequent healthcare for all preterm infants born in the UK in a year is estimated at almost £3 billion 10. In the United States the cost is thought to be in excess of $26 billion per year 15.

20 Introduction

Physiology of parturition

1.3 Normal term parturition

During pregnancy the myometrium is held in a state of functional quiescence, and the uterine cervix is long and closed. At term gestation a transformation towards increased uterine contractility, cervical remodelling, and fetal membrane activation occurs. This is mediated by a complex interplay between inflammatory and endocrine pathways, culminating in the spontaneous onset of labour and delivery of the fetus.

As pregnancy progresses, this shift from uterine quiescence to one of contractility occurs most notably in the third trimester. It results from progressive synthesis of myometrial gap junctions, oxytocin receptors and prostaglandin receptors within the myometrium 16, which accumulatively increase the sensitivity of myocytes to extracellular calcium. The steadily increased myometrial contractility potential, achieved by term ensures the smooth muscle is primed for stimulated contractions by circulating oxytocin, a powerful endogenous uterotonic produced by the posterior pituitary17. In the late third trimester, up regulation of inflammatory mediators occurs and it is this balance between suppressor and pro-inflammatory cytokines that is thought to be central to timing of spontaneous birth 18.

Chemotactic cytokines, in particular Interleukin (IL)-1β, IL-6, IL-8, tumour necrosis factor (TNF)- α, and monocyte chemotactic protein 1 (MCP1) play a key role in promoting recruitment of leukocytes, macrophages and neutrophils to the myometrium and cervix with the onset of labour 19. Pro-inflammatory cytokines also stimulate prostaglandin synthesis from arachidonic by cycle- oxygenase (COX) enzymes, through up regulation of the inducible isoform, COX-2 in the myometrium 20. In addition to pro-inflammatory cytokines, mechanical stretch of the lower uterine segment 21 and increased circulating oxytocin are the primary inducers of prostaglandin synthesis and therefore are promoters of cervical maturation and uterine contractility 22.

21 Introduction

1.4 Oestrogen and progesterone

Progesterone, a reproductive steroid hormone is the primary pro-gestational hormone in mammals. The anti-inflammatory and immunomodulatory effects of progesterone are central to maternal-fetal immunotolerance 23, 24. Initially produced by the corpus luteum until the 10th gestational week, the placenta takes over as the major site of both progesterone and oestrogen production in pregnancy, and serum levels increase progressively until delivery 25. In pregnancy, progesterone’s actions have primarily thought to be the maintainence of myometrial and cervical quiescence, however evidence suggests progesterone may also play an immunomodulatory role in maternal-fetal tolerance 23. Progesterone’s maintainence of myometrial quiescence is mediated through the actions of two major progesterone receptor (PR) subtypes. These are PR-A and PR-B, and they have opposing pro- and anti-inflammatory downstream effects. The action of progesterone in pregnancy is thought to be regulated through PR-A/B expression ratios at the cell surface 26, 27. Throughout pregnancy progesterone largely exerts an anti-inflammatory effect via PR-B dominant signalling 26. Downstream effects include reduced nitric oxide, prostaglandin and cytokine production 28, 29, thereby maintaining reduced myometrial contractility 30. Progesterone also exerts a quiescent effect on the cervix, limiting prostaglandin-induced collagenous remodelling of the cervical fibroblast 31-33. This is employed in clinical practice to induce labour; administration of the progesterone-receptor antagonist, Mifepristone (RU486) effectively induces cervical ripening as early as the second trimester of pregnancy in cases of late miscarriage or terminations of pregnancy 34. At the onset of normal physiological term parturition, a switch from PR-B to PR-A expression occurs, resulting in a shift to pro-inflammatory actions26, frequently referred to as a functional progesterone withdrawal 24.Circulating oestrogen induces myometrial contractility through stimulating the synthesis of prostaglandins to promote cervical ripening and myometrial gap junctions, thereby enhancing nerve conduction for coordinated labour contractions24. Oestrogen exerts pro-contractile effects through acting on the oestrogen receptor, ERα. Throughout pregnancy progesterone inhibits expression of the ERα in the myometrium, therefore holding the myometrium in a state refractory to circulating oestrogen24. Diminished inhibition following functional progesterone withdrawal increases expression of ERα, and in doing so induces functional oestrogen activation, transforming the myometrium to a contractile state 24, 26. In humans a ‘functional progesterone withdrawal’ is therefore considered a key-initiating event for labour onset 24, 26.

22 Introduction

1.5 Cervical remodelling

During pregnancy, the cervix protects the growing fetus and uterine cavity from the vaginal microbial environment through providing an anatomical barrier by way of cervical mucus plug and an immunological one through production of cytokines and antimicrobial properties 35.

In early pregnancy 85% of the cervix is composed of an extracellular matrix (ECM) dominated by collagen fibres type I (66%) and type III (33%). Elastin, fibroblasts, smooth muscle cells, and white cells are also scattered among the cervical tissue 35. Structural support is provided to the growing fetus through the high tensile strength of type I and III collagen. The cervix remains long and closed for the vast duration of pregnancy and essentially undergoes extensive structural remodelling towards term to enable effective dilation in response to uterine contractions during labour. This process, clinically referred as cervical ripening, occurs with activation of an inflammatory signalling cascade. An influx of cytokines including IL-6, IL-8, IL- 1β, and TNFα into the cervix recruits activated macrophages, T lymphocytes, mast cells and eosinophils to the cervical extra-cellular matrix (ECM) 36-38. The powerful chemotactor IL-8 stimulates extravasation of neutrophils that release degradative enzymes matrix metalloproteinases (MMPs -1, -8, -9, and 13) into the ECM 38. In conjunction with increased prostaglandin production, these MMPs cause physiological collagenous remodelling. Cervical collagen concentration at term is reduced by 30-50% compared to mid-gestation 35, 39 (Figure 1-3).

23 Introduction

Figure 1-3 A demonstration of the process of cervical remodelling that occurs with advancing gestation in the lead up to normal labour at term. Reproduced from Word et al 2007 35

1.5.1 Cervical angiogenesis In response to inflammatory-driven collagenous remodeling, the cervix simultaneously undergoes major vascular change. Cytokine activation of vascular endothelial growth factor (VEGF) initiates local cervical angiogenesis 40, 41. In a positive feedback relationship, VEGF further increases cytokine expression, particularly IL-6, which in turn attracts further VEGF to the cervix 36, 41 leading to rapid increase in vasodilation and new blood vessel formation. Increased blood flow aids delivery of activated cells and enzymes, accelerating collagenous breakdown, while increased vascular permeability increases the water content in the ECM. These physiological changes are clinically evident as cervical swelling, softening and ripening in preparation for dilation and expulsion of the fetus.

24 Introduction

Adaptive and innate immunity in pregnancy

During pregnancy the maternal immune system undergoes an evolutionary adaptive step enabling tolerance of the developing embryo, 50% of which is derived from paternal genes, whilst recognising pathogenic infection 42, 43. This is mediated by two main arms of the immune system: the innate and the adaptive immune response. The innate immune response provides an immediate, non-specific response towards invading microorganisms, whereas the adaptive response forms a slower antigen specific reaction 42, 44, 45. The innate response is a primitive defence mechanism involving phagocytosis (e.g. neutrophils, monocytes) mediated by non- specific pattern-recognition receptors, such as Toll-like receptors, and acute phase response proteins. The adaptive system takes several days to mount a response, involving T and B cell memory with high affinity, target-specific receptors 44, and is thought to be modulated in pregnancy to tolerate paternal derived fetal antigens (Figure 1-4).

Figure 1-4 The two arms of innate and adaptive immunity. The more primitive, innate arm mounts an immediate, antigen-non-specific response after exposure to foreign antigens. Adaptive immunity, reliying on a slower antigen-specific response,. Reproduced from Luppi et al44

There is evidence that in pregnancy, women may develop a tolerance to paternal antigens through the adaptive immune system; in simplified terms, the maternal T cells are conditioned to a state of ‘unresponsiveness’ to fetal cells 44. Specifically pregnancy is thought to promote conversion of undifferentiated T helper cells (Th0), towards an increased Th2 profile, which is associated with predominantly anti-inflammatory cytokines, interleukin (IL)-4 and IL-1045. In contrast a Th1 profile is associated with an array of pro-inflammatory mediators including interferon-gamma (IFN)-γ, IL-2 and tumour necrosis factor-alpha (TNF-α), which are the major effectors against pathogen invasion45.

25 Introduction

Pathophysiology of preterm birth

Spontaneous preterm birth is a consequence of the same terminal pathways as physiological term labour, but may be initiated by multiple pathological aetiologies including infection, inflammation, uteroplacental infarction or haemorrhage, uterine over-distension, uterine structural abnormalities and cervical weakness (Figure 1-5) 46.

Figure 1-5 The pathological processes implicated in the preterm parturition syndrome. Reproduced from Romero et al 2014 47

26 Introduction

1.6 Infection and inflammation

Infection-induced inflammatory parturition pathways are thought to be the causal driver of around 40% of preterm births, and as many as 80% of early preterm births before 28 weeks gestation 38, 48. The mechanism underlying microbial induced preterm labour is thought to be exposure of the amniotic cavity to pathogenic bacteria, as microorganisms identified in the amniotic fluid of preterm births exhibit similar profiles to those found in the vagina 49. As such the primary sources of infection are thought to be ascending bacteria from the lower genital tract 50 (Figure 1-6).

Figure 1-6 Ascending vaginal bacteria enter the amniotic cavity triggering a local inflammatory response. This causes production of prostaglandins and recruitment of proteases, which in turn enhance myometrial contractility, cervical remodelling and induce membrane rupture. Reproduced from Romero et al 2014 50.

1.6.1 Preterm birth and the immune response to infection

Toll like receptors (TLRs) are an important component of pattern recognition receptors within innate immunity 51. In pregnancy TLRs are expressed by leucocytes, placental, decidual and myometrial cells 45, 51. Specifically TLR-2 and TLR-4 detect gram-positive and gram-negative bacteria-derived lipopolysaccharide (LPS) respectively 52. TLRs that detect invading bacteria at the maternal-fetal interface stimulate a NF-ҡB mediated inflammatory

27 Introduction response, which causes the release of cytokines IL-1β, TNF-α and chemokines IL-8 and MCP- 152, 53. This inflammatory response to infection triggers an inflammatory-driven parturition cascade, which initiates recruitment of degradative proteases and prostaglandins to the cervix, and culminates in premature collagenous and vascular remodeling, and ultimately in preterm birth (Figure 1-7) 36-38, 54.

Pathological Trigger, eg Bacterial derived LPS

Detection by toll like receptors, TLR-2 and TLR-4

Activation of NF-ҡB pathway

Release of inflammatory mediators IL-1β, TNF-α, IL-6, Il-8, MCP-1

Chemokine recruitment to the cervix and myometrium of: White cells Prostaglandins Matrix metalloproteinases

Cervical dilation and Myometrial contractions Membrane activation effacement

Figure 1-7 Bacterial induced activation of the inflammatory parturition pathway culminating in premature cervical ripening, myometrial contractility and membrane activation and rupture.

The expression of TLRs in the placenta, particularly TLR-4, increases with advancing gestation, and expression is predominantly localised to the outer trophoblast layer 52. This suggests that pregnancy becomes more responsive to pathogenic stimuli towards term, but also that an immune response to microbial invasion is mounted once microbes have breached this protective layer into the sterile uterine cavity 52. Similarly, expression of TLR-4 is comparatively reduced, or absent from epithelial cells in the lower genital tract (i.e. the vagina)55. This may contribute to a degree of tolerance of colonisation by abnormal bacteria in the vagina by some women.

28 Introduction

1.7 Structural factors

1.7.1 Uterine overdistension Mechanical stretch of the myometrium in response to fetal uterine growth at term is considered an important stimulus for labour onset. In vitro studies have demonstrated that uterine stretch up regulates expression of IL-856, contraction-association proteins, in particular oxytocin receptors21 as well as the COX-2 enzyme and therefore prostaglandin synthesis57. Uterine stretch is thus thought to contribute to elevated preterm birth rates among twin and higher order pregnancies 58 as well as in conditions such as polyhydramnios and uterine anomalies.

1.7.1a Multiple pregnancy Twin pregnancies account disproportionately for the number of babies born preterm. In 2011, 53.1% of multiple births were premature in England and Wales, compared to only 5.6% of singleton pregnancies 59. In addition, twins have a six times higher neonatal mortality than singletons (2.5 vs. 15.4 deaths per 1,000 live births) 59. The burden of twin pregnancies on health services has increased substantially over the last 30 years. Between 1980 and 2011 the number of twin births rose 76% in the United States 2, a consequence of rising maternal age and associated use of assisted reproductive techniques, particularly in high income countries. The underlying mechanism driving spontaneous preterm labour in multiple pregnancy is likely a combination the endocrine effects of increased corticotrophin-releasing hormone production from a larger placental mass in twins 60, as well as the effect of uterine stretch upon contraction- associated inflammatory mediators within the uterus.

1.7.1b Uterine anomalies Uterine anomalies, an umbrella term for unicornuate, bicornuate, septated uterus and uterus didelphys, are caused by a defective fusion of the mullerian ducts during embryogenesis. With a prevalence of about 2-4%, they remain largely unrecognised among reproductive age women until pregnancy conception 61. Associated with recurrent miscarriage as well as preterm birth61- 64, the underlying mechanisms for adverse pregnancy outcome are poorly understood, but are thought to relate to the premature activation of uterine stretch due to smaller intrauterine capacity63.

29 Introduction

1.7.2 Excisional cervical treatment for cervical intra-epithelial neoplasia Persistent infection with high-risk oncogenic Human Papillomavirus (HPV) subtypes, most commonly 16 and 18, is a necessary pre-cursor for development of cervical intra-epithelial neoplasia (CIN) and the progression to invasive cervical cancer65. Screening programs have significantly reduced the incidence of cervical cancer through the sensitive detection and treatment of its precancerous form, CIN.

Figure 1-8 Excisional cervical treatment for cervical intra-epithelial neoplasia (CIN) in non-pregnant women. Methods of cervical treatment including cold knife conization (as shown), large loop excision of the transformation zone (LLETZ), and laser conization, are associated with an increased risk of preterm birth in subsequent pregnancy.

Meta-analyses 66-68 and large retrospective linkage studies 69-71 suggest that excisional methods of cervical treatment (cold knife conization, laser conization, and large loop excision of the transformation zone (LLETZ)) are associated with an increased risk of adverse reproductive sequelae in subsequent pregnancy. These include a 2-fold increased risk of preterm birth <37 weeks, low birth weight, premature rupture of membranes, and perinatal mortality 66-71. The underlying mechanism for spontaneous preterm birth in women with cervical treatment remains uncertain. Hypotheses include a mechanical weakness secondary to loss of cervical tissue at excisional treatment and also the “dose-response” effect; the larger the excision or proportion of cervix removed, the more serious the effect may be from treatment 72-75. Other hypotheses include immunomodulation relating to underlying HPV infection that subsequently augments parturition pathways 73, a compromised barrier to ascending infection, and CIN directly contributing to preterm birth risk 71, 72.

30 Introduction

1.8 Disorders of placentation

Abnormal placentation is a pathological feature commonly associated with pre-eclampsia and is present in about 30% of women with spontaneous preterm birth 50. The vascular endothelial dysfunction that occurs at the uteroplacental interface is characteristic of defective placentation. This vascular endothelial dysfunction increases the risk of preterm birth due to abnormal decidual haemostasis and propensity for haemorrhage 76, 77. A decidual haemorrhage, clinically referred to as an abruption, is the partial or complete separation of the placenta from myometrium 78 and clinically manifests as vaginal bleeding, abdominal pain, contractions and frequently fetal compromise78. Activation of the clotting cascade following an acute bleed results in thrombin formation at the decidual layer. Thrombin, a powerful uterotonic, stimulates myometrial contractility through increasing intracellular myometrial calcium comparable to the effects of oxytocin 77, thereby contributing to preterm birth risk among these women.

31 Introduction

Prediction of spontaneous preterm birth

Identification of pregnancies most likely to deliver preterm allows for timely and targeted antenatal preparation in attempt to prolong pregnancy and optimise neonatal outcome. These include commencement of preventative interventions, ensuring appropriate administration of prophylactic corticosteroids for lung maturation, commencement of tocolytics, and mobilisation for proximity to neonatal intensive care units. Accurate prediction of pregnancies at highest risk of preterm birth remains a major clinical challenge. Currently predictive tools include 1) measurement of fetal fibronectin glycoprotein - a bedside test, primarily reserved for women symptomatic of preterm labour presenting after 24 weeks gestation, and 2) assessment of the cervix by transvaginal ultrasound – frequently employed as a screening test in the setting of antenatal preterm surveillance clinics, and usually performed prior to 24 weeks gestation.

1.9 Fetal fibronectin

Fetal fibronectin is an extracellular glycoprotein produced by amniocytes throughout pregnancy. It acts as biological glue, fixing the blastocyst in place at implantation and cementing fetal membranes together until the mid-third trimester 79. Fetal fibronectin is detectable in the vagina early in pregnancy but is normally undetectable between 22-35 weeks until membranes ‘ripen’, when it becomes detectable again 79. If any disruption to the fetal membranes occurs during this time, fetal fibronectin is detectable in the posterior fornix of the cervix, and as such has been employed as a commercial screening tool to identify high-risk pregnancy for preterm delivery between 22 to 35 weeks gestation80. The major clinical utility of fetal fibronectin screening is its high negative prediction (99%) for preterm birth when detectable levels are <50ng/ml 79. Positive predictive values are less impressive, although this has improved with the advent of quantitative bedside machines 81, 82. When sampled between 22 and 28 weeks gestation, concentrations of fetal fibronectin <50, 50–199, 200–499, and >500 ng/mL are associated with rates of spontaneous preterm birth before 34 weeks of 11%, 15%, 34% and 48% respectively 82. While this aides acute antenatal management of an individual, utility in population based screening is limited as respective sensitivities are 73%, 47%, 29% and 9% 82. Furthermore screening at this late gestation (22 to 28 weeks) is too late for safe preventative surgical intervention by way of cervical cerclage.

32 Introduction

1.10 Transvaginal ultrasound of the cervix

1.10.1 Cervical length screening The process of cervical remodelling during pregnancy, also referred to as ‘ripening’, precedes myometrial contractions by several weeks. Premature onset of cervical ripening can be detected using transvaginal ultrasound for the measurement of cervical length 83 (Figure 1-9).

A shortened cervix relative to gestational age is therefore considered to be a clinical phenotype. Based on work in the late 1990’s by Iams et al 83 and later Heath et al 84, pregnant women may be clinically stratified by spontaneous preterm birth risk according to their second trimester cervical length (CL) measurement 85. A normal CL, considered to be over the 10th centile (>25mm) taken at 24 weeks gestation, is reassuring; typically 90% will deliver at term 86. A short cervix identified in the context of a history of clinical risk factors, namely a prior spontaneous preterm birth or prior midtrimester miscarriage 87 has a predictive value for delivery before 35 weeks ranging from 22 to 50% 88.

Figure 1-9 Transvaginal cervical length measurements (A) a long, closed cervix and (B) a shortened cervix. Source: anonymised transvaginal scans from St Mary’s Hospital Imperial College NHS Trust prematurity clinic.

The predictive accuracy of cervical length screening is dependent on CL thresholds as well as gestational age at measurement 88, 89. At 24 weeks a CL ≤25mm has relative risk (RR) of 7 for preterm birth <37weeks, while a shorter CL, below the 1st centile (≤13mm) increases this risk to RR of 1483, but with a compromise of screening sensitivity 88. As physiological shortening of the cervix occurs with advancing gestational age, earlier screening associates with longer CL measurements, while later screening with short CL measurements89. Figure 1-10 clearly demonstrates this association, as the 10th centile differs from 34mm at 18 week screening, to 25mm at 24 week screening.

33 Introduction

100 Gestation at cervical measurement 18 weeks

80 20 weeks 22 weeks 24 weeks 60

Percentile 40

20 10th percentile

0 15 20 25 30 35 40 45 50 55 60 Cervical length (mm)

Figure 1-10 Distribution of cervical length measurements in over 6000 twins pregnancies screened at 18, 20, 22 and 24 weeks gestation. Reproduced from supplementary material in Kindinger et al 201590

Gestational age at measurement therefore impacts on screening accuracy; the greatest risk for preterm birth is among women with a short cervix at an early gestational age, compared to the same measurement at later screening 89. For example, given a CL of 10mm, the predicted probability of preterm birth before 32 weeks is 55% if measured at 16 weeks, compared to 28% if measured at 24weeks 89. Screening programs therefore frequently commence CL monitoring in the early second trimester as this balances the low sensitivity and high specificity of early CL screening (<16weeks), with improved sensitivity at the compromise of specificity at later screening (>22weeks) 88, 89.

Although thresholds vary, women identified to have a shortened cervix, are considered at high risk of preterm birth and may be offered an intervention by way of cervical cerclage or progesterone supplementation. Many women undergo intervention unnecessarily, as screening is associated with varying rates of false positives depending on the CL threshold employed 88, 89. Improved targeting of pregnancies which are most likely to benefit from intervention is therefore required. Despite the poor sensitivity of CL screening (60-70%) 88, no test has yet surpassed its reproducibility, cost-effectiveness, and technical ease of use for preterm birth prediction. The challenge in developing an improved predictive tool involves consideration of the multifactorial underlying aetiology of preterm birth.

34 Introduction

1.10.2 Cervical volume In the weeks and months prior to the onset of term and preterm parturition, the cervix increases in water content (and therefore volume) as it undergoes a morphological transition during collagenolysis and angiogenesis. It is therefore possible that preterm birth prediction using CL measurements may be improved through addition of quantifiable cervical volume and vascularity. This has been the focus of some enquiry over the last few years, made possible with the advent of sophisticated three-dimensional (3-D) sonographic technology 91 (Figure 1-11).

Figure 1-11 Three dimensional cervical volume image of (A) a closed cervix and (B) an opening cervix

Cervical volume (CV) in pregnancy increases with advancing gestation and with parity 92-94, consistent with expected clinical findings. CV has been shown to predict preterm birth with similar accuracy as CL measurements, albeit in a limited number of cross-sectional studies 92, 93, 95. The limitation of CV as a screening tool is the added technical difficultly of 3-D scan acquisition and analysis.

35 Introduction

1.10.3 Cervical vascularity

The introduction of 3-D/4-D volumetric studies to power doppler has enabled the construct of a virtual vascular tree within the cervix. Vascular and blood flow can be measured through the ‘histogram’ facility of VOCALTM power doppler software, generating two indices: vascularization index (VI) and blood flow index (FI). Vascularization index (VI) indicates the density of blood vessels within volume of interest as a percentage. It is calculated as a ratio of colour-coded voxels (smallest unit of volume) to all voxels. Flow index (FI) describes the signal intensity of the color voxels, reflecting the rate of blood flow within the vessels 96. Importantly, these VOCALTM indices are reliable and reproducible 97, 98. However similar to volumetric studies, published reports on the vascularity of the cervix in pregnancy are limited 95, 99-101. While there is some suggestion that VI increases in women symptomatic of preterm labour 95, further investigation into the predictive value of cervical vascularity indices are required. Moreover, as evidenced by the relation between VEGF and inflammation in the cervix 41, the relation between cervical vascularity and vaginal infection has not been established, nor the impact of the cerclage on cervical vascularity.

36 Introduction

Prevention of preterm birth

Early diagnosis of high risk pregnancies enables timely and targeted interventions for the prevention of preterm birth. Currently there are two main prevention strategies 1) insertion of a cervical cerclage, and 2) progesterone therapy by way of daily vaginal pessaries. Both treatments are usually commenced prior to 24 weeks. For women presenting in clinically established threatened preterm labour after 24 weeks, tocolysis may be administered to inhibit uterine contractions. For women with suspected infection in pregnancy, antibiotics may be prescribed at any gestational age. These are discussed below.

1.11 Cervical cerclage

Cervical cerclage was first described by Shirodkar in 1955 as a surgical procedure involving the suturing of the cervix in a “purse-string fashion” for the repair of the ‘incompetent cervix’ (Figure 1-12). In its original form, the operation involved dissection of the bladder from cervix and a biological suture material, fascia lata, was used. However the disadvantage of this technique was that the suture could not be readily removed. In 1957 McDonald described a simpler procedure involving the placement of a non-absorbable purse string stitch around the cervix with no dissection102. In current clinical practice, a modified Shirodkar procedure involving bladder dissection, insertion of a non-absorbable suture which is subsequently removed at 36-37 weeks, has shown no benefit over the McDonald suture with respect of birth outcomes 103, and choice of techniques remains largely determined by operator preference.

Figure 1-12 Cervical cerclage placement around the uterine cervix (A). The cerclage suture is inserted (B) and tied (C) in a purse-string fashion around the cervix

37 Introduction

Cervical cerclage reduces the risk of spontaneous preterm birth by approximately 20%104, although its mechanism of action is still not fully understood. It is thought that the cerclage provides structural reinforcement to a weakened cervix, maintains cervical length and supports the endocervical mucus plug as a barrier to pathogens and ascending infection 105, 106.

Cervical cerclage is now commonly performed in the management of women who have had a previous preterm delivery or second-trimester pregnancy loss or in those who have undergone excisional procedures for cervical intraepithelial neoplasia (CIN). Globally an estimated 2 million procedures are performed annually for the prevention of preterm birth 107, 108, with geographic variations; approximately 3% of pregnancies in the USA and 1% of women in the UK receive a cerclage antenatally 109, 110. The value of cervical cerclage in reducing the risk of late pregnancy loss or preterm delivery continues to be debated. The MRC/RCOG trial of cervical cerclage involved randomisation of women whose obstetrician was uncertain of the benefit of cerclage to either cerclage or expectant management. The rate of preterm delivery was lower in the cerclage group, although this effect was only significant in those who had had three previous second-trimester losses or preterm births108. Overall the number of cerclages needed to prevent one preterm birth prior to 33 weeks was 25 but with very wide confidence intervals (95% CI 12 to 300) 108. Furthermore the cerclage was associated with doubling of the risk of puerperal pyrexia. However a large meta-analysis has since shown that, in singleton pregnancies with a prior preterm birth and a short cervix (≤25mm) before 24 weeks, cerclage is associated with a significant reduction in the risk of preterm birth before 35 weeks (RR 0.77, 95% CI 0.55– 0.89), and significant reduction in composite perinatal morbidity (RR 0.64, 95% CI 0.45–0.91) 111. Many obstetricians now use measurement of cervical length by transvaginal ultrasound to target cervical cerclage. The Royal College of Obstetricians and Gynaecologists recommend cerclage for women with a history of 3 or more spontaneous preterm birth or a short cervical length ≤25mm 112.

1.11.1 Cerclage suture material There is no consensus on the type of suture material that should be inserted for the cervical cerclage. A recent survey of UK obstetricians revealed that 86% of consultants have a preference for MersileneTM suture material, a braided multifilament suture, over a monofilament alternative such as Nylon or Prolene 113. Braided sutures consist of non-absorbable polyester ethylene terephthalate fibres braided together to form a 5mm-wide mesh tape. They are

38 Introduction characteristically high in tensile strength and due to a high coefficient of friction, are thought to provide structural support to a weakened cervix. Monofilament sutures are made of a single strand of non-absorbable polyamide polymer. Due to their simple structure they are less likely to harbour organisms leading to infection, and provide less resistance through tissues. Because of this monofilament sutures have a tendency to slip and therefore require a greater number of throws to secure a knot than braided sutures 106.

Figure 1-13 Suture materials used for cerclage insertion. Microscopic depictions of braided multifilament and monofilament suture material constructs, adjacent to a photograph of their macroscopic appearance, to provide a comparison of actual suture material size. Reproduced from www.surgicalspecialties.com/material-guide

Investigation into the impact and efficacy of various suture materials for cervical cerclage is limited to one RCT comparing two braided suture materials: Ethibond™ and Mersilene™ 114. In this study no difference in rates of preterm birth among either braided suture material. A RCT powered to compare outcomes in patients receiving multifilament or monofilament has yet to be undertaken.

Israfil-Bayli et al have hypothesized that the increased risk of infection and inflammation associated with braided suture materials could be a significant confounding factor leading to the underestimation of the true effectiveness of the cervical cerclage procedure115. As a preliminary test for their hypothesis they conducted a retrospective analysis of a small cohort of pregnant women who had cervical cerclage using either Nylon (monofilament) or Mersilene (braided) suture material in their units, finding that Nylon was associated with a 5% pregnancy loss rate compared to 12% for Mersilene™ 107. A prospective RCT comparison of braided and monofilament suture material is therefore clinically indicated.

39 Introduction

1.12 Progesterone supplementation

Progesterone’s mode of action in preterm birth prevention is thought to be through maintenance of myometrial quiescence via downstream anti-inflammatory signalling 26, as well as inhibition of premature cervical ripening. Progesterone supplementation is not associated with exposure to surgical risks of cerclage insertion including vaginal bleeding, pyrexia and caesarean delivery 104. In an indirect comparison individual patient data meta-analysis, progesterone was reported to be as effective as the cervical cerclage in preterm birth prevention among singleton pregnancies with a prior history of preterm birth, as well as those complicated by cervical shortening 116. The number needed to treat to prevent one spontaneous preterm delivery <33weeks is 11 patients with a short cervix117, which compares favourably to administration of antenatal corticosteroids to prevent respiratory distress syndrome and neonatal death 118.

Reports of progesterone’s effectiveness from several large trials are conflicting 116, 119-126. Fonseca et al and Hassan et al describe up to a 45% reduction in preterm birth <33 weeks when progesterone was prescribed to women with a short cervix CL <15mm119 and 10-20mm120 respectively, as well as a reduction in composite neonatal morbidity 125. In contrast, Van Os et al reported no improvement in neonatal morbidity or prematurity among healthy pregnancies found to have a CL <30mm 121. Most recently Opptimum, the largest RCT comparing progesterone versus placebo among high risk pregnancies with a CL <25mm concluded no reduction in the risk of preterm birth <34weeks or adverse neonatal outcome 124, further adding to the debate surrounding the effectiveness of progesterone in preterm birth prevention.

40 Introduction

1.13 Tocolytics

Tocolytic drugs are used in clinical practice to supress uterine contractions in women with threatened preterm labour. While tocolytics themselves effectively prolong gestation for up to 2 weeks127, they do not improve neonatal morbidity or mortality128. They are primarily employed to delay labour long enough for the administration of antenatal corticosteroids for fetal maturation and if applicable, transfer for proximity to an appropriate neonatal tertiary care centre. There are many different classes of tocolytics with varying mechanisms of action, route of administration, side effect and cost effectiveness. The primary groups include: NSAIDS such as indomethacin which inhibit prostaglandins; β-agonist such as terbutaline and calcium channel blockers such as nifedipine which reduce intracellular calcium in myocytes; and oxytocin receptor antagonists such as atosiban, which compete with endogenous oxytocin for receptors in the myometrium 128.

1.14 Antibiotics

Vaginal bacterial infection has long been considered an important risk factor for preterm birth. The ORACLE I trial demonstrated erythromycin antibiotic treatment for preterm prelabour rupture of membranes (PPROM), was associated with reduced preterm birth rates, use of surfactant, neonatal oxygen dependence, and fewer major ultrasound cerebral abnormalities 129. Co-amoxiclav, the comparison antibiotic, also prolonged pregnancy, but was associated with increased neonatal necrotising enterocolitis 129. In an assessment of longer-term outcomes, neither antibiotic improved childhood health compared to placebo 130.

The ORACLE II study131 demonstrated antibiotics were not beneficial in women who spontaneously laboured preterm with intact membranes. Neither erythromycin or co-amoxiclav reduced neonatal mortality, respiratory disease, or major cerebral abnormality 131, and in a 7- year follow-up, an increased incidence of cerebral palsy was observed among the children who had received antibiotics 132.

1.14.1 Antibiotic treatment for Bacterial Vaginosis Bacterial vaginosis (BV) is a common condition characterised by a depletion of Lactobacilli in the vagina 133 and overgrowth of various anaerobic bacteria, in particular Gardnerella vaginalis and Atopobium vaginae 134. Clinical diagnoses are made using Amsel criteria (a diagnosis

41 Introduction requires 3 of the following: vaginal pH >4.5, abnormal discharge, observation of clue cells attached to vaginal epithelium, or fishy odour with addition of potassium hydroxide) 135 or Nugent scoring of gram stains 136.

Although sometimes asymptomatic, BV is the most common cause of symptoms including offensive odour, discharge and vaginal irritation in reproductive age women 137. Moreover BV is associated with adverse reproductive health outcomes including pelvic inflammatory disease 138, transmission of sexually transmitted infections139, and HIV 140, 141, and when detected in pregnancy is associated with a 2-fold increase in preterm birth 142, 143.

Clindamycin and metronidazole are traditionally used for the treatment of BV 144, however the recurrence rate of BV post-antibiotic treatment is high 145, 146. It is thought that this relates to a failed restoration of Lactobacilli dominance following antibiotic treatment, leaving the vaginal environment susceptible to re-colonisation by BV associated pathogens 147.

Controversy surrounds management of BV in pregnancy 142, 143, 148. A recent meta-analysis suggested some benefit of clindamycin for the treatment of BV prior to 22 weeks gestation, with a reported reduction in rates of late preterm birth (before 37 weeks), but not early preterm birth (before 33 weeks), low birth weight, neonatal intensive care admissions, or maternal or neonatal infections 148. Other studies largely indicate antenatal antibiotic treatment for BV do not effectively reduce preterm births, and currently the consensus is that there is little evidence for screening of BV in pregnancy, as this does not improve outcome 142, 143.

42 Introduction

The vaginal microbiome

Bacteria have traditionally been cultured using classical microbiology techniques. These are limited in that the majority of bacteria cannot be cultured in vitro meaning that results do not accurately reflect the full spectrum of microorganisms present in a given sample. Major advances in DNA sequencing technology by way of Next Generation Sequencing (NGS) platforms have revealed a vast array of bacterial communities that resist traditional cultivation methods. Most commonly, this approach involves the amplification and sequencing specific regions of the bacterial 16S ribosomal RNA gene, which is widely pervasive among bacterial species with a highly conserved flank regions ideal for targeted binding of PCR primers. The 16S rRNA gene includes nine ‘hyper-variable’ regions (V1 – V9) that differ among bacterial species 149. The sequence of these hypervariable regions therefore provides information useful for the identification of microbial taxonomy following PCR amplification and sequencing. Bacterial taxonomy may then be allocated to amplified gene regions down to genera and species levels, providing comprehensive characterisation of microbial profiles 150. This characterisation of the full set of microbial genomes in a given sample of environment is referred to as the ‘microbiome’.

The NIH Human Microbiome Project was set up in 2008 with the goal of comprehensively characterising the microbiome at five sites in the body: oral, lung, skin, gut and vagina. The project findings revolutionised our understanding of microbial community structures across these body sites and their associations with disease processes 150, 151. Specifically the complex microbial ecosystem in the gut microbiome and its important contribution to systemic health continues to provide valuable insight into immune development and human disease processes 152, 153.

1.15 Vaginal microbiota and reproductive health

The ecological community of bacteria that colonise the vagina, collectively referred to as the vaginal microbiome, also play a central role in reproductive health154. Vaginal microbiota in reproductive age women primarily consists of Lactobacillus spp.155, which are broadly thought to promote health through the production of various bacteriostatic and bactericidal compounds as well as lactic acid to maintain a low vaginal pH 156. Collectively, these excretion products produce a local chemical environment that is hostile to other bacterial species and potential pathobionts, but which can be tolerated by Lactobacillus species 157. Importantly this provides

43 Introduction protection against diseased states such as BV, and STI’s 53, 158, 159, and pathogenic viruses including HIV 160 and the carcinogenic human papilloma virus (HPV) 161-163. Bacterial 16s rRNA gene sequencing has revealed a more complex picture of the vaginal microbiota than previously thought to exist155. Vaginal microbiota when assessed at species taxonomic level, commonly cluster in into five predominant communities. These may be referred to and described as community state types (CSTs) 155, although other classification systems do exist. Four of the five CSTs are highly abundant in specific Lactobacillus spp.; CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), and CST V (L. jensenii), while CST IV (vaginal dysbiosis) is characteristically deficient in Lactobacillus spp. and frequently abundant in diverse, often anaerobic and enteric bacteria species 155.

1.15.1 Hormonal influences Vaginal bacterial community composition differs both within and between women, and appears to be largely influenced by circulating hormones, in particular oestrogen. Oestrogen mediates intracellular deposition of glycogen in the vaginal epithelium 164, and glycogen is broken down by α-amylase present in the host vaginal mucosa to complex sugars including maltose, maltotriose, and maltotetraose. These sugars can be preferentially used as energy substrates by Lactobacillus spp. thus supporting their colonisation and domination of the vaginal microbial niche 165. Mirroring oestrogen availability, the composition of vaginal bacterial communities typically shift towards increased Lactobacillus abundance with the onset of menarche 166, and fluctuate with menstruation in a cyclical manner 167. Declining oestrogen levels during menopause associates with a shift from Lactobacillus spp. dominant CSTs I, II and V, towards a high prevalence of L. iners (CST III) around peri-menopause and ultimately CST IV in postmenopausal states, accompanied by atrophy of the vaginal epithelium and reduced cervicovaginal secretions 168.

1.15.2 Environmental influences Certain environmental factors influence vaginal microbial stability, including smoking 169, vaginal douching, and sexual intercourse 158. These habits are all associated with a lower prevalence of vaginal Lactobacillus spp. and a greater proportion of anaerobic bacteria (CST IV). Ethnicity may also be a source in variation of vaginal microbiota composition; Black and Hispanic women have a higher prevalence of CST IV than Caucasians and Asians 155, 170, 171.

44 Introduction

1.16 Pregnancy and vaginal microbiota

Aagaard and colleagues 172, as part of the Human Microbiome Project, were the first to employ NGS techniques to study the vaginal microbiome in pregnancy. Highlighting unexpected variations in the vaginal microbiome of a cross section of pregnant and non-pregnant women, they showed that pregnancy associated with comparatively increased vaginal Lactobacillus spp. stability and dominance with reduced microbial diversity compared to a non-pregnant state 172. Further studies in healthy pregnancy revealed that species diversity diminishes towards mid- trimester viability, characterised by Lactobacillus spp, dominance with high stability 171, 173-175. L. crispatus in particular has the lowest associated diversity and greatest stability throughout pregnancy171, whereas L. iners is associated with lower stability and increased likelihood to transition toward an atypical or dysbiotic microbiome 168, 176, 177. The microbial stability in mid- pregnancy is thought to be important in inhibiting preterm labour through protecting against ascending infection 173, 178.

A dysbiotic microbiome in pregnancy, characterised by reduction in Lactobacillus spp., is frequently associated with a reciprocal increase in potential pathobionts, including E. coli, group- B Streptococcus, and Bacterial vaginosis (BV)-associated bacteria including Gardnerella vaginalis, Prevotella and Bacteroides spp., and Mycoplasma 179. This dysbiosis has been associated with increased expression of inflammatory mediators in the cervicovaginal fluid 53, 180, and it has been suggested that in pregnancy, this pathogen-induced inflammatory vaginal environment compromises the integrity of the cervical epithelial barrier, thereby accelerating the downstream inflammatory-driven signalling cascade, culminating in preterm birth 51.

1.16.1 Vaginal microbiota and preterm birth Reports thus far implicating specific microbial communities in the pathogenesis of preterm birth have been conflicting. Digiulio et al 181 recently reported that increased vaginal microbial diversity in the 2nd trimester is a risk factor for preterm birth; however this study was limited to only 6 women experiencing preterm birth, with a mean gestation of delivery was 36 weeks, among a total cohort of 40 women. Petricevic et al182 in contrast demonstrated a dominance of L. iners (CST III) among their 13 preterm births from a cohort of 111 low risk pregnancies, while Romero et al 54 reported a lack of any association between vaginal microbiota and preterm birth risk albeit in an almost homogenous African American population.

45 Introduction

An association between an abnormal or atypical placental microbiome and spontaneous preterm labour has also been reported, leading to the suggestion that haematogenous seeding of oral bacteria to placenta deposits contributes to infection-driven preterm birth 183.

1.17 Vaginal microbiota and the neonate

As many as a quarter of infection-association preterm births are subclinical and diagnosed at placental histology, and therefore not recognised or treated184. This has significant implications for preterm fetuses that are at risk of inflammation-induced injury due to their immature innate immune system which starts to develop in utero from about 24 weeks. During this preterm period until about 36 weeks, the ability of the fetus to induce an adequate immune response is impaired 185, and so exposure to inflammation in utero is associated with additional morbidity, beyond adjustments for gestational age 13, 186.

The influence of maternal vaginal microbiota on neonatal outcome extends beyond delivery, as these bacteria are important sources of pioneering neonatal gut microbiota 187. Early work suggests neonatal gut microbiota is largely influenced by mode of delivery 188, whereby babies born vaginally exhibit gut bacterial communities that closely resemble their mother’s vaginal tract, while cases of caesarean section birth reflect skin microbiota 188. The longer-term implications of this early neonatal gut microbiota on childhood health are gradually being revealed. Dysbiotic vaginal microbiota has been implicated in necrotizing enterocolitis, a severe inflammatory bowel disease of premature infants 189, and babies born by caesarean, are more susceptible to severe asthma requiring hospitalisation 190 as well as development of allergies 191. Indeed early gut microbiota has longer term implications for health immunity 152. Vaginal microbial composition, pregnancy, and neonatal outcomes as well as longer-term health are therefore crucially interconnected.

46 Introduction

Justification for project

To date, there has been no examination of the association between the cervix as a mechanical and immunological barrier to ascending bacterial infection, and co-existing vaginal microbiota during pregnancy. Furthermore, the preventative measures for preterm birth employed in current clinical practice, namely the cervical cerclage and progesterone, have not been investigated for their impact on vaginal microbial composition or stability. The work presented in this thesis therefore aims to address this, focusing on the cervix in relation to the vaginal microenvironment in pregnancies both at low- and high-risk for preterm birth.

1.18 Project aims and hypotheses

Aims 1. Characterise the longitudinal shift in vaginal microbiota, cervical volume and cervical vascularity parameters in healthy pregnancies delivering at term.

2. Assess the vaginal microbiota and cervical parameters in pregnancies at high risk of spontaneous preterm birth and compare results between women with a prior preterm birth and those with pre-pregnancy excisional cervical treatment for CIN.

3. Investigate the impact of preventative interventions including cervical cerclage and progesterone treatment on vaginal microbiota and cervical anatomy during pregnancy.

47 Introduction

Hypotheses 1. Cervical volume (CV), vascularisation (VI) and blood flow (FI) indices increase with advancing gestation as the cervix undergoes cervical remodeling and angiogenesis.

2. Current screening for spontaneous preterm birth by way of cervical length (CL) measurement before 24 weeks may be improved through the addition of ultrasound parameters, CV, VI and FI.

3. Vaginal dysbiosis associates with cervical shortening, and altered CV, VI and FI as assessed using transvaginal ultrasound.

4. Prevalence of vaginal dysbiosis is greater among pregnant women with excisional-CIN treatment, than other high-risk groups, in particular those with a prior preterm birth. This increased incidence of dysbiosis in the CIN-treatment cohort contributes to their preterm birth risk.

5. Cervical cerclage-associated outcomes are related to the current preference for braided suture material clinical practice, over a monofilament alternative and the use of braided suture disrupts stability of the vaginal microbiome, inducing local inflammation and premature cervical vascularisation.

6. Progesterone pessary for preterm birth prevention promotes Lactobacillus spp. dominance in high-risk pregnancy.

48 Materials and Methods

2 MATERIALS AND METHODS

49 Materials and Methods

Retrospective study

A retrospective study was conducted to assess the impact of cerclage suture material on two primary clinical outcomes: 1) term versus preterm birth, and 2) viable versus non-viable fetus at delivery.

2.1 Study design

Retrospective outcome data was collected from singleton pregnancies receiving a cervical cerclage for their preterm birth risk over a ten-year period between January 2003 and 2013 across five UK hospitals in London (St Mary’s Hospital, Queen Charlottes and Chelsea Hospital, Chelsea and Westminster Hospital), Cambridge and Birmingham. Details regarding suture material used, outcomes of viable birth at term (≥37+0 weeks), viable preterm birth (between 24+0 and 36+6 weeks’ gestation), and non-viable birth (still birth or miscarriage >16+0 weeks’ gestation) were collected. Additional metadata assessed included maternal age, ethnicity, parity, history of prior preterm birth or midtrimester miscarriage, history of excisional cervical treatment, clinical indication (ultrasound indicated or elective), and cervical length and gestational age at cerclage insertion. Assessment of differences in outcomes of viability and preterm birth between cerclage suture material groups (braided versus monofilament) was performed using the Fisher exact test for categorical variables and Wilcoxon test for continuous variables.

The distributions of hospital location, maternal age, parity and gestation at cerclage insertion were identified as possible source of inhomogeneity between the monofilament and the braided groups. A linear mixed-effects model incorporating maternal age, parity and hospital location as fixed effects and indication for cerclage (ultrasound versus elective) as a random effect was used to compare braided versus monofilament suture material for the two primary outcomes (viability and preterm birth).

50 Materials and Methods

Prospective studies

The prospective studies described in this thesis focus on the assessment of the vaginal microbiota and cervix in pregnancy at risk of preterm birth. They were performed as longitudinal observational cohort studies, with individual arms assessing the impact of preventative interventions.

2.2 Study design

2.2.1 Prospective recruitment Pregnant women were recruited from two maternity hospital sites across Imperial College Healthcare NHS Trust: St Mary’s Hospital (SMH) and Queen Charlottes and Chelsea Hospital (QCCH), from January 2013 to January 2015.

Participants were recruited in two groups: 1. Healthy pregnancies without an underlying risk factor for preterm birth were recruited from routine midwifery-led antenatal booking appointments. These women were considered low risk controls. 2. Pregnant women with an underlying risk factor for preterm birth were identified at their first attendance at the prematurity clinics across the two sites. These women were considered the high-risk cohort.

51 Materials and Methods

2.3 Eligibility criteria

2.3.1 Inclusion criteria Participants were eligible for inclusion in the low risk control group if they did not have any underlying medical comorbidities and were considered appropriate for routine midwifery-led antenatal care. Eligibility criteria for the high-risk group included women with a previous spontaneous mid- trimester miscarriage or preterm birth between 16+0 and 37+0 weeks gestation in a singleton pregnancy immediately prior to the index pregnancy. A second inclusion group were women with previous excisional cervical treatment of depth >1cm for CIN grade II or III75, immediately prior to the index pregnancy. This included loop electrosurgical excision procedure (LEEP), large loop excision of the transformation zone (LLETZ) and cold knife conization (CKC).

2.3.2 Exclusion criteria Exclusion criteria for study participation included multiple pregnancy (twins or higher order), uterine anomalies, iatrogenic preterm birth <37 weeks in the previous or index pregnancy, laser and ablative therapy and CIN grade ≤1, HIV positive women, women who had received antibiotic therapy within the preceding 2 weeks, and women who had had sexual intercourse or vaginal bleeding in the preceding 48 h.

2.4 Research timeline

Every participant was followed up in a longitudinally at five gestational time points: A. 12+0 weeks B. 16+0 weeks C. 22+0 weeks D. 28+0 weeks E. 34+0 weeks

2.5 Ethical approval

The study was favourably reviewed by NRES Committee London - City & East. REC reference 12/LO/2003.

52 Materials and Methods

Procedures

At each time-point, cervicovaginal fluid (CVF) was sampled from the posterior fornix under direct visualisation. Two swabs were taken, snap frozen and stored at -80 ᵒC: 1. A BBL™ CultureSwab™ MaxV Liquid Amies swab (Becton, Dickinson and Company, Oxford, UK) for 16s rRNA gene sequencing 2. A Transwab® MW170 with rayon bud type (Medical Wire & Equipment, Corsham, UK) for cytokine analysis.

Following this, a transvaginal scan (TVS) was performed for cervical length (CL), cervical volume (CV), vascularization index (VI), and flow index (FI). To avoid introduction of operator variability, all scans were performed by a single clinician, trained and experienced in TV scanning.

Metadata collected for all participants included maternal age, BMI, ethnicity, gestation age at research visit and subsequent interventions for preterm birth.

53 Materials and Methods

2.6 Preventative interventions for preterm birth

For the duration of the study, both units employed a policy of continued CL screening, with indication for intervention being a CL ≤25mm at TVS measured at <24 weeks gestation.

2.6.1 Comparison of high and low risk pregnancies In the first part of the study from January 2013 to January 2014, the aim was to assess longitudinal variations in vaginal microbiota and cervical parameters among high and low risk groups. Therefore participation in the study did not influence or dictate preventative interventions for perceived preterm birth risk, namely cervical cerclage or vaginal progesterone supplementation.

2.6.2 Impact of an Intervention The second part of the study aimed to assess the impact of preventative interventions of vaginal microbiota and cervical morphology. Women identified to have a short cervix (CL ≤25mm taken <24 weeks gestation) were allocated to receive a cervical cerclage or progesterone pessary as described below. All women receiving these preventative intervention continued longitudinal follow up according to specified research time-points.

2.6.2a Cerclage suture material To investigate the impact of cerclage suture material, women identified to have a short CL ≤25mm at transvaginal ultrasound, were randomised to receive either a braided Mersilene® (n=25) or monofilament Ethilon® (n=24) cerclage. To ensure no bias, allocation of suture material was according to time at recruitment of participant: from January to March 2014 women received a monofilament cerclage, and from April to July 2014 a braided cerclage was inserted. All cervical cerclages were performed by the same obstetrician at the same site, and all were electively removed between 36+0 and 36+6 weeks, unless spontaneous labour occurred prior to this.

2.6.2b Progesterone supplementation Following this, between August and December 2014 all recruits with a short cervix (≤25mm) were allocated to receive a vaginal progesterone pessary as a preventative intervention. Women were prescribed 400mg every night to continue until 34+0 weeks.

54 Materials and Methods

Figure 2-1 Diagram of the study structure, including recruitment cohorts, research time-points, procedures and data collection and insertion of preventative interventions for preterm birth risk.

55 Materials and Methods

Cervical assessment

At each research attendance a transvaginal scan was performed for ultrasound assessment of the cervix including cervical length, volume and vascularity indices. Women were asked to empty their bladders and lie in dorsal lithotomy. A Voluson E ultrasound scanner and a 2.8–10- MHz transvaginal transducer with a 146◦ field of view (General Electric Healthcare, USA) with both 2D and 3D/4D modalities was used. The transvaginal probe was inserted, with care not to exert undue pressure on the cervix. With the cervix occupying two thirds of the screen, 2D cervical length measurements were taken 192 (Figure 2-2A).

The scan mode was then changed to 3D and the region of interest defined using the volume box, with a set angle of 90°. A sagittal plane of volume acquisition sweep was performed. The 4D power Doppler mode was set, with preinstalled ultrasound settings: frequency 3–9 MHz, pulse repetition frequency 0.6 kHz, gain 32bD, power -4.0bD, wall motion filter ‘low 1’, with medium persistence to achieve high sensitivity and normal line density. A further sweep in the same plane was than performed for cervical vascularity. The acquired data was stored on external hard drive for later analysis using the VOCALTM (Virtual Organ Computer-aided AnaLysis) software programme (Figure 2-2B).

Figure 2-2 Transvaginal ultrasound scans demonstrating (A) 2D cervical length measurement and (B) 3D cervical volume indicating cervical blood vessels

56 Materials and Methods

2.7 Virtual Organ Computer-aided AnaLysis (VOCAL) technology

Virtual Organ Computer-aided AnaLyis (VOCALTM, GE Medical systems) technology, is considered the most reliable and reproducible tool for the assessment of cervical volume (CV) in pregnancy 91. VOCALTM uses three-dimensional ultrasound, and is more accurate than standard 2D volumetric measurements (calculated from cervical length and diameter) 193, 194. For cervical volumetric analysis the contours of each sectional plane are manually drawn, rotating the cervix through 30° rotations. Three-dimensional cervical vascular trees were constructed, and the histogram facility was used to calculate indices of vasculature; Vascularity index, VI and blood flow index, FI.

Figure 2-3 A volumetric cervical scan with a cerclage in situ, demonstrating the planes of data acquisition for volumetric reconstruction.

57 Materials and Methods

2.8 Statistical analyses

Longitudinal data for cervical length (CL), cervical volume (CV), vascularity index (VI) and blood flow (FI) were compared in women experiencing preterm (<37+0 weeks) and term (≥37+0 weeks) birth.

Statistical tests for significance included one way ANOVA with Bonferroni multiple comparison post-test, two tailed t-tests, and fisher exact tests where appropriate. Linear regression assessed for correlation between cervical parameters (CL, CV, VI and FI) and gestational age at birth.

Receiver operator curves where used to assess CV, VI, and FI parameters for accuracy in prediction of preterm birth <34 weeks and <37 weeks. These predictions were compared to the predictive accuracy of CL measurements as taken routinely in preterm surveillance clinics.

58 Materials and Methods

Vaginal Microbiota: 16S ribosomal RNA gene sequencing

For 16s rRNA gene profiling, vaginal samples stored at -80◦C were defrosted and bacterial DNA isolated from the cervicovaginal fluid (CVF). Hypervariable V1-V3 regions of the 16S rRNA gene were amplified and sequenced.

2.9 Materials

- Liquid Amies swab (BBL™ CultureSwab™, Becton, Dickinson and Company) - Glass beads (Sartorius Stedim Cat No BBI-8541400) - Micro Dismembrator (Sartorius)

QIAamp DNA Mini Kit (Qiagen Catalogue No 51304)

Enzymes - Lysozyme (Sigma L6876 – chicken egg white). Prepared for concentration 10mg/ml in filter sterilised 10mM Tris.HCl pH 8.0 - Mutanolysin (Sigma M9901/10KU). Prepared for concentration 25U/μl by dissolving 10,000units in 400 μl sterile water - Lysostaphin (Sigma L9043). Prepared for concentration 4000U/ml by dissolving 23,915units in 20mM NaOAc

Buffers and solutions - Filter sterilised PBS (phosphate-buffered saline) - NaOAc (sodium acetate) - Tris.HCl pH 8.0 10mM - TE50 (10 mM Tris–HCl, 50 mM EDTA, pH 8.0) - Ethanol (100%) - DNase / RNase free water

59 Materials and Methods

Enzyme cocktail mix (per sample) 130μl PBS (filter sterilised) 50μl Lysozyme (10mg/ml) 6μl Mutanolysin 3μl Lysostaphin 41μl TE50

Polymerase chain reaction (PCR) NEB OneTaq® DNA Polymerase (BioLabs) dNTP 10mM (Sigma cat D7295) 5 x Buffer (BIO-37045) 6 x loading dye – PROMEGA G190A 1006034 Agarose (Electrophoresis grade from Invitrogen cat no 15510-019) SYBR Safe DNA Stain (Invitrogen Cat No S33102) Molecular weight marker (Bioline Hyperladder 100bp Cat No BIO-33056) TBE buffer (Tris/Borate/EDTA)

5 x NEB Buffer 10μl 5 x Buffer 1μl 10mM DNTP 1μl Universal Forward Primer 10picomol/μl 1μl Universal Reverse Primer 10picomol/μl 0.25μl NEB one Taq Polymerase 10picomol/μl 31.75μl H20 (make up to 50 μl) 5μl DNA

Forward Reverse

Universal 195 GCCTTGCCAGCCCGCTCAGTCAGAGTTTGATC GCCTCCCTCGCGCCATCAGACACACTGCATGCTGCCTC Primers CTGGCTCAG CCGTAGGAGT

60 Materials and Methods

2.10 Isolation of DNA from vaginal samples

Swabs were thawed on ice and re-suspended in Amies Liquid transport medium. Debris was removed by centrifuging at 7000 x g 10 min. The supernatant was aspirated and the pellet resuspended in 230μl Enzyme cocktail mix. Cells were transferred to a sterile DNase/RNase free 2ml tube where an enzymatic lysis step was carried out for 1h at 37°C incubation. Samples underwent mechanical disruption through oscillation with acid washed glass beads at 1000 x g for 1 min using a Mikro-Dismembrator (Sartorius UK Ltd, Surrey, United Kingdom). The resulting lysate was further processed and purified using QIAamp DNA Mini Kit (Qiagen, Manchester, UK). Briefly as per manufacturers guidelines, 20μl of Proteinase K and 200μl lysis AL buffer was added per 200μl of lysate, and incubated at 56°C for 30 min. 200μl Ethanol was added per sample, the lysate transferred to the QIAamp Spin Column, taking care to avoid aspirating the beads. The columns were centrifuged at 7000 x g for 1 min, then washed with 500μl AW1 buffer at 7000 x g for 1 min and with 500μl AW2 buffer at 18 600 x g for 3 min. Finally the DNA was eluted in 100μl AE buffer at 7000 x g for 1 min.

2.10.1 Polymerase chain reaction (PCR) amplification Following extraction, presence of bacterial DNA was confirmed by PCR amplification using the universal forward and reverse primers 195. Gel electrophoresis was performed using 1-1.5% Agarose gels. For this, agarose was microwaved in 1x TBE buffer until dissolved. Sybr Safe gel stain was added (1μl per 100mls Agarose gel solution) and the gel was allowed to set. Loading dye was added to each DNA extract and a molecular weight marker was included on each gel. In each gel a positive control and two negative controls (one blank swab, and one pure water) was included. PCR products were electrophoretically separated at 140 V for 10 to 20 min. The gel was transilluminated and photographed with UV light (Figure 2-4).

61 Materials and Methods

Figure 2-4 PCR gel demonstrating molecular weight marker, detectable DNA amplified in samples 1 – 24, and controls (+ve = positive, -ve1 =water, -ve2 =blank swab). This was performed for all samples prior to 16S rRNA gene sequencing to ensure the presence of extracted DNA.

2.11 DNA sequencing

Forward and reverse fusion primers were used to amplify the V1-V3 hypervariable regions of 16S rRNA genes in preparation for sequencing on an Illumina MiSeq platform (Illumina, Inc. San Diego, California). The forward primer was made up of an Illumina i5 adapter (5′-3′) (AATGATACGGCGACCACCGAGATCTACAC), 8 bp barcode, primer pad (Forward: TATGGTAATT), and the 28F-GAGTTTGATCNTGGCTCAG primer 196. The reverse fusion primer consisted of an (5′-3′) Illumina i7 adapter (CAAGCAGAAGACGGCATACGAGAT), 8 bp barcode, primer pad (Reverse: AGTCAGTCAG), and the reverse primer (519R- GTNTTACNGCGGCKGCTG) 197. Sequencing was conducted at Research and Testing Laboratories LLC (Texas, USA).

62 Materials and Methods

2.12 Statistical analyses

2.12.1 Mothur and MiSeq pipeline Sequence data was processed and analysed using the MiSeq SOP Pipeline of the Mothur package 198 with the Silva bacterial database (www.arb-silva.de/) used for sequence alignment. Sequence classification was performed using the RDP database reference sequence files and the Wang method 199 and taxonomy assignments determined using the RDP MultiClassifier script and USEARCH with 16S rRNA gene sequences from the cultured representatives from the RDP database 200 for species level taxonomies. Data was re-sampled and normalized to the lowest read count in Mothur (n=725). The Shannon index was used to analyse the alpha diversity (i.e. diversity of bacterial species within individuals)

2.12.2 Re-sequencing A total of 621 swabs were selected for bacterial DNA sequencing. Due to the large number of samples this was performed across two sequence runs. To test reproducibility between sequence runs, 15 samples from each run were re-sequenced together on a third sequence run and the resulting data compared (Appendix 10.2)

63 Materials and Methods

2.12.3 Statistical analysis of vaginal microbial communities The software programme Statistical Analysis of Metagenomic Profiles (STAMP) was used to determine differences in microbial profiles between normal and high risk pregnancy groups, and the impact of interventions (cerclage or progesterone) on longitudinal profiles 201. Principal components analysis (PCA) was initially used as unsupervised multivariate method to provide an overview of the degree of microbial variance between samples among the cohorts. Ward linkage hierarchical clustering analyses was then performed using genera and species taxonomic data with a clustering density threshold of 0.75. Genera sequence data classified individual samples into a ‘Normal’, ‘Intermediate’, or ‘Dysbiotic’ microbiome according to lactobacillus abundance; >90%, 30-90% and <30% lactobacillus respectively. Hierarchical clustering of species data assigned samples into community state types (CSTs) consistent with those described by Ravel et al155: CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), CST IV (diverse species) and CST V (L. jensenii).

Microbial profiles (at genera and species level) were assessed for differences according to ethnicity, cervical length, gestational age at sampling and outcomes of outcomes of early preterm birth <34+0 weeks, late preterm birth 34+0 to 36+6 weeks, and term birth ≥37+0 weeks. Proportions of sequences and mean species abundance among groups were tested using Fisher’s exact test, Welch’s t-test, Kruskal-Wallis, with Dunn’s and Bonferroni multiple testing corrections where appropriate. Accuracy parameters for predictions of preterm birth: sensitivity (sens), specificity (spec), positive predictive values (PPV) and negative predictive values (NPV) were calculated based on the number of samples assigned to CSTs.

2.12.4 LEfSe analysis Linear discriminative analysis (LDA) Effect Size (LEfSe) 202 analysis was used to identify discriminating features from differing taxonomic levels (phylum to species) characteristic of the differences between cerclage suture material types. This allows for identification of differentially abundant features that emphasise both statistical significance and biological relevance.

64 Materials and Methods

Quantitative PCR

2.13 Materials

- 1x SYBR Green Jumpstart Taq Ready Mix (Sigma-Aldrich) - 5ul of bacterial DNA isolated from the vaginal swabs - 0.8μM final concentration of forward and reverse primers

Oligonucleotide Forward Reverse primers

A. vaginae203 5’-TAGGCGGTTTGTTAGGTCAGGA-3’ 5’-CCTACCAGACTCAAGCCTGC-3’ G. vaginalis204 5’-GGAAACGGGTGGTAATGCTGG-3’ 5’-CGAAGCCTAGGTGGGCCATT-3’

Quantitative PCR Due to potential limitations in the V1-V3 primers used for the MiSeq 16s rRNA gene sequencing, detection of specific bacterial species associated with bacterial vaginosis (BV), namely Atopobium vaginae and Gardnerella vaginalis, was performed using targeted quantitative PCR. Assays were SYBR green based and performed on Applied Biosystem’s StepOnePlus. The thermocycle profile used was 95ºC for 2 min, followed by 40 cycles at 95ºC for 15 sec and 65 ºC for 1 min. For both A. vaginae and G. vaginalis assays, vaginal samples and corresponding standards (A. vaginae and G. vaginalis DNA) were run in duplicates and the average numbers were used to calculate 16S rRNA gene copies per 5μl of vaginal DNA.

65 Materials and Methods

Assessment of cytokines in cervico-vaginal secretions

To assess the impact of cerclage insertion on cervicovaginal inflammation, vaginal swabs collected immediately prior to cerclage insertion, as well as those taken at the first sampling time-point post-cerclage (4 weeks), were assessed for inflammatory cytokine concentrations.

2.14 Magnetic Luminex Screen Assay for Cytokines

A Human Magnetic Luminex Screen Assay (Luminex Corporation, Austin, Texas) was used to detect 15 pre-specified analytes in the cervicovaginal fluid. A multiplex bead assay was employed to provide a quantitative assessment of a large number of target analytes simultaneously. Traditional cytokine assays using ELISA (enzyme-linked immunosorbent assays) rely upon an immobilised antibody-coated surface and are limited to detection of single analyte. In comparison, Luminex assay captures analyte ligands onto antibody-coated beads in suspension, and is therefore advantageous as it requires smaller volumes to simultaneously test multiple cytokines from a single sample. Luminex is also considered more sensitive than ELISA especially for samples with a low concentration of target-analyte 205.

2.14.1 Analytes Target analytes were selected according to evidence of involvement in inflammatory change related to preterm birth, cervical ripening, and angiogenesis. A justification of each analyte included on the assay is provided in the table below.

66 Materials and Methods

Target analyte Relevant biological function

Granulocyte colony- A pro-inflammatory cytokine that is upregulated in the cervix during labour206 and is considered a possible mediator of stimulating factor (G-CSF) preterm birth 207.

Granulocyte-macrophage Plays an important role in angiogenesis through induction of 208 colony-stimulating factor VEGF expression and has shown to be elevated in CVF prior to cervical shortening 209. (GM-CSF) Interferon (IFN)-γ A cytokine central to innate adaptive immunity, it is primarily produced by NK and T cells, and has been shown to be elevated in preterm labour 45

Interleukin (IL)-1β Expression is up-regulated in the cervix, myometrium and amnion, driving COX-2 and prostaglandin production, during term and infection induced preterm labour 19.

IL-2 A pro-inflammatory cytokine that is upregulated in pregnancy and plays a significant role in T cell proliferation and in promotion of angiogenesis 210

IL-4 An anti-inflammatory cytokine produced by the placenta and amnion, is induced by progesterone and inhibits Th1 pro- inflammatory milieu during pregnancy211.

IL-6 A well characterised pro-inflammatory cytokine central to cervical remodelling 38. It is considered highly specific, but not sensitive for preterm labour when detected in CVF during pregnancy 212.

IL-8 Promotes cervical ripening through recruiting activated cells (particularly neutrophils) to the cervical extracellular matrix. Elevated IL-8 in CVF in pregnancy is associated with a reduction in Lactobacillus spp. and infection-induced preterm birth 180, 213

IL-10 An anti-inflammatory cytokine detectable in cervicovaginal fluid, it inhibits excessive inflammation in pregnancy through down regulation of pro-inflammatory cytokines including IL-8, IL-6, TNFα, IL-1β, and matrix-metalloproteinases 38, 206

67 Materials and Methods

Intercellular Adhesion A cell surface glycoprotein expressed by the vascular endothelium, macrophages and lymphocytes, and Molecule 1 (ICAM-1) upregulated by inflammatory cytokines including IL-1β and TNF-α, in response to inflammation. Elevated serum levels in pregnancy has been associated with preterm birth 214

Matrix metalloproteinase An interstitial fibroblast collagenase that breaks down the extracellular matrix in the cervix as part of cervical 1 (MMP-1) remodelling in parturition 215

Monocyte chemotactic Promotes migration and infiltration of monocytes/ macrophages to the cervix, is elevated in preterm birth 206 protein (MCP)-1

Regulated on Activation A versatile chemokine, it regulates the inflammatory response Normal T Expressed and to bacterial invasion in many organs, and is increased in Secreted/Chemokine amniotic fluid in preterm labour 216 ligand 5 (RANTES /CCL5) Tumour necrosis factor A pro-inflammatory cytokine commonly involved in systemic inflammatory responses to infection, is also central to (TNF)-α parturition. In conjunction with the actions of IL-1 and IL-6 is stimulates arachidonic acid thereby increasing prostaglandin production, and cervical remodelling 217

Vascular endothelial A protein that stimulates angiogenesis and vasculogenesis in response to tissue inflammation and hypoxia. In the cervix, it growth factor (VEGF) is important for the vascular remodelling that occurs as part of term and preterm labour 41, 208

68 Materials and Methods

2.15 Analyte extraction

2.15.1 Materials - PBS - Protease inhibitor (Sigma- Aldrich P8340) - One Flat-bottom 96-well Microplate - Foil plate sealers - Luminex Human Premised Multi-analyte Kit (Catalog Number LXSAHM)

2.15.2 Methods

Collected CVF was suspended by vortexing swabs with 350μl of standard PBS solution supplemented with protease inhibitor. The suspended CVF and swab was centrifuged at 400 x g for 3 min, the supernatant collected into a new 2ml eppendorf before being centrifuged again 400 x g for 5 min to remove any cell debris. Cell-free supernatants were then analysed by Human Magnetic Luminex Screen Assay (15-plex) (Luminex Corporation, Austin, Texas) with a Bioplex®200system (Biorad technologies).

As per manufacturer guidelines, standards were serially diluted by reconstituting 100μl of each of 15 standard cocktails (unique to the pre-specified analytes) with Calibrator Diluent RD6-52 to create a standard curve. Samples were run in duplicates across two 96-well plates, including six standards and two blanks. The first plate was a 1 in 1 dilution and the second plate, 1 in 50 dilution to ensure detection of analytes within the standards ranges as specified by Luminex Human premixed analyte kit.

To each well of a 96 well microplate, 50μl of cell-free supernatant and 50μl of the magnetic bead cocktail (beads coated with 15 pre-specified analyte-specific capture antibodies) was added. The plates were securely covered with a foil plate sealer, and incubated for 2 h at room temperature on a horizontal orbital microplate shaker set at 800 rpm. The microplate was washed 3 times by applying the magnet securely at the bottom of the microplate, the supernatant was removed, and 100μl of Wash Buffer added and removed for each of the three washed with the aid of the magnetic microplate. A total of 50μl of the Biotin Antibody Cocktail was then added to each well before the plate was securely covered and incubated for 1 h at room temperature on the shaker set at 14 x g (800 rpm). Plates were again washed as

69 Materials and Methods previously described before 50μl of Streptavidin-PE was added to each well to bind to the biotinylated antibody. The plate was again securely covered and incubated on the shaker at 14 x g for 30 min at room temperature. Any unbound Streptavidin-PE was removed at a further wash step. Magnetic beads in each well were then resuspended in 100μl of Wash Buffer and the plate incubated for 2 min on the shaker set at 14 x g.

Plates were then read on a Biorad Bioplex®200system where concentrations of the analytes were determined through analysis of the spectral properties of the beads and the amount of fluorescence emitted. The standard ranges of the two plates were compared for consistency across runs (Figure 2-5).

Sample concentrations below the limit of detection for the assay were excluded from further analysis. Where the analytes were detectable, but outside their specified standard range concentrations were calculated by extrapolation of the standard curve. Only analytes detectable in the standard range in at least 10% of samples were further analysed.

70 Materials and Methods

Figure 2-5 Comparison of the standard curves from first and second multiplex Luminex plates of 1 in 1 and 1 in 50 dilutions. Comparisons of the standard curves calculated from immunofluorescence demonstrate highly reproducible results.

71 Materials and Methods

2.16 Statistical analyses

Analyte concentrations were compared for differences in expression pre- and 4 weeks post- cerclage insertion using Wilcoxon signed ranked test. The Mann-Whitney test was applied to examine differences among suture material (braided versus monofilament) at 4 weeks post cerclage. Fold change in analyte expression was calculated and plotted as a volcano plot comparing significant differences pre- and post-cerclage according to suture material. A P-value of <0.05 was considered statistically significant.

Analyte concentrations were correlated with genera level classification of samples into normal, intermediate and severe. The Mann Whitney test was used to assess differences in the mean detection of cytokine expression between “normal” and “dysbiotic” samples.

72 Ultrasound assessment of the cervix in pregnancy

3 ULTRASOUND ASSESSMENT OF THE CERVIX IN PREGNANCY

73 Ultrasound assessment of the cervix in pregnancy

Chapter abstract

Hypothesis The onset of parturition involves cervical remodeling. This is clinically evident as the cervix softens and enlarges towards the third trimester of pregnancy. Measurement of cervical length (CL) before 24 weeks can detect premature cervical shortening and is currently used as a tool to identify pregnancies at risk of preterm birth. This study hypothesizes that the predictive accuracy of CL may be improved through the addition of measurements of cervical volume (CV), indices of vascularity (VI) and blood flow (FI).

Aims 1. Establish a reference range of CV, VI and FI in healthy pregnancy and compare these to pregnancies at risk of preterm birth. 2. Determine the predictive values of CV, VI and FI before 24 weeks as a screening test for preterm birth.

Methods Healthy pregnant women and those attending prematurity surveillance clinics for their risk of preterm birth were prospectively recruited at 12 weeks and followed up longitudinally at 16, 22, 28 and 34 weeks gestation. At each time-point a transvaginal scan was performed using 2D and 3D/4D modalities on a Voluson E8 ultrasound machine. Data collected on cervical length (CL), volume (CV) and vascularity indices (VI and FI) were analysed using Virtual Organ Computer-aided AnaLysis (VOCALTM) software.

Results After loss to follow up, 295 women consented to longitudinal transvaginal scans, including 96 healthy pregnancies and 199 at risk of preterm birth. CL decreased with advancing gestation from 12 to 34 weeks, while CV, VI and FI all increased (P<0.01). Compared to term births, women who delivered preterm (<37 weeks, n=43, 22%) had higher CV, VI and FI, most significantly from 28 weeks. CV, VI and FI predicted preterm birth <34 weeks better than birth <37 weeks. Overall predictive accuracies of CV, VI and FI did not substantially improve current CL screening performance.

Conclusion Although prediction of preterm birth is comparable to CL, the technical complexity and increased time required for acquisition of CV, VI and FI measurements limits their use for preterm birth surveillance in current clinical settings.

74 Ultrasound assessment of the cervix in pregnancy

Introduction

The uterine cervix serves as a mechanical and chemical barrier during pregnancy, protecting the uterus and fetus from ascending vaginal infection 35. In normal pregnancy, advancing gestation is associated with inflammatory-driven collagenous remodelling of the cervix in preparation for labour 36, 37. In conjunction with collagenous remodeling, cytokine activation recruits vascular endothelial growth factor (VEGF) to the cervix, simultaneously initiating cervical angiogenesis 40. In a positive feedback loop, VEGF increases expression of the cytokine IL-6, which in turn recruits VEGF to the cervix, thereby accelerating vascular remodelling 36, 41. These changes are detectable at transvaginal ultrasound (TVS), where cervical length (CL) decreases and volume and vascularity increase with gestational age and parity 92-94.

In the case of spontaneous preterm birth, pathological triggers of inflammatory driven pathways can induce premature cervical ripening. Based on work in the 1990’s by Iams et al 83 and later Heath et al 84, it was established that premature cervical ripening may be identified as shortened CL at TVS, many weeks or even months prior to the onset of clinical symptoms of preterm birth. Second trimester CL measurements have since been established as a reliable and predictive tool for preterm birth, and are now frequently employed for preterm birth surveillance in clinical practice88. Currently pregnant women found to have a short cervix measuring ≤25mm at TVS before 24 weeks gestation are considered at highest risk of preterm birth 83.

Recently, the addition 3D/4D power doppler software to TVS has enabled further cervical parameters including cervical volume (CV) and vascularity to be quantified in pregnant and non- pregnant women. Using the histogram facility of Virtual Organ Computer-aided AnaLysis (VOCAL)TM two indices of vascularity are frequently used; the vascularization index (VI) and blood flow index (FI). Vascularisation index (VI) indicates the density of blood vessels as a percentage within volume of interest. It is calculated as a ratio of colour-coded voxels (smallest unit of volume) to all voxels. Flow index (FI) describes the signal intensity of the color voxels, reflecting the rate of blood flow within the vessels 96. Importantly, these VOCALTM indices have been demonstrated to be reliable and reproducible 97, 98.

Although CV, VI and FI have been examined in healthy pregnancy 100, 101, the few studies reporting on their prediction for preterm birth are mostly limited to observational cross-sectional

75 Ultrasound assessment of the cervix in pregnancy cohorts in low-risk pregnancy 92, 93, 100. In 2011, Park et al 93 screened a cross-section of 391 low-risk pregnancies at 20-24 weeks and demonstrated that a threshold of CV <20cm3 carries similar predictive accuracy for preterm birth as CL ≤25mm 93. Similar results were reported by Barber et al 92 soon after. In 2014, Diego et al 95 were the first group to study CV in a population considered at-risk of preterm birth with a short CL ≤25mm. They reported that those symptomatic of preterm labour had a lower CV compared to the asymptomatic group 95. The only two studies to investigate longitudinal changes in cervical volume were performed in low- risk pregnancies 100, 101 and neither reported significant improvement in prediction of preterm labour using CV over CL.

As with cervical volumetric studies in pregnancy, published literature on 3D/4D power Doppler angiography of the cervix is limited and conflicting 95, 99-101. While Basgul et al 100 found that cervical vascularity was not affected by advancing gestational age, Yilmaz et al 101, reported cervical vascularity increases from the 2nd trimester onwards, most notably among parous women. The latter study was an analysis of 111 low-risk pregnancies, of whom 9% delivered preterm with no significant variation in vascularity when compared to term births. In a more recent study of 70 high-risk women, De Diego et al 95 reported an elevated VI in women symptomatic of preterm labour after 24 weeks. To date, no study has performed a longitudinal characterisation of CV, VI and FI comparing low- and high-risk pregnancies. Furthermore the predictive accuracy of these parameters for preterm birth have not been assessed in comparison to current CL measurements.

76 Ultrasound assessment of the cervix in pregnancy

Aims 1. To characterise CV, VI and FI in healthy pregnancy and compare these to pregnancies at risk of spontaneous preterm birth. 2. To establish the predictive accuracy of CV, VI and FI for subsequent preterm birth, and compare these to current screening by way of CL measurements.

Hypothesis This study hypothesised that CV, VI and FI increase with advancing gestational age as the cervix undergoes cervical remodeling and angiogenesis. A premature increase in volume and vascularity precedes spontaneous preterm delivery, effectively differentiating low- and high-risk pregnancy cohorts. Furthermore these parameters, if effectively quantifiable, should improve the predictive value of current CL screening.

77 Ultrasound assessment of the cervix in pregnancy

Study design

Patient recruitment and sample collection

Pregnant women attending prematurity surveillance clinics for their preterm birth risk across two tertiary London maternity units between January 2013 and December 2014 were prospectively recruited at their first clinic attendance (approximately 12 weeks gestation). A healthy control group were recruited from low-risk midwifery-led antenatal clinic at their booking appointment. Both groups were followed up longitudinally at 16, 22, 28 and 34 weeks gestation. At each time- point a transvaginal scan was performed for measurements of cervical length (CL), cervical volume (CV) and indices of vascularisation (Vascularisation Index, VI and Flow index, FI), calculated using the VOCALTM (Virtual Organ Computer-aided AnaLysis) software programme (as described in the Methods chapter).

Metadata collected for all participants included age, BMI, ethnicity, gestation age at sampling and subsequent interventions for preterm birth. Exclusion criteria for study participation included multiple pregnancy (twins or higher order), uterine anomalies and iatrogenic preterm birth <37 weeks. Eligibility criteria for the high-risk group included a previous spontaneous preterm birth <37weeks of a singleton pregnancy immediately prior to the index pregnancy, or excisional conization including LEEP, LLETZ, CKC for CIN grade II or III. For the duration of the study, both units employed a policy of continued CL screening, with indication for intervention being a CL ≤25mm at TVS measured at <24 weeks gestation. Participation in this study did not influence subsequent clinical care or dictate preventative interventions, namely cervical cerclage or vaginal progesterone supplementation, for perceived preterm birth risk.

78 Ultrasound assessment of the cervix in pregnancy

Figure 3-1 A volumetric cervical scan of a shortened cervix with evidence of cervical funnelling. The planes of data acquisition for volumetric reconstruction are demonstrated.

Figure 3-2 A cervical scan with power doppler demonstrating planes of data acquisition for vascularity indices.

79 Ultrasound assessment of the cervix in pregnancy

Statistical analyses of scan and sequence data Cervical length (CL), cervical volume (CV) and vascularity indices (VI and FI) were assessed for the effect of advancing gestational age at measurement, differences between high- and low-risk cohorts and outcomes of term (≥37weeks) and preterm (<37 weeks) birth. Statistical tests for significance included Fisher’s exact test for uneven, small sample sizes of categorical variables, with chi-squared tests where data were of sufficient power. Two-tailed t-tests compared continuous variables and one-way ANOVA with Bonferroni post-test was used for multiple comparisons where applicable. Linear regression was used to assess correlation between cervical parameters (CL, CV, VI and FI) and gestational age at birth. Receiver operator curves assessed the predictive accuracy of CV, VI, and FI for preterm birth <34 weeks and <37 weeks. These predictions were compared to the predictive accuracy of CL measurements as taken routinely in preterm surveillance clinics.

80 Ultrasound assessment of the cervix in pregnancy

Results

3.1 Recruitment

A total of 315 pregnant women were prospectively recruited. After 6% loss-to-follow up (20/315), a remaining total of 295 women participated in the study. This included 96 healthy low-risk pregnancies and 199 high-risk pregnancies (Table 3-1). Transvaginal scans were collected at five antenatal screening time-points: 12, 16, 22, 28 and 34 weeks gestation. Of these, 53% (n=167) also consented to vaginal sampling for microbial assessment by 16S rRNA gene sequencing as discussed in Chapter 4, Results 2 (Table 3-1).

Table 3-1 Prospective recruitment for transvaginal scans and vaginal swabs. Total population Low-risk High-risk Recruitment n/N % n/N % n/N %

Initial recruitment 315 108/315 34% 207/315 66%

Loss to follow up/ miscarriage <13weeks 20/315 6% 12/108 11% 8/207 4%

Final recruitment Consented to: Transvaginal scan 295/315 94% 96/108 89% 199/207 96% only Transvaginal scan 167/315 53% 34/108 31% 133/207 64% & vaginal swabs

Low-risk = healthy pregnancy with no known risk factor for preterm birth; High-risk = pregnancy with a risk factor for preterm birth

81 Ultrasound assessment of the cervix in pregnancy

3.2 Participant demographics

Patient characteristics of the high- (n=199) and low-risk (n=96) cohorts, and their delivery outcomes are provided in Table 3-2. Age and BMI were comparable among groups, but there were more Caucasians (67% vs. 54%, P=0.04) and primps (59% vs. 38%; P=0.001) as well as fewer Black women (14% v 31%; P=0.001) and smokers (6% vs. 12%; P=0.2) in the low-risk compared to high-risk groups. A cerclage was inserted for cervical shortening in 43% of the high-risk women (85/199). Overall, mean birth gestation was 33+6 weeks (SD± 3+6 weeks, range 24+4 – 36+6 weeks). Preterm birth <37 weeks occurred in 22% (43/199) of the women, of which 8% (15/199) delivered early preterm (<34+0 weeks), and 14% (28/199) late preterm (34+0 to 36+6 weeks, Table 3-2)

Table 3-2 Characteristics of low and high-risk pregnancies recruited to transvaginal scan with intervention and delivery outcomes Total population Low-risk High-risk

N=295 N=96 N=199 Age (years) Mean ±SD 32.9 ±5.2 32.7 ±5.2 33.0 ±4.6 BMI Mean ±SD 24.7 ±4.3 24.4 ±4.3 24.8 ±4.2

Ethnicity, n/N % Caucasian 171/295 58% 64/96 67% 107/199 54%* Asian 49/295 17% 19/96 20% 30/199 15% Black 75/295 25% 13/96 14% 62/199 31%**

Parity, n/N % Para 0 133/295 45% 57/96 59% 76/199 38%** Para ≥ 1 162/295 55% 39/96 41% 123/199 62%

Smoker, n/N % 29/295 10% 6/96 6% 23/199 12%

Cerclage Intervention, n/N % 85/295 29% n/a 85/199 43%

Early PTB, <34+0 w 15/295 5% 0 0% 15/199 8% Gestation at Late PTB, 34+0 to <37+0 w 28/295 9% 0 0% 28/199 14% birth, n/N % +0 Total PTB <37 w 43/295 15% 0 0% 43/199 22% Term ≥37+0 w 252/295 85% 96 100% 156/199 78% *P<0.05, **P<0.01; two tailed Fisher exact test. Low-risk = healthy pregnancy with no known risk factor for preterm birth; High-risk = pregnancy with a risk factor for preterm birth, PTB = preterm birth, w= weeks.

82 Ultrasound assessment of the cervix in pregnancy

3.3 Correlation of cervical length, volume and vascularity

Linear regression was used to assess correlations among cervical parameters, length (CL), volume (CV), vascularisation (VI) and flow (FI) indices. CL was increasingly positively correlated with CV (P<0.05; Figure 3-3A) as gestational age advanced from 12 to 34weeks (12 weeks: r=0.2, P=0.02, 34 weeks: r=0.5, P<0.001, Figure 3-3A). Cervical vascularisation indices, VI (Figure 3-3B, D) and FI (Figure 3-3C, E) did not correlate significantly with either CL or CV.

83 Ultrasound assessment of the cervix in pregnancy

A Cervical length (CL) v Cervical Volume (CV) 120 Screening gestation

) 100 (weeks) 3

80 12w 16w 60 22w 40 28w 34w

Cervical volume (cm 20

0 0 20 40 60 80 Cervical length (mm) Cerivcal length (CL) BCVI FI 20 60

50 15

40 10 30

5 FI Index, Flow 20 Vascularisation Index, VI

0 10 0 20 40 60 80 0 20 40 60 80 Cervical length (mm) Cervical length (mm)

Cerivcal volume (CV) DEVI FI 20 60

50 15

40 10 30

5 FI Index, Flow 20 Vascularisation Index, VI Vascularisation Index,

0 10 0 20 40 60 80 0 20 40 60 80 Cervical volume (cm 3) 3 Cervical volume (cm ) Figure 3-3 The correlation between cervical length and volume improves with advancing gestation, while neither correlate with cervical vascularity. (A) CL positively correlates with CV from 12 weeks (r=0.2, P=0.02) to 34 weeks (r=0.5, P<0.001). Cervical length does not correlate with cervical vascularity (B) or blood flow (C) irrespective of gestation at screening. Similarly vascularity indices do not correlate with cervical volume (D, E) (r= Pearson correlation, CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, w= weeks gestation at cervical screening)

84 Ultrasound assessment of the cervix in pregnancy

3.3.1 Gestation at screening and gestation at birth

Correlation between CL and gestation at birth improves with later gestation at measurement (Figure 3-4A). Correlation at 12 weeks (r=0.07, P=0.44), 16 weeks (r=0.15, P=0.09) and 22 weeks (r=0.02, P=0.08) do not reach significance, however positive correlations are observed at 28 weeks (r=0.29, P=0.003) and 34 weeks (r=0.30, P=0.001). There is no correlation between CV, VI and FI and gestation at birth, irrespective of gestation at screening (Figure 3-4 B, C, D).

Gestation at screening 12w 16w 22w 28w 34w

A CL B CV

60 100 )

3 80 40 60

40 20

Cervical length (mm) 20 Cervical volume (cm

0 0 22 24 26 28 30 32 34 36 38 40 42 22 24 26 28 30 32 34 36 38 40 42 Gestation at birth (weeks) Gestation at birth (weeks)

CDVI FI 20 60

15 50

10 40

5 30 Vascularisation Index, VI Vascularisation Index, VI

0 20 22 24 26 28 30 32 34 36 38 40 42 22 24 26 28 30 32 34 36 38 40 42 Gestation at birth (weeks) Gestation at birth (weeks) Figure 3-4 Linear regression demonstrates that the correlation between cervical length measurement and gestation at birth is dependent on gestation at screening, reaching significance at 28 weeks (A) CL and birth gestation at 12 weeks, r=0.07 (P=0.44), 16 weeks, r=0.15 (P=0.08), 22 weeks, r=0.02 (P=0.09), 28 weeks r=0.29 (P=0.003) and 34 weeks r=0.30 (P=0.001). Gestation at screening does not significantly affect the correlation between cervical volume (B), vascularity (C), blood flow (D) and gestation at birth. (CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, r= Pearson correlation co-efficient)

85 Ultrasound assessment of the cervix in pregnancy

3.4 Cervical length and volume

Longitudinal change in CL and CV was assessed from 12 to 34 week screening time-points, initially in 96 healthy pregnancies. These data were used as baseline measurements and considered low-risk controls for later comparison to high-risk recruits. Figure 3-5 demonstrates trends in CL and CV among healthy control pregnancies. These results showed that the cervix shortens in length with advancing gestation, while a corresponding increase in overall cervical volume is observed: mean CL decreased from 45mm (±8.6) at 12 weeks to 32mm (±7.1) at 34 weeks (P<0.001; ANOVA), and CV increased from 29.2cm3 (±8.7) to 54cm3 (±20.8) cm3 over the same period (P<0.001; ANOVA; Table 3-3; Figure 3-5)

Table 3-3 Mean cervical length, volume, vascularisation and flow indices at cervical at longitudinal screening time-points in 96 healthy pregnancies. 3 Scan CL (mm) CV (cm ) VI FI time-points Mean ±SD Mean ±SD Mean ±SD Mean ±SD 12w 45 ±8.6 29.2 ±8.7 1.9 ±1.9 28.9 ±5.4 16w 40 ±5.2 36.0 ±13.3 3.0 ±2.4 30.8 ±5.3 22w 39 ±4.3 45.6 ±15.9 3.6 ±2.8 31.5 ±6.2 28w 34 ±5.7 51.0 ±17.7 3.9 ±3.3 31.9 ±6.5 34w 32 ±7.1 54.2 ±20.8 4.1 ±3.5 32.5 ±4.7 CL = cervical length, CV= cervical volume, VI = vascularisation index, FI = flow index w= weeks, SD = standard deviation

86 Ultrasound assessment of the cervix in pregnancy

3 A Cervical length (mm) Cervical Volume (cm ) 50 60

45 50

40 3) 40 35 CL (mm) CV (cmCV

30 30

25 20 12w 16w 22w 28w 34w Scan timepoints (weeks)

BCANOVA ANOVA *** 120 *** 100 *** *** *** *** ns *** 100 80

80

60 3) 60

40 (cmCV CL (mm) 40

20 20

0 0 12w 16w 22w 28w 34w A 12wmem B mem 16wC mem 22wD 28wmem E 34wmem A VOL b VOL c VOL d VOL e VOL

Scan timepoints (weeks) Scan timepoints (weeks)

Figure 3-5 (A) Mean (±SEM) cervical length (CL, mm) and cervical volume (CL, cm3) in healthy pregnancy at low-risk of preterm birth (n=96) at scan time-points 12 to 34 weeks. (B) Cervical length decreases from 12 weeks (mean CL 45mm) to 34 weeks (CL 32mm; P<0.001), while cervical volume increases (29cm3 to 54cm3, P<0.001; C). (***P<0.001; 1 way ANOVA, Bonferroni post-test.)

87 Ultrasound assessment of the cervix in pregnancy

3.5 Cervical vascularity

Cervical vascularisation (VI) and blood flow (FI) were shown to increase with advancing gestation in healthy pregnancy (P<0.001; ANOVA; Table 3-3, Figure 3-6). VI increased at all time-points when compared to 12 weeks (mean VI at 12 weeks 1.9 ±1.9, 16 weeks 3.0 ±2.4 and 34 weeks 4.1 ±3.5, P<0.001; Figure 3-6A,C). Despite this, vasculature blood flow within the cervix only increased significantly at 28 weeks (mean FI 32.5 ±4.7) relative to 12 week measurements (mean FI 28.9 ±5.4, P<0.01; Bonferroni multiple comparison, Figure 3-6B, D, Table 3-3).

A 5 B 36 Vascularisation Index Flow Index

4 34

3 32

2 30 Flow Index, FI Index, Flow 1 28 Vascularisation VI Index,

0 26 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints (weeks) Scan timepoints (weeks)

*** ANOVA *** ANOVA C 20 *** D 60 ** *** ns *** ns 50 15

40 10

30 5 FI Index, Flow

Vascularisation VI Index, 20

0 A VI12w wholeb VI 16w wholec VI 22w wholed VI 28w wholee VI 34w whole A FI12w wholeb FI 16w wholec FI 22w wholed FI 28w wholee FI 34w whole Scan timepoints (weeks) Scan timepoints (weeks)

Figure 3-6 In normal pregnancy (n=96) both mean (±SEM) cervical vascularisation (VI) and blood flow indices (FI) increased with advancing gestation (A, B). From 12 to 34 weeks, mean VI increased from 1.9 to 4.9 (P<0.001; C), and mean FI increased from 28.9 to 32.5 (P<0.001, D). (***P<0.001, **P<0.01; 1 way ANOVA, Bonferroni post-test)

88 Ultrasound assessment of the cervix in pregnancy

3.6 Parity and cervical morphology

Cervical morphology among healthy low-risk primiparous women (primips, n=57/96, 59%) were compared to multiparous women, who delivered a baby beyond 24 weeks gestation (multips, n=39/96, 41%). At all timepoints assessed, CL did not differ significantly between primips and multips (Figure 3-7A). At 12 and 16 weeks CV was lower in primips than multips (mean 27cm3 vs 35 cm3 at 12 weeks, P=0.014, and 31cm3 vs 40cm3 at 16 weeks, P=0.002, respectively, Figure 3-7B). From mid-gestation (22 weeks) onwards, CV was comparable among primips and multips. There was a trend for lower vascularity (VI) among primips but this did not reach significance (mean VI 2.3 vs 3.2 multips at 12 weeks, P=0.7, and 4.1 vs 4.1 at 34 weeks, P=0.6, Figure 3-7C). Cervical FI was unaffected by parity (Figure 3-7D).

ABCL CV Primip Multip Primip Multip ns 60 50 ns ns

ns ** 50 40 * ns 40 ns

ns 3) ns 30 30 20 CV CV (cm CL (mm) 20 10 10

0 0 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints Scan timepoints

C VI D FI Primip Multip Primip Multip 7 35 ns ns ns ns ns ns ns 6 ns 30 ns 5 25 ns 4 20

3 15

2 FI Index, Flow 10

Vascularisation Index, VI 1 5

0 0 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints Scan timepoints

Figure 3-7 Parity does not significantly influence CL (A), VI (C) or FI (B). Primiparous women had a lower CV than multiparous women at 12 weeks (mean 27 v 35 cm3) and 16 weeks (mean 31 v 40 cm3, P<0.05; t-test). (CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, w= weeks gestation at cervical screening. ns=non-significant, *P<0.05, **P<0.01; t-test)

89 Ultrasound assessment of the cervix in pregnancy

3.7 A comparison of high and low-risk pregnancy

Low-risk controls (n=96) were compared to high-risk pregnancies (n=199) for differences in CL, CV, VI and FI. Longer CL measurements were observed in low-risk than high-risk women throughout pregnancy: mean CL was 45mm versus 38mm at 12 weeks, (P<0.001) and 32mm versus 27mm respectively at 34 weeks (P=0.03, t-test; Figure 3-8A, Table 3-4). CV was lower in the high-risk women, most notably from 22 weeks (mean 38.8cm3 v 45cm3 low-risk, P<0.008; Figure 3-8B). There was a trend increase in VI and FI among high-risk women compared to low- risk controls (Figure 3-8 C, D, Table 3-4).

Table 3-4 A comparison of cervical length, volume, vascularisation and flow indices among high- and low-risk women at cervical screening from 12 to 34 weeks gestation.

Scan time-points (weeks) Cervical Cohort parameter 12w 16w 22w 28w 34w Mean ±SD Mean ±SD Mean ±SD Mean ±SD Mean ±SD # Low-risk 45 8.9 40 9.1 38 7.0 34 7.2 32 6.9 CL (mm) # High-risk 38*** 8.6 36* 7.7 34** 8.4 31* 7.2 27** 7.6 # 3 Low-risk 28.0 6.3 33.0 7.6 45.4 16.5 50.8 15.9 57.6 21.3 CV (cm ) High-risk 29.8 12.8 33.2 13.0 38.8** 17.1 43.6** 15.9 49.1 18.9 # Low-risk 1.9 1.4 2.9 1.8 3.2 1.6 4.3 2.2 4.5 2.5 # VI High-risk 2.6* 2.0 3.5 2.6 3.8 2.6 4.5 3.5 4.8 3.3 # Low-risk 28.2 3.8 30.6 5.1 30.6 4.5 32.2 5.8 33.2 4.3 # FI High-risk 29.9 4.4 31.0 5.1 32.0 5.5 33.3 5.9 34.0 5.7 # *P<0.05, **P<0.01, ***P<0.001; unpaired t-test for high v low-risk, #P<0.05; 1-way ANOVA for longitudinal 12w to 34w. CV= cervical volume, VI = vascularisation index, FI = flow index, w= weeks, SD = standard deviation Low-risk = healthy pregnancy with no known risk factor for preterm birth, High-risk = pregnancy with a risk factor for preterm birth.

90 Ultrasound assessment of the cervix in pregnancy

ABCL CV 50 *** 60 ns * ** 40 ** ** * ** 40 ns ) 30 3 ns

CL (mm) 20 CV (cmCV 20

10 Low risk controls Low risk controls High risk for preterm birth High risk for preterm birth 0 0 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints (weeks) Scan timepoints (weeks)

CDVI FI 6 38 ns ns 36 ns ns ns 34 4 ns ns ns 32 * ns 30 2 Flow Index, FI Index, Flow 28 Low risk controls Low risk controls Vascularisation Index, VI 26 High risk for preterm birth High risk for preterm birth 0 24 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints (weeks) Scan timepoints (weeks)

Figure 3-8 (A) High-risk pregnancies (n=199) had lower mean cervical lengths compared to low-risk controls (n=96) from 12 weeks (38 v 45mm, P<0.001) to 34 weeks (27 v 32mm, P<0.01). (B) Mean cervical volume was lower among high versus low-risk women, most notably from 22 weeks (P<0.01). (C,D) Cervical vascularity and blood flow were not significantly different among groups at longitudinal measurements. (CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, w= weeks gestation at cervical screening) (ns=non-significant, *P<0.05, **P<0.01; t-test)

91 Ultrasound assessment of the cervix in pregnancy

3.8 Gestation at birth

Data collected on cervical parameters were then assessed for differences in outcomes of preterm birth <37 weeks (n=43/295, 15%) and term birth (252/295, 85%). High-risk pregnancies that went on to deliver preterm <37 weeks demonstrated a trend for lower CL and CV than term births, mostly notably from 28 weeks. At recruitment, CL was comparable among preterm and term groups (mean CL 38mm vs 41mm at 12 weeks respectively, P=0.2; Figure 3-9A, Table 3-5). By 28 weeks this difference reached significance (26mm vs 31mm respectively, P=0.005). CV was similar among the two delivery cohorts from 12 to 28 weeks, after which point women that went on to deliver preterm had lower CV measurements (40cm3 preterm v 49cm3 term, P=0.03, Figure 3-9B; Table 3-5).

A relative increase in VI was also observed when preterm births were compared to term births. This difference was significant from 16 weeks (mean VI 3.1 term vs 4.7 preterm, P=0.0013) and persisted until 34 weeks (mean VI 3.9 vs 5.4 respectively, P=0.04, Figure 3-9C, Table 3-5). Despite an increase in VI from 16 weeks, FI only appeared to increase among preterm deliveries at 28 and 34 weeks when compared to term births (Figure 3-9D).

Table 3-5 Comparison of term (>37 weeks, n=252) and preterm birth (<37 weeks, n=43) for measurement of cervical length, volume, vascularisation and flow indices. Scan time-points (weeks) Cervical Delivery parameter outcome 12w 16w 22w 28w 34w Mean ±SD Mean ±SD Mean ±SD Mean ±SD Mean ±SD Term 41 1.2 36 0.8 34 0.7 31 0.7 28 0.7 CL (mm) Preterm <37w 38 2.1 34 1.9 31 2.4 26 1.9 23 2.5 Term 30.0 15.2 34.2 14.2 41.0 15.6 46.4 17.0 49.3 19.3 CV (cm3) Preterm <37w 30.6 12.1 37.6 15.2 39.8 19.3 42.2 19.0 39.9 20.7 Term 2.2 2.2 3.1 2.7 3.7 3.3 3.9 4.6 3.9 3.2 VI Preterm <37w 2.7 2.9 4.7 4.0 5.3 4.8 5.0 4.7 5.4 5.0 Term 30.0 5.2 31.2 5.4 32.1 5.9 32.2 6.1 32.1 5.5 FI Preterm <37w 29.8 4.8 31.7 6.3 32.8 6.4 33.8 4.6 34.5 5.1 CV= cervical volume, VI = vascularisation index, FI = flow index, Term = birth >37 weeks, preterm = birth <37 weeks w= weeks, SD = standard deviation

92 Ultrasound assessment of the cervix in pregnancy

ABCL CV

50 60

40 * 50 ** )

30 ** 3 40

CL (mm) 20 CV (cmCV

30 10 Term birth Term birth Preterm birth <37 weeks Preterm birth <37 weeks 0 20 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints (weeks) Scan timepoints (weeks)

CDVI FI

8 38 * * * * 34 6 **

4 30 Flow Index, FI Index, Flow 2 26 Term birth Term birth Vascularisation Index, VI Preterm birth <37 weeks Preterm birth <37 weeks 0 22 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Scan timepoints (weeks) Scan timepoints (weeks)

Figure 3-9 A comparison of CL, CV, VI and FI among term (>37 weeks, n=252) and preterm births (<37 weeks, n=43) at longitudinal measurments from 12 to 34 weeks. (A) Cervical length was lower in women delivering preterm than term from 12 weeks (mean CL 38mm v 41m, P=0.2), persisting to 34 weeks (mean CL 23mm v 28mm, P=0.009). (B) Cervical Volume decreased in those delivering preterm from 28 weeks onwards (CV at 34weeks, 40cm3 v 49cm3, P=0.03). Cervical vascularisation, VI (C) and blood flow, FI (D) were both increased among preterm births. Elevated vascularisation was evident from 16 weeks (preterm mean VI 3.1 v term 4.7, P=0.001), while increased blood flow occurred later at 34 weeks (mean FI 35 v 32 respectively, P=0.001). CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, w= weeks gestation at cervical screening. (*P<0.05, **P<0.01; t-test)

93 Ultrasound assessment of the cervix in pregnancy

3.9 Prediction of preterm birth

Data collected from TVS scans were assessed for the prediction of preterm birth <34 weeks (n=15, 5%) and <37 weeks (n=43, 15%) at clinically relevant screening time-points: 12, 16, and 22 weeks. The predictive value of CL, CV, VI and FI were tested using receiver operator curves (ROCs), and demonstrated better prediction of birth <34 weeks than for birth <37 weeks.

CL and CV provided comparable predictive accuracies for preterm birth, and both improved with later gestation at screening. At 22 weeks, the area under the curves (AUCs) for preterm birth <34weeks were 0.80 for CL (P<0.001) and 0.71 for CV (P=0.06; Table 3-6, Figure 3-10A-D). Predicative performance provided by 12 week screening for birth <34 weeks was comparatively poor (AUCs 0.56 for CL, P=0.1 and 0.54 for CV, P=0.6; Figure 3-10A, C). Prediction of preterm birth <37weeks improved with later gestation at screening; AUCs at 12, 16 and 22 weeks for CL were 0.51, 0.55, 0.70 (Figure 3-10B), and for CV 0.53, 0.54, 0.61 respectively (Figure 3-10D).

Neither VI or FI provided good prediction of preterm birth <34 weeks or <37 weeks, and unlike CL and CV screening performance, did not improve with later gestation at measurement. For preterm birth <34 weeks, AUCs at 12 weeks to 22 weeks were VI 0.60 to 0.69, and FI 0.56 to 0.62 (Figure 3-10E, G) and for <37 weeks, VI 0.53 to 0.66 and FI 0.54 to 0.56 (Figure 3-10F, H; Table 3-6).

94 Ultrasound assessment of the cervix in pregnancy

Preterm birth <34 weeks Preterm birth <37 weeks

A CL B CL

100 100

80 80

60 60

40 22w (AUC 0.80) 40 22w (AUC 0.70) Sensitivity % Sensitivity % 16w (AUC 0.63) 16w (AUC 0.55) 20 20 12w (AUC 0.56) 12w (AUC 0.51) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1-specificity (False postive rate, %) 1-specificity (False postive rate, %)

C CV D CV

100 100

80 80

60 60

40 22w (AUC 0.71) 40 22w (AUC 0.61) Sensitivity % Sensitivity % 16w (AUC 0.59) 20 20 16w (AUC 0.54) 12w (AUC 0.54) 12w (AUC 0.53) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1-specificity (False postive rate, %) 1-specificity (False postive rate, %)

E VI F VI

100 100

80 80

60 60

40 22w (AUC 0.69) 40 22w (AUC 0.59) Sensitivity % Sensitivity % 16w (AUC 0.58) 16w (AUC 0.53) 20 20 12w (AUC 0.60) 12w (AUC 0.66) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1-specificity (False postive rate, %) 1-specificity (False postive rate, %)

G FI H FI

100 100

80 80

60 60

40 40 22w (AUC 0.61) 22w (AUC 0.55) Sensitivity % Sensitivity % 16w (AUC 0.56) 16w (AUC 0.54) 20 20 12w (AUC 0.62) 12w (AUC 0.56) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1-specificity (False postive rate, %) 1-specificity (False postive rate, %)

95 Ultrasound assessment of the cervix in pregnancy

Figure 3-10 Receiver operator curves (ROC) demonstrate predictive performance of CL, CV, VI and FI collected from 295 women at 12, 16 and 22 weeks for preterm birth <34 weeks (5%, 15/295) and <37 weeks (15%, 43/295). Optimal predictive accuracy for preterm birth <34 weeks is provided by CL at 22 week screening (AUC 0.80) (A), followed by CV at 22 weeks (AUC 0.71) (C). Prediction of preterm birth <37 weeks by CL and CV was poor, but did improve slightly with later gestation at screening (B, D). VI and FI did not predict preterm birth <34 or <37 weeks (E, F, G, H). (CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, AUC = area under the ROC)

Table 3-6 Areas under the CL, CV, VI, and FI ROC curves for prediction of preterm birth <34 weeks and <37 weeks Areas under the ROC Scan time- 3 CL (mm) CV (cm ) VI FI points <34w <37w <34w <37w <34w <37w <34w <37w 12w 0.56 0.51 0.54 0.53 0.60 0.66 0.62 0.56 16w 0.63 0.55 0.59 0.54 0.58 0.53 0.56 0.54 22w 0.80 0.70 0.71 0.61 0.69 0.59 0.61 0.55 CL = cervical length, CV= cervical volume, VI = vascularisation index, FI = flow index ROC = receiver operator curves, w = weeks

96 Ultrasound assessment of the cervix in pregnancy

Discussion

This study aimed to establish a reference range for cervical length, volume and vascularity in healthy pregnancy and compare these to women at risk of preterm birth. In doing so it aimed to determine their predictive accuracies for thresholds of preterm birth <34 weeks and <37 weeks gestation.

As was hypothesised and consistent with previous findings, CV, VI and FI increased with advancing gestational age 93, 94, 101. Uniquely this study demonstrates comparable prediction of CV and CL measurements for preterm birth <34 and <37 weeks. The accuracy of this prediction was affected by gestational at age measurement, whereby improved accuracy was achieved at later gestation (22 weeks) and for earlier preterm birth (<34 weeks). This is a well-established phenomenon; the sensitivity of CL is good at late screening gestations (eg 22 weeks), while specificity improves with an earlier gestation at screening (eg 12 weeks) 88, 89, 218. It is a novel, although not unsurprising revelation in this study, that CV follows this trend.

In this, the largest longitudinal study of VI and FI in high-risk pregnancies, vascularity parameters did not provide additional or improved prediction for preterm birth over CL measurements when assessed before 24 weeks. My findings demonstrate that the poor prediction of VI and FI is primarily due to late gestational increases in cervical vascularity. Although women delivering <37 weeks demonstrated increased VI and FI values compared to both low-risk controls and term births, these differences were only apparent from 28 weeks onwards. As the limit of viability is 24 weeks gestation, a clinically useful screening tool requires the detection of quantifiable cervical change before a 24 week threshold. This ensures safe surgical intervention without the risk of iatrogenic preterm birth.

The strength of my study was the use of a high-risk patient cohort, not previously observed in such large numbers. This was demonstrated by a high spontaneous preterm birth rate (22%), and low mean gestation at birth (32+6 weeks), as well as high rates of cerclage insertion for cervical shortening (43% of high-risk pregnancies). The high rate of preterm birth is higher than would be expected in the setting of a preterm surveillance clinic 219 and may represent an unintentional source of recruitment bias whereby women with poor obstetric histories may have been perceived to be clinically interesting and therefore preferentially recruited.

97 Ultrasound assessment of the cervix in pregnancy

Limitations

The main limitation of the study was that women found to have a shortened cervix received a cerclage. Insertion of a cerclage alters the anatomy of cervix to increase cervical length, by squeezing and elongating it 220-222. Therefore CL measurements as well as CV, VI and FI prior to 24 weeks are likely to be affected by cerclage insertion. The predictive abilities of CV, VI and FI may also have been impacted on by cerclage insertion. This is because the highest risk pregnancies, with the shortest cervical lengths, may have had their gestational period extended through cerclage insertion. Consequently, false positive rates of potential thresholds for CV, VI and FI would have been increased. To fully assess this, a prospective study aimed at addressing the impact of the cerclage on cervical parameters was performed and described in Chapter 4. Due to an ethical obligation to intervene in women with a shortened cervix in high- risk pregnancy, interventions such as cervical cerclages are an unavoidable limitation of clinical preterm birth research. A control group with a short cervix, but without an intervention, does not exist in current clinical practice.

During analysis of the scans, delineation of the ‘region of interest’ using VOCALTM was found to be technically challenging as volumetric values are dependent on the visualisation of clear tissue planes. Scan analysis was time consuming, required training, experience and interpretable results were not available in the clinic setting as ultrasound data had to be downloaded and analysed on a separate software at a later date. These are relevant when considering the practical application of CV as a clinical screening tool, and would potentially impede on implementation across maternity units on a larger scale. The measurement of CL in comparison is dependent on visualising the highly echogenic cervical canal on ultrasound. This is technically easy, reproducible and provides immediate results that are universally interpretable in the setting of a clinic. Consequently, CL measurements have a technical and practical advantage over CV in a context of preterm surveillance clinics.

In summary

This study of cervical volume and vascularity has provided useful insight into the longitudinal morphological changes in the cervix in both high- and low-risk pregnancies. I have demonstrated that CV is comparable to CL for the prediction of spontaneous preterm birth, but

98 Ultrasound assessment of the cervix in pregnancy found that there are considerable technical challenges in acquiring interpretable CV values. I have also demonstrated that cervical vascularity indices, VI and FI, were prematurely increased in women subsequently delivering preterm. These changes occurred late in the second trimester and therefore did not add clinically useful information for preterm screening programs, which primarily are performed before 24 weeks. This study therefore concludes that CV, VI and FI are not a clinically viable addition to current preterm birth surveillance, when compared to CL measurements as screening tool for preterm birth risk.

99 Vaginal microbiota and the cervix in pregnancy

4 VAGINAL MICROBIOTA AND THE CERVIX IN PREGNANCY

100 Vaginal microbiota and the cervix in pregnancy

Chapter abstract

Hypothesis Ascending bacteria from the vagina is thought to be an important cause of spontaneous preterm birth. What remains uncertain is the propensity for some bacteria to trigger parturition cascades while others do not. The interaction between the vaginal microbiota and the cervix as a barrier to infection in pregnancy appears key to preterm birth pathogenesis.

Aim To characterise the interaction between the cervical morphology and vaginal microbiota in pregnancy.

Methods Pregnancies at high- and low-risk of preterm birth were prospectively recruited and high vaginal swabs and transvaginal ultrasound scans of the cervix taken longitudinally at 12 16, 22, 28 and 34 weeks gestation. MiSeq sequencing of 16S rRNA gene amplicons (V1-V3) was used to characterise vaginal microbiota. Vaginal microbiota profiles were correlated with paired cervical length, volume and vascularity data, and assessed for correlations with subsequent preterm birth.

Results A total of 167 women consented to matched vaginal swabs and transvaginal scans at each research time-point. A short CL≤25mm prior to 24 weeks gestation associated with a high abundance of L. iners in corresponding vaginal microbiota. Term births were associated with high proportions of L. crispatus sequences, most significantly from 16 weeks (P<0.001), persisting to 28 weeks (P=0.015). L. iners in comparison was highly abundant from 12 weeks among women subsequently experiencing early preterm births <34+0 weeks (P<0.001). This equated to a preterm birth prediction of 67% sensitivity and 71% specificity. The addition of a CL≤25mm threshold to L. iners dominance improved screening performance to 87% sensitivity and 68% specificity.

Conclusion Understanding the interactions between the cervix and vaginal microbiota provides important insight to preterm birth aetiology. Vaginal L. crispatus dominance appears protective against preterm birth, while L. iners dominance is a risk factor.

101 Vaginal microbiota and the cervix in pregnancy

Introduction

Ascending bacteria from the vagina through the cervix into the uterine cavity is thought to be major cause of spontaneous preterm birth 6, 48. The host-microbe interactions within the vagina play a fundamental role in reproductive health as well as disease. Unlike other body sites where bacterial diversity is thought to be beneficial to health, the dominance of single Lactobacillus species within the vagina is associated with a healthy phenotype. The Lactobacilli-dominant niche of the vagina is promoted by their capacity to anaerobically metabolise oestrogen- glycogen depositions in the vaginal epithelium into complex sugars and subsequently lactic acid and in doing so, maintain a low vaginal pH <4.5 223. This, in addition to production of various bacteriocidal compounds, ensures inhibition of pathobionts known to associate with subfertility, poor pregnancy outcome 142, contraction of sexual transmitted infections 158, and HIV 140. Next generation sequencing (NGS) of the bacterial 16S rRNA gene has led to the identification of 5 major vaginal community state types 155 (CSTs); L. crispatus (CST I), L. gasseri (CST II), L. iners (CST III) and L. gasseri (CST V). CST IV is characterised by a lack of Lactobacilli and enrichment in diverse anaerobic bacteria (CST IV, species diverse). Pregnancy is associated with high Lactobacillus species stability and reduced dysbiosis, compared to non-pregnant women. This mechanism likely protects the immunocompromised pregnant mother from prolonged exposure to pathogenic bacteria, as well as protects the fetus from potential exposure to vaginal microbiota, which is a major determinant of both neonatal and long-term health 187, 188. Abnormal vaginal microbiota confers a risk for preterm birth although there are conflicting reports from small case series implicating high L. iners abundance (CST III) 182 as well as BV-like Lactobacilli-deficient states (CST IV) in preterm birth pathogenesis 181. The cervix acts as a mechanical and immunological barrier to ascending vaginal bacteria 35. Premature cervical ripening, a prerequisite for the spontaneous expulsion of the preterm fetus, may be triggered by exposure of the amniotic cavity to pathogenic bacteria ascending from the vagina 224. The presence of lipopolysaccharides in the normally sterile uterine cavity induces a pro-inflammatory cytokine response 53 thereby activating parturition pathways, triggering a prostaglandin release and culminating cervical collagenous and vascular remodelling, softening and dilation 38. The relationship between vaginal microbial communities and the uterine cervix in pregnancy is largely unknown. There have been no studies reporting on the interaction between microbiota and the cervix in pregnancy. This study aims to address this relationship using 16S rRNA sequencing techniques, matched to 3D/4D transvaginal ultrasound assessment of the cervix, in both healthy pregnancy and those with an underlying risk factor for preterm birth.

102 Vaginal microbiota and the cervix in pregnancy

Aims 1. Characterise vaginal microbial community structure in high-risk pregnancy from 12 to 34 weeks, using 16s rRNA gene sequencing. 2. Examine the association between cervical volume, length and vascularity and corresponding vaginal microbiota with respect to preterm birth outcome.

Hypothesis Healthy pregnancy is typified by increasing vaginal Lactobacillus spp. dominance, which is thought to be central to the protection of gestational tissues to ascending infection 171, 172. Vaginal dysbiosis is associated with a pro-inflammatory vaginal environment 180 and compromised integrity of the cervical epithelium 51. Thus it is hypothesized that vaginal dysbiosis associates with cervical shortening, altered cervical volume and vascularity, which may be measured at transvaginal ultrasound.

103 Vaginal microbiota and the cervix in pregnancy

Study design

Patient recruitment and sample collection As described in the Methods (Chapter 2) pregnant women were recruited in two groups: healthy low-risk pregnancies, and women with an underlying risk factor for preterm birth. All women were followed up longitudinally at 12, 16, 22, 28 and 34 weeks gestation. At each time-point, cervico-vaginal fluid was sampled from the posterior fornix under direct visualisation, using a BBL™ CultureSwab™ MaxV Liquid Amies swab (Becton, Dickinson and Company, Oxford, UK) for later 16S rRNA gene sequencing. Following this a transvaginal scan was performed, as previously described (Chapter 3, Results 1). Bacterial DNA was extracted from vaginal swabs and 16S rRNA gene sequencing performed. Resulting microbiota profiles were then correlated to ultrasound data for cervical length (CL), cervical volume (CV), and vascularisation indices (Vascularisation Index, VI and Flow index, FI).

As previously stipulated, participation in this study did not influence clinical care or dictate preventative interventions (cerclage or progesterone) for perceived preterm birth risk. The clinical indication for intervention was CL ≤25mm at TVS measured at ≤23+6 week’s gestation.

Exclusion criteria for study participation included multiple pregnancy (twins or higher order), uterine anomalies, iatrogenic preterm birth <37 weeks, laser and ablative therapy and CIN grade ≤1, HIV positive women and women who had had sexual intercourse or vaginal bleeding in the preceding 48 hours. Eligibility criteria for high risk group included a previous spontaneous preterm <37+0 weeks in a singleton pregnancy immediately prior to the index pregnancy or excisional conisation including LEEP, LLETZ, CKC for CIN grade II or III.

Statistical analyses of scan and sequence data Bacterial gene sequence data were assessed for differences in total number of species observed and the alpha diversity using the Shannon index with advancing gestational age using a two-tailed t test and ANOVA. Examination of statistical differences between vaginal microbiota was performed using the Statistical Analysis of Metagenomic Profiles (STAMP) software package 201. Vaginal bacterial communities were assessed at genera and species taxonomic level. Samples were classified into community state types (CSTs) as described by Ravel et al

104 Vaginal microbiota and the cervix in pregnancy

155, using ward linkage hierarchical clustering analysis (HCA) of species sequence data with a clustering density threshold of 0.75.

Fisher exact tests and two-way ANOVA with Bonferroni correction assessed differences in proportions of assigned CSTs as a function of ethnicity, gestational age at sampling, and gestation at birth as categorised into early preterm birth (<34+0 weeks), late preterm birth (34+0 to 36+6 weeks), and term birth (≥37+0 weeks). The association between microbial profiles and cervical parameters (CL, CV, VI and FI), and gestation at birth were tested by ANOVA, Fisher’s exact test, t-tests and linear regression where appropriate.

105 Vaginal microbiota and the cervix in pregnancy

Results

4.1 Recruitment

After loss to follow up, 53% of the initial 315 women prospectively recruited (n=167) consented to simultaneous vaginal sampling for microbial assessment by 16S rRNA gene sequencing and transvaginal ultrasound assessment of the cervix (Table 4-1).

Table 4-1 Prospective recruitment for transvaginal scans and vaginal swabs Total population Low-risk High-risk Recruitment n/N % n/N % n/N %

Initial recruitment 315 108/315 34% 207/315 66%

Loss to follow up/ miscarriage <13weeks 20/315 6% 12/108 11% 8/207 4%

Final recruitment Consented to: Transvaginal scan only 295/315 94% 96/108 89% 199/207 96%

Transvaginal scan & 167/315 53% 34/108 31% 133/207 64% vaginal swabs

Low-risk = healthy pregnancy with no known risk factor for preterm birth; High-risk = pregnancy with a risk factor for preterm birth

106 Vaginal microbiota and the cervix in pregnancy

4.2 Participant demographics

Significant participation bias was demonstrated when comparing participants consenting to ‘scan only’ or ‘swab and scan’ studies. In the low-risk cohort, women consenting to swabs in addition to scans were more likely to be Caucasian rather than black, nulliparous (P<0.01), and non-smokers (P=0.08). Similarly in the high-risk cohort, women were less likely to be black, or smokers (P<0.05) if they consented to swabs and scans (Table 4-2).

Table 4-2 Demographic according to consent to participate in ‘scan only study’ or ‘scan and swabs study’ Low risk, Low risk, High risk, High risk, Consented to: scan only scan & swabs scan only scan & swabs n/N, % 62/96 65% 34/96 35% 66/199 33% 133/199 67%

Age (years) Mean ±SD 32.9 ±5.6 32.1 ±4.7 33.1 ±4.7 32.1 ±4.7 BMI Mean ±SD 24.0 ±4.8 25.1 ±4.5 24.5 ±4.4 25.1 ±4.5

Ethnicity, n/N % Caucasian 37/62 60% 27/34 79%** 25/66 38% 82/133 62%* Asian 12/62 19% 7/34 21% 11/66 17% 19/133 14% Black 13/62 21%** 0/34 0% 30/66 45%** 32/133 24%

Parity, n/N % Para ≥ 1 10/62 16% 29/34 85%** 40/66 61% 83/133 62%

Smoker, n/N % 6/62 10% 0/34 0% 18/66 27% 5/133 4%*** ***P0.001, **P<0.01, *P0.05 for scan only v swab and scan; fishers exact

Of the 167 pregnancies consenting to longitudinal sampling of cervicovaginal fluid (CVF), 34 were healthy low-risk pregnancies (20%, 34/167), and 133 (80%) were considered at risk of preterm birth. The patient characteristics of these high and low risk groups and their delivery outcomes are provided in Table 4-3. The overall preterm birth rate <37 weeks was 19% (32/167). All cases of preterm birth were from women in the high-risk cohort. Early (<34+0 weeks) and late preterm births (34+0 to 36+6 weeks) occurred in 11% (15/133) and 13% (17/133) respectively. Mean gestation at birth among the 32 preterm births was 32+6 weeks (SD±3+6 weeks, range 24+4-36+6 weeks). Black women had the greatest number of preterm deliveries <37+0 weeks (38%, 12/32; P<0.05) compared to Caucasians (15%, 16/109) and Asians (15%, 4/26; P<0.05).

107 Vaginal microbiota and the cervix in pregnancy

Table 4-3 Patient demographics of low risk and high risk pregnancies consenting to vaginal microbial swabs Total population Low-risk High-risk

N=167 N=34 N=133 Age (years) Mean ±SD 31.9 ±5.2 31.6 ±5.2 32.1 ±4.7 BMI Mean ±SD 24.7 ±4.6 22.8 ±3.3 25.1 ±4.5

Ethnicity, n/N % Caucasian 109/167 65% 27/34 28% 82/133 62% Asian 26/167 16% 7/34 7% 19/133 14% Black 32/167 19% 0/34 0% 32/133 24%**

Parity, n/N % Para 0 55/167 33% 5/34 5% 50/133 38%* Para ≥ 1 112/167 67% 29/34 30% 83/133 62%

Smoker, n/N % 5/167 3% 0/34 0% 5/133 4%

Cerclage Intervention, n/N % 38/167 23% n/a 38/133 29% Ultrasound indicated 31/167 19% n/a 31/133 23% History indicated 7/167 4% n/a 7/133 5%

Early PTB, <34+0 w 15/167 9% 0/34 0% 15/133 11% Late PTB, 34+0 to <37+0 w 17/167 10% 0/34 0% 17/133 13% Gestation at +0 birth, n/N % Total PTB <37 w 32/167 19% 0/34 0% 32/133 24%

Term ≥37+0 w 135/167 81% 34/34 100% 101/133 76%

**P<0.01, *P<0.05; fisher exact. Low-risk = healthy pregnancy with no known risk factor for preterm birth; High-risk = pregnancy with a risk factor for preterm birth; PTB = preterm birth, w= weeks.

Cervical shortening to ≤25mm prior to 24 weeks occurred in 23% (31/133), all of whom went on to receive an ultrasound indicated cervical cerclage. In addition, 7 women (5%) received a history indicated cervical cerclage due to a poor obstetric history (3 or more previous preterm births) (Table 4-3). As the impact of cerclage insertion on vaginal microbiota and cervical volume and vascularity was an unknown, all data from sampling time-points post-cerclage insertion (n=107 samples) were excluded from the analyses in this chapter (Table 4-4). The impact of the cerclage on the microbiome and cervix will be specifically addressed later in Chapter 4. The remaining total of 505 vaginal samples with matched ultrasound data collected from longitudinal sampling at 12 to 34 weeks gestation were included in this study (Table 4-4).

108 Vaginal microbiota and the cervix in pregnancy

Table 4-4 A total of 505 vaginal microbial samples with matched ultrasound data made up from low- and high-risk patient groups at longitudinal sampling, were included in the study. Total Total Gestation at Low-risk High-risk Samples excluded post cerclage* samples sampled sampling included (weeks) n, Mean n, Mean n, Mean ±SD Min Max ±SD Min Max N= ±SD Min Max N= samples (weeks) samples (weeks) samples (weeks) 12w 16 12.4 ±1.6 8.0 14.7 87 13.1 ±1 11 14.9 103 0 103

16w 30 17.2 ±1.7 16.0 18.4 126 16.9 ±0.8 15.1 18.4 156 18 16.9 ±1 15.4 18.4 138 22w 22 21.9 ±1.2 20.0 23.7 126 21.7 ±1.1 20 23.7 148 33 21.6 ±1.1 20 24.3 115 28w 22 28.0 ±0.9 26.7 30.0 85 27.7 ±0.7 25.86 29.7 107 30 27.9 ±0.8 26.1 29.3 77 34w 20 34.0 ±0.9 32.0 36.1 78 33.6 ±0.9 31.43 36 98 26 33.6 ±1 31.9 35.3 72 Total 110 502 612 107 505 samples, n w= weeks; Low-risk = healthy pregnancy with no known risk factor for preterm birth; High-risk = pregnancy with a risk factor for preterm birth *Cervical cerclage inserted in 38 women

109 Vaginal microbiota and the cervix in pregnancy

4.3 Community state type classification

505 samples were classifed into community state types (CSTs) 155 according to ward hierarchical clusting analysis (HCA) of species data: CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), CST IV (diverse species) and CST V (L. jensenii) (Figure 4-1).

L. crispatus (CST I) dominant vaginal communities were most prevalent (43%, n=217), followed by L. iners (CST III, 29%, n=144), L. gasseri (CST II, 12%, n=60) and L. jensenii (CST V, 9%, n=46). The remaining 8% (n=38) of samples collectively made up the CST IV group, which consisted of a diverse mix of microbial species and were characteristically deficient in Lactobacillus spp. (Table 4-5).

Table 4-5 Community state type (CST) assignment of samples according to ward hierarchical clustering analysis of microbial species data.

Community state type, CST Species n %

I L. crispatus 217 43%

II L. gasseri 60 12%

III L. iners 144 29%

IV Diverse species 38 8%

V L. jensenii 46 9%

TOTAL samples 505

110 Vaginal microbiota and the cervix in pregnancy

Figure 4-1 Bacterial species diversity of 505 vaginal samples. (A) Heatmap of ward hierarchical clustering analysis and (B) Principal component analysis (PCA) of species sequence data identified 5 major clusters of samples corresponding to 5 community state types (CSTs): CST I (L. crispatus, n=127), CST II (L. gasseri, n=60), CST III (L. iners, n=144), CST IV (diverse species, n=38), and CST V (L. jensenii, n=46).

111 Vaginal microbiota and the cervix in pregnancy

4.4 Ethnicity

The effect of ethnicity on CST distribution among Black (n=32), Asian (n=26) and Caucasian women (n=109), was assessed at the first sampling time-point (mean 14+4 ±2.2 weeks).

Black women accounted disproportionately for CST IV samples that were lacking in Lactobacillus spp (47%, 7/15; RR 3.7 [95% CI 1.4-9.4]; P=0.006). Comparatively, 6% of Caucasians (7/109) and 4% of Asians (1/26) were classified as CST IV (Table 4-6, Figure 4-2). L. crispatus (CST I) was most prevalent among Caucasian women (51%; 55/109), while L. iners (CST III) was highly abundant in Black (50%, 16/32) and Asian women (42%, 11/26; P<0.05; Table 4-6, Figure 4-2).

Table 4-6 Community state type classification according to ethnicity

Black Asian Caucasian Total CST Species n=32 n=26 n=109 n=167

I L. crispatus 5/32 (16%) 8/26 (31%) 55/109 (50%)* 68/167 (41%)

II L. gasseri 1/32 (3%) 3/26 (12%) 13/109 (12%) 17/167 (10%)

III L. iners 16/32 (50%)* 11/26 (42%)* 25/109 (23%) 52/167 (32%)

IV Diverse species 7/32 (22%)* 1/26 (4%) 7/109 (6%) 15/167 (9%)

V L. jensenii 3/32 (9%) 3/26 (12%) 9/109 (8%) 15/167 (9%)

TOTAL 32/167 (19%) 26/167 (16%) 109/167 (65%) 167/167

CST = community state type based on ward HCA of species data; w= weeks *P value <0.05, chi squared

112 Vaginal microbiota and the cervix in pregnancy

Figure 4-2 Ethnicity significantly influences community state type distribution. (A) Caucasians accounted for 81% (55/68) of L. crispatus dominant samples, while Black women accounted disproportionately for CST IV dominated microbiota (47%, 7/15; RR 3.7, 95% CI 1.4-9.4; P=0.006). Comparatively, 6% of Caucasians (7/109) and 4% of Asians (1/26) were classified as CST IV. (B) L. iners was highly prevalent in Black (50%, 16/32) and Asian women (42%, 11/26) compared to Caucasians (23%, 25/109, P<0.05).

113 Vaginal microbiota and the cervix in pregnancy

4.5 Advancing gestational age

The impact of advancing gestation on vaginal microbial community stability was assessed through observation of the number of species, Shannon Index of alpha diversity and shifts in CST classification from 12 to 34 weeks.

There was a non–significant reduction in both the number of species and alpha diversity observed at 12 weeks towards mid-gestation (22 weeks), before diversifying again towards 34 weeks (Table 4-7, Figure 4-3A, B).

Table 4-7 Number of bacterial species observed and corresponding alpha diversity according to gestation at sampling.

Gestation at sampling No. Species Observed Shannon index of alpha diversity (weeks) Mean ±SD Min Max Mean ±SD Min Max

12w 5 ±3.6 1 21 0.36 ±0.40 0 1.86

16w 5 ±3.3 1 16 0.29 ±0.36 0 1.69

22w 4 ±3.3 1 18 0.33 ±0.39 0 1.77

28w 5 ±4.4 1 26 0.40 ±0.43 0 2.14

34w 6 ±5.8 1 35 0.39 ±0.45 0 1.95

P value (1 way ANOVA) 0.12 0.23

114 Vaginal microbiota and the cervix in pregnancy

ABSpecies observed Shannon index

40 P=0.12; A NOV A 2.5 P=0.23; A NOV A

2.0 30

1.5

obs 20 S 1.0

10 Shannon Index 0.5

0 0.0 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Gestation at screening (weeks) Gestation at screening (weeks)

*** C 100

80

60

40 number of samples

20

0 12w 16w 22w 28w 34w Gestation at screening (weeks)

CST, species: I L. crispatus II L. gasseri III L. iners IV Diverse species V L. jensenii

Figure 4-3 Vaginal microbial species shifts with advancing gestation. Gestational age did not significantly impact on the number of species observed (A) or Shannon Index of alpha diversity (B). L. crispatus (CST I) and L. gasseri (CST II) dominance increased from 37% and 10% respectively at 12 weeks, peaking at 46% and 17% respectively at 22 weeks. A corresponding reduction in proportions of women assigned to CST III (L. iners) was observed, declining from 37% at 12weeks to 20% at a 22-week nadir (***P<0.001; two-way ANOVA. (CST= community state type, w=weeks)

115 Vaginal microbiota and the cervix in pregnancy

When classified into CSTs (Figure 4-1A), and grouped into gestational age at sampling, there was a demonstrable increase in both L. crispatus and L. gasseri towards the mid-gestational point. Assignment to CST I (L. crispatus) increased from 37% at 12 weeks to 46% at 22 weeks, while CST II (L. gasseri) increased from 10% to 17% respectively (Table 4-8, Figure 4-3C). A corresponding decrease in the prevalence of CST III (L. iners) was observed towards mid- gestation (37% at 12 weeks versus 20% at 22 weeks) before increasing towards the end of the third trimester (31% at 34 weeks) (P<0.001; 2-way ANOVA). A initial trend reduction in the proportions of women classified as CST IV (diverse species) from 12 weeks (10%) to 16weeks (6%) was observed before levels remained consistently low thereafter (8% at 22 weeks, 8% at 28 weeks and 7% at 34 weeks; Table 4-8, Figure 4-3C).

Table 4-8 Community state type distribution according to gestational time-point at sampling

Community state type, Species Gestation at CST IV, sampling CST I, CST II, CST III, Diverse CST V, (weeks) L. crispatus L. gasseri L. iners species L. jensenii TOTAL n % n % n % n % n % n 12w 38 37% 10 10% 38 37% 10 10% 7 7% 103 16w 59 43% 17 12% 38 28% 8 6% 16 12% 138 22w 53 46% 19 17% 23 20% 9 8% 11 10% 115 28w 34 44% 6 8% 23 30% 6 8% 8 10% 77 34w 33 46% 8 11% 22 31% 5 7% 4 6% 72 Total 217 60 144 38 46 505 Two way ANOVA: P=0.008 Gestation factor; P<0.001 CST factor

116 Vaginal microbiota and the cervix in pregnancy

4.6 Correlation between cervical assessment and vaginal microbiota

Cervical length (CL), volume (CV), vascularity (VI) and blood flow (FI) distribution were assessed for association with co-existing vaginal microbiota. Values of CL (Figure 4-4A), CV (Figure 4-4B), and VI (Figure 4-4C) did not vary with community state types (ANOVA, Table 4-9). Elevated FI associated with CST II (mean 33.6 ±5.2) and CST IV (mean 32.2 ±6.3) but not CST I (mean 31.2 ±5.0), CST III (mean 31.7 ±4.9) or CST V (mean 30 ±4.6; P=0.004; Figure 4-4D, Table 4-9). This comparison does not take into account gestational age at sampling however. This is an important consideration, as gestation has a significant influence on measurements of CL, CV and VI (as previously demonstrated in Chapter 3). An assessment of samples taken prior to 24 weeks (a clinically relevant threshold for the purpose of preterm birth surveillance) was therefore performed.

Table 4-9 Mean cervical length, volume, vascularisation and flow indices according to corresponding community state type.

Community state type, Species

Cervical CST I, CST II, CST III, CST IV, CST V, parameter L. crispatus L. gasseri L iners Diverse species L. jensenii n=217 n=60 n=144 n=38 n=46 P value Mean SD Mean SD Mean SD Mean SD Mean SD CL (mm) 34.8 9.2 35.5 12.0 33.4 10.5 33.9 9.9 34.0 7.4 0.584 CV (cm3) 37.9 17.1 40.6 14.5 38.4 17.0 37.8 17.1 33.2 10.5 0.230 VI 3.3 2.4 4.1 2.6 4.0 2.8 4.1 3.3 3.6 2.7 0.079 FI 31.2 5.0 33.6 5.2 31.7 4.9 32.2 6.4 30.1 4.6 **0.004 CL = cervical length, CV= cervical volume, VI = vascularisation index, FI = flow index, SD= standard deviation **P value; 1-way ANOVA

117 Vaginal microbiota and the cervix in pregnancy

Community state type I L. crispatus II L. gasseri III L. iners IV mixed species V L. jensenii n= 217 n= 60 n= 144 n= 38 n= 46

ABCervical length Cervical volume

100 ns 150 ns

80

100

60 3)

40 CL (mm) CV (cm CV 50

20

0 0 I II III IV V I II III IV V Community state type Community state type

CDVascularisation Index Flow Index ns ** 20 50

15 40

10

30 5 FI Index, Flow Vascularisation Index, VI VascularisationIndex,

0 20 I II III IV V I II III IV V Community state type Community state type

Figure 4-4 Correlation of cervical length (CL), volume (CV), vascularity (VI) and blood flow (FI) with co-existing vaginal microbiota. There was no significant association between community state type and corresponding measurements of CL (P=0.58), CV (P=0.23) or VI (P=0.08). However, highest mean FI measurements were associated with CST II (L. gasseri; mean 33.6 ±5.2) and CST IV (diverse species; mean 32.2 ±6.3) and lowest with CST V (L. jensenii; mean 30 ±4.6; P=0.004, 1 way ANOVA).

118 Vaginal microbiota and the cervix in pregnancy

4.7 Vaginal microbiota and a short cervix

CL screening prior to 24 weeks gestation is used clinically to identify pregnancies at risk of preterm birth. Detection of a cervix below the 10th centile (CL ≤25mm) is universally employed to define a short cervix, conferring highest risk of subsequent preterm birth 83. Of 167 study participants, attendance prior to 24 weeks totalled 356 research time-points (n=103 at 12 weeks, n=138 at 16 weeks and n=115 at 22 weeks). During this time, 19% of women (31/167) were found to have a short CL ≤25mm; detected in 7 women at 12 weeks, 12 women at 16 weeks and 12 women at 22 weeks (Table 4-10). The 356 vaginal microbiome samples taken during this period (prior 24 weeks gestation) were grouped according to corresponding CL measurements, into short (CL ≤25mm) and non-short (CL >25mm) (Table 4-10).

Table 4-10 Number of women screened at by transvaginal ultrasound and found to have a short cervix at 12, 16 and 22 week time-points.

Screening time-point (weeks), n CL threshold Total women, n Total samples, n 12w 16w 22w ≤25mm 31 7 12 12 31 >25mm 136 96 126 103 325 TOTAL 167 103 138 115 356

CL = cervical length, w=weeks

119 Vaginal microbiota and the cervix in pregnancy

A greater proportion of L. iners sequence reads were detected in those women with a short cervix (44%) compared to those without (25%, P=0.047; Welches t test; Figure 4-5B) corresponding to a higher proportion of women with a short cervix having a CST III (L. iners) dominated microbial profile (49%, 15/31 versus 27%, 86/325 in the non-short group; P=0.01, Table 4-11).

In contrast, women with a normal cervical length exhibited a greater proportion of sequence reads attributable to L. crispatus (41%) than those with a short cervix (26%, P=0.049; Welches t test; Figure 4-5A). Although not reaching significance, this equated to a greater number of women assigned to CST I in the non-short cervix group (42% v 29% short group; P=0.2, Table 4-11).

Figure 4-5 Proportions of L. crispatus and L. iners sequences in women with short (≤25mm) and non-short cervices (>25mm) at screening before 24 weeks. (A) Mean L. crispatus sequence reads were higher in women with a long cervix (>25mm; P<0.05), (B) while mean L. iners sequences were higher in the short cervix group (P<0.05; Welches t-test).

120 Vaginal microbiota and the cervix in pregnancy

Table 4-11 Vaginal microbiota profiles of samples taken prior to 24 weeks: A comparison of community state types associated with short (≤25mm) and non short (>25mm) cervical lengths Cervical length threshold Screening <24weeks CL ≤25mm CL >25mm P value* n= 31 women n= 136 women Total samples taken, N % 31 8.7% 325 91.3%

GA at screening (w), Mean (range) 17.8 (11.7-23.6) 17.4 (11-23.7) ns

CL (mm), Mean (range) 21.7 (5-25) 38.2 (26-68) <0.001

Community state type n/N % n/N % I, L. crispatus 9/31 29.0% 139/325 42.8% 0.182 II, L. gasseri 2/31 6.5% 44/325 13.5% 0.400

III, L. iners 15/31 48.4% 86/325 26.5% *0.013 IV, Diverse species 2/31 6.5% 25/325 7.7% 1.000 V, L. jensenii 3/31 9.7% 31/325 9.5% 1.000 GA = gestational age, w= weeks, CL = cervical length (mm), P values= Fishers exact test

121 Vaginal microbiota and the cervix in pregnancy

4.7.1 Gestation at sampling, the cervix and vaginal microbiota

Vaginal microbiota were assessed at specific gestation time points (n=101 at 12 weeks, n=138 at 16 weeks, and n=115 at 22 weeks) to account for potential confounding of increasing gestational age. A short cervix was significantly associated with specific microbial communities at 12 and 16 weeks (P<0.05), but not at 22 weeks (Figure 4-6A). L. iners most strongly associated with a short CL ≤25mm at 12 weeks (71% v 31% non-short group; P=0.09; Fishers exact; Figure 4-6A, Table 4-12). Thereafter, L iners dominance declined slightly, but remained the most prevalent species in the short cervix group at 16 weeks (50% 6/12), higher than the non-short group (27%, 34/126; P=0.10). The correlation between a short cervix and vaginal microbiota did not continue at 22 weeks (P=0.06, Figure 4-6A).

L crispatus abundance increased with advancing gestation, most notably in the long cervix group, where species dominance increased from 38% at 12 weeks to 48% at 22 weeks. This remained higher than women with a short cervix (CST I: 14% at 12weeks and 33% at 22 weeks; non-significant; Figure 4-6A, B, Table 4-12).

122 Vaginal microbiota and the cervix in pregnancy

A Gestation at screening

I L. crispatus III L. iners II L. gasseri IV mixed species V L. jensenii

12 weeks 16 weeks 22 weeks * * ns 100 100 100

80 80 80

60 60 60

40 40 40 % of women of % % of women of % % of women of %

20 20 20

0 0 0 <25mm >25mm <25mm >25mm <25mm >25mm n=7 n=96 n=12 n=126 n=12 n=103 Cervical length Cervical length Cervical length

B Longitudinal microbial profiles in CL <25mm and >25mm groups

I L. crispatus III L. iners II L. gasseri IV mixed species V L. jensenii

CL 25mm, n=31 CL >25mm, n=325

80 80

60 60

40 40 % of women % of women 20 20

0 0 12w 16w 22w 12w 16w 22w Gestation at measurement (weeks) Gestation at measurement (weeks)

Figure 4-6 (A) Microbial profiles associated with a short cervix (CL≤25mm) at 12 and 16 weeks gestation differed significantly from those with a non-short cervix (>25mm) (P<0.05; 2way ANOVA). (A) The short group associated with a higher prevalence of L. iners compared to the non-short cervix groups at 12 weeks (71% v 34%; P=0.09) and 16 weeks (50% v 27%, P=0.10; two tailed Fishers exact). (B) L. iners abundance declined with advancing gestation in both the short and non-short groups. L. crispatus, the most prevalent species among the long cervix group, increased from 38% to 48% at 12 to 22 weeks. (CL= cervical length, w=weeks)

123 Vaginal microbiota and the cervix in pregnancy

Table 4-12 Vaginal microbiota profiles of short (CL≤25mm) compared to non-short (>25mm) cervices at screening 12, 16 and 22 weeks Cervical length threshold Screening time-point P value* ≤25mm >25mm 12 weeks N, % 7 6.8% 96 93.2%

GA at screening (w), Mean (range) 12.9 (11.7-14) 13.0 (11-14.9) ns

CL (mm), Mean (range) 23.0 (20-25) 41.0 (27-81) <0.001 Community state type n/N % n/N % I, L. crispatus 1/7 14.3% 37/96 38.5% 0.256 II, L. gasseri 0/7 0.0% 10/96 10.4% 1.000

III, L. iners 5/7 71.4% 33/96 34.4% 0.098 IV, Diverse species 1/7 14.3% 9/96 9.4% 0.522 V, L. jensenii 0/7 0.0% 7/96 7.3% 1.000

16 weeks N, % 12 8.7% 126 91.3%

GA at screening (w), Mean (range) 16.8 (15.1-18.3) 17.0 (15.4-18.4) ns CL (mm), Mean (range) 22.0 (15-25) 38.0 (26-65) <0.001 Community state type n/N % n/N % I, L. crispatus 4/12 33.3% 53/126 42.1% 0.760 II, L. gasseri 0/12 0.0% 17/126 13.5% 0.361

III, L. iners 6/12 50.0% 34/126 27.0% 0.106 IV, Diverse species 0/12 0.0% 8/126 6.3% 1.000 V, L. jensenii 2/12 16.7% 14/126 11.1% 0.631

22 weeks N, % 12 10.4% 103 89.6%

GA at screening (w), Mean (range) 21.6 (20-23.5) 21.2 (20-23.7) ns

CL (mm), Mean (range) 20.0 (5-25) 36.0 (26-68) <0.001 Community state type n/N % n/N %

I, L. crispatus 4/12 33.3% 49/103 47.6% 0.381 II, L. gasseri 2/12 16.7% 17/103 16.5% 1.000

III, L. iners 4/12 33.3% 19/103 18.4% 0.254 IV, Diverse species 1/12 8.3% 8/103 7.8% 1.000 V, L. jensenii 1/12 8.3% 10/103 9.7% 1.000 GA = gestational age, w= weeks, CL = cervical length (mm), P values= *Fishers exact

124 Vaginal microbiota and the cervix in pregnancy

4.8 Comparisons of term and preterm birth outcomes

4.8.1 Demographics

Cervical data matched to vaginal microbial profiles were grouped according to gestation at birth. The demographics of the term (≥37+0 weeks; 81%, 135/167), and preterm groups (<37 weeks; 19%, 32/167) are provided in Table 4-13. A higher proportion of Caucasian women delivered at term (69% v 50% preterm, P=0.06), while greater numbers of black women delivered preterm (38% v 15% term, P=0.005). Cervical lengths were lower among preterm compared term births from 12 weeks (30mm v 40mm) to 34 weeks (22mm v 27mm), although numbers were not powered to reach significance.

Table 4-13 Demographics of term versus preterm birth <37 weeks gestation Term birth, Preterm birth, ≥ 37 weeks < 37 weeks 135/167 (81%) 32/167 (19%)

Ethnicity, n/ N % Asian 22/135 16% 4/32 13% Black 20/135 15% 12/32 38%** Caucasian 93/135 69% 16/32 50% CL at sampling time-point, Mean (mm) ±SD

12w 40 ±11 38 ±10 16w 36 ±8 37 ±10 22w 35 ±7 33 ±14 28w 31 ±7 29 ±8 34w 27 ±7 22 ±9 Gestation at delivery, n/N % Early PTB, <34+0 w n/a 15/167 9% Late PTB, 34+0 to <37+0 w n/a 17/167 10%

Term, ≥37+0 w 135/167 81% n/a **P <0.01 Fisher exact: Term V Preterm birth group PTB = preterm birth, w = weeks gestation, CL = cervical length (mm)

125 Vaginal microbiota and the cervix in pregnancy

4.8.2 Vaginal Microbiota and gestation at birth

4.8.2a Vaginal bacterial genera and preterm birth

Microbial profiles among term birth (≥37+0 weeks, n=135), late preterm birth (34+0-36+6 weeks, n=15), and early preterm birth (<34+0 weeks, n=17) were first assessed at genera taxonomic level using hierarchical clustering analysis (Figure 4-7A).

At the first sampling time-point for each study participant (mean 14+4 weeks ±2.2), no relationship between genera classification and subsequent preterm birth (<37weeks) was observed. Of women experiencing preterm birth, 91% (29/32) had Lactobacillus spp. dominance indicating spontaneous preterm delivery is not associated with vaginal dysbiosis in the early second trimester (Figure 4-7A).

Further evidence for this was provided by richness and diversity metrics. Total number of species observed did not differ significantly among term (mean 6 ±0.5), late preterm (mean 4.3 ±0.8), and early preterm births (mean 4.8±1.2; P=0.38, ANOVA; Figure 4-7B). Consistent with this Shannon index of alpha diversity values did not differ between term (0.4±0.5), late preterm birth (0.2±0.4), and early preterm births (0.3±0.5, P=0.11, ANOVA; Figure 4-7C).

126 Vaginal microbiota and the cervix in pregnancy

Figure 4-7 Vaginal dysbiosis in the early second trimester (mean 14 weeks) does not associate with subsequent preterm birth. (A) Heatmap of ward hierarchical clustering of microbial genera from 167 women classifies samples according to subsequent gestation at delivery. Women delivering early <34+0 weeks (n=15, red) and late preterm 34+0 to 36+6 weeks (n=17, orange) have a predominantly Lactobacillus spp. dominated vaginal microbiota, as do women delivering at term >37+0 weeks (n=135, grey). (B, C) There is no correlation between the number of vaginal microbial species observed (B), or alpha diversity (C) and gestation at birth (ns: non-significant, Mann Whitney)

127 Vaginal microbiota and the cervix in pregnancy

4.8.2b Vaginal species and preterm birth

Vaginal microbiota profiles were further assessed at differing gestational time-points using species taxonomic data (Figure 4-8). Bacterial species abundance in women subsequently experiencing late preterm birth (34+0 to 36+6 weeks) were similar to those of term births (>37+0 weeks) regardless of gestation at time of sampling (Figure 4-8). However, significant differences in bacterial species were detected in those women experiencing early preterm birth (<34+0 weeks, P<0.05; 2-way ANOVA). In comparison, late preterm and term births were associated with high L. crispatus levels, most significantly from 16 weeks (P=0.0003; Figure 4-8A. This trend persisted until 28 weeks gestation (P=0.015; Figure 4-8A). In contrast, L. iners was significantly higher among the early preterm birth group, compared to both late preterm and term birth groups, through sampling from 12 to 28 weeks (P=0.0006, Figure 4-8C). L. gasseri abundance was not correlated with gestation at delivery (Figure 4-8B). Proportions of L. jensenii were most abundant among late preterm births (34-37weeks; P<0.05; ANOVA; Figure 4-8D).

128 Vaginal microbiota and the cervix in pregnancy

Figure 4-8 Box and Whisker plots demonstrating proportions of gene sequences from (A) L. crispatus, (B) L. gasseri, (C) L. iners, (D) L. jensenii according to gestation at sampling (12, 16, 22, and 28 weeks) and categorised gestation at birth: <34 weeks (red), 34-37 weeks (orange), and >37 weeks (grey). From 16 weeks L. crispatus abundance was greater among term births (>37 weeks) (A; P<0.01; ANOVA), while abundance of L. iners was elevated among preterm births (<34 weeks) from 12 week sampling (C; P<0.001; ANOVA).

129 Vaginal microbiota and the cervix in pregnancy

4.8.2c Community state types and preterm birth

The 505 sequenced samples were then classified into CSTs according to ward hierarchical clustering analysis (Figure 4-1). Distribution of CSTs differed significantly throughout antenatal sampling among women experiencing early preterm birth <34 weeks and birth > 34weeks (Figure 4-9A, Table 4-14). Early preterm birth was associated with a high abundance of L. iners (CST III) and deficiency in L. crispatus at 12 weeks (73% L. iners v 9% L. crispatus, P<0.01). This persisted from 16 weeks (64% L. iners v 9% L. crispatus, P<0.01) to 28 weeks (50% L. iners v 0% L. crispatus, P<0.05; Figure 4-9B, Table 4-14). Term birth, in contrast, was associated with high prevalence of CST I (L. crispatus) at 12 weeks (42%), 16 weeks (46%), 22 weeks (49%) and 28weeks (49%) compared to L. iners (32%, 25%, 19% and 28% at respective gestations; P<0.01, 2-way ANOVA, Bonferroni multiple comparison post-test; Figure 4-9B, Table 4-14).

130 Vaginal microbiota and the cervix in pregnancy

A Gestation at sampling

12 weeks, n=103 16 weeks, n=138 22 weeks, n=115 28 weeks, n=77 * * * * Community state type (CST): 100 100 100 100 CSTCST II L.CrispatusL. crispatus 80 80 80 80 CSTIICST II L.GasseriL. gasseri 60 60 60 60 CSTCST IIIIII L.InersL. iners 40 40 40 40 % of women of % % of women of % % of women of % % of women of % CSTCST IVIV Diverse Div ers species e s pec ies 20 20 20 20

CSTCST VV L.JenseniiL. jensenii 0 0 0 0 <34w 34-37w >37w <34w 34-37w >37w <34w 34-37w >37w <34w 34-37w >37w n=11 n=13 n=79 n=11 n=13 n=114 n=7 n=12 n=96 n=6 n=10 n=61 Gestation at birth (weeks) Gestation at birth (weeks) Gestation at birth (weeks) Gestation a t birth (weeks)

B Gestation at birth

Preterm birth <34 weeks , n=15 Preterm birth 34 +0 to 36+6 weeks, n=17 Term birth >37 weeks , n=135 Community state type (CST): 80 *** 80 *** 80 *** CSTCST II L.CrispatusL. crispatus

60 60 60 CSTIICST II L.GasseriL. gasseri

CST III L.Iners 40 40 40 CST III L. iners % of women of % % of women of % % of women of % CSTCST IVIV Diverse Div ers species e s pec ies 20 20 20 CSTCST VV L.JenseniiL. jensenii

0 0 0 12w 16w 22w 28w 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Gestation at sampling (weeks) Gestation at sampling (weeks) Gestation at sampling (weeks) Figure 4-9 CST III (L. iners) dominance between 12 and 28 weeks associated with subsequent early preterm birth (<34+0 weeks, n=15), whilst CST I (L. crispatus) dominance associated with term birth (>37weeks, n=135). (A) At 12 weeks 73% of early preterm birth <34+0 weeks were assigned to CST III (8/11), compared to 32% of term births (25/79, P<0.05), while L. crispatus (CST I) was dominant in 9% (1/11) and 42% (33/79) respectively (P<0.05; ANOVA, Bonferroni post- test). (B) This trend persisted throughout 16, 22 and 28 weeks sampling where women delivering early preterm (<34+0 weeks) were characteristically dominant in L. iners, and deficient in L. crispatus (P<0.001). Term delivery in contrast was associated with high L. crispatus abundance with advancing pregnancy (P<0.001; 2way ANOVA, Bonferroni post-test). (w=weeks, CST= community state type).

131 Vaginal microbiota and the cervix in pregnancy

Table 4-14 Vaginal microbial profiles as a function of gestation at sampling and subsequent gestation at birth

Community state type, Species

Categorised Gestation CST IV gestation at at CST I CST II CST III Diverse CST V birth sampling L. crispatus L. gasseri L. iners species L. jensenii TOTAL (weeks) (weeks) n/N % n/N % n/N % n/N % n/N % N

12w 1/11 9% 1/11 9% 8/11 73%†** 0/11 0% 1/11 9% 11

Preterm 16w 1/11 9% 1/11 9% 7/11 64%†** 0/11 0% 2/11 18% 11 <34 weeks 22w 1/7 14% 1/7 14% 3/7 43% 0/7 0% 2/7 29% 7

28w 0/6 0% 1/6 17% 3/6 50%* 0/6 0% 2/6 33% 6

12w 4/13 31% 1/13 8% 5/13 38% 1/13 8% 2/13 15% 13

16w 5/13 38%* 1/13 8% 3/13 23% 1/13 8% 3/13 23% 13 Preterm 34+0-36+6 22w 5/12 42%* 2/12 17% 2/12 17% 1/12 8% 2/12 17% 12 weeks 28w 4/10 40%* 0/10 0% 3/10 30% 1/10 10% 2/10 20% 10

34w 4/8 50%* 1/8 13% 1/8 13% 1/8 13% 1/8 13% 8

12w 33/79 42%† 8/79 10% 25/79 32% 9/79 11% 4/79 5% 79

16w 53/114 46%†** 15/114 13% 28/114 25% 7/114 6% 11/114 10% 114 Term >37 weeks 22w 47/96 49%†*** 16/96 17% 18/96 19% 8/96 8% 7/96 7% 96

28w 30/61 49%** 5/61 8% 17/61 28% 5/61 8% 4/61 7% 61

34w 29/63 46% 7/63 11% 20/63 32% 4/63 6% 3/63 5% 63 ***P<0.001**P<0.01,*P<0.05; 2way ANOVA, Bonferroni post-tests for comparison of CST I v III at gestation at sample †P<0.05; 2way ANOVA, Bonferroni post-tests for CST comparison of gestation at birth <34w v 34-47 v >37w. CST = community state type based on ward HCA of species data; w= weeks

132 Vaginal microbiota and the cervix in pregnancy

4.8.2d Survival curves Kaplan-Meier survival curves compared duration of the pregnancy according to vaginal CST when sampled at 12, 16 and 22 weeks (Figure 4-10). L. crispatus dominance associated with longer duration of pregnancy compared to L. iners at all three sampling time-points. Of the women assigned to CST I at 12 weeks, 94% remained pregnant at 34 weeks, and 84% at 37 weeks. This was significantly more than CST III at 12 weeks (76% at 34 weeks and 58% at 37 weeks; P=0.01, Figure 4-10A). Based on 16 week samples, 88% of CST I and 63% CST III were undelivered at 37 weeks (P=0.02, Figure 4-10B) and at 22 weeks, 80% and 56% respectively (P=0.003; Gehan-Breslow-Wilcoxon Test, Figure 4-10C).

A 12 weeks B 16 weeks

100 100

80 80

60 60

40 40

Percent survival Percent CST I L. crispatus survival Percent CST I L. crispatus 20 P=0.01 20 P=0.02 CST III L. iners CST III L. iners 0 0 0 4 8 12 16 20 24 28 32 36 40 0 4 8 12 16 20 24 28 32 36 40 Weeks Gestation Weeks Gestation

C 22 weeks

100

80

60

40

Percent survival Percent CST I L. crispatus 20 P=0.003 CST III L. iners 0 0 4 8 12 16 20 24 28 32 36 40 Weeks Gestation Figure 4-10 Kaplan-Meier survival curves demonstrate earlier gestation at delivery associated with L. iners than L. crispatus dominance at 12 weeks (P=0.01), 16 weeks (P=0.02), and 22 weeks (P=0.003; Gehan-Breslow-Wilcoxon Test). Of those abundant in L. crispatus at 12 weeks, 94% remained undelivered at 34 weeks, compared to 76% of L. iners (P=0.01). At 22 weeks 97% of L. crispatus and 82% of L. iners were undelivered at 34 weeks (P=0.003; Gehan-Breslow-Wilcoxon Test)

133 Vaginal microbiota and the cervix in pregnancy

4.9 Cervical length, the vaginal microbiota and gestation at birth

Linear regression was then used to compare dominance of L. iners or L. crispatus as a function of cervical length measurements taken before 24 weeks against gestation at birth. In the presence of an L. iners dominated microbiome, cervical shortening correlated with earlier gestation at birth (r=0.32, P=0.002). This was not the case in the presence of an L. crispatus dominant microbiome (r=0.15, P=0.06; Figure 4-11)

42

40

38

36

34

32

30 CST I r=0.15 28 Gestation at birth Gestation (weeks) at birth CST III r=0.32* P<0.01 26

24 60 55 50 45 40 35 30 25 20 15 10 5 0

Cervical length (mm) Figure 4-11 Linear regression plotting cervical length measured before 24 weeks, against subsequent gestation at birth. As cervical length shortens, L. iners dominance associates with an earlier gestation at birth (r=0.32, P=0.002), while L. crispatus dominance does not (r=0.15, P=0.06). (Linear regression, Spearman rank correlation.)

Overall, a short CL ≤25mm associated with earlier gestation of birth when corresponding microbiota was classified as CST III (L. iners, mean 35+2 weeks, n=13) than CST I (L. crispatus mean 38+6 weeks, n=11; P=0.03, t test; Figure 4-12A). Importantly, CST III was also a risk factor for preterm birth in women with a longer cervix >25mm compared to CST I (CST III mean birth at 36+2 weeks, n=69, versus CST I mean birth at 38+5 weeks, n=106; P<0.001, t test, Figure 4-12A).

134 Vaginal microbiota and the cervix in pregnancy

Gestation at screening significantly influences this relationship however. Given a short CL≤25mm, earlier gestation at detection correlates with earlier gestation at birth, but only in the presence of an L. iners dominated microbiome (r=0.37, P=0.06; Figure 4-12B). In the case of L. crispatus dominance, a short CL≤25mm does not increase the risk of preterm birth irrespective of gestation at detection (r=-0.38, P=0.25; Figure 4-12B). This is a clinically significant finding, as currently detection of a short cervix at early gestation at screening (<16weeks) confers high specificity and significantly elevated risk of subsequent preterm birth. These findings imply the predictive accuracy of CL measurements is dependent on corresponding microbiota.

Cervical length screening <24 weeks: L. crispatus v L. iners

A 42 ns

40 * *** L. crispatus 38 L. iners

36 Other species

34

32 Gestation at birth (weeks) birth at Gestation 30 <25mm >25mm Cervical length

BCCL: 25mm CL > 25mm

40 40

36 36

32 32

28 28 Gestationat birth (weeks) Gestationat birth (weeks) 24 24 12 16 20 24 12 16 20 24 Gestation at screening (weeks) Gestation at screening (weeks) Figure 4-12 (A) At screening <24weeks, detection of a short CL≤25mm with L. crispatus-dominance associates with later gestation at birth (mean 38+6 weeks ± 0.4, n=11), than with L. iners-dominance (mean 35+2 weeks ±1.2, n=13; P=0.03, t test). (B) Linear regression analysis demonstrates gestation at screening positively correlates with gestation birth when CL ≤25mm in association with L. iners (r=0.37, P=0.06), but not L. crispatus (r=-0.38, P=0.25; spearman correlation). (C) A long cervix (>25mm) in with L. iners-dominance associates with earlier birth (mean 36+2 weeks ± 0.5, n=69) than L. crispatus irrespective of gestation at screening (38+5 weeks ± 0.2, n=106; P<0.001, t test),. (CL= cervical length)

135 Vaginal microbiota and the cervix in pregnancy

4.10 The prediction of preterm birth

4.10.1 The predictive value of vaginal microbiota Findings thus far indicate that microbial profiles at 12 and 16 weeks differ among women going onto deliver before and after 34+0 weeks. Vaginal microbiota at these time-points were therefore assessed for their prediction of subsequent preterm birth.

L. iners dominance at 12 weeks predicted preterm birth <34+0 weeks with 73% sensitivity and 67% specificity, and at 16 weeks with 64% and 76% respectively. Also, an absence of L. iners conferred high negative prediction value (NPV) (95% and 96% at 12 and 16 weeks) (Table 4-15). Detection of L jensenii and/or L. iners further improved the detection rate (sensitivity) of preterm birth <34 weeks to 82% at 12 weeks (Table 4-15).

High L. crispatus abundance was overall protective against preterm birth, associated with 91% specificity and above 97% positive predictive value (PPV) for birth after 34 weeks (Table 4-15).

136 Vaginal microbiota and the cervix in pregnancy

Table 4-15 Prediction accuracies of community state types for preterm birth <34weeks and birth >34 weeks at screening time-points 12 and 16 weeks.

Community state type, Species

Screening gestation CST IV, CST I, CST II, CST III, CST V, CSTs Diverse L. crispatus L. gasseri L. iners L. jensenii III &V species 12 Preterm birth Sens 9% 9% 73% 0% 9% 82% weeks <34w Spec 60% 90% 67% 89% 93% 61%

PPV 3% 10% 21% 0% 14% 20%

NPV 85% 89% 95% 88% 90% 97%

Birth >34w Sens 32% 14% 36% 9% 10% 46%

Spec 91% 91% 27% 100% 91% 18%

PPV 97% 94% 84% 100% 92% 86%

NPV 11% 9% 4% 9% 9% 3%

16 Preterm birth Sens 9% 9% 64% 0% 18% 82% weeks <34w Spec 54% 87% 76% 94% 89% 65%

PPV 2% 6% 18% 0% 13% 17%

NPV 87% 92% 96% 92% 93% 98%

Birth >34w Sens 42% 12% 30% 7% 9% 39%

Spec 91% 91% 36% 100% 82% 18%

PPV 98% 94% 86% 100% 86% 86%

NPV 11% 8% 4% 8% 7% 2%

CST = community state type based on ward HCA of species data; w= weeks; Sens=sensitivity; Spec=specificity; PPV/ NPV= positive/negative predictive values

137 Vaginal microbiota and the cervix in pregnancy

4.10.2 The predictive value of vaginal microbiota and cervical length Given the significant interaction effect between CL, the vaginal microbiome and subsequent birth gestation, predictive accuracies of assigned CSTs in conjunction with CL and CV measurements were determined using receiver operating curves (ROC) for outcomes of preterm birth <34 weeks. Figure 4-13 demonstrates improved predictive accuracy of cervical parameters length and volume in the presence of L. crispatus compared to L. iners. At 12, 16 and 22 week sampling the areas under the ROC curve (AUC) for CL measurements were 0.92, 0.90 and 0.82 in the presence of an L. crispatus dominant vaginal microbiome. This is compared to AUCs 0.52, 0.61 and 0.61 in the presence of L. iners dominance at respective time-points (Figure 4-13 A, C and E). Cervical volume demonstrated a similar trend; high L. iners abundance negated the predictive value of CL and CV measurements, while it was maintained in the presence of L. crispatus dominance (Figure 4-13B, D, F). The observed poor predictive value of CL and CV in the presence of L. iners dominance likely relates to the pathogenicity of L. iners alone; as L. iners is strongly associated with preterm birth irrespective of cervical length measurements, the ability of CL and CV measurements to reach high specificity becomes increasingly difficult without compromising sensitivity (the true positive rate). For example, at 12 weeks a CL=29mm achieves high specificity (>80%) when associated with L. crispatus (with 92% sensitivity). To achieve comparable specificity with L. iners, CL>44mm is required, and this is associated with 30% sensitivity (Figure 4-13A). Therefore in the presence of L. crispatus, CL appears to retain its predictive accuracy, but in the presence of L. iners the risk of preterm birth negates any reassurance gained by a long CL. This trend continues to 22 weeks (Figure 4-13E) Therefore in order to identify pregnancies most at risk, a combination of L. iners and CL measurements with a threshold of ≤25mm was employed (Table 4-16). At the first screening time point (14+4 weeks gestation, n=167), the detection of L. iners and/or a CL ≤25mm provides a much improved combined sensitivity (detection rate) for preterm birth <34 weeks (87%) compared to L iners alone (67%). This is at a small compromise of specificity: 68% for L. iners and/or CL≤25mm compared to 71% for L. iners alone. Prediction of preterm birth <37 weeks is also improved through the addition of a CL measurements ≤25mm (Table 4-16). The lowering of the CL threshold to ≤20mm does not further improve preterm birth prediction for either <34 weeks (80% sens, 70% spec) and or <37 weeks (53% sens, 71% spec) (Table 4-16).

138 Vaginal microbiota and the cervix in pregnancy

ROC curves for prediction of preterm birth <34weeks

Cervical length Cervical volume

AB12 weeks 12 weeks

100 100

80 80

60 60

40 40

Sensitivity % L. crispatus (AUC 0.95) Sensitivity % L. crispatus (AUC 0.92) 20 L. iners (AUC 0.53) 20 L. iners (AUC 0.52) Other species (AUC 0.70) Other species (AUC 0.65) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1 - Specificity% 1 - Specificity%

CD16 weeks 16 weeks 100 100

80 80

60 60

40 40 Sensitivity % L. crispatus (AUC 0.90) Sensitivity % L. crispatus (AUC 0.85)

20 L. iners (AUC 0.61) 20 L. iners (AUC 0.63)

Other species (AUC 0.64) Other species (AUC 0.81) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1 - Specificity% 1 - Specificity%

EF22 weeks 22 weeks 100 100

80 80

60 60

40 40

Sensitivity % L. crispatus (AUC 0.82) Sensitivity % L. crispatus (AUC 0.93)

20 L. iners (AUC 0.61) 20 L. iners (AUC 0.76)

Other species (AUC 0.71) Other species (AUC 0.69) 0 0 0 20 40 60 80 100 0 20 40 60 80 100 1 - Specificity% 1 - Specificity%

Figure 4-13 Cervical length and volume measurements predict preterm birth <34weeks with high accuracy given a vaginal microbial dominance of L.

139 Vaginal microbiota and the cervix in pregnancy

crispatus at 12, 16 and 22 week screening. Measurement of CL and CV in the presence of L. iners provides poor preterm birth prediction, while improved predictive accuracy is achieved with L. crispatus.

Table 4-16 Predictive accuracies of L. iners in combination with CL measurement for preterm birth at initial screening <16 weeks.

L. iners and/or L. iners and/or Prediction of: L. iners CL ≤25mm CL ≤20mm

Sens 67% 87% 80% Preterm birth Spec 71% 68% 70% <34 weeks PPV 19% 20% 20% NPV 96% 97% 96% Sens 50% 56% 53%

Preterm birth Spec 72% 69% 71% <37 weeks PPV 30% 30% 30% NPV 86% 87% 86% CL=cervical length; Sens=sensitivity; Spec=specificity; PPV/ NPV= positive/negative predictive values

140 Vaginal microbiota and the cervix in pregnancy

Discussion

Healthy vaginal microbiota has traditionally been synonymous with low bacterial diversity and Lactobacilli spp. dominance. This study finds that the benefit of Lactobacilli spp. dominance in pregnancy is species specific, where dominance of L. crispatus is advantageous, while L. iners is a risk factor for preterm birth. In this chapter, I describe the predictive role that the second trimester vaginal microbial composition plays in differentiating early from late preterm birth (before and after 34+0 weeks). A high abundance of L. crispatus is highly specific for term birth, with a false positive rate of just 2% in the high-risk population analysed.

Significantly, this study also finds that vaginal dysbiosis in the second trimester does not contribute to preterm birth risk; the majority of preterm births in this cohort (91%) occurred despite a dominance of Lactobacillus spp. This finding is in contrast to a recent report by Digiulio et al 181, who associated a deficiency of Lactobacillus spp. with spontaneous preterm births, albeit in a small study population (6/40 pregnancies delivered at relatively late mean preterm gestation of 36 weeks). However, in support of my findings, Petricevic et al 182, in a low risk cohort of 111 pregnancies, described a high prevalence of L. iners among their 13 preterm births (none whom delivered before 33 weeks). Their study was limited by the use of denaturing gradient gel electrophoresis (DGGE) for the characterization of only major Lactobacillus species and could not identify other pathobionts in the samples. In contrast to these reports, Romero et al 174 recently reported no association between preterm birth and vaginal microbiota composition in a small cohort in 22 pregnancies, of which the majority (90%) were Afro-Caribbean 225. Ethnicity has been previously shown to significantly influence vaginal microbial composition, whereby higher rates of CST IV are observed in black women 155, 226. High background prevalence of a dysbiotic microbiome in the Romero cohort, and its small sample size, likely accounts for the reported inconsistences in their findings.

L. crispatus, the most stable vaginal commensal 177, 182, 227, is rarely be displaced through its ability to inhibit of anaerobic pathogens in the vagina 228, 229, and has been shown to negatively associate with clinical BV 177, STIs and HIV 140. The findings are consistent with the observations within my study, whereby L. crispatus dominance maintains high stability in pregnancy, and provided protection against preterm birth. Women delivering before 34 weeks displayed a relative deficiency of L. crispatus, with a greater abundance of L. iners particularly prior to 22 weeks. The differing protective potential of Lactobacilli species may be in part explained by their

141 Vaginal microbiota and the cervix in pregnancy

capacity to produce L- and D-lactic acid isomers 230. L. iners is unique among the Lactobacilli species, in that it lacks the ability to synthesise D-lactic acid 230. Therefore women exhibiting a vaginal microbiome dominated by L. iners tend to have an increased concentration of L-lactic acid, which has been shown to promote expression of vaginal extracellular matrix metalloproteinase inducer (EMMPRIN) and subsequent activation of matrix metalloproteinase-8 (MMP8), which may subsequently modulate cervical integrity230. Conversely, no such relationship is observed in women with vaginal microbial communities dominated by L. crispatus, which instead preferentially excretes high levels of D-lactic acid. L iners is also unique from the other major vaginal Lactobacillus spp. in that it does not produce hydrogen peroxide 231. Unlike other Lactobacillus spp., L iners has been shown to induce pro-inflammatory signalling in-vitro 53, 232. Consequently L. iners, associated with higher vaginal pH, creates an environment permissible to colonisation by BV-associated vaginal pathogens, in which it largely tolerates co-existence 133, 233, 234. L. iners has frequently been reported as an intermediary in the transition towards CST IV-associated states 168, and the findings of this study add to a growing body evidence highlighting the potential negative impact L. iners dominance can impart on reproductive health outcomes.

Interaction between vaginal microbiota and cervical length in pregnancy

A second novel finding within this study was the identification of vaginal microbiota correlated with a short cervix in pregnancy. Currently measurement of CL is the primary tool employed for preterm birth surveillance. Traditionally, CL screening is performed at 20 to 24 weeks, as measurements at this gestation are highly sensitive for preterm birth 83. Although a short cervix is universally defined as CL ≤25mm before 24 weeks 83, earlier screening provides improved specificity for preterm birth, reflecting an earlier activation of the parturition cascade driving premature cervical shortening and ripening 38. Here I show that L. iners dominance is associated with a short cervix in the early second trimester; a clinical finding in itself highly specific for preterm birth 89.

Interestingly, gestational age at screening appeared to impact on this relationship. In the presence of L. iners dominance, earlier gestation at detection of a short cervix correlated more strongly with preterm birth than later gestation at screening. This is consistent with performance of CL screening; high specificity is achieved with early screening (e.g. <16 weeks), and high sensitivity at late screening (e.g. 22 weeks). This study reveals that this is not the case in the

142 Vaginal microbiota and the cervix in pregnancy

context of L. crispatus dominance. Risk of preterm birth is not increased with earlier detection of a short cervix. Importantly, this provides insight into differing underlying mechanisms or cervical shortening, and the protection L. crispatus provides in the context of high-risk pregnancy.

The vaginal microbiome and prediction of preterm birth

Preterm birth is defined at birth <37 weeks gestation, however major risk of morbidity and mortality is associated with birth <34 weeks 6. Prediction and prevention of these early preterm births remains a major challenge. At optimal screening performance (20-24 weeks), the use of CL measurements for prediction of preterm birth <34 weeks (using thresholds of CL <15mm, <25mm, <35mm) provides sensitivities and of 59%, 67% and 79% and specificities of 85%, 78% and 60% respectively 84, 86, 88, 89. Shortcomings of measuring CL between 20 and 24 weeks are the increased complication risk associated with intervention at this late gestation 219, and neglect of late miscarriages who deliver with cervical insufficiency prior to this gestation.

This study finds vaginal microbiota provides comparable sensitivity as CL measurement for prediction of preterm birth <34 weeks, but at an earlier gestational, therefore potentially enabling timely interventions. Prior to 16 weeks, L. iners dominance predicted preterm birth <34+0 weeks with 67% sensitivity and 71% specificity. The addition of CL measurements improved screening performance to 87% sensitivity and 68% specificity for CL ≤25mm and 80% and 70% for CL ≤20mm respectively.

Furthermore this study reveals that a long cervix does not protect against preterm birth in the context of L. iners dominance. Higher thresholds of CL were required to reach high negative predictions in the presence of L. iners than L crispatus. The vaginal microbiota may therefore explain the false negative rate associated with a CL threshold of ≤25mm. In the future, a combined screening tool utilising both CL and vaginal microbiota to identify women at risk may be of clinical use.

Limitations

Studies have shown that L. iners is often present together with dysbiosis-associated pathobionts, namely of S. agalactiae (Group B streptococcus, GBS) E. coli, G. vaginalis, A. vaginae and Candidiasis 235. A major limitation of this study is that these pathobionts were either

143 Vaginal microbiota and the cervix in pregnancy

not, or minimally detected by 16S sequencing 236. It is therefore possible that these pathobionts play a role in the pathogenesis of L. iners-associated preterm birth, but that this may have been overlooked through the use of NGS sequencing. Species-specific qPCR of these bacteria 237 may be provide some insight with respect to their presence co-existence, and pathogenesis in pregnancy.

There was significant participation bias among women that consented to scans, but not swabs. Ethnicity and smoking status were two of the features that differed significantly among these groups, whereby non-smoking Caucasians were more likely to consent to swabs. Both of these factors influence vaginal microbiota composition 155, 169, and so it is possible that the presented data are skewed through participation bias. A further limitation of this study is the contribution of multiple confounders influencing vaginal microbiome composition, in particular ethnicity. A more robust assessment of the reported relationship between vaginal microbiota, cervical length and preterm birth would be to treat Black ethnicity as a confounder in a multivariate statistical model. Furthermore, given the longitudinal nature of the data, added power and clinical relevance may be gained through analysing both species abundance and scan data as longitudinal time series, rather than multiple independent events as per current cross-sectional analysis.

The exclusion of samples post-cerclage insertion is a large does introduce a potentially severe bias. To assess the impact of this bias, sensitivity analyses of microbiota post cerclage insertion, compared to pre-cerclage insertion may have been useful. Unfortunately the DNA from these samples, although extracted, were not sequenced, therefore this assessment could not be easily estimated.

In summary This study finds that individual Lactobacilli species play an integral role in determining outcome in pregnancies at high risk preterm birth, and detection of vaginal microbial composition in the early second trimester may be used to stratify preterm birth risk. Specifically L. crispatus dominance is associated with term birth, while high L. iners abundance warrants close surveillance.

144 A comparison of previous preterm birth and cervical treatment in pregnancy

5 A COMPARISON OF PREVIOUS PRETERM BIRTH AND CERVICAL TREATMENT IN PREGNANCY

145 A comparison of previous preterm birth and cervical treatment in pregnancy

Chapter abstract

Hypothesis Cervical cancer screening programs have been highly effective in identifying pre-cancerous cervical intra- epithelial neoplasia (CIN) among reproductive age women. As a result, excisional cervical treatment (CT) for CIN in has increased, as have the adverse sequelae of treatment. This includes an elevated risk of preterm birth in subsequent pregnancy. The hypothesis of this work was that underlying mechanisms for preterm birth in women receiving treatment for CIN involve increased vaginal dysbiosis and/or deficient cervical volume secondary to loss of cervical tissue at treatment and hence compromised mechanical strength of the cervix.

Aims To characterise and compare vaginal microbiota and cervical morphology of women: 1) With a prior spontaneous preterm birth <37+0 weeks (PTB), without excisional CIN treatment 2) In their first pregnancy post-excisional treatment for CIN (CT)

Methods Two groups of high-risk pregnancies (PTB and CT women) were recruited and followed up longitudinally at 12, 16, 22 and 28 weeks gestation. High vaginal swabs for 16s rRNA gene sequencing and transvaginal scans for cervical length (CL) and volume (CV) measurements were performed at every time-point.

Results 133 pregnant women (n=66 CT, n=67 previous PTB) consented to both HVS and transvaginal scans, of which 24% (32/133) delivered preterm <37+0 weeks. Fewer CT women delivered preterm (11%, 7/66) than women with a previous PTB (37%, 25/67). This was despite persistently lower CL and CV in the CT group compared to the prior PTB group throughout pregnancy (P<0.05). Corresponding vaginal microbiome profiles in the CT group displayed a greater prevalence of L. crispatus dominance than the previous PTB group (61% v 20%, P<0.001); who had higher prevalence of both vaginal dysbiosis (13% v 4%, P<0.01) and L. iners dominance (41% v 17%, P<0.001).

Conclusion Excisional cervical treatment is associated with a reduction in cervical volume and length in pregnancy, but greater L. crispatus dominance and lower preterm birth rates compared to women with a prior preterm birth, who are instead characterised by higher L. iners prevalence and dysbiosis.

146 A comparison of previous preterm birth and cervical treatment in pregnancy

Introduction

Preterm birth is a syndrome of multiple aetiologies 50. Women with a prior preterm birth <37 weeks are at increased risk of subsequent preterm birth (OR 7.4, 95% CI 5.94–8.22) 238. The introduction of widespread cervical cancer screening programmes have increased detection of and treatment for pre-cancerous CIN, effectively decreasing the incidence of cervical cancer by 44% over the last 25 years 239. More recently, CIN treatment has been associated with elevated rates of preterm birth in subsequent pregnancy, which has become the focus of growing concern 66, 71. Excisional methods of treatment (cold knife conisation, large loop excision of the transformation zone (LLETZ), and laser conisation) in particular are associated with adverse pregnancy outcome 69, 70. This includes low birth weight, up to a 2-fold increase preterm birth 66, 71 and a 2.8-fold increased risk of perinatal death, mostly relating to a 4-fold increase in very early preterm birth <28 weeks 240.

Figure 5-1 Excisional cervical treatment for cervical intra-epithelial neoplasia (CIN) in non-pregnant women. Methods of excisional cervical treatment include cold knife conization (CKC) as shown, large loop excision of the transformation zone (LLETZ) and laser conization.

The underlying mechanism for preterm birth in these women remains uncertain. A proposed hypothesis is mechanical weakness of the cervix secondary to loss of cervical tissue at the time of excisional treatment. There is evidence for a persistent deficit of cervical tissue in non- pregnant women as the volume of cervix removed at LLETZ is reportedly proportional to subsequent CL and CV at 6 month follow up 241. Importantly this deficit appears to persist into subsequent pregnancy; Poon et al reported pregnancies post-LLETZ had shorter CL at 22 weeks gestation than untreated women 218. Furthermore there is evidence for dose response effect of cervical treatment; the larger the proportion of cervix removed, the larger and more serious the implications are in pregnancy 72-75, 242.

147 A comparison of previous preterm birth and cervical treatment in pregnancy

An alternate hypothesis for increased preterm birth post-CIN treatment is the augmentation of immunomodulatory pathways of parturition resulting from underlying human papilloma virus (HPV) infection, a pre-requisite for CIN 73. In mice models HPV infection of the cervix during pregnancy reduces the capacity of the female reproductive tract to prevent bacterial infection of the uterus 243 - a major cause of preterm birth 6, 48. Extrapolating this to a human model, it may be postulated that underlying immune-regulation plays a role in changing susceptibility to the co-existing vaginal microbiota. In non-pregnant women, there appears to be an interaction between the vaginal microbiome and HPV infection, whereby vaginal dysbiosis associates with slowest regression of HPV 244 as well as high grade CIN severity 245. As yet, the vaginal microbiota in women post- CIN treatment in pregnancy has not yet been characterised.

The findings described in Chapter 4 reveal the pathogenesis of L. iners dominated microbiota as early as 12 weeks, with respect to both premature cervical shortening and subsequent preterm birth risk. What remains uncertain is whether the vaginal microbial profiles differ in women with a previous preterm birth, and those with excisional cervical treatment. These distinct groups, both at risk of preterm birth, are likely to have substantially different cervical anatomy and vaginal microbiota contributing to their pregnancy outcome.

148 A comparison of previous preterm birth and cervical treatment in pregnancy

Aim To characterise the vaginal microbiome at cervical morphology (length, volume and vascularity) in women with pre-pregnancy excisional cervical CIN treatment (CT) and compare this to women with a previous preterm birth (PTB), to better understand the anatomical and microbiological mechanism underlying their risk of preterm birth.

Hypotheses 1. Vaginal dysbiosis is more prevalent among women with a history of CT than previous PTB groups. 2. Pregnancies going on to deliver preterm birth in the CT group associate with vaginal dysbiosis, whilst preterm birth in the prior PTB group is primarily associated with L. iners- dominance. 3. Rates of preterm birth in the CT women associate with lower cervical volumes and lengths than the prior PTB group.

149 A comparison of previous preterm birth and cervical treatment in pregnancy

Study design

Patient recruitment and sample collection This chapter assesses two groups of high-risk pregnancies categorised according to their indication for referral to prematurity clinic: 1) pregnant women with prior excisional cervical treatment (CT group) and 2) women with a prior spontaneous preterm birth <37weeks (PTB group). Women were recruited at their first clinic attendance (approximately 12 weeks gestation) and followed up longitudinally at 16, 22, 28 and 34 weeks gestation. As described in Methods (Chapter 2), at each time point cervicovaginal fluid was sampled for 16s rRNA gene sequencing and data on cervical parameters (CL, CV, VI and FI) were collected from a transvaginal scan, and analysed using the VOCALTM (Virtual Organ Computer-aided AnaLysis) software programme.

Eligibility criteria for the CT group was a first singleton pregnancy following an excisional conization of depth ≥10mm75 for CIN grade II or III (including LEEP, LLETZ and CKC). Eligibility for the prior PTB group was a spontaneous preterm birth <37+0 weeks of a singleton pregnancy in the previous pregnancy. As severity of the CIN histology and time since excision have not been shown to associate with the risk of preterm birth75, this data was not analysed for the CT group. Exclusion criteria for study participation are as described in Chapter 2.

150 A comparison of previous preterm birth and cervical treatment in pregnancy

Figure 5-2 Flow chart of study design detailing recruitment cohorts, sampling time-points, procedures and data analysis

151 A comparison of previous preterm birth and cervical treatment in pregnancy

Statistical analyses of scan and sequence data Cervical length (CL), volume (CV) and vascularity indices (VI and FI) were assessed for differences between CT and PTB cohorts with advancing gestational age. Statistical tests for significance included one-way ANOVA with Bonferroni multiple comparison post-test, two tailed t-tests, and fisher exact and chi-squared tests where applicable. Linear regression was used to assess potential relationships between cervical parameters (CL, CV, VI and FI) and gestational age at birth according to cohort CT versus PTB.

Bacterial gene sequence data were examined for differences in total number of species observed and alpha diversity (as assessed using the Shannon index) between CT and prior PTB groups. Examination of statistical differences between vaginal microbiota was performed using the STAMP software package 201. Vaginal bacterial communities were assessed at species taxonomic level and classified into community state types (CSTs) as previously described 155. Two way ANOVA and fisher exact tested assessed differences in proportions of assigned CSTs among CT and prior PTB groups, and as a function of categorised gestation at birth (early preterm birth <34+0 weeks, late preterm birth 34+0 to 36+6 weeks, and term birth ≥37+0 weeks).

152 A comparison of previous preterm birth and cervical treatment in pregnancy

Results

5.1 Cervical assessment by transvaginal ultrasound

5.1.1 Participant demographics A total of 199 women were recruited from preterm surveillance clinics at a mean of 12+6 weeks. Of these, 113 (57%) were in their first pregnancy post-excisional cervical treatment for CIN (CT group), and 86 women (43%) had had a previous spontaneous preterm birth <37weeks (PTB group). The demographics of these groups are provided in Table 5-1. The CT group consisted of more Caucasians than PTB group (72% vs. 30%, P<0.001), while a greater number of black (47%) and Asian (23%) women were in the PTB group (P<0.01). Cervical cerclage intervention, indicated by cervical shortening to ≤25mm before 24 weeks, was higher in women with a previous preterm birth (51%) compared to previous cervical treatment (CT) (36%, P=0.03).

Time from excisional treatment to conception of the index pregnancy in the CT group ranged from 18 months to 5 years (mean 3.1 years) and did not associate with rates of preterm birth in the index pregnancy.

The overall preterm birth rate <37 weeks was 22% (43/199). Women in the PTB group accounted for significantly higher proportion of preterm births; 37% (32/86) delivered <37weeks compared to 10% (11/113) of the CT group (RR 3.8, 95% CI 2.0-7.1; P<0.001; Figure 5-3). In particular, rates of early preterm birth (<34 weeks) were higher in the PTB group compared to none in the CT group (17% vs. 0%, P<0.001, Figure 5-3, Table 5-1).

153 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-1 Patient demographics of pregnancies with prior excisional cervical treatment and prior preterm birth <37 weeks who consented to transvaginal scans Prior CT Prior PTB <37 weeks TOTAL, High risk

N=113 N=86 N=199 Age (years) Mean ±SD 33.6 ±4.1 32.4 ±5.8 33.0 ±4.6 BMI Mean ±SD 24.0 ±3.8 25.6 ±5.6 24.8 ±4.2

Ethnicity, n/N % Caucasian 81/113 72%*** 26/86 30% 107/199 54% Asian 10/113 9% 20/86 23%** 30/199 15% Black 22/113 19% 40/86 47%*** 62/199 31%

Parity, n/N % Para 0 76/113 67% 0/86 0% 76/199 38% Para ≥ 1 37/113 33% 86/86 100% 123/199 62%

Smoker, n/N % 11/113 10% 12/86 14% 23/199 12%

Cerclage Intervention, n/N % 41/113 36% 44/86 51%* 85/199 43%

Early PTB, <34+0 w 0/113 0% 15/86 17% 15/199 8% Late PTB, 34+0 to <37+0 w 11/113 10% 17/86 20% 28/199 14% Gestation at birth, n/N % Total PTB <37+0 w 11/113 10% 32/86 37%*** 43/199 22%

Term ≥37+0 w 102/113 90% 54/86 63% 156/199 78% CT= Prior excisional cervical treatment, PTB = preterm birth, w= weeks

***P<0.001, **P<0.01; two tailed Fisher exact test

100 *** Gestation at birth 80 >37w

60 34-37w % 40 <34w

20 *** 0 PriorPTB Pre-pregnancyCT preterm birth cervical treatment Figure 5-3 Rates of early preterm birth (<34+0 weeks) and late preterm birth (34+1 to <37+0 weeks) were higher in the prior PTB group (20% and 17% respectively) than the CT group (0% and 10% respectively). (P value= Fisher’s exact, CT= previous excisional cervical treatment, PTB= previous spontaneous preterm birth <37weeks, w= weeks)

154 A comparison of previous preterm birth and cervical treatment in pregnancy

5.1.2 Ultrasound assessment of cervical length and volume

Cervical length (CL) and volume (CV) measurements were compared between CT (n=113) and prior PTB women (n=86) at each of the sampling time points from 12 to 34 weeks. These were also compared to control group of low-risk pregnancies (n=96), as previously described in Chapter 3. PTB and low-risk women demonstrated similar CL measurements throughout pregnancy, while the CT women had significantly shorter cervices (P<0.001, ANOVA, Figure 5-4A, B, Table 5-2). At 12 weeks, mean CL was 45mm in low-risk women compared to 35mm in CT women (P<0.001), decreasing to 32mm and 27mm respectively at 34 weeks (P<0.05 Figure 5-4B). The difference in length between PTB and CT women was most apparent at 12 weeks where mean CL was 5mm (95% CI 0.8-7.9) shorter among CT women than PTB women (35 v 40 mm, P=0.016). This difference was 4mm at 16 weeks (95% CI 0.8-7.9, CT 34 v PTB 38mm), 4mm at 22 weeks (95% CI 0.8-7.5, CT 32 v PTB 36mm, P=0.02), and 3mm at 28 weeks (95% CI 0.4-6.6, CT 33 v 30mm PTB, P=0.03). By 34 weeks cervical lengths among CT and PTB groups were comparable (CT 27mm v PTB 28mm; P=0.7, mean difference 0.7mm, 95% CI -2.7 to 4.1).

Similar trends were observed with measurements of cervical volume. Throughout pregnancy CT women exhibited lower CVs than both low-risk and PTB women (P<0.001; Figure 5-5A, Table 5-2), with the deficit most notable at 12 weeks (CV 26 ±13cm3 in the CT group versus 33 ±11cm3 in the PTB group, P=0.008). Differences persisted until 28 weeks at which point they were no longer significant (CT: 41 ±15cm3 v PTB: 41 ±16cm3, P=0.2, Figure 5-5B, Table 5-2)

155 A comparison of previous preterm birth and cervical treatment in pregnancy

A Cervical Length *** 50 *** ** 40 ** * * CL (mm) 30 Previous excisional CT Previous preterm birth Low risk 20 12w 16w 22w 28w 34w

Gestational age (weeks)

B ** ** * * ns ns ns ns 80 *** ** ** * ns * *

60

40 CL (mm)

20

0

CT CT CT CT CT PTB Low PTB Low PTB Low PTB Low PTB Low

12 weeks 16 weeks 22 weeks 28 weeks 34 weeks

Previous excisional CT Previous preterm birth Low risk controls

Figure 5-4 (A) Cervical length (CL) was lower among CT (n=113) than PTB women (n=86), as well as low-risk controls (n=96) throughout screening (P<0.001; ANOVA, Bonferroni post-test). All groups demonstrated a decline in CL with advancing gestation (P<0.001, ANOVA). (B) The most significant difference in CL was observed at 12 weeks (mean CL, CT 35 v PTB 40mm v low 45mm, P<0.001; ANOVA). By 34 weeks CT and PTB groups were comparable (mean CL 27mm v 28mm, P=0.7), however CL in CT women remained shorter than in low-risk women (32mm, P<0.05, t-test). (CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, Low = low-risk controls, w = weeks, CL = cervical length.)

156 A comparison of previous preterm birth and cervical treatment in pregnancy

A Cervical Volume *** 60 ns * 50 **

3) 40 * **

30 CV (cm

Previous excisional CT 20 Previous preterm birth Low risk 10 12w 16w 22w 28w 34w

Gestational age (weeks)

120 ns * *** ** ns B ns ns ns ns ns ** * * ns ns

80 ) 3

CV (cmCV 40

0

CT CT CT CT CT PTB Low PTB Low PTB Low PTB Low PTB Low

12 weeks 16 weeks 22 weeks 28 weeks 34 weeks

Previous excisional CT Previous preterm birth Low risk controls

Figure 5-5 (A) Cervical volume (CV) increased with advancing gestation age among all cohorts: CT (n=113) PTB (n=86), and low-risk controls (n=96) (P<0.001, ANOVA Bonferroni post-test). (B) Mean CV was lower in CT compared to PTB groups at 12 weeks (26cm3 v 33cm3, P=0.008), 16 weeks (31cm3 v 35cm3, P=0.02), and 22 weeks (36cm3 v 43cm3, P=0.01). By 34 weeks, CV among CT, PTB, and low-risk controls were comparable (CT: 47±21cm3, PTB: 52±16cm3 and low: 57±21cm3; P=0.14; ANOVA). (CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, Low = low-risk controls, w = weeks, CV = cervical volume)

157 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-2 Mean cervical length, volume, vascularisation and flow indices of PTB, CT and low-risk control groups at screening 12 to 34 weeks

Scan time points Cervical Risk 12w 16w 22w 28w 34w parameter factor Mean SD Mean SD Mean SD Mean SD Mean SD CL (mm) CT 35* 5.2 34** 5.7 32** 6.7 30* 6.3 27 6.6 # PTB 40 10.6 38 9.0 36 9.9 33 8.0 28 8.9 # Low 45 8.9 40 9.1 38 7.0 34 7.2 32 6.9 # CV (cm3) CT 26** 13.9 31* 10.8 36* 14.1 41.0 15.0 47.3 20.6 # # PTB 33 10.8 35 12.4 43 17.3 47 16.6 52 16.2 # Low 28 6.3 33 7.6 45 16.5 51 15.9 58 21.3 # VI CT 2.0** 1.3 3.0 2.3 3.6* 2.9 3.8 3.2 4.7 3.2 # PTB 3 2.3 4 2.8 4 2.0 5 3.7 5 3.4 # Low 2 1.4 3 1.8 3 1.6 4 2.2 4 2.5 FI CT 29.4 4.1 31.3 5.4 32.0 6.1 32.2 6.0 33.8 6.1 # # PTB 30 4.6 31 4.8 32 4.6 34 5.4 34 5.2 # Low 28 3.8 31 5.1 31 4.5 32 5.8 33 4.3 CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, Low = low-risk control, W = weeks, CL = cervical length, CV= cervical volume, VI = vascularisation index, FI = flow index *P<0.05, **P<0.01; unpaired t-test for CT v PTB, #P<0.05; 1-way ANOVA for longitudinal 12w to 34w.

5.1.3 Ultrasound assessment of cervical vascularity A trend towards higher density of cervical blood vessels (VI) was observed until 28 weeks gestation in the prior PTB group compared to the CT group. (VI at 12 weeks: PTB 3.2 ±1.3 vs. CT 2 ±1.3, P=0.004, and at 28 weeks: PTB 5.1 ±3.7 vs. CT 3.8 ±3.2, P=0.04; t-test, Figure 5-6A, C, Table 5-2). Mean VI values among PTB women were also higher than the low-risk group (Table 5-2), which in turn were comparable to values in the CT group throughout pregnancy (Table 5-2, Figure 5-6C). This may be explained by differences in parity, as there were more multips in the PTB group than the CT and low-risk groups, although differences in VI based parity alone did not achieve significance (as observed in Chapter 3).

In contrast to VI, there were no clear differences in cervical blood flow (Flow Index, FI) between previous CT, previous PTB and low-risk control cohorts (Figure 5-6B, D, Table 5-2).

158 A comparison of previous preterm birth and cervical treatment in pregnancy

ABVascularisation Index Flow index *** ** 6 ns ns 36 ns ns

5 ns ns ns ns 4 32 *** ns 3

2 28 Flow Index, FI Index, Flow Previous excisional CT Previous excisional CT 1 Vascularisation Index, VI Index, Vascularisation Previous preterm birth Previous preterm birth Low risk Low risk 0 24 12w 16w 22w 28w 34w 12w 16w 22w 28w 34w

Gestational age (weeks) Gestational age (weeks)

C Vascularisation Index ns 20 ns ns ns ns ns ns ns ns ** *** ns * * ns

15

10

5 Vascularisation Index, VI Index, Vascularisation

0

CT CT CT CT CT PTB Low PTB Low PTB Low PTB Low PTB Low

12 weeks 16 weeks 22 weeks 28 weeks 34 weeks

D Flow index 60 ns ns ns ns ns ns ns ns ns ns ns ns ns ns ns

50

40

30 Flow Index, FI Index, Flow

20

10

CT CT CT CT CT PTB Low PTB Low PTB Low PTB Low PTB Low

12 weeks 16 weeks 22 weeks 28 weeks 34 weeks

Previous excisional CT Previous preterm birth Low risk controls Figure 5-6 (A) VI and (B) FI increased with advancing gestation in the prior PTB group (n=86), the CT group (n=113), as well as low-risk controls (n=96) (P<0.001, ANOVA, Bonferroni post-test). (C) VI was higher among PTB than CT women until 28 weeks (t-test), and was similar among CT and low-risk controls throughout gestation. (D) FI did not differ among groups at any gestational age (t-test). (VI=vascularisation Index, FI=flow index, CT=previous excisional cervical treatment, PTB=previous preterm birth <37weeks, Low-risk=healthy pregnancy, w=weeks).

159 A comparison of previous preterm birth and cervical treatment in pregnancy

5.1.4 The cervix and gestation at birth Despite significantly shorter CLs among the CT group compared to the PTB women, rates of preterm birth <37 weeks were substantially higher among PTB women (preterm birth <37weeks: CT 10%, 11/113, vs. PTB 37%, 32/86; P<0.001, Figure 5-3).

Cervical measurements (CL, CV, VI and FI) at sampling points before the clinically relevant screening threshold of <24 weeks, were therefore assessed for associations with subsequent gestation at birth. Mean CL was shorter in the CT group compared to the prior PTB group among term births (mean CL: CT 34mm v PTB 38 mm, P<0.001) and preterm births (32 v 37mm, P=0.08, Figure 5-7A, Table 5-3). Mean CL did not differ significantly when comparing term and preterm births among CT women (34mm v 32mm respectively, P=0.17) or among prior PTB women (38mm v 37mm, respectively, P=0.6).

Similarly, mean cervical volumes were lower in CT than prior PTB women, irrespective of gestation at birth (term: CT 32cm3 v PTB 38cm3; P<0.001, and preterm: CT 26cm3 v PTB 35cm3; P=0.045, Figure 5-7B, Table 5-3). FI did not differ substantially among groups with respect to gestation at birth. However VI was lower in women delivering at term in the CT group (mean VI: CT 2.9 v PTB 3.8 P=0.005, Figure 5-7C). This was similar to the observation made in Chapter 3 when comparing term and preterm births among high and low-risk women, indicating a potential lack of power to reach significance.

160 A comparison of previous preterm birth and cervical treatment in pregnancy

Combined screening <24weeks Previous PTB Previous excisional CT

A Cervical Length, CL B Cervical Volume, CV

45 45

40 *** 40 P=0.08 *** * )

3 35 35 30 CL (mm) CV (cmCV

30 25

25 20 <37w >37w <37w >37w Gestation at birth (weeks) Gestation at birth (weeks)

C Vascularisation Index, VI D Flow index, FI

4.5 34

ns ** ns 4.0 32 ns

3.5 30 Flow Index, FI Index, Flow 3.0 28 Vascularisation Index, VI

2.5 26 <37w >37w <37w >37w Gestation at birth (weeks) Gestation at birth (weeks)

Figure 5-7 CL, CV, VI and FI (mean ±SD) at screening <24 weeks in PTB (n=86) and CT (n=113) women, as a function of term (>37weeks) versus preterm birth (<37weeks). (A) Mean CL was lower in CT compared to prior PTB women among term births (CT: n=102/113, mean CL 34mm v PTB: n=54/86, mean CL 38mm, P=0.08) and preterm births (CT: n=11/113, mean CL 32mm v PTB: n=32/86, mean CL 37mm P<0.001). (B) Similarly CV was lower among CT compared to previous PTB women (Term birth: mean CV 32 v 38cm3, P<0.001, and preterm birth 26 v 35cm3, P=0.04). (C) VI was also lower in the CT group, most notably among those going on to term birth (mean VI 2.9 v 3.8 PTB, P=0.005), although FI did not differ among groups. (CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, *P<0.05, **P<0.01, ***P<0.001; unpaired t-test two tailed)

161 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-3 A comparison of cervical parameters at screening <24 weeks for outcomes of term and preterm birth <37weeks among previous preterm births (PTB) and previous excisional cervical treatment (CT).

Cervical measurement *P value Cervical Delivery outcome CT v parameter Previous CT Previous PTB PTB Mean ±SD Min Max Mean ±SD Min Max PTB <37w 32 9 5 44 37 12 5 68 0.084 CL (mm) Term birth >37w 34 6 21 46 38*** 8 20 59 ***<0.001 PTB <37w 26 11 12 41 35 15 10 83 *0.045 CV (cm3) Term birth >37w 32 16 10 90 38*** 13 16 81 ***<0.001 PTB <37w 3.3 2.2 0.2 8 3.6 2.5 0.1 15 0.573 VI Term birth >37w 2.9 2.4 0.0 12 3.8** 2.4 0.3 16 **0.005 PTB <37w 30.5 4.5 13 42 30.7 4.5 20 40 0.857 FI Term birth >37w 31.0 5.6 19 46 31.1 4.8 20 42 0.903 CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, w= weeks, SD= standard deviation CL = cervical length, CV= cervical volume, VI = vascularisation index, FI = flow index, *P<0.05, **P<0.01; unpaired t-test two-tailed

162 A comparison of previous preterm birth and cervical treatment in pregnancy

5.1.4a Correlation between a short cervix ≤25mm and gestation at birth Women were grouped according to CL measurements into short (≤25mm) and non-short (>25mm) cervix groups 83. As was clinically indicated, all women found to have a short cervix before 24 weeks (39%, 78/199) had an ultrasound-indicated cervical cerclage. Despite cerclage insertion, detection of a short cervix in the previous PTB group (43%, 37/86) was associated with earlier gestation at birth (mean 35+3 weeks) than the CT group (36%, 41/113, mean gestation at birth 38+6 weeks, P<0.001; Figure 5-8A, Table 5-4). Indicating a short cervix confers higher preterm risk in the PTB than CT group. A normal cervix (CL>25mm, 61%, 121/199) was also associated with comparably later gestation at birth in the CT (mean 39+4 weeks) than the prior PTB group (mean 36+5 weeks, P<0.001; Figure 5-8A, C; Table 5-4).

Linear regression analysis demonstrated a positive correlation between early detection of a short cervix and earlier gestation at birth in the prior PTB (r=0.4, P=0.03), but not the CT group (r=0.2, P=0.3; Figure 5-8B). Given a longer cervix (>25mm), gestation at screening did not correlate with gestation at birth (CT: r=0.03, P=0.7 and PTB: r=0.08, P=0.4; Figure 5-8C).

Table 5-4 A comparison of gestation at birth in women with a short (CL<25mm) and non-short cervix (CL >25mm) detected at screening < 24 weeks gestation.

Threshold of Cervical length *P value, CL ≤ 25 vs. CL ≤25mm CL >25mm >25mm CT, n=113 n= 41 72 Mean ±SD 38+6 weeks ±1.8 39+4 weeks ±1.7 *0.02 Gestation at birth (weeks) Range (34.3 - 42.0) (34.7 - 42) PTB, n=86 n= 37 49 Mean ±SD 35+3 weeks ±4.1 36+5 weeks ±3.5 0.08 Gestation at birth (weeks) Range (24.5 - 41) (22.4 - 41.1) *P value, CT vs. PTB ***<0.001 ***<0.001 CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, w = weeks, SD = standard deviation *P value= unpaired t-test two-tailed previous CT v previous PTB

163 A comparison of previous preterm birth and cervical treatment in pregnancy

ABCL screening <24 weeks CL 25mm PTB, n=86 v CT, n=113 42 40 42 P=0.02 38 40 *** *** 36 34 38 32 30 36 28 CT: r = 0.2

Gestation at birth (weeks) 26 PTB: r = 0.4* 34 24 P=0.08 12 16 20 24 32 Gestation at screening (weeks) Gestation at birth (weeks) birth at Gestation 30 C CL > 25mm <25mm >25mm n=78 n=121 42 Cervical length 40 38 36 34 Previous excisional CT 32 Previous preterm birth 30 28 CT: r = 0.03

Gestation at birth (weeks) 26 PTB: r = 0.08 24 12 16 20 24 Gestation at screening (weeks)

Figure 5-8 A comparison of gestation at birth in CT (n=113), and PTB women (n=86), when grouped according to CL measurements ≤25mm and >25mm at screening before 24 weeks. (A) A short CL ≤25mm in PTB women (n=37/86) was associated with earlier gestation at birth (mean 35+3 weeks) than CT women (n=41/113, mean 38+6 weeks, P<0.001). This was also true of women with a normal CL (>25mm): mean birth gestation in the prior PTB (n=49/86, 36+5 weeks) versus CT group (n=72/113, 39+3 weeks, P<0.001). (B) Linear regression demonstrated a positive correlation between gestation at detection of a short cervix and earlier gestation at birth in the prior PTB (r=0.4, P=0.03), but not the CT group (r=0.2, P=0.3). (C) Given a longer CL >25mm, gestation at screening does not correlate with gestation birth in either CT (r=0.03, P=0.7) or prior PTB groups (r=0.08, P=0.4). (***P<0.001; t-test, r= Pearson correlation co-efficient, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, CL = cervical length, mm)

164 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2 The vaginal microbiota in cervical treatment and prior preterm birth groups

5.2.1 Participant demographics

A total of 66 women with a pre-pregnancy excisional CIN treatment (CT) and 67 women a prior preterm birth <37weeks (PTB) consented to serial high vaginal swabs matched to transvaginal scans. Patient characteristics of these groups are provided in Table 5-5.

Table 5-5 Patient demographics of high risk pregnancies with a prior excisional cervical treatment (CT) and prior preterm birth (<37weeks) consenting to vaginal microbial swabs Prior CT Prior PTB <37 weeks TOTAL, High risk

N=66 N=67 N=133 Age (years) Mean ±SD 32.5 ±4.1 31.7 ±5.8 33.0 ±4.6 BMI Mean ±SD 24.7 ±3.9 25.4 ±5.5 24.8 ±4.2

Ethnicity, n/N % Caucasian 53/66 80%*** 29/67 43% 82/133 62% Asian 6/66 9% 13/67 19% 19/133 14% Black 7/66 11% 25/67 37%*** 32/133 24%

Parity, n/N % Para 0 50/66 76% 0/67 0% 50/133 38% Para ≥ 1 16/66 24% 67/67 100%*** 83/133 62%

Smoker, n/N % 3/66 5% 2/67 3% 5/133 4%

Cerclage Intervention, n/N % 16/66 24% 22/67 33% 38/133 29%

Early PTB, <34+0 w 0/66 0% 15/67 22% 15/133 11% +0 +0 Gestation at Late PTB, 34 to <37 w 7/66 11% 10/67 15% 17/133 13% birth, n/N % Total PTB <37+0 w 7/66 11% 25/67 37%*** 32/133 24% Term ≥37+0 w 59/66 89% 42/67 63% 101/133 76%

CT=Prior excisional cervical treatment, PTB= preterm birth, w= weeks; ***P<0.001 Fisher exact CT v PTB birth groups

165 A comparison of previous preterm birth and cervical treatment in pregnancy

In the total population, cervical shortening ≤25mm detected before 24 weeks occurred in 23% (31/133) of women, all of whom went on to receive an ultrasound indicated cervical cerclage. In addition 7 women (5%) received a history indicated cervical cerclage following a poor obstetric history. A total of 505 vaginal swabs were taken with matched cervical length, volume and vascularity data at longitudinal time-points (12 to 34 weeks). As in Chapter 4, all samples taken after cerclage insertion were excluded from the analyses, as the impact of the cerclage may skew microbial assessments (n=38 women and 107 samples). A total of 395 vaginal samples taken from 133 high-risk women, with matched ultrasound data, were included in this study (Table 5-6). The impact of cerclage insertion on the vaginal microbiome will be addressed specifically in Chapter 6.

Table 5-6 Gestational age at vaginal sampling in high-risk pregnancies with a prior excisional cervical treatment (CT) and prior preterm birth (<37weeks)

Prior CT, Prior PTB <37 weeks, Total Gestation at n=66 women n=67 women included* sampling n, Mean n, Mean n, ±SD Min Max ±SD Min Max samples (weeks) samples (weeks) samples 12w 41 13+0 ±0.9 11+3 14+5 46 13+3 ±1.0 11+0 14+6 87 16w 52 17+0 ±0.8 16+0 18+3 56 16+6 ±0.8 15+1 18+3 108 22w 53 21+5 ±1.2 20+0 23+5 40 21+5 ±1.0 20+0 23+3 93 28w 27 27+6 ±0.8 25+6 29+5 28 27+5 ±0.7 26+0 29+2 55 34w 28 33+6 ±0.9 31+4 35+5 24 33+4 ±1.0 31+6 36+0 52 Total 201 194 395 samples, n *samples excluded post cerclage insertion include n=38 women, 107 samples; CT= Prior excisional cervical treatment, PTB = preterm birth, w= weeks

166 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2.2 Assessment of bacterial richness and alpha diversity

The total number of species observed (Sobs) were compared between PTB and CT groups at each sampling time-point (12, 16, 22, 28 and 34 weeks). At 12 weeks there were no differences in the number of species observed between PTB (n=67, mean Sobs 4.5) and CT women (n=66, mean Sobs 4.0, P=0.5, t-test; Figure 5-9A). From 16 weeks onwards, the prior PTB group exhibited increased numbers of species, which continued to 34 weeks (mean Sobs 8.1, P=0.16; ANOVA). The CT group in contrast did not demonstrate any increase in species observed. At 16 weeks there were significantly more species observed in the PTB compared to CT groups (mean 3.9 v 5.6, P=0.03). This difference persisted to 34 weeks (mean 4.2 v 8.1, P=0.02; t-test, Figure 5-9A, Table 5-7). Correspondingly, alpha diversity increased from 12 to 34 weeks gestation in the PTB, but not the CT group (Shannon Index at 12 weeks: mean CT 0.39 v PTB 0.29, P=0.1, and at 34 weeks 0.30 v 0.54 respectively, P=0.06; Figure 5-9B, Table 5-7) demonstrating low microbial stability with high diversity in the prior PTB group, and high stability with low diversity in the CT group. Overall, both richness and diversity in the CT group were comparable to the low-risk women previously described in Chapter 4 (Table 5-7).

Previous preterm birth Previous excisional CT

Species observed Alpha diversity index

12 0.8 ns 10 * ns 0.6 ns ns 8 ns ns * * 6 obs ns 0.4 S 4

Shannon Index Shannon 0.2 2

0 0.0

12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Gestation at sampling (weeks) Gestation at sampling (weeks) Figure 5-9 (A) The observed number of species increased with advancing gestation in the prior PTB group (n=67, mean Sobs 4.5 at 12 weeks to 8.1 at 34 weeks, P=0.16; ANOVA), but not in CT women (n=66). From 16 weeks there were significantly more species observed in PTB than CT group (mean 3.9 v 5.6, P=0.03) persisting to 34 weeks (mean 4.2 v 8.1, P=0.02; t-test). (B) Correspondingly, the Shannon index of alpha diversity was higher in the PTB than CT group throughout gestation (at 12 weeks, mean 0.39 v 0.29, P=0.1, and at 34 weeks 0.30 v 0.54, P=0.06). (ns=nonsignificant, *P<0.05; t-test, two tailed. Sobs =species observed, w=weeks)

167 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-7 Number of bacterial species observed and corresponding Shannon index of alpha diversity in CT (n=66), PTB (n=67) and low-risk groups (n=30)

Species observed Shannon Index of Alpha Diversity Gestation at Samples sampling taken, n Mean ±SD Range Mean ±SD Range

12 weeks 41 4.0 3.1 (1 - 17) 0.29 0.38 (0 - 1.46) 16 weeks 52 3.9 3.4 (1 - 20) 0.27 0.41 (0 - 1.77) Prior CT 22 weeks 53 4.0 2.7 (1 - 14) 0.30 0.36 (0 - 1.15) 28 weeks 27 3.9 2.2 (1 - 10) 0.30 0.32 (0 - 1.11) 34 weeks 28 4.2 3.4 (1 - 14) 0.30 0.32 (0 - 1.12) 12 weeks 46 4.5 3.9 (1 - 21) 0.39 0.39 (0 - 1.58) 16 weeks 56 5.6* 4.6 (1 - 22) 0.41 0.47 (0 - 1.69) Prior PTB 22 weeks 40 5.7* 5.2 (1 - 24) 0.44 0.47 (0 - 1.77) 28 weeks 28 6.0 6.2 (1 - 26) 0.50 0.57 (0 - 2.14) 34 weeks 24 8.1* 9.0 (1 - 41) 0.54 0.61 (0 - 1.96) 12 weeks 16 4.9 3.6 (1 - 16) 0.31 0.49 (0 - 1.46) 16 weeks 30 3.9 3.0 (1 - 15) 0.22 0.36 (0 - 1.58) †Low-risk 22 weeks controls 22 4.1 3.7 (1 - 12) 0.27 0.49 (0 - 1.13) 28 weeks 22 3.9 2.9 (1 - 12) 0.35 0.35 (0 - 1.04) 34 weeks 20 4.1 5.5 (1 - 18) 0.36 0.43 (0 - 1.2) CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, †Low-risk controls as previously described in Chapter 4. SD = standard deviation *P <0.05; t-test two-tailed CT v PTB at screening gestation

168 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2.3 Distribution of vaginal community state types (CSTs) in cerivcal treatment and prior preterm birth groups

16S rRNA sequence reads of bacterial species from 395 longitudinal vaginal samples (133 women; CT, n=66 and PTB, n=67 PTB) were classifed into community state types (CSTs) 155 based on ward hierarchical clustering (Figure 5-10). CST I (L. crispatus) was the most prevalent CST observed in the total cohort (41%, 161/395). This was followed by CST III (L. iners, 29%, 115/395), CST II (L. gasseri, 13%, 52/395) and CST V (L. jensenii, 9%, 35/395). The remaining 8% (32/395) were classified as CST IV and were characterised by Lactobacillus spp. difficiency and high microbial diversity.

Women in the prior PTB group exhibited a substantially different distribution of vaginal microbial profiles compared to the CT group (Figure 5-11, Table 5-8). A greater proportion of prior PTB women were dysbiotic (CST IV: PTB 13% v CT 4%, P=0.002) or L. iners dominant (CST III: PTB 41% v CT 17%, P<0.001) compared to CT women. L. crispatus dominance (CST I) in contrast was more prevalent among CT women (CT 61% v PTB 20%, P<0.001; Fisher's exact; Figure 5-11, Table 5-8). In the CT group, no correlation was observed between microbiral profiles and time from excisional treatment.

169 A comparison of previous preterm birth and cervical treatment in pregnancy

170 A comparison of previous preterm birth and cervical treatment in pregnancy

Figure 5-10 Heatmaps demonstrating ward hierarchical clustering of bacterial species data from 395 vaginal swabs taken from CT (n=66, 201 samples) and prior PTB groups (n=67, 194 samples) according to sampling gestation: 12 weeks (n=87 samples), 16 weeks (n=108 samples), 22 weeks (n=93 samples), 28 weeks (n=55 samples) and 34 weeks (n=52 samples). L. crispatus was highly abundant among CT women, dominant in 61% of samples (123/201; P<0.001), while L. iners was most prevalent among PTB women, dominant in 41% of samples (80/194; P<0.001, fisher exact) (w= weeks, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

171 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-8 Classification of vaginal samples into community state types, comparing CT and prior PTB groups

Community state type & Species

Cohort CST I, CST II, CST III, CST IV, CST V, Total L. crispatus L. gasseri L. iners Diverse species L. jensenii n/N % n/N % n/N % n/N % n/N % N=

PTB 38/194 20% 34/194 17% 80/194 41%*** 25/194 13%** 17/194 9% 194

CT 123/201 61%*** 18/201 9% 35/201 17% 7/201 4% 18/201 9% 201 P value* <0.001 0.02 <0.001 0.002 1.0

TOTAL 161/395 41% 52/395 13% 115/395 29% 32/395 8% 35/395 9% 395 CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks; CST = community state type P value*= fisher exact

Figure 5-11 Distribution of community state types among PTB (n=67, 194 samples) and CT women (n=66, 201 samples). CST III (L. iners) was more prevalent in the prior PTB group (PTB 41% v CT 17%, P<0.001), as was vaginal dysbiosis (CST IV: PTB 13% v CT 4%, P=0.002). CST I (L. crispatus) was more abundant among CT women (61% v 20% PTB, P<0.001 Fisher’s exact). (CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

172 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2.4 Longitudinal analysis of vaginal CSTs in cerivcal treatment and prior preterm birth groups Gestational age at sampling influenced the composition of the vaginal microbiome in both CT and PTB women. Both groups demonstrated a shift towards increasing L. crispatus and L. gasseri dominance with declining L. iners abundance at 22 weeks. Following this, increased diversity was associated with advancing gestation in that distributions of CSTs at 12 weeks were similar to those observed at 34 weeks. Among PTB women, the number assigned to CST III (L. iners) decreased from 52% at 12 weeks to 25% at 22 weeks before increasing again to 50% at 34 weeks (P=0.01). A corresponding shift in L. crispatus dominance from 17% at 12 weeks to 23% at 22 weeks was observed before returing to 17% of women 34 weeks (P=0.6), while L. gasseri dominance shifted from 11% to 28% and finally to 17% at the respective time- points (P=0.9).

Despite increased CST I at 22 weeks in the PTB group, L. crispatus dominance remained higher in the CT than the prior PTB group throughout sampling. At 12 weeks, 56% of CT women were assigned to CST I compared to 17% of PTB women (P<0.001), increasing to 58% and 23% respectively at 22 weeks (P<0.001) and 61% and 17% respectively at 34 weeks (P<0.001). Comparatively there was a higher proportion of CST III (L. iners) in the PTB group at all pregnancy time points; 12 weeks (PTB 52% v CT 22%; P=0.004), 16 weeks (PTB 41% v CT 13%; P=0.002), 22 weeks (PTB 25% v CT 17%; P=0.4), 28 weeks (PTB 39% v CT 19%; P=0.1), and 34 weeks (PTB 50% v CT 18%; P=0.02; Fisher’s exact; Figure 5-12, Table 5-9).

173 A comparison of previous preterm birth and cervical treatment in pregnancy

Previous PTB Previous CT

# P=0.03; ANOVA P=0.06; ANOVA 100

80 * ** ** 60 *** ** ** *** *** 40 Percentage

20

0

12w 16w 22w 28w 34w 12w 16w 22w 28w 34w Gestation at screening (weeks) Gestation at screening (weeks)

CST, species: I L. crispatus II L. gasseri III L. iners IV Diverse species V L. jensenii

Figure 5-12 Gestational age has a greater impact upon vaginal microbiota composition in women with a prior PTB (n=67; P=0.03; ANOVA) than those with previous CT (n=66, P=0.06; ANOVA). The PTB group had a greater prevalence of L. iners (CST III) than CT women, from 12 weeks (PTB 52% v CT 22%, P=0.004) to 34 weeks (50% v 18%; P=0.02), while L. crispatus (CST I) was comparatively more prevalent among the CT group (P<0.01). (*P<0.05, **P<0.01, ***P<0.001 for CT v PTB at comparative time-point; fisher exact. #P<0.05; ANOVA two way. w= weeks, CST = community state types, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

174 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-9 Classification of vaginal samples into community state types, comparing CT and PTB groups according gestation at sampling.

Community state type & Species Gestation at sampling CST I, CST II, CST III, CST IV, CST V, Total & L. crispatus L. gasseri L. iners Diverse species L. jensenii Cohort n/N % n/N % n/N % n/N % n/N % N=

PTB 8/46 17% 5/46 11% 24/46 52%** 6/46 13% 3/46 7% 46

CT 23/41 56%*** 3/41 7% 9/41 22% 3/41 7% 3/41 7% 41

12 weeks 12 weeks TOTAL 31/87 36% 8/87 9% 33/87 38% 9/87 10% 6/87 7% 87 P value* <0.001 0.7 0.004 0.6 1.0

PTB 11/56 20% 11/56 20% 23/56 41%** 6/56 11% 5/56 9% 56

CT 35/52 67%*** 4/52 8% 7/52 13% 1/52 2% 5/52 10% 52

16 weeks 16 weeks TOTAL 46/108 43% 15/108 14% 30/108 28% 7/108 6% 10/108 9% 108 P value* <0.001 0.09 0.002 0.1 1.0

PTB 9/40 23% 11/40 28% 10/40 25% 6/40 15% 4/40 10% 40

CT 31/53 58%*** 6/53 11% 9/53 17% 2/53 4% 5/53 9% 53 weeks

22 TOTAL 40/93 43% 17/93 18% 19/93 20% 8/93 9% 9/93 10% 93 P value* <0.001 0.06 0.4 0.07 1.0

PTB 6/28 21% 3/28 11% 11/28 39% 4/28 14% 4/28 14% 28

CT 17/27 63%** 2/27 7% 5/27 19% 1/27 4% 2/27 7% 27

28 weeks 28 weeks TOTAL 23/55 42% 5/55 9% 16/55 29% 5/55 9% 6/55 11% 55 P value* 0.003 1.0 0.1 0.4 0.7

PTB 4/24 17% 4/24 17% 12/24 50%* 3/24 13% 1/24 4% 24

CT 17/28 61%** 3/28 11% 5/28 18% 0/28 0% 3/28 11% 28

34 weeks 34 weeks TOTAL 21/52 40% 7/52 13% 17/52 33% 3/52 6% 4/52 8% 52 P value* 0.002 0.7 0.02 0.09 0.6

PTB 38/194 20% 34/194 18% 80/194 41%*** 25/194 13%** 17/194 9% 194

CT 123/201 61%*** 18/201 9% 35/201 17% 7/201 3% 18/201 9% 201 All P value* <0.001 0.02 <0.001 0.002 1.0 samples samples

TOTAL 161/395 41% 52/395 13% 115/395 29% 32/395 8% 35/395 9% 395 POPULATION CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks; CST = community state type P value*= fisher exact: CT v PTB at each sampling time point

175 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2.5 Vaginal microbiota and gestation at birth L. iners dominance (CST III) was associated with early gestation at birth in both PTB (mean 34.2 weeks) and CT groups (38.2 weeks) (Figure 5-13, Table 5-10). Among the PTB group, women with CST III delivered significantly earlier than those with CST I (L. crispatus, mean 37.8 weeks, P=0.003), CST II (L. gasseri, mean 37.3 weeks, P=0.01) and CST IV (diverse species, mean 37.9 weeks, P=0.01). In the CT group, CST III women also delivered earlier than CST I (L. crispatus, mean 39.3 weeks, P=0.07) and CST II (L. gasseri, mean 39.7 weeks, P=0.03). The smallest difference in gestation at birth among CT and PTB groups was observed with L. crispatus dominance (mean difference 1.1 week, 95% CI 0.2-2), where both groups delivered at term on average (CT, mean 39.3 weeks v PTB, 37.8 weeks, P=0.02; Figure 5-13, Table 5-10).

Table 5-10 A comparison of gestation at birth between previous PTB and CT women, as a function of corresponding community state type. Gestation at delivery (weeks) *P Previous PTB Previous CT value CST Species CT v Mean ±SD Min Max Mean ±SD Min Max PTB I L. crispatus 37.8* ±1.8 33.0 39.0 39.3 ±1.7 34.7 42.0 0.02 II L. gasseri 37.3* ±3.7 27.9 41.1 39.7 ±1.2 37.7 41.1 0.03 III L. iners 34.2*** ±5.1 22.4 41.0 38.2 ±1.7 35.7 41.0 0.004 Diverse IV 37.9* ±1.3 36.3 40.0 38.8 ±0.4 38.0 39.0 0.06 species V L. jensenii 36.4* ±3.4 27.4 38.0 39.1 ±1.8 36.0 41.4 0.02 All 36.3 ±4.0 22.4 41.1 39.2 ±1.6 34.7 42.0

***P<0.001, *P<0.05; t test CT v PTB for CST. CT= previous excisional cervical treatment, PTB= previous preterm birth <37w, w = weeks, CST= community state type

176 A comparison of previous preterm birth and cervical treatment in pregnancy

# 42 * * * 40 ** ns

38 Term, 37weeks 36

34

32

30

Gestation at birth (weeks) birth at Gestation ## ##* Previous excisional CT 28 ** * Previous PTB 26

CST III L. iners CST I, L. crispatusCST II L. gasseri CST V L. jensenii CST IV Diverse species

Figure 5-13 Mean (±SD) gestation at birth as a function of dominant CST prior to 24 weeks gestation. In the PTB group, CST III was associated with significantly earlier gestation at birth than CST I (##P=0.003), CST II (#P=0.01) and CST IV (#P=0.01; t-test). CST III was also associated with the greatest difference in birth gestation between CT and PTB groups (mean 38.2 v 34.2 weeks; P<0.001). The smallest difference in birth gestation occurred with CST I (mean CT 39.3 v PTB 37.8 weeks, P=0.02) and CST IV (38.8 v 37.9 weeks; P=0.06). (***P<0.001,**P<0.01, *P<0.01; t-test for CT v PTB gestation at birth, CT= excisional cervical treatment, PTB= previous preterm birth <37 weeks, CST = community state type).

177 A comparison of previous preterm birth and cervical treatment in pregnancy

Longitudinal shifts in CST distributions associated with term and preterm birth labour were observed in both prior PTB and CT women (Figure 5-14 and Table 5-11).

In the PTB group, women delivering at term demonstrated declining L. iners abundance towards mid-gestation. L. iners dominance was observed in 50%, 38%, 26%, 41% and 58% of women at 12, 16, 22, 28 and 34 weeks respectively (P<0.01; 2way ANOVA, Figure 5-14A). This trend was not observed in women delivering early preterm (<34 weeks), who had higher proportions of L. iners at respective sampling gestations (73% at 12 weeks, 64% at 16 weeks, 43% at 22 weeks and 50% at 28 weeks, Figure 5-14E). Late preterm birth exhibited mixed microbial profiles with longitudinal sampling.

The CT group consisted of higher L. crispatus abundance than PTB groups, particularly among those delivering at term (dominant in 60%, 68%, 57%, 64% and 58% of women at 12, 16, 22, 28 and 34 weeks respectively, Figure 5-14B). There were no early preterm births among the CT group, however those delivering late preterm (34 to 37 weeks) demonstrated a relative reduction in CST I (dominant in 33%, 60%, 67%, 50% and 50% at 12, 16, 22, 28 and 34 weeks respectively) with an increase in CST III (50%, 20%, 33%, 50% and 50% at respective time- points, Figure 5-14D, Table 5-11). Assigned CSTs in samples from individual patients and their corresponding birth gestations are demonstrated in the Figure 5-15.

178 A comparison of previous preterm birth and cervical treatment in pregnancy

Previous PTB, n=67 Previous CT, n=66

A Te rm birth >37 we e ks (n=42/67, 63%) B Te rm birth >37 we e ks (n=59/66, 89%)

# (P=0.04) ns (P=0.06) 100 100

80 80

60 60 *** *** 40 (P<0.001) 40 (P<0.001) % of women of % women of % 20 20

0 0

12w 16w 22w 28w 34w 12w 16w 22w 28w 34w

C Late pre te rm 34 +0 to 36+6 weeks (n=10/67, 15%) D Late pre te rm 34 +0 to 36+6 weeks (n=7/66, 11%) ns (P=0.5) ns (P=0.8) 100 100

80 80

60 60 ns *** 40 (P=0.6) 40 (P<0.001) % of women of % women of % 20 20

0 0

12w 16w 22w 28w 34w 12w 16w 22w 28w 34w

Gestation at sampling (weeks)

E Early preterm <34 weeks (n=15/67, 22%)

ns (P=0.4) Community state type (CST): 100 CST I L.CrisL. patus crispatus 80 CSTCSTII II L.Gas L. sgasseri eri

60 CST IIIIII L.Iners L. iners *** 40 (P<0.001) CST IV IV Div Diverse ers e s pec species ies % of women of % 20 CST VV L.JensL. jensenii enii

0 P value= 2way ANOVA 12w 16w 22w 28w # Gesational age factor Gestation at sampling (weeks) * CST factor Figure 5-14 Microbial profiles differed according to term (>37 weeks), late preterm (34-37weeks) and early preterm birth (<34weeks) in PTB (n=67) and CT (n=66) cohorts. Greater proportions of L. crispatus (CST I) were present in women delivering at term compared to early preterm birth, in both the PTB (A, E) and the CT groups (B, D). (P values = 2 way ANOVA: # impact of gestation at sample, *impact of community state type. CT= excisional cervical treatment, PTB= previous preterm birth <37 weeks, CST = community state type)

179 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-11 Vaginal microbial profiles as a function of gestation at sampling and subsequent gestation at birth for previous PTB and previous CT

Community state type & Species Cohort & Sampling CST I, CST II, CST III, CST IV, CST V, birth TOTAL, time point L. crispatus L. gasseri L. iners Diverse species L. jensenii gestation N= n/N % n/N % n/N % n/N % n/N % 12w 5/28 18% 3/28 11% 14/28 50% 5/28 18% 1/28 4% 28

16w 8/37 22% 9/37 24% 14/37 38% 5/37 14% 1/37 3% 37

22w 7/27 26% 8/27 30% 7/27 26% 5/27 19% 0/27 0% 27

Term >37w >37w Term 28w 5/17 29% 2/17 12% 7/17 41% 3/17 18% 0/17 0% 17

34w 3/19 16% 3/19 16% 11/19 58% 2/19 11% 0/19 0% 19

12w 2/7 29% 1/7 14% 2/7 29% 1/7 14% 1/7 14% 7 w w +6

-36 16w 2/8 25% 1/8 13% 2/8 25% 1/8 13% 2/8 25% 8 +0 22w 1/6 17% 1/6 17% 1/6 17% 1/6 17% 2/6 33% 6

Previous PTB PTB Previous 28w 1/6 17% 1/6 17% 1/6 17% 1/6 17% 2/6 33% 6

Preterm 34 Preterm 34w 1/4 25% 1/4 25% 0/4 0% 1/4 25% 1/4 25% 4

12w 1/11 9% 1/11 9% 8/11 73% 0/11 0% 1/11 9% 11

16w 1/11 9% 1/11 9% 7/11 64% 0/11 0% 2/11 18% 11

22w 1/7 14% 1/7 14% 3/7 43% 0/7 0% 2/7 29% 7

Preterm <34w <34w Preterm 28w 0/6 0% 1/6 17% 3/6 50% 0/6 0% 2/6 33% 6

TOTAL, PTB group 38/194 20% 34/194 18% 80/194 41% 25/194 13% 17/194 9% 194

12w 21/35 60% 3/35 9% 6/35 17% 3/35 9% 2/35 6% 35 16w 32/47 68% 4/47 9% 6/47 13% 1/47 2% 4/47 9% 47 22w 27/47 57% 6/47 13% 7/47 15% 2/47 4% 5/47 11% 47

28w 14/22 64% 2/22 9% 3/22 14% 1/22 5% 2/22 9% 22 Term >37w >37w Term

34w 14/24 58% 3/24 13% 4/24 17% 0/24 0% 3/24 13% 24

12w 2/6 33% 0/6 0% 3/6 50% 0/6 0% 1/6 17% 6 w w +6 Previous CT CT Previous 16w 3/5 60% 0/5 0% 1/5 20% 0/5 0% 1/5 20% 5 -36 +0 22w 4/6 67% 0/6 0% 2/6 33% 0/6 0% 0/6 0% 6

28w 2/4 50% 0/4 0% 2/4 50% 0/4 0% 0/4 0% 4

Preterm 34 Preterm 34w 2/4 50% 0/4 0% 2/4 50% 0/4 0% 0/4 0% 4 TOTAL, CT group 123/201 61% 18/201 9% 34/201 17% 7/201 3% 18/201 9% 201

CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, CST = community state type based on ward HCA of species data; w= weeks

180 A comparison of previous preterm birth and cervical treatment in pregnancy

Gestation at Gestation at Gestation at sampling Gestation at sampling ABPatient number birth (weeks) Patient number birth (weeks) 12W 16W 22W 28W 34W <34 34-37 >37 12W 16W 22W 28W 34W <34 34-37 >37 CST Species

1 >>>>>> 1 >>> > > > I L. Crispatus

2 >>>>> 2 >>>> > II L. gasseri

3 >>>>> 3 >> > > > > III L. iners

4 >>>>> 4 >>>> > IV Diverse species

5 >>> >>>> 5 >>>> > V L. Jensenii

6 >>>>>>> 6 >>>> > n/a Cerclage

7 >>>>> 7 >>>>>>>

8 >>>>>> 8 >>>> >>

9 >>>>>> 9 >>>>>>>

10 >>>>> 10 >>>>>>>

11 >> > >>> > 11 >>>>>>>

12 >> > > > > 12 >>>>>>>

13 >> > >>> > 13 >>>>>>>

14 >>>>> 14 >>>>>>>

15 >>>> > 15 >>>>>>>

16 >>>> > 16 >>>>>>>

17 >>>> > 17 >>>> >>

18 >> > > > > > 18 >>>> >>

19 >> >> > > > > 19 >>>> >>

20 >>>>>>>> 20 >> > > > >>

21 >>>> > 21 >> > > > >>

22 >>>> > 22 >> > > > >>

23 >> > > > > 23 >> > > > >>

24 >>>> > 24 >> > > > >>

25 >> > > > > 25 >> > > > >>

26 >> > > > >> 26 >> > > > >>

27 >>>> >> 27 >> > > > >>

28 >>>>>>> 28 >> > > >> > >

29 >>>> >> 29 >> > > >> > >

30 >>> > > >> 30 >> >>> > >>

31 >>>>>>> 31 >>>>>>>>

32 >> > > > >> 32 >>>>>>>>

33 >>>>>>> 33 >>>>>>>>

34 >>>> >> 34 >>>> >>

35 >>>>> >> 35 >>>> >>

36 >> > > > > >> 36 >>>> >>

37 >> > > > > >> 37 >>>> >>

38 >> > >>> > > > 38 >> > > > >>

39 >> > > > >> 39 >> > > > >>

40 >> > >>> >> 40 >> > > > >>

41 >>>> >> 41 >>>> > > >> >>>>> >> > >>> Previous preterm birth <37 weeks <37 birth preterm Previous 42 >> 42 >> Previous excisionalPrevious cervicaltreatment 43 >>>>> >> 43 >>>> >>

44 >>>>> >> 44 >>>> >>

45 >>>> >> 45 >>>> >>

46 >> > > > >> 46 >>>>> >>

47 >>>>> >> 47 >> > >>>> > >

48 >>>> >> 48 >>>>>>>>

49 >>>> >> 49 >>>>>>>>

50 >> > > > > >> 50 >>> > > >>

51 >> >> > > >> 51 >> > >>>> > >

52 >>>> >> 52 >>>> > > >>

53 >> > > > >> 53 >>>> > >> >>

54 >> > > > >> 54 >>>> >>

55 >>>>>> >> 55 >>>> >>

56 >>>> >> 56 >> > > >> > >

57 >>>>>>> 57 >>>> >>

58 >>>>>>>>> 58 >>>> > > >>

59 >>>>>>> 59 >>>>>>>

60 >>>>> >> 60 >>>> >>

61 >> > > > >> 61 >>>>>>>>

62 >> > > > > >> 62 >> > >>> >>

63 >>>>>>> 63 >>>>>>>>

64 >>>>>>> 64 >>> > > >>

65 >>>> >> 65 >> > > >> > >

66 >>>> >> 66 >>>> > > >> 67 >>>>> >> Figure 5-15 Individual patient journey of assigned community state types (CSTs) for 67 PTB women (A), and 66 CT women (B) and their associated categorised gestation at birth. CST I= blue, CST II= green, CST III= red, CST IV= orange, CST V= light blue. Grey= sample not processed following cerclage insertion. (w= weeks gestation, CT= excisional cervical treatment, PTB= previous preterm birth <37 weeks)

181 A comparison of previous preterm birth and cervical treatment in pregnancy

5.2.6 Vaginal microbiota before 24 weeks and preterm birth prediction Chapter 4 demonstrated the predictive value of vaginal microbiota at screening before 24 weeks for subsequent gestation at birth. Here I compare microbial profiles of CT and PTB groups separately. Although underpowered, in the PTB group CST III (L. iners) associates with the highest rates of subsequent preterm birth <37 weeks (40%) compared to any other CST (OR 1.4, [95% CI 0.7 to 2.8], P=0.4). This provides high specificity for preterm birth (68%), but poor sensitivity (40%) due to the high prevalence of L. iners among term births. This is also reflected by a relatively low negative predictive value (NPV 62%). Lowest preterm rates in the PTB group were associated with CST I (29%; OR 0.69 [95% CI 0.3 to 1.7], P=0.5), CST II (23%; OR 0.5 [95% CI 0.3 to 1.3], P=0.2) and CST IV (17%; OR 0.33 [95% CI 0.1 to 1.2], P=0.1).

In the CT group, women harbouring high levels of L. iners before 24 weeks were also associated with the highest preterm birth rates (24%; OR 3.2 [95% CI 1.0 to 9.6], P=0.07), while low rates of preterm birth associated with dominance of CST I (10%, OR 0.69 [95% CI 0.3 to 1.9], P=0.6). Again, these were underpowered and therefore did not reach significance.

Interestingly, L. iners dominance provides a greater preterm birth risk in the CT group than in the PTB group (CT group RR 2.4 for preterm birth [95% CI 1.1 to 5.2], compared to PTB group RR 1.2 [95% CI 0.8 to 1.8]). This is associated with 91% specificity, 24% sensitivity and 85% NPV for preterm birth <37weeks.

182 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3 Associations between vaginal microbiota and cervical parameters

The relationship between the vaginal microbiome and cervical parameters was assessed among CT women (n=66) and compared to the PTB group (n=67).

Irrespective of gestation at sampling, a significant association between CL and CST was observed in the PTB group (P=0.002), but not the CT group (P=0.1, ANOVA, Figure 5-16A), supporting differing underlying aetiologies of cervical shortening among these women.

Among the prior PTB group, L. iners dominance (CST III) associated with a shorter CL (mean 31mm) than L. crispatus (38mm, P<0.001) or L. gasseri (37mm P=0.002; t test). In the CT group, mean CL was also marginally lower when associated with L. iners (30mm) when compared to L. crispatus (32mm, P=0.08) (Figure 5-16A).

Lower CVs were associated with L. iners when compared to L. crispatus in the PTB group (36cm3 v 41cm3, P=0.04), and when compared to L. gasseri in the CT group (L. iners: 30cm3 v L. gasseri: 40cm3, P=0.02; t-test, Figure 5-16B). VI and FI did not correlate with the vaginal microbiome in the PTB group, however in the CT group L. gasseri had a higher mean VI (5.6 ±3.3) and FI (30.9 ±4.9), than CST IV (VI 1.8 ±1.6, P=0.01) and L. jensenii (VI, 2.7 ±1.6 ; FI 28.15 ±3.4, P=0.01; Figure 5-16C, D).

183 A comparison of previous preterm birth and cervical treatment in pregnancy

ACervical length PTB CT ** P=0.002 ANOVA ns P=0.10 ANOVA 80 80 ** ns *** ns 60 60

40 40 CL (mm) CL (mm) 20 20

0 0 I II III IV V I II III IV V

BCervical volume PTB CT ns P=0.16 ANOVA ns P=0.15 ANOVA 100 ns 100 * ns * 80 80

3) 60 3) 60

40 40 CV (cmCV (cmCV 20 20

0 0 I II III IV V I II III IV V

CVascularisation Index PTB CT ns P=0.4 ANOVA ns P=0.06 ANOVA 20 20 * * 15 15

10 10

5 5

Vascularisation Index, VI 0 Vascularisation Index, VI 0 I II III IV V I II III IV V

DFlow Index PTB CT ns P=0.8 ANOVA ns P=0.05 ANOVA 50 50 **

40 40

30 30 Flow Index, FI Index, Flow FI Index, Flow 20 20

I II III IV V I II III IV V Community state type Community state type

CST I L.crispatus CST II L. gasseri CST III L.Iners CST IV Diverse species CST V L. jensenii

184 A comparison of previous preterm birth and cervical treatment in pregnancy

Figure 5-16 Association between vaginal CST and corresponding CL (A), CV (B), VI (D) and FI (E) in 395 sampling time points in PTB and CT groups. CL varied with corresponding vaginal microbiota in PTB women (P=0.002, ANOVA): a shorter CL associated with L. iners (mean 31 mm) and a longer CL with L. crispatus (38mm, P<0.001) and L. gasseri (37mm P=0.002; t test). CL was not influenced by vaginal microbiota in CT women (P=0.1, ANOVA). (B) L. iners associated with lower CV compared to L. crispatus in PTB women (36cm3 v 41 cm3, P=0.04) and to L. gasseri among CT women (30cm3 v 40cm3, P=0.02, t-test). There was no correlation between vaginal CST and the cervical VI (D) or FI (E). (CST = community state type, CL= cervical length, CV= cervical volume, VI = vascularisation index, FI= flow index, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

185 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3.1 Cervical length and vaginal microbiota before 24 weeks CL measurements from CT and PTB participants taken at research time points <24 weeks were assessed for associations with vaginal microbiota and subsequent gestation at birth. As previously described in the larger ‘scan-only’ cohort (Figure 5-8), a short cervix was more strongly associated with earlier gestation at birth in the prior PTB compared to the CT group (mean 35.1 v 38.4 weeks, P=0.02; t-test, Figure 5-17A).

A total of 288 swab samples with matching cervical ultrasound data from 133 women were collected (146 samples from 66 women in the CT group, and 142 samples from 67 women in the PTB group). When categorized into community state types, this was compiled of 119 samples dominant in L. crispatus (CST I) and 82 dominant in L. iners (CST III), while CST II, IV and V made up the remaining 87 samples (n=38, n=24 and n=25 respectively; Table 5-12).

A total of 31 samples were collected from women found to have a short CL ≤25mm (CST I, n=13; CST II n=2; CST III n=13; CST IV, n=2 and CST V n=1) all of whom received a cervical cerclage. A total of 257 samples (102 women) were collected from women with non-short (>25mm) cervices (Table 5-12). Hereafter, CSTs I and III were specifically selected for correlation with cervical lengths given their predictive values for term and preterm birth respectively (as described in Chapter 4).

186 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-12 A comparison of PTB and CT women according to microbiota profiles of short (CL≤25mm) compared to non-short (>25mm) cervices at screening <24weeks Cohort Screening <24weeks Prior PTB Prior CT n= 67 women n= 66 women Total n/N % n/N % samples

Total samples taken 142/288 49.3% 146/288 50.7% 288

Cervical length ≤25mm 15/142 11% 16/146 11% 31 >25mm 127/142 89% 130/146 89% 257

Community state type & CL CST I, ≤25mm 3/142 2% 10/146 7% 13 L. crispatus >25mm 27/142 19% 79/146 54% 106 TOTAL 30/142 21% 89/146 61% 119 CST III, ≤25mm 9/142 6% 4/146 3% 13 L. iners >25mm 48/142 34% 21/146 14% 69 TOTAL 57/142 40% 25/146 17% 82 Other CSTs ≤25mm 3/142 2% 2/146 1% 5 >25mm 52/142 37% 30/146 21% 82 TOTAL 55/142 39% 32/146 22% 87 Combined 142/142 100% 146/146 100% 288 TOTAL CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, w = weeks, CST= community state type, CL = cervical length

L. crispatus was more strongly associated with a later gestation at delivery than L. iners (Figure 5-17B). This was true of both a short CL ≤25mm (mean gestation at birth L. iners 38.6 v L. crispatus 35.5 weeks, P=0.03), and normal CL >25mm (mean L. iners 38.7 v L. crispatus 36.3 weeks P<0.001; Table 5-13). This trend continued among CT women: following detection of a short CL ≤25mm, L. crispatus was associated with longer gestational age (mean gestation at birth 38.8 weeks) compared to L. iners (mean 36.3 weeks, P=0.03; t-test; Figure 5-17C, Table 5-13). A similar trend was observed in the PTB group; mean gestation at birth associated with L. crispatus dominance was 36.5 weeks and L. iners, 35.0 weeks (P=0.15; t-test; Figure 5-17C, Table 5-13).

187 A comparison of previous preterm birth and cervical treatment in pregnancy

Cervical length screening <24 weeks CL screening <24 weeks A CT v PTB B L. crispatus v L. iners

42 42 P=0.05 P=0.8

40 *** 40 * * *** 38 38

36 36

34 34

32 P=0.08 32 P=0.5 Gestation at birth (weeks) birth at Gestation Gestation at birth (weeks) birth at Gestation 30 30 <25mm >25mm <25mm >25mm Samples: n=31 n=257 Samples: n=31 n=257

Previous excisional CT Previous PTB L. crispatus L. iners Other species (n=66 women, 146 samples) (n=67, 142 samples) (n=119 n=82 n=87 samples)

C Previous CT Previous PTB P=0.6 *** P=0.8 *

* ** P=0.5 P=0.16 42 40 38 Term, 37 weeks 36 34 32 30 28 26 Gestation at birth Gestation(weeks) at birth 24 22

CST I CST I CST I CST I CST III CST III CST III CST III CL 25mm >25mm 25mm >25mm Samples: n=16 n=130 n=15 n=127

CL <25mm, L. crispatus CL >25mm, L. crispatus CL <25mm, L. iners CL >25mm, L. iners

Figure 5-17 Gestation at birth comparing CL ≤25mm (n=31 women) and >25mm (n=102 women) at screening <24 weeks. (A) CT women (n=66) delivered later

188 A comparison of previous preterm birth and cervical treatment in pregnancy

than PTB women (n=67) irrespective of CL measurement. (B) L. iners dominance associated with earlier gestation at birth (mean 35.5 weeks) than L. crispatus given a short CL ≤25mm (39.0 weeks, P=0.03) as well as a long CL >25mm (36.3 weeks CT v 38.7 weeks PTB, P<0.001). (C) Gestation at birth following detection of a short CL ≤25mm was later when associated with L. crispatus in both CT (mean 38.8 weeks) and PTB women (36.5 weeks), while L. iners associated with earlier birth gestation (CT: mean 36.3 weeks, P=0.03, and PTB: mean 35.0 weeks, P=0.15; t- test) (CL= cervical length, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

189 A comparison of previous preterm birth and cervical treatment in pregnancy

Table 5-13 A comparison of gestation at birth between previous PTB and CT, as a function of vaginal microbiome and cervical length threshold ≤25mm, taken at screening before 24 weeks gestation

Gestation at delivery (weeks) Cervical length Previous CT Previous PTB All samples *P value CST Species (mm) Mean ±SD Min Max Mean ±SD Min Max Mean ±SD Min Max CT v PTB

CL ≤25 38.8 ±2 36 41 36.5 ±1 36 37 38.6 ±2 36 41 0.05 I L. crispatus CL >25 39.4 ±2 35 42 37.0 ±2 33 39 38.7 ±2 33 42 ***<0.001

CL ≤25 36.3 ±1.6 34.3 38.0 35.0 ±5.3 24.7 40.4 35.5 ±4.5 24.7 40.4 0.6 III L. iners CL >25 38.1 ±1.9 34.3 41.0 35.5 ±5.1 22.4 41.0 36.3 ±4.5 22.4 41.0 *0.031

other CL ≤25 38.4 ±0.5 38.0 38.7 35.1 ±4.4 27.4 38.0 36.0 ±3.9 27.4 38.7 0.4 species CL >25 39.4 ±1.4 36.0 41.4 37.0 ±3.4 27.9 41.1 37.9 ±3.1 27.9 41.4 ***<0.001 CL ≤25 *0.03 0.15 *0.025

P value CST I v III CL >25 **0.005 0.16 ***<0.0001

CL ≤25 38.4 ±1.7 34.3 41.1 35.2 ±3.7 24.7 40.4 36.8 ±0.7 24.7 41.1 *0.015 Total CL >25 39.1 ±1.7 34.3 42.0 36.4 ±3.9 22.4 41.1 38.2 ±0.0 22.4 42.0 ***<0.001

All <25mm v >25mm P value 0.050 0.080 ns CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, w = weeks, CST= community state type, CL = cervical length (mm) *P<0.05,**P<0.01; unpaired t-test two-tailed previous CT v previous PTB, and CST I v CST III

190 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3.2 The value of cervical length Only 12% (3/25) of the prior PTB group dominant in L. crispatus at their first sampling time-point went onto deliver preterm <37weeks. This compared to 43% (3/7) of the CT group, and 60% (15/25) of the prior PTB group who were L. iners dominant at first sampling. While L. crispatus appears protective against preterm birth, this data indicates that it is not completely inhibitory, particularly within the CT group.

Therefore I next aimed to determine how significant the role a shortened cervix plays in women post-excisional CT, compared to the ‘protectiveness’ offered by specific vaginal microbiota. For this, CL measurements from L. crispatus-associated preterm births were compared to L. crispatus-associated term births. These were also compared CL measurements of L. iners- associated preterm and term births (Figure 5-18).

Shortest mean CLs were observed in CT women with L. crispatus-associated preterm births (mean CL 28mm). This was significantly lower than CT women with L. crispatus associated- term birth (34mm, P=0.008, Figure 5-18C). Longest CLs were associated with term births in PTB women dominated by either L. crispatus (mean 39mm) or L. iners (mean 38mm, Figure 5-18C, Table 5-14). This indicates that although L. crispatus may protect against preterm birth, if the cervix is significantly shortened or damaged by cervical treatment, the pregnancy remains at risk of preterm birth (Figure 5-18C).

191 A comparison of previous preterm birth and cervical treatment in pregnancy

Screening <24weeks: Cervical length CT v PTB

P=0.16 A 70 P=0.11 B 42 P=0.13 60 *** 40

50 38

40 36

30 34 CL (mm) 20 32

10 30

Gestation at birth (weeks) at birth Gestation 28 0 CTPTB CT PTB 26

Preterm < 37weeks Term > 37weeks 24 60 40 20 0 CT, preterm <37 weeks CT, term >37 weeks Cervical length (mm) PTB, preterm <37 weeks PTB, term >37 weeks Previous excisional CT Previous PTB

C Previous CT Previous PTB

P=0.2 50 ** P=0.4 *

40 L. crispatus, 30 Preterm birth <37weeks L. crispatus, 20 Term birth >37weeks CL (mm) L. iners, 10 Preterm birth <37weeks

L. iners, 0 Term birth >37weeks

<37w >37w <37w >37w <37w >37w <37w >37w

L. iners L. iners L. iners L. iners L. crispatusL. crispatus L. crispatusL. crispatus

Figure 5-18 (A) Comparison of CL measurements for outcomes of term versus preterm birth among CT (n=66 women, 146 scans) and PTB women (n=67, 142 scans). CL was lower in the CT compared to PTB women; term birth, mean CL 34mm CT v 38 mm PTB (P<0.001) and 31 v 35mm respectively for preterm birth (P=0.13). (B) Linear regression indicates a positive correlation between CL and gestation at birth among the CT (r=0.18, P=0.03), and PTB women (r=0.11, ns) (C) Shortest CLs are observed in CT women with L. crispatus dominant-preterm births (mean CL 28mm), shorter than L. crispatus dominant-term births in CT women (34mm, P=0.008). L. crispatus dominant-preterm births in PTB women also had

192 A comparison of previous preterm birth and cervical treatment in pregnancy

shorter CL than corresponding term births (mean 32 v 39mm, P=0.02). Longest CL’s were observed in PTB women with either L. crispatus or L. iners dominant-term births (mean 39 and 38mm respectively). (*P<0.05, **P<0.01, ***P<0.001; t-test. r value = Pearson correlation co-efficient, ns= nonsignificant, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, CL = cervical length, mm)

Table 5-14 Comparison of PTB and CT women according to microbial profiles at screening <24 weeks, and corresponding cervical lengths and volumes.

3 Birth outcome Cervical volume (cm ) Cervical length (mm) CST N= (weeks) Mean ±SD Range Mean ±SD Range

CST I, Preterm birth <37w 9 21.7* ±9.3 (11.7 - 43.4) 29†† 9.9 (5 - 39) L. crispatus Term > 37w 80 29.7 ±12.8 (10.3 - 63.4) 34†† 5.5 (23 - 46)

CST III, Preterm birth <37w 6 30.3 ±9.7 (10.8 - 37) 31 5.2 (25 - 39) L. iners Term > 37w 19 33.1* ±13.2 (13.2 - 53.7) 33 6.3 (21 - 43) Previous CT

CST I, Preterm birth <37w 8 31.4 ±10.7 (19.9 - 48) 32† 6.7 (25 - 46) L. crispatus Term > 37w 20 35.9 ±13.1 (23.3 - 74.9) 39† 7.7 (22 - 54) Preterm birth <37w 22 34.0 ±13.1 (18.4 - 65.7) 35 9.3 (15 - 55) CST III, L. iners Term > 37w 35 37.8 ±9.5 (17.5 - 51.7) 38 8.4 (22 - 57) Previous PTB *P=0.02; t-test for L. crispatus preterm birth v L. iners term birth, †P=0.02, ††P=0.008; t-test for L. crispatus term v preterm birth, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, w = weeks, CST= community state type.

193 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3.3 Correlation between the cervix, vaginal microbiota and gestation at birth In a linear regression cervical length, measured before 24 weeks, was plotted against gestation at birth for CT and prior PTB groups (Figure 5-19). There was a significant and positive correlation between a shortened cervix and earlier gestation at birth, when associated with L. iners dominance in the PTB group (r=0.21, P=0.02; Pearson correlation) and CT group (r=0.35, P=0.003). Dominance of L. crispatus did not associate with CL and gestation at birth in the prior PTB group (r=0.23, P=0.07), however a significant correlation was observed in the CT group (r=0.14, P=0.004; Figure 5-19).

A Previous PTB B Previous CT 42 42 40 40

38 38 36 36

34 34 32 32 30 30 28 CST I r=0.23 28 CST I r=0.14** P=0.6 P=0.4 CST III r=0.21* CST III r=0.35**

Gestation at birth (weeks) 26 Gestation at birth (weeks) 26 24 24 60 55 50 45 40 35 30 25 20 15 10 5 0 60 55 50 45 40 35 30 25 20 15 10 5 0 Cerivcal length (mm) Cerivcal length (mm)

Figure 5-19 Linear regression correlation between CL measured before 24 weeks, corresponding vaginal microbiome, and subsequent gestation at birth. As cervical length shortens, L. iners dominance associates with a significantly earlier gestation at birth in both PTB (r=0.21, P=0.02) and CT (r=0.35, P=0.003) groups. L. crispatus dominance provided a correlation with shortening CL and earlier gestation at birth in the CT group (r=0.14, P=0.004, but not in the PTB group (r=0.23, P=0.07). (*P<0.05, ***P<0.01; Pearson correlation co-efficient)

194 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3.3a Impact of gestation at screening Linear regression demonstrated that among prior PTB women, early onset of cervical shortening (CL ≤25mm) correlates with earlier gestation at birth (r=0.5, P=0.07; Figure 5-20 A, B). The addition of L. iners dominance further increases this positive correlation (r=0.7, P=0.03; Figure 5-20C). This relationship does not extend to CT women, in whom gestation at birth remains unchanged irrespective of gestational age at detection of a short cervix; 12 weeks is comparable to 24 weeks with respect to eventual gestation at birth (P=0.2 Figure 5-20A, B). This remains the case with L. crispatus dominance (r=-0.4, P=0.18; Figure 5-20C). However in women with L. iners dominance, a positive and significant correlation exists between gestation at screening and gestation at birth in those with a short cervix (r=0.9, P=0.04).

195 A comparison of previous preterm birth and cervical treatment in pregnancy

A B CL 25mm CL > 25mm Prior PTB v CT Prior PTB v CT 42 42

40 40

38 38

36 36

34 34

32 32

30 30

28 28 CT: r = -0.47 CT: r = 0.1 Gestation at birth (weeks) at Gestation birth 26 (weeks) at Gestation birth 26 PTB: r = 0.49 PTB: r = 0.01 24 24 12 16 20 24 12 16 20 24 Gestation at screening (weeks) Gestation at screening (weeks)

Previous excisional CT Previous PTB

CDCL 25mm CL > 25mm Prior PTB v CT L. crispatus v L. iners Prior PTB v CT L. crispatus v L. iners

42 42

40 40

38 38

36 36

34 34

32 32

30 30 Gestation at birth (weeks) at Gestation birth Gestation at birth (weeks) at Gestation birth 28 28 CST I: PTB r = 0.08, CT r = 0.08 26 CST I: PTB r = -0.89, CT r = -0.4 26 CST III: PTB r = -0.04, CT r = 0.07 CST III: PTB r = 0.71*, CT r = 0.96* 24 24 12 16 20 24 12 16 20 24 Gestation at screening (weeks) Gestation at screening (weeks)

Previous PTB, L. crispatus Previous CT, L. crispatus Other species Previous PTB, L. iners Previous CT, L. iners Figure 5-20 Earlier gestation at screening positively correlates with earlier gestation at birth when a short cervix ≤25mm is associated with PTB women and/or L. iners dominance. (A) A positive correlation exists between earlier gestation screening and gestation at birth given a short cervix, ≤25mm (n=31) in the PTB group (n=142 screening time-points, r=0.49, P=0.07), but not the CT group (n=146, r=-0.47, P=0.2). (C) The combination of PTB women with a short CL≤25mm, and an L. iners dominated microbiome provides a significantly positive correlation (n=9, r=0.71, P=0.04), as do CT women with a short cervix and L. iners dominance (n=4, r=0.96, P=0.04). The same cohorts in the presence of L. crispatus do not provide any correlation between birth and screening (PTB: n=3, r=-0.89, P=0.3 and CT: n=10, r=-0.4, P=0.3), and neither does a longer CL >25mm regardless or risk factor or co- existing vaginal microbiota (B, D,) (**P<0.01, *P<0.05, r= Pearson correlation co- efficient, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, CL= cervical length, mm)

196 A comparison of previous preterm birth and cervical treatment in pregnancy

5.3.4 Cervical volume and vaginal microbiota before 24 weeks The association between cervical volume and vaginal microbiota with respect to gestation at delivery was assessed in 288 samples collected at varying time points before 24 weeks where matched transvaginal scans and vaginal swabs where available. Women with CT (n=66, 146 samples) were compared to PTB women (n=67, 142 samples). CV’s in the CT group were significant lower than the prior PTB group, both among women going on to deliver preterm <37 weeks (CT: mean 23 cm3 v PTB: 34cm3, P=0.003) and at term (31 v 38cm3 respectively, P<0.001, Figure 5-21A). Linear regression showed that lower CV correlates with earlier gestation at birth in both groups (CT: r=0.23, P=0.006 and PTB: r=0.18, P=0.03, Figure 5-21B).

Screening <24weeks: Cervical volume, CT v PTB

A B 42

100 P=0.09 P=0.07 40 ** *** 80 38

) 36

3 60

34 40 CV (cmCV 32 20 30

0 (weeks) birth at Gestation CTPTB CT PTB 28

Preterm < 37weeks Term > 37weeks 26 Previous CT r= 0.23** Previous PTB r= 0.18* CT, preterm birth <37 weeks CT, term birth >37 weeks 24 PTB, preterm birth <37 weeks PTB, term birth >37 weeks 100 80 60 40 20 0 Cervical volume (cm 3)

Figure 5-21 (A) At assessment prior to 24 weeks, CV was lower in the CT group (n=66 women, 146 scans) compared to the PTB group (n=67, 142 scans); Mean CV for term birth, CT 31 v PTB 38cm3, P<0.001, and for preterm birth 23 v 34cm3 respectively, P=0.003). There were no significant differences in CV when comparing preterm and term birth within the CT group (P=0.09) or PTB group (P=0.07). (B) Linear regression demonstrates lower CV correlates with earlier gestation at birth in both CT (n=146 screening time-points, r=0.23, P=0.006), and PTB groups (n=142, r=0.18, P=0.03) (**P<0.01, ***P<0.001; t-test, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks, CV = cervical volume)

197 A comparison of previous preterm birth and cervical treatment in pregnancy

Similar to trends observed in cervical length, smallest cervical volumes associated most strongly with L. crispatus dominant-preterm births, and largest volumes with L. iners dominant-term births (Figure 5-22). This trend was most marked in the CT group, where volumes associated with L. crispatus dominant-preterm births were lower (mean CV 21.7cm3) than both L. crispatus dominant-term births (29.7 cm3, P=0.06) and L. iners dominant-term births (33.1 cm3, P=0.02, Figure 5-22A, Table 5-14).

A similar trend was observed in women with a previous PTB, where reduced CVs were observed in ‘L. crispatus dominant-preterm births’ (mean CV 31.7cm3) compared to increased CVs observed in term births with either L. crispatus (35. cm3, P=0.3) or L. iners (37.8 cm3, P=0.12). CVs were comparable among ‘the term, L. crispatus’ and ‘preterm, L. iners’ group, both in CT women (29.7 v 30.1cm3) and prior PTB women (35.9 v 34.0cm3). This suggests an interactive and contributory role of both CV and vaginal microbiota and birth gestation; when CVs are comparable, L. crispatus provides a protective mechanism against preterm birth, while L. iners contributes additional risk. Equally, a large CV provides some protection against L. iners associated preterm birth, while a small volume is risk for preterm birth despite L. crispatus protection (P=0.02, Figure 5-22A, Table 5-14).

L. iners dominance in women with previous PTB (Figure 5-22C) was positively correlated to decreasing CV and earlier gestational age at birth (r=0.26, P=0.05). However no such relationships were observed in women with L. crispatus dominance (r=0.17, P=0.4). In CT women a significant and positive correlation between lower CV and gestation at birth in both L. crispatus and L. iners dominated bacterial communities was detected (L. crispatus: r=0.45, P<0.001 and L. iners: r=0.61, P=0.001) (Figure 5-22B).

198 A comparison of previous preterm birth and cervical treatment in pregnancy

Screening <24weeks: Cervical volume

A Previous CT Previous PTB 60 P=0.06 P=0.6 P=0.3 P=0.2 * P=0.1

40 ) 3

CV (cmCV 20

0

<37w >37w <37w >37w <37w >37w <37w >37w

L. iners L. iners L. iners L. iners L. crispatusL. crispatus L. crispatusL. crispatus L. crispatus, Preterm birth <37weeks L. iners, Preterm birth <37weeks L. crispatus, Term birth >37weeks L. iners, Term birth >37weeks

B Previous CT C Previous PTB

42 42

40 40

38 38

36 36

34 34

32 32

30 I L. crispatus 30 III L. iners 28 28 Gestation at birth (weeks) Gestation at birth (weeks) birth at Gestation CST I: r =0.45*** CST I: r =0.17 26 CST III: r =0.61*** 26 CST III: r =0.26 24 24 80 60 40 20 0 80 60 40 20 0 3 Cervical volume (cm 3) Cervical volume (cm ) Figure 5-22 Lowest CV was observed in L. crispatus dominant-preterm births and largest CV in L. iners dominant-term birth. (A) In the CT group L. crispatus dominant-preterm births had lower CVs than L. crispatus dominant-term births (mean CV 21.7 v 29.7 cm3, P=0.06) as well as L. iners dominant-term births (33.1 cm3, P=0.02). Similarly in the PTB group, CVs associated with

199 A comparison of previous preterm birth and cervical treatment in pregnancy

L. crispatus-preterm birth were lower than L. crispatus-term birth (CV 31.7cm3 v 35.9 cm3, P=0.3) and L. iners term birth (37.8 cm3, P=0.12). L. crispatus-term birth and L. iners-preterm birth had comparable cervical volumes (CT: 29.7 v 30.1cm3, and PTB: 35.9 v 34.0cm3). (B) Linear regression analysis demonstrates a significant and positive correlation between lower CV and earlier gestation at birth in CT women (L. crispatus: r=0.45, P<0.001 and L. iners: r=0.61, P=0.001). (C) In PTB women, high abundance of L. iners provides a greater positive correlation between decreasing CV and earlier birth gestation (r=0.26, P=0.05) than L. crispatus dominance (r=0.17, P=0.4). (P value = t-test, r = Pearson correlation, CT= previous excisional cervical treatment, PTB= previous preterm birth <37weeks)

200 A comparison of previous preterm birth and cervical treatment in pregnancy

5.4 Preterm birth prediction by vaginal microbiota and cervical length

The predictive accuracy provided by vaginal microbiota at screening before 24 weeks is improved through the addition of cervical length measurements. Alone, L. iners dominance is associated with the highest risk of preterm in both prior PTB (Likelihood ratio, LR 1.2) and CT groups (LR 2.6). L. iners dominance also provides high specificity (PTB 68% and CT 91%), but poor sensitivity (PTB 40% and CT 24%) for subsequent preterm birth <37 weeks.

The detection of a short CL ≤25mm in combination with high L. iners abundance significantly increases the risk of preterm birth <37 weeks, particularly among CT women (RR 8.5, 95% CI 2.1 to 35, P=0.01). This is associated with improved sensitivity at marginal compromise in specificity; LR 5.0, sensitivity 57% and specificity 89%. In the prior PTB group, a short CL ≤25mm with L. iners dominance also improves sensitivity 62% and specificity 68%, increasing associated preterm birth risk, although this does not reach significance (RR 2.7, 95% CI 0.95 to 7.9, P=0.07).

Little can be concluded from the combination of a short CL ≤25mm in combination with L. crispatus dominance, as this was a rare occurrence before 24 weeks (2% of PTB samples, 3/142 of PTB samples, and 7% of CT samples, 10/146).

Reassurance may be gained from the absence of L. iners and a short cervix as this was associated with high negative predicative values in both groups (NPV: CT 98% and PTB 95%). Equally, reassuring is the combination of a long CL >25mm and vaginal microbiota dominated L. crispatus; in CT women LR is 0.7 (RR 0.58, 95% CI 0.18 to 1.2, P=0.08) and in PTB women LR is 0.8 (RR 0.77, 95% CI 0.37 to 1.6, P=0.6).

201 A comparison of previous preterm birth and cervical treatment in pregnancy

Discussion

In this chapter I describe the cervical anatomy and microbial profiles underlying preterm birth in pregnant women following excisional cervical treatment, and compare these to women with a prior preterm birth. In rejection of my hypothesis, I describe lower rates of vaginal dysbiosis and microbial instability, associated with proportionally higher abundance of L. crispatus dominance in the cervical treatment cohort compared to women with a prior preterm birth. This microbial stability was associated with lower rates of preterm birth before 37 weeks, despite lower cervical volume in the CT women. A significant loss of cervical volume did remain a risk factor for preterm birth in the cervical treatment group, this was not entirely negated by an L. crispatus dominant microbiome. A total of 40% of preterm births occurred despite L. crispatus dominance and these women exhibited cervical volumes substantially lower than the remaining study population. This highlights the contributory roles provided by both the mechanical support of the cervix, as well as vaginal microbiota when considering aetiologies of preterm birth. Dominance of L. iners associated with cervical shortening before 24 weeks, although this finding was isolated to women with a prior preterm birth.

Cervical morphology in pregnancy: cervical treatment and prior preterm birth Evidence for regeneration of the cervix following excisional treatment is conflicting 74, with some reports that the cervical volume deficit at 6 months post-treatment is proportional to the size of the excised tissue 241. In this chapter I demonstrate that a deficit of cervical volume post-cervical treatment is most significant at 12 weeks, declining with advancing gestation until at least the end of the second trimester (28 weeks). By 34 weeks the cervical morphology was comparable to normal pregnancy indicating substantial cervical remodeling had occurred by this stage of pregnancy. At 22 weeks, a gestational age at which preterm birth screening is frequently performed, the cervical treatment group demonstrated cervical lengths 4mm shorter on average than controls (32 v 36mm, P<0.01), a finding consistent with a previous cross-sectional study of 473 pregnancies post treatment 218. In that study Poon et al218 reported, with mean difference in CL of 2mm at 20-24 weeks (32mm cervical treatment vs. 34mm controls, P<0.001), that cervical treatment associated with high rates of spontaneous birth before 34 weeks. Substantiating this association is a growing body of evidence to support elevated preterm birth rates in pregnancy post cervical treatment 67, 71, 73, 246.

202 A comparison of previous preterm birth and cervical treatment in pregnancy

The proportional association between size of excised tissue and subsequent preterm birth risk75 have led some to hypothesize that the loss of cervical tissue mechanically weakens the cervix which leads to preterm birth 242, 247. The findings in my study do support this, as the smallest cervical volumes were associated with preterm birth risk, but importantly the composition of vaginal microbiota also contributes to this risk.

Microbial profiles in pregnancy: cervical treatment and to prior preterm birth This chapter reports differences in the vaginal microbial composition of two distinct populations at risk of preterm birth. Women with a prior spontaneous preterm birth demonstrated low microbial stability and high species diversity with advancing gestation. This was associated with high abundance of L. iners at longitudinal assessment, compared to women with pre-pregnancy excisional cervical treatment and normal term controls 171. Compared to other commensal Lactobacillus spp., L. iners colonisation is associated with higher vaginal pH 230, 231, poorer inhibition of anaerobic bacteria colonisation and transition to BV-like states 133, 233, 234. High abundance of L. iners has previously been implicated in preterm birth 182, as discussed in Chapter 4. The cervical treatment group in contrast exhibited high microbial stability, with low diversity and a microbial composition highly abundant in L. crispatus. It is likely that the dominance of L. crispatus, a known inhibitor of anaerobic pathogens in the vagina 228, played a major contributory role in the effective protection against preterm birth in this cervical treatment cohort, despite a shortened cervix in pregnancy.

The high microbial stability observed in the cervical treatment cohort was in contrast to the original hypothesis. Based on reports in non-pregnant women where CST IV-states associate with HPV infection 162, slow HPV clearance 244, and high grade CIN severity 245, it was hypothesised that a higher incidence of vaginal dysbiosis would be observed in pregnancy post- excisional treatment. Instead, cervical treatment women demonstrated rates of dysbiosis comparable to healthy low-risk pregnancy 171, which were markedly lower than women with a prior preterm birth. It is possible however that the low incidence of dysbiosis in my cohort reflected an effective clearance of HPV among the study participants. Future work may focus on antenatal persistence of HPV in pregnancy, given that mice models indicate HPV infection in pregnancy impedes cervical defense against bacterial invasion of the uterus 243. Brotman et al reported in non-pregnant women that rapid HPV clearance was associated with L. gasseri dominance 244. Interestingly among the cervical treatment women in our pregnant cohort, those dominant in L. gasseri had the largest cervical volumes, as well as the longest

203 A comparison of previous preterm birth and cervical treatment in pregnancy

duration of pregnancy. Numbers were small however (n=3/67); none delivered preterm or had a short cervix therefore few conclusions could be drawn. These findings may justify further investigation into the potential benefits of L. gasseri for a therapeutic preterm birth intervention in the future, such a probiotic.

The contribution of cervix and vaginal microbiota to risk of preterm birth Through incorporation of vaginal microbiota and cervical length measurements it is possible to provide cohort specific risk-profiles for cervical treatment and prior preterm birth groups. This understandably, relates to their substantially differing underlying aetiologies for preterm birth. In women with a prior preterm birth the combination of L. iners and a short CL ≤25mm before 24 weeks was associated with an increased risk of preterm birth (RR 2.7, P=0.07). In the cervical treatment group, the background incidence of L. iners dominance is relatively low, therefore combination of L. iners with a short cervix substantially increases the risk of preterm birth (RR 8.5, 95% CI 2.1 to 35).

Limitations A major limiting factor in this study was the lack of cervical treatment women delivering early preterm (<34 weeks). This may suggest that the population included in this study was not representative of a truly high-risk cohort. Given the relatively rare incidence of preterm birth before 34 weeks in cervical treatment women (1% of pregnancies) 218, a much larger sample size would be required to assess this.

As discussed in Chapters 3 and 4, a further limitation was the insertion of cervical cerclages for women with the shortest cervical lengths. It would have been unethical to have not intervened for a short cervix, and although some women still delivered preterm despite intervention, it is impossible to determine how many of the terms births had their pregnancy prolonged through cerclage insertion. This is an important limitation as the predictive accuracies, and in particular sensitivities, of L. iners dominant microbiota were likely underestimated in the cervical treatment group.

Also previously discussed in Chapter 4, a multivariate statistical model, taking into account the longitudinal nature of sequence and scan data, would be a more robust assessment of the reported relationship between vaginal microbiota, cervical length, ethnicity, and subsequent gestational age a birth among the high risk cohorts.

204 A comparison of previous preterm birth and cervical treatment in pregnancy

A further limitation of this project was the lack of availability data detailing depth of the excisional treatment among CT women. In retrospect this data would have been useful to correlate depth of excision and subsequent cervical volume in pregnancy. An attempt was made to collect this data retrospectively, but proved to very difficult to ascertain for several reasons. Primarily this detail was not accurately recorded in procedural notes or histology reports where treatment was performed locally, and where women had had their treatment elsewhere, hospitals either did not respond to information requests or could not locate the documentation required.

In summary This is the first study to investigate the anatomical and microbiological mechanism underlying the preterm birth in high-risk pregnant populations. I have provided evidence to refute the hypothesis that mechanical damage is a primary driver of preterm birth 75, 247, demonstrating that, despite shorter cervical lengths, rates of preterm birth in women with previous CT were lower than women with a prior preterm birth. I have also described that integrity of the cervix does still play an important role in preterm birth prevention, more-so among women post- cervical treatment, than those with a prior preterm birth. In particular, a large cervical volume appears to provide some protection against L. iners-associated preterm birth pathogenesis in this cohort. Having previously highlighted the protective role attributable to L. crispatus in preventing preterm birth, some women post-cervical treatment still delivered preterm despite L. crispatus dominance. Through assessment of corresponding cervical volumes, it was revealed that these women had significantly lower cervical volumes than L. crispatus-term births, as well as those with L. iners associated preterm births. This provides crucial and important mechanistic insight into the contribution of mechanical damage that drives preterm birth among this distinct cohort of women.

Overall, a vaginal microbiome dominated by L. crispatus appears protective against preterm birth in high-risk pregnancy, while prevalence of L. iners particularly in the presence of a short cervix strongly associates with preterm birth. This is likely to account for lower preterm birth rates among women post-cervical treatment than with a prior preterm birth, despite comparatively reduced cervical lengths and volumes in pregnancy.

205 Cervical cerclage, vaginal microbiota and preterm birth prevention

6 CERVICAL CERCLAGE, VAGINAL MICROBIOTA AND PRETERM BIRTH PREVENTION

206 Cervical cerclage, vaginal microbiota and preterm birth prevention

Chapter abstract

Hypothesis Two million cervical cerclages are inserted annually to prevent preterm birth, yet the procedure is associated with an increased risk of infection. Braided, as opposed to monofilament suture material is predominately used without evidence base. In this study I hypothesised that braided suture promotes pathobiont colonisation of the vagina, which contributes to poor pregnancy outcome.

Aims The aim of this study was to investigate the impact of cerclage suture material on the vaginal microbiome, the expression of inflammatory mediators in cervicovaginal fluid and the effect on cervical vascularity as an indicator of cervical remodelling, in high risk pregnancy.

Methods Initially a retrospective analysis of cervical cerclages over a 10 year period across 5 UK university hospitals was performed. Pregnancy outcomes were compared for braided versus monofilament suture material using mixed linear effects models and controlling for confounders. Following this, in a prospective study, pregnant women with a short CL (≤25mm) were randomised to a braided (n=25) or monofilament cerclage (n=24). Sampling of vaginal fluid and matched transvaginal scans for cervical vascularity was performed pre-, and 4, 8, 12 and 16 weeks post-cerclage. 16S rRNA gene sequencing characterised the vaginal microbiome and multiplex cytokine arrays assessed corresponding inflammatory mediator expression.

Results Retrospective analysis of 671 pregnancy outcomes revealed that braided cerclage was associated with increased intrauterine death (16% v 5%, P<0.0001) and preterm birth (28% v 17%, P<0.001) compared to monofilament cerclage. In the prospective study, insertion of a braided cerclage caused a 5-fold shift towards vaginal dysbiosis (P<0.01), which was associated with expression of pro-inflammatory mediators IL-1β, IL-6, IL-8, TNF-α and MMP-1, and premature induction of cervical vascularisation (P<0.05). In contrast, monofilament cerclage, associated with reduced vaginal dysbiosis (0.7-fold, P<0.01), maintained Lactobacillus spp. stability and did not promote inflammation or cervical vascular remodelling.

Conclusion Braided cervical cerclage induces significant disruption of vaginal microbial stability driving inflammation and premature cervical remodelling. These effects were not observed with monofilament, potentially explaining improved outcomes associated with its use in pregnancy.

207 Cervical cerclage, vaginal microbiota and preterm birth prevention

Introduction

The cervical cerclage, a purse string suture inserted around the uterine cervix (Figure 6-1), is clinically indicated for women with a short cervical length (≤25mm detected before 24 weeks gestation), or those with a history of three or more spontaneous preterm births 104, 111, 112. Globally, an estimated 2 million procedures are performed annually for the prevention of preterm birth 107, 108. The mechanism of action of the cerclage is uncertain, although it is thought to provide mechanical support to the cervix, as well as support the cervical mucosal plug as a barrier to ascending infection 105, 107, 248. Despite effectively reducing the risk of preterm birth by 20%, cerclage insertion does not significantly improve neonatal morbidity or mortality, and procedure itself is associated with a doubled risk of puerperal sepsis 104, 108, 249.

Figure 6-1 Cervical cerclage placement around the uterine cervix (A). The cerclage suture is inserted (B) and tied (C) in a purse-string fashion around the cervix.

Currently, there is no consensus as to the optimal cerclage technique 103, and braided suture materials, rather than monofilament alternatives, are predominately used for the procedure by 86% of obstetricians without evidence base 113 (Figure 6-2) .Braided sutures consist of non- absorbable polyester fibres weaved together. They are characteristically high in tensile strength and thought to provide a secure structural support to a weakened cervix 106. Monofilament sutures, consisting of a single strand of non-absorbable polyamide polymer, are less likely to harbour organisms than braided materials 250, but are thought to provide less secure knots due to their tendency to slip 106. The only study to investigate the impact of suture material on pregnancy outcome was an RCT comparing two types of braided suture, Ethibond™ and Mersilene™ 114. Excluding all monofilament cerclages, they reported no difference in rates of preterm birth among types of braided suture 114.

208 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-2 Suture materials used for cerclage insertion: braided and monofilament sutures. Reproduced from www.surgicalspecialties.com/material-guide.

When observed in other body sites, braided suture is associated with an intense, prolonged inflammatory tissue reaction, and when compared to monofilament, has a higher incidence of wound infection 251. Cerclage efficacy has previously been correlated to co-existing inflammation where elevated inflammatory mediators in cervicovaginal fluid such as IL-8, associates with earlier gestation at birth 252. Therefore it may be postulated that the inflammatory reaction associated with braided suture in other surgical arenas, adversely influences pregnancy outcome when inserted as a cerclage. In particular, the reported risk of puerperal sepsis and chorioamnionitis associated with cerclage insertion 108 may relate to the preferential use of braided suture material over monofilament in current clinical practice 113. This led to the hypothesis that braided cerclage disrupts vaginal microbial stability, inducing a local inflammatory response, thereby triggering premature cervical remodelling, where a monofilament alternative would not. This study aimed to address this through matched observations of the microbiome and cervical vascularity in women receiving an ultrasound indicated cerclage for cervical shortening.

209 Cervical cerclage, vaginal microbiota and preterm birth prevention

Aim

To investigate the impact of cerclage suture material on the structure of the vaginal microbiome, the expression of inflammatory mediators in cervicovaginal fluid, and the effect on cervical vascularity, as a marker of cervical ripening. Specifically this study aimed to compare monofilament and braided suture materials.

Hypotheses 1. Braided cervical cerclage induces greater vaginal microbial disruption in-vivo, as assessed by 16S rRNA gene sequencing. 2. Braided cerclage induces a greater inflammatory response detectable in the cervicovaginal fluid, than monofilament cerclage that is correlated with vaginal dysbiosis. 3. Suture-dependant inflammation influences cervical vascularity indices as measured at transvaginal ultrasound.

210 Cervical cerclage, vaginal microbiota and preterm birth prevention

Study design

6.1 Retrospective study

Retrospective outcome data was collected from singleton pregnancies receiving a cervical cerclage for preterm birth risk over a ten-year period between January 2003 and 2013 across five UK hospitals (B'ham = Birmingham women's hospital; C&W = Chelsea and Westminster Hospital, Cam = Cambridge University Hospitals; QCCH = Queen Charlotte's and Chelsea Hospital; SMH = St Mary's Hospital). Details regarding suture material used, outcomes of viable term birth (≥37+0 weeks), viable preterm birth (between 24+0 and 36+6 weeks’ gestation), and non-viable birth (still birth or miscarriage >16+0 weeks’ gestation) were collected. Additional metadata assessed included maternal age, ethnicity, parity, history of prior preterm birth or midtrimester miscarriage, history of excisional cervical treatment, clinical indication for cerclage (elective or ultrasound indicated), cervical length and gestational age at cerclage insertion.

Statistical analyses Assessment of differences in outcomes of viability and preterm birth between cerclage suture material groups was performed using the Fisher’s exact test for categorical variables and Wilcoxon test for continuous variables. Based on these distributions, possible sources of inhomogeneity between the monofilament and the braided suture material groups were identified. A linear mixed-effects model incorporating maternal age, parity, previous preterm birth, hospital location as fixed effects and indication for cerclage (ultrasound versus elective) as a random effect was used to compare braided versus monofilament suture material for the two primary outcomes (viability and preterm birth).

211 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.2 Prospective study

Patient recruitment and sample collection Women at risk of preterm birth requiring a sonographically indicated cervical cerclage were prospectively recruited from preterm surveillance clinics. Inclusion criteria were pregnant women with a CL ≤25mm measured by transvaginal scan at ≤ 23+6 weeks’ gestation with history of spontaneous preterm birth (<37+0 weeks). Eligible women were randomised to either braided Mersilene® (n=25) or monofilament Ethilon® (n=24) cerclage suture material. The same obstetrician performed the procedure, using a modified Shirodkar cerclage.

Participants were recruited prior to cerclage insertion, and followed up longitudinally at 4, 8, 12 and 16 weeks post-insertion (Figure 6-3). At each time point two high vaginal swabs were collected, immediately followed by a transvaginal scan for cervical vascularisation assessment. The collected cervicovaginal fluid was analysed using 16S rRNA gene sequencing for characterisation of the vaginal microbiome, and Human Magnetic Luminex Screen Assay (15- plex) (Luminex Corporation, Austin, Texas) with a Bioplex®200 system (Biorad Laboratories Ltd.) to assess corresponding expression of inflammatory cytokines pre and at 4 weeks post cerclage insertion (as previously described in Materials and Methods).

Figure 6-3 Pregnant women with a short CL (≤25mm) at TVS, were randomised to receive either monofilament or braided cervical cerclage. Prior to cerclage insertion, a TVS and two swabs (one for 16S rRNA gene sequencing and one for multiplex cytokine array) were taken. All women were followed up for serial swabs and TVS at 4, 8, 12 and 16 weeks post-cerclage insertion. (CL= cervical length; w = weeks, TVS = transvaginal ultrasound scan)

212 Cervical cerclage, vaginal microbiota and preterm birth prevention

Fifteen pre-specified analytes were selected for the multiplex cytokine array according to evidence of involvement in inflammatory change related to preterm birth, cervical ripening, and angiogenesis: granulocyte colony-stimulating factor (G-CSF), granulocyte-macrophage colony- stimulating factor (GM-CSF), intercellular Adhesion Molecule 1 (ICAM-1), interferon (IFN)-γ, interleukin (IL)-1β, IL-2, IL-4, IL-6, IL-8, IL-10, matrix metalloproteinase 1 (MMP-1), monocyte chemotactic protein (MCP)-1, tumor necrosis factor (TNF)-α, Regulated on Activation Normal T Expressed and Secreted/Chemokine ligand 5 (RANTES/CCL5) and Vascular endothelial growth factor (VEGF).

Due to potential limitations in the V1-V3 primers used for the MiSeq 16S rRNA gene sequencing, detection of specific bacterial species associated with bacterial vaginosis (BV), namely Atopobium vaginae and Gardnerella vaginalis, were examined further by quantitative PCR using 5μl of the bacterial DNA previously isolated from the vaginal swabs for sequencing.

Statistical analyses of scan and sequence data Bacterial 16S rRNA sequence data was assessed at genera and species taxonomic levels. Ward linkage HCA of bacterial genera, using a clustering density threshold of 0.75, classified samples into a normal (>90% Lactobacillus spp.), intermediate (30-90% Lactobacillus spp.) and dysbiotic microbiomes (<30% Lactobacillus spp.). Species data was assessed by ward linkage HCA, and classified, as previously described, into community state types (CSTs) 155. To identify potential associations between suture material and differing degrees of dysbiosis, an alternative classification of the species data was performed, as described by Borgdorff et al 159 into communities characterized by healthy Lactobacillus spp. dominance, L. iners, or intermediate or severe dysbiosis.

The effect of suture material and time from cerclage insertion on bacterial genera, number of species observed (richness) and alpha diversity were assessed using One-way ANOVA, Kruskal-Wallis, and Dunn’s multiple comparisons where appropriate. The linear discriminant analysis (LDA) effect size (LEfSe) method 202 was used to characterise differentially abundant taxonomic features of the two suture materials pre- and 4 weeks post-cerclage insertion. An alpha value of 0.01 was used for factorial Kruskal-Wallis test between classes and a threshold of 3.0 used for logarithmic LDA score for discriminative features. All sequence data has been uploaded to the European Nucleotide Archive’s (ENA) Sequence Read Archive (SRA) and made available to the public (accession number PRJEB11895).

213 Cervical cerclage, vaginal microbiota and preterm birth prevention

Quantitative PCR values for Atopobium vaginae and Gardnerella vaginalis were compared pre- and post- cerclage insertion using a paired t-test within suture groups. Mann Whitney compared the mean fold change in copy numbers with cerclage insertion for braided and monofilament groups.

The Wilcoxon signed rank test was used to compare cytokine analyte concentrations pre- and 4 weeks post-cerclage insertion. The Mann-Whitney test was applied to examine differences among suture material. Analyte expression was classified according to corresponding vaginal dysbiosis (at genera level) and the Mann Whitney test used to compare cytokine expression in the presence of a “normal” or “dysbiotic” microbiome. Cervical vascularisation indices (VI) were assessed for the impact of suture material, longitudinal changes with from time from cerclage insertion, and as function of corresponding bacterial classification, using Kruskal-Wallis and ANOVA multiple comparison analyses where appropriate. Linear regression analyses provided correlations between cervical vascularity and the number of observed species, and the Shannon index of alpha diversity, according to suture material.

214 Cervical cerclage, vaginal microbiota and preterm birth prevention

Retrospective study results

6.3 Results of a ten year audit.

A total of 671 women receiving cervical cerclage in pregnancy were identified from 5 UK university hospitals within a 10-year period. Of the total population 327 (49%) received braided suture and 344 (51%) monofilament suture material for their cervical cerclage (Table 6-1; Figure 6-4). Consistent with clinical practice, suture material varied across hospital sites depending on operator preference (Table 6-1; Figure 6-4).

Linear mixed effects modelling demonstrated suture material was the major factor influencing outcomes of both non-viable pregnancy (defined as miscarriage before 24 weeks or intrauterine death) and preterm birth, after adjustment for potential confounders including hospital location, maternal age, parity, history of a previous preterm birth, ethnicity and indication for cerclage insertion (obstetric history or ultrasound indicated) (Fig. S1, Table S2). Specifically, braided suture was strongly associated with increased risk of non-viable pregnancy when compared with monofilament suture (16% v 5%; relative risk (RR), 3.24; 95% confidence interval [CI], 1.91 to 5.48; P<0.0001; Figure 6-5A). Braided cerclage was also associated with significantly increased risk of preterm birth (28% v 17%; RR, 1.64; 95% CI, 1.23 to 2.19; P=0.0007; Figure 6-5A).

There was a significant interaction between gestation age at cerclage insertion and outcomes of preterm birth according to suture material type (Figure 6-6, Table 6-2). In particular braided cerclage when inserted as an ultrasound indicated cerclage was associated with increased preterm birth rates, but not as an elective cerclage (P<0.0004; Table 6-3). The gestational age at insertion of a monofilament cerclage did not influence gestation at birth or viability.

215 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-1 Demographics of women receiving braided versus monofilament cervical cerclage across five university maternal units, collected retrospectively

Braided Monofilament P value

n/N, % 327/671 49% 344/671 51%

Maternal age (years), median [95% CI] 34 [22.7-44] 34 [22.5-42] 0.031b

Ethnicity, n/ N % 0.075a Asian 72/327 22% 75/344 22% Black 78/327 24% 59/344 17% Caucasian 161/327 49% 193/344 56%

Parity, median [95% CI] 1 [0-4] 1 [0-4] 0.015b 0 175/327 54% 218/344 63% ≥1 152/327 46% 126/344 37%

Previous PTB/ MTL, median [95% CI] 1 [0-5] 1 [0-6] 0.136b

Hospital B'ham 80/327 24% 12/344 3% 1.00E-07a C&W 65/327 20% 94/344 27% Cam 36/327 11% 49/344 14% QCCH 13/327 4% 187/344 54% SMH 133/327 41% 2/344 1%

Indication for cerclage 0.063a Elective (Obstetric history) 160/327 49% 143/344 42% Ultrasound CL ≤25mm 167/327 51% 201/344 58%

GA at cerc (w), median [95% CI] 15 [11-23] 16 [10-23] 0.042b CL at cerc (mm), median [95% CI] 20 [5-25] 21 [5-25] 0.220b

GA at birth (w), median [95% CI] 37 [18-41] 38 [22-41] 5.03E-07 aP value = Fisher’s exact test, bP value = Wilcoxon test. GA = gestational age (weeks), PTB = spontaneous preterm birth <37 weeks, MTL= mid-trimester loss <24weeks, CL = cervical length (mm); w= weeks; B'ham = Birmingham women's hospital; C&W = Chelsea and Westminster Hospital, Cam = Cambridge University Hospitals; QCCH =Queen Charlotte's and Chelsea Hospital; SMH = St Mary's Hospital

216 Cervical cerclage, vaginal microbiota and preterm birth prevention

ABMaternal age Gestation at cerclage insertion CCL at cerclage insertion 50 * * 30 P=0.2 24 25

40 20 20

16 15 30 10 Age (ye ars) 12

CervicalLength (mm) 5 20 (weeks) age Gestational 8 0 Braided Monofilament Braided Monofilament Braided Monofilament

DEFParity Cerclage indication Hospital location P=0.06 500 * 400 250 ***

400 200 300

300 150 200 n= n= n= 200 100

100 100 50

0 0 0 Para 0 Para 1 Elective Ultrasound

C&W Cam SMH B'ham QCCH Braided Monof lament

Figure 6-4 A comparison of demographics of women receiving braided and monofilament cerclage suture material. (A) Women receiving a monofilament cerclage had a lower maternal age than those receiving a braided cerclage (P=0.03; Wilcoxon). (B) Gestation at cerclage insertion was earlier in the braided than monofilament group (P=0.04), however CL at insertion did not differ significantly among groups (P=0.2; Wilcoxon; C). (D) There were more nulliparous women in the monofilament than braided group (P=0.01; fisher exact). (E) Indication for cerclage (elective following poor obstetric history versus ultrasound indicated for a short CL ≤25mm) was not significantly different among groups (P=0.06; fisher exact). (F) Selected suture material for cerclage differed with hospital location as would be expected with variations in clinical practice (P<0.001; Chi-squared). (B'ham = Birmingham women's hospital; C&W = Chelsea and Westminster Hospital, Cam = Cambridge University Hospitals; QCCH =Queen Charlotte's and Chelsea Hospital; SMH = St Mary's Hospital; CL= cervical length (mm); Para 0/≥1= none/≥1 previous deliveries beyond 24 weeks gestation, Para 1 Indication for cerclage, Elective = obstetric history includes three or more spontaneous preterm birth; Indication for cerclage, Ultrasound = CL ≤25mm on transvaginal ultrasound)

217 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-5 Braided suture material for cervical cerclage is associated with higher rates of non-viable pregnancy and preterm birth. (A) Retrospective comparison of 10 years of birth outcomes for women receiving a cerclage based on suture material (monofilament, n=346 vs braided, n= 332) revealed higher rates of non-viable births (intrauterine death or delivery <24 weeks) in women receiving a braided cerclage compared to a monofilament alternative (16% vs 5% respectively; ***P<0.0001, Fisher’s exact test). (B) Preterm birth rates (24-37 weeks gestation) were also higher in women receiving braided cerclage compared to monofilament cerclage (28% braided vs 17% monofilament; ***P<0.0001, Fisher’s exact test).

218 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-6 Odds ratio for outcomes of (A) non-viable fetus and (B) viable preterm birth<37 weeks based on using monofilament (n= 344) rather than braided suture (n= 327) from retrospectively collected cerclage outcomes across 5 UK maternity units. Monofilament cerclage is associated with reduced odds of a non-viable pregnancy OR 0.91 (95%CI 0.86-0.95) and of preterm birth OR 0.65 (95%CI 0.50- 0.83) compared to braided cerclage. Early gestation at cerclage insertion also reduces risk of preterm birth (OR 0.94, 95% CI 0.91-0.97), but does not impact on pregnancy with viability (OR 0.99, 95% CI 0.99-1.03). Parity and maternal age do not significantly influence pregnancy outcome given a monofilament cerclage.

Table 6-2 Odds ratio for outcomes of viability and preterm birth based using monofilament rather than braided suture

Pregnancy outcome Non-viable fetus Preterm birth OR (95% CI) OR (95% CI)

Monofilament suture material 0.91 (0.86-0.95) 0.65 (0.50-0.83) Maternal Age 1.01 (0.99-1.01) 1.02 (0.99-1.04) Parity 0.99 (0.97-1.01) 0.97 (0.90-1.06) Gestation at cerclage insertion 0.99 (0.99-1.03) 0.94 (0.91-0.97) OR = odds ratio; 95% CI - 95% confidence interval

219 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-3 A comparison of pregnancy outcomes: Viable vs Non-viable birth, and Term versus Preterm birth, as a function of suture material type (braided and monofilament) in retrospectively collected data across 5 hospitals

Braided suture, n=327 Monofilament suture, n=344

Pregnancy outcome Non-viable Viable PTB Term Non-viable Viable PTB Term P value P value P value P value n=49 n=278 n=93 n=185 n=17 n=237 n=60 n=267 Hospital location, n (%) 0.187 0.657 0.022 0.001 B'ham 6 (7.5%) 74 (92.5%) 22 (30.1%) 52(69.9%) 0 (0%) 12 (100%) 6 (50%) 6 (50%) C&W 11 (16.9%) 54 (83.1%) 18 (33.3%) 36 (66.7%) 6 (6.4%) 88 (93.6%) 12 (13.6%) 76 (86.4%) Cambridge 8 (22.2%) 28 (77.8%) 7 (25.9%) 21 (74.1%) 5 (10.2%) 44 (89.8%) 15 (34.1%) 29 (65.9%) QCCH 2 (15.4%) 11 (84.6%) 3 (27.3%) 8 (72.7%) 5 (2.7%) 182 (97.3%) 27 (14.8%) 155 (85.2%) SMH 22 (16.5%) 111 (83.5%) 43 (38.7%) 68 (61.3%) 1 (50%) 1 (50%) 0 (0%) 1 (100%) Maternal age, 34 [23-42] 34.5 [23-44] 0.687 34 [23-44] 35 [22-44] 0.006 32 [25-38] 34 [22-42] 0.160 35 [24-44] 34 [22-42] 0.156 median [95% CI]

Ethnicity, n (%) 0.059 0.120 0.487 0.034 Asian 5 (6.9%) 67 (93.1%) 18 (26.9%) 49 (73.1%) 4 (5.3%) 71 (94.7%) 9 (12.7%) 62 (87.3%) Black 16 (20.5%) 62 (79.5%) 17 (27.4%) 45 (72.6%) 1 (1.7%) 58 (98.3%) 17 (29.3%) 41 (70.7%) Caucasian 24 (14.9%) 137 (85.1%) 53 (39.3%) 82 (60.7%) 11 (5.7%) 182 (94.3%) 28 (15.4%) 154 (84.6%) Parity, 0 [0-4] 0 [0-4] 0.243 0.5 [0-4] 0 [0-4] 0.899 1 [0-4] 0 [0-4] 0.138 0 [0-4.55] 0 [0-4] 0.991 median [95% CI]

Previous PTB, 1 [0-5] 1 [0-5] 0.498 1 [0-5.7] 1 [0-5] 0.881 3 [0-5.2] 1 [0-5.8] 0.006 1 [0-4.525] 1 [0-6] 0.238 median [95% CI]

GA at cerc insertion (w), 17 [9-21] 15 [11-22] 0.156 16 [12-23] 14 [11-22] 4.88E-04 18 [11-22] 16 [10-23] 0.299 16 [12-23] 16 [10-22] 0.282 median [95% CI]

CL at cerc insertion (mm), 19 [4-24] 20 [7-25] 0.038 16 [5-25] 22 [14-25] 1.44E-06 15 [5-23] 21 [6-25] 0.086 16 [5-25] 21 [9-25] 0.001 median [95% CI]

Indication for cerclage 0.019 2.90E-08 0.450 0.194 Elective (obstetric history) 16 (10%) 144 (90%) 26 (18.3%) 116 (81.7%) 9 (6.3%) 134 (93.7%) 20 (14.9%) 114 (85.1%) Ultrasound CL ≤25mm 33 (19.8%) 134 (80.2%) 67 (50%) 67 (50%) 8 (4%) 193 (96%) 40 (20.7%) 153 (79.3%) GA = gestational age (weeks), PTB = spontaneous preterm birth <37 weeks, CL = cervical length (mm); cerc = cervical cerclage; B'ham = Birmingham women's hospital; C&W = Chelsea and Westminster Hospital, Cambridge = Cambridge University Hospitals; QCCH =Queen Charlotte's and Chelsea Hospital; SMH = St Mary's Hospital

220 Cervical cerclage, vaginal microbiota and preterm birth prevention

Prospective study results

6.4 Participant demographics

Women with a prior preterm birth attending prematurity surveillance clinic that were sonographically identified to have a short cervix (≤25mm) were randomised to receive a cerclage using either braided Mersilene® (n=25) or monofilament Ethilon® (n=24) suture material. Demographics including maternal age, ethnicity, BMI, gestation and cervical length at cerclage insertion were comparable among suture material groups (Table 6-4). A greater number of women receiving a braided cerclage delivered preterm <37+0 weeks compared to monofilament, although this was not powered to reach significance (32% v 24%, P=0.75; Fishers’ exact; Figure 6-7).

Figure 6-7 A greater proportion of women delivered preterm <37+0 weeks among those receiving a braided (32%) compared to a monofilament cerclage (24%, P=0.75; Fishers’ exact)

221 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-4 Patient characteristics for women randomised to receive monofilament and braided cervical cerclages

Monofilament suture Braided suture Total P value

n / N (%) 24/49 (49%) 25/49 (51%) 49/49

Age, years Mean ±SD (range) 32.8 ± 3.0 (27-39) 33.9 ± 3.8 (25-42) 33.5 ± 3.5 (25-42) 0.58

BMI Mean ±SD (range) 24.1 ± 4.2 (18-35) 26 ± 3.6 (21-36) 25.1 ± 4.5 (18-36) 0.71

Ethnicity, n (%) Caucasian 16 (67%) 11 (44%) 27 (55%) 0.15 Asian 2 (8%) 7 (28%) 9 (18%) 0.07 Black 6 (25%) 7 (28%) 13 (27%) 1.00

Parity (%) P 0 12 (50%) 13 (52%) 25 (51%) 1.00

Smoker (%) 1 (4%) 2 (8%) 3 (6%) 1.00

Cerclage insertion, mean ±SD

GA at insertion (w) 17+6 ± 2.8 18+1 ± 3 17+0 ± 2.9 0.41 CL at insertion (mm) 18 ± 5.1 19 ± 4.5 19 ± 5.3 0.72

GA at delivery, n/N (%) +0 <34 w 4/24 (16%) 0/25 (0%) 4 (8%) 0.05 34+1-36+6 w 2/24 (8%) 8/25 (32%) 10 (20%) 0.07 +0 ≥37 w 18/24 (76%) 17/25 (68%) 35 (71%) 0.75 BMI= body mass index; CL = cervical length (mm); GA = gestational age; w= weeks

222 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.5 Cervical assessment

Vascularity index (VI) 101 was used to assess morphological differences in the cervix according to suture material. Prior to cerclage insertion, VI was comparable among braided and monofilament cerclages (Mean VI 3.6 ±0.36 v 3.4 ±0.34; ns; Figure 6-8). Following cerclage insertion, braided cerclage was strongly associated with increased cervical vascularisation compared to pre-cerclage values (P<0.05; ANOVA repeated measures), as well as in comparison to the monofilament group from 4 weeks post-procedure (P<0.01; K-W test; Figure 6-8, Figure 6-1, Table 6-5,). This difference in VI among suture material groups persisted until 16 weeks post-insertion (P<0.01). In contrast, cervical VI in the monofilament group remained stable with advancing gestation (Table 6-5, Figure 6-1).

223 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-8 Braided cerclage induces premature cervical vascularisation. Cervical vascularisation index (VI), as assessed using transvaginal ultrasound, was greater in patients receiving braided cerclage compared to monofilament cerclages at 4, 8, 12 and 16 weeks post-insertion (**P<0.01, ***P<0.001; Kruskal-Wallis test, #P<0.05; ANOVA multiple comparison to pre cerclage).

Table 6-5 Vascularisation Index as a function of time from cerclage insertion: monofilament versus braided cerclage suture material

Monofilament Braided Time from cerclage Mean VI ±SD Pre-cerclage 3.4 ±0.34 3.6 ±0.36

4w 3.3 ±0.43 7.9 ±1.09** 8w 3 ±0.47 7.1 ±0.79** 12w 2.7 ±0.41 8.5 ±1.08*** 3 ±0.51 8.1 ±0.86** Post cerclage Post cerclage 16w w= weeks from cerclage insertion; VI= Vascularity Index; **P<0.01, ***P<0.001; Kruskal-Wallis test Monofilament v Braided suture.

224 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.6 Impact of suture material on vaginal microbial structure

Women attended a mean of 4.2 research time-points providing a total of 189 samples for characterisation of the vaginal microbiota by 16S rRNA gene sequencing.

6.6.1 Principal component analyses The effect of cerclage on the vaginal microbiome was initially examined using unsupervised principal component multivariate analysis (PCA) to assess for microbial variance at genera taxonomic level according to suture material. Pre-cerclage variance among monofilament and braided groups was minimal (Figure 6-9A). Following cerclage insertion, braided suture material use was associated with larger microbial variance than a monofilament suture material (Figure 6-9B)

Figure 6-9 Principal component analysis comparing variance in microbial profiles among women receiving monofilament and braided cerclages. (A) Pre- cerclage insertion microbial profiles did not differ substantially. (B) Following cerclage insertion, those with a braided cerclage exhibited greater microbial variance than monofilament cerclages.

225 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.7 Hierarchical clustering of bacterial class, genera and species

Heatmaps were then generated to assess microbial diversity according to suture material and ‘time’ from cerclage insertion: pre cerclage, and 4 weeks, 8 weeks, 12 weeks and 16 weeks post-cerclage insertion.

6.7.1 Bacterial class Heatmaps plotting the relative abundance of bacteria at class taxonomic level demonstrated separate clustering of braided suture from monofilament cerclage. A notable increase in microbial diversity was observed in the braided suture compared to the monofilament group from 4 weeks following cerclage insertion, persisting to 16 weeks (Figure 6-10).

Figure 6-10 Heatmap of hierarchical clustering analysis using ward linkage of microbial class data comparing braided to monofilament cerclage sample, according to time from cerclage insertion: Pre-cerclage insertion, and 4, 8, 12, and 16 weeks post insertion. Braided suture clustered separately from monofilament cerclage and induced greater bacterial diversity from 4 weeks following cerclage insertion.

226 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.7.2 Bacterial genera Genera level Ward linkage hierarchical clustering analysis classified individual samples according to relative abundance at bacteria genera into normal (>90% Lactobacillus spp.), intermediate (30-90% Lactobacillus spp.) or dysbiotic (<30% Lactobacillus spp.; Figure 6-11).

Figure 6-11 All samples (n=189) collected from women receiving a monofilament (n=24 women) and braided (n=25 women) cerclage were classified using ward- linkage analysis of vaginal bacterial genera into three groups; normal (>90% Lactobacillus spp. abundance), intermediate (30-90% Lactobacillus spp. abundance) or dysbiotic (<30% Lactobacillus spp. abundance)

227 Cervical cerclage, vaginal microbiota and preterm birth prevention

Prior to cerclage insertion, monofilament and braided groups consisted of similar numbers of women harbouring either intermediate or dysbiotic vaginal profiles (monofilament 16.7% v braided 17.3%, P=0.7, Figure 6-12, Table 6-6). Insertion of braided suture caused a significant shift in the number of women classified as ‘dysbiotic or intermediate’, from 17.3% pre-cerclage to 60% at 4 weeks post cerclage. This trend towards a higher proportion of women with intermediate/dysbiosis persisted throughout sampling at 8, 12 and 16 weeks post-cerclage insertion (45%, 47%, and 50% respectively; P<0.01, ANOVA, Figure 6-12).

In contrast, the insertion of a monofilament cerclage had minimal impact on the diversity of the vaginal microbial communities at genera level; the proportion of women classified as having a ‘normal’ Lactobacillus spp dominant microbiome increased from 83% pre-cerclage, to 85% at 4 weeks post cerclage, peaking at 94% at 8 weeks after cerclage insertion (Figure 6-12, Table 6-6).

This reflected a 5-fold increase in microbial dysbiosis with insertion of a braided suture at 4 weeks compared to a 0.7-fold change with a monofilament cerclage. Compared to pre-cerclage levels, respective fold change in dysbiosis at 8, 12 and 16 weeks post-cerclage, was 4, 7 and 6 for braided suture, and 0.3, 0.7 and 0.7 respectively for monofilament (Figure 6-12). The negative fold change observed in the monofilament group was associated with increasing Lactobacillus spp. dominance (Figure 6-13); a trend consistent with observations expected in a healthy pregnant population. In contrast, insertion of a braided suture was characterized by a depletion in Lactobacillus spp. dominance (Figure 6-13), and increased levels of bacteria associated with bacterial vaginosis (BV) including Prevotella (P=0.02), Finegoldia (P=0.02), and Dialister (P=0.04) species (Figure 6-14). There was no significant increase in Gardnerella or Atopobium, two genera of bacteria specifically associated with BV. Quantitative PCR investigated this further as reported in Section 6.7.3b.

228 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-12 Braided suture material for cervical cerclage is associated with vaginal microbiome dysbiosis. Braided cerclage was associated with a 5-fold increase in microbial dysbiosis within 4 weeks of insertion that persistent until at least 16 weeks while no change was observed in women receiving a monofilament cerclage (#P<0.01; Bonferroni corrected two sided Welch’s t-test: monofilament vs braided at comparative time points, **P<0.05; chi-squared: pre vs post cerclage, ns=non-significant)

Figure 6-13 Bar charts representing relative abundance of Lactobacillus (genera) within individual samples compare suture material and time from insertion. Monofilament cerclage demonstrated maintenance of high Lactobacillus abundance following insertion of monofilament cerclage (P=0.9; ANOVA). In contrast, proportions of Lactobacillus in samples collected following braided cerclage, declined at 4 weeks and persisted until 16 weeks (P=0.043; ANOVA).

229 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-6 Bacterial genera classification according to time from cerclage based on suture material using ward-linkage hierarchical clustering: Normal: >90% Lactobacillus abundance, Intermediate: 30-90% Lactobacillus abundance, Dysbiotic: <30% Lactobacillus abundance

Microbiome classification of bacterial genera* Normal Intermediate Dysbiotic

Monofilament 83.3% 4.2% 12.5% Pre-cerclage Braided 82.6% 13.0% 4.3% TOTAL 83.0% 8.5% 8.5% P. value 1.000 0.348 0.608 Post cerclage: Monofilament 85.0% 0.0% 15.0% 4w Braided 40.0% 35.0% 25.0% TOTAL 62.5% 17.5% 20.0% P. value 0.008 0.008 0.695 Monofilament 94.1% 0.0% 5.9% 8w Braided 55.0% 25.0% 20.0% TOTAL 73.0% 13.5% 13.5% P. value 0.010 0.049 0.348 Monofilament 88.2% 0.0% 11.8% 12w Braided 52.6% 10.5% 36.8% TOTAL 69.4% 5.6% 25.0% P. value 0.031 0.489 0.128 Monofilament 84.6% 0.0% 15.4% 16w Braided 50.0% 12.5% 37.5% TOTAL 65.5% 6.9% 27.6% P. value 0.114 0.487 0.238 *Based on Hierarchical clustering Analysis. P value = Fishers exact Normal: >90% Lactobacillus abundance, Intermediate: 30-90% Lactobacillus abundance, Dysbiotic: <30% Lactobacillus abundance w= weeks from cerclage insertion

230 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-14 Bar plots of relative abundance of Prevotella (A), Finegoldia (B) and Dilaster (C) at genera taxonomic level in individual women taken pre- and post- cerclage, comparing braided to monofilament cerclage. Prior to cerclage insertion, the vaginal microbiome was dominated by Lactobacillus (genera) (Figure 6-13), and levels of Prevotella (P=0.7), Finegoldia (P=0.9) and Dilaster (P=0.3) were comparable among suture material groups. (A) Mean sequence proportions of Prevotella spp increased significantly in braided compared to monofilament cerclages at 4 weeks (P=0.02), 8 weeks (P=0.03), 12 weeks (P=0.01) and 16 weeks (P=0.01). (B) Finegoldia spp. abundance increased in braided compared to monofilament at 4 weeks post-cerclage, (P=0.02) persisting at 8 weeks (P=0.02), 12 weeks (P=0.05) and 16 weeks (P=0.03). (C) Dilaster spp. abundance was also associated with a significant increase in braided compared to monofilament at 4 weeks (P=0.04), 8 weeks (P=0.04), 12 weeks (P=0.03) and 16 weeks (P=0.017) post insertion. (Welch’s t-test; mean sequence proportions monofilament v braided). Horizontal lines indicate mean relative abundance in those patients receiving braided (yellow) or monofilament (blue) suture.

231 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.7.3 Bacterial species Hierarchical clustering of species data assigned samples into community state types (CSTs) 155: CST I (L. crispatus, n=59, 31%), CST II (L. gasseri n=19, 10%), CST III (L. iners n=43, 23%), CST IV (diverse species n=48, 25%), and CST V (L. jensenii n=20, 11%) (Figure 6-15, Figure 6-16A).

Braided cerclage increased the proportion of women with CST IV, characterised by reduced levels of Lactobacillus spp. and increased diversity of anaerobic bacteria, from 13% pre- cerclage, to 45% at 4 weeks and 50% at 16 weeks post-cerclage (P=0.02). This microbial disruption was not observed in women receiving a monofilament cerclage, who instead demonstrated maintenance of high Lactobacillus spp. abundance (CSTs I, II, III and V) and stability throughout longitudinal sampling (P=0.9; Figure 6-16A, Table 6-7).

To identify degrees of dysbiosis that were not identified by CST classification, an alternate species level classification was performed, as described by Borgdorff et al 159. Samples were classified into Normal Lactobacillus spp. (n=73, 39%), L. iners (n=30, 16%), intermediate dysbiosis (n=57, 30%) and severe dysbiosis (n=29, 15%; Figure 6-16B). Although L. iners has previously been observed as an intermediary towards dysbiosis, there was no significant change in L. iners abundance in association with insertion of a monofilament or braided suture material (Figure 6-16B, Table 6-8). There was a trend to increased proportions of intermediate and severe dysbiosis among those receiving a braided compared to monofilament cerclage, however this did not reach significance (Figure 6-16B, Table 6-8).

232 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-15 Ward hierarchical clustering analysis of species sequence data. Five major clusters were identified corresponding to community state types (CSTs): CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), CST IV (Lactobacillus spp.-depleted, diverse species), and CST V (L. jensenii). Vaginal bacterial communities classified into CST IV could be further divided into those characteristically enriched with or without Atopobium vaginae, Streptococcus agalactiae and Shuttleworthia satelles.

233 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-16Longitudinal assessment of vaginal bacterial community structure following suture insertion. (A) Longitudinal distribution of Community State Types I, II, III, IV, and V in women receiving monofilament or braided cerclage. Braided cerclage is associated with a shift from Lactobacillus spp. dominance (CST I, II, III, and V) towards CST IV after cerclage insertion. Monofilament maintained microbial stability with CST I and III dominance pre- and post- cerclage insertion. (B) Longitudinal distribution of alternative classification of species taxonomic data into normal, L. iners dominant, intermediate dysbiosis, and severe dysbiosis. Braided cerclage is associated with a shift towards intermediate and dysbiotic microbiomes at the species level; however, L. iners dominance is not influenced by insertion of a monofilament or braided cerclage.

234 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-7 Bacterial species classification into community state types according to time from cerclage: braided versus monofilament suture material Community state type* CST I CST II CST III CST IV CST V Diverse Lactobacillus spp. L. crispatus L. gasseri L. iners L. jensenii species Pre-cerclage Monofilament 29% 0% 38% 17% 17% Braided 30% 22% 26% 13% 9% TOTAL 30% 11% 32% 15% 13% P value 1.000 0.023 0.534 0.075 0.662 Post cerclage: Monofilament 40% 5% 25% 15% 15% 4w Braided 25% 15% 15% 45% 0%

TOTAL 33% 10% 20% 30% 8% P value 0.500 0.605 0.690 0.176 0.487 8w Monofilament 41% 6% 29% 12% 12% Braided 20% 10% 20% 50% 0% TOTAL 30% 8% 24% 32% 5% P value 0.279 1.000 0.703 0.017 0.204 12w Monofilament 41% 6% 24% 12% 18% Braided 26% 11% 16% 47% 0% TOTAL 33% 8% 19% 19% 19% P value 0.483 1.000 0.684 0.031 0.095 16w Monofilament 46% 8% 15% 15% 15% Braided 19% 19% 13% 50% 0% TOTAL 31% 14% 14% 34% 7% P value 0.226 0.606 1.000 0.114 0.192 TOTAL POPULATION 31% 10% 23% 25% 11% *Based on Hierarchical clustering Analysis. P value = Fishers exact w= weeks from cerclage insertion

235 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-8 Species classification into normal, L. iners dominant, intermediate and severe dysbiosis according to time from cerclage. 1 2 3 4 Microbial classification* Intermediate Severe Normal L. iners dysbiosis dysbiosis Monofilament 42% 21% 33% 4% Pre-cerclage Braided 52% 17% 22% 9% TOTAL 47% 19% 28% 6% P value 0.564 1.000 0.517 0.601 Monofilament 45% 20% 20% 15% Post-cerclage: Braided 4 w 25% 15% 40% 20% TOTAL 35% 17% 30% 18% P value 0.320 1.000 0.300 0.700 8 w Monofilament 53% 18% 24% 6% Braided 20% 20% 50% 10% TOTAL 35% 19% 38% 8% P value 1.000 1.000 0.170 1.000 12 w Monofilament 47% 18% 24% 12% Braided 26% 11% 32% 32% TOTAL 36% 14% 28% 22% P value 0.299 0.650 0.717 0.232 16 w Monofilament 54% 0% 31% 15% Braided 25% 13% 25% 38% TOTAL 38% 7% 28% 28% P value 0.143 1.000 1.000 0.220 TOTAL POPULATION 39% 16% 30% 15% *Based on hierarchical clustering analysis into normal, L. iners dominant, intermediate dysbiosis, and dysbiotic. P value = Fisher’s exact test for monofilament v braided. w = weeks from cerclage insertion

236 Cervical cerclage, vaginal microbiota and preterm birth prevention

To identify bacteria specifically associated with braided suture insertion, linear discriminant analysis with effect size (LEfSe) 202 analysis was performed pre-cerclage and at 4 weeks post cerclage. Braided cerclage led to enriched numbers of Gram-negative bacteria at 4 weeks (Figure 6-17A, B), while monofilament increased in relative abundance of Bacilli (Figure 6-17).

Figure 6-17 Bacterial taxonomic groups that discriminate between monofilament and braided cerclage. (A) Significant, differentially abundant microbial clades and nodes according to suture material four weeks post-insertion were identified using LEfSe analysis and presented as a cladogram. (B) Linear Discriminant Analysis (LDA) was then used to estimate the effect size of each differentially abundant species. The vaginal microbiome of patients receiving a monofilament cerclage was enriched with bacilli whereas those receiving a braided cerclage were comparatively enriched in Bacteroides spp., Actinobacteria and Clostridia

Consistent with these observations, indices of bacterial community richness (Figure 6-18A, Table 6-9) and alpha-diversity (Figure 6-18B) were increased in samples collected following braided suture compared to monofilament with the greatest differences observed at 16 weeks post-cerclage (P<0.05; Kruskal-Wallis and Dunn’s multiple comparisons).

237 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-18 (A) Comparison of total number of bacterial species observed reveal a significant increase post-braided cerclage compared to a monofilament cerclage. (B) Alpha diversity was significantly increased at 16 weeks post braided cerclage insertion compared to monofilament (*P<0.05; Kruskal-Wallis, Dunn’s multiple comparisons post-hoc monofilament vs. braided, #P<0.05; Bonferroni multiple comparison to pre cerclage sample).

Table 6-9 Species observed and alpha diversity according to time from cerclage: Braided versus Monofilament suture material Monofilament Braided Mean ±SD Species observed Pre-cerclage 8.1 ±2.01 9.1 ±1.68 4w 8.4 ±1.51 18.9 ±2.65 8w 6.8 ±1.51 19.2 ±3.33 12w 7.4 ±1.77 20.4 ±3.78 16w 5.8 ±1.25 27.2 ±6.05 Shannon Alpha Pre-cerclage 1.5 ±0.13 1.4 ±0.13 Diversity Index 4w 1.3 ±0.09 2 ±0.2 8w 1.2 ±0.07 2.4 ±0.41 12w 1.3 ±0.08 2.6 ±0.6 16w 1.3 ±0.11 2.547 ±0.49 w= weeks from cerclage insertion

238 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.7.3a Species-specific changes associated with suture material use Given the relative pathogenicity of L. iners and beneficial effects of L. crispatus (as described in Chapters 2 and 3), mean sequence proportions were compared among suture groups as a function of time from cerclage. Braided cerclage was associated with a reduction in proportions of L. crispatus (Figure 6-19A) and L. iners (Figure 6-19B), relative to monofilament suture material, however neither change reached significance.

Figure 6-19 Box plots reveal no significant difference in L. crispatus (A) and L. iners (B) when comparing mean proportions of sequences in braided and monofilament cerclages, pre cerclage, and at 4 weeks, 8 weeks, and 12 weeks post cerclage (P value = Welch’s t test braided v monofilament)

239 Cervical cerclage, vaginal microbiota and preterm birth prevention

Other bacteria clinically associated with preterm birth were also considered. Observed proportions of S. agalactiae (Group B streptococcus) did not differ substantially with insertion of cerclage material when braided and monofilament groups were compared (difference in mean sequences 0.06, P=0.3). E. coli was not detected but this may reflect a limitation of the primer set used for 16S rRNA sequencing 149.

6.7.3a Atopobium vaginae and Gardnerella vaginalis Atopobium vaginae and Gardnerella vaginalis are specific organisms known to associate with bacterial vaginosis (BV). Braided cerclage induced a trend towards increasing A. vaginae and G. vaginalis sequence reads at 4 weeks post-cerclage compared to pre-cerclage levels (Figure 6-20A, C). Compared to monofilament cerclage, this represented a greater mean fold change in A. vaginae and G. vaginalis sequences, although neither were significant (Figure 6-20B, D). This may be due to the V1-V3 primers (used for 16s rRNA sequencing) under-representing proportions of A. vaginae and G. vaginalis. Quantitative PCR was therefore performed on the samples of sequenced bacterial DNA to further investigate this (Table 6-10; Figure 6-21). Compared to pre-cerclage levels, insertion of braided suture associated with a trend increase in mean PCR copy numbers for both A. vaginae (P=0.07, Figure 6-21A) and G. vaginalis (P=0.05; Figure 6-21C, Table 6-10). In contrast, no change in levels of G. vaginalis or A. vaginae were detected following monofilament cerclage.

240 Cervical cerclage, vaginal microbiota and preterm birth prevention

16s rRNA gene sequences

ABA. vaginae A. vaginae

0.2856P=0.3 0.7236P=0.7 0.5324 P=0.5 800 25

20 600

15 400 10

200 Sequence reads, n Sequence reads, n reads, Sequence 5

0 0 Pre Post Pre Post Monofilament Braided Cerclage insertion

CDG. vaginalis G. vaginalis

0.3287P=0.3 0.7712P=0.8 0.5996 P=0.6 15 1.5

1.0 10

0.5

5

Sequence reads, n 0.0 Sequence reads, n

0 -0.5 Pre Post Pre Post Monofilament Braided Cerclage insertion

Monofilament suture Braided suture

Figure 6-20 16s rRNA gene sequence reads for A. vaginae and G. vaginalis. Compared to pre-cerclage levels, insertion of a monofilament or braided suture did not significantly increase proportions of A. vaginae (A) or G. vaginalis (C). Mean fold change in sequence proportions was comparatively, but no significantly greater in braided compared to monofilament groups for A. vaginae (B; P=0.5) and G. vaginalis (D; P=0.6; t-test).

241 Cervical cerclage, vaginal microbiota and preterm birth prevention

Quantitative PCR

ABA. vaginae A. vaginae P=0.81 0.4194P=0.42 P0.0695=0.07 0.8027 40000000 50000

40000 30000000

30000 20000000 20000

10000000 Mean Fold change, Fold Mean PCR copy numbers PCR copy numbers 10000

0 0 Pre Post Pre Post Monofilament Braided Cerclage insertion

CDG. vaginalis G. vaginalis

0.5802P=0.58 P0.0511=0.05 0.4055P=0.41 1.010 08 30000

8.010 07

20000 6.010 07

4.010 07 10000 Mean Fold change, Fold Mean PCR copy numbers 2.010 07 PCR copy numbers

0.0 0 Pre Post Pre Post Monofilament Braided Cerclage insertion

Monofilament suture Braided suture Figure 6-21 Quantitative PCR demonstrated increasing A. vaginae and G. vaginalis in braided but not monofilament suture at 4 weeks post cerclage. (A) Compared to pre-cerclage levels, insertion of braided suture associated with an increase in mean copy numbers of A. vaginae (594,326 pre-cerclage v 5,081,000 post-cerclage; P=0.07) while monofilament did not (1,227,000 pre-cerclage v 320,477 post- cerclage; P=0.4; paired t –test). This reflected in a greater fold change of bacterial copy numbers with insertion of braided suture (mean 25,903) than monofilament (mean 113, P=0.8; t-test B). (C) G. vaginalis also increased with braided (961,805 pre-cerclage v 10,170,000 post-cerclage; P=0.05) but not monofilament cerclage (1,445,000 pre-cerclage v 2,989,000 post-cerclage; P=0.58; paired t -test). (D) This is reflected in a mean fold change of bacterial copy numbers of 12861 for braided compared to 266 monofilament respectively (P=0.4; t-test). (PCR= polymerase chain reaction)

242 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-10 Mean bacterial counts of Atopobium vaginae and Gardnerella vaginalis pre- and immediately post-cerclage insertion, as quantified by PCR and 16s rRNA gene sequence.

Mean bacterial count Fold change in read count Pre-cerclage Post-cerclage *P #P Mean ±SD Mean ±SD Mean ±SD value value Atopobium vaginae

PCR copy numbers Monofilament 1,227,000 5,084,000 320,477 1,104,000 0.419 113 356 0.803 Braided 594,326 2,630,000 5,081,000 10,690,000 0.070 25,903 96,387

16s rRNA gene Monofilament 38 155 33 152 0.286 0 1 0.532 sequence Braided 22 102 13 38 0.724 12 38 Gardnerella vaginalis

PCR copy numbers Monofilament 1,445,000 4,086,000 2,989,000 12,180,000 0.580 266 829 0.406 Braided 961,805 1,983,000 10,170,000 21,130,000 0.051 12,861 57,101

16s rRNA gene Monofilament 0.2 0.9 0.1 0.4 0.329 0.0 0.1 0.600 sequence Braided 0.4 1.9 0.6 2.9 0.771 0.6 2.9 *P value = paired t-test, pre v post cerclage; #P value = Mann Whitney, fold change monofilament v braided

243 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.7.3b Correlation between qPCR and dysbiosis qPCR copy numbers for both A. vaginae and G. vaginalis correlated with classification of an intermediate or dysbiotic vaginal microbiome as determined by 16s rRNA gene sequence analysis of genera data (Figure 6-22). This indicated that although V1-V3 primers may underestimate abundance of A. vaginae and G. vaginalis, these species do play a contributory role in cerclage-induced vaginal dysbiosis.

16s rRNA gene sequence microbiome classification Normal Intermediate Dysbiotic

*** (K-W)*** (K-W) 1.010 09 *** * **

1.010 08

1.010 07

1.010 06

1.010 05

1.010 04

1.010 03 qPCR copy numbers (LOG) qPCR

1.010 02

1.010 01

1.010 00 A. vaginae G. vaginalis

Figure 6-22 Atopobium vaginae and Gardnerella vaginalis qPCR copy numbers grouped according corresponding to microbial classification into Normal, Intermediate, Dysbiotic (as determined by HCA clustering of genera sequence data). A normal microbiome correlated with lower copy numbers of both A. vaginae and G. vaginalis, and an intermediate/dysbiotic microbiome correlated relatively greater qPCR copy numbers (***P<0.001, **P<0.01, *P<0.05: Kruskal-Wallis test, Dunn's multiple comparisons)

244 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.8 Expression of inflammatory cytokines

6.8.1 Detection of cytokines in the cervicovaginal fluid The inflammatory response to cerclage insertion was investigated by using a multiplex immunoassay to assess expression of 15 inflammatory analytes in cervicovaginal fluid (CVF) quantified in vaginal samples taken in conjunction with microbial sampling at pre- and 4 weeks post-cerclage insertion. The analytes were selected according to evidence of involvement in inflammatory change related to preterm birth, cervical ripening, and angiogenesis (IL-1β, IL-2, IL-4, IL-6, IL-8, IL-10, G-CSF, GM-CSF, INF-γ, MCP-1, MMP-1, and TNFα, VEGF, ICAM-1). Detectable levels of IL-1β, IL-2, IL-4, IL-6, IL-8, and VEGF were within the standard ranges as determined by the Luminex immunoassay kit (Figure 6-23, Table 6-11). Concentrations of the remaining analytes were calculated from extrapolating the immunofluorescence readings off the Luminex standard curves (Figure 6-24, Table 6-11).

245 Cervical cerclage, vaginal microbiota and preterm birth prevention

Standard range Indiv idual res ults

ABCDEIL-1 IL-2 IL-8 VEGF IL-6 5000 40000 3000 3500 8000 3000 2500 2000 4000 6000 1500 30000 1000 4000 2000 500 3000 2000 100 20000 150 2000 100 1000 10000 50 1000 50

0 0 0 0 0

FGHIG-CSF ICAM-1 MMP-1 TNF- JRANTES 10000 2000000 8000 4000 6000 8000 1500000 6000 3000 6000 4000 4000 2000 4000 1000000 2000 2000 2000 1000 500000 100 20000 100 50 100 80 15000 80 40 80 60 60 30 60 10000 40 40 20 40 20 5000 20 10 20 0 0 0 0 0

KLMNOIL-4 IL-10 IFN-gamma GM-CSF MCP-1 5000 10000 2000 8000 8000 4000 6000 3000 8000 1500 6000 2000 6000 4000 1000 1000 4000 4000 2000 50 2000 500 2000 200 40 100 10 10 150 30 8 8 100 6 6 20 50 4 4 50 10 2 2 0 0 0 0 0 Figure 6-23 Detectable expression of inflammatory mediators in the CVF of vaginal swabs using Luminex multiplex immunoassay. Detectable levels of (A) Interleukin (IL)-1β, (B) IL-2 (D) IL-8, (C) VEGF and (K) IL-4 were within the standard ranges, while concentrations of the remaining analytes were calculated from extrapolating the immunofluorescence readings off the standard curve. (IL-6, IL-10, IL-β; ICAM (Intercellular Adhesion Molecule), IFN (Interferon)-gamma, G-CSF (Granulocyte-colony stimulating factor), GM-CSF (Granulocyte-macrophage colony- stimulating factor), MMP1 (Matrix metalloproteinase)-1, MCP-1 (monocyte chemotactic protein), RANTES (Regulated on Activation, Normal T Expressed and Secreted), TNF-α (Tumour necrosis factor) and VEGF (Vascular endothelial growth factor))

246 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-11 Comparison of analyte detection versus standard range detected by Luminex kit Standard Range Analyte results ANAYLYTE Min Max Min IQR 1 Median IQR 3 Max TNF α 6 3,410 0.6 3.0 3.2 3.8 47.9 IL-8 8 42,881 0 3079 6661 22166 53372 VEGF 12 2,899 60 788 1186 1459 2930 IL-4 5 4,173 3.9 8.3 9.5 10.1 18.3 IL-6 10 3,403 1.2 3.5 12.2 55.9 7345 ICAM-1 7,812 1,895,099 62 180 1161 3187 167276 MMP-1 29 6,883 4.3 5.3 14.3 22.6 285 MCP-1 20 7,744 3.7 14.7 19.5 23 171 IL-1β 17 4,083 1.0 57.5 521.3 1662 4818 IL-2 32 8,072 23.0 42.3 50.2 59 112 G-CSF 34 8,224 3.7 5.3 21.1 136 18965 IL-10 13 3,288 1.1 1.2 1.4 1.4 4.3 IFN-γ 7 1,756 1.0 1.4 1.5 2.7 5.9 GM-CSF 32 7,836 3.8 3.8 4.2 4.6 9.8 RANTES/CCL15 21 5,228 3.0 5.1 5.5 6.4 48.1 IL= interleukin; ICAM= Intercellular Adhesion Molecule; IFN=Interferon; G-CSF= Granulocyte-colony stimulating factor; GM-CSF= Granulocyte-macrophage colony-stimulating factor; MMP= Matrix metalloproteinase; MCP/CCL= monocyte chemotactic protein/chemokine ligand; RANTES= Regulated on Activation, Normal T Expressed and Secreted; TNF= Tumor necrosis factor; VEGF= Vascular endothelial growth factor.

9000 TNF alpha IL-8 IL-10 8000 VEGF IFN-gamma IL-4 GM-CSF IL-6 MMP-1 7000 MCP-1 IL-1beta RANTES 6000 IL-2 G-CSF 5000

4000

3000

Concentration(ng/dl) 2000

1000

0 123456 Standards

Figure 6-24 Standard curves generated from Luminex multiplex assay samples of standard concentrations for the following analytes: IL (interleukin)-2, IL-4, IL-6, IL- 8, IL-10, IL-β; ICAM (Intercellular Adhesion Molecule), IFN (Interferon)-gamma, G- CSF (Granulocyte-colony stimulating factor), GM-CSF (Granulocyte-macrophage colony-stimulating factor), MMP1 (Matrix metalloproteinase)-1, MCP-1 (monocyte chemotactic protein), RANTES (Regulated on Activation, Normal T Expressed and Secreted), TNF-α (Tumour necrosis factor) and VEGF (Vascular endothelial growth factor).

247 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.8.2 Impact of cerclage insertion on cytokine expression Expression of IL-1β, a key inflammatory mediator in preterm parturition 19, increased significantly in the CVF of women receiving a braided cerclage compared to pre-cerclage levels (P<0.001; Wilcoxon signed rank test), and relative to monofilament cerclage (P<0.001; Mann Whitney, Figure 6-25A).

Expression of the pro-inflammatory cytokines IL-6, IL-8 and TNFα, associated with preterm cervical ripening 252, as well as MMP-1, a matrix metalloproteinase central to cervical collagenous breakdown and remodelling 215, were all increased following insertion of braided, but not monofilament cerclage (P<0.05; Wilcoxon signed rank test; Figure 6-25). Levels of anti- inflammatory cytokine IL-2 and IL-4, did not follow this trend, and remained uninfluenced by suture material (Figure 6-25). Furthermore, VEGF (a promoter of cervical angiogenesis), although detectable and within the standard range, was not influenced by cerclage insertion irrespective of suture material (Figure 6-26A).

Fold change in cytokine expression with cerclage insertion was observed most significantly with braided cerclage, but not monofilament (Figure 6-26B, Table 6-12). Braided suture was associated with a significant increase in pro-inflammatory cytokines IL-1β, IL-6, IL-8, IL-10, TNF- α, MCP-1, MMP-1, IFN-γ, and RANTES (P<0.05) but not IL-2, IL-4, G-CSF or VEGF (Figure 6-26B)

248 Cervical cerclage, vaginal microbiota and preterm birth prevention

A Analytes within standard range

IL-2 IL-4 IL-6 ns IL-8 ns ns ns ns * ns 150 20 1200 150000 ***

1000 15 100 800 100000

10 600

50 400 50000 5 200

0 0 0 0 Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post

VEGF ns ns 4000

3000

2000

1000

0 Pre Post Pre Post

B Analytes extrapolated from standard curve

G-CSF GM-CSF IFN- ICAM ns ns ns ns 20000 * 12 * 8 *** 200000 *

10 15000 6 150000

8 10000 4 100000 6

5000 2 50000 4

0 2 0 0 Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post

IL-1 IL-10 MCP-1 MMP-1 ns ns ns 15000 *** 5 * 200 ** * 300 **

4 150 10000 200 3 100 2 5000 100 50 1

0 0 0 0 Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post Pre Post

RANTES TNF- ns 60 * * 50 *

40

40 30

20 20

10

0 0 Pre Post Pre Post Pre Post Pre Post

249 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-25 Relative to pre-cerclage levels, braided suture was associated with a significant increase in pro-inflammatory cytokines at 4 weeks post-insertion, while monofilament cerclage insertion was not. (A) Expression of analytes detectable within the standard range of Luminex multiplex assay kit. (B) Expression of analytes, calculated from extrapolation of the standard concentration curves (*P<0.05, **P<0.01, ***P<0.001; Wilcoxon signed rank test).

250 Cervical cerclage, vaginal microbiota and preterm birth prevention

A Fold change in analyte expression, pre- and 4 weeks post-cerclage. Monofilament v Braided

TNF IL-8 IL-4 VEGF IL-6 ** ns ns * 6 * 15000 1.0 300 800

10000 0.5 600 4 200 5000 0.0 400 2 0 -0.5 200 100 0 -5000 -1.0 0 Difference in analyte

-2 -10000 -1.5 0 -200

ICAM-1 MMP-1 MCP-1 IL-1 IL-2 ** * *** 20000 60 10 4000 *** 6 ns

40 3000 4 0 10000 20 2000 2 -10

0 1000 0 0 -20 -20 0 Difference in analyte -2

-10000 -40 -30 -1000 -4

G-CSF IL-10 IFN-gamma GM-CSF RANTES 2500 ns 0.6 ** 1.0 ** 1.5 * 1.5 **

1.0 2000 0.4 0.5 1.0 0.5 1500 0.2 0.0 1000 0.0 0.5 -0.5 0.0 500 Difference in analyte Difference in analyte -1.0

0 -0.2 -0.5 0.0 -1.5

B Braided suture IL-1 Monofilament suture

IFN- IL-8 0.001 MMP-1 MCP-1 nge, P value P nge, 0.01 GM-CSF RANTES ICAM-1 IL-6 MCP-1 G-CSF RANTES IL-10 TNF- 0.05

IL-10 VEGF ICAM-1 IL-8 IL-2

Significance of fold cha IL-1 IFN- IL-4 VEGF G-CSF TNF- IL-2 MMP-1 GM-CSF IL-6 IL-4

10 -1 10 1 10 2 decrease increase

Fold change in analyte (LOG10)

251 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-26 (A) Mean fold change in cytokine expression detected pre and 4 weeks post-cerclage comparing monofilament and braided suture. Braided suture was associated with a significant increase in pro-inflammatory cytokines IL-1β, IL-6, IL-8, IL-10, TNF-α, MCP-1, MMP-1, IFN-γ, and RANTES but not IL-2, IL-4, G-CSF or VEGF (*P<0.05, **P<0.01, ***P<0.001; Mann Whitney for fold change in cytokine concentration Monofilament v Braided). (B) The most significant fold change in cytokine expression, pre and 4 weeks-post cerclage insertion was observed among braided cerclage, while monofilament did not induce a significant fold increase in cytokine expression. (P value= Mann Whitney. IL (interleukin)-2, IL-4, IL-6, IL-8, IL- 10, IL-β; ICAM (Intercellular Adhesion Molecule), IFN (Interferon)-gamma, G-CSF (Granulocyte-colony stimulating factor), GM-CSF (Granulocyte-macrophage colony- stimulating factor), MMP1 (Matrix metalloproteinase)-1, MCP-1 (monocyte chemotactic protein), RANTES (Regulated on Activation, Normal T Expressed and Secreted), TNF-α (Tumour necrosis factor) and VEGF (Vascular endothelial growth factor)

252 Cervical cerclage, vaginal microbiota and preterm birth prevention

Table 6-12 Mean fold change in analyte concentration detectable in cervicovaginal fluid, pre and post cerclage: Monofilament versus braided suture material

Mean fold change in analyte expression pre and post cerclage Monofilament Braided Analyte (pg/ml) Fold change P value Fold change P value TNF-α 0.9 0.687 1.5 0.029 IL-8 2.4 0.266 5.5 0.001 VEGF 1.7 0.578 1.2 0.218 IL-4 1.0 0.977 1.1 0.300 Il-6 2.3 0.754 31.6 0.018 ICAM-1 1.3 0.224 10.8 0.018 MMP-1 1.5 0.586 4.0 0.002 MCP-1/CCL2 0.7 0.003 1.4 0.026 IL-1β 2.9 0.281 14.2 0.000 IL-2 1.0 0.384 1.1 0.670 G-CSF 2.4 0.644 49.7 0.027 IL-10 1.0 0.085 1.2 0.022 IFN-γ 1.0 0.477 1.4 0.001 GM-CSF 1.0 1.000 1.2 0.014 RANTES/CCL5 0.9 0.046 1.1 0.017 IL= interleukin; ICAM= Intercellular Adhesion Molecule; IFN=Interferon; G-CSF= Granulocyte- colony stimulating factor; GM-CSF= Granulocyte-macrophage colony-stimulating factor; MMP= Matrix metalloproteinase; MCP/CCL= monocyte chemotactic protein/chemokine ligand; RANTES= Regulated on Activation, Normal T Expressed and Secreted; TNF= Tumor necrosis factor; VEGF= Vascular endothelial growth factor.

253 Cervical cerclage, vaginal microbiota and preterm birth prevention

6.9 Correlation between cervical vascularity, vaginal microbiota and cytokine expression.

Mean cytokine profiles were grouped according to corresponding microbial sequence classification (based on ward HCA of bacterial genera into Normal (>90% Lactobacillus abundance), Intermediate (30-90% Lactobacillus abundance), and Dysbiotic <30% Lactobacillus abundance).

Vaginal dysbiosis was associated with a significant increase in ICAM-1 (P<0.01), IL-1β (P<0.001), MMP-1 (P<0.01), MCP-1 (P<0.01), TNF-α (P<0.01), GM-CSF (P<0.01) and IFN-γ (P<0.05) and a non-significant increases in IL-6 (P=0.7) and G-CSF (P=0.06; Figure 6-27). Anti- inflammatory cytokines, IL-2 and IL-4, were not elevated in the presence of vaginal dysbiosis, while IL-10, also anti-inflammatory, was increased with dysbiosis (P<0.05, Mann Whitney; Figure 6-27)

254 Cervical cerclage, vaginal microbiota and preterm birth prevention

Figure 6-27 Mean cytokine profiles when grouped by corresponding genera microbial classification (Normal, Intermediate and Dysbiotic). Dysbiosis was associated with increased pro-inflammatory cytokines ICAM-1, IL-1β, IL-6, MMP-1, MCP-1, TNF-α, GM-CSF, IFN--γ and anti-inflammatory IL-10 expression but not IL-2 or IL-4. (***P<0.001, **P<0.01, *P<0.05: Mann Whitney for Normal vs Dysbiotic)

255 Cervical cerclage, vaginal microbiota and preterm birth prevention

Comparison of vaginal bacterial communities with cervical vascularity revealed a strong relationship between vaginal dysbiosis and increased vascularity however, this was limited to those patients receiving a braided suture. When VI was plotted against the number of species observed and alpha diversity, there was a significant positive correlation among women with a braided cerclage between increasing number of species corresponding cervical vascularity (VI: R2=0.09, P=0.002) (Figure 6-28A), as well as alpha diversity (VI: R2=0.14, P=0.001) (Figure 6-28A). VI among monofilament cerclages did not correlate with species numbers or alpha diversity increase (Figure 6-28A, B).

Figure 6-28 Observed increases in cervical VI were driven primarily by braided suture insertion. Linear regression analyses demonstrated a positive correlation 2 between (A) VI and number of species observed in braided (R =0.09, P=0.002) but 2 not monofilament (R =0.001, P=0.75) cerclages. (B) A similar relationship was 2 observed between suture material and alpha diversity index (braided; R =0.14, 2 P=0.001 and monofilament; R =0.02, P=0.14) indicating an interplay between braided suture, increased cervical vascularity and vaginal microbial dysbiosis.

256 Cervical cerclage, vaginal microbiota and preterm birth prevention

Discussion

This study showed that braided suture used for an ultrasound indicated cerclage induces vaginal dysbiosis, increases expression of inflammatory cytokines, induces premature cervical vascular remodeling and is associated with increased rates of preterm birth and non-viable pregnancy. In contrast, the use of monofilament suture has minimal impact upon the vaginal microbiome and inflammatory pathways associated with premature onset of parturition. These findings provide mechanistic insight into the impact of the cervical cerclage, and more specifically suture material, in high-risk pregnancies. Furthermore, these observations are of clinical significance, as the majority of obstetricians have a non-evidenced based preference for braided suture material at cerclage insertion 113.

Suture material and vaginal bacterial community structure Insertion of braided suture for cerclage was associated with a dramatic and persistent shift towards vaginal dysbiosis characterised by reduced Lactobacillus species, increased diversity and a trend for enrichment of with BV-associated bacteria, Atopobium vaginae and Gardnerella vaginalis, as well as those associated with poor pregnancy outcomes including Peptoniphilus harei253, species of Bacteriodes254, Prevotella255 and Clostridium256. The inherent capacity of braided suture material to facilitate bacterial growth has been previously described in vivo 250, however this study data provides evidence for preferential pathobiont colonization over commensal vaginal species when in-situ, as a cervical cerclage. Moreover monofilament cerclage was associated with high Lactobacillus spp. dominance and stability, characteristic of the healthy vaginal microbial profiles normally seen in low-risk pregnancy 172.

This study highlights potential limitations of the V1-V3 primer set, as proportions of A. vaginae and G. vaginalis appear to be underrepresented among women with significant cerclage- induced vaginal dysbiosis. Subsequent qPCR of the sequenced samples confirmed elevated A. vaginae and G. vaginalis at 4 weeks post-braided cerclage insertion, supporting the contributory role of these species in vaginal dysbiosis. Recently studies of non-pregnant women with BV, employing next generation V3-V4 sequencing, have highlighted the heterogeneous microbial communities that associate with BV 133. In support of my findings, they described BV as not dominated by a single taxon such as G. vaginalis and/or A. vaginae, but rather characterised by heterogeneous bacterial communities with high species richness and diversity, and abundant in a variety of bacteria including genera of Clostridium, Prevotella and Dilaster 133. The virulence of

257 Cervical cerclage, vaginal microbiota and preterm birth prevention

G. vaginalis is thought to relate to its biofilm-producing capacity and adherence to vaginal epithelial cells 134. Future work may investigate the presence of a biofilm on the surface of suture material, following removal of the cervical cerclage prior to labour.

The inflammatory response to suture material Parturition is a cascade of inflammatory events culminating in cervical dilation, myometrial contractility and expulsion of the fetus. At term, circulating non-infection related, pro- inflammatory cytokines IL-1β, IL-6, IL-8 and TNFα play a key role in the onset of cervical remodelling and prostaglandin production leading to labour 19. A reduction in vaginal Lactobacillus spp. is associated with elevated inflammatory mediators in cervicovaginal fluid 180, supporting evidence for the importance of vaginal microbial stability for pregnancy health. Indeed, infection-related inflammation has been implicated in the pathogenesis of preterm birth 206, and therefore in this study it was hypothesised the inflammatory response would correlate with cerclage suture-induced microbial disruption. In support of the hypothesis, I found that braided suture-induced vaginal dysbiosis was associated with increased expression of pro- inflammatory cytokines. In particular, expression of IL-1β, an important inflammatory mediator of preterm parturition 212, as well as IL-6, IL-8, and TNFα, all drivers of preterm cervical ripening212, 213, and MMP-1, a matrix metalloproteinase central to collagenous remodelling preceding preterm birth215, were all increased following braided cerclage insertion. Importantly my findings reveal that these associations are isolated to the braided cerclage whereas the globally less frequently used monofilament suture demonstrated minimal local inflammatory response to cerclage insertion.

This work therefore provides a crucial link between the causative role that the braided suture cerclage plays in infection-induction inflammation, which is thought to prematurely trigger parturition pathways. Indeed, when classified according to the corresponding microbial environment at the time of sampling, compared to normal microbial profiles, a dysbiosis was associated with a significant increase in pro-inflammatory cytokines ICAM-1, IL-1β, IL-6, MMP- 1, MCP-1, TNF-α, GM-CSF, and IFN-γ, but not anti-inflammatory IL-2 or IL-4. Although a causal link is yet to be established, this potentially explains the increased rates of non-viable pregnancy associated with braided cerclage, as exposure of the fetal brain in-utero to elevated levels of inflammatory cytokines has been implicated in the pathogenesis of adverse neonatal outcome and fetal white matter brain injury in utero 13.

258 Cervical cerclage, vaginal microbiota and preterm birth prevention

The cerclage and the cervical vascularity In the lead up to parturition, cervical ripening is characterised by distinct vascular remodelling, particularly angiogenesis, vasodilation and vascular permeability 41, 95, 101. In a pro-inflammatory and positive feedback relationship, activated IL-6 and TNF-α recruit VEGF to the cervix, which in turn up regulates expression of IL-6 and TNF-α 36, 41. This culminates in rapid vasodilation and new blood vessel formation. The increased blood flow that ensues aids delivery of activated cells and enzymes to the cervix, accelerating collagenous breakdown and cervical ripening in preparation for parturition 36, 41. The findings of this study provide evidence for an association between cerclage induced infection, cervical vascularisation and preterm birth. This is supported by previous work in humans and mice reporting LPS-induced inflammation up regulates VEGF in the both the cervix and amnion, suggesting a role for infection triggering premature cervical angiogenesis 41. Although in my study, detectable VEGF was not upregulated following cerclage insertion, cervical vascularity was clearly elevated among women receiving the braided cerclage. In contrast, monofilament suture appeared to provide an unexpected degree of vascularity impedance, relative to the increasing VI observed in healthy pregnancy with advancing gestational age (Chapter 1). This may be due to a tourniquet effect of the cerclage, inhibiting delivery of pro-inflammatory cytokines, VEGF and enzymes.

In summary In summary, this data provides evidence that cervical cerclage using braided suture is associated with increased rates of preterm birth and non-viable pregnancy. I indicate these adverse pregnancy outcomes are likely caused by the promotion of vaginal bacterial dysbiosis following insertion of braided suture material, resulting in activation of local tissue inflammation and in premature cervical remodelling. As monofilament suture has minimal impact on the host microbiome or inflammation in pregnancy, and associates with improved pregnancy outcome, we advocate its use for cervical cerclage. Further clinical trials addressing the impact of cerclage suture material, powered for outcomes of preterm birth, neonatal morbidity and mortality are therefore urgently required. These results are of great clinical significance, as the role of suture material in cerclage efficacy has largely been neglected in clinical research thus far. Existing studies rarely detail the suture material used for cervical cerclage, and the sole investigation into the impact of suture material on cerclage outcomes, randomised women to one of two braided suture materials (Ethibond™ versus Mersilene™); they reported no difference in preterm birth outcome 114. Evidence from my study indicates a complete re- evaluation of existing literature is required.

259 Progesterone supplementation and vaginal microbiota

7 PROGESTERONE SUPPLEMENTATION AND VAGINAL MICROBIOTA

260 Progesterone supplementation and vaginal microbiota

Chapter abstract

Hypothesis Given the anti-inflammatory properties of progesterone on myometrium and cervix, and its efficacy for preterm birth prevention, it was hypothesised that progesterone supplementation promotes Lactobacillus dominance and prevents vaginal dysbiosis, consistent with that of healthy pregnancy.

Aim This study aimed to investigate the impact of progesterone intervention on vaginal microbial structure in high-risk women with a short cervix (≤25mm).

Methods Women with a prior preterm birth (<37weeks) attending cervical length (CL) screening before 18 weeks gestation were prospectively recruited. Women with a short CL (≤25mm before 18 weeks) were prescribed vaginal progesterone 400mg/OD to continue until 34 weeks. Women without cervical shortening were considered controls. High vaginal swabs for 16S rRNA sequencing were taken at initial screening (<18 weeks) and longitudinally at 22, 28 and 34 weeks follow up. Microbial profiles were compared among progesterone and control (non-short cervix) groups.

Results Of the 70 women recruited, 3 were lost to follow up, 25 received progesterone for a short cervix, and the remaining 42 were considered controls. Rates of preterm birth (<37weeks) were higher in the progesterone (32%, 8/25) versus control groups (5%, 2/42; P=0.004). Before progesterone intervention, a short cervix was associated with slightly higher prevalence of vaginal dysbiosis compared to controls (12% v 2%, P=0.14). Progesterone intervention was not associated with any significant modulation of vaginal bacterial communities during pregnancy. Similarly, no significant differences in vaginal microbiota were detected between patients receiving progesterone who delivered preterm and those who delivered at term.

Conclusion Progesterone supplementation does not adversely impact upon the vaginal microbiome in high-risk pregnancy with a short cervix, however nor does it attenuate L. iners-associated preterm birth. Progesterone’s mode of action in preterm birth prevention is not via modulation of vaginal microbial communities.

261 Progesterone supplementation and vaginal microbiota

Introduction

Pregnant women with a short cervix, ≤25mm before 24 weeks gestation are considered at highest risk of preterm birth83. Early diagnosis of these pregnancies enables timely and targeted intervention by either cervical cerclage or vaginal progesterone therapy 50. While both prevention strategies display comparable efficacy 117, progesterone supplementation is increasing used as it negates the surgical risks associated with cerclage insertion 104 and has not been associated with adverse neonatal effects 124. In pregnancy, progesterone maintains myometrial quiescence through the downstream anti- inflammatory effects via Progesterone-B (PR-B) receptor mediated signalling 26, 27. Labour onset is thought to occur following a functional decline in progesterone action, precipitated through switching in the expression of the progesterone receptor isoforms from PR-B to PR-A 24. Increasing myometrial PR-A receptor expression opposes the PR-B receptor-mediated anti- inflammatory effects, while promoting pro-inflammatory gene expression and therefore parturition 26. Increased infection-induced pro-inflammatory response resulting in preterm birth is associated with decreased progesterone receptor expression levels at the maternal fetal interface in decidual cells 224. The mechanism of action of vaginal progesterone supplementation in the prevention of preterm birth is thought to involve its capacity to promote anti-inflammatory, pro-relaxant pathways in the uterus 26, 28, 30, as well as direct inhibition of prostaglandin induced collagenous remodeling and softening of the cervix 31, 33. In clinical studies, vaginal progesterone has been shown to attenuate the rate of cervical shortening 122, but it’s efficacy in preterm birth prevention is cervical-length dependant 119, 121, 125, 126. Most beneficial to high risk pregnancies with a short cervix ≤25mm 117, administration of progesterone to women with a longer cervix has not been shown to not improve pregnancy outcome 121. Studies on the impact of progesterone on the composition of the hormonally-influenced 167, 168 vagina microbiota are limited. In non-pregnant women Borgdorff et al 257 found that both injectable progestin contraception and combined oral contraception (progestin and oestrogen) does not significantly alter vaginal microbiota, although larger systematic reviews indicate a decrease in the clinical incidence of BV 258. However, hormonal contraceptives do increase the risk of HIV transmission 259, and it is has been hypothesised that this susceptibility relates to a progesterone-induced modulation of the local inflammatory immune response to infection or alternately thinning of the vaginal epithelial barrier 260. The impact of vaginal progesterone pessaries on the composition of vaginal microbial communities in pregnancies at risk of preterm birth is unknown.

262 Progesterone supplementation and vaginal microbiota

Aim To assess the longitudinal impact of progesterone pessaries on the vaginal microbiome in high- risk pregnancy with a short cervix using 16S rRNA gene sequencing.

Hypothesis Progesterone supplementation does not disrupt Lactobacillus spp. dominance and stability, characteristic of healthy pregnancy.

263 Progesterone supplementation and vaginal microbiota

Study design

Patient recruitment and sample collection Given the significant association between second trimester vaginal microbial composition, cervical length and subsequent preterm birth <34 weeks (demonstrated in Chapter 4) a prospective longitudinal study aimed at assessing the impact of progesterone on pregnancy at risk of preterm birth with a short cervix was undertaken.

Women with a prior preterm birth <37weeks were prospectively recruited from preterm surveillance clinics. At initial screening ≤18 weeks, all women underwent vaginal swab sampling for 16S rRNA sequencing, followed by CL measurement at transvaginal ultrasound. Based on CL measurement, women were then allocated into one of two groups: 1) women with a short cervix ≤25mm were treated with vaginal progesterone 400mg OD at night to continue until 34 completed gestational weeks, 2) Women with a normal CL >25mm were used as controls and did not receive progesterone or cerclage for the remainder of their pregnancy. Both progesterone and control groups were recruited at ≤18 weeks and vaginal swab samples were collected longitudinally at 22, 28 and 34 weeks gestation.

Figure 7-1 Women with a prior preterm were recruited at ≤18 weeks for HVS sampling and a transvaginal scan. Women with a CL ≤25mm were prescribed vaginal progesterone 400mg/OD until 34 weeks, and women with a long cervix (CL >25mm) were ‘controls’. Longitudinal HVS sampling was performed at 22, 28 and 34 weeks for g of 16S rRNA gene sequencing. (w=weeks gestation, HVS= high vaginal swab)

264 Progesterone supplementation and vaginal microbiota

As per current clinical practice, progesterone was administered vaginally to ensure enhanced bioavailability and the absence of undesirable side-effects such as sleepiness, fatigue, and headaches that have been associated with oral administration 261. Eligibility criteria included women with a singleton pregnancy and a prior spontaneous preterm birth <37weeks, who had not undergone CL screening, or received either progesterone or cerclage intervention prior to recruitment. Bacterial DNA extraction and 16s rRNA gene sequencing was performed as described in materials and methods. Data was re-sampled and normalized to the lowest read count in Mothur (n=725).

Statistical analyses Sequence data was assessed at bacterial species level, and assigned into community state types (CSTs) 155 using ward linkage hierarchical clustering analysis (HCA) species taxonomy with a clustering density threshold of 0.75, as previously described. The effect of progesterone on species data was assessed at 22, 28 and 34 weeks using Kruskal-Wallis, and Dunn’s multiple comparisons where appropriate. Significance of differences in CST classification according to HCA clustering of species data, between progesterone and control groups, and outcomes of preterm delivery using Fisher’s exact test were calculated at each sampling time- point (18, 22, 28 and 34 weeks). Welch’s t-test compared differences in mean species abundance among groups at corresponding time-points. Gestation at delivery was defined as outcomes of early preterm birth <34+0 weeks, late preterm birth 34+0 to 36+6 and term birth ≥ 37+0 weeks.

265 Progesterone supplementation and vaginal microbiota

Results

7.1 Demographics

A total of 70 pregnant women with a prior spontaneous preterm birth consented to recruitment and had a CL measurement at initial screening ≤18 weeks. Of these, 3 were lost to follow up and 25 women were found to have a short CL ≤25mm and were administered Progesterone 400mg OD nocté until 34 weeks. The 42 remaining women did not shorten their cervix or receive any subsequent preventative intervention (progesterone or cerclage) for their preterm birth risk, and were considered the controls for the remainder of the study.

234 high vaginal samples were collected from control and progesterone groups attending longitudinal follow up (22, 28 and 34 weeks) at matched gestational ages Table 7-1.

Table 7-1 Gestational age at sampling of control and progesterone groups

Cohort High risk controls Progesterone Total

Mean ±SD Mean ±SD n Range n Range n (weeks) (weeks) Recruitment, 18w 42 15+0 ±2.8 (12+1 - 18+2) 25 15+5 ±2.0 (12+0 - 17+6) 67

Follow up, 22w 40 22+2 ±1.5 (19+1 - 25+6) 22 21+3 ±1.5 (19+1 - 23+6) 62 Follow up, 28w 37 28+0 ±1.0 (26+1 - 30+0) 21 28+0 ±0.8 (27+0 - 30+3) 58 (weeks) (weeks) sampling sampling Gestation at Follow up, 34w 29 34+1 ±1.0 (32+1 - 36+2) 19 33+5 ±1.0 (32+0 - 34+6) 48 Total samples, n 148 85 234

266 Progesterone supplementation and vaginal microbiota

Demographics of the progesterone and control populations are provided in Table 7-2. As per study design, the mean CL at commencement of progesterone was significantly lower in the ‘short CL’ than the ‘non-short control’ group (22mm v 32mm, P<0.05), at comparable screening gestations (15+5 v 15+0 respectively). Respective rates of preterm birth (<37weeks) were higher in the progesterone (32%, 8/25) versus control groups (5%, 2/42; P=0.004; Table 7-2)

Table 7-2 Participant demographics for control and progesterone groups

High-risk controls Progesterone Total population (CL >25mm) (CL ≤25mm)

n/N (%) 42/67 (63%) 25/67(37%) 67/67 (100%)

Age, years Mean ±SD (range) 29 ± 3.1 (24-34) 30 ±3.8 (25-36) 29 ±3.3 (24-36)

BMI Mean ±SD (range) 24.7 ±5.3 (19-48) 25.2 ±4.7 (18.4-35) 24.9 ±5.0 (18.4-48)

Ethnicity, n/N (%) Caucasian 32/42 (76%) 18/25 (72%) 50/67 (75%) Asian 4/42 (10%) 3/25 (12%) 7/67 (10%) Black 6/42 (14%) 4/25 (16%) 10/67 (15%)

Smoker n/N (%) 2/42 (5%) 0/25 (0%) 2/67 (3%)

Screening for Progesterone GA (w), median (range) 15+0 (12+1 - 18+2) 15+6 (12+0 - 18+6) 15+3 (12+0 - 18+6) CL (mm), median (range) 32 (26 - 43) 22 (13 - 25) 28 (13 - 43)

Gestation at delivery, n/ N (%) Early PTB, <34+0 w 1/42 (2%) 4/25 (16%) 5/67 (7%) Late PTB, 34+0 to <37+0 w 1/42 (2%) 4/25 (16%) 5/67 (7%)

Total PTB <37w 2/42 (5%) 8/25 (32%)* 10/67 (15%) Term, ≥37+0 w 40/42 (95%) 17/25 (68%) 57/67 (85%) PTB = preterm birth, GA = gestational age, w= weeks, CL = cervical length (mm) *P=0.004 (Fishers exact)

267 Progesterone supplementation and vaginal microbiota

7.2 Vaginal microbiota and progesterone

Bacterial 16S rRNA sequence data were assessed at species taxonomic level. Samples were assigned to community state types (CSTs) 155, using ward linkage HCA at each sampling time- points (<18, 22, 28 and 34 weeks): CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), CST IV (diverse species) and CST V (L. jensenii) (Figure 7-2, Table 7-3)

Figure 7-2 Ward hierarchical clustering of vaginal species at sampling prior to progesterone intervention at 18 weeks (w), and longitudinally at 22, 28 and 34 weeks comparing control and progesterone groups.

268 Progesterone supplementation and vaginal microbiota

Prior to progesterone intervention, no significant difference in the distribution of CSTs between to the two patient cohorts was observed. However there was a trend for a greater number of women deficient in Lactobacillus spp. (CST IV) in the Progesterone (short CL group) compared to controls (12% v 2%, P=0.14) and specifically, fewer women with bacterial communities dominated by L. crispatus (CST I: 43% v 32%, P=0.2; Figure 7-3A, Table 7-3).

Vaginal progesterone supplementation had no effect upon vaginal bacterial community state structure throughout pregnancy (Figure 7-3A, Table 7-3), nor measurements of species richness (Figure 7-3B) or alpha diversity (Figure 7-3C). At the first sampling after commencement of progesterone (22 weeks gestation) there was a slight reduction in the number of species observed in the progesterone group compared to pre-treatment observations (Mean 3.8 v 4.7; P=0.26), and compared to gestational age-matched controls (mean 4.3, P=0.46; Figure 7-3B,C; Table 7-4).

Progesterone was associated with a non-significant trend reduction in mean L. iners abundance, with a corresponding increase in L. crispatus, a trend mirroring microbial composition shifts in the control (Figure 7-4A, B, Table 7-5).

The control group demonstrated higher mean L. crispatus abundance throughout pregnancy than the progesterone group (Figure 7-4B); the number of women with a CST I dominated microbiome increased from 43% at sampling prior to 18 weeks, to 66% at 34 weeks, with a corresponding reduction in L. iners dominance (33% reducing to 21% respectively; Table 7-3). Comparatively, in women receiving progesterone at gestationally aged matched time-points, CST I dominance increased from 32% to 47%, and CST III dominance fell from 36% to 24% with advancing gestation from <18 weeks to 34 weeks respectively (Table 7-3, Figure 7-3A).

269 Progesterone supplementation and vaginal microbiota

Table 7-3 CST classification of microbial samples according to gestation at sampling. Progesterone versus control groups Community state type I II III IV V Gestation at Diverse Cohort L. crispatus L. gasseri L. iners L. jensenii sampling species <18 weeks Control 43% 12% 33% 2% 10% Short Cervix 32% 12% 36% 12% 8% TOTAL 39% 12% 34% 6% 9% P value* 0.40 1.00 1.00 0.14 1.00

22 weeks Control 51% 13% 26% 5% 5% Progesterone 32% 18% 32% 9% 9% TOTAL 44% 15% 28% 7% 7% P value* 0.18 0.71 0.77 0.61 0.61

28 weeks Control 55% 8% 26% 3% 8% Progesterone 41% 9% 32% 9% 9% TOTAL 50% 8% 28% 5% 8% P value* 0.42 1.00 0.77 0.55 1.00

34 weeks Control 66% 3% 21% 3% 7% Progesterone 47% 12% 24% 12% 6% TOTAL 59% 7% 22% 7% 7% P value* 0.61 0.55 1.00 0.55 1.00

TOTAL POPULATION 47% 11% 29% 6% 8% *P value = two tailed fisher exact test; Comparison of control v short cervix/progesterone groups Community state type according to ward hierarchical clustering of bacterial species ; Short cervix = CL <25mm going onto receive progesterone

270 Progesterone supplementation and vaginal microbiota

Figure 7-3 Vaginal progesterone treatment does not alter the structure of the vaginal microbiome in pregnancy (A) Compared to controls (n=42), progesterone supplementation (n=25) had no significant impact upon microbial community profiles with advancing gestation. Similarly, no effect of progesterone treatment upon (B) the number of species observed or (C) the corresponding Shannon index of alpha diversity was observed. Fewer women requiring progesterone had an L. crispatus dominated microbiome compared to controls (8/25, 32% v 18/42, 43%, P=0.4), however progesterone treatment was associated with increased L. crispatus abundance with advancing gestation.

271 Progesterone supplementation and vaginal microbiota

Table 7-4 Number of bacterial species observed in high risk control and progesterone groups according to gestation at sampling

Gestation at Samples Species observed sampling taken, n Mean Std. Dev Min Max <18 w 42 4.4 3.1 1 17 22 w 40 4.3 2.8 1 14 28 w 37 4.1 2.6 1 12 controls High risk 34 w 29 4.0 2.9 1 14

<18 w 25 4.7 2.9 1 12 22 w 22 3.8 2.6 1 12 28 w 21 4.2 3.8 1 18 34 w Progesterone 19 4.5 4.3 1 18 No significant difference: Dunns multiple comparison test, and Kruskal-Wallis w= weeks gestation

272 Progesterone supplementation and vaginal microbiota

7.3 Longitudinal microbial profiles

The distribution of vaginal CSTs during pregnancy were longitudinally assessed in both progesterone and control cohorts (Figure 7-4, Figure 7-5). Regardless of intervention, L. crispatus (CST I) dominance was associated with high stability throughout pregnancy with 92% (24/26) of women maintaining high L. crispatus abundance across all sampling time points. In contrast, significantly lower stability was observed in the 23 women exhibiting a L. iners dominated microbiome at the first sampling; 74% of these women shifted to an alternative CST at some stage during their pregnancy (n=17, P<0.0001; Fisher’s exact test). Similar levels of CST-shifting were observed in the vaginal progesterone (9/25; 36%) and control groups (10/43; 23%, P=0.277; Fisher’s exact test; Figure 7-5; Table 7-5)

273 Progesterone supplementation and vaginal microbiota

Figure 7-4 Gestational age was associated with a non significant trend decline in (A) mean L. iners abundance from <18w to 34 weeks in both the controls (P=0.8) and progesterone groups(P=0.7), with a corresponding increase in (B) mean L. crispatus abundance in progesterone groups (P=0.4; Kruskal-Wallis; Table 7-5).

274 Progesterone supplementation and vaginal microbiota

Table 7-5 P values for differences between mean microbial abundance with (A) advancing gestational age from <18 to 34 weeks among Progesterone and Control groups and (B) a cross-sectional comparison of Progesterone versus Control groups at specific gestational ages.

A P values for differences in mean species abundance at longitudinal sampling from <18 week to 34 week sampling

L. crispatus L. gasseri L. iners L. jensenii Control 0.358 0.68 0.853 0.965 Progesterone 0.395 0.393 0.740 0.742 P value* Total 0.146 0.659 0.826 0.995 *P value: uncorrected Kruskal-Wallis, Turkey Kramer post hoc

B P values for differences in mean species abundance comparing Control versus Progesterone groups

Gestation <18 w 22 w 28 w 34 w L. crispatus 0.193 0.136 0.458 0.178 L. gasseri 0.849 0.704 0.733 0.268 L. iners 0.622 0.620 0.726 0.808 value* P L. jensenii 0.759 0.799 0.855 0.437

Control 42 39 37 29 Progesterone 25 22 21 19 n=

sample, TOTAL 67 61 58 58 *P value: Two tailed Welch's t-test

275 Progesterone supplementation and vaginal microbiota

Figure 7-5 Longitudinal profiling of community state types for progesterone (n=25) and control groups (n=42). Progesterone supplementation was commenced after the first sampling time-point (<18 weeks). Each longitudinal sample was assigned to a CST (Figure 2A) as indicated by the colour-coded rectangle, and categorised as a function of delivery gestation.

276 Progesterone supplementation and vaginal microbiota

7.4 Vaginal microbiota and gestation at birth

Microbial profiles differed significantly among women delivering before and after 34 weeks, but not 37 weeks (Figure 7-6A, B). This association is consistent with findings described in Chapter 4, where samples from late preterm (34-37 weeks) and term births (>37 weeks), were similarly abundant in L. crispatus, while early preterm births (<34weeks) were characteristically dominant in L. iners and low in L. crispatus (Figure 7-6D, Figure 7-7D, E, Table 7-6). This is demonstrated in the hierarchical clustering of samples (Figure 7-6A Heatmap), and the patient journey (Figure 7-5), where early and late preterm births are clearly differentiated. Unlike the findings in Chapter 4, cervical length did not correlate with vaginal microbiota, however this is likely due to the small sample size in this study (Figure 7-6A, C)

277 Progesterone supplementation and vaginal microbiota

278 Progesterone supplementation and vaginal microbiota

Figure 7-6 Preterm birth <34+0 weeks is associated with L. iners dominance at longitudinal sampling irrespective of progesterone intervention. Clustering (A) and PCA (B) analyses of species taxonomic data of samples taken <18weeks (prior to intervention), were classified according to cervical length <25mm (B) and corresponding gestation at delivery (C). No significant association between CST III and cervical length was observed (B), however women delivering preterm <34+0 weeks (C) exhibited CST III (L. iners) dominance. (D) At longitudinal sampling of 67 women (inclusive of control and progesterone groups), L. iners was highly abundant (P<0.05), and L. crispatus deficient (P=0.14), in women delivering <34+0 weeks (n=5) compared to >34+0 weeks (n=62), at comparable sampling time-points (two- tailed Fisher’s exact).

279 Progesterone supplementation and vaginal microbiota

Table 7-6 CST classification of microbial samples according to gestation at sampling. Preterm <34 weeks (n=5) versus birth >34 week (n=62). Community state type I II III IV V Gestation at Gestation at Diverse L. crispatus L. gasseri L. iners L. jensenii sampling birth species <18 weeks PTB < 34w 0% 0% 80% 0% 20% Birth >34w 42% 13% 31% 6% 8% TOTAL 39% 12% 34% 6% 9% P value* 0.15 1.00 0.04 1.00 0.38

22 weeks PTB < 34w 0% 0% 60% 0% 40% Birth >34w 48% 16% 25% 7% 4% TOTAL 44% 15% 28% 7% 7% P value* 0.06 1.00 0.13 1.00 0.03

28 weeks PTB < 34w 0% 0% 100% 0% 0% Birth >34w 54% 9% 23% 5% 9% TOTAL 50% 8% 28% 5% 8% P value* 0.11 1.00 0.00 1.00 1.00

34 weeks PTB < 34w 0% 0% 100% 0% 0% Birth >34w 61% 7% 18% 7% 7% TOTAL 59% 7% 22% 7% 7% P value* 0.17 1.00 0.04 1.00 1.00

TOTAL POPULATION 47% 11% 29% 6% 8%

* P value = two tailed fisher exact test; Comparison of PTB <34 weeks v birth >34weeks group PTB= preterm birth; w= weeks; Community state type according to ward hierarchical clustering of bacterial species

280 Progesterone supplementation and vaginal microbiota

281 Progesterone supplementation and vaginal microbiota

Figure 7-7 Comparison of mean L. iners (A) and L. crispatus (B) abundance in Control (n=42) versus Progesterone groups (n=25). (A) Prior to Progesterone intervention at <18week sampling, women with a short CL ≤25mm had greater abundance of L. iners compared to controls, and lower L. crispatus (B) although this did not reach significance. L. iners abundance declined in both control and progesterone groups towards 34+0 weeks sampling while mean L. crispatus abundance increased. Inclusive of control and progesterone groups, preterm birth <34 weeks was associated with higher mean L. iners abundance at longitudinal sampling (C; P<0.05), and lower mean L. crispatus abundance (D; P<0.001) than deliveries >34weeks, at matched gestational age at sampling throughout follow up (Welch’s t-test; Table 7-7)

Table 7-7 P value for differences in mean species abundance among women delivering <34weeks compared to >34 weeks in the total cohort of 67 women Gestation <18 w 22 w 28 w 34 w L. crispatus <0.001 <0.001 <0.001 <0.001 L. gasseri 0.002 <0.001 0.042 0.028 L. iners 0.039 0.050 0.008 0.362 P value* L. jensenii 0.641 0.166 0.537 0.943

PTB <34 5 5 4 3 Term >34w 62 56 54 56 n=

sample, TOTAL 67 61 58 58 *P value: Two tailed Welch's t-test

282 Progesterone supplementation and vaginal microbiota

7.5 Progesterone and gestation at birth

In women receiving progesterone, marked differences in longitudinal CST distributions were observed among women delivering before and after 34 weeks (Figure 5). At sampling <18 weeks, L. iners-dominance was observed in 100% (4/4) of women receiving progesterone who subsequently delivered preterm <34 weeks compared to 16% (4/25) of births >34 weeks. This trend persisted with advancing gestation at sampling; L. iners-dominance was observed in 50% (2/4) at 22 weeks and 100% (4/4) at 28 weeks (Figure 7-8, Figure 7-9, Tables 7-8).

283 Progesterone supplementation and vaginal microbiota

Figure 7-8 Ward hierarchical clustering of vaginal species in the progesterone group only (n=25) at sampling prior to progesterone intervention at 18 weeks (A), and longitudinally at 22 (B), 28 (C) and 34 weeks (D) comparing gestation at birth.

284 Progesterone supplementation and vaginal microbiota

Progesterone only n=25: Birth <34 versus >34weeks

Figure 7-9 Preterm birth despite vaginal progesterone, was associated with L. iners dominance throughout pregnancy. Longitudinal sampling of 25 women receiving progesterone for a short cervix showed L. iners dominance was strongly associated subsequent preterm birth <34 weeks (n=4; (P<0.05). *Delivery samples were collected within 2 weeks of delivery between 28 and 34 weeks.

Tables 7-8 P value for differences in mean species abundance among women receiving progesterone for a short cervix (n=25) delivering <34weeks (n=4) compared >34 weeks (n=21) Gestation <18 w 22 w 28 w 34 w L. crispatus 0.003 0.005 0.003 0.002 L. gasseri 0.082 0.040 0.330 0.089 L. iners 0.116 0.114 0.000 0.316 P value* L. jensenii 0.672 0.254 0.532 0.695

PTB<34w 4 4 4 3 Term >34w 21 18 17 17 n=

sample, TOTAL 25 22 21 20 *P value: Two tailed Welch's t-test

285 Progesterone supplementation and vaginal microbiota

Discussion

Vaginal microbial profiles have previously been shown to closely associate with endogenous hormonal fluctuations 167, 168. Prior to the work described in this chapter, the impact of progesterone supplementation on vaginal microbial communities was only described in non- pregnant women, receiving either injectable or oral preparations 257. To date, no assessment of the impact of vaginal progesterone supplementation on vaginal microbiota in pregnancy has been undertaken. The evidence for vaginal progesterone supplementation in preterm birth prevention is now compelling 262, particularly in those with a short CL ≤25mm 117, 121. Defining a targeted group for progesterone intervention using CL thresholds of ≤25mm was a major strength of this study, as it enabled provision of the first evidence for the role of progesterone supplementation on bacterial communities in a truly high risk pregnant population.

The impact of progesterone on vaginal microbiota Previously (Chapters 4 and 5) I demonstrated a significant association between vaginal microbial composition and both preterm birth and cervical shortening, strongly supporting a pathogenic role of L. iners dominance in these groups. Unlike other Lactobacillus spp., there is evidence that L. iners induces pro-inflammatory signalling in vitro 53, 232. In this study it was postulated that the pathogenic and inflammatory effects of L. iners might be attenuated by the anti-inflammatory actions of progesterone. Longitudinal 16S rRNA sequence data of the vaginal microbiome in women with a prior preterm birth and short cervix demonstrated that progesterone supplementation induces a mild decline in L. iners and a corresponding increase in L. crispatus abundance. However, a similar shift in CST profiles was observed in the control group at gestationally aged matched sampling time-points, and so it could not be ascertained if this effect was exclusively due to progesterone action. Importantly, my data demonstrate, as observed in non-pregnant cohorts 257, that progesterone treatment, or the daily vaginal insertion of a pessary, does not have a detrimental impact of the structure of vaginal microbial communities during pregnancy. This is reassuring from a clinical perspective, as patients and clinicians who may be concerned about the ‘infection risk’ associated with use of vaginal pessaries during high risk pregnancy can be reassured.

286 Progesterone supplementation and vaginal microbiota

Progesterone supplementation, vaginal microbiome and preterm birth The second important finding in the study was that women receiving progesterone who went on to deliver preterm had a disproportionately high L. iners dominance of vaginal bacterial communities. This may indicate a lack of protection provided by progesterone against preterm birth in this subpopulation of L. iners-dominant women with a short cervix. The study size however was underpowered to draw firm conclusions. Further larger studies are required to ascertain whether progesterone truly is preferentially effective among pregnancies dominant in alternate community state types. If so, this has significant clinical implications. In several studies, progesterone has been promoted as reducing both cervical shortening and rates of preterm birth 119, 125, 126. In the US, progesterone is increasingly advocated over cervical cerclage as the primary preventative intervention 262. Recently however, the Opptimum study and others, refute the effective of progesterone, reporting no reduction in preterm birth in those receiving targeted for a short cervix 121, 124. Postulating that co-existing vaginal microbiota plays a role in the effectiveness of progesterone prophylaxis, may help explain the conflicting results of these large RCTs. Furthermore, assessment of vaginal microbiota in the setting of a high risk prematurity clinic may help guide targeted intervention in the future; whereby progesterone treatment is reserved for non-L. iners dominant women with a short cervix, while a cerclage or probiotic treatment may be more effective in the subpopulation abundant in L. iners.

In summary Vaginal progesterone therapy for a short cervix in high-risk pregnancies had no effect upon the observed vaginal community state structure. The action of vaginal progesterone in the prevention of preterm birth is therefore unlikely to be via modulation of vaginal microbial communities. While progesterone supplementation maintains vaginal Lactobacillus spp. stability, it does not appear to diminish the preterm birth risk associated with L. iners dominance. These findings need to be replicated on a larger scale before firm clinical conclusions may be drawn.

287 Summary Discussion

8 SUMMARY DISCUSSION

288 Summary Discussion

Summary of the clinical problem

Preterm birth is the leading cause of mortality under the age of five 7, and survivors of preterm birth are at risk of long-term multi-organ morbidity8. Ascending bacterial infection from the vagina into the uterine cavity is the most common source of intrauterine inflammation that leads to spontaneous preterm birth 6, 48, 50. Preterm babies exposed to infection are at significantly increased risk of subsequent neurodevelopmental disability 13. During pregnancy, the cervix acts as both a physical and immuno-modulatory barrier protecting the fetus and gestational tissues from the external environment 35. Lipopolysaccharide (LPS)-induced inflammation in the cervix has been shown to disrupt the cervical epithelial barrier 51, causing influx of activated neutrophils and matrix-metalloproteinases, which in turn cause premature cervical ripening 38. The host-microbe interactions within the vagina therefore play a fundamental role in reproductive health, as well as pathology.

Prior to the work presented in this thesis there had been no examination of the association between the cervical phenotype and vaginal microbiota during pregnancy. Furthermore, the impact of preventative measures such as the cervical cerclage and progesterone pessaries had not been established in women at risk of spontaneous preterm birth. Overall, this study aimed to assess the relationship between cervical morphology, in particular cervical vascularity, volume and length and the co-existing vaginal microbiota, in both low- and high-risk populations for preterm birth.

289 Summary Discussion

Clinical relevance and future work

8.1 Preterm birth surveillance in high-risk pregnancy

Currently transvaginal cervical length measurement is the primary screening tool employed for preterm birth surveillance 88, where a 10th centile threshold (≤25mm before 24 weeks) is considered to be a short cervix 83, 84. The gestational age at which measurements are taken significantly affects the calculation of risk for spontaneous preterm birth, whereby improved sensitivity is achieved at later screening (>22 weeks) at the compromise of specificity 89. Identification of a short cervix at this late gestation is associated with an increased risk of complication at insertion of a cervical cerclage 263. Consistent with observations in previous cross-sectional cohorts 95, work in this thesis has demonstrated that assessment of additional cervical parameters by 3D/4D ultrasound technology, namely cervical volume and vascularity, do effectively differentiate subsequent term from preterm birth, when assessed antenatally. Despite cervical volume providing comparable screening performance to length for the prediction of preterm birth, the clinical application of this technology is limited due the technical complexity of acquiring and interpreting volume and vascularity data. Measurement of 2D cervical length in comparison is an easier skill to acquire, and provides instant, interpretable and reproducible results in the clinic setting. Therefore cervical length screening is unlikely to be surpassed or improved by the addition of cervical volume or vascularity assessments for use in preterm birth surveillance in current or future clinical practice. Subsequent work in this thesis employing 16S rRNA gene sequencing to characterise vaginal microbiota in pregnancy revealed that assessment of vaginal microbial composition in the early second trimester may be used to stratify preterm birth risk. In my study the combination of vaginal microbiome and cervical length measurements appeared to improve screening performance, and although numbers were small, high sensitivities were achieved at comparatively earlier gestations 88, 89. This is clinically important as early identification of the highest risk pregnancies enables timely interventions, at an earlier and therefore safer gestation. My work thus far indicates that second trimester L. crispatus dominance, a known inhibitor of pathobiont colonisation 228, is predictive of term birth, while high L. iners abundance is associated with increased risk of preterm birth and warrants close surveillance. Larger cross- sectional studies are required to validate these findings.

290 Summary Discussion

Future work may focus on exploring detection of vaginal microbial composition in the early second trimester that may be used to stratify pregnancy risk and guide frequency of cervical length surveillance, or indeed focus on a combined screening tool utilising both cervical length and vaginal microbiota to identify women at highest risk. A system such as this would require a technique for rapid detection of specific vaginal microbial profiles however. Metabolic profiling of urine, serum or cervicovaginal fluid maybe the focus of future exploratory studies to identify biomarkers specific to vaginal microbial communities or their interaction with the maternal host. The ultimate aim would be to develop a bedside test for use in a clinical setting such as prematurity surveillance clinics. My presented findings have led to the hypothesis that women who have dysbiosis consistently during pregnancy may not mount a strong immune response to dysbiosis-associated bacteria. This may explain why they have persistent dysbiosis during pregnancy (rather than shifting to a more stable Lactobacilli-dominant microbiome, as would be expected 172), and why they do not go on to deliver preterm. In contrast, women who are L. iners-dominated and then develop dysbiosis are at higher risk. If routine vaginal microbiome screening were to be introduced during pregnancy, it will likely not be clear if a diagnosed dysbiosis is chronic (low risk) or acute (high risk). Therefore it may be useful to include all women who screen positive for dysbiosis, as well as L. iners. This is feasible, as dysbiosis accounts for a small proportion of women in pregnancy, as demonstrated in my cohort, and by others 171. It is currently unclear why some women are susceptible to the pathogenesis of L. iners and BV- dysbiosis, while others appear to tolerate these microbiota and do not deliver preterm. Such cases may represent the absence of an inflammatory response invoked by response to vaginal infection, or a lack of virulence potential encompassing certain pathogenic organisms 176. Future work on the expression of mediators in cervicovaginal fluid of women known to have L. iners and/or BV and go onto deliver preterm may provide some insight into this.

291 Summary Discussion

8.2 Preterm birth risk following excisional cervical treatment

Based on previous associations between vaginal dysbiosis, HPV persistence and CIN severity 244, 245, it was hypothesised that women with cervical treatment would harbour a greater proportion of abnormal vaginal microbiota than other pregnancy cohorts. Instead, I demonstrated that pregnancy post-excisional CIN treatment was associated with comparatively higher proportions of L. crispatus than women with a previous preterm birth, with fewer women experiencing preterm birth, despite lower cervical lengths and volumes throughout pregnancy. This reinforces the concept that vaginal microbiota may be drivers of pathogenic cervical shortening, but only a specific cohort of women; in the cervical treatment group, a short cervix was likely secondary to iatrogenic excision of cervical tissue rather than pathological microbial- induced inflammation. Therefore cervical shortening did not correlate with vaginal microbial profiles in this cohort. A second important finding was that, unlike the prior preterm birth group, L. crispatus is not entirely protective against subsequent preterm birth in cases where the cervix has been substantially shortened by excisional treatment. This supports a hypothesis for mechanical damage and previous evidence that the depth of cervical excision is proportional to adverse pregnancy outcome 75. Current preterm birth surveillance policies that advocate blanket thresholds of cervical length (such as ≤25mm) for targeted preventative interventions are unlikely to accurately capture women at risk in the two distinct groups described; as evidenced from my findings, cervical and microbial profiles among women with a prior preterm birth and those with excisional cervical treatment differed substantially throughout pregnancy. Certainly an individualised assessment of vaginal microbiota and cervical length would appear to be beneficial, although larger studies are required to fully assess feasibility and predictive accuracies of such models.

Future investigation of antenatal persistence of HPV also may be an important consideration with respect to preterm birth risk, as viral infections of the cervix during pregnancy have been shown to impede defense against bacterial invasion of the uterus in mice-models 243. Indeed, it is possible that the low preterm birth rates among my study population reflected effective clearance of HPV prior to sampling. In this sub population of pregnant women post-cervical treatment, L. gasseri may be considered a potential probiotic. In this study, L. gasseri dominance was associated with the largest cervical volumes, as well as the longest duration of pregnancy. Furthermore, in non-pregnant women L. gasseri as has been associated with rapid

292 Summary Discussion

HPV clearance when compared to other CSTs 244. Clinical effectiveness, feasibility and safety trials for a L. gasseri probiotic may therefore be justified.

8.3 Investigation and treatment of vaginal infection in pregnancy

Bacterial vaginosis (BV) has long been associated with adverse reproductive 138-141 and pregnancy outcomes 142. A major challenge in the management of BV is a high relapse and recurrence rate following standard antibiotic treatment with Metronidazole and Clindamycin 145, 146. Moreover, evidence suggests that screening and treating BV in pregnancy does not significantly reduce rates of preterm birth 142. This quandary may relate to emerging evidence suggesting an interconnected pathogenesis of L. iners and BV-associated dysbiosis in non- pregnant women 176. The relatively recently described L. iners 264 has a tendency to transition to and from BV-associated states 177, 233, and can tolerate co-existence with BV 133. In addition L. iners has been identified as the prominent vaginal species following BV-antibiotic treatment 265. Indeed Wang et al 147 proposed that the failure of antibiotic treatment for the eradication of BV relates to a lack of Lactobacillus spp. restoration post-treatment. Macklaim et al266 supported these postulations, indicating that the current gold standard of antibiotics to treat BV is not sufficient to restore bacterial homeostasis in the vagina, and that a regime of probiotics, post- antibiotic was a preferential treatment regime. Describing a regime of Tinidazole with L. reuteri and L. rhamnosus, they noted an increase in proportions of L. iners as well as L. crispatus post treatment 266. Future work may focus on developing a rapid technique for the identification of L. iners, and investigate the effect of probiotic on community stability in non-pregnant working towards pregnant cohorts. Ultimately the efficacy of a probiotic for the treatment of L. iners-BV pathogenesis as a preterm birth intervention would be pivotal to preterm birth research.

8.4 Progesterone versus cerclage for preterm birth prevention

In this thesis I have demonstrated the important role of the vaginal microbiota with respect to the efficacy of intervention. While progesterone supplementation maintains vaginal Lactobacillus spp. stability, it did not appear to diminish the preterm birth risk associated with L. iners dominance. Study participant numbers were small however, but if this phenomenon is real, there may be fundamental implications for preterm birth prevention strategies. Specifically, assessment of vaginal microbiota in the setting of a high risk prematurity clinic may help guide targeted intervention, whereby progesterone treatment is reserved for non- L. iners dominant

293 Summary Discussion

women with a short cervix, while a cerclage or probiotic treatment may be more effective in the subpopulation abundant in L. iners. Larger studies are therefore required to confirm these preliminary findings. The work in this thesis has also highlighted the impact of cerclage suture material on the vaginal microenvironment and cervical phenotype. My findings demonstrate for the first time a significant association between use of braided suture and prematurely induced cervical vascularity, increased species diversity and adverse pregnancy outcome. These findings have significant clinical implications as braided suture material is predominately used for cerclage rather than a monofilament alternative, without an evidence base113. Although a causal link is yet to be established, this may potentially explain the increased rates of non-viable pregnancy associated with braided cerclage insertion, as well as the high incidence of puerperal sepsis 104, 107, 108. Importantly, the work presented in this thesis was not powered for the clinical outcome of preterm birth. A large clinical RCT specifically powered for outcomes of preterm birth and viability is therefore required to validate my findings. Since the commencement of my PhD thesis, the C-STITCH (Cerclage Suture Type for an Insufficient Cervix and its effect on Health outcomes) study, a multicenter NIHR funded study aimed at addressing the clinical impact of braided and monofilament suture materials has commenced recruitment. The findings of this study may justify a change in clinical practice away from braided suture material.

294 Summary Discussion

Strengths and limitations

Strengths Employing 16S rRNA gene sequencing to assess vaginal bacterial communities avoided the limitations associated with traditional culture based methods. This provided a comprehensive species-specific characterisation of high- and low-risk pregnancy cohorts. This was a major strength of the study as, in contrast to previous studies 181, it highlighted the clinical significance of individual Lactobacillus species rather than generalised dysbiosis, with respect to preterm birth risk. My work represents the largest characterisation of vaginal microbiota using 16S gene sequencing in a pregnant cohort at risk of preterm birth to date. The high-risk nature of the study population was demonstrated by high rates spontaneous preterm birth rates within the studies, ranging from 11% to 34%, compared to a background UK incidence of 7%1. Importantly, this enabled the differential characterisation of microbial profiles associated with clinically meaningful outcomes of early (<34 weeks) and late (34>37 weeks) preterm birth. Another strength of this project was the availability of transvaginal ultrasound scan data paired to 16S rRNA gene sequence data. Observation of cervical length data ensured identification of a truly high-risk population (those found to have a short cervix) in which to characterise and compare vaginal microbial composition. In particular, using CL thresholds of ≤25mm to define a targeted group for intervention (cerclage or progesterone) ensured an accurate reflection of clinical practice. Furthermore matched ultrasound and microbial data provided a unique illustration of the bacterial drivers of pathogenic cervical shortening; a finding likely to have significant clinical impact.

Limitations During the volumetric scan acquisition and data analysis, it became evident that operator experience with the VOCALTM software and programme is clearly advantageous. VOCALTM Doppler values are influenced by inconsistencies in the machine settings, movement artifact and operator technique 267. Operator technique in particular affected vascularity indices, where the distance between the probe and defined volume of interest can vary. As yet there are no internationally standardised or recommended machine settings recommended for use in 3D volume and power Doppler modes. GE Healthcare advice was therefore sought, and a standardised setting applied for all acquisitions of images. During analysis of the scans, delineation of the ‘region of interest’ using VOCALTM was also technically difficult, and volumetric values were dependent on the visualisation of clear tissue planes.

295 Summary Discussion

There was significant participation bias among women that consented to swabs as well as scans (only 31% of the low risk group, and 64% of the high risk group consented to both). Ethnicity and smoking status were two of the features that differentiated the cohorts consenting to ‘scans only’ or ‘swabs and scans’; non-smoking Caucasians were more likely to consent to the later, whereas smokers and Black women were less likely. Both ethnicity and smoking status are important factors influencing vaginal microbiome composition 155, 169. It is therefore possible that the presented data are skewed towards a non-dysbiotic population, simply through the demographics of participation. Future studies should attempt to identify reasons for lack of participation, potentially through questionnaires at the time of consent, to better understand barriers to research participation

Studies have shown that L. iners is often present together with dysbiosis-associated anaerobes, pathobionts (such as GBS, E. coli, Candida) 235, and pathogens (sexually transmitted pathogens, including the understudied Mycoplasma genitalium) 140, and is the first Lactobacillus to grow back after antibiotic treatment 265. This is in contrast to L. crispatus, which is not easily displaced and hardly ever occurs together with dysbiosis-associated anaerobes, pathobionts or pathogens 177, 227. A major limitation of this thesis is that these pathobionts were not taken into account: bacterial pathobionts and pathogens are often present in much lower abundance than Lactobacilli and anaerobes and are therefore not reliably identified by 16S sequencing, this is particularly true of GBS which is frequently underestimated 236. In my study, E. coli was detected in very few samples, and no significant shifts in the levels of S. agalactiae (Group B streptococcus) were observed. 16S sequencing also underestimated G. vaginalis and A. vaginae (bacteria central to bacterial vaginosis), which was highlighted through comparison of the same DNA samples analysed with targeted qPCR. Species-specific qPCR of GBS and E. coli 237 would most likely be provide better insight as to the role these pathobionts play in L. iners pathogenesis. Furthermore Candida is a yeast and is therefore not identified by 16S sequencing at all. Further work could address these issues through employing qPCR of stored DNA specimens from the described studies, and comparing them to existing sequence data. Employing qPCR may also address the possibility that bacterial load 268 (rather than type of bacteria) play a role in adverse pregnancy outcome, as sequence data simply represents relative abundance.

296 Summary Discussion

It is also possible that sequence reads, and therefore described microbial profiles were limited by the decision to target V1-V3 hypervariable regions for amplification in this study. V1-V3 primer sets were selected as it was considered that together, these regions provide a comprehensive differentiation of all bacterial species to the genus level, and in particular, good resolution of different Lactobacillus species 149. Others have employed V3-V4 primer sets 52, 53, 133, 149, 172, or even used V6 alone 266, as it this is thought these may provide better differentiation of certain gram-negative bacteria, as well as BV-associated bacteria 149. Currently there is no platform available for combined V1-V6 sequencing. The additional sequencing of V4-V6 within this project would have potentially improved the characterisation of dysbiosis, but cost and time restrictions prevented this. Additionally, as the primary focus was a comprehensive assessment of vaginal Lactobacilli communities, V1-V3 primers were deemed appropriate 149.

Further limitations relate to sample size among the studies. Pre-determined power calculations were largely not possible as the studies described were novel designs. In this regard, a limiting factor was the lack of cervical treatment women delivering early preterm (<34 weeks). Given the relatively rare incidence of preterm birth before 34 weeks (1% of pregnancies) 218, a much larger sample size would be required to investigate this. Similarly the primary limitation of the intervention studies was the small number of participants receiving progesterone (n=25) or cerclage (n=50), and the lack of an equivalent control group with a short CL ≤25mm, not receiving any intervention. This was an unavoidable limitation of this, and indeed any research study in human preterm birth, due a clinical and ethical obligation to intervene in women identified to have a short cervix.

A further limitation of this project was the lack of availability data detailing depth of the excisional treatment among CT women. This data would have been useful to correlate depth of excision and subsequent cervical volume in pregnancy. An attempt was made to collect this retrospectively, but proved to very difficult to ascertain for several reasons. Primarily this detail was not accurately recorded in procedural notes or histology reports, and where women had had their treatment elsewhere, hospitals either did not respond to information requests or could not locate the documentation required.

297 Summary Discussion

Final conclusions

The work presented in this thesis provides a novel description of the interaction between the cervical phenotype and vaginal microbiota in pregnancy. Additionally, the impact of intervention measures to prevent preterm birth, progesterone and cervical cerclage, have been described. My final conclusions from the work presented in this thesis are:

1. The addition of 3D-4D capabilities for transvaginal sonographic assessment of the cervix does not provide additional clinical benefit for preterm birth surveillance.

2. Individual Lactobacilli species play an integral role in determining pregnancy outcome in women at risk of preterm birth. Microbial profiles of women delivering early preterm differ substantially from those delivering late preterm, which most closely resembled term births. As such, detection of vaginal microbial composition in the early second trimester may be used to stratify preterm birth risk.

3. Cervical shortening is significantly associated with corresponding vaginal microbial dominance of L. iners in high-risk pregnancy. The interaction between the cervix, vaginal microbiota and subsequent preterm birth is largely influenced by underlying risk factors however; interpretation of risk given a short cervix or vaginal community state type should be tailored appropriately given a history of excisional cervical treatment or prior spontaneous preterm birth.

4. Progesterone supplementation maintains vaginal Lactobacillus spp. stability, although does not attenuate L. iners-associated preterm birth risk. The action of vaginal progesterone in the prevention of preterm birth is therefore unlikely to be via modulation of vaginal microbial communities.

5. Increased rates of non-viable pregnancy and preterm birth following cervical cerclage using braided suture may be driven largely by the suture material itself, which was shown to adversely impact upon the vaginal microbiome and its interplay with the maternal innate immune response. In contrast monofilament suture has minimal impact on the host microbiome or inflammation in pregnancy, and was associated with improved pregnancy outcome.

298 References

9 REFERENCES

1. ONS. Gestation-specific infant mortality in England and Wales, 2011. In: Statistics OfN, editor. www.ons.gov.uk/ons/child-health/gestation-specific-infant-mortality-in-englang-and- wales/20112013. 2. Martin JA, Hamilton B, Ventura S, Osterman M, Mathews MS. Births: Final Data for 2011. Natl Vital Stat Rep. 2013;62(1):1-90. 3. Slattery MM, Morrison JJ. Preterm delivery. Lancet. 2002;360(9344):1489-97. 4. Moore T, Hennessy EM, Myles J, Johnson SJ, Draper ES, Costeloe KL, et al. Neurological and developmental outcome in extremely preterm children born in England in 1995 and 2006: the EPICure studies. Bmj. 2012;345:e7961. 5. Statistics NM. Hospital Episode Statistics. Health & Social Care Information Centre: 2013. 6. Liu L, Oza S, Hogan D, Perin J, Rudan I, Lawn JE, et al. Global, regional, and national causes of child mortality in 2000-13, with projections to inform post-2015 priorities: an updated systematic analysis. Lancet. 2015;385(9966):430-40. 7. WHO WHO. Causes of child mortality who.int2015 [cited 2015 01.11.2015]. Available from: http://www.who.int/gho/child_health/mortality/causes/en/. 8. Costeloe K, Group ES. EPICure: facts and figures: why preterm labour should be treated. BJOG : an international journal of obstetrics and gynaecology. 2006;113 Suppl 3:10-2. 9. Mercer BM. Preterm premature rupture of the membranes. Obstet Gynecol. 2003;101(1):178-93. 10. Petrou S. The economic consequences of preterm birth during the first 10 years of life. BJOG : an international journal of obstetrics and gynaecology. 2005;112 Suppl 1:10-5. 11. Marlow N, Wolke D, Bracewell MA, Samara M, Group ES. Neurologic and developmental disability at six years of age after extremely preterm birth. N Engl J Med. 2005;352(1):9-19. 12. Wu YW, Colford JM, Jr. Chorioamnionitis as a risk factor for cerebral palsy: A meta- analysis. JAMA. 2000;284(11):1417-24. 13. Hagberg H, Mallard C, Ferriero DM, Vannucci SJ, Levison SW, Vexler ZS, et al. The role of inflammation in perinatal brain injury. Nat Rev Neurol. 2015;11(4):192-208. 14. Saigal S, Doyle LW. An overview of mortality and sequelae of preterm birth from infancy to adulthood. Lancet. 2008;371(9608):261-9. 15. McCormick MC, Litt JS, Smith VC, Zupancic JA. Prematurity: an overview and public health implications. Annu Rev Public Health. 2011;32:367-79. 16. Wray S. Uterine contraction and physiological mechanisms of modulation. Am J Physiol. 1993;264(1 Pt 1):C1-18. 17. Shojo H, Kaneko Y. Characterization and expression of oxytocin and the oxytocin receptor. Mol Genet Metab. 2000;71(4):552-8. 18. Romero R, Espinoza J, Goncalves LF, Kusanovic JP, Friel LA, Nien JK. Inflammation in preterm and term labour and delivery. Semin Fetal Neonatal Med. 2006;11(5):317-26. 19. Osman I, Young A, Ledingham MA, Thomson AJ, Jordan F, Greer IA, et al. Leukocyte density and pro-inflammatory cytokine expression in human fetal membranes, decidua, cervix and myometrium before and during labour at term. Mol Hum Reprod. 2003;9(1):41-5.

299 References

20. Slater D, Allport V, Bennett P. Changes in the expression of the type-2 but not the type-1 cyclo-oxygenase enzyme in chorion-decidua with the onset of labour. Br J Obstet Gynaecol. 1998;105(7):745-8. 21. Terzidou V, Sooranna SR, Kim LU, Thornton S, Bennett PR, Johnson MR. Mechanical stretch up-regulates the human oxytocin receptor in primary human uterine myocytes. J Clin Endocrinol Metab. 2005;90(1):237-46. 22. Hertelendy F, Zakar T. Prostaglandins and the myometrium and cervix. Prostaglandins Leukot Essent Fatty Acids. 2004;70(2):207-22. 23. Lissauer D, Eldershaw SA, Inman CF, Coomarasamy A, Moss PA, Kilby MD. Progesterone promotes maternal-fetal tolerance by reducing human maternal T-cell polyfunctionality and inducing a specific cytokine profile. European journal of immunology. 2015;45(10):2858-72. 24. Mesiano S. Myometrial progesterone responsiveness and the control of human parturition. J Soc Gynecol Investig. 2004;11(4):193-202. 25. Stjernholm Y, Sahlin L, Malmstrom A, Barchan K, Eriksson HA, Ekman G. Potential roles for gonadal steroids and insulin-like growth factor I during final cervical ripening. Obstet Gynecol. 1997;90(3):375-80. 26. Tan H, Yi L, Rote NS, Hurd WW, Mesiano S. Progesterone receptor-A and -B have opposite effects on proinflammatory gene expression in human myometrial cells: implications for progesterone actions in human pregnancy and parturition. J Clin Endocrinol Metab. 2012;97(5):E719-30. 27. Mesiano S, Wang Y, Norwitz ER. Progesterone receptors in the human pregnancy uterus: do they hold the key to birth timing? Reprod Sci. 2011;18(1):6-19. 28. Hardy DB, Janowski BA, Corey DR, Mendelson CR. Progesterone receptor plays a major antiinflammatory role in human myometrial cells by antagonism of nuclear factor-kappaB activation of cyclooxygenase 2 expression. Mol Endocrinol. 2006;20(11):2724-33. 29. Loudon JA, Elliott CL, Hills F, Bennett PR. Progesterone represses interleukin-8 and cyclo-oxygenase-2 in human lower segment fibroblast cells and amnion epithelial cells. Biol Reprod. 2003;69(1):331-7. 30. Anderson L, Martin W, Higgins C, Nelson SM, Norman JE. The effect of progesterone on myometrial contractility, potassium channels, and tocolytic efficacy. Reprod Sci. 2009;16(11):1052-61. 31. Carbonne B, Dallot E, Haddad B, Ferré F, Cabrol D. Effects of progesterone on prostaglandin E(2)-induced changes in glycosaminoglycan synthesis by human cervical fibroblasts in culture. Mol Hum Reprod. 2000;6(7):661-4. 32. Nold C, Maubert M, Anton L, Yellon S, Elovitz MA. Prevention of preterm birth by progestational agents: what are the molecular mechanisms? Am J Obstet Gynecol. 2013;208(3):223 e1-7. 33. Yellon SM, Dobyns AE, Beck HL, Kurtzman JT, Garfield RE, Kirby MA. Loss of progesterone receptor-mediated actions induce preterm cellular and structural remodeling of the cervix and premature birth. PLoS One. 2013;8(12):e81340. 34. Clark K, Ji H, Feltovich H, Janowski J, Carroll C, Chien EK. Mifepristone-induced cervical ripening: structural, biomechanical, and molecular events. Am J Obstet Gynecol. 2006;194(5):1391-8.

300 References

35. Word RA, Li XH, Hnat M, Carrick K. Dynamics of cervical remodeling during pregnancy and parturition: mechanisms and current concepts. Seminars in reproductive medicine. 2007;25(1):69-79. 36. Christiaens I, Zaragoza DB, Guilbert L, Robertson SA, Mitchell BF, Olson DM. Inflammatory processes in preterm and term parturition. Journal of Reproductive Immunology. 2008;79(1):50-7. 37. Nold C, Anton L, Brown A, Elovitz M. Inflammation promotes a cytokine response and disrupts the cervical epithelial barrier: a possible mechanism of premature cervical remodeling and preterm birth. Am J Obstet Gynecol. 2012;206(3):208.e1-7. 38. Ekman-Ordeberg G, Dubicke A. Preterm Cervical Ripening in humans. Facts Views Vis Obgyn. 2012;4(4):245-53. 39. Timmons B, Akins M, Mahendroo M. Cervical remodeling during pregnancy and parturition. Trends Endocrinol Metab. 2010;21(6):353-61. 40. Mowa CN, Jesmin S, Sakuma I, Usip S, Togashi H, Yoshioka M, et al. Characterization of vascular endothelial growth factor (VEGF) in the uterine cervix over pregnancy: effects of denervation and implications for cervical ripening. J Histochem Cytochem. 2004;52(12):1665- 74. 41. Nguyen BT, Minkiewicz V, McCabe E, Cecile J, Mowa CN. Vascular endothelial growth factor induces mRNA expression of pro-inflammatory factors in the uterine cervix of mice. Biomed Res. 2012;33(6):363-72. 42. Trowsdale J, Betz AG. Mother's little helpers: mechanisms of maternal-fetal tolerance. Nat Immunol. 2006;7(3):241-6. 43. Veenstra van Nieuwenhoven AL, Heineman MJ, Faas MM. The immunology of successful pregnancy. Hum Reprod Update. 2003;9(4):347-57. 44. Luppi P. How immune mechanisms are affected by pregnancy. Vaccine. 2003;21(24):3352-7. 45. Sykes L, MacIntyre DA, Yap XJ, Teoh TG, Bennett PR. The Th1:th2 dichotomy of pregnancy and preterm labour. Mediators Inflamm. 2012;2012:967629. 46. Romero R, Espinoza J, Kusanovic JP, Gotsch F, Hassan S, Erez O, et al. The preterm parturition syndrome. BJOG : an international journal of obstetrics and gynaecology. 2006;113 Suppl 3:17-42. 47. Romero R, Yeo L, Chaemsaithong P, Chaiworapongsa T, Hassan SS. Progesterone to prevent spontaneous preterm birth. Semin Fetal Neonatal Med. 2014;19(1):15-26. 48. Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes of preterm birth. The lancet. 2008;371(9606):75-84. 49. Romero R, Gonzalez R, Sepulveda W, Brandt F, Ramirez M, Sorokin Y, et al. Infection and labor. VIII. Microbial invasion of the amniotic cavity in patients with suspected cervical incompetence: prevalence and clinical significance. Am J Obstet Gynecol. 1992;167(4 Pt 1):1086-91. 50. Romero R, Dey SK, Fisher SJ. Preterm labor: one syndrome, many causes. Science. 2014;345(6198):760-5. 51. Nold C, Anton L, Brown A, Elovitz M. Inflammation promotes a cytokine response and disrupts the cervical epithelial barrier: a possible mechanism of premature cervical remodeling and preterm birth. Am J Obstet Gynecol. 2012;206(3):208 e1-7. 52. Koga K, Mor G. Toll-like receptors at the maternal-fetal interface in normal pregnancy and pregnancy disorders. Am J Reprod Immunol. 2010;63(6):587-600.

301 References

53. Anahtar MN, Byrne EH, Doherty KE, Bowman BA, Yamamoto HS, Soumillon M, et al. Cervicovaginal bacteria are a major modulator of host inflammatory responses in the female genital tract. Immunity. 2015;42(5):965-76. 54. Romero R, Hassan SS, Gajer P, Tarca AL, Fadrosh DW, Bieda J, et al. The vaginal microbiota of pregnant women who subsequently have spontaneous preterm labor and delivery and those with a normal delivery at term. Microbiome. 2014;2:18. 55. Girling JE, Hedger MP. Toll-like receptors in the gonads and reproductive tract: emerging roles in reproductive physiology and pathology. Immunol Cell Biol. 2007;85(6):481-9. 56. Loudon JA, Sooranna SR, Bennett PR, Johnson MR. Mechanical stretch of human uterine smooth muscle cells increases IL-8 mRNA expression and peptide synthesis. Mol Hum Reprod. 2004;10(12):895-9. 57. Mohan AR, Sooranna SR, Lindstrom TM, Johnson MR, Bennett PR. The effect of mechanical stretch on cyclooxygenase type 2 expression and activator protein-1 and nuclear factor-kappaB activity in human amnion cells. Endocrinology. 2007;148(4):1850-7. 58. Leguizamon G, Smith J, Younis H, Nelson DM, Sadovsky Y. Enhancement of amniotic cyclooxygenase type 2 activity in women with preterm delivery associated with twins or polyhydramnios. Am J Obstet Gynecol. 2001;184(2):117-22. 59. Statistics OfN. Gestation-specific Infant Mortality in England and Wales, 2011. In: Care HaS, editor. 2013. 60. Smith R, Smith JI, Shen X, Engel PJ, Bowman ME, McGrath SA, et al. Patterns of plasma corticotropin-releasing hormone, progesterone, estradiol, and estriol change and the onset of human labor. J Clin Endocrinol Metab. 2009;94(6):2066-74. 61. Chan YY, Jayaprakasan K, Zamora J, Thornton JG, Raine-Fenning N, Coomarasamy A. The prevalence of congenital uterine anomalies in unselected and high-risk populations: a systematic review. Hum Reprod Update. 2011;17(6):761-71. 62. Saravelos SH, Cocksedge KA, Li TC. Prevalence and diagnosis of congenital uterine anomalies in women with reproductive failure: a critical appraisal. Hum Reprod Update. 2008;14(5):415-29. 63. Airoldi J, Berghella V, Sehdev H, Ludmir J. Transvaginal ultrasonography of the cervix to predict preterm birth in women with uterine anomalies. Obstet Gynecol. 2005;106(3):553-6. 64. Fox NS, Roman AS, Stern EM, Gerber RS, Saltzman DH, Rebarber A. Type of congenital uterine anomaly and adverse pregnancy outcomes. The journal of maternal-fetal & neonatal medicine : the official journal of the European Association of Perinatal Medicine, the Federation of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2013. 65. Munoz N. Human papillomavirus and cancer: the epidemiological evidence. J Clin Virol. 2000;19(1-2):1-5. 66. Kyrgiou M, Koliopoulos G, Martin-Hirsch P, Arbyn M, Prendiville W, Paraskevaidis E. Obstetric outcomes after conservative treatment for intraepithelial or early invasive cervical lesions: systematic review and meta-analysis. Lancet. 2006;367(9509):489-98. 67. Arbyn M, Kyrgiou M, Simoens C, Raifu AO, Koliopoulos G, Martin-Hirsch P, et al. Perinatal mortality and other severe adverse pregnancy outcomes associated with treatment of cervical intraepithelial neoplasia: meta-analysis. Bmj. 2008;337:a1284. 68. Kyrgiou M, Mitra A, Arbyn M, Stasinou SM, Martin-Hirsch P, Bennett P, et al. Fertility and early pregnancy outcomes after treatment for cervical intraepithelial neoplasia: systematic review and meta-analysis. Bmj. 2014;349:g6192.

302 References

69. Albrechtsen S, Rasmussen S, Thoresen S, Irgens LM, Iversen OE. Pregnancy outcome in women before and after cervical conisation: population based cohort study. Bmj. 2008;337:a1343. 70. Noehr B, Jensen A, Frederiksen K, Tabor A, Kjaer SK. Loop electrosurgical excision of the cervix and subsequent risk for spontaneous preterm delivery: a population-based study of singleton deliveries during a 9-year period. Am J Obstet Gynecol. 2009;201(1):33.e1-6. 71. Bruinsma FJ, Quinn MA. The risk of preterm birth following treatment for precancerous changes in the cervix: a systematic review and meta-analysis. BJOG : an international journal of obstetrics and gynaecology. 2011;118(9):1031-41. 72. Castanon A, Brocklehurst P, Evans H, Peebles D, Singh N, Walker P, et al. Risk of preterm birth after treatment for cervical intraepithelial neoplasia among women attending colposcopy in England: retrospective-prospective cohort study. Bmj. 2012;345:e5174. 73. Kyrgiou M, Arbyn M, Martin-Hirsch P, Paraskevaidis E. Increased risk of preterm birth after treatment for CIN. Bmj. 2012;345:e5847. 74. Founta C, Arbyn M, Valasoulis G, Kyrgiou M, Tsili A, Martin-Hirsch P, et al. Proportion of excision and cervical healing after large loop excision of the transformation zone for cervical intraepithelial neoplasia. BJOG : an international journal of obstetrics and gynaecology. 2010;117(12):1468-74. 75. Noehr B, Jensen A, Frederiksen K, Tabor A, Kjaer SK. Depth of cervical cone removed by loop electrosurgical excision procedure and subsequent risk of spontaneous preterm delivery. Obstet Gynecol. 2009;114(6):1232-8. 76. Buhimschi CS, Schatz F, Krikun G, Buhimschi IA, Lockwood CJ. Novel insights into molecular mechanisms of abruption-induced preterm birth. Expert Rev Mol Med. 2010;12:e35. 77. Elovitz MA, Ascher-Landsberg J, Saunders T, Phillippe M. The mechanisms underlying the stimulatory effects of thrombin on myometrial smooth muscle. Am J Obstet Gynecol. 2000;183(3):674-81. 78. Hall DR. Abruptio placentae and disseminated intravascular coagulopathy. Semin Perinatol. 2009;33(3):189-95. 79. Peaceman AM, Andrews WW, Thorp JM, Cliver SP, Lukes A, Iams JD, et al. Fetal fibronectin as a predictor of preterm birth in patients with symptoms: a multicenter trial. Am J Obstet Gynecol. 1997;177(1):13-8. 80. Foster C, Shennan AH. Fetal fibronectin as a biomarker of preterm labor: a review of the literature and advances in its clinical use. Biomark Med. 2014;8(4):471-84. 81. Ridout A, Carter J, Shennan A. Clinical utility of quantitative fetal fibronectin in preterm labour. BJOG : an international journal of obstetrics and gynaecology. 2016. 82. Abbott DS, Hezelgrave NL, Seed PT, Norman JE, David AL, Bennett PR, et al. Quantitative fetal fibronectin to predict preterm birth in asymptomatic women at high risk. Obstet Gynecol. 2015;125(5):1168-76. 83. Iams JD, Goldenberg RL, Meis PJ, Mercer BM, Moawad A, Das A, et al. The length of the cervix and the risk of spontaneous premature delivery. National Institute of Child Health and Human Development Maternal Fetal Medicine Unit Network. N Engl J Med. 1996;334(9):567- 72. 84. Heath VC, Southall TR, Souka AP, Elisseou A, Nicolaides KH. Cervical length at 23 weeks of gestation: prediction of spontaneous preterm delivery. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 1998;12(5):312-7.

303 References

85. Berghella V, Baxter JK, Hendrix NW. Cervical assessment by ultrasound for preventing preterm delivery. The Cochrane database of systematic reviews. 2013;1:CD007235. 86. Crane JM, Hutchens D. Transvaginal sonographic measurement of cervical length to predict preterm birth in asymptomatic women at increased risk: a systematic review. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2008;31(5):579-87. 87. McManemy J, Cooke E, Amon E, Leet T. Recurrence risk for preterm delivery. Am J Obstet Gynecol. 2007;196(6):576.e1-6; discussion .e6-7. 88. Grimes-Dennis J, Berghella V. Cervical length and prediction of preterm delivery. Curr Opin Obstet Gynecol. 2007;19(2):191-5. 89. Berghella V, Roman A, Daskalakis C, Ness A, Baxter JK. Gestational age at cervical length measurement and incidence of preterm birth. Obstet Gynecol. 2007;110(2 Pt 1):311-7. 90. Kindinger LM, Poon LC, Cacciatore S, MacIntyre DA, Fox NS, Schuit E, et al. The effect of gestational age at cervical length measurements in the prediction of spontaneous preterm birth in twin pregnancies: an individual patient level meta-analysis. BJOG : an international journal of obstetrics and gynaecology. 2015. 91. Rovas L, Sladkevicius P, Strobel E, Valentin L. Intraobserver and interobserver reproducibility of three-dimensional gray-scale and power Doppler ultrasound examinations of the cervix in pregnant women. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2005;26(2):132-7. 92. Barber MA, Medina M, Cabrera F, Romero A, Valle L, Garcia-Hernandez JA. Cervical length vs VOCAL cervical volume for predicting pre-term delivery in asymptomatic women at 20-22 weeks' pregnancy. Journal of obstetrics and gynaecology : the journal of the Institute of Obstetrics and Gynaecology. 2012;32(7):648-51. 93. Park IY, Kwon JY, Hong SC, Choi HM, Kwon HS, Won HS, et al. Usefulness of cervical volume by three-dimensional ultrasound in identifying the risk for preterm birth. Ultrasound Med Biol. 2011;37(7):1039-45. 94. Jo YS, Jang DG, Kim N, Kim SJ, Lee G. Comparison of cervical parameters by three- dimensional ultrasound according to parity and previous delivery mode. Int J Med Sci. 2011;8(8):673-8. 95. De Diego R, Sabrià J, Vela A, Rodríguez D, Gómez MD. Role of 3-dimensional power Doppler sonography in differentiating pregnant women with threatened preterm labor from those with an asymptomatic short cervix. J Ultrasound Med. 2014;33(4):673-9. 96. Alcazar JL. Three-dimensional power Doppler derived vascular indices: what are we measuring and how are we doing it? Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2008;32(4):485-7. 97. Raine-Fenning N, Campbell B, Collier J, Brincat M, Johnson I. The reproducibility of endometrial volume acquisition and measurement with the VOCAL-imaging program. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2002;19(1):69-75. 98. Raine-Fenning NJ, Campbell BK, Clewes JS, Kendall NR, Johnson IR. The reliability of virtual organ computer-aided analysis (VOCAL) for the semiquantification of ovarian, endometrial and subendometrial perfusion. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2003;22(6):633- 9.

304 References

99. Rovas L, Sladkevicius P, Strobel E, Valentin L. Reference data representative of normal findings at three-dimensional power Doppler ultrasound examination of the cervix from 17 to 41 gestational weeks. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2006;28(6):761-7. 100. Basgul A, Kavak ZN, Bakirci N, Gokaslan H. Three-dimensional ultrasound power Doppler assessment of the cervix: comparison between nulliparas and multiparas. Journal of perinatal medicine. 2007;35(1):48-50. 101. Yilmaz NC, Yigiter AB, Kavak ZN, Durukan B, Gokaslan H. Longitudinal examination of cervical volume and vascularization changes during the antepartum and postpartum period using three-dimensional and power Doppler ultrasound. Journal of perinatal medicine. 2010;38(5):461-5. 102. McDonald IA. Incompetant cervix as a cause of recurrent abortion. BJOG. 1963;70(1):105-9. 103. Odibo AO, Berghella V, To MS, Rust OA, Althuisius SM, Nicolaides KH. Shirodkar versus McDonald cerclage for the prevention of preterm birth in women with short cervical length. American journal of perinatology. 2007;24(1):55-60. 104. Alfirevic Z, Stampalija T, Roberts D, Jorgensen AL. Cervical stitch (cerclage) for preventing preterm birth in singleton pregnancy. The Cochrane database of systematic reviews. 2012;4:CD008991. 105. Hein M, Valore EV, Helmig RB, Uldbjerg N, Ganz T. Antimicrobial factors in the cervical mucus plug. Am J Obstet Gynecol. 2002;187(1):137-44. 106. Ethicon.Inc. Ethicon Wound Closure Manual. In: Dunn DL, editor. www.ethicon.com2007. p. 1-119. 107. Israfil-Bayli F, Toozs-Hobson P, Lees C, Slack M, Ismail KM. Pregnancy outcome after elective cervical cerclage in relation to type of suture material used. Med Hypotheses. 2013;81(1):119-21. 108. Quinn M. Final report of the MRC/RCOG randomised controlled trial of cervical cerclage. Br J Obstet Gynaecol. 1993;100(12):1154-5. 109. Menacker F, Martin JA. Expanded health data from the new birth certificate, 2005. Natl Vital Stat Rep. 2008;56(13):1-24. 110. Al-Azemi M, Al-Qattan F, Omu A, Taher S, Al-Busiri N, Abdulaziz A. Changing trends in the obstetric indications for cervical cerclage. Journal of obstetrics and gynaecology : the journal of the Institute of Obstetrics and Gynaecology. 2003;23(5):507-11. 111. Berghella V, Rafael TJ, Szychowski JM, Rust OA, Owen J. Cerclage for short cervix on ultrasonography in women with singleton gestations and previous preterm birth: a meta-analysis. Obstet Gynecol. 2011;117(3):663-71. 112. Shennan A, To M. RCOG Green-top Guideline No. 60. Royal College of Obstetricians and Gynaecologists2011. 113. Israfil-Bayli F, Toozs-Hobson P, Lees C, Slack M, Daniels J, Vince A, et al. Cervical cerclage and type of suture material: a survey of UK consultants' practice. The journal of maternal-fetal & neonatal medicine : the official journal of the European Association of Perinatal Medicine, the Federation of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2014;27(15):1584-8. 114. Berghella V, Szychowski JM, Owen J, Hankins G, Iams JD, Sheffield JS, et al. Suture type and ultrasound-indicated cerclage efficacy. The journal of maternal-fetal & neonatal medicine : the official journal of the European Association of Perinatal Medicine, the Federation

305 References

of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2012;25(11):2287-90. 115. Israfil-Bayli F, Toozs-Hobson P, Lees C, Slack M, Ismail K. Cerclage outcome by the type of suture material (COTS): study protocol for a pilot and feasibility randomised controlled trial. Trials. 2014;15:415. 116. Conde-Agudelo A, Romero R, Nicolaides K, Chaiworapongsa T, O'Brien JM, Cetingoz E, et al. Vaginal progesterone vs. cervical cerclage for the prevention of preterm birth in women with a sonographic short cervix, previous preterm birth, and singleton gestation: a systematic review and indirect comparison metaanalysis. Am J Obstet Gynecol. 2013;208(1):42 e1- e18. 117. Romero R, Nicolaides K, Conde-Agudelo A, Tabor A, O'Brien JM, Cetingoz E, et al. Vaginal progesterone in women with an asymptomatic sonographic short cervix in the midtrimester decreases preterm delivery and neonatal morbidity: a systematic review and metaanalysis of individual patient data. Am J Obstet Gynecol. 2012;206(2):124 e1-19. 118. Roberts D, Dalziel S. Antenatal corticosteroids for accelerating fetal lung maturation for women at risk of preterm birth. The Cochrane database of systematic reviews. 2006(3):CD004454. 119. Fonseca EB, Celik E, Parra M, Singh M, Nicolaides KH, Group FMFSTS. Progesterone and the risk of preterm birth among women with a short cervix. N Engl J Med. 2007;357(5):462- 9. 120. Hassan SS, Romero R, Vidyadhari D, Fusey S, Baxter JK, Khandelwal M, et al. Vaginal progesterone reduces the rate of preterm birth in women with a sonographic short cervix: a multicenter, randomized, double-blind, placebo-controlled trial. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2011;38(1):18-31. 121. van Os MA, van der Ven AJ, Kleinrouweler CE, Schuit E, Kazemier BM, Verhoeven CJ, et al. Preventing Preterm Birth with Progesterone in Women with a Short Cervical Length from a Low-Risk Population: A Multicenter Double-Blind Placebo-Controlled Randomized Trial. American journal of perinatology. 2015;32(10):993-1000. 122. O'Brien JM, Defranco EA, Adair CD, Lewis DF, Hall DR, How H, et al. Effect of progesterone on cervical shortening in women at risk for preterm birth: secondary analysis from a multinational, randomized, double-blind, placebo-controlled trial. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2009;34(6):653-9. 123. Grobman WA, Thom EA, Spong CY, Iams JD, Saade GR, Mercer BM, et al. 17 alpha- hydroxyprogesterone caproate to prevent prematurity in nulliparas with cervical length less than 30 mm. Am J Obstet Gynecol. 2012;207(5):390.e1-8. 124. Norman JE, Marlow N, Messow CM, Shennan A, Bennett PR, Thornton S, et al. Vaginal progesterone prophylaxis for preterm birth (the OPPTIMUM study): a multicentre, randomised, double-blind trial. Lancet. 2016. 125. Hassan S, Romero R, Vidyadhari D, Fusey S, Baxter J, Khandelwal M, et al. Vaginal progesterone reduces the rate of preterm birth in women with a sonographic short cervix: a multicenter, randomized, double‐blind, placebo‐controlled trial. Ultrasound in Obstetrics & Gynecology. 2011;38(1):18-31. 126. O'Brien JM, Adair CD, Lewis DF, Hall DR, Defranco EA, Fusey S, et al. Progesterone vaginal gel for the reduction of recurrent preterm birth: primary results from a randomized, double-blind, placebo-controlled trial. Ultrasound in obstetrics & gynecology : the official

306 References

journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2007;30(5):687- 96. 127. Gyetvai K, Hannah ME, Hodnett ED, Ohlsson A. Tocolytics for preterm labor: a systematic review. Obstet Gynecol. 1999;94(5 Pt 2):869-77. 128. Haas DM, Caldwell DM, Kirkpatrick P, McIntosh JJ, Welton NJ. Tocolytic therapy for preterm delivery: systematic review and network meta-analysis. Bmj. 2012;345:e6226. 129. Kenyon SL, Taylor DJ, Tarnow-Mordi W, Group OC. Broad-spectrum antibiotics for preterm, prelabour rupture of fetal membranes: the ORACLE I randomised trial. ORACLE Collaborative Group. Lancet. 2001;357(9261):979-88. 130. Kenyon S, Pike K, Jones DR, Brocklehurst P, Marlow N, Salt A, et al. Childhood outcomes after prescription of antibiotics to pregnant women with preterm rupture of the membranes: 7-year follow-up of the ORACLE I trial. Lancet. 2008;372(9646):1310-8. 131. Kenyon SL, Taylor DJ, Tarnow-Mordi W, Group OC. Broad-spectrum antibiotics for spontaneous preterm labour: the ORACLE II randomised trial. ORACLE Collaborative Group. Lancet. 2001;357(9261):989-94. 132. Kenyon S, Pike K, Jones DR, Brocklehurst P, Marlow N, Salt A, et al. Childhood outcomes after prescription of antibiotics to pregnant women with spontaneous preterm labour: 7-year follow-up of the ORACLE II trial. Lancet. 2008;372(9646):1319-27. 133. Srinivasan S, Hoffman NG, Morgan MT, Matsen FA, Fiedler TL, Hall RW, et al. Bacterial communities in women with bacterial vaginosis: high resolution phylogenetic analyses reveal relationships of microbiota to clinical criteria. PLoS One. 2012;7(6):e37818. 134. Machado A, Cerca N. Influence of Biofilm Formation by Gardnerella vaginalis and Other Anaerobes on Bacterial Vaginosis. J Infect Dis. 2015;212(12):1856-61. 135. Amsel R, Totten PA, Spiegel CA, Chen KC, Eschenbach D, Holmes KK. Nonspecific vaginitis. Diagnostic criteria and microbial and epidemiologic associations. Am J Med. 1983;74(1):14-22. 136. Nugent RP, Krohn MA, Hillier SL. Reliability of diagnosing bacterial vaginosis is improved by a standardized method of gram stain interpretation. Journal of clinical microbiology. 1991;29(2):297-301. 137. Klebanoff MA, Schwebke JR, Zhang J, Nansel TR, Yu KF, Andrews WW. Vulvovaginal symptoms in women with bacterial vaginosis. Obstet Gynecol. 2004;104(2):267-72. 138. Haggerty CL, Hillier SL, Bass DC, Ness RB, Evaluation PID, Clinical Health study i. Bacterial vaginosis and anaerobic bacteria are associated with endometritis. Clin Infect Dis. 2004;39(7):990-5. 139. Brotman RM, Klebanoff MA, Nansel TR, Yu KF, Andrews WW, Zhang J, et al. Bacterial vaginosis assessed by gram stain and diminished colonization resistance to incident gonococcal, chlamydial, and trichomonal genital infection. J Infect Dis. 2010;202(12):1907-15. 140. Borgdorff H, Tsivtsivadze E, Verhelst R, Marzorati M, Jurriaans S, Ndayisaba GF, et al. Lactobacillus-dominated cervicovaginal microbiota associated with reduced HIV/STI prevalence and genital HIV viral load in African women. Isme Journal. 2014;8(9):1781-93. 141. Low N, Chersich MF, Schmidlin K, Egger M, Francis SC, van de Wijgert JH, et al. Intravaginal practices, bacterial vaginosis, and HIV infection in women: individual participant data meta-analysis. PLoS Med. 2011;8(2):e1000416. 142. Guaschino S, De Seta F, Piccoli M, Maso G, Alberico S. Aetiology of preterm labour: bacterial vaginosis. BJOG : an international journal of obstetrics and gynaecology. 2006;113 Suppl 3:46-51.

307 References

143. Donders GG, Van Calsteren K, Bellen G, Reybrouck R, Van den Bosch T, Riphagen I, et al. Predictive value for preterm birth of abnormal vaginal flora, bacterial vaginosis and aerobic vaginitis during the first trimester of pregnancy. BJOG : an international journal of obstetrics and gynaecology. 2009;116(10):1315-24. 144. Hay PE. Therapy of bacterial vaginosis. J Antimicrob Chemother. 1998;41(1):6-9. 145. Hay P. Recurrent Bacterial Vaginosis. Curr Infect Dis Rep. 2000;2(6):506-12. 146. Bradshaw CS, Morton AN, Hocking J, Garland SM, Morris MB, Moss LM, et al. High recurrence rates of bacterial vaginosis over the course of 12 months after oral metronidazole therapy and factors associated with recurrence. J Infect Dis. 2006;193(11):1478-86. 147. Wang B, Xiao BB, Shang CG, Wang K, Na RS, Nu XX, et al. Molecular analysis of the relationship between specific vaginal bacteria and bacterial vaginosis metronidazole therapy failure. Eur J Clin Microbiol Infect Dis. 2014;33(10):1749-56. 148. Lamont RF, Nhan-Chang CL, Sobel JD, Workowski K, Conde-Agudelo A, Romero R. Treatment of abnormal vaginal flora in early pregnancy with clindamycin for the prevention of spontaneous preterm birth: a systematic review and metaanalysis. Am J Obstet Gynecol. 2011;205(3):177-90. 149. Chakravorty S, Helb D, Burday M, Connell N, Alland D. A detailed analysis of 16S ribosomal RNA gene segments for the diagnosis of pathogenic bacteria. J Microbiol Methods. 2007;69(2):330-9. 150. Ma J, Prince A, Aagaard KM. Use of whole genome shotgun metagenomics: a practical guide for the microbiome-minded physician scientist. Seminars in reproductive medicine. 2014;32(1):5-13. 151. Turnbaugh PJ, Ley RE, Hamady M, Fraser-Liggett CM, Knight R, Gordon JI. The human microbiome project. Nature. 2007;449(7164):804-10. 152. Yatsunenko T, Rey FE, Manary MJ, Trehan I, Dominguez-Bello MG, Contreras M, et al. Human gut microbiome viewed across age and geography. Nature. 2012;486(7402):222-7. 153. Lozupone CA, Stombaugh JI, Gordon JI, Jansson JK, Knight R. Diversity, stability and resilience of the human gut microbiota. Nature. 2012;489(7415):220-30. 154. Hickey RJ, Zhou X, Pierson JD, Ravel J, Forney LJ. Understanding vaginal microbiome complexity from an ecological perspective. Transl Res. 2012;160(4):267-82. 155. Ravel J, Gajer P, Abdo Z, Schneider GM, Koenig SS, McCulle SL, et al. Vaginal microbiome of reproductive-age women. Proc Natl Acad Sci U S A. 2011;108 Suppl 1:4680-7. 156. Reid G, Younes JA, Van der Mei HC, Gloor GB, Knight R, Busscher HJ. Microbiota restoration: natural and supplemented recovery of human microbial communities. Nat Rev Microbiol. 2011;9(1):27-38. 157. Linhares IM, Summers PR, Larsen B, Giraldo PC, Witkin SS. Contemporary perspectives on vaginal pH and lactobacilli. Am J Obstet Gynecol. 2011;204(2):120 e1-5. 158. Brotman RM. Vaginal microbiome and sexually transmitted infections: an epidemiologic perspective. J Clin Invest. 2011;121(12):4610-7. 159. Borgdorff H, Tsivtsivadze E, Verhelst R, Marzorati M, Jurriaans S, Ndayisaba GF, et al. Lactobacillus-dominated cervicovaginal microbiota associated with reduced HIV/STI prevalence and genital HIV viral load in African women. ISME J. 2014;8(9):1781-93. 160. Atashili J, Poole C, Ndumbe PM, Adimora AA, Smith JS. Bacterial vaginosis and HIV acquisition: a meta-analysis of published studies. AIDS. 2008;22(12):1493-501.

308 References

161. King CC, Jamieson DJ, Wiener J, Cu-Uvin S, Klein RS, Rompalo AM, et al. Bacterial vaginosis and the natural history of human papillomavirus. Infect Dis Obstet Gynecol. 2011;2011:319460. 162. Gillet E, Meys JF, Verstraelen H, Verhelst R, De Sutter P, Temmerman M, et al. Association between bacterial vaginosis and cervical intraepithelial neoplasia: systematic review and meta-analysis. PLoS One. 2012;7(10):e45201. 163. Guo YL, You K, Qiao J, Zhao YM, Geng L. Bacterial vaginosis is conducive to the persistence of HPV infection. Int J STD AIDS. 2012;23(8):581-4. 164. Farage M, Maibach H. Lifetime changes in the vulva and vagina. Arch Gynecol Obstet. 2006;273(4):195-202. 165. Spear GT, French AL, Gilbert D, Zariffard MR, Mirmonsef P, Sullivan TH, et al. Human alpha-amylase present in lower-genital-tract mucosal fluid processes glycogen to support vaginal colonization by Lactobacillus. J Infect Dis. 2014;210(7):1019-28. 166. Gajer P, Brotman RM, Bai G, Sakamoto J, Schutte UM, Zhong X, et al. Temporal dynamics of the human vaginal microbiota. Sci Transl Med. 2012;4(132):132ra52. 167. Chaban B, Links MG, Jayaprakash TP, Wagner EC, Bourque DK, Lohn Z, et al. Characterization of the vaginal microbiota of healthy Canadian women through the menstrual cycle. Microbiome. 2014;2:23. 168. Brotman RM, Shardell MD, Gajer P, Fadrosh D, Chang K, Silver MI, et al. Association between the vaginal microbiota, menopause status, and signs of vulvovaginal atrophy. Menopause. 2014;21(5):450-8. 169. Brotman RM, He X, Gajer P, Fadrosh D, Sharma E, Mongodin EF, et al. Association between cigarette smoking and the vaginal microbiota: a pilot study. Bmc Infectious Diseases. 2014;14. 170. Zhou X, Brown CJ, Abdo Z, Davis CC, Hansmann MA, Joyce P, et al. Differences in the composition of vaginal microbial communities found in healthy Caucasian and black women. ISME J. 2007;1(2):121-33. 171. MacIntyre DA, Chandiramani M, Lee YS, Kindinger L, Smith A, Angelopoulos N, et al. The vaginal microbiome during pregnancy and the postpartum period in a European population. Sci Rep. 2015;5:8988. 172. Aagaard K, Riehle K, Ma J, Segata N, Mistretta TA, Coarfa C, et al. A metagenomic approach to characterization of the vaginal microbiome signature in pregnancy. PLoS One. 2012;7(6):e36466. 173. Hyman RW, Fukushima M, Jiang H, Fung E, Rand L, Johnson B, et al. Diversity of the Vaginal Microbiome Correlates With Preterm Birth. Reprod Sci. 2013. 174. Romero R, Hassan SS, Gajer P, Tarca AL, Fadrosh DW, Nikita L, et al. The composition and stability of the vaginal microbiota of normal pregnant women is different from that of non- pregnant women. Microbiome. 2014;2(1):4. 175. Walther-António MR, Jeraldo P, Berg Miller ME, Yeoman CJ, Nelson KE, Wilson BA, et al. Pregnancy's Stronghold on the Vaginal Microbiome. PLoS One. 2014;9(6):e98514. 176. Africa CW, Nel J, Stemmet M. Anaerobes and bacterial vaginosis in pregnancy: virulence factors contributing to vaginal colonisation. Int J Environ Res Public Health. 2014;11(7):6979-7000. 177. Verstraelen H, Verhelst R, Claeys G, De Backer E, Temmerman M, Vaneechoutte M. Longitudinal analysis of the vaginal microflora in pregnancy suggests that L. crispatus promotes

309 References

the stability of the normal vaginal microflora and that L. gasseri and/or L. iners are more conducive to the occurrence of abnormal vaginal microflora. BMC Microbiol. 2009;9:116. 178. Jefferson KK. The bacterial etiology of preterm birth. Adv Appl Microbiol. 2012;80:1- 22. 179. Witkin SS. The vaginal microbiome, vaginal anti-microbial defence mechanisms and the clinical challenge of reducing infection-related preterm birth. BJOG : an international journal of obstetrics and gynaecology. 2015;122(2):213-8. 180. Sakai M, Ishiyama A, Tabata M, Sasaki Y, Yoneda S, Shiozaki A, et al. Relationship between cervical mucus interleukin-8 concentrations and vaginal bacteria in pregnancy. Am J Reprod Immunol. 2004;52(2):106-12. 181. DiGiulio DB, Callahan BJ, McMurdie PJ, Costello EK, Lyell DJ, Robaczewska A, et al. Temporal and spatial variation of the human microbiota during pregnancy. Proc Natl Acad Sci U S A. 2015;112(35):11060-5. 182. Petricevic L, Domig KJ, Nierscher FJ, Sandhofer MJ, Fidesser M, Krondorfer I, et al. Characterisation of the vaginal Lactobacillus microbiota associated with preterm delivery. Sci Rep. 2014;4:5136. 183. Aagaard K, Ma J, Antony KM, Ganu R, Petrosino J, Versalovic J. The placenta harbors a unique microbiome. Sci Transl Med. 2014;6(237):237ra65. 184. Romero R, Gomez R, Chaiworapongsa T, Conoscenti G, Kim JC, Kim YM. The role of infection in preterm labour and delivery. Paediatr Perinat Epidemiol. 2001;15 Suppl 2:41-56. 185. Melville JM, Moss TJ. The immune consequences of preterm birth. Front Neurosci. 2013;7:79. 186. Inomata K, Mizobuchi M, Tanaka S, Iwatani S, Sakai H, Yoshimoto S, et al. Patterns of increases in interleukin-6 and C-reactive protein as predictors for white matter injury in preterm infants. Pediatr Int. 2014;56(6):851-5. 187. Merrifield CA, Lewis MC, Berger B, Cloarec O, Heinzmann SS, Charton F, et al. Neonatal environment exerts a sustained influence on the development of the intestinal microbiota and metabolic phenotype. ISME J. 2015. 188. Dominguez-Bello MG, Costello EK, Contreras M, Magris M, Hidalgo G, Fierer N, et al. Delivery mode shapes the acquisition and structure of the initial microbiota across multiple body habitats in newborns. Proc Natl Acad Sci U S A. 2010;107(26):11971-5. 189. Sim K, Shaw AG, Randell P, Cox MJ, McClure ZE, Li MS, et al. Dysbiosis anticipating necrotizing enterocolitis in very premature infants. Clin Infect Dis. 2015;60(3):389-97. 190. Arrieta MC, Stiemsma LT, Dimitriu PA, Thorson L, Russell S, Yurist-Doutsch S, et al. Early infancy microbial and metabolic alterations affect risk of childhood asthma. Sci Transl Med. 2015;7(307):307ra152. 191. Bager P, Wohlfahrt J, Westergaard T. Caesarean delivery and risk of atopy and allergic disease: meta-analyses. Clin Exp Allergy. 2008;38(4):634-42. 192. Kagan KO, Sonek J. How to measure cervical length. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2015;45(3):358-62. 193. Bega G, Lev-Toaff A, Kuhlman K, Berghella V, Parker L, Goldberg B, et al. Three- dimensional multiplanar transvaginal ultrasound of the cervix in pregnancy. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2000;16(4):351-8.

310 References

194. Farrell T, Leslie JR, Chien PF, Agustsson P. The reliability and validity of three dimensional ultrasound volumetric measurements using an in vitro balloon and in vivo uterine model. BJOG : an international journal of obstetrics and gynaecology. 2001;108(6):573-82. 195. Lane DJ. 16S/23S rRNA sequencing. New York: Wiley; 1991. 196. Sundquist A, Bigdeli S, Jalili R, Druzin ML, Waller S, Pullen KM, et al. Bacterial flora- typing with targeted, chip-based Pyrosequencing. BMC Microbiol. 2007;7:108. 197. Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE, et al. A core gut microbiome in obese and lean twins. Nature. 2009;457(7228):480-4. 198. Kozich JJ, Westcott SL, Baxter NT, Highlander SK, Schloss PD. Development of a dual- index sequencing strategy and curation pipeline for analyzing amplicon sequence data on the MiSeq Illumina sequencing platform. Appl Environ Microbiol. 2013;79(17):5112-20. 199. Wang Q, Garrity GM, Tiedje JM, Cole JR. Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Appl Environ Microbiol. 2007;73(16):5261-7. 200. Edgar RC. Search and clustering orders of magnitude faster than BLAST. Bioinformatics. 2010;26(19):2460-1. 201. Parks DH, Beiko RG. Identifying biologically relevant differences between metagenomic communities. Bioinformatics. 2010;26(6):715-21. 202. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, et al. Metagenomic biomarker discovery and explanation. Genome Biol. 2011;12(6):R60. 203. Fredricks DN, Fiedler TL, Thomas KK, Mitchell CM, Marrazzo JM. Changes in vaginal bacterial concentrations with intravaginal metronidazole therapy for bacterial vaginosis as assessed by quantitative PCR. Journal of clinical microbiology. 2009;47(3):721-6. 204. Zozaya-Hinchliffe M, Lillis R, Martin DH, Ferris MJ. Quantitative PCR assessments of bacterial species in women with and without bacterial vaginosis. Journal of clinical microbiology. 2010;48(5):1812-9. 205. Elshal MF, McCoy JP. Multiplex bead array assays: performance evaluation and comparison of sensitivity to ELISA. Methods. 2006;38(4):317-23. 206. Dubicke A, Fransson E, Centini G, Andersson E, Byström B, Malmström A, et al. Pro- inflammatory and anti-inflammatory cytokines in human preterm and term cervical ripening. Journal of reproductive immunology. 2010;84(2):176-85. 207. Whitcomb BW, Schisterman EF, Luo X, Chegini N. Maternal serum granulocyte colony- stimulating factor levels and spontaneous preterm birth. J Womens Health (Larchmt). 2009;18(1):73-8. 208. Eubank TD, Roberts R, Galloway M, Wang Y, Cohn DE, Marsh CB. GM-CSF induces expression of soluble VEGF receptor-1 from human monocytes and inhibits angiogenesis in mice. Immunity. 2004;21(6):831-42. 209. Chandiramani M, Seed PT, Orsi NM, Ekbote UV, Bennett PR, Shennan AH, et al. Limited relationship between cervico-vaginal fluid cytokine profiles and cervical shortening in women at high risk of spontaneous preterm birth. PLoS One. 2012;7(12):e52412. 210. Bae J, Park D, Lee YS, Jeoung D. Interleukin-2 promotes angiogenesis by activation of Akt and increase of ROS. J Microbiol Biotechnol. 2008;18(2):377-82. 211. Chatterjee P, Chiasson VL, Bounds KR, Mitchell BM. Regulation of the Anti- Inflammatory Cytokines Interleukin-4 and Interleukin-10 during Pregnancy. Front Immunol. 2014;5:253.

311 References

212. Taylor BD, Holzman CB, Fichorova RN, Tian Y, Jones NM, Fu W, et al. Inflammation biomarkers in vaginal fluid and preterm delivery. Hum Reprod. 2013;28(4):942-52. 213. Sakai M, Sasaki Y, Yoneda S, Kasahara T, Arai T, Okada M, et al. Elevated interleukin-8 in cervical mucus as an indicator for treatment to prevent premature birth and preterm, pre-labor rupture of membranes: a prospective study. Am J Reprod Immunol. 2004;51(3):220-5. 214. Chen X, Scholl TO. Maternal biomarkers of endothelial dysfunction and preterm delivery. PLoS One. 2014;9(1):e85716. 215. Dubicke A, Akerud A, Sennstrom M, Hamad RR, Bystrom B, Malmstrom A, et al. Different secretion patterns of matrix metalloproteinases and IL-8 and effect of corticotropin- releasing hormone in preterm and term cervical fibroblasts. Mol Hum Reprod. 2008;14(11):641- 7. 216. Hamilton SA, Tower CL, Jones RL. Identification of chemokines associated with the recruitment of decidual leukocytes in human labour: potential novel targets for preterm labour. PLoS One. 2013;8(2):e56946. 217. Fortunato SJ, Menon R, Lombardi SJ. Role of tumor necrosis factor-alpha in the premature rupture of membranes and preterm labor pathways. Am J Obstet Gynecol. 2002;187(5):1159-62. 218. Poon LC, Savvas M, Zamblera D, Skyfta E, Nicolaides KH. Large loop excision of transformation zone and cervical length in the prediction of spontaneous preterm delivery. BJOG : an international journal of obstetrics and gynaecology. 2012;119(6):692-8. 219. Kindinger L, Teoh T. Preterm delivery – who is most at risk? An audit of a preterm surveillance clinic. BJOG : an international journal of obstetrics and gynaecology. 2013;120(s3):50. 220. Guzman ER, Houlihan C, Vintzileos A, Ivan J, Benito C, Kappy K. The significance of transvaginal ultrasonographic evaluation of the cervix in women treated with emergency cerclage. Am J Obstet Gynecol. 1996;175(2):471-6. 221. O'Brien JM, Hill AL, Barton JR. Funneling to the stitch: an informative ultrasonographic finding after cervical cerclage. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2002;20(3):252-5. 222. Song RK, Cha HH, Shin MY, Choi SJ, Oh SY, Kim JH, et al. Post-cerclage ultrasonographic cervical length can predict preterm delivery in elective cervical cerclage patients. Obstet Gynecol Sci. 2016;59(1):17-23. 223. Mirmonsef P, Krass L, Landay A, Spear GT. The role of bacterial vaginosis and trichomonas in HIV transmission across the female genital tract. Curr HIV Res. 2012;10(3):202- 10. 224. Guzeloglu-Kayisli O, Kayisli UA, Semerci N, Basar M, Buchwalder LF, Buhimschi CS, et al. Mechanisms of chorioamnionitis-associated preterm birth: interleukin-1beta inhibits progesterone receptor expression in decidual cells. J Pathol. 2015. 225. Alexander GR, Kogan M, Bader D, Carlo W, Allen M, Mor J. US birth weight/gestational age-specific neonatal mortality: 1995-1997 rates for whites, hispanics, and blacks. Pediatrics. 2003;111(1):e61-6. 226. Fettweis JM, Brooks JP, Serrano MG, Sheth NU, Girerd PH, Edwards DJ, et al. Differences in vaginal microbiome in African American women versus women of European ancestry. Microbiology. 2014;160(Pt 10):2272-82. 227. Jespers V, de Wijgert JV, Cools P, Verhelst R, Verstraelen H, Delany-Moretlwe S, et al. The significance of Lactobacillus crispatus and L-vaginalis for vaginal health and the negative

312 References

effect of recent sex: a cross-sectional descriptive study across groups of African women. Bmc Infectious Diseases. 2015;15. 228. Ghartey JP, Smith BC, Chen Z, Buckley N, Lo Y, Ratner AJ, et al. Lactobacillus crispatus dominant vaginal microbiome is associated with inhibitory activity of female genital tract secretions against Escherichia coli. PLoS One. 2014;9(5):e96659. 229. Borgdorff H, Armstrong SD, Tytgat HL, Xia D, Ndayisaba GF, Wastling JM, et al. Unique Insights in the Cervicovaginal Lactobacillus iners and L. crispatus Proteomes and Their Associations with Microbiota Dysbiosis. PLoS One. 2016;11(3):e0150767. 230. Witkin SS, Mendes-Soares H, Linhares IM, Jayaram A, Ledger WJ, Forney LJ. Influence of vaginal bacteria and D- and L-lactic acid isomers on vaginal extracellular matrix metalloproteinase inducer: implications for protection against upper genital tract infections. MBio. 2013;4(4). 231. Wilks M, Wiggins R, Whiley A, Hennessy E, Warwick S, Porter H, et al. Identification and H2O2 production of vaginal lactobacilli from pregnant women at high risk of preterm birth and relation with outcome. J Clin Microbiol. 2004;42(2):713-7. 232. Doerflinger SY, Throop AL, Herbst-Kralovetz MM. Bacteria in the vaginal microbiome alter the innate immune response and barrier properties of the human vaginal epithelia in a species-specific manner. J Infect Dis. 2014;209(12):1989-99. 233. Macklaim JM, Fernandes AD, Di Bella JM, Hammond JA, Reid G, Gloor GB. Comparative meta-RNA-seq of the vaginal microbiota and differential expression by Lactobacillus iners in health and dysbiosis. Microbiome. 2013;1(1):12. 234. Tamrakar R, Yamada T, Furuta I, Cho K, Morikawa M, Yamada H, et al. Association between Lactobacillus species and bacterial vaginosis-related bacteria, and bacterial vaginosis scores in pregnant Japanese women. BMC Infect Dis. 2007;7:128. 235. Cools P, Jespers V, Hardy L, Crucitti T, Delany-Moretlwe S, Mwaura M, et al. A Multi- Country Cross-Sectional Study of Vaginal Carriage of Group B Streptococci (GBS) and Escherichia coli in Resource-Poor Settings: Prevalences and Risk Factors. PLoS One. 2016;11(1):e0148052. 236. van de Wijgert J, Borgdorff H, Verhelst R, Crucitti T, Francis S, Verstraelen H, et al. The Vaginal Microbiota: What Have We Learned after a Decade of Molecular Characterization? Plos One. 2014;9(8). 237. van de Wijgert JH, Jespers V. Incorporating microbiota data into epidemiologic models: examples from vaginal microbiota research. Ann Epidemiol. 2016;26(5):360-5. 238. Martius JA, Steck T, Oehler MK, Wulf KH. Risk factors associated with preterm (<37+0 weeks) and early preterm birth (<32+0 weeks): univariate and multivariate analysis of 106 345 singleton births from the 1994 statewide perinatal survey of Bavaria. Eur J Obstet Gynecol Reprod Biol. 1998;80(2):183-9. 239. UK CR. Cervical Cancer: European Age-Standardised Incidence Rates per 100,000 Population, Females, Great Britain: 1979-2013 [cited 2016 June]. Available from: http://www.cancerresearchuk.org/sites/default/files/cstream-node/inc_asr_gb_cervix_2.pdf. 240. Ortoft G, Henriksen T, Hansen E, Petersen L. After conisation of the cervix, the perinatal mortality as a result of preterm delivery increases in subsequent pregnancy. BJOG. 2010;117(3):258-67. 241. Carcopino X, Maycock JA, Mancini J, Jeffers M, Farrar K, Martin M, et al. Image assessment of cervical dimensions after LLETZ: a prospective observational study. BJOG : an international journal of obstetrics and gynaecology. 2013;120(4):472-8.

313 References

242. Khalid S, Dimitriou E, Conroy R, Paraskevaidis E, Kyrgiou M, Harrity C, et al. The thickness and volume of LLETZ specimens can predict the relative risk of pregnancy-related morbidity. Bjog. 2012;119(6):685-91. 243. Racicot K, Cardenas I, Wunsche V, Aldo P, Guller S, Means RE, et al. Viral infection of the pregnant cervix predisposes to ascending bacterial infection. J Immunol. 2013;191(2):934- 41. 244. Brotman RM, Shardell MD, Gajer P, Tracy JK, Zenilman JM, Ravel J, et al. Interplay between the temporal dynamics of the vaginal microbiota and human papillomavirus detection. J Infect Dis. 2014;210(11):1723-33. 245. Mitra A, MacIntyre DA, Lee YS, Smith A, Marchesi JR, Lehne B, et al. Cervical intraepithelial neoplasia disease progression is associated with increased vaginal microbiome diversity. Sci Rep. 2015;5:16865. 246. Sasieni P, Castanon A, Landy R, Kyrgiou M, Kitchener H, Quigley M, et al. Risk of preterm birth following surgical treatment for cervical disease: executive summary of a recent symposium. BJOG : an international journal of obstetrics and gynaecology. 2015. 247. Kyrgiou M, Valasoulis G, Stasinou SM, Founta C, Athanasiou A, Bennett P, et al. Proportion of cervical excision for cervical intraepithelial neoplasia as a predictor of pregnancy outcomes. Int J Gynaecol Obstet. 2015;128(2):141-7. 248. Romero R, Espinoza J, Erez O, Hassan S. The role of cervical cerclage in obstetric practice: can the patient who could benefit from this procedure be identified? Am J Obstet Gynecol. 2006;194(1):1-9. 249. Drakeley AJ, Roberts D, Alfirevic Z. Cervical stitch (cerclage) for preventing pregnancy loss in women. The Cochrane database of systematic reviews. 2003(1):CD003253. 250. Henry-Stanley MJ, Hess DJ, Barnes AM, Dunny GM, Wells CL. Bacterial contamination of surgical suture resembles a biofilm. Surg Infect (Larchmt). 2010;11(5):433-9. 251. Imagawa H, Nakano S, Kawachi K, Takano S, Tsunooka N, Shikata F. A prospective randomized study of sternal closure: comparison of Mersilene tape versus standard wire closure. Ann Thorac Cardiovasc Surg. 2004;10(6):362-6. 252. Sakai M, Shiozaki A, Tabata M, Sasaki Y, Yoneda S, Arai T, et al. Evaluation of effectiveness of prophylactic cerclage of a short cervix according to interleukin-8 in cervical mucus. Am J Obstet Gynecol. 2006;194(1):14-9. 253. Wang X, Buhimschi CS, Temoin S, Bhandari V, Han YW, Buhimschi IA. Comparative microbial analysis of paired amniotic fluid and cord blood from pregnancies complicated by preterm birth and early-onset neonatal sepsis. PLoS One. 2013;8(2):e56131. 254. Buhimschi CS, Bhandari V, Hamar BD, Bahtiyar MO, Zhao G, Sfakianaki AK, et al. Proteomic profiling of the amniotic fluid to detect inflammation, infection, and neonatal sepsis. PLoS Med. 2007;4(1):e18. 255. Hillier SL, Krohn MA, Cassen E, Easterling TR, Rabe LK, Eschenbach DA. The role of bacterial vaginosis and vaginal bacteria in amniotic fluid infection in women in preterm labor with intact fetal membranes. Clin Infect Dis. 1995;20 Suppl 2:S276-8. 256. Han YW, Shen T, Chung P, Buhimschi IA, Buhimschi CS. Uncultivated bacteria as etiologic agents of intra-amniotic inflammation leading to preterm birth. Journal of clinical microbiology. 2009;47(1):38-47. 257. Borgdorff H, Verwijs MC, Wit FW, Tsivtsivadze E, Ndayisaba GF, Verhelst R, et al. The impact of hormonal contraception and pregnancy on sexually transmitted infections and on cervicovaginal microbiota in african sex workers. Sex Transm Dis. 2015;42(3):143-52.

314 References

258. van de Wijgert JH, Verwijs MC, Turner AN, Morrison CS. Hormonal contraception decreases bacterial vaginosis but oral contraception may increase candidiasis: implications for HIV transmission. AIDS. 2013;27(13):2141-53. 259. Ralph LJ, McCoy SI, Shiu K, Padian NS. Hormonal contraceptive use and women's risk of HIV acquisition: a meta-analysis of observational studies. Lancet Infect Dis. 2015;15(2):181- 9. 260. Blish CA, Baeten JM. Hormonal contraception and HIV-1 transmission. Am J Reprod Immunol. 2011;65(3):302-7. 261. Cicinelli E, de Ziegler D, Bulletti C, Matteo MG, Schonauer LM, Galantino P. Direct transport of progesterone from vagina to uterus. Obstet Gynecol. 2000;95(3):403-6. 262. Conde-Agudelo A, Romero R. Vaginal Progesterone to Prevent Preterm Birth in Pregnant Women with a Sonographic Short Cervix: Clinical and Public Health Implications. Am J Obstet Gynecol. 2015. 263. Liddiard A, Bhattacharya S, Crichton L. Elective and emergency cervical cerclage and immediate pregnancy outcomes: a retrospective observational study. JRSM Short Rep. 2011;2(11):91. 264. Vasquez A, Jakobsson T, Ahrne S, Forsum U, Molin G. Vaginal lactobacillus flora of healthy Swedish women. Journal of clinical microbiology. 2002;40(8):2746-9. 265. Ferris MJ, Norori J, Zozaya-Hinchliffe M, Martin DH. Cultivation-independent analysis of changes in bacterial vaginosis flora following metronidazole treatment. Journal of clinical microbiology. 2007;45(3):1016-8. 266. Macklaim JM, Clemente JC, Knight R, Gloor GB, Reid G. Changes in vaginal microbiota following antimicrobial and probiotic therapy. Microb Ecol Health Dis. 2015;26:27799. 267. Raine-Fenning NJ, Nordin NM, Ramnarine KV, Campbell BK, Clewes JS, Perkins A, et al. Evaluation of the effect of machine settings on quantitative three-dimensional power Doppler angiography: an in-vitro flow phantom experiment. Ultrasound in obstetrics & gynecology : the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2008;32(4):551-9. 268. Jespers V, Crucitti T, van de Wijgert J, Vaneechoutte M, Delany-Moretlwe S, Mwaura M, et al. A DNA tool for early detection of vaginal dysbiosis in African women. Res Microbiol. 2016;167(2):133-41.

315 Appendix

10 APPENDIX

10.1 Twins individual patient meta-analysis

Singletons and twins represent two distinct aetiologies for preterm birth risk. Prior to commencement of prospective recruitment of participants for longitudinal vaginal sampling and transvaginal cervical scans, an individual patient level meta-analysis was performed on retrospectively collected CL data in twin pregnancy. Given the close interaction between cervical parameters and the vaginal microbiome, an individual patient level meta-analysis was performed to determine the relationship between cervical length, gestation at screening and gestation at birth in twin pregnancy. This interaction has been characterised previously in singleton pregnancy 89, however the relationship between these variables in twins remains uncertain. Furthermore it was unknown whether the association was comparable to singleton pregnancy, and if twin pregnancy data was suitable for inclusion in the prospective study.

316 Appendix

10.1.1 Study design Corresponding authors of internationally published studies reporting on cervical length (CL) measurements and preterm birth screening were contacted. Individual patient level data were requested, specifically the exact gestational age at CL screening, the corresponding CL measurement and gestation at birth. A total of 4409 twin pregnancies were included from 12 studies across 7 countries worldwide (Figure 10-1)

Figure 10-1 Search strategy flow chart for twin pregnancy individual patient level meta-analysis

Cervical length and gestational age were treated as continuous variables to assess the interaction between these two variables with respect to gestation at birth.

317 Appendix

10.1.2 Findings There was a significant and non-linear correlation between cervical length, gestation at screening and risk of preterm birth (Figure 10-2, Figure 10-3). Given an equivalent CL measurement and gestation at screening, this risk of preterm birth <37 weeks was higher in twins90 than in singletons89 (Figure 10-3, Figure 10-4). The same threshold for a short cervix in singletons, considered to be the 10th centile at 25mm 83, cannot reliably be applied to twins. Furthermore underlying aetiologies for preterm labour vary between singleton and twin pregnancies. Therefore the decision was made to exclude twins from recruitment for prospective study. As a result only singleton pregnancies were recruited for the investigation of the vaginal microbiome and cervical length (as detailed in the inclusion and exclusion criteria in this Methods chapter).

Figure 10-2 Association between gestation at birth and cervical length measurements according to gestation at screening in twin pregnancy

318 Appendix

Figure 10-3 Predicted probability of birth at < 28+0 weeks (A), 28+1-32+0 weeks (B), 32+1-36+0 weeks (C) and >36+1 weeks (D) based on cervical length measurements (x axis) and gestational age at ultrasound screening

Figure 10-4 Predicted probability of delivery <35 weeks by cervical length (mm) and gestation at measurement in singleton pregnancy. Reproduced from: Berghella et al 200789

319 Appendix

10.2 Re-sequencing of 16S rRNA gene sequences

A total of 621 swabs were selected for bacterial 16s rRNA gene sequencing. Due to the large number of samples this was performed across 2 sequence runs. To test reproducibility and variance between sequence runs, 15 samples from each run were re-sequenced together on a third sequence run and the resulting data compared.

PCA analysis, scatterplot correlations and hierarchical clustering analysis at class taxonomic level demonstrate the reproducibility of sequence reads from the original (blue) and resequenced (orange) runs (Figure 10-5, Figure 10-6, Figure 10-7). The majority of samples were highly consistent with original sequence results. Figure 10-8 plots relative and mean abundance of individual women among runs 1 and 2 for Actinobacteria, Clostridia and Bacilli. The symmetry of these graphs indicates depth of sequencing for common and rarer bacteria are comparative across runs 1 and 2. Scatterplot correlations plotted an R2 of 1.00 for original and resequenced runs, confirming 100% correlation among runs (Figure 10-7). This demonstrated minimal error was introduced at sequencing, and reads were reliable and reproducible going forward. It was therefore considered that the two sequence runs can be reliably combined and compared for data analysis in the remainder of the study.

320 Appendix

Figure 10-5 PCA (class level) of resequenced samples. Blue represents original sequence run and orange represents repeat sequence run. Dots are labelled with the original number used for sequencing. As indicated by the overlap between the original and the repeated sequence runs, high reproducibility was obtained indicating that data collected across the two sequence runs were comparable.

321 Appendix

Figure 10-6 Heatmap of ward hierarchical clustering for resequenced samples at class taxonomic level: run 1 (blue) and run 2 (orange). Samples were classified as a ‘Normal’, ‘Intermediate’, or ‘Atypical’ microbiome according to lactobacillus abundance; >90%, 30 to 90% and <30% Lactobacillus respectively. Reproducibility of the runs is demonstrated as the samples are plotted in close proximity for first and second sequencing results.

322 Appendix

Figure 10-7 Scatterplot correlation graph for original (blue) and re-sequenced samples (orange), demonstrated 100% correlation (R2=1.0)

Figure 10-8 Bar graphs comparing individual participants at class taxonomic level for proportion of sequences of (A) Actinobacteria, (B) Clostridia and (C) Bacilli in run 1 (blue) and run 2 (orange). This demonstrates high reproducibility among runs.

323 Peer reviewed puplications

11 PEER REVIEWED PUPLICATIONS

324 Kindinger et al. Microbiome (2017) 5:6 DOI 10.1186/s40168-016-0223-9

RESEARCH Open Access The interaction between vaginal microbiota, cervical length, and vaginal progesterone treatment for preterm birth risk Lindsay M. Kindinger1,2,3, Phillip R. Bennett1,2, Yun S Lee1, Julian R. Marchesi4,5,6, Ann Smith5, Stefano Cacciatore1, Elaine Holmes4,6, Jeremy K. Nicholson4,6, T. G. Teoh1,3 and David A. MacIntyre1*

Abstract Background: Preterm birth is the primary cause of infant death worldwide. A short cervix in the second trimester of pregnancy is a risk factor for preterm birth. In specific patient cohorts, vaginal progesterone reduces this risk. Using 16S rRNA gene sequencing, we undertook a prospective study in women at risk of preterm birth (n = 161) to assess (1) the relationship between vaginal microbiota and cervical length in the second trimester and preterm birth risk and (2) the impact of vaginal progesterone on vaginal bacterial communities in women with a short cervix. Results: Lactobacillus iners dominance at 16 weeks of gestation was significantly associated with both a short cervix <25 mm (n =15,P<0.05) and preterm birth <34+0 weeks (n =18;P<0.01; 69% PPV). In contrast, Lactobacillus crispatus dominance was highly predictive of term birth (n = 127, 98% PPV). Cervical shortening and preterm birth were not associated with vaginal dysbiosis. A longitudinal characterization of vaginal microbiota (<18, 22, 28, and 34 weeks) was then undertaken in women receiving vaginal progesterone (400 mg/OD, n =25)versuscontrols(n = 42). Progesterone did not alter vaginal bacterial community structure nor reduce L. iners-associated preterm birth (<34 weeks). Conclusions: L. iners dominance of the vaginal microbiota at 16 weeks of gestation is a risk factor for preterm birth, whereas L. crispatus dominance is protective against preterm birth. Vaginal progesterone does not appear to impact the pregnancy vaginal microbiota. Patients and clinicians who may be concerned about “infection risk” associated with the use of a vaginal pessary during high-risk pregnancy can be reassured. Keywords: Vaginal microbiome, Progesterone, Lactobacillus, Preterm birth, Cervical length

Background preterm birth [3–5]. Maternal host-vaginal microbial in- Preterm birth before 37 weeks of gestation is now the teractions throughout pregnancy are likely to play a fun- leading cause of death among children under the age of damental role in reproductive health outcomes. Unlike five [1]. An estimated 15 million babies are born preterm other body sites where high bacterial diversity is consid- each year, and in the USA alone, the annual healthcare ered beneficial to health [6, 7], a healthy vaginal commu- costs associated with those babies that survive is in ex- nity structure in pregnancy is dominated by only one, or cess of $25 billion [2]. Ascending bacterial infection a few, Lactobacillus species [8, 9]. These species provide from the vagina through the cervix into the uterine cav- protection against pathobiont colonization through ity is considered to be a major cause of spontaneous excretion of lactic acid and production of antimicrobial compounds [10]. Assessment of vaginal microbial com- * Correspondence: [email protected] munity structure can be performed using a variety of 1 Imperial College Parturition Research Group, Division of the Institute of next generation sequencing and PCR-based platforms. Reproductive and Developmental Biology, Department of Surgery and Cancer, Faculty of Medicine, Imperial College London, Hammersmith The resulting data is typically analyzed using multivari- Campus, London W12 0NN, UK ate clustering approaches that permit comparison of Full list of author information is available at the end of the article

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Kindinger et al. Microbiome (2017) 5:6 Page 2 of 14

individual taxa or community compositions [11–13]. A remodeling, softening, and dilation [31]. Premature cer- commonly used classification scheme involves hierarch- vical ripening is detectable by transvaginal ultrasound ical clustering analysis of 16S rRNA gene sequencing (TVS) several weeks prior to the onset of the clinical data into community state types (CSTs) as first described symptoms of preterm labor. Second trimester transvagi- by Ravel and colleagues [13]. CSTs are typically domi- nal cervical length measurements are considered a nated by one of four Lactobacillus species; Lactobacillus reliable and predictive tool for preterm birth and are fre- crispatus (CST I), Lactobacillus gasseri (CST II), quently used for preterm birth surveillance [32]. Preg- Lactobacillus iners (CST III), and Lactobacillus jensenii nant women with a short cervix, <25 mm before (CST V). CST IV describes microbial communities 24 weeks of gestation are considered at highest risk of largely devoid of Lactobacillus species and enriched preterm birth [33]. Early diagnosis of these pregnancies mainly in anaerobic bacteria (CST IV). enables timely and targeted intervention by either The composition of vaginal CSTs appears to be influ- cervical cerclage or vaginal progesterone therapy [34]. enced by endogenous hormones, fluctuating with men- While both prevention strategies display comparable ses [14], the use of oral contraceptives [15] and onset of efficacy [35], progesterone supplementation is increas- menopause [16], and estrogen supplementation in post- ingly used as it negates the surgical risks associated with menopausal states [17]. In pregnancy, elevated concen- cerclage insertion such as maternal pyrexia, vaginal trations of circulating estrogen drive glycogen accumula- infection, bleeding, and subsequent requirement for tion in the vaginal epithelium, which is broken down by cesarean section [36, 37] and has not been associated host α-amylase to complex sugar products such as mal- with any adverse neonatal effects [38, 39]. totetraose, maltotriose, and maltose providing carbon The mechanism of action of vaginal progesterone in sources preferentially utilized by Lactobacillus species the prevention of preterm birth involves its capacity to [18]. This substrate availability leads to increased Lacto- promote anti-inflammatory and pro-relaxant pathways bacillus species abundance and stability with advancing in the uterus thereby reducing uterine contractility [40– gestation and a decline in the number of women harbor- 43]. Endogenous progesterone is thought to largely exert ing microbial communities void of lactobacilli, with a anti-inflammatory downstream effects via progesterone dramatic change to reduced Lactobacillus dominance receptor B-dominant signaling [43] to inhibit nitric and increased diversity following estrogen withdrawal in oxide, prostaglandin, and cytokine production [41, 42]. the post-partum period [8, 19]. There is an association This ultimately reduces myometrial contractility thereby between Lactobacillus spp. depletion and vaginal dysbio- inhibiting premature onset of labor [40]. Progesterone sis and poor pregnancy outcomes including preterm also exerts a quiescent effect on the cervix by limiting birth [19] and late miscarriage [20], which appears to be prostaglandin-induced collagenous remodeling of the patient cohort dependent [9]. Thus, the interplay be- cervical fibroblast [44–46]. Progesterone supplementa- tween hormonal and metabolic signaling at the vaginal tion is therefore prescribed primarily for its systemic mucosa interface may act as a protective mechanism for anti-inflammatory actions to maintain myometrial and the immuno-modulated pregnant mother, against expos- cervical quiescence in high-risk pregnancies. In clinical ure to pathogenic bacteria [21]. This may influence not studies, vaginal progesterone has been shown to attenu- only immediate pregnancy outcomes but also longer im- ate the rate of cervical shortening [47, 48], but its effi- munological health in the neonate such as allergy and cacy in preterm birth prevention is cervical-length asthma [22, 23]. This vaginal dysbiosis is present in dependent [48–51] with most benefit observed in high- between 2 and 27% of the population in pregnancy but risk pregnancies with a short cervix <25 mm [35]. Ad- does not always appear to be pathogenic [8, 24]. Recent ministration of vaginal progesterone to women with a evidence implicates L. iners dominance rather than long cervix has not been shown to improve pregnancy dysbiosis for preterm birth risk in pregnancy [25]. L. outcome [51]. Studies on the impact of progesterone on iners is a vaginal commensal that has relatively recently the composition of the vagina microbiota are limited. In been shown to be associated with dysbiosis [26, 27] and non-pregnant women, Borgdorff et al. [52] found that has been suggested as marker of microbial imbalance both injectable progestin contraception and combined leading to BV [28]. oral contraception (progestin and estrogen) do not The cervix serves as both a mechanical and chemical significantly alter vaginal microbiota, but may increase barrier to ascending bacteria [29]. Premature cervical the risk of HIV transmission [53]. It is has been hypoth- ripening, a prerequisite for the expulsion of the fetus, esized that this susceptibility relates to a progesterone- may be triggered by exposure of the amniotic cavity to induced modulation of the local inflammatory immune pathogenic bacteria ascending from the vagina [30], response to infection or alternately thinning of the vagi- which drives a pro-inflammatory cytokine response, trig- nal epithelial barrier [54]. The impact of vaginal proges- gering prostaglandin release, and untimely cervical terone pessaries on the composition of vaginal microbial Kindinger et al. Microbiome (2017) 5:6 Page 3 of 14

communities in pregnancies at risk of preterm birth is Longitudinal sampling following progesterone unknown. intervention We hypothesized that progesterone supplementation A longitudinal study was undertaken in a new pregnancy would promote Lactobacillus spp. dominance and stabil- cohort to assess the effect of progesterone therapy on ity, given progesterone’s anti-inflammatory properties, pregnancy with a short cervix. Women with a prior pre- and efficacy for preterm birth prevention. We therefore term birth <37 weeks were prospectively recruited from undertook a prospective study in women at high-risk of the same preterm surveillance clinics. At initial screen- preterm birth to assess (1) the relationship between ing ≤18 weeks, all women underwent vaginal swab vaginal microbiota in the second trimester and the risk sampling for 16S rRNA gene sequencing, followed by of preterm birth and (2) the impact of vaginal progester- CL measurement at TVS, as previously described. Based one therapy on the vaginal microbiota in women with a on CL measurement, women were then allocated into short cervix. one of two groups. Women with a short cervix <25 mm were treated with vaginal progesterone 400 mg OD at Methods night to continue until 34 completed weeks gestation. The study was approved by NHS National Research Women with a CL ≥25 mm were used as controls and Ethics Service (NRES) Committee London - City and did not receive progesterone or cerclage for the remain- East (REC 12/LO/2003), and all participants provided der of their pregnancy. Both progesterone and control written, informed consent at enrolment. A workflow of groups were recruited at ≤18 weeks, and vaginal swab the study is provided in Additional file 1. samples were collected longitudinally at 22, 28, and 34 weeks of gestation. Patient details and sample collection for cross-sectional Eligibility criteria for participation in both cross- recruitment sectional and longitudinal studies included women with To investigate the association between the vaginal a singleton pregnancy and a prior spontaneous preterm microbiome, cervical length, and preterm birth outcome, birth <37 weeks+0 days, who had not undergone CL a cross-sectional cohort of singleton pregnancies at screening or received either progesterone or cerclage higher than normal risk of preterm birth (due to a his- intervention prior to recruitment. Exclusion criteria in- tory of previous spontaneous preterm birth <37 weeks+0 cluded multiple pregnancy, preterm pre-labor rupture of days) were prospectively recruited from the preterm sur- membranes (PPROM), iatrogenic preterm birth, HIV veillance clinics at two tertiary London maternity units positive women, and women who had had sexual inter- between January 2013 and August 2014. At initial at- course or vaginal bleeding in the preceding 48 h. In the tendance of preterm surveillance clinics at 16 weeks of longitudinal study, any women receiving cervical cerc- gestation, cervico-vaginal fluid was sampled from the lage in either the progesterone or control groups were posterior fornix under direct visualization, using a BBL™ excluded as this may adversely impact on vaginal micro- CultureSwab™ MaxV Liquid Amies swab (Becton, biota [37]. Dickinson and Company, Oxford, UK). The vaginal swabs were placed immediately on ice before being DNA extraction and 16S rRNA gene sequencing transferred and stored at −80 °C within 5 min of collec- DNA extraction from BBL™ CultureSwab™ was per- tion. A cervical length (CL) measurement was taken by formed as previously described [8]. Forward and reverse transvaginal scan (TVS) in supine position, with an fusion primers were used to amplify the V1-V3 hyper- empty bladder, taking care to avoid undue pressure on variable regions of 16S rRNA genes. The forward primer the cervix. Metadata collected included gestation age at was made up of an Illumina i5 adapter (5′-3′) (AATGA- sampling, subsequent interventions for preterm birth, TACGGCGACCACCGAGATCTACAC), 8 bp barcode, gestation at birth, BMI, ethnicity, and antibiotics within primer pad (forward: TATGGTAATT), and the 28F- the week preceding sampling. Participation in this study GAGTTTGATCNTGGCTCAG primer [55]. The reverse did not influence subsequent clinical care or dictate pre- fusion primer consisted of an (5′-3′) Illumina i7 adapter ventative interventions (cervical cerclage or vaginal (CAAGCAGAAGACGGCATACGAGAT), 8 bp barcode, progesterone supplementation) for preterm birth risk. primer pad (reverse: AGTCAGTCAG), and the reverse For the duration of the study, both units employed a primer (519R-GTNTTACNGCGGCKGCTG). Sequen- policy of CL screening every 3 weeks until 25 weeks, cing was performed on an Illumina MiSeq platform with the indication for intervention being a CL <25 mm (Illumina, Inc. San Diego, California). Sequence data was at TVS measured at ≤23+6 weeks gestation. In this processed and analyzed using the MiSeq SOP Pipeline of cross-sectional arm of the study, the choice of interven- the Mothur package [56] with the Silva bacterial data- tion for a short cervix (cerclage or progesterone) was at base (www.arb-silva.de/) used for sequence alignment. the discretion of the attending clinician. Sequence classification was performed using the RDP Kindinger et al. Microbiome (2017) 5:6 Page 4 of 14

database reference sequence files and the Wang method prediction of preterm birth <34+0 weeks according to [57] and taxonomy assignments determined using the CST classification at 16-week sampling. RDP MultiClassifier script and USEARCH with 16S rRNA gene sequences from the cultured representatives Longitudinal cohort from the RDP database [58] for species level taxonomies. A linear mixed-effects model incorporating gestational Data was re-sampled and normalized to the lowest read age, maternal age, BMI, ethnicity (Asian, Black, or Cau- count in Mothur (n =725)[59]. casian), and cohort (progesterone with short cervix ver- sus control with normal cervix) as fixed-effects and the Statistical analyses anonymized patient ID as uncorrelated random-effect Examination of statistical differences between vaginal was used to assess the impact of progesterone interven- microbiota was performed using the Statistical Analysis tion on CST distribution and relative abundance in spe- of Metagenomic Profiles (STAMP) software package cies present in >5% of all samples. The contributions of [60]. To classify vaginal bacterial communities into com- fixed-effects terms (P value and F statistics) were calcu- munity state types (CSTs), hierarchical clustering ana- lated using ANOVA with Satterthwaite approximation lysis (HCA) species taxonomy was performed using for degrees of freedom. For each fixed-effects term, a ward linkage with a clustering density threshold of 0.75. false discovery rate adjustment (Benjamin-Hochberg) Samples were classified into five CSTs; I (L. crispatus), II was applied to correct P values. All data and computa- (L. gasseri), III (L. iners), IV (mixed bacterial species), tional approaches used for this study are provided in and V (L. jensenii)asdescribedbyRaveletal.[13]. additional information (Additional files 2, 3, 4, 5, and 6).

Cross-sectional cohort Results In the cross-sectional cohort, a total number of species Cross-sectional study patient cohort demographics observed and the Shannon index of alpha diversity were A total of 161 pregnant women attending prematurity calculated and compared across gestation at birth: <34 surveillance clinics for their first appointment (mean 16 +0 weeks, 34+0 to 36+6 weeks, and ≥37+0 weeks, using a +6 weeks gestation, Table 1) consented to a vaginal swab 2-way ANOVA. followed by a transvaginal scan for cervical length meas- Examination of the relationship between vaginal bac- urement. Spontaneous preterm birth <37 weeks oc- terial communities (or CSTs) and cervical length or pre- curred in 34 women (21%) (mean 32+6 weeks, SD ± 3 term birth was assessed using a Fisher exact test as to +6 weeks, range 24+4–36+6 weeks). Rates of preterm birth provide an exact, robust P value that is less sensitive to <37+0 weeks were higher in Black women (37%, 11/30) small sample sizes than alternative approaches such as than Caucasians (17%, 18/104) and Asians (19%, 5/27; P Chi-squared (reference). Fisher’s exact test was also used < 0.05). Subsequent cervical shortening to below 25 mm to examine individual CST assignments (in 5 × 2 contin- occurred in 66% (91/161), all of whom went on to re- gency tables for cervical length and requirement for fu- ceive an intervention (ultrasound indicated cervical cerc- ture interventions and 3 × 25 contingency tables for lage, n = 71 or vaginal progesterone, n = 20). birth gestation and ethnicity). Gehan-Breslow-Wilcoxon test was used to compare pregnancy survival (duration The vaginal microbiome at 16 weeks in high-risk of gestation) for CST I compared to CST III [61]. pregnancy A logistic mixed-effects regression analysis incorporat- Using hierarchical clustering analysis (HCA) of normal- ing gestational age at sample, maternal age, and BMI as ized genera taxonomy read counts, vaginal swab samples fixed effects and ethnicity as a random effect was per- were classified into three categories; normal (>90% formed to assess the relationship between pregnancy Lactobacillus spp., 147/161, 91%), intermediate (50–90% outcome (birth <34 and >34 weeks) and CSTs as well as Lactobacillus spp., 5/161, 3%), and dysbiotic (<10% individual species abundance. Analyses were performed Lactobacillus spp., 9/161, 6%; Additional file 7A). No re- in R using ANOVA and the “glm” (generalized linear lationship was observed between genera level structure model) R function to analyze the table of deviance for and subsequent gestational age at delivery. Dominance CSTs and for individual species (present in >5% of sam- of Lactobacillus species occurred in equal proportions of ples). Species abundances were log-transformed, ad- patients experiencing preterm <37 weeks (31/34; 91%) justed for confounders, and false discovery rate or term birth (116/127; 91%). Of those women delivering adjustment (Benjamini-Hochberg) was applied to correct preterm, 3/34 (9%) harbored a dysbiotic or intermediate P values for each analysis [62]. microbiome at 16 weeks compared to 6/127 (5%) who Accuracy parameters, sensitivity (sens), specificity delivered at term. Consistent with these findings, mea- (spec), positive predictive values (PPV), and negative surements of species richness (total number of species predictive values (NPV) were calculated for the observed; Additional file 7B) and alpha diversity Kindinger et al. Microbiome (2017) 5:6 Page 5 of 14

Table 1 Patient demographics for a cross-section of 161 participants Term birth >37 weeks Preterm birth <37 weeks Total n/N (%) 127/161 (79%) 34/161 (21%) 161/161 (100%) BMI Mean ± SD (range) 24.3 ± 4.4 (18–48) 24.3 ± 4.4 (18.4–35) 24.3 ± 4.4 (18–48) Ethnicity, n/N (%) Caucasian 86/127 (68%) 18/34 (53%) 104/161 (65%) Asian 22/127 (17%) 5/34 (15%) 27/161 (17%) Black 19/127 (15%) 11/34 (32%)* 30/161 (19%) Smoker, n/N (%) 8/127 (6%) 3/34 (9%) 11/161 (7%) Gestation at sample (weeks) Mean ± SD (range) 17+0 ± 1.0 (13+1–18+4)16+4 ± 1.4 (12+1–18+4) 16+6 ± 1.1 (12+1–18+4) Cervical length (mm) Mean ± SD (range) 32.5 ± 1.0 (18–50) 30.6 ± 6.4 (10–40) 32 ± 5.6 (10–50) Intervention n/N (%) No intervention 60/127 (47%) 10/32 (29%) 70/161 (43%) Cerclage 51/127 (40%) 20/32 (59%) 71/161 (44%) Progesterone 16/127 (13%) 4/32 (12%) 20/161 (12%) Gestation at delivery, n/N (%) Early PTB, <34+0 weeks na 18/34 (53%) 18/161 (11%) Late PTB, 34+0 to <37+0 weeks na 16/34 (47%) 16/161 (10%) Term, ≥37+0 weeks 127/127 (100%) na 127/161 (79%) PTB preterm birth, na not applicable *P < 0.05 Fisher’s exact term vs. preterm birth groups

(Shannon index; Additional file 7C) at 16 weeks were com- (>37+0 weeks) (Fig. 1b, Table 2). Specifically, an L. iners- parable between women experiencing term (≥37+0 weeks, dominated microbiome was significantly over- n =127),latepreterm(34+0–36+6 weeks, n =16),andearly represented in women delivering <34+0 weeks (67%) preterm (<34+0 weeks, n =18)delivery. compared to late preterm (31%) and term (29%; P= Hierarchical clustering of species data permitted clas- 0.003, Fisher’s exact). In contrast, L. crispatus dominance sification of samples into community state types (CSTs): associated with subsequent term birth (46 vs. 11% early I(L. crispatus), II (L. gasseri), III (L. iners), IV (diverse preterm birth; P=0.009, Fisher’s exact, Fig. 1c, Table 2), species), and V (L. jensenii) (Fig. 1a). The most prevalent and comparatively longer duration of pregnancy than L. CST observed in the patient cohort was CST I (L. cris- iners Fig. 1d. A logistic regression mixed-effects model patus, 40%), followed by CST III, (L. iners, 34%), CST II demonstrated that the association between gestation at (L. gasseri, 10%); CST V (L. jensenii, 9%), and CST IV birth and CST at 16 weeks persisted after accounting for (diverse, 6%; Table 2). L. crispatus (CST I) was most ethnicity, maternal age, BMI, and gestation at sampling abundant among Caucasian women (P = 0.008), while (P = 0.04; ANOVA; Additional file 8). When individual Black women had greater numbers of CST III (L. iners; species were assessed by mixed-effects modeling follow- P = 0.049 and CST IV (P = 0.033) (Fisher’s exact, Table 2). ing correction for potential confounders, both L. crispa- Short CL <25 mm was significantly associated with L. tus and L. iners were significantly correlated with birth iners (CST III) dominance (9/15, 60%) compared to outcome with L. crispatus positively associated with those women with a CL >25 mm (45/101, 31%; P = 0.04; delivery >34 weeks (P = 0.009, q = 0.048) and L. iners Fisher’s exact, Fig. 1b). positively associated with delivery <34 weeks (P = 0.001, q = 0.006; Additional file 9). The association of the vaginal microbiome at 16 weeks of The analysis of outcomes stratified by ethnicity did not gestation and risk of preterm birth identify a significant difference between groups although Major differences were detected in vaginal microbial this may be due to relatively small sample sizes. The ma- communities at 16 weeks in women subsequently jority of women delivering >34 weeks with L. crispatus delivering early preterm (<34+0 weeks) compared those dominance (CST I) were Caucasian (Caucasian 50/95, delivering late preterm (34+0 to 36+6 weeks) or at term 53%, Asian 7/23, 30%, and Black women 6/25, 24%), but Kindinger et al. Microbiome (2017) 5:6 Page 6 of 14

Fig. 1 L. iners dominance is associated with a short cervix and preterm birth risk while L. crispatus is protective. a Heatmap of vaginal species data correlated community state types of samples (n = 161) with ethnicity, cervical length <25 mm, subsequent cerclage or progesterone intervention, and gestation at birth. b A short cervix <25 mm at 16 weeks was associated with a higher prevalence of L. iners (9/15, 60%) than longer cervical length (45/146, 31%, P=0.04, two-tailed Fisher’s exact). c L. iners dominance was associated with early preterm birth <34+0 weeks (12/18, 67%), but not late preterm birth, 34+0 to 36+6 weeks (5/16, 31%) or term birth (37/127, 29%, P=0.003). A greater proportion of term births had L. crispatus dominance at 16 weeks (63/127, 46%) than both late preterm (5/16, 31%) and early preterm births <34+0 weeks (2/18, 11%; P=0.009; Fisher’s exact). d A Kaplan-Meier survival curve demonstrated that L. iners (n = 54) dominance at 16 weeks is associated with earlier gestation at delivery than a microbiome dominated by L. crispatus (n = 65, P=0.02; Gehan-Breslow-Wilcoxon test) this was not significant. In those women delivering <34 crispatus abundance was strongly predictive of birth with L. iners dominance (CST III), similar proportions >34 weeks gestation (89% specificity and 97% PPV; were represented across ethnic groups (Caucasian 6/9, Table 3). 67%, Asian 3/4, 75%, and Black women 3/5, 60%; Additional file 10). Effect of progesterone intervention on vaginal microbial Calculation of predictive accuracies for preterm birth communities in high-risk pregnancy using CST assignments at 16 weeks provided sensitivity Given the significant association between vaginal micro- and specificity values comparable to screening using cer- bial composition at 16 weeks and cervical length and/or vical length [32]: L. iners dominance predicted preterm subsequent preterm birth <34+0 weeks, we next con- birth <34+0 weeks with 67% sensitivity and 71% specifi- ducted a longitudinal study of the vaginal microbiome in city (Table 3). While its absence provided a 94% negative women receiving vaginal progesterone supplementation predictive value (NPV), the PPV of L. iners dominance for a short cervix (<25 mm). A total of 67 pregnant for preterm birth <34 weeks was 22%. High relative L. women were eligible and consented to recruitment, of Kindinger et al. Microbiome (2017) 5:6 Page 7 of 14

Table 2 Distribution of community state types according to ethnicity and gestation at birth CST, Species Total population CST I, L. crispatus CST II, L. gasseri CST III, L. iners CST IV, diverse species CST V, L. jensenii n/N (%) 161 (100%) 65/161 (40%) 17/161 (11%) 54/161 (34%) 11/161 (7%) 14/161 (9%) Ethnicity Caucasian 104/161 (65%) 52/104 (50%)* 13/104 (13%) 26/104 (25%) 5/104 (5%) 8/104 (8%) Asian 27/161 (17%) 7/27 (26%) 3/27 (11%) 13/27 (48%)* 1/27 (4%) 3/27 (11%) Black 30/161 (19%) 6/30 (20%) 1/30 (3%) 15/30 (50%)* 5/30 (17%)* 3/30 (10%) Gestation at birth <34 weeks 18/161 (11%) 2/18 (11%)** 1/18 (6%) 12/18 (67%)** 1/18 (6%) 2/18 (11%) 34–37 weeks 16/161 (10%) 5/16 (31%) 2/16 (13%) 5/16 (31%) 2/16 (13%) 2/16 (13%) >37 weeks 127/161 (79%) 58/127 (46%) 14/127 (11%) 37/127 (29%) 8/127 (6%) 10/127 (8%) CST community state type based on ward HCA of species data *P<0.05, **P<0.01; for comparison of birth <34 vs. >34 weeks, two-tailed Fisher’s exact which 25 were found to have a short CL <25 mm and or L. crispatus with advancing gestation when compared received progesterone until 34 weeks of gestation. The to controls (Fig. 2d, c; Additional file 13). remaining 42 women did not experience cervical short- The dynamics of individual vaginal CSTs during preg- ening or receive any subsequent preventative interven- nancy were then longitudinally assessed in both proges- tion, and hence were used as controls. Demographics of terone and control cohorts (Fig. 3). Regardless of the two groups were comparable although, as per study intervention, a L. crispatus (CST I)-dominated micro- design, the mean CL at commencement of progesterone biome was associated with high stability throughout was significantly lower in the “short CL” than the “nor- pregnancy with 92% (24/26) of women maintaining L. mal” control group (22 vs. 32 mm, P<0.05) at compar- crispatus dominance across all sampling time points. In able screening gestations (15+5 vs. 15+0, respectively; contrast, significantly lower stability was observed in the Table 4). Respective rates of preterm birth (<37 weeks) 23 women exhibiting a L. iners-dominated microbiome were higher in the progesterone (32%, 8/25) versus at the first sampling with 17 (74%) of these women ex- control groups (5%, 2/42; P=0.004). A total of 234 high periencing a shift to an alternative CST at some stage vaginal samples were collected from longitudinal follow- during their pregnancy (P<0.0001). Similar levels of up (22, 28, and 34 weeks) at matched gestational ages CST-shifting were observed in those women receiving among groups (Additional file 11). vaginal supplementation (9/25; 36%) and control The distribution of CSTs in the progesterone and con- patients (10/43; 23%) (P=0.3). trol groups at each sampling time point is provided in When gestational age at sampling, maternal age, BMI, Additional file 12. Prior to progesterone intervention, no ethnicity, and cohort were incorporated into a linear significant difference in the distribution of CSTs between mixed-effects model, progesterone treatment did not to the two patient cohorts was observed (Fig. 2; have a significant impact upon CSTs apart from CST II; Additional file 12). Vaginal progesterone supplementa- however, this difference did not withstand multiple test- tion had no effect upon vaginal bacterial community ing correction (Additional file 14). When further state structure throughout pregnancy (Fig. 2a) nor were assessed by relative abundance of individual species, species richness or alpha diversity measurements altered there were no significant differences in proportions of L. (Fig. 2b, c). Progesterone supplementation did not sig- gasseri, or any other species in the control compared to nificantly impact on mean relative abundance of L. iners progesterone cohorts (Additional file 15).

Table 3 Predictive accuracies of microbial species dominance at 16 weeks for gestation <34 weeks CST Species Preterm birth <34 weeks Birth >34 weeks Sens/DR (%) Spec (%) PPV (%) NPV (%) Sens/DR (%) Spec (%) PPV (%) NPV (%) I L. crispatus 11 56 3 83 44 89 97 17 II L. gasseri 6 89 6 88 11 94 94 12 III L. iners 67 71 22 94 29 33 78 6 IV Diverse 6 93 9 89 7 94 91 11 V L. jensenii 11 92 14 89 8 89 86 11 CST community state type based on ward HCA of species data, Sens/DR sensitivity or detection rate, Spec specificity, PPV/NPV positive/negative predictive values Kindinger et al. Microbiome (2017) 5:6 Page 8 of 14

Table 4 Participant demographics for control and progesterone groups High risk controls Progesterone Total n/N (%) 42/67 (63%) 25/67(37%) 67/67 (100%) Age, years Mean ± SD (range) 32 ± 5.5 (21–40) 32 ± 3.9 (22–38) 32 ± 5.0 (21–40) BMI Mean ± SD (range) 24.7 ± 5.3 (19–48) 25.2 ± 4.7 (18.4–35) 24.9 ± 5.0 (18.4–48) Ethnicity, n/N (%) Caucasian 32/42 (76%) 18/25 (72%) 50/67 (75%) Asian 4/42 (10%) 3/25 (12%) 7/67 (10%) Black 6/42 (14%) 4/25 (16%) 10/67 (15%) Smoker n/N (%) 2/42 (5%) 0/25 (0%) 2/67 (3%) Screening for progesterone GA (weeks), median, range 15+0 (12+1–18+2)15+6 (12+0–18+6)15+3 (12+0–18+6) CL (mm), median, range 32 (26–43) 22 (13–25) (13–43) Gestation at delivery, n/N (%) Early PTB, <34+0 weeks 1/42 (2%) 4/25 (16%) 5/67 (7%) Late PTB, 34+0 to <37+0 weeks 1/42 (2%) 4/25 (16%) 5/67 (7%) Term, ≥37+0 weeks 40/42 (95%) 17/25 (68%) 57/67 (85%) PTB preterm birth, GA gestational age, CL cervical length (mm)

In women receiving progesterone, marked differences (odds ratio 2.3) [66]. Using similar methodology, in the longitudinal CST distributions were observed in Donders and colleagues recently reported that a women delivering <34 weeks compared to those deliver- lactobacilli-dominated vaginal microbiome in the first ing >34 weeks (Fig. 4). At 18 weeks of gestation, L. iners trimester was associated with a 75% lower risk of deliv- dominance was observed in 100% (4/4) of women who ery before 35 weeks of gestation (0.26; 95% confidence subsequently delivered <34 weeks of gestation compared interval (CI) 0.12–0.56] compared to women harboring to 24% (5/21) in women delivering >34 weeks. At deliv- a vaginal microbiome void of Lactobacillus species (OR ery, L. iners dominance was observed in 50% (2/4) at 2.4; 95% CI 1.2–4.8) [67]. Using culture-independent 22 weeks and 100% (4/4) at 28 weeks (Figs. 3 and 4). characterization of vaginal bacterial communities in a high-risk pregnant population, we show that the per- Discussion ceived benefit of lactobacilli dominance in pregnancy is This study represents the largest next generation species specific; L. crispatus is advantageous and associ- sequencing-based analysis of vaginal microbiota in preg- ated with term delivery whereas L. iners is associated nancies at risk of preterm birth to date. We demonstrate with increased risk of preterm delivery. Furthermore, L. a significant association between L. iners dominance of iners is associated more specifically with a risk of early the vaginal microbiome at 16 weeks of gestation with (<34 weeks) rather than late (34–37 weeks) preterm subsequent preterm birth and conversely show that L. birth. High relative abundance of L. crispatus is highly crispatus dominance correlates with reduced risk or pre- specific for term birth, with a false positive rate (1 speci- term birth. Moreover, we show that the insertion of a ficity) of just 3% in our population of women at high risk progesterone pessary for prevention of preterm birth has because of a previous preterm birth. In this population, no adverse impact on vaginal microbial communities. second trimester dominance of L. iners carries a 67% de- A healthy vaginal microbiome in non-gravid and tection rate (i.e., sensitivity) for preterm birth before gravid subjects is often described as being synonymous 34 weeks; a screening sensitivity comparable to cervical with low bacterial diversity and Lactobacillus species length, the current and primary screening tool used for dominance [63–66]. Examination of vaginal microbiota preterm birth surveillance [32, 68]. Consistent with our at the time of delivery using culture and/or microscopy- findings, Petricevic and colleagues recently reported an based techniques has shown that Lactobacillus species over-representation of L. iners dominance in vaginal dominance is negatively associated with delivery before swab samples collected from 13 preterm births derived 37 weeks of gestation (odds ratio 0.2) whereas bacterial from a low-risk cohort of 111 pregnancies, and none of dysbiosis is positively associated with preterm delivery whom delivered before 33 weeks of gestation [25]. Kindinger et al. Microbiome (2017) 5:6 Page 9 of 14

Fig. 2 Vaginal progesterone treatment does not alter structure of the vaginal microbiome. a Compared to controls (n = 42), progesterone supplementation (n = 25) had no significant impact upon microbial community profiles with advancing gestation. Similarly, no effect of progesterone treatment upon b the number of species observed or c the corresponding Shannon index of alpha diversity was observed (2-way ANOVA). Fewer women requiring progesterone had a L. crispatus dominated microbiome compared to controls (8/25, 32 vs. 18/42, 43%, P=0.4); however, progesterone treatment was associated with increased relative L. crispatus abundance with advancing gestation. Advancing gestational age from 18 to 34 weeks was not associated with a significant shift in mean relative abundance of L. iners (d)orL. crispatus (e) in either the controls or progesterone groups (Kruskal-Wallis, Dunn’s multiple comparison)

However, this study was limited by the use of denaturing dysbiosis (CST IV) [13, 24]. In contrast to these findings, gradient gel electrophoresis (DGGE) for the Digiulio and co-workers [19] reported, in a small yet characterization of only major Lactobacillus species and densely sampled cohort of women experiencing preterm could not identify other pathobionts in the samples. birth (n = 15), which vaginal bacterial diversity does correl- While our study reveals a clear relationship between ate with risk of preterm delivery [19]. The clinical rele- relative abundance of vaginal Lactobacillus species and vance of these findings however are difficult to establish risk of subsequent preterm birth, our data indicate that considering the small sample size and the heterogeneous Lactobacillus spp. depletion or vaginal dysbiosis in the nature of the cohort; only five women delivered preterm second trimester does not appear to contribute to pre- spontaneously and almost half delivered within 1 week of term birth risk. The role of early gestational vaginal dys- term dates (>36 weeks+3/7 days). biosis in the pathology of preterm birth is controversial. Our presented data provide some suggestion that vagi- In agreement with our findings, a recent longitudinal nal microbiota in Black women may not play as an im- analysis of the vaginal microbiome by Romero and col- portant contributory role to preterm birth pathogenesis leagues in 18 women experiencing preterm birth as Caucasians and Asians. We did not however have (<34 weeks gestation) reported no association between sufficient power to demonstrate the significance of this, preterm birth and vaginal microbial dysbiosis when com- but this may be worth examining in future studies. pared to controls experiencing term delivery (n =72)[24]. In our study, we also observed a high rate of CST- However, in their study, 95% (17/18) of preterm birth shifting in women with an L. iners-dominated micro- samples and 86% of control samples were collected from biome in the second trimester compared to women with African American women who exhibit a higher pregnant an L. crispatus-dominated microbiome. L. iners has been and non-pregnant background prevalence of vaginal reported as an intermediary between lactobacilli Kindinger et al. Microbiome (2017) 5:6 Page 10 of 14

Other evidence for a role of bacterial dysbiosis in the pathology of preterm birth includes the long recognized association between bacterial vaginosis (BV) and in- creased risk of preterm birth; however, evidence suggests that screening and treating BV in pregnancy reduces preterm delivery in certain cohorts [73], but not in others [74]. We propose an alternate concept, which is that it is the presence of L. iners that promotes risk of early preterm birth, but because an L. iners-dominated vaginal microbiome has less stability, there is a tendency for transition to BV-associated CST-IV [69, 75]. Indeed, L. iners is the prominent vaginal species following anti- biotic treatment for BV [76]. Older studies aimed at de- tecting BV, which could not differentiate Lactobacillus species and concluded that it was BV rather than species of Lactobacillus that conferred the risk. Recent investigations into the protective role of Lacto- bacillus species in the context of reproductive health have revealed major species-specific differences in the capacity to prevent pathobiont colonization and viral infections [12, 72, 77, 78] that are driven largely by maternal host-bacterial metabolite interactions at the va- ginal mucosal interface. For example, although lactic acid-producing bacteria including Lactobacillus spp. produce both the D- and L-lactic acid isomers [79], the chirality of the isomer has major functional implications. In women exhibiting a vaginal microbiome dominated by L. iners, an increased ratio of L to D-lactic acid has previously been shown to promote expression of vaginal extracellular matrix metalloproteinase inducer (EMM- PRIN) and the activation of matrix metalloproteinase-8 (MMP8), which may subsequently modulate cervical in- tegrity [80]. Conversely, no such relationship has been observed in women with vaginal microbial communities dominated by L. crispatus, which instead preferentially Fig. 3 Longitudinal profiling of community state types for excretes high levels of D-lactic acid and greater overall progesterone (n = 25) and control groups (n = 42). Progesterone levels of lactic acid than L. iners [80]. Apart from modu- supplementation was commenced after the first sampling time lating local tissue inflammation, recent studies have also point (<18 weeks). Each longitudinal sample was assigned to a CST implicated lactate isomers in vaginal mucosal trapping (Fig. 1a) as indicated by the color-coded rectangle and categorized as a function of delivery gestation mechanisms. High concentrations of D-lactic acid are associated with L. crispatus dominance and enhanced trapping of HIV-1 virions in cervico-vaginal mucosa whereas low concentrations of D-lactic acid associated dominance and CST IV-associated states and is the pre- with L. iners dominance permits comparatively rapid dif- dominant microbiome in peri-menopausal women as fusion of virions through cervico-vaginal mucosa [77]. they transition through to postmenopausal dominance Collectively, these data suggest mechanisms by which L. of anaerobic bacteria [16]. Interactions between L. iners iners dominance of vaginal microbial communities dur- and the maternal host likely provides a vaginal mucosal ing pregnancy may lead to the modulation of local tissue environment permissible to colonization by BV- inflammation and remodeling pathways and to disrup- associated pathogens, a setting in which it tolerates co- tion of chemical and mechanical mucosal barriers pro- existence well [69, 70]. Unlike other Lactobacillus species, tective against ascending infection and increase the risk L. iners also induces secretion of pro-inflammatory cyto- of preterm birth. Such mechanisms may account for the kines when human vaginal epithelial cells are observed in observed association between L. iners dominance and a vitro, whereas L. crispatus does not [71, 72]. short cervical length (<25 mm) seen in our study at Kindinger et al. Microbiome (2017) 5:6 Page 11 of 14

Fig. 4 Preterm birth, despite vaginal progesterone, is associated with L. iners dominance throughout pregnancy. Longitudinal sampling of 25 women receiving progesterone for a short cervix showed L. iners dominance was associated with all women who subsequently delivered preterm <34+0 weeks (n = 4; (**P<0.05; Fisher’s exact). Single asterisk indicates the delivery samples collected within 2 weeks of delivery between 28 and 34 weeks

16 weeks of gestation, which itself is highly specific for birth risk [33]; therefore, once detected, clinicians are preterm birth [81]. ethically obliged to provide a preventative intervention Considering the potential pro-inflammatory roles such as progesterone. Consequently, a “placebo” inter- played by L. iners in the vagina during pregnancy, we vention for a short cervix could not be included for postulated that any associated poor pregnancy outcomes study in this clinical study. As such, the control women might be attenuated by the anti-inflammatory actions of are not true controls as their cervical lengths were all progesterone [45, 82]. However, in this study, no effect greater than 25 mm at entry. A further potential con- of vaginal progesterone therapy upon the frequency of founding factor was the impact of ethnicity of vaginal vaginal community state structure was observed across microbiota and gestation at birth, although we demon- pregnancy indicating that the mode of action of proges- strated this not to be significant in our cohort. terone in the prevention of preterm birth is not through modulation of the vaginal microbiome. The data also Conclusions show, however, that there is no detrimental effect upon Our data indicate that specific Lactobacillus species have the vaginal microbiome of either progesterone itself or differing associations with outcome in pregnancies at of the daily vaginal insertion of a pessary. high risk of preterm birth. Detection of vaginal microbial A particular strength of our study is that we character- composition in the early second trimester may be used ized the vaginal microbiome in a comparatively large to stratify preterm birth risk; L. crispatus dominance is patient cohort at high risk of preterm birth. This highly predictive of term birth, while high L. iners rela- strength was demonstrated by a high spontaneous pre- tive abundance is associated with increased risk of pre- term birth rate (n = 34/161). The mean gestation at birth term birth and warrants heighted surveillance. Increased of 32+6 weeks within our preterm birth cohort, and a diversity of vaginal microbiota at 16 weeks of gestation distribution of gestational ages ranging from 24 to is not associated with increased risk of preterm birth. 36 weeks, enabled the characterization of microbial pro- The use of progesterone therapy for preterm birth pre- files associated with both early (<34 weeks) and late vention does not appear to adversely affect the relative (34 > 37 weeks) preterm birth, providing a broader abundance of vaginal Lactobacillus species or species di- observational base for microbial-host interactions in versity, indicating that progesterone’s mode of action pregnancy. The primary limitation was the small number during pregnancy is likely not via modulation of vaginal of women receiving progesterone (n = 25) and the lack microbial communities. Patients and clinicians who may of an equivalent control group with a short CL <25 mm be concerned about the “infection risk” associated with not receiving any intervention or receiving a placebo. A the use of vaginal pessaries during high-risk pregnancy short CL significantly increases subsequent preterm can be reassured. Kindinger et al. Microbiome (2017) 5:6 Page 12 of 14

Additional files Acknowledgements We thank all participants of the study and members of Women’s Health Research Centre, Imperial College Health NHS. Additional file 1: Work flow of methodology for cross-sectional and longitudinal studies. Funding Additional file 2: Detailed methodology for statistical analyses. This work was supported by the National Institute for Health Research (NIHR) Additional file 3: Species read count data for cross-sectional cohort. Comprehensive Biomedical Research Centre at Imperial College London Additional file 4: Patient metadata for cross-sectional cohort. (Grant Ref P45272) and by the Genesis Research Trust (Grant Ref P51389). DAM is supported by a Career Development Award from the Medical Additional file 5: Species read count data for longitudinal cohort Research Council (MR/L009226/1). (progesterone treatment). Additional file 6: Patient metadata for longitudinal cohort Availability of data and materials (progesterone treatment). The sequence datasets and relevant metadata supporting the conclusions of Additional file 7: Preterm birth does not associate with vaginal this study can be obtained at the European Nucleotide Archive’s (ENA) dysbiosis at 16 weeks of gestation. (A) Heatmap of ward hierarchical Sequence Read Archive (SRA) (accession number PRJEB11895 and clustering of microbial genera from 161 women sampled at 16 weeks of PRJEB12577). Further details of all statistical analyses performed in this gestation, classified according to subsequent gestation at delivery. manuscript are provided in Additional file 2. Species-level read count data Women delivering preterm both <34+0 weeks (n = 18, red) and 34+0 to used for these analyses along with the relevant metadata are provided in 36+6 weeks (n = 16, orange) had a predominantly Lactobacillus Additional files 3, 4, 5, and 6. species-dominated vaginal microbiome, as did women experiencing term births >37+0 weeks (n = 127, gray). (B) No correlation between richness Authors’ contributions (number of bacterial species observed; Sobs) (C), nor alpha diversity as LMK, PRB, and DAM conceived and designed the experiments. LMK and YSL measured using the Shannon index with gestation at birth was observed. collected samples and performed experiments. LMK, JRM, AS, SC, and DAM (ns non-significant, 2-way ANOVA). performed the data analysis and generated the figures and tables. LMK, PRB, and DAM wrote the first draft of the manuscript. LMK, PRB, YSL, JRM, AS, SC, Additional file 8: Linear mixed-effects model of cross-sectional data EH, JKN, TGT, and DAM critically reviewed the data and the manuscript. All assessing the correlation between CST, gestational age at sample, authors read and approved the final manuscript. maternal age, and BMI on pregnancy outcome. Additional file 9: Linear mixed-effects model assessing the impact of Competing interests relative species abundance at 16 weeks on subsequent gestation at birth. The views expressed in the submitted article are of the authors and not an Additional file 10: Distribution of CSTs as a function of ethnicity and official position of the institutions or funders. The authors declare that they gestation at birth in cross-sectional cohort. have no competing interests. Additional file 11: Gestational age at vaginal sampling in high-risk control and progesterone groups. Consent for publication Not applicable. Additional file 12: Community state type classification of samples as a function of gestation at sampling: progesterone versus control groups. Ethics approval and consent to participate Additional file 13: Comparison of mean L. iners and L. crispatus relative The study was approved by NHS National Research Ethics Service (NRES) abundance in control (n = 42) versus progesterone groups (n = 25) and as Committee London - City and East (REC 12/LO/2003), and all participants a function of birth before and after 34 weeks. (A) Prior to progesterone provided written, informed consent at enrolment. intervention at <18-week sampling, women with a short CL <25 mm had greater relative abundance of L. iners compared to controls, and lower L. Author details crispatus (B) although this did not reach significance. L. iners abundance 1Imperial College Parturition Research Group, Division of the Institute of declined in both control and progesterone groups towards 34 weeks of Reproductive and Developmental Biology, Department of Surgery and sampling while mean L. crispatus abundance increased (ANOVA, K-W, Cancer, Faculty of Medicine, Imperial College London, Hammersmith Dunn’s multiple comparison). Inclusive of control and progesterone Campus, London W12 0NN, UK. 2Queen Charlotte’s Hospital, Imperial College groups, preterm birth <34 weeks was associated with higher mean L. Healthcare NHS Trust, London, UK. 3Department of Obstetrics and iners abundance at longitudinal sampling (C; P<0.05), and lower mean L. Gynaecology, St Mary’s Hospital, Imperial College Healthcare NHS Trust, crispatus abundance (D; P<0.001) than deliveries >34 weeks, at matched London, UK. 4Centre for Digestive and Gut Health, Department of Surgery gestational age at sampling throughout follow-up (Welch’s t test). and Cancer and the Institute of Global Health Innovation, Faculty of 5 Additional file 14: Mixed-effects model assessing the impact of Medicine, Imperial College London, London, UK. School of Biosciences, 6 progesterone treatment on CST profile at longitudinal sampling, Cardiff University, Cardiff, UK. Division of Computational Systems Medicine, incorporating potential contributing confounders (gestational age at Department of Surgery and Cancer, Faculty of Medicine, Imperial College sample, maternal age, BMI, and ethnicity). London, London, UK. Additional file 15: Mixed-effects model comparing relative species Received: 26 July 2016 Accepted: 15 December 2016 abundance in control and progesterone groups.

References 1. Causes of child mortality [http://www.who.int/gho/child_health/mortality/ Abbreviations causes/en/]. Accessed 7 July 2016. bp: Base pairs; BV: Bacterial vaginosis; CI: Confidence interval; CL: Cervical 2. McCormick MC, Litt JS, Smith VC, Zupancic JA. Prematurity: an overview and length; CST: Community state type; DDGE: Denaturing gradient gel public health implications. Annu Rev Public Health. 2011;32:367–79. electrophoresis; EMMPRIN: Extracellular matrix metalloproteinase inducer; 3. Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes of HCA: Hierarchical cluster analysis; HIV: Human immunodeficiency virus; preterm birth. Lancet. 2008;371(9606):75–84. MMP: Metalloproteinase; NHS: National Health Service; NPV: Negative 4. Liu L, Oza S, Hogan D, Perin J, Rudan I, Lawn JE, Cousens S, Mathers C, Black predictive value; OD: Omne in die (once per day); OR: Odds ratio; RE. Global, regional, and national causes of child mortality in 2000-13, with PPV: Positive predictive value; RDP: Ribosomal database project; projections to inform post-2015 priorities: an updated systematic analysis. REC: Research ethics committee; rRNA: Ribosomal ribonucleic acid; Lancet. 2015;385(9966):430–40. SD: Standard deviation; STAMP: Statistical Analysis of Metagenomic Profiles; 5. Romero R, Dey SK, Fisher SJ. Preterm labor: one syndrome, many causes. TVS: Transvaginal scan Science. 2014;345(6198):760–5. Kindinger et al. Microbiome (2017) 5:6 Page 13 of 14

6. Eckburg PB, Bik EM, Bernstein CN, Purdom E, Dethlefsen L, Sargent M, Gill 26. Tamrakar R, Yamada T, Furuta I, Cho K, Morikawa M, Yamada H, Sakuragi N, SR, Nelson KE, Relman DA. Diversity of the human intestinal microbial flora. Minakami H. Association between Lactobacillus species and bacterial Science. 2005;308(5728):1635–8. vaginosis-related bacteria, and bacterial vaginosis scores in pregnant 7. Turnbaugh PJ, Ley RE, Mahowald MA, Magrini V, Mardis ER, Gordon JI. An Japanese women. BMC Infect Dis. 2007;7:128. obesity-associated gut microbiome with increased capacity for energy 27. Srinivasan S, Hoffman NG, Morgan MT, Matsen FA, Fiedler TL, Hall RW, Ross harvest. Nature. 2006;444(7122):1027–31. FJ, McCoy CO, Bumgarner R, Marrazzo JM, et al. Bacterial communities in 8. MacIntyre DA, Chandiramani M, Lee YS, Kindinger L, Smith A, Angelopoulos women with bacterial vaginosis: high resolution phylogenetic analyses N, Lehne B, Arulkumaran S, Brown R, Teoh TG, et al. The vaginal reveal relationships of microbiota to clinical criteria. PLoS One. microbiome during pregnancy and the postpartum period in a European 2012;7(6), e37818. population. Sci Rep. 2015;5:8988. 28. Africa CW, Nel J, Stemmet M. Anaerobes and bacterial vaginosis in 9. Romero R, Hassan SS, Gajer P, Tarca AL, Fadrosh DW, Nikita L, Galuppi M, pregnancy: virulence factors contributing to vaginal colonisation. Int J Lamont RF, Chaemsaithong P, Miranda J, et al. The composition and Environ Res Public Health. 2014;11(7):6979–7000. stability of the vaginal microbiota of normal pregnant women is different 29. Word RA, Li XH, Hnat M, Carrick K. Dynamics of cervical remodeling during from that of non-pregnant women. Microbiome. 2014;2(1):4. pregnancy and parturition: mechanisms and current concepts. Semin 10. Reid G, Younes JA, Van der Mei HC, Gloor GB, Knight R, Busscher HJ. Reprod Med. 2007;25(1):69–79. Microbiota restoration: natural and supplemented recovery of human 30. Guzeloglu-Kayisli O, Kayisli UA, Semerci N, Basar M, Buchwalder LF, microbial communities. Nat Rev Microbiol. 2011;9(1):27–38. Buhimschi CS, Buhimschi IA, Arcuri F, Larsen K, Huang JS, et al. Mechanisms 11. van de Wijgert JH, Jespers V. Incorporating microbiota data into of chorioamnionitis-associated preterm birth: interleukin-1beta inhibits epidemiologic models: examples from vaginal microbiota research. Ann progesterone receptor expression in decidual cells. J Pathol. 2015. Epidemiol. 2016;26(5):360–5. 31. Ekman-Ordeberg G, Dubicke A. Preterm cervical ripening in humans. Facts 12. Borgdorff H, Tsivtsivadze E, Verhelst R, Marzorati M, Jurriaans S, Ndayisaba Views Vis Obgyn. 2012;4(4):245–53. GF, Schuren FH, van de Wijgert JH. Lactobacillus-dominated cervicovaginal 32. Grimes-Dennis J, Berghella V. Cervical length and prediction of preterm microbiota associated with reduced HIV/STI prevalence and genital HIV viral delivery. Curr Opin Obstet Gynecol. 2007;19(2):191–5. load in African women. ISME J. 2014;8(9):1781–93. 33. Iams JD, Goldenberg RL, Meis PJ, Mercer BM, Moawad A, Das A, Thom E, 13. Ravel J, Gajer P, Abdo Z, Schneider GM, Koenig SS, McCulle SL, Karlebach S, McNellis D, Copper RL, Johnson F, et al. The length of the cervix and the Gorle R, Russell J, Tacket CO, et al. Vaginal microbiome of reproductive-age risk of spontaneous premature delivery. National Institute of Child Health women. Proc Natl Acad Sci U S A. 2011;108 Suppl 1:4680–7. and Human Development Maternal Fetal Medicine Unit Network. N Engl J 14. Chaban B, Links MG, Jayaprakash TP, Wagner EC, Bourque DK, Lohn Z, Med. 1996;334(9):567–72. Albert AY, van Schalkwyk J, Reid G, Hemmingsen SM, et al. Characterization 34. Iams JD. Prevention of preterm parturition. N Engl J Med. 2014;370(19):1861. of the vaginal microbiota of healthy Canadian women through the 35. Romero R, Nicolaides K, Conde-Agudelo A, Tabor A, O’Brien JM, Cetingoz E, menstrual cycle. Microbiome. 2014;2:23. Da Fonseca E, Creasy GW, Klein K, Rode L, et al. Vaginal progesterone in 15. van de Wijgert JHHM, Verwijs MC, Turner AN, Morrison CS. Hormonal women with an asymptomatic sonographic short cervix in the midtrimester contraception decreases bacterial vaginosis but oral contraception may increase decreases preterm delivery and neonatal morbidity: a systematic review and candidiasis: implications for HIV transmission. Aids. 2013;27(13):2141–53. metaanalysis of individual patient data. Am J Obstet Gynecol. 2012;206(2): 16. Brotman RM, Shardell MD, Gajer P, Fadrosh D, Chang K, Silver MI, Viscidi RP, 124. e121-119. Burke AE, Ravel J, Gravitt PE. Association between the vaginal microbiota, 36. Alfirevic Z, Stampalija T, Roberts D, Jorgensen AL. Cervical stitch (cerclage) menopause status, and signs of vulvovaginal atrophy. Menopause. for preventing preterm birth in singleton pregnancy. Cochrane Database 2014;21(5):450–8. Syst Rev. 2012;4, CD008991. 17. Shen J, Song N, Williams CJ, Brown CJ, Yan Z, Xu C, Forney LJ. Effects of low 37. Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Smith A, McDonald JA, dose estrogen therapy on the vaginal microbiomes of women with Terzidou V, Cook JR, Lees C, Israfil-Bayli F, et al. Relationship between atrophic vaginitis. Sci Rep. 2016;6:24380. vaginal microbial dysbiosis, inflammation, and pregnancy outcomes in 18. Spear GT, French AL, Gilbert D, Zariffard MR, Mirmonsef P, Sullivan TH, Spear cervical cerclage. Sci Transl Med. 2016;8(350):350ra. 102. WW, Landay A, Micci S, Lee BH, et al. Human alpha-amylase present in 38. Resseguie LJ, Hick JF, Bruen JA, Noller KL, O’Fallon WM, Kurland LT. lower-genital-tract mucosal fluid processes glycogen to support vaginal Congenital malformations among offspring exposed in utero to progestins, colonization by Lactobacillus. J Infect Dis. 2014;210(7):1019–28. Olmsted County, Minnesota, 1936-1974. Fertil Steril. 1985;43(4):514–9. 19. DiGiulio DB, Callahan BJ, McMurdie PJ, Costello EK, Lyell DJ, Robaczewska A, 39. Northen AT, Norman GS, Anderson K, Moseley L, Divito M, Cotroneo M, Sun CL, Goltsman DS, Wong RJ, Shaw G, et al. Temporal and spatial Swain M, Bousleiman S, Johnson F, Dorman K, et al. Follow-up of children variation of the human microbiota during pregnancy. Proc Natl Acad Sci U exposed in utero to 17 alpha-hydroxyprogesterone caproate compared S A. 2015;112(35):11060–5. with placebo. Obstet Gynecol. 2007;110(4):865–72. 20. Leitich H, Kiss H. Asymptomatic bacterial vaginosis and intermediate flora as 40. Anderson L, Martin W, Higgins C, Nelson SM, Norman JE. The effect of risk factors for adverse pregnancy outcome. Best Pract Res Clin Obstet progesterone on myometrial contractility, potassium channels, and tocolytic Gynaecol. 2007;21(3):375–90. efficacy. Reprod Sci. 2009;16(11):1052–61. 21. Chandiramani M, Bennett PR, Brown R, Lee YS, MacIntyre DA. Vaginal 41. Hardy DB, Janowski BA, Corey DR, Mendelson CR. Progesterone receptor microbiome-pregnant host interactions determine a significant proportion plays a major antiinflammatory role in human myometrial cells by of preterm labour. Fetal Matern Med Rev. 2014;25(1):73–8. antagonism of nuclear factor-kappaB activation of cyclooxygenase 2 22. Merrifield CA, Lewis MC, Berger B, Cloarec O, Heinzmann SS, Charton F, expression. Mol Endocrinol. 2006;20(11):2724–33. Krause L, Levin NS, Duncker S, Mercenier A, et al. Neonatal environment 42. Loudon JA, Elliott CL, Hills F, Bennett PR. Progesterone represses interleukin- exerts a sustained influence on the development of the intestinal 8 and cyclo-oxygenase-2 in human lower segment fibroblast cells and microbiota and metabolic phenotype. ISME J. 2015. amnion epithelial cells. Biol Reprod. 2003;69(1):331–7. 23. Dominguez-Bello MG, Costello EK, Contreras M, Magris M, Hidalgo G, Fierer 43. Tan H, Yi L, Rote NS, Hurd WW, Mesiano S. Progesterone receptor-A and -B N, Knight R. Delivery mode shapes the acquisition and structure of the have opposite effects on proinflammatory gene expression in human initial microbiota across multiple body habitats in newborns. Proc Natl Acad myometrial cells: implications for progesterone actions in human pregnancy Sci U S A. 2010;107(26):11971–5. and parturition. J Clin Endocrinol Metab. 2012;97(5):E719–30. 24. Romero R, Hassan SS, Gajer P, Tarca AL, Fadrosh DW, Bieda J, 44. Carbonne B, Dallot E, Haddad B, Ferré F, Cabrol D. Effects of progesterone Chaemsaithong P, Miranda J, Chaiworapongsa T, Ravel J. The vaginal on prostaglandin E(2)-induced changes in glycosaminoglycan synthesis by microbiota of pregnant women who subsequently have spontaneous human cervical fibroblasts in culture. Mol Hum Reprod. 2000;6(7):661–4. preterm labor and delivery and those with a normal delivery at term. 45. Nold C, Maubert M, Anton L, Yellon S, Elovitz MA. Prevention of preterm Microbiome. 2014;2:18. birth by progestational agents: what are the molecular mechanisms? Am J 25. Petricevic L, Domig KJ, Nierscher FJ, Sandhofer MJ, Fidesser M, Krondorfer I, Obstet Gynecol. 2013;208(3):223. e221-227. Husslein P, Kneifel W, Kiss H. Characterisation of the vaginal Lactobacillus 46. Yellon SM, Dobyns AE, Beck HL, Kurtzman JT, Garfield RE, Kirby MA. Loss of microbiota associated with preterm delivery. Sci Rep. 2014;4:5136. progesterone receptor-mediated actions induce preterm cellular and Kindinger et al. Microbiome (2017) 5:6 Page 14 of 14

structural remodeling of the cervix and premature birth. PLoS One. 2013; 67. Donders GG, Van Calsteren K, Bellen G, Reybrouck R, Van den Bosch T, 8(12), e81340. Riphagen I, Van Lierde S. Predictive value for preterm birth of abnormal 47. Facchinetti F, Paganelli S, Comitini G, Dante G, Volpe A. Cervical length vaginal flora, bacterial vaginosis and aerobic vaginitis during the first changes during preterm cervical ripening: effects of 17-alpha- trimester of pregnancy. BJOG. 2009;116(10):1315–24. hydroxyprogesterone caproate. Am J Obstet Gynecol. 2007;196(5):453. e451- 68. Berghella V, Bega G, Tolosa JE, Berghella M. Ultrasound assessment of the 454; discussion 421. cervix. Clin Obstet Gynecol. 2003;46(4):947–62. 48. Hassan S, Romero R, Vidyadhari D, Fusey S, Baxter J, Khandelwal M, 69. Rampersaud R, Planet PJ, Randis TM, Kulkarni R, Aguilar JL, Lehrer RI, Ratner Vijayaraghavan J, Trivedi Y, Soma‐Pillay P, Sambarey P. Vaginal progesterone AJ. Inerolysin, a cholesterol-dependent cytolysin produced by Lactobacillus reduces the rate of preterm birth in women with a sonographic short iners. J Bacteriol. 2011;193(5):1034–41. cervix: a multicenter, randomized, double-blind, placebo-controlled trial. 70. Macklaim JM, Fernandes AD, Di Bella JM, Hammond JA, Reid G, Gloor GB. Ultrasound Obstet Gynecol. 2011;38(1):18–31. Comparative meta-RNA-seq of the vaginal microbiota and differential 49. Fonseca EB, Celik E, Parra M, Singh M, Nicolaides KH, Group FMFSTS. expression by Lactobacillus iners in health and dysbiosis. Microbiome. Progesterone and the risk of preterm birth among women with a short 2013;1(1):12. cervix. N Engl J Med. 2007;357(5):462–9. 71. Doerflinger SY, Throop AL, Herbst-Kralovetz MM. Bacteria in the vaginal 50. O’Brien JM, Adair CD, Lewis DF, Hall DR, Defranco EA, Fusey S, Soma-Pillay P, microbiome alter the innate immune response and barrier properties of Porter K, How H, Schackis R, et al. Progesterone vaginal gel for the reduction of the human vaginal epithelia in a species-specific manner. J Infect Dis. recurrent preterm birth: primary results from a randomized, double-blind, 2014;209(12):1989–99. placebo-controlled trial. Ultrasound Obstet Gynecol. 2007;30(5):687–96. 72. Anahtar MN, Byrne EH, Doherty KE, Bowman BA, Yamamoto HS, Soumillon 51. van Os MA, van der Ven AJ, Kleinrouweler CE, Schuit E, Kazemier BM, M, Padavattan N, Ismail N, Moodley A, Sabatini ME, et al. Cervicovaginal Verhoeven CJ, de Miranda E, van Wassenaer-Leemhuis AG, Sikkema JM, bacteria are a major modulator of host inflammatory responses in the Woiski MD, et al. Preventing preterm birth with progesterone in women female genital tract. Immunity. 2015;42(5):965–76. with a short cervical length from a low-risk population: a multicenter 73. Joergensen JS, Kjaer Weile LK, Lamont RF. The early use of appropriate double-blind placebo-controlled randomized trial. Am J Perinatol. 2015; prophylactic antibiotics in susceptible women for the prevention of preterm 32(10):993–1000. birth of infectious etiology. Expert Opin Pharmacother. 2014;15(15):2173–91. 52. Borgdorff H, Verwijs MC, Wit FW, Tsivtsivadze E, Ndayisaba GF, Verhelst R, 74. Guaschino S, De Seta F, Piccoli M, Maso G, Alberico S. Aetiology of preterm Schuren FH, van de Wijgert JH. The impact of hormonal contraception and labour: bacterial vaginosis. BJOG. 2006;113 Suppl 3:46–51. pregnancy on sexually transmitted infections and on cervicovaginal 75. Verstraelen H, Verhelst R, Claeys G, De Backer E, Temmerman M, microbiota in african sex workers. Sex Transm Dis. 2015;42(3):143–52. Vaneechoutte M. Longitudinal analysis of the vaginal microflora in 53. Ralph LJ, McCoy SI, Shiu K, Padian NS. Hormonal contraceptive use and pregnancy suggests that L. crispatus promotes the stability of the normal women’s risk of HIV acquisition: a meta-analysis of observational studies. vaginal microflora and that L. gasseri and/or L. iners are more conducive to Lancet Infect Dis. 2015;15(2):181–9. the occurrence of abnormal vaginal microflora. BMC Microbiol. 2009;9:116. 54. Blish CA, Baeten JM. Hormonal contraception and HIV-1 transmission. Am J 76. Ferris MJ, Norori J, Zozaya-Hinchliffe M, Martin DH. Cultivation-independent Reprod Immunol. 2011;65(3):302–7. analysis of changes in bacterial vaginosis flora following metronidazole 55. Sundquist A, Bigdeli S, Jalili R, Druzin ML, Waller S, Pullen KM, El-Sayed YY, treatment. J Clin Microbiol. 2007;45(3):1016–8. Taslimi MM, Batzoglou S, Ronaghi M. Bacterial flora-typing with targeted, 77. Nunn KL, Wang YY, Harit D, Humphrys MS, Ma B, Cone R, Ravel J, Lai chip-based pyrosequencing. BMC Microbiol. 2007;7:108. SK. Enhanced trapping of HIV-1 by human cervicovaginal mucus is 56. Kozich JJ, Westcott SL, Baxter NT, Highlander SK, Schloss PD. Development associated with Lactobacillus crispatus-dominant microbiota. MBio. 2015; of a dual-index sequencing strategy and curation pipeline for analyzing 6(5):e01084–01015. amplicon sequence data on the MiSeq Illumina sequencing platform. Appl 78. Brotman RM, Shardell MD, Gajer P, Tracy JK, Zenilman JM, Ravel J, Gravitt PE. Environ Microbiol. 2013;79(17):5112–20. Interplay between the temporal dynamics of the vaginal microbiota and 57. Wang Q, Garrity GM, Tiedje JM, Cole JR. Naive Bayesian classifier for rapid human papillomavirus detection. J Infect Dis. 2014;210(11):1723–33. assignment of rRNA sequences into the new bacterial taxonomy. Appl 79. Linhares IM, Summers PR, Larsen B, Giraldo PC, Witkin SS. Contemporary Environ Microbiol. 2007;73(16):5261–7. perspectives on vaginal pH and lactobacilli. Am J Obstet Gynecol. 2011; 58. Edgar RC. Search and clustering orders of magnitude faster than BLAST. 204(2):120. e121-125. Bioinformatics. 2010;26(19):2460–1. 80. Witkin SS, Mendes-Soares H, Linhares IM, Jayaram A, Ledger WJ, Forney LJ. 59. Schloss PD, Westcott SL, Ryabin T, Hall JR, Hartmann M, Hollister EB, Influence of vaginal bacteria and D- and L-lactic acid isomers on vaginal Lesniewski RA, Oakley BB, Parks DH, Robinson CJ, et al. Introducing mothur: extracellular matrix metalloproteinase inducer: implications for protection open-source, platform-independent, community-supported software for against upper genital tract infections. MBio. 2013;4(4):e00460–13. describing and comparing microbial communities. Appl Environ Microbiol. 81. Berghella V, Roman A, Daskalakis C, Ness A, Baxter JK. Gestational age at 2009;75(23):7537–41. cervical length measurement and incidence of preterm birth. Obstet 60. Parks DH, Beiko RG. Identifying biologically relevant differences between Gynecol. 2007;110(2 Pt 1):311–7. metagenomic communities. Bioinformatics. 2010;26(6):715–21. 82. Gotkin JL, Celver J, McNutt P, Shields AD, Howard BC, Paonessa DJ, 61. Breslow N. Covariance analysis of censored survival data. Biometrics. 1974; Napolitano PG. Progesterone reduces lipopolysaccharide induced 30(1):89–99. interleukin-6 secretion in fetoplacental chorionic arteries, fractionated 62. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical cord blood, and maternal mononuclear cells. Am J Obstet Gynecol. and powerful approach to multiple testing. J R Stat Soc Ser B Methodol. 2006;195(4):1015–9. 1995;57(1):289–300. 63. Donders GG, Bosmans E, Dekeersmaecker A, Vereecken A, Van Bulck B, Spitz B. Pathogenesis of abnormal vaginal bacterial flora. Am J Obstet Gynecol. 2000;182(4):872–8. 64. Gupta K, Stapleton AE, Hooton TM, Roberts PL, Fennell CL, Stamm WE. Inverse association of H2O2-producing lactobacilli and vaginal Escherichia coli colonization in women with recurrent urinary tract infections. J Infect Dis. 1998;178(2):446–50. 65. Hillier SL, Krohn MA, Rabe LK, Klebanoff SJ, Eschenbach DA. The normal vaginal flora, H2O2-producing lactobacilli, and bacterial vaginosis in pregnant women. Clin Infect Dis. 1993;16 Suppl 4:S273–81. 66. Martius J, Krohn MA, Hillier SL, Stamm WE, Holmes KK, Eschenbach DA. Relationships of vaginal Lactobacillus species, cervical Chlamydia trachomatis, and bacterial vaginosis to preterm birth. Obstet Gynecol. 1988; 71(1):89–95. RESEARCH ARTICLE Preterm Birth Prevention Post-Conization: A Model of Cervical Length Screening with Targeted Cerclage

Lindsay M. Kindinger1,2, Maria Kyrgiou1,3, David A. MacIntyre1, Stefano Cacciatore1, Angela Yulia1,4, Joanna Cook1, Vasso Terzidou1,4, T. G. Teoh1,2, Phillip R. Bennett1,3

1 Institute of Reproductive and Developmental Biology, Department of Surgery and Cancer, Imperial College, London, United Kingdom, 2 Department of Obstetrics and Gynaecology, St Mary’s Hospital, Imperial College Healthcare NHS Trust, London, United Kingdom, 3 Department of Obstetrics and Gynaecology, Queen Charlotte’s and Chelsea—Hammersmith Hospital, Imperial College Healthcare NHS a11111 Trust, London, United Kingdom, 4 Department of Obstetrics and Gynaecology, Chelsea and Westminster Hospital NHS Trust, London, United Kingdom

* [email protected]

Abstract Women with a history of excisional treatment (conization) for cervical intra-epithelial neo- Citation: Kindinger LM, Kyrgiou M, MacIntyre DA, Cacciatore S, Yulia A, Cook J, et al. (2016) Preterm plasia (CIN) are at increased risk of preterm birth, perinatal morbidity and mortality in sub- Birth Prevention Post-Conization: A Model of sequent pregnancy. We aimed to develop a screening model to effectively differentiate Cervical Length Screening with Targeted Cerclage. pregnancies post-conization into low- and high-risk for preterm birth, and to evaluate the PLoS ONE 11(11): e0163793. doi:10.1371/journal. impact of suture material on the efficacy of ultrasound indicated cervical cerclage. We ana- pone.0163793 lysed longitudinal cervical length (CL) data from 725 pregnant women post-conization Editor: Rogelio Cruz-Martinez, Fetal Medicine and attending preterm surveillance clinics at three London university Hospitals over a ten year Surgery, MEXICO period (2004–2014). Rates of preterm birth 37 weeks after targeted cerclage for Received: February 14, 2016 CL25mm were compared with local and national background rates and expected rates for Accepted: September 14, 2016 this cohort. Rates for cerclage using monofilament or braided suture material were also Published: November 3, 2016 compared. Of 725 women post-conization 13.5% (98/725) received an ultrasound indicated

Copyright: 2016 Kindinger et al. This is an open cerclage and 9.7% (70/725) delivered prematurely, 37weeks; 24.5% (24/98) of these access article distributed under the terms of the despite insertion of cerclage. The preterm birth rate was lower for those that had monofila- Creative Commons Attribution License,which ment (9/60, 15%) versus braided (15/38, 40%) cerclage (RR 0.7, 95% CI 0.54 to 0.94, P = permits unrestricted use, distribution, and 0.008). Accuracy parameters of interval reduction in CL between longitudinal second tri- reproduction in any medium, provided the original author and source are credited. mester screenings were calculated to identify women at low risk of preterm birth, who could safely discontinue surveillance. A reduction of CL 10% between screening timepoints pre- Data Availability Statement: We have uploaded our data to datadryad.org under the details: DOI: dicts term birth, 37weeks. Our triage model enables timely discharge of low risk women, 10.5061/dryad.r7r01 Data files: Kindinger eliminating 36% of unnecessary follow-up CL scans. We demonstrate that preterm birth in data_Uploaded PlosOne. women post-conization may be reduced by targeted cervical cerclage. Cerclage efficacy is Funding: This study was supported in part by the however suture material-dependant: monofilament is preferable to braided suture. The Genesis Research Trust (P51389), the British introduction of triage prediction models has the potential to reduce the number of unneces- Society of Colposcopy Cervical Pathology Jordan/ Singer Award (P47773), and by the Imperial sary CL scan for women at low risk of preterm birth. Healthcare NHS Trust Biomedical Research Centre (Grant Ref P45272). MK received support and a research award from the British Society of

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 1 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Colposcopy and Cervical Pathology and the Introduction Imperial College Healthcare Charity. Women with a history of excisional treatment (conization) for cervical intra-epithelial neopla- Competing Interests: There is no conflict of interest related to this work. sia (CIN) have a shorter cervical length (CL) in pregnancy than those without treatment [1], significantly increasing their risk of preterm birth <37weeks, perinatal morbidity and mortality [2–8]. Although the underlying mechanisms are unclear, hypotheses include immunomodula- tion relating to HPV infection affecting biochemical pathways to parturition, and ‘mechanical weakness’ secondary to loss of cervical tissue [9]. Second trimester CL measurements in preg- nancy post-conization are as predictive for preterm birth [1, 10, 11]astheyareforthegeneral obstetric population, in whom a cut-off of 25 mm is widely used as a treatment threshold [10, 12–14]. Pregnancies post-conization account for an increasing proportion of referrals to pre- term CL surveillance clinics; from none in 1999, to more than 40% of referrals in 2012 [15]. The antenatal management of those found to have a short cervix is largely unstandardized and remains clinician and unit-dependant. Evidence for progesterone treatment is lacking, and while a cerclage is often inserted to mechanically support the deficient cervix [16], increasing evidence from retrospective case series suggest that cerclage insertion is either of no benefit, or may even worsen outcome through increased preterm birth rates [17–20]. However these were studies in cohorts in whom cerclage was performed using braided suture material; a reflection of current global clinical practice. In the UK Mersilene™, a non-absorbable braided polyester suture, is used by over 80% of obstetricians in preference to a monofilament alternative such as Nylon or Prolene [21], despite a lack of evidence-base. The only study to report on the effect of suture material on cerclage efficacy compared two types of braided suture, Ethibond™ and Mer- silene™, and excluded all monofilament cerclages. This study showed no difference in preterm birth rates [22]. The multifilament structure of braided suture favours bacterial colonisation over monofila- ment suture [23–25], and is associated with poorer wound healing [26–28]. This may explain the doubled risk of puerperal pyrexia associated with cerclage insertion [29], and the lack of any observedbenefit of a cerclage for preterm birth prevention post-conization [17–20]. Hypothesizing that suture material effects cerclage efficacy post-conization, we compared rates of preterm birth for monofilament versus braided suture material in a retrospective observa- tional study of cerclage procedures across three hospitals, over ten years. Despite accounting for a large proportion of referrals for labour intensive and costly ultra- sound-directed preterm surveillance, the majority of pregnancies post-conization will deliver at term (80–86%) without intervention [2, 5]. Therefore a secondary aim of this study to develop a triage prediction model to clearly differentiate high-risk women, in whom targeted intervention may be beneficial, from those at low-risk of preterm birth who could discontinue surveillance in a safe and timely manner.

Methods Aretrospectiveobservationalstudywasconductedinthepretermsurveillanceclinicsatthree London University maternity units (Queen Charlottes Hospital, St Marys Hospital, Chelsea Westminster Hospital) from January 2004 to January 2014. We included all women during their first singleton pregnancy after excisional cervical treatment for CIN of depth 12mm ! [30] (including Cone biopsy, LLETZ and LEEP). We included women that required an ultra- sound-indicated cerclage, but excluded those undergoing a pre-planned or history-indicated cervical cerclage. Women were only eligible if they had no other risk factors for preterm birth; any women with a prior preterm delivery (<37 weeks), mid-trimester miscarriage (>13 weeks), uterine anomaly or a multi-fetal pregnancy in the index pregnancy preterm were excluded. Women were included for analysis if CL measurements were available for at least

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 2 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

one of three screening time-points; A: 13+0–15+6 weeks, B: 16+0–18+6 weeks, and C: 20+0–22+6 weeks. This was retrospectiveanalysis of previously collected,anonymized data and did not require ethics approval. All hospitals implemented a pre-specified surveillance protocol, applicable to all women attending prematurity clinics across sites for the duration of the study. This included serial mid-trimester CL measurements and cervical cerclage for those with a cervix shorter than 25mm before 24 weeks of gestation. CL measurements were taken at trans-vaginal ultrasound (TVUS) with an empty bladder, avoiding undue pressure on the cervix, and fundal pressure applied to illicit any further cervicalshortening. The primary objectives were to assess the overall efficacy of cerclage in women with a short- ened cervix post-conization, and the effect of suture material on preterm birth rates. The over- all preterm birth rates in the observed cerclage study population were therefore compared to background preterm birth rates in the general obstetric population at the three university hos- pitals, and to national preterm birth rates for England and Wales as taken from The Office of National Statistics Data UK 2005 (mid-point in thestudyperiod).Fisher’sexacttestwasused to assess whether the rate of preterm birth <37 weeks differs in women with a monofilament versus braided cerclage. The second objective was to develop a triage prediction model to differentiate women at low-risk for preterm birth in whom CL surveillance is unnecessary, from women at high-risk. The study cohort was classified into two groups. Group 1 were ‘low-risk’; they delivered at term, did not demonstrate cervical shortening and did not receive cerclage intervention. Group 2wereconsidered‘high-risk’astheyeitherhadanultrasound-indicatedcerclageforcervical shortening below 25mm, or delivered prematurely (less than 37 weeks of gestation). To determine if these pre-defined high and low risk groups demonstrated differing patterns of cervical shortening at screening, Group 1 and Group 2 were assessed for the percentage change in CL (%CL) between serial screening time-points (A, B and C), using Kruskal-Wallis and Mann Whitney tests. Only CL measurements made before the insertionof cerclage were included in the analysis. We calculatedthe accuracyof different thresholds for percentage CL ( 5%, 10%, 20%, ! ! ! 30%, 40%) in predicting preterm birth (<37 weeks) and/or the need for cerclage, as well as ! ! the accuracy of single CL measurements ( 15mm, 20mm, 25mm, 30mm and 35mm).

Accuracy parameters included sensitivity (S), specificity (Sp), positive (PPV), negative predic- tive value (NPV) and likelihoodratios (LR). Receiver operator curves(ROC) were plotted and used to determine optimum CL thresholds. A triage prediction model was developed using decision tree analysis (R package “rpart”[31]) to identify pregnancies at low-risk of preterm birth that could safely discontinue serial surveillance. The model was designed to predict term birth (beyond 36+6 weeks) without requirement for cerclage with high sensitivity and a low false positive rate (5%), ensuring a minimal number of women that delivered preterm were falsely classified as ‘low-risk’ within the model. These highly conservative margins ensured identification of only the lowest risk women suitable for discharge from surveillance. To assess the prediction ability, the model was built on two independent sets and tested on the remaining set. The whole procedure was repeated three times testing each hospital location. The model consists of 3 steps. In the first step, CL at timepoint A was used as predictor. Two thresholds were defined to classify the par- ticipant as high risk, low risk, or requiring further screening at timepoint B and/or C. In the second step, the CL at time-point B and change in CL between A and B were used as predictors. In the third step, the classification tree was used as final predictor using as variables CL at time-point A, time-point B and time-point C, and all the variations between timepoints.

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 3 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

The produced algorithm was then applied retrospectively to our cohort to calculate the number of unnecessary ultrasound scans. Based on an assumption that each patient attended for at least three transvaginal scans, we estimated the total number of transvaginal scans that could have been prevented. Potential sources of bias were considered. Notably this included selectionof participants, as only data on women referred to prematurity clinics were included. Selectionof suture material for cerclage may have been biased as the retrospective design precluded randomization.

Results A total of 725 pregnant women post-conization were included in the analysis. The patient char- acteristics and the distribution into low-risk term birth (Group 1) and those delivering preterm or receiving an ultrasound indicated cerclage (Group 2) are describedin Table 1. Age, BMI, and ethnicity were comparable amongst the two groups. Smoking was more prevalent in Group 2 (33/144, 23%) than in Group 1 (41/581, 7%). Ultrasound-indicated cerclage was

Table 1. Patient characteristics for women that delivered at term without cerclage (Group 1) and women that had cerclage or delivered prema- turely (Group 2). Term birth(without intervention)Group 1, PTB 37w or cerclage insertion Group 2, Total population N = 581 N = 144 N = 725 Age, years. Mean (SD, range) 33.8 (4.2, 24–49) 33.7 (3.6, 26–44) 33.8 (4.1, 24–49) BMI. Mean (SD, range) 24.4 (4.1, 18–40) 23.5 (3.5, 18–34) 24.1 (3.9, 18–40) Ethnicity, n/N (%) Caucasian 381/581 (66%) 98/144 (68%) 479/725 (66%) Asian 96/581(16%) 20 /144 (14%) 116/725 (16%) Black 104/581 (18%) 26 /144 (18%) 130/725 (18%) Parity, n/N (%) 0 422 /581 (76%) 124 /144 (86%) 566/725 (78%) 1 139/581 (24%) 20 /144 (14%) 160/725 (22%) ! Smoker, n/N (%) 41/581 (7%) 33 /144 (23%)* 87/725 (12%) Preterm birth, n/N (%) N/A 70/144 (49%) 70/725 (9.7%) Cervical Cerclage for CL 25mm, n/ N (%) Cerclage inserted N/A 98/144 (68%) 98/725 (13.5%) Preterm birth, with cerclage N/A 24/98 (24%) 24/725 (3%) Term birth, with cerclage N/A 74/98 (76%) 74/725 (10%) CL (mm) at timepoints, Mean (SD) [n] A: 13+0–15+6 34 (4.2) [481] 28 (6.3)* [129] 32 (5.3) [610] B: 16+0–18+6 33 (4.4) [493] 27 (6.7)* [102] 32 (5.4) [595] C: 20+0–22+6 32 (4.4) [492] 25 (7.3)* [62] 31 (5.2) [554] % CL between timepoints, Mean (SD) [n] A-B 3% (8) [426] 11% (13)* [96] 4% (10) [522] B-C 2% (9) [452] 18% (20)* [55] 4% (12) [507] A-C 5% (11) [413] 24% (20)* [52] 7% (14) [465]

Group 1 = delivery 37weeks without intervention; Group 2 = preterm birth 37weeks and/or cerclage. BMI = body mass index; CL = cervical length (mm); % CL = percentage change in CL (mm) between screening time points; GA = gestational age; PTB = preterm birth 37 weeks; Screening timepoints: A: 13 +0–15+6 weeks, B: 16+0–18+6 weeks, C: 20+0–22+6 weeks; SD = standard deviation; W = weeks *P 0.05 for comparisons Group 1 vs Group 2 doi:10.1371/journal.pone.0163793.t001

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 4 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Table 2. Preterm birth rates: A in the UK; B in the local population (three study units); C estimated in women post-cervical treatment based on UK rates; D in this study cohort. GA at birth A. UK (2005) B. Local population C. Post-treatment estimate D. Our cohort 37 w 11% 13% 22% 9% 34 w 3% 3% 6% 2%

GA = gestational age; w = weeks *based on Relative Risk (RR) reported in Kyrgiou Lancet 2006; Arbyn BMJ 2008 doi:10.1371/journal.pone.0163793.t002

inserted in 14% (98/725) and preterm birth <37 weeks occurred in 10% (70/725), 24 of these despite insertion of cerclage (24/70, 34%). Fourteen delivered at <34 weeks (14/725, 2%), of which 6 (6/14, 43%) had cerclage. The background rate of preterm birth in 2005 (the midpoint of this study) across the three study units was 13% (<37 weeks) and 3% (<34 weeks). The rates in England and Wales were 11% (<37 weeks) and 3% (<34 weeks) over the same year (Table 2). For women who received a cerclage, the mean CL and gestational age at cerclage insertion was 20mm (SD = 4.0) and 18+6 weeks (SD = 3.9) respectively. This did not differ significantly across hospital sites. Eight women (1%) who reached the CL threshold declined cerclage. Two of these delivered prematurely (<37 weeks). Monofilament suture material was inserted in 61% (60/98) of cerclages, and the remaining received braided suture (39%, 38/98). Choice of suture material was entirely at the discretion of the operator. Although the mean CL and gestation at insertion was similar for the two suture groups, the rate of preterm birth (<37 weeks) was lower for monofilament (15%, 9/60) com- pared to Braided cerclages (40%, 15/38) (RR 0.7, 95% CI 0.54 to 0.94, P = 0.008) (Fig 1).

Fig 1. Gestation at delivery in women with an ultrasound-indicated cerclage for CL 25mm before 24weeks: a comparison of suture material braided versus monofilament. Preterm birth 37weeks was significantly higher (P = 0.08) in women with braided cerclages, compared to monofilament cerclages. This is difference is most notable among those delivering late preterm birth (34-37weeks). doi:10.1371/journal.pone.0163793.g001

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 5 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Table 3. Neonatal outcome as a function of cerclage suture material. Braided, n = 38 Mean SD Monofilament, n = 60 Mean SD No cervical shortening/ no cerclage, n = 627 Mean (range) (range) SD (range) Gestation at birth (w) 37.3 3.4 (25–42) 38.4 2.8 (27–42) 39.1 1.8(29–42) Birth weight (g) 2890 873 (621–4210) 3173 692 (1260–4340) 3348 566 (1450–5074) Apgars 1 minute 8 2.3 (2–10) 8 1.5 (3–10) 9 1.3 (1–10) 10 minute 9 1.7 (5–10) 10 0.9* (4–10) 10 0.6 (7–10) Admission to NICU, n/ 5/38, 13% 5/60, 8% 8/627, 1.3% N,%

SD = standard deviation w = weeks gestation, g = grams, NICU = neonatal intensive care unit *P = 0.03 t-test; Braided v monofilament doi:10.1371/journal.pone.0163793.t003

Respective mean gestational ages at birth were 38.4 weeks (±2.8, range 27 to 42 weeks) and 37.3 weeks (±3.4, range 25 to 42 weeks; P = 0.056) for monofilament and braided groups. Neo- natal outcome was comparably worse for those receiving braided suture material than monofil- ament, who had lower mean birthweights (2890 vs. 3173 grams, P =0.1),lowerApgarscoresat 10 minutes of age (8 vs.10, P = 0.03), and higher rates of admission to neonatal intensive care (5/38, 13% vs. 5/60, 8%, P = 0.05; Table 3).

Difference in serial CL measurements (CL) For women who underwentcerclage, there was no differencein the mean CL of those deliver- ing before or after 37 weeks (P = 0.4; Table 4), indicating that the CL at insertion of cerclage did not predict preterm birth post-cerclage. The largest reduction in CL (%CL) was observed in those requiring a cerclage, regardless of eventual gestation at birth (P<0.05) (subgroups of Group 2) (Fig 2, Table 4). The difference in CL was greatest for comparisons made screening after 20 weeks (timepoint C: 20+0–22+6 weeks), indicating that thesewomenathigh-riskstart

Table 4. Mean CL (mm)(SD) at screening time-points A, B, C and mean percentage CL (SD) between screening time-points A-B, B-C, and A-C for Group 1 and 2. Screening timepoints Group 1 Birth 37weeks Group 2 PTB 37weeks Group 2 subgroups Total scanned, (w) without cerclage N = 581 and/or cerclage N = 144 PTB (no PTB with Term with N = 725 cerclage) N = 46 cerclage N = 24 cerclage N = 74 Mean CL (mm) (SD) [n] A: 13+0–15+6 33.6 (4.2) [481] 28.0* (6.3) [129] 32.3 (6.0) [39] 26.8* (5.4) [22] 25.8* (5.4) [68] 610 B: 16+0–18+4 32.8 (4.4) [493] 26.8* (6.7) [102] 32.3 (6.2) [35] 24.5* (5.2) [15] 23.8* (5.2) [52] 595 C: 20+0–22+6 31.8 (4.4) [492] 25.1* (7.3) [62] 30.0 (4.5) [35] 16.4* (5.9) [8] 19.7* (4.4) [19] 554 Difference between Mean % CL (SD) [n] screening timepoints A-B 3% (8) [426] 11%* (13) [96] 4% (8) [31] 18%* (12) [16] 12%* (15) [49] 522 B-C 2% (9) [452] 18%* (20) [55] 6% (11) [30] 39%* (23) [9] 30%* (15) [16] 507 A-C 5% (11) [413] 24%* (20) [52] 10% (12) [28] 46%* (21) [8] 36%* (15) [16] 465

CL = cervical length (mm); % CL = percentage change in CL (mm) between screening time points; GA = gestational age; Group 1 = delivery 37weeks without intervention; Group 2 = preterm birth 37weeks and/or cerclage; ns = not significant; PTB = preterm birth 37 weeks; Screening time points: A: 13 +0–15+6 weeks, B: 16+0–18+6 weeks, C: 20+0–22+6 weeks; SD = standard deviation; Term = birth 37 weeks; W = weeks *P0.05 for comparisons of mean CL & % CL for Group 1 vs Group 2, and Group 1 vs Group 2 subgroups, according to screening timepoints A, B and C and A-B, B-C and A-C respectively. doi:10.1371/journal.pone.0163793.t004

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 6 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Fig 2. Mean difference in CL (mean % CL) between time-points A: 13+0–15+6 weeks, B: 16+0–18+6 weeks, C: 20+0–22+6 weeks (A-B, B-C, and A-C) according to delivery outcome and cerclage insertion. In women receiving a cerclage, mean CL started above 25mm at timepoint A, and went on to shorten, most significantly at timepoint C. The greatest difference in CL is observed between timepoints B-C and A-C in those that received a cerclage and went on to deliver preterm 37weeks, followed by term delivery with a cerclage. (% CL = percentage change in CL (mm) between screening time points; PTB = preterm birth 37 weeks; Screening time points = A: 13+0–15+6 weeks, B: 16+0–18+6 weeks, C: 20+0–22+6 weeks; SD = standard deviation; Term = birth 37 weeks; W = weeks). doi:10.1371/journal.pone.0163793.g002

with a reassuring CL before 20 weeks (above 25mm), and go onto shorten (Fig 2, Table 4). The sensitivity of single CL measurements in predicting preterm birth improves with advancing screening gestation (from A to C) and increasing CL thresholds (Table 5). A100%NPVforpretermbirth>37 weeks is associated with CL >50mm, >50mm, and >37mm at screening timepoints A, B and C respectively. Observation of CL reduction before 20 weeks was highly specific for prediction of preterm birth <37 weeks and/or cerclage (99% specificity when CL 30% at timepoints A-B; Table 5), although sensitivity remained poor. ! The optimum balance between sensitivity and specificity was a CL 10% at A-C, associated ! with high negative prediction (95%).

Triage Screening Model Based on this analysis, we developed a Triage Prediction Model using discriminatory analyses to identify women with a history of cervical treatment at low risk of preterm birth <37 weeks

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 7 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Table 5. Sensitivity, specificity, likelihood ratios, and positive and negative predictive values for cerclage intervention and/or preterm birth 37 weeks, for screening time-points A, B and C, and percentage difference in CL between screening time-points A, B and C. Screening time-points (w) CL threshold (mm) S (%) Sp (%) PPV (%) NPV (%) LR A: 13+0–15+6 20 12 100 100% 81% 27.9

25 30 99 85% 84% 20.6

30 73 74 44% 91% 2.9

35 94 29 26% 95% 1.3

40 97 4.6 22% 95% 1

50 100 0.2 21% 100% 1

B: 16+0–18+6 20 17 100 95% 85% 85.3

25 41 98 84% 89% 25.5

30 78 68 34% 94% 2.5

35 92 24 20% 94% 1.2

40 99 5.1 18% 96% 1

50 100 0.2 17% 100% 1

C: 20+0–22+6 20 29 99 82% 92% 35.1

25 51 95 57% 94% 10.4

30 75 56 18% 95% 1.7

35 97 21 13% 98% 1.2

37 100 13 13% 100% 1.2

Difference between screening time-points CL threshold (mm), % S (%) Sp (%) PPV (%) NPV (%) LR A-B 5% 66 60 27% 89% 1.6 ! 10% 49 83 39% 88% 2.8 ! 20% 20 97 59% 84% 6.5 ! 30% 10 99 77% 83% 14.8 ! 40% 1 100 100% 82% 1 ! B-C 5% 73 62 19% 95% 1.9 ! 10% 56 84 30% 94% 3.5 ! 20% 44 98 69% 93% 17.9 ! 30% 24 100 87% 91% 53.4 ! 40% 13 100 100% 90% 57.5 ! A-C 5% 79 47 16% 95% 1.5 ! 10% 73 69 23% 95% 2.3 ! 20% 48 91 40% 93% 5.4 ! 30% 37 98 70% 92% 18.9 ! 40% 19 99 77% 91% 26.5 ! CL = cervical length (mm); % CL = percentage change in CL (mm) between screening time points; LR = Likelihood ratio; NPV = negative predictive value; PPV = positive predictive value; S = sensitivity; Screening time points: A: 13+0–15+6 weeks, B: 16+0–18+6 weeks, C: 20+0–22+6 weeks; Sp = specificity; W = weeks. doi:10.1371/journal.pone.0163793.t005

and/or cerclage and applied this in our cohort (Fig 3). At timepoint A (13+0–15+6 weeks), a CL threshold of <19mm identified 13% of the women who went onto deliver preterm (9/70). At the same screening timepoint, 40 of the 581 low risk women (7%) were identified using a CL threshold 42mm. Of the remaining 676 women with CL between 19mm and <42mm, 24% ! ! (135/144) went on to require either a cerclage or deliver preterm, however these were not dis- tinguishable from the remaining 541 low risk women. This group were considered at interme- diate risk and required subsequent screening at timepoint B (16+0–18+6 weeks). At timepoint

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 8 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Fig 3. Triage Screening Model for pregnancies post excisional cervical treatment for the prevention of preterm birth 37weeks. A triage screening model was developed using decision tree analyses to determine optimum thresholds of CL and % change in CL between screening timepoints A (13+0–15+6 weeks), B (16+0–18+6 weeks) and C (20+0–22+6 weeks), to ensure appropriate allocation of resources. This model identifies pregnancies at low-risk of preterm birth enabling safe and timely discharged from cervical length surveillance (green dot). Similarly early identification of high-risk pregnancies allows for timely cerclage intervention (red dot). Serial CL surveillance can therefore reserved for pregnancies considered at intermediate risk, requiring further observation. CL = cervical length (mm); CLAB% = percentage change in CL (mm) between screening time points; w = weeks. doi:10.1371/journal.pone.0163793.g003

B, a CL <23mm or a reduction in CL of 21% from timepoint A identified a further 38/144 ! (26%) high risk women going onto deliver preterm or receive a cerclage, and a CL 38mm or a ! reduction in CL of <6% at the same screening timepoint identified 89 (15%) low risk women. The remaining 552 women were considered at intermediate risk, of which 100 (18%) would require either a cerclage or would go onto deliver preterm. This intermediate risk group then went on to require screeningat timepoint C (20+0–22+6 weeks), where a further discriminatory

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 9 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

analysis was performed based on initial CL at timepoint A. CL <28mm at timepoint A with 5% reduction in CL or a CL 24mm at timepoint C, identified 38 and 16 women at risk of ! preterm birth respectively. A CL 24mm at timepoint A, but <26mm at timepoint B identified ! 46 women at risk of preterm birth, while the remaining 452 women were low risk. Assuming all patients attended initial screening at timepoint A and attended for at least three TVUS, we computed that 23% of follow up scans would have been unnecessary. More specifically, 6% (40/725) of women would have been discharged immediately after initial screening, and further 22% (156/725) would require only one additional scan. This would equate to a substantial reduction in unnecessary follow up scans (236/1450). Furthermore, after discharge of low-risk women and cerclage insertion for high-risk women, only 36% of the population (n = 258/725) would have benefitted from screening at all three timepoints; applica- tion of this model would ensure focused allocation of resources to this specific group most likely to shorten with the highest risk of preterm birth.

Discussion The antenatal management of women with a previous cervical treatment for CIN varies, while evidence on how to best manage this population is lacking. Many obstetricians believe that pre- term birth following cervical excision is a result of ‘cervical weakness’, which can be corrected by cerclage. Evidence thus far indicates that this may not be the case and therefore the risk of preterm labour could plausibly be unaffected or even worsened by cervical cerclage[17–20, 32]. As these reports were exclusively in cohorts in which braided suture was used for cerclage, we hypothesized the reported lack of cerclage efficacy in pregnancy post-conization, relates to the effects of ‘foreign’ material (cerclage) on the vaginal microenvironment[23–25]andthe immune system[33]. Braided suture is the current material of choice for the cervical cerclage despite a lack of evidence to support its preferential use [17, 21, 22]. This is the first study to reveal an advantage of using a monofilament suture in pregnancies post-conization for CIN with a shortened cervix to <25mm. Furthermore, with the proviso that the cerclage is of mono- filament suture, ours is the first study to indicate that the policy of targeted cervical cerclage is beneficial in preterm birth prevention. In our cohort of women post-conization the rate of pre- term birth was below the baseline rate for the general obstetric population across the three study hospitals sites and below the overall rates for England and Wales. We estimated the expected rates of preterm birth in our cohort without specific management as a 1.5–2 fold increase [2, 5, 34]. Our management protocol therefore reduces the risk of preterm birth in pregnancy post-conization to a rate similar to that of the general background obstetric population. Measurement of serial second trimester CL is increasing used for surveillanceof women at risk of preterm birth to balance the high specificity of early screening (<16weeks), with improved sensitivity at later screening (>22weeks) [1, 10, 11]. Observation of the rate of cervi- cal change between screening is also predictive of preterm birth in general obstetric popula- tions [35–40]. The only study assessing cervical change in women post-LLETZ concluded that it does not predict preterm birth [41]. Although this appears to contradict our findings, this may explained by that study’s design. All women with cervical shortening deemed sufficient to warrant an ultrasound indicated cerclage (the highest risk of preterm birth) were excluded, while a 17% rate of preterm birth in the remaining population questions the quality and reli- ability of the screening program. We excluded from our cohort women with additional risk fac- tors for preterm birth (previous preterm deliveries, uterine anomalies, multiple pregnancies etc.) to ensure that our screening algorithm applied specifically to those whose only risk was prior cervical conisation.

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 10 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

This study finds that in pregnancy post-conization the observation of change in second tri- mester cervical length is most valuable in the management of women whose CL is between 25mm and 30mm at screening. This challenging group makes up a significant proportion of the workload in our prematurity clinic; the positive predictive values of a single CL measuring between 25mm and 30mm do not warrant the risks of cerclage, yet the negative predictive val- ues are not reassuring enough to discharge women from surveillance. The observation of per- centage change in CL between screening time-points in these women provides clinically relevant information. A small (eg <10%) or large (eg >30%) reduction in length between screening time points justifies either a timely discharge or intervention respectively. Further- more, this study provides optimal gestational ages for screening for both serial and single CL measurements to achieve high positive and negative predictive values for preterm birth. The Triage Prediction Model demonstrates that if CL measurement screening begins before 16 weeks, only 36% of women will require screening at all three second trimester timepoints. It ensures timely identification of women at low risk who may be safely discharged, thereby focusingallocation of resources to the women at highest risk of preterm birth. The Triage Pre- diction Model provides clinicians with a user-friendly, cost-effective tool for preterm birth pre- vention in pregnancies following excisional cervical treatment. While we propose this model of management, its efficacy and cost effectiveness needs to be prospectively assessed in furtherobservationalstudies before we can comment on the true validity of the model’s application in the general obstetric population. Preventative strategies other than the cerclage, including the cervical pessary and progesterone have not yet been eval- uated in this discrete clinical group with cervical damage [13, 42, 43]. The strength of this study is that describes the largest multi-centre cohort of a homogenous population of women after cervical treatment attending for intensive antenatal surveillance in specialist prematurity clinics over a 10-year study period. The longitudinal construct of the study ensures that the sequential assessment of cervical change is from a single person and reduces potential error from individual participant variability. The limitations of the study relate to its retrospective design and the lack of a direct compar- ison population (women with prior excisional cervicaltreatment that did not attend screening). This was an important source of bias, particularly in the earlier years of the study, when the association between conization and preterm birth was not well-established, and preterm clinic referrals were unlikely to reflect all women post-conization within the population. Another source of bias was a lack of randomization of suture material at cerclage insertion. Operators may have been biased towards one suture material depending their own pre-conception as to superior suture material and how high risk they perceived the participant. This bias was con- sidered minimal as all operators appeared to use only one suture material throughout the study period. A further limitation was a lack of available data with respect to specific neonatal out- come including oxygen supply, mechanical ventilation, neonatal sepsis, and neonatal brain lesions such as leucomalacia or interventricular haemorrhage. Although the Shirodkhar and MacDonald cerclage techniques have been shown to be equally efficacious [44], we were also unable to incorporate cerclage techniques into our analy- sis as this was not always clearly specified in all operation notes; another limitation of retro- spective data collection. Due to an ethical obligation to intervene in women with a shortened cervix in pregnancy, a control group with a short cervix but without a cerclage in which to compare preterm birth rates, does not exist in current clinical practice. This is an unavoidable limitation of clinical preterm birth research.

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 11 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

Conclusion In pregnancy following excisional cervicaltreatment for CIN, insertion of a non-absorbable monofilament cerclage for a shortened cervix effectively reduces the preterm birth risk and is preferable to a braided cerclage. Using our Triage Prediction Model pregnancies at low-risk of preterm birth may be identifiedand discharged from preterm cervicallength surveillancein a cost effective and timely manner.

Acknowledgments We acknowledge the prematurity clinic doctors across the hospital sites over the years: J Lou- don, L Sykes, L Alabi-Isama, M Chandiramani, B Jones and S Chatfield.

Author Contributions Conceptualization: LK MK PB. Data curation: LK JC AY VT TGT PB. Formal analysis: LK SC MK DM PB. Funding acquisition: LK MK DM PB. Methodology:LK MK PB. Project administration: LK. Resources: LK JC AY VT TGT PB. Supervision: MK TGT PB. Writing – originaldraft: LK MK PB. Writing – review & editing: LK MK DM JC AY VT TGT PB.

References 1. Poon LC, Savvas M, Zamblera D, Skyfta E, Nicolaides KH. Large loop excision of transformation zone and cervical length in the prediction of spontaneous preterm delivery. BJOG: an international journal of obstetrics and gynaecology. 2012; 119(6):692–8. doi: 10.1111/j.1471-0528.2011.03203.x PMID: 22329452. 2. Kyrgiou M, Koliopoulos G, Martin-Hirsch P, Arbyn M, Prendiville W, Paraskevaidis E. Obstetric out- comes after conservative treatment for intraepithelial or early invasive cervical lesions: systematic review and meta-analysis. Lancet. 2006; 367(9509):489–98. doi: 10.1016/S0140-6736(06)68181-6 PMID: 16473126. 3. Arbyn M, Kyrgiou M, Simoens C, Raifu AO, Koliopoulos G, Martin-Hirsch P, et al. Perinatal mortality and other severe adverse pregnancy outcomes associated with treatment of cervical intraepithelial neoplasia: meta-analysis. Bmj. 2008; 337:a1284. PMID: 18801868; PubMed Central PMCID: PMCPMC2544379. doi: 10.1136/bmj.a1284 4. Kyrgiou M, Mitra A, Arbyn M, Stasinou SM, Martin-Hirsch P, Bennett P, et al. Fertility and early preg- nancy outcomes after treatment for cervical intraepithelial neoplasia: systematic review and meta-anal- ysis. Bmj. 2014; 349:g6192. PMID: 25352501; PubMed Central PMCID: PMCPMC4212006. doi: 10. 1136/bmj.g6192 5. Bruinsma FJ, Quinn MA. The risk of preterm birth following treatment for precancerous changes in the cervix: a systematic review and meta-analysis. BJOG: an international journal of obstetrics and gynae- cology. 2011; 118(9):1031–41. doi: 10.1111/j.1471-0528.2011.02944.x PMID: 21449928. 6. Albrechtsen S, Rasmussen S, Thoresen S, Irgens LM, Iversen OE. Pregnancy outcome in women before and after cervical conisation: population based cohort study. Bmj. 2008; 337:a1343. PMID: 18801869; PubMed Central PMCID: PMCPMC2544429. doi: 10.1136/bmj.a1343

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 12 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

7. Noehr B, Jensen A, Frederiksen K, Tabor A, Kjaer SK. Loop electrosurgical excision of the cervix and subsequent risk for spontaneous preterm delivery: a population-based study of singleton deliveries during a 9-year period. Am J Obstet Gynecol. 2009; 201(1):33.e1-6. doi: 10.1016/j.ajog.2009.02.004 PMID: 19345930. 8. Kyrgiou M, Athanasiou A, Paraskevaidi M, Mitra A, Kalliala I, Martin-Hirsch P, et al. Adverse obstetric outcomes after local treatment for cervical preinvasive and early invasive disease according to cone depth: systematic review and meta-analysis. Bmj. 2016; 354:i3633. doi: 10.1136/bmj.i3633 PMID: 27469988; PubMed Central PMCID: PMCPMC4964801. 9. Kyrgiou M, Arbyn M, Martin-Hirsch P, Paraskevaidis E. Increased risk of preterm birth after treatment for CIN. Bmj. 2012; 345:e5847. doi: 10.1136/bmj.e5847 PMID: 22951549. 10. Grimes-Dennis J, Berghella V. Cervical length and prediction of preterm delivery. Curr Opin Obstet Gynecol. 2007; 19(2):191–5. doi: 10.1097/GCO.0b013e3280895dd3 PMID: 17353688. 11. Berghella V, Roman A, Daskalakis C, Ness A, Baxter JK. Gestational age at cervical length measure- ment and incidence of preterm birth. Obstet Gynecol. 2007; 110(2 Pt 1):311–7. doi: 10.1097/01.AOG. 0000270112.05025.1d PMID: 17666605. 12. Owen J, Hankins G, Iams JD, Berghella V, Sheffield JS, Perez-Delboy A, et al. Multicenter randomized trial of cerclage for preterm birth prevention in high-risk women with shortened midtrimester cervical length. Am J Obstet Gynecol. 2009; 201(4):375.e1-8. doi: 10.1016/j.ajog.2009.08.015 PMID: 19788970; PubMed Central PMCID: PMCPMC2768604. 13. Alfirevic Z, Owen J, Carreras Moratonas E, Sharp AN, Szychowski JM, Goya M. Vaginal progesterone, cerclage or cervical pessary for preventing preterm birth in asymptomatic singleton pregnant women with a history of preterm birth and a sonographic short cervix. Ultrasound in obstetrics & gynecology: the official journal of the International Society of Ultrasound in Obstetrics and Gynecology. 2013; 41 (2):146–51. doi: 10.1002/uog.12300 PMID: 22991337. 14. Berghella V, Pereira L, Gariepy A, Simonazzi G. Prior cone biopsy: prediction of preterm birth by cervi- cal ultrasound. Am J Obstet Gynecol. 2004; 191(4):1393–7. doi: 10.1016/j.ajog.2004.06.087 PMID: 15507971. 15. Kindinger L, Teoh T. Preterm delivery–who is most at risk? An audit of a preterm surveillance clinic. BJOG: An International Journal of Obstetrics & Gynaecology. 2013; 120(Supplement 3):50. doi: 10. 1111/1471-0528.12496 16. Sharp AN, Alfirevic Z. Provision and practice of specialist preterm labour clinics: a UK survey of prac- tice. BJOG: an international journal of obstetrics and gynaecology. 2014; 121(4):417–21. doi: 10.1111/ 1471-0528.12512 PMID: 24314110. 17. Rafaeli-Yehudai T, Kessous R, Aricha-Tamir B, Sheiner E, Erez O, Meirovitz M, et al. The effect of cer- vical cerclage on pregnancy outcomes in women following conization. The journal of maternal-fetal & neonatal medicine: the official journal of the European Association of Perinatal Medicine, the Federa- tion of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2014; 27 (15):1594–7. doi: 10.3109/14767058.2013.871254 PMID: 24289749. 18. Shin MY, Seo ES, Choi SJ, Oh SY, Kim BG, Bae DS, et al. The role of prophylactic cerclage in prevent- ing preterm delivery after electrosurgical conization. J Gynecol Oncol. 2010; 21(4):230–6. doi: 10. 3802/jgo.2010.21.4.230 PMID: 21278884; PubMed Central PMCID: PMCPMC3026301. 19. Nam KH, Kwon JY, Kim YH, Park YW. Pregnancy outcome after cervical conization: risk factors for preterm delivery and the efficacy of prophylactic cerclage. J Gynecol Oncol. 2010; 21(4):225–9. doi: 10.3802/jgo.2010.21.4.225 PMID: 21278883; PubMed Central PMCID: PMCPMC3026300. 20. Odibo AO, Farrell C, Macones GA, Berghella V. Development of a scoring system for predicting the risk of preterm birth in women receiving cervical cerclage. J Perinatol. 2003; 23(8):664–7. doi: 10. 1038/sj.jp.7211004 PMID: 14647165. 21. Israfil-Bayli F, Toozs-Hobson P, Lees C, Slack M, Daniels J, Vince A, et al. Cervical cerclage and type of suture material: a survey of UK consultants’ practice. The journal of maternal-fetal & neonatal medi- cine: the official journal of the European Association of Perinatal Medicine, the Federation of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2014; 27(15):1584–8. doi: 10.3109/14767058.2013.870551 PMID: 24283438. 22. Berghella V, Szychowski JM, Owen J, Hankins G, Iams JD, Sheffield JS, et al. Suture type and ultra- sound-indicated cerclage efficacy. The journal of maternal-fetal & neonatal medicine: the official jour- nal of the European Association of Perinatal Medicine, the Federation of Asia and Oceania Perinatal Societies, the International Society of Perinatal Obstet. 2012; 25(11):2287–90. doi: 10.3109/ 14767058.2012.688081 PMID: 22545593. 23. Henry-Stanley MJ, Hess DJ, Barnes AM, Dunny GM, Wells CL. Bacterial contamination of surgical suture resembles a biofilm. Surg Infect (Larchmt). 2010; 11(5):433–9. doi: 10.1089/sur.2010.006 PMID: 20673144; PubMed Central PMCID: PMC2967823.

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 13 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

24. Fowler JR, Perkins TA, Buttaro BA, Truant AL. Bacteria adhere less to barbed monofilament than braided sutures in a contaminated wound model. Clin Orthop Relat Res. 2013; 471(2):665–71. doi: 10. 1007/s11999-012-2593-z PMID: 23001503; PubMed Central PMCID: PMCPMC3549181. 25. Kindinger LM, MacIntyre DA, Lee YS, Marchesi JR, Smith A, McDonald JA, et al. Relationship between vaginal microbial dysbiosis, inflammation, and pregnancy outcomes in cervical cerclage. Sci Transl Med. 2016; 8(350):350ra102. doi: 10.1126/scitranslmed.aag1026 PMID: 27488896. 26. Mehta P, Patel P, Olver JM. Functional results and complications of Mersilene mesh use for frontalis suspension ptosis surgery. Br J Ophthalmol. 2004; 88(3):361–4. PMID: 14977770; PubMed Central PMCID: PMCPMC1772030. doi: 10.1136/bjo.2002.009951 27. Slack M, Sandhu JS, Staskin DR, Grant RC. In vivo comparison of suburethral sling materials. Int Uro- gynecol J Pelvic Floor Dysfunct. 2006; 17(2):106–10. doi: 10.1007/s00192-005-1320-7 PMID: 15995792. 28. Van Winkle W, Hastings JC, Barker E, Hines D, Nichols W. Effect of suture materials on healing skin wounds. Surg Gynecol Obstet. 1975; 140(1):7–12. PMID: 1108243. 29. Quinn M. Final report of the MRC/RCOG randomised controlled trial of cervical cerclage. Br J Obstet Gynaecol. 1993; 100(12):1154–5. PMID: 8297859. 30. Noehr B, Jensen A, Frederiksen K, Tabor A, Kjaer SK. Depth of cervical cone removed by loop electro- surgical excision procedure and subsequent risk of spontaneous preterm delivery. Obstet Gynecol. 2009; 114(6):1232–8. doi: 10.1097/AOG.0b013e3181bf1ef2 PMID: 19935024. 31. Breiman L. Classification and regression trees. Belmont, Calif.: Wadsworth International Group; 1984. 32. Oh HY, Kim BS, Seo SS, Kong JS, Lee JK, Park SY, et al. The association of uterine cervical micro- biota with an increased risk for cervical intraepithelial neoplasia in Korea. Clin Microbiol Infect. 2015. doi: 10.1016/j.cmi.2015.02.026 PMID: 25752224. 33. Chandiramani M, Seed PT, Orsi NM, Ekbote UV, Bennett PR, Shennan AH, et al. Limited relationship between cervico-vaginal fluid cytokine profiles and cervical shortening in women at high risk of sponta- neous preterm birth. PLoS One. 2012; 7(12):e52412. doi: 10.1371/journal.pone.0052412 PMID: 23300664; PubMed Central PMCID: PMCPMC3530581. 34. ONS. Gestation-specific infant mortality in England and Wales, 2008. In: Statistics OfN, editor. http:// www.ons.gov.uk/ons/child-health/gestation-specific-infant-mortality-in-englang-and-wales/20082010. 35. Behrendt N, Gibbs RS, Lynch A, Hart J, West NA, Iams JD. Rate of change in cervical length in women with vaginal bleeding during pregnancy. Obstet Gynecol. 2013; 121(2 Pt 1):260–4. http://10.1097/ AOG.0b013e31827d8e1b. PMID: 23344274. doi: http://10.1097/AOG.0b013e31827d8e1b 36. Fox NS, Rebarber A, Klauser CK, Peress D, Gutierrez CV, Saltzman DH. Prediction of spontaneous preterm birth in asymptomatic twin pregnancies using the change in cervical length over time. Am J Obstet Gynecol. 2010; 202(2):155.e1-4. Epub 2009/10/23. doi: 10.1016/j.ajog.2009.09.004 PMID: 19846054. 37. Moroz LA, Simhan HN. Rate of sonographic cervical shortening and the risk of spontaneous preterm birth. Am J Obstet Gynecol. 2012; 206(3):234.e1-5. doi: 10.1016/j.ajog.2011.11.017 PMID: 22189048. 38. Souka AP, Papastefanou I, Michalitsi V, Salambasis K, Chrelias C, Salamalekis G, et al. Cervical length changes from the first to second trimester of pregnancy, and prediction of preterm birth by first- trimester sonographic cervical measurement. Journal of ultrasound in medicine: official journal of the American Institute of Ultrasound in Medicine. 2011; 30(7):997–1002. PMID: 21705733. 39. Dilek TU, Yazici G, Gurbuz A, Tasdelen B, Gulhan S, Dilek B, et al. Progressive cervical length changes versus single cervical length measurement by transvaginal ultrasound for prediction of pre- term delivery. Gynecologic and obstetric investigation. 2007; 64(4):175–9. doi: 10.1159/000106486 PMID: 17664877. 40. Naim A, Haberman S Fau—Burgess T, Burgess T Fau—Navizedeh N, Navizedeh N Fau—Minkoff H, Minkoff H. Changes in cervical length and the risk of preterm labor. (0002–9378 (Print)). 41. Pils S, Eppel W, Seemann R, Natter C, Ott J. Sequential cervical length screening in pregnancies after loop excision of the transformation zone conisation: a retrospective analysis. BJOG: an international journal of obstetrics and gynaecology. 2014; 121(4):457–62. doi: 10.1111/1471-0528.12390 PMID: 24148580. 42. Arabin B, Alfirevic Z. Cervical pessaries for prevention of spontaneous preterm birth: past, present and future. Ultrasound in obstetrics & gynecology: the official journal of the International Society of Ultra- sound in Obstetrics and Gynecology. 2013; 42(4):390–9. doi: 10.1002/uog.12540 PMID: 23775862; PubMed Central PMCID: PMCPMC4282542. 43. Romero R, Nicolaides K, Conde-Agudelo A, Tabor A, O’Brien JM, Cetingoz E, et al. Vaginal progester- one in women with an asymptomatic sonographic short cervix in the midtrimester decreases preterm

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 14 / 15 Preterm Birth Prevention Post-Conization: A Model of CL Screening and Cerclage Intervention

delivery and neonatal morbidity: a systematic review and metaanalysis of individual patient data. Am J Obstet Gynecol. 2012; 206(2):124.e1-19. doi: 10.1016/j.ajog.2011.12.003 PMID: 22284156; PubMed Central PMCID: PMCPMC3437773. 44. Odibo AO, Berghella V, To MS, Rust OA, Althuisius SM, Nicolaides KH. Shirodkar versus McDonald cerclage for the prevention of preterm birth in women with short cervical length. American journal of perinatology. 2007; 24(1):55–60. doi: 10.1055/s-2006-958165 PMID: 17195146.

PLOS ONE | DOI:10.1371/journal.pone.0163793 November 3, 2016 15 / 15 ACCEPTED MANUSCRIPT Science Translational Medicine (2016) 8 (350), 350ra102

Title: Relationship between vaginal microbial dysbiosis, inflammation, and pregnancy outcomes in cervical cerclage.

One Sentence Summary: Cervical cerclage using braided suture material disrupts vaginal microbial stability and increases inflammation.

Authors: Lindsay M. Kindinger1,2,3, David A. MacIntyre1*, Yun S. Lee1, Julian R. Marchesi4,5, Ann Smith5, Julie A. K. McDonald4, Vasso Terzidou1,6, Joanna R. Cook1, Christoph Lees1,2,7, Fidan Israfil-Bayli8, Yazmin Faiza9, Philip Toozs-Hobson8, Mark Slack9, Stefano Cacciatore1, Elaine Holmes4,10, Jeremy K. Nicholson4,10, TG Teoh3 and Phillip R. Bennett1,2*. Affiliations: 1 Imperial College Parturition Research Group, Division of the Institute of Reproductive and Developmental Biology, Imperial College London, W12 0NN, UK 2 Queen Charlotte’s Hospital, Imperial College Healthcare NHS Trust, London, W12 0HS, UK 3 St Mary’s Hospital, Imperial College Healthcare NHS Trust, London, W2 1NY, UK 4 Centre for Digestive and Gut Health, Imperial College London, W2 1NY, UK 5 School of Biosciences, Cardiff University, CF103AX, UK 6 Chelsea & Westminster Hospital, Imperial College Healthcare NHS Trust, London, SW10 9NH, UK 7 Department of Development and Regeneration, KU Leuven, 3000, Belgium 8 Birmingham Women’s Hospital NHS Foundation Trust, Edgbaston, Birmingham, B15 2TG, UK 9 Urogynaecology Department, Addenbrooke’s Hospital, Cambridge University Hospitals NHS Foundation Trust, Cambridge, CB2 0QQ, UK 10 Division of Computational Systems Medicine, Department of Surgery and Cancer, Imperial College London, SW7 2AZ, UK

*Corresponding authors: Dr David MacIntyre ([email protected]) and Prof Phillip Bennett ([email protected]) Imperial College Parturition Research Group, Division of the Institute of Reproduction and Developmental Biology, Imperial College London, Hammersmith Campus, London, W12 0NN, UK.

Abstract: Preterm birth, the leading cause of death in children under five, may be caused by inflammation triggered by ascending vaginal infection. About two million cervical cerclages are performed annually to prevent preterm birth. The procedure is thought to provide structural support and maintain the endocervical mucus plug as a barrier to ascending infection. Two types of suture material are used for cerclage: monofilament or multifilament braided. Braided sutures are most frequently used, though no evidence exists to favor them over monofilament sutures. In this study we assessed birth outcomes in a retrospective cohort of 678 women receiving cervical cerclage in 5 UK university hospitals and showed that braided cerclage was associated with increased intrauterine death (15% v 5%, P = 0.0001) and preterm birth (28% v 17%, P = 0.0006) compared to monofilament suture. To understand the potential underlying mechanism, we performed a prospective, longitudinal study of the vaginal microbiome in women at risk of preterm birth because of short cervical length (≤25 mm) who received braided (n=25) or monofilament (n=24) cerclage under otherwise comparable circumstances. Braided suture induced a persistent shift towards vaginal microbiome dysbiosis characterized by reduced Lactobacillus spp. and enrichment of pathobionts. Vaginal dysbiosis was associated with inflammatory cytokine and interstitial collagenase excretion into cervicovaginal fluid and premature cervical remodeling. Monofilament suture had comparatively minimal impact upon the vaginal microbiome and its interactions with the host. These data provide in vivo evidence that a dynamic shift of the human vaginal microbiome toward dysbiosis correlates with preterm birth.

Main Text: Introduction Each year, preterm birth (PTB, defined as delivery before 37 weeks of gestation) causes over one million deaths worldwide (1). Although preterm birth has multiple etiologies, infection is thought to be a causal mechanism in up to 50% of cases (2). It has been postulated that microbiota may spread hematogenously (3), or ascend from the vagina along mucosal surfaces. During a healthy pregnancy, the uterus and fetus are protected from ascending infection from the vagina by the cervix, which acts as a functional and physical barrier to bacteria and pathogens (4). In the vagina, Lactobacillus spp. stability and dominance are central to reproductive health. Pregnancy induces a shift in the vaginal microbiome from a temporally dynamic community structure in non-pregnant women (5) towards stable, Lactobacillus spp. dominance that inhibits growth of pathobionts (6, 7). The vaginal microbiome at delivery also acts as an important source of pioneering microbiota for the neonatal gut microbiome, thus implicating it in long-term health outcomes (8, 9). An association between vaginal dysbiosis during pregnancy, characterized by reduced quantities of Lactobacillus spp., and preterm birth has long been recognized (2, 10); women diagnosed with bacterial vaginosis have a 2- and 6-fold increased risk of preterm birth and late miscarriage, respectively (10). Recent analyses of the vaginal microbiome in pregnancy using culture-independent methods lend further support to associate vaginal dysbiosis and preterm birth (11). Cervical cerclage (12, 13) and progesterone supplementation (14) are the only widely used clinical strategies for the prevention of preterm birth, with an estimated two million cerclage procedures performed annually (15). Cervical cerclage reduces the risk of preterm birth by approximately 20% in women with a history of spontaneous preterm birth and/or a short cervical length (16, 17), and its use in these circumstances is recommended by both the American and the UK Royal College of Obstetricians and Gynecologists (12, 13). Its mechanisms of action are uncertain, but it is thought to provide mechanical support to a weakened cervix (18) as well as to support the cervical mucosal plug as a barrier to ascending infection (4). However, cervical cerclage is associated with increased risk of infection (17, 19). The procedure involves placing a purse-string stitch around the cervix with or without a dissection of the bladder away from the cervix although there is no evidence for a benefit of bladder dissection (20). Two different suture materials, braided or monofilament, are used for the procedure, with braided preferred by 80% of surgeons without an evidence base (21, 22). Braided suture is composed of non-absorbable polyester ethylene terephthalate fibers braided together to form a 5 mm wide mesh tape. The tape is characteristically high in tensile strength and is thought to provide a secure structural support to a weakened cervix because of its high coefficient of friction (18). Monofilament sutures are made of a single strand of non- absorbable polyamide polymer and because of their simple structure, they provide less mechanical resistance when passed through tissue. As a result, they have a tendency to slip and therefore require a greater number of throws to secure the knot than a braided suture (18), which is why braided suture is usually preferred.

However, an association between braided suture use and increased risk of infection in other disciplines (23, 24) has informed the hypothesis that pregnancy outcome after cervical cerclage may be influenced by suture material (15). Aiming to assess the impact of cerclage suture on vaginal microbiota, we hypothesized that the braided suture material promotes pathobiont colonization of the vagina, resulting in activation of inflammatory parturition pathways and premature cervical ripening. In this study we have undertaken a retrospective analysis of pregnancy outcomes in women receiving a clinically indicated cervical cerclage over a ten-year period across five university hospitals in England (UK). After this, a prospective cohort of women at risk of preterm birth were randomized to either braided or monofilament suture material. Longitudinal profiling of their vaginal microbiome in the context of cerclage insertion was undertaken using 16S rRNA gene sequencing. Cytokine expression profiling of matched cervico-vaginal fluid samples and cervical vascular assessment by 4-D ultrasound were concurrently performed as measures of physiological responses to cerclage insertion and changing microbial composition.

Results

Retrospective assessment of suture materials’ effects on pregnancy outcomes A total of 671 women receiving cervical cerclage during pregnancy were identified from five UK university hospitals within a 10-year period. Of these, 327 (49%) received a braided suture material and 344 (51%) received monofilament suture for their cervical cerclage. In women receiving a braided cerclage, higher rates of non-viable births (delivery <24 weeks or intrauterine death) were observed compared to those receiving the monofilament alternative (15% vs 5% respectively; P < 0.0001, Fig. 1A). Increased rates of preterm birth (24-37 weeks gestation) were also observed in women receiving braided cerclage (28% braided vs 17% monofilament; P < 0.0001, Fisher’s exact test, Fig. 1A). Comparison of available demographics demonstrated that consistent with known clinical practice (21, 22), preference of suture material varied across hospital sites (table S1). Linear mixed effects modeling excluding hospital location, which is linked to the choice of suture material, demonstrated that although history of a previous preterm birth was a significant contributor to non-viable births, braided suture was the primary driver of the observed outcome independent of potential confounders including maternal age, ethnicity, and parity (fig. S1), but not hospital location. The effect of previous preterm birth was lost when hospital location was included in the model (table S2). In both analyses, suture material was the major variable influencing the risk of preterm birth independent of maternal age, ethnicity, parity, and history of a previous preterm birth.

Data on gestational age at cerclage insertion and corresponding cervical length were available for women receiving an ultrasound-indicated cerclage (for CL ≤25 mm). Distribution of cervical length was comparable among monofilament and braided groups (P = 0.2; Mann- Whitney, table S1), and as would be expected, a shorter cervix significantly contributed to the risk of adverse outcome (P = 0.019). A sub-analysis using linear mixed effects regression models demonstrated that suture material remained a significant contributing factor for both preterm birth (P = 0.00002) and non-viable pregnancy (P = 0.006) among ultrasound- indicated cerclages after adjusting for potential confounders including gestational age at insertion (table S2).

Baseline characteristics of the prospective study subjects Women who were attending prematurity surveillance clinic after a history of preterm birth and were sonographically identified as having a short cervix (≤25 mm) were prospectively recruited and randomized to receive a cerclage using either braided Mersilene (n=25) or monofilament Ethilon (n=24) suture material. Demographics among suture material groups, including gestation and cervical length at cerclage insertion were comparable (Table 1).

Suture materials’ impact on the vaginal microbiome The data set consisted of 2,792,842 high quality gene sequences, with a mean sequence read count of 13,825 per sample (range 689 to 1,396,421). Using bacterial genera sequence data, samples were classified according to their vaginal bacterial communities as normal (>90% Lactobacillus spp.), intermediate (30-90% Lactobacillus spp.), or dysbiotic (<30% Lactobacillus spp.) (Fig. 1B and table S3). Before cerclage insertion, prevalence of intermediate and dysbiotic microbiomes in monofilament and braided patient groups was similar (16.7% v 17.3%, P = 0.7, Fig. 1C). These were higher than the rates of dysbiosis observed in the background low-risk pregnant population at the same gestational age not requiring intervention who had normal pregnancy outcomes (6%, Fisher’s exact, P = 0.03) (6). Insertion of the braided cerclage caused a dramatic shift towards dysbiosis at 4 weeks after the procedure, which persisted until the final follow up time point, at 16 weeks after cerclage insertion (P = 0.008, ANOVA, Fig. 1C, table S4). When assessed at the species level (fig. S3, fig. S4, and tables S4 and S5), braided cerclage insertion was associated with an increasing proportion of women with community state type (CST) IV, characterized by reduced numbers of Lactobacillus spp. and increased diversity of anaerobic bacteria (6, 25). Before the insertion of braided cerclage, 13% were classified as CST IV, increasing to 45% at 4 weeks and 50% at 16 weeks after cerclage (P = 0.02, fig. S4A, table S4). In contrast, microbial disruption was not observed in women receiving monofilament cerclage, who instead demonstrated maintenance of high Lactobacillus spp. abundance (CSTs I, II, III, and V) and stability throughout longitudinal sampling (P = 0.9; fig. S4A and table S4). To identify degrees of dysbiosis that were not identified by CST classification, we also undertook alternate species level classification based upon dominance of vaginal bacterial communities by Lactobacillus species associated with stability and health. L. iners dominance was more frequently associated with transition to dysbiosis, as well as intermediate and severe dysbiosis (26). Although L. iners has previously been observed as an intermediary towards dysbiosis, there was no significant change in L. iners abundance in association with insertion of a monofilament or braided suture material (fig. S4B and table S5).

An in vitro adhesion assay showed that braided suture cultured with the vaginal commensal L. jensenii or the pathobiont E. coli resulted in a 16- and 20-fold greater bacterial load adherence per unit length (cm), respectively, compared to monofilament suture (P = 0.03 and P = 0.0003, respectively, Student’s t-test; fig. S5).

To identify bacteria specifically associated with braided suture insertion, we performed linear discriminant analysis with effect size (LEfSe)(27) on the 16S rRNA sequence data collected before and 4 weeks after the procedure. Although no differences were identified between patient groups before insertion, braided cerclage resulted in enriched numbers of Gram- negative bacteria at 4 weeks (Fig. 2A and B). This correlated with a 5-fold increase in the number of dysbiotic samples collected after braided cerclage insertion compared to monofilament cerclage (P = 0.04, ANOVA; Fig. 2C). Use of braided cerclage was characterized by increased numbers of bacteria associated with bacterial vaginosis, including species of Prevotella (P = 0.02), Finegoldia (P = 0.02), and Dialister (P = 0.04), and reduced Lactobacillus spp. (Fig. 2A, B and fig. S6). Targeted qPCR suggested that insertion of braided suture was associated with an increase in mean copy numbers of both A. vaginae (594,326 before cerclage v 5,081,000 after cerclage; P = 0.07) and G. vaginalis (961,805 before cerclage v 10,170,000 after cerclage; P = 0.05; fig. S7, table S6). In contrast, no change in the amount of G. vaginalis or A. vaginae was detected after monofilament cerclage. Consistent with these observations, indices of bacterial community richness (Fig. 2D) and alpha-diversity (Fig. 2E) were increased in samples collected after braided suture compared to monofilament, with the greatest differences observed at 16 weeks after cerclage (P = 0.02; ANOVA, Bonferroni multiple comparison).

Inflammatory response to cerclage insertion Insertion of braided, but not monofilament cerclage, increased the release of inflammatory cytokines into cervico-vaginal fluid, including IL-1β, IL-6, IL-8, TNFα, and MMP-1 (Fig. 3A-F). No change was detected in anti-inflammatory cytokines IL-4 (Fig. 3G) IL-2, or IL-10 (table S7). We observed a strong association between severe dysbiosis (<30% Lactobacillus spp.) and increased cervico-vaginal fluid concentrations of pro-inflammatory cytokines ICAM-1, IL-1β, IL-6, MMP-1, MCP-1, TNF-α, GM-CSF, and IFN-γ, as well as the anti- inflammatory cytokine IL-10 when compared to women harboring Lactobacillus spp. dominated microbiomes (Fig. 3H).

Impact on cervical vasculature after cerclage

We constructed three-dimensional cervical vascular trees from ultrasound data and assessed the indices of cervical vasculature (Vascularity index, VI) (28) for morphological differences in the cervix according to suture material. The vascularity was not significantly different between the two groups before cerclage insertion (Fig. 4A). Braided cerclage was strongly associated with increased cervical vascularization 4 weeks after the procedure, and this relationship persisted until 16 weeks after insertion (P = 0.0003; ANOVA, Bonferroni multiple comparison, Fig. 4A). A positive correlation of cervical vascularity with both the number of bacterial species (Fig. 4B, R2 = 0.09, P = 0.002) and alpha diversity (Fig. 4C, R2 = 0.14, P = 0.001) was observed in women receiving a braided cerclage. No relationship between monofilament suture, cervical vascularization, and indices of microbial diversity or richness was observed.

Discussion

Braided suture material is commonly used in preference to monofilament for cervical cerclage (21), because it is assumed that braided suture provides a more secure cerclage that is less likely to slip or tear the cervix (18); however, this assumption is not evidence-based. Here we show that use of braided suture material is associated with an increased risk of preterm birth and non-viable pregnancy, although this will need to be confirmed in a prospective randomized study. Braided cervical cerclage induces vaginal dysbiosis, increases excretion of inflammatory cytokines into the cervico-vaginal fluid, and induces premature cervical vascular remodeling. In contrast, monofilament suture has minimal impact upon the vaginal microbiome and inflammatory pathways associated with premature onset of parturition. These findings have clinical relevance for cerclage procedures in pregnancy and wider implications for braided suture use in other surgical procedures, particularly in potentially contaminated sites. Based upon an approximated two million cervical cerclages per annum (15), 80% of which are performed using braided suture (21), we estimate that a global shift to monofilament suture use for cervical cerclage would prevent 170,000 preterm births (number needed to treat NNT 9.4; 95% CI 5.9 to 22.6) and 172,000 fetal losses (NNT, 9.3; 95% CI 6.6 to 16.0) per annum world-wide. Although cervical cerclage is effective in preventing preterm birth in singleton pregnancies with a previous preterm birth (17, 29), our data show that braided cervical cerclage increases vaginal dysbiosis and inflammation, and likely accounts for the doubled risk of puerperal sepsis (19) with no improvement in neonatal outcome (17) after cerclage insertion. Moreover, increasing evidence suggests that the braided cerclage is of no benefit, and may even be detrimental in other groups that are at high risk for preterm birth, such as multiple pregnancies (30) and women with a shortened cervix after excisional treatment for cervical intra-epithelial neoplasia (31-34). The role of suture material in cerclage efficacy has largely been neglected, with studies rarely detailing the suture material used for the procedure (35). To date, only one randomized controlled trial has been conducted to examine the impact of suture material on pregnancy outcomes, but this study was limited to the comparison of two braided materials, and no differences in preterm birth rates were observed (22). Re-evaluation of existing literature on cervical cerclage use in pregnancy for the prevention of preterm birth is required in light of our findings. Our study is limited by the retrospective nature of the comparisons of pregnancy outcome. However, the cohort size is large, and preterm delivery rates for each suture material were similar at each of the five centers. The size of the prospective study was limited by practicalities and cost of intensive multiple investigations; however, the experimental data provide supporting evidence for the mechanism of poorer outcomes in patients treated with braided suture cerclage. Analysis of the vaginal microbiome was undertaken using three alternate approaches. We primarily used a genera-based classification of normal, intermediate, and severe dysbiosis, which demonstrated that the principal effect of braided cerclage is to reduce lactobacillus numbers and induce vaginal dysbiosis. We next classified samples at the species level into 5 previously described CSTs, however, this analysis was limited to the consideration of only one dysbiotic group. Therefore, an alternate species level classification considering two levels of dysbiosis (intermediate and severe) was also undertaken. The inherent capacity of braided suture material to facilitate bacterial growth has been previously described in other surgical arenas (36, 37); however, our in vivo and in vitro data provide evidence for preferential pathobiont colonization over commensal vaginal species that are important for reproductive health outcomes. Vaginal dysbiosis associated with braided cerclage insertion was characterized by reduced Lactobacillus species and increased diversity and enrichment of bacteria associated with poor pregnancy outcomes, including Peptoniphilus harei (38), species of Bacteriodes (38, 39), Prevotella (40), and Clostridium (41). However, it is recognized that sequencing of specific hypervariable regions of the 16S rRNA gene can result in underrepresentation of key vaginal microbiota such as G. vaginalis, which together with A. vaginae is characteristic of bacterial vaginosis (BV) (42, 43), a condition associated with adverse reproductive health outcomes including pelvic inflammatory disease (44), HIV transmission (45), and preterm birth(46). Using targeted qPCR, we showed that dysbiosis associated with braided cerclage insertion may involve increased abundance of G. vaginalis. The virulence of G. vaginalis is thought to relate to its biofilm-producing capacity and adherence to vaginal epithelial cells (42, 47). It is possible that such characteristics may promote biofilm formation on the surface of cerclage suture material, and this should be examined in future studies. Other bacteria clinically associated with adverse pregnancy outcome include S. agalactiae (Group B streptococcus) and E. coli. However, we did not observe any changes in the amounts of S. agalactiae. E. coli was not detected in our data set, but this may reflect a limitation of the primer set used for 16S rRNA sequencing (48, 49). Increased bacterial diversity in the vagina corresponded to the induction of a pro- inflammatory cytokine profile in cervico-vaginal fluid involving known mediators of cervical vascular remodeling, including IL-1β, IL-6, IL-8, and TNFα (50-52). Concentrations of MMP-1, a matrix metalloproteinase central to collagenous remodeling preceding preterm birth (53, 54), were also increased after insertion of braided cerclage. Increased cervical vascularity occurs before term parturition (28, 55), and an association between increased cytokine excretion and cervical angiogenesis, vasodilation, and vascular permeability has been previously described (56). Consistent with a role in untimely cervical remodeling preceding preterm birth, increased cervical vascularity was observed as early as 4 weeks after insertion in women receiving a braided cerclage. Our study therefore provides a human model for understanding how pregnant host-vaginal microbial interactions may underpin poor pregnancy outcomes. Cerclage-induced inflammation resulting in premature weakening of the cervix could also provide a mechanism for high rates of intrauterine death in women receiving a braided cerclage, because in-utero exposure of the fetus to elevated concentrations of circulating pro-inflammatory cytokines is known to associate with fetal brain injury (57, 58) and stillbirth (59-61). In summary, our data provide evidence that cervical cerclage using braided suture associates with increased rates of preterm birth and non-viable pregnancy. Promotion of vaginal bacterial dysbiosis after insertion of braided suture material likely contributes to these adverse pregnancy outcomes through activation of local tissue inflammation and premature cervical remodeling. Because monofilament suture has minimal impact on the host microbiome or inflammation in pregnancy and associates with improved pregnancy outcome, we advocate its use for cervical cerclage. Further clinical trials addressing the impact of cerclage suture material, powered to assess outcomes of preterm birth, neonatal morbidity, and mortality, are therefore urgently required.

Materials and Methods

The study was approved by NHS National Research Ethics Service (NRES) Committee London - City & East (REC 12/LO/2003), and all participants provided written informed consent. Study design We initially performed a retrospective data collection to assess outcomes of cervical cerclages in singleton pregnancies considered at risk of preterm birth over a ten-year period between January 2003 and 2013 across five UK hospitals in London, Cambridge, and Birmingham. Cases were identified from operating theatre logs, and all case notes were reviewed where possible. Details regarding cerclage suture insertion, suture material used, outcomes of preterm birth (between 16+0 and 36+6 weeks’ gestation), and non-viable birth (still birth or miscarriage >16+0 weeks’ gestation) were collected. Other metadata collected included maternal age, parity, previous spontaneous preterm birth / midtrimester miscarriage, indication for cerclage (ultrasound indicated or elective), and cervical length at cerclage insertion. After this, we prospectively recruited pregnant women at risk of preterm birth with sonographic indications for cervical cerclage at preterm surveillance clinics from January 2013 until August 2014 at a single London site (Queen Charlotte’s and Chelsea Hospital). Inclusion criteria were pregnant women with history of spontaneous preterm birth (<37+0 weeks) and a cervical length (CL) measurement below the 10th centile (≤25 mm) on transvaginal scan at ≤ 23+6 weeks’ gestation in the index pregnancy. A normally distributed cervical length range not associated with preterm birth risk is typically 35 mm ± 8.3 mm (mean ± SD) (62). Exclusion criteria included multiple pregnancy, previous iatrogenic preterm births, HIV positive status, and sexual intercourse or vaginal bleeding in the preceding 48 hours. Eligible women were randomized to either braided Mersilene (n=25) or monofilament Ethilon (n=24) cerclage suture material. The same obstetrician performed the procedure using the MacDonald technique (63). Participants were recruited before cerclage insertion and followed up longitudinally at 4, 8, 12, and 16 weeks after insertion. At each time point, cervico-vaginal fluid was sampled from the posterior fornix under direct visualization using 2 swabs for later 16s rRNA gene sequencing and cytokine analysis: a BBL CultureSwab MaxV Liquid Amies swab (Becton, Dickinson and Company) and a Transwab MW170 with rayon bud type (Medical Wire & Equipment), respectively. Both swabs were immediately placed on ice and snap frozen at -80°C. A transvaginal ultrasound scan was then immediately performed to assess cervical vascularization in the dorsal lithotomy position with an empty bladder, taking care to avoid undue pressure on the cervix.

DNA extraction and 16S rRNA sequencing DNA extraction from the BBL CultureSwab was performed as previously described (6). Integrity of the extracted bacterial DNA was confirmed by PCR amplification using universal forward and reverse primers (6). The V1-V3 hypervariable regions of 16S rRNA genes were amplified for sequencing using a forward and reverse fusion primer. The forward primer was constructed with (5′-3′) the Illumina i5 adapter (AATGATACGGCGACCACCGAGATCTACAC), an 8 bp barcode, a primer pad (forward: TATGGTAATT), and the 28F-GAGTTTGATCNTGGCTCAG primer (64). The reverse fusion primer consisted of (5′-3′) the Illumina i7 adapter (CAAGCAGAAGACGGCATACGAGAT), an 8 bp barcode, a primer pad (reverse: AGTCAGTCAG), and the reverse primer (519R-GTNTTACNGCGGCKGCTG). Sequencing was performed on an Illumina MiSeq platform (Illumina, Inc.) at Research and Testing Laboratory (Lubbock, TX, USA). Resulting sequence reads were analyzed using the MiSeq SOP Pipeline of the Mothur package (65), which is designed to analyze a multiplexed set of samples. Sequence alignment was performed using the Silva bacterial database (www.arb- silva.de/), and classification of sequences was undertaken using the RDP database reference sequence files and the Wang method (66). OTU taxonomies (phylum to genus) were determined using the RDP MultiClassifier script. Species level taxonomies were determined using USEARCH with 16S rRNA gene sequences from the cultured representatives from the RDP database (67). Sequence alignment data describing the capacity of the V1-V3 amplicons to discriminate Lactobacillus spp. are provided in Table S8 and Fig. S8. Data were re- sampled and normalized to the lowest read count in Mothur (n=689). Quantitative PCR Targeted quantitative PCR was carried out to detecct 16S rRNA genes from Atopobium vaginae and Gardnerella vaginalis. The assays were SYBR green based and performed on Applied Biosystem’s StepOnePlus. PCR reaction mixes are as follows: 1x SYBR Green Jumpstart Taq Ready Mix (Sigma-Aldrich), 5 µl of bacterial DNA isolated from the vaginal swabs, and 0.8 µM final concentration of forward and reverse primers. Oligonucleotide primers used for A. vaginae were: forward- 5’-TAGGCGGTTTGTTAGGTCAGGA-3’; reverse- 5’-CCTACCAGACTCAAGCCTGC-3’ (68) and for G. vaginalis; forward- 5’- GGAAACGGGTGGTAATGCTGG-3’; reverse- 5’-CGAAGCCTAGGTGGGCCATT-3’) (69). Thermocycle profile was 95ºC for 2 minutes, followed by 40 cycles at 95ºC for 15 sec and 65 ºC for 1 min. For both assays, vaginal samples and corresponding standards (A. vaginae and G. vaginalis DNA) were run in duplicates, and mean numbers were used to calculate 16S rRNA gene copies per 5 µl of vaginal DNA. In vitro adhesion assay Propensity of bacteria to adhere to braided and monofilament suture material was assessed using an in vitro adhesion assay (70). Briefly, 1 cm segments of sterile braided Mersilene or monofilament Ethilon were prepared with a sterile blade and tweezers. Suture fragments were placed into a sterile screw-capped tube and incubated with 1 ml of 100% ethanol for 1 hour at room temperature. Fragments were washed 3x using 1 mL of sterile water before inoculation with E. coli (Nissle 1917) using LB broth and LB plates or Lactobacillus jensenii (Cultech Ltd.) using MRS broth and MRS plates. For each bacterial isolate we tested the following groups: inoculated Mersilene thread fragments (n=3), uninoculated Mersilene thread fragment (sterility control, n=1), inoculated Ethilon thread fragments (n=3), uninoculated Ethilon thread fragment (sterility control, n=1). An overnight bacterial culture was centrifuged at 3,000 x g for 10 min, and the supernatant was discarded. Cell pellets were resuspended in fresh broth to a final concentration of 107 CFU/mL and 1 mL of this cell suspension was added to each of the suture fragments. For sterility controls, 1 mL of sterile broth was added to an uninoculated Mersilene and Ethilon thread fragment. Suture thread fragments were incubated at 37ºC for 24 h before being washed 3x with sterile PBS and transferred into tubes containing 1 ml of sterile PBS. Each tube was vortexed 3 x for 30 seconds to detach bacterial cells. The cell suspension was vigorously passed through a 25G needle 10 times to break up cell clumps. Aliquots of 100 µL were collected from the cell suspension and used for colony counts on LB or MRS agar plates. After incubation, we used plate colony counts to calculate the CFU/mL of cell suspension. Thread fragment length was accurately measured after cell suspension plating to calculate CFU/cm.

Cytokine analysis

The transwab cervico-vaginal fluid samples were thawed on ice and resuspended in 350 µL of phosphate-buffered saline solution with protease inhibitor (5 µl/ml; Sigma Aldrich). The suspension was centrifuged at 3000 x g for 2 min and the supernatant collected into a new microcentrifuge tube before repeating the centrifugation step to ensure removal of any cellular debris. Cell-free supernatants were analyzed by Human Magnetic Luminex Screen Assay (15-plex) (Luminex Corporation) with a Bioplex 200 system (Biorad Laboratories Ltd.). Analyte-specific Luminex Screening Assays were performed for 15 analytes: interleukin (IL)-1β, IL-2, IL-4, IL-6, IL-8, IL-10, granulocyte colony-stimulating factor (G- CSF), granulocyte-macrophage colony-stimulating factor (GM-CSF), monocyte chemotactic protein (MCP)-1, tumor necrosis factor (TNF)-α, interferon (IFN)-γ, regulated on activation normal T expressed and secreted/chemokine ligand 5 (RANTES/CCL5), vascular endothelial growth factor (VEGF), intercellular adhesion molecule 1 (ICAM-1), and matrix metalloproteinase 1 (MMP-1). Analytes were selected according to evidence of involvement in inflammatory change related to preterm birth, cervical ripening, and angiogenesis. Samples were analyzed on 96-well plates at two dilutions (1:1 and 1:50) optimized for detection of analytes within the range of the standards as specified by Luminex Human premixed analyte kit.

Cervical vascularization assessment

Voluson E ultrasound (GE Healthcare) in 3D/4D mode with medium persistence, high sensitivity, and normal line density was used for transvaginal cervical vascular assessment. A sagittal plane of volume acquisition, set at 90°, was analyzed using Virtual Organ Computer- aided AnaLysis software program (VOCAL, GE Medical Systems) (28). The “histogram facility” of the software was used to calculate the vascularization index (VI) within the defined volume.

Statistical analysis Assessment of differences in outcomes of viability and preterm birth between cerclage suture material groups (braided versus monofilament) was performed using the Fisher exact test for categorical variables and Mann-Whitney for continuous variables. We used a linear mixed- effects model incorporating suture material group, maternal age, parity, previous preterm birth, and hospital location as fixed effects and ethnicity (Asian, Black, or Caucasian) as a random effect to compare braided versus monofilament suture material for the two primary outcomes (viability and preterm birth). The contributions of fixed-effects terms (p-value and F statistics) were calculated using the analysis of variance (ANOVA) with Satterthwaite approximation for degrees of freedom. Examination of statistical differences between vaginal microbiota was performed at bacterial genera and species levels using the Statistical Analysis of Metagenomic Profiles (STAMP) software package (71). Ward linkage hierarchical clustering analysis (HCA) of bacterial genera was performed using a clustering density threshold of 0.75. Samples were classified according to the percentage of Lactobacillus spp. reads as a proportion of the total number of reads per sample into the following groups: normal (>90% Lactobacillus spp.), intermediate (30-90% Lactobacillus spp.), or dysbiotic (<30% Lactobacillus spp.). Bacterial species data were classified into community state types (CSTs) as described by Ravel et al (25): CST I (L. crispatus), CST II (L. gasseri), CST III (L. iners), CST IV (mixed bacterial species), and CST V (L. jensenii). To identify potential associations between suture material and differing degrees of dysbiosis, an alternative classification of the species data was performed as described by Borgdorff et al. (26), grouping bacteria into communities characterized by healthy Lactobacillus spp. dominance, L. iners, or moderate or severe dysbiosis. The effects of suture material and time from cerclage insertion on bacterial genera, number of species observed and alpha diversity were assessed using One-way ANOVA, Kruskal-Wallis, and Dunn’s multiple comparisons where appropriate. Linear discriminant analysis (LDA) effect size (LEfSe) method (27) characterized differentially abundant taxonomic features of the two suture materials before and 4 weeks after cerclage insertion. An alpha value of 0.01 was used for factorial Kruskal-Wallis test between classes, and a threshold of 3.0 was used for logarithmic LDA score for discriminative features.

The Wilcoxon signed rank test compared cytokine analyte concentrations before and 4 weeks after cerclage insertion. The Mann-Whitney test was used to test for differences among suture material types. Analyte expression was classified according to the corresponding microenvironments (Fig. 1C), and the Mann-Whitney test compared cytokine expression in the presence of a normal or dysbiotic microbiome. Cervical vascularization was compared according to suture material from the time of cerclage insertion and as a function of the corresponding bacterial classification, using Kruskal-Wallis and ANOVA multiple comparison analyses where appropriate. We used linear regression to assess for correlations between cervical vascularity, the number of observed species, and Shannon index of alpha diversity, according to suture material.

Supplementary information

Figure S1. Linear mixed effects modeling of retrospective outcome data.

Figure S2. Gestation at birth as a function of suture material used in the prospectively recruited cohort.

Figure S3. Ward hierarchical clustering analysis of species sequence data.

Figure S4. Longitudinal assessment of vaginal bacterial community structure following suture insertion.

Figure S5. In vitro adherence assay of suture material with E. coli or L. jensenii.

Figure S6. Longitudinal comparison of bacterial genera increased following braided suture insertion.

Figure S7. Quantitative PCR assessment of A. vaginae and G. vaginalis at 4 weeks after cerclage

Figure S8. V1-V3 hypervariable region sequence alignment against major vaginal Lactobacillus species

Table S1. Patient demographics for retrospective study cohort.

Table S2. Contributing confounder analysis for non-viable pregnancy and preterm birth <37 weeks.

Table S3. Bacterial genus classification according to time from cerclage

Table S4. Bacterial species classification into community state types according to time from cerclage.

Table S5. Species classification into normal, L. iners dominant, intermediate and severe dysbiosis according to time from cerclage.

Table S6. Mean bacterial counts of Atopobium vaginae and Gardnerella vaginalis before and 4 weeks after cerclage insertion, as assessed by quantitative PCR.

Table S7. Mean fold change in analyte expression detectable in cervico-vaginal fluid before and after cerclage insertion.

Table S8. DNA identity for the V1-V3 region used in the analysis for the 4 main lactobacilli in CST I, II, III, and V. References and notes

1. L. Liu, S. Oza, D. Hogan, J. Perin, I. Rudan, J. E. Lawn, S. Cousens, C. Mathers, R. E. Black, Global, regional, and national causes of child mortality in 2000-13, with projections to inform post-2015 priorities: an updated systematic analysis. Lancet 385, 430-440 (2015); published online EpubJan (10.1016/S0140-6736(14)61698-6). 2. R. L. Goldenberg, J. F. Culhane, J. D. Iams, R. Romero, Epidemiology and causes of preterm birth. The lancet 371, 75-84 (2008). 3. K. Aagaard, J. Ma, K. M. Antony, R. Ganu, J. Petrosino, J. Versalovic, The placenta harbors a unique microbiome. Sci Transl Med 6, 237ra265 (2014); published online EpubMay 21 (10.1126/scitranslmed.3008599). 4. M. Hein, E. V. Valore, R. B. Helmig, N. Uldbjerg, T. Ganz, Antimicrobial factors in the cervical mucus plug. Am J Obstet Gynecol 187, 137-144 (2002); published online EpubJul ( 5. P. Gajer, R. M. Brotman, G. Bai, J. Sakamoto, U. M. Schutte, X. Zhong, S. S. Koenig, L. Fu, Z. S. Ma, X. Zhou, Z. Abdo, L. J. Forney, J. Ravel, Temporal dynamics of the human vaginal microbiota. Sci Transl Med 4, 132ra152 (2012); published online EpubMay 2 (10.1126/scitranslmed.3003605). 6. D. A. MacIntyre, M. Chandiramani, Y. S. Lee, L. Kindinger, A. Smith, N. Angelopoulos, B. Lehne, S. Arulkumaran, R. Brown, T. G. Teoh, E. Holmes, J. K. Nicoholson, J. R. Marchesi, P. R. Bennett, The vaginal microbiome during pregnancy and the postpartum period in a European population. Sci Rep 5, 8988 (2015)10.1038/srep08988). 7. R. Romero, S. S. Hassan, P. Gajer, A. L. Tarca, D. W. Fadrosh, L. Nikita, M. Galuppi, R. F. Lamont, P. Chaemsaithong, J. Miranda, T. Chaiworapongsa, J. Ravel, The composition and stability of the vaginal microbiota of normal pregnant women is different from that of non-pregnant women. Microbiome 2, 4 (2014)10.1186/2049-2618-2-4). 8. T. Yatsunenko, F. E. Rey, M. J. Manary, I. Trehan, M. G. Dominguez-Bello, M. Contreras, M. Magris, G. Hidalgo, R. N. Baldassano, A. P. Anokhin, A. C. Heath, B. Warner, J. Reeder, J. Kuczynski, J. G. Caporaso, C. A. Lozupone, C. Lauber, J. C. Clemente, D. Knights, R. Knight, J. I. Gordon, Human gut microbiome viewed across age and geography. Nature 486, 222-227 (2012); published online EpubJun 14 (10.1038/nature11053). 9. M. G. Dominguez-Bello, E. K. Costello, M. Contreras, M. Magris, G. Hidalgo, N. Fierer, R. Knight, Delivery mode shapes the acquisition and structure of the initial microbiota across multiple body habitats in newborns. Proc Natl Acad Sci U S A 107, 11971-11975 (2010); published online EpubJun 29 (10.1073/pnas.1002601107). 10. H. Leitich, H. Kiss, Asymptomatic bacterial vaginosis and intermediate flora as risk factors for adverse pregnancy outcome. Best Pract Res Clin Obstet Gynaecol 21, 375-390 (2007); published online EpubJun (10.1016/j.bpobgyn.2006.12.005). 11. D. B. DiGiulio, B. J. Callahan, P. J. McMurdie, E. K. Costello, D. J. Lyell, A. Robaczewska, C. L. Sun, D. S. Goltsman, R. J. Wong, G. Shaw, D. K. Stevenson, S. P. Holmes, D. A. Relman, Temporal and spatial variation of the human microbiota during pregnancy. Proc Natl Acad Sci U S A 112, 11060- 11065 (2015); published online EpubSep 1 (10.1073/pnas.1502875112). 12. A. C. o. O. a. G. ACOG, Practice Bulletin No. 142: Cerclage for the Management of Cervical Insufficiency. Obstetrics & Gynecology 123, 372-379 (2014)10.1097/01.AOG.0000443276.68274.cc). 13. A. Shennan, M. To, RCOG Green-top Guideline No. 60. Royal College of Obstetricians and Gynaecologists, (2011). 14. E. B. Fonseca, E. Celik, M. Parra, M. Singh, K. H. Nicolaides, F. M. F. S. T. S. Group, Progesterone and the risk of preterm birth among women with a short cervix. N Engl J Med 357, 462-469 (2007); published online EpubAug (10.1056/NEJMoa067815). 15. F. Israfil-Bayli, P. Toozs-Hobson, C. Lees, M. Slack, K. M. Ismail, Pregnancy outcome after elective cervical cerclage in relation to type of suture material used. Med Hypotheses 81, 119-121 (2013); published online EpubJul (10.1016/j.mehy.2013.04.003). 16. V. Berghella, T. J. Rafael, J. M. Szychowski, O. A. Rust, J. Owen, Cerclage for short cervix on ultrasonography in women with singleton gestations and previous preterm birth: a meta-analysis. Obstet Gynecol 117, 663-671 (2011); published online EpubMar ( 17. Z. Alfirevic, T. Stampalija, D. Roberts, A. L. Jorgensen, Cervical stitch (cerclage) for preventing preterm birth in singleton pregnancy. Cochrane Database Syst Rev 4, CD008991 (2012)10.1002/14651858.CD008991.pub2). 18. Ethicon.Inc, D. L. Dunn, Ed. (www.ethicon.com, 2007), pp. 1-119. 19. M. Quinn, Final report of the MRC/RCOG randomised controlled trial of cervical cerclage. British journal of obstetrics and gynaecology 100, 1154-1155 (1993); published online EpubDec ( 20. A. O. Odibo, V. Berghella, M. S. To, O. A. Rust, S. M. Althuisius, K. H. Nicolaides, Shirodkar versus McDonald cerclage for the prevention of preterm birth in women with short cervical length. Am J Perinatol 24, 55-60 (2007); published online EpubJan (10.1055/s-2006-958165). 21. F. Israfil-Bayli, P. Toozs-Hobson, C. Lees, M. Slack, J. Daniels, A. Vince, K. M. Ismail, Cervical cerclage and type of suture material: a survey of UK consultants' practice. J Matern Fetal Neonatal Med 27, 1584-1588 (2014); published online EpubOct (10.3109/14767058.2013.870551). 22. V. Berghella, J. M. Szychowski, J. Owen, G. Hankins, J. D. Iams, J. S. Sheffield, A. Perez-Delboy, D. A. Wing, E. R. Guzman, V. U. T. Consortium, Suture type and ultrasound-indicated cerclage efficacy. J Matern Fetal Neonatal Med 25, 2287-2290 (2012); published online EpubNov (10.3109/14767058.2012.688081). 23. P. Mehta, P. Patel, J. M. Olver, Functional results and complications of Mersilene mesh use for frontalis suspension ptosis surgery. Br J Ophthalmol 88, 361-364 (2004); published online EpubMar ( 24. M. Slack, J. S. Sandhu, D. R. Staskin, R. C. Grant, In vivo comparison of suburethral sling materials. Int Urogynecol J Pelvic Floor Dysfunct 17, 106-110 (2006); published online EpubFeb (10.1007/s00192-005-1320-7). 25. J. Ravel, P. Gajer, Z. Abdo, G. M. Schneider, S. S. Koenig, S. L. McCulle, S. Karlebach, R. Gorle, J. Russell, C. O. Tacket, R. M. Brotman, C. C. Davis, K. Ault, L. Peralta, L. J. Forney, Vaginal microbiome of reproductive-age women. Proc Natl Acad Sci U S A 108 Suppl 1, 4680-4687 (2011); published online EpubMar 15 (10.1073/pnas.1002611107). 26. H. Borgdorff, E. Tsivtsivadze, R. Verhelst, M. Marzorati, S. Jurriaans, G. F. Ndayisaba, F. H. Schuren, J. H. van de Wijgert, Lactobacillus-dominated cervicovaginal microbiota associated with reduced HIV/STI prevalence and genital HIV viral load in African women. ISME J 8, 1781-1793 (2014); published online EpubSep (10.1038/ismej.2014.26). 27. N. Segata, J. Izard, L. Waldron, D. Gevers, L. Miropolsky, W. S. Garrett, C. Huttenhower, Metagenomic biomarker discovery and explanation. Genome Biol 12, R60 (2011)10.1186/gb-2011-12- 6-r60). 28. N. C. Yilmaz, A. B. Yigiter, Z. N. Kavak, B. Durukan, H. Gokaslan, Longitudinal examination of cervical volume and vascularization changes during the antepartum and postpartum period using three- dimensional and power Doppler ultrasound. Journal of perinatal medicine 38, 461-465 (2010); published online EpubSep (10.1515/JPM.2010.087). 29. D. Abbott, M. To, A. Shennan, Cervical cerclage: a review of current evidence. The Australian & New Zealand journal of obstetrics & gynaecology 52, 220-223 (2012); published online EpubJun (10.1111/j.1479-828X.2012.01412.x). 30. T. J. Rafael, V. Berghella, Z. Alfirevic, Cervical stitch (cerclage) for preventing preterm birth in multiple pregnancy. Cochrane Database Syst Rev 9, CD009166 (2014)10.1002/14651858.CD009166.pub2). 31. T. Rafaeli-Yehudai, R. Kessous, B. Aricha-Tamir, E. Sheiner, O. Erez, M. Meirovitz, M. Mazor, A. Y. Weintraub, The effect of cervical cerclage on pregnancy outcomes in women following conization. J Matern Fetal Neonatal Med 27, 1594-1597 (2014); published online EpubOct (10.3109/14767058.2013.871254). 32. M. Y. Shin, E. S. Seo, S. J. Choi, S. Y. Oh, B. G. Kim, D. S. Bae, J. H. Kim, C. R. Roh, The role of prophylactic cerclage in preventing preterm delivery after electrosurgical conization. J Gynecol Oncol 21, 230-236 (2010); published online EpubDec (10.3802/jgo.2010.21.4.230). 33. K. H. Nam, J. Y. Kwon, Y. H. Kim, Y. W. Park, Pregnancy outcome after cervical conization: risk factors for preterm delivery and the efficacy of prophylactic cerclage. J Gynecol Oncol 21, 225-229 (2010); published online EpubDec (10.3802/jgo.2010.21.4.225). 34. A. O. Odibo, C. Farrell, G. A. Macones, V. Berghella, Development of a scoring system for predicting the risk of preterm birth in women receiving cervical cerclage. J Perinatol 23, 664-667 (2003); published online EpubDec (10.1038/sj.jp.7211004). 35. J. D. Iams, Clinical practice. Prevention of preterm parturition. N Engl J Med 370, 254-261 (2014); published online EpubJan 16 (10.1056/NEJMcp1103640). 36. M. J. Henry-Stanley, D. J. Hess, A. M. Barnes, G. M. Dunny, C. L. Wells, Bacterial contamination of surgical suture resembles a biofilm. Surgical infections 11, 433-439 (2010); published online EpubOct (10.1089/sur.2010.006). 37. S. Katz, M. Izhar, D. Mirelman, Bacterial adherence to surgical sutures. A possible factor in suture induced infection. Annals of surgery 194, 35-41 (1981); published online EpubJul ( 38. X. Wang, C. S. Buhimschi, S. Temoin, V. Bhandari, Y. W. Han, I. A. Buhimschi, Comparative microbial analysis of paired amniotic fluid and cord blood from pregnancies complicated by preterm birth and early-onset neonatal sepsis. PLoS One 8, e56131 (2013)10.1371/journal.pone.0056131). 39. C. S. Buhimschi, V. Bhandari, B. D. Hamar, M. O. Bahtiyar, G. Zhao, A. K. Sfakianaki, C. M. Pettker, L. Magloire, E. Funai, E. R. Norwitz, M. Paidas, J. A. Copel, C. P. Weiner, C. J. Lockwood, I. A. Buhimschi, Proteomic profiling of the amniotic fluid to detect inflammation, infection, and neonatal sepsis. PLoS Med 4, e18 (2007); published online EpubJan (10.1371/journal.pmed.0040018). 40. S. L. Hillier, M. A. Krohn, E. Cassen, T. R. Easterling, L. K. Rabe, D. A. Eschenbach, The role of bacterial vaginosis and vaginal bacteria in amniotic fluid infection in women in preterm labor with intact fetal membranes. Clin Infect Dis 20 Suppl 2, S276-278 (1995); published online EpubJun ( 41. Y. W. Han, T. Shen, P. Chung, I. A. Buhimschi, C. S. Buhimschi, Uncultivated bacteria as etiologic agents of intra-amniotic inflammation leading to preterm birth. J Clin Microbiol 47, 38-47 (2009); published online EpubJan (10.1128/JCM.01206-08). 42. J. L. Patterson, A. Stull-Lane, P. H. Girerd, K. K. Jefferson, Analysis of adherence, biofilm formation and cytotoxicity suggests a greater virulence potential of Gardnerella vaginalis relative to other bacterial-vaginosis-associated anaerobes. Microbiology 156, 392-399 (2010); published online EpubFeb (10.1099/mic.0.034280-0). 43. J. P. Trama, K. E. Pascal, J. Zimmerman, M. J. Self, E. Mordechai, M. E. Adelson, Rapid detection of Atopobium vaginae and association with organisms implicated in bacterial vaginosis. Mol Cell Probes 22, 96-102 (2008); published online EpubApr (10.1016/j.mcp.2007.08.002). 44. C. L. Haggerty, S. L. Hillier, D. C. Bass, R. B. Ness, P. I. D. Evaluation, i. Clinical Health study, Bacterial vaginosis and anaerobic bacteria are associated with endometritis. Clin Infect Dis 39, 990-995 (2004); published online EpubOct 1 (10.1086/423963). 45. T. E. Taha, D. R. Hoover, G. A. Dallabetta, N. I. Kumwenda, L. A. Mtimavalye, L. P. Yang, G. N. Liomba, R. L. Broadhead, J. D. Chiphangwi, P. G. Miotti, Bacterial vaginosis and disturbances of vaginal flora: association with increased acquisition of HIV. AIDS 12, 1699-1706 (1998); published online EpubSep 10 ( 46. S. L. Hillier, R. P. Nugent, D. A. Eschenbach, M. A. Krohn, R. S. Gibbs, D. H. Martin, M. F. Cotch, R. Edelman, J. G. Pastorek, 2nd, A. V. Rao, et al., Association between bacterial vaginosis and preterm delivery of a low-birth-weight infant. The Vaginal Infections and Prematurity Study Group. N Engl J Med 333, 1737-1742 (1995); published online EpubDec 28 (10.1056/NEJM199512283332604). 47. A. Machado, N. Cerca, Influence of Biofilm Formation by Gardnerella vaginalis and Other Anaerobes on Bacterial Vaginosis. J Infect Dis 212, 1856-1861 (2015); published online EpubDec 15 (10.1093/infdis/jiv338). 48. S. Chakravorty, D. Helb, M. Burday, N. Connell, D. Alland, A detailed analysis of 16S ribosomal RNA gene segments for the diagnosis of pathogenic bacteria. J Microbiol Methods 69, 330-339 (2007); published online EpubMay (10.1016/j.mimet.2007.02.005). 49. S. Srinivasan, D. N. Fredricks, The human vaginal bacterial biota and bacterial vaginosis. Interdisciplinary perspectives on infectious diseases 2008, 750479 (2008)10.1155/2008/750479). 50. G. Rizzo, A. Capponi, A. Vlachopoulou, E. Angelini, C. Grassi, C. Romanini, The diagnostic value of interleukin-8 and fetal fibronectin concentrations in cervical secretions in patients with preterm labor and intact membranes. Journal of perinatal medicine 25, 461-468 (1997). 51. M. Sakai, Y. Sasaki, S. Yoneda, T. Kasahara, T. Arai, M. Okada, H. Hosokawa, K. Kato, Y. Soeda, S. Saito, Elevated interleukin-8 in cervical mucus as an indicator for treatment to prevent premature birth and preterm, pre-labor rupture of membranes: a prospective study. American journal of reproductive immunology 51, 220-225 (2004); published online EpubMar (10.1111/j.1600-0897.2004.00145.x). 52. B. D. Taylor, C. B. Holzman, R. N. Fichorova, Y. Tian, N. M. Jones, W. Fu, P. K. Senagore, Inflammation biomarkers in vaginal fluid and preterm delivery. Hum Reprod 28, 942-952 (2013); published online EpubApr (10.1093/humrep/det019). 53. A. Dubicke, A. Akerud, M. Sennstrom, R. R. Hamad, B. Bystrom, A. Malmstrom, G. Ekman- Ordeberg, Different secretion patterns of matrix metalloproteinases and IL-8 and effect of corticotropin-releasing hormone in preterm and term cervical fibroblasts. Mol Hum Reprod 14, 641- 647 (2008); published online EpubNov (10.1093/molehr/gan060). 54. G. Weiss, L. T. Goldsmith, Mechanisms of relaxin-mediated premature birth. Ann N Y Acad Sci 1041, 345-350 (2005); published online EpubMay (10.1196/annals.1282.055). 55. R. De Diego, J. Sabrià, A. Vela, D. Rodríguez, M. D. Gómez, Role of 3-dimensional power Doppler sonography in differentiating pregnant women with threatened preterm labor from those with an asymptomatic short cervix. J Ultrasound Med 33, 673-679 (2014); published online EpubApr (10.7863/ultra.33.4.673). 56. B. T. Nguyen, V. Minkiewicz, E. McCabe, J. Cecile, C. N. Mowa, Vascular endothelial growth factor induces mRNA expression of pro-inflammatory factors in the uterine cervix of mice. Biomed Res 33, 363-372 (2012); published online EpubDec ( 57. J. Van Steenwinckel, A. L. Schang, S. Sigaut, V. Chhor, V. Degos, H. Hagberg, O. Baud, B. Fleiss, P. Gressens, Brain damage of the preterm infant: new insights into the role of inflammation. Biochem Soc Trans 42, 557-563 (2014); published online EpubApr (10.1042/BST20130284). 58. H. Hagberg, C. Mallard, D. M. Ferriero, S. J. Vannucci, S. W. Levison, Z. S. Vexler, P. Gressens, The role of inflammation in perinatal brain injury. Nat Rev Neurol 11, 192-208 (2015); published online EpubApr (10.1038/nrneurol.2015.13). 59. C. Blackwell, The Role of Infection and Inflammation in Stillbirths: Parallels with SIDS? Front Immunol 6, 248 (2015)10.3389/fimmu.2015.00248). 60. E. Tolockiene, E. Morsing, E. Holst, A. Herbst, A. Ljungh, N. Svenningsen, I. Hagerstrand, L. Nystrom, Intrauterine infection may be a major cause of stillbirth in Sweden. Acta Obstet Gynecol Scand 80, 511-518 (2001); published online EpubJun ( 61. R. L. Goldenberg, C. Thompson, The infectious origins of stillbirth. Am J Obstet Gynecol 189, 861-873 (2003); published online EpubSep ( 62. J. D. Iams, R. L. Goldenberg, P. J. Meis, B. M. Mercer, A. Moawad, A. Das, E. Thom, D. McNellis, R. L. Copper, F. Johnson, J. M. Roberts, The length of the cervix and the risk of spontaneous premature delivery. National Institute of Child Health and Human Development Maternal Fetal Medicine Unit Network. N Engl J Med 334, 567-572 (1996); published online EpubFeb (10.1056/NEJM199602293340904). 63. I. A. McDonald, Incompetant cervix as a cause of recurrent abortion. BJOG: An International Journal of Obstetrics & Gynaecology 70, 105-109 (1963). 64. A. Sundquist, S. Bigdeli, R. Jalili, M. L. Druzin, S. Waller, K. M. Pullen, Y. Y. El-Sayed, M. M. Taslimi, S. Batzoglou, M. Ronaghi, Bacterial flora-typing with targeted, chip-based Pyrosequencing. BMC microbiology 7, 108 (2007)10.1186/1471-2180-7-108). 65. J. J. Kozich, S. L. Westcott, N. T. Baxter, S. K. Highlander, P. D. Schloss, Development of a dual- index sequencing strategy and curation pipeline for analyzing amplicon sequence data on the MiSeq Illumina sequencing platform. Applied and environmental microbiology 79, 5112-5120 (2013); published online EpubSep (10.1128/AEM.01043-13). 66. Q. Wang, G. M. Garrity, J. M. Tiedje, J. R. Cole, Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and environmental microbiology 73, 5261- 5267 (2007); published online EpubAug (10.1128/AEM.00062-07). 67. R. C. Edgar, Search and clustering orders of magnitude faster than BLAST. Bioinformatics 26, 2460- 2461 (2010); published online EpubOct 1 (10.1093/bioinformatics/btq461). 68. D. N. Fredricks, T. L. Fiedler, K. K. Thomas, C. M. Mitchell, J. M. Marrazzo, Changes in vaginal bacterial concentrations with intravaginal metronidazole therapy for bacterial vaginosis as assessed by quantitative PCR. J Clin Microbiol 47, 721-726 (2009); published online EpubMar (10.1128/JCM.01384-08). 69. M. Zozaya-Hinchliffe, R. Lillis, D. H. Martin, M. J. Ferris, Quantitative PCR assessments of bacterial species in women with and without bacterial vaginosis. J Clin Microbiol 48, 1812-1819 (2010); published online EpubMay (10.1128/JCM.00851-09). 70. O. de Castro Costa Neto, L. A. Lobo, N. L. Iorio, M. de Fatima Carvalho Vasconcelos, L. C. Maia, P. N. Tannure, A. G. Antonio, Oral bacteria adherence to suture threads: an in vitro study. Oral Maxillofac Surg 19, 275-280 (2015); published online EpubSep (10.1007/s10006-015-0487-4). 71. D. H. Parks, R. G. Beiko, Identifying biologically relevant differences between metagenomic communities. Bioinformatics 26, 715-721 (2010); published online EpubMar 15 (10.1093/bioinformatics/btq041).

Acknowledgments: We thank all participants of the study and members of Women's Health Research Centre, Imperial College Health NHS Trust. Funding: Supported by the National Institute for Health Research (NIHR) Comprehensive Biomedical Research Centre at Imperial College London (Grant Ref P45272) and by the Genesis Research Trust (Grant Ref P51389). DAM is supported by a Career Development Award from the Medical Research Council (MR/L009226/1). Author contributions L.M.K., D.A.M., C.L., P.T-H., M.S., T.G.T. and P.R.B. conceived and designed the retrospective study. L.M.K., D.A.M., T.G.T and P.R.B. conceived and designed the prospective study. Retrospective data collection and collation was performed by L.M.K., V.T., J.R.C., C.L., F.I-B., Y.F., P.T-H., M.S., T.G.T., and P.R.B. Prospective patient enrollment and sample collection and transvaginal scans were undertaken by L.M.K. and J.R.C. Experiments were performed by L.M.K., Y.S.L. and J.A.K.M. Data analysis was performed by L.M.K., D.A.M., J.R.M., A.S., S.C., E.H., J.K.N and P.R.B. All figures and tables were generated by L.M.K., D.A.M., S.C., and J.R.M. The manuscript was written by L.M.K., D.A.M. and P.R.B, and critically reviewed by all authors. Competing interests: P.R.B. serves as a consultant for ObsEva, a company that works in the field of preterm birth. P.T.-H. works as an advisor to Allergan, SEP, Astellas and Boston Scientific in areas not relevant to the current study. All other authors declare that they have no competing interests. Data and materials availability: Public access to sequence data and accompanying metadata can be obtained at the European Nucleotide Archive’s (ENA) Sequence Read Archive (SRA) (PRJEB11895).

Figure legends: Figure 1. Braided suture material for cervical cerclage is associated with worse outcomes. (A) Retrospective comparison of 10 years of birth outcomes for women receiving a cerclage based on suture material (monofilament, n=344 vs braided, n= 337) revealed higher rates of non-viable births (delivery <24 weeks or intrauterine death) in women receiving a braided cerclage compared to a monofilament alternative (15% vs 5% respectively; P = 0.0001, Fisher’s exact test) and increased rates of preterm birth (24-37 weeks gestation) in women receiving braided cerclage (28% braided vs 17% monofilament; P = 0.0006, Fisher’s exact test) (see table S3 for details). (B) Ward-linkage analysis of vaginal bacterial genera of cervical vaginal fluid samples (n=197) collected longitudinally before and after insertion of a monofilament (n=24) or braided (n=25) cervical cerclage permitted classification of vaginal bacterial communities into three groups: normal (>90% Lactobacillus abundance), intermediate (30-90% Lactobacillus abundance), or dysbiotic (<30% Lactobacillus abundance). (C) Braided cerclage was associated with a 5-fold increase in microbial dysbiosis within 4 weeks of insertion, which persistent until at least 16 weeks, whereas no change was observed in women receiving a monofilament cerclage (P values = Fishers exact test before v after cerclage, and 2 way ANOVA for monofilament vs braided at comparable time points). Figure 2. Bacterial taxonomic groups discriminate between monofilament and braided cerclage. (A) Differentially abundant microbial clades and nodes according to suture material four weeks after insertion were identified using LEfSe analysis and presented as a cladogram. (B) Linear Discriminant Analysis (LDA) was used to estimate the effect size for each differentially abundant species. The vaginal microbiome of patients receiving a monofilament cerclage was enriched with bacilli, whereas those receiving a braided cerclage were comparatively enriched in Bacteroides spp. and Clostridia. (C) Relative abundance bar charts for individual samples highlight maintenance of Lactobacillus genus stability after insertion of monofilament cerclage (P = 0.9; ANOVA), in contrast to the decreased numbers after braided cerclage (P = 0.043; ANOVA) (P values = corrected two-sided Welch’s t-test for monofilament vs. braided). (D) Comparison of total number of bacterial species observed reveals an increase after braided cerclage compared to a monofilament cerclage. (E) Alpha diversity was increased at 16 weeks after braided cerclage insertion compared to monofilament. (*P values = corrected two-sided Welch’s t-test for monofilament vs. braided, #P values = ANOVA Bonferroni multiple comparison to before-cerclage samples).

Figure 3. Braided suture induces cytokine release into the cervico-vaginal fluid. (A) Cytokines were detected using a multiplex assay before and 4 weeks after cerclage insertion. An increase in pro-inflammatory cytokines was detected in the cervico-vaginal fluid after braided cerclage (see table S7 for details). Relative to the concentrations before cerclage, braided suture was associated with an increase in pro-inflammatory cytokines (B) IL-1β, (C) IL-6, (D) IL-8, (E) TNF-α, and (F) MMP-1 but not (G) IL-4 (*P values = Wilcoxon signed rank test for cytokine concentration before and after cerclage). Similar changes were observed when the braided cerclage samples were compared for cytokine concentrations 4 weeks after monofilament cerclage (#P values = Mann-Whitney for fold change monofilament vs. braided). (H) Mean cytokine profiles grouped by the corresponding microbial classification (normal, intermediate, and dysbiotic) revealed that dysbiosis is associated with increased expression of pro-inflammatory cytokines ICAM-1, IL-1β, IL-6, MMP-1, MCP-1, TNF-α, GM-CSF, and IFN-γ and anti-inflammatory IL-10, but not G-CSF, IL-8, VEGF, RANTES, IL-2, or IL-4 (P value = Mann-Whitney for normal vs dysbiotic). Figure 4. Braided cerclage induces premature cervical vascularization. (A) Cervical vascularization index (VI), as assessed by transvaginal ultrasound, was greater in patients receiving braided cerclage compared to monofilament cerclage at 4, 8, 12, and 16 weeks after insertion (*P value = Welches corrected t-test for monofilament vs braided, #P value = ANOVA, Bonferroni multiple comparison for before vs after cerclage). (B) Linear regression analyses demonstrated a positive correlation between VI and the number of species observed 2 2 in braided (R =0.09, P = 0.002) but not monofilament (R =0.001, P = 0.75) cerclages. (C) A 2 similar relationship was observed between VI and alpha diversity index (braided: R = 0.14, P 2 = 0.001 and monofilament: R =0.02, P = 0.14), indicating an interplay between braided suture, increased cervical vascularity, and vaginal microbial dysbiosis. Tables Table 1. Patient characteristics for women randomized to receive monofilament and braided cervical cerclages BMI= body mass index; CL = cervical length (mm); GA = gestational age; w= weeks; SD=standard Monofilament suture Braided suture Total population n (%) 24/49 (49%) 25/49 (51%) 49/49

Age, Mean ±SD (range) years 32.8 ± 3.0 (27-39) 33.9 ± 3.8 (25-42) 33.5 ± 3.5 (25-42)

BMI, Mean ±SD (range) 24.1 ± 4.2 (18-35) 26 ± 3.6 (21-36) 25.1 ± 4.5 (18-36)

Ethnicity, n (%)

Caucasian 16 (67%) 11 (44%) 27 (55%)

Asian 2 (8%) 7 (28%) 9 (18%)

Black 6 (25%) 7 (28%) 13 (27%)

Parity, n (%)

Para 0 12 (50%) 13 (52%) 25 (51%) Smoking, n (%) 1 (4%) 2 (8%) 3 (6%) Cerclage insertion

GA at insertion, mean ±SD w 17+6 ± 2.8 18+1 ± 3 17+0 ± 2.9

CL at insertion, mean ±SD mm 18 ± 5.1 19 ± 4.5 19 ± 5.3

GA at delivery, n (%)

<34+0 w 4/24 (16%) 0/25 (0%) 4 (8%)

34+1-36+6 w 2/24 (8%) 8/25 (32%) 10 (20%)

≥37+0 w 18/24 (75%) 17/25 (68%) 35 (71%) deviation

ACCEPTED MANUSCRIPT: BRITISH JOURNAL OF OBSTETRICS AND GYNAECOLOGY BJOG. 2016 May; Vol.123 (6):877-84. doi: 10.1111/1471-0528.13575.

Title: The effect of gestational age at cervical length measurements in the prediction of spontaneous preterm birth in twin pregnancies: an individual patient level meta-analysis Lindsay M. Kindinger 1,2, Liona C Poon 2,3, Stefano Cacciatore 1, David A MacIntyre 1, Nathan S Fox 4, Ewoud Schuit5, Ben Mol 6, Sophie Liem S7, Arianne C Lim7, Vicente Serra 8, Alfredo Perales9, Frederik Hermans7, Ara Darzi10, Phillip Bennett1, Kypros H Nicolaides3, TG Teoh2.

1. Institute of Reproductive and Developmental Biology, Imperial College London 2. Fetal Medicine Unit, St Mary’s Hospital, Imperial College Healthcare NHS Trust, London 3. Harris Birthright Research Centre for Fetal Medicine, Kings College Hospital, London 4. Maternal Fetal Medicine Associates, PLLC, New York, USA 5. Stanford Prevention Research Center, Stanford University, USA 6. The Robinson Research Institute, School of Paediatrics and Reproductive Health, University of Adelaide, Australia 7. Department of Obstetrics and Gynaecology, Academic Medical Centre, University of Amsterdam, The Netherlands 8. Maternal-Fetal Medicine Unit, Instituto Valenciano de Infertilidad, University of Valencia, Spain 9. Department of Pediatrics, Obstetrics and Gynecology, La FE, University and Polytechnic Hospital, University of Valencia, Spain 10. Department of Academic Surgery, St Marys Hospital, Imperial College Healthcare NHS Trust, London

Correspondence: Dr Lindsay Kindinger [email protected] Institute of Reproductive and Developmental Biology Imperial College London, Hammersmith Hospital Campus Du Cane Road London W12 0NN

Abstract

Objective: To assess the effect of gestational age (GA) and cervical length (CL) measurements at transvaginal ultrasound (TVUS) in the prediction of preterm birth in twin pregnancy.

Design: Individual patient data (IPD) meta-analysis

Setting: International multicentre study

Population: Asymptomatic twin pregnancy

Methods: MEDLINE and EMBASE searches were performed and IPD obtained from authors of relevant studies. Multinomial logistic regression analysis determined probabilities for birth at ≤28+0, 28+1 to 32+0, 32+1 to 36+0, and ≥36+1 weeks as a function of GA at screening and CL measurements.

Main outcome measure: Predicted probabilities for preterm birth at ≤28+0, 28+1 to 32+0, and 32+1 to 36+0.

Results: A total of 6188 CL measurements were performed on 4409 twin pregnancies in twelve studies. Both GA at screening and CL had a significant and non-linear effect on GA at birth. The best prediction of birth ≤28+0 weeks was provided by screening at ≤18+0 weeks (p<0.001), whereas the best prediction of birth between 28+1 and 36+0 weeks was provided by screening at ≥24+0 weeks (p<0.001). 100% NPV for birth ≤28+0 weeks is achieved at CL 65mm and 43mm at ultrasound GA ≤18+0 weeks and 22+1 to 24+0 weeks respectively.

Conclusion: In twin pregnancy, prediction of preterm birth depends on both CL and the GA at screening. When CL <30 mm, screening ≤18+0 weeks is most predictive for birth at ≤28+0 weeks. Later screening at > 22+0 weeks is most predictive of delivery at 28+1 to 36+0 weeks. In twins, we recommend CL screening in twins to commence from ≤18+0 weeks. Running title: Gestation at screening and predicting preterm birth in twins Key words: Twin pregnancy, Preterm birth, Prematurity, CL, Individual patient meta-analysis Introduction

Preterm birth is the leading cause of perinatal death and handicap in survivors (1). The rate of preterm birth in twins is almost 10 times higher than in singletons (2). Extensive studies in singleton pregnancies have established an inverse relation between mid-gestation sonographically measured cervical length (CL) and gestational age (GA) at birth (3, 4), hence CL provides effective screening for spontaneous preterm birth (sPTB). These studies in singletons have also highlighted the relation between CL and sPTB is affected by the GA at screening; a short CL (<15mm) at early GA at screening (<16 weeks) has a higher risk of sPTB compared to the finding of the same CL at later screening (≥24 weeks) (5).

Studies on the prediction of sPTB in twin pregnancy using CL are small in size and are not easily comparable. In these studies, there are large variations in number of patients examined (56-1,135), GA at screening (range 16-28 weeks), cut-offs for defining short CL (range 15-35 mm) and the GA thresholds for defining sPTB (range 28-37 weeks) (6-28). The largest study examined 1,135 pregnancies attending for routine antenatal care at 22-24 weeks and reported that the risk of sPTB increases with decreasing CL but even for women with a long cervix the risk is still substantially higher than in singleton pregnancies (17). The study also showed that monochorionic and dichorionic twins have a similar incidence of early sPTB, once severe twin- to–twin transfusion syndrome has been excluded (17).

Previous meta-analyses have confirmed the association between CL and the rate of sPTB in twins. Their methodology grouped these varied thresholds of CL and GA at birth, as defined in the original studies, (29, 30). As such they were unable to evaluate the effect of GA at screening on the prediction of sPTB.

The objective of this study is to assess the effect of both GA at screening and CL in the prediction of sPTB in twin pregnancy. This meta-analysis of individual patient data (IPD) provides a novel assessment in which both CL and GA at screening are treated as continuous variables

Methods

Literature Search

Searches of MEDLINE, EMBASE, the Cochrane Central Register of Controlled Trials and Research Registers of ongoing trials were performed to identify relevant publications from inception to December 2014. Keywords searched for were "multiple pregnancy”, “preterm birth", “cervical length” and their related terms. Identified studies were assessed for inclusion if they reported on CL measurements in the prediction of gestation at birth in twin pregnancy.

Among the studies, the observed thresholds for CL cut-offs varied, as did GA at screening and GA cut-offs for defining sPTB. To optimise analysis of a large sample size, individual patient data were sought from eligible studies. Corresponding authors were contacted via email and/or telephone and requests were made for original anonymised data for every individual in their study, specifically: (1) the exact GA at CL screening; (2) the CL measurement in millimetres; (3) the exact GA at birth in weeks and days. A request was also made for any subsequent and unpublished data at initial and follow up correspondence. Additional variables maternal age, ethnicity, body mass index, smoking, chorionicity and parity were requested where available.

Eligibility criteria and study selection The inclusion criteria were twin pregnancies with transvaginal measurement of CL at 15+0 to 28+6 weeks. Any studies unable to provide the specified original data for every individual were excluded. Further exclusion criteria were pregnancies with major fetal anomalies, iatrogenic PTB, twin-to-twin transfusion syndrome, intrauterine death, insertion of cervical cerclage or pessary and pre-pregnancy excisional cervical treatment.

Two authors (LK and LP) independently assessed eligible studies for methodological quality. In the case of eligible randomised controlled trials (RCTs) for PTB prevention (progesterone, pessary or cerclage), the study protocols were reviewed for randomisation method and outcome reporting bias. Where the study method or participation randomisation was unclear, authors were contacted for written clarification. To rule out any treatment, effect IPD from RCTs were only included for pregnancies randomised to ‘no-intervention’. All manuscripts were reviewed to ensure a standardised technique of CL attainment, ensuring the ultrasound examination was transvaginal with an empty bladder.

Statistical analysis Clinical heterogeneity was assessed by reviewing differences in patient characteristics across studies. Statistical heterogeneity was assessed using forest plots, the I2 measure and the Cochran’s Q-test (31). GA at measurement of CL were categorised into ≤18+0 weeks, 18+1 to 20+0 weeks, 20+1 to 22+0 weeks, 22+1 to 24+0 weeks and ≥ 24+1 weeks. In the first analysis both CL and GA at birth were considered as continuous variables. In the second analysis, outcomes of GA at birth were grouped into the following categories of clinical significance: very early preterm (≤28+0 weeks), early preterm (28+1 to 32+0 weeks), late preterm (32+1 to 36+0 weeks), and term (≥36+1 weeks). Welch’s T-test assessed statistical differences between groups, and spearman’s rank correlation assessed the relationship between continuously variables. Bonferroni was used for multiple correction and the threshold for significance was corrected to p<0.05. A univariate statistical analysis was performed to assess if the following clinical parameters were associated with GA at birth: maternal age, ethnicity, smoking, BMI, chorionicity, parity and study location.

A multinomial logistic regression model was generated where predicted probabilities of very early preterm, early preterm, late preterm, and term birth were calculated as a function of categorised GA at ultrasound and CL, considered as a continuous variable. Two independent sets (a training and validation set) were built from the entire cohort. The training set consisted of women randomly selected from each of the four GA at birth categories, and the remainder were included in the validation set. These were selected using the standard function “sample” of the R statistical software package (32). Pregnancies with repeat CL measurements from different screening gestations were accounted for as participants allocated to the training and validation sets were not duplicates of the same pregnancy. We used multinomial log-linear model (via neural networks) to fit the data from the training set using the function “multinom” of the R package “nnet”. A model validation (33) was then performed, where predicted probabilities for categorised GA at birth were tested against observed proportions of GA at birth. Confusion matrices for predicted and true GA at birth were calculated to indicate the sensitivity and specificity of CL measurements in predicting sPTB.

Results

The search identified 1048 citations and following review of the articles, 23 studies met the inclusion criteria (Figure 1). The authors of 12 publications on a combined total of 3989 twin pregnancies provided IPD and were included in the study (6-17, 34) (Table 1). One group, Fox et al (9) provided additional unpublished data on 420 twin pregnancies. Therefore, our study population included a total of 6188 transvaginal scans, performed on 4409 twin pregnancies. 657 women had repeat CL measurements at various screening gestations. Corresponding authors of 11 studies, publishing on a combined total of 1846 twin pregnancies (29.5% of all eligible participants) either did not respond (5 studies) (24-28), were unable to locate or declined to provide IPD following initial consent to contribute (6 studies) (18-23) (Table S1). The corresponding authors of these 11 articles were contacted by phone and email, up to four times over an eight month period to maximise response rate.

Demographic and clinical characteristics of the patient cohorts are provided in Table 1. The respective mean and median at screening GA was 22+3 and 22+0 weeks (Figure S1), and GA at birth was 35+5 and 36+5 weeks. Figure S2 demonstrates the distribution of GA at birth, where 0.005%, 2.9%, 9.3%, and 36.4% delivered at ≤22+0, ≤28+0, ≤32+0, ≤36+0 weeks respectively.

The index of heterogeneity among studies, I2 (total heterogeneity/total variability), was 0% and the Cochran’s Q-test for CL was not significant (p=0.923). A graphical representation of the distribution of the GA at screening and CL data among all studies is provided in Figure S3.

In the first analysis, IPD from all eligible studies were assessed for an association between CL and GA at birth. A multinomal logistic regression model with validation was performed from training (n=400) and validation (n=4009) sets (Table S2). BMI was the only additional variable shown to significantly correlate with GA at birth (Table S3), however when incorporated into a multinomial logistic regression model with CL and GA at ultrasound, prediction for GA at birth was not improved. As demonstrated in Figure S4, both CL and GA at screening have a significant and non-linear effect on predicted GA at birth (p<0.001). In addition there is an overall significant interaction between GA at screening and the measurement of CL (p<0.001). The implications of a short CL vary depending on GA at screening, indicated by the gradient of the curves (Figure S4). A short CL taken at ≤20+0 weeks, indicates a probability of birth significantly earlier than if the same CL was measured at a later GA.

The second analysis provides the predicted probabilities of GA at birth within each of the clinically significant categories (≤28+0, 28+1 to 32+0, 32+1 to 36+0 and ≥36+1 weeks) based on CL and GA at ultrasound (≤18+0, 18+0 to 20+0, 20+1 to 22+0, 22+1 to 24+0 and ≥24+1 weeks). These are shown in Table 2 and illustrated in Figure 2. The odds of birth at ≤28+0 weeks, compared to ≥36+0 weeks increases by 0.77 with every one millimetre decrease in CL (p=0.002). This linear relationship does not persist for sPTB ˃28+0 weeks, where there is no significant association between CL and the odds of sPTB between 28+1 and 36+0 weeks (Figures 2b and 2c).

Screening at ≤18+0 weeks is most significantly predictive for sPTB ≤28+0 weeks, irrespective of CL (p<0.001) (Figure 2a). In contrast, to predict sPTB at 28+1 to 32+0 weeks, screening at ≤18+0 week is most predictive if CL≥20mmm; when CL <15mm, a later GA at screening is superior (Figure 2b). For sPTB ≥32+1 weeks (Figure 2c), most predictive screening is at ≥26+1 weeks. For example, for a CL of 15 mm at 24+0 weeks, the probabilities of sPTB at ≤28+0, 28+1 to 32+0 and 32+1 to 36+0 weeks are 27%, 38% and 28% respectively, as seen in Table 2 and Figure 2. If the same CL of 15 mm is obtained earlier, at 18+0 weeks, there is higher probability of sPTB ≤28+0 weeks (45%), than 28+1 to 32+0 weeks (38%), or 32+1 to 36+0 weeks (12%). 100% negative prediction for sPTB <28+0 weeks was achieved for CL thresholds: 65mm at ≤18+0 weeks, 55mm at 18+0 to 20+0 weeks, 48 mm at 20+1 to 22+0 weeks, and 43 mm at 22+1 to 24+0 weeks.

In a third analysis, the predictive accuracy of the meta-regression model for sPTB was assessed. The model demonstrates improved accuracy of CL and GA at screening in predicting term compared to preterm birth; 68.2% of those predicted to deliver at ≥36+1 weeks were correctly classified (true negative rate), compared to 26.2%, 13.3% and 36.2% who were correctly predicted to deliver at ≤28+0, 28+1 to 32+0 and 32+1 to 36+0 weeks respectively (true positive rate) (Table S4).

Discussion

Main findings

This IPD meta-analysis has shown for the first time the importance of considering GA at screening in the prediction of sPTB in twins from the measurement of CL. Using this model, probabilities may be projected for GA at birth given any CL measurement and GA at screening. These two variables provide accurate prediction of probabilities for sPTB ≤28+0 weeks, at 28+1 to 32+0 weeks and at 32+1 to 36+0 weeks.

There is an inverse and linear relation between CL and sPTB ≤28+0 weeks. Uniquely we demonstrate this relation changes to a non-linear association when predicting later sPTB ≥28+0 weeks.

Strengths and Limitations

The main strength of our study reside in the large number of cases examined and the methodology of the IPD meta-analysis in which both CL and GA at screening are treated as continuous variables. A further strength is the differentiation of predicted GA at birth into very early (≤28+0), early (28+1 to 32+0) and late (32+1 to 36+0) preterm birth. This is clinically preferable to single thresholds such as sPTB <34 weeks as described in previous studies. Provided with the risks of very early, early and late preterm birth, a personalised and cost-effective antenatal management plan may be implemented, including optimal timing of corticosteroid administration and mobilisation to appropriate neonatal units.

The limitations of the study are the lack of data on other variables beyond CL and GA that are known to associate with an increased risk for sPTB. Our model found all other variables (maternal age, smoking, ethnicity, BMI, chorionicity and parity) did not predict sPTB. In addition, while some authors commented on the presence of clinical symptoms at the time of screening, these data were not available for all individuals, therefore could not be tested.

Of the 23 eligible studies, we were unable to obtain IPD from 11 studies, equivalent to 29.5% of the potential participants (n=1849/6258). The studies included and excluded were largely dependent on availability of raw data; several authors of older studies (21-23) did not have their data stored electronically, and were either unable to locate or transfer hard copies of patient data (accounting for 20% of unavailable IPD). Data collection may therefore be biased towards more recently published studies. Indeed a quarter of included IPD were from studies published in the last 2 years (8, 12, 14, 15). A further source of potential bias is the inclusion of unpublished IPD. We had originally requested additional unpublished data from all corresponding authors, however only one author (9) offered a contribution of 450 additional participants. At heterogeneity assessment, this data was comparable to the published IPD. As our study is a collation of observational data in a multicentre international setting, it is unlikely that the missing data, nor the addition of unpublished data will have had a significant impact on the study findings overall.

A further limitation of our study was the inability to evaluate earlier screening <16 weeks or the predictive value of rate of change in cervical length, as we were restricted by availability of previously collected data. Likewise the miscarriage rate <22 weeks (0.005%), is likely to be underestimated; as the median GA at ultrasound was 22 weeks, many delivering <22 weeks were not included in the original observational studies as they had not met the gestational criteria to receive a transvaginal ultrasound. Consequently this analysis cannot comment on the predictive value of CL screening for late miscarriage <22 weeks.

Future studies may consider evaluating the predictive value of CL measurement at earlier screening gestation <16 weeks, as well as the predictive value of rate of CL change in sequential measurements in twin pregnancies.

Interpretation in light of other evidence

Previous meta-analyses in twins confirmed the association between CL and sPTB, using thresholds of CL and GA set by the original study authors (29, 30, 35). This is the first study evaluating individual patient data in the prediction of preterm in twins where CL and GA at birth are investigated as continuous variables. Our findings conclude that risk of sPTB is dependent on the GA at which the CL is obtained. Furthermore, given the GA at screening and CL, we can differentially predict the probability of early versus late preterm birth; a novel and valuable clinical tool.

Berghella et al emphasised the importance of considering GA at screening in the prediction of sPTB in singletons (5). A comparison between the twins in this IPD meta-analysis study and Berghella’s singleton population, indicates a higher risk of sPTB in twins when comparative GA at screening and CL measurements are taken.

Currently, effective interventions for sPTB in twin pregnancies are limited. The focus of prematurity surveillance in twins remains in the antenatal preparation of targeted pregnancies considered most at risk. Timing, in particular for corticosteroid administration, is key to optimising neonatal outcome (36). Therefore this study provides justified evidence for serial CL screening in twin pregnancy as, depending on GA at measurement, early and late sPTB may be differentially predicted. We recommend commencement of initial CL screening at ≤18+0 weeks with repeat screening at >22+0 weeks; this best identifies pregnancies at risk of rarer prematurity-associated mortality (most common with early sPTB, ≤28+0 weeks), as well as the more prevalent prematurity-associated morbidity from later sPTB (28+0 to 36+0 weeks).

Conclusion This IPD meta-analysis of twelve international twin cohorts has shown that both CL and GA at screening contribute to the prediction of GA at sPTB. To optimise prediction of preterm birth at ≤28+0 weeks, CL screening should commence before18+0 weeks. At this stage, any CL <30 mm has a higher risk of sPTB at ≤28+0 weeks in twins than in singletons. Prediction of later sPTB between 28+0 and 36+0 weeks, improves with later GA at measurement ≥22+0 weeks. We therefore recommend CL screening in twins to commence from ≤18+0 weeks.

Acknowledgements We are grateful for the provision of patient level data from the following authors M. Aboulghar, B. Arabin, M. Brizot, Hofmeister, K. Klein, C. Sauvanaud and L. Sperling.

Disclosure of interest: There is no conflict of interest related to this work. Contribution to authorship: LK, LP, PB, KN, AD, TGT contributed to the concept of the study. LK and LP planned the study protocol and collection of data. SC, LK and LP analysed the data. NF, ES, BM, Al, AP, VS, LS, FM, FH provided published data. LK, LP, SC, KN, PB, DM and TGT wrote the manuscript. All authors reviewed and approved the final version of the manuscript. Funding: The study was supported by the Imperial Healthcare NHS Trust Biomedical Research Centre (Grant Ref P45272). Details of ethics approval: This was an analysis of previously collected, anonymised data and did not require ethics approval.

Tables

Table 1. Study characteristics included in the IPD metaanalysis

Table 2. Predicted probability categories

Table S1. Eligible studies, not included in the IPD metaanalysis

Table S2. Model validation characteristics

Table S3. Variables predictive of GA at birth

Table S4. Cross tabulation of predictive accuracy: Percentage of deliveries within predicted gestational age groups.

Figures

Figure 1. Search strategy flow chart

Figure 2a-d. Predicted probability of birth at ≤ 28+0 weeks (a), 28+1-32+0 weeks (b), 32+1-36+0 weeks (c) and ≥ 36+1 weeks (d) based on cervical length measurements (x axis) and gestational age at ultrasound screening

Figure S1. Box and whisker plot of GA at screening (weeks) and CL measurements (mm)

Figure S2. Gestational age distribution at birth.

Figure S3. Distribution of CL measurements and GA at ultrasound by study

Figure S4. Association between GA at birth and CL measurements according to GA at screening

References

1.# Centre#HaSCI.#NHS#Maternity#Statistics#–#2012613.#Hospital#Episode#Statistics.# http://www.hscic.gov.uk/catalogue/PUB12744:#2013.# 2.# ONS.#Gestation6specific#infant#mortality#in#England#and#Wales,#2011.#In:#Statistics#OfN,#editor.# www.ons.gov.uk/ons/child6health/gestation6specific6infant6mortality6in6englang6and6wales/20112013.# 3.# Iams#JD,#Goldenberg#RL,#Meis#PJ,#Mercer#BM,#Moawad#A,#Das#A,#et#al.#The#length#of#the#cervix# and#the#risk#of#spontaneous#premature#delivery.#National#Institute#of#Child#Health#and#Human# Development#Maternal#Fetal#Medicine#Unit#Network.#N#Engl#J#Med.#1996;334(9):567672.# 4.# Heath#VC,#Southall#TR,#Souka#AP,#Elisseou#A,#Nicolaides#KH.#Cervical#length#at#23#weeks#of# gestation:#prediction#of#spontaneous#preterm#delivery.#Ultrasound#in#obstetrics#&#gynecology#:#the# official#journal#of#the#International#Society#of#Ultrasound#in#Obstetrics#and#Gynecology.#1998;12(5):3126 7.# 5.# Berghella#V,#Roman#A,#Daskalakis#C,#Ness#A,#Baxter#JK.#Gestational#age#at#cervical#length# measurement#and#incidence#of#preterm#birth.#Obstet#Gynecol.#2007;110(2#Pt#1):31167.# 6.# Aboulghar#MM,#Aboulghar#MA,#Mourad#L,#Serour#GI,#Mansour#RT.#Ultrasound#cervical# measurement#and#prediction#of#spontaneous#preterm#birth#in#ICSI#pregnancies:#a#prospective#controlled# study.#Reprod#Biomed#Online.#2009;18(2):2966300.# 7.# Arabin#B,#Roos#C,#Kollen#B,#van#Eyck#J.#Comparison#of#transvaginal#sonography#in#recumbent#and# standing#maternal#positions#to#predict#spontaneous#preterm#birth#in#singleton#and#twin#pregnancies.# Ultrasound#in#obstetrics#&#gynecology#:#the#official#journal#of#the#International#Society#of#Ultrasound#in# Obstetrics#and#Gynecology.#2006;27(4):377686.# 8.# Brizot#ML,#Hernandez#W,#Liao#AW,#Bittar#RE,#Francisco#RP,#Krebs#VL,#et#al.#Vaginal#progesterone# for#the#prevention#of#preterm#birth#in#twin#gestations:#a#randomized#placebo6controlled#double6blind# study.#Am#J#Obstet#Gynecol.#2015.# 9.# Fox#NS,#Rebarber#A,#Klauser#CK,#Peress#D,#Gutierrez#CV,#Saltzman#DH.#Prediction#of#spontaneous# preterm#birth#in#asymptomatic#twin#pregnancies#using#the#change#in#cervical#length#over#time.#Am#J# Obstet#Gynecol.#2010;202(2):155.e164.# 10.# Hofmeister#C,#Brizot#MeL,#Liao#A,#Francisco#RP,#Zugaib#M.#Two6stage#transvaginal#cervical#length# screening#for#preterm#birth#in#twin#pregnancies.#Journal#of#perinatal#medicine.#2010;38(5):479684.# 11.# Klein#K,#Gregor#H,#Hirtenlehner6Ferber#K,#Stammler6Safar#M,#Witt#A,#Hanslik#A,#et#al.#Prediction# of#spontaneous#preterm#delivery#in#twin#pregnancies#by#cervical#length#at#mid6gestation.#Twin#research# and#human#genetics#:#the#official#journal#of#the#International#Society#for#Twin#Studies.#2008;11(5):55267.# 12.# Liem#S,#Schuit#E,#Hegeman#M,#Bais#J,#de#Boer#K,#Bloemenkamp#K,#et#al.#Cervical#pessaries#for# prevention#of#preterm#birth#in#women#with#a#multiple#pregnancy#(ProTWIN):#a#multicentre,#open6label# randomised#controlled#trial.#Lancet.#2013;382(9901):134169.# 13.# Lim#AC,#Schuit#E,#Papatsonis#D,#van#Eyck#J,#Porath#MM,#van#Oirschot#CM,#et#al.#Effect#of#176alpha# hydroxyprogesterone#caproate#on#cervical#length#in#twin#pregnancies.#Ultrasound#in#obstetrics#&# gynecology#:#the#official#journal#of#the#International#Society#of#Ultrasound#in#Obstetrics#and#Gynecology.# 2012;40(4):426630.# 14.# Sauvanaud#C,#Equy#V,#Faure#C,#Boussat#B,#Hoffmann#P,#Sergent#F.#[Transvaginal#sonographic# cervical#length#and#prediction#of#preterm#delivery#in#twin#pregnancies#with#preterm#labor].#J#Gynecol# Obstet#Biol#Reprod#(Paris).#2013;42(5):488692.# 15.# Serra#V,#Perales#A,#Meseguer#J,#Parrilla#JJ,#Lara#C,#Bellver#J,#et#al.#Increased#doses#of#vaginal# progesterone#for#the#prevention#of#preterm#birth#in#twin#pregnancies:#a#randomised#controlled#double6 blind#multicentre#trial.#BJOG#:#an#international#journal#of#obstetrics#and#gynaecology.#2013;120(1):5067.# 16.# Sperling#L,#Kiil#C,#Larsen#LU,#Qvist#I,#Bach#D,#Wojdemann#K,#et#al.#How#to#identify#twins#at#low#risk# of#spontaneous#preterm#delivery.#Ultrasound#in#obstetrics#&#gynecology#:#the#official#journal#of#the# International#Society#of#Ultrasound#in#Obstetrics#and#Gynecology.#2005;26(2):138644.# 17.# To#MS,#Fonseca#EB,#Molina#FS,#Cacho#AM,#Nicolaides#KH.#Maternal#characteristics#and#cervical# length#in#the#prediction#of#spontaneous#early#preterm#delivery#in#twins.#Am#J#Obstet#Gynecol.# 2006;194(5):136065.# 18.# Goya#M,#Pratcorona#L,#Merced#C,#Rodó#C,#Valle#L,#Romero#A,#et#al.#Cervical#pessary#in#pregnant# women#with#a#short#cervix#(PECEP):#an#open6label#randomised#controlled#trial.#Lancet.# 2012;379(9828):180066.# 19.# Wennerholm#UB,#Holm#B,#Mattsby6Baltzer#I,#Nielsen#T,#Platz6Christensen#J,#Sundell#G,#et#al.#Fetal# fibronectin,#endotoxin,#bacterial#vaginosis#and#cervical#length#as#predictors#of#preterm#birth#and# neonatal#morbidity#in#twin#pregnancies.#Br#J#Obstet#Gynaecol.#1997;104(12):13986404.# 20.# Ehsanipoor#RM,#Haydon#ML,#Lyons#Gaffaney#C,#Jolley#JA,#Petersen#R,#Lagrew#DC,#et#al.# Gestational#age#at#cervical#length#measurement#and#preterm#birth#in#twins.#Ultrasound#in#obstetrics#&# gynecology#:#the#official#journal#of#the#International#Society#of#Ultrasound#in#Obstetrics#and#Gynecology.# 2012;40(1):8166.# 21.# Goldenberg#RL,#Iams#JD,#Miodovnik#M,#Van#Dorsten#JP,#Thurnau#G,#Bottoms#S,#et#al.#The#preterm# prediction#study:#risk#factors#in#twin#gestations.#National#Institute#of#Child#Health#and#Human# Development#Maternal6Fetal#Medicine#Units#Network.#Am#J#Obstet#Gynecol.#1996;175(4#Pt#1):1047653.# 22.# Guzman#ER,#Walters#C,#O'reilly6Green#C,#Kinzler#WL,#Waldron#R,#Nigam#J,#et#al.#Use#of#cervical# ultrasonography#in#prediction#of#spontaneous#preterm#birth#in#twin#gestations.#Am#J#Obstet#Gynecol.# 2000;183(5):110367.# 23.# Gibson#JL,#Macara#LM,#Owen#P,#Young#D,#Macauley#J,#Mackenzie#F.#Prediction#of#preterm# delivery#in#twin#pregnancy:#a#prospective,#observational#study#of#cervical#length#and#fetal#fibronectin# testing.#Ultrasound#in#obstetrics#&#gynecology#:#the#official#journal#of#the#International#Society#of# Ultrasound#in#Obstetrics#and#Gynecology.#2004;23(6):56166.# 24.# Asnafi#N,#Basirat#Z,#Hajian6Tilaki#K,#Dadvar#S.#Assessment#of#cervical#length#by#transvaginal# ultrasonography#to#predict#preterm#delivery#in#twin#pregnancy.#The#journal#of#maternal6fetal#&#neonatal# medicine#:#the#official#journal#of#the#European#Association#of#Perinatal#Medicine,#the#Federation#of#Asia# and#Oceania#Perinatal#Societies,#the#International#Society#of#Perinatal#Obstet.#2013;26(14):143568.# 25.# Boyer#A,#Cameron#L,#Munoz6Maldonado#Y,#Bronsteen#R,#Comstock#CH,#Lee#W,#et#al.#Clinical# significance#of#amniotic#fluid#sludge#in#twin#pregnancies#with#a#short#cervical#length.#Am#J#Obstet# Gynecol.#2014;211(5):506.e169.# 26.# Senat#MV,#Porcher#R,#Winer#N,#Vayssière#C,#Deruelle#P,#Capelle#M,#et#al.#Prevention#of#preterm# delivery#by#17#alpha6hydroxyprogesterone#caproate#in#asymptomatic#twin#pregnancies#with#a#short# cervix:#a#randomized#controlled#trial.#Am#J#Obstet#Gynecol.#2013;208(3):194.e168.# 27.# Imseis#HM,#Albert#TA,#Iams#JD.#Identifying#twin#gestations#at#low#risk#for#preterm#birth#with#a# transvaginal#ultrasonographic#cervical#measurement#at#24#to#26#weeks'#gestation.#Am#J#Obstet#Gynecol.# 1997;177(5):1149655.# 28.# Vayssière#C,#Favre#R,#Audibert#F,#Chauvet#MP,#Gaucherand#P,#Tardif#D,#et#al.#Cervical#length#and# funneling#at#22#and#27#weeks#to#predict#spontaneous#birth#before#32#weeks#in#twin#pregnancies:#a# French#prospective#multicenter#study.#Am#J#Obstet#Gynecol.#2002;187(6):15966604.# 29.# Lim#AC,#Hegeman#MA,#Huis#In#TVMA,#Opmeer#BC,#Bruinse#HW,#Mol#BW.#Cervical#length# measurement#for#the#prediction#of#preterm#birth#in#multiple#pregnancies:#a#systematic#review#and# bivariate#meta6analysis.#Ultrasound#in#obstetrics#&#gynecology#:#the#official#journal#of#the#International# Society#of#Ultrasound#in#Obstetrics#and#Gynecology.#2011;38(1):1067.# 30.# Conde6Agudelo#A,#Romero#R,#Hassan#SS,#Yeo#L.#Transvaginal#sonographic#cervical#length#for#the# prediction#of#spontaneous#preterm#birth#in#twin#pregnancies:#a#systematic#review#and#metaanalysis.#Am# J#Obstet#Gynecol.#2010;203(2):128#e1612.# 31.# Higgins#JP,#Thompson#SG,#Deeks#JJ,#Altman#DG.#Measuring#inconsistency#in#meta6analyses.#Bmj.# 2003;327(7414):557660.# 32.# Ripley#BD.#Stochastic#simulation1987.# 33.# Venables#W,#Ripley#B.#Modern#applied#statistics#with#S.#4th#Edition#ed:#Springer;#2002.# 34.# Lim#AC,#Bloemenkamp#KW,#Boer#K,#Duvekot#JJ,#Erwich#JJ,#Hasaart#TH,#et#al.#Progesterone#for#the# prevention#of#preterm#birth#in#women#with#multiple#pregnancies:#the#AMPHIA#trial.#BMC#Pregnancy# Childbirth.#2007;7:7.# 35.# Schuit#E,#Stock#S,#Rode#L,#Rouse#DJ,#Lim#AC,#Norman#JE,#et#al.#Effectiveness#of#progestogens#to# improve#perinatal#outcome#in#twin#pregnancies:#an#individual#participant#data#meta6analysis.#BJOG#:#an# international#journal#of#obstetrics#and#gynaecology.#2015;122(1):27637.# 36.# Roberts#D,#Dalziel#S.#Antenatal#corticosteroids#for#accelerating#fetal#lung#maturation#for#women# at#risk#of#preterm#birth.#The#Cochrane#database#of#systematic#reviews.#2006(3):CD004454.#