Long‐term follow‐up of human papillomavirus type replacement among young pregnant Finnish females before and after a community‐randomised HPV vaccination trial with moderate coverage

Large scale human papillomavirus (HPV) vaccination against the most oncogenic high‐risk human papillomavirus (HPV) types 16/18 is rapidly reducing their incidence. However, attempts at assessing if this leads to an increase of nonvaccine targeted HPV types have been hampered by several limitations, such as the inability to differentiate secular trends. We performed a population‐based serological survey of unvaccinated young women over 12 years. The women were under 23‐years‐old, residents from 33 communities which participated in a community‐randomised trial (CRT) with approximately 50% vaccination coverage. Serum samples were retrieved pre‐CRT and post‐CRT implementation. Seropositivity to 17 HPV types was assessed. HPV seroprevalence ratios (PR) comparing the postvaccination to prevaccination era were estimated by trial arm. This was also assessed among the sexual risk‐taking core group, where type replacement may occur more rapidly. In total, 8022 serum samples from the population‐based Finnish Maternity Cohort were retrieved. HPV types 16/18 showed decreased seroprevalence among the unvaccinated in communities only after gender‐neutral vaccination (PR16/18A = 0.8, 95% CI 0.7‐0.9). HPV6/11 and HPV73 were decreased after gender‐neutral vaccination (PR6/11A = 0.8, 95% CI 0.7‐0.9, PR73A = 0.7, 95% CI 0.6‐0.9, respectively) and girls‐only vaccination (PR6/11B = 0.8, 95% CI 0.7‐0.9, PR73B = 0.9, 95% CI 0.8‐1.0). HPV68 alone was increased but only after girls‐only vaccination (PR68B = 1.3, 95% CI 1.0‐1.7, PRcore68B = 2.8, 95% CI 1.2‐6.3). A large‐scale, long‐term follow‐up found no type replacement in the communities with the strongest reduction of vaccine HPV types. Limited evidence for an increase in HPV68 was restricted to girls‐only vaccinated communities and may have been due to secular trends (ClinicalTrials.gov number: NCT00534638).


| INTRODUCTION
Oncogenic human papillomavirus (HPV), the necessary cause of cervical cancer, 1 is a well-established causal agent of several anogenital and oropharyngeal cancers. There are currently three efficacious HPV-vaccines licensed which target the two most high-risk HPV types HPV16 and 18, (the two first-generation vaccines, Cervarix and Gardasil), or five additional high-risk HPV types, HPV31, 33,45,52 and 58 (the nonavalent vaccine, Gardasil9). 2 Since 2007, these vaccines have been gradually implemented in national vaccination programs. 3 However, there are a total of 12 high-risk HPV types which are classified by the International Agency for Research on Cancer (IARC) as carcinogenic to humans, HPV16, 18,31,33,35,39,45,51,52, 56, 58 and 59, with a further eight types classified as possibly (or probably) carcinogenic, HPV26, 53, 66, 68, 67, 70, 73 and 83. 4 As such, concern has been flagged over a decade ago, whether such selective vaccination could induce HPV type replacement to occur. 5 By removing selected HPV types, vaccination may disrupt the dynamic equilibrium among HPV types. Subsequently, the vacant niche may become superceded by one or more of the nonvaccine types. This vaccine-induced evolutionary response in the niche habitation by the nonvaccine types has already been described in analogous situations, for example, following vaccination against Streptococcus pneumoniae, and is commonly known as type replacement. 6,7 Several studies have evaluated the occurrence of HPV type replacement, using differing methodologies; comparison of the HPV prevalence between the postvaccination and prevaccination era, [8][9][10] between vaccinated and unvaccinated persons in the postvaccination era 11 or via individually randomised HPV vaccination trials. 12 Importantly, a recent meta-analysis of the above-mentioned studies found an increasing trend in the pooled nonvaccine targeted (and noncrossprotected) HPV types, 13 further to an earlier meta-analysis which found possible increases in HPV 39 and 52. 14 However, there are multiple major limitations when evaluating the occurrence of HPV type replacement. When conducting postvaccination era surveillance and comparing the HPV prevalence prevaccination and postvaccination era, it is difficult to distinguish possible increases due to vaccine induced-type replacement from that due to secular trends. On the other hand, evaluating type replacement by means of negative vaccine effectiveness in the postvaccination era, may be an unsuitable measure for both the identification and prediction of type replacement occurrence. That is, type replacement may be subdued in the vaccinated by vaccine-induced cross-protection, while manifesting with lesser limitation in the unvaccinated due to the indirect impacts of community-wise vaccination. 15 Last but not least, when evaluating type replacement via individually randomised clinical trials, it is likely that any estimates will underestimate the probability of type replacement, as the vaccination-induced selective pressure stems from too small a proportion of the population in comparison to that present after community-wise vaccination. 16 Further to these, following vaccination against several other pathogen types, there has been a transitory "honeymoon period" immediately following vaccination implementation, before arriving at a new endemic equilibrium. 17,18 In the context of HPV vaccination, a recently published modelling study has found that there may be a HPV type replacement "honeymoon period" following vaccination, wherein nonvaccine types may at first appear to remain stable or even decrease before rebounding due to type replacement after a certain

What's new?
Vaccination efforts have decreased the prevalence of oncogenic HPV types, such as HPV 16/18. This may create space for other types to expand, but it is difficult to distinguish the effect of the vaccine from long-term temporal trends. Here, the authors conducted a community-randomized trial in 33 communities. Each group of 11 communities received either gender-neutral HPV vaccination, girls-only HPV vaccination, or gender neutral hepatitis vaccination. In the girls-only arm, vaccination reduced the prevalence of HPV 6/11, and HPV 68 increased in prevalence. However, this effect may have been due to secular trends. incubation period since vaccination initiation. 15 Thus, previous inconclusive findings of any HPV type replacement may have been premature to identify type replacement occurrence.
We now evaluate the occurrence of HPV type replacement in a decade following a large population-based community-randomised HPV vaccination trial with close to 50% vaccination coverage (beginning in October 2007), by conducting a survey of HPV seroprevalence in unvaccinated Finnish female community residents over the pretrial and posttrial era. Herpes simplex virus type II (HSV-2) serology is an established marker of sexual risk taking, thereby we now utilise this as a proxy of core group membership, 19 to further investigate the occurrence of type replacement also within the core-group (the assortative subgroup of the population with high sexual contact rates), as it has been suggested that HPV type replacement may first manifest within this group. 20 2 | MATERIALS AND METHODS

| Study design and materials
The material of our study comprises of longitudinal population-based biobank follow-up of the Finnish community-randomised HPV vaccination trial evaluating the comparative effectiveness of girls-only or gender-neutral HPV vaccination (NCT00534638). [21][22][23] Briefly, in 2007, 33 Finnish communities were stratified according to preascertained HPV16/18 seroprevalence, 24 into those with low, moderate or high seroprevalence. From each seroprevalence strata, the communities were then randomised to one of three trial arms, with an allocation ratio of 11:11:11. 21 All Swedish or Finnish speaking 1992 to 1995 born adolescents who were residing in the trial commu-

| Laboratory analyses
The retrieved FMC serum samples were analysed for the presence of serum antibodies to HPV6, 11, 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68, 73 and herpes simplex virus type II (HSV-2) using heparin bound HPV pseudovirion and HSV-2 glycoprotein G 2 Luminex assay. 22,28,29 A negative control panel of serum samples from children under the age of 12-years-old was used to calculate HPV type-specific seropositivity cut-off values, by computing the median fluorescence intensities of the negative control serum panel plus three SDs. 22 For this study, the assay panel was extended to include also HPV51, 66 and HSV-2, more details of which are described in the Supporting Information Methods and reference 22.

| Statistical analyses
The magnitude of within-arm clustering was measured by calculating the intracluster correlation coefficient (ICC) using observations from 2005 to 2010, among all subjects and subjects stratified by HSV-2 seropositivity. The ICC of vaccine targeted and nonvaccine targeted HPV was estimated using Fleiss and Cuzick's estimator and the accompanying 95% confidence intervals estimated using Zou and Donners modified Wald test. 30,31 To evaluate the occurrence of HPV type replacement, the abso- were adjusted for era-specific community-wise maternal smoking as a proxy measure of general risk-taking behaviour. To evaluate the occurrence of type replacement among the core group, with-in arm PRs were similarly estimated among subjects who were HSV-2 seropositive (as a proxy measure of core-group membership). To take account for a possible delay between niche clearance and type replacement occurrence, with-in arm PRs were further stratified by the postvaccination era numerator, into the first or second postvaccination era, 2011 to 2013 or 2014 to 2016, respectively, compared to the entire prevaccination era. To assess whether observed increased with-in arm PRs were due to secular trends or type replacement, type specific with-in arm PRs from the intervention arms were compared to the with-in arm PRs from the control arm using the methodology of Altman and Bland. 32 As a sensitivity analysis, probabilistic bias analysis was used to take account of misclassification owing to the known lack of seroconversion in a proportion of individuals following HPV infection. Previously 33 and currently ascertained (see Appendix for HPV51 and 66) specificity and sensitivity values of this heparin-bound HPV pseudovirion serology to identify cumulative infections were used. As a further sensitivity analysis, the HPV type-specific PRs were additionally estimated stratified by subjects age at the time of sample donation (into those aged 14-to 19-years-old and 20-to 22years-old).
The statistical analyses were conducted using R software package   In the prevaccination era reporting for the 1992-1995 birth cohorts is merged to avoid reporting identifiable data due to small count numbers (n < 5).  The shape of the birth cohort distribution remained the same in the prevaccination to postvaccination era, but with a shift towards the younger cohorts in the postvaccination era (from those born 1992 and younger, to those born in 1988 and younger; Table 1). The HSV-2 seroprevalence was somewhat reduced between the prevaccination and postvaccination eras among subjects from Arm B (19.2%-14.0%), although not notably altered in subjects from Arms A or C ( Table 1). The vaccination coverage among the 1992 to 1995 females was comparatively similar among the intervention Arms A and B (Figure 1).

| HPV seroprevalence, postvaccination vs prevaccination era
Among the subjects, the crude seroprevalence, P, of vaccine targeted HPV16 and 18 was notably high among all three intervention arms T A B L E 2 Seroprevalence ratios (95% confidence intervals) comparing the seroprevalence in the postvaccination era, 2011-2016, to that in the prevaccination era among, A, all subjects and B, among the core-group (identified by herpes simplex virus type 2 seropositivity) to P HPV66 = 14.4% and from P HPV73 = 12.8% to P HPV73 = 9.3%; Figure 3).
When comparing the postvaccination era to the prevaccination era among all the subjects, the smoking adjusted seroprevalence ratios of both HPV16 and 18 were reduced among Arm A only (Table 2A). was again found to be further increased in the second postvaccination era, but not significantly different to that found in Arm C (Table S3).

| Comparing HPV seroprevalence changes by trial arm
When comparing the within-arm PRs between the Arms A or B to Arm C, we found no notable increases in the nonvaccine HPV types' T A B L E 3 HPV type-specific seroprevalence ratios comparing the seroprevalence in the postvaccination era, to that in the prevaccination era  ratios of seroprevalence ratios (RPRs; Figure S1). When comparing Arm B to Arm C core-groups, although the HPV68 RPR was increased, the confidence intervals overlapped the null (RPR = 1.98, 0.73-5.40), and for the latest subjects, the RPR approximated 1 ( Figure S2, Table S4).

| DISCUSSION
Of all the nonvaccine targeted HPV types measured, only the sero- When comparing this postvaccination vs prevaccination increase to that in the counterfactual control Arm C, this increase among Arm B was found to disappear, except in the core-group. This might suggest that the observed increase in Arm B may have been partially due to secular trends rather than type replacement. Furthermore, the core-group observations lacked statistical significance, meaning that the observed increase may have been due to chance.
Although the finding of increased HPV68 may be explainable due to aforementioned reasons other than type replacement, this possibility deserves comment. Previous modelling studies assuming that HPV types compete via the hosts immune system, have demonstrated that the occurrence of type replacement, and the ability to observe it at an early stage of the postvaccination era, is a trade-off between vaccineinduced cross-protection and naturally acquired cross-immunity. 15 HPV68 is from the alpha 7 species, and although phylogenetically related to HPV18 present in the bivalent vaccine, has not been shown following vaccination programs or clinical trials to be cross-protected by the vaccine. 34 Although a large degree of vaccine-induced crossprotection has been demonstrated against types phylogenetically related to vaccine targeted HPV16 and 18, alpha 9 and alpha 7 species, respectively, the vaccine has been much more successful at eliciting cross-neutralising antibodies to HPV types from alpha 9, than it has to HPV types from alpha 7. 35 From the alpha 7 species, only HPV45 has been shown to be cross-protected by the bivalent vaccine. 13 Thus, the HPV68 increase will likely not be mitigated by vaccine-induced cross-protection. Furthermore, the fact that the observed increase in HPV68 after girls-only vaccination stems more from the second postvaccination era, is in line with a proposed honeymoon period postvaccination before type replacement might occur. 15 We noted a decrease in both the occurrence of low-risk type HPV6, HPV 66 and the possibly high-risk type HPV73. This reduction in HPV6 was found to exactly follow the patterns that would be expected if the bivalent vaccine had induced HPV6 herd effect; the observed reduction was the greatest postgender-neutral vaccination and when comparing to the control Arm C was not explainable due to secular trends. HPV6 is a very common low-risk HPV type, responsible for a large proportion of genital warts, and not phylogenetically related to either of the vaccine-targeted types. Nevertheless, this finding of a possible HPV 6 herd effect postbivalent HPV vaccination, is consistent with previous findings of bivalent vaccine efficacy against HPV6, 36 reported postvaccine era reductions in genital warts when using the bivalent vaccine, 37,38 and also with findings of HPV6 specific vaccine-induced cross-neutralising antibodies among individuals vaccinated with the bivalent HPV vaccine. 35  to be causally associated with the development of invasive cervical cancer. 39 HPV73 is from the alpha 11 species group, which is of particular interest given that the alpha 11 species group is phylogenetically close to the alpha 9 species, and the degree of cross-protection is correlated to the phylogenetic distance to the vaccine types. 40 Contrary to this, the observed decrease of HPV66 in Arm A was unexpected. Previous studies have not documented any decrease in HPV66 following the induction of the bivalent HPV vaccine. In several such studies, HPV66 was not included in the laboratory assay, 41,42 and where HPV66 was evaluated, there was no notable vaccine efficacy observed against a HPV infection endpoint. 43,44 Thus, the currently observed decrease in HPV66 necessitates further study before any causal relationship may be asserted, to guard against a chance finding among the multiple HPV type comparisons which are commonplace in such HPV type replacement studies.
Even with a decade of follow-up, our study may still be limited in its ability to evaluate HPV type replacement. A recent modelling study found that there may be a honeymoon period of 10 years after the start of HPV vaccination in a population, before HPV type replacement starts to occur. 15 However, this is only considering the scenario where prevaccination competition occurs via naturally acquired crossimmunity, 15 thus if type competition occurs via another mechanism, such as for resources (eg, competition for available micro-abrasions), then it may be that this honeymoon period either does not apply or is altered. Therefore, whether or not our survey is limited by the followup may be subject to the biology of HPV type competition.
Furthermore, our study may also suffer from bias due to misclassification of the outcome, cumulative HPV infection. As previously described, 22 45 When using HPV serology, on the other hand, it allows for the identification of those individuals who have had true persistent infections, and is a measure of cumulative HPV infection exposure. 46 Serum IgG antibodies induced by natural infection specific to HPV types have previously been shown to be stable over several years of follow-up among women. 47 However, this method will also incorrectly identify a proportion of individuals as negative who have previously had a HPV infection but have not seroconverted; among a sample of Swedish women with clinically confirmed HPV16 infections only 65% were found to be HPV16 seropositive. 48 Therefore, by using type-specific HPV serology as a measure of cumulative infection our comparative measures of HPV occurrence are likely deviated towards the null.
There are many difficulties in designing a study with the ability to evaluate type replacement of any kind. The community-randomised trial design of our study with both pretrial and posttrial outcome measurements is best placed to evaluate type replacement and avoid many of these common problems. 15 The pretrial and posttrial mea- Although our study is highly generalizable to the wider pregnant Finnish population under the age of 23 due to the population-based nature of the study, it may be limited in its generalisability to all females under the age of 23 years old. Although the prevalence of maternal smoking (an indicator of general risk-taking behaviour) is high in our study population, it is almost identical to that found in previous studies of pregnant females of similar age over for the total population of Finland. 49 However, it is likely that our study population has above and below average sexual risk-taking behaviours compared to the general population given that the average age of first pregnancy in Finland is currently 29-years-old. 50 Despite the presence of our control Arm C to control for secular trends, our study still may not have been able to completely distinguish the magnitude of the increase in HPV68 due to type replacement from that due to secular trends, for example, if the parallel trend assumption between the Arms does not hold. Finally, although our study with moderate vaccination coverage may mimic typical vaccination coverages achieved in many national vaccination programs, it is limited in its transportability to scenarios with greater vaccination coverage.
In conclusion, no clear indications of type replacement were found, as of yet. Possible increases in HPV68 after girls-only vaccination may have resulted from secular trends. Continued monitoring in the postvaccination era to confirm or refute possible HPV type replacement by HPV68 and all other nonvaccine targeted HPV types is necessary.