How polio vaccination affects poliovirus transmission

A fundamental complexity of polio eradication is that the elimination of wild poliovirus (WPV) alters the risk-benefit profile of using oral polio vaccine (OPV)---as WPV is eliminated, OPV produces an increasing proportion of the paralytic disease burden since, in rare instances, OPV causes paralysis in vaccine recipients and generates circulating vaccine-derived polio outbreaks (cVDPV) in under-immunized populations. Therefore, to secure the success and long-term stability of polio eradication after the elimination of WPV, OPV use should eventually cease. Type 2 OPV (OPV2) was withdrawn from routine immunization (RI) in April 2016, but detection of type 2 cVDPV has necessitated the use of OPV2 in outbreak response. Thus the world today: RI with OPV2 has stopped, but OPV2 is needed to interrupt outbreaks, and any future OPV2 use several years hence will take place in a population with an unprecedented lack of type 2 immunity. To better understand the complex risk landscape of OPV cessation, we reproducibly summarized data spanning 75 years of polio literature detailing how vaccination affects individual-level susceptibility to infection and viral shedding. We then examined individual-level immunity in the context of close-contact transmission data from the USA and India to quantify the impacts of vaccination on transmission. Our results demonstrate that in settings with inadequate sanitation: (1) OPV has been effective in all populations because it blocks transmission locally, (2) cross-immunity against type 2 produced by bivalent OPV is insufficient to block OPV2 transmission, (3) IPV boosting after prior OPV or WPV exposure is effective for interrupting transmission in settings where inadequate or waning immunity permits significant re-infection, and (4) OPV transmission is limited more by population immunity than attenuation and so the risk of seeding new cVDPV with OPV use will increase substantially a few years after OPV cessation. We conclude with discussion of the implications for policy decisions about IPV and OPV use and vaccine research.


Introduction
1 are readily transmissible. This transmissibility provides additional passive immunization that 8 enhances the effectiveness of OPV for generating herd immunity. However, Sabin OPV can in rare 9 instances cause paralytic poliomyelitis [5] and establish endemic circulation of vaccine-derived 10 poliovirus (cVDPV) [6]. Thus, to complete the task of poliovirus eradication, Sabin OPV 11 vaccination must eventually cease [7]. 12 The dual role of Sabin OPV as both a vaccine and a source of transmissible poliovirus is 13 responsible for key uncertainties surrounding the ability of the Global Polio Eradication Initiative 14 (GPEI) to achieve and sustain eradication. To date, polio outbreaks have taken place in regions of 15 low immunity against infection surrounded by regions of high immunity [8], OPV campaigns 16 implemented in outbreak response have been effective for interrupting transmission [3], and cVDPV 17 epidemics have been rare [9]. However, within a few years of global OPV cessation, a birth cohort 18 will accumulate with an unprecedented lack of immunity against poliovirus infection. If poliovirus 19 outbreaks occur after cessation due to accidental or deliberate re-introduction [10][11][12], or sustained 20 silent transmission [2,[13][14][15] as has recently been observed in Nigeria following the April 2016 global 21 type 2 OPV cessation [2], will cVDPV emergences following OPV use remain rare? 22 The answer to that question requires a quantitative understanding of deeper questions about how 23 the facts of individual-level immunity, viral infectivity, and transmission dynamics fit together to 24 explain the epidemiology of poliovirus transmission. Immunity derived from multiple vaccination 25 with trivalent OPV (tOPV) reduces poliovirus shedding after oral exposure by a few orders of 26 magnitude [4], but what is the quantitative relationship between shedding and transmission? How 27 does the relationship vary among populations with different levels of fecal-oral exposure? After OPV 28 cessation, will the effectiveness of routine immunization (RI) with inactivated polio vaccine (IPV) 29 against transmission in high income countries [16][17][18][19][20] generalize to all settings to prevent cVDPV 30 emergence from OPV use, given the limited effectiveness of IPV alone against fecal shedding [21] and 31 the proven ability of WPV to transmit in an IPV-only country [15,22]? Post-OPV2-cessation, how immunity" therein). 131 The concept of OPV-equivalent humoral antibody titer unifies shedding and acquisition data for 132 different vaccine schedules. Following the results of Behrend et al [30], we assumed that the typical 133 immunologically-naive individual with no history of poliovirus exposure ("unvaccinated") and no 134 measurable humoral immunity ("seronegative") is defined to have an OPV-equivalent humoral 135 antibody titer equal to one: N Ab = 1, that the maximum median homotypic OPV-equivalent titer is 136 N Ab = 2048 (= 2 11 ), and that homotypic antibody titers for each serotype are independent. 137 Sources of data on individual-level shedding and acquisition 138 Almost all relevant studies on OPV shedding, acquisition, and transmission published prior to 2012 139 were reviewed by Duintjer Tebbens et al [9]. Digitized data on shedding duration and fecal viral load 140 were taken from the supplementary material in Behrend et al [30], corrected where discrepancies 141 were noticed, and studies missing or involving bOPV were added [23][24][25]. Dose response data were 142 first digitized and made publically-available here. The analyses are broadly inclusive of published 143 data, but this paper does not represent a systematic review with pre-specified exclusion criteria. 144 Whole studies and trial arms were excluded if they reported evidence of substantial uncontrolled or 145 unmeasured natural exposure to either wild poliovirus or vaccines strains by contact prior to OPV 146 challenge or WPV infection [38][39][40][41][42][43][44] or when data across vaccination regimens or serotypes could not 147 be disaggregated [45]. We included OPV challenge studies in which low levels of natural exposure 148 were possible but not common. A summary of all included data describing vaccination regimens, 149 OPV challenge formulation or WPV exposure, ages, and available shedding and dose reponse data, 150 and possible natural exposure is provided in Table S2 [23-25, 33, [46][47][48][49][50][51][52][53][54][55][56][57][58][59][60][61][62]. For a deeper discussion of 151 data quality from the included studies, see Duintjer Tebbens et al [9]. with the text correspond to two-tailed Fisher's exact tests. All model parameters describing 159 prevalences were fit by maximum likelihood assuming binomial sampling (at each time point when 160 relevant), and 95% confidence intervals were estimated by parametric bootstrap with 1000 replicates. 161 Models for positive-definite quantities (concentration of poliovirus in stool, antibody titer) were 162 estimated by ordinary least squares on log(quantity), and 95% confidence intervals assume 163 log-normality. To estimate bootstrap confidence intervals of parameters that are 164 conditionally-dependent on previously estimated parameters, we propagated uncertainty by 165 independently resampling parameters from the 95% confidence intervals assuming normality 166 (log-normality where appropriate) prior to resampling the data and re-estimating the parameters immunity were also included [46,59]. As described in detail below, RI regimen is predictive of 179 shedding duration. Conditional on RI regimen, data exploration revealed no associations of shedding 180 duration with age at OPV challenge or the precise RI schedule. Three studies of shedding duration 181 after WPV exposure contained adequate longitudinal data to estimate the duration of shedding after 182 wild poliovirus exposure in previously unimmunized individuals (either paralytic cases or individuals 183 with known serology) [34, 63,64]; no data are available to test for serotype differences in WPV 184 shedding duration. 185 Immunologically-naive and maximally-immune individuals. With respect to shedding 186 after OPV challenge, there were no significant differences in shedding duration between unvaccinated 187 and confirmed seronegative children, and thus both subject types constitute the class of immunologically-naive individuals. There were also no significant differences in Sabin shedding 189 duration by serotype. Conditional on vaccine take, the maximum likelihood estimate of the median 190 shedding duration in an immunologically-naive individual shedding any Sabin strain is 191 30.3 (23. 6, 38.6) days, shorter than the median shedding duration of WPV, 43.0 (35.7, 51.7) days 192 (Fig. 1). The Sabin shedding duration given vaccine take associated with maximum antibody titer 193 (N Ab = 2048) is 6 (4, 10) days, as defined by the data from the tOPVx3 arm of Asturias et al [24]. 194 That trial arm was chosen to define maximal immunity because its subjects had the shortest interval 195 between the RI and OPV challenge (4 weeks), shortest median shedding duration and lowest titer based on the data in Fig. 1, is plausibly distributed as: 202 P shedding at t N Ab ; infected at t = 0 = 1 used without derivation in references [36,65]. 207 Effects of routine immunization on shedding duration and pre-challenge immunity. 208 To enable quantitative comparisons between different RI regimens of the effect of pre-challenge 209 immunity on shedding after OPV challenge, we estimated the median shedding durations and 210 inferred pre-challenge OPV-equivalent antibody titers for all represented RI regimens and serotypes 211 using the maximum likelihood model for shedding duration after OPV challenge in Eq. (1) and 212 aggregated data for each RI regimen. Results across all RI regimens are shown in Fig. S2  of tOPVx3 data; waning is discussed in detail later). As a rule of thumb, an additional dose of OPV 226 increases the modeled OPV-equivalent antibody titer by roughly a factor of 10.

227
Routine immunization with IPV only. There is no cumulative reduction of shedding duration 228 with the number of pre-challenge IPV doses (Fig. 3). All IPV-only RI regimens with significant 229 inferred pre-challenge immunity are supported by data from IPV trials conducted in otherwise 230 tOPV-using or pre-WPV-eradication settings [51,55,56,60], whereas the studies showing no impact 231 from IPV examined shedding in the youngest cohort studied in an OPV-using setting [23] or an 232 established IPV-only setting [58]. These data are consistent with the hypothesis that IPV-only 233 vaccination has no impact on shedding duration [21], in agreement with molecular evidence that IPV 234 produces no mucosal immunity in the absence of prior exposure to live poliovirus [66]. We discussed 235 IPV boosting in the dose response section. OPV-equivalent humoral antibody titer Heterotypic immunity provided by bOPV against mOPV2 challenge. In preparation for 237 the recent global switch from tOPV to bOPV in routine immunization and the need to understand 238 how the switch could impact type 2 immunity, recent studies have examined the heterotypic 239 immunity against mOPV2 challenge provoked by RI schedules containing bOPV and possibly one or 240 two doses of IPV [23,24]. In settings where primary risks associated with OPV vaccination outweigh 241 transmission risks, current RI regimens use at least one dose of IPV followed by at least one dose of 242 bOPV, and in settings where type 1 transmission risk is of high concern, three doses of bOPV are 243 recommended with at least one dose of IPV concurrent with later doses of bOPV. OPV2-equivalent humoral antibody titer . Effects of pre-challenge bOPV vaccination on shedding duration and inferred homotypic OPV-equivalent pre-challenge antibody titers against type 2 poliovirus. Median shedding durations derived from trial data (A) decrease and modeled pre-challenge homotypic OPV-equivalent humoral antibody titers (B) are shown for RI regimens containing bOPV. All bOPV schedules provide statistically significant increases in immunity to type 2 relative to unvaccinated individuals, and the inferred immunity against type 2 shedding after recieving bOPV is roughly equivalent to receiving one dose of tOPV.
It is also interesting to note that while our immune correlate, OPV-equivalent humoral antibody 253 titer, is a statistical construct that is not intended to be directly measurable for vaccination regimens 254 in which IPV provides primary homotypic serologic immunity, our inferred median values for the 255 OPV-equivalent humoral antibody titer against Sabin  of trial arms challenged subjects with mOPV2 (mOPV1, n = 5; mOPV2, n = 11; mOPV3, n = 5). 267 Data exploration revealed no systematic differences in viral load by serotype ( Fig. S3 and online). 268 We are not aware of similar data for WPV shedding.

269
Immunologically-naive individuals. aggregated data for immunologically-naive individuals (Fig. S4). Viral loads for all immunity levels 289 were well-fit by the product of the immunologically-naive temporal profile and the peak   Routine immunization with IPV only. We found no significant differences in fecal 300 concentration between seronegative children and IPV-only children when looking across trials.

301
However, one study in Cuba that did not meet our inclusion criteria because only one sample was 302 collected per subject reported that IPV in RI reduced fecal concentration by a factor of 3 one week 303 after OPV challenge [70]. That difference is small relative to the factor of 10 difference following   [48,51,53,55] that measured the probability of shedding after exposure from doses delivered 313 in oral droplets ranging from 10 to 10 6 TCID50 and clearly described the pre-challenge immune 314 histories of their subjects. Three studies challenged with Sabin 1, none used Sabin 3, and one 315 unusual human-passage study challenged with Sabin 2 and type 2 poliovirus derived from Sabin 2 316 after five days of replication in children [48]. We also included data from studies that only tested The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint samples per dose; we examined further evidence for differences in infectivity by serotype and vaccine 321 or wild-type in later sections on transmission-informed dose response models.

322
Effects of pre-challenge immunity on dose response. Healthy individuals with no immunity 323 against infection are susceptible to oral OPV doses of roughly 10 TCID50 or greater (Fig. 7A-B).  parsimonious description of all the OPV challenge data was provided by:   TCID50. (D) Dose response model fit to subjects with no immunity against infection ("unvaccinated" or "IPVx3" with no history of live poliovirus exposure). These individual-level data reveal no significant differences by serotype, but household transmission data provide additional evidence to resolve serotype differences ( In contrast, the data for dose response studies repeat the pattern from shedding duration that 353 IPV-only vaccination produces no immunity against infection. As mentioned in our analysis of 354 shedding duration, the distinction between having no prior experience with live poliovirus and IPV 355 boosting likely explains the studies that show some impact of IPV on intestinal infection. In the 356 dose response data, the two trial arms we reviewed with IPV booster doses after one year of age 357 show reduced susceptibility [51,55] (Fig. 7A). Both studies took place in tOPV-using communities, 358 one reported direct evidence of transmission-acquired infection in the cohort prior to the 359 booster [51], and shedding durations for these trial arms were also reduced, consistent with the 360 influence of live poliovirus exposure (Fig. 3).

361
Waning immunity against infection 362 By examining the OPV-equivalent antibody titers inferred from shedding duration or from dose 363 response in studies with subjects of many ages, we built a composite picture of waning immunity 364 against infection. To estimate waning rates, we considered data for individuals that were likely 365 maximally immune after their last poliovirus exposure prior to OPV challenge, either due to 366 immunization with 3 or more doses of tOPV [24,25,55] or natural immunity [46,59], and for which 367 the interval between the last immunization and OPV challenge was known or could be reasonably  The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint last immunization and mOPV challenge. We fit a power law to the OPV-equivalent antibody titers, 371 where t is measured in months between last immunization and mOPV challenge, N Ab (1) is the 372 baseline immunity one month post-immunization, and we found the exponent to be people differs from the interpretation given in Abbink et al [59] in which the lack of correlation 380 between the speed of serologic immune response and shedding were used to argue that memory 381 immunity provides no protection from shedding. Their data support the hypothesis that 382 post-challenge memory response does not discriminate differences in shedding, but they did not have 383 a control group of never-exposed subjects to compare deeply waned immunity with true naive 384 immunity. As seen through metastudy, the observed shedding durations in the seronegative elderly 385 are reduced relative to the shedding durations in unvaccinated and seronegative children, and are 386 compatible with the hypothesis that previously-exposed elderly people retain waned but persistent 387 immunity against infection, as has been suspected previously [28,74]. inferior at all ages that induced by multiple doses of tOPV, even decades after immunization.  The relevant results of the study are summarized here.

420
The majority of index children had prior serological immunity either due to maternal antibodies 421 or prior IPV vaccination, but none had any evidence for prior exposure to live poliovirus. The 422 authors reported that they found no statistically significant differences in poliovirus fecal shedding 423 between IPV recipients and unvaccinated children. This paper follows their lead and treats all index 424 subjects as a single cohort, although a small reduction in shedding in older children may be apparent 425 shed significantly more than those who received either mOPV1 or mOPV3, and shedding was similar 430 for mOPV1 and mOPV3 (mean prevalence over 5 weeks: type 1 vs. type 2 p < 0.001; type 1 vs. . Type 2 transmitted at highest intensity, both because of elevated shedding in infants relative to types 1 and 3, and due to the higher infectivity of type 2 ( Table 1).
Siblings of index children were not directly vaccinated, but became infected with Sabin strains 433 21/53 . CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint via transmission. Shedding due to transmission is significantly higher in siblings under 5 years of age 434 than in children ages 5 to 9 for all serotypes (mean prevalence by age: type 1 p < 0.001; type 2 435 p < 0.001; type 3 p = 0.002). All model results and statistical comparisons in this paper for siblings 436 and contacts are based on the age under 5 years cohort (see Supplement for discussion of the 437 more-detailed age breakdown presented in the original paper). Shedding rates were very low in 438 parents and children age 10 years and older (< 2%) [33], and so it is likely the transmission was 439 direct from index child to sibling and was not mediated by infected caretakers. Shedding due to 440 transmission-acquired type 2 was significantly more common than for types 1 and 3, and shedding 441 due to transmission was similar for types 1 and 3 (mean prevalence: type 1 vs type 2 p = 0.002; type 442 1 vs type 3 p = 0.33).

443
Primary extrafamilial contacts of siblings exhibited a similar pattern of increased type 2 shedding 444 and comparable type 1 and 3 shedding (type 1 vs type 2 p < 0.001; type 1 vs type 3 p = 0.73).

445
Although the authors did not describe the relationships between siblings and extrafamilial contacts 446 in detail, it is likely that the contacts were close friends of the siblings and were directly infected by 447 the siblings, as the authors also describe a smaller set of more socially-distant "secondary 448 extrafamilial contacts" who "were drawn from the neighborhoods or schools attended by the siblings" 449 and who were infected at lower rates than the primary contacts [33]. No detailed demographic 450 decomposition by trial arm was reported for the contacts, and so we assumed the contact and sibling 451 demographics were the same, and estimated shedding fractions by age cohort for the contacts were 452 derived as described in the supplement. an extrafamilial contact is similarly exposed to sibling fecal matter, which in turn doses them with 469 poliovirus in proportion to the probability the sibling is shedding at its viral concentration. The free 470 parameters specific to the transmission study were the pre-challenge immunity of each subject type, 471 the daily fecal dose (micrograms of stool) between infants and siblings, the daily dose between 472 siblings and contacts, and the mOPV take rates for each serotype. Transmission-informed dose response model parameters for OPV. When we assumed 481 that there were no serotype-specific differences in dose response (Eq. (5)), we found that the 482 estimated fecal exposure was highest for type 2, followed by type 1 and then type 3. Rather than 483 attributing serotype differences in the daily probability of transmission to differences in stool 484 exposure, it is more reasonable to assume that fecal exposure did not vary by serotype but dose 485 response does. Given equivalent fecal exposure across arms and assuming that our previous results 486 are most appropriate for Sabin 1 (Fig. 7), we found that the maximum likelihood estimates of the β 487 parameters in the dose response model in Eq. (5) are β S1 = 14 (3,59)    and all WPV serotypes were present in roughly equal amounts, and so we cannot examine 513 differences in WPV infectivity by serotype.

514
With all the dose response and transmission data examined, we summarized the differences in 515 infectivity between the Sabin strains and wild poliovirus by the estimated oral dose that would infect 516 50% of people with no pre-challenge immunity (HID50). We found that it is likely that WPV is 517 between three and twelve times more infectious than Sabin 2, which is in turn roughly twice as 518 infectious as Sabin 1 and Sabin 3 (Table 1).  shedding duration, we inferred that the median OPV-equivalent antibody titer of contacts with 6+ 539 doses is N Ab = 512, similar to what would be expected after two or three doses in clinical trial 540 settings (Fig. 2). Under our model for shedding duration after WPV infection, the observation that 541 50% of contacts reporting zero to two doses of prior OPV were positive for WPV in stool in the 10 542 weeks after the onset of paralyis in the index cases (which occurs two to five weeks after initial 543 infection [76]) implies that essentially 100% must have been infected and shed for the maximal 544 duration (Fig. 1). This inference follows from the structure of the model: most of the incidence in 545 contacts occurs before the onset of paralysis, and the median WPV shedding duration of 43 days in 546 naive individuals indicates that many contacts clear their infections before stool collection. Thus, we 547 inferred that the contacts with 0-2 reported tOPV doses had no functional immunity (N Ab = 1), 548 reflecting the poor efficacy of OPV in northern India during the time of the study [77]. To explain 549 the 12% rate of positivity averaged over the fifteen weeks after index case infection in infected young children transmit to household contacts who may in turn transmit to their intimate 558 social contacts outside the home (Fig. 10A). Our model of close-contact transmission within 559 households and between close extrafamilial contacts allowed us to estimate parameters that are 560 closely tied to detailed transmission study data, but its description of transmission intensity in terms 561 of fecal-oral exposure does not easily compare to more common models of disease transmission. The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint young children (Fig. 10B). We defined the local reproduction number as where p hh is the total probability that an index child transmits in one household transmits through 566 an older sibling contact to an extrafamilial contact in another household, and N h is the number of 567 intimate extra-familial contacts (and is thus a measure of the number of contact households 568 exposed). The dependence on immunity and daily fecal-oral exposure levels in p hh is calculated from 569 the detailed model of household to extrafamilial contact transmission described above (see America childhood social networks [78]. For an upper bound in Uttar Pradesh and Bihar, we 577 assumed N h = 10 based on scaling the typical USA social network size in proportion to the two-fold 578 larger typical classroom sizes in northern India [79,80]. Colored bars indicate range of immunity provided by IPV-only immunization (red), one dose of tOPV or heterotypic type 2 immunity from bOPVx3 and zero or more doses of IPV (blue), and homotypic immunity from three or more doses of OPV or IPV boosting on previous OPV (green). Gradient indicates approximate waning one, two through five, and more than five years since last immunization. (B) Relative transmission probability from an index case to an intimate contact vs. shedding index for immunized members of the pair in a setting with moderate fecal-oral transmission (daily fecal-oral exposure 46 µg per day). Red, both members have same OPV-equivalent immunity; green, index immunized and contact naive; blue, index naive and contact immunized. (C) Relative transmission probability from an index case to an intimate contact vs. shedding index for immunized members of the pair in a setting with very high fecal-oral transmission (daily fecal-oral exposure 340 mg per day). In all cases, the transmission probability is sub-linear with declining shedding index. In very high transmission settings, substantial immunity in both index and contact is required to appreciably reduce transmission.
Protection from shedding. Our results consolidate data and discussions collected over decades 595 into one figure (Fig. 11A). Relative to unvaccinated individuals, homotypic immunity in children 596 who have been fully immunized with OPV reduces shedding by roughly three orders of magnitude. 597 Heterotypic immunity from bOPVx3 provides a roughly one order of magnitude reduction in The fast waning removes homotypic protection equivalent to roughly one dose of OPV, and so the 611 multi-dose schedules in routine immunization ensure that adults who were fully immunized in 612 childhood remain well-protected throughout their child-bearing years, as is compatible the 613 observations in detailed above from Houston [33] and elsewhere [62]. The limited data on heterotypic 614 type 2 waning after bOPV vaccination is consistent with similar waning dynamics, and so it is 615 reasonable to expect that the protection against type 2 shedding provided by bOPV wanes to 616 neglible levels within a few years of vaccination.

617
Variations in OPV efficacy. Throughout this paper, we have assumed that three doses of tOPV 618 is sufficient to provide maximal immunity. However, OPV take [30, 72,73] and immunogenicity [77]  we found that the OPV-equivalent immunity in individuals who report receiving six or more doses of 624 tOPV had immunity equivalent to two to three doses of tOPV in healthy trial subjects. Low efficacy 625 can manifest itself as failures of OPV take or reduced antibody titer responses [82]. Because our  The impacts of pre-exposure immunity on person-to-person transmission 632 In Fig. 11B&C, we show how pre-exposure immunity against infection, manifested as reductions in 633 shedding index, reduce the probability of person-to-person transmission between pairs of intimate 634 contacts. In settings like Houston 1960 with moderate fecal-oral transmission (estimated daily 635 fecal-oral exposures less than one milligram per day), small changes in pre-exposure immunity 636 significantly reduce transmission. When both the index shedder and the exposed contact have 637 similar immunity, the transmission probability declines linearly with shedding index over the first 638 two orders of magnitude. When the pair has heterogeneous immunity in which one is immunized 639 while the other is not, transmission is more common when the shedder is naive rather than the 640 recipient. In these settings, transmission is driven by immunologically-naive individuals, and while 641 re-infection of previously immunized people may be common, the re-infected individuals play a 642 smaller role in transmission to unimmunized contacts (and to each other) than would be expected 643 from prevalence alone.  surveillance data in Uttar Pradesh and Bihar, there is a 65% chance that a WPV infection in a 661 re-infected child in one household will propagate to re-infect a child in another household. Thus, if 662 the original child has unvaccinated intimate contacts in two or more households, then local R Eff > 1 663 (Eq. (7)) and WPV transmission is likely to sustain itself (as was observed during the period of the 664 2003-2008 study our model is calibrated to). If the same WPV exposure occurs shortly after an 665 mOPV campaign in this low efficacy setting, the probability of transmission along this contact chain 666 would be reduced to 33% due to the combined effects of mOPV vaccination on shedding and 667 acquisition in all children. However, if the WPV exposure occurs shortly after an IPV campaign, 668 transmission along this contact chain would be reduced to 19% solely due to the substantially higher 669 effectiveness of IPV to boost immunity against re-infection in the previously immunized children. The totality of evidence for predicting how Sabin 2 transmission potential will change after OPV 673 cessation is summarized in Fig. 13. In the tOPV era, wherever tOPV was delivered in sufficient 674 quantities to achieve durable immunity against infection, transmission of Sabin 2 was rare even 675 among intimate contacts in settings with very poor sanitation and high population densities.

676
At the time of writing, the world is six months into the bOPV era, in which infants who have 677 received bOPV in routine immunization are surrounded by older tOPV-immunized household 678 contacts (Fig. 13 B,F,J). While the shift to bOPV has increased the transmission potential of Sabin 679 2, well-vaccinated communities are likely still protected from Sabin 2 transmission-local R eff 1 at 680 all levels of fecal-oral exposure, and so the community reproduction number almost certainly remains 681 below one.

682
However, many households will soon have multiple children who receive only bOPV (and possibly 683 one or more doses of IPV), and so the ability of Sabin 2 to transmit in poor sanitation settings will 684 be very different (Fig. 13 C,G,I). Because of the substantially weaker immunity against type 2 685 infection provided by bOPV, re-introduced Sabin 2 will eventually transmit through well-vaccinated 686 communities with inadequate sanitation as if they had not been vaccinated at all (Fig. 13 D,H,L). In 687 moderate transmission settings, the protection from bOPV will still be substantial, but only because 688 Sabin 2 local R eff 1 even in the absence of immunity in children.  reproduction numbers depend on poliovirus infectiousness, ranging from Sabin 3 to WPV (Fig. 14). In moderate transmission settings, both reversion and low immunity are essential for cVDPV emergence to occur with non-negligible probability when Sabin virus is introduced to a population. In contrast, in very high transmission settings, cVDPV emergence is only prevented by widespread high immunity throughout the child population. Genetic reversion is of minor importance in comparison to population immunity and cVDPV would likely be common in low immunity settings.
Israel. The crucial quantitive differences in transmission between the Sabin strains and WPV Low transmission settings. At daily stool exposures below 1 to 10 micrograms of stool per day 720 (roughly ten-fold less than we estimated for low-SES families in Houston in 1960), our model shows 721 that the fecal-oral route cannot sustain poliovirus transmission, wild or vaccine. This observation 722 supports the long-held hypothesis that oral-oral transmission is important in settings with high 723 socioeconomic status and corresponding good sanitation, supported by many observations that IPV 724 alone-an effective blocker of oral shedding [64,85]-can block transmission and prevent outbreaks 725 from importation in middle-and high-SES communities [8,85]. 726 High transmission settings. In contrast, our model indicates that epidemic Sabin transmission 727 "out of the vial" in settings like Uttar Pradesh and Bihar in the 2000s can only be prevented by high 728 immunity against infection. High immunity was common prior to OPV cessation: we inferred from 729 reported estimates of shedding duration that the typical OPV-equivalent immunity in healthy 730 children who reported six or more doses of tOPV is roughly N Ab = 512. This conclusion of high 731 immunity despite high WPV risk, due to both comprehensive vaccination and endemic WPV 732 transmission, is supported by serosurveys as well [86]. With that immunity, we estimate local 733 R eff = 0.05 for Sabin 2. However, once families have more than one child who received only bOPV 734 immunization, because of the ability of both children to shed high quantities of poliovirus, we 735 estimate the local reproduction number of Sabin 2 will jump to R eff = 9.3, bounded by the number 736 of intimate contacts (assumed N h = 10) and not by immunity against infection. Demographic 737 transitions [87] will reduce local R eff roughly in proportion to declining birth rates, but local R eff 738 will remain above protective levels without significant reductions in fecal-oral exposure in currently 739 high-transmission settings. We have covered a great deal of information in pursuit of a single question: when all the evidence is 749 considered, should OPV cessation fundamentally change how we think about the Sabin polio 750 vaccine? We conclude that the answer is yes. The Sabin strains became the preferred live-attenuated 751 vaccine because of their low neurovirulence and high vaccine efficacy [4,88]. In pursuit of both goals, 752 we arrived at vaccine strains that are 1,000-10,000 times less neurovirulent than wild polioviruses [5], 753 but only five to ten times less infectious. The Sabin vaccine strains are fundamentally infectious 754 polioviruses with very low virulence-less capable of causing paralysis but not categorically different 755 from the naturally-occurring low-virulence strains once found in the wild [88]. However, like for the 756 naturally-occurring low-virulence strains, neurovirulence is not a stable phenotype. In the absence of 757 widespread population immunity, the difference between vaccine and wild virus is small in settings 758 where community transmission of the vaccine strains is possible.

759
The safety and effectiveness of tOPV in the pre-eradication era. Prior to the advent of national outbreak, as well as the estimated community reproduction number of WPV near 1 [15], 783 indicate that transmission intensity during the WPV1 outbreak was similar to that once common in 784 low-SES parts of the USA. In such circumstances, mass vaccination with OPV is necessary to 785 eliminate WPV transmission, but the Sabin strains are not capable of widespread transmission prior 786 to complete genetic reversion, and so OPV cessation can resume safely after outbreak interruption. 787 The instability of OPV cessation in low-income countries. In settings with low-SES and 788 accompanying inadequate sanitation, our synthesis of the evidence accumulated over decades makes 789 clear that the ability of the Sabin strains to transmit has been limited by population immunity and 790 not by the attenuated phenotype. In the absence of substantial population immunity, the Sabin 791 strains will be capable of widespread transmission "out of the vial" prior to any genetic reversion, 792 and so, with respect to transmission, the Sabin strains will behave like wild poliovirus. If Sabin OPV 793 has to be used in outbreak response once population immunity has fallen below historically 794 protective levels, synchronized vaccination campaigns with coverage sufficiently high to saturate 795 local contact networks can limit cVDPV risk within target populations. However, export into 796 unvaccinated populations will only be limited by the weakness of social ties that are difficult to 797 characterize or control. WPV outbreaks demonstrate that poliovirus is able to rapidly traverse the 798 world despite immunity in the majority of people [8,90,91]. If mOPV used in outbreak response 799 campaigns is found to seed more cVDPV outbreaks than can be prevented with further campaigns, 800 then OPV will need to be re-introduced to routine immunization to maintain the pre-cessation 801 successes of the polio eradication program-successes that have limited poliomyelitis to only 802 hundreds of paralytic cases per year [2,5], down from hundreds of thousands [4].

803
Timeline for cascading cVDPV2 outbreak risk. Global Sabin 2 cessation began in April 804 2016, when it was believed that all persistent cVDPV2 transmission had ceased [7]. At the same 805 time, a cVDPV2 lineage thought eliminated was detected in environmental surveillance in Borno, 806 Nigeria, and it is continuing to circulate at the time of writing in September 2016 [7,92]. In response, 807 37/53 . CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint mOPV2 campaigns are being conducted in the Lake Chad region, but due to the impacts of violent 808 conflict on surveillance and vaccination, there is currently no timeline for the expected interruption 809 of the cVDPV2.

810
Six months after tOPV cessation, the conditions supporting Sabin 2 transmission have likely not 811 changed substantially. Type 2 immunity throughout sub-Saharan Africa was rapidly increased in 812 preparation for tOPV cessation [7,93,94]. Almost all households contain only at most one child born 813 after cessation, and our model of transmission among intimate contacts predicts that one 814 unimmunized child is insufficient to support local epidemic transmission. At this time, focal mOPV2 815 campaigns are not likely to seed cVDPV2 outside of the target population. However, our models 816 show that once families have more than one child born after tOPV cessation, the potential for Sabin 817 2 transmission will increase substantially to near WPV levels. The median birth spacing in most 818 currently-OPV-using countries is between 24 and 36 months [95]. Thus, we predict that in 819 2018-2019, the risk of establishing cVDPV2 in the many regions of the developing world that have 820 not received mOPV2 campaigns will increase substantially. The meager cross-immunity of bOPV 821 (and one or more doses of IPV) against type 2 does not alter this conclusion.  [97], twelve independent VDPV lineages were 828 detected in the following one to three years in association with the restoration of tOPV campaigns, 829 including the largest known outbreak of cVDPV2 in history [98].

830
The importance of IPV. We echo the recommendation of the WHO Strategic Advisory Group 831 of Experts on Immunization to introduce IPV into global routine immunization [7,99]. While IPV in 832 RI will have little direct impact on transmission in settings with poor sanitation, IPV boosting 833 campaigns will remain effective interventions against transmission until a few years after cessation 834 when a lack of primary immunity in young children will dominate transmission everywhere.

835
Expanded IPV coverage will prevent paralysis from polio outbreaks and OPV use. Long-term, if 836 poliovirus transmission cannot be interrupted globally without using OPV in RI, then IPV followed 837 by OPV provides superior protection against paralysis and all the immunological benefits of OPV 838

38/53
. CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint vaccination without the associated risks. In all scenarios, in a population well-immunized with IPV, 839 live poliovirus re-introduction does not pose substantial risks to human health.

840
Conflicting with this recommendation is a severe IPV shortage. Despite coordinated global effort 841 to introduce at least one dose of IPV everywhere, currently 50 countries have either not done so or 842 are experiencing stockouts that will not be replenished until at least the end of 2017 [7]. Our Given the evolving landscape of cVDPV2 risk, the IPV shortage should be viewed as a global public 851 health emergency, and all efforts to achieve adequate supply must be accelerated. These efforts 852 should include attempting to procure donations and shift allocations from middle-and high-income 853 countries that currently give four or more doses of IPV in RI.

854
Sufficient OPV manufacturing capacity must be maintained until OPV cessation is 855 secure. Even in the absence of IPV supply, we reiterate that resuming OPV vaccination is vastly 856 preferable to the widespread resurgence of WPV or cVDPV transmission. It is thus imperative that 857 sufficient Sabin manufacturing capability be maintained until we can be confident that all polio 858 transmission has ceased or that Sabin vaccine is no longer necessary. This recommendation is in 859 conflict with the limitations that the GAPIII policy on poliovirus containment places on vaccine 860 manufacturers in low-income countries [100], and so it is critical that the GPEI coordinate with 861 vaccine manufacturers now to insure that adequate supply for OPV re-introduction in RI will be 862 available if necessary. One creative route to securing sustained OPV supply capacity would be to 863 develop contigency plans to repurpose Sabin IPV supplies into OPV should the need arise [101]. potential of WPV transmission, we hope to emphasize that the conclusion that cVDPV risk will rise 885 well above historical norms within a few years of OPV cessation follows from the deeply understood 886 epidemiology of polio and not the particular assumptions of our model. It follows from the simple 887 facts that the Sabin strains are infectious polioviruses by design and that doses acquired via 888 fecal-oral exposure can be much higher in the poorest parts of the developing world than they were 889 in the countries where Sabin OPV was first studied and where OPV cessation has already been 890 successful. OPV cessation must be secured quickly for OPV re-introduction to be avoided. If OPV 891 re-introduction is required, securing IPV supply is critical for achieving the dual goals of maintaining 892 population immunity against transmission without risks of paralysis. Given the costs and complexity 893 of sustaining dual vaccination with IPV and OPV, both financial and regulatory support for new 894 polio vaccines must be sustained until viable candidates are found to guarantee the stability of polio 895 eradication for the indefinite future.   Shedding duration after OPV challenge Figure S1 shows the reverse cumulative shedding duration distributions that describe estimates of the probability an individual is still shedding after successful OPV take. Each curve represents the sample-size weighted average of the curves from the individual studies; disaggregated data is provided in the supplementary data files. All original data were presented as estimates of prevalence over time, sampled on discrete days that often differ across studies, and variations in sample size due to missing data or dropout were often impossible to reconstruct. These distributions are thus not proper Kaplan-Meier estimates of the survival functions, and due to different censoring patterns across studies, the average curves are not guaranteed to decrease monotonically, although deviations from monotonicity are rare and only found with small total sample sizes.  Concentration of poliovirus in stool after OPV challenge Figure S3 shows the geometric mean poliovirus concentration in stool (TCID50/g) for all included trial arms.
Seronegative, unvaccinated, IPVx2, and IPVx3 data also appear in Figure 5, and all type 2 data appear age-adjusted to 12 months in Figure 6.  Figure S4 shows the average concentration, age-adjusted to 120 months for all immunologically-naive trial arms (seronegative, unvaccinated, IPVx2, and IPVx3) and the best fit temporal profile described in equation (4). Determining the interval between last immunization and OPV challenge to assess waning immunity against infection For individuals from tOPVx3 vaccine trials, intervals between last immunization and mOPV challenge ranged from 1 month [21] to 6 months [11]. To assess waning of tOPV-based immunity in older children, one study in Uttar Pradesh compared mOPV vaccine take rates in children 1, 5, or 10 years of age [19] who had previously recieved an unknown but high number of tOPV doses. To estimate the likely interval between last immunization and challenge, we assumed that children are offered up to 5 doses in the first year of life (3 RI plus 5 campaigns at 60% coverage), corresponding to roughly 2.5 months on average between last vaccination and mOPV challenge at 1 year of age. We assumed campaigns delivered 3 doses per year in ages two through four, corresponding to roughly 4 months between last vaccination and challenge at 5 years of age, and no doses after 5 years of age, corresponding to 5 years since last vaccination and challenge at 10 years of age. For this study, OPV-equivalent immunity was inferred via vaccine take rates using equation (5). Data on adult shedding after natural immunity were taken from studies in the Netherlands. From the study by Verlinde et al [1] in 1959, the average seropositive subject in the study was 20 years of age, and we assumed that their last infection was 5 years earlier at 15 years of age when maximum seropositivity was first achieved in the population. From the study by Abbink et al [15] from 2005 that measured shedding in elderly individuals upon mOPV challenge, we assumed last exposure was 45 years earlier in 1960, at roughly the year in which widespread endemic transmission ceased in the Netherlands. We included data for both seropositive and seronegative adults from the Abbink et al study because seronegative adults showed evidence of memory immunity and reduced shedding durations in comparison to immunologically-naive children.
Houston 1960: detailed exploration of shedding fraction by age As shown in Fig. S5, older index children shed slightly less after mOPV challenge than younger children for types 2 and 3 (type 1 p = 0.105; type 2 p = 0.016; type 3 p = 0.025). The reduction may be influenced by three effects. One possibility is a small fraction of older children experienced natural exposure prior to the start of the study and so have more immunity than suggested by their lack of vaccination history. We believe this is unlikely as it is unlikely wild polio transmission would affect a small fraction of a geographically, demographically, and immunologically similar cohort. A second possibility is a small influence from prior IPV 8/16 . CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint immunization. As described in Table 1 of Benyesh-Melnick et al [8], older index children were more likely to have received at least one dose of IPV. However, it should be noted that the original authors who had access to the individual-level data reported that they found no significant differences between IPV and unvaccinated index subjects, as is compatible with our metastudy. A third possibility is that stool concentrations of poliovirus are higher in young infants who are not yet eating solid foods and may not have mature immune systems, and so stool culture may be more sensitive to shedding in younger children. This third possibility is suggested by the observations summarized in equation (2) that shedding in children under 8 months of age shed systematically higher viral concentrations than comparable children 15 to 18 months of age [6,11].  Figure S5. Fraction shedding by cohort and age range as originally reported. Observed fraction shedding and estimated 95% binomial confidence interval for each serotype, subject type, and reported age cohort.
There were no statistically significant differences in shedding among the age groups under 12 months, 12 to 23 months, 24 to 35 months, or 36 to 59 months for any serotype. However, there was significantly less shedding in the 60 to 107 month age group relative to the 36 to 59 age group (p < 0.001 for all serotypes).

9/16
. CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint As stated in the main text, shedding in siblings age 60 to 107 months (5 to 9 years) is significantly below that of ages less than 5 years for all serotypes (type 1 p < 0.001; type 2 p < 0.001; type 3 p = 0.002).
No breakdown by age was presented by Benyesh-Melnick et al [8] for the extrafamilial contacts of the siblings. However, because the contacts are demographically similar to the siblings and age is a significant factor for poliovirus acquisition via transmission in this setting, we used age-adjusted shedding rates in this paper. To estimate the unreported shedding fraction in the age under 5 contact cohort, we adjusted the total reported shedding counts for each serotype as follows: (estimated contacts shedding under 5) = (total contacts shedding) × (fraction siblings shedding under 5) (estimated contacts under 5) = (total contacts) × (fraction siblings under 5) .
The estimated counts were rounded to the nearest integer and confidence intervals presented are based on the rounded estimated counts. The estimated age 5 to 9 years contact data was constructed similarly.

Index-sibling-extrafamilial contact transmission model
The index-sibling-extrafamilial contact model was implemented in Matlab (supplementary file primarySecondaryTertiaryDoseModel.m) and all parameters are given in Table S1. For each of the three individuals along the transmission chain, based on specified age and pre-exposure immunity, the model calculates daily incidence (the probability of becoming infected each day), prevalence (the probability of shedding poliovirus in stool each day), and quantity shed (TCID50 per gram).
All infections in index (primary) children begin on day t = 1 due to exposure on day t = 0. For our analysis of the Houston Sabin transmission study, infection started with vaccination with dose = 10 6 TCID50. The incidence due to vaccination is given by where p Sx is the study-specific mOPVx (x = 1, 2, 3) vaccine take for N Ab = 1 and the second term is given by the dose response model in equation (5). For our analyses of WPV surveillance around index cases, P index (infection) = 1 by definition. In index cases, the prevalence after vaccination is thus P index (shedding at t) = P index (infected at t = 1) P shedding at t N Ab,index ; infected at t = 1 , 10/16 . CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint where the first term is from eq. (S1) and the second is the shedding duration model in equation (1).
Incidence in siblings (secondary) derives from exposure to index (primary) shedding as: P sibling (infected at t) = P sibling transmission at t index shedding P index shedding at t N Ab,index ; infected at t = 1 (S3) with P sibling transmission at t index shedding = β(t) t−1 t =1 (1 − β(t )) β(t) = P infection dose at t, N Ab,sibling (S4) where T is is the daily fecal-oral exposure between index and sibling per day (grams of stool), index fecal concentration (TCID50 per gram) is determined by the fecal concentration model in equation (4), β(t) is the infection probability determined by the dose response model with sibling pre-exposure immunity, and P sibling transmission at t index shedding is the probability the sibling is infected on day t given contact with a shedding index child.
Sibling (secondary) prevalence follows from convolving the daily incidence with the shedding duration distribution: P sibling (shedding at t) = tc t =1 P sibling (infected at t ) P shedding at (t − t ) N Ab,sibling ; infected at t . (S5) The cutoff time is for computational convenience; t c = 35 days for Houston and t c = 100 days for Louisiana, and Uttar Pradesh and Bihar.
Extrafamilial contact (tertiary) shedding derives from exposure to sibling (secondary) shedding, after summing over all days at which the sibling may have been infected, as: P contact transmission at t sibling shedding since t × P sibling (infected at t ) P shedding at (t − t ) N Ab,sibling ; infected at t , (S6)

11/16
. CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. . https://doi.org/10.1101/084012 doi: bioRxiv preprint with P contact transmission at t sibling shedding since t = β(t) β(t) = P infection dose at t, N Ab,sibling (S7) (dose at t) = T sc × sibling concentration (t − t ) N Ab,sibling ; sibling age , where T sc is the daily fecal-oral exposure between sibling and extrafamilial contact per day, and (t − t ) is the interval since the sibling became infected. This convolution over sibling incidence accounts for the dependence of the daily dose received by extrafamilial contacts on the start of sibling infection. Extrafamilial contact prevalence follows from incidence in the same manner as for the sibling: P contact (shedding at t) = tc t =1 P contact (infected at t ) P shedding at (t − t ) N Ab,contact ; infected at t .

Local reproduction number
We introduced the local reproduction number to summarize the transmission among clusters of households, based on the index-sibling-contact model described above. In the main text, we define the local reproduction number in equation (7) as: where p hh is the total probability that an index child transmits in one household transmits through its older sibling contacts to an extrafamilial contact in another household, and N h is the number of intimate extra-familial contacts (and is thus a measure of the number of households exposed). In terms of index to extrafamilial contact transmission, we define p hh as the cumulative incidence in an extrafamilial contact given infection in an index case: with t c = 100 days.

12/16
. CC-BY-ND 4.0 International license certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (which was not this version posted October 27, 2016. While there is no "bOPV" in use for Sabin 1 and 3, plots with those immune states are included to show how transmission varies by serotype for equivalent incomplete immunity. Differences in transmissibility by serotype and attenuation are due to differences in infectivity (Table 1). In Houston in 1960, WPV was endemic in a pre-OPV population. In settings with higher fecal-oral exposure and larger intimate contact networks, similar and greater levels of transmissibility will be common for all Sabin strains after OPV cessation. In the absence of immunity against infection, the virological differences in transmissiblity between the Sabin strains and WPV are small in comparison to the structural differences between settings.