Threshold protective levels of serum IgG to Shigella lipopolysaccharide: re-analysis of Shigella vaccine trials data

Objectives Establishing a correlate of protection is essential for the development and licensure of Shigella vaccines. We examined potential threshold levels of serum IgG to Shigella lipopolysaccharide (LPS) that could predict protection against shigellosis. Methods We performed new analyses of serologic and vaccine efficacy (VE) data from two randomized vaccine-controlled trials of the Shigella sonnei–Pseudomonas aeruginosa recombinant exoprotein A (rEPA) conjugate conducted in young adults and children aged 1–4 years in Israel. Adults received either S. sonnei–rEPA (n = 183) or control vaccines (n = 277). Children received the S. sonnei–rEPA conjugate (n = 1384) or S. flexneri 2a–rEPA conjugate (n = 1315). VE against culture-proven shigellosis was determined. Sera were tested for IgG anti–S. sonnei LPS antibodies. We assessed the association of various levels of IgG anti–S. sonnei LPS antibodies with S. sonnei shigellosis risk using logistic regression models and the reverse cumulative distribution of IgG levels. Results Among adults, four vaccinees and 23 controls developed S. sonnei shigellosis; the VE was 74% (95% CI, 28–100%). A threshold of ≥1:1600 IgG anti–S. sonnei LPS titre was associated with a reduced risk of S. sonnei shigellosis and a predicted VE of 73.6% (95% CI, 65–80%). The IgG anti–S. sonnei LPS correlated with serum bactericidal titres. In children, a population-based level of 4.5 ELISA Units (EU) corresponding to 1:1072 titre, predicted VE of 63%, versus 71% observed VE in children aged 3–4 years. The predicted VE in children aged 2–4 years was 49%, consistent with the 52% observed VE. Conclusion Serum IgG anti–S. sonnei LPS threshold levels can predict the degree of VE and can be used for the evaluation of new vaccine candidates.


Introduction
Shigella is a leading cause of diarrhoea and is associated with a substantial burden globally. In the low-and middle-income countries, >250 million cases of shigellosis and >212 000 deaths are estimated to occur annually, mostly in children [1]. In these settings, Shigella flexneri accounts for most cases [2], whereas in the high-income countries, Shigella sonnei is the most common [1,3].
Shigella infection confers, although of limited duration, serotype-specific immunity, pointing to the O-specific polysaccharide (O-SP) as the protective antigen [3e5].
No licensed vaccine against Shigella is currently available, but several vaccine candidates have been tested in clinical trials in the last decades, including O-SPebased conjugate vaccines [6]. The first generation of Shigella conjugate vaccines was developed at the U.S. National Institutes of Health and evaluated in double-blind randomized vaccine-controlled trials in Israel [7,8]. A randomized clinical trial (RCT) of the S. sonneierEPA conjugate among military recruits [7] showed significant protection against S. sonnei shigellosis; the vaccine efficacy (VE) was 74% (p 0.007). A subsequent RCT of S. sonnei and S. flexneri 2a O-SPerEPA conjugates among children aged 1e4 years showed 71% VE (p 0.043) of the S. sonneierEPA conjugate in the children aged 3e4 years but not in younger ones. Immunogenicity and efficacy were age-related and correlated [8].
The second generation of Shigella conjugates and other subunit vaccines are presently in clinical development [9e12].
The identification of an immunological correlate of protection for shigellosis is essential for the development and licensure of Shigella vaccines and can inform Go/No-Go decisions in the evaluation of these vaccines.
Accumulated evidence indicates that serum IgG antibodies against Shigella lipopolysaccharide (LPS) can be a mechanistic immunological correlate of protection [13,14].
Here, leveraging serology and efficacy data from the mentioned above RCTs, we evaluated whether defined serum IgG antieS. sonnei LPS levels induced by the S. sonneierEPA vaccine could predict the degree of protection against homologous Shigella infection.

Study design and population
New analyses of data from the RCT in young adults For a detailed description of the trial, please see supplementary material. We reanalysed the serological results of IgG antieS. sonnei LPS levels in vaccinees and controls of units AeC, in which S. sonnei shigellosis cases occurred 70e155 days post-vaccination [15].
The results were expressed as endpoint titres; geometric mean titres (GMTs) and !4-fold increases in titre post-vaccination over baseline were calculated [7]. We assessed the associations of various titres of IgG antieS. sonnei LPS as potential thresholds, with the incidence of culture-proven S. sonnei shigellosis and whether they could predict the VE of the S. sonneierEPA conjugate.
Serum bactericidal activity (SBA) was examined in 20 remaining archived sera of vaccinees on day 17 post-vaccination as described [9].
We estimated the potential threshold of serum IgG antieS. sonnei LPS that might be associated with S. sonnei shigellosis incidence using multiple approaches to verify the robustness of our findings, especially given the small number of shigellosis incident cases during the trial.
The primary analysis included bivariate and multivariable logistic regression models to assess the Prentice criteria for a valid immunological substitute endpoint [16] (please see supplementary material). The incidence of culture-proven S. sonnei shigellosis was the dependent variable. The independent variables comprised vaccination status (vaccine vs. control) and serum IgG levels expressed in titres (1:800, 1:1200, 1:1600, 1:1800, 1:2000, and 1:3200), or !4-fold rise in titre on day 17. We hypothesized that the serum IgG levels will blunt the association between vaccination and shigellosis, demonstrating that just antibodies were those that protect against the disease. The multivariable model also included an interaction term between vaccination and antibody levels and adjusted for baseline antibody titres as a putative confounder [17,18]. Firth's correction was used in the regression model [19] given the small number of shigellosis cases.
Sera on day 17 post-vaccination were obtained from 371 of 450 (82%) participants. Therefore, we employed the predictive mean matching multiple imputation method using the observed information on vaccination status and baseline serum IgG antieS. sonnei LPS levels to replace the missing values of day 17 antibody level. Multiple imputations were deemed accurate because the sample in this RCT was homogenous regarding age, sex, and risk of exposure.
Mantel-Haenszel analysis was employed to examine the effect of vaccination on disease in stratification by levels of antibodies !1:1600 or <1:1600, 17 days post-vaccination.
We also used the method developed by Siber et al. [20] to estimate the level of IgG titre that defines a threshold correlate of the protection model assuming perfect/complete versus zero protection against the disease [20] (S. sonnei shigellosis). We compared the VE estimated by the calculated immunological threshold (VE IM ) with the observed VE calculated using vaccination status only, to evaluate the accuracy of the threshold correlate of the protection model.
The VE IM formula is as follows: Consistency between VE IM and the observed VE supports the utility of using a simple threshold correlate of the protection model [20,21]. The Taylor series [21] was used to calculate the 95% CIs for the predicted VEs.
We also used a nonparametric method [22] to estimate IgG thresholds corresponding to any specified risk, including zero risk, to provide a 95% CI for the estimated threshold levels. This method accounts for missing day 17 serology data by inverse probability of observation weights and accounts for loss to follow-up of some participants.
Proportion of vaccinated individuals having a level of IgG anti À S:sonnei LPS below the threshold Proportion of unvaccinated individuals ðcontrolsÞ with a level of IgG anti À S:sonnei LPS below the threshold Â 100 Data were analysed using the SPSS Statistics, version 28 (IBM Corp., Armonk, NY, USA), SAS Software, version 9.4 (SAS Institute Inc., Cary, NC, USA), and R version 4.1.0 (The R Foundation for Statistical Computing, Vienna, Austria).

New analyses of data from the RCT in children
For a detailed description of the trial, please see supplementary material. We applied the approach reported by Siber et al. [20] as described above in data analysis of the paediatric RCT [8], in which serological results were available for 10% of randomly selected participants. The paucity of sera did not allow individual-level assessment of the association between IgG antieS. sonnei LPS and S. sonnei shigellosis. The VE IM of the S. sonneierEPA conjugate was compared with the observed VE by age group.

Commutability of IgG antieS. sonnei LPS measurement units in the two trials
In the RCT of young adults, IgG antieS. sonnei LPS levels were expressed as endpoint titres, whereas in the paediatric RCT, they were expressed in EU related to a standard prepared at the U.S. National Institutes of Health from convalescent sera of S. sonnei shigellosis patients assigned 100 EU. To attain commutability between the two units, the conversion of the endpoint titres to EU (standing for a percentage of standard) was performed (Figs. S1 and S2).

Results
Overall, 460 participants were randomized in units AeC in the young adults' trial, 183 and 277 in the S. sonneierEPA vaccine and control groups, respectively. Sera were obtained from 177 vaccinees and 273 controls on day 0 and from 146 vaccinees and 225 controls on day 17 (Fig. 1).
On day 17, the GMT of IgG antieS. sonnei LPS was significantly lower among participants who subsequently developed S. sonnei shigellosis than in those without shigellosis (Table 1).
A multivariable logistic regression model (complete-case analysis and using an imputed data set) that included vaccination status and IgG antieS. sonnei LPS levels as predictors and S. sonnei shigellosis as the dependent variable confirmed our hypothesis that antieS. sonnei LPS levels blunt associations between vaccination and shigellosis (Table S4), thus supporting the inference that the efficacy of S. sonnei conjugate is mediated via the serum IgG antieS. sonnei LPS. The results were similar when using a !4-fold increase in IgG antieS. sonnei LPS on day 17 as the independent variable (Table S5). The results of Mantel-Haenszel stratified analysis by day 17 IgG antieS. sonnei LPS levels were similar (Table S6).
The estimated probability of disease as a function of IgG antieS. sonnei LPS titres from logistic regression models showed good separation between participants with and without shigellosis at a titre of 1:1600 (Fig. S3).
Using the nonparametric method reported by Donovan et al. [22] to estimate IgG threshold levels corresponding to any specified risk, including zero risk, while accounting for missing data, showed

S. sonnei-rEPA vaccinees N=183
Total randomized in units A-C N=460

Controls N=277
Controls with sera on day 0 N=273

Controls with sera on day 17 N=225
Cases of S. sonnei shigellosis with sera on day 0 N=27 Cases of S. sonnei shigellosis with sera on days 0 and 17 N=16 Fig. 1. Flowchart of study participants.
that overall there was a decreasing risk of disease for all participants with IgG antieS. sonnei LPS increasing threshold levels (Fig. S4).

Reverse cumulative distribution of antibody levels approach
The reverse cumulative distribution of ln antibody levels on day 17 post-vaccination showed that the threshold of 7.36 (ln of endpoint titre 1:1600) was consistent with the 74% (95% CI, 28e100%) observed VE in this trial (Table S7 and Fig. S5).

Putative antibody threshold of protection in children
The VE of the S. sonneierEPA conjugate at 2 years postvaccination was 71.1% (95% CI, 4.43e92.0%) in children aged 3e4 years, but it was lower and not statistically significant in the age groups 2e3 years and 1e2 years. A pooled analysis of children aged 2e4 years yielded a VE of 51.6% (95% CI, 1.0e76.0%) ( Table S8). The VE IM could be assessed on the basis of the proportion of children vaccinated with the S. sonneierEPA conjugate and control vaccine who had IgG antieS. sonnei LPS levels below the different candidate thresholds. These results are shown for two age groups: 3e4 and 2e4 years (Table S9 and Fig. 3). The cutoff of 1.5 (ln of 4.5 EU) yielded a VE IM of 63% (95% CI, 42e76%), slightly lower than the 71% observed VE for children aged 3e4 years. The same cut-off predicted a VE IM of 49% (95% CI, 35e59%), consistent with the 52% observed VE for the children aged 2e4 years (Tables S8 and S9).

Discussion
The main findings of our study are the identified protective threshold levels of IgG antieS. sonnei LPS against S. sonnei shigellosis in young adults and children. In the RCT comprising adults, a threshold of 1:1600 (~6.6 EU) IgG antieS. sonnei LPS titre at day 17 post-vaccination reduced the risk of shigellosis and predicted an S. sonneierEPA VE IM of 73.6%, consistent with the observed 74% VE. A post-vaccination level of 4.5 EU of IgG antieS. sonnei showed 63% VE IM in children aged 3e4 years, slightly lower than the observed 71% S. sonneierEPA conjugate VE and VE IM of 49% similar to the observed 52% VE in children aged 2e4 years.
The findings from this re-analysis of data from S. sonneierEPA conjugate RCTs corroborate with results from our observational studies regarding the pivotal role of pre-existing serum IgG antieS. sonnei LPS in the protection against the homologous disease [23,24] and collectively highlight the usefulness of IgG anti-LPS as a correlate of protection [13] that can foster the development and licensure of Shigella vaccines.
In the RCT comprising young adults, where we used individuallevel data, confirming the Prentice criteria [16], we found that the full effect of the vaccine is mediated via IgG antieS. sonnei LPS by documenting that the protection by the S. sonneierEPA vaccine is annulled after adjusting for the day 17 post-vaccination IgG antieS. sonnei LPS in the logistic regression model. This was confirmed by the stratified Mantel-Haenszel analysis.
In children, we used the population-based predicted VE IM following the methodology employed by Siber et al. [20] to identify the minimal protective antibody levels that supported the licensure of new pneumococcal and meningococcal vaccines without conducting large efficacy trials [20,21].
The conversion of endpoint titres to EU enabled the identification of a 'dose-response' relationship between the antibody levels and protection levels across the two trials. The 6.6 EU threshold level (corresponding to 1600 endpoint titre) was measured 17 days post-vaccination in young adults and predicted 73.6% VE against culture-proven S. sonnei shigellosis occurring 71e155 days postvaccination. The 4.5 EU threshold level was identified at the population level in children aged 2e4 years at a median time of 18.5 weeks after vaccination, whereas more than half of the cases of S. sonnei shigellosis occurred in the second year of follow-up. At this time after vaccination, the 4.5 EU threshold yielded a VE IM of 49%, consistent with the observed VE in this merged age group (52%).
We found a strong significant correlation between IgG antieS. sonnei LPS of vaccinees and SBA. The antibody-mediated complement-dependent bacteriolysis was proposed as the most plausible mechanism by which IgG antieS. sonnei LPS can protect against Shigella infection [8,25,26]. ELISA for serum IgG antibodies against Shigella LPS is an easy and inexpensive assay which can be practically applied to large RCTs of candidate vaccines. The strong correlation between vaccine-induced IgG anti-LPS levels and SBA titres indicates that the functional activity of ELISA-measured binding antibodies can be inferred.
The main strength of our study is the utilization of data from the only two available S. sonneierEPA conjugate field efficacy RCTs to link between vaccination, vaccine-induced antibody level, and homologous shigellosis in the two studies and populations. The individual-level data in the RCT comprising adults enabled the assessment of a protective Shigella LPS antibody threshold using multiple analytical approaches that yielded consistent findings. Moreover, we were able to link between serum IgG anti-LPS and SBA, thus providing mechanistic insights for protection.
Our study has limitations, including the lack of day 17 postvaccination sera in 18% of the adult RCT volunteers and the availability of sera only for 10% of the participants in the paediatric RCT, the latter allowed only for using the reverse cumulative distribution analytical approach in children. Lastly, the number of S. sonnei shigellosis cases was small. To address these limitations, we used several analytical approaches, including Firth's correction, in logistic regression models and multiple imputations for missing values on day 17 post-vaccination sera, which yielded consistent results.
There is a need to harmonize internationally between the ELISA assays employed and generate an International Standard Serum. This will enable assay results to reach a maximum external validity for the threshold levels of serum IgG antibodies to Shigella LPS associated with protection [27].
The findings of this study on serum IgG antieS. sonnei LPS as a correlate of protection, including putative thresholds predicting field VE, are important in the regulatory process of licensure of the candidate Shigella tetravalent conjugate and other injectable subunit O-SPebased vaccine candidates [27].

Transparency declaration
The study was supported by Investment ID OPP1195433 from Bill & Melinda Gates Foundation, USA. The authors declare that they have no conflicts of interest.