Durability of Immune Response After COVID-19 Booster Vaccination and Association With COVID-19 Omicron Infection

This cohort study investigates antibody waning after second and third doses of the BNT162b2 (BioNTech/Pfizer) COVID-19 vaccine.

This supplemental material has been provided by the authors to give readers additional information about their work.

SARS-CoV-2 IgG Assay
Samples from vaccinated participants were tested before receipt of the third dose using the SARS-CoV-2 Receptor Binding Domain (RBD) IgG assay (Beckman-Coulter, CA, U.S.A.), or after receipt of the third dose using the SARS-CoV-2 IgG II Quant (Abbott, IL, USA) test.
These commercial tests were performed according to the manufacturer's instructions. To present all IgG Antibody levels in Binding Antibody Units (BAU) per the World Health Organization (WHO) standard measurements we imputed the Abbott-based BAU values from the Beckman-Coulter assay results, based on an independent sample of individuals with both Abbott BAU and Beckman-Coulter levels.

Avidity
To measure the quality of IgG antibodies we used urea as a chaotropic reagent and test the strength of interaction between the IgG and the viral antigen (the RBD). Specifically, a 96well microtiter Polysorb plate (Nunc, Thermo, Denmark) was coated overnight at 4°C with 50μl per well of 1μg/ml of RBD antigen. After blocking with 5% skimmed milk at 25°C for 60 minutes, serum samples were diluted at 1:100, 1:400, and 1:1000 with 3% skimmed milk and added to antigen-coated wells. The plate was incubated at 25°C for 120 minutes, and following washing each sample, was incubated either with the addition of 6M urea or PBS for 10 min. After washing, a goat anti-human IgG horseradish peroxidase (HRP) conjugate (Jackson ImmunoResearch, PA, USA Code: 109-035-088) (diluted 1:15000) was be added to each well for 60 min. After washing, incubation of TMB Substrate Solution (Abcam) for 5 min, and the addition of stop solution (2N HCl), the OD of each well was measured at 450nm using a microplate reader (Sunrise, Tecan). Avidity index was calculated as the ratio (in percentage) between sample OD with 6M urea and sample OD with PBS.

SARS-CoV-2 Pseudovirus (psSARS-2) Neutralization Assay
To test the overall neutralizing ability of each serum against the WT virus and specifically to compare with neutralizing levels of the SHEBA HCW following one, two, and three vaccine doses we used Pseudovirus (psSARS-2) Neutralization as previously described 1 . SARS-CoV-eMethods 5. Memory Immune Response To investigate the memory response we isolated peripheral blood mononuclear cell (PBMC) using Ficoll density gradient centrifugation and analyzed T cell activation as described previously 4 .
T cell activation was assessed by IFN-γ ELISpot assay. Specifically, IFN-γ -secreting cells were enumerated using Elispot IFN-γ kits (IFN-γ kit, AID Autoimmun Diagnostika GmbH, Strassberg, Germany) according to manufacturer instructions. For antigen stimulation, 50 μl of SARS-CoV-2 peptide pools (S-complete, Miltenyi Biotech) was used. Test medium was used as negative control and Phytohaemagglutinin (PHA) was used as a positive control. IFN-γsecreting cells frequency was quantified using the AID ELISpot Reader (Strassberg, Germany).
The unspecific background (mean SFU from negative control wells) was subtracted from experimental readings. eMethods 6. Imputation of Binding Antibody Units and IgG Linear Mixed Model

S6.1 Imputation of Binding Antibody Units based on Beckman-Coulter Assay Results
We developed a method for imputing Abbott IgG levels from Beckman-Coulter IgG levels using data on 215 selected serum samples, taken from individuals who had not received a booster dose and were not included in the HCW cohort and were measured by both methods. We fitted a cubic polynomial regression model where the log (to the base e) of IgG measured in BAU units by the Abbott kit was regressed on the log (to the base e) of IgG measured by the Beckman-Coulter kit, its squared value, and its cubed value. The fitted regression equation, using the glm procedure in R, was: logIGG_Abbott = 4.506 + 0.6634 × logIGG_Beckman -0.0852× (logIGG_Beckman) 2 + 0.0403× (logIGG_Beckman) 3  The regression equation shown above was used to impute the values of Abbott BAU for samples taken after the second vaccine dose. To avoid extrapolation, any Beckman-Coulter

Output from the glm procedure in R
IgG value that was lower than the minimum value of the calibration sample (0.06) was assigned to that minimum (occurring in 0.1% of the HCW sample) and any value above the maximum value (55.29) was assigned to that maximum (occurring in 2.2% of the HCW sample). Abbott BAU levels following the third vaccine dose were measured directly and did not require imputation.

Doses
We modeled the natural log-transformed IgG in a mixed-effects linear model with subjectlevel random effects. Separate models were run for the second and third vaccine doses. We fit a linear slope from 30 days after vaccination onwards. At the same time, using the statistical m described below, IgG tests taken before 30 days were used to estimate the "peak" IgG attained. We fitted a model that included age group (<45, 45-64, ≥65), sex, time (measured in days from day 30 post-vaccine onwards; all measurements before day 30 were counted as day 0), and all age-sex-time interactions (including the 3-way interaction) as fixed effects, and subject-level random intercept. For the third dose, the three-way interaction between age, sex, and time was small and not statistically significant, and therefore omitted.
In summary, for the second dose, we used the following statistical model: where is the IgG level of subject i measured on the j th occasion, is an overall intercept, ( − 30) + is the time of measurement since 30 days post-vaccination and is set equal to 0 for the first thirty days post-vaccination, is the gender of subject i (coded 0 for female and 1 for male), is vector denoting the age-group of subject i (coded [0,0] for <45y, [1,0] for 45-64y and [1,1] for 65+y), the 's are the regression coefficients, and , , are the subject random intercept, the subject random slope, and the residual error, respectively, all assumed to be independently distributed normally with mean zero, the residual error independently of the intercept and slope. The same model was used for levels following the third dose, except that the term ( − 30) + was omitted.
Expected peak IgG levels, rates of decline, and level at day 140 following vaccine for each age-sex profile were estimated from the coefficients obtained from the model fit and averaged over the cohorts, with weights according to the their sample proportions. Specifically, for the second dose, the log peak for subject i 's age-sex profile was estimated from + + + , the rate of change per day on the log scale from + + + , and the log level at 140 days from + 110 + + + 110 + 110 + 110 . The same expressions were used for kinetics parameters following the third dose, except that the terms involving were omitted.
For comparisons between the kinetics following the second and third doses, the weighted averages of peak level, rate of decline and level at 140 days for the third vaccine dose were standardized to the distribution of age and sex of those in the second dose cohort. Ratios of these averaged parameters for the third dose to the averaged parameters for the second dose were then computed together with a 95% confidence interval.
The confidence intervals were computed by first estimating the standard error of the ratio This delta method assumes statistical independence of the numerator and denominator; for its computation, the method required estimates of the standard errors of the numerator and denominator of the ratio. These were computed as follows. For the denominator (the second dose estimate), a bootstrap procedure was used to account for the uncertainty due to the imputation: first a bootstrap of the calibration sample was taken and the imputation equation was re-computed; then a bootstrap of the HCW sample for the second dose was taken, the linear mixed model was run, and the age-sex standardized peak, slope or 14-day level was estimated; from the distribution of these parameters across the bootstrap samples, a standard error was computed. The standard error of the numerator (the third dose estimate) was derived directly from the model-based variance-covariance matrix of the regression coefficients estimates from the linear mixed model.
There are two assumptions in the above method that may not hold true. Firstly, the normality of the sampling distribution of the ratio, and secondly the statistical independence of the numerator and denominator. To check on whether these lead to inaccurate confidence intervals we also ran a method that was entirely based on bootstrap sampling and not dependent on the above assumptions. As previously, a bootstrap sample of the calibration sample was first taken and the calibration equation was recomputed and applied to the Beckman-Coulter readings following the second dose. This step was followed by taking a stratified bootstrap sampling of the HCW cohort using three strata: (i) those HCWs contributing to follow-up following only the second dose; (ii) those HCWs contributing to follow-up following only the third dose; and (iii) those HCWs contributing to follow-up following both the second and the third dose. Separate linear mixed models were conducted following each dose; for the second dose including strata (i) and (iii), and for the third dose including strata (ii) and (iii). The peak level, rate of decline and level at 140 days were computed for each dose as described above, and the ratio of third to second dose calculated for each of these parameters. This bootstrap sampling was repeated 1000 times, and the confidence intervals for the ratios was computed using the bootstrap percentile method, after checking that the bootstrap distributions appeared symmetric. In the table below we compare the confidence intervals based on the delta method (that were reported in the paper) with those based on the bootstrap method. It can be seen that there is excellent agreement between the two methods, so we have retained the results from the delta method in our revised version of the paper.

Level Kinetics Following Third Vaccine Dose
To Interactions between age (<65, ≥65), sex, time and omicron infection were tested and retained if significant at the 5% level. When a higher-order interaction was retained all its lower-order interactions were also retained. In this way, we included in the model all main effects and the following interactions: age-omicron-time; age-time; omicron-time; sex-time. From the parameter estimates of this model, ratios of average peak levels and rates of waning between omicron-infected and uninfected persons were computed separately for each age group.
Standard errors of the ratios between omicron-infected and uninfected were computed for these parameters (peak level and rate of waning) using the delta method.