Quantifying Antibody Persistence After a Single Dose of COVID‐19 Vaccine Ad26.COV2.S in Humans Using a Mechanistic Modeling and Simulation Approach

Abstract

Understanding persistence of humoral immune responses elicited by vaccination against coronavirus disease 2019 (COVID‐19) is critical for informing the duration of protection and appropriate booster timing. We developed a mechanistic model to characterize the time course of humoral immune responses in severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2)–seronegative adults after primary vaccination with the Janssen COVID‐19 vaccine, Ad26.COV2.S. The persistence of antibody responses was quantified through mechanistic modeling‐based simulations. Two biomarkers of humoral immune responses were examined: SARS‐CoV‐2 neutralizing antibodies determined by wild‐type virus neutralization assay (wtVNA) and spike protein‐binding antibodies determined by indirect spike protein enzyme‐linked immunosorbent assay (S‐ELISA). The persistence of antibody responses was defined as the period of time during which wtVNA and S‐ELISA titers remained above the lower limit of quantification. A total of 442 wtVNA and 1,185 S‐ELISA titers from 82 and 220 participants, respectively, were analyzed following administration of a single dose of Ad26.COV2.S (5 × 10¹⁰ viral particles). The mechanistic model adequately described the time course of observed wtVNA and S‐ELISA serum titers and its associated variability up to 8 months following vaccination. Mechanistic model‐based simulations show that single‐dose Ad26.COV2.S elicits durable but waning antibody responses up to 24 months following immunization. Of the estimated model parameters, the production rate of memory B cells was decreased in older adults relative to younger adults, and the antibody production rate mediated by long‐lived plasma cells was increased in women relative to men. A steeper waning of antibody responses was predicted in men and in older adults.

Study Highlights.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

• Recent evidence shows an association between neutralizing and binding antibody titers elicited by coronavirus disease 2019 (COVID‐19) vaccine and vaccine efficacy up to ~ 4 months post‐immunization. However, the utility of these measurements for predicting long‐term vaccine efficacy remains unknown.

WHAT QUESTION DID THIS STUDY ADDRESS?

• A mechanistic modeling‐based approach was used to predict the long‐term neutralizing and binding antibody persistence after single‐dose Ad26.COV2.S in severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2)–seronegative volunteers. The study provides mechanistic insight useful for vaccine development.

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

• A single dose of Ad26.COV2.S was predicted to elicit antibody responses up to 24 months following immunization. Of the estimated model parameters, the production rate of memory B cells decreased in older adults, and the antibody production rate mediated by long‐lived plasma cells increased in women. A steeper antibody waning was predicted in men and in older adults.

HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

• These findings are a step toward translating neutralizing and binding antibody measurements into correlates of protection of COVID‐19 vaccine‐induced immunity.

The severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2) causing coronavirus disease‐2019 (COVID‐19) is highly transmissible and pathogenic, ¹ posing an ongoing major global health threat. Vaccination remains a cornerstone for controlling the COVID‐19 pandemic, ² including the uptake of potential booster doses in the context of emerging variants of concern. ³

The Janssen COVID‐19 vaccine, Ad26.COV2.S (JNJ‐78436735), is a monovalent vaccine composed of a recombinant, replication‐incompetent adenovirus type 26 (Ad26) vector constructed to encode prefusion‐stabilized full‐length SARS‐CoV‐2 spike (S) protein. ⁴ , ⁵ As of April 2022, more than 50 million doses of this vaccine have been administered worldwide. ⁶ Robust immune responses following immunization with single‐dose 5 × 10¹⁰ viral particles (vp) of Ad26.COV2.S were reported in an interim analysis of a phase I/IIa trial. ⁵ Quantifiable SARS‐CoV‐2 neutralizing and spike binding antibodies elicited by single‐dose Ad26.COV2.S were detectable in 99% of participants aged 18–55 years and 96% of participants aged ≥ 65 years at day 28 following immunization. ⁵ Durable antibody responses through at least 8 months were observed following single‐dose Ad26.COV2.S in both younger (18–55 years) and older (≥ 65 years) populations. ⁷ However, published data are limited on the persistence of antibody responses beyond 8 months following vaccination with Ad26.COV2.S.

Vaccine‐induced neutralizing and binding antibodies targeting Spike epitopes are recognized as the most accurate biomarkers for predicting protection (i.e., correlates of protection (CoPs)) against SARS‐CoV‐2. ⁸ Preliminary evidence from a phase III ENSEMBLE trial reported significant correlations between binding and neutralizing antibody titers and vaccine efficacy (VE) against moderate to severe‐critical COVID‐19 through 83 days after single‐dose Ad26.COV2.S. ⁹ Similarly, increasing levels of binding and neutralizing antibodies were associated with greater VE against symptomatic COVID‐19 through ~ 4 months after 2 doses of a mRNA‐1273 vaccine. ¹⁰ However, published data are limited on the utility of antibody markers to predict protection against COVID‐19 beyond 4 months. In addition, the emergence of new SARS‐CoV‐2 lineages that are associated with lower Ad26.COV2.S efficacy and reduced antibody responses relative to the reference strain ¹¹ , ¹² further complicates the determination of CoPs. Determining reproducible CoPs against SARS‐CoV‐2 in relation to clearly defined end points and with improved SARS‐CoV‐2 variant coverage would facilitate the measurement of persistence of VE, ¹³ which could inform decisions about appropriate booster scheduling.

Mechanistic modeling is recognized as a powerful tool for describing the relation between vaccine‐delivered antigen and antibody response. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ For example, this approach was used to quantify long‐term persistence of binding antibodies after heterologous prime‐boost immunization with Ad26‐ and MVA‐vectored vaccines against Ebola. ¹⁵ By using a model based on the assumption of the importance of two distinct antibody‐secreting cell populations – the short‐lived and long‐lived plasma cells ¹⁶ – the study adequately predicted the persistence of humoral responses, and identified key factors that induced variability in humoral immune responses. ¹⁵

Building on prior modeling research of humoral immune response dynamics in the setting of Ebola ¹⁴ , ¹⁵ and hepatitis B ¹⁷ vaccines, we developed an extended 6‐compartment mechanistic model to characterize the time course of neutralizing and binding antibody responses to a single dose of Ad26.COV2.S in SARS‐CoV‐2–seronegative human volunteers. The mechanistic model incorporated key factors responsible for describing the dynamics of vaccine‐elicited antibody response: antigen, memory B cells, short‐lived and long‐lived plasma cells, and antibodies both in serum and peripheral sites. ¹⁴ , ¹⁵ , ¹⁷ , ¹⁸ Mechanistic modeling‐based simulations were used to quantify the persistence of antibody responses – defined as the time during which neutralizing and binding antibody titers remained above the lower limit of quantification (LLOQ) – up to 24 months after immunization with Ad26.COV2.S.

METHODS

Data collection

Data from SARS‐CoV‐2–seronegative adults (those who had not been previously infected by SARS‐CoV‐2 or vaccinated against COVID‐19), aged 18–55 and ≥ 65, who had been randomized in 2 clinical trials to receive a single dose of 5 × 10¹⁰ vp Ad26.COV2.S, were used to conduct the mechanistic modeling analysis. A pooled dataset using cohorts 1a and 3 from the phase I–IIa COV1001 study (NCT04436276) ¹⁹ and group 5 from the phase IIa COV2001 study (NCT04535453) ²⁰ was created and utilized for the analyses. The analyses did not include participants in the age range between 56 and 64 years. Both studies were reviewed and approved by institutional review boards. All participants provided written informed consent, which included the possibility of further testing and evaluation. These trials adhere to the principles of the Declaration of Helsinki.

Blood samples were collected at days 1 (day of immunization), 15, 29, 57, 71, 85, and 239 in COV1001, and days 1, 15, 29, 57, 64, 71, 85, and 169 in COV2001. Two biomarkers of antibody response were used. Binding antibodies specific to the pre‐fusion conformation of SARS‐CoV‐2 spike protein were determined by spike protein enzyme‐linked immunosorbent assay (S‐ELISA). Neutralizing antibodies were determined by wild‐type virus neutralizing assay (wtVNA) in a subset of participants in each study.

Immunogenicity assays

Wild‐type virus neutralizing assay

Neutralizing antibodies capable of inhibiting wild‐type virus infections were quantified using a wild‐type virus microneutralization assay performed by Public Health England using the Victoria/1/2020 strain. Assay values were determined as a 50% maximal inhibitory concentration (IC₅₀) and ranged between 58 (LLOQ) and 12,800 (upper limit of quantification (ULOQ)).

Spike protein enzyme‐linked immunosorbent assay

The SARS‐CoV‐2 antigen was a stabilized pre‐fusion spike protein ((2P), Δfurin, T4 foldon, His‐Tag), derived from the first clinical isolate of the Wuhan strain (Wuhan, 2019, whole genome sequence NC_045512). Assay LLOQ and ULOQ values were 50.3 and 15,797.9 EU/mL, respectively.

Further details on the wtVNA and S‐ELISA assays were previously reported, ⁵ and are provided in the Supplementary Materials. Seropositivity was defined as sera above the LLOQ in both assays.

Structural model

A 6‐compartment mechanistic model (Figure 1 ) was developed considering the key elements of humoral immune response ¹⁴ , ¹⁵ , ¹⁷ : the antigen (Ag, S‐protein), the memory B cells (M), the short‐lived plasma cells (S), the long‐lived plasma cells (L), the antibody in serum (Ab), and the antibody in peripheral sites (P). After vaccination with Ad26.COV2.S, a certain amount of antigen, S‐protein, is generated in the body. In the absence of antigen measurements, a K‐PD (kinetics of drug action) model was used to relate vaccine dosing to the antigen effects on humoral response ²¹ through a virtual antigen compartment, as described by Eq. 1:

$$\frac{\mathrm{dAg}}{\mathrm{\text{d}}t}=-k_{\mathrm{Ag}}·\mathrm{Ag}$$ (1)

Figure 1.

Schematic of the mechanistic model for humoral immune response after a single dose of Ad26.COV2.S in severe acute respiratory syndrome coronavirus‐2 (SARS‐CoV‐2)–seronegative volunteers. The mechanistic model builds on models of vaccine‐induced antibody dynamics as described previously. ¹⁴ , ¹⁵ , ¹⁷

The presence of antigen stimulates the production of memory B cells and their differentiation into short‐lived and long‐lived plasma cells. Because memory B cells are also self‐replicating, their proliferation is a dual process with antigen‐dependent and antigen‐independent pathways, ¹⁷ , ²² characterized by the rate constants k _pAg and k _pM, respectively. In the absence of antigen, the memory B cell population continues to grow, whereas the antigen concentrations decay over time toward zero. ²³ However, the growth of memory B cells is assumed to be limited to a maximum, N, according to a logistic growth function, ¹⁷ , ²⁴ , ²⁵ as described by the Eq. 2:

$$\begin{array}{c}\frac{\mathrm{d}M}{\mathrm{d}t}=\left(k_{\mathrm{pAg}}·\mathrm{Ag}+k_{\mathrm{pM}}·M\right)\left(1-\frac{M}{N}\right)-\left[\left(k_{\mathrm{pS}}+k_{\mathrm{pL}}\right)·\mathrm{Ag}+k_{\mathrm{eM}}\right]·M\hfill \end{array}$$ (2)

where k _pS and k _pL represent differentiation rates of the memory B cells into short‐lived and long‐lived plasma cells, ¹⁴ , ²⁵ which are also stimulated by the antigen presence; and k _eM represents the first‐order rate constant of the memory B cell elimination, which is antigen‐independent. Short‐lived B cells are differentiated cells that together with long‐lived B cells are responsible for the antibody production. Their production was assumed to be a second‐order process, depending directly on both the amount of antigen and the memory B cells available, as described by Eqs. 3 and 4:

$$\frac{\mathrm{d}S}{\mathrm{d}t}=k_{\mathrm{pS}}·\mathrm{Ag}·M-k_{\mathrm{eS}}·S$$ (3)

$$\frac{\mathrm{d}\mathit{L}}{\mathrm{d}\mathit{t}}=k_{\mathrm{pL}}·\mathrm{Ag}·M-k_{\mathrm{eL}}·L$$ (4)

where k _eS and k _eL represent the first‐order elimination rate constants of short‐lived and long‐lived plasma cells, respectively.

Although Eqs. 3 and 4 are mathematically identical, the dynamics that they describe are different. Short‐lived plasma cells are expected to have a half‐life of a few days, ²⁶ whereas long‐lived plasma cells are expected to persist for months or even years in absence of cell division. ²² , ²⁵ The serum antibody production was assumed to be directly proportional to the amount of short‐lived and long‐lived plasma cells, whereas the antibody disposition was characterized by a first‐order elimination and linear distribution to a nonspecific peripheral compartment. These factors are described in Eqs. 5 and 6:

$$\begin{array}{c}\frac{\mathrm{d}\mathit{Ab}}{\mathrm{d}t}=k_{\mathrm{AbS}}·S+k_{\mathrm{AbL}}·L-k_{\mathrm{eAb}}·\mathrm{Ab}-k_{\mathrm{AbP}}·\mathrm{Ab}+k_{\mathrm{PAb}}·P\hfill \end{array}$$ (5)

$$\frac{\mathrm{d}P}{\mathrm{d}t}=k_{\mathrm{AbP}}·\mathrm{Ab}-k_{\mathrm{PAb}}·P$$ (6)

In these equations, k _AbS and k _AbL represent the first‐order production rate constant of antibodies from short‐lived and long‐lived plasma cells, respectively; k _AbP and k _PAb describe the nonspecific distribution and return of antibodies from the peripheral compartment, respectively; and k _eAb is the first‐order elimination rate constant of the serum antibodies.

Structural model parameters

Given that only serum antibody titers data were available for modeling, the humoral immune response model was overparameterized. Consequently, the parameters in the model were grouped into three groups: structurally nonidentifiable, fixed, and identifiable parameters.

After applying a sequence of nine transformations to the system of differential equations (described in the Supplementary Materials), an identical model to the one described above was obtained, resulting in three structurally nonidentifiable parameters: k _pAg, k _pS, and k _pL. In particular, k _pS and k _pL were confounded with the antibody production rates from short‐lived and long‐lived plasma cells (k _AbS and k _AbL, respectively). The same applied to k _pAg. Although k _pAg, k _pS, and k _pL could all be set to 1 and consequently removed from the model, we kept them to preserve the model's mechanistic structure. The sum of k _pS and k _pL was estimated and their ratio fixed to the value previously reported ¹⁴ to reduce the number of estimated parameters, and k _pAg was arbitrarily set to 1 (cells/vp/day) due to its nonidentifiability.

The fixed model parameters included the first‐order equilibration rate constant representing the apparent decay of the S‐protein (k _Ag), which was fixed to 0.065 days^‐1. ¹⁴ The growth of memory B cells was assumed to be limited to a maximum, N, fixed to 3.019 · 10⁴ cells, based on data from participants who recovered after natural SARS‐CoV‐2 infection and were vaccinated with 2 doses of mRNA vaccine. ²⁷ The parameter k _eM was fixed to 0 year^‐1 because of the time frame of the collected data. The value of k _eL was fixed to 0 day^‐1 in all models except the final updated model for S‐ELISA (as detailed in the Supplementary Materials). The first‐order elimination rate constant of short‐lived plasma cells (k _eS) was fixed to 0.578 days^‐1. ¹⁵ Antibody disposition parameters – central‐to‐peripheral rate (k _AbP), peripheral‐to‐central rate (k _PAb), and antibody decay rate (k _eAb) – were fixed based on the published estimates. ¹⁸

In addition to the structurally nonidentifiable parameters, the experimental design impacted the parameter estimation. For any ordinary differential equation model with p unknown constant parameters, 2p + 1 are the minimum quantifiable observations needed for the identification of the p parameters if observations are absolutely accurate. ²⁸ , ²⁹ It has been also observed that increasing the number of time points would not help in improving parameter identifiability once the system enters its steady state. ²⁹ This criterion, together with the Fisher information matrix showing no large correlation between parameters, led to the estimation of four structural parameters: k _pM, the sum of k _pS and k _pL, k _AbS, and k _AbL.

Building of the final model and the final updated model

Because the sample size was greater in participants with S‐ELISA than wtVNA data, the S‐ELISA data were modeled first. Subsequently, a similar model structure was applied to the wtVNA dataset, where all systemic parameters were kept fixed, whereas antibody‐dependent parameters (k _AbS and k _AbL) were re‐estimated. Therefore, interindividual variability (IIV) in model parameters was quantified using an exponential error model for k _pM, k _AbS, and k _AbL when modeling S‐ELISA, and for k _AbS and k _AbL when modeling wtVNA data. An additive error model was also used to quantify the residual unexplained variability after logarithmic transformation of the measured antibody serum titers with a different estimate of the variance for S‐ELISA and wtVNA assays. Further details on IIV and additive error models are available in the Supplementary Materials.

Covariate analysis

Once the suitable structural model was identified, a covariate analysis was performed and the effects of age, body weight, sex, race, and country were evaluated. Prior to categorical analysis, levels of categorical covariates that were under‐represented (e.g., < 10% of the total or < 30 participants) were lumped together with other categories. The covariate screening was guided by graphical assessment and stepwise linear regression of the relationships between the empirical Bayesian estimates of the random effects and the covariates. Only covariates that were statistically significant (P < 0.01) and/or had a coefficient of determination r ² > 0.10 with model parameters were further tested one‐by‐one in nonlinear mixed‐effects modeling to evaluate whether they were incorporated in the mechanistic model following the forward‐inclusion (P < 0.01) and backward‐elimination (P < 0.005) process. ³⁰ Further details are provided in the Supplementary Materials.

Model evaluation and selection

To identify the mechanistic model that best describes the time course of the serum antibody data, a series of models were evaluated. For each nested model, the improvement in the fit was assessed by the likelihood ratio test (P ≤ 0.01), based on the change in the minimum objective function value. ³¹ , ³² In addition, the reduction in IIV and residual variability, the precision and correlation in parameter estimates, diagnostic plots, predictive checks, and shrinkage were also examined.

Goodness‐of‐fit plots and visual predictive check (VPC) were used to evaluate the developed model. ³³ A nonlinear mixed‐effects model analysis was conducted using NONMEM 7.4 (ICON plc). ³⁴ The first‐order conditional estimation method (with INTERACTION option) estimation method was used.

Model‐based simulations

Model‐based simulations at population level were performed using the parameter estimates of the fixed and random effects included in the final model. The uncertainty around the fixed‐effect parameter estimates, the random effect distribution, and the residual error were considered when conducting the simulations. The time course of wtVNA and S‐ELISA titers every 4 weeks up to 2 years after vaccination was simulated for a cohort of 250,000 participants. Participant age was sampled from the age distribution observed in the COV3001 study, ³⁵ whereas men and women were sampled by 1:1 ratio.

The persistence of antibody responses at individual level was defined as the time from vaccination until the last wtVNA or S‐ELISA titer above the LLOQ before two consecutive monthly measurements of wtVNA or S‐ELISA titers below the LLOQ. If the two consecutive titers were found below the LLOQ, all titers from the second one onward were counted as below the LLOQ. The persistence of antibody responses at the population level was defined as the time at which 95%, 90%, 75%, 50%, 25%, 10%, and 5% of participants showed the persistence of antibody responses at individual level. If the defined predicted proportion of participants above LLOQ fell between the two sampled months, the earlier month was chosen.

RESULTS

Population

Samples were obtained from participants who received a single injection of 5 × 10¹⁰ vp of Ad26.COV2.S in 2 studies: COV1001 and COV2001. The number of titers available for modeling in the original dataset (up to day 85) was 1,007 for S‐ELISA from 220 participants and 381 for wtVNA from 82 participants. After adding S‐ELISA and wtVNA measurements from 2 additional visits on days 169 and 239, the total number of titers available for modeling in the updated dataset was 1,185 for S‐ELISA from 220 participants and 442 for wtVNA from 82 participants. Table 1 summarizes participant demographics by assay type. Among participants with S‐ELISA titers, the median age was 52.5 years (range, 18–82; 46.8% women). Among participants with wtVNA titers, the median age was 47.5 years (range, 18–75; 42.7% women).

Table 1.

Summary demographics for the continuous and categorical covariates

		Total S‐ELISA	Total wtVNA
		N = 220	N = 82
Country, n (%)
United States		75 (34.1)	0 (0)
Belgium		70 (31.8)	47 (57.3)
The Netherlands		18 (8.2)	11 (13.4)
Spain		24 (10.9)	24 (29.3)
Germany		33 (15)	0 (0)
Race, n (%)
White, non‐Hispanic, or Latino		190 (86.4)	61 (74.4)
Black, of African heritage, or African American		0 (0)	0 (0)
White, Hispanic, or Latino		24 (10.9)	19 (23.2)
Asian		2 (0.9)	1 (1.2)
Native Hawaiian or Other Pacific Islander		0 (0)	0 (0)
American Indian or Alaskan Native		3 (1.4)	0 (0)
Other		1 (0.5)	1 (1.2)
Sex, n (%)
Male		117 (53.2)	47 (57.3)
Female		103 (46.8)	35 (42.7)
Age, years
Mean (SD)		51.6 (18.8)	48.5 (19.2)
Median		52.5	47.5
Range		(18.0; 82.0)	(18.0; 75.0)
Weight, kg
Mean (SD)		73.5 (12.5)	73.3 (12.8)
Median		72.1	72.7
Range		(48.0; 113)	(48.7; 102)
Height, cm
Mean (SD)		172 (9.94)	172 (9.85)
Median		172	172
Range		(148; 198)	(148; 196)
Body mass index, kg/m2
Mean (SD)		24.7 (3.09)	24.5 (3.22)
Median		24.4	24.2
Range		(16.6; 29.9)	(18.4; 29.8)
Race (dichotomic), n (%)
White, non‐Hispanic, or Latino		190 (86.4)	61 (74.4)
Other		30 (13.6)	21 (25.6)
Age (dichotomic), n (%)
< 60		116 (52.7)	47 (57.3)
≥ 60		104 (47.3)	35 (42.7)

S‐ELISA, spike protein enzyme‐linked immunosorbent assay; wtVNA, wild‐type virus neutralization assay.

The S‐ELISA and wtVNA models were first built with data available up to day 85. This structural model is referred to as the “final model” in the analysis. The S‐ELISA and wtVNA models were re‐run with additional data at days 169 and 239 by keeping the previously estimated parameters fixed, and by additionally estimating antibody‐dependent parameter k _eL to capture the waning of binding antibody titer. Residual variability and IIV were re‐estimated. This structural model is referred to as the “final updated model” in the analysis.

The observed S‐ELISA and wtVNA titers from the original dataset vs. time after vaccination are presented in Figure S1 . The similar median serum titer time profile by study justified pooling COV1001 and COV2001 datasets for modeling.

S‐ELISA and wtVNA mechanistic models

Table 2 shows the fixed and estimated parameter values of the final model and the final updated model for S‐ELISA and wtVNA datasets, including statistically significant covariate effects. All parameters were estimated with adequate precision with a relative standard error of ≤ 40% and ≤ 47% for the S‐ELISA and wtVNA models, respectively.

Table 2.

Fixed and estimated parameter values of the final model and the final updated model of the humoral immune response after a single dose of Ad26.COV2.S

Parameter	Unit	Assay	Estimate	95% CI
Ad26 antigen decay, k Ag	day‐1	S‐ELISA	0.065	Fixed 14
		wtVNA
Mem B production rate (antigen‐dependent, k pAg)	Cells/vp/day	S‐ELISA	1	Fixed following reparameterization of the model
		wtVNA
Mem B production rate (antigen‐independent, k pM) for age < 60 years	day‐1	S‐ELISA	0.523	0.381 to 0.665
		wtVNA	0.523	Fixed from S‐ELISA model
Mem B decay, k eM	year‐1	S‐ELISA	0	Fixed due to short follow‐up
		wtVNA
Maximum mem B produced (N)	Cells	S‐ELISA	3.019.104	Fixed 27
		wtVNA
Sum of S and L production rate – mem B dependent, k pSpL = k pS + k pL	day‐1/vp	S‐ELISA	0.691.10−5	(0.517 to 0.865).10−5
		wtVNA	0.691.10−5	Fixed from S‐ELISA model
Ratio of S and L production rate – mem B dependent, k pS/k pL	day‐1/vp	S‐ELISA	2.5/0.011	Fixed 14
		wtVNA
S production rate – mem B dependent, k pS	day‐1/vp	S‐ELISA	k pS = k pSpL/(1 + (0.011/2.5))	Derived
		wtVNA
S elimination half‐life, k eS	day‐1	S‐ELISA	0.578	Fixed 15
		wtVNA
L production rate – mem B dependent, k pL	day‐1/vp	S‐ELISA	k pL = k pSpL–k pS	Derived
		wtVNA
L elimination half‐life, k eL	day‐1	S‐ELISA	0; 16.8·10−4 d	Fixed due to short follow‐up; (9.45 to 24.15) ∙ 10−4 d
		wtVNA	0; 0	Fixed due to short follow‐up; due to unidentifiability
Ab production rate by S, k AbS	day‐1 · (EU/mL)/Cells	S‐ELISA	0.530	0.377 to 0.683
		wtVNA	0.603	0.477 to 0.729
Ab production rate by L for males, k AbL	day‐1 · (EU/mL)/Cells	S‐ELISA	4.43	3.70 to 5.16
		wtVNA	1.86	1.22 to 2.50
Central‐to‐peripheral rate, k AbP	day‐1	S‐ELISA	0.2075	Fixed 18
		wtVNA
Peripheral‐to‐central rate, k PAb	day‐1	S‐ELISA	0.2716	Fixed 18
		wtVNA
Ab decay rate, k eAb	day‐1	S‐ELISA	0.0556	Fixed 18
		wtVNA
Age (dichotomic, k pM_AGE) – decrease for age ≥ 60 years (1 + estimate)	–	S‐ELISA	−0.441	−0.614 to −0.268
		wtVNA	−0.441	Fixed from the S‐ELISA model
Sex (k AbL_SEX) – increase for female (1 + estimate)	–	S‐ELISA	0.671	0.140 to 1.20
		wtVNA	1.10	0.087 to 2.11
Inter‐individual variability			CVa (%)	Shrinkageb (%)
On kpM		S‐ELISA	100; 78.5d	43.6; 48d
On kAbS		S‐ELISA	145; 147d	18.9; 15.9d
		wtVNA	101; 102d	12.5; 11.2d
On kAbL		S‐ELISA	162; 141d	5.0; 4.5d
		wtVNA	120; 104d	17.0; 13.0d
Proportional errorc		S‐ELISA	8.28; 11.7d	23.1; 19.4d
		wtVNA	17.5; 19.1d	16.9; 15.7d

CI, confidence interval; CV, coefficient of variation; S‐ELISA, spike protein enzyme‐linked immunosorbent assay; vp, viral particles; wtVNA, wild‐type virus neutralization assay.

^a CV% derived as sqrt(exp(ω ²)‐1).

^b Shrinkage derived as 1 − SD(η _i)/sqrt(ω ²).

^c Proportional error refers to original values (i.e., additive error on log‐transformed values).

^d Values estimated from the re‐running of the model with additional data from visits on days 169 and 239 (i.e., the final updated model).

Age and sex were statistically significant covariates. Older adults (≥ 60 years) showed a 44% (95% confidence interval (CI), 27–61%) decrease in k _pM relative to younger adults (18–55 years) in both S‐ELISA and wtVNA datasets. Moreover, women showed a 67% (95% CI, 14–120%) increase in k _AbL relative to men in the S‐ELISA model, and a 110% (95% CI, 9–210%) increase in k _AbL relative to men in the wtVNA model.

Plotting the population‐ and individual‐predicted titers against observed S‐ELISA and wtVNA titers indicated a normal random scatter around the identity line (Figures S2 a and S3 a). Additionally, no trend evidencing model inadequacy was indicated by plotting conditional weighted residuals by population predicted S‐ELISA titers and time (Figure S2 b). Differently from the S‐ELISA dataset, the longer follow‐up wtVNA titers were maintained up to 8 months (Figure S3 b).

As confirmed by VPC plots, the final updated model adequately described the time course of S‐ELISA (Figure 2 a ) and wtVNA serum titers (Figure 2 b ) and its associated variability in seronegative participants up to 8 months after vaccination.

Figure 2.

Visual predictive check plots for the final updated model of antibody response after single‐dose Ad26.COV2.S in SARS‐CoV‐2–seronegative volunteers. (a) S‐ELISA. (b) wtVNA. Blue dots represent observed data (log₁₀ transformed). Continuous and dashed red lines represent median, 5^th and 95^th percentiles of the observed data. Black continuous line and red shaded area represent median of the simulated data with its 95% CI. Blue dashed lines and blue shaded area represent 5^th and 95^th percentiles of the simulated data with its 95% CI. Horizontal blue dashed lines represent the LLOQ and ULOQ specific to each assay. CI, confidence interval; IC₅₀, 50% maximal inhibitory concentration; LLOQ, lower limit of quantification; SARS‐CoV‐2, severe acute respiratory syndrome coronavirus‐2; S‐ELISA, spike protein enzyme‐linked immunosorbent assay; ULOQ, upper limit of quantification; wtVNA, wild‐type virus neutralization assay.

S‐ELISA and wtVNA model‐based simulations

As shown in Table 3 , by 8 months following a single dose of Ad26.COV2.S, there was a high agreement between the observed proportion of participants with the measurable binding and neutralizing antibodies in COV1001 (95.0% and 87.0%, respectively) and model predictions (95.2% and 88.4%, respectively). The predicted proportion of participants with measurable titers at 24 months was 81.1% for S‐ELISA and 80.4% for wtVNA. Similarly, 75% of participants were predicted to show measurable antibody titers beyond 24 months for both S‐ELISA and wtVNA datasets.

Table 3.

Predicted persistence of wtVNA and S‐ELISA antibody responses in SARS‐CoV‐2‐seronegative participants after single‐dose Ad26.COV2.S, and observed persistence at 8 months following immunization in COV1001

Predicted % above the LLOQ for the overall population			Observed % (95% CI)a above the LLOQb		Predicted time when % of participants is above LLOQ
Time, months	S‐ELISA (%)	wtVNA	S‐ELISA (N = 133)	wtVNA (N = 45)	Percentage (%)	S‐ELISA, months	wtVNA, months
8	95.2	88.4	95% (91%, 98%)	87% (74%, 94%)	95	8	4
12	92.4	84.9			90	15	6
18	87.6	82.1			75	>24	>24
24	81.1	80.4			50	>24	>24

The predictions are based on the final updated model. The persistence of antibody responses is shown as the proportion of participants with measurable wtVNA and S‐ELISA titers up to 2 years after a single dose of Ad26.COV2.S and as time up to which 95%, 90%, 75% and 50% of participants were above LLOQ.

CI, confidence interval; LLOQ, lower limit of quantification; N, number of participants in the final analysis dataset (8‐month timepoint only); SARS‐CoV‐2, severe acute respiratory syndrome coronavirus‐2; S‐ELISA, spike protein enzyme‐linked immunosorbent assay; wtVNA, wild‐type virus neutralization assay.

^a 95% CI for binomial probabilities based on Wilson's method.

^b Observed proportions after 8 months following single‐dose Ad26.COV2.S in COV1001.

The predicted proportions of participants with measurable antibody titers up to 2 years after single‐dose Ad26.COV2.S stratified by age and sex are depicted in Figure 3 for S‐ELISA and Figure 4 for wtVNA. A steeper antibody waning was predicted in men and in older participants (≥ 60 years) in both datasets. Specifically, 76.3% of older and 82.5% of younger participants were predicted to have measurable neutralizing antibodies 24 months following vaccination; similarly, 77.7% of older and 82.8% of younger participants were predicted to show measurable binding antibodies after 24 months (Table S1 ). Furthermore, 69.4% of men and 91.4% of women were predicted to have measurable neutralizing antibodies 24 months following vaccination; similarly, 74.9% of men and 87.2% of women were predicted to show measurable binding antibodies after 24 months (Table S2 ).

Figure 3.

Predicted proportions of SARS‐CoV‐2–seronegative participants with measurable S‐ELISA antibody titers up to 2 years after single‐dose Ad26.COV2.S stratified by age and sex. SARS‐CoV‐2, severe acute respiratory syndrome coronavirus‐2; S‐ELISA, spike protein enzyme‐linked immunosorbent assay.

Figure 4.

Predicted proportions of SARS‐CoV‐2–seronegative participants with measurable wtVNA antibody titers up to 2 years after single‐dose Ad26.COV2.S stratified by age and sex. SARS‐CoV‐2, severe acute respiratory syndrome coronavirus‐2; wtVNA, wild‐type virus neutralization assay.

DISCUSSION

The open 6‐compartment mechanistic model considered both the biphasic production and disposition of antibodies and included key elements of humoral immune responses ¹⁴ , ¹⁵ , ¹⁷ , ¹⁸ : antigen, memory B cells, short‐lived and long‐plasma cells, and antibodies in serum and at peripheral sites. The mechanistic model adequately described the time course of wtVNA and S‐ELISA titers and its associated variability in seronegative volunteers up to 8 months after immunization with a single dose of Ad26.COV2.S.

The main mechanistic predictors of the time course of immune response were elements of the humoral response related to the antibody production. The peak magnitude was in large proportion dependent on the production of short‐lived plasma cells, whereas the persistence of antibody responses was driven by slow and long‐lasting production of long‐lived plasma cells known to dominate over antibody terminal phase. ²² , ²⁵ These two processes depended on the ability of memory B cells to be activated, to grow exponentially over time, and to differentiate into short‐lived or long‐lived plasma cells after antigen presentation. By contrast, the antibody disposition did not appear to be a limiting factor to the antibody dynamics in the current model.

Age and sex were identified as statistically significant covariates. The mechanistic model predicted a 67% increase in the rate of antibody production mediated by long‐lived plasma cells in women relative to men, consistent with the literature reporting stronger vaccine‐induced antibody responses in women, ³⁶ sex‐dependence in humoral response to SARS‐CoV‐2 following 2 doses of mRNA vaccine, ³⁷ and higher IgG antibody concentrations in female patients affected with severe COVID‐19. ³⁸ Furthermore, model simulations by sex revealed a slower decay in women relative to men, with a later antibody peak. This supports the finding that the persistence of antibody responses was predominantly driven by the long‐lasting production of long‐lived plasma cells, and that durability of antibody responses was associated with the magnitude of peak response to vaccination. Correspondingly, the model simulations predicted a more pronounced waning of antibody response in men relative to women: at 24 months, the predicted proportions of men vs. women were 69.4% vs. 91.4% for above‐LLOQ neutralizing antibody titer and 74.9% vs. 87.2% for above‐LLOQ binding antibody titer. By contrast, a phase III trial of single‐dose Ad26.COV2.S reported consistent VE in men and women against moderate to severe‐critical COVID‐19 ≥ 28 days after immunization, ¹¹ raising the possibility that longer follow‐up might be needed for the waning of immune response to be reflected in vaccine efficacy.

A predicted reduction in the antigen‐independent production rate of memory B cells in older adults is in line with prior research reporting decreasing memory B cell response with age among SARS‐CoV‐2–naïve participants following mRNA vaccination. ²⁷ Accordingly, the predicted proportions of measurable titers at 24 months were 76.3% for neutralizing antibodies and 77.7% for binding antibodies in older adults, and 82.5% for neutralizing antibodies and 82.8% for binding antibodies in younger adults.

Antibody persistence was determined by estimating the percentage of participants with neutralizing and binding antibody titers measured against the reference strain that remained above the LLOQ for 24 months. These findings carry implications for the identification of CoPs that are viewed as critical for demonstrating protective immune responses in future clinical trials due to practical constraints of including/retaining placebo participants, and support the primacy of using vaccine‐induced antibody titers as CoPs against SARS‐CoV‐2 infection. ⁸ The results from the correlate analyses from phase III COVID‐19 vaccine efficacy trials demonstrated utility of binding and neutralizing antibodies as CoPs in both mRNA ¹⁰ and Ad26.COV2.S ⁹ vaccine recipients up to 4 months post‐immunization. Although waning of vaccine‐induced antibody titers has been observed and could limit the utility of protective antibody thresholds measured shortly after vaccination, ⁸ future mechanistic modeling studies could be designed to perform longitudinal analysis of CoPs, focusing on investigating the correlation between VE and antibody titers at multiple timepoints. Another avenue for future research includes assessing the robustness of protective antibody thresholds relative to SARS‐CoV‐2 variants.

Given the ability of SARS‐CoV‐2 to adapt rapidly to immune pressure, evidence of waning immunity, ⁸ , ³⁹ , ⁴⁰ and the emergence of new variants with increased transmission, ⁴¹ booster dose(s) have been shown to increase antibody response and protection. ⁴² , ⁴³ Real‐world data reveal durable effectiveness of a single dose of Ad26.COV2.S against COVID‐19 infections and related hospitalizations in a seronegative cohort during the time of Delta variant predominance in the United States. ⁴⁴ In addition, real‐world data show effectiveness of a homologous booster dose of Ad26.COV2.S against severe COVID‐19 during a surge of the Omicron variant in South Africa. ⁴⁵ Mechanistic modeling research aimed at describing the persistence of antibody responses after multiple dose levels and booster regimens of Ad26.COV2.S is ongoing.

In conclusion, the mechanistic model adequately described the observed time course of neutralizing and binding antibody titers in SARS‐CoV‐2–seronegative participants up to 8 months following a single dose of Ad26.COV2.S. The rate of antibody production mediated by long‐lived plasma cells was predicted to be higher in women vs. men, whereas the rate of antigen‐independent memory B cell production was predicted to decrease in older relative to younger adults. The mechanistic modeling‐based simulations provide evidence that one dose of Ad26.COV2.S would elicit durable antibody responses between 8 and 24 months after immunization. This modeling data can inform hypotheses about the long‐term dynamics of humoral immune responses following homologous and heterologous booster doses in different subgroups of Ad26.COV2.S recipients.

FUNDING

The clinical studies and the analyses presented here were supported by research funding from Janssen Research & Development.The work was partially funded by the Office of the Assistant Secretary for Preparedness and Response, Biomedical Advanced Research and Development Authority, under Government Contract No. 75A50120C00034; by the National Institute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health, under Public Health Service Grant UM1 AI068635 to the HVTN SDMC; and by the Intramural Research Program of the NIAID Scientific Computing Infrastructure at Fred Hutch, under ORIP Grant S1OD028685. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

CONFLICT OF INTEREST

All authors are employees of Janssen Research and Development (a Johnson & Johnson company) and may hold stock in Johnson & Johnson.

AUTHOR CONTRIBUTIONS

A.D., M.B., and J.J.P.R. wrote the manuscript. J.S., M.L.G., G.S., V.C., J.R.G., and R.M.W.H. designed the research. J.S., M.L.G., G.S., V.C., J.R.G., and R.M.W.H. performed the research. A.D. and M.B. analyzed the data. M.N. and B.V. contributed new reagents/analytical tools.

Supporting information

Table S1

Table S2

Supplementary Materials

Figure S1

Figure S2

Figure S3