We conducted a prospective cohort study “MARkers of Vaccine Efficacy and Longevity in SARS-CoV-2 (MARVELS)” in healthy and immunocompromised children aged 5 to 12 years old. The study protocol was approved by the National Healthcare Group Domain Specific Review Board (NHG DSRB) (2021/00945, 2021/00984, 2021/01040). Children were recruited from the general population, through advertisements around the National University Hospital, Singapore and in the community with written parental consent and written subject assent. All children had no history of SARS-CoV-2 infection prior to enrollment, and were negative for anti-nucleocapsid antibodies. Demographic characteristics including gender are parent- or self-reported, and not considered for enrollment. Decision to receive the vaccination was based on parental discretion, and healthy children were inoculated with two doses of 10mcg monovalent BNT162b2 at days 0 and 21. Immunocompromised children all received two 10mcg doses of monovalent BNT161b2 at days 0 and 21, and the inclusion of a third dose of 10mcg monovalent BNT162b2 as part of the primary vaccination regime, 2 months after dose 2, was based on discretion of the primary physician looking after the child. The interval between the first and second dose was extended to 3 months if the child acquired natural SARS-CoV-2 infection after dose 1. Among children who completed their primary vaccination regime, several received a booster dose at 5 months or later after the last dose, also at parental discretion for healthy children, and physician discretion for immunocompromised children. Healthy children were recruited between 20 December 2021 and 8 March 2022, while immunocompromised children were recruited between 19 December 2021 and 1 April 2022.

All children underwent venipuncture 10 days after dose 1, and at 3, 6 and 12 months after the last dose of primary vaccination. In addition, immunocompromised children who received 3 doses for primary vaccination also underwent venipuncture 6 weeks after dose 2.

Patients were scored for total immunosuppression using a total immunosuppression score developed by the authors and largely based on literature (2, 3). This score is calculated by first assigning each immunosuppressive agent a score per dose by weight (mg/m²/day), and then summing together the scores for all immunosuppressive agents that each subject was on at recruitment ( Supplementary Table 2 ).

Parents were trained to administer a SARS-CoV-2 ART testing at home if the child developed symptoms suggestive of COVID-19 anytime during the follow up period, according to prevailing national guidelines for active surveillance and community treatment during the COVID-19 pandemic in Singapore. ART kits were certified by the Health Sciences Authority of Singapore and freely distributed by the Ministry of Health. ART+ symptomatic SARS-CoV-2 infections were reported to the study team within 72 hours, and a symptom diary was filled up by the study team based on a phone interview conducted within 72 hours.

T cell responses were quantified with a cytokine release assay, a validated method of quantifying T cell responses that has good correlation to T cell ELISPOT (1, 7). Fresh peripheral blood was stimulated with 55 overlapping 15-mer peptide pools covering the immunogenic regions of the Wuhan-Hu-1 SARS-CoV-2 S protein (representing 40.5% of the whole S protein) (GenScript) before and after vaccination. Similar overlapping peptide pool of SARS-CoV-2 N protein (spanning the entire N protein) was also used to stimulate peripheral blood to test for prior SARS-CoV-2 infection. Freshly drawn whole blood was mixed with RPMI and stimulated with the indicated N or S peptide pool at 2ug/ml, or with 1.25% DMSO as a control. After 16 hours of incubation, the supernatant (plasma) was collected and stored at -80°C until analysis. Cytokine concentrations in the plasma was quantified using an Ella machine (ProteinSimple) with microfluidic multiplex cartridges that measured Th1 specific cytokines interferon-γ (IFN-γ) and interleukin-2 (IL-2) for both adults and children. In addition, Th2 specific cytokines interleukin-4, interleukin-5 and interleukin-13, and other cytokines tumor necrosis factor-α, Granzyme-B and interleukin-10 were quantified for children, according to the manufacturer’s instructions (ProteinSimple). The levels of cytokines present in the plasma of DMSO controls were subtracted from the corresponding N or S stimulated samples. Technical duplicates were used for this assay.

Peripheral blood was collected from all individuals in heparin-containing tubes, and PBMCs from all collected blood samples were isolated by Ficoll-Paque density gradient centrifugation. PBMCs were cryopreserved in liquid nitrogen until analysis.

S+ MBCs were quantified using flow cytometry as previously described (1). Briefly, thawed PBMCs were first enriched for B cells using pan B cell isolation kit (Miltenyi, Germany), according to manufacturer’s guidelines. Biotinylated full-length Wuhan-Hu-1 S proteins (Miltenyi, Germany) were incubated with fluorescently labeled streptavidin (SA) for 15 minutes at room temperature (1). Cells were stained with an antibody cocktail containing CD3, CD19, CD21, CD27, CD38, CD138, CD71, IgA, IgG, IgD, IgM and 7-AAD for 30minutes at 4°C prior to acquisition on the LSR Fortessa flow cytometer (BD). S+ MBC were defined as live CD3-CD19+IgD-CD27+CD38-/+S bispecific cells, while S+ PB were defined as live CD3-CD19+IgD-CD27+CD38++/S bispecific cells.

N-specific antibodies, anti-S IgG, as well as N-reactive T cell responses, were used to identify subjects who were asymptomatically infected. N-reactive T cell responses were selected as a large proportion of SARS-CoV-2-convalescent individuals develop T cell responses against N, and its absence among pre-pandemic donor samples demonstrates its low cross-reactivity with seasonal coronaviruses (8). At each time point, any 1) newly positive anti-N antibodies, 2) 4-fold rise in anti-S IgG in the absence of vaccination, or 3) significant increase in IFN-γ or interleukin-2, were taken to represent an interim asymptomatic SARS-CoV-2 infection. Significant increase in IFN-γ or interleukin-2 was defined as either 1) 10-fold increase of cytokine level from baseline, 2) 10-fold increase of cytokine level from the last visit, or 3) cytokine level 10 times above the threshold of positivity. As there is non-specific T cell activation up to 4 weeks post vaccination, N-reactive T cell responses at Day 10 post dose 1 were not used for identification of asymptomatically infected individuals (7). As N-reactive T cell responses were not used as an exclusion criterion for enrolment in this study, children with asymptomatic SARS-CoV-2 infections as evidenced by high N-reactive T cell responses as baseline were included in the study; these were excluded when studying vaccine-only immunity but included when studying hybrid immunity.

Clinical data was collected in REDCap version 14.2.2, and exported for analysis into Microsoft Excel. Python 3.9.2 was used to combined immunological parameters and clinical data. Statistical analyses were conducted using GraphPad Prism 10.0.2 and IBM SPSS 29. Two-tailed Mann-Whitney U test was used for comparison of unpaired continuous data between two groups. Spearman correlation was used to analyze the association between non-parametric continuous variables. Logrank (Mantel-Cox) test was used to compare groups for Kaplan-Meier survival curves. p<0.05 level of confidence was accepted for statistical significance. Figures were created using GraphPad Prism 10.0.2 and BioRender. All box and whiskers plots show median (center line), interquartile range (box) and range (whiskers).

Immunocompromised children (n = 19) aged 5-12 years were recruited between 19 December 2021 and 1 April 2022 and followed up until 1 year after the end of the primary vaccination regime ( Figure 1A ). These children had neither clinical history, nor serological evidence of prior SARS-CoV-2 infection. Immunocompromised children were either post-solid organ transplantation, had autoimmune disease or were on current or recent (completed less than 3 months ago) chemotherapy for acute lymphoblastic leukemia ( Supplementary Table 1 ). Most of the immunosuppressive medications were T cell immunosuppressants, with total immunosuppression score ranging from 0.2 to 9.7. The healthy children cohort (n = 116) were of similar ages and enrolled during the same period as the immunocompromised children, as previously described ( Figure 1A ) (1).

All healthy children received 2 doses of monovalent 10 mcg BNT162b2 21 days apart. Among the immunocompromised children, 15 (79%) received 3 doses of 10mcg BNT162b2 mRNA SARS-CoV-2 vaccine, while 4 (21%) received only 2 doses within 3 months of the first dose ( Figure 1A ). 1 child (chemotherapy group) dropped out of the study after receiving 2 doses of vaccine, and thus did not contribute data after the post-dose 1 time point. Otherwise, no subjects were excluded from analyses at any time point. Post-dose 2 data were combined from the 4 immunocompromised children who received a 2-dose primary vaccination regime, and 15 children who received a 3-dose primary regime, for both “post-dose 2” and “pre-dose 3” analyses. Several healthy and immunocompromised children went on to receive a 3rd and 4th dose, respectively, between month 6 and 12, according to parental discretion ( Figure 1A ).

We first compared the adaptive immune response to SARS-CoV-2 spike (S) protein after each of the first 2 doses of BNT162b2 between immunocompromised and healthy children. At each time point, children who acquired antigen rapid test positive (ART+) symptomatic SARS-CoV-2 infection, or asymptomatic SARS-CoV-2 (evidenced by serological or T cell responses), were excluded from analysis so that all comparisons were in children with vaccine-only immunity. After 2 doses, immunocompromised children had significantly lower anti-S IgG antibody titers and neutralizing antibodies titers against Wuhan-Hu-1 – measured by surrogate virus neutralization assay (sVNT) – compared to healthy children ( Figures 1B, C ). As we previously demonstrated that full protection against symptomatic BA.2 infection required a pVNT50 titer of 1000, we applied this titer as the protective pVNT50 threshold (1). We found no significant difference in pVNT50 titers against Beta and Delta variants, while pVNT50 titers were below the threshold of protection for all Omicron subvariants ( Figure 1D ).

To determine the impact of 2-dose vaccination on S+ MBCs and S-reactive T cell responses, we used the 80% sensitivity point on our receiver operating characteristic (ROC) curve that predicted protection against symptomatic SARS-CoV-2 infection (1). S+ MBCs were comparable after dose 2 in both cohorts ( Figure 1E ). However, S-reactive T cell responses (measured by interferon (IFN)γ using a previously reported cytokine release assay (1, 7), were significantly lower in immunocompromised children after 2 doses ( Figure 1F ). Unlike in healthy children where 60% achieved the threshold of protection, only 20% of immunocompromised children reached this threshold ( Figure 1F ). While age, weight and gender did not correlate with the measured immune responses, T cell response at day 10 post-dose 1 significantly correlated with total immunosuppression score ( Figure 1G , Supplementary Figures 1, 2 , Supplementary Table 2 ). This suggests that the reduced immunogenicity observed was more applicable to more severely immunosuppressed children.

Since a two-dose primary vaccination regime for immunocompromised children failed to produce S+ MBC and S-reactive T cells that correlated with protection, we analyzed the effect of the third dose by comparing the pre- to post-dose 3 parameters ( Figures 1H, J ). Expectedly, sVNT50 titers were significantly lower than that in healthy children ( Figure 1H ). As in healthy children, BA.2 pVNT50 titers remained below the protective threshold for the majority (67%) of immunocompromised children ( Figure 1I ). However, S-reactive T cell responses reached levels comparable to healthy children and importantly, above the threshold of protection ( Figure 1J ). Taken together, a third dose given as part of the primary vaccination regime to immunocompromised children in our cohort significantly improved vaccine immunogenicity.

Coincident with our study, community transmission of COVID-19 in Singapore fortuitously switched from low to high in December 2021, when Omicron became the predominant variant. In both healthy and immunocompromised children who received at least 1 dose of vaccine, proportion of asymptomatic infections was similar (26% in healthy, 30% in immunocompromised); symptomatic SARS-CoV-2 infections were all mild.

With endemicity of SARS-CoV-2, hybrid immunity will be the most frequent type of protection against COVID-19 globally. We thus compared hybrid immunity between immunocompromised and healthy children by using 12 immunocompromised children who had hybrid immunity at month 3 after 3 doses of vaccine and acquired one to two episodes of natural infection, and 54 healthy children who had hybrid immunity at month 3 after 2 doses of vaccine and acquired one to two episodes of natural infection. Children who acquired symptomatic/asymptomatic SARS-CoV-2 infection, or received booster doses of vaccine before each time point were excluded from analysis to allow us to assess the kinetics of hybrid immunity in the absence of further antigenic encounter.

Healthy and immunocompromised children demonstrated comparable anti-S IgG, neutralizing antibody titers, S+ MBCs and S-reactive T cell responses at month 3 ( Figures 2A–E ). However, by month 12, only 58% of immunocompromised children still had BA.2 neutralizing titers above the threshold of protection ( Figure 2F ). As we have previously shown that in the absence of protective titers of neutralizing antibodies, MBC and T cell responses are strongly predictive of protection against symptomatic infection, we analyzed the durability of these responses. Immunocompromised children demonstrated significantly more decline in S+ MBC and T cell responses than healthy children by month 12 ( Figures 2G, H ); the majority (84% for MBC and 79% for T cells) had MBC and T cell responses below the threshold of protection ( Figures 2G, H ). Taken together, we observed that, despite an initial robust formation of hybrid immun ity in immunocompromised children, the durability of immunological memory reduced significantly. This suggests that a booster dose given between 6 and 12 months after completion of primary vaccination regime may be beneficial in this population.