Abstract
Post vaccine immunity following COVID-19 mRNA vaccination may be driven by extrinsic, or controllable and intrinsic, or inherent health factors. Thus, we investigated the effects of extrinsic and intrinsic on the peak antibody response following COVID-19 primary vaccination and on the trajectory of peak antibody magnitude and durability over time. Participants in a longitudinal cohort attended visits every 3 months for up to 2 years following enrollment. At baseline, participants provided information on their demographics, recreational behaviors, and comorbid health conditions which guided our model selection process. Blood samples were collected for serum processing and spike antibody testing at each visit. Cross-sectional and longitudinal models (linear-mixed effects models) were generated to assess the relationship between selected intrinsic and extrinsic health factors on peak antibody following vaccination and to determine the influence of these predictors on antibody over time. Following cross-sectional analysis, we observed higher peak antibody titers after primary vaccination in females, those who reported recreational drug use, younger age, and prior COVID-19 history. Following booster vaccination, females and Hispanics had higher peak titers after the 3rd and 4th doses, respectively. Longitudinal models demonstrated that Moderna mRNA-1273 recipients, females, and those previously vaccinated had increased peak titers over time. Moreover, drug users and half-dose Moderna mRNA-1273 recipients had higher peak antibody titers over time following the first booster, while no predictive factors significantly affected post-second booster antibody responses. Overall, both intrinsic and extrinsic health factors play a significant role in shaping humoral immunogenicity after initial vaccination and the first booster. The absence of predictive factors for second booster immunogenicity suggests a more robust and consistent immune response after the second booster vaccine administration.
Introduction
Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) infection has had rapid and widespread international effects since its onset in late December 2019 (1–5). Aside from the socioeconomic burdens and turmoil generated from its spread, early access to health interventions and prophylactic measures for the virus were limited and ineffective at controlling the spread. Instead, government-imposed lockdowns and social distancing were utilized as preventative measures (4), but due to high transmissibility and virulence the U.S. Food and Drug Administration (FDA) granted emergency use authorization (EUA) for nearly 400 coronavirus disease-19 (COVID-19) related medical products in the United States, including diagnostic tests, medical devices, drugs, and vaccines (4, 6–8), and helped distribute vaccines worldwide.
Prior work has shown that post-vaccine immunity is driven by various factors, including prior SARS-CoV-2 infection, vaccine manufacturer, heterologous booster administration, time since vaccination, and the magnitude of peak antibody titers after vaccination (9–18). Furthermore, both systemic and local symptoms following SARS-CoV-2 infection and vaccination have been associated with higher peak antibody responses following primary vaccination (5). Overall, COVID-19 vaccines and boosters have been demonstrated to safely curb infection and severe SARS-CoV-2 outcomes, but more work is needed to understand drivers for post-vaccination heterogeneity. The effect of intrinsic health factors (IHFs) and extrinsic health factors (EHFs) on COVID-19 vaccine immunogenicity is unknown. IHFs are characteristics that relate to an individual and cannot be changed or manipulated (e.g., age, biological sex, or race), while EHFs relate to health comorbidities and behaviors (e.g., diet, smoking, or drug consumption) that can be altered or controlled.
Accounting for IHFs and EHFs when evaluating post-vaccine immunogenicity in large populations could result in more personalized analyses from a public health approach (5). For example, prior research has shown that individuals with diabetes mellitus and obesity exhibit dampened immune responses to COVID-19 and hepatitis B vaccines, compared to controls (19, 20). Apart from chronic disease, age, and biological sex, human leukocyte antigen types have also been identified as factors that can cause heterogeneous immune reactions to vaccination due to senescence and variable cytokine production among different groups (21–24). Despite these findings, the evaluation of IHFs and EHFs have yet to be determined in a longitudinal cohort following COVID-19 vaccination. Thus, the purpose of this study is to investigate the effects of IHFs and EHFs on peak antibody (Ab) titers following COVID-19 primary vaccination (PV) and booster vaccination (BV), and to investigate how these factors influence Ab magnitude and trajectory over time.
Methods
Study design
Participants were enrolled in our Institutional Review Board (IRB) approved (#20201026), longitudinal, observational SARS-CoV-2 immunity study (n=228) known as “CITY” (COVID-19 ImmuniTY study) as part of the PARIS/SPARTA study (25). Recruitment for the study began in October 2020 and ended in February 2023. The study was conducted at the University of Miami Miller School of Medicine, and study subjects participated in visits every 3 months for a total of 2 years from the time of enrollment. The primary objective of the CITY study was to characterize differential antibody kinetics among SARS-CoV-2 uninfected and infected individuals in a high-risk, ethnically diverse cohort. The “high-risk” designation for inclusion referred to hazardous occupational exposure (e.g., healthcare workers) but also to advanced age or other sociodemographic characteristics known to increase risk of SARS-CoV-2 morbidity and mortality. Following written informed consent, participants provided demographic details to include social and lifestyle habits and relevant past medical history that would preclude them to more severe SARS-CoV-2 infection-related outcomes. They also provided details regarding their past medical history to include conditions like hypertension, hypercholesterolemia, previous SARS-CoV-2 infection, diabetes, and COVID-19 vaccine manufacturer, lifestyle factors like alcohol or drug consumption, and other details such as age, biological sex, educational level, employment status, marital status, race, ethnicity, and sexual orientation. Those who suffered from SARS-CoV-2 infection with a documented positive nucleic acid amplification test (NAAT+) prior to entry provided evidence details regarding their past SARS-CoV-2 infection associated symptoms during the baseline visit to the study team.
At all regularly scheduled visits, participants prospectively answered symptom questionnaires to screen for new or recent SARS-CoV-2 infection, and blood samples were collected for serum and peripheral blood mononuclear cell (PBMC) processing. Plasma was stored at -80°C and PBMCs were cryopreserved in liquid N2 (26). All participants agreed to sample banking for future research use (Fig 1).
SARS-CoV-2 ELISAs
SARS-CoV-2 enzyme-linked immunosorbent assays (ELISAs) were performed according to a well described assay developed by the Icahn School of Medicine at Mount Sinai (27, 28). 96-well plates were collected at 4°C with wild-type Wuhan SARS-CoV-2 spike protein (2 mg/ml) solution and incubated overnight. Plates were blocked with 3% non-fat milk prepared in phosphate buffered saline with 0.1% Tween 20 (PBST) and incubated at room temperature for an hour. After blocking, serial dilutions of heat-inactivated serum samples were added to the plates and incubated for two hours at room temperature. Plates were washed in triplicate with 0.1 % PBST and 50 µl of a 1:3,000 dilution of goat anti-human immunoglobulin G-horseradish peroxide (HRP) conjugated secondary antibody was added. After an hour of incubation, 100 µl o-phenylenediamine dihydrochloride (SIGMAFASTTM-OPD, Sigma Aldrich) solution was added to each well for 10 minutes. The reaction was stopped by the addition of 50 µl of 3 M hydrochloric acid at each well. The optical density at 490 nm (OD490) was measured using a Synergy 4 (BioTek) plate reader. The background value (OD490 = 0.15) allowed to report discrete antibody endpoint titers in the values of 1:100, 1:200, 1:400, 1:800, 1:1600, 1:3200, 1:6400, 1:12800, 1:25600, 1:51200, 1:102400, and 1:204800. The limit of detection was set at 1:100.
Statistical analyses
For all analyses, regardless of the statistical approach, Ab titers were log2 transformed. Intrinsic and extrinsic factors (Table 1) were included in the cross-sectional and longitudinal analyses detailed below.
Cross-sectional analysis
For the cross-sectional analysis, all respondents had participated in additional visits with associated Ab titers measured within 2 weeks [± 5 days] following primary and booster vaccination. Cross-sectional analysis was conducted separately for post-vaccine timepoints following PV, BV1, and BV2. We included only those (n = 118) who had received their primary vaccination (i.e., two doses of Pfizer BNT162b2 or two doses of Moderna mRNA-1273) after their inclusion into the study and strictly within the two-dose schedule recommended by the Centers for Disease Control and Prevention (CDC). This approach minimized potential discrepancies in vaccination effects based on administration time.
The underlying assumptions for parametric tests (e.g., tests for homogeneity of variance and normality) were conducted for Ab titers among IHF and EHF groups following each post-vaccination sub cohort of interest (i.e., PV1, BV1, and BV2) prior to conducting hypothesis testing. Differences in mean Ab were analyzed using parametric (one-way analysis of variance (ANOVA), Student’s t-test) or non-parametric (Games-Howell, Welch’s t-test) tests, based on Levene’s test results. Lastly, Pearson’s correlation was used to examine the correlation between age and peak Ab titer. All statistical analyses were conducted via Minitab (Minitab® 21.3.1) at a significance level of α = 0.05, and tests for groups with sample sizes <2 were excluded from analysis.
Longitudinal analysis
The selection of covariates of interest for the longitudinal analysis was informed by the aforementioned cross-sectional analysis conducted at post-primary vaccination timepoints. Preliminary analyses utilized a range of potential predictors to evaluate their association with the response variable, log2 Ab titers over time. From this subset, only those predictors that exhibited a significant association log2 Ab were carried forward as covariates of interest for the longitudinal analysis. This method ensured that the longitudinal model was informed by empirically significant predictors, thus increasing the potential explanatory power and relevance of the model to the observed data. Other proven covariates of interest (5), including COVID-19 vaccine manufacturer, were included during the model generation phase as well.
Following the above, linear mixed-effects models were constructed to investigate the growth curve of log2 antibody titers over time following primary vaccination (Dose 1 and 2), as well as booster vaccination (Dose 3 and Dose 4) utilizing the nlme and lme4 package in R Studio. The models were designed to incorporate both fixed and random effects, with the fixed effects representing the average population response and the random effects capturing subject-specific deviations. The process of model generation was initiated by constructing a maximal model, where the time since vaccination – represented as a continuous variable – was incorporated as additional polynomial terms to account for potential non-linear growth patterns. This maximal model considered the most complex growth trajectory from the outset.
To account for the repeated measures structure of the data, various correlation structures including compound symmetry, first-order autoregressive, and general autoregressive were compared. Model comparison was performed using the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC), with the model yielding the lowest values being selected as the optimal model.
Results
Cohort characteristics
Extrinsic health factors - summary
As seen in Table 1, among those enrolled (n = 228), the average age in the “CITY” cohort was 48.6 ± 16.56 years, with 66 (34.74%) participants classified as young, 84 (44.21%) as middle aged, and 40 (21.05%) as elderly (n = 190).
Table 2 displays a summary table for the cohort extrinsic health factors. At baseline, n = 180 (78.95%) participants self-reported as alcohol consumers and 47 (20.61%) did not (n = 227). There were 9 (3.96%) participants with a history of cardiovascular disease (CVD), 216 (95.16%) with no history of CVD, and 2 (0.88%) that were unsure (n = 227). Also, 9 (3.96%) participants were diabetic, while 215 (94.72%) were not, and 3 (1.32%) were unsure (n = 227). Moreover, 16 (7.08%) participants were recreational drug users and 210 (92.92%) were not (n = 226). Among the cohort, 39 (17.18%) of the participants had hypercholesterolemia, while 173 (76.21%) did not, and 14 (6.17%) were unsure (n = 227). Lastly, 37 (16.30%) participants reported hypertension, 185 (81.50%) did not, and 5 (2.20%) were unsure (n = 227).
Ninety-nine (43.42%) participants indicated previous SARS-CoV-2 infection at entry, while 126 (55.26%) did not, and 3 (1.32%) were unsure (n=228). Of the 163 participants in the cohort that received PV, 8 (4.91%) received Johnson & Johnson Ad26.COV2.S, 58 (35.58%) received Moderna mRNA-1273, and 97 (59.51%) received Pfizer BNT162b2. Of the 97 participants that received the first booster (BV1), 17 (17.53%) received the full Moderna mRNA-1273 dose, 24 (24.74%) received the half-dose Moderna mRNA-1273, and 56 (57.73%) received Pfizer BNT162b2. Of those 32 participants that received a second booster (BV2), 16 (50%) received a full Moderna mRNA-1273 dose, 1 (3.12%) received the Moderna mRNA-1273 half dose, and 15 (46.88%) received a Pfizer BNT162b2 booster. No bivalent boosters were included in the analysis.
Intrinsic health factors - summary
Table 3 displays a summary table for the cohort intrinsic health factors. One hundred and thirty-seven (60.00%) participants were female and 91 (40.00%) were male (n = 228). Furthermore, it was reported that 23 (10.10%) participants had an associate degree or a technical degree, 49 (21.49%) had a bachelor’s degree, 12 (5.26%) had a high school diploma or equivalent, 3 (1.32%) had less than a high school diploma, 111 (48.68%) had a master’s degree or higher, and 29 (12.72%) had other educational levels (n = 227).
Additionally, 2 (0.87%) participants were disabled, 188 (82.46%) were employed, 24 (10.54%) were retired, and 12 (5.26%) were unemployed (n=226). Nineteen (8.33%) were divorced, 11 (4.82%) participated in a domestic partnership, 113 (49.57%) were married, 79 (34.65%) were single, and 6 (2.63%) were widowed (n = 228). In addition, 12 (5.26%) participants identified as Asian, 12 (5.26%) as African American, 183 (80.27%) as White, and 21 (9.21%) as part of another race (n = 228). Ninety (39.47%) participants identified as Hispanic, and 138 (60.53%) participants identified as non-Hispanic. Lastly, 3 (1.32%) participants identified as bisexual, 15 (6.58%) as gay, 206 (90.35%) as heterosexual, 3 (1.32%) as lesbian, and 1 (0.43%) as another sexual orientation (n = 228).
Primary vaccination
Overall, 148 participants were included in the PV sub-cohort, with an average peak titer of 13.29 ± 1.67 following vaccination. Table 4 details all significant differences observed across extrinsic and intrinsic health factor covariates in the PV sub-cohort or that had significance in the BV1 and BV2 sub-cohorts. Other measures examined for the purpose of this analysis are documented in S1 Table.
Extrinsic health factors
Five (3.42%) individuals included in the PV sub-cohort had CVD. while the remaining 141 (96.58%) did not. The mean log2 Ab of participants with CVD (12.84 ± 3.11) and without CVD (13.29 ± 1.59) were not statistically different (T = 0.59; p = 0.555). Four who received PV (2.72%) had diabetes and 143 (97.28%) did not. The mean log2 Ab of participants with diabetes (12.89 ± 2.87) and participants without diabetes (13.31 ± 1.61) were not statistically different (T = 0.50; p = 0.619). Twenty-seven (19.42%) had hypercholesterolemia while 112 (80.58%) did not. The mean log2 Ab of participants with hypercholesterolemia (13.01 ± 1.96) and without hypercholesterolemia (13.31 ± 1.60) were not statistically different (T = 0.83; p = 0.405). Furthermore, 22 (15.07%) had hypertension and 124 (84.93%) did not have hypertension. The variances of log2 Ab between participants with hypertension and participants without hypertension were statistically different, (F (1,145) = 8.15; p = 0.005) and indicated that individuals without hypertension had lower and more homogenous Ab response after PV. However, the mean log2 Ab of participants with hypertension (12.69 ± 2.36) and participants without hypertension (13.34 ± 1.50) were not statistically different (T = 1.24; p = 0.225).
Fifty-four (36.49%) of those that received PV reported prior SARS-CoV-2 infection at study entry and 94 (63.51%) did not. The variances of log2 Ab between participants with a history of COVID-19 and no history of COVID-19 were statistically different (F (1, 147) = 4.92; p = 0.028), as were the peak responses among participants with a COVID-19 history (14.39 ± 1.15) and no history of COVID-19 (12.62 ± 1.59) (T = - 7.76; p < 0.005). Furthermore, 149 participants received PV, where 54 (36.24%) received a Moderna vaccine and 95 received a Pfizer BNT162b2. The mean log2 Ab of participants with the Moderna mRNA-1273 vaccine (13.50 ± 1.71) and the Pfizer BNT162b2 vaccine (13.14 ± 1.64) were not statistically different, as shown in Fig 2 (T = 1.26; p = 0.211). Of the 148 participants who received PV, 119 (80.41%) were alcohol consumers and 29 (19.59%) were not alcohol consumers. The mean log2 Ab of alcohol consumers (13.39 ± 1.58) and non-alcohol consumers (12.75 ±1.97) were not significantly different (T = - 1.87; p = 0.063) after PV. Moreover, of the 148 participants that received PV, 9 (6.08%) were recreational drug users and 139 (93.92%) were not. The mean log2 Ab of recreational drug users (14.31 ± 0.70) and non-users (13.20 ± 1.70) were statistically different (T = -1.95; p = 0.050) after PV.
Intrinsic health factors
Of the 149 participants that received PV, 55 (36.91%) of them were young, 60 (40.27%) were middle aged, and 34 (22.82%) were elderly. The mean log2 Ab of low age participants (13.57 ± 1.41; Tukey = A) was statistically different (Fig 3) than elderly participants (12.61 ± 1.78 ; Tukey = B), but not middle-aged participants (13.36 ± 1.75 ; Tukey = A, B) after PV (F = 3.73 ; p = 0.026). Pearson correlation demonstrated a negative correlation between age and log2 Ab after PV (r = -0.206; p = 0.012). Additionally, 86 (57.72%) participants were female and 63 (42.28%) were male. The variances of the log2 Ab between female and male participants were significantly different (F (1,148) = -4.68; p = 0.032) after PV. Furthermore, male participants exhibited a lower (12.87 ± 12.87) was statistically lower (T = 2.47; p = 0.015) than that of female (13.56 ± 1.46) participants after PV (Fig 4).
Furthermore, of the 149 participants that received PV, 12 (8.05%) reached an education level of an associate degree or technical degree, 30 (20.13%) held a bachelor’s degree, 9 (6.04%) a high school diploma, 75 (50.34%) a master’s degree or higher, and 23 (15.44%) reported reaching an unlisted level. Peak titers were not associated with differences between any group (F = 1.87) ; (p = 0.120), including those with an associate degree or technical degree (13.56 ± 1.44 ; Tukey = A), a bachelor’s degree (13.68 ± 1.69 ; Tukey = A), a high school diploma (13.98 ± 1.58 ; Tukey = A), a master’s degree or higher (13.16 ± 1.64 ; Tukey = A), and other educational levels (12.64 ± 1.76 ; Tukey = A). One-hundred and twenty-five (84.46%) were employed, 17 (11.49%) were retired, and 6 (4.05%) were unemployed. The mean log2 Ab of employed (13.36 ± 1.63, Tukey = A), retired (12.35 ± 1.96; Tukey = A), and unemployed (13.98 ± 0.82, Tukey = A) participants were statistically different (F = 3.37 ; p = 0.037) after PV, though the authors speculate that this finding is due to a strong correlation between age and the retired group.
Twelve (8.05%) participants were divorced, 7 (4.70%) participated in a domestic partnership, 81 (54.36%) were married, 47 (31.54%) were single, and 2 (1.35%) were widowed. The post PV Ab between divorced (13.81 ± 1.24, Tukey = A), domestic partnership (13.07 ± 1.51, Tukey = A), married (13.04 ± 1.82, Tukey = A), and single (13.54 ± 1.49, Tukey = A) participants were not statistically different (F = 1.36; p = 0.257) after PV. Only 2 (1.34%) identified as bisexual, 10 (6.71%) identified as gay, 134 (89.94%) identified as heterosexual, 2 identified as lesbian (1.34%), and 1 (0.67%) identified as another sexual orientation. Furthermore, the mean log2 Ab between gay (13.34 ± 0.45) and heterosexual (13.24 ± 0.15) participants were not significantly different (T = 0.19; p = 0.853) after PV.
Also, of the 149 participants that received PV, 9 (6.04%) were Asian, 6 (4.03%) were African American, 14 (9.40%) identified under another race, and 120 (80.54%) were White, though the mean peak response (13.09 ± 1.88, Tukey = A; 13.98 ± 0.82, Tukey = A; 13.43 ± 1.37, Tukey = A; 13.23 ± 1.72, Tukey = A; respectively) were not statistically different (F = 0.46 ; p = 0.712) after PV. Moreover 57 (38.26%) participants identified as Hispanic, and 92 (61.74%) identified as non-Hispanic. The mean log2 Ab between Hispanic (13.57 ± 1.77) and non-Hispanic (13.08 ± 1.58) participants were not statistically different (T = 1.77; p = 0.079) after PV.
Longitudinal analysis following primary vaccination
A linear mixed-effects model (Fig 5) was employed to investigate the fixed effects of days since full vaccination (including linear, quadratic, and cubic time), COVID-19 vaccine manufacturer, COVID-19 status, and biological sex on the log-transformed titer levels (antibody), while accounting for random effects associated with each study participant. The fixed effects of linear, cubic, and quadratic time revealed that days since full vaccination (i.e., >14 days following the 2nd dose) and its second and third powers were negatively associated with log2 antibody titers (estimate = -4.653e^-03; F(1,524) = 96.889; p = <0.001, estimate = -2.208e^-04; F(1,524) = 118.393; p <0.001, and estimate = 9.467e^-07; F(1,24) = 44.654; p < 0.001, respectively). Participants who had been vaccinated with Pfizer BNT162b2 had a lower antibody response over time compared to those who received the Moderna mRNA-1273 vaccine (estimate = -9884e^- 01; F(1,108) = 36.900; p < 0.001).
Additionally, COVID-19 positive participants had higher antibody levels compared to COVID-19 negative individuals (estimate = 1.464; F(1,108) = 64.917; p < 0.001), consistent with well-described trends for hybrid immunity (29). The analysis also suggested a difference in antibody levels based on biological sex, with males having lower levels compared to females (estimate = -3.759e^01; F(1,108) = 4.856; p = 0.030). No other variable was found to significantly influence log2 antibody titers over time (S2 Table).
Booster vaccine I
Following Booster Vaccine I (BV1), 148 participants averaged a mean peak titer of 17.74 ± 1.14. Table 5 details significant differences observed across all extrinsic and intrinsic health factors in the BV1 sub-cohort or that were noted to be significant in the PV and BV2 sub-cohorts. Other measures are documented in S3 Table.
Extrinsic health factors
Of the 95 participants that received BV1, 3 (3.16%) had CVD and 92 (96.84%) did not have CVD. Additionally, the mean log2 Ab of participants with CVD (14.64 ± 1.00) and participants without CVD (14.75 ± 1.15) did not significantly vary (T = 0.16; p = 0.872) after BV1. Three (3.12%) had diabetes and 93 (96.88%) did not, but diabetic (13.98 ± 1.53) and non-diabetic (14.77 ± 1.14) participants did not have observable differences (T = 1.19; p = 0.239) in peak titers. Of the 89 participants that received BV1, 23 (25.84%) had hypercholesterolemia and 66 (74.16%) did not. Eighteen (18.95%) participants had hypertension and 77 (81.05%) did not. Among participants with hypercholesterolemia (14.73 ± 0.95) or without (14.81 ± 1.25), and hypertension (14.81 ± 0.22) or without (14.75 ± 1.12), we did not observe differences in peak BV1 response (T = 0.28; p = 0.781, T = -0.21; p = 0.836).
Thirty (30.61%) participants who received BV1 were previously infected with SARS-CoV-2 and 68 (69.39%) were not. Furthermore, the mean log2 Ab of participants with a COVID-19 history (14.81 ± 1.02) and no history (14.70 ± 1.20) were not statistically different (T = -0.46; p = 0.649) after BV1. In the BV1 sub-cohort, 17 (17.52%) received a full-dose Moderna mRNA-1273 vaccine, while 24 (24.74%) received a half-dose Moderna mRNA-1273 vaccine, and 56 (56.74%) received a Pfizer BNT162b2 vaccine, but the mean log2 Ab in participants with full-dose Moderna mRNA-1273 (14.82 ± 1.55 ; Tukey = A), half-dose Moderna mRNA-1273 (15.14 ± 0.78 ; Tukey = A), and Pfizer BNT162b2 (14.54 ± 1.10 log2 Ab ; Tukey = A) did not have significantly distinct peak responses (F = 2.51 ; p = 0.087).
Interestingly, the Pfizer BNT162b2 BV1 group exhibited a very homogeneous peak response following vaccination (14.54; SE = 0.02) while ultimately it was not statistically different from the other groups. Of the 97 participants that received BV1, 74 (76.29%) were alcohol consumers and 23 (23.71%) were not. The peak response of alcohol consumers (14.85 ± 1.01) and non-consumers (14.43 ± 1.48) were not significantly different (T = -1.55; p = 0.124). Four (4.12%) participants were recreational drug users and 93 (95.88%) were not, and the peak response was not significantly different (T = -0.26; p = 0.794) between drug users (14.89 ± 0.96) and non-users (14.74 ± 1.15).
Intrinsic health factors
Twenty-four (24.49%) young participants were in the BV1 sub-cohort, while 43 (43.88%) were of middle age, and 34 (34.69%) were elderly. The peak response among different age groups was not significantly different (F = 1.30; p = 0.277), including young (15.06 ± 0.97; Tukey = A), middle-aged (14.62 ± 0.96, Tukey = A), and elderly participants (14.46 ± 1.44; Tukey = A). Of the 98 participants that received BV1, 50 (51.02%) were female and 48 (49.98%) were male, and a significantly lower peak response (T = -2.05; p = 0.043) was observed in males (14.50 ± 1.01) in comparison to females (14.96 ± 1.22).
Of the 98 participants that received BV1, 10 (10.20%) obtained an associate degree or technical degree, 21 (21.43%) a bachelor’s degree, 2 (2.04%) a high school diploma, 48 (48.98%) a master’s degree or higher, and 17 (17.35%) reported an unlisted level. The mean log2 Ab between participants with an associate degree or technical degree (14.84 ± 1.62; Tukey = A), a bachelor’s degree (14.64 ± 1.38; Tukey = A), a master’s degree or higher (14.81 ± 0.91; Tukey = A), and other educational levels (14.64 ± 1.23; Tukey = A) were not statistically different (F = 0.17; p = 0.916). Seventy-eight participants (80.41%) were employed, 14 (14.43%) were retired, and 5 (5.16%) were unemployed. We observed that the variances of log2 Ab between employed, retired and unemployed participants were statistically different (F (2,147) = 4.41; p = 0.015) after BV1, though the peak responses were not significantly different (F = 0.66; p = 0.541) between these groups (14.82 ± 0.94, Games-Howell = A; 14.29 ± 1.91, Games-Howell = A; 15.04 ± 0.89; Games-Howell = A; respectively).
In the BV1 sub-cohort, 8 (8.16%) participants were divorced, 4 (4.08%) participated in a domestic partnership, 61 (62.24%) were married, 22 (22.45%) were single, and 3 (3.07%) were widowed. The mean log2 Ab between divorced (15.02 ± 1.06; Tukey = A), domestic partnership (14.39 ± 0.96; Tukey = A), married (14.68 ± 1.22 ; Tukey = A), single (14.83 ± 1.01; Tukey = A), and widowed (14.98 ± 1.16; Tukey = A) participants were not statistically different (F = 0.31; p = 0.869) after BV1. One (1.02%) identified as bisexual, 6 (6.12%) identified as gay, 90 (91.84%) identified as heterosexual, and 1 identified as lesbian (1.02%), but the mean log2 Ab between gay (14.31 ± 0.82) and heterosexual (14.74 ± 1.15) participants were not significantly different (T = -0.90; p=0.368) following BV1.
Finally, of the 98 participants that received BV1 8 (8.16%) were Asian, 3 (3.06%) were African American, 5 (5.11%) identified under another race, and 82 (83.67%) were White. The peak responses between participants among different races were not statistically different (F = 0.41; p = 0.747), including Asian (14.77 ± 0.84 ; Tukey = A), African American (15.32 ± 0.58; Tukey = A), Other (15.04 ± 0.55 ; Tukey = A), and White (14.69 ± 1.21; Tukey = A). Of the 98 participants that received BV1, 34 (34.69%) identified as Hispanic, and 64 (65.31%) identified as non-Hispanic. The mean log2 Ab between Hispanic (14.97 ± 0.98) and non-Hispanic (14.61 ± 1.21) participants were not statistically different (T = 1.47; p = 0.144) after BV1.
Longitudinal analysis following booster vaccination I
Linear mixed effects models were generated to examine the effect of IHFs and EHFs on log2 antibody titers over time following BV1 (Fig 6). Overall, we found that the quadratic (estimate = -6.657e-04; F(1, 72) = 4.007; p = 0.049) and cubic (estimate = 1.827e-06; F(1, 72) = 10.142; p = 0.002) terms for days elapsed since the first booster were observed to be a significant predictor of log2 antibody titers over time, elucidating a non-linear decay pattern following BV1. Curiously, this analysis also revealed that drug use positively impacted log2 antibody titers, (estimate = 2.527e+00; F(1, 24) = 8.917; p = .006). Following post-hoc testing, booster type (i.e., Pfizer BNT162b2 vs. Moderna mRNA-1273 (full dose) vs. Moderna mRNA-1273 (half dose) was found to significantly affect the trajectory of log2 antibody titers, (estimate = 3.997416e-01; F(2, 24) = 4.755; p = .018), with those who received a half-dose of the Moderna mRNA-1273 booster displaying increased log2 antibody titers boasting significantly higher titers than who received the Pfizer BNT162b2 booster.
Notably, factors such as ethnicity, gender, age at entry, race, and the linear term for the number of days since the first boost did not significantly contribute to the variation in log2 antibody titers, all p > .05 (S4 Table).
Booster vaccine II
In the cohort, 24 participants received BV2 and averaged 15.06 ± 1.61 Ab following vaccination. Table 6 details the extrinsic and intrinsic health factors that observed a significant difference within groups in the BV2 cohort or that had significance in the PV and BV1 sub-cohorts. Other measures are documented in S5 Table.
Extrinsic health factors
In the BV2 sub-cohort 2 (9.09%) had CVD and 20 (90.91%) did not have CVD, 2 (8.70%) had diabetes and 21 (91.30%) did not have diabetes, and 10 (47.62%) had hypercholesterolemia and 11 (52.38%) did not have hypercholesterolemia. The peak response of participants with hypercholesterolemia (15.54 ± 0.99) and without (14.92 ± 2.01) were not statistically different (T = -0.89; p = 0.383) after BV2. Five (21.74%) had hypertension and 18 (78.26%) did not, but the peak responses between those with hypertension (15.84 ± 0.37) and without (14.92 ± 1.74) were not statistically different (T = -1.13; p = 0.270).
Three (12.50%) participants from the BV2 sub-cohort were infected with SARS-CoV-2 prior to enrollment and 21 (87.50%) were not. The mean log2 Ab of participants with prior SARS-CoV-2 infection (14.98 ± 1.16) and no prior infection (15.07 ± 1.69) were not statistically significant (T = 0.13; p = 0.908). Thirteen (54.17%) received a full-dose Moderna mRNA-1273 vaccine, while 11 (45.83%) received a Pfizer BNT162b2 vaccine, but the peak responses between Moderna mRNA-1273 (15.41 ± 1.74) and Pfizer BNT162b2-vaccinated (14.64 ± 1.41) participants were not significantly different (T = 1.17; p = 0.253). Of the 23 participants that received BV2, 17 (73.91%) were alcohol consumers and 6 (26.09%) were not. The mean log2 Ab of alcohol consumers (14.94 ± 1.21) and non-alcohol consumers (15.64 ± 2.53) were not significantly different (T = 0.91; p = 0.371) after BV2. Two (8.70%) participants were drug users and 21 (91.30%) were not.
Intrinsic health factors
Of the 24 participants that received BV2, 7 (29.17%) were middle aged and 17 (70.83%) were elderly. The peak response of middle-aged participants (15.36 ± 1.50) and elderly participants (14.96 ± 1.69) was not significantly different (T = -0.57; p = 0.573). Ten (41.67%) participants were female and 14 (58.33%) were male, and the peak response in male (15.07 ± 0.42) and female (15.04 ± 1.78) participants was not statistically different (T = -0.04; p = 0.967) after BV2.
Two (8.33%) participants had an associate degree or technical degree, 5 (20.83%) had a bachelor’s degree, 13 (54.17%) had a master’s degree or higher, and 4 (16.67%) had an unlisted degree. The peak responses among participants in this group were not statistically different (F = 0.64, p = 0.539), including those with a bachelor’s degree (14.44 ± 2.39; Tukey = A), a master’s degree or higher (14.03 ± 1.26; Tukey = A), and other educational levels (15.64 ± 1.41; Tukey=A). Of the 24 participants that received BV2, 14 (58.33%) were employed, 9 (37.50%) were retired, and 1 (4.17%) was unemployed. The Ab of employed (15.29 ± 1.50) and retired (14.87 ± 1.86) participants were not statistically different (T = 0.60; p = 0.556) after BV2.
In the BV2 cohort, 3 (12.50%) participants were divorced, 1 (4.17%) participated in a domestic partnership, 16 (66.67%) were married, 2 (8.33%) were single, and 2 (8.33%) were widowed. The mean log2 Ab between divorced (15.98 ± 0.58) and married (14.77 ± 1.75) participants were not statistically different (T = 1.61; p = 0.261). Two (8.33%) identified as gay and 22 (91.67%) identified as heterosexual.
Four (16.67%) identified as Hispanic, and 20 (83.33%) identified as non-Hispanic. The peak response between Hispanic (16.64 ± 0.82) and non-Hispanic (14.74 ± 1.55) participants were statistically different (T = 2.35; p = 0.028) after BV2. Of the 24 participants that received BV2, 2 (8.33%) were Asian, 2 (8.33%) were African American, 2 (8.33%) identified under another race, and 18 (75.01%) were White.
Longitudinal analysis following booster vaccination II
Lastly, linear mixed effects models were used to generate a growth curve of the log2 Ab response following BV2. Again, the quadratic term for the number of days since the 2nd booster emerged as a significant predictor of log2 antibody titers, (estimate = -6.628^e-05; F(1, 104) = 7.916; p = .006), suggestive of a non-linear decay pattern following vaccination. No other covariate of interest was found to significantly affect the trajectory of Ab over time, including linear or cubic time, gender, booster type, or drug use (S6 Table). Overall, these results suggest that a 4th dose of an mRNA COVID-19 vaccine provide an additive effect on the immune response, dampening the role of underlying IHF and EHFs that may influence the post-vaccine immune response.
Discussion
In this work, we sought to evaluate the effects of intrinsic and extrinsic health factors on the peak antibody response as well as characterize antibody trajectory over time following COVID-19 mRNA primary vaccination and booster vaccination. Through cross-sectional analysis, we found that EHFs influenced Ab production after primary vaccination. Participants that reported recreational drug use, including marijuana, cocaine, and amphetamines had marginally higher log2 Ab compared to non-users after PV, but not BV1 or BV2. Previous work has shown that recreational drug use is immunomodulatory, increasing the susceptibility to infectious diseases and microbes while impairing antibody production following vaccination (30–33). Our findings support the opposite, though this warrants further investigation given the small sample size of drug users in our cohort. This was further substantiated in linear mixed-effects models, which demonstrated that participants with reported drug use had higher titers over time only after BV1.
Our cross-sectional analyses were inherently limited in their ability to control for all other variables that may influence humoral responses and were utilized in the context of this work to capture only the peak antibody response. Further, unlike mixed effects models, a cross-sectional approach does not permit the assessment of the impact of individual covariates while controlling for the effect of time elapsed since vaccination. We also found that individuals without hypertension had a lower and more homogeneous Ab response following PV, but not BV1 or BV2, which may be attributed to previously reported changes in T cell immunometabolism (i.e., aberrant T cell activation, differentiation, and proliferation) (34–38). Studies evaluating the effect of hypertension on antibody titers have provided variable results, with some supporting hypertension as a factor driving vaccine immunogenicity (37, 38), and others refuting it (39). We also observed significantly higher mean peak log2 Ab in participants with previous SARS-CoV-2 infection compared to those without a history after only PV, consistent with previous findings (5, 40, 41). Lastly, no significant difference in the peak log2 Ab was found between participants that received Moderna mRNA-1273 and Pfizer BNT162b2 vaccines after vaccination, although this was not the case over time.
Young (low age) individuals had higher peak Ab compared to elderly participants, but not middle-aged participants only after PV, which may be attributed to age-related defects in lymphocytes like mature lymphocyte senescence or decreased production and proliferation of lymphocytes and their associated biomarkers (42–46). Furthermore, we observed significantly lower Ab in retired participants following PV, which likely corresponds to the advanced age of those participants. Also, we found that non-Hispanic participants had lower Ab than Hispanic after BV2, but our sample size is limited and therefore may be limited in generalizability. Lastly, we found that females had significantly higher Ab following PV and BV1, but not BV2. Differences in immune responses based on biological sex may be linked to phylogenetic preservation and subsequent reproductive success associated with more robust humoral immune reaction in females (47). Accordingly, the decreased peak Ab responses we observed in males is supported by previously documented sex-related humoral and cellular-mediated differences (40, 47–49).
The overall findings from the cross-sectional approach indicate, at least in part, the role of EHFs and IHFs in generating differential humoral immunes response following COVID-19 vaccination. Notably, these analyses did not control for other variables, which increases the likelihood of introducing confounding factors in the analysis. Thus, cross-sectional analyses should be considered useful in identifying differences in immune response following vaccination but are perhaps limited in evaluating how these change over time. Additionally, controlling for other factors can help determine more robust relationships between Ab and IHFs/EHFs. Because the immune system follows a dynamic response following activation both through time and due to inter-individual differences, adopting models that account for this and can control for these differences may be used to corroborate cross-sectional analysis. Thus, distinct linear mixed-effects models were used to investigate the factors influencing the log-transformed titer levels following primary and booster vaccination. The number of days elapsed since primary vaccination was negatively associated with antibody titer magnitude, suggesting that as time passes since full vaccination, the titer levels decrease. This time-dependent decay of antibodies was observed to have significant nonlinear components, particularly following BV1 and BV2.
We also found that COVID-19 positive individuals had higher antibody levels following primary vaccination compared to COVID-19 negative individuals, and male participants had lower antibody titers compared to females. The type of vaccine received also affected antibody magnitude and durability; those who received the Pfizer BNT162b2 vaccine had lower titers than those who were given the Moderna mRNA-1273 vaccine. Differential antibody production between Moderna mRNA-1273 and Pfizer BNT162b2 vaccines observed in our cohort are consistent with previous findings, which could be due to varied anti-spike activity generated by each vaccine type (50, 51). We also observed that Moderna mRNA-1273 half-dose recipients had higher post-booster Ab compared to those who received Pfizer BNT162b2 vaccination over time, but this effect was not observed following BV2. Interestingly, post-BV2 Ab were not significantly affected by any predictive factors. This lack of predictors, including vaccine manufacturer, suggest a more robust and homogeneous humoral immune response following BV2.
Conclusions
Overall, this study aimed to examine the effects of intrinsic and extrinsic health factors on Ab following vaccination and through time using cross-sectional and longitudinal models. Differences in antibody titers among the cohort were observed after primary vaccination for individuals with a history of drug use, prior SARS-CoV-2 infection, hypertension, age, and biological sex. Differential titer production was also observed across females following BV1 and Hispanic participants following BV2. Linear-mixed effects models used in this study aimed to control for random and fixed effects like population responses and individual Ab variations over time and revealed trends in the decline of Ab over time following full vaccination, as well as effects by COVID-19 vaccine manufacturer, prior SARS-CoV-2 infection, and biological sex. Drug users and half-dose Moderna mRNA-1273 recipients had higher peak antibody titers over time following the first booster, while no predictive factors significantly affected post-second booster antibody responses. The absence of predictive factors for second booster immunogenicity suggests a more robust and consistent immune response after the second booster vaccine administration.
Data Availability
All relevant de-identified data is available upon request.
Conflict of Interest
Florian Krammer has been consulting for Curevac, Seqirus and Merck and is currently consulting for Pfizer, Third Rock Ventures, Avimex and GSK. He is named on several patent regarding influenza virus and SARS-CoV-2 virus vaccines, influenza virus therapeutics and SARS-CoV-2 serological tests. Some of these technologies have been licensed to commercial entities and Dr. Krammer is receiving royalties from these entities. Dr. Krammer is also an advisory board member of Castlevax, a spin-off company formed by the Icahn School of Medicine at Mount Sinai to develop SARS-CoV-2 vaccines. The Krammer laboratory has received funding for research projects from Pfizer, GSK and Dynavax and three of Dr. Krammer’s mentees have recently joined Moderna. All other authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Funding
This work is part of the PARIS/SPARTA studies funded by the NIAID Collaborative Influenza Vaccine Innovation Centers (CIVIC) contract 75N93019C00051.
Supporting information
S1 Table. Other Measured Extrinsic and Intrinsic Health Factors Among the Primary Vaccination Sub-Cohort. ǂǂǂǂ: Levene’s test found a statistically significant difference in the variances between participants with and without hypertension (F (1,145) =8.15, p=0.005).
S2 Table. Linear Mixed Effects Model (LMM) Evaluating the Relationship Between Post-Primary Vaccination Antibody Titers and Time, COVID-19 Vaccine Manufacturer, Prior COVID-19 Infection Status, and Biological Sex.
S3 Table. Other Measured Extrinsic and Intrinsic Health Factors Among the First Booster Vaccination Sub-Cohort.
S4 Table. Linear Mixed Effects Model (LMM) Evaluating the Relationship Between BV1 Antibody Titers and Time, COVID-19 Vaccine Manufacturer, Prior COVID-19 Infection Status, and Biological Sex.
S5 Table. Other Measured Extrinsic and Intrinsic Health Factors Among the Second Booster Vaccination Sub-Cohort.
S6 Table. Linear Mixed Effects Model (LMM) Evaluating the Relationship Between BV2 Antibody Titers and Time, COVID-19 Vaccine Manufacturer, Prior COVID-19 Infection Status, and Biological Sex.
Acknowledgments
We thank Margaret Roach, Elizabeth Varghese, Maria Pallin, Celeste Sanchez, and Ailet Reyes for their technical contributions. We also thank Aria Nawab and Daniel Muniz for their assistance in sample collection, and Alexander Kizhner for his assistance with data management and in generating the early statistical analysis.