Abstract
Background We studied whether comorbid conditions impact strength and duration of immune responses after SARS-CoV-2 mRNA vaccination in a US-based, adult population.
Methods Sera (pre-and-post-BNT162b2 vaccination) were tested serially up to 6 months after two doses of vaccine for SARS-CoV-2-anti-Spike neutralizing capacity by pseudotyping assay in 91 Veterans and 33 healthcare workers; neutralizing titers were correlated to clinical variables with multivariate regression. In 36 participants, post-booster effect was measured.
Results After completion of the primary vaccine series, neutralizing antibody IC-50 titers were high at one month (14-fold increase from pre-vaccination), declined at six months (3.3-fold increase), and increased at one month post-booster (52.5-fold increase). Age >65 years (β=-0.94, p=0.001) and malignancy (β=-0.88, p=0.002) significantly reduced strength of response at 1 month. Both strength and durability of response at 6 months, respectively, were negatively impacted by end-stage renal disease [(β=-1.10, p=0.004); (β=-0.66, p=0.014)], diabetes mellitus [(β=-0.57, p=0.032); (β=-0.44, p=0.028)], and systemic steroid use [(β=-0.066, p=0.032); (β=-0.55, p=0.037)]. Interestingly, the booster neutralizing antibody titer response was unaffected by clinical factors.
Conclusion Multiple clinical factors impacted the strength and duration of post-vaccination serum neutralizing antibodies in this adult population. Response to the booster dose was universally robust, however. This suggests that the antibody response to the booster dose benefits from a sustained and effective anti-Spike memory immune response.
1. Introduction
Severe Acute Respiratory Syndrome Coronavirus-2 (SARS-CoV-2) virus has infected more than 265 million people worldwide as of December, 2021, and has caused over five million deaths.1 Three vaccines have been approved by the United States Food and Drug Administration, based on safety and efficacy.2 The mRNA vaccines encode a codon-optimized version of spike glycoprotein (S) of SARS-CoV-2, which is the viral protein that elicits neutralizing antibodies that block viral entry into cells and subsequent replication. 3–5 Neutralizing antibodies to the spike protein of SARS-CoV-2 correlate with immunity against the virus after vaccination, reducing rates of COVID-related infection, hospitalization, and death.6,7 Neutralizing antibody titers have been shown to decline over a period of 6 months after completion of the initial vaccine series. However, neither the actual duration of humoral protection, nor the clinical factors that impact the strength and duration of this protection are well described.8–13 Unlike anti-SARS-CoV-2 antibodies, Spike-specific T-cell responses may be sustained at least up to 6 months after both infection and vaccination.14 Poorer concordance between neutralizing antibodies and T cell responses of the adaptive immune system have been associated with severity of disease in individuals ≥ 65 years old.15 It remains to be seen whether other clinical factors and co-morbidities negatively impact the strength and duration of the humoral and cellular immune responses after vaccination.
In light of the emerging data on the waning protection of SARS-CoV-2 mRNA vaccines and the clinical benefits of a third, booster dose, the CDC is now recommending a boost six months after completion of the initial mRNA vaccine series for all adults.16,17 Although advancing age and a personal history of malignancy are associated with decreased immunogenicity to vaccination, those with end-stage renal disease or diabetes mellitus had promising neutralizing antibody responses in other studies.18–21 A better understanding of which clinical factors impact post-vaccine immune responses can help guide additional dose requirements over time. We present our results from a longitudinal study on the evaluation of the clinical variables that impact strength and duration of neutralizing antibodies in the sera of individuals vaccinated with Pfizer-BioNTech SARS-CoV-2 mRNA vaccine. In a subset of subjects, we evaluated response to the six-month booster dose, and its correlation with clinical factors.
2. Methods
2.1 Study Design
Starting in December 2020, we conducted a prospective longitudinal study to collect both clinical information and peripheral blood samples from recipients of BNT162b2 (Pfizer-BioNTech) mRNA vaccine, including veterans and healthcare workers at the Veterans Affairs Connecticut Healthcare System (VACHS) located in West Haven, CT, USA. Venous blood was obtained within 48 hours before the first and second doses of vaccine, and then at one month (up to 1.5 months), three months (up to 3.5 months), and six months (+/- 2 weeks) after the second dose. In a subset of participants who received the booster shot at least six months after the second dose, another sample was collected one month after the booster. Venous blood was processed to obtain serum and plasma. which were cryopreserved at -80°C. Clinical and demographic variables were collected via retrospective review of medical records, and included age, race, ethnicity, sex, body mass index (BMI), medical comorbidities (over 80 variables including cardiovascular, pulmonary, oncological, renal, hepatic disease and others), laboratory values (hemoglobin, serum creatinine, hemoglobin A1c (HbA1c), RT-PCR test results for COVID-19 before or during study participation), and concomitant medications. Renal function was calculated by estimating glomerular filtration rate (GFR) by using Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation.22
2.2 Ethics
The study was approved by the Institutional Review Board at the VACHS. Written informed consent was obtained from each participant.
2.3 Evaluation of anti-SARS-CoV-2 neutralizing antibodies
Sera were tested for neutralization activity against SARS-CoV-2 using a single cycle infectivity assay with Spike-pseudotyped virus particles as described previously.23 SARS-CoV-2 spike (codon-optimized WA1 or Wuhan-1) pseudotyped lentiviral cores expressing firefly luciferase (FFLUC) were produced. These pseudotyped particles were titered on 293T-hACE2 cells seeded in 96 well plates. Each serum sample was tested in duplicate by premixing 75 microliter of four-fold serial dilutions (1:9.21 to 1:603587) with 30 microliter of pseudotyped virus, then adding 95 microliter of this mixture to the 293T-hACE2 cells. After an overnight incubation, 165 microliter of fresh medium was added. After another 48 hours, cells were lysed and luciferase luminometry performed. Curve-fitting using GraphPad PRISM was employed to calculate the neutralization titer half-maximal inhibitory concentration (IC-50 value) for each sample at each time-point.
2.4 Statistical analysis
Variable Selection
Participant morbidity data were grouped based on organ system involved, and clinical characteristics (Table S1). Undetectable IC-50 values (<9.21 μg/mL) were changed to 9.21 for statistical analyses. The IC-50 values were then log-normalized. Paired t-test analysis was used to compare log (IC-50) at various time points: one, three, and six months with baseline log2IC-50 values before vaccination.
To assess the neutralization assay over time and as a comparison with baseline values, the fold-change (Figure 1B) was calculated by dividing the IC-50 at each timepoint by the baseline IC-50. As shown in the y-axis of Figure 1B, the fold changes were found to follow a logarithmic distribution and were thus log-normalized. These steps to calculate log2fold-change (LFC) are shown in the equation below and were used in all univariate and multivariate analyses (Figures 2-4).
(1A) Swarmplot/boxplot of IC-50 titers with unit (IC-50 titer) plotted over time before and after vaccination, 2nd dose, and booster dose. Boxplot shows median, lower/upper quartile, and extremes. (1B) Swarmplot/boxplot of fold change over pre-vaccination IC-50 titers with unit FC[IC-50] plotted over time after the first dose of vaccine. P-values are derived from paired t-test. Outliers, values further than 1.5 times the interquartile range (Q1 – 1.5*IQR and Q3 + 1.5*IQR), are shown beyond the boxplot.
Scatterplot and Pearson correlation coefficients between pre-vaccination IC-50 titers and fold-change response at one-month and six-months.
Log2ΔIC-50 was also considered as an alternate outcome variable in which the pre-vaccination IC-50 was subtracted from the IC-50 at each timepoint. Ultimately, LFC was chosen to adequetely normalize the outcome for parametric statistical tests and consistent reporting with prior studies. However, using LFC results in a measurement artifact in which higher pre-vaccination IC-50 values are associated with lower increases in LFC.24
In our analysis, we used the highest LFC as a measure of ‘strength’ of the humoral response. Additionally, we calculated the LFC of titers at six months compared with titers at pre-vaccination to represent the ‘duration’ of the response.24 Positive LFC values represent increase in IC-50 compared with pre-vaccination IC-50 and negative values represent a decrease in IC-50.
Univariate analysis
Continuous variables (Xi) were used in analyses with outcomes log2(FC) (neutralization titers). Pearson or Spearman correlation tests were utilized depending on the distribution of the response variable:
In this equation, Y represents log2FC. R values reported represent the strength and direction of correlation. Positive values closer to 1 represent stronger positive correlation; negative values closer to -1 represent stronger negative correlation between clinical variable and response.
For dichotomous categorical variables, Shapiro-Wilk test was conducted to assess for each subgroup. If both subgroups maintained normal distributions (p>0.10) of the outcome variable, then we reported the results of two-sided student’s (independent) on t-test (one-sided). If, however, at least one subgroup did not maintain a normal distribution (p<0.10), then we used the non-parametric test, Wilcoxon rank-sum test (one-sided). All univariate analyses were conducted using the Python (v3.8.5) statsmodels package.25
Multiple regression analysis
Variables (Xi) were added and subtracted in a stepwise manner to optimize the Akaike information criterion, an established metric quantifying the trade-off between underfitting and overfitting the data.26,27
To account for the artifactual negative correlation between pre-vaccination IC-50 and log2FC values, log2pre-vaccination IC-50 values was also added as a variable in our analysis. Reported are the coefficient estimates representing the slope of each variable, βi. Positive and negative values (full range between negative infinity and positive infinity) represent positive and negative association between clinical variable and response. The intercept represents y-intercept of the model (i.e., younger participants without co-morbidities and higher GFR response variables). Goodness of fit was analyzed using Q-Q plots. Multiple regression analysis was conducted using R (v4.1.1) with the MASS package.
2.5 Visualizations
All univariate association figures were created using the Python package seaborn. Forest plots, illustrating multiple linear regression models, and associated Q-Q plots was generated using R package forest plot.
2.6 Role of the Funding Source
The funders of the study had no role in the study design, data collection, data analysis, writing of the report, or in the decision to submit results for publication.
3. Results
We included 124 participants (91 veterans, 33 healthcare workers), of whom 74% were male, 66% were white, 93% non-hispanic, and 60% were over the age of 65 years (Table 1 and Table S2). Thirteen participants had COVID-19 diagnosed by RT-PCR prior to the first dose of vaccine and were thus excluded from the univariate and multivariate analyses. Three participants developed COVID-19 between months 3 and 6 after the second dose of vaccine, and their 6 month IC-50 values were excluded from analyses.
3.1 Time since vaccination and anti-SARS-CoV-2 neutralizing antibody titers
Among the time-points tested, IC-50 values were highest at one month after the second dose of vaccine, with a median fold change (FC) of 14.1 compared to pre-vaccination titers, thereafter declining to a median FC of 5.6 and 3.3 at three months and six months, respectively (Figure 1).
Consistent with previous reports24, we found that higher pre-vaccination IC-50 values were associated with a lower FC response at one month (R=-0.27, p=0.005) and six month timepoints (R=-0.63, p<0.0001) (Figure 2). Furthermore, we found that individuals with stronger response strength (higher FC at one month) also had maintained duration of vaccination response (higher FC at six months) (R=0.72, p<0.0001).
IC-50 values measured following the booster (third dose of the vaccination) showed a marked increase in IC-50 values with a median FC of 52.5 compared with pre-vaccination IC-50 values (p<0.0001). We found that individuals had a significant improvement in response even when compared with their measurements at 1 month after the second dose (p<0.0001) (Figure 1).
3.2 Correlation of neutralizing antibody response to clinical variables
Univariate analysis
To understand the role of demographic and clinical factors in the strength and duration of the humoral response, univariate analysis was conducted after excluding participants with prior COVID-19 diagnoses (Figure 3A). IC-50 was highest for all participants one month after the second dose of vaccine. The strength of the humoral response inversely correlated with age (R=-0.27, p=0.005). Age did not have a significant correlation with duration (IC-50 at 6 months) of the vaccination response (R=-0.01, p=0.918) (Figure 3A). Interestingly, female sex was significantly associated with improved strength of the response (ΔFC=7.8, p=0.014), but not the duration of the response (ΔFC=6.5, p=0.503) (Figure 3A). Lastly, there was no difference in strength or duration of response between white and black participants (ΔFC=-2.9, p=0.727; ΔFC =-1.2, p = 0.103 respectively; Figure S2).
(3A) Categorical analysis between significant (p<0.05) and trending variables (p<0.10) of vaccination response at one month (left) and six months following 2nd dose of the vaccine. (3B) Scatterplot illustrating continuous variables: age, glomerular filtration rate, and hemoglobin A1c plotted against vaccination response at one month (red crosses) and six months (blue circles). Correlation analyses of hemoglobin A1c was conducted separately for patients with and without diabetes. Colored lines represent lines of best fit, with shading showing 95% confidence intervals. NS represents non-significant or trending association.
To analyze the clinical variables, we first categorized co-morbidities based upon organ systems. For instance, all pulmonary diagnoses except malignancy were categorized as lung disease (Table S1). We then assessed whether each co-morbidity group impacted immune response (log2FC).
Of the co-morbidities studied, study participants with diabetes mellitus were found to have the highest reduction in strength of response (ΔFC=-7.2, p=0.003), while also having a reduced duration of response (ΔFC=1.0, p=0.013) (Figure 3A). Prior diagnosis of malignancy (solid organ or hematological) (ΔFC=-7.8, p=0.001) and chronic heart disease (ΔFC=-5.3, p=0.025) were independently associated with significant reductions in strength of humoral response. In assessing duration, neither malignancy (ΔFC=-0.6, p=0.126) nor chronic heart disease (ΔFC=-0.8, p=0.238) were significant as clinical factors. Lastly, lung disease trended, without meeting statistical significance, towards reduced humoral response strength (ΔFC=-5.8, p=0.068) and duration (ΔFC=-0.7, p=0.086). Clinical variables found not to be associated with response strength or duration included liver disease, cerebrovascular disease, current steroid use, dementia, and potentially immunosuppressed status from HIV/organ or stem-cell transplant/autoimmune rheumatological disease.
In addition to age, the effect of three additional continuous variables, GFR, BMI, and HbA1c was assessed through univariate analysis (Figure 3B). GFR was significantly associated with response strength, with poor renal function correlating to lower IC-50 at one month (R=0.28, p=0.003); there was, however, no significant association between GFR and duration (R=0.08, p=0.407). In subjects with diabetes mellitus, HbA1c showed a significant inverse association with strength (R=-0.37, p=0.043) and duration (R=-0.48, p=0.007) of the response (Figure 3B). Unsurprisingly, HbA1c did not affect immune response in subjects in the absence of diabetes mellitus. BMI did not correlate with strength (R= 0.10, p=0.314) or duration (R=-0.01, p=0.908) of vaccination response.
We also conducted analyses with age, GFR, HbA1c and BMI as categorical variables. Public health policies and guidelines for prioritization of vaccination often rely on correlation with vaccine responses to clinical factors as categorical variables. Older age, as a categorical variable (>65 years), significantly reduced vaccination response strength (ΔFC=-8.9, p=0.001) and duration (ΔFC=-1.0, p=0.043).
Lastly, we sought to identify if clinical factors with sufficient power in our subset of 36 participants negatively affected the IC-50 one month after booster (third dose) of vaccine. Post-booster neutralization titers (log2IC-50 values) did not differ based on age (R=0.00, p=0.985), GFR (R=-0.08, p=0.643), diabetes mellitus (p=0.926) or malignancy (p=0.217) (Figure S1). Steroid use could not be correlated because of limited sample size.
Multiple Linear Regression
To account for several factors in the analysis, we conducted a stepwise multiple linear regression both for the response strength (Figure 4A) and response duration (Figure 4B). BMI was intentionally excluded due to concerns for collinearity with chronic heart disease. Because of the strong negative correlation between pre-vaccination IC-50 values and response measured as log2FC, the baseline log2IC-50 value was added to the stepwise regression. Unsurprisingly, higher pre-vaccination IC-50 maintained a significant association with lower fold-change IC-50 values at one month and six months, consistent with our univariate analyses, as well as, previously reported neutralization assays.24
Forest plots showing multiple linear regression variables of response strength at (a) one month and (b) response duration at six months. Intercept represents baseline comparison values: age less than 65 years, GFR greater than 60 mL/min/1.73m2, no steroid use, and no comorbidities.
Significant clinical factors associated with decreased strength of response at one month after vaccine series completion, included age greater than 65 years (β=-0.94, p=0.001), malignancy (β=-0.88, p=0.002), GFR below 30 mL/min/1.73m2 (β=-1.10, p=0.004), current steroid use (β=-0.066, p=0.032), and diabetes mellitus (β=-0.57, p=0.032). Liver disease showed an insignificant trend (β=-0.57, p=0.153) towards lower peak response (Fig 3a). GFR < 30 mL/min/1.73m2 (β=-0.66, p=0.014), current steroid use (β=-0.55, p=0.037), and diabetes mellitus (β=-0.44, p=0.028) were significantly associated with decreased duration of response. Meanwhile, age greater than 65 (β=-0.36, p=0.065) and history of malignancy (β=-0.36, p=0.064) showed a trend towards reduced duration without meeting significance (Fig 4B).
Clinical factors that did not demonstrate significant contributions to our multivariable model were removed in our stepwise regression. These factors included sex, race, chronic heart disease, lung disease, and dementia. Liver disease was also not found to sustain a trend in our duration model.
4. Discussion
SARS-CoV-2 vaccines have played a key role in curbing the spread, morbidity, and mortality from COVID-19. With rising concerns regarding the duration of protection after vaccination, efforts to understand the optimal timing of booster shots are underway. It is imperative to understand clinical factors that impact the strength and duration of the immune responses to these vaccines, and the role of adaptive immune responses on duration of protection. Previous studies have examined one or two clinical factors to assess the effect of vaccination on immune response.18–21 In this study, with full access to medical records, we evaluated multiple clinical factors with multivariable regression in order to identify which clinical variables had the most significant negative (or positive) impact on the strength and duration of immune response.
This study was conducted on a U.S.-based population of veterans and healthcare workers, which offers a unique study demographic. About half of our sample population for this cohort was > 65 years old and the majority was male with multiple co-morbidities. The predominantly older male population typical of Veterans is also a predisposed demographic group to suffer severe/fatal COVID-19, especially if there are coexistent co-morbid conditions.28, 29 It is critical to determine how protective and persistent the vaccine response will be in individuals at high risk for contracting severe COVID-19. We included healthcare workers in this analysis to enhance the diversity of the study population in terms of demographic and clinical factors.
In this cohort, we found that among the time-points tested, IC-50 values peaked at one-month after completion of the primary series of BNT162b2 mRNA vaccine, with the six-month median level at one-fourth of the peak level. The neutralizing antibody IC-50 peak level was negatively impacted by advanced age, malignancy, renal disease, diabetes mellitus, and active use of systemic steroids within the study period. IC-50 values declined over time in all subjects over the period of 6 months after primary vaccination series completion. The clinical factors with the most significant negative impact on the duration of response were end-stage renal disease, diabetes mellitus, and steroid use during the study period. Our data suggest that age and malignancy play a role in reducing the initial peak response, but may not significantly impact the duration of response at 6 months. Given that the strength of neutralizing antibody titers has been shown to be highly correlated with immune protection,7 these findings support the use of booster doses, with prioritization for specific, vulnerable populations.
Although neutralizing antibody titers decline over time, B- and T-cell responses appear to persist.14 Antibody titers to mRNA vaccination at six months have been shown to correlate with T-cell responses after the first vaccine dose30, but further studies are needed to examine whether clinical factors impact T-cell responses similarly to neutralizing antibody titers. Interestingly, in our study, the clinical factors that impacted strength and duration of humoral immune responses after the primary series of the BNT162b2 vaccine, did not affect the responses one month after the booster dose. Antibody titers after a booster dose of mRNA vaccine have been shown to correlate with pre-existing B-cell frequency.30 Our data supports the hypothesis that a booster dose of vaccine may benefit from a memory T- and B-cell response that allows for efficient, rapid, and high-level production of neutralizing antibodies. Further study will be needed to understand whether additional booster doses are needed over time, whether immune memory will offer sufficient protection, and whether different clinical factors will impact that immune memory.
This study has some important limitations. Our sample size is relatively small, and similar clinical correlations in larger samples may better inform vaccine responses based upon select clinical variables. We evaluated the neutralizing antibody titers over a six month time period after completion of the primary vaccine series. Evaluation beyond that time point in individuals who did not receive the booster dose may reveal further associations between clinical variables and duration of response. Similarly, we defined strength of the response based on the highest level of IC-50 obtained by each individual. Since the time-points when we collected samples were predetermined, it is possible that the true peak response occurs before or after the 1 month time-point. While this may affect the fold-change values, it will likely not alter the correlation of strength of response with clinical factors. Another limitation of our study is that we excluded subjects from statistical analysis if they had known history of COVID-19 before vaccination. The number of these subjects in our study was too small (13) to conduct multiple regression analysis. Further investigation of clinical factors that associate with vaccine responses in individuals with prior SARS-CoV-2 infection would be valuable, especially when compared with those who were not previously infected. Lastly, this work does not include memory B-cell and memory T-cell responses and how those correlate to clinical factors and IC-50 values of neutralizing antibody titers. The excellent booster-dose responses among all subjects tested suggest that the memory cells may not be as impacted by clinical or demographic factors, but this needs to be investigated in larger studies.
In conclusion, while age >65 years negatively impacts the strength of the humoral immune response as quantified by levels of neutralizing antibodies, clinical comorbidities of end-stage renal disease, diabetes mellitus, and steroid use negatively impact both the strength and duration of the humoral immune response against SARS-CoV-2. Neutralizing antibody IC-50 titers after the booster dose, however, are robust regardless of these clinical factors, suggesting protection despite reduction in prior IC-50 levels. This may be related to underlying persistent immunologic memory and durability of SARS-CoV-2 specific T and B cells, an area that needs further exploration.
Data Availability
All data produced in the present work are contained in the manuscript.
Declaration of Interests
None of the authors declare any conflict of interests:
Funding
Funding for this study was provided by the National Institutes of Health (awards R01 AI150334 to RES). The analysis was supported by Global Health Equity Scholars (FIC D43TW010540 to RS), the NIH (5T32AI007517-20 to LP) and the Yale School of Medicine Medical Student Research Fellowship (AHS).
Data sharing
De-identified individual participant-level data that underlie the results reported in this manuscript (text, figures, tables, supplementary material) and supporting documents (informed consent form, study protocol, statistical analysis plan) will be shared after publication and finalization of completed study report for ≥1 year. This study is ongoing, and all individual participant data cannot be available until evaluation of immune responses several months after vaccination. URL links will be provided for access to these materials.
Figure legends
(a) Scatterplot with line of best fit illustrating age and GFR with log2IC-50 values one month after the booster. Correlation coefficient (R) and p-values shown in each legend. (b) Categorical analysis between clinical diagnoses of diabetes mellitus and cancer with post-booster neutralizing antibody titers.
(a) Categorical analysis between clinical variables not associated (p>0.10) with vaccination response at one month (left) and six months (right) following 2nd dose of the vaccine. (b) Scatterplot illustrating continuous variables: BMI plotted against vaccination response at one month (red crosses) and six months (blue circles). Colored lines represent lines of best fit with shading showing 95% confidence intervals.
Acknowledgements
We thank our study administrative team, David Ardito and Lucienne Levy, for their tireless dedication, and to the Veterans and healthcare workers who have donated their samples to make this study possible.
Bibliography
