Lack of antibodies to SARS-CoV-2 in a large cohort of previously infected persons

Abstract Background Reports suggest that some persons previously infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) lack detectable IgG antibodies. We aimed to determine the proportion IgG seronegative and predictors for seronegativity among persons previously infected with SARS-CoV-2. Methods We analyzed serologic data collected from health care workers and first responders in New York City and the Detroit metropolitan area with history of a positive SARS-CoV-2 reverse transcriptase polymerase chain reaction (RT-PCR) test result and who were tested for IgG antibodies to SARS-CoV-2 spike protein at least 2 weeks after symptom onset. Results Of 2,547 persons with previous confirmed SARS-CoV-2 infection, 160 (6.3%) were seronegative. Of 2,112 previously symptomatic persons, the proportion seronegative slightly increased from 14 to 90 days post symptom onset (p=0.06). The proportion seronegative ranged from 0% among 79 persons previously hospitalized to 11.0% among 308 persons with asymptomatic infections. In a multivariable model, persons taking immunosuppressive medications were more likely to be seronegative (31.9%, 95% confidence interval [CI] 10.7%-64.7%), while participants of non-Hispanic Black race/ethnicity (versus non-Hispanic White) (2.7%, 95% CI 1.5%-4.8%), with severe obesity (versus under/normal weight) (3.9%, 95% CI 1.7%-8.6%), or with more symptoms were less likely to be seronegative. Conclusions In our population with previous RT-PCR confirmed infection, approximately one in 16 persons lacked IgG antibodies. Absence of antibodies varied independently by illness severity, race/ethnicity, obesity, and immunosuppressive drug therapy. The proportion seronegative remained relatively stable among persons tested up to 90 days post symptom onset.


Introduction
Many published studies using in-house and commercial assays have examined development of antibodies to SARS-CoV-2 following infection. These studies demonstrate that IgG, IgM, and IgA antibodies targeted to the virally encoded surface spike (S) and nucleocapsid (N) proteins develop nearly simultaneously in a high proportion of patients within 2 weeks after symptom onset. [1][2][3][4][5][6][7][8] However, accumulating data indicate that IgG antibodies may not be detected in some persons following infection, particularly among those with milder illnesses. In one study of 20 persons with mild disease subsequently tested with five commercial assays 28-54 days after symptom onset, 10-30% tested negative depending on the assay, with no obvious differences whether the assays tested antibodies targeted to the S or N proteins. [9] In another study of 27 patients with mild disease tested with two commercial assays 8-12 weeks post symptom onset, 22% were seronegative.
[10] A third study of 24 patients with mild disease tested with a commercial assay detecting total antibody, 13% were seronegative 21-24 days post symptom onset. [11] However, these results contrast with another study of 624 patients with mild disease tested with an in-house assay, of whom 99% tested positive. [12] Unlike patients with mild disease, studies consistently show that all or nearly all hospitalized patients develop antibodies to the S and N proteins. [11,[13][14][15][16][17] Fewer studies have examined longer-term persistence of anti-SARS-CoV-2 antibodies.
Serum IgA and IgM antibodies appear to rapidly wane over weeks or a few months; [18,19] whereas, evidence suggests longer-term persistence of IgG antibodies. [17][18][19][20][21] Nevertheless, in one study 40% and 13% of asymptomatic and symptomatic persons, respectively, lacked IgG antibodies approximately 10 weeks after the initial positive RT-PCR result or symptom onset. [22] The lack of antibody in that study resulted both from persons not developing A c c e p t e d M a n u s c r i p t 5 antibody as well as from waning antibody among those who did. In another study, IgG antibodies targeting the receptor binding domain (RBD) decreased with an estimated half-life of 36 days. [23] In yet another study of seven asymptomatic patients, two (29%) lacked IgG antibodies approximately 8 weeks after initial infection. [24] To further determine the proportion of persons without detectable IgG antibodies after SARS-CoV-2 infection and predictors for lack of antibody, we studied serologic data from 2,547 persons with previous confirmed SARS-CoV-2 infection. We identified these persons during a large-scale serologic survey of healthcare personnel and first responders in two large U.S. metropolitan areas.

Overall approach
Most published studies of SARS-CoV-2 antibody development have been limited by sample size and enrollment of symptomatic persons who have sought health care. To obtain a large sample across a broad spectrum of illness severities ranging from asymptomatic to hospitalization, we conducted a serosurvey for anti-SARS-CoV-2 IgG antibodies among healthcare personnel and first responders and determined those who had had a previous positive SARS-CoV-2 RT-PCR test result as described below. These persons comprised the final study population from which comparisons of antibody prevalence by demographic and clinical variables permitted evaluation of their contribution to antibody development and persistence.
A c c e p t e d M a n u s c r i p t 6 Population tested In the Detroit metropolitan area, healthcare personnel at 27 hospitals who worked in   emergency departments, intensive care units, general inpatient units, surgical units, and who provided support services to those areas, such as radiology, were offered antibody testing. [25] In addition, first responders and public safety personnel in emergency medical services agencies overseen by seven local Medical Control Authorities were invited to participate. In New York City, all police, firemen, medical examiner and corrections staff, and staff at 11 hospitals and seven outpatient facilities were offered testing.

Data collection
The study period was from May 18 through June 13 in the Detroit metropolitan area and   To help assess the generalizability of the serosurvey results with other SARS-CoV-2 IgG antibody tests, a sample of 126, 39, and 80 specimens with signal-to-cutoff ratios (S/CO) ratios of <0.05 (low negative), 1.0-1.5 (low positive), and >1.5 (higher positive), respectively, were selected from the overall serosurvey population for testing with the Abbott Architect SARS-CoV-2 IgG assay, which detects IgG antibodies targeted at viral nucleoprotein; the ORTHO Clinical Diagnostics VITROS Immunodiagnostic Products Anti-SARS-CoV-2 Total Antibody Test, which detects total Ig antibodies targeted at the S1 protein domain; and an ELISA test developed at CDC that detects total Ig antibodies directed at spike protein.

Overall antibody prevalence
Of the 36,918 persons surveyed, 2,613 (7.1%) self-reported having tested positive for SARS-CoV-2 by RT-PCR in a nasal or throat swab or saliva sample (Figure 1). Of these 2,613 persons, samples from four persons were not tested for antibody due to lipemia, two persons A c c e p t e d M a n u s c r i p t 9 presented twice for testing, 2,387 tested antibody positive, and 220 tested antibody negative.
Of the 220 persons who tested antibody negative, 32 with only a record of a negative RT-PCR result and 28 with no testing record at the health department were excluded from further analysis. Thus, the final data set consisted of 2,547 persons with confirmed previous SARS-CoV-2 infection, of whom 160 (6.3%) were seronegative (Figure 1). A total of 2,164 persons were from New York and 383 were from Detroit, and the average age was 40.1 years (standard deviation [SD] 10.4 years).
Factors related to seronegativity Of 2,547 persons in the final data set, 1,995 (78.3%) reported at least one symptom listed in Table 2 and an exact date of symptom onset. Antibody testing occurred a mean of 63.6 days (SD 16.0 days) after symptom onset. The mean time since symptom onset was slightly longer among those who tested antibody negative (66.5 days, SD 17.3 days; p=0.06) than those who tested positive (63.4 days, SD 15.9 days; p=0.06) and the proportion testing negative increased slightly as the duration from symptom onset to antibody testing date increased (Table 1). Among the 265 persons who reported no symptoms and who could recall the RT-PCR testing date, a mean of 53.5 days (SD=19.7 days) elapsed between RT-PCR and antibody testing. For this group, seroprevalence did not differ significantly by the interval between RT-PCR and antibody testing dates (supplementary Table 1).
Although the absence of antibodies did not differ significantly by age and sex, non-Hispanic Black participants were less likely than non-Hispanic White participants to lack antibodies (3.2% versus 6.1%, respectively; A c c e p t e d M a n u s c r i p t 10 Persons who reported immunosuppressive therapy or medications, such as cancer treatment, were much more likely to be seronegative; approximately one in four lacked antibodies (26.7% versus 6.2% not taking immunosuppressive therapy or medications, p=0.001; Table   1). However, antibody prevalence did not differ according to the presence of immunosuppressing conditions, such as human immunodeficiency virus infection or autoimmune disease. Persons who sought medical care for COVID-19 symptoms were less likely to lack antibodies compared to those who didn't seek care (5.2% versus 8.0%, p=0.005). Seroprevalence was not related to the presence of diabetes, hypertension, chronic heart disease, chronic kidney disease, chronic liver disease, chronic obstructive pulmonary disease, or hypertension (data not reported).
Several measures of previous illness severity were related to absence of antibodies.
Asymptomatic persons were more likely to lack antibodies (11.0%) than those previously symptomatic (5.6%, p=<0.001), with seronegativity decreasing as the number of symptoms increased ( Table 2). Among individual symptoms, history of fever and loss of taste or smell had the strongest relationship with the presence of antibody. None of 79 persons who reported previous COVID-19-related hospitalization were seronegative compared to 6.5% of those not hospitalized (p=0.02; Table 1). In a multivariable model, characteristics that remained significantly associated with lacking antibodies included taking immunosuppressive medications (31.9% versus 6.2% for persons not taking such medications), non-Hispanic White and Hispanic race/ethnicity (6.4% and 8.6%, respectively versus 2.7% for non-Hispanic Black race/ethnicity), under/normal weight status (9.4% versus 5.4% for obesity) and fewer symptoms (persons with 0-2 symptoms had higher risk of lacking antibodies compared with persons with 6-9 symptoms) (Figure 2, supplementary

Discussion
Among 2,547 persons with confirmed previous SARS-CoV-2 infection, 6.3% lacked anti-SARS-CoV-2 IgG antibodies. This result can be compared with an Icelandic study in which 1,215 persons with qPCR-diagnosed infection were tested approximately 3 months after recovery using pan-Ig assays measuring anti-N or anti-S1-RBD antibodies. [19] Negative antibody results were obtained in either one or both assays in 9.0% and 4.9%, respectively. In another seroprevalence study of New York City healthcare personnel, 6.5% of 2,044 persons with a history of a PCR-positive infection were seronegative.
[29] However, the interval between symptom onset and antibody testing was not reported and seven different assays were used at various time points.
A c c e p t e d M a n u s c r i p t 12 In our study population, the proportion lacking antibodies increased slightly as the time between symptom onset and antibody testing increased, although this trend was on the margins of statistical significance (p=0.06). This finding along with the fact that fewer than 10% were seronegative up to 90 days after symptom onset suggested a minimal temporal rate of seroreversion over the time period studied. This is consistent with another study that showed that IgG antibodies to the receptor binding domain and N protein slightly decreased over 6-7 months post illness onset. [20] In the Icelandic study of 1,215 patients, the proportion positive testing positive with two pan-Ig antibody levels remained stable over 3 months and IgG anti-N and anti-S1 antibody levels decreased slightly after 6 weeks from diagnosis. [19] Our data affirm previous observations that IgG antibodies are less likely to develop in persons with milder disease. [9-11, 17, 24, 30, 31] In our study, 11% of those asymptomatically infected lacked antibodies; whereas, none of 79 persons previously hospitalized lacked IgG antibodies. Between these two extremes, those with fewer symptoms were less likely to have antibodies. Our large sample also enabled identification of other factors presumptively related to the presence of antibodies following infection. In a multivariable model controlling for potential confounding factors such as symptom frequency and seeking medical care, non-Hispanic Black race/ethnicity and obesity were associated independently with higher prevalence of antibodies following infection; taking immunosuppressive medication was inversely associated with the presence of antibodies.
While serologic surveys have often found that persons of non-Hispanic Black race/ethnicity have higher seroprevalence than non-Hispanic whites, we are not aware of other studies that have examined racial and ethnic differences in development of antibodies following infection. Our finding that non-Hispanic Whites were over twice as likely to lack antibody (adjusted 6.4%) than non-Hispanic Blacks (adjusted 2.7%) following infection requires A c c e p t e d M a n u s c r i p t 13 confirmation by other studies. In addition, we are also not aware of other studies showing that persons with lower body mass index are less likely to have detectable IgG antibodies following infection. However, this finding is consist with another large study that found lower pan-Ig antibody levels in persons with lower body mass index. [19] While it is plausible that immunosuppressive drugs would inhibit antibody development, it is not clear why race/ethnicity or obesity would independently predict the presence or absence of IgG antibodies following infection.
We observed a high concordance of ORTHO IgG antibody assay results with the ORTHO total antibody and CDC assays, each of which measures antibody to spike protein. In contrast, the Abbott assay, which measures IgG antibody to nucleoprotein, appeared to have lower sensitivity, particularly among the low positives on the ORTHO IgG assay. It should be noted that the 39 low positives represented only 1.6% of the 2,387 positive samples. Among these samples, the similar intervals between PCR and antibody testing among those testing positive and negative on the Abbott assay suggested that the lower sensitivity was not due to waning antibody. Our results suggested that if an orthogonal testing algorithm were employed (Ortho IgG antibody assay followed by the Abbott assay) approximately 5% of the positive results would have been excluded; however, our study design did not permit assessment of potential improvements in positive predictive value with this approach.
This study had several limitations. The sample was a working population that may have been healthier than the population-at-large, had few adults aged over 65 years, and did not include children. There may have been inaccuracies in self-reporting of information, such as the dates of symptom onset and RT-PCR specimen collection. We could not ascertain whether specific immunocompromising conditions or medications influenced antibody development since our survey did not gather this level of detail. Persons who died would not be included in this survey, although this number was expected to be small. Our observation that 6.3% had no  A c c e p t e d M a n u s c r i p t      A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t 27 Figure 2