key: cord-1002531-5nhxrnrx
authors: Hughes, Ellen C; Amat, Julien A R; Haney, Joanne; Parr, Yasmin A; Logan, Nicola; Palmateer, Norah; Nickbakhsh, Sema; Ho, Antonia; Cherepanov, Peter; Rosa, Annachiara; McAuley, Andrew; Broos, Alice; Herbert, Imogen; Arthur, Ursula; Szemiel, Agnieszka M; Roustan, Chloe; Dickson, Elizabeth; Gunson, Rory N; Viana, Mafalda; Willett, Brian J; Murcia, Pablo R
title: SARS-CoV-2 serosurveillance in a patient population reveals differences in virus exposure and antibody-mediated immunity according to host demography and healthcare setting
date: 2020-12-26
journal: J Infect Dis
DOI: 10.1093/infdis/jiaa788
sha: e58bd6b065481991a5a32dd4d04adad5e5352b6b
doc_id: 1002531
cord_uid: 5nhxrnrx

Identifying drivers of SARS-CoV-2 exposure and quantifying population immunity is crucial to prepare for future epidemics. We performed a serial cross-sectional serosurvey throughout the first pandemic wave among patients from the largest health board in Scotland. Screening of 7480 patient sera showed a weekly seroprevalence ranging from 0.10% to 8.23% in primary and 0.21% to 17.44% in secondary care, respectively. Neutralisation assays showed that around half of individuals who tested positive by ELISA assay, developed highly neutralising antibodies, mainly among secondary care patients. We estimated the individual probability of SARS-CoV-2 exposure and quantified associated risk factors. We show that secondary care patients, males and 45-64-year-olds exhibit a higher probability of being seropositive. The identification of risk factors and the differences in virus neutralisation activity between patient populations provided insights into the patterns of virus exposure during the first pandemic wave and shed light on what to expect in future waves.

A c c e p t e d M a n u s c r i p t Background SARS-CoV-2 was first reported in China in December 2019 and spread rapidly across multiple countries. The first COVID-19 case in Scotland was confirmed on February 28 th , 2020, the country entered lockdown on March 23 rd and restrictions were eased on May 28 th [1] . Serological surveys are instrumental in determining infection rates at the population scale [2] . Assays based on the detection of anti-SARS-CoV-2 IgG antibodies, which are typically detectable 7 to 21 days post-infection [3] , can identify past viral exposure even in asymptomatic individuals. In-house assays commonly utilise an indirect-ELISA format, with recombinant S protein, S1 subunit of the S protein or the receptor binding domain (RBD) used as antigens. Virus neutralisation assays (VNAs) provide insights into the effectiveness of the humoral immune response. Neutralisation titers obtained with pseudotype-based tests are similar to those obtained with live virus [4] , and two pseudotype-based methodologies are commonly used: HIV-based pseudotypes and VSV-based pseudotypes. Both methodologies produce similar results [5] .

Models that link patient information (e.g. age, sex, and time of sampling) with exposure and immunity enable the identification of factors associated with SARS-CoV-2 infection [6] . NHS Greater Glasgow and Clyde (NHSGGC) is the largest health board in Scotland and reported the highest number of COVID-19 cases (n=3876) and deaths (n=1280) in the country between March 1 st and May 24 th , [7] . We performed a serial cross-sectional study amongst primary and secondary care patients in NHSGGC to estimate levels of exposure to SARS-CoV-2 since the introduction of the virus in Scotland and up to calendar week 21 (starting on May 18th, 2020). Using a Bayesian framework, we combined serological and patient information to estimate an individual's probability of testing positive for SARS-CoV-2 across various age groups, time and healthcare settings. We also performed neutralisation assays to estimate the fraction of exposed individuals who developed an effective antibody response. Finally, we combined serological data with publicly available information on deaths to estimate the case fatality ratio.

A c c e p t e d M a n u s c r i p t

Ethical approval was provided by NHSGGC Biorepository (application 550). Random residual biochemistry serum samples (n= 7480) from primary (general practices) and secondary (hospitals) healthcare settings were collected by the NHSGGC Biorepository between 16 th of March and 24 th of May 2020. Associated metadata included date of collection, patient sex and age, partial post code of the patient and sample origin (primary or secondary care). All serum samples were inactivated at 56 o C for 30' before being tested.

ELISA testing S1 and RBD antigens were prepared as described previously [8] . The SARS-CoV-2 RBD and S1 constructs, spanning SARS-CoV-2 S (UniProt ID P0DTC2) residues 319-541 (RVQPT…KCVNF) and 1-530 (MFVFL…GPKKS), respectively, were produced with Cterminal twin Strep tags. Proteins were produced by transient expression in Expi293F cells grown in FreeStyle-293 medium (Thermo Fisher Scientific). Proteins were harvested at two timepoints, 3-4 and 6-8 days post-transfection. Twin Strep-tagged proteins were captured on Streptactin XT (IBA LifeSciences) and purified by size exclusion chromatography through Superdex 200 (GE Healthcare). Purified SARS-CoV-2 antigens, concentrated to 1-5 mg/ml by ultrafiltration were aliquoted and snap-frozen in liquid nitrogen prior to storage at -80 o C.

Assays to detect IgG antibodies against recombinant S1 and RBD antigens of SARS-CoV-2 were performed as described [9] . 96-well plates (Immulon 2HB, Fisher Scientific) were coated overnight with S1 or RBD antigen (50ng/well). After washing three times with PBS/0.05%Tween-20 (all subsequent wash steps followed the same protocol), sera were diluted 1:100 in PBS/0.05%Tween20 (v/v) supplemented with 10% (v/v) casein (Vector laboratories, c/o 2BScientific) and incubated for one hour at room temperature before a second wash. Anti-human IgG horseradish peroxidase-conjugated secondary antibody (Bethyl laboratories) diluted 1:3000 in PBS/0.05%Tween-20/casein was then added and A c c e p t e d M a n u s c r i p t incubated for one hour before a third wash. 3,3′,5,5′-tetramethylbenzidine (TMB) (Sigma-Aldrich/Merck) was added and incubated for 10' in the dark. The reaction was stopped by adding an equal volume of 1M H 2 SO 4 . Absorbance was read immediately at 450 nm on a Labsystems Multiskan Ascent plate reader. Duplicates of pooled known-positive and knownnegative controls were included on each plate. Raw absorbance values were corrected using the equation, "(sample absorbance-negative control mean)/negative control mean". analyzed on a Perkin Elmer EnSight multimode plate reader (Perkin Elmer). Sera were considered to have high neutralising activity if at a 1:50 dilution they reduced infection by HIV(SARS-CoV-2) pseudotypes by ≥90% [11] .

The number of laboratory-confirmed cases was obtained from the Scottish Government Website (https://www.gov.scot/coronavirus-covid-19/) and the West of Scotland Specialist Virology Centre. The number COVID-19 associated deaths was obtained from the National Records of Scotland website (https://www.nrscotland.gov.uk/covid19stats).

A c c e p t e d M a n u s c r i p t

Multivariable logistic regression models were used to investigate associations between neutralisation at a 1:50 dilution and corrected OD values, care-type, age group and sex in ELISA positive samples (n=216). Separate models were run for samples positive to S1 and RBD (Supplementary Tables 2 and 3 ). Univariate analyses comparing the mean corrected OD, or percent neutralisation, between ELISA positive samples from primary and secondary care types were undertaken using Mann-Whitney U tests. To determine a sample size for estimating the prevalence of partial postcode districts a simple calculation, assuming a random sample from a large population, was used. An assumed prevalence (p) of 10%, and a confidence of 95%, substituted into the equation n=1.96 2 p(1-p)/d 2 (where d=precision=0.05), resulted in a sample size of 138. Statistical analyses and data visualisation were undertaken in R [12] , version 3.6.1. Models were run using lme4 [13] .

An SSM was developed to estimate the weekly probability of infection of the patient population and to evaluate the impact of the different demographic factors affecting the probability of an individual being seropositive for SARS-CoV-2. The model followed methods previously published [14] and comprised two coupled parts; a population-level process, and an observation or individual-level process. The population process captured the weekly exposure dynamics through a linear predictor comprising a temporal trend and autocovariates (i.e. first-and second-order AR components capable of reconstructing potential exposure cycles). This results in a weekly probability of infection that reflects the average chance of being infected in a given week after adjusting for individual covariates in the observation process. The observation process confronted the population probabilities by using individual-level data (i.e. binary observed serology data from each patient) in a Bernoulli trial that adjusted seropositivity according to the sensitivity and specificity of the test and estimated an individual's probability of infection based on the population-level A c c e p t e d M a n u s c r i p t dynamics but also through a series of individual covariates such as sex, age, care type and week of sample collection. We noted that since further adjusting for population size resulted in differences of ~0.1% in group-based seroprevalence estimates, for simplicity this was omitted from the final SSM. We ran the model in JAGS for 100K iterations and 50K burn-in to achieve full convergence. Priors and the model code are provided with supplementary material.

An IFR was calculated for each age group by estimating the fraction of SARS-CoV-2 confirmed deaths relative to the number of people exposed. The latter variable was approximated using the adjusted seroprevalence, multiplied by the corresponding group 

A total of 7480 residual biochemistry serum samples from patients living in NHSGGC were tested for the presence of IgG antibodies against the S1 subunit of the SARS-CoV-2 spike protein and its receptor binding domain (RBD) using two ELISA assays [9] . Of these, 6635 met the inclusion criteria and were used for further analysis. Samples spanned a 10-week period, starting on March 16 th , 2020 and covered all NHSGGC districts and all age groups, except for children and young adults under 18 years of age for whom insufficient samples were available ( Figure 1 describes the sample inclusion criteria and sample sizes). The underrepresentation of samples from paediatric patients reflected the reduction in general practitioner appointments, the prioritisation of COVID-19 suspected cases during this period, avoidance of attending medical facilities of parents to protect children from the virus, and likely reduced risk of non-COVID-19 infections and injuries (the most common reason for A c c e p t e d M a n u s c r i p t emergency attendances in children) due to physical distancing as well as the lower incidence of clinical signs in children [15, 16] .

The overall unadjusted seroprevalence in our patient population was 7.81% (95% CI 7.17-8.48, Figure 2A) . Seroprevalence was higher in the 45-64 year olds, in males, and in patients attending secondary care services (Figure 2A) . A steady increase in seroprevalence from week commencing (w/c) March 16th up to w/c April 13th in both primary and secondary care settings was observed. However, while seroprevalence in the secondary care subpopulation was higher, and started to decrease from w/c April 13th, seroprevalence in primary care remained at a similar level after w/c 13th April to the end of our study period ( Figure 2B ). For some age groups (45-64y and 64-74y) seroprevalence was higher in men ( Figure 2C) , perhaps driven by a sex bias in SARS-CoV-2-associated hospitalisation [17] since men admitted to secondary care services had a higher seroprevalence (10.73%, 95% CI 9.40-12.17) than women (7.60%, 95% CI 6.51-8.81; Figure 2C ). This difference between sexes was not observed among primary care patients (6.06%, 95% CI 4.73-7.63 for men and 5.40%, 95% CI 4.29-6.71 for women, Figure 2C ). Patient seroprevalence was also calculated in a subset (20/61) of districts in which sample numbers provided sufficient power to estimate prevalence. Estimated seroprevalence ranged from 3.83% (95% CI 1.67-7.40) to 12.94% (95% CI 8.29-18.94) (Supplementary Table 1 ) suggesting that there may be geographically-driven differences in infection risk. However, sample size limitations prevented more detailed analysis. Our Bayesian state-space model [14] was used to adjust the crude patient seroprevalence rates for the sensitivity and specificity of the assays and to determine the factors associated with seropositivity in the study population. The model converged well and provided a good fit to the data ( Figure 3A and Supplementary Figure 2 ).

Although the test had high sensitivity (95.31%, 95% CI 90.08-98.26%) and specificity (97.20%, 95% CI 94.76-98.71%), the adjusted overall seroprevalence (5.29%, 95% CI 0.13-15.10) was approximately half the crude estimates ( Figure 3A and Table 1 ). The analysis indicated that patients receiving secondary care were twice as likely (odds ratio 2.2, 95% CI:

A c c e p t e d M a n u s c r i p t 1.6-3.1) to be seropositive compared to those in primary care ( Figure 3B ). Male patients were 1.39 times (95% CI 1.1-1.8) more likely to be seropositive, and individuals belonging to the 45-64y age group were 2.2 times (95% CI 1.5-3.3) more likely to be seropositive than those in the 18-44y age group. However, belonging to the older age groups (65y+) did not significantly increase the probability of being seropositive ( Figure 3B ). Nonetheless, considering the adjusted seroprevalences per age group, and their associated population size and SARS-CoV-2 related deaths, we estimated a higher infection fatality ratio (IFR) in older age groups (Table 1) , which is consistent with a previous UK-based study [18] . The probability of infection at the population level ( Figure 3C Figure 4B ) and mean percent neutralisation when compared with sera from antibody-positive patients in primary care ( Figure 4C ), implying that disease severity is associated with a stronger and more effective antibody-mediated response.

Multivariable logistic regression models confirmed that increasing absorbance values on ELISA were significantly associated with neutralisation (OR=1.15, 95% CI 1.10-1.21, p=<0.001), and that samples derived from secondary care had significantly higher odds of having neutralising ability compared to primary care (OR=6.77, 95% CI 2.68-18.75, p=<0.001) ( Supplementary Tables 2 and 3 ).

Serological surveys are key to informing strategies aimed at controlling the spread of disease. Our study showed that SARS-CoV-2 exposure during the first wave of the pandemic remained broadly consistent over time (likely due to lockdown conditions), but heterogeneous among different groups of the Glasgow patient population. After adjustment for test sensitivity and specificity, the overall seroprevalence in the patient population of NHSGGC (5.29%) was similar to reports from community-based cross-sectional studies carried out during an equivalent period in other European cities such as Geneva [19] and Madrid [20] . However, as our study relied on analyses of residual biochemistry samples from a population of individuals seeking healthcare including -but not exclusively-people who are more likely to be symptomatic with SARS-CoV-2 infection than the general population, generalisation beyond the study population requires caution. For example, male patients had A c c e p t e d M a n u s c r i p t a significantly higher risk of being seropositive in our study, although this was not a feature of the previous community-based studies, likely reflecting a sex bias in COVID-19 presentation [21] or differences in social behaviour that led to increased exposure [22] . It is important to note that 38% of samples were derived from patients attending primary care, and this proportion remained stable during the studied period. Under normal circumstances, such samples would provide a cost-effective method of obtaining samples for serosurveillance that are broadly representative of the wider community [23] . However, the unprecedented changes to routine healthcare guidelines and health-seeking behaviour [16] during the first wave of the pandemic are likely to have altered the structure of this population considerably. Patients in primary care were well enough to be managed in the community and so might be subject to similar exposure conditions as the general population.

At the same time, groups who continued to be seen in primary care for blood sampling, including pregnant women and those with chronic conditions, may have shielded during this period and thus have had lower exposure than the general population. The prevalence in this group may therefore be lower than the expected community prevalence. Conversely, the probability of exposure for individuals from secondary care might be higher than expected in the general population due to the prioritization of severe COVID-19 cases in hospital settings during this period. In addition, some patients may have been in the early stages of infection and may not have seroconverted at the time of sampling, resulting in an underestimation of seroprevalence in both health care settings. Overall, and with the aforementioned caveats, the seroprevalence observed in the primary care subpopulation may be a better representation of the general population than that observed in secondary care.

Neutralisation assays provided insight into post-exposure antibody-mediated immunity.

HIV(SARS-CoV-2) pseudotype-based neutralisation assays display a high correlation with live virus-based assays [4] . Although we found a significant correlation between antibody levels and neutralising activity, we also found, in agreement with other studies [24] , that exposure to SARS-CoV-2 resulted in heterogenous responses. As samples from secondary However, similar results linking disease severity and immune response were reported [25] [26] [27] . Neutralising ability observed in a small number of ELISA-negative sera suggests that the presence of epitopes outside the SARS-CoV-2 S1 or receptor binding domains may contribute to the neutralising response. We note that while there is evidence linking the presence of neutralising antibodies with protection [28] , any inferences between antibody levels and protective immunity should be interpreted with caution. The determinants of a protective immune response to SARS-CoV-2 are unknown and recent studies have suggested that T cell responses play an important role in SARS-CoV-2 immunity [29] . It has been postulated that between 43 and 70% of the population needs to be immune to SARS-CoV-2 to reach herd immunity [30, 31] . Achieving such levels without vaccination is unlikely in the short term given that seroprevalence, even amongst secondary care patients who 

COVID-19): Scotland's route map

The important role of serology for COVID-19 control

Antibody Responses to SARS-CoV-2 in Patients With Novel Coronavirus Disease

Characterisation of SARS-CoV-2 Lentiviral Pseudotypes and Correlation between Pseudotype-Based Neutralisation Assays and Live Virus-Based Micro Neutralisation Assays

Establishment of the WHO International Standard and Reference Panel for anti-SARS-CoV-2 antibody

Seroepidemiologic Study Designs for Determining SARS-COV-2 Transmission and Immunity

Pre-existing and <em>de novo</em> humoral immunity to SARS-CoV-2 in humans

A serological assay to detect SARS-CoV-2 seroconversion in humans

Vpr is required for efficient replication of human immunodeficiency virus type-1 in mononuclear phagocytes

Efficient generation of vesicular stomatitis virus (VSV)-pseudotypes bearing morbilliviral glycoproteins and their use in quantifying virus neutralising antibodies

R: A language and environment for statistical computing

Fitting linear mixed-effects models using lme4

Dynamics of a morbillivirus at the domesticwildlife interface: Canine distemper virus in domestic dogs and lions

COVID-19 in children and adolescents in Europe: a multinational, multicentre cohort study

Impact of COVID-19 on accident and emergency attendances and emergency and planned hospital admissions in Scotland: an interrupted time-series analysis

Features of 20 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: prospective observational cohort study

Antibody prevalence for SARS-CoV-2 in England following first peak of the pandemic: REACT2 study in 100,000 adults

Seroprevalence of anti-SARS-CoV-2 IgG antibodies in Geneva, Switzerland (SEROCoV-POP): a population-based study

Prevalence of SARS-CoV-2 in Spain (ENE-COVID): a nationwide, population-based seroepidemiological study

Disparities in Age-specific Morbidity and Mortality From SARS-CoV-2 in China and the Republic of Korea

Using social and behavioural science to support COVID-19 pandemic response

Ten years of serological surveillance in England and Wales: methods, results, implications and action

Serological Assays Estimate Highly Variable SARS-CoV-2

Neutralizing Antibody Activity in Recovered COVID19 Patients

Dynamics of neutralizing antibody titers in the months after SARS-CoV-2 infection

SARS-CoV-2 infection severity is linked to superior humoral immunity against the spike

Defining the features and duration of antibody responses to SARS-CoV-2 infection associated with disease severity and outcome

Neutralizing antibodies correlate with protection from SARS-CoV-2 in humans during a fishery vessel outbreak with high attack rate

Robust T Cell Immunity in Convalescent Individuals with Asymptomatic or Mild COVID-19

A mathematical model reveals the influence of population heterogeneity on herd immunity to SARS-CoV-2

COVID-19 herd immunity: where are we?

We thank Clare Orange, David Murray, Christine Willshire, Lisa Jarvis, and Giada Mattiuzzo for providing the serum samples required to carry out this work. We also thank Matt Turnbull and Suzannah Rihn (MRC-University of Glasgow Centre for Virus Research) for providing the pSCRPSY-hACE2 plasmid.

The authors declare no conflicts of interest. 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