Immunoglobulin-G enzyme-linked immunosorbent assay predicts neutralising antibody response 1 in convalescent SARS-CoV-2 patients 2

32 Severe acute respiratory coronavirus 2 (SARS-CoV-2) has spread globally since its emergence in 33 2019. Most SARS-CoV-2 infections generate immune responses leading to rising levels of 34 immunoglobulins (Ig) M, A and G which can be detected using diagnostic tests including enzyme- 35 linked immunosorbent assays (ELISA). Whilst implying previous SARS-CoV-2 infection, the detection 36 of Ig by ELISA does not guarantee the presence of neutralising antibodies (NAb) that can prevent the 37 virus infecting cells. Plaque reduction neutralisation tests (PRNT) detect NAb but are not amenable 38 to mass testing as they take several days and require use of viable SARS-CoV-2 in high 39 biocontainment laboratories. We evaluated the ability of IgG and IgM ELISAs targeting SARS-CoV-2 40 spike subunit 1 (S1) and nucleocapsid protein (NP) at predicting the presence and magnitude of NAb 41 determined by PRNT. SARS-CoV-2 IgG ELISA correlated well with NAb and was highly sensitive 42 (93.8% [95% CI 79.2-99.2]) and specific (88.9% [95% CI 51.8-99.7%]) at predicting the presence of 43 NAb. There was not a strong correlation between IgM ELISA and PRNT result. IgG ELISA provides a 44 useful, high throughput method of predicting the presence of neutralising antibodies, with higher 45 ELISA results increasing the likelihood of having a greater NAb titre.


Serum samples 83
Anonymized EDM serum samples from hospital patients with SARS-CoV-2 infection confirmed by 84 reverse transcription -quantitative polymerase chain reaction (RT-qPCR) were used for this study 85 and were selected from 645 EDM serum samples that were collected from a pool of 177 patients 86 treated at St George's Hospital, London UK [15]. Where possible, samples were selected from 87 patients at least 10 days post-RT-qPCR confirmation. Samples were grouped based on their 88 normalised optical density (NOD) values derived from an anti-SARS-CoV-2 IgG ELISA [15] into 89 "negative NOD" values (< 0; indicating the patient had not seroconverted), "low NOD" (0 to 0.5), 90 "medium NOD" (0.9 to 1.1); and "high NOD" (> 1.5). The final sample available from all patients was 91 chosen for this study. The narrow "medium NOD" window was purposely selected to reduce sample 92 numbers in this grouping, as the grouping 0.5 to 1.5 contained 5-6 times more samples than the 93 other groupings. A single sample was then selected from any patient with at least 3 samples with 94 NOD values remaining in one NOD grouping (i.e. indicating a stable antibody response). The serum 95 sample selected for any given patient was that collected furthest from the swab taken for 96 confirmation of SARS-CoV-2 infection (and at least 10 days post-swab). The approach resulted in 9, 9, 97 11 and 12 patient samples in each group (41 single patient serum samples in total). 98 All participants were confirmed as positive for SARS-CoV-2 using RT-PCR from nose/throat swabs (in 99 Sigma Virocult®, Corsham, UK) and Roche RNA extraction kits (Magnapure, West Sussex, UK) 100 followed by Altona Diagnostics RealStar® SARS-CoV-2 RT-PCR (S and E target genes, Hamburg, 101 Germany) or Roche cobas® SARS-CoV-2 Test (E and ORF target genes). 102

Enzyme-linked immunosorbent assays to detect anti-SARS-CoV-2 IgM 103
Anti-SARS-CoV-2 IgM ELISAs (Mologic, Bedfordshire, UK), which targets the nucleocapsid (NP) and 104 spike protein subunit 2 (S2) antigens, were used to measure antibodies, as per the manufacturer's 105 instructions. Briefly, sera were diluted and incubated on a pre-coated plate (30 minutes) at room 106 temperature and then washed three times. Conjugated antibody (anti-human IgM) was then applied 107 to each well and incubated (30 minutes) at room temperature. Following washing (x4), TMB 108 substrate was added and incubated for 10 minutes at room temperature before addition of stop 109 solution. Optical densities (OD) were read at 450nm within 10 minutes of addition of the stop 110 solution. 111

Plaque reduction neutralisation tests 112
Vero E6 cells were seeded into 24-well cell culture plates at a density of 250,000 cells/ml and 113 incubated (24 hours, 37°C, 5% CO 2 ). The following day serum samples were heated to inactivate 114 complement (56°C for 1 hour). Heat-inactivated serum samples were 2-fold serially diluted in 115 infection media (DMEM with 2% v/v FBS and 1:1000 50mg/ml gentamicin). Under biosafety level 3 116 conditions, SARS-CoV-2 isolate REMRQ0001/Human/2020/Liverpool [16] was added to an equal 117 volume of diluted patient serum, at a titre of 800 pfu/ml, to achieve 12 final serum dilutions from 118 1:20 to 1:40960 for each patient sample. Following incubation (1 hour, 37°C), the virus-serum 119 mixture (100µL) was inoculated onto Vero E6 cells and incubated (1 hour, 37°C, 5% CO 2 ) before 120 applying an overlay of infection media containing agarose (0.4% w/v). Infected cells were then 121 incubated (48 hours, 37°C, 5% CO 2 ). The assays were fixed with formaldehyde (37% w/v), stained 122 with crystal violet solution (0.25% w/v) and allowed to air dry. The PRNT 80 was determined as the 123 lowest dilution of serum that produced a ≥ 80% reduction in the number of plaques compared to 124 controls that contained no patient serum. The investigators were blinded to the ELISA status of the 125 samples when performing the PRNTs. 126

Western blots 127
Western blots were conducted to investigate the antigen binding profiles associated with the 128 neutralising responses. Recombinant spike subunit 1 protein (S1), spike subunit 2 protein (S2) and IgG, IgM and timing of serum sampling post-symptom onset with PRNT 80 . Briefly, the PRNT 80 outcome 150 for the i-th patient, denoted by , was modelled as an ordinal variable taking the values from k=1, 151 corresponding to a titre less than 1:20, to k=10, for a titre of 1:2560 or more. We assumed that the 152 observed titre values corresponded to the discretization of a continuous antibody distribution, * , 153 which was assumed to follow a log-Gaussian distribution with mean and variance 2 . 154 This can be summarised as: 155 where 1 = 0, 2 = 20, 3 = 40 and so forth up 10 = 2560 and with the convention 11 = ∞. 157 The probability in (1)  Twenty-five (61.0%) of the 41 patients were male, and the median age was 63 (IQR 55-71) years. 169 Seventeen patients (41.5%) were classified as white, 13 (31.7%) non-white, and 11 (26.8%) were of 170 unknown or 'other' ethnicity. Twenty-three patients (56.1%) had one or more comorbidities. Ten 171 patients (24.4%) were obese, with a body mass index (BMI) >30. These patients come from a subset 172 already described [15]. 173 Of the thirty-nine patients for whom symptom data was available, 35 (89.7%) were symptomatic at 174 the time of their initial swab. Thirty-three of these patients (94.3%) had one or more of the classic 175 triad of symptoms: cough, fever, and shortness of breath, 12 (34.3%) had gastrointestinal symptoms, 176 including 9 patients with diarrhoea. Three patients (7.3%) had an incidental positive swab, taken prior to admission to a rehabilitation facility, and three patients were swabbed following contact 178 with a patient who had tested positive. The median interval between the onset of symptoms and 179 date of the first positive swab was 4 days (IQR 3-7 days). Seven (17%) patients died within 28 days of 180 their first positive swab. The median timing of serum sampling post symptom onset was 29 days 181 (range 13-60 days). The sample timing post-symptom onset was not available for six patients. 182 The correlation of IgG ELISA and neutralising responses are shown in Figure 1. Using   It was assumed that the relationship between IgG titre and PRNT 80 was unlikely to be linear across all 203 ranges of observed IgG. Therefore, the model predictions for PRNT 80 based on IgG were considered 204 for the first, median, and third quartiles of the IgG titres (see Figure 2). For the first quartile, there 205 was a low probability (<0.2) of observing neutralising titres ≥ 1:20 (Figure 2A). For a median IgG titre, 206 the highest probability (0.15) was for PRNT 80 1:160 ( Figure 2B). IgG ELISA results in the 3 rd quartile 207 had a generally higher probability of a higher PRNT 80 than median or 1 st quartile IgG titres, albeit 208 with increasing uncertainty at higher PRNT 80 values ( Figure 2C). Across the range of IgG titres 209 observed, we predicted that increasing IgG corresponded to PRNT 80 , but with greater uncertainty in 210 predicting the neutralising response at higher IgG titres (Figure 2D). 211 Antibody binding to all three of the S1, S2 and NP antigens, or to both the S2 and NP antigens, were 212 seen in all samples that had high neutralising titres (PRNT 80 ≥1:80) (Figure 3). In samples with low 213 PRNT 80 (≤1:40) there was greater variability in antigen binding, with a larger proportion 214 demonstrating antibody binding to single antigens, or combinations of two antigens involving S1. 215 One IgM and IgG positive sample by ELISA showed binding to only NP. Three of the six samples that 216 did not achieve PRNT 80 ≥1:20 demonstrated no binding to S1, S2 or NP. 217

Discussion 218
Our data show that IgG ELISA (NOD 450 cut-off 0.2) can predict the presence of a NAb titre of ≥PRNT 80 219 1:40 with high sensitivity (93.8%) and specificity (88.9%), and therefore can be used as a proxy of 220 neutralising response to SARS-CoV-2 in convalescent patients. Our ordinal outcomes model 221 demonstrates higher NAb titres correlate with increasing IgG titres. These findings are supported by 222 previous studies, which report a significant correlation between anti-spike and anti-RBD IgG titres 223 with the neutralisation titres established by microneutralisation tests, PRNTs and pseudotyped virus 224 neutralisation assays [6,14,[17][18][19]. However, our findings also demonstrate considerable IgG NOD 225 variation for samples within the same PRNT 80 category, indicating it is not possible to make accurate 226 quantifiable predictions of the expected PRNT 80 based on the IgG ELISA titres alone. 227 The IgM ELISA NOD was not significantly associated with the NAb titres and was less sensitive and 228 specific at identifying a neutralising response (≥PRNT 80 1:40) than the IgG ELISA. Previously, IgM 229 ELISAs have been reported to be more predictive of neutralising titres than IgG [13,17,20]. Given 230 the short duration of IgM expression, the timing of serum sampling post-infection is an important 231 determinant of these relationships [13,17,20]. Furthermore, as serum samples were selected to 232 provide a range of IgG titres, it is possible that a larger sample size could have provided a wider 233 range of IgM titres to detect a relationship. IgA titres correlate well with neutralisation titres [21]. As 234 we did not measure IgA, it is possible that its neutralising activity may explain why some samples 235 had greater PRNT 80 values than expected from the IgG titres. 236 Anti-NP antibodies are not considered to have protective activity, despite correlating with 237 neutralising titres [2]. Our western blot analysis revealed that the majority of serum samples with 238 neutralising activity had antibodies directed against the NP, S1 and S2. This finding suggests that 239 anti-NP antibodies are raised as part of a suite of antibodies, and whilst not directly neutralising, are 240 indicative of the presence of other neutralising immunoglobulins. In general, the inclusion of the NP 241 in an ELISA does not seem to prevent accurate prediction of the presence of a neutralising response. 242 For example, the Roche Elecsys Anti-SARS-CoV-2 ELISA, which targets only the NP has a similar 243 performance predicting neutralisation as the Abbott SARS-CoV-2 IgG ELISA, which targets both the 244 NP and S1 [22]. However, one sample with a PRNT 80 <1:20 which was positive by IgG and IgM ELISA 245 in our study, only showed binding to the NP on the western blot. This highlights how isolated NP 246 binding is capable of producing ELISA positive results which are not associated with neutralisation. 247 The main limitation of our study is the small sample size. As the relationship between serological 248 assays and neutralising titres appears to be variable, a larger sample size would have sufficient 249 power to reveal overall trends and minimise the effects of outliers. Moreover, our cohort only 250 includes hospitalised patients with severe COVID-19 and therefore the findings of this study may not 251 be applicable to individuals with mild and asymptomatic infections, who may foster different 252 antibody responses. 253 In conclusion, IgG ELISA targeting SARS-CoV-2 S2 and NP can predict the presence of a NAbs to SARS-254 CoV-2 in convalescent patients hospitalised with severe COVID-19. 255    Figure 1C shows