Sera were obtained from distinct groups of subjects from well-validated cohorts that are described below (3, 5). Group 1: Non-vaccinated individuals without any reported COVID infection (NEG). Sera were obtained from subjects in May 2020, prior to widespread PCR testing and before the introduction of vaccines against SARS-CoV-2 infection. Sera were screened using a clinically validated, CE marked, ELISA assay that measures the IgG, IgA and IgM (IgGAM) response to the S glycoprotein (16, 17) (manufactured by The Binding Site (TBS; product code: MK654), Birmingham). This assay, described below in the section on detecting antibodies to S and N, has been clinically validated and reported to have a sensitivity of 98.6% (95% CI, 92.6-100.0%) and a specificity of 98.3% (95% CI, 96.4-99.4%) (16). Group 2: Individuals without evidence of infection (as determined by an absence of anti-N antibodies), vaccinated 28-35 days previously with BNT162b2 vaccine (VACC). Group 3: Individuals without evidence of infection (as determined by an absence of anti-N antibodies) who had received their second dose of BNT162b2 vaccine at least 28 days previously (DOUBLE VACC). Group 4: These sera were obtained in May 2020 from a cohort of healthcare workers from the University Hospitals Birmingham Foundation Trust, who had previously self-isolated a minimum of 28 days previously because they experienced symptoms suggestive of COVID-19, and had not been hospitalized for any of these symptoms. In May 2020 widespread PCR testing was not available, and thus most of these samples were not from individuals with prior confirmed PCR tests. Since the only predefined exclusion criteria was participation in existing SARS-CoV-2 vaccine trial or current COVID-19 symptomatology and the time was prior to the introduction of vaccines, anti-S antibodies could be used as a reliable surrogate of previous infection. Anti-S IgGAM were determined using the clinically validated anti-S glycoprotein ELISA described above (non-hospitalised convalescents, NHC). Group 5: Non-vaccinated convalescent, PCR confirmed SARS-CoV-2 infection patients who required ITU treatment, samples taken a minimum of 4 months after ITU discharge (ITU-CONV).

Ethical approval for obtaining samples for groups 1 -4 was provided by the London – Camden and Kings Cross Research Ethics Committee reference 20/HRA/1817. Ethical approval for obtaining samples for group 5 was provided by the North West ethics committee, Preston CIA UPH IRAS approval reference REC 20\NW\0240.

To generate the spike glycoprotein used in this study, HEK293F cells were transiently transfected with a pα-H plasmid containing the near full-length sequence for the Wuhan SARS-CoV-2 spike (GenBank: MN908947). The spike glycoprotein used here contains 1208 amino acids and includes all the S1 and most of the S2 domain (18, 19). The protein has been modified so that there are an additional four prolines present in addition to the two which are normally expressed (2P) to stabilize recombinant spike (18). The 6 prolines are present at positions 817, 892, 899, 942, 986 and 987 (18). This so-called HexaPro spike glycoprotein expresses as a metastable recombinant SARS-CoV-2 prefusion ectodomain. Extensive comparisons between the native, 2P and HexaPro spike glycoproteins demonstrate that they have comparable native-like protein architecture, have similar antigenic properties including the induction of neutralizing antibodies (20–22) and have similar glycosylation profiles (18, 19, 23, 24).

The HEK293F cells were cultured in Freestyle 293 Expression medium (Fisher Scientific) and maintained at a density of 0.2 x 10⁶ cells/mL at 37°C, 8% CO₂ and 125 rpm shaking. Prior to transfection, two solutions of 25 mL Opti-MEM (Fisher Scientific) medium were prepared. The expression plasmid encoding SARS-CoV-2 HexaPro was added to the first solution to give a final concentration of 310 µg/L. To the other solution, 1 mg/mL pH7 polyethylenimine (PEI) max reagent was added to generate a ratio of 3:1 PEI max:plasmid DNA. Both solutions were combined and incubated at room temperature for 30 minutes. Cells were transfected at a density of 1 x 10⁶ cells/mL and incubated for 7 days at 37°C, 8% CO₂ and 125 rpm shaking.

Cells were centrifuged at 3041g for 30 minutes at 4°C and supernatant was applied to a 500 mL Stericup-HV sterile vacuum filtration system (Merck) with a pore size of 0.22 µm. Purification of HexaPro S protein was undertaken using an ÄKTA Pure system (Cytiva). A 5 mL HisTrap Excel column (Cytiva) charged with Ni(II) was equilibrated using 10 column volumes (CV) of wash buffer (50 mM Na₂PO₄, 300 mM NaCl, pH 7). Supernatant was then loaded onto the column at a flow rate of 5 mL/min and washed with 10 CV of washing buffer containing 50 mM imidazole. Protein was eluted from the column in 3 CV of elution buffer (300 mM imidazole in washing buffer) and buffer exchanged to phosphate buffered saline (PBS) and concentrated using a Vivaspin column (MWCO 100 kDa) (Cytiva).

The nickel purified eluate was concentrated to 1 mL in PBS and injected into a Superdex 200 pg 16/600 column (Cytiva) to further purify trimeric S protein using size exclusion chromatography (SEC). The column was washed with PBS at 1 mL/min for 2 hours where fractions corresponding to the correct peak on the size exclusion chromatogram were collected and concentrated to ~1 mL as above.

Nucleocapsid was generated as a recombinant protein from E. coli by the Protein Expression Facility at the University of Birmingham (17).

Antibody ELISAs were carried out as previously described (17), with 50 µl per dilution used. In brief, 96 well high-binding plates (Corning) were coated with 0.1 µg S or N protein in PBS and incubated overnight at 4°C. PBS-0.1% Tween 20 was used to wash plates 3 times, and between all subsequent steps. Plates were blocked with 2% (w/v) BSA in PBS-0.1% (v/v) Tween 20 for 1 hr at room temperature (RT). Serum was diluted 1:40 and incubated for 1 hr at RT. HRP-conjugated anti-human secondary antibodies were added for 1 hr at RT. For combined anti-IgG, IgA and IgM (IgGAM) the antibodies came in a combined pre-diluted form from The Binding Site (EACONJ654). The individual constituent HRP-labelled secondary antibodies used in this are polyclonal rabbit anti-human IgG (1:16,000), polyclonal rabbit anti-human IgA (1:2000) and polyclonal rabbit anti-human IgM (1:8000). Individual immunoglobulin isotypes were detected using HRP-conjugated monoclonal antibodies: mouse anti-human IgM (clone AF6, 1:2000), mouse anti-human IgG1 (clone MG6.41, 1:3000), mouse anti-human IgG3 (clone MG5.161, 1:1000). All monoclonal antibodies were produced at the University of Birmingham. Plates were developed for up to 20 minutes using 100 µl TMB Core (Bio-Rad) and the reaction was stopped with 50µl 0.2M H₂SO₄. Optical density (OD) was read at 450 nm using a SpectraMax ABS Plus plate reader.

Plates were coated as above. Plates were washed three times with PBS-0.1% Tween 20 – this wash step was carried out between all subsequent steps. Blocking was carried out for 1 hr at RT with 2% BSA in PBS-0.1% Tween 20. Test serum was heat inactivated at 56°C for 30 minutes, before being diluted 1 in 5 with 2% BSA supplemented with 5 mM calcium chloride and 5 mM magnesium chloride. 50 µl was added to the antigen-coated plate and incubated for 1hr at 37°C. After washing, 50 µl COVID negative normal human serum (same source used throughout all assays, containing no detectable S or N specific antibodies as measured by IgGAM ELISA) at a dilution of 1:40 (in 2% BSA plus 5 mM calcium chloride and 5 mM magnesium chloride) was added to each well for 1hr at RT. 100µl of rabbit anti-C1q FITC antibody (Invitrogen PA5-16601) at a 1:200 dilution in PBS-0.1% Tween 20 was added and incubated at 37°C for 1 hr. HRP conjugated swine anti-rabbit (Dako P0399) at a 1:2000 dilution was then incubated for 1 hr. The assay was amplified using the Perkin Elmer ELAST amplification kit as per manufacturer’s instructions, with an optimised dilution of streptavidin, 1:800, incubated for 20 minutes. Plates were developed using 100 µl TMB Core (Bio-Rad) for 10 minutes, before being stopped with 50 µl 0.2MH₂SO₄. OD was measured as described above.

Microtiter plates were coated and washed as described above and blocked with Starting Block (ThermoFisher) for 10 min. Test serum was heat inactivated at 56°C for 30 minutes, before being diluted 1 in 5 with Starting Block supplemented with 5 mM calcium chloride and 5 mM magnesium chloride. 50 µl was added to the antigen-coated plate and incubated for 1hr at 37°C. After washing, 50 µl COVID negative normal human serum (same source used throughout all assays, containing no detectable S or N specific antibodies as measured by IgGAM ELISA) at a dilution of 1:40 (in 2% Starting Block plus 5 mM calcium chloride and 5 mM magnesium chloride) was added to each well for 1 hr at 37°C. The following anti-human monoclonal complement antibodies (100ul, diluted in PBS-0.1% Tween 20) were added and incubated at 37°C for 1 hr: mouse anti-C4b, 1:22,500 (Invitrogen, LF-MA0198); mouse anti-C3b, 1:10,000 (Invitrogen MA1-70053); mouse anti-C5b, 1:10,000 (Invitrogen DIA 011-01-02). HRP conjugated goat anti-mouse at a 1:4000 (Southern Biotech 1010-05) was then incubated at RT for 1 h. Plates were developed and read as described above.

Statistical analysis was carried out using GraphPad Prism 9.0. Kruskal-Wallis followed by Dunn’s post-hoc test for multiple groups was used to calculate p values. Statistical significance was accepted at P<0.05. Spearman correlation was carried out on the appropriate data sets.

Total IgGAM antibody responses to trimeric S and N were assessed in five different groups: individuals without any reported COVID-19 infection (NEG); post first BNT162b2 vaccine, infection-naïve individuals (VACC); post second BNT162b2 vaccine, infection-naïve individuals (DOUBLE VACC); convalescing non-hospitalised patients (NHC) and convalescing patients who had been hospitalised and required ITU treatment (ITU-CONV). The VACC, DOUBLE VACC, NHC, ITU-CONV groups all had significantly higher anti-S glycoprotein IgGAM responses than the NEG group ( Figure 1A ), whereas IgGAM levels against N in the two convalescent groups were higher than the NEG, VACC and DOUBLE VACC groups ( Figure 1B ). There were no significant differences in the anti-N responses between the NEG group and the VACC and DOUBLE VACC groups, consistent with individuals within these groups not having had prior SARS-CoV-2 infections. Similar results were observed when specific IgG1 responses, an IgG isotype efficient at fixing complement, were assessed ( Figure 1C ). No differences in anti-S IgGAM and IgG1 antibody responses were observed between the VACC and patient groups. Anti-S IgGAM and IgG1 responses were higher in the ITU-CONV group compared to the NHC group, but no differences were observed for anti-N responses in these two groups ( Figure 1D ). Modest IgM and IgG3 responses to S and N were detected in some individuals ( Supplementary Figures 1A, B and Supplementary c 1 ).

To determine if the complement protein C1q can bind to SARS-CoV-2-specific immunoglobulins in vitro we developed a solid phase C1q-binding assay. In these antigen-specific assays, the test serum from COVID-19 patients or vaccinees is heat-inactivated and standardisation of complement is provided by using sera from non-infected, non-vaccinated subjects. Results from this assay showed that C1q binding mirrored IgG1 levels for both S and N antigens, with the lowest signals for S seen in the NEG group ( Figure 2A ), and for N in the NEG, VACC and DOUBLE VACC groups ( Figure 2B ). The lack of C1q binding detected for N when sera from the VACC and DOUBLE VACC groups were tested is expected and consistent with a lack of prior infection by SARS-CoV-2 in these groups ( Figure 2B ), and we have included these two groups in our downstream analyses of the response to N for their value as control groups. No difference in C1q binding was observed between sera from the two convalescent groups ( Figures 2A, B ). Plotting IgG1 responses against C1q responses shows a positive correlation between the amount of IgG1 antibody and the amount of C1q binding detected ( Figures 2C, D ). Therefore, C1q binding reflects the serological response to both antigens.

To determine whether C1q binding reflected downstream activation of the complement cascade, we examined whether complement breakdown products could be detected. Deposition of C4b, a major component of the classical pathway C3 convertase, and the effector molecules C3b and C5b were assessed. In the absence of S or N-specific antibodies, C4b, C3b and C5b breakdown products were not detected, but they were detected in the presence of specific antibodies, indicating involvement of the classical complement pathway ( Figure 3 ). When S was used as the assay antigen, the highest median levels of C4b, C3b and C5b deposition detected were in the DOUBLE VACC and ITU-CONV groups ( Figure 3A ), whereas the VACC and NHC groups showed similar lower levels of downstream activation. The median levels of C4b, C3b and C5b deposition detected when N was used as the assay antigen were similar between the NHC and ITU-CONV groups ( Figure 3B ). In contrast, for the NEG, VACC and DOUBLE VACC groups, where anti-N antibody responses and C1q binding were not detected, there was no downstream activation of the complement cascade. Therefore, in this assay differences in complement activation by antibody can be detected dependent upon what patient group and antigen were examined.

In contrast to the linear association between anti-S IgG1 and C1q detection, there was a non-linear association between IgG1 and C4b, where C4b was only detectable beyond a threshold level of IgG1. A correlation between C4b and IgG1 to N was also observed, as was a threshold response, although the threshold response was less clear for N than for S ( Figure 4A ). Similar threshold responses for IgG1 to S and N were also observed if C3b or C5b were plotted against IgG1 ( Supplementary Figures 2A, B ). When correlations were performed for C1q vs C4b ( Figure 4B ), C4b vs C3b ( Figure 4C ) and C3b vs C5b ( Figure 4D ) for both antigens, then clear correlations were observed. This suggests that in this assay threshold levels of IgG1 are needed to activate downstream complement components for S and N.

IgM is the most efficient antibody isotype for fixing and activating complement (25) yet most sera tested had only low levels of anti-S or anti-N IgM ( Supplementary Figure 1A ). This was expected due to the convalescent nature of our infected groups, and the minimum one month period after vaccination that sera were obtained from our vaccinated groups. This suggested IgG1 or IgG3 might compensate for IgM when IgM is not present at high levels. Therefore, we examined the level of complement fixation (C1q binding) and activation (C3b deposition) associated with antigen-specific IgM and how this correlated with antigen-specific IgG1 and IgG3 levels ( Figure 5 and Supplementary Figure 3 ). To do this the IgM to S or N was correlated with C1q and C3b. There were modest correlations between IgM against S and C1q and C3b (r = 0.4613, p < 0.0001 and r = 0.4753, p < 0.0001 respectively; Figures 5A, C ). To N protein, no correlation was identified, although only a few sera had elevated levels of IgM (IgM vs C1q, r = 0.1661, p = 0.1155 and IgM vs C3b r = 0.09955, p = 0.3478; Figures 5B, D ) and if the 2 vaccinated groups (lacking anti-N responses) were excluded from the analysis for N, then the correlation became stronger (IgM vs C1q, r = 0.3019. p = 0.0162; IgM vs C3b, r = 0.03509, p = 0.0048). These results are consistent with IgM being able to activate complement when present, but not always being present at sufficiently high levels to do so. C1q binding and C3b deposition was detected for many sera in which IgM responses to S or N were low. We hypothesized that sera which exhibited the lowest IgM responses, but which could still fix and activate complement had higher levels of IgG1 or IgG3. We therefore divided the sera with the lower IgM levels into 2 groups, one group in which we detected the highest C1q or C3b levels (termed IgM^loC1q^hi and IgM^loC3b^hi respectively) and another group where we detected the lowest C1q and C3b levels (termed IgM^loC1q^lo and IgM^loC3b^lo respectively). The C1q or C3b response was then plotted against the IgG1 or IgG3 response. In the IgM^loC1q^hi and IgM^loC3b^hi groups higher levels of IgG1 were detected than in the IgM^loC1q^lo and IgM^loC3b^lo groups ( Figures 5A–D and Supplementary Figure 3 ), and this difference was observed for both S and N. IgG3 levels were also higher in the IgM^loC1q^hi and IgM^loC3b^hi groups than in the IgM^loC1q^lo and IgM^loC3b^lo groups ( Figures 5A–D and Supplementary Figure 3 ). Therefore, complement fixation and activation is observed in the presence of IgG1 or IgG3 when IgM levels are low.