Ethics Approval and Consent

Virus Watch was approved by the Hampstead NHS Health Research Authority Ethics Committee: 20/HRA/2320, and conformed to the ethical standards set out in the Declaration of Helsinki. All participants provided informed consent for all aspects of the study.

Participants

Participants (n = 2010) were a sub-cohort of Virus Watch (n = 58,628), a household longitudinal cohort study of SARS-CoV-2 infections in England and Wales running since June 2020. Recruitment and methodology of the full cohort have been described in detail elsewhere^21,22.

Recruitment criteria into the Virus Watch study were residence in England or Wales, household size up to six people with consent or assent of all household members, access to an email address, and ability to complete English-language surveys. The source population were households from the general population across all national regions of England and Wales. Households were recruited using several methods in order to achieve a target sample of > 50,000 people from the general population and representation of socioeconomically and ethnically diverse group; these included postal recruitment using a probability sample derived from the Royal Mail Post Office Address File, as well as including letter and SMS-based recruitment supported by general practices, and social media campaigns. Participants completed a detailed baseline questionnaire about demographic and clinical features for all household members, and subsequently completed weekly questionnaires about acute symptoms, SARS-CoV-2 tests, and vaccinations, and monthly questionnaires about detailed psychosocial and clinical topics tailored to the phase of the pandemic. A sub-cohort (n = 19,555) of participants over 18 years of age and resident in England also completed monthly finger-prick antibody testing for SARS-CoV-2 antibodies (see Outcomes section below) with samples collected during the period between 24/02/2021-23/03/2022; participants in the current study were drawn from this sub-cohort.

Further inclusion criteria for the current study were:

1. 1.

   returned at least one finger-prick antibody sample with a valid result for anti-S and/or anti-N antibodies,

2. 2.

   completed survey(s) about new-onset long-term symptoms which covered symptom development between February 2020 and March 2023, and had binary classifiable PCC status (see Exposure section below),

3. 3.

   had their first recorded SARS-CoV-2 infection detected via polymerase chain reaction (PCR) or lateral flow test (LFT) before end of serological follow-up,

4. 4.

   mild-moderate infection (i.e., convalesced in the community without hospitalisation).

Exposure

We conducted a nested longitudinal case-control study, as we were interested in the difference between antibody seropositivity and trajectories over time by PCC status. The primary exposures were binary PCC status and the interaction between PCC status and time since immunogenic events (i.e., infection and – where relevant – vaccination).

PCC Status

Participants were classified as having developed PCC if they reported one or more new-onset long-term symptoms following polymerase chain reaction (PCR) or lateral flow test (LFT) confirmed primary infection, with symptoms meeting the World Health Organisation consensus definition for PCC: interfering long-term symptom(s) which cannot be explained by another diagnosis with an onset within three months of SARS-CoV-2 infection and a duration of at least two months²³. Participants who had new-onset symptoms with a duration of less than two months or that developed outside of the three-month period following infection were excluded as they did not meet the WHO consensus definition of PCC but may have an immunologically distinct profile compared to those with an acute infection and no symptoms. There were insufficient numbers of these participants for detailed analysis (n = 49 not meeting duration criterion and n = 118 not meeting onset date criterion); participants not meeting onset date criterion were also excluded to prevent misclassification due to asymptomatic or undetected infection. If participants reported at least one symptom meeting the WHO consensus definition, they were classified as having PCC regardless of the duration or timing of any other reported symptoms. Participants were classified as not having developed PCC if they completed all long-term symptom surveys and never reported new-onset symptoms at any point during follow-up. Complete follow-up was required for acute-only participants to minimise misclassification of undetected new-onset symptoms.

To determine PCC status, participants were sent a questionnaire about new-onset long-term symptoms as part of the Virus Watch monthly surveys. These surveys requested participants to indicate whether they had experienced any new-onset long-term symptoms that could not be explained by another diagnosis or health state (e.g. other chronic illness or pregnancy) and provide onset dates and duration of interfering symptoms (see¹⁸ for further detail), which were used to determine PCC status according to the criteria above. The survey did not specify that these symptoms were linked to a SARS-CoV-2 infection, to avoid perceptions of PCC influencing participants’ answers. The survey was sent online to the Virus Watch cohort four times: in February 2021, May 2021, March 2022, and March 2023. Participants were asked to report new-onset long-term symptoms that developed within the previous year – covering the period between February 2020 and March 2023, which included the full period associated with antibody testing; only the May 2021 survey had a different recall period (from February 2020) as it was intended to supplement any non-response to the initial February 2021 survey regarding the previous year. While there was consequently some overlap in survey periods, symptoms could be matched by their onset date and consequently tracked if these overlapped across surveys.

Time since immunogenic events

Time was defined as the number of days between an immunogenic event and the antibody test date, with the event varying depending on the outcome of interest (anti-N and anti-S). The date range for sample inclusion began at 0 (i.e. day of immunogenic event) and the upper limit was determined depending on the outcome, with anti-N models capped at 365 days while the anti-S models were capped at a maximum of 200 days due to stratification (see Statistical Analysis section below). For logistic models (see Statistical Analysis section), time since immunogenic event was categorised into the following bands previously used in Virus Watch research related to seroconversion to facilitate interpretation of odds ratios by time period¹⁰: 0–29 days, 30–59 days, 60–89 days, 90–119 days, 120–269 days, 270+ days. In linear regression models, time in days was used to produce estimated antibody trajectories.

Infection was defined as evidence of first infection based on PCR or LFT based on linkage to UK national testing records or study-specific testing records. All participants had results available from linkage and also self-reported any SARS-CoV-2 tests taken across the study period in the weekly survey. PCR and LFT testing was also provided by the Virus Watch study during several periods, with the protocol varying over time (please see ref. ¹⁸ for details).

COVID-19 vaccination status was determined based on linkage to UK national vaccination records as well as self-reported vaccinations collected as part of the weekly survey, and was coded as (0, 1, 2, 3 doses). Samples from participants who received additional doses – which were only available to a minority of the UK population²⁴ – were excluded after the third dose.

Supplementary Fig. 1 illustrates data collection and classification of PCC status; serological sampling is described further below and associated analyses are illustrated in Supplementary Fig. 2.

Outcome

The outcomes of interest were anti-N and anti-S antibody levels, based on self-collected capillary blood samples (400–600 µl) collected between 24/02/2021-23/03/2022. Participants collected samples at home using test kits produced by the company Thriva, and returned kits using prepaid priority postage. Serological testing was conducted in UK Accreditation Service accredited-laboratories using the Roche Elecsys Anti-SARS-CoV electrochemiluminescence assays targeting total immunoglobulin (predominantly IgG, but also IgA and IgM) to the nucleocapsid (N) protein and the receptor binding domain in the S1 subunit of the spike protein²⁵. Further details of the laboratory testing process for Virus Watch samples are detailed in previous Virus Watch papers^10,26.

Antibody levels were expressed as semi-quantitative numeric values in form of cut-off indices (COIs) and log-transformed to base 10. For anti-N antibodies, the manufacturer-recommended seropositivity threshold was ≥1.0, with a sensitivity of 97.2-99.5% and specificity of 99.8%^27,28,29. Base-10 log transformed anti-N levels were included for samples taken between 0-365 days following PCR or LFT-confirmed primary infection. Samples that were collected following PCR- or LFT-confirmed reinfections were excluded due to the impact of reinfection on both antibody levels and unknown impact on long Covid symptomology; investigation into reinfections was beyond the scope of this analysis. Samples that were anti-N seropositive within 5 days following infection or that demonstrated a four-fold rise in levels between sequential samples taken beyond 120 days following primary infection were also excluded to remove otherwise undetected reinfections, based on established timelines of conversion and trajectories³⁰.

For anti-S antibodies, the manufacturer recommended seropositivity threshold was ≥ 0.8, with a sensitivity of 97.9-98.8% and a specificity of 100%^27,28,29. Anti-S levels were subject to detection limits that changed over time to allow investigation into quantitative antibody levels in the highly vaccinated UK population, with limits changing from 250 u/mL between 24/02/2021 – 30/06/2021 (excluding a two-day pilot of the protocol change to increase detection limits), to 25000 u/mL between 01/07/2021 – 01/01/2022, and to 100000 u/mL between 01/01/2022 – 21/03/2022; the assay remained consistent and increased detection limits were obtained through dilution. Samples for anti-S were only included from 01/07/2021 due to a large number of samples reaching the low initial detection limit of 250 u/mL prior to this time point. The later change in the detection threshold was addressed through stratification (see Statistical Analysis section below). Samples were included if they occurred prior to primary infection (i.e., for vaccination only models) or following primary infection and prior to any confirmed reinfection. Samples following confirmed or suspected reinfection as described above were excluded for the remainder of follow-up. As with anti-N, anti-S levels were log-transformed to base 10.

Stratification variables and covariates

The following variables based on data collected in an online demographic survey upon study registration were used to test for effect modification and stratify models and/or included as covariates in models: self-reported sex at birth (male or female), binary comorbidity status (presence of any condition on the UK NHS/government list denoting extreme clinical vulnerability or clinical vulnerability at COVID-19³¹, binary variant of infection based on infection date in reference to dominant variant in participants’ region of residence (pre-Omicron versus Omicron), and vaccination status at the time of infection (unvaccinated, two doses, or three doses). Please see the Statistical Analysis section for further details.

Statistical analysis

We used binary logistic mixed models to investigate how PCC status influenced the probability of seroconversion for anti-N. A random term was included to account for individuals submitting multiple samples. Separate models were constructed to evaluate probability of ever demonstrating anti-N seropositivity across the full follow-up period, as well as models evaluating anti-N seropositivity during the following time periods, to assess between-group differences in trajectories of seropositivity: 0-29 days, 30-59 days, 60-89 days, 90-119 days, 120-269 days, 270+ days. Results were expressed as odds ratios and predicted probabilities based on average marginal effects to facilitate between-group comparison over time. Seroconversion was investigated for anti-N only as this was the primary outcome and non-conversion is a more prominent feature of anti-N response¹⁰; only 3 participants in the current study did not seroconvert for anti-S.

We used linear mixed models to investigate how PCC status influenced the trajectory of log anti-N and anti-S antibody levels. The exposure was the interaction between PCC status and time since immunologic event, and the outcome was log anti-N/anti-S antibody levels. Anti-N was modelled for all participants across 365 days of follow-up, and a sensitivity analysis was conducted including only samples from participants who seroconverted for N during the study period. As anti-S antibodies respond to both vaccination status and infection status and the combination of these events may differentially affect antibody levels, models were stratified according to these characteristics. Only samples following the increase of the cap to 25000 u/ML were included, as a substantial number of samples with the early cap (250uML) reached this threshold, possibly precluding accurate estimations of levels and between-group differences. The sample period corresponded to periods of two-dose vaccination onwards in the Virus Watch study population, so anti-S models were constructed to investigate response to two-dose and three-dose vaccination as follows: pre-infection (i.e. vaccination response only), hybrid immunity with vaccination before the infection, and hybrid immunity with vaccination after the infection. A schematic diagram of these models illustrating the timing of immunogenic events is provided in Supplementary Fig. 2. All anti-S models were capped at 200 days follow-up due to data availability and UK vaccination schedules, except for the post-second-dose anti-S model (infection most recent event), which was capped at 90 days due to low sample availability beyond this point.

We tested models with time modelled as a linear term, a quadratic term, and with a B-spline with a single knot at 120 days for anti-N¹⁰, 30 days for post-second-dose anti-S models³², and 14 days for post-third-dose anti-S models³³. Models were selected based on Bayesian Information Criterion values. The spline models were used for anti-N and for anti-S post-second dose (pre-infection and post-infection with vaccination as the most recent outcome); time was modelled using a linear term for the remaining anti-S models.

Conceptual models and effect modification

Conceptual models underlying these analyses are presented in Supplementary Fig. 3a for anti-N and Supplementary Fig. 3b and 3c for anti-S antibodies. These analyses did not aim to estimate the causal effect of PCC status on antibody responses. Rather, differences in antibody responses to infection by PCC status were investigated to provide evidence for differential immune processes and/or viral persistence, which are proposed mechanisms for PCC development, and which could not be directly measured here (denoted as node ‘U’ in Supplementary Fig. 3a-c). Consequently, antibody levels provide evidence as a proxy for a hypothesised process underlying PCC and are not themselves a traditional causal exposure (i.e., a relationship between antibody responses and PCC would provide evidence of an unmeasured underlying immune process, but antibodies themselves are not believed to cause PCC directly). Pre-infection demographic and clinical features are therefore proposed to influence these unmeasured mechanisms, and are consequently conceptualised as effect modifiers rather than confounders. This is because they are proposed to influence the strength and direction of the effect (i.e. effect modification) rather than inducing a potentially spurious relationship (i.e. confounding)³⁴ and were consequently investigated using stratification following methodological recommendations³⁵. Infection severity was limited to mild-moderate community infections within this study based on inclusion criteria and study composition. The impact of variant of infection and vaccination on infection severity was accounted for using stratification and adjustment in anti-N-related models, with further description and justification provided below. The direct effect of vaccination on anti-S antibody responses was addressed through stratification as described in the previous section.

We consequently assessed effect modification by sex at birth, comorbidity status, binary variant (pre-Omicron vs Omicron), and vaccination status at the time of infection (unvaccinated, 2 doses, or 3 doses) for anti-N seroconversion models covering the full study period. Other vaccination status categories could not be included due to insufficient anti-N samples. We included interaction terms adding sex, comorbidity status, variant, and vaccination status and evaluated evidence of the interaction providing additional explanatory power to the model using likelihood ratio tests. Given the temporal overlap between variant of infection and vaccination status and their independent relevance to infection severity, we adjusted the vaccination-stratified model for variant and vice versa. Other demographic and clinical features, such as granular clinical conditions and smoking status, could not be included as stratification variables due to limitations on sample size and/or data availability.

Predicted antibody trajectories were similarly presented stratified by sex and comorbidity based on the inclusion of an interaction term. Due to small subgroup sizes over time according to variant and vaccination status for anti-N samples, we were not able to provide stratified waning trajectories for anti-N according to variant and vaccination status. However, given the relevance of these factors both to the immune response to infection and as a potential proxy for infection severity, we conducted an adjusted analysis of the anti-N trajectory accounting for these two factors. We also adjusted the age- and sex-stratified models for variant and vaccination status.

Interaction tests were conducted for anti-N models only as anti-S models were already stratified according to vaccination status and samples were not sufficient to meaningfully assess three-way interaction for antibody trajectory across all models and timepoints. We lacked the sample size to assess effect modification for more granular variables, such as specific pre-Omicron variants of infection.

Pre-infection, vaccination-related anti-S responses were also investigated by PCC status (i.e., following later infection) to evaluate evidence for any pre-infection differences in immune response following challenge with the SARS-CoV-2 spike protein. The associated conceptual model is illustrated in Supplementary Fig. 3c. As described for the infection-related models, demographic and clinical features are appropriately conceptualised as effect modifiers within this framework.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.