Patient and public involvement

There was no patient or public involvement in the conceptualisation or design of this review.

Identification of studies

Keyword-structured searches were performed in MEDLINE, Embase, COVID-19 Primer and the Public Health England library [12] for articles published from 01/01/2020-26/06/2020. A sample search strategy is in S1 Appendix in S1 File. Subject area experts were consulted to identify relevant papers not captured through the database searches.

Definitions, inclusion and exclusion criteria

We included studies in all human and animal populations, and in all settings (laboratory, community and clinical—encompassing primary, secondary and tertiary care centres) relevant to our research questions. We excluded the following study designs: case reports, commentaries, correspondence pieces or letter responses, consensus statements or guidelines and study protocols.

We focused on studies reporting measured titres (total antibody, IgA, IgG and/or IgM) with follow-up duration of greater than 28 days (which we defined as the limit of the acute phase of illness). Shorter follow-up studies were included if they reported on protective immunity, or immune response correlates. We defined “correlates” as encompassing, among other factors: primary illness severity—proxied by the WHO’s distinction between “mild”, “moderate”, “severe” and “critical” illness [13]; subject age; gender; the presence of intercurrent or co-morbid disease e.g. diabetes, cardiovascular and/or chronic respiratory disease; and ethnicity.

Selection of studies

Studies were independently screened for inclusion on title, abstract and full text by two members of the research team (working across four pairs), with arbitration of disagreements by one review lead.

Data extraction, assessment of study quality, and data synthesis

Data were extracted in duplicate from each included study. Extraction was performed directly into a dedicated Excel template (S2 Appendix in S1 File). Pre-prints of subsequently published peer reviewed papers were included and results extracted where substantial differences in reported data were identified; if little difference was observed only the peer-reviewed version was retained.

Critical appraisal for each included study was performed in duplicate using a version of the MetaQAT 1.0 tool, adapted for improved applicability to basic science and laboratory-based studies. MetaQAT was selected for its simplicity and versatility in application to studies of all design types [14]. Principal adaptations to the MetaQAT tool are described in S3 Appendix in S1 File.

The adapted MetaQAT tool was used to gather both qualitative (narrative) feedback on study quality and scaled responses (yes/no/unclear) for answers to key questions around study reliability, internal and external validity, and applicability, among other fields. These data provided the basis for quality assessments for each paper included in the review.

Study heterogeneity precluded formal meta-analysis. Results were instead synthesised narratively.

Ethical approval

This was a systematic review based on analysis of openly published secondary data and did not involve humans. No ethical approval was required.

General characteristics of included studies

150 studies were included, of which 108 (72%) contained data pertaining to antibody response, and 70 (47%) to protective immunity (descriptive statistics for included studies are given in Table 1). The vast majority (108 or 72%) focused on hospitalised patients (i.e. higher severity disease). Eleven studies (7%) considered antibody responses in asymptomatic individuals in the community and only five (3%) investigated protective immunity in this group. Most studies had significant methodological limitations. Assays used to detect and quantify antibody response were diverse, with target antigens including spike (S), S1 and S2 subunits, receptor binding domain (RBD) and nucleocapsid (N). Details of assays used, and an overview of strengths and limitations of these is provided in S4 Appendix in S1 File.

Table 1

Summary of characteristics of included studies.

	Antibody response	Protective Immunity	All papers
Article type
Pre-print	57	43	81
Peer reviewed paper	51	26	68
Report	0	1	1
Study designs
Cohort	58	18	66
Case control	20	14	26
Case series	15	8	21
Basic science	4	22	24
Narrative review	4	2	4
Systematic review with meta-analysis	5	2	5
Systematic review without meta-analysis	1	1	1
Non-randomised trial	1	2	3
Subjects
Human	104	58	137
Animal	1	6	6
Both	3	6	7
Country of origin
China	41	20	54
USA	15	14	21
Europe excl UK	29	9	35
UK	7	5	10
Other countries	5	4	7
Multiple populations	10	6	11
Lab or animal based*	1	12	12
Study setting
Hospital patients	70	33	85
Mixed hospital and community	18	11	24
Community	13	5	18
Unclear	6	12	14
Animal only study	1	9	9

Note that some studies addressed both Ab and protective responses, and hence are counted in both of the columns relating to these topics above.

Kinetics of the antibody response

Time to seroconversion

The majority of individuals in the included studies mounted a SARS-CoV-2-specific antibody response during the acute phase of illness, with many studies reporting 100% seroconversion. Overall seroconversion rates depended on the time point at which testing was conducted in the disease course, the populations under study, the serology assay platforms used and their specific target proteins. Studies considered time to seropositivity for total antibody and/or individual antibody classes (IgA/IgG/IgM) (Fig 2), although this was often not clearly defined with respect to symptom onset or first positive PCR test. In addition, whilst some studies described specific target proteins of assays used, others were either non-specific or not described. This limited assessment of dynamics of antibodies against specific viral targets, in particular anti-N versus anti-S, the latter of which may be more closely related to protection.

Fig 2

Forest plot of median time to seroconversion by severity across included studies.

Central points in the forest plot represent the median reported by each study overall. Included studies reported the distribution of times to seroconversion around the median in different ways: lines with whiskers represent ranges (from maximum to minimum around the median); lines without whiskers represent interquartile ranges around the median; for a small number of studies only point estimates were provided.

A number of studies reported seroconversion for total antibody (combined IgG, IgM and/or IgA) [15–21], however the focus of findings presented is for specific antibody isotypes. For IgG, mean or median time to seroconversion ranged from 12–15 days post symptom onset [7, 9, 15, 22–26], with wide variation in first to last detection of IgG from four to 73 days post symptom onset although reporting methods varied by study [15, 27–33]. For IgM, mean or median time to seroconversion ranged from four to 14 days post symptom onset [7, 9, 15, 22–24, 26, 31, 34], again with variations in reporting methods, study quality, and sample size giving rise to uncertainty around findings. Time to seroconversion for IgA was measured in fewer studies, ranging from four to 24 days post symptom onset, although most were within four to 11 days [23, 35, 36], with some outliers, including two reports of 24 days to first detection [37, 38].

Antibody dynamics over time

IgG dynamics appeared to follow a pattern of peak, plateau, and persistence at lower levels (Fig 3). After appearance, IgG titres rose to a peak between three and seven weeks post symptom onset [7, 23, 30, 42–48], with studies recording the presence of IgG in and beyond weeks four [40, 49], five [50], six [23, 51], seven [52, 53], and eight [17, 45, 54–56] post symptom onset. Some studies reported a plateau in virus-specific IgG beyond week three but levels beyond the peak were not well described [32, 57–59]. A decrease in antibody levels was reported in the eighth week post symptom onset by two studies [17, 38], while another reported a decline from the second month after symptom onset [58]. Evidence from a cohort of 40 UK patients suggests a decline in titres after eight weeks [58], although persistence of virus-specific IgG has been described at varying levels up to 12 weeks post symptom onset [43], the longest follow up period among included studies. Dates of last detection were limited by the length of the study follow-up period, rather than confirmation of disappearance of detectable antibody titres.

Fig 3

Schematic showing the scale of IgG/IgM/IgA/Neutralising Ab response over time from disease onset.

Note that the y-axis is illustrative only and therefore no scale is given: this figure gives an indicative overview of findings from all included studies with relative peaks and decline indicated.

IgM dynamics follow a ‘rise and fall’ pattern, with a peak two to five weeks post symptom onset [7, 26, 30, 34, 43, 46, 47, 53, 60] then decline over time to below the detection limit [38, 43, 61]. Beyond the peak, IgM is consistently reported to decrease from as early as two to three weeks [53, 60], to as late as eight weeks [55] post symptom onset, with the majority of studies reporting this decline to occur at between three to five weeks [40, 43, 61, 62]. Virus-specific IgM became undetectable in almost all cases by around six weeks after disease onset in two small but well-conducted cohort studies [53, 63].

Fewer studies describe IgA dynamics compared to IgM or IgG. IgA levels are reported to peak between 16–22 days post symptom onset, although there is no consensus on trends over time [23, 60].

Correlates of antibody response

Key findings regarding correlates of the antibody response to SARS-CoV-2 infection are summarised in Table 2. Included papers addressed clinical factors (disease severity, co-morbid disease status and symptom profile) and demographic factors (age, sex and ethnicity) although results for many of these factors were conflicting or inconclusive. Across all papers, the definitions of comparator groups were highly variable, including disease severity classifications (severe/mild), outcomes (deceased/mild), and treatment categories (ICU/Non-ICU). The lack of consistency in methods, comparison groups and study design means it is not possible to determine whether or how disease severity affects, or is affected by, the antibody response. Most studies showed no association between antibody response and age or sex, and, when taken together, studies that did show associations had inconclusive results and lacked statistical analysis to relate these findings to disease severity. There were virtually no data to describe the immune response according to ethnicity.

Table 2

Summary of evidence on correlates of antibody response from studies included in this review.

Category	Correlate	Dimension or sub-population	Findings
Clinical	Disease severity	Longitudinal trends in Ab production	Most studies report no relationship between time to seroconversion for IgG or IgM and disease severity [22, 30, 35, 53, 56, 64]. A number of studies report earlier antibody response to more severe disease, including specifically for IgG [7]; a shorter time to peak antibody titre [19, 65–67], and that IgG and IgM persist for longer in severe disease compared to milder cases [68, 69]. Conversely, several studies report an earlier antibody response to milder disease [26, 43], including specifically for IgM [70] and IgA [69, 71]. One study reported shorter time to peak for IgG in mild cases [69], and another reported IgG persisting for longer compared to severe cases [72]. One study reports lower seroconversion rate amongst asymptomatic compared to symptomatic cases [73]. Two other studies on this association were limited by lack of detail on methodology and participant selection [74, 75].
		Antibody titre	Ten studies reported that more severe cases have higher IgG titres [23, 34, 39, 42, 62, 76–80], three studies reported no difference between mild and severe cases [44, 52, 81]. Five studies reported severe cases have higher IgM titres [23, 34, 39, 45, 53], two studies reported no difference in between mild and severe cases [63, 81], and three studies reported severe cases have lower IgM titres [80, 82, 83]. Three studies report severe cases have higher IgA titres [23, 69, 78], one study reports no difference between mild and severe cases [82].
	Co-morbid disease		Two studies report an association between co-morbidities and seroconversion or antibody positivity [63, 84], with one finding immunocompromised individuals developed a lower response [85]. One study reports antibody responses to be independent of co-morbidities, for both IgA and IgG [78]. One study on this association was limited by sampling bias leading to low quality [86].
	Symptom profile		Several studies report an association between COVID-19 symptoms and: seropositivity [87–89], higher titres of IgG [29, 79, 90] and anti-RBD and anti-S antibodies [91]. Two studies reported that asymptomatic healthcare workers did develop antibodies [92, 93]. Fever appears to have a consistent relationship with seropositivity and antibody titres [29, 89, 91], although other symptoms such as ageusia have also been associated [87, 89].
Demographic	Sex		Four studies report no association between antibody titres (either IgG or IgM) and sex [21, 78, 94, 95]. One study reports a higher IgG titre in women, in ‘severe’ patients only [96]. One study reports a delayed peak of antibody response in men [97]. Two studies report a higher proportion of women tested positive for antibodies [87, 95]. One study found higher anti-RBD and anti-S antibodies in male plasma donors [91]. Another study found higher concentrations of IgM in male cases [34].
	Age	Older adults	Five studies found no association between antibody response and age, for both antibody positivity [63] or IgM/IgG titre [21, 58, 78, 98]. One study found that seroconverters were older than non-seroconverters [84] and another that the concentration of IgG was related to age [34]. Three studies reported that older people had higher titres of IgA and IgG [23], IgM [94], and anti-S and anti-RBD IgG [91].
		Children	Two studies found children generally developed a detectable antibody response to SARS-CoV-2 infection [99, 100]. Two studies found children with pneumonia generally mounted lower IgG [101] and IgA responses [102]. One study found no significant difference between the levels of antibody in children and adults [98]. Two studies reported most neonates born to COVID-19 positive mothers had raised IgM [103, 104], and COVID-19 recovered donor breast milk was found to have reactive IgA in one study [105].
	Ethnicity		One study reported non-white ethnicity was associated with higher antibody levels than white ethnicity [84].

Protective immunity

Correlates of neutralising antibody production

Clinical and demographic correlates of the neutralising antibody response are described in Table 3. Neutralising antibody responses correlated with disease severity in all studies in which this association was tested [7, 43, 49, 108, 117, 126, 127, 135–137]. Importantly, the few studies that investigated asymptomatic cases found those individuals were considerably less likely to develop detectable serum neutralising antibody responses than cases with symptoms. With regard to age and sex, evidence was mixed and a limitation across all papers was a lack of statistical adjustment for severity.

Table 3

Summary of evidence on correlates of neutralising antibody response from studies included in this review.

Category	Correlate	Dimension or sub-population	Findings
Clinical	Disease severity	Longitudinal trends in Ab production	One study reported asymptomatic cases with neutralising antibodies were more likely to lose detectable neutralising antibodies in the convalescent phase [90].
		Antibody titre	Ten studies reported a higher titre of neutralising antibodies in more clinically severe cases [7, 43, 49, 108, 126, 127, 134–137]. One study reported undetectable neutralising activity in plasma from a majority of asymptomatic cases [126].
Demographic	Sex	N/A	Five studies reported neutralising antibody response was positively correlated with male sex [108, 110, 127, 135], although it is unclear how this is related to disease severity [49]. One study reported a positive correlation between neutralising antibody formation and female sex [126].
	Age	Older adults	Two studies reported increasing neutralising antibody response with increasing age [20, 133]. Two studies reported no association between neutralising antibodies and age [110, 126].
		Children	Two studies reported children can develop a neutralising antibody response [48, 100].
	Ethnicity	N/A	One study reported that individuals with Hispanic/Latino ethnicity were more likely to have detectable neutralising antibody responses (although this study was based on a convenience sample in an atypical cohort—US service personnel on a warship) [107].

Summary of findings

This review narratively synthesis the findings from 150 studies published by the end of June 2020. The pace of production of evidence over time in regard to SARS-CoV-2 has been exceptional, and further evidence is now available with which to contextualise these earlier contributions to our understanding of SARS-CoV-2 infection. In the discussion that follows we summarise key findings from the included studies and highlight in addition (where relevant) evidence published since our searches were completed.

Most people who experience symptomatic SARS-CoV-2 infection undergo seroconversion to produce a detectable, specific antibody response in the acute phase (≤28 days). The kinetics of the antibody response to SARS-CoV-2 follow typical immunological paradigms: virus-specific IgM rises in the acute phase to a peak around two to five weeks following disease onset, then declines over a further three to five weeks before becoming undetectable in many cases; IgG peaks later (three to seven weeks following disease onset), then plateaus, persisting for at least eight weeks with some evidence suggesting a moderate decline over that period. However, understanding of IgG dynamics over time is limited by the understandably short duration of follow up in studies published for inclusion in this review. Studies with longer follow are now be starting to show persistence of IgG for at least three to four months [161, 162].

Evidence suggests the majority of those infected with SARS-CoV-2 develop nAbs—a finding consistent with that for SARS-CoV-1 and MERS-CoV [7]. The magnitude of this response appears to correlate with disease severity, although not necessarily the kinetics. Neutralising antibodies are initially detectable from around seven to ten days, peaking at around three weeks and then declining. Further evidence is required to evaluate comprehensively the apparent waning of the nAb response over time: in the studies included here although nAb may be detectable, higher quality studies suggest that titres are generally low, and the response is short lived. Although this is supported by recently published data (beyond the date cut off for inclusion in this study) from a UK cohort of healthcare professionals with follow up to three months [163], durable neutralising antibody response up to seven months have now been described [164]. Ongoing vaccine research has highlighted a need for evidence of longer-term protection due to nAbs, and the titres at which these effects are achieved—neither of which were fully addressed by studies included in this review. A number of potent, specific nAbs have been identified—in line with findings for other HCoVs. This is particularly the case for neutralising anti-RBD antibodies [165], and is consistent with data emerging from vaccine development studies showing that protective antibodies can be induced [166–168].

Data on correlates of the antibody response in this review were incomplete, inconsistent or contradictory. It is not possible to draw robust conclusions on the associations of antibody response with age, sex, ethnicity or comorbidities, and although disease severity positively correlated with higher IgG antibody titres in a number of studies, distinguishing causation from correlation is not possible. More recent evidence shows that while responses may be of lower magnitude in milder cases, they are still elicited and may be protective—although evidence on response patterns in milder illness remains in short supply [164, 169].

The size of the detectable nAb response appears to be associated with male sex (although the effect of disease severity was not controlled for); this is a surprising finding given the now well-recognised association between male sex and poor COVID-19 outcomes [170]. A recent study has characterised sex differences in the cytokine and cellular response, although identified no differences in anti-S1 IgG or -IgM between male and female cases [171]. With regard to age, recent data published beyond the cut off for this review has shown distinct antibody responses in children and adults, which may inform future understanding of the disease course in different age groups [172]. This is a welcome addition to the literature considering the lack of studies on children included identified for our review.

Available data on protection following primary infection for this review were limited to small scale animal models which consider re-exposure rather than reinfection. Primary infection appears to provide a degree of protection to reinfection up to day 43 post primary inoculation but no further data were available at later time points. The success of passive transfer of nAbs for protection against SARS-CoV-2 infection appears to be dose dependent, although no data exist around the importance of affinity, isotype or immunoglobulin subclass. Given the possible reduction in nAb titre over time the protection they provide could be limited. However, more recent evidence from a large cohort of UK health professionals has shown anti-S IgG generated following natural infection may protect from re-infection up to six months [173], consistent with emerging data on the durability of neutralising antibodies. Further evidence on longer durability and correlates is required.

Some emerging evidence has suggested pre-existing humoral immunity arising from infection with seasonal HCoVs, although these data remains difficult to interpret [159]. Overall, our review found limited evidence of cross reactivity between SARS-CoV-2 and other HCoVs, but cross-neutralisation is rare and when it does occur, fails to fully neutralise the SARS-CoV-2 virus.

Strengths and limitations

This study presents, to our knowledge, the first comprehensive overview and critical appraisal of studies investigating the antibody response to SARS-CoV-2 infection over the first 6 months of the pandemic. The results presented here have a range of implications for treatment and policy, as well as providing a useful basis for building further research. Our findings are nevertheless limited both by aspects of the review methodology and by shortcomings in the included literature. The comprehensiveness of systematic reviews is always dependent on search strategy, and some results relevant to the research question may have been missed; as with all systematic reviews, our findings cannot account for unpublished negative results; and importantly, the pace of evidence production on SARS-CoV-2 means that systematic review research is inherently at risk of missing new, divergent data.

Limitations of the underlying evidence base were considerable. A majority of included studies were of variable but generally moderate quality. Study populations were highly variable, as were the assays used, along with the rigour with which they were described, verified and validated against their target populations. There are efforts in the UK to standardise laboratory SARS-CoV-2 assays use through the National External Quality Assessment Service (NEQAS), but these are early stage and no comparable international initiatives yet exist to support comparability of research findings. Longitudinal follow-up for durations greater than 50–60 days was rare, although we note this was limited by the timeline of the review for this novel pathogen. Many studies did not perform statistical analysis of findings; in particular, studies of putative correlates of immune response usually failed to control for the effects of potential confounders. Small sample sizes were common, as were study populations selected by convenience which, although common for clinical cohort studies, are prone to bias. Additionally, a large body of the evidence drew from pre-print publications which have not been subject to peer-review. While efforts were made to account for this during synthesis and reporting, reporting standards in these publications were highly variable and there is no validated system at this time for weighting evidence from pre-print publications relative to peer-reviewed papers.

Finally, a substantial proportion of included studies failed to appropriately make statements on ethical approval for studies or the use of consent for participation. There were two underlying factors behind this. First, papers uploaded to pre-print servers routinely failed to include an ethical statement: although this may be updated by the time of the peer reviewed publication, the wide use of pre-print material in the current phase of the pandemic suggests ethical statements should be routinely made in these uploaded manuscripts. Second, a number of studies where ethical statements were made reported that informed consent for participation was waived due to the pandemic situation: such measures limit the confidence with which such research can be used, particularly in an academic field where the use of informed consent should be routine, and where the ethical conduct of studies with human subjects should be the norm.

Implications for policy

We identify two main policy implications arising from this work. At individual level, continuing uncertainty concerning the nature of the humoral response to SARS-CoV-2 makes it difficult to determine what the practical meaning of serologically-detected antibody response is with respect to sterilising immunity. Short follow-up periods, as well as the use of binary (positive/negative) serological tests in many studies continue to limit what can be said about the granularity of the immune response over time—and by implication, how best to interpret the results of serological testing with respect to individual susceptibility to infection [174]. We did not identify any studies considering risk of re-infection with SARS-CoV-2, which might provide an alternative perspective on susceptibility to infection. Further studies published following the completion of this review shed additional light in this area, although this question remains under active investigation.

At a population-level, important policy implications arising from these data on antibody response relate to both surveillance and control. Serological tests have been evaluated predominantly in acutely unwell, hospitalised patients (without appropriate validation against mild disease or in people with asymptomatic infection) and seroprevalence estimates from this work should therefore be treated with caution. A recent Cochrane review emphasises the risk of false-positive and false-negative results under different population prevalence scenarios [175]. However, in the UK, nationally validated assays have been evaluated with convalescent samples from community participants and a number of large-scale sero-surveys now use these [176–178]. Clear understanding of the kinetics of the response, particularly for the specific N and S antigens, is important for the interpretation of seroprevalence studies. Serological tests remain variable in performance, and a major constraint in interpreting findings across different studies.

With regard to control, the evidence here for lasting protective immunity, or lack thereof, post-infection, may suggest it is too early to recommend the use of ‘immunity passports’. A range of promising data have been identified to support further investigation of treatment with convalescent plasma or immunoglobulin. The basic science underlying the antibody/virus/host cell interaction is starting to be described, with promising findings related to vaccine development: most recent vaccine data shows they are able to generate robust humoral responses [166]. For vaccines strategies for implementation will also require a thorough understanding of the likely impact in different population groups. Initial findings presented here give useful context to this, although further research is needed.

Onward research questions

The limited amount of data on antibody dynamics for mild and asymptomatic cases, which are likely to make up a significant proportion of infections, is a particularly important gap in the literature that will need to be addressed to improve understanding and definition of the varied clinical phenotypes associated with SARS-CoV-2 infection, although progress is starting to be made in this area. Investigating the relationship between antibody response and correlates including age, sex, ethnicity and disease severity through high-quality, large-sample studies using well validated assays and incorporating appropriate statistical testing of results should be prioritised.

Mutations affecting infectivity and potential resistance to vaccines are an emerging threat. There has been considerable recent attention on the likely impact of mutations to the S protein arising from infection in non-human hosts transmitting back to humans. A range of mutations, and their resistance to neutralising monoclonal antibodies, have been characterised [179]; however, further work in the area will be essential for understanding the roll out of vaccines in populations around the world.

Evidence on immunity beyond six months following primary infection or vaccination is urgently needed. Evidence of immunity following vaccination is being explored through various vaccine trials (e.g. ChAdOx1 nCoV-19) [166]. However, further longitudinal studies of those already infected with SARS-CoV-2 is required to examine the degree of protection arising from prior infection.