More than a month after the World Health Organisation declared COVID19 a global pandemic on 11th March we are still missing one crucial piece of data from our modelling: the fraction of people who have established productive immunity against the SARS-CoV2 virus. While lockdowns may slow the rate of viral transmission in the short term, our longer time battle with the virus will depend on how quickly the population develops sufficient immunity (whether through natural exposure or a vaccine, once one becomes available).

If enough people have become immune, and critically can neither suffer symptoms nor act as a carrier for the virus, then immunity will replace social distancing as the key breaker of transmission chains. This is the concept of “herd immunity” – whether it is the “goal” of government policies or not, it is the only endgame in town given that we cannot remain in lockdown forever.

Understanding the current status of the human herd, however, is harder than it looks. Current tests for SARS-CoV2 are exclusively measures of the viral nucleic acid, exploiting the exquisite sensitivity and specificity of PCR-based amplification strategies. This gives a clear indication of current (or very recent) infection, but it tells us nothing about the cumulative number of people exposed, nor the degree or nature of their immune response to the virus. Added to that, most countries have adopted a testing strategy focused on people with symptoms, giving us a relatively poor understanding of the degree of spread among people without symptoms.

If there has been extensive spread of SARS-CoV2 in people who experienced no symptoms, then the fraction of the population with protective immunity may be much higher than predicted from nucleic acid testing among those with symptoms. Without a better estimate of who has been exposed at any point since the virus jumped into the human population we have only a poor estimate of where we are on the journey to herd immunity.

The solution is a test for the immune response to the virus (often called a “serology test”). At a stroke, such a test will tell us how many people have ever been exposed to the virus as well as quantifying for us the strength and nature of their immune response to it. This serology test then needs to be applied to a truly random sample of people in each population.

If data from a serology test applied to a random sample of people is so crucial to understanding where we are, why haven’t we got one in widespread use already?

Because there is more to designing and validating a serology test than meets the eye. The team at RxCelerate have been creating such assays for more than twenty years, and that experience quickly demonstrates the challenge that lies ahead.

The first issue in designing a serology test lies in the selection of the antigen. The optimum antigen is probably the whole inactivated virus, since it contains all the protein and carbohydrate antigenic determinants and therefore ensures that reactivity against any of them are detected by the assay. But manufacture of inactivated virus at scale is challenging, as is ensuring batch to batch consistency, which is why most serology tests use recombinant protein components (just one or two of the most antigenic protein components from the virus coat). Recombinant antigens are much easier to produce in bulk with the necessary consistency, but risk missing immune responses to other antigenic determinants on the virus.

For most viruses, this trade-off is acceptable since we know which determinants drive the strongest immune reactions in most people and as a result very few people have a strong reaction to other antigens not present in the recombinant mix without also having a reaction to the chosen antigens. False negatives are therefore rare – and with SARS-CoV2 very occasional false negatives (that is, people who the serology test says have never encountered the virus but actually have) are not a major clinical concern, since they will come to no harm continuing to protect themselves from infection.

Selecting an appropriate antigen mix, however, is probably the easier part. The next problem relates to the complexity of the human immune response: antibodies come in multiple isotypes, with an initial response of IgM followed by higher affinity IgG a week or two later. Even this over-simplifies what happens, because IgM-positive plasma cells also express IgD, but the levels of IgD vary almost 1,000-fold between individuals for reasons that we don’t properly understand.

This complexity is usually ignored – by detecting the bound immunoglobulin using specific reagents that only detect IgM (and hence detect a recent immune response) or only IgG (and so detect a fully mature immune response). But that only makes the problem invisible. IgD antibodies in the serum sample still bind to the recombinant antigen – and compete with the IgM and IgG also in the sample. The result is some very complex dilution curves (at best), or else a significant risk of false negatives.

The problem is actually even worse, because IgA is often also present, as are IgG1, G2, G3 and G4 subtypes of IgG – so depending on the particular detection antibody used, you either see inhibitory competition or, with a broad anti-human immunoglobulin detection reagent, a complex mix of bound antibody species.

That doesn’t sound important until you try and dilute the serum sample in order to estimate the strength of the antibody response the individual had mounted (a process immunologists called “titre-ing”). If you can dilute the sample a lot and still detect bound antibody, we say the individual has a high titre (or strong) response. If the signal disappears as the sample is diluted, that indicates progressively lower titre and hence weaker response. If we are trying to predict who is protected against re-infection with the virus, this titre information can be crucial.

Yet the competition between different isotypes can make titre estimates difficult or impossible. Sometimes the signal in the assay even INCREASES as the sample is diluted (because the low affinity competing IgD doesn’t bind at higher dilutions, revealing a higher affinity IgG response that was previously hidden).

Believe it or not, it doesn’t even end there: because IgM antibodies have five “arms”, compared to the two “arms” of the classical Y-shaped IgG, these penta-valent IgM molecules bind stronger than you would predict from affinity alone (this is called “avidity” – imagine holding on to a bar while hanging over a chasm; if you had five arms rather than two you would be less likely to ever let ago with all of them at once!).

The bottom-line is that it is no simple task to devise an assay that gives clean, readily interpretable titres for the immune response to the chosen antigens. Rushing to deliver an assay without properly understanding how it performs for many different individuals at different stages of their immune response to the virus, and in particular how well titres are estimated in these different samples, would be a dangerous error.

But the challenges in designing the test, though not to be under-estimated, pale into insignificance compared to the challenge of interpreting the results.

Of course, it depends what question you want to answer with the serology test. At the simplest level, the question “how many people have been infected by the virus at any point in time?” is relatively straightforward. Using a test that detects IgM and IgG with relatively undiluted serum will provide a pretty good estimate (likely missing only those infected within the last 7-10 days).

But the real question we would like to address is “who has been infected by the virus and developed a productive immune response sufficient to ensure they will neither suffer symptomatic re-infection nor act as an asymptomatic carrier?”. The answer to that question has been proposed as the basis of “immunity passports” or certificates allowing such people to be released from lockdown and return to work if not to normal life.

Running a serology test, however, doesn’t really answer that question. Do antibodies against different antigens all result in protective immunity? What titre for the immune response would be considered sufficient to confer such protection? Whatever cut-off in whichever assay is chosen, it seems certain some people will have a “positive” test but lack a neutralising (and therefore protective) immune response. But how many people? And which ones? Answering those questions will require clinical studies following a large cohort of people after testing for some considerable time.

All these questions can and will be answered for COVID19, just as they have to a very large extent for other infections. But once you understand what is involved, it becomes much clearer why, just a couple of months after the pandemic broke, we still do not have a validated, well understood serology test ready for mass deployment and clinical use to end the lockdown, at least for the fortunate few who do have protective immunity at this time.

There is no “serology test” – there are millions of possible serology tests, which need to be carefully compared and characterised to select ones that do more than just show you if some antibody against the virus is present. Reliable titres are going to be essential in the next phase of understanding this new viral infection – identifying who has protective immunity. They are also going to be essential tools in the trials of vaccine candidates that are reaching the clinic right now – and which offer the biggest and best hope for a route back to something approaching normal life.

Rolling out a bad test will do more harm than good. So will widespread use of a test for which the cut-off for declaring a positive response has not be validated clinically. Releasing people from lockdown, believing they have protective immunity, not only risks causing them harm when they subsequently do become infected – it risks re-igniting the rapid and uncontrolled expansion of the virus in a second-wave that we might naively think we were protected against. False-positives are much more relevant than false negatives in this setting.

Next time someone says “why haven’t we got a serology test? How hard can it be?” you can point them to DrugBaron and remind them that actually it is a substantial challenge. And one that will take time to overcome, even for experienced scientists like the RxCelerate team.

RxCelerate Ltd is an outsourced drug development platform based near Cambridge, UK. We specialize in delivering an entire road map of drug development services from discovery and medicinal chemistry through to formal preclinical development and clinical up to Phase IIa. In the last five years, we have witnessed dramatic changes in the drug development …