Using correlates to accelerate vaccinology

There have been many reports of large-scale vaccine studies showing that various severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccines give almost complete protection against severe COVID-19 and incomplete but dwindling protection against infection of the nose and lungs. Given the costs and difficulty of field studies involving thousands of people, which are necessary to show vaccine efficacy, the hunt for immune correlates of protection (COPs; laboratory measurements that predict the outcomes of large-scale studies) has become intense. On pa ge 43 of this issue, Gilbert et al. (1) report the use of a technique called mediation analysis to examine data from a trial of the mRNA-1273 vaccine from Moderna to infer that virus neutralizing antibody (VN-Ab) accounts for ∼60% of protection. They propose that VN-Abs might provide a reliable COP, which could be used to support the approval of future COVID-19 vaccines, bypassing the need for large trials.

Vaccines create resistance to infection by creating a set of immunological barriers that are deployed at different phases of pathogen entry and replication. These immune responses control engagement of the virus with target host cells in the nose and lungs, suppress the propagation of infection within the mucosa, modulate the inflammatory response to viral invasion, and inhibit the dissemination of the virus with consequent extrapulmonary involvement in those with severe disease (see the figure). Understanding the immunology that underlies protection is of great value to enable monitoring of vaccine duration and protective efficacy, as well as in aiding the development of new or improved vaccines that induce key elements of the protective immune response. Any single measure of immunity, however convenient, is never going to reflect this multilayered response. Therefore, can one measure sufficiently reflect the full palette of immune responses and thus reliably predict protection?

The discovery of reliable COPs can be game-changing in the production and licensure of vaccines. For example, influenza vaccines are updated twice a year according to the prevalent and projected circulating strains of influenza virus, based on the ability of an updated vaccine to induce an antibody response in a small group of volunteers, bypassing the need for large, slow, and expensive field trials. The use of such a correlate depends on a clear definition of the type of protection that is required and a solid relationship between the measure and the desired outcome.

In assessing the effects of vaccines designed to prevent COVID-19, VN-Abs, as described by Gilbert et al., are a logical and measurable target. VN-Abs coat virus particles, thereby preventing virus entry into cells. However, protection induced by two doses of messenger RNA (mRNA) vaccine may not last, declining to nonprotective levels in some vaccinees after 5 or 6 months (2). By focusing on an early serological COP, the opportunity may be missed to develop vaccines that mediate broad and durable protection. T cell responses may be important in supporting VN-Ab–mediated protection and helping to maintain long-term protection, but each component of immune memory exhibits distinct kinetics, which makes measures of such protection challenging (3). Cell-mediated immunity has the added benefit of conferring protection against a wide variety of viral strains and may provide additional protection against emerging variants (4). The protective role of mucosal T cells, including resident memory T cells (5), emphasizes the importance of studying immune responses in relevant tissue sites, not just in the blood (6).

There have been other attempts to define COPs against SARS-CoV-2 infection and disease across a range of vaccines (7, 8). These studies generally support the premise that circulating VN-Ab is a good indicator of protection, not only against severe disease but also (if amounts are very high) against viral replication in the mucosa (with or without symptoms), thereby reducing transmission. This does not mean that it is circulating antibody that is causing protection: Antibody in the serum does not normally diffuse (or get transported) into mucosal fluids. Protection against superficial infection depends on mucosal antibodies [including immunoglobulin A (IgA)] and possibly on antiviral T cells that reside within the linings of the respiratory tract. T cells floating free in the blood cannot have antiviral action; they only act through direct contact with infected cells. It could be that serum VN-Ab initially correlates with protection but fades and is replaced by other forms of immunity (such as T cells) that become important later. So, is the COP stable over time?

With the potential for a disconnection to develop between a COP and the desired effect of future vaccines, such correlates require constant scrutiny and reevaluation. For example, the live attenuated influenza virus vaccine that has been widely adopted for use in children does not induce acceptable concentrations of serum N-Abs. Instead, protective immune responses that have been demonstrated to result from the use of this vaccine require techniques of site-specific mucosal monitoring (9). COPs should therefore be validated with vaccines that work in a variety of ways, not just one type of vaccine. COPs should also be robust in diverse settings, including after natural infection and at delayed time points.

There are clear advantages and pitfalls to the use of correlates and surrogates in diverse medical fields (10). For example, measuring viral load to determine the effect of antiviral drugs in those living with hepatitis C or HIV is a practical shortcut in tailoring treatment to prevent disease and early death. The relationship between viral suppression and clinical benefit is so clear that there is no need to doubt viral load as a surrogate for long-term benefit. However, the relationship between a marker and the desired outcome can be insecure or even misleading. For example, the use of forced airflow measurements in evaluating the reversibility of airflow obstruction induced by β2 agonists in patients with asthma might distract clinicians and patients from also treating airway inflammation that is fundamental to pathogenesis. Indeed, long-acting β2 agonists may enhance inflammation (11) and so must be paired with anti-inflammatory treatments, such as inhaled steroids. Another example is in the treatment of osteoporosis: The measurement of bone density is widely accepted as a predictive marker of future bone fractures, but measures that increase bone density may not necessarily reduce fractures and might even increase them (12).

An accurate and stable COP can save vaccine developers from performing large and expensive trials to demonstrate the efficacy of new or updated vaccines, can predict the effects of established vaccines against new variants of the pathogen, and may expedite regulatory approval of updated or improved vaccines. For example, the emergence of the Omicron variant of SARS-CoV-2 (13) required urgent investigation into the likely protective efficacy of existing vaccines and possibly the redesign of vaccines to match Omicron’s mutated spike protein. A reliable COP could allow this to be done without large field trials, accelerating vaccine rollout and saving many lives. However, uncritical adoption of a COP may have the perverse effect of focusing future vaccine development on meeting the correlate rather than preventing infection and reducing disease and death. All is well if the correlate is reliable in the face of new viral variants and vaccines, but direct clinical outcomes need to be kept firmly in focus.