FROM THEORY TO PRACTICE: How Correlates of Protection Actually License Vaccines

A journey through history’s successes (and stumbles)

Oct 12, 2025 12 min read Serological Modelling

So let’s get practical. How have correlates of protection actually been used to license vaccines? What can we learn from both the successes and failures? And what does this mean for the next pandemic or the next vaccine waiting in the pipeline?

The Early Days: Stumbling Toward Success

Remember Edward Jenner from last time? While he had no clue about correlates of protection (the term wouldn’t exist for another two centuries), the seeds were planted early. By the mid-20th century, researchers began noticing patterns: people with certain antibody levels didn’t get sick, even when exposed to disease.

The breakthrough came when scientists realised they could standardise this observation. Instead of just noticing correlations, they could establish thresholds—specific antibody levels that predicted protection with reasonable confidence.

Smallpox

Smallpox vaccination provides our earliest example, though we have to squint a bit to see it. Researchers in the 1960s and 70s, during the final push for eradication, observed that vaccinia antibody levels correlated with protection against smallpox. This wasn’t used for licensing (the vaccine was already widely deployed), but it helped answer a crucial question: had someone been successfully vaccinated?

The vaccinia-neutralising antibody response became an informal correlate, used to assess take rates and guide revaccination strategies. It wasn’t rigorous by modern standards, but it worked. The disease was eradicated in 1980, and those early observations laid groundwork for thinking systematically about immune markers and protection.

Hepatitis B

The hepatitis B vaccine story is where correlates of protection really came of age. In the 1980s, researchers established that anti-HBs (antibodies to hepatitis B surface antigen) levels above 10 mIU/mL provided protection against infection. This threshold was validated across multiple studies and populations, and it stuck.

Here’s why this mattered: once that threshold was established with the original plasma-derived vaccine, subsequent recombinant vaccines could be licensed based on demonstrating similar antibody responses, what we call “immunobridging.” No need to wait years watching thousands of people to see who got infected. Just show that your new vaccine generates antibodies above 10 mIU/mL, and you’re good to go.

This was a Qin/Gilbert Level 2 surrogate before that framework even existed: validated across trials, populations, and vaccine platforms. It’s still used today, nearly 40 years later. That’s the power of a well-established correlate.

Key papers : Hadler et al. (1986) and Coursaget et al. (1986) provided early validation of the 10 mIU/mL threshold; subsequent bridging studies by Greenberg et al. (1996) demonstrated how this could accelerate new vaccine approvals. (All links to references are at the end of the post!)

Measles

Measles provides another success story. Hemagglutination inhibition (HI) titers above 120-200 mIU/mL correlate strongly with protection. This correlate has been used to bridge new measles-containing vaccines, evaluate vaccine effectiveness in outbreak settings, and guide public health policy on booster doses.

Measles demonstrates both the power and limitations of correlates. While HI titers predict individual protection reasonably well, the relationship between population-level antibody responses and herd immunity is more complex. You can have breakthrough infections even in highly vaccinated populations if immunity wanes. The correlate tells us about individual protection but doesn’t capture all the epidemiological dynamics.

Key references : LeBaron et al. (2020) systematically reviewed the evidence for the 120 mIU/mL threshold; Chen et al. (1990) provided foundational data on antibody levels and protection during outbreaks; FDA and ACIP documents from 2022 detail the immunobridging approach used for PRIORIX approval.

The Modern Era: Correlates Go Mainstream

Meningococcal Vaccines: When Antibodies Do the Heavy Lifting

Meningococcal vaccines provide perhaps the clearest example of mechanistic correlates in action. Serum bactericidal antibodies (SBA)—antibodies that can directly kill bacteria in the presence of complement—are not just correlated with protection; they are the protective mechanism.

This has been validated through multiple lines of evidence: people with complement deficiencies (who can’t perform SBA activity even with antibodies) have dramatically increased meningococcal disease risk; passive transfer of antibodies provides protection; and SBA titers predict protection across ages and vaccine types.

Because the mechanism is so clear, regulators accept SBA titers as a primary endpoint for licensing. This has allowed rapid development of vaccines against new meningococcal strains without massive efficacy trials each time. When MenB vaccines were developed in the 2000s, they were licensed based largely on SBA responses, saving years of development time.

Key references : Goldschneider et al. (1969) established the foundational relationship between SBA and protection; Borrow et al. (2001) provided comprehensive reviews validating SBA as a surrogate endpoint.

Pneumococcal Vaccines: The Complicated Case

Pneumococcal conjugate vaccines present a more complex story. Antibody levels (measured as IgG concentration) correlate with protection, and a threshold of 0.35 μg/mL has been widely used for licensure decisions. But here’s the catch: this threshold was never validated using Prentice criteria or even formal Qin/Gilbert methods.

Instead, it emerged from observational studies and expert consensus. It’s been tremendously useful—allowing new pneumococcal vaccines with additional serotypes to be licensed through immunobridging rather than massive efficacy trials. But strictly speaking, we don’t know if it’s a Level 2 surrogate or just a convenient Level 1 surrogate that happens to work well enough.

This pragmatism isn’t a bad thing. Sometimes “good enough” is actually optimal when the alternative is delaying life-saving vaccines by years. But it reminds us that regulatory decisions often balance statistical rigour with practical public health needs.

Key references : Siber et al. (2007) and the WHO pneumococcal vaccine position papers discuss both the utility and limitations of the 0.35 μg/mL threshold.

HPV Vaccines: Modern Immunobridging

Human papillomavirus (HPV) vaccines showcase how correlates can accelerate development in elegant ways. The original HPV vaccine trials took years because the endpoint was cervical precancerous lesions, which develop slowly. But once those trials established that neutralising antibodies predicted protection, subsequent vaccines could be licensed much faster.

When 9-valent HPV vaccines were developed (covering more cancer-causing strains), they didn’t need decade-long cancer prevention trials. Researchers showed that antibodies against the new strains were non-inferior to antibodies against the original strains—a bridging strategy based on the validated correlate. This probably accelerated deployment by 5-10 years, preventing thousands of cancer cases.

The HPV story also illustrates an important concept: correlates need not be perfect to be useful. We still don’t know the exact antibody level that guarantees protection, and cell-mediated immunity probably plays a role too. But knowing that antibodies are strongly predictive was sufficient for regulatory purposes.

Key references : Einstein et al. (2014) and Joura et al. (2015) describe the immunobridging strategy for 9-valent HPV vaccines.

Chikungunya: When Correlates Work (Until They Don’t)

The chikungunya vaccine story showcases both the promise and peril of approving vaccines based on correlates alone. In November 2023, the FDA approved IXCHIQ (VLA1553), the first chikungunya vaccine, using an accelerated pathway based entirely on immunological correlates—no traditional efficacy trial required.

Here’s why this was innovative: chikungunya outbreaks are unpredictable and transient, making traditional Phase 3 efficacy trials practically impossible. So researchers established a surrogate endpoint based on animal models. In non-human primates, neutralising antibody titers (measured by micro-plaque reduction neutralization test, μPRNT50) above 150 prevented viremia after viral challenge. A sero-epidemiological study in the Philippines provided supporting evidence: people with neutralising titers above a certain threshold showed 100% protection from symptomatic infection.

The FDA accepted μPRNT50 ≥150 as the surrogate endpoint for protection. In the pivotal trial, 98.9% of vaccinated participants achieved seroprotective titers. This looked like a triumph—a vaccine approved in record time based on validated immune correlates, potentially preventing millions of cases of debilitating arthritis.

But then reality struck. In May 2025, less than 18 months after approval, the FDA and CDC recommended pausing IXCHIQ use in people over 60 following reports of serious adverse events including cardiac and neurologic complications. By August 2025, the FDA suspended the vaccine’s license entirely after reports of severe chikungunya-like illness in vaccinated people—including 21 hospitalisations, 3 deaths, and one death from encephalitis directly attributed to the vaccine strain.

What went wrong? The correlate predicted protection , but it didn’t predict safety . The live-attenuated vaccine itself was causing disease in some recipients, particularly older adults. The immune response data were solid, but they told only half the story. A second chikungunya vaccine (VIMKUNYA, a virus-like particle vaccine) recieved FDA approval in February 2025 and remains available, also licensed based on similar correlate data but without the safety issues associated with live virus.

The chikungunya saga illustrates a crucial lesson: correlates can accurately predict efficacy while missing critical safety signals. Accelerated approval pathways based on surrogate endpoints are tremendously valuable for rapid vaccine development, but they require robust post-marketing surveillance to catch problems that smaller trials might miss. Sometimes the stumbles happen after approval, not before.

Key references : Schnyder et al. (2024) and Reisinger et al. (2024) describe the innovative development pathway and correlate validation for VLA1553; FDA safety communications from May and August 2025 detail the post-marketing adverse events.

Learning from Failure: When Correlates Mislead

Not every correlate story ends happily. Sometimes what looks like a clear immune marker turns out to be anything but.

HIV Vaccines: The Antibody Trap

HIV vaccine development has been humbling. For years, researchers chased neutralising antibodies as the obvious correlate—after all, they work for most viral vaccines. But HIV is different. The virus mutates rapidly, hides from antibodies, and attacks the very immune cells meant to fight it.

Some HIV vaccine candidates actually increased infection risk in people who developed certain antibody responses—a correlate of risk in the wrong direction. The STEP trial and Phambili trial in 2007 provided devastating examples: certain immune responses predicted higher infection rates, not lower.

This taught researchers crucial lessons: correlates must be validated, not assumed; the wrong immune response can be worse than no response; and what works for one pathogen may fail catastrophically for another.

Key references : Gray et al. (2010) reviewed the STEP/Phambili trial failures; Haynes et al. (2012) analysed immune correlates in the modestly successful RV144 trial.

TB Vaccines: Still Searching

Tuberculosis vaccines remain frustratingly elusive, partly because we lack validated correlates of protection. BCG (the existing TB vaccine) provides variable protection, and we don’t fully understand why. Is it antibodies? T cells? Trained immunity? Location of immune responses in lung tissue?

Without clear correlates, new TB vaccine candidates must undergo lengthy efficacy trials, watching thousands of people for years to see who develops disease. This delays development enormously. Researchers are actively searching for correlates using systems biology and machine learning, but we’re not there yet.

The TB story reminds us that correlates are luxuries, not guarantees. For some diseases, the immunity is simply too complex or poorly understood to reduce to a single measurable marker.

The Future: Where Do We Go From Here?

Beyond Antibodies: Multi-parameter Correlates

The future of correlates isn’t about finding the marker but about integrating multiple markers. Systems serology, transcriptomics, and machine learning are revealing that protection often involves coordinated responses: antibodies plus specific T cell types plus innate immune training plus mucosal immunity.

We’re moving toward multi-parameter correlates that combine, say, neutralising antibody levels, antibody Fc effector functions, memory B cell frequencies, and CD4+ T cell responses. These complex correlates are harder to standardise and validate, but they’re also more accurate and mechanistically informative.

Personalised Correlates

One-size-fits-all correlates may give way to population-specific or even personalised correlates. A 25-year-old and an 85-year-old may need different immune profiles for equivalent protection. Immunocompromised patients certainly do. Future correlates might be conditional: “For population X with characteristics Y, immune response Z predicts protection.”

Real-time Surveillance

Imagine a world where we continuously monitor population immunity through serological surveillance, using validated correlates to predict when boosters are needed or when herd immunity is waning. We’re already seeing glimpses of this with serosurveys. As correlates improve and surveillance infrastructure expands, public health responses could become increasingly predictive rather than reactive.

Accelerating the Next Pandemic Response

Perhaps the most important lesson: establishing correlates during inter-pandemic periods is crucial preparation. If we’d had validated influenza correlates before 2009, the H1N1 pandemic response would have been faster. If we’d understood coronavirus immunity better before 2020, vaccine development would have been even more rapid.

Investment in correlate research during “peacetime” pays enormous dividends during emergencies. Every pathogen with pandemic potential—influenza, coronaviruses, filoviruses, paramyxoviruses—deserves correlate studies now, before we desperately need them.

The Bottom Line: Imperfect but Invaluable

Correlates of protection aren’t perfect. They’re often incompletely validated, mechanistically fuzzy, and context-dependent. The terminology remains a mess (sorry, did my best to clarify it last time). Regulators must balance rigorous validation against public health urgency.

But despite these limitations, correlates have transformed vaccine development. They’ve cut years from licensure timelines, enabled rapid responses to emerging variants, and allowed rare-disease vaccines that would otherwise be impossible. From hepatitis B in the 1980s to chikungunya in the 2020s, correlates have proven their worth.

As we face future infectious disease threats, our ability to rapidly identify, validate, and apply correlates of protection may be one of our most powerful tools. The history we’ve reviewed here—both successes and failures—provides a roadmap. We know what works, what doesn’t, and what questions still need answering.

Edward Jenner would be amazed. Not only do we understand why vaccines work, but we can predict their efficacy with increasing precision, often before testing them in large populations. That’s the power of correlates of protection, fumbling terminology and all.