Universal Vaccines

Established A universal vaccine is one that covers a whole viral family, or every drifting variant of a single virus, without being reformulated every year. Established None has been approved anywhere, for any pathogen. A 2026 review of the influenza programme states the position without hedging: despite encouraging preclinical findings and favourable early-phase results, no universal influenza vaccine has yet received regulatory approval. Every other claim in this brief is downstream of that.

Established The target is defined by the thing it replaces. Seasonal influenza vaccines are rebuilt annually because haemagglutinin's head domain drifts; a universal vaccine would instead aim at the conserved parts of the virus that drift cannot easily move — the HA stalk, the neuraminidase active site, internal proteins such as nucleoprotein and matrix protein 1, or the extracellular domain of M2. Established The same logic extends outward: a pan-coronavirus vaccine would target sites shared across the sarbecovirus clade, and an HIV vaccine would have to reach conserved surface sites that the virus hides behind glycan and hypervariable loops. Established The three cases are usually discussed together and they are not equally hard: influenza has a conserved target and an imprinting problem, sarbecoviruses have a conserved target and no human data, and HIV has a conserved target that the virus has spent its evolutionary history making unreachable.

Frontier The argument of this brief is that the binding constraint is not antigen design but measurement. Established The field can build immunogens that produce broad antibody and T-cell responses in humans; it has done so repeatedly in phase 1. Established What it cannot do is say which of those responses predicts protection, because the only regulatorily accepted immune correlate for influenza vaccination is a haemagglutination-inhibition titre of 1:40, and that assay measures antibodies against the HA head — the exact domain a universal vaccine is designed not to target. Frontier A field whose yardstick is orthogonal to its goal will keep producing encouraging phase 1 studies and no product, which is precisely what it has done.

Established Start with the completed trial that failed, because the field's structure follows from it. Established BiondVax's Multimeric-001 (M-001) is a recombinant protein carrying nine conserved epitopes drawn from influenza haemagglutinin, nucleoprotein and matrix protein. Established It ran the only pivotal efficacy trial a universal influenza candidate has ever completed: 12,400 participants, registration NCT03450915, completed August 2020, topline results published in October 2020. Established The reference article that indexes this literature carries the sentence verbatim — the pivotal phase III study “did not show any statistical difference from the placebo group.” Established M-001 had already passed seven phase 2 trials showing safety and immunogenicity, two of them in the published record (Atsmon and colleagues in Vaccine, 2014; van Doorn and colleagues in Medicine, 2017). Frontier The failure was therefore not predicted by any immunological readout the field had. That is not one company's bad luck; it is a statement about what the field's instruments can and cannot see.

Established The consequence is that there is no validated correlate of protection for any universal approach. Established The HAI 1:40 convention is a regulatory artefact of the seasonal era: it was adopted because head-directed antibody happens to track protection against a matched strain, and it is the only correlate a regulator currently accepts. Established A stalk-directed, neuraminidase-directed or T-cell-directed vaccine can be highly effective and score nothing on it. Frontier The 2026 review that states this also states the way out — future evaluation will likely require a standardised panel of immune endpoints integrating multiple complementary measures of protection — which is an admission that the panel does not yet exist. Speculative Candidate correlates are named in the literature (stalk-specific antibody-dependent cellular cytotoxicity, neuraminidase inhibition titre, mucosal IgA, lung-resident memory T-cell frequency) and none has been shown to predict efficacy across two independent trials, because there have not been two independent efficacy trials to test them against.

Established Now the scale, which is routinely misreported. Established The best-designed universal influenza trial to date is the chimeric-haemagglutinin phase 1 published in Nature Medicine in 2021: observer-blind, randomised, placebo-controlled, and 66 healthy US adults aged 18 to 39 across five arms — live-attenuated prime followed by AS03-adjuvanted inactivated boost (n=20), live-attenuated followed by unadjuvanted inactivated (n=15), two AS03-adjuvanted inactivated doses (n=16), and placebo (n=15). Established Its conclusion, in the authors' words, is that the regimen was safe and induced a broad, strong, durable and functional immune response targeting the conserved, immunosubdominant stalk of the haemagglutinin. That is a real result and it is an immunogenicity result in sixty-six people.

Established The rest of the register has the same shape. Established A 2026 review enumerates the registered trials candidate by candidate: BPL-1357, the whole-inactivated intranasal NIH candidate, phase 1 randomised and placebo-controlled (NCT05027932), 45 healthy adults aged 18 to 55, acceptable safety profile; the H1 stabilised-stem ferritin nanoparticle H1ssF, phase 1 dose-escalation (NCT03814720), safe, with cross-group-1 neutralising antibodies spanning H1, H5, H6 and H9; the mRNA-LNP version of H1ssF, phase 1 (NCT05755620), 50 healthy adults aged 18 to 49 at three dose levels of 10, 25 and 50 micrograms; G1mHA, phase 1/2a (NCT05901636), 170 healthy adults aged 18 to 45 at 45 or 135 micrograms; FluMos-v1, phase 1 open-label dose-escalation (NCT04896086) in healthy adults, safe and well tolerated; DeltaFLU, phase 1 dose-escalation, 48 volunteers across five dose levels, eight per level; FLU-v, a phase 2b randomised double-blind placebo-controlled study (NCT02962908) reporting significant increases in interferon-gamma- and granzyme-B-secreting cells by day 42; and the largest of them, OVX836's phase 2 (NCT05569239) at 2,850 participants. Read the shape of that list: every entry is phase 1 or phase 2, the largest is 2,850 people, and the only pivotal efficacy study the field has ever run is M-001's, which failed.

Established The immunology underneath is genuinely strong, and that is the tension. Established The HA stalk mutates far less than the head, and stem-directed immunisation was shown to induce broadly neutralising antibodies in mice by Wei and colleagues in Science in 2010. Established Human broadly neutralising antibodies exist: FI6, described in 2011, binds all sixteen influenza A HA subtypes through the stalk, and Stadlbauer and colleagues extended the conserved-target argument to the neuraminidase active site in 2019. Established Better still, the human immune system will make these antibodies on its own under the right conditions — Wrammert and colleagues showed in 2011 that broadly cross-reactive antibodies dominated the human B-cell response to 2009 pandemic H1N1 infection, because that virus presented a novel head on a familiar stalk. Frontier The existence proof is not in doubt. The elicitation problem is the entire field. Knowing that a target exists tells you nothing about whether a vaccine can reliably induce antibodies to it in an outbred, immunologically experienced population.

Established Two structural obstacles sit between the existence proof and a product. Established The first is immunodominance: the head is loud and the stalk is quiet, so a conventional immunogen spends the immune response on the part that drifts. Established The second is imprinting — original antigenic sin — whereby the response to any new influenza exposure is biased toward the strain first encountered in childhood. Frontier A universal vaccine given to adults is not writing on a blank slate; it competes with decades of head-directed memory. Frontier The chimeric-HA strategy is the most intellectually serious answer available: sequentially immunise with haemagglutinins carrying exotic heads the recipient has never seen, so that the stalk is the only constant and the only thing memory can boost. Established It has a completed phase 1 at n=66 and no efficacy data. Frontier A quieter gap sits beside it: H1ssF's demonstrated breadth is across H1, H5, H6 and H9, which are all group 1. Almost all current human data are group 1 only, and a vaccine that does not cover group 2 — H3, H7 — is not universal in any sense a public health agency would accept.

Established Coronaviruses are the newest and, on animal data, the strongest branch. Established Mosaic-8b displays receptor-binding domains from eight different sarbecoviruses — a SARS-CoV-2 Beta RBD plus seven others from clade 1a, 1b and 2 spikes — on a 60-mer mi3 protein nanoparticle. The design rationale is avidity: an antibody that can bridge two different RBDs on the same particle must be binding a site conserved between them, so the geometry selects for cross-reactivity without anyone having to know in advance which epitopes are conserved. Frontier The rabbit study reports five New Zealand White females immunised, monoclonals screened against a twelve-pseudovirus panel spanning SARS-CoV-2 D614G, five variants of concern, three non-SARS-CoV-2 sarbecoviruses and two chimeric spikes, with three monoclonals binding every sarbecovirus RBD tested. Frontier And the protection paper is stronger than the sceptical summary usually allows. Frontier Cohen and colleagues, in Science in 2022, report challenge protection in mice and macaques: stronger responses to mosaic-8 than to homotypic particles, equivalent neutralisation of SARS-CoV-2 variants including Omicron, protection against multiple heterologous sarbecovirus challenges where homotypic controls protected only against the matched strain, and epitope mapping confirming increased targeting of conserved sites. Established There is still no human efficacy data of any kind. Both halves belong in the same sentence; omitting the macaque result makes the sceptical case look uninformed, and omitting the absent human data makes the enthusiastic case dishonest.

Established HIV is where the antigenic-space argument bites hardest, and the record is a list of large, well-run failures. Established RV144 in Thailand enrolled 16,395 participants (8,197 vaccine, 8,198 placebo) on an ALVAC prime and AIDSVAX B/E boost and reported roughly 31 per cent efficacy at 3.5 years — the index source used here prints the figure as about 30 per cent, adjusted to 26 per cent, and the primary NEJM paper was not opened for this brief, so treat the number as a range rather than a constant. The correlates analysis found IgG against the V2 loop associated with 43 per cent lower infection risk and IgA associated with 54 per cent greater risk. Established STEP (HVTN 502), roughly 3,000 participants on a Merck Ad5 vector, was halted in September 2007 because vaccination appeared associated with an increased risk of infection in some recipients. HVTN 505 was stopped for futility in 2013; HVTN 702 Uhambo stopped in early 2020; HVTN 705 Imbokodo reported no significant reduction in August 2021; HVTN 706 Mosaico was terminated for ineffectiveness in January 2023. Established The AMP trials (HVTN 703/704), which infused the VRC01 monoclonal rather than vaccinating, found it did not prevent acquisition overall — and produced the field's most valuable output, a serum-neutralisation threshold framework that at least tells you what titre would have to be reached.

Frontier The mainline HIV response to that record is germline targeting, and it is honest about what it is. Established eOD-GT8 60mer is a self-assembling nanoparticle designed to bind and expand naive B cells carrying VRC01-class precursors, and its cGMP manufacture by transient transfection is published. Frontier The strategy explicitly concedes that no single shot can work: it is a sequential programme in which a prime engages the precursor and a designed series of boosts shepherds affinity maturation toward a known broadly neutralising signature. Frontier The IAVI G001 phase 1 is the trial behind the widely quoted claim that a large majority of vaccine recipients developed VRC01-class precursor responses; the citation is confirmed (Leggat and colleagues, Science, 1 December 2022, DOI 10.1126/science.add6502) but the paper itself could not be opened for this brief and no response rate is printed here. Frontier An mRNA version, mRNA-1644, has been in first-in-human trials with Moderna since January 2022. Frontier The unresolved limit is allelic. VRC01-class precursors depend on particular IGHV1-2 alleles that most, but not all, humans carry. Speculative A vaccine that requires the right germline is a pharmacogenomic product, and nobody has assembled the parallel programme — germline-targeting immunogens for several bnAb classes with disjoint allele requirements — that genuine coverage would need.

Frontier The most striking recent result in the whole subject is a T-cell vaccine that ignores antibodies entirely. Frontier Malouli and colleagues immunised Mauritian cynomolgus macaques with cytomegalovirus vectors expressing M1, NP and PB1 from the 1918 H1N1 virus, then challenged them with aerosolised A/Vietnam/1203/2004 H5N1. Frontier All six unvaccinated animals died within seven days with acute respiratory distress; 54.5 per cent — six of eleven — of vaccinated animals survived. The protection runs through lung-resident effector memory CD4+ T cells, and the antigenic gap crossed is eighty-six years and two subtypes. Established It is n=17 in total, in one species, with no human data.

Speculative That result is the strongest steelman for the minority position that universality should be sought in T cells rather than antibodies. Established Antibody-based universality requires a conserved surface epitope, which is exactly where selection pressure is strongest; T-cell epitopes sit on internal proteins that drift far less. The standard objection is that T cells rarely prevent infection — they limit severity. Speculative The reply is that this is a definitional dispute, not a scientific one: if “universal” means prevention of severe disease and death across all subtypes rather than sterilising immunity, the T-cell route already has the better cross-subtype animal data. The field has not made that choice explicitly, and it should.

Speculative The persistent-vector heresy follows immediately. Frontier CMV vectors work because they persist and continually restimulate effector memory, which is also why they alarm regulators: a self-perpetuating, genetically modified, lifelong-infecting vector given to healthy people is a different safety proposition from a protein or an mRNA. Speculative What would have to be true is a conditional kill switch or a rationally attenuated vector with demonstrated non-transmissibility. Speculative The technology for that exists in a literature nobody has joined to this one: recoded organisms engineered to depend on synthetic amino acids, with measured escape frequencies below 4.85 x 10^-12. Handwave A vaccine vector that cannot replicate outside a host supplied with a non-natural amino acid is a coherent object on paper and has never been built.

Frontier Mucosal and transmission-blocking approaches are the most neglected serious idea. Frontier The EU-funded FLUniversal consortium is developing an intranasal universal candidate, published in 2025. Speculative A mucosal vaccine that blocked onward transmission of any strain would deliver population-level universality without individual-level breadth, because it removes the substrate on which antigenic drift operates. The obstacle is regulatory rather than immunological: vaccines are licensed on individual protection, there is no validated correlate for mucosal transmission blocking, and no endpoint rewards it.

Frontier Neuraminidase is the cheap partial win the field ignores because it is not universal. Established Broadly protective human antibodies against the neuraminidase active site are documented, NA immunity tracks reduced severity, and NA content is not standardised in current seasonal vaccines. Standardising it would plausibly buy a measurable breadth improvement within the existing regulatory framework, and it gets little attention precisely because it does not answer the headline question.

Frontier The M2e line is the field's clearest example of a serious approach that stalled quietly rather than being refuted. Established The extracellular domain of matrix protein 2 is short, highly conserved across influenza A, and exposed on the virion surface, which makes it an obvious universal target on paper. Acambis announced an M2e candidate, ACAM-FLU-A, in 2008, and there was limited follow-through. Established Non-antibody routes have the same archaeology: ITS FP01, built from six peptide antigens drawn from PA, PB1, PB2, NP and M1, entered phase 1, as did Inovio's DNA candidates VGX-3400X and SynCon. Frontier None was published as a failure, because a programme that runs out of money does not generate a negative result — it generates silence. A field that cannot see its own discontinued arms will systematically overestimate the novelty of whatever it is currently doing, and the M2e story should be read beside every claim that a new conserved target has just been found.

Frontier Multivalent mRNA is adjacent but conceptually different: the twenty-subtype mRNA influenza approach and Moderna's mRNA-1010 seek breadth by including everything rather than by focusing the response on conserved sites. No source reachable for this brief covered either programme, so nothing quantitative about them is asserted here; the distinction, however, is real and worth holding onto, because “covers many strains because it contains many antigens” and “covers many strains because it targets what they share” fail in different ways. Frontier On the HIV side, a multiclade env-gag virus-like-particle mRNA vaccine has elicited tier-2 neutralising antibodies and reduced heterologous SHIV acquisition risk in macaques, which is the meaningful animal bar. Frontier And a frontier datum of a different kind: US funding cuts in 2025 stopped an HIV vaccine programme that was about to enter trials. The binding constraint on a scientific programme is sometimes a budget line.

Frontier The workback plan for a genuinely universal influenza vaccine has eight links, and the first one is not a technology.

Frontier (1) A validated correlate of protection for stalk-directed, neuraminidase-directed and mucosal immunity — an assay whose readout predicts efficacy across two independent field or challenge trials. Without it, links 2 through 8 are unmeasurable and every phase 1 remains uninformative about phase 3. Frontier (2) A controlled human challenge platform licensed for universal candidates, with heterologous challenge strains and a reproducible placebo attack rate. Frontier (3) Demonstrated redirection of immunodominance in imprinted adults — a measured shift in the ratio of stalk-directed to head-directed antibody in people with known exposure histories. Frontier (4) Durability, meaning maintenance of whatever correlate link 1 produces at twelve, twenty-four and thirty-six months. Frontier (5) Breadth across group 1 and group 2 influenza A, plus influenza B, demonstrated by protection against an H3 or H7 challenge after a group 1 immunogen. Speculative (6) An escape study: serial passage under stalk-directed antibody pressure, measuring the replicative fitness of escape mutants relative to wild type. Frontier (7) Efficacy in an unselected population including the elderly and the immunocompromised, in whom seasonal vaccines already underperform. Established (8) Manufacturing at seasonal-vaccine cost parity.

Frontier The binding link is (1), a scientific unknown, with (2) as the institutional lever that makes everything downstream affordable. Frontier Note the asymmetry that ought to embarrass the field: link (6), the escape study, is the one that determines whether the goal is even coherent, it is cheap by the standards of any other item on this list, and it is on nobody's critical path — because it could return an answer nobody wants. Frontier For HIV the equivalent chain is shorter and the binding link is further along: prime, guided boost, serum tier-2 breadth, protection, allele coverage, deliverable schedule. Established The binding link there is the third: nobody has induced broad tier-2 neutralisation in humans by vaccination, ever. Everything after it is contingent, and everything before it has already been claimed.

Frontier There is a fifth bottleneck that is not a link in either chain, and it is trial economics. Established An influenza field efficacy trial has to be large enough to accumulate endpoints in a season whose attack rate nobody controls, which sets a floor of ten thousand participants or more and a cost that only a government or a very large company can carry. That floor is what makes the correlate question load-bearing rather than academic: with a validated correlate, candidates can be ranked in phase 1 and only the best goes to a field trial; without one, the field trial is the first informative experiment, and almost nobody can run it. The whole subject is arranged around an instrument that does not exist.

Frontier This subject depends on four bodies of work it does not itself produce. Frontier The first is systems serology and assay standardisation — the boring, unglamorous business of making a stalk-specific antibody-dependent cellular cytotoxicity assay or a neuraminidase-inhibition titre mean the same thing in two laboratories on two continents. Without inter-laboratory comparability there is no correlate, only local numbers.

Frontier The second is structure-based immunogen design: stabilised stems, headless mini-stems, chimeric constructs and nanoparticle scaffolds are all products of structural biology, and the mosaic strategy is an explicitly geometric idea about how many distinct antigens fit on a 60-mer. Frontier The third is B-cell lineage sequencing and deep mutational scanning, which is what makes germline targeting checkable rather than hopeful — you can watch whether a boost pushed a lineage along the intended trajectory, and you can map which residues an elicited antibody actually depends on. Frontier The fourth is cohort infrastructure: to test imprinting redirection you need people whose influenza exposure history is known, which means serum banks and birth-cohort studies that were funded for other reasons decades ago. None of these four is a vaccine programme, and none is funded as though a vaccine programme depended on it.

Frontier Two further dependencies are physical rather than intellectual. Established A controlled human challenge programme needs GMP-manufactured challenge viruses of defined sequence and known attack rate, and the strains a universal candidate most needs to be tested against — heterologous, group 2, ideally drifted — are not the strains for which characterised challenge stocks exist. Making them is a regulatory and manufacturing project of its own, with no obvious owner. Frontier And the animal models carry more weight here than the field admits: ferrets for transmission, macaques for lethal heterologous challenge, and neither is a human. A discipline whose human endpoint is unmeasurable has quietly delegated its decisions to two animal models whose predictive validity has itself never been formally established for this class of vaccine.

Speculative The single most informative unfunded experiment is a fitness-cliff study. Speculative Serially passage influenza in the presence of stalk-directed monoclonal or polyclonal pressure, isolate escape mutants, and measure their replicative fitness against wild type in primary human airway cultures and in ferrets. If escape costs little, universal stem-directed immunity has a half-life and the field should be designing for replacement rather than permanence; if escape is severely costly, the stem is a genuine fitness cliff and the whole enterprise is on firmer ground than anyone can currently claim. Handwave Nobody has published a definitive version, and the reason is not cost.

Frontier Second: a group 2 challenge after a group 1 immunogen. Established The published human breadth data are group 1 (H1, H5, H6, H9). A ferret or macaque study immunising with an H1 stabilised stem and challenging with H3 or H7 would settle in months a question the clinical programme has been carrying for a decade.

Frontier Third: an imprinting-redirection readout in adults. Measure stalk-directed to head-directed antibody ratios before and after chimeric-HA immunisation, stratified by birth cohort, and report the ratio rather than the titre. Frontier Fourth: a controlled human influenza challenge trial of two or three universal candidates head to head at n of order 50 to 100 per arm, using the same heterologous challenge strain. That is two orders of magnitude cheaper than a field efficacy trial and it is the only design under which a correlate can be validated against an observed outcome in a controlled setting. Frontier Fifth: a first-in-human study of mosaic-8b. The macaque protection data exist; the human immunogenicity data do not, and the gap has now stood for three years.

Frontier Sixth: a lineage-tracking readout for the HIV boost series. Frontier Sequence B-cell receptor repertoires before and after each immunogen in a germline-targeting sequence and report the fraction of responding lineages that moved toward a known broadly neutralising signature, rather than reporting only that a precursor response occurred. The first step of that programme is the one that has been claimed; the second is the one that decides whether the strategy is real, and it is a sequencing readout rather than a clinical endpoint, so it can be run inside a phase 1.

Speculative Seventh, and cheapest of all: a retrospective correlate hunt on stored M-001 samples. Frontier A failed efficacy trial with 12,400 participants is the only dataset in existence that pairs a universal-style immunogen with observed infection outcomes at scale. If serum from that trial survives, every candidate correlate the field has proposed can be tested against real endpoints in a single analysis. Handwave Whether those samples exist, and who now controls them, is not documented in any source reachable for this brief — which is itself the kind of institutional fact that decides what science is possible.

Established The manufacturing problems in this field are unusually well solved, which is why they are not the bottleneck. Established Self-assembling protein nanoparticles — ferritin for H1ssF, the 60-mer mi3 for mosaic-8b, eOD-GT8 60mer for HIV — are produced by established routes, and the cGMP manufacture of eOD-GT8 60mer by transient transfection is published in the open literature. mRNA-LNP delivery removes the antigen-production step entirely and has been applied to H1ssF and to eOD-GT8 alike.

Frontier What remains genuinely hard is conjugation chemistry at scale for mosaic particles, where eight different RBDs must be attached to one scaffold in controlled stoichiometry, and where the immunological argument depends on the particles being genuinely mixed rather than a population of near-homotypic ones. Frontier Intranasal delivery is a device problem as much as a formulation problem: dose deposition in the nasal mucosa varies with device, posture and patient, and the mucosal candidates inherit that variability. Speculative Persistent viral vectors need engineering the field has not attempted — attenuation with demonstrated non-transmissibility, or a metabolic dependency that confines replication. And the terminal engineering requirement is unromantic: a universal vaccine that costs more per dose than a quadrivalent seasonal shot will be bought by rich health systems for pandemic insurance and by nobody else.

Frontier Adjuvants and schedules are engineering constraints disguised as immunology. Established The chimeric-HA phase 1 depended on AS03 in three of its five arms, and a squalene-based adjuvant is a supply chain, a safety file and a licensing negotiation as much as it is a formulation choice. Sequential regimens compound the problem: a strategy that requires a prime and two or three designed boosts months apart is immunologically coherent and operationally brutal, because completion rates fall with every additional visit. Speculative A germline-targeting HIV regimen that works perfectly in a trial and requires four clinic visits over eighteen months is not obviously deployable in the populations that need it most. Stockpiling adds a further requirement that nobody designs for at the outset: a pandemic-insurance vaccine has to be stable for years in storage, which favours protein nanoparticles over mRNA-LNP on current formulation science and cuts against the platform that is fastest to design.

Established Four adjacent technologies do most of the work here. Structure-based protein design supplies the stabilised stems and the scaffolds. mRNA platforms supply speed: an immunogen sequence can be in a phase 1 within months, which is why the 2023 mRNA universal influenza candidate reached the clinic with twenty-four volunteers so quickly. Frontier Deep mutational scanning and pseudovirus panels supply the epitope maps that tell you whether a mosaic particle actually redirected the response to conserved sites — the mosaic-8b work depends on exactly this. Frontier Monoclonal antibodies are the field's shadow alternative: the AMP trials showed that infusing a broadly neutralising antibody is a testable prevention strategy in its own right, and long-acting antibodies may reach some of the same public health goals without solving elicitation at all.

Speculative One adjacency is close to unexplored and belongs to xenobiology. Speculative A live-attenuated HIV-1 vaccine built through genetic code expansion — a virus made dependent on an unnatural amino acid, so that it replicates only where that amino acid is supplied — was published in 2014. It is a live-attenuated vaccine with a hard-coded off switch, it is twelve years old, and essentially nobody has pursued it toward the clinic. Speculative The same containment logic, drawn from recoded organisms with measured escape frequencies below 4.85 x 10^-12, is what a persistent CMV vector would need. Handwave Two literatures, one obvious join, no programme.

Frontier The most instructive adjacency is a success story from next door. Established Respiratory syncytial virus vaccines failed for half a century until structural biology showed that the fusion protein had to be stabilised in its prefusion conformation, after which licensed products followed. That is the strongest available argument that structure-based immunogen design converts an intractable target into a tractable one, and it is the analogy universal-vaccine advocates reach for. Frontier The disanalogy matters just as much: RSV had a correlate — neutralising titre against a conformationally correct target — and the universal influenza programme does not. The lesson to draw is that structure solved the antigen problem in RSV because the measurement problem was already solved, which is the reverse of the situation here.

Established Universal vaccines are a textbook case of a public good that markets fund only during emergencies. Established The visible funding is small and episodic: Osivax received a 19.5 million dollar BARDA contract in January 2025 for a nucleoprotein-targeting oligomeric particle platform; Centivax raised 45 million dollars in July 2025; the FLUniversal intranasal consortium is EU-funded through a public-private partnership. Against that, an influenza field efficacy trial needs tens of thousands of participants and a season in which the right viruses happen to circulate — which is why M-001 needed 12,400 people and why almost nothing else can afford the same design. Frontier The result is a field that can afford to generate immunogenicity data forever and can almost never afford to test whether the immunogenicity means anything.

Frontier The second institutional gap is ethical infrastructure rather than money. Speculative A controlled human influenza challenge programme could screen candidates at n of order 50 to 100 and would make the correlate question answerable. Challenge models exist and have been run. Frontier What does not exist is a standing ethics framework, agreed across the jurisdictions that would host such trials, for deliberately infecting healthy volunteers with a heterologous influenza strain in service of a vaccine that does not yet have an indication. That is arguably the single highest-leverage unfunded intervention in this subject, and it is institutional, not scientific.

Frontier A third gap is regulatory imagination: there is no licensure pathway that rewards transmission blocking, because vaccines are licensed on individual protection. Frontier And a fourth is political fragility — a promising HIV vaccine programme was stopped in 2025 by US funding cuts, which is a reminder that in this field the critical path can be severed by an appropriations decision.

Frontier A fifth, less obvious institutional problem is that the target definition is owned by a funder rather than by a regulator. The criteria a candidate is judged against — how much breadth, how much durability, what minimum efficacy — are set out in an agency strategic plan, not in a licensure guideline, which means the goalpost can move without any public process and that two programmes can both be on target against different definitions. Speculative What is missing is a regulatory guidance document stating the endpoints on which a universal influenza vaccine would actually be approved. An advance market commitment of the kind used for pneumococcal vaccines, priced against a defined endpoint, would do more for this field than another immunogen, because it would tell developers what they are being paid to demonstrate.

Frontier Human challenge trials are the ethical centre of this subject and the argument is closer than either side admits. Established Deliberately infecting a healthy volunteer with influenza carries real if small risk, is done under confinement, and is paid. Speculative The case for it here is unusually strong: the alternative is a 12,400-person field trial that costs a fortune, depends on the weather of a flu season, and — as M-001 demonstrated — can return a null result that no earlier study anticipated. A challenge programme trades a small, consented, closely monitored risk to a few dozen people against the diffuse cost of a field failing to learn anything for another decade.

Established The asymmetry with HIV is total and worth stating plainly: there is no HIV challenge model and there will not be one, because there is no cure. Everything HIV vaccinology knows about efficacy it has had to learn from field trials in populations at high risk of acquiring the virus, which is why the failures are so expensive and so slow.

Speculative Two further problems deserve naming. An allele-restricted vaccine — one that works for people carrying the right IGHV1-2 germline — would be the first pharmacogenomic vaccine, and the distributional politics of that have never been discussed. Handwave And a lifelong, self-perpetuating CMV vector given to healthy people raises a consent question that has no good analogue: the recipient cannot later withdraw from the intervention, because the intervention lives in them.

Frontier Two practical ethics questions are decidable and worth deciding now. Frontier The first is payment: challenge-trial volunteers are compensated for confinement and inconvenience, and the level at which compensation becomes undue inducement is a judgement that currently varies by ethics committee rather than by rule. Speculative The second is third-party risk, which challenge trials in respiratory viruses uniquely carry — a deliberately infected volunteer is a potential source of onward transmission, so confinement duration is an ethical parameter and not merely a logistical one. Neither question is unanswerable; both are unanswered, and each unanswered question is a reason a committee gives for saying no. The equity question sits behind them all: a universal influenza vaccine developed to insure wealthy countries against a pandemic, and priced accordingly, would be a worse outcome for global health than a cheap seasonal vaccine that is actually distributed.

Established The annual influenza reformulation cycle is a tax the world pays in surveillance, manufacturing and mismatch. Every year a strain-selection committee guesses, factories are retooled, and in bad-match years effectiveness falls well below what the platform can deliver. Speculative A vaccine that removed the guess would decouple immunisation from surveillance, which matters most exactly where surveillance is weakest. Speculative It would also change what a vaccination programme is: a single course in childhood, boosted rarely, rather than an annual campaign that has to be re-sold to a sceptical public every autumn.

Speculative The pandemic argument is larger. Speculative A stockpiled universal influenza vaccine, or a deployed pan-sarbecovirus one, converts the first months of a novel outbreak from a race to design and manufacture into a distribution problem. That is a change in the character of pandemic risk rather than a change in its magnitude, and it is the strongest reason to fund the field on national-security grounds rather than health-economics ones.

Handwave There is also a biosecurity argument that runs the other way and is rarely made: broad, pre-existing immunity against a whole viral family reduces the expected payoff from engineering a member of that family. A universal vaccine is, on this reading, a partial disarmament of a category of weapon. Speculative It is a real argument and it is not how any funder currently scores the work.

Speculative The deepest civilizational question is whether universality is permanent or rented. Speculative If conserved epitopes hold under population-scale immune pressure, a universal vaccine is infrastructure — something built once and maintained, like sanitation. If they erode, it is a subscription: a treadmill running at a slower speed than the annual one, but a treadmill. The two futures look identical for the first decade after licensure and completely different by the third, and no one is currently running the experiment that would distinguish them.

These horizons track when a specific, checkable thing would have to be true, not when a press release will say it is.

• 10 yr: Frontier A validated correlate of protection for at least one non-HAI readout, established in a controlled human challenge setting, is the realistic ceiling for the decade. Frontier Expect one or two universal influenza candidates to reach phase 2b or phase 3, and expect a first-in-human mosaic-8b study. Speculative A licensed universal influenza vaccine within ten years would require an efficacy trial that no one has yet funded to begin within about five, which is why the honest ten-year forecast is a measurement result rather than a product.
• 25 yr: Speculative A licensed influenza vaccine covering group 1 and group 2 influenza A with multi-season durability is plausible on this horizon and is not assured. Speculative A pan-sarbecovirus vaccine is a better bet than a pan-coronavirus one, because the sarbecovirus clade is where the conserved-epitope argument has animal protection behind it. Frontier For HIV, twenty-five years is the honest horizon for the first demonstration that a designed immunogen sequence induces broad tier-2 neutralisation in humans — a result, not a product.
• 50 yr: Speculative An HIV vaccine, if germline targeting works, arrives here rather than earlier, and probably as a multi-dose sequential regimen with allele-dependent coverage rather than as a shot. By this point the escape question should have an empirical answer, and the field will know whether universality has a half-life. Handwave Persistent-vector T-cell vaccines with engineered containment are a fifty-year technology at the earliest, because the regulatory precedent does not exist.
• 100 / 250+ yr: Handwave On this horizon the interesting object is not a universal vaccine but a universal platform: family-level immunogen design as a routine engineering step applied to any emerging pathogen within weeks. Handwave Whether that is coherent depends entirely on the fitness-cliff question — if conserved epitopes erode under population-scale pressure, the far future is a treadmill running faster, not a finish line. Handwave The version worth arguing about at this range is not a better vaccine but the end of the category: continuous, individualised immune programming in which the distinction between a vaccine and a therapeutic antibody stops being useful.

• Depends on Depends on structure-based immunogen design, protein nanoparticle scaffolds such as ferritin and the 60-mer mi3, mRNA-LNP delivery, deep mutational scanning and pseudovirus neutralisation panels, and above all on standardised immunological assays that mean the same thing in two laboratories on two continents. It depends on cohort and serum-bank infrastructure built decades ago for other purposes, because imprinting cannot be studied in people whose exposure history is unknown. It depends on GMP-manufactured heterologous challenge viruses, which do not exist for most of the strains that matter. And for the persistent-vector branch it depends on a containment technology that currently lives in the xenobiology literature: recoded organisms with measured escape frequencies, applied to a vaccine vector rather than to a chassis organism.
• Requires (not on this map) Two constraints sit outside the science and bind harder than anything inside it. The first is capital that survives the interval between emergencies: an influenza field efficacy trial needs tens of thousands of participants and a cooperative season, which is why the only completed one cost a company its future, and why current funding arrives in 20-to-45-million-dollar increments that cannot buy such a trial. The second is a standing, cross-jurisdictional ethics framework for controlled human infection with heterologous influenza strains. A challenge platform would let candidates be screened at n of order 50 to 100 and would make the correlate question answerable at all; the models exist, the governance does not, and no amount of immunology substitutes for it.
• Enables A validated correlate of protection is the enabling result, and it would unblock every early-phase programme in the subject at once, because it converts phase 1 immunogenicity from a safety exercise into a ranking. It would be reusable across influenza, the sarbecoviruses and other respiratory targets. A licensed universal influenza vaccine would decouple immunisation from annual strain surveillance, which matters most where surveillance is weakest, and would turn the first months of a novel influenza pandemic from a design-and-manufacture race into a distribution problem. Success in mucosal transmission blocking would create a licensure category that does not currently exist, and would make population-level protection purchasable without individual-level breadth.
• Adjacent Adjacent to long-acting monoclonal antibodies, which pursue the same public health goal without solving elicitation at all, and whose AMP-trial threshold framework is the closest thing the HIV field has to a correlate. Adjacent to xenobiological containment, which is the missing safety technology both for persistent viral vectors and for the genetic-code-expanded live-attenuated vaccine published in 2014 and pursued by nobody. Adjacent to systems serology and B-cell lineage sequencing, which is where a correlate would actually be found and where a germline-targeting boost series can be checked rather than assumed. And adjacent, instructively, to the RSV prefusion-F programme, which shows what structure-based design achieves when the measurement problem has already been solved.

Established “A universal flu vaccine is nearly here.” Established The only pivotal efficacy trial ever completed — M-001, 12,400 participants — failed in October 2020, after seven phase 2 trials that all reported safety and immunogenicity. The correct summary of the field's state is not that it is close, but that it has never once tested whether its immunological successes translate, and the one time it did the answer was no.

Established “Universal flu candidates have been tested in tens of thousands of people.” Established The chimeric-HA phase 1 enrolled 66. BPL-1357's phase 1 enrolled 45; the H1ssF mRNA-LNP phase 1 enrolled 50; DeltaFLU's enrolled 48; G1mHA's phase 1/2a enrolled 170; the 2023 mRNA candidate enrolled 24. Established A figure of ten thousand participants has circulated against the NIAID stabilised-stem candidate; NCT03814720 is a first-in-human phase 1 dose-escalation and nothing else. Frontier A caution about that correction itself: the reference-page text reported to carry the ten-thousand figure is not present on re-fetch, so the error belongs to general coverage and to citation drift, not to any one page.

Frontier “Discovering a broadly neutralising antibody means a vaccine is coming.” Established FI6, described in 2011, binds all sixteen influenza A HA subtypes and produced worldwide headlines announcing the prospect of a universal flu vaccine. Fifteen years later there is no product. Frontier Finding a target and eliciting a response to it are different problems, and popular coverage collapses them every time.

Frontier “RV144 showed a 30 per cent effective HIV vaccine.” Established The modified-intention-to-treat efficacy was about 31 per cent at 3.5 years, it was not durable, and it has never been replicated. The correlates analysis found one antibody response associated with 43 per cent lower risk and another with 54 per cent greater risk. Frontier The index source used here prints the efficacy as about 30 per cent adjusted to 26 per cent; the honest form is a range around thirty, from a single unreplicated trial.

Established “HIV vaccine trials have been safe, just ineffective.” Established STEP was halted in 2007 because vaccination appeared associated with an increased risk of HIV infection in some recipients. That is a different category of result from futility and it changed how the field thinks about vector immunity.

Frontier “T-cell vaccines do not work.” Frontier A CMV-vectored influenza vaccine took uniformly lethal heterologous H5N1 challenge in macaques from zero of six surviving to six of eleven, using antigens from a 1918 virus against a 2004 challenge strain. It is one study, in one species, with no human data — and it is enough to retire the blanket claim.

Frontier “Mosaic nanoparticle vaccines have been shown to protect people against future coronaviruses.” The correction here runs in both directions, and most sceptical write-ups get the first half wrong. Frontier The published mosaic-8b work includes challenge protection in mice and macaques against diverse sarbecoviruses, where homotypic controls protected only against matched strains — not merely rabbit immunogenicity and pseudovirus panels. Established And there is still no human efficacy data of any kind. The sceptical point is about the species, not about the quality of the immunology.

Frontier “Universal means one shot for life against everything.” Established Nobody in the field means that. Working definitions concern breadth across subtypes and durability of at least a season or a year, and the criteria themselves are set by funding agencies rather than derived from immunology. Speculative The version in circulation — at least 75 per cent effectiveness, protection across group 1 and group 2 influenza A, durability of at least one year, suitability for all ages — is widely quoted and could not be confirmed against the originating agency document for this brief. Speculative Treat those four criteria as unsourced until someone opens the plan.

Frontier Enthusiast-side: “mRNA solved vaccine design, so universal vaccines are now an engineering problem.” Established mRNA solved delivery and manufacturing speed. It does nothing whatever about immunodominance, imprinting, correlate validation or antigenic escape, which are the actual blockers. Established The 2023 mRNA universal influenza candidate went into a twenty-four-person phase 1.

Frontier Sceptic-side: “Decades of failure prove it is impossible.” Established Broadly neutralising antibodies against both influenza and HIV demonstrably exist in humans, and influenza bnAbs arise from natural infection under the right antigenic conditions. The immune system can make the required response; the problem is elicitation, which is an engineering problem sitting inside a scientific unknown, not a proof of impossibility.

Speculative The deepest objection, and the one the field does not answer: universality may be a property of a system at a point in time rather than a property of a vaccine. Speculative A pathogen's antigenic space is not a fixed target; it is explored under selection from the immune response you induce. Conservation in nature reflects functional constraint plus the absence of immune pressure, and immunising a population against a conserved epitope removes the second term. Handwave On the strong version of this view, “universal” is not achievable at all, only rentable. Frontier The best reply is measles: six decades of a stable vaccine against an RNA virus, because the H protein is genuinely constrained — existence proofs matter. The second-best reply is that stalk-directed broadly neutralising antibodies occur in humans after natural infection and no escape lineage has swept. Speculative The right framing is neither triumph nor despair but a number: what is the half-life of universality, and is it longer than the deployment cycle? A vaccine good for twenty years is operationally universal. Handwave Nobody models the problem this way, the experiment that would answer it is cheap, and it remains unfunded.