Cross-protection against highly pathogenic avian influenza H5N1 virus from seasonal influenza vaccines: a systematic review and meta-analysis of ferret studies Authors: IShin Tseng, Ying-Chi Huang, Wei-Liang Shih, Chung-Hsi Chou, Sten H. Vermund, Chi-Tai Fang Published: 2026 ABSTRACT ABSTRACTThe recent surge in spillover events of highly pathogenic avian influenza A(H5N1) clade 2.3.4.4b to humans and mammals in North America has raised urgent pandemic concerns. Human H5N1 vaccines are unavailable in most countries. We synthesized data from ferret challenge trials to evaluate whether widely available seasonal influenza vaccines confer cross-protection against lethal H5N1 infection. We systematically searched PubMed, Embase, and Web of Science for ferret studies of lethal H5N1 challenge published up to 5 July 2025 (PROSPERO #CRD42024520346). Random-effects meta-analyses were conducted to compare vaccine efficacy (VE) of seasonal influenza vaccines and H5N1 vaccines against H5N1-related mortality. Seroprotection was defined as a neutralizing antibody titre of ≥1:40. We identified 35 studies (157 trials). Seasonal influenza vaccines without N1 did not confer significant cross-protection (five trials; VE 14.8%, 95% CI –3.6 to 30.0). In contrast, VE was 73% for N1-containing seasonal influenza vaccines (19 trials; 95% CI 54–84) and 77% for H5N1 vaccines overall (133 trials; 95% CI 72–82) (p = 0.52). The VE of N1-containing seasonal influenza vaccines was modestly lower than that of H5N1 vaccines with seroprotection (88%; 66 trials; 95% CI 84–91; p = 0.009), but comparable to H5N1 vaccines that did not achieve seroprotection (63%; 67 trials; 95% CI 52–71; p = 0.29). The VE of seasonal influenza vaccines against H5N1 was robust across sensitivity analyses, with no evidence of publication bias (p = 0.99). Seasonal influenza vaccines significantly reduce H5N1-associated mortality in ferret trials, suggesting the cross-protection potential of currently available vaccines. Human studies are warranted. Introduction The recent surge in spillover of highly pathogenic avian influenza (HPAI) A(H5N1) clade 2.3.4.4b virus to mammals in North America has raised global alarm [1]. In addition to sporadic transmission from avian species to carnivores [2], sustained mammal-to-mammal transmission has occurred among cattle across 18 U.S. states, as of 5 October 2025 [3,4]. Reports have also documented cases of cattle-to-human transmission and bird-to-human transmission, including one fatality, underscoring the zoonotic potential of this virus [5]. The possibility of human-to-human transmission is becoming increasingly concerning. A recent U.S. study found that three of 150 surveyed bovine veterinarians had serological evidence of recent infection despite no known exposure to infected cattle [6]. Given the historically high fatality rates of H5N1 infection in humans [7], the spillover of H5N1 clade 2.3.4.4b into human populations could initiate a global pandemic with serious consequences [8]. The availability of an effective H5N1 vaccine is critical to prevent severe disease and mortality during a future H5N1 pandemic [9]. However, H5N1 vaccines have been comparatively less immunogenic in humans, typically requiring the use of aluminum [10] or MF59 adjuvants [11]. Several H5N1 vaccines have completed phase 3 clinical trials using immunogenicity endpoints [12], yet to date, only Finland has initiated H5N8 vaccination for persons at increased risk [13]. Although some nations have stockpiles, no human vaccine specific for the prevention of HPAI H5N1 is commercially available in the United States or elsewhere [14]. mRNA-based vaccines offer the advantage of rapid production without the need for egg-adapted seed strains [15]. Nevertheless, on 5 August 2025, the United States federal government announced the cancellation of its H5 mRNA vaccine development programme [16]. Given the limited availability of H5N1 vaccines and their associated safety concerns during the pre-pandemic stage, alternative strategies, such as leveraging cross-protection from seasonal influenza vaccines, are being explored. In vivo mouse studies suggest that seasonal influenza vaccination provides cross-protection against H5N1 via H5 cross-reactive antibodies [17]. Seasonal influenza vaccination has been shown to elicit cross-protective antibodies against H5N1 in 2.5% to 15% of non-elderly adults [18], and to boost titres to seroprotective levels in individuals with pre-existing H5N1 antibodies [19]. (In recent years, the human seasonal influenza vaccine has been trivalent or quadrivalent (2013–2024): an influenza A (H1N1) virus, an influenza A (H3N2) virus, and one or two influenza B lineage viruses. For example, in the 2010–2011 season, the product contained A/California/7/2009 (H1N1)-like, A/Perth/16/2009 (H3N2)-like, and B/Brisbane/60/2008-like viruses, while in 2025–2026, it had A/Victoria/4897/2022 (H1N1) pdm09-like, A/Croatia/10136RV/2023 (H3N2)-like, and B/Austria/1359417/2021 (B/Victoria lineage)-like viruses.) Seasonal influenza vaccines may confer cross-protection against H5N1, as demonstrated in an ex vivo human immunology study [20] and a ferret challenge study [21], through N1-targeted T cell-mediated immunity and antibody-dependent mechanisms. Potentially challenging is evidence that recent seasonal influenza vaccination can interfere with seroconversion following H5N1 vaccination, raising concerns about its impact on immune responses [22]. Human challenge trials, which have long been used to study the pathogenesis and transmission of seasonal influenza [23,24], would be required to provide definitive conclusions, but they pose insurmountable ethical and practical challenges because of the high fatality rate associated with H5N1 infection. Likewise, randomized clinical trials are not feasible given the low incidence of H5N1 outside pandemic conditions. Ferrets, given their similarity to humans in influenza pathogenesis and transmission, have been established as the standard animal model for studying influenza viruses [25]. Challenge trials using ferret models offer an excellent alternative to human trials for evaluating the cross-protective efficacy of seasonal influenza vaccination against HPAI H5N1. However, past ferret challenge trials have had small sample sizes, limiting the generalizability of individual studies. Given that substantial uncertainty remains regarding the extent and consistency of cross-protection conferred by seasonal influenza vaccines against H5N1, we conducted a systematic review and meta-analysis of mortality data from controlled ferret challenge trials to determine the extent of cross-protection provided by seasonal influenza vaccines against HPAI, compared with that of H5N1 vaccines. Methods Search strategy and selection criteria Our systematic review and meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The study protocol, part of a systematic review and meta-analysis project on vaccines against HPAI H5N1 across experimental animal models, was prospectively registered in PROSPERO (CRD42024520346). We searched PubMed, Embase, and Web of Science for articles published between 1 January 1997, and 5 July 2025, that reported lethal HPAI H5N1 challenge trials in ferrets assessing vaccine efficacy (VE), defined here as reduction in H5N1-associated mortality, for seasonal influenza or H5N1 vaccines, using placebo controls as the comparison group and with no language restrictions. The following search string was applied to article titles: [(avian influenza) OR (bird flu) OR (HPAI) OR (H5N1) OR (H5Nx) OR (influenza A)] AND [(efficacy) OR (protect) OR (protection) OR (immunization) OR (prevent)] AND [(vaccine) OR (vaccination) OR (hemagglutinin)]. Full search strategies are provided in Supplementary Table 1. This query was applied uniformly across all databases. Records were excluded if they were duplicates, meta-analyses, reviews, reader comments, or protocols. Two reviewers (IST and YCH) independently conducted the literature search, applied the inclusion and exclusion criteria, identified eligible trials from each article, and assessed the risk of bias using the Collaborative Approach to Meta-analysis and Review of Animal Data in Experimental Studies (CAMARADES) checklist [26]. Because multiple trials within a study shared identical methodology and conditions, quality was assessed at the study level. Any discrepancies, including continuous variables with a difference exceeding 5%, were resolved through discussion with a third reviewer (CTF). We compared VE against lethal H5N1 challenge between seasonal influenza vaccines and H5N1 vaccines. We included studies if they were experimental trials conducted in ferrets that compared mortality between vaccinated and placebo groups (unvaccinated, saline/PBS control, or vector control) following lethal H5N1 challenge, with at least 50% mortality in the placebo group. We imposed no restriction on the type of seasonal influenza or H5N1 vaccine and treated commercially available formulations and vaccines prepared for experimental use equivalently. Studies were excluded if the challenge strain was not H5N1, if the challenge dose was sublethal or could not be determined because of missing survival data, or if the vaccine strain was not H5N1, H1N1, H3N2, or influenza B. For publications that included multiple trials using different challenge doses, trials with sublethal challenge were excluded. Data extraction Data from eligible trials (experiments) were extracted using a standardized form. Extracted variables comprised: (1) study characteristics (first author, publication year, and number of trials per publication); (2) experimental animal characteristics (baseline influenza serostatus and sample size per trial); (3) intervention characteristics (vaccine type [seasonal influenza or H5N1] and dosing regimen [priming or booster]); (4) challenge H5N1 clade/strain; (5) time interval (days) from the last dose of vaccination to challenge; and (6) outcome measures (mortality counts and post-vaccination geometric mean titres [GMTs]). When outcome data were presented only in graphical format, values were digitized using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer). Outcome measures The primary outcome was mortality, defined as death from H5N1 infection or euthanasia performed for very severe illness following H5N1 challenge. The secondary outcome was seroprotection against H5N1 – measured using the challenge strain, vaccine strain, or a heterologous strain as the test strain – after study vaccination, defined as a neutralizing antibody titre of ≥1:40 by hemagglutination inhibition (HI) or microneutralization (MN) assays. When multiple HI time points were reported, the pre-challenge measurement closest to challenge was extracted. If HI data were unavailable, MN results were used. Antibody measurements were prioritized in the following order: HI with horse red blood cells, HI with chicken red blood cells, and MN with Madin–Darby canine kidney cells. Meta-analytic framework VE against mortality following H5N1 challenge was estimated by comparing mortality risk between vaccinated and control groups. VE was calculated as 1 − RR, where RR is the mortality risk ratio. Trials from the same publication were treated as separate observations when they represented distinct experimental comparisons with different vaccine regimens or challenge conditions. Because substantial clinical and methodological heterogeneity across ferret challenge studies was anticipated, we used a random-effects meta-analysis with the Mantel–Haenszel method and further explored heterogeneity through prespecified subgroup and sensitivity analyses. To examine the role of N1-targeted immunity in cross-protection, subgroup analyses of seasonal influenza vaccines were conducted according to whether the study vaccines contained the N1 antigen. Because immunogenicity varied substantially across H5N1 vaccine studies, subgroup analyses of H5N1 vaccines were performed according to whether the study vaccines achieved seroprotection. Differences in pooled VE estimates between vaccine categories were assessed by testing for subgroup differences within a random-effects model using Cochran’s Q statistic. Relative efficacy was estimated through a frequentist network meta-analysis with a random-effects model. Heterogeneity was assessed using the I² statistic. To explore potential sources of heterogeneity, additional subgroup analyses were conducted according to challenge H5N1 clade, vaccine platform (virus-like particle [VLP], DNA, inactivated, or live vaccines), adjuvant use, booster use, and study period (before versus after the 2009 H1N1 pandemic, for N1-containing seasonal influenza vaccines). To evaluate the robustness of the estimated cross-protective efficacy of seasonal influenza vaccines, we performed a leave-one-out sensitivity analysis. Because sample sizes in animal experiments are constrained to a narrow range under the Replacement, Reduction, and Refinement (3R) principles, conventional methods for detecting publication bias, such as funnel plots and Egger’s test, were not considered appropriate. Instead, we evaluated potential small-study effects, including possible publication bias, by testing whether VE varied systematically according to trial arm size. Subgroup differences were assessed using Cochran’s Q statistic across sample-size categories: small (≤4 animals per arm), medium (5–9 animals per arm), and large (10–12 animals per arm). The certainty of evidence (CoE) was evaluated using the GRADE approach [27]. Statistical analysis All statistical analyses were performed using R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) with the meta and netmeta packages. Statistical significance was defined as a two-sided p value of less than 0.05. Results We identified 2,274 potentially eligible studies (Figure 1). After removal of duplicates and screening of titles and abstracts, 94 articles underwent full-text review, of which 59 were excluded for reasons listed in Supplementary Table 2, including four conference abstracts without available full text. We included 35 studies reporting 157 eligible trials (Supplementary Table 3). These trials involved a total of 911 ferrets in the vaccinated groups and 880 ferrets in the placebo control groups. Trials were conducted between 2005 and 2024, most commonly in 2009 (40 trials) and 2012 (20 trials). Figure 1.Study selection. Of 2,274 potentially eligible items identified, 1,349 were screened by title and abstract after removing duplicates and non-original research articles. Of these, 1,255 were excluded for not involving experimental trials or for using animal models other than ferrets. After full-text review, 59 of the remaining 94 studies were excluded, with reasons listed in Supplementary Table 2. In total, 35 studies comprising 157 eligible trials were included in the meta-analysis (Supplementary Table 3). WOS = Web of Science.A flowchart summarizing screening of 2273 database records down to 35 included ferret trials with exclusion counts at each step.The figure shows a flowchart outlining study selection for a ferret trial meta analysis. At the top, a large box states that 2273 potentially eligible articles are identified through database search, subdivided as 1315 from Web of Science, 504 from Embase, and 454 from PubMed. A smaller box to the right records 1 potentially eligible article identified from other sources. Arrows from both sources lead to a box labeled 925 duplicates removed. A downward arrow leads to a box stating that 1349 articles are screened by title and abstract. From there, a rightward arrow leads to a box listing 1255 articles excluded, broken down as 1177 using experimental animals that are not ferrets, 39 meta analyses, 24 reviews, 6 comments including replies, 1 press release, 1 editorial, 4 protocols, and 3 retractions. A downward arrow from the screening box leads to a box saying 94 articles underwent full text review. A rightward arrow leads to a box summarizing 59 studies excluded: 44 using challenging strains not H5N1, 4 conference abstracts without full text, 4 with missing survival data, 5 with no lethal challenge, and 2 with no challenging test performed. The final box at the bottom reports that 35 studies, containing 157 trials, are included in the meta analysis. Study selection. Of 2,274 potentially eligible items identified, 1,349 were screened by title and abstract after removing duplicates and non-original research articles. Of these, 1,255 were excluded for not involving experimental trials or for using animal models other than ferrets. After full-text review, 59 of the remaining 94 studies were excluded, with reasons listed in Supplementary Table 2. In total, 35 studies comprising 157 eligible trials were included in the meta-analysis (Supplementary Table 3). WOS = Web of Science. Among the 157 trials, 24 evaluated vaccines derived from human strains, and 133 evaluated vaccines derived from avian strains, all of which were H5N1 or included an H5N1 valency. Seasonal influenza vaccine types included inactivated vaccines (whole-virus, split-virion, or subunit; 15 trials), live vaccines (wild-type or attenuated; 4), VLP vaccines (3), and DNA vaccines (2). H5N1 vaccine types included recombinant subunit vaccines (58 trials), inactivated vaccines (whole-virus or split-virion; 46), VLP vaccines (11), DNA vaccines (9), mRNA vaccines (7), and live wild-type H5N1 vaccines (2). The challenge H5N1 viruses belonged to A/Vietnam/1203/2004 (clade 1; 93 trials), A/Indonesia/05/2005 (clade 2.1; 26), A/Vietnam/1194/2004 (clade 1; 17), A/bald eagle/Florida/22–006544–004/2022 (clade 2.3.4.4b; 8), A/EM/Korea/W149/2006 (clade 2.2; 6), A/Whooper Swan/Mongolia/244/2005 (clade 2.2; 3), A/Egypt/2321–NAMRU3/2007 (clade 2.2.1; 2), A/Cygnus cygnus/Germany/R65/2006 (clade 2.2; 1), and A/bald eagle/Florida/W22-134-OP/2022 (clade 2.3.4.4b; 1). The interval between vaccination and challenge ranged from 7 to 140 days, with most challenges occurring 14–28 days after vaccination. The VE of seasonal influenza vaccines differed according to the presence of N1 antigens (Figure 2). Seasonal influenza vaccines lacking the N1 antigen – including monovalent H3N2 vaccine (three trials), purified H1 antigen (one), and purified H3 antigen (one), all specifically prepared for experimental use – did not confer significant cross-protection against H5N1 (VE 14.8%, 95% CI –3.6 to 30.0; I² = 47%). In contrast, N1-containing seasonal influenza vaccines (VE 73%, 95% CI 54–84; I² = 62%; CoE: moderate; Figure 2) showed protection slightly lower than that of H5N1 vaccines overall (VE 77%, 95% CI 72–82; I² = 54%; CoE: moderate; p = 0.52; Supplementary Figure 1). Figure 2.Vaccine efficacy against H5N1-related mortality for seasonal influenza vaccines. Forest plot showing vaccine efficacy among seasonal influenza vaccine trials, stratified by whether the vaccine contained the N1 antigen. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals.A forest plot chart of vaccine efficacy against H5N1 mortality for seasonal influenza vaccines, stratified by whether the vaccine contained the N1 antigen. N1-containing vaccines showed significant protective effects, whereas those without N1 did not.The figure shows a forest plot chart summarizing vaccine efficacy against H5N1 related mortality from seasonal influenza vaccine trials. Rows list individual trials by first author and year, grouped into 2 sections: neuraminidase equals N1 and neuraminidase equals others or none. For each trial, the left columns show vaccinated events, vaccinated total, control events, and control total. To the right, each trial has a square centered on a horizontal confidence interval bar along a horizontal axis labeled vaccine efficacy ranging from about minus 50 to 100 percent sign, with tick marks at 0 and intermediate values. Squares for neuraminidase equals N1 trials mostly appear on the positive efficacy side, many between about 35 and 95 percent sign, while the bars vary in length. Squares for neuraminidase equals others or none cluster closer to 0, with some bars crossing into negative values. Diamonds represent random effects model summary estimates for the neuraminidase equals N1 group, the neuraminidase equals others or none group, and all trials combined. Text below each group reports heterogeneity statistics including I squared, tau squared, and p value. All data are approximate. Vaccine efficacy against H5N1-related mortality for seasonal influenza vaccines. Forest plot showing vaccine efficacy among seasonal influenza vaccine trials, stratified by whether the vaccine contained the N1 antigen. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals. None of the seasonal influenza vaccines achieved H5N1 seroprotection. In contrast, 66 (50%) of the 133 H5N1 vaccine trials achieved seroprotection. H5N1 vaccine trials that included booster doses were substantially more likely to achieve seroprotection than priming-only trials (55.1% [65 of 118] vs 6.7% [1 of 15]; p = 0.002). The likelihood of seroprotection did not differ by test strain (p = 0.82) or by assay method (p = 0.53) (Supplementary Results). The pooled VE of H5N1 vaccines with seroprotection were 88% (66 trials; 95% CI 84–91; I² = 0%; CoE: high; Figure 3), significantly higher than that of H5N1 vaccines without seroprotection (63%; 67 trials; 95% CI 52–71; I² = 52%; CoE: moderate; Figure 4; p < 0.01; Supplementary Figure 2). However, among H5N1 trials with seroprotection, no difference was observed across subgroups using different numbers of boosters (p = 0.79; Supplementary Figure 3), indicating that once seroprotection was achieved, additional boosters did not further improve VE. Figure 3.Vaccine efficacy against H5N1-related mortality for H5N1 vaccines with seroprotection. Forest plot showing vaccine efficacy among H5N1 vaccine trials that achieved seroprotection. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals.A forest plot showing vaccine efficacy against H5N1 related mortality in seroprotective trials, with pooled estimate near 88 percent.The figure shows a forest plot summarizing H5N1 vaccine efficacy against H5N1 related mortality among trials that reach seroprotection. The left side lists individual trials, identified by first author surname, year, and a trial number in open and ) when multiple trials are reported in one article. Next to each trial, a table column labeled Vaccinated shows events and total participants, and a column labeled Control shows events and total participants. To the right, a central column labeled Vaccine efficacy displays a square for each trial on a horizontal axis from approximately negative 200 to 200 percent vaccine efficacy with tick marks every 50 percent. Horizontal lines extending from each square show the 95 percent confidence interval, with most squares clustered between roughly 60 and 100 percent. A diamond at the bottom represents the pooled random effects model estimate, centered near 88 percent vaccine efficacy with narrow confidence limits. A weight column on the far right lists the contribution of each trial to the meta analysis as percentages. All data are approximate. Figure 4.Vaccine efficacy against H5N1-related mortality for H5N1 vaccines without seroprotection. Forest plot showing vaccine efficacy among H5N1 vaccine trials that did not achieve seroprotection. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals.A forest plot chart of H5N1 vaccine efficacy without seroprotection, with most trial estimates near 60 percent and wide intervals.The figure shows a forest plot chart summarizing vaccine efficacy against H5N1 related mortality for H5N1 vaccines that do not reach seroprotection. The left side lists individual trials by first author and year, such as Hoffmann et al 2005, Govorkova et al 2006, Baras et al 2008, Lalor et al 2008, Mahmood et al 2008, Forrest et al 2009, Middleton et al 2009, Song et al 2009, Rao et al 2010, Gustin et al 2011, Layton et al 2011, Liu et al 2012, Vela et al 2012, Verma et al 2012, Mann et al 2014, Baz et al 2015, Cox et al 2015, Major et al 2015, Liu et al 2016, Wong et al 2017, Smith et al 2019, and Wang et al 2019. Next columns list vaccinated and control event counts and totals as small integers, typically between 0 and 8. The central plot area shows for each trial a square marker on a horizontal efficacy axis from about minus 100 to 100 percent, with horizontal lines for the 95 percent confidence intervals. Most squares lie on the protective side of the vertical zero line, but several intervals cross zero. Square sizes vary, indicating different weights. At the bottom, a diamond represents the random effects pooled efficacy estimate, centered around 63 percent with its confidence limits. Text below reports heterogeneity with I squared near 52. All data are approximate. Vaccine efficacy against H5N1-related mortality for H5N1 vaccines with seroprotection. Forest plot showing vaccine efficacy among H5N1 vaccine trials that achieved seroprotection. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals. Vaccine efficacy against H5N1-related mortality for H5N1 vaccines without seroprotection. Forest plot showing vaccine efficacy among H5N1 vaccine trials that did not achieve seroprotection. One published study may report multiple trials, which were labelled in the order described in the article with numbers in parentheses. Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals. N1-containing seasonal influenza vaccines had a VE of 73% (19 trials; 95% CI 54–84; I² = 62%; CoE: moderate; Figure 2), modestly lower than H5N1 vaccines with seroprotection (p < 0.01; Supplementary Figure 4) but comparable to those without seroprotection (p = 0.29; Supplementary Figure 5). In a network meta-analysis using seasonal influenza vaccination as the reference (Figure 5), H5N1 vaccination with seroprotection was associated with a substantially lower risk of H5N1-associated mortality (RR = 0.37, 95% CI 0.22–0.63), whereas H5N1 vaccination without seroprotection conferred no additional benefit compared with seasonal influenza vaccination (RR = 1.25, 95% CI 0.77–2.03). Placebo was associated with markedly higher mortality relative to seasonal vaccination (RR = 2.98, 95% CI 1.94–4.60). Figure 5.Network meta-analysis of vaccine efficacy. Relative risk of H5N1-related mortality after H5N1 challenge in ferrets vaccinated with H5N1 vaccines (stratified by seroprotection status) or placebo, relative to those vaccinated with seasonal influenza vaccines.A forest plot chart comparing relative risk of H5N1 related mortality for seroprotected H5N1 vaccine, non seroprotected vaccine, and placebo.The figure shows a forest plot chart that compares relative risk of H5N1 related mortality in ferrets across 3 treatments. The left side lists treatment groups as H5N1 vaccine achieving seroprotection, H5N1 vaccine not achieving seroprotection, and placebo. The right side displays a horizontal axis labeled relative risk, with a numeric scale from 0.5 to 2 and a tick at 1 marking no difference from the reference. Each treatment has a square marker with a horizontal confidence interval bar. The H5N1 vaccine achieving seroprotection appears left of 1 with its confidence interval fully below 1. The H5N1 vaccine not achieving seroprotection is plotted near 1 with a confidence interval that crosses 1. The placebo group is positioned to the right of 1 with its confidence interval entirely above 1. To the far right, numeric values are listed for each treatment: relative risk 0.37 with confidence interval 0.22 to 0.63 for H5N1 vaccine achieving seroprotection, relative risk 1.25 with confidence interval 0.77 to 2.03 for H5N1 vaccine not achieving seroprotection, and relative risk 2.98 with confidence interval 1.94 to 4.59 for placebo. All data are approximate. Network meta-analysis of vaccine efficacy. Relative risk of H5N1-related mortality after H5N1 challenge in ferrets vaccinated with H5N1 vaccines (stratified by seroprotection status) or placebo, relative to those vaccinated with seasonal influenza vaccines. Additional subgroup analyses showed that VE did not differ significantly across H5N1 challenge virus clades (p = 0.60; Supplementary Figure 6). For H5N1 vaccines, VE did not differ significantly by the interval from vaccination to challenge (<14 days, 89.5%; 14–28 days, 76.4%; >28 days, 75.2%; p = 0.10; Supplementary Figure 7). For N1-containing seasonal influenza vaccines, VE also did not differ significantly by the interval from vaccination to challenge (14–28 days, 72.0%; >28 days, 74.0%; p = 0.89; Supplementary Figure 8). Moreover, the VE of N1-containing seasonal influenza vaccines did not differ significantly by platform (VLP, 76.0%; DNA vaccines, 81.8%; inactivated vaccines, 66.5%; live vaccines, 85.6%; p = 0.62, Supplementary Figure 9), adjuvant use or booster use (p = 0.40 and p = 0.26, respectively; Supplementary Figures 10 and 11), or study period before versus after 2009 – when H1N1pdm09 emerged and was subsequently incorporated into seasonal influenza vaccines (76.3% vs. 72.0%, p = 0.77; Supplementary Figure 12). To assess the robustness of these findings, we conducted leave-one-out sensitivity analyses. The pooled VE of N1-containing seasonal influenza vaccines varied between 73% (95% CI 51–83) and 76% (95% CI 57–89) (Supplementary Figure 13), and that of H5N1 vaccines with seroprotection varied between 87.8% (95% CI 83.5–90.9) and 88.5% (95% CI 84.4–91.5) (Supplementary Figure 14). The quality assessment of the 35 included studies yielded a median score of 5.6 (IQR 5–6) out of nine items (Supplementary Figures 15 and 16), with minimal differences between seasonal influenza vaccine studies and H5N1 vaccine studies (median 5.43 vs 5.66; p = 0.53). VE estimates did not differ across group-size categories for N1-containing seasonal influenza vaccines (p = 0.99; Figure 6) or for H5N1 vaccines with seroprotection (p = 0.27; Supplementary Figure 17). By contrast, among H5N1 vaccine trials without seroprotection, VE declined with increasing study size, raising concern about possible small-study effects, including potential publication bias (75% for ≤4 animals per arm, 63% for 5–9 per arm, and 35% for 10–12 per arm; p = 0.008; Supplementary Figure 18). Figure 6.Publication bias assessment of N1-containing seasonal influenza vaccine trials. Forest plot showing vaccine effectiveness (VE) by trial arm sample size (small, medium, or large). Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals. No significant difference in VE was observed among subgroups (p = 0.99).A forest plot showing N1 influenza vaccine effectiveness by small, medium, and large trial arm sizes, with pooled effects centered near 70 percent.The figure shows a forest plot summarizing N1 containing seasonal influenza vaccine effectiveness by trial arm size category. The left side lists studies grouped under number equals small, number equals medium, and number equals large. For each study, columns show vaccinated events and totals, control events and totals, calculated vaccine effectiveness, confidence interval, and weight. To the right, squares with horizontal whiskers represent study level vaccine effectiveness on a horizontal scale from about minus 400 to 400. A vertical reference line runs at 0. Diamonds depict pooled random effects estimates for each group and for all studies combined. Small studies cluster around vaccine effectiveness values between roughly 35 and 90 percent. Medium studies cluster mostly between about 46 and 92 percent. Large studies include one estimate close to 96 percent and one near 0 percent, producing a wide pooled diamond. At the bottom, an overall pooled diamond lies around 73 percent with a relatively narrow span. All data are approximate. Publication bias assessment of N1-containing seasonal influenza vaccine trials. Forest plot showing vaccine effectiveness (VE) by trial arm sample size (small, medium, or large). Diamonds represent pooled estimates under random-effects models; error bars indicate 95% confidence intervals. No significant difference in VE was observed among subgroups (p = 0.99). Data for survival rate are shown in Figure 7. Vaccinated ferrets had median survival rates above 90% in seasonal influenza vaccine trials, H5N1 vaccine trials with seroprotection, and those without seroprotection, compared with 0% in the corresponding control groups (p < 0.001 for all comparisons). A summary of findings for VE of seasonal influenza vaccines and H5N1 vaccines (overall, with seroprotection, and without seroprotection) is presented in Supplementary Figure 19. Figure 7.Survival rates in vaccinated and placebo-treated ferrets. (A) H5N1 vaccine trials achieving seroprotection, (B) H5N1 vaccine trials without seroprotection, and (C) N1-containing seasonal influenza vaccine trials. Boxes show interquartile ranges, horizontal lines indicate medians, whiskers represent ranges, and dots denote outliers. Statistical comparisons between vaccine and placebo groups were performed using Student’s t-test (**** p < 0.001). Vaccinated ferrets showed significantly higher survival in all categories.Three box plots comparing survival rate in placebo and vaccine ferret groups, with higher survival for vaccine across all categories.The figure shows three side by side box charts comparing survival rate in ferrets that receive placebo versus vaccine. Each chart has survival rate on the vertical axis, scaled from 0.0 to 1.0 with tick marks at 0.3, 0.6, and 0.9. The horizontal axis of each chart lists 2 groups labeled placebo and vaccine. The left chart is titled H5N1 vaccines achieving seroprotection and displays a low survival range for placebo centered near 0.1 and a high survival range for vaccine centered near 0.95, with several individual vaccine data points scattered between about 0.6 and 0.95. The middle chart is titled H5N1 vaccines no seroprotection and again shows a low placebo survival cluster near 0.1 and a taller vaccine box spanning roughly 0.6 to just above 0.9. The right chart is titled Seasonal influenza vaccines containing N1 and shows placebo survival at 0.0 with a few points up to about 0.5, while the vaccine box sits near 0.85 to 0.95. A bracket above each chart links placebo and vaccine with four asterisks indicating a reported p value less than 0.001. A legend at the right identifies placebo and vaccine boxes. All data are approximate. Survival rates in vaccinated and placebo-treated ferrets. (A) H5N1 vaccine trials achieving seroprotection, (B) H5N1 vaccine trials without seroprotection, and (C) N1-containing seasonal influenza vaccine trials. Boxes show interquartile ranges, horizontal lines indicate medians, whiskers represent ranges, and dots denote outliers. Statistical comparisons between vaccine and placebo groups were performed using Student’s t-test (**** p < 0.001). Vaccinated ferrets showed significantly higher survival in all categories. Discussion Human challenge trials are widely regarded as the gold standard for evaluating vaccine efficacy, but these are impossible since H5N1 infection in humans has been associated with a historic case fatality rate of 52% [28], markedly higher than those of SARS (10%), MERS (37%), and COVID-19 (0.007–10.1%, depending on age) [29]. Furthermore, current low incidence of H5N1 infection makes randomized clinical trials with clinical endpoints impossible. Pre-licensure human H5N1 vaccine trials have relied solely on immunogenicity endpoint. Given the ethical impossibility of conducting human H5N1 challenge studies, and the lack of feasibility for randomized clinical trials with clinical endpoints, meta-analyses of ferret challenge trials provide the best available evidence for assessing vaccine efficacy against H5N1. Ferret lung physiology closely mimics that of humans both for influenza pathogenesis and viral transmissibility. To our knowledge, this is the first systematic review and meta-analysis of ferret challenge studies evaluating the efficacy of seasonal influenza vaccines against lethal H5N1 infection. Our findings show that seasonal influenza vaccines confer significant protection against H5N1-associated mortality in the ferret model. The VE against lethal H5N1 challenge was 73% (95% CI 54–84) for seasonal influenza vaccines containing the N1 antigen. Although this VE was lower than that of H5N1 vaccines achieving seroprotection (88%; p < 0.01), the difference in magnitude does not materially alter the overall interpretation that seasonal influenza vaccines provided substantial cross-protection against H5N1 mortality in ferrets. Our findings indicate that N1-containing seasonal influenza vaccines conferred 73% efficacy against H5N1-associated mortality in ferrets, despite not achieving seroprotection by conventional HI or MN assays targeting the H5 antigen [20,30]. In contrast, seasonal influenza vaccines lacking N1 did not confer significant cross-protection. These observations are consistent with an earlier ex vivo immunology study [20] and a ferret challenge study [21], which together suggest that seasonal influenza vaccination may confer cross-protection against H5N1 through N1-directed cellular and antibody-dependent mechanisms in the absence of HA-matched immunity. This interpretation is further supported by experimental evidence in ferrets that prior H1N1 infection reduced replication and transmission of H5N1 challenge virus more effectively than prior H5N7 infection, further implicating N1-mediated immunity [31]. Taken together, these prior mechanistic studies provide a biological framework for our pooled findings and strengthen the plausibility that N1-directed immunity is a major contributor to the cross-protection observed here. An unresolved question arising from the observed N1-directed cross-protection is the extent to which antigenic or genetic relatedness between the vaccine N1 and the challenge-virus N1 is required, particularly given the well-established strain-specific immune response patterns in ferrets. None of the included studies directly evaluated genetic or antigenic relatedness between the N1 neuraminidase of the seasonal H1N1 vaccine strain and that of the H5N1 challenge virus. Future studies addressing this question could clarify the degree of N1 matching necessary for cross-protection and help inform more rigorous characterization and quality control of the N1 component in seasonal influenza vaccine production. We observed that seasonal influenza vaccines lacking N1 were not associated with statistically significant cross-protection; however, this estimate was based on only five trials and may therefore have been underpowered to detect smaller protective effects. Accordingly, these findings should not be interpreted as definitive evidence of no cross-protection. Additional immune mechanisms, such as vaccine-induced cell-mediated immunity or limited cross-reactive antibody responses not captured by standard neutralization assays, may also contribute. In humans, a prospective study reported that seasonal influenza vaccination increased H5N1 antibody titres in a subset of participants, with seroprotection achieved in 2.5–15.0% of non-elderly adults [18], suggesting that seasonal vaccines can elicit limited cross-reactive antibodies against H5 antigens. Similarly, seasonal influenza vaccination has been reported to induce cross-reactive antibodies recognizing the 2009 A(H1N1)pdm09 strain [32]. Together, these findings highlight the multifaceted nature of vaccine-induced cross-protection and underscore the need to clarify the underlying immunological mechanisms to inform the design of broader and more effective influenza vaccines against emerging H5N1 viruses. Our results do not diminish the importance of H5N1 vaccines in the face of emerging H5N1 threats [8]. However, in the pre-pandemic setting, broad seasonal influenza vaccination coverage could represent a safe, widely available, and practical interim measure while strain-matched H5N1 vaccines are being developed and deployed. Current recommendations encouraging seasonal influenza vaccination aim primarily to prevent co-infection with H5N1 and seasonal influenza viruses, thereby reducing opportunities for reassortment and the emergence of novel strains with enhanced transmissibility or virulence. Beyond these indirect benefits, our findings in the ferret model raise the possibility that seasonal influenza vaccines may also provide some direct cross-protection against H5N1-associated mortality. However, whether such protection occurs in humans remains uncertain and requires confirmation in human observational and serological studies. It is important to note that all trials included in our study were forced lethal-challenge experiments in immunologically naïve animals with mortality as the primary outcome. Therefore, these ferret-derived findings may not be directly generalizable to real-world prevention of H5N1 infection, transmission, or disease severity in humans, whose diverse histories of influenza infection and vaccination are likely to shape baseline immune memory. Nevertheless, supporting evidence from a previous study showed that H1N1-specific memory B cells induced by seasonal influenza vaccination cross-reacted with H5N1 in 45% of adult participants [33]. This finding suggests that immunity induced by seasonal influenza vaccination may facilitate a more rapid secondary immune response upon H5N1 exposure. Although our study did not assess the effect of vaccination on viral load or transmission in the ferret model, emerging evidence suggests that seasonal influenza vaccination and prior influenza A(H1N1)pdm09 virus infection may offer some degree of protection against H5N1 viral replication, shedding, and possibly transmission in ferrets [21,31]. Even partial cross-reactive immunity could still have public health relevance if it reduces susceptibility or disease severity and thereby helps mitigate early spread, buying time for the development and deployment of strain-matched H5N1 vaccines. Although our analysis included a clade 2.3.4.4b challenge strain and did not detect significant differences in vaccine effectiveness across H5N1 clades, the available data remain insufficient to establish whether the observed cross-protection extends to the rapidly evolving 2.3.4.4b variants currently circulating in cattle and other mammals. An important limitation in the context of pandemic preparedness is that the apparent cross-protective effect identified in our study is most consistent with immunity directed against N1. Accordingly, these findings may be more relevant to emerging avian influenza viruses bearing N1 neuraminidase, such as H5N1, than to non-N1 viruses such as H5N8 [34]. If a future epidemic or pandemic were caused by a non-N1 avian influenza virus, the relevance of N1-directed cross-protection from seasonal influenza vaccination would be expected to be substantially lower. Our study has several limitations. First, ferret studies did not use blinding or randomization protocols. However, the primary outcome – mortality – is unlikely to have been influenced by subjective bias. Moreover, unlike humans, ferrets used in experimental settings are confirmed influenza-negative before challenge, reducing inter-individual variability and limiting the need for strict randomization. Second, although unpublished full reports eligible for inclusion were not identified, some potentially relevant data may remain unavailable. This limitation should be considered when interpreting the findings. In particular, animal challenge studies are inherently susceptible to selective reporting. Because there is no registry for ferret experiments, we cannot completely exclude publication bias. Nevertheless, the observed cross-protection conferred by seasonal influenza vaccines is unlikely to be solely an artefact of publication bias, as five ferret trials reporting no cross-protection by seasonal influenza vaccines lacking the N1 antigen were included, and subgroup analysis by sample size – which detected possible publication bias for H5N1 vaccines without seroprotection – did not indicate such bias for N1-containing seasonal influenza vaccines. Third, multiple trials from the same publication were included as separate observations when they represented distinct experimental comparisons; however, residual non-independence within studies cannot be fully excluded. Fourth, under the 3R ethical principle, ferret trials reported within the same study often shared a common control group. However, this is not a methodological problem from biological viewpoint, as mortality rate in the control group is, by definition, not a random variable. The strain and inoculum dose of the challenge H5N1 virus determine the mortality rate in animal experiments. To assess VE against H5N1-related mortality, researchers must ensure that the challenge strain and dose are lethal. This is the purpose of using a control group in such experiments. The inherently low variance in mortality reduced the need for additional control animals to achieve precise VE estimation. Fifth, although VE did not differ significantly across platforms among N1-containing seasonal influenza vaccines but could still vary in magnitude, the diversity of vaccine platforms used in ferret studies complicates direct extrapolation to traditional non-adjuvanted trivalent inactivated seasonal influenza vaccines, which may be less immunogenic than newer platforms [21]. Thus, while current seasonal influenza vaccines generally contain an H1N1 component and therefore include N1, our findings should not be interpreted as indicating that all such vaccines would necessarily confer the same degree of protection against H5N1. Nonetheless, the landscape of commercial seasonal influenza vaccines is evolving, with cell culture-based, adjuvanted, recombinant, and live-attenuated formulations offering enhanced immunogenicity. An additional limitation is that some included trials used reassortant challenge viruses incorporating H5 on a PR8 (N1) backbone rather than fully wild-type H5N1 viruses. Because influenza pathogenicity and lethality can be influenced by internal gene constellations, mortality observed in these experiments may partly reflect properties of the PR8-derived internal genes rather than those of naturally circulating H5N1 viruses alone. This may have introduced heterogeneity into the pooled estimates. Moreover, differences in serological assays, including HI assays using different red blood cells and MN assays, may have reduced comparability of seroprotection assessments across H5N1 trials and thereby introduced bias into the pooled estimate. Finally, VE estimates from ferret trials, in which viral challenge was administered predominantly 14–28 days after vaccination, may not be directly generalizable to humans, because in pandemic preparedness scenarios the interval between vaccination and exposure to H5N1 would likely be much longer, potentially resulting in waning VE. Human investigations – such as observational cohort studies comparing vaccinated and unvaccinated high-risk individuals, and serological investigations – are needed to validate these findings and clarify their implications for human populations. In conclusion, this systematic review and meta-analysis of available ferret H5N1 lethal-challenge studies indicates that seasonal influenza vaccines were associated with substantial protection against H5N1-associated mortality in this animal model. Although such cross-protection cannot replace the strain-specific immunity provided by H5N1 vaccines during a pandemic, it may represent a potential interim benefit in pre-pandemic preparedness. However, the relevance of these findings to humans remains uncertain. Human observational and serological studies are needed to determine whether the cross-protective effects observed in ferrets translate into clinically meaningful protection in human populations. Supplementary Material ICMJE_Vermund.docx Supplementary Appendix.pdf ICMJE_WL Shih.docx ICMJE_YC Huang.docx ICMJE_IS Tesng.docx ICMJE_CT Fang.docx ICMJE_CH Chou.docx