Persistence of Neutralizing Antibody Responses Among Yellow Fever Virus 17D Vaccinees Living in a Nonendemic Setting

Abstract

Background

The once-in-a-lifetime recommendation for vaccination against yellow fever virus (YFV) has been controversial, leading to increased scrutiny of the durability of immunity after 17D vaccination.

Methods

This is a cross-sectional analysis of 17D vaccinees living in nonendemic Portland, Oregon. Neutralization assays were used to determine YFV immunity. The relationships between 17D immunity and vaccination history, demographics, and travel were evaluated using nominal logistic regression.

Results

Seventy-one of 92 (77.2%) subjects were YFV seropositive (90 percent plaque reduction neutralization test ≥1:10) at all timepoints, and 24 of 38 (63.8%) were YFV seropositive at ≥10 years after single-dose vaccination. No relationship was found between YFV immunity and time in endemic countries, other flavivirus immunity, or demographics. Subjects were most likely to become seronegative between 3 and 12 years postvaccination (logistic regression, odds ratio [OR] = 1.75; 95% confidence interval [CI], 1.12–2.73). A comparison of our results and 4 previous studies of YFV nonendemic vaccinees found that overall, 79% (95% CI, 70%–86%) of vaccinees are likely to be seropositive ≥10 years postvaccination.

Conclusions

These results suggest that 1 in 5 17D vaccinees will lack neutralizing antibodies at ~10 years postvaccination, and a booster vaccination should be considered for nonendemic vaccinees before travel to regions where there is a high risk of YFV transmission.

(See the Major Article by Ndeffo-Mbah and Pandey on pages 2026–34, and the Editorial Commentary by Barrett on pages 1922–4.)

Yellow fever virus (YFV) is the prototype flavivirus—transmitted by Aedes spp and Haemogogus spp mosquitoes—and it is closely related to dengue virus (DENV) and Zika virus (ZIKV). Historically, YFV caused explosive epidemics in Europe, the Americas, and Africa [1, 2], and it remains endemic in 34 African and 13 South American countries. Over the past decade, YFV cases have surged dramatically, with an estimated 130 000 severe cases and 78 000 fatalities in Africa in 2013 [3], 7334 sylvatic-urban cases in Angola and the Democratic Republic of Congo in 2015–2016 [4], and ongoing sylvatic outbreaks in Brazil [5, 6]. Ominously, city parks in São Paulo were closed in 2017 after at least 1 park-dwelling monkey died from YFV [7], whereas the Republic of Congo reported at least 1 confirmed case of YFV in the heart of the port city of Pointe-Noire in July 2018 [8]. In addition, 11 YFV cases were imported by travelers into China from Angola in 2016 [9], potentially exposing the largely unvaccinated Asian population to epidemic transmission. Taken together, the rising incidence of YFV outbreaks into previously well controlled endemic settings and introductions into unvaccinated populations in nonendemic [6] settings demonstrate the enormous risk YFV still poses to the global population.

Although well regarded as highly immunogenic and durable, the yellow fever vaccine 17D, introduced in 1937, has been administered as a prime followed by boosting every 10 years to maintain YFV-neutralizing antibodies [10], which are thought to be both necessary and sufficient for protection [11, 12]. However, rising global demand for vaccine doses to control YFV outbreaks has resulted in unprecedented worldwide shortages of 17D, and a production shutdown of the US supply has led to US reliance on European 17D supplies at a limited number of vaccination centers under an investigational new drug protocol [13]. These critical shortages led to a re-evaluation of the 10-year boost recommendation and exploration of strategies to extend vaccine stocks such as fractional vaccine dosing in emergency settings [14–16]. In 2013, the World Health Organization (WHO) Strategic Advisory Group of Experts (SAGE) on vaccination concluded that there was “no demonstrated need” for a 10-year boost [17], a conclusion shared by the Centers for Disease Control and Prevention (CDC) Advisory Committee on Immunization Practices (ACIP) [18] and the World Health Assembly, both of whom removed the 10-year boost recommendations from their guidelines in 2015 and 2016, respectively [19].

The policy changes were based in part on data from prior cross-sectional serosurveys, summarized in the WHO Grading of Recommendations Assessment, Development and Evaluation (GRADE) (Table 3), that showed an average of 88% of vaccinees remained seropositive for 10 or more years after single-dose vaccination [18]. However, these serosurveys were heterogeneous in both methods and results, calling into question the generalizability of the overall conclusion [2]. Specifically, when stratified and pooled by endemic/nonendemic regions, 92.7% (359 of 387) of endemic subjects were seropositive by various criteria [20–23], but only 83.7% (282 of 337) of nonendemic subjects were seropositive [24–28]. This disparity may be due to asymptomatic boosting by YFV exposure in endemic subjects, unreported YFV vaccine boosts in endemic subjects, infection with other flaviviruses, age differences of vaccinees, or other unknown confounders/effect modifiers. Irrespective, these data argue that 17D immunity may differ between endemic and nonendemic populations, with nonendemic studies representing the worst-case scenario for vaccine efficacy with potentially serious implications for both vaccine deployment and vaccine safety.

To gain more detailed insight into long-term persistence of neutralizing antibodies induced by 17D in nonendemic vaccinees, we established a cohort of 92 17D-vaccinated adults living in the Portland, Oregon area. We performed neutralization testing on their sera against YFV and other arboviruses, and we subsequently evaluated the roles that subject age, time since 17D vaccination, lifetime residence and travel histories, and serologic evidence of other flavivirus infections play in 17D immunity in our cohort. We then evaluated our results alongside the GRADE (Table 3) nonendemic studies. The results are expected to contribute to what is known about 17D-induced immunity in nonendemic populations, our understanding of how YFV-neutralizing antibody titers are affected by time since vaccination, prior or subsequent flavivirus infection and travel history, and inform how clinicians should consider revaccination in travelers who reside primarily in nonendemic countries.

METHODS

Human Research Ethics

The study was reviewed and approved by the Oregon Health & Science University Institutional Review Board (IRB no. 10212). Informed consent was obtained from subjects on initiation of their participation in the study.

Study Population

Subjects were YFV-vaccinated individuals selected from a study of long-term immunity after infection with the arthropod-borne DENV, ZIKV, and chikungunya virus (CHIKV). Subjects were aged ≥18 years living in the Portland, Oregon metropolitan area with proven or suspected arboviral infection. The subjects were recruited from area college campuses, hospitals, clinics, and international nongovernmental organization offices. Known and suspected arboviral infections, lifetime travel histories, and subject-reported YFV vaccination history were documented. Yellow fever virus vaccination history was confirmed by vaccination record, electronic medical record (EMR), or subject recall corroborated by travel to YFV-endemic risk area within the time frame of reported vaccination.

Neutralization Assays

Ninety percent plaque reduction neutralization test (PRNT₉₀) titers were used to characterize sera. Assays are prepared in duplicate. Sera were heat-inactivated for 30 minutes at 56°C, diluted 4-fold from a starting dilution of 1:5, and mixed with an equal volume of ~25 plaque-forming units of vaccine strain 17D (YF-VAX; Sanofi-Pasteur), giving a starting serum dilution of 1:10. Virus-dilution mixes without sera were prepared as controls. After 1 hour of incubation, virus mixes were inoculated into individual wells of 24-well plates seeded with Vero E6 cells, incubated for 1 hour, and overlaid with 1% methylcellulose. Plates were incubated for 6 days at 37°C, 5% CO₂. The overlay was then removed, monolayers were fixed with 80% methanol and stained with 2% crystal violet, and plaques were enumerated by visual review of each well. Proportion of virus neutralized per well was calculated, and the serum dilution that neutralizes 90% of control input virus (PRNT₉₀) was determined by sigmoidal dose-response curve fitting using GraphPad Prism, version 7.0.

Statistical Methods

All statistical calculations were conducted in JMP, version 14.0 (SAS Institute Inc., Cary, NC, 1989–2018) or using R (R Core Team, 2018) in RStudio (RStudio Team, 2016). Neutralizing antibody decay rate was estimated using standard least squares regression of log₁₀-transformed PRNT₉₀ and years postvaccination. Serostatus odds were estimated using nominal logistic regression. For serostatus odds by years postvaccination, study subjects were divided into 3 nonoverlapping subgroups by years postvaccination, corresponding to early, middle, and late years postvaccination. Subgroups evaluated included the following: early, 0–3 through 0–5 years postvaccination; middle, 3–10 through 5–14 years postvaccination; and late, >10 through >14 years postvaccination. Serostatus odds by years-postvaccination were then estimated within each subgroup using nominal logistic regression (Supplemental Figure 1). Comparison between the current study results and prior studies was executed using a random effects model with double-arcsine transformation to estimate the proportion and 95% confidence intervals (CIs) for seropositive subjects adjusted for study size (R, General Package for meta-analysis, 2018).

RESULTS

Subject Demographics

A total of 227 subjects living in and around Portland, Oregon were recruited into a larger study of arbovirus immunity, from which we identified 92 subjects who were vaccinated with 17D for travel, work, or other related indications. Subject demographics are summarized in Table 1. Of the 92 subjects who received 17D vaccination in connection with travel, 64 of 92 (70%) subsequently traveled to YFV-endemic regions. One native Brazilian spent 48 years in YFV-endemic settings (primarily Brazil), 1 native Venezuelan resided 24 years in Venezuela, and 1 native US subject spent 8 years in Brazil. The remaining 61 subjects spent between 1 and 365 weeks in YFV-endemic countries with a median of 10.5 weeks (interquartile range [IQR], 4.0–56.6).

Other Flavivirus Infections

Because 17D vaccinated subjects often travel to areas endemic for other arboviruses, we used neutralization assays to assess for evidence of DENV, ZIKV, and CHIKV infections (Table 2). Forty-four of 92 subjects (48%) had serologic evidence of other arboviral infection by detection of neutralizing antibodies against DENV, ZIKV, and/or CHIKV (Table 2). Dengue virus was the most common infection with 35 seropositive subjects. Eight subjects were ZIKV seropositive and 5 were CHIKV seropositive. Sixteen subjects also reported a history of Japanese encephalitis vaccination, but the type of vaccine, live-attenuated or inactivated, and number of doses was not documented by this study and was not included in the subsequent analyses.

17D Serostatus

Study subjects reported vaccination between 1 month and 61 years before enrollment, with a median 9.5 years postvaccination at enrollment. Overall, 71 of 92 subjects (77%) were YFV seropositive (PRNT₉₀ ≥1:10), with 45 of 53 (85%) seropositive among vaccine card/EMR confirmed subjects, with a median 5.6 (IQR, 2.5–12.4) years postvaccination, and 26 of 39 (67%) among verbal/travel history subjects, with a median 14.8 (IQR, 8.4–30.8) years postvaccination. Nine subjects reported having had at least 1 booster dose, giving 62 of 83 (75%) seropositive for single-dose vaccinees. The 9 boosted subjects were censored from statistical analyses.

We found a significant correlation between subject YFV PRNT₉₀ titer and years-postvaccination, with PRNT₉₀ titers waning over time in a power model manner (titer and years postinfection log-transformed) (Decay rate = −0.351 logPRNT₉₀/log-year, P = .0062) (Figure 1). To evaluate the relationship between serostatus and years-postvaccination in more detail, vaccinees were divided into 3 subgroups by years-postvaccination, and nominal logistic regression was used to model the relationship between serostatus—seropositive or seronegative at PRNT₉₀ ≥1:10—and years-postvaccination (Supplemental Figure 1, Table 3). Although antibody titers decline over the first 3–5 years postvaccination (Figure 1), seropositivity rates remained high, with no significant odds of loss of seropositivity with the first 3 years postvaccination (Figure 1, Supplemental Figure 1). Subjects 3–12 years (N = 37) postvaccination had significant odds of loss of seropositivity per year postvaccination (odds ratio [OR] = 1.75; 95% CI, 1.12–2.73), whereas subjects >12 years postvaccination did not (OR = 0.99; 95% CI, .94–1.04). This analysis suggests that most who lose YFV seropositivity will do so between 3 and 12 years postvaccination, and that YF serostatus at >12 years postvaccination is unlikely to change over greater time periods.

We next asked whether there were individual exposures or modifiers that correlated with YFV immune status, including total weeks in YFV-endemic countries, age, age at vaccination, gender, and coexistent flavivirus immunity, by univariate logistic regression. We were surprised to find that in univariate analyses, none of these variables had a significant effect on the odds of becoming seronegative, either overall or by years-postvaccination subgroup (Table 4).

Other Nonendemic Studies

In our study, 29 of 43 (67.4%) subjects were seropositive ≥10 years postvaccination, 5 of whom, all seropositive, were boosted at least once. Among single-dose vaccinees, 24 of 38 (63.2%) were seropositive ≥10 years after single-dose vaccination. To put our single nonendemic cohort study in context with other nonendemic YFV immunity studies, we evaluated our proportion of subjects with PRNT₉₀ ≥1:10 alongside proportions in comparable nonendemic GRADE (Table 3) publications [28] cited by the CDC vaccination recommendations (Supplemental Table 1) [18]. Similar to our study, each of these other nonendemic studies used serum dilution cutoffs of 1:8 or 1:10 and 90% or 80% neutralization cutoff thresholds. Because of potential variability among estimated neutralization titers between laboratory tests, we examined a proportion of subjects neutralized at a threshold 90% neutralization titers and did not directly compare PRNT titers between studies. For Poland et al [27], we used the reported PRNT₉₀ serum dilution cutoff of 1:8 rather than the 1:2 cutoff cited by GRADE (Table 3) because the difference between the 1:8 and 1:10 cutoffs was expected to fall within the margin of error for the PRNT assay (validated by reanalyzing our own data at a dilution cutoff of 1:8 rather than 1:10, finding no additional seropositive subjects in our cohort). Coulange Bodillis et al [24] originally reported 80 of 84 seropositive using an 80% PRNT (PRNT₈₀) ≥1:10 cutoff. Because 80% neutralization titers may be more sensitive for neutralizing antibodies than 90% neutralization titers, a serum dilution cutoff of PRNT₈₀ ≥1:10 may overestimate seropositivity relative to a PRNT₉₀ ≥1:10, depending on subject neutralization titers. The authors made available PRNT₈₀ values for their subjects, 9 of whom had PRNT₈₀ = 1:10. These subjects were conservatively classified as seronegative at PRNT₉₀ <1:10, giving a revised maximum 71 of 84 seropositive at PRNT₉₀ ≥1:10. After regression analysis, we found that a cutoff PRNT₉₀ = 1:10 was approximately equivalent to a cutoff PRNT₈₀ = 1:20 (95% CI, 1:18–1:22) (Supplemental Figure 2). The remaining studies—Niedrig et al [29] (N = 51) and Lindsey et al (N = 66) [30]—used a PRNT₉₀, but Niedrig et al originally reported 37 of 51 (73%) seropositive at a PRNT₉₀ >1:10, rather than ≥1:10. When including subjects with PRNT₉₀ ≥1:10, 45 of 51 (88%) were seropositive (Supplemental Table 1). When we analyzed data across all 5 studies, we obtained a summary proportion estimate of 78.7% seropositive ≥10 years postvaccination (Figure 2). Inspection of Figure 2 shows substantial heterogeneity across the studies, and the heterogeneity estimate of 69% is consistent with moderate heterogeneity. In this comparison, there are both unknown and known potentially confounding variables for which we could not practically control, particularly the manner in which vaccination history was established for each study; however, our study and Coulange Bodilis et al [24] used a mix of vaccination records and subject report, Niedrig et al [29] exclusively used vaccine record, Lindsey et al [30] used dates reported on CDC sample submission sheets, and Poland et al [27] used US Military service history as a surrogate for vaccination record. Each approach is expected to have different levels of certainty and different degrees of positive and negative selection bias. Nevertheless, subsequent analysis using studentized residuals to assess for statistically significant outliers found that no single study was an outlier when compared with the other studies at an absolute z-value ≥2. Even so, our calculated heterogeneity (I²) indicates that the summary proportion of 78.7% should be interpreted with caution and may not be generalizable to generic vaccinated nonendemic populations.

DISCUSSION

Our evaluation of long-term immunity after 17D vaccination in a nonendemic cohort is among the more comprehensive cross-sectional analysis to date of long-term YFV immunity in a cohort with diverse vaccination and related exposure histories. However, our single cohort study has several limitations: (1) it was relatively small and not powered sufficiently to identify subtle effects from potential confounders; (2) we had relatively few vaccinees at early time points (<3 years postvaccination, N = 13), which made evaluation of our potential confounders impossible for this subgroup; and (3) we could draw no conclusions regarding the effect of confounders on early vaccinees.

Nevertheless, several of our results address, in a limited manner, knowledge gaps about the durability of YFV-neutralizing antibodies after vaccination and the potential effects of likely confounders of such analyses. Several groups have speculated that both endemic and nonendemic 17D immunity studies may be difficult to compare when it is not possible to account for potential unrecorded confounding variables such as past flavivirus exposure, age at vaccination, and travel in endemic countries [24, 26, 30]. In this study, we specifically evaluated the effect of these potential confounders, and, although all are biologically plausible, none showed a trend towards influencing long-term serostatus in nonendemic vaccinees (Table 3). Going forward, we suggest that comparisons across nonendemic cohorts are unlikely to be substantially biased by the influence of these otherwise plausible confounding effects.

The comparisons of results between different laboratories should also be interpreted with caution because we are comparing results across different laboratories using different methods. Although substantial variability in absolute PRNT titers may arise due to the serial dilution scheme used for the assay, limiting our primary comparison to the proportion of subjects that met a threshold PRNT₉₀ ≥1:10 would be expected to minimize variability because the measurement is being assessed at the starting, or near starting, dilution of the laboratory assays for all studies. Nevertheless, we also assessed heterogeneity between PRNT titers obtained by each study, by comparing geometric mean PRNT titers (GMT) across time points for which data were available from the publications [27, 30] or from the authors (generously provided by Matthias Niedrig [29] and Paul-Henri Consigny [24]; summarized in Supplemental Table 2). Despite the PRNT assays being performed in diverse laboratory settings, we found no evidence of significant variation in PRNT titers between studies (Supplemental Table 2). Although we were unable to compare the PRNT₈₀ values from Coulange Bodillis et al [24] with the other studies that used PRNT₉₀ values, the point estimate-adjusted proportion seropositive at PRNT₉₀ ≥1:10 for Coulange Bodillis was comparable to the other studies (Figure 2), and the 95% CIs for the Coulange Bodillis study fell within the 95% CIs for the other 4 studies. These comparisons, taken together, suggest that it is unlikely that the differences in seropositivity at a threshold dilution of 1:10 are significantly affected by variation in PRNT methods between laboratories.

More important, and consistent with most other studies, we confirm that, for our nonendemic cohort, almost all vaccinees remained seropositive at early time points after vaccination. Although we found that the seropositive proportion was most likely to fall steadily between 3 and 12 years postvaccination, we suspect that most of the conversions from seropositive to seronegative still occurred <10 years postvaccination. This is supported by at least 2 other studies out of YFV-endemic Brazil [31, 32] and the Niedrig et al [29] data. The Brazilian studies examined YFV seropositivity earlier than 10 years postvaccination, finding that neutralizing titers reach a steady state of ~80% seropositive before 10 years: Campi-Azevedo et al [31], using a PRNT₅₀ ≥1:50 neutralization threshold, reported that 83.1% subjects vaccinated 5–9 years prior were seropositive compared with 93.9% vaccinated 1–4 years prior and 76.1% vaccinated ≥10 years prior, arguing that most titer decay occurs between 5 and 10 years postvaccination. de Menezes Martins et al [32] also found that by 8 years postvaccination, 82.4% of full-dose vaccinees remain seropositive. Finally, subgroup nominal logistic regression analysis of the Niedrig et al [29] PRNT₉₀ cohort data found the odds of seroconversion to be greatest between years 3 and 9 postvaccination (OR = 1.63; 95% CI, 1.03–2.59) (Supplemental Table 3). These data, taken together, suggest that careful prospective studies of vaccinees from primary vaccination through the first 10 years postvaccination should be expected to precisely estimate antibody decay and loss of seropositivity after vaccination because this time period is when individuals are most likely to go from seropositive to negative. Rather than leading to a return to a global recommendation of every 10-year boost, we believe these studies provide rationale for data-driven recommendations of who needs only a single dose of vaccine, who will need boosting, and, if needed, when they need a boost.

Finally, the conclusion of our random-effects model study comparison—78.7% seropositive ≥10 years postvaccination—suggests that approximately 1 in 5 vaccinees will be seronegative at 10 years postvaccination. In fact, our finding that 63.2% of single-dose vaccinees ≥10 years postvaccination was echoed by the Brazilian Collaborative Group for Studies on Yellow Fever Vaccines who found 68.7% (68 of 99) single-dose vaccinees seropositive at their threshold PRNT₅₀ ≥1:50 at 10 to 22 years postvaccination [33]. In contrast, a recently published single-center prospective 17D fractional dose study found a much higher seropositivity rate of 98% (39 of 40 subjects, PRNT₈₀ ≥1:10) at 10 years postvaccination [34]. Although encouraging, this single-center experience may be expected to have limited generalizability or external validity, given that the vaccine was administered in a research context, at one center, to a carefully qualified adult population, rather than capturing the inherent variability in real-world vaccine administration that cross-sectional studies capture. The results of our combined study comparison suggest that in the community, seropositivity rates are likely to be lower than that reported by this single-center study.

The policy implications of our overall result depend on the context to which they are applied and should not be overgeneralized. Although the herd immunity threshold for YFV is not firmly established [35], single-dose vaccination may be sufficient to prevent urban transmission of yellow fever. However, for individuals traveling or working in sylvatic settings where nonhuman primates serve as the virus reservoir, herd immunity does not play a role in protecting from sylvatic infections like those reported in both Brazil and the Democratic Republic of the Congo. For individuals likely to face exposure to wild-type sylvatic YFV, the risk of lost immunity is potentially lethal, and we would argue that boosting, even before 10 years postprime, should be strongly considered even if, or especially if, the odds of loss of immunity are 1 in 5. This risk is highlighted by publicly available Pan-American Health Organization data that show at least 7.12% of reported sylvatic yellow fever cases in Peru, Brazil, Columbia, and Bolivia were in vaccinated individuals [36].

CONCLUSIONS

Rather than being considered a critique of current guidelines, the results of our study should instead be seen as rationale for conducting more detailed investigations into why 17D works so well in many vaccinees but may fail to provide long-term immunity for a significant number of other vaccinees. We see such investigations as critical to providing data-driven recommendations and, more importantly, effective and directed use and potentially improvement of an invaluable vaccine.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We thank the study participants. We also thank Drs. Matthias Niedrig and Paul-Henri Consigny for generously providing the data from their publications. Finally, we thank Matt Collins and Paul-Henri Consigny for reviewing the manuscript.

Financial support. This work was supported by federal funds from the National Institute of Allergy and Infectious Diseases (Grant R21 AI135537-01; to W. B. M.) and the National Center for Advancing Translational Science Clinical and Translational Science Awards (Grant UL1 TR000128), Oregon Clinical and Translational Research Institute, Takeda Vaccines (IISR 2016–101586; to W. B. M.), Sunlin and Priscilla Chou Foundation (to W. B. M.), and Oregon National Primate Research Center grant, 8P51 OD011092 (MKS).

Potential conflicts of interest. I. J. A., an employee of Najít Technologies, Inc., Oregon Health & Science University (OHSU), and M. K. S. have a financial interest in Najit Technologies, Inc., a company that may have a commercial interest in the results of this research and technology. This potential individual and institutional conflict of interest has been reviewed and managed by OHSU. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References