We conducted this investigation during May 12, 2014–December 16, 2016, simultaneously sampling from Rio de Janeiro and 2 municipalities of Minas Gerais state (Alfenas and Ribeirão das Neves), Brazil. We assigned participants to groups on the basis of official vaccination records. The study included 421 samples collected from 326 healthy adults 18–77 years of age, initially categorized into 3 arms: primary vaccination, secondary vaccination, and multiple vaccination (Figure 1). We designed the primary vaccination and secondary vaccination arms as 2 complementary independent approaches, each including a longitudinal (95 paired samples) and a cross-sectional investigation (231 unpaired samples). Study groups were coded to indicate participants’ vaccination status (NV for nonvaccinated persons, PV for those who had had primary YF vaccination only, RV for those who had been revaccinated) and time since last vaccination, given in days or years (e.g, d0 for day zero).

We collected whole blood samples from each participant. We used samples of 5 mL without anticoagulant for plaque-reduction neutralization test (PRNT) and samples of 20 mL in heparin for 17DD-YF phenotypic and functional analyses.

This study was composed of 2 independent but complementary approaches: a longitudinal investigation and a cross-sectional investigation. We performed statistical analyses for the longitudinal investigation using paired t-test to compare NV(d0) versus PV(d30–45) groups for primary vaccination, as well as #PV(y>10) (primary vaccination >10 years ago) versus RV(d30–45) for secondary vaccination. For the cross-sectional design, we made the transversal comparisons among groups using analysis of variance adjusted to multiple comparisons and set statistical significance at p<0.05. (We did not highlight nonsignificant differences in the figures.) We used the χ² test to compare seropositivity rates between NV and PV groups and also between #PV and RV groups.

We performed biomarker signature analysis as described previously by Luiza-Silva et al. (15). In brief, we calculated the global median value of 17DD-YF Ag/CC Index for each phenotypic and functional biomarker and used that value as the cutoff to identify each biomarker as low index (below global median) or high index (above global median). We considered only biomarkers observed in >50% of study participants for comparative analysis among groups.

We conducted Venn diagram analysis (http://bioinformatics.psb.ugent.be/webtools/Venn) to select common biomarkers among subgroups. We overlaid biomarker signatures for comparative analysis of time-dependent changes of biomarker sets observed after vaccination.

Data analysis demonstrated that primary vaccination triggered significant levels of 17DD-YF–specific neutralizing antibodies (p<0.0001), reaching a seropositivity rate of 96% (Figure 2). Of note, we observed progressive decrease in PRNT levels (p<0.0001) and in the PRNT seropositivity rates along the time compared with PV(d30–45). The seropositivity rate declined to ≈71% by 10 years after primary vaccination (Figure 2, panel B). Booster vaccination significantly increased the PRNT levels (p<0.0001) and raised the seropositivity rate from 69% to 100% (Figure 2, panels A, B). The secondary vaccination was accompanied by higher seropositivity rate after 10 years upon booster dose, regardless of the decrease in PRNT levels (p<0.0001) observed over time compared with #PV(d30–45) (Figure 2, panel B).

Comparative analysis of NV(d0) versus PV(d30–45) after 1 or 2 doses of 17DD-YF vaccine demonstrated that primary vaccination is followed by an increase of memory T cells, including eEfCD4 (p<0.05), EMCD4 (p<0.05), and EMCD8 (p = 0.0006), and all B-cell subsets evaluated, NCD19 (p = 0.01), nCMCD19 (p = 0.001), and CMCD19 (p = 0.001). We also observed a decrease of NCD8 (p = 0.02), eEfCD8 (p = 0.02), and CMCD8 (p = 0.001). The results showed that cellular immunity, eEfCD4 (p = 0.01), EMCD4 (p = 0.01), and EMCD8 (p = 0.009); and all B-cell subsets, NCD19 (p = 0.003), nCMCD19 (p = 0.02), and CMCD19 (p = 0.0002), clearly wane over 10 years, compared with 30–45 days after primary vaccination (Figure 3).

Secondary vaccination with 17DD-YF was able not only to increase the level of memory T-cell subsets, eEfCD4 (p<0.05), EMCD4 (p<0.05), and EMCD8 (p = 0.04) at 30–45 days after the second dose but also to sustain the maintenance of eEfCD4 and EMCD8 levels even after >10 years upon booster vaccination compared with 30–45 days after secondary vaccination. We observed no substantial changes in memory B-cell subsets upon secondary vaccination (Figure 3).

Data analysis for in vitro 17DD-YF antigen recall revealed that primary vaccination induced significant increases of functional memory biomarkers in CD4+ (tumor necrosis factor [TNF]–α/p = 0.04, IFN-γ/p = 0.04 and IL-5/p = 0.0001) and CD8+ T cells (TNF-α/p = 0.0003, IFN-γ/p = 0.008 and IL-5/p = 0.0002) as well as in B cells (TNF-α/p<0.05 and IL-5/p = 0.016) (Figure 4). We observed a decrease of IL-10+CD4+ T cells (p = 0.04) at 30–45 days after primary vaccination and a clear decrease of TNF-α, IFN-γ, and IL-5 produced by CD4+ and CD8+ T cells. We also saw a decrease in TNF-α and IL-5 from B cells over time, particularly at >10 years, compared with 30–45 days after primary vaccination (p<0.05 in all cases). Conversely, we detected an increase of IL-10+ B cells over time after primary vaccination (Figure 4).

Secondary vaccination was able to restore T cell functional memory mediated by IFN-γ and B-cell functional memory by TNF-α. We observed sustained production of IFN-γ by CD8+ T cells even >10 years after booster vaccination (Figure 4).

We constructed overlays of biomarker signatures of NV(d0) versus PV(d30–45) in the primary-vaccination arm of the study as well as #PV(y>10) versus RV(d30–45) in the secondary-vaccination arm to select those attributes eligible as universal memory-related biomarkers (Appendix Figure 1). Venn diagram analysis revealed 3 common attributes (EMCD4, EMCD8, and IFNCD8) that we tagged for follow-up analysis after primary, secondary, or multiple 17DD-YF vaccination (Appendix Figure 1).

After we selected the universal set of follow-up attributes, we compared biomarker signatures over time after 17DD-YF primary or secondary vaccination (Figure 5). Data analysis demonstrated that all 3 biomarkers were observed in PV(d30–45) and PV(y1–5). EMCD8 was maintained in PV(y>5–9) but not in PV(y>10). Comparative analysis between #PV(y>10) and RV(d30–45) demonstrated restoration of universal memory-related biomarkers (EMCD4, EMCD8, and IFNCD8) upon revaccination. Moreover, we identified EMCD8 in a high proportion of secondary vaccinees across the time periods after booster dose (Figure 5).

We used individual PRNT and EMCD8 profiles to assemble a memory matrix and calculated the resultant YF-specific memory. We classified each volunteer by positive results above the cutoff threshold as EMCD8, PRNT, none, or both (Figure 6). Data analysis demonstrated that primary vaccination leads to a resultant memory in 97% of volunteers, with 3% of failure in the PV(d30–45) group. However, an increase in the proportion of primary vaccinees with “none” positive attributes, neither PRNT nor EMCD8 biomarkers, was observed, reaching a critical value of 30% at ≥10 years after primary vaccination (Figure 6, panel A).

The comparison between #PV(y>10) and RV(d30–45) demonstrated that secondary vaccination restored the resultant memory in 100% of the volunteers, which was different from primary vaccination results. All secondary vaccines evaluated simultaneously for PRNT and EMCD8 profile presented a preserved resultant memory. In particular, at >10 years after secondary vaccination, 100% of vaccinees had 1 or both biomarkers detectable (Figure 6, panel B).

We analyzed 17DD-YF memory status triggered by multiple vaccinations (Figure 7). The results demonstrated that all vaccinees who received multiple shots had PRNT levels above the cutoff limit (Figure 7, panel A). The proportion of vaccinees with EMCD8 above the cutoff limit was higher for RV(y1–5) and RV2+(y1–5) than for PV(y1–5) (Figure 7, panel B). Likewise, the overall resultant memory for RV(y1–5) and RV2+(y1–5) was higher than that for PV(y1–5) – (Figure 7, panel C).

The volunteers from the #PV(y≥10) group were further categorized into subgroups, PRNT– and PRNT+, according to their PRNT results before revaccination. The levels of humoral and cellular biomarkers, as well as the magnitude of baseline fold changes in neutralizing antibodies, were higher in PRNT– than in PRNT+ vaccines (Figures 8, 9). An increase in PRNT titer by a factor >4 at follow-up further demonstrated the relevance of booster doses to restore the immunological memory of these subjects (Figure 8). Furthermore, the booster dose had higher impact on cellular immunity and biomarker signature of PRNT– than PRNT+ primary vaccinees, and both groups restored the EMCD8 biomarker compared with the #PV(y>10) group (Figure 9).

We analyzed the results to determine whether DENV seropositivity influences the humoral and cellular memory after secondary vaccination (Appendix Figures 2, 3). We found DENV seropositivity in 28% of volunteers in the secondary vaccination arm. The results demonstrated that DENV seropositivity did not influence PRNT seropositivity over time after secondary vaccination (Appendix Figure 2). Moreover, DENV seropositivity did not affect cellular immunity (EMCD8), which remained above the cutoff limit in all subgroups of secondary vaccinees (Appendix Figure 3).