1. Epidemiology and Global Health
Download icon

Genetic epidemiology of dengue viruses in phase III trials of the CYD tetravalent dengue vaccine and implications for efficacy

  1. Maia A Rabaa Is a corresponding author
  2. Yves Girerd-Chambaz
  3. Kien Duong Thi Hue
  4. Trung Vu Tuan
  5. Bridget Wills
  6. Matthew Bonaparte
  7. Diane van der Vliet
  8. Edith Langevin
  9. Margarita Cortes
  10. Betzana Zambrano
  11. Corinne Dunod
  12. Anh Wartel-Tram
  13. Nicholas Jackson
  14. Cameron P Simmons Is a corresponding author
  1. Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Vietnam
  2. University of Oxford, United Kingdom
  3. Sanofi Pasteur, France
  4. Sanofi Pasteur, United States
  5. Sanofi Pasteur, Colombia
  6. Sanofi Pasteur, Uruguay
  7. Sanofi Pasteur, Singapore
  8. Peter Doherty Institute, University of Melbourne, Australia
Research Article
  • Cited 0
  • Views 689
  • Annotations
Cite as: eLife 2017;6:e24196 doi: 10.7554/eLife.24196

Abstract

This study defined the genetic epidemiology of dengue viruses (DENV) in two pivotal phase III trials of the tetravalent dengue vaccine, CYD-TDV, and thereby enabled virus genotype-specific estimates of vaccine efficacy (VE). Envelope gene sequences (n = 661) from 11 DENV genotypes in 10 endemic countries provided a contemporaneous global snapshot of DENV population genetics and revealed high amino acid identity between the E genes of vaccine strains and wild-type viruses from trial participants, including at epitope sites targeted by virus neutralising human monoclonal antibodies. Post-hoc analysis of all CYD14/15 trial participants revealed a statistically significant genotype-level VE association within DENV-4, where efficacy was lowest against genotype I. In subgroup analysis of trial participants age 9–16 years, VE estimates appeared more balanced within each serotype, suggesting that genotype-level heterogeneity may be limited in older children. Post-licensure surveillance is needed to monitor vaccine performance against the backdrop of DENV sequence diversity and evolution.

https://doi.org/10.7554/eLife.24196.001

eLife digest

Each year, about 100 million people—mostly children in tropical parts of Asia and Latin America—are infected with the dengue virus. It has been difficult to produce a vaccine against the virus, because there are four different types of the virus, and people respond to infections with different types in an unusual way. Once a person is infected with one type of dengue, they are protected from future infections with that type. However, if that person later becomes infected with a different type, they are more likely to experience severe illness. As a result, a dengue vaccine must simultaneously protect against all four types of the virus to be safe and effective.

The first dengue vaccine has recently become available. Clinical studies of the vaccine show that it can protect against all four virus types, but that the protection against certain types and in some age groups varies. Complicating matters, the four types of the dengue virus have continued to evolve since scientists first began developing the vaccine. Therefore, scientists are concerned that the vaccine may not be as effective against the newly evolved subtypes.

To find out, scientists would have to carefully compare the genetics of the strains used to develop the vaccine with the strains currently circulating. They would also have to see how well the vaccine protects against current strains.

Now, Rabaa et al. show that there is a high level of genetic similarity between the viruses used to create the vaccine, and dengue viruses that caused infections in people participating in clinical studies of the vaccines. The analyses also showed that in children between the ages of 2 and 16, the vaccine is more effective against one subtype of the dengue type-4, compared to the other circulating subtype. In children between the ages of 9 and 16, who are eligible to receive the vaccine in some countries, the vaccine was largely equally effective across the various subtypes.

In addition to providing reassurance that the vaccine is working against currently circulating types, Rabaa et al. provide a valuable snapshot of the genetic diversity of dengue viruses. This snapshot will help scientists develop more effective dengue vaccines and treatments. More studies following vaccinated people are needed to ensure that the current vaccine remains effective as circulating strains of the virus evolve.

https://doi.org/10.7554/eLife.24196.002

Introduction

Dengue is the commonest arboviral disease of humans and has been a major public health problem in tropical Asia and Latin America for decades (Stanaway et al., 2016). Reducing the population of competent mosquito vectors of dengue viruses has been the central aim of disease control efforts, but these have had little success in eliminating or stopping the spread of dengue globally. Effective dengue vaccines will be essential tools to achieving dengue control. Accordingly, the licensure of the first tetravalent dengue vaccine (chimeric yellow fever–dengue virus tetravalent dengue vaccine (CYD-TDV), Sanofi Pasteur) together with recommendations from The World Health Organisation’s Strategic Advisory Group of Experts (SAGE) on Immunization on its use in highly endemic countries, has provided the first prospects of an integrated public health approach to disease control (WHO, 2016).

Dengue vaccine development has been challenging, in part because dengue viruses (DENV) exist as four phylogenetically and antigenically distinct serotypes (DENV-1 to −4). Within each virus serotype exists considerable genetic diversity at local, national and continental scales (Holmes, 2008). Subtle antigenic differences can also be measured amongst members of the same virus serotype and are speculated to be of epidemiological and clinical importance (Katzelnick et al., 2015). The virus population dynamics of DENV in hyperendemic areas are complex, often involving the emergence and extinction of viral lineages against a backdrop of multiple virus types co-circulating and oscillating in their relative prevalence. Human population immunity and intrinsic virus fitness in mosquitoes and humans are potential drivers of DENV evolution in these settings (Holmes and Burch, 2000). Acting to balance high mutation rates of DENV within individual hosts, the vector-human transmission cycle subjects viral populations to strong purifying selection, whereby emergent virus variants that are less fit for disseminated infection of both humans and mosquitoes are lost from the viral population (Holmes, 2003).

CYD-TDV was found to be safe and efficacious for use in children 9 years of age and older, with efficacy varying according to age, baseline serostatus and virus serotype (Capeding et al., 2014; Villar et al., 2015). Furthermore, a trend toward reduced efficacy against DENV-2 was observed in the Asian phase III trial compared to the Latin American trial (Hadinegoro et al., 2015). This finding suggested that the efficacy of CYD-TDV might be affected by sub-serotype (i.e. genotype) level diversity in DENV populations, often associated with geographical boundaries. Beyond the epidemiological factors identified in previous studies of CYD-TDV efficacy, the performance of dengue vaccines could also be influenced by the evolving nature of DENV populations in endemic settings. For example, the possibility that circulating DENV populations could ‘escape’ vaccine-elicited immune responses was nominated as one of several possible explanations for the relatively low efficacy of CYD-TDV against DENV-2 in a phase IIb trial in Thailand (Sabchareon et al., 2012). Two phase III efficacy trials of CYD-TDV, involving more than 31,000 children between the ages of 2–14 years in the Asia–Pacific region (CYD14 trial) and between the ages of 9–16 years in Latin America (CYD15 trial) (Hadinegoro et al., 2015) enable, for the first time, a post hoc investigation of vaccine efficacy versus DENV population diversity. Thus, the aims of the present study were threefold. First, to document the genetic distance between the components of the CYD-TDV formulation and the DENV strains detected amongst cases in the CYD14 and CYD15 trials. Second, to perform focused analysis of the level of sequence conservation between CYD-TDV vaccine strains and wild-type DENV at epitope locations targeted by potent virus neutralising human monoclonal antibodies (mAbs). Lastly, we aimed to explore if a more complex genotype-specific efficacy pattern existed in the CYD14 and CYD15 trials, notwithstanding the limitations inherent to post hoc analysis. Collectively, these data provide insights into the characteristics of the CYD-TDV product relative to contemporary DENV populations and provide preliminary insight into genotype-level vaccine efficacy that can serve as a baseline for future research.

Results

Acquisition of DENV envelope gene sequences

433 acute serum samples from 595 virologically-confirmed dengue (VCD) cases in CYD14 and 512 samples from 662 VCD cases in CYD15 were selected for investigation on the basis of subject consent, viremia level and sample volume considerations (Figure 1A and B, respectively). From CYD14, 314 complete DENV envelope (E) gene nucleotide sequences (1485 nt for DENV-1,–2, −4; 1479 nt for DENV-3) were acquired directly from 433 serum samples (72.5%, including three mixed infections), with a subset of 299/433 (69.1%) samples also having a complete premembrane (prM) nucleotide sequence. From CYD15, 333 complete DENV E gene nucleotide sequences were acquired directly from 512 serum samples (65.0%, including eight mixed infections), with a subset of 313/512 (61.1%) samples also having a complete prM nucleotide sequence. The proportion of serum samples that yielded an E gene sequence was similar between control and dengue vaccine recipients within each study (Supplementary file 1a). The probability of acquiring an E gene sequence from serum samples was positively associated with the DENV viremia level (Figure 1—figure supplement 1).

Figure 1 with 1 supplement see all
Sequencing flow chart for samples obtained in CYD-TDV trials.

(A) CYD14, (B) CYD15.

https://doi.org/10.7554/eLife.24196.003

Phylogenetic profile of CYD-TDV vaccine strains and DENV detected in CYD14 and CYD15 trials

Full and partial E gene sequences determined directly from serum samples collected in CYD14 and CYD15 trials (253 DENV-1, 191 DENV-2, 107 DENV-3 and 110 DENV-4) were aligned with E gene sequences corresponding to the CYD-TDV vaccine strains and sequences from GenBank for which the year and country of sampling were known. Maximum likelihood trees representing subsampled E gene sequence datasets allowed the classification of CYD14/15 viruses to the major intra-serotype lineages (genotypes) previously described for DENV (Figure 2—figure supplements 14). At the country level, CYD14/15 viruses were closely related to publicly available DENV sequences acquired from the same country, an indicator of ongoing local evolution. Figure 2 shows the genotypes detected in the CYD14/15 virus populations according to their country of sampling. Collectively, these data define the population genetics of viruses responsible for dengue cases in the CYD14/15 trials and provide a unique contemporaneous snapshot of DENV diversity in ten endemic countries.

Figure 2 with 4 supplements see all
Distribution of DENV serotypes and genotypes sequenced in CYD14 and CYD15 vaccine trials by country.

Numbers in parentheses indicate the total number of samples of each genotype for which complete or partial E gene sequences were obtained.

https://doi.org/10.7554/eLife.24196.009

Sequence differences between CYD-TDV vaccine strains and circulating wild-type viruses

We quantified the differences between the E gene amino acid sequences in the components of the tetravalent CYD-TDV formulation and viruses from VCD cases in the CYD14 and CYD15 trials. The mean level of E gene amino acid sequence difference between vaccine strains and viruses from VCD cases in CYD14 and CYD15 was <3% for all serotypes (Figure 3 and Supplementary file 1b). To define the nature of these sequence differences, the amino acid positions that varied between CYD-TDV vaccine strains and the E gene sequences sampled in CYD14/15 trials and in the subsampled GenBank sequences were annotated adjacent to the subsampled maximum likelihood phylogenetic trees for each serotype. The DENV-2 E gene phylogeny (incorporating the vaccine strain) of relevance to the CYD14 trial is shown in Figure 4A and for CYD15 in Figure 4B. The equivalent annotated phylogenies for DENV-1,–3 and −4 are shown in Figure 4—figure supplements 16. These data reveal that positions of amino acid non-identity between CYD-TDV vaccine strains and wild-type viruses were dispersed across the E protein and do not cluster to any particular structural domain.

Figure 3 with 1 supplement see all
Average genotype-specific amino acid identity of DENV isolated in CYD-TDV trials compared to the vaccine strain of the corresponding DENV serotype.

Black bars indicate the IQR of the full sample set. Coloured dots show the geographic regions from which each genotype was collected – red: CYD14, maritime SE Asia; blue: CYD14, mainland SE Asia; grey: CYD15, Americas. Black dots indicate the genotype of the serotype-specific CYD-TDV vaccine component.

https://doi.org/10.7554/eLife.24196.014
Figure 4 with 6 supplements see all
Amino acid differences between the DENV-2 E gene vaccine sequence, DENV-2 viruses isolated in CYD14 and CYD15 vaccine trials, and representative subsets of publically available DENV-2 sequences from the vaccine trial sites.

(A) CYD14 DENV-2 phylogeny, (B) CYD15 DENV-2 phylogeny. Coloured tips on the trees show sequences isolated in the CYD-TDV trials (country of origin coloured as indicated in the key) and the vaccine sequence (purple star); grey tips indicate publicly available sequences isolated from other studies in the countries of interest. Columns to the right indicate amino acid sites at which variation was observed in two or more CYD14/CYD15 sequences. Numbers at the top of columns indicate the amino acid site within the E gene. Bars at the top of the figures indicate the E gene domain of the site. Amino acids at variable sites in the E gene sequence of the vaccine component are shown in colour. For all other sequences, a lack of colour indicates an amino acid identical to that of the vaccine component at that site.

https://doi.org/10.7554/eLife.24196.016

Human mAb epitope sequences in vaccine and wild-type viruses

We examined amino acid sequence identity between vaccine strains and wild-type CYD14/15 virus sequences at twelve B cell epitopes. The twelve epitopes represent some of the best structurally defined epitopes in DENV that are targeted by potent virus neutralising human mAbs and are thus of particular interest in vaccine development and immune correlate assays (Fibriansah et al., 2014; Cockburn et al., 2012a; Smith et al., 2013; Fibriansah et al., 2015a, 2015b; Teoh et al., 2012; Rouvinski et al., 2015; Costin et al., 2013; Cockburn et al., 2012b). Sequence analyses indicated limited variation at these epitope regions in the CYD14/15 sequences, as well as in a global database of wild-type virus sequences (Figure 5 and Figure 5—figure supplement 1). The conservation of these epitope sequences between the decades-old ‘donor’ viruses from which the CYD-TDV product was derived and contemporary virus populations suggests that these amino acid sites are not highly prone to evolutionary drift.

Figure 5 with 1 supplement see all
Sequence conservation between the DENV-2 vaccine component and wild-type DENV-2 viruses at epitope locations targeted by virus neutralising human mAbs.

Amino acid targets for five neutralising human mAbs (Fibriansah et al., 2015b; Rouvinski et al., 2015) are coloured as indicated in the key (top) and compared to the vaccine sequence and wild-type sequences obtained within the CYD14 and CYD15 trials (middle), as well as complete E gene sequences from wild-type DENV-2 available on GenBank (bottom). Sites are indicated at the top of columns. For wild-type virus populations, the darker the block, the greater the proportion of sequences with an amino acid differing from the target amino acid at that site. When disagreement between amino acids was observed between epitope targets (as at E67 and E71), wild-type sequences were compared to 2D22 as a reference, denoted by an asterisk.

https://doi.org/10.7554/eLife.24196.023

Vaccine efficacy by DENV serotype and genotype

Given the high degree of overall amino acid sequence identity, including at key epitope positions, between the E protein found in CYD-TDV vaccine strains and contemporary wild-type CYD14/15 viruses, we postulated that vaccine efficacy would be largely independent of virus genotype. We report two levels of intention to treat genotype-level efficacy from the CYD14 and CYD15 trials: the observed estimates and the observed+imputed estimates. The observed estimate refers to vaccine efficacy in the population of VCD cases who had serum samples yielding an E gene sequence that was empirically assigned a genotype. The observed+imputed estimates used the observed genotype data plus imputation to give genotype assignments to VCD cases where the serotype was known but genotype information was absent. Imputation was likely to be accurate because data from this study (Figure 2) indicated eight out of the ten study countries had only a single genotype of each serotype in circulation during the study period. Publicly available sequence data largely mirror the genotype distributions observed in this study; greater diversity is found in some Asian countries relative to those detected in this study, likely because the publicly available sequences are collated at the country level, whereas the CYD14/15 sequences represent those circulating only within the geographically limited trial populations (Supplementary file 1c). The count of observed and imputed genotypes is summarised in Supplementary file 1d.

Estimates of genotype-level vaccine efficacy amongst the observed and observed+imputed case populations are described in Table 1 (all ages) and Table 2 (participants 9–16 years of age). For completeness, we also show the observed genotype-level vaccine efficacy for participants < 9 years of age in Supplementary file 2 but do not consider it in the main analyses because this age-class was only present in the CYD14 trial and is below the age for which the licensed vaccine is now indicated (i.e. ≥ 9 years) (WHO, 2016). For each serotype, a Cox proportional hazards regression model (expressing the hazard function) was used to estimate vaccine efficacy (derived as 100* [1- Hazard Ratio]) with vaccine group, genotype and the interaction between vaccine group and genotype included as covariates. The parameter estimates and the 95% confidence intervals of the interactions are given in Table 3 (all ages) and Table 4 (participants 9–16 years of age).

Table 1
Observed and imputed efficacy of CYD-TDV in all participants who received ≥1 injection (intention to treat) by serotype and genotype.
https://doi.org/10.7554/eLife.24196.025

CYD dengue vaccine groupControl groupVaccine efficacy
Observed
Vaccine Efficacy with imputation for missing genotype data
CasesPerson-years at riskDensity incidence (95% CI)CasesPerson-years at riskDensity incidence (95% CI)%(95% CI)%(95% CI)
Serotype 163.1(52.7; 71.2)54.7(45.4; 62.3)
Genotype I CYD14CYD15137420.1 (0.1; 0.2)1867960.3 (0.2; 0.4)58.8(18.3; 79.5)57.4(29.7; 74.2)
Genotype IV CYD1440137420.3 (0.2; 0.4)5167960.8 (0.6; 1.0)61.3(41.5; 74.5)53.3(37.2; 65.3)
Genotype V CYD1553270160.2 (0.1; 0.3)76134340.6 (0.4; 0.7)65.3(50.9; 75.7)54.9(40.7; 65.6)
p-value*0.86140.9912
Serotype 239.1(18.9; 54.3)43.0(29.4; 53.9)
American/Asian CYD1548270350.2 (0.1; 0.2)50134610.4 (0.3; 0.5)52.2(28.9; 67.9)50.2(32.6; 63.2)
Asian I CYD14CYD28137660.2 (0.1; 0.3)1468560.2 (0.1; 0.3)0.3(−94.9; 46.6)19.8(−30.0; 49.6)
Cosmopolitan CYD1428137660.2 (0.1; 0.3)2168560.3 (0.2; 0.5)33.8(−18.0; 62.2)43.8(16.1; 62.2)
p-value*0.14690.2493
Serotype 375.1(62.9; 83.3)71.6(63.0; 78.3)
Genotype I CYD14913835<0.1 (0.0; 0.1)1468950.2 (0.1; 0.3)67.9(26.9; 86.6)58.1(25.2; 76.8)
Genotype II CYD14CYD0138350.0 (0.0; 0.0)46895<0.1 (0.0; 0.1)100.0(69.3; 100.0)85.8(41.1; 97.9)
Genotype III CYD14413835<0.1 (0.0; 0.1)768950.1 (0.0; 0.2)71.6(6.1; 92.6)68.4(19.8; 88.4)
Genotype III CYD152327060<0.1 (0.1; 0.1)47134590.3 (0.3; 0.5)75.7(60.5; 85.5)74.2(64.3; 81.4)
Genotype III CYD14 + CYD152740896<0.1 (0.0; 0.1)54203540.3 (0.2; 0.3)75.2(61.0; 84.6)73.7(64.3; 80.8)
p-value*0.37510.2561
Serotype 474.1(61.7; 82.5)76.9(69.5; 82.6)
Genotype I CYD1419138260.1 (0.1; 0.2)1868740.3 (0.2; 0.4)47.4(−0.9; 72.5)58.3(29.9; 75.2)
Genotype II CYD14CYD813826<0.1 (0.0; 0.1)2468740.3 (0.2; 0.5)83.5(64.8; 93.1)83.8(69.3; 91.5)
Genotype II CYD15CYD1127063<0.1 (0.0; 0.1)31134420.2 (0.2; 0.3)82.4(66.0; 91.5)80.8(71.2; 87.3)
Genotype II CYD14 + CYD15CYD1940890<0.1 (0.0; 0.1)55203160.3 (0.2; 0.4)82.9(71.7; 90.1)81.8(74.3; 87.1)
p-value*0.00720.0086
  1. Cases: number of subjects with at least one sequenced symptomatic virologically-confirmed dengue episode during the active phase of follow-up.

    Density incidence: data indicate cases per 100 person-years at risk.

  2. *The p-value was obtained by testing the heterogeneity of genotype distribution between groups (within each serotype) using a Chi2 (or Fisher’s exact test).

    CYD Genotype of the serotype-specific CYD-TDV vaccine component.

Table 2
Observed and imputed efficacy of CYD-TDV for subjects 9 years and older who received ≥1 injection (intention to treat) by serotype and genotype
https://doi.org/10.7554/eLife.24196.026

CYD dengue vaccine groupControl groupVaccine efficacy
Observed
Vaccine Efficacy with imputation for missing genotype data
CasesPerson-years at riskDensity incidence (95% CI)CasesPerson-years at riskDensity incidence (95% CI)%(95% CI)%(95% CI)
Serotype 167.7(56.1; 76.3)58.4(47.7; 66.9)
Genotype I CYD14CYD66683<0.1 (0.0; 0.2)833060.2 (0.1; 0.5)62.8(−6.8; 87.8)69.0(33.8; 85.5)
Genotype IV CYD14866830.1 (0.1; 0.2)1933060.6 (0.3; 0.9)79.2(54.1; 91.4)64.0(39.7; 78.5)
Genotype V CYD1553270160.2 (0.1; 0.3)76134340.6 (0.4; 0.7)65.3(50.9; 75.7)54.9(40.7; 65.6)
p-value*0.52130.5400
Serotype 248.6(27.4; 63.7)47.1(31.3; 59.2)
American/Asian CYD1548270350.2 (0.1; 0.2)50134610.4 (0.3; 0.5)52.2(28.9; 67.9)50.2(32.6; 63.2)
Asian I CYD14CYD1266870.2 (0.1; 0.3)933300.3 (0.1; 0.5)33.6(−62.7; 71.9)34.6(−27.4; 65.7)
Cosmopolitan CYD1456687<0.1 (0.0; 0.2)433300.1 (0.0; 0.3)37.8(−151; 83.5)40.3(−41.4; 74.3)
p-value*0.77360.7253
Serotype 376.0(62.3; 84.7)73.6(64.4; 80.4)
Genotype I CYD1446715<0.1 (0.0; 0.2)633470.2 (0.1; 0.4)66.8(−16.3; 91.5)61.2(−4.1; 86.1)
Genotype II CYD14CYD067150.0 (0.0; 0.1)33347<0.1 (0.0; 0.3)100.0(55.4; 100.0)80.1(7.6; 97.1)
Genotype III CYD1416715<0.1 (0.0; 0.1)23347<0.1 (0.0; 0.2)75.1(−160; 98.8)75.1(−27.4; 96.6)
Genotype III CYD152327060<0.1 (0.1; 0.1)47134590.3 (0.3; 0.5)75.7(60.5; 85.5)74.2(64.3; 81.4)
Genotype III CYD14 + CYD152433775<0.1 (0.0; 0.1)49168060.3 (0.2; 0.4)75.7(60.8; 85.3)74.3(64.7; 81.4)
p-value*0.59280.6985
Serotype 485.2(74.6; 91.4)83.2(76.2; 88.2)
Genotype I CYD1436716<0.1 (0.0; 0.1)1233270.4 (0.2; 0.6)87.6(60.9; 97.2)86.2(63.6; 94.8)
Genotype II CYD14CYD36716<0.1 (0.0; 0.1)1433270.4 (0.2; 0.7)89.4(67.7; 97.6)89.6(70.5; 96.3)
Genotype II CYD15CYD1127063<0.1 (0.0; 0.1)31134420.2 (0.2; 0.3)82.4(66.0; 91.5)80.8(71.2; 87.3)
Genotype II CYD14 + CYD15CYD1433779<0.1 (0.0; 0.1)45167690.3 (0.2; 0.4)84.6(72.6; 91.8)82.6(74.7; 88.1)
p-value*1.00000.6678
  1. Cases: number of subjects with at least one sequenced symptomatic virologically-confirmed dengue episode during the active phase of follow-up.

    Density incidence: data indicate cases per 100 person-years at risk.

  2. *The p-value was obtained by testing the heterogeneity of genotype distribution between groups (within each serotype) using a Chi2 (or Fisher’s exact test).

    CYD Genotype of the serotype-specific CYD-TDV vaccine component.

Table 3
Estimation of the interaction between genotype and vaccine group for symptomatic VCD detected during the active phase of follow-up by serotype in all participants who received >= 1 injection (intention to treat) (CYD14/CYD15).

The estimate of the interaction term between genotype and vaccine group is derived from Cox proportional hazards regression models including the vaccine group, the genotype and the interaction.

https://doi.org/10.7554/eLife.24196.027
Estimated interaction with observed vaccine efficacyEstimated interaction with vaccine efficacy with imputation
SerotypeParameterParameter estimate95%Parameter estimate95%
 Serotype 1Genotype IV vs Genotype I−0.058[−0.858; 0.743]0.095[−0.475; 0.665]
Genotype V vs Genotype I−0.167[−0.936; 0.603]0.067[−0.492; 0.625]
 Serotype 2American/Asian vs Asian I−0.732[−1.486; 0.022]−0.471[−1.032; 0.089]
Cosmopolitan vs Asian I−0.404[−1.259; 0.451]−0.344[−0.966; 0.267]
 Serotype 3Genotype II vs Genotype I−12.748[−729.203; 703.707]−1.079[−2.754; 0.596]
Genotype III vs Genotype I−0.251[−1.208; 0.705]−0.459[−1.116; 0.198]
 Serotype 4Genotype II vs Genotype I−1.114[−1.943; −0.285]−0.8184[−1.434; −0.203]
Table 4
Estimation of the interaction between genotype and vaccine group for symptomatic VCD detected during the active phase of follow-up by serotype in subjects older than 9 years of age who received >= 1 injection (intention to treat) (CYD14/CYD15).

The estimate of the interaction term between genotype and vaccine group is derived from Cox proportional hazards regression models including the vaccine group, the genotype and the interaction.

https://doi.org/10.7554/eLife.24196.028
Estimated interaction with observed vaccine efficacyEstimated interaction with vaccine efficacy with imputation
SerotypeParameterParameter estimate95%Parameter estimate95%
 Serotype 1Genotype IV vs Genotype I−0.574[−1.917; 0.768]0.153[−0.760; 1.066]
Genotype V vs Genotype I−0.061[−1.177; 1.054]0.385[−0.416; 1.186]
 Serotype 2American/Asian vs Asian I−0.327[−1.277; 0.624]−0.270[−0.987; 0.448]
Cosmopolitan vs Asian I−0.064[−1.637; 1.510]−0.089[−1.151; 0.972]
 Serotype 3Genotype II vs Genotype I−13.019[−943.634; 917.597]−0.664[−2.578; 1.250]
Genotype III vs Genotype I−0.309[−1.665; 1.047]−0.405[−1.443; 0.633]
 Serotype 4Genotype II vs Genotype I0.226[−1.174; 1.626]0.244[−0.789; 1.277]

For DENV-1, vaccine efficacy estimates against the three different genotypes were highly similar in the all ages group and in participants 9–16 years of age (Tables 1 and 2). Additionally, the genotype interaction parameter estimates in the all ages group (Table 3) were close to zero and had reasonably tight 95% confidence bounds. This suggests it is unlikely that an interaction exists between genotype and vaccine efficacy, but if such an interaction does exist, it is small. Amongst participants 9–16 years of age, the interaction parameter estimates had 95% confidence intervals that bounded zero and were wider than the all ages group, making conclusions relatively difficult to draw.

For DENV-2, vaccine efficacy estimates against the American-Asian genotype (50.2%; 95% CI: 32.6–63.2%) and the Cosmopolitan genotype (43.8%; 95% CI: 16.1–62.2%) were similar, and both were higher than against the Asian I genotype (19.8%; 95% CI: −30.0–49.6%) in the all ages group (Table 1). However, the genotype interaction estimates had 95% confidence intervals that, although reasonably tight, included zero in the all ages group (Table 3) and in participants ≥ 9 years (Table 4). We note however that the upper bound of the confidence interval was very close to zero for the Asian/American versus Asian I genotype interaction (all ages, Table 3), leaving open the possibility that true heterogeneity may exist.

Within DENV-3, the confidence intervals for the interaction estimates were very wide when comparing genotype II versus genotype I in the all ages population (Table 3) and in participants ≥ 9 years (Table 4), and thus no conclusions could be drawn from these data. For DENV-3 genotype III versus genotype I, the 95% confidence intervals around the interaction estimates (Tables 3 and 4) were much tighter but nonetheless passed through zero. This suggested an interaction between genotype and vaccine efficacy remained possible but unlikely, and that more data would be needed to address this question.

Against DENV-4, vaccine efficacy was significantly lower against genotype I (58.3%, 95% CI: 29.9–75.2%), which circulates endemically only in Asia, compared to the globally distributed genotype II (81.8%, 95% CI: 74.3–87.1%, across CYD14/CYD15) in the all ages population (p=0.009) (Table 1). Confidence intervals around estimates of the interaction between genotype I and genotype II and vaccine group exclude zero, consistent with a lower efficacy against genotype I relative to genotype II (Table 3). However, when efficacy against DENV-4 genotype I versus genotype II was considered only in participants ≥ 9 years, efficacy was similar between genotypes (Table 2) and confidence intervals for interaction estimates included zero (Table 4). An important caveat is that relatively wide 95% confidence intervals for all interaction estimates amongst participants ≥ 9 years suggests limited power to detect differences in vaccine efficacy (Table 4), i.e. this study was generally underpowered to assess heterogeneity in this age subgroup analysis.

Genotype-specific vaccine efficacy versus amino acid identity

A visualization of genotype-specific vaccine efficacy versus amino acid identity of trial viruses to CYD-TDV components is shown in Figure 3—figure supplement 1. These data illustrate the absence of a direct relationship between vaccine efficacy and genetic similarity between wild-type and vaccine strains of DENV.

Discussion

The pivotal Phase III efficacy trials of CYD-TDV and longer term follow-up have revealed the complex efficacy profile of this vaccine (Capeding et al., 2014; Hadinegoro et al., 2015; L'Azou et al., 2016). These trials also highlighted generic challenges in dengue vaccine development, e.g. the goal of balanced immunity to four DENV types, achieving efficacy in naïve and partially immune populations, and the need for long-term safety evaluation. A potential additional layer of complexity stems from the ongoing evolution of DENV populations in endemic countries and whether vaccines derived from viruses that circulated decades ago are ‘fit for purpose’ as immunogens against contemporary virus populations. Here, we demonstrate that the DENV E protein components in the CYD-TDV formulation shared high-level amino acid sequence identity, including at prominent B cell epitopes targeted by virus neutralising human mAbs, with viruses sampled in the CYD14 and CYD15 trials. Additionally, within the constraints of the available sample size and confounding factors discussed later, we found limited statistical evidence of genotype-specific differences in the efficacy profile.

With a total of 253 DENV-1, 191 DENV-2, 107 DENV-3 and 110 DENV-4 E gene sequences generated from CYD14 and CYD15, this study provides a contemporary characterization of DENV population genetics in ten highly endemic countries. Across all four serotypes, the E gene phylogenies positioned the CYD14 and CYD15 sequences together with those from geographically identical locations, consistent with long-term endemic circulation of a single virus genotype within each viral serotype in nearly all locations. Additionally, the phylogeographical profile demonstrates within-region sharing of virus genotypes but little inter-regional mixing (distinct viral population profiles can be distinguished within three major geographical categories: mainland Southeast Asia, maritime Southeast Asia, the Americas); only DENV-4 genotype II was found in multiple countries in both regions. The number of genotypes, and hence genetic diversity, detected for any given serotype was greater in Southeast Asia than in Latin America, as expected given the long history of hyperendemicity, within-serotype diversity, and high forces of infection in Southeast Asia (Halstead, 2006; Holmes, 2009; Rodríguez-Barraquer et al., 2014; Imai et al., 2015). Collectively, these data are informative for vaccine and drug development and design of molecular diagnostics. They also serve as virus population baseline profiles from which to monitor DENV evolution in countries where a selective pressure such as CYD-TDV might be widely introduced.

Indonesia, with two DENV-1 genotypes, and the Philippines, with two DENV-4 genotypes, were the only countries with genotype co-circulation detected within the trial. As each of these viral populations appears closely related to previously sequenced viruses from their respective country, these likely represent true local circulation in the population rather than recent importations of novel viruses. In general, however, long-term co-circulation of multiple genotypes within a single serotype in one location is rare and these viral populations may instead represent a cross-section of the viral populations during a process of genotype replacement, in which an endemic viral population is rapidly replaced (often completely) by a novel population imported from another geographical region (Zhang et al., 2005; Lambrechts et al., 2012; Loroño-Pino et al., 2004). This process may also be responsible for the presence of DENV-3 genotype III in Thailand, while neighboring Vietnam harbors only genotype II viruses. Genotype II was the dominant DENV-3 lineage in Thailand and mainland Southeast Asia from the early 1980s until at least 2010 (Rabaa et al., 2013), but was not detected in Thailand in this study. Recent reports of DENV-3 genotype III infections elsewhere in mainland Asia suggest that this lineage may be moving through the region, potentially replacing genotype II (Lao et al., 2014; Jiang et al., 2012). Ongoing virological surveillance in the CYD14 and CYD15 trial populations, as well as populations vaccinated post-licensure, will be used to further investigate the relationships between the vaccine, virus evolution and local DENV genotype variation.

Antigenic differences between viruses of the same serotype have been observed with neutralizing polyclonal and monoclonal antibodies in laboratory assays, and these differences have been postulated to be relevant to clinical epidemiology and vaccine development (Katzelnick et al., 2015; Kochel et al., 2002; OhAinle et al., 2011). Arguing against a critical role for within-serotype sequence diversity is the common acceptance that natural infection with one serotype elicits life-long clinical immunity to that serotype in the vast majority of instances (Waggoner et al., 2016). With respect to CYD-TDV, immunization of non-human primates elicited antibodies that neutralised geographically, phylogenetically and clinically diverse DENV serotypes and genotypes (Barban et al., 2012). That CYD-TDV immunisation induced similar measured levels of efficacy against all genotypes of DENV-1 is supportive of the concept that clinical immunity elicited by CYD-TDV to this serotype was pan-genotype in nature (Tables 1 and 2). Analyses also suggest potential pan-genotype immunity in the case of DENV-3, although the relatively low numbers of DENV-3 cases detected within the CYD14 trials in Southeast Asia result in this study being underpowered to assess potential heterogeneity. Further research is warranted to understand if smaller differences exist in genotype-level vaccine efficacy than could be measured in these trials and that might be relevant to programmatic use of vaccine.

In DENV-4, efficacy against genotype I (found in Asia) in the all ages population was significantly lower than that against genotype II in both Asia and the Americas. Subgroup analyses of genotype and age-stratified vaccine efficacy are inevitably speculative because of diminishing sample sizes and wide confidence intervals around the interaction estimates. Nonetheless we observed that in participants 9–16 years of age, the age group eligible for the licensed vaccine, the vaccine efficacy point estimates were similarly high (>80%) between DENV-4 genotypes.

DENV-2 is of particular interest because CYD-TDV efficacy is lowest against this serotype. In a previous, single-centre phase IIb trial (CYD23) in Thai children (Sabchareon et al., 2012), where all the circulating DENV-2 viruses belonged to the Asian I genotype, efficacy was just 3.5% (-59.8; 40.5) against this serotype/genotype. In CYD14/15, with a larger sample size, efficacy against the DENV-2 Asian I genotype in the all ages population (19.8% (-30.0; 49.6)) was lower, but not significantly so, than that against other DENV-2 genotypes. Although heterogeneity could not be confirmed, further analysis of the interaction between genotype and vaccine group suggested a potentially decreased efficacy profile of the DENV-2 Asian I genotype in the all ages population compared to the American/Asian genotype, which currently circulates only in the Americas. As above for DENV-4, it is speculative to examine subgroups, but for the age group 9–16 years of age, the age group eligible for the now licensed vaccine, the efficacy against DENV-2 Asian I genotype (34.6% (-27.4; 65.7) was comparable to that seen against the other DENV-2 genotypes (Asian/American and Cosmopolitan), albeit with inevitably wide confidence intervals around the point estimates. The basis for reduced efficacy against DENV-2 Asian I genotype and DENV-4 genotype I in the all ages population could be complex and linked to uncharacterised differences in how vaccine-elicited immunity acts on these virus genotypes. We note that DENV-2 Asian I genotype and DENV-4 genotype I viruses were only detected in Asia (CYD14), where younger trial participants were included compared to Latin America (CYD15). While a sole impact of age in vaccine-elicited immunity may account for a proportion of this difference, high DENV diversity within Asia resulted in the detection of additional DENV-2 and DENV-4 genotypes within the CYD14 study (DENV-2 Cosmopolitan genotype and DENV-4 genotype II), against which no evidence of decreased vaccine efficacy was shown. It will be of interest to monitor efficacy against DENV-2 and DENV-4 genotypes in post-marketing effectiveness studies of CYD-TDV. Interestingly, while the parental strain of the CYD DENV-2 component is in fact based on an historical Asian I strain, contemporary DENV-2 Asian I populations diverge from the parental CYD strain at multiple amino acid residues, some of which are postulated to increase transmission fitness in some circumstances (Vu et al., 2010). Additional investigations, which might include animal model studies coupled with virological surveillance in the post-vaccine licensure period, could assist further understanding and precision of estimates of vaccine efficacy against different genotypes of DENV-2.

Examination of amino acid sequences among circulating viruses, vaccine components, and epitope sequences targeted by potent, virus neutralising human mAbs provides a framework to predict and possibly understand genotype-specific vaccine performance. These analyses generally underscore the similarity between vaccine components and circulating viruses. Where there were differences at epitope sequence locations, we did not observe a measurable effect in the genotype-level vaccine efficacy. For example, mismatches between circulating DENV-1 and the vaccine component at IF4 epitope sites (Fibriansah et al., 2014) (sites E155, E161, and E171; Figure 5—figure supplement 1) are present across individual genotypes, yet the point estimates of genotype-specific vaccine efficacy were not measurably different. In DENV-4, there was evidence of lower vaccine efficacy against genotype I viruses and also amino acid mismatches between the vaccine component and genotype I virus sequences at known 5H2 epitope positions (E155 and E160; Figure 5—figure supplement 1) (Cockburn et al., 2012a). Further research will be needed to understand the significance of these differences for clinical immunity.

Several mismatches between circulating DENV-2 viruses and the vaccine component were observed at important epitope sites (Figure 5), but these were shared by two circulating genotypes in most cases (sites E71, E149 and E226). Comparing wild-type DENV-2 sequences sampled in CYD14/CYD15 to the CYD-TDV vaccine component, only site E83 showed a mismatch in a high proportion of the Asian I population alone. Sequence data obtained from GenBank confirm that, while there is some variability at this site among all contemporaneous DENV-2 lineages, the Asian I lineage is defined by this amino acid difference at E83. An important caveat to these sequence comparisons is that amino acid differences between vaccine strains and wild-type viruses at known B cell epitopes does not necessarily imply that antigenicity (or immunogenicity) is altered – functional assays of antibody binding will be needed for this.

Our study had several limitations. These were post hoc analyses and, inevitably, sample sizes became small for genotype-level vaccine efficacy estimates and in particular the age-class (9–16 year olds) subgroup analysis. This manifests as wide 95% confidence intervals around the genotype-level point estimates of vaccine efficacy. We generally did not obtain E gene sequences from VCD cases with low viremia and, hence, whether some rare genotypes are not represented in the population of E gene sequences is unknown. In countries where only a single genotype was detected, it was assumed that no undetected genotypes were circulating concurrently in the same location, while publicly available data indicate greater diversity in some Asian countries during this period than detected in this study (Serotypes 1 and 3; Supplementary file 1c). This assumption may have affected the imputed estimates of vaccine efficacy within CYD14, but because vaccine efficacy was ultimately estimated across the entire study population rather than at the country level, the impact of this assumption is expected to be limited. Baseline serostatus is an important determinant of the efficacy profile of CYD-TDV (Hadinegoro et al., 2015). It is possible that some of the genotype-level efficacy results are confounded by the baseline serostatus of vaccine recipients in particular countries, but because only 10% of all participants were characterized immunologically at baseline it is not possible to explore this further (Capeding et al., 2014). Such questions may be addressed in post-licensure research. Finally, our results and conclusions may be specific to this particular vaccine given its unique composition. Other vaccines in development employ different donor viruses and have different compositions and, hence, deserve their own evaluations; in any large-scale trial of a candidate DENV vaccine, continual monitoring will be key to understanding the landscape and evolution of circulating DENV populations and further elucidating the potential relationships between virological factors, vaccine efficacy and post-immunization transmission dynamics. Despite these limitations, the results described here improve the understanding of CYD-TDV vaccine performance. Post-licensure research is needed to further understand the complex profile of this vaccine, and to monitor the impact of vaccination programs on the evolution of DENV populations.

Materials and methods

Sample sets

Briefly, viremic serum samples from VCD cases detected during the active phase of surveillance in CYD14 (Indonesia, Malaysia, the Philippines, Thailand, Vietnam; collected between June 2011 and December 2013) and CYD15 (Brazil, Colombia, Honduras, Mexico, Puerto Rico; collected between June 2011 and April 2014) were eligible for inclusion in this study (Figure 1). Samples with very low viremia or low sample volume that were highly unlikely to be fit for purpose with respect to nucleic acid isolation, amplification and sequencing of prM/E genes were excluded from further investigation, as were samples for which consent was not obtained.

Sequencing methodology

A total of 433 and 512 viremic serum samples were available and subjected to sequencing from CYD14 and CYD15, respectively. Viral RNA was extracted from serum using the MagNA Pure 96 DNA and Viral NA Small Volume Kit (Roche, Mannheim, Germany). PrM and E gene regions were amplified by PCR using 16 different primer pairs, with universal tails at the 5’ end to allow the addition of 454 sequencing-specific nucleotides and isolate-specific multiplex identifiers (MIDs) in a second PCR round, ‘barcode incorporation PCR’. The first PCR round was performed in 20 µl reaction volumes using the FastStart High Fidelity Reaction Kit (Roche) with the addition of 0.25 µM of each PCR primer. The target genes were amplified by PCR in 96 well plates, with the following cycling conditions: denaturation at 94°C for 2 min followed by 40 cycles of PCR, with cycling conditions of 30 s at 94°C, 1 min at 57°C for DENV-1; 60°C for DENV-2; 56°C for DENV-3; 55°C for DENV-4, 60 s at 72°C and 72°C in 5 min for final extension. After PCR, the amplicons were purified using magnetic AMPure XP beads (Agencourt, Woerden, The Netherlands).

Barcode incorporation PCR

The purified first round PCR amplicons were re-amplified to incorporate 454 sequencing-specific nucleotides and isolate-specific MIDs. For this, we used fusion primers that are composed of three parts: 454 sequencing-specific adapter nucleotides, MID sequences and the sequence target of interest on the DNA sample. The second PCR reactions were performed in 10 µl reaction volumes using the FastStart High Fidelity Reaction Kit (Roche) with the addition of 0.1 µM of each fusion primer. The thermo cycling conditions were: denaturation at 94°C for 2 min followed by 30 cycles of PCR, with cycling conditions of 30 s at 94°C, 30 s at 57°C, 60 s at 72°C and 72°C in 5 min for final extension.

Sample pooling

Amplicons spanning the same genomic coordinates, but from different viruses, were pooled. Amplicon pools were measured using Quant-iT PicoGreen dsDNA Assay Kit (Invitrogen, Carlsbad, California) after purification by magnetic AMPure XP beads. In preparation for 454 sequencing, the concentration of the pooled amplicons was adjusted to 106 copies/mL. The purified amplicons were the pooled into one library tube at a concentration of 5.105copies/mL.

Emulsion PCR and 454 sequencing

An emulsion-based clonal amplification (emPCR) was performed according to the manufacturer’s instructions as described in the emPCR Amplification Method Manual - Lib-A, revision June 2010 (Roche). DNA sequencing was performed using the GS Junior Titanium Sequencing Kit and the GS Junior Titanium PicoTiterPlate using the Sequencing Method Manual, revision June 2010 (Roche).

Data analysis

GS Mapping software (Roche) was used for primer trimming and alignment of reads against reference sequences. Briefly, each read per amplicon was mapped to a reference sequence (DENV-1/VN/BID-V2732/2007, GenBank accession number GQ199773.1; DENV-2/VN/BID-V1873/2007, GenBank accession number FJ461321.1; DENV-3/VN/BID-V1933/2008, GenBank accession number KF955460.1; DENV-4/KH/BID-V2055/2002 (GenBank accession number KF955510.1). Sequence quality was high; the Phred scores for E gene sequences are provided in Supplementary file 3. Technical controls were included in all sequencing runs, and showed 100% sequence concordance across the prM and E gene in all cases. CYD-TDV prME sequences from CYD1-CYD4 vaccine components are deposited in GenBank under accession numbers KX239894-KX239897, respectively. All sequences obtained from study subjects are deposited in GenBank under accession numbers KY818060-KY818289, KY851378-KY851758, and KY882502-KY882554.

Sequence analysis

A total of 664 DENV prM and/or E gene sequences were obtained using the above protocol (DENV-1, 253; DENV-2, 191; DENV-3, 108; DENV-4, 112) (Figure 1, Figure 1—source datas 14). Sequences were manually aligned using Geneious (v7.1.7; RRID:SCR_010519) and validated on both nucleotide and amino acid levels. Due to the availability of a larger, less geographically biased public database of E gene sequences compared to prM/E, we focused analysis on the E gene only (1485 nucleotide/495 amino acids for DENV-1,–2, −4; 1479 nucleotide/493 amino acids for DENV-3), excluding three sequences for which only the prM sequence could be obtained. Thus for each serotype, all full and partial E gene sequences (DENV-1, 253; DENV-2, 191; DENV-3, 107; DENV-4, 110) were aligned to large datasets of publicly available E gene sequences from GenBank for which the country and year of sampling are known. Maximum likelihood (ML) phylogenies were inferred for nucleotide sequences using RAxML (v8.0, http://www.exelixis-lab.org/; RRID:SCR_006086) under the GTRGAMMAI model and were visually assessed to determine the genotype of each virus obtained from CYD14 and CYD15. Visual inspection further indicated that all viruses sequenced in this study fell into expected lineages corresponding to previously sampled sequences from the countries from which they were isolated.

To investigate DENV sequences from the CYD14 and CYD15 studies in the context of the viruses circulating in their respective countries and make datasets more tractable, datasets were subsampled to include all CYD14 and CYD15 E gene sequences and up to three randomly selected, publicly available sequences per country per year from the countries involved in this study, along with up to five representative sequences of each known genotype (Goncalvez et al., 2002; Twiddy et al., 2002; Wittke et al., 2002; Patil et al., 2012), regardless of the country from which they were isolated (Total number of taxa used for phylogenetic reconstruction, CYD14: DENV-1, 317; DENV-2, 279; DENV-3, 222; DENV-4, 207. Total number of taxa used for phylogenetic reconstruction, CYD15: DENV-1, 236; DENV-2, 252; DENV-3, 159; DENV-4, 119). Evolutionary models for each dataset were determined using jModeltest (v2.0; RRID:SCR_015244) (Posada, 2009). ML trees were then inferred from these nucleotide sequences using RAxML (v8.0) under the GTRGAMMAI model with 500 bootstrap replications, and p-uncorrected sequence identity (pairwise comparison of genetic differences across the nucleotide and amino acid alignments) was determined using Geneious (v7.1.7). To investigate potentially novel amino acid residues or changes suggestive of selection, all amino acid sites showing a difference between two or more CYD14/CYD15 sequences relative to the vaccine or circulating viruses of the same genotype were mapped to the aforementioned phylogenies using Phandango (https://jameshadfield.github.io/phandango; RRID:SCR_015243). All phylogenies were visualized and annotated using FigTree (v1.4.2; RRID:SCR_008515) and Phandango. Gene annotations were done using the GR7 sequence viewer.

Epitope mapping

To assess the diversity of DENV amino acid sequences and vaccine strains at sites targeted by virus neutralising human mAbs, the serotype-specific E gene alignments used for genotype determination were trimmed to include only sites at which a relevant epitope has previously been identified (Fibriansah et al., 2014; Cockburn et al., 2012a; Smith et al., 2013; Fibriansah et al., 2015a; 2015b; Teoh et al., 2012; Rouvinski et al., 2015; Costin et al., 2013; Cockburn et al., 2012b). For each site, the sequences were compared to vaccine components, strains isolated in CYD14 and CYD15, and publicly available sequences to determine the frequency at which viral sequences matched mAbs targets across all known human DENV lineages.

Statistical determination of genotype-specific vaccine efficacy

Vaccine efficacy against symptomatic VCD cases according to each genotype during the active phase (i.e. from D0 to Month 25) was calculated using the number of cases (i.e., children/adolescents with one or more episodes of VCD) and the person-time at risk in all participants who received at least one injection according to intention to treat. The incidence density was derived as the number of cases per 100 person-years at risk in each group. A Cox regression model was used to estimate vaccine efficacy (derived as 100* [1- Hazard Ratio]) with vaccine group included as a covariate and 95% CI. To further investigate the interaction between vaccine efficacy and genotype, an additional Cox proportional hazards regression model (expressing the hazard function) was used to estimate vaccine efficacy with vaccine group, genotype and the interaction between vaccine group and genotype included as covariates.

VCD cases with missing genotype were imputed using multiple imputation techniques (logistic regression) by serotype with the country included in the model. The twenty imputed (completed) datasets were then analyzed separately and the resultant estimates combined using Rubin’s variance rules and their multivariate generalizations (Rubin, 1987). Analyses were run based on the available data (i.e. no imputed values) and on the imputed data (raw imputation data are shown in Supplementary file 1d).

A Chi² test (or Fisher’s exact test) was used to test the heterogeneity of genotype distribution between vaccine groups. The alternative procedure for pooling chi-square distributed statistics that was proposed by Rubin (Rubin, 1987) and further investigated by Li et al. was used on imputed data (Li et al., 1991). A p-value of less than 0.10 was considered to indicate statistical significance. All statistical analyses were performed using SAS (Version 9.3; RRID:SCR_008567).

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
    The Evolution and Emergence of RNA Viruses
    1. EC Holmes
    (2009)
    Oxford, United Kingdom: Oxford University Press.
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
    Significance levels from repreated p-values with multiply-imputed data
    1. KH Li
    2. XL Meng
    3. TE Raghunathan
    4. DB Rubin
    (1991)
    Statistica Sinica 1:65–92.
  24. 24
    Introduction of the American/Asian genotype of dengue 2 virus into the yucatan state of mexico
    1. MA Loroño-Pino
    2. JA Farfán-Ale
    3. AL Zapata-Peraza
    4. EP Rosado-Paredes
    5. LF Flores-Flores
    6. JE García-Rejón
    7. FJ Díaz
    8. BJ Blitvich
    9. M Andrade-Narváez
    10. E Jiménez-Ríos
    11. CD Blair
    12. KE Olson
    13. W Black
    14. BJ Beaty
    (2004)
    The American journal of tropical medicine and hygiene 71:485–492.
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
    Meeting of the strategic advisory group of experts on immunization, April 2016 – conclusions and recommendations
    1. WHO
    (2016)
    Releve Epidemiologique Hebdomadaire 91:266–284.
  41. 41
  42. 42

Decision letter

  1. Marc Lipsitch
    Reviewing Editor; Harvard TH Chan School of Public Health, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Genetic epidemiology of dengue virus in phase III trials of the CYD tetravalent vaccine and implications for efficacy" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Wendy Garrett as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This further analysis of two dengue vaccine trials reports sequencing and phylogenetic results for viruses that were isolated from vaccinated and control cases in the trials, examines the protein location of the variation along the alignment and with respect to antibody epitopes, and concludes that serotype-specific vaccine efficacy is relatively homogeneous across the genetic variation within each serotype (a conclusion challenged by the reviewers). It provides very interesting data on the relationship of vaccine viruses to circulating viruses and some information on the relationship between this and vaccine efficacy, but important issues remain with the analysis and interpretation.

Essential revisions:

It is eLife's policy to provide one list of changes requested for reconsideration of a revised manuscript. We do that in what follows, and would like point by point responses to the overall decision. Because there is some extra detail in the individual reviewer comments, I have chosen to include them in their entirety for your benefit. I have lifted language from them to inform the main decision comments, so you will see some redundancy. You do not need to write point-by-point responses to each reviewer’s comments.

Overall decision:

We invite a revised resubmission that addresses the following essential points (A-D):

A) The most important theme in the comments from all reviewers is, in brief, that the authors have interpreted an absence of evidence to reject the null hypothesis of homogeneous efficacy against each serotype (across genotypes), as evidence in favor of the null. Statistically, this is invalid, and from a scientific and practical perspective it seems to give an unduly rosy picture of the efficacy of the vaccine and unduly low weight to concerns about evolution to escape the protection of the vaccine. As reviewer 1 writes: "In this manuscript authors Rabaa and colleagues investigate the important issue of whether the CYD14 and CYD15 phase III dengue vaccine trials showed differential vaccine efficacy based on whether circulating viruses "matched" the vaccine product. Generally, I very much appreciated the authors' attention to detail and in-depth presentation of exactly what amino acids differ at what sites between vaccine strains and circulating viruses along with detailed presentation of genotype-by-VE results. However, I felt the authors did not go far enough in their analyses to properly support the primary conclusion of the paper that:

"The finding that CYD-TDV shows relatively consistent efficacy profiles against all genotypes of DENV-1 and -3 suggests that if any antigenic differences exist in these virus populations they are not measurably important to vaccine performance."

The authors’ logic is roughly as follows:

1) Measure overall AA difference between vaccine strains and circulating viruses in the ENV protein.

2) Also measure AA differences at relevant epitope sites between vaccine strains and circulating viruses (where "relevant epitope sites" are defined by monoclonals).

3) Find that overall AA similarity and AA similarity at epitope sites is high (>97%).

4) Separately find that there are no statistically significant differences in examining genotype-by-VE.

5) Conclude that within-serotype antigenic differences are not "measurably important to vaccine performance.

I agree with the authors that points 1-4 are useful avenues for investigating the hypothesis at hand (and I fully agree with the methods used to support points 1-4). However, I do not agree that points 1-4 are sufficient to reach the conclusion in point 5.

Regarding point 3, there are examples from influenza in which a single amino acid change reduces VE by ~50% (the 159Y mutation in H3N2 viruses in the 2014-2015 season). It is conceivable that VE would vary across dengue genotypes even if the dynamic range of AA similarity is between 97% and 98.5%.

Additionally, regarding point 4, estimates of genotype-specific VE often have a CI of >40% (Table 2). I realize this is dictated by the quantity of data available, but it is still an open question of whether there is statistical power to declare that there is no across-genotype differences in VE. To be convinced, I need a power calculation. What magnitude of across-genotype effect can the authors statistically reject?”

Required revisions for point A:

1) Authors should explore quantitatively the relationship between VE and AA difference, as for example here:http://www.cehd.umn.edu/ci/rationalnumberproject/images/90_3/figure_1.jpg

2) A strong (adequately powered) test of the null can provide confidence bounds on the degree of heterogeneity that include 0 and are narrow, which is not to say that the null is true but that a strong departure from the null is inconsistent with the data. A weak test of the null will provide wide bounds that may include 0 and also very large degrees of heterogeneity. The similarity of point estimates for different genotypes, alone, provides almost no information about the existence of heterogeneity or lack thereof. Therefore the authors need to provide some evidence about how strong their evidence is for near-heterogeneity of the VE – and based on the wide confidence intervals it seems not very strong for some genotypes. I can think of two ways to do this; the authors could choose one or perhaps take a different approach to provide equivalent information:a) A post-hoc power calculation that would show how large a difference in genotype-specific efficacy (or per-AA dropoff in efficacy) the sample is powered to detect.b) Alternatively, a statement of the confidence intervals for the interaction between vaccination and genotype (or vaccination and AA distance from the vaccine strain) that would measure the plausible amounts of heterogeneity observed.

Together with one of these or something similar, the phrasing in the Abstract and the Discussion about the likely homogeneity of effect should be changed to reflect the findings – again, based on the CI it seems the findings will likely be that the data are consistent with some or moderate heterogeneity for ST1, with no to very much heterogeneity for ST2 (underpowered), with considerable heterogeneity for ST4, and it's hard to say for ST3 as the 2 studies are well powered for different genotypes.

B) The manuscript should address in more detail the level of uncertainty in other findings, including the imputation.

C) The manuscript should more fully address the details of the reduced efficacy of the vaccine among seronegatives and younger children.

D) Little support is given for the particular epitopes considered in the analysis of divergence at mAb targets, except with references made in the legend of supplemental figures. Specifically, can the authors provide a rationale and/or reference data or literature to support the choice of the twelve epitope sequences that they state are "well characterised and potent virus neutralising human mAbs" (subsection “Human mAb epitope sequences in vaccine and wild-type viruses”).

Reviewer #1:

In this manuscript authors Rabaa and colleagues investigate the important issue of whether the CYD14 and CYD15 phase III dengue vaccine trials showed differential vaccine efficacy based on whether circulating viruses "matched" the vaccine product. Generally, I very much appreciated the authors' attention to detail and in-depth presentation of exactly what amino acids differ at what sites between vaccine strains and circulating viruses along with detailed presentation of genotype-by-VE results. However, I felt the authors did not go far enough in their analyses to properly support the primary conclusion of the paper that:

"The finding that CYD-TDV shows relatively consistent efficacy profiles against all genotypes of DENV-1 and -3 suggests that if any antigenic differences exist in these virus populations they are not measurably important to vaccine performance."

The authors’ logic is roughly as follows:

1) Measure overall AA difference between vaccine strains and circulating viruses in the ENV protein.

2) Also measure AA differences at relevant epitope sites between vaccine strains and circulating viruses (where "relevant epitope sites" are defined by monoclonals).

3) Find that overall AA similarity and AA similarity at epitope sites is high (>97%).

4) Separately find that there are no statistically significant differences in examining genotype-by-VE.

5) Conclude that within-serotype antigenic differences are not "measurably important to vaccine performance".

I agree with the authors that points 1-4 are useful avenues for investigating the hypothesis at hand (and I fully agree with the methods used to support points 1-4). However, I do not agree that points 1-4 are sufficient to reach the conclusion in point 5.

Regarding point 3, there are examples from influenza in which a single amino acid change reduces VE by ~50% (the 159Y mutation in H3N2 viruses in the 2014-2015 season). It is conceivable that VE would vary across dengue genotypes even if the dynamic range of AA similarity is between 97% and 98.5%.

Additionally, regarding point 4, estimates of genotype-specific VE often have a CI of >40% (Table 2). I realize this is dictated by the quantity of data available, but it is still an open question of whether there is statistical power to declare that there is no across-genotype differences in VE. To be convinced, I need a power calculation. What magnitude of across-genotype effect can the authors statistically reject?

Major suggestions for improvement to bridge to the conclusion at point 5:

1) It is fully possible that the authors' results can be explained with a consistent relationship between AA distance between vaccine strain and genotype and genotype-specific vaccine efficacy while not showing an obvious difference of VE across individual genotypes. I'm imagining a situation like shown here:

http://www.cehd.umn.edu/ci/rationalnumberproject/images/90_3/figure_1.jpg

One sees a consistent relationship between the x variable (distance) and the y variable (VE) even if group-specific y values are hard to pin down.

I would strongly encourage the authors to do a similar analysis. Is there an overall correlation between amino acid distance (overall and at epitope sites) and mean estimates of genotype-specific VE?

2) The authors' primary conclusion needs to be justified with a power calculation. With the quantity of data that the authors have (6 cases and 6683 person-years-at-risk in the vaccine group of serotype 1 / genotype 1 in CYD14, etc.) what level of relationship between AA distance and VE do the authors have statistical power to perceive? This could be approached by generating simulated datasets with a specified effect size and then testing whether the statistical approach has power to detect this effect size.

This would allow the authors to reach the more justified and defined conclusion that within-serotype antigenic differences must be less than X in magnitude rather than antigenic differences are not "measurably important to vaccine performance".

Without these additional analyses, I do believe the paper's major conclusions can be supported.

Reviewer #2:

In this manuscript, Rabaa and coathors demonstrate pan-genotype and pan-serotype efficacy for the CYD DENV vaccine by generating 454 sequenced data for 314 and 333 viral E genes from patient samples deriving from the CYD14 and CYD15 phase 3 clinical trials, respectively. The samples derive from children with a range of ages, hailing from eleven countries, and the authors examine genotype-specific efficacy for 10 clades of the E gene.

This is a well-written and largely clearly presented manuscript. The authors acknowledge the two most significant weaknesses of the analysis: the strong and well-documented bias with regard to generation for samples with high viral titer, and the power limitations of sample size for the calculation of genotype-specific efficacy given the large number of serotypes and genotypic clades examined (not to mention the significant genetic diversity within each genotypic clade).

The authors could address the former issue more directly by looking for association between successfully amplified viral genotypes and viral titer; if there are no trends or genotype associations within the genotype-able range of titer, that would be reassuring.

The latter issue is more difficult to address with the available data. Larger sample size with less stratification would be desirable. The authors describe part of the motivation for this study as a desire to determine whether a vaccine 'escape response' could result from genotype-specific efficacy. Are there models or data to suggest what magnitude of genotype-specific efficacy could induce an escape response in a DENV population, and if so, does the present study have power to detect selective pressures of that relevant magnitude?

Specific points:

Subsection “Sequence differences between CYD-TDV vaccine strains and circulating wild-type viruses” and elsewhere: What does X% sequence divergence mean? How large is the E gene construct in the CYD vaccine, and how many amino acid positions were different in the observed samples?

Subsection “Vaccine efficacy by DENV serotype and genotype”, second paragraph: It would be useful in this discussion of genotype-specific efficacy to mention which genotype clades are represented in the CYD vaccine. Otherwise, it is necessary to manually annotate Tables 1 and 2 after referencing and interpreting the phylogenetic tree figures.

Subsection “Data Analysis”: Was there any validation of genotype/sequencing accuracy performed? Even if phred scores were high, were there issues with homopolymeric sequences, a notorious weak point of 454 sequencing? Were real or spurious insertions or deletions observed? Were any technical replicates performed on samples to evaluate data via consistency?

Reviewer #3:

The manuscript "Genetic epidemiology of dengue virus populations in phase III trials of the CYD tetravalent dengue vaccine and implications for efficacy" presents an analysis of dengue viruses that were detected during a vaccine trial. The study characterizes dengue viruses that were isolated from vaccinated and individuals who experienced infection during the trial. The study compares parental viruses of the vaccine to circulating viruses and estimates genotype specific efficacy rates. The study provides important information about the performance of the vaccine as well as interesting virus sequences (for interests even independent of the vaccine trial) in providing a snapshot of dengue viruses circulating in multiple locations at the same time.

The manuscript makes an important contribution to the field. My main concerns are that 1) little support is given for the particular epitopes considered, except with references made in the legend of supplemental figures 2) consideration of uncertainty in each of the analyses is insufficient including the imputation and the details of the reduced efficacy of the vaccine among seronegatives and younger children should be an additional focus of the manuscript. Overall, the strengths of the manuscript outweigh the weaknesses.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Genetic epidemiology of dengue virus in phase III trials of the CYD tetravalent vaccine and implications for efficacy" for further consideration at eLife. Your revised article has been evaluated by Wendy Garrett (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

From the Reviewing Editor:

I remain concerned that underpowered and thus uninformative analyses are presented as results. The qualifications stated in the revised version help this somewhat, but the paper continues to give the impression of finding that efficacy was equivalent between genotypes except for DENV4.

The comment in the decision letter was that the authors need to perform either post-hoc power calculations or report confidence intervals for the evidence for interaction, consistent with the very strong recent statement from the American Statistical Association that p values alone are not a grounds for scientific conclusion, and recommendation: "In view of the prevalent misuses of and misconceptions concerning p-values, some statisticians prefer to supplement or even replace p-values with other approaches. These include methods that emphasize estimation over testing, such as confidence, credibility, or prediction intervals; Bayesian methods; alternative measures of evidence, such as likelihood ratios or Bayes Factors; and other approaches such as decision-theoretic modeling and false discovery rates. All these measures and approaches rely on further assumptions, but they may more directly address the size of an effect (and its associated uncertainty) or whether the hypothesis is correct." (http://amstat.tandfonline.com/doi/abs/10.1080/00031305.2016.1154108)

The authors declined a post-hoc power calculation on the advice of a statistician, but did not address the other suggestion to calculate confidence intervals for the interactions, which would be trivial to do. In the case of DENV2 in particular, it seems that the calculation of these intervals would lay bare that the study simply offers little evidence on the question, rather than as the authors say a difference that was not statistically significant.

I repeat this request for CI for the interactions and believe it needs to be addressed before acceptance. As an example of what I believe this will show: The statement in the Abstract currently is:

"In post-hoc analysis of all CYD14/15 trial participants, the only statistically significant genotype-level VE association was within DENV-4, where efficacy against genotype I was lower than to genotype II. In trial participants age 9-16 years, no genotype-level associations with VE were observed, although this subgroup analysis had less precision due to smaller sample sizes."

I suspect that the confidence bounds for interaction will be tight for DENV-4 and exclude 0: Thus for DENV-4, the authors can conclude that efficacy against genotype I was lower than for Genotype II.

I expect the confidence bounds for interaction will be extremely wide for DENV-2, including both zero and very large differences: if true this means the study results gave little information about efficacy differences between genotypes.

I expect the confidence bounds for interaction will be relatively tight around DENV 1, indicating that the study provides evidence that if there is heterogeneity, it is relatively small.

For DENV3 I expect the results to be intermediate between the cases for DENV1 and DENV2, meaning the study results are consistent with heterogeneity from zero to moderate.

These are the legitimate conclusions. Stating a trend, but noting lack of significance and small sample size clouds the issue. The goal of a study like this is to narrow the possible states of the world consistent with the data. For DENV4 that has been done to strongly suggest heterogeneity exists and may be substantial. For DENV1 that has been done to strongly suggest that heterogeneity is absent or at most small. For DENV2 that has not been done to an appreciable extent, and DENV3 is in between.

https://doi.org/10.7554/eLife.24196.055

Author response

[…] Required revisions for point A:

1) Authors should explore quantitatively the relationship between VE and AA difference, as for example here:http://www.cehd.umn.edu/ci/rationalnumberproject/images/90_3/figure_1.jpg

As suggested, we have generated a supplementary figure (Figure 3—figure supplement 1) to represent the relationship between VE and AA identity (between the individual vaccine strains and the wild-type viruses sampled in the trial). This figure is illustrative only and the confidence intervals around the VE estimates preclude any attempt at correlation analyses. This approach nonetheless provides a useful precedent to the field in how to present such data in future dengue vaccine trials.

2) A strong (adequately powered) test of the null can provide confidence bounds on the degree of heterogeneity that include 0 and are narrow, which is not to say that the null is true but that a strong departure from the null is inconsistent with the data. A weak test of the null will provide wide bounds that may include 0 and also very large degrees of heterogeneity. The similarity of point estimates for different genotypes, alone, provides almost no information about the existence of heterogeneity or lack thereof. Therefore the authors need to provide some evidence about how strong their evidence is for near-heterogeneity of the VE – and based on the wide confidence intervals it seems not very strong for some genotypes. I can think of two ways to do this; the authors could choose one or perhaps take a different approach to provide equivalent information:a) A post-hoc power calculation that would show how large a difference in genotype-specific efficacy (or per-AA dropoff in efficacy) the sample is powered to detect.b) Alternatively, a statement of the confidence intervals for the interaction between vaccination and genotype (or vaccination and AA distance from the vaccine strain) that would measure the plausible amounts of heterogeneity observed.

Together with one of these or something similar, the phrasing in the Abstract and the Discussion about the likely homogeneity of effect should be changed to reflect the findings – again, based on the CI it seems the findings will likely be that the data are consistent with some or moderate heterogeneity for ST1, with no to very much heterogeneity for ST2 (underpowered), with considerable heterogeneity for ST4, and it's hard to say for ST3 as the 2 studies are well powered for different genotypes.

The main critique is that we have not adequately qualified our interpretation of our finding that vaccine efficacy did not differ between genotypes of the same serotype in the trial data. We agree that the absence of a measurable association does not prove the null, i.e. that there is no true difference in VE between genotypes. With respect to the suggestion that we perform post hoc power calculations – we consulted with independent statisticians and there is clearly not a consensus opinion; indeed the literature suggests post hoc power calculations can have significant drawbacks and may imply greater certainty around these estimates (1, 2). Our response therefore is to modify the manuscript text to be appropriately more conservative in our interpretation. In the revised manuscript, we have placed greater focus on the uncertainty in the point estimates of VE, revealed by the size of the 95% CIs, and repeatedly described the uncertainty inherent to VE estimates from sometimes small sample sizes.

Line by line edits to the manuscript addressing this critique and resulting in a more balanced perspective of the results:

1) Abstract: We removed text suggesting that our findings were broadly reassuring that efficacy would not be influenced by within-serotype genetic diversity, instead we simply state our observations relating to DENV-4 and qualify the subgroup analysis herewith “In post hoc analysis of all CYD14/15 trial participants, the only statistically significant genotype-level VE association was within DENV-4, where efficacy against genotype I was lower than to genotype II. In trial participants aged 9-16 years, no genotype-level associations with VE were observed, although this subgroup analysis had less precision due to smaller sample sizes. Post-licensure surveillance is needed to monitor vaccine performance against the backdrop of DENV sequence diversity and evolution.”

2) Introduction, “Lastly, we aimed to explore if a more complex genotype-specific efficacy pattern existed in the CYD14 and CYD15 trials, notwithstanding the limitations inherent in post hoc analysis.”

3) Results, “However, when the genotype-specific efficacy was considered only in participants ≥ 9 years, DENV-4 vaccine efficacy was comparable across all groups (80.8-89.6%) and no significant heterogeneity was detected between genotypes (p=0.67), although the sample sizes were small and the confidence intervals around the point estimates wide (Table 2).”

4) Discussion, “That CYD-TDV immunisation induced similar measured levels of efficacy against all genotypes of DENV-1 and -3 is so far supportive of the concept that clinical immunity elicited by CYD-TDV to these serotypes was pan-genotype in nature (Table 1 and 2). Nonetheless further research is warranted to understand if smaller differences exist in genotype-level vaccine efficacy than could be measured in these trials and that might be relevant to programmatic use of vaccine.”

5) Discussion, “Subgroup analyses of genotype and age-stratified vaccine efficacy are inevitably speculative because of diminishing sample sizes and wide confidence intervals around the point estimates.”

6) Discussion, “As above for DENV-4, it is speculative to examine subgroups, but for the age group 9-16 years of age, the age range eligible for the now licensed vaccine, the efficacy against DENV-2 Asian I genotype (34.6% (-27.4; 65.7) was comparable to that seen against the other DENV-2 genotypes (Asian/American and Cosmopolitan), albeit with inevitably wide confidence intervals around the point estimates.”

7) Discussion, “These were post hoc analyses and inevitably sample sizes became small for genotype level vaccine efficacy estimates and in particular the age-class (9-16 year olds) subgroup analysis. This manifests as wide 95% confidence intervals around the genotype-level point estimates of vaccine efficacy.”

8) Discussion, “Despite these limitations, with the exception of DENV-4, the results described here do not reveal large and significant differences in vaccine efficacy between genotypes of the same serotype in the CYD14/15 trials. However, these data do not preclude the possibility that a true difference exists, just that a larger sample size than that available in this study would be required to detect this.”

9) Discussion: We removed text suggesting that the lack of heterogeneity in efficacy estimates is an indication that efficacy is homogenous across diverse genotypes, replacing it with a more clear statement of our findings only, that “the results described here do not reveal a significant difference in VE between genotypes of the same serotype in the CYD14/15 trials. However these data do not preclude the possibility that a true difference exists, just that a larger sample size than that available in this study would be required.”

B) The manuscript should address in more detail the level of uncertainty in other findings, including the imputation.

The critique suggests there is uncertainty in the imputation of genotype identity. We recognize some uncertainty and have acknowledged this in the manuscript. However, as described in the manuscript, we believe that uncertainty in the imputation is low overall and particularly so for Latin America. In all of the CYD15 Latin American countries, not only did we identify just one genotype per serotype in each country (with the exception of a single isolate of genotype I DENV-4, a likely importation from Southeast Asia), but the virus genotype imputation was entirely supported by contemporary independent descriptions of the local DENV population genetics, i.e. the genotype identity of ~200 DENV sequences submitted independently to Genbank between 2010-2014 by other investigators were identical to the country-level genotype assignments reported for the CYD15 patients in this study, and thus supported our imputation approach (Supplementary file 3).

For CYD14, Supplementary file 3 provides genotype identity to viruses submitted to Genbank between 2010 and 2014. In all instances the genotype we imputed for the country level data CYD14 results was the majority sequence genotype in that country, thus generally supporting the validity of the imputation methods we used for missing data CYD14 genotype data.

C) The manuscript should more fully address the details of the reduced efficacy of the vaccine among seronegatives and younger children.

It is not possible to do any meaningful virus genotype analyses in the “sero-cohort” because these presented only 10% of the total number of participants (see the respective CYD trial manuscripts). We note our inability to assess this due to small sample sizes in the last paragraph of the Discussion. For the benefit of readers, we show the genotype level vaccine efficacy results for children under 9yrs of age in Supplementary file 5 and make mention of this in the Results text. We do not elaborate greatly on these results because only the CYD14 trial enrolled participants in this age range and thus the numbers of cases per genotype is small and inevitably this yields very large 95% CI around the efficacy point estimates.

D) Little support is given for the particular epitopes considered in the analysis of divergence at mAb targets, except with references made in the legend of supplemental figures. Specifically, can the authors provide a rationale and/or reference data or literature to support the choice of the twelve epitope sequences that they state are "well characterised and potent virus neutralising human mAbs" (subsection “Human mAb epitope sequences in vaccine and wild-type viruses”).

We have now added the appropriate references for the mAbs in the Results section in addition to their original referencing in the Materials and methods section. We have also added text to explain why these represent prominent neutralizing mAb epitopes under investigation by the dengue vaccine community.

References:

1) J. M. Hoenig, D. M. Heisey, The abuse of power: the pervasive fallacy of power calculations for data analysis, Am. Stat. 55, 19–24 (2001).

2) R. V Lenth, Post hoc power: tables and commentary, Iowa City Dep. Stat. Actuar. Sci. Univ. Iowa (2007).

[Editors' note: further revisions were requested prior to acceptance, as described below.]

[…] These are the legitimate conclusions. Stating a trend, but noting lack of significance and small sample size clouds the issue. The goal of a study like this is to narrow the possible states of the world consistent with the data. For DENV4 that has been done to strongly suggest heterogeneity exists and may be substantial. For DENV1 that has been done to strongly suggest that heterogeneity is absent or at most small. For DENV2 that has not been done to an appreciable extent, and DENV3 is in between.

We appreciate the reviewers’ concern for our assessment of uncertainty and the language that they used to push us in the right direction. As suggested, we have now adjusted our analyses to include estimates of the interaction between vaccine efficacy and genotype, along with 95% confidence intervals defining the uncertainty around these estimates. We have added this to the Materials and methods, which now indicate “A Cox regression model was used to estimate vaccine efficacy (derived as 100* [1- Hazard Ratio]) with vaccine group included as a covariate and 95% CI. To further investigate the interaction between vaccine efficacy and genotype, an additional Cox regression model was used to estimate vaccine efficacy with vaccine group, genotype and the interaction between vaccine group and genotype included as covariates.”

We have added two tables with these results, Table 3 (all age groups) and Table 4 (9-16 years of age). The results included in these tables are generally as expected by the reviewer. For DENV-4, they confirm the decreased efficacy of genotype I in the all age group, and confirm a lack of heterogeneity in the ≥ 9 year olds. For DENV-1, tight confidence intervals suggest limited uncertainty around our conclusion of little to no heterogeneity. For DENV-3, wide confidence intervals suggest that we have limited power to assess heterogeneity in vaccine efficacy in these groups. In the case, of DENV-2, we interestingly find confidence bounds including, but very close to zero, in comparison of the Asian I genotype to the American/Asian genotype in the all ages group, which suggests that although our power to detect a difference is limited, there may be an interaction that we are just beyond the limit of detecting. We have added these results to the manuscript in the Results section (subsection “Vaccine efficacy by DENV serotype and genotype”, last four paragraphs) and in the Discussion (fourth, fifth and sixth paragraphs).

To clarify this uncertainty further, we have also altered the Abstract to read: “Post-hoc analysis of all CYD14/15 trial participants revealed a statistically significant genotype-level VE association within DENV-4, where efficacy was lowest against genotype I. In subgroup analysis of trial participants age 9-16 years, VE estimates appeared more balanced within each serotype, suggesting that genotype-level heterogeneity may be limited in older children.”

https://doi.org/10.7554/eLife.24196.056

Article and author information

Author details

  1. Maia A Rabaa

    1. Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam
    2. Centre for Tropical Medicine, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    mrabaa@oucru.org
    Competing interests
    No competing interests declared
    ORCID icon 0000-0003-0529-2228
  2. Yves Girerd-Chambaz

    Research and Development, Sanofi Pasteur, Lyon, France
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    Competing interests
    Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
    ORCID icon 0000-0002-5121-686X
  3. Kien Duong Thi Hue

    Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam
    Contribution
    Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  4. Trung Vu Tuan

    Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam
    Contribution
    Investigation, Methodology, Writing—review and editing
    Competing interests
    No competing interests declared
  5. Bridget Wills

    1. Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam
    2. Centre for Tropical Medicine, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom
    Contribution
    Supervision, Writing—review and editing
    Competing interests
    No competing interests declared
  6. Matthew Bonaparte

    Global Clinical Immunology, Sanofi Pasteur, Swiftwater, United States
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Matthew Bonaparte: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  7. Diane van der Vliet

    Research and Development, Sanofi Pasteur, Lyon, France
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Diane van der Vliet: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  8. Edith Langevin

    Research and Development, Sanofi Pasteur, Lyon, France
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Edith Langevin: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  9. Margarita Cortes

    Research and Development, Sanofi Pasteur, Bogota, Colombia
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Margarita Cortes: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  10. Betzana Zambrano

    Research and Development, Sanofi Pasteur, Montevideo, Uruguay
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Betzana Zambrano: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  11. Corinne Dunod

    Research and Development, Sanofi Pasteur, Lyon, France
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Corinne Dunod: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  12. Anh Wartel-Tram

    Medical Affairs and Public Policy Asia AP, Sanofi Pasteur, Singapore, Singapore
    Contribution
    Data curation, Writing—review and editing
    Competing interests
    Anh Wartel-Tram: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  13. Nicholas Jackson

    Research and Development, Sanofi Pasteur, Lyon, France
    Contribution
    Conceptualization, Data curation, Funding acquisition, Project administration, Writing—review and editing
    Competing interests
    Nicholas Jackson: Employee of Sanofi Pasteur, a company engaged in the development of CYD-TDV
  14. Cameron P Simmons

    1. Oxford University Clinical Research Unit, Hospital for Tropical Diseases, Ho Chi Minh City, Vietnam
    2. Centre for Tropical Medicine, Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom
    3. Department of Microbiology and Immunology, Peter Doherty Institute, University of Melbourne, Victoria, Australia
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Investigation, Visualization, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    csimmons@unimelb.edu.au
    Competing interests
    No competing interests declared
    ORCID icon 0000-0002-9039-7392

Funding

Sanofi Pasteur (Direct funding to perform viral sequencing)

  • Cameron P Simmons

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

The authors would like to acknowledge all investigators, Sanofi Pasteur’s Global Clinical Immunology team and volunteers involved in the clinical evaluation of CYD-TDV. Many thanks also to all Sanofi Pasteur Dengue Vaccine team members and in particular to Helena Aurell, Alain Bouckenooghe, Laurent Chambonneau, Danaya Chansinghakul, Diana Leticia Coronel, Remi Forrat, Carina Frago, Etienne Gransard, Thelma Laot, Josemund Menezes, Fernando Noriega, Leon Ochiai, Eric Plennevaux, Paula Perroud, Enrique Rivas, Melanie Saville, Jo-Ann West and Jean Lang for their support. In Vietnam, thanks to Vi Tran Thuy and Huy Huynh Le Anh. We would also like to thank Hao Chung The and Christine Boinett for assistance with figures. A special thanks to Bruno Guy for helpful discussions and critical reading of the manuscript.

Ethics

Clinical trial registration ClinicalTrials.gov Identifiers: NCT01373281 & NCT01374516

The CYD14 and CYD15 studies were conducted in compliance with Good Clinical Practice guidelines, the principles of the Declaration of Helsinki, and the regulations of the relevant countries. Each study was approved by the appropriate ethics review committee and regulatory agencies as per local rules. Written informed consent was obtained from a parent or guardian for all participants in the two trials, with assent obtained depending on the age of the participant. Metadata for viruses sequenced in this study included the date and country of isolation; to maintain anonymity, no other defining characteristics of study participants were attached to viral data.

Reviewing Editor

  1. Marc Lipsitch, Reviewing Editor, Harvard TH Chan School of Public Health, United States

Publication history

  1. Received: December 13, 2016
  2. Accepted: August 3, 2017
  3. Version of Record published: September 5, 2017 (version 1)

Copyright

© 2017, Rabaa et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 689
    Page views
  • 117
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading