Nonsecretor Histo–blood Group Antigen Phenotype Is Associated With Reduced Risk of Clinical Rotavirus Vaccine Failure in Malawian Infants

Abstract Background Histo–blood group antigen (HBGA) Lewis/secretor phenotypes predict genotype-specific susceptibility to rotavirus gastroenteritis (RVGE). We tested the hypothesis that nonsecretor/Lewis-negative phenotype leads to reduced vaccine take and lower clinical protection following vaccination with G1P[8] rotavirus vaccine (RV1) in Malawian infants Methods A cohort study recruited infants receiving RV1 at age 6 and 10 weeks. HBGA phenotype was determined by salivary enzyme-linked immunosorbent assay (ELISA). RV1 vaccine virus shedding was detected by quantitative real-time polymerase chain reaction (qRT-PCR) in stool collected on alternate days for 10 days post-immunization. Plasma rotavirus–specific immunoglobulin A was determined by ELISA pre- and post-immunization. In a case-control study, HBGA phenotype distribution was compared between RV1-vaccinated infants with RVGE and 1:1 age-matched community controls. Rotavirus genotype was determined by RT-PCR. Results In 202 cohort participants, neither overall vaccine virus fecal shedding nor seroconversion differed by HBGA phenotype. In 238 case-control infants, nonsecretor phenotype was less common in infants with clinical vaccine failure (odds ratio [OR], 0.39; 95% confidence interval [CI], 0.20–0.75). Nonsecretor phenotype was less common in infants with P[8] RVGE (OR, 0.12; 95% CI, 0.03–0.50) and P[4] RVGE (OR, 0.17; 95% CI, 0.04–0.75). Lewis-negative phenotype was more common in infants with P[6] RVGE (OR, 3.2; 95% CI, 1.4–7.2). Conclusions Nonsecretor phenotype was associated with reduced risk of rotavirus vaccine failure. There was no significant association between HBGA phenotype and vaccine take. These data refute the hypothesis that high prevalence of nonsecretor/Lewis-negative phenotypes contributes to lower rotavirus vaccine effectiveness in Malawi.

Introduction of rotavirus vaccines into childhood immunization programs has reduced global child deaths from diarrheal disease [1], but current vaccines are less effective in low-income, high-mortality countries than in higher-income settings [2]. Multiple explanations for this disparity have been proposed, but definitive data are lacking [3]. A widely proposed hypothesis is that histo-blood group antigen (HBGA) phenotype could affect the replication of live rotavirus vaccines in the gut, potentially explaining observed population differences in rotavirus vaccine immunogenicity and effectiveness [4][5][6][7][8][9].
HBGA are complex carbohydrates expressed on the surface of red blood cells and mucosal epithelial cells. Secretion of HBGA, as free oligosaccharides in saliva and other exocrine secretions, is determined by expression of the FUT2 gene.
Mutations of FUT2 result in a nonfunctional enzyme and "nonsecretor" phenotype. A combination of FUT2 and FUT3 gene expression determines the Lewis HBGA phenotype [10].
Both the monovalent human rotavirus vaccine Rotarix (RV1) and pentavalent human-bovine reassortant vaccine Rotateq are based on attenuated P [8] strains. HBGA-associated resistance to P [8] vaccine virus replication could therefore diminish vaccine response. Evidence to support this hypothesis is limited and inconsistent, and no data are available from sub-Saharan Africa [5,6,8]. Malawi is a low-income country that introduced RV1 nationally in 2012. Malawi has high rotavirus genotypic diversity, with around 20% of RVGE caused by P [6] strains [21]. Rotavirus vaccine effectiveness in the first year of life is estimated at 70% [22]. In this population, we sought to test the hypothesis that intrinsic resistance of Lewis-negative/ nonsecretors to G1P [8] infection results in reduced immunoglobulin A (IgA) response, reduced vaccine virus replication, and impaired clinical protection against severe RVGE following G1P [8] rotavirus vaccine.

METHODS
The relationships between HBGA phenotype, vaccine virus replication, and rotavirus-specific IgA response were determined in a longitudinal cohort study. The relationship between HBGA phenotype and clinical rotavirus vaccine failure was determined by a cross-sectional case-control study. The University of Malawi College of Medicine (P.09/14/1624) and University of Liverpool (00758) research ethics committees provided ethical approval for both studies.

Longitudinal Cohort Study
Healthy infants attending a vaccination clinic in Blantyre, Malawi, were consecutively recruited from April 2015 to August 2016, prior to first RV1 immunization, following informed parental consent. Blood samples were taken prior to the first RV1 dose (at approximately 6 weeks of age) and 2 weeks following the second RV1 dose (at approximately 12 weeks of age). Stool samples were taken on days 4, 6, 8, and 10 post-immunization.

Case-control Study
Infants aged between 10 weeks and 1 year with severe gastroenteritis, defined as Vesikari score ≥11 [23], were consecutively recruited from January 2015 to January 2017, with informed parental consent, from a secondary referral hospital and 3 primary healthcare centers in Blantyre, Malawi. Stools were tested for rotavirus by rapid immunochromatography test (RotaStrip, Coris Bioconcept, Belgium). Infants who tested rotavirus positive were recruited as RVGE cases (vaccine failures). Age-matched community controls without diarrhea (for at least 1 week prior to recruitment), born within ±30 days of RVGE cases, were recruited from randomly generated locations within the healthcare catchment areas of each recruitment site in a 1:1 ratio. All cases and controls had received 2 doses of RV1 vaccine, confirmed by hand-held health records.

Data Collection and Anthropometry
Socioeconomic and demographic data were collected by structured interview. Nutritional status was determined by measurement of length, weight, and mid-upper arm circumference (MUAC; a measure of wasting) at time of recruitment compared to World Health Organization age-determined z scores [24].

Laboratory Methods
For detailed laboratory methods see Supplementary Methods. HBGA phenotyping was determined by detection of antigens A, B, H, and Lewis a and b in saliva by enzyme-linked immunosorbent assay (ELISA), using specific monoclonal antibodies, detected by peroxidase conjugated anti-mouse IgM. Infants with detectable salivary A, B, or H antigen were classified as secretors. Where detection of A, B, and H antigens was negative or borderline, secretor status was confirmed by ELISA to detect lectin antigen [25]. Infants who were positive for either Lewis a or Lewis b antigen were classed as Lewis positive, and those negative for both Lewis antigens as Lewis negative. FUT2 genotype was determined for infants of nonsecretor phenotype with enough blood available. DNA was extracted from whole blood using the Qiagen DNA Blood Mini Kit (Qiagen, Germany) in accordance with manufacturer's instructions. FUT2 was amplified by polymerase chain reaction (PCR) and restriction fragment length polymorphism was used to identify inactivating mutations.
RV-specific IgA was determined by a custom antibody-sandwich ELISA [26]. Quantification was made by comparison to a standard plasma [27], reported as geometric mean concentration (GMC) in units per milliliter.
Nucleic acid was extracted from stool using the Qiagen Viral RNA Mini-Kit (Qiagen, Germany). Reverse transcription using random primers was used to generate complementary DNA [28]. RV1 shedding was determined by vaccine-specific non-structural protein 2 (NSP2) real-time PCR (RT-PCR) [29] and confirmed by VP6 quantitative RT-PCR (qRT-PCR) [30] (S1), with a cycle threshold (Ct) cutoff value for positivity of <40 cycles. In case-control study participants, including community controls, rotavirus infection was defined as VP6 ≥100 copies/ mL by qRT-PCR. In both cases and in asymptomatic rotavirus infections in controls, rotavirus genotyping was undertaken using 2-stage RT-PCR [31].

Statistical Analyses
All statistical analyses were performed in StataIC version 13.1 (StataCorp, College Station, TX).

Cohort Study
RV1 vaccine virus shedding was defined as 2 or more NSP2positive, VP6-positive samples post-immunization. NSP2positive, VP6-negative samples were considered negative. NSP2-negative, VP6-positive samples were assumed to reflect wild-type infection. A minimum of 2 post-immunization samples were required for inclusion in shedding analysis. Seropositivity was defined as RV-specific IgA >20 U/mL. Seroconversion was defined as a change from seronegative pre-immunization to seropositive post-immunization or at least a 4-fold rise in RV-specific IgA concentration post-immunization among infants seropositive at baseline. The relationships between HBGA phenotype (defined categorically on secretor and Lewis status) and these categorical outcomes were assessed by log-binomial regression. The relationships between HBGA phenotype and continuous variables (peak vaccine virus shedding, RV-specific IgA GMC) were determined by Wilcoxon rank sum test.
For the cohort study, a sample size of 200 was estimated to achieve 80% power to detect a risk ratio of 0.5 (vs equal risk, alpha 0.05).

Case-control Study
The odds of specific HBGA phenotype (defined categorically on secretor and Lewis status) were compared between cases and matched community controls by conditional logistic regression. With 1:1 controls, a sample size of 123 cases was estimated to achieve 80% power to detect an odds ratio (OR) of 2.5 (vs equal odds, alpha 0.05).

Genotyping Analysis
In an additional case-control analysis, the distribution of HBGA phenotype by genotype-specific RVGE was compared to community controls. This stratified analysis was unmatched, as there were too few matched pairs for meaningful analysis. Separate analyses determined distribution of HBGA phenotype in P [8], P [4], and P[6] RVGE compared to community controls by logistic regression. Rotavirus cases where genotype could not be confirmed were excluded.
A descriptive analysis of HBGA phenotype distribution in genotype-specific asymptomatic rotavirus infection in community controls was made.

HBGA Phenotype, RV1 Fecal Shedding, and Seroconversion
A total of 293 infants were recruited to the cohort study. Of these, 243 infants in the first dose period, 214 infants in the second dose period, and 202 infants in both dose periods provided at least 2 stool samples. Both pre-and post-immunization samples for RV-specific IgA were provided by 196 infants. Demographic characteristics were similar in those with complete data compared to those with incomplete data (Supplementary Tables 1 and 2).
Compared to secretor infants, nonsecretors had significantly reduced risk of vaccine virus fecal shedding in the first dose period, but not in the second. The overall risk of vaccine virus fecal shedding in infants with data for both dose periods did not differ between nonsecretors and secretors (Table 1).
In a stratified analysis comparing shedding by sampling day, nonsecretors had significantly reduced risk of vaccine virus shedding (4/49, 8%) compared to secretors (51/182, 28%) on day 10 following the first vaccine dose. Risk of vaccine virus shedding was not significantly different between nonsecretors and secretors on other sampling days in the first dose period or on any day in the second dose period (Supplementary Table 3). There was no difference in peak level of vaccine virus shedding, as determined by NSP2 Ct value, by secretor status (Table 1).
When Ct values were compared by sample day, median Ct values in nonsecretors were higher (viral load lower) compared to secretors on days 6 and 8 following the first vaccine dose, but not on any other sample day (Supplementary Table 4). There was no difference in vaccine virus fecal shedding between Lewis-negative and Lewis-positive infants by any categorical or quantitative measure ( Table 2, Supplementary Tables  3 and 4).
Among infants with detectable post-immunization RV-specific IgA, there was no difference in GMC between secretors and nonsecretors or between Lewis-positive and Lewis-negative infants (Tables 1 and 2).
In a sensitivity analysis where secretor/nonsecretor status was recategorized by confirmatory FUT2 genotyping and phenotype at age 10 weeks, there remained no association between nonsecretor status and either vaccine virus shedding or seroconversion (Supplementary Table 5). Concordance between genotype and phenotype was 90%.
There was no difference in vaccine virus shedding or seroconversion when secretor phenotype was stratified by Lewis phenotype (Supplementary Table 6). In a subanalysis of secretor infants, there was no association between ABO phenotype and either vaccine virus shedding or seroconversion (Supplementary  Tables 7 and 8).

Case-control Study
A total of 119 eligible severe RVGE cases and 119 age-matched community controls were recruited. Median MUAC was lower in RVGE cases (13.1 cm; interquartile range [IQR], 12.4-14 cm) than in community controls (13.8 cm; IQR, 13.2-14.5 cm; P < .01). No other differences in anthropometric or socioeconomic characteristics between cases and controls were observed (Supplementary Table 9).

HBGA Phenotype Distribution in Infants With RV1 Clinical Vaccine Failure
The prevalence of nonsecretor phenotype was significantly lower in infants with clinical RV1 vaccine failure (14/119, 12%) compared to community controls (33/119, 28%). The odds of nonsecretor phenotype were more than 60% lower in RV1 vaccine failures than in age-matched community controls (Table  3). In a sensitivity analysis where secretor/nonsecretor status was recategorized by FUT2 genotyping, the distribution of nonsecretor phenotype in RV1 vaccine failures and controls was unchanged (OR, 0.36; 95% CI, 0.17-0.74; Supplementary Table 10). Concordance between genotype and phenotype was 86%.
There was no association between Lewis phenotype and RV1 vaccine failure (Table 3).

HBGA Phenotype and Asymptomatic Rotavirus Infection
Asymptomatic rotavirus infection was common; 52/119 (54%) of community controls had detectable rotavirus above 100 copies/mL, with a median viral load of 628 (IQR, 258-2008) copies/ mL. Due to low viral load, full genotype was only available in 21 asymptomatic infections and partial genotype in an additional 7 ( Figure 1B).

DISCUSSION
Contrary to our initial hypothesis, nonsecretor phenotype was significantly less prevalent in infants with clinical vaccine failure.   We found limited evidence that nonsecretor phenotype was associated with reduced vaccine take. The proportion of infants with RV1 vaccine virus shedding in the first dose period was lower in nonsecretors compared to secretors, with lower quantitative shedding on some sample days. However, the overall risk of vaccine virus shedding and peak shedding level did not differ. The proportion of infants with post-immunization RV-specific IgA seroconversion was lower in nonsecretors compared to secretors but not significantly so. Nonsecretor phenotype was associated with protection against both P [8] and P [4] RVGE, the 2 most common rotavirus strains in Malawi. Similarly, against our initial hypothesis, there was no observed association between Lewis-negative phenotype and either rotavirus vaccine take or clinical vaccine failure. Lewis-negative phenotype was less common in infants with P [8] and P [4] gastroenteritis but more common in infants with P [6] gastroenteritis, the third most common strain in this study population. These opposing effects may have brought the association between Lewis phenotype and rotavirus vaccine failure toward the null. The lower point estimate of seroconversion in nonsecretor infants (13% compared to 27% in secretor infants) is consistent with previous studies. Bucardo et al [6] in Nicaragua reported similar findings, while Kazi et al [8] in Pakistan reported lower seropositivity following 3 doses of RV1 in nonsecretors. Our finding that nonsecretor infants are relatively protected from RVGE is consistent with data from Bangladesh where nonsecretor phenotype was associated with a decreased risk of rotavirus diarrhea in unvaccinated infants [5]. This study did not demonstrate a significant association between nonsecretor phenotype and risk of rotavirus vaccine failure, but numbers of vaccine failures were small. Our findings are also consistent with surveillance data from the United States where nonsecretors were at greatly reduced risk of vaccine failure [19], although notably in this population, 91% of gastroenteritis cases were due to P [8] infection.
Nonsecretor phenotype distribution was similar in infants with asymptomatic rotavirus infection compared to the general study population. This could suggest that nonsecretor phenotype provides relative protection against rotavirus disease, but not against asymptomatic infection. This "partial resistance" might explain the limited effect of nonsecretor phenotype on vaccine virus shedding. Asymptomatic infection could potentially allow further boosting of protective immunity [32]. We are the first to report on the relationship between HBGA phenotype and asymptomatic rotavirus infection. Although the number of infants with asymptomatic infection was high, as observed in other low-income settings [26,33,34], the number of genotyped asymptomatic infections was small, and conclusions should be considered within this context. However, our findings are consistent with data from Lee et al (2018) in Bangladesh, in a prospective cohort including mild diarrhea, where P [8] infection was not associated with secretor phenotype [5]. Most prior studies on the relationship between HBGA phenotype and rotavirus have focused on hospitalized RVGE. Additional data on mild and asymptomatic infections are required to confirm this partial resistance hypothesis.
Our study has several limitations. The lower than expected seroconversion rate may have limited analytic power. Exposure to wild-type rotavirus may have increased post-immunization seropositivity. However, since nonsecretors are protected against wild-type infection, any bias would be toward reduced post-immunization RV-specific IgA in this group. Subtle differences in vaccine virus shedding may have been underestimated by semiquantitative measures (Ct value), and borderline results might be clearer in a larger population. Our study relied primarily on salivary HBGA phenotyping by ELISA, which may be less sensitive than genotyping, although concordance between genotyping and phenotyping was high. Furthermore, sensitivity analysis using FUT2 genotyping strengthened the observed protective association between nonsecretor type and odds of clinical vaccine failure.
In summary, we found little evidence in this population that nonsecretor phenotype was significantly associated with reduced vaccine take. Any possible phenotypic disadvantage in vaccine response was clearly outweighed by nonsecretors' relative resistance to wild-type P [8] and P [4] infections, even in this population in which P [6] RVGE was common (>20%). A similar balance would likely exist in other countries with a similar or lower proportion of P [6] RVGE. Recent data show that other sub-Saharan African countries have a prevalence of P [6] RVGE that is similar to the prevalence in Malawi, while the .006 Abbreviation: CI, confidence interval; OR, odds ratio; RVGE, rotavirus gastroenteritis. a Odds ratio of nonsecretor/Lewis-negative phenotype in genotype-specific RVGE cases compared to community controls. P value determined by logistic regression.
prevalence in all other world regions is substantially lower [35,36]. While the prevalence of P [6] could vary over time, we contend that HBGA phenotype is highly unlikely to contribute to current population differences in rotavirus vaccine effectiveness between high-and low-income countries.

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