Predicting Vibrio cholerae infection and symptomatic disease: a systems serology study

Abstract

Background: Vibriocidal antibodies are currently the best characterised correlate of protection against cholera and are used to gauge immunogenicity in vaccine trials. Although other circulating antibody responses have been associated with a decreased risk of infection, the correlates of protection against cholera have not been comprehensively compared. We aimed to analyse antibody-mediated correlates of protection from both V cholerae infection and cholera-related diarrhoea.

Methods: We conducted a systems serology study that analysed 58 serum antibody biomarkers as correlates of protection against V cholerae O1 infection or diarrhoea. We used serum samples from two cohorts: household contacts of people with confirmed cholera in Dhaka, Bangladesh, and cholera-naive volunteers who were recruited at three centres in the USA, vaccinated with a single dose of CVD 103-HgR live oral cholera vaccine, and then challenged with V cholerae O1 El Tor Inaba strain N16961. We measured antigen-specific immunoglobulin responses against antigens using a customised Luminex assay and used conditional random forest models to examine which baseline biomarkers were most important for classifying individuals who went on to develop infection versus those who remained uninfected or asymptomatic. V cholerae infection was defined as having a positive stool culture result on days 2-7 or day 30 after enrolment of the household's index cholera case and, in the vaccine challenge cohort, was the development of symptomatic diarrhoea (defined as two or more loose stools of ≥200 mL each, or a single loose stool of ≥300 mL over a 48-h period).

Findings: In the household contact cohort (261 participants from 180 households), 20 (34%) of the 58 studied biomarkers were associated with protection against V cholerae infection. We identified serum antibody-dependent complement deposition targeting the O1 antigen as the most predictive correlate of protection from infection in the household contacts, whereas vibriocidal antibody titres ranked lower. A five-biomarker model predicted protection from V cholerae infection with a cross-validated area under the curve (cvAUC) of 79% (95% CI 73-85). This model also predicted protection against diarrhoea in unvaccinated volunteers challenged with V cholerae O1 after vaccination (n=67; area under the curve [AUC] 77%, 95% CI 64-90). Although a different five-biomarker model best predicted protection from the development of cholera diarrhoea in the challenged vaccinees (cvAUC 78%, 95% CI 66-91), this model did poorly at predicting protection against infection in the household contacts (AUC 60%, 52-67).

Interpretation: Several biomarkers predict protection better than vibriocidal titres. A model based on protection against infection among household contacts was predictive of protection against both infection and diarrhoeal illness in challenged vaccinees, suggesting that models based on observed conditions in a cholera-endemic population might be more likely to identify broadly applicable correlates of protection than models trained on single experimental settings.

Funding: National Institute of Allergy and Infectious Diseases and National Institute of Child Health and Human Development, National Institutes of Health.

Figure 1:. Risk of infection in household contacts of an index cholera case

Error bars are 95% CIs. Arrows indicate where the upper confidence interval extends beyond the graph. Odds ratios for household contacts of an index cholera case becoming infected for every two-fold increase in baseline (day 2) antibody titres, after adjusting for age and household clustering. Odds ratios and 95% CIs are shown for each biomarker analysed independently. Biomarkers were excluded for logistic model fitted values that were very close to 1. ADCD=antibody-dependent complement deposition. ADCP=antibody-dependent cellular phagocytosis. ADNP=antibody-dependent neutrophil phagocytosis. CtxB=cholera toxin B subunit. CT-HT=cholera holotoxin. OSP=O-specific polysaccharide. TcpA=toxin coregulated pilus.

Figure 2:. Biomarkers important for classifying household contacts by infection outcome

Top 15 biomarkers important in classifying household contacts of index cholera cases as infected (ie, becoming stool culture positive) versus uninfected (ie, remaining stool culture negative). Biomarkers are ranked by importance scores calculated using conditional random forest classification models. ADCD=antibody-dependent complement deposition. ADCP=antibody-dependent cellular phagocytosis. CtxB=cholera toxin B subunit. CT-HT=cholera holotoxin. OSP=O-specific polysaccharide. TcpA=toxin coregulated pilus.

Figure 3:. Predicting infection outcome among household contacts using different subsets of biomarkers

(A) Cross-validated receiver operator curves for classifying household contacts of index cholera cases that remain uninfected versus became infected using random forest models with different subsets of biomarkers and age. Five biomarkers corresponds to five of the top biomarkers selected via conditional importance (ADCD, CtxB IgM, TcpA IgG2, Ogawa OSP IgG1, and Sialidase IgG1). True and false positive rates were calculated using leave-one-out cross-validation. (B) cvAUC corresponding to the models in A. Error bars are influence-curve-based 95% CIs for the cvAUC estimates. ADCD=antibody-dependent complement deposition. CtxB=cholera toxin B subunit. cvAUC=cross-validated area under the curve. OSP=O-specific polysaccharide. TcpA=toxin coregulated pilus.

Figure 4:. Biomarkers important for classifying vaccinees by whether they developed diarrhoea following Vibrio cholerae challenge

Top 15 biomarkers important in classifying vaccinees by developing either no qualifying diarrhoea or mild-to-severe diarrhoea following V cholerae challenge. Biomarkers were selected using serum collected from participants on the day of challenge. Biomarkers are ranked by importance scores calculated using conditional random forest classification models. ADCD=antibody-dependent complement deposition. CtxB=cholera toxin B subunit. CT-HT=cholera holotoxin. OSP=O-specific polysaccharide. TcpA=toxin coregulated pilus.

Figure 5:. Predicting whether vaccinees develop diarrhoea using different subsets of biomarkers

(A) Cross-validated receiver operator curves for classifying vaccinees that develop diarrhoea versus those that do not using random forest models with different subsets of biomarkers and age. Five biomarkers correspond to five of the top biomarkers selected via conditional importance (CT-HT IgA, CtxB IgA, CT-HT IgA2, TcpA IgA, and Sialidase IgA2). True and false positive rates were calculated using leave-one-out cross-validation. (B) cvAUC corresponding to the models in A. Error bars are influence-curve-based 95% CIs for the cvAUC estimates. ADCD=antibody-dependent complement deposition. CtxB=cholera toxin B subunit. CT-HT=cholera holotoxin. cvAUC=cross-validated area under the curve. TcpA=toxin coregulated pilus.