A genotypic method for the identification of HIV-2 coreceptor usage - - PowerPoint PPT Presentation

A genotypic method for the identification of HIV-2 coreceptor usage

Matthias Döring

Max Planck Institute for Informatics AREVIR meeting May 26, 2017

HIV-2 is prevalent in West Africa and Europe

May 26, 2017 2

Ibe S, Sugiura W. Recombinant Forms of HIV-2. Encyclopedia of AIDS. 2014.

HIV-2 has a milder course of infection than HIV-1

May 26, 2017 3

commons.wikimedia.org/wiki/File:Hiv-timecourse.png

Course of infection for HIV-1 HIV-2 CD4 count HIV-2 viral load Course of infection for HIV-2

Coreceptors are necessary for HIV cell entry

May 26, 2017 4 Engelman and Cherepanov. The structural biology of HIV-1: mechanistic and therapeutic insights. Nature Review Microbiology. 2012; 10, 279-290. Maraviroc

Why treat HIV-2 with coreceptor antagonists?

May 26, 2017 5

NRTIs NNRTIs NRTI + NNRTI PIs INI CCR5 Antagonist FI

Determining coreceptor usage before presccribing

May 26, 2017 6

Courtesy of Nico Pfeifer

R5 X4-capable

The advantages of genotypic approaches

• Interpretable results
• Cost efficiency
• Reduced technological bias
• No standardized assay for HIV-2

May 26, 2017 7

Examples of technological bias

Identifier of X4-capable Isolate

• No. of R5

isolates

• No. of X4-

capable isolates V3 loop of the X4-capable sequence Decision DQ870430 21 1 CKRPGNKTVVPITLMSGLVFHSQPINKRPRQAWC R5 NARI-12 5 1 CKRPGNKTVLPITLMSGLVFHSQPINTRPRQAWC R5 GU204945 3 1 CKRPGNKTVRPITLLSGRRFHSQVYTVNPKQAWC Exclude 310248 1 1 CRRPGNKTVVPITLMSGLVFHSQPINKRPRQAWC X4-capable May 26, 2017 8 HIV-2 samples with discordant annotations of coreceptor usage but identical V3 loops

Approach of geno2pheno[coreceptor-hiv2]

May 26, 2017 9

Reproduction of known markers and novel markers

May 26, 2017 10

X4-probabilities of X4-capable variants as predicted by geno2pheno[coreceptor-hiv2]

Visseaux et al. Molecular Determinants of HIV-2 R5-X4 Tropism in the V3 Loop: Development of a New Genotypic Tool. J Infect Dis. 2012; 205:111–120.

Top-scoring (75% of total weight) features of the predictive model

Performance comparison to Visseaux et al.

May 26, 2017 11

10-fold nested CV performance on the test data set (N = 84) P-value (McNemar’s test): 0.37 No significant difference at 𝛽 = 0.05

Why should you use our tool?

SVM Visseaux et al. Sensitivity 73.5% 85.3% Specificity 96% 94.0%

Döring M, Borrego P, Büch J, Martins A, Friedrich G, Camacho RJ, et al. A genotypic method for determining HIV-2 coreceptor usage enables epidemiological studies and clinical decision support.

• Retrovirology. 2016;13:85.

Validation on an independent test set

• geno2pheno[coreceptor-hiv2]: 9/9 correct
• Rules-based approach: 7/9 correct

May 26, 2017 12

Performance on nine novel HIV-2 samples

Interpretation for ROD10 (H18L + K29T)

May 26, 2017 13

Web service

May 26, 2017 14

Overview of results CSV output Visualization coreceptor-hiv2.geno2pheno.org

Conclusions

May 26, 2017 15

Tool enables epidemiological studies

www.linkedin.com

First web service for HIV-2 coreceptor prediction

• Discriminatory features occur at the end
• Individual amino acids are highly predictive
• Net charge is highly predictive

Features of HIV-2 coreceptor usage in the V3 loop

Thank you for your attention and thanks to …

May 26, 2017 16

Rolf Kaiser Institute for Virology, University of Cologne Rega institute, KU Leuven Ricardo Camacho University of Lisbon Pedro Borrego Max Planck Institute for Informatics, Saarbrücken Thomas Lengauer Max Planck Institute for Informatics, Saarbrücken Nico Pfeifer University of Lisbon Nuno Taveira Georg Friedrich Max Planck Institute for Informatics, Saarbrücken Achim Büch Max Planck Institute for Informatics, Saarbrücken

Backup Slides

May 26, 2017 17

Properties of HIV-2

May 26, 2017 18

kpbs.org

Reeves JD, Doms RW. Human immunodeficiency virus type 2. J Gen Virol. 2002; 83:1253–1265.

albanydailystar.com

Cpz: Chimpanzee MM: Sooty mangabey

Adapted from Ibe S, Sugiura W Recombinant Forms of HIV-2. Encyclopedia of AIDS. 2014.

High prevalence Low prevalence

Marlink et al. Reduced rate of disease development after HIV-2 infection as compared to HIV-1. Science. 1994; 265:1587–90.

Local prevalence Relation to HIV-1 Milder course of infection

Decline of CD4+ cell count in HIV-2 is slower

May 26, 2017 19

N = 32 N = 31

Disease-free: CD4+ cell count ≥ 400 copies per 𝜈𝑚

Marlink et al. Reduced rate of disease development after HIV-2 infection as compared to HIV-1.

• Science. 1994; 265:1587–90.

Viral coreceptor usage and maraviroc treatment

May 26, 2017 20

Treatment with maraviroc

• R5-viruses can use only CCR5
• X4-capable viruses can use

CXCR4

• R5-viruses are inhibited
• X4-capable variants can

still replicate www.aidsinfo.nih.gov www.aidsinfo.nih.gov

R5 X4-capable

www.aidsmap.com

Broad use of HIV-2 coreceptors in vitro

May 26, 2017 21

van der Ende et al. Journal of General Virology. 2000 Bron et al. Journal of Virology. 1997

Coreceptor use in vivo

“CCR5 and CXCR4 appear to be the major coreceptors for HIV-2 infection of PBMC.”

— Mörner et al. AIDS Resarch and Human Retroviruses. 2002

Promiscuity of the HIV-2 ROD strain Coreceptor use of multiple isolates

Clinical staging of HIV-infection

May 26, 2017 22

HIV staging scheme from the CDC WHO staging

Primary Infection

• Acute
• Asymptomatic

Stage 1

• Asymptomatic
• Lymphadenopa

thy

Stage 2

• Weight loss
• Respiratory

infections

Stage 3

• Tuberculosis
• Chronic

diarrhea

Stage 4

• HIV wasting

syndrome

• Pneumocystis

pneumonia

AIDS

Interactions between HIV and cell receptors

May 26, 2017 23

HIV surface Host membrane

Delhalle et al. Phages and HIV-1: from display to interplay. Int J Mol Sci. 2012;13(4):4727-94

The clinical data pyramid

Phenotypes Clinical data Genotypes

May 26, 2017 24

Few data points Many data points

Support vector machines: separable case

May 26, 2017 26

Learning with support vector machines

May 26, 2017 27

Separate the two classes: Maximize the margin:

Learning with support vector machines

May 26, 2017 28

Inseparable case: Cannot be fulfilled!

Learning with support vector machines

May 26, 2017 29

Optimization problem Classification rule

R5 X4-capable

SVM optimization problem

November 23, 2015 30

Inseparable case with margin M Inseparable case with cost parameter C

Kernel SVM function

November 23, 2015 31

Nu-support vector classification

May 26, 2017 32

Schölkopf B, Smola AJ, Williamson RC, Bartlett PL. New Support Vector Algorithms. Neural Comput. 2000; 12:1207-1245.

Motivation for nu-SVM: interpretability

• 𝜉 is bounded: 𝜉 ∈ [0,1]
• Upper bound for the ratio of “errors”
• Lower bound for the ratio of support

vectors

Contingency tables and other measures

November 23, 2015 33

X4-capable R5 X4-capable TP FP R5 FN TN

Reference Prediction

Sensitivity 𝑈𝑈𝑈 = 𝑈𝑈 𝑈𝑈 + 𝐺𝐺 Specificity 𝐺𝐺 𝑈𝐺 + 𝐺𝑈 = 1 − 𝐺𝑈 𝑈𝐺 + 𝐺𝑈 = 1 − 𝐺𝑈𝑈

Structure of confusion tables

Mc Nemar’s Test

November 23, 2015 34

X4-capable R5 X4-capable a b R5 c d

Visseaux et al. SVM Structure for McNemar’s test

Hypothesis

𝐼0: 𝑞𝑐 = 𝑞𝑑 𝐼1: 𝑞𝑐 ≠ 𝑞𝑑 Test the marginal homogeneity: 𝑞𝑏 + 𝑞𝑐 = 𝑞𝑏 + 𝑞𝑑 and 𝑞𝑑 + 𝑞𝑒 = 𝑞𝑐 + 𝑞𝑒

Test statistic

𝜓2 = 𝑐 − 𝑑 2 𝑐 + 𝑑

Distribution

Chi-squared distribution with 1 df

Rejection of the null hypothesis The SVM does not predict very different labels from the approach by Visseaux et al.

Construction of the learning data set

Group Tropism Count A R5 61 A X4-capable 46 B R5 12 B X4-capable 5 D X4-capable 1 U R5 1

May 26, 2017 35

Distribution of genotype-phenotype pairs N=126 (74 R5, 52 X4-capable)

AUCs of different kernel functions are similar

May 26, 2017 36

Best-performing model

Linear SVM AUC = 0.94

Results from 10 runs of 10-fold CV

Interpretation

Marginal role of higher-

• rder interactions

Distribution of transformed decision values

May 26, 2017 37

Choosing an FPR-based classification cutoff

Setting an FPR cutoff at 5% is reasonable!

X4-capable R5