[PPT] - Rega Institute for Medical Research Introduction Rega Institute for PowerPoint Presentation

SLIDE 1

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Rega Institute for Medical Research

Kristof Theys

Clinical and Epidemiological Virology Katholieke Universiteit Leuven

April, 10th Arevir Meeting, Bonn

SLIDE 2

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 3

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 4

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Clinical and Epidemiological Virology

Bioinformatics group

Annemie Vandamme, Raphael Sangeda, Ana Abecasis, Pieter Libin, Philippe Lemey and Kristof Theys

Research interests

Factors that influence therapy response of HIV infected patients

viral resistance against HIV drugs

Molecular evolution and epidemiology of HIV

subtype diversity of HIV tracing the origin of HIV

Combined approach enables to study the evolution of HIV during selective pressure of HIV

SLIDE 5

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Rega analysis tools

Rega algorithm

genotypic resistance interpretation system

Rega HIV subtyping tool Sequence analysis tools

Alignment Translation to amino acids Resistance interpretation

All accessible at http://jose.med.kuleuven.be All integrated in RegaDB

SLIDE 6

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 7

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Therapy Response

Many factors influence therapy outcome

adherence dosis drug interactions metabolism - absorption complexity - toxicity antiviral resistance

Keyrole for antiviral resistance

cause and consequence of failure Avoid development of resistance

irreversible proces cross-resistance

SLIDE 8

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Selecting the optimal combination

Life-long HIV antiviral treatment

Planning successful drug sequencing strategies

Dual requirements for combination Short term

be potent (virological suppression) forgiving (minimally affected by adherene) high genetic barrier** to resistance

Long term

minimally cross-resistance

Sequencing strategies most applicable for drug naive

SLIDE 9

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Rega’s objectives

Optimalisation is necessary for succesful lifelong therapy Time to failure of combination therapy Optimal therapy options after therapy failure

SLIDE 10

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Rega’s methods

Better understanding how resistance develops and evolves under the selective pressure of (combination) therapy.

Influences on resistance evolution

Impact of genetic variability (inter/intra) Interactions between mutations

Genetic barrier of a drug/combination

Time to therapy failure

Evolutionary distance to resistance

Genetic barrier

Potency of the combination (activity)

Phenotype, adherence, . . .

SLIDE 11

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 12

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Applied techniques and applications

Bayesian network learning for resistance development

role of mutations and polymorphisms in treatment failure influence of subtype diversity on resistance improve genotypic interpretation systems

Estimation of fitness landscapes

model of HIV evolution under the selective pressure of treatment predict evolution of HIV

Calculate genetic barrier

define the genotypic correlates which influence genetic barrier

Prediction of therapy response

genotypic predictors based on estimated fitness

SLIDE 13

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 14

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Bayesian Network Analysis

Probabilistic Graphical Models Finds minimal set of direct dependencies that together explain most observed correlation Represent direct dependencies in a directed graph A Bayesian Network refactors the JPD of multivariable data in product of CPDs: P(A1, . . . , An) =

n

i

P(Ai|parents(Ai)) by assuming conditional independencies Refactoring of JPD

P(30N, 71V, 88D, 90M) = P(71V|30N)xP(88D|30N)xP(30N)xP(90M)

SLIDE 15

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Bayesian Network learning

On amino acid sequence data

A Bayesian Network can be “learned“ from data Can be used as a blue-print for amino acid interactions

for an emperical fitness function

Can be interpreted semantically

mutations connected to drug node

presence is directly influenced by drug major mutation

mutations connected to each other

presence is influenced by other mutation minor mutation in resistance pathway

mutations connected to polymorphisms

presence is influenced by natural variation

SLIDE 16

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Bayesian Network of Nelfinavir resistance

SLIDE 17

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Outline

1

Introduction Rega Institute for Medical Research Treatment of HIV-1 infection

2

Understanding HIV evolution under selective pressure of HIV therapy General overview of applied techniques Bayesian Network Learning Estimation of Fitness Landscape

SLIDE 18

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Darwinian Fitness (during treatment)

Defines the ability to replicate in a given environment

dependent on virus (replication capacity) dependent on environment (CTL response, therapy, ...)

Fitness is the driving force behind evolution

Selective pressure of therapy influences relative fitness

f population

Shift in quasispecies distribution and selection of resistant strains

Viral Resistance is the Outcome of Viral Replication, Mutation and Selection

SLIDE 19

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Estimating a fitness function

. . . to imagining a fitness landscape

SLIDE 20

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Fitness Landscape

Objective

A model for HIV evolution during treatment

Mutation, fitness and selection

Main components

An estimated in vivo fitness landscape A mathematical model of evolution

Concept

Observing evolution provides information on fitness High correlation between prevalence and selective advantage

Simulated evolution over the landscape

predict evolution genetic barrier

SLIDE 21

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Fitness Landscape

Principle

Fitness function

Fitness function structure

using interactions indicated by Bayesian Network learning (structure)

Fitness function parameters

comparing two sequence populations sequences PN from class-naive patients sequences PT from patients treated with a single inhibitor in its class so that simulated evolution of a naive population over the fitness landscape resembles the treated population

SLIDE 22

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Fitness function structure

Bayesian Network: P(A1, . . . , An) =

n

i

P(Ai|parents(Ai)) Fitness function: F(A1, . . . , An) =

n

i

F(Ai|parents(Ai))

Bayesian network structure refactors the JPD P(30N, 71V, 88D, 90M) = P(71V|30N)xP(88D|30N)xP(30N)xP(90M) Corresponding relative fitness function F(30N, 71V, 88D, 90M) = F(71V|30N)xF(88D|30N)xF(30N)xF(90M) F(88D|30N) represents contribution of mutation 88D depending on presence of mutation 30N

SLIDE 23

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Estimating a fitness function

model parameters HIV intra−host evolution Data Fitness landscape parameter estimation Naive sequences mutation biases µ Ne Bayesian Network learning Bayesian Network Fitness landscape Parameter Adjustment Evolution Simulator Fitness function structure Phylogeny Treated sequences

SLIDE 24

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Fitness Landscape

validation

Once the fitness model has been builded

fixed fitness structure fixed fitness values for each term in the function

Number of ways to validate the model

correlate in vitro resistance versus estimated fitness Predict evolution under treatment for a single sequence

Summarize in a graph nodes corresponds to predicted sequences arcs corresponds to predicted mutations

Statistical test to correlate predicted evolution versus

bserved evolution

SLIDE 25

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Fitness Landscape

validation

Baseline sequence, Treated sequence

SLIDE 26

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Individualised genetic barrier to resistance

Genetic Barrier

Evolutionary distance of the virus to escape the inhibition of the drugs Simulate evolution from a baseline sequence, over such a fitness landscape, together with a criteria for resistance, allows the estimation of the genotypic barrier to resistance

Simulate evolution until considered "resistant" (N=100) Measure of resistance defined by Rega algorithm Average time as estimate for genetic barrier. number of mutations MR number of simulated generations GR

SLIDE 27

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Individualised genetic barrier to resistance

Investigated the association of estimated genetic barrier with risk of development of NFV resistance at virological failure

population of sequences, susceptible to NFV association of MR, GR or log F using regression a higher estimated genetic barrier was associated with lower

dds for development of resistance at failure

thus, variation in individualised genetic barrier to NFV resistance may impact options after failure Similar for other drugs??

find genotypic correlates

impact of protease mutations and polymorphisms on estimated genetic barrier to NFV resistance

SLIDE 28

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Genotypic correlates that influence the barrier

88D (0.1 %) 33F (0.1 %) 36V (0.3 %) 75I (0.2 %) 20T (0.1 %) 71V (3.1 %) 10F (0.4 %) 13V (17 %) 0.0 0.5 1.0 1.5 2.0 41K 12P 89M 37A 69Y 17D 89I (26 %) (3.3 %) (1.3 %) (1.6 %) (2.4 %) (0.3 %) (0.1 %) 10I 12K 45R 36I 64M 35D 10V 20R 77I 70R 71T 62V 64V 72V (8.0 %) (1.0 %) (1.3 %) (1.4 %) (2.3 %) (2.4 %) (3.1 %) (5.8 %) (14 %) (26 %) (29 %) (23 %) (18 %) (9.3 %) Fold change genetic barrier

SLIDE 29

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Clinical utility of estimated Fitness Landscape for Prediction of virological outcome

To treatment with zidovudine + lamivudine + nelfinavir

Collaboration with EuroSIDA (A. Cozzi-Lepri) Objective: proof-of-principle validation of using fitness functions to predict treatment outcome Prediction of virological outcome at two end-points

VL change at week 12 (176 TCEs)

censored for VL below detection limit of the assays ignoring treatment changes

risk of virological failure at week 48 (90 TCEs)

virological failure defined as a VL >400 copies/ml

nly those still on the same treatment

In comparison with expert genotypic IS (GGS)

Rega v7.1, ANRS 2006.07 and HIVDB 4.2.9

SLIDE 30

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Prediction of virological outcome

Genotypic predictions

Estimated fitness landscape

Nelfinavir selective pressure as function of the PRO sequence Azt plus 3tc selective pressure as function of the RT sequence

Statistical analysis

Linear regression model (prediction VL reduction) Logistic regression model (risk of VF)

SLIDE 31

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Prediction of virological outcome

Genotypic predictions

Genotypic predictions of model

Genetic barrier to resistance

simulating evolution until considered resistant repeat 100 times average time used as estimate

Criteria for resistance

major resistance mutation (MRM) increase in fitness that exceeds threshold

Measures of time

the number of mutations the number of simulated generations

Quantitative predictions

MR and GR (MRM) MF and GF (fitness) log F (susceptibility)

SLIDE 32

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape

Prediction of virological outcome

Conclusions

Expert systems and fitness functions showed comparable performance

fitness functions are perhaps more sensitive to detect reduced response to nelfinavir best performance when using fitness function only for simulation of evolution (MR, GR) combine with a virtual phenotype system

Promising approach for prediction for non-B subtypes

expert knowledge is based on subtype B

interpretation systems show more disagreement in non-B subtypes

most patients in EuroSIDA have subtype B virus but fitness functions are not biased towards subtype B

SLIDE 33

HIV-1 Resistance Evolution

K. Theys

Introduction

Rega Institute for Medical Research Treatment

Methods

Design Bayesian Network Fitness Landscape