Selected amino acid changes in HIV-1 subtype-C gp41 are associated with specific gp120(V3) signatures in the regulation of co-receptor usage

VirusResearchxxx (2012) xxx–xxx

Contents

lists

available

at

SciVerse

ScienceDirect

Virus

Research

j o

u

r

n

a l

h o m

e p a

g e :

w w w . e l s e v i e r . c o m / l o c a t e / v i r u s r e s

Selected

amino

acid

changes

in

HIV-1

subtype-C

gp41

are

associated

with

specific

gp120

V3

signatures

in

the

regulation

of

co-receptor

usage

1 2

Salvatore

Dimonte

a

,

∗

_,

_Muhammed

_Babakir-Mina

b

,

c

_,

_Fabio

_Mercurio

a

_,

_Domenico

_Di

_Pinto

a

_,

Q1

Francesca

Ceccherini-Silberstein

a

_,

_Valentina

_Svicher

a

_,

_{Carlo-Federico}

_Perno

a

,

b

,

d

3 4

a_{UniversityofRomeTorVergata,viaMontpellier1,00133Rome,Italy}

5

b_{LaboratoryofMolecularVirology,FoundationPolyclinicTorVergata,viaOxford81,00133Rome,Italy}

Q2

6

c_{FoundationofTechnicalEducationinSulaimaniyah,IraqiKurdistanRegion,Iraq}

7

d_{NationalInstituteofInfectiousDiseases(INMI)L.Spallanzani,viaPortuense292,00149Rome,Italy}

8 9

a

r

t

i

c

l

e

i

n

f

o

10 11 Articlehistory: 12 Received6April2012 13

Receivedinrevisedform13June2012

14 Accepted15June2012 15 Available online xxx 16 Keywords: 17 HIV-1 18 Subtype-C 19 gp41 20 gp120V3loop 21 Genotype 22 Tropism 23 Mutations 24 Cluster 25

a

b

s

t

r

a

c

t

ThemajorityofstudieshavecharacterizedthetropismofHIV-1subtype-Bisolates,butlittleisknown aboutthedeterminantsoftropisminothersubtypes.So,thegoalofthepresentstudywastogenetically characterizetheenvelopeofviralproteinsintermsofco-receptorusagebyanalyzing356full-lengthenv sequencesderivedfromHIV-1subtype-Cinfectedindividuals.Theco-receptorusageofV3sequences wasinferredbyusingtheGeno2PhenoandPSSMalgorithms,andalsoanalyzedtothe“11/25rule”.All reportedenvsequenceswerealsoanalyzedwithregardtoN-linkedglycosylationsites,netchargeand hydrophilicity,aswellasthebinomialcorrelationphicoefficienttoassesscovariationamonggp120V3 andgp41signaturesandtheaveragelinkagehierarchicalagglomerativeclusteringwerealsoperformed. AmongenvsequencespresentinLosAlamosDatabase,255and101sequencespredictedasCCR5and CXCR4wereselected,respectively.TheclassicalV3signaturesatpositions11and25,andotherspecific V3andgp41aminoacidchangeswerefoundstatisticallyassociatedwithdifferentco-receptorusage. Furthermore,severalstatisticallysignificantassociationsbetweenV3andgp41signatureswerealso observed.ThedendrogramtopologyshowedaclusterassociatedwithCCR5-usagecomposedbyfivegp41 mutatedpositions,A22V,R133M,E136G,N140L,andN166QthatclusteredwithT2VV3andG24TV3 (boot-strap=1).Conversely,aheterogeneousclusterwithCXCR4-usage,involvingS11GRV3,13–14insIG/LGV3, P16RQV3,Q18KRV3,F20ILVV3,D25KRQV3,Q32KRV3alongwithA30Tgp41,S107Ngp41,D148Egp41,A189Sgp41 wasidentified(bootstrap=0.86).

OurresultsshowthatasobservedforHIV-1subtype-B,alsoinsubtype-Cspecificanddifferentgp41 andgp120V3aminoacidchangesareassociatedindividuallyortogetherwithCXCR4and/orCCR5usage. Thesefindingsstrengthenpreviousobservationsthatdeterminantsoftropismmayalsoresideinthe gp41protein.

© 2012 Published by Elsevier B.V.

1.

Introduction

26

Ninety

percent

of

HIV-1-infected

people

worldwide

harbors

27

non-B-subtype

variants,

and

consequently

the

vast

majority

of

28

cases

of

infections

are

due

to

these

viruses

(

Arien

et

al.,

2007

).

Glob-29

ally,

the

C-subtype

is

the

most

prevalent

circulating

viral

clade

and

30

accounts

for

nearly

half

of

infections,

followed

by

A,

B,

G

subtypes

31

and

the

recombinant

form

CRF02-AG

and

CRF01-AE

(

Hemelaar

32

et

al.,

2011

).

33

The

higher

rate

of

non-synonymous

mutations

tends

to

occur

in

34

regions

of

the

HIV-1

env

gene

and

is

submitted

to

strong

selective

35

∗ Correspondingauthor.Tel.:+390672596564;fax:+390672596039. E-mailaddress:salvatore.dimonte@uniroma2.it(S.Dimonte).

pressure

from

the

immune

system

(

Choisy

et

al.,

2004;

Lemey

et

al.,

36

2006;

Mikhail

et

al.,

2005;

Zhang

et

al.,

2005

).

A

structure

of

partic-

37

ular

importance

in

this

process

is

the

third

variable

loop

(V3)

of

the

38

surface

glycoprotein

gp120

which

is

essential

for

HIV-1

co-receptor

39

usage

(

de

Jong

et

al.,

1992;

Fouchier

et

al.,

1992;

Huang

et

al.,

40

2005

).

In

most

European

countries,

HIV

tropism

is

identified

with

41

tropism

phenotype

testing.

New

data

support

genotype

analysis

42

of

the

V3

for

the

identification

of

HIV-1

tropism

(

Vandekerckhove

43

et

al.,

2011

).

44

HIV-1

enters

into

the

host

cell

by

binding

gp120

to

CD4

recep-

45

tor

on

a

target

cell,

leading

to

conformational

changes

within

gp120

46

that

allows

for

the

engagement

of

a

second

host

cell

receptor

(co-

47

receptor)

(

Alkhatib

et

al.,

1996;

Choe

et

al.,

1996;

Deng

et

al.,

1996;

48

Doranz

et

al.,

1996;

Dragic

et

al.,

1996;

Feng

et

al.,

1996;

Trkola

49

et

al.,

1996;

Wu

et

al.,

1996

).

About

20

G-protein-coupled

receptors

50

0168-1702/$–seefrontmatter © 2012 Published by Elsevier B.V.

2 S.Dimonteetal./VirusResearchxxx (2012) xxx–xxx

(GPCRs)

have

been

shown

to

act

in

vitro

as

co-receptors

(

Neil

et

al.,

51

2005;

Shimizu

et

al.,

2009;

Simmons

et

al.,

2000

),

but

only

CCR5

52

and

CXCR4

are

considered

essential

and

apparently

relevant

in

53

HIV

pathogenesis

(

Berger

et

al.,

1998;

Simmons

et

al.,

2000;

Zhang

54

et

al.,

1998

).

Moreover,

the

interaction

with

co-receptor

induces

the

55

arrest

of

the

gp41

transitions

at

a

pre-hairpin

intermediate

stage

56

that

leads

to

the

insertion

of

the

fusion

peptide

into

the

target

cell

57

membrane

and

ultimately

to

virus-cell

fusion

activity

(

Eckert

and

58

Kim,

2001;

Wyatt

and

Sodroski,

1998

).

59

The

gp41

is

a

transmembrane

glycoprotein

that

retains

the

60

gp120

on

viral

surface

with

non-covalent

interactions

(

Helseth

61

et

al.,

1991

)

and

some

studies

indicate

that

several

mutations

in

62

gp41

were

involved

to

be

significantly

associated

with

co-receptor

63

usage

(

Dimonte

et

al.,

2011a;

Huang

et

al.,

2008;

Stawiski

et

al.,

64

2009;

Thielen

et

al.,

2009,

2010

),

beyond

the

primary

classical

65

determinants

of

gp120

including

particularly

positions

11

and

25

66

in

V3-loop

(

de

Jong

et

al.,

1992;

Fouchier

et

al.,

1992;

Resch

et

al.,

67

2001

),

and

secondly

by

other

flanking

domains

(as

V1,

V2,

C3,

C4

68

and

V5)

(

Carrillo

and

Ratner,

1996;

Huang

et

al.,

2008,

2011;

Koito

69

et

al.,

1995;

Labrosse

et

al.,

2001;

Lin

et

al.,

2011;

Pastore

et

al.,

70

2006;

Svicher

et

al.,

2011b;

Suphaphiphat

et

al.,

2007

).

Both

CCR5

71

and

CXCR4

co-receptors

interact

with

the

same

region

of

the

sur-72

face

gp120

viral

protein

that

encompasses

not

only

the

V3

loop

but

73

also

specific

regions

from

the

V1/V2

and

the

C4

domains

(

Sierra

74

et

al.,

2007

).

75

Moreover,

few

amino

acid

substitutions

and

an

increasing

net

76

charge

of

the

V3-loop

were

sufficient

to

confer

a

change

from

CCR5

77

to

CXCR4

in

cellular

tropism

(

de

Jong

et

al.,

1992;

De

Wolf

et

al.,

78

1994

).

On

the

other

hand,

the

previous

studies

have

defined

the

loss

79

of

a

Potential

N-linked

Glycosylation

Site

(PNGS)

at

V3

positions

6–8

80

(

Pollakis

et

al.,

2001

),

as

a

close

association

between

the

V3-loop

N-81

linked

glycosylation

motifs

(sequons)

and

CXCR4

usage

(

Clevestig

82

et

al.,

2006

).

83

Molecular

mechanisms

underlying

the

transition

from

CCR5

84

to

CXCR4

usage

of

clade

C

virus

remain

poorly

known.

With

the

85

recent

introduction

of

HIV-1

chemokine

receptor

antagonists

on

86

the

market

as

components

of

antiretroviral

therapy,

it

is

increas-87

ingly

important

to

properly

screen

co-receptor

usage

for

all

infected

88

patients

prior

to

therapy

(

Hunt

and

Romanelli,

2009;

Sayana

and

89

Khanlou,

2009

).

Hence,

simple

and

efficient

processes

for

routinely

90

characterizing

and

monitoring

HIV-1

co-receptor

usage

are

needed

91

to

replace

slow

and

resource-intensive

phenotypic

assays.

Existing

92

methods

do

not

consider

the

other

gp120

regions,

mainly

for

lim-93

ited

data

available,

although

incorporating

the

V2-loop

is

known

94

to

improve

prediction

methods

based

on

V3

sequence

informa-95

tion

(

Prosperi

et

al.,

2009

),

and

key

genetic-elements

in

V1,

V2,

and

96

C4

domains

tightly

and

differentially

modulate

HIV-1

dependency

97

on

CXCR4

or

CCR5,

irrespective

of

V3

genetic-background

(

Svicher

98

et

al.,

2011a

).

Nevertheless,

genotypic

determinants

of

co-receptor

99

usage

located

outside

V3

could

also

explain

some

of

the

mispredic-100

tions

(

Raymond

et

al.,

2010

).

101

In

this

study,

large

datasets

of

HIV-1

gp120

V3

and

gp41

C-102

subtype

sequences

were

analyzed

to

genetically

characterize

them

103

in

terms

of

co-receptor

usage.

In

addition,

according

to

CCR5

104

and/or

CXCR4

usage,

the

association

between

amino

acid

signa-105

tures,

average

hydrophilicity,

net

charge,

and

number

of

N-linked

106

glycosylation

sites

were

defined

for

the

V3

and

the

gp41.

107

2.

Materials

and

methods

108

2.1.

Sequence

analysis

109

The

analysis

included

312

HIV-1

C-subtype

env

full-length

110

sequences

and

other

44

HIV-1

C-subtype

V3

sequences,

retrieved

111

from

the

Los

Alamos

Database

(overall

from

356

infected

112

individuals

at

all

stages

of

infection,

with

one

isolate

per

single

113

patient)

(

http://www.hiv.lanl.gov

)

(

Table

S1

).

The

treatment

sta-

114

tus

for

the

individuals

is

not

available

in

the

Los

Alamos

Database.

115

The

multiple

sequence

alignments

of

V3

and

gp41

segments

were

116

performed

by

using

ClustalX

(

Thompson

et

al.,

1997

)

and

manually

117

edited

with

the

Bioedit

software

(

Hall,

1999

).

Published

env

con-

118

sensus

sequences

of

pure

HIV-1

subtypes

(A,

B,

C,

D,

F1,

F2,

G,

H,

J,

119

and

K)

were

used,

and

multi-aligned

sequences

were

subjected

to

120

phylogenetic

inference

through

the

Neighbor-Joining

method

and

121

Kimura

two-parameter

model

implemented

in

the

MEGA

4

pack-

122

age

(

Tamura

et

al.,

2007

).

One

thousand

bootstrap

replicates

were

123

used

to

assess

the

phylogenetic

robustness

of

the

clusters.

124

2.2.

Tropism

prediction

125

Within

all

356

env-sequences,

the

V3

region

was

extrapolated

126

and

submitted

for

tropism

prediction

to

Geno2Pheno

algorithm

127

(

http://coreceptor.bioinf.mpi-inf.mpg.de

)

and

to

the

Position

Spe-

128

cific

Scoring

Matrices

(PSSM)

algorithm

(

http://fortinbras.us/cgi-

129

bin/fssm/fssm.pl

)

(

Vandekerckhove

et

al.,

2011

).

130

Geno2Pheno

was

preferred

because

it

features

an

adjustable

131

cutoff.

Beyond

tropism

prediction,

it

assigns

to

each

V3

sequence

132

a

score,

called

False

Positive

Rate

(FPR),

ranging

from

0%

to

100%,

133

which

represents

the

probability

for

a

sequence

to

belong

to

a

CCR5-

134

virus.

According

to

FPR

values,

arbitrarily

we

selected

sequences

135

with

FPR

≤5%

(indicating

a

strong

CXCR4

prediction)

and

sequences

136

with

FPR

≤80%

(indicating

a

strong

CCR5

prediction)

for

CXCR4-

and

137

CCR5-tropic

viruses,

respectively.

These

sequences,

together

with

138

the

related

gp41

sequences,

were

then

used

for

all

at

the

rest

of

the

139

study.

140

For

Fortinbras

PSSM,

an

easy

and

rapid

bioinformatic

method

for

141

viral

tropism

estimation

written

by

the

original

WebPSSM

devel-

142

oper,

the

subtype-C

specific

matrix

(that

recently

was

provided)

143

was

used

(

Jensen

et

al.,

2003

).

144

2.3.

N-linked

glycosylation

motifs

prediction

145

We

assessed

the

N-linked

glycosylation

motifs

(sequons)

in

all

146

356

V3

HIV-1

C-subtype

sequences

using

the

LANL

N-glycosite

147

program

(

http://www.hiv.lanl.gov

)

(

Table

S1

).

The

sequons

were

148

governed

by

the

amino

acid

order

asparagine-X-threonine/serine-

149

Y

(N-X-S/T-Y)

(

Marshall,

1972

),

where

X

can

be

any

amino

acid

150

except

proline

(P)

in

the

threonine

(T)

context

(

Gavel

and

von

151

Heijne,

1990;

Kasturi

et

al.,

1997;

Mellquist

et

al.,

1998

)

and

also

152

not

tryptophan

(W),

aspartic

acid

(D)

or

glutamine

(E)

in

a

serine

(S)

153

context

(

Kasturi

et

al.,

1997

).

These

parameters

provided

us

with

a

154

high

probability

of

oligosaccharide

addition

(

Gavel

and

von

Heijne,

155

1990;

Kasturi

et

al.,

1997;

Mellquist

et

al.,

1998;

Shakin-Eshleman

156

et

al.,

1996

)

and

the

criteria

for

evaluating

each

sequon

as

a

possible

157

N-linked

glycosylation

site.

Sequences

exhibiting

ambiguities

in

a

158

site

were

included

from

this

calculation.

159

2.4.

V3-loop

amino

acid

physical

and

chemical

properties

160

Determinations

of

the

net

charge

and

the

average

of

161

hydrophilicity

of

the

V3-loop

for

each

sequence

at

pH

7.0

were

162

determined

using

a

desktop-based

bioinformatics

system

Peptide

163

Property

Calculator

from

Innovagen

(

http://www.innovagen.se

).

164

All

possible

permutations

were

assessed

when

amino

acid

mixtures

165

were

found

at

some

codons

of

V3.

To

compare

the

values

between

C-

166

and

B-subtype,

the

V3

B-subtype

sequences

(one

per

patient)

with

167

available

phenotypic

determination

of

HIV-1

tropism

(114

CXCR4-

168

and

582

CCR5-tropic

viruses,

respectively;

Los

Alamos

Database)

169

S.Dimonteetal./VirusResearchxxx (2012) xxx–xxx 3

values

of

hydrophobicity

and

surface

probability

of

gp120

V3-loop

171

region

were

calculated.

172

2.5.

Verification

of

tropism

prediction

173

To

further

support

the

correlation

of

V3

and

gp41

mutations

174

with

different

co-receptor

usage

and

the

correlation

among

these

175

Env

amino

acid

signatures,

all

sequences

available

from

Los

Alamos

176

Database

with

pure

phenotype

and/or

co-receptor

determinations

177

have

been

considered

(for

V3:

423

CCR5-

and

48

CXCR4-using

178

viruses,

respectively;

for

gp41:

106

CCR5-

and

19

CXCR4-using

179

viruses,

respectively)

(

Table

S1

).

180

2.6.

Statistical

analysis

181

To

analyze

gp41

and

V3

mutations,

we

calculated

the

frequency

182

of

all

mutations

in

the

353

gp41

amino

acids

and

35

V3

amino

183

acids,

using

the

env

selected

sequences.

Fisher

exact

tests

were

used

184

to

determine

whether

the

differences

in

frequency

between

the

2

185

groups

of

patients

were

statistically

significant

(isolates

with

strong

186

CCR5

and

CXCR4

prediction,

respectively).

187

The

Benjamini–Hochberg

method

has

been

used

to

iden-188

tify

results

that

were

statistically

significant

in

the

presence

of

189

multiple-hypothesis

testing

(

Benjamini

and

Hochberg,

1995

).

A

190

false

discovery

rate

of

0.05

was

used

to

determine

statistical

sig-191

nificance.

To

identify

significant

patterns

of

pairwise

associations

192

between

V3

and

gp41

mutations,

we

calculated

the

ϕ

coefficient

193

and

its

statistical

significance

for

each

pair

of

mutations.

A

positive

194

and

statistically

significant

correlation

between

mutations

at

two

195

specific

positions

(0

<

ϕ

<

1;

P

<

0.05)

indicates

that

the

latter

mutate

196

in

a

correlated

manner

in

order

to

confer

an

advantage

in

terms

of

197

co-receptor

selection

and

that

the

co-occurrence

of

these

muta-198

tions

is

not

due

to

chance.

Moreover,

to

analyze

the

covariation

199

structure

of

mutations

in

more

detail,

we

performed

average

link-200

age

hierarchical

agglomerative

clustering

(

Dimonte

et

al.,

2011b;

201

Svicher

et

al.,

2009

).

Mann–Whitney

U

tests

have

been

used

to

202

assess

statistically

significant

differences

among

all

the

pairwise

203

mutations

associated.

Statistical

tests

have

been

corrected

for

204

multiple-hypothesis

testing

by

using

Benjamini–Hochberg

method

205

at

a

false

discovery

rate

of

0.05

(

Benjamini

and

Hochberg,

1995

).

206

Using

again

the

nonparametric

Mann–Whitney

U

tests,

we

com-207

pared

the

mean

changes

in

the

mean

net

charge

and

in

the

mean

208

hydrophilicity

respectively,

in

255

CCR5-

and

101

CXCR4-using

209

viruses

V3

amino

acid

sequences.

210

3.

Results

and

discussion

211

The

genotypic

algorithms

built

from

B-subtype

virus

data

are

212

questioned

whether

they

correctly

predict

the

tropism

of

non-B

213

viruses

(

Garrido

et

al.,

2008

),

despite

recent

observations

suggest-214

ing

that

they

performed

well

for

predicting

the

tropism

of

HIV-1

215

clade

C

virus

(

Raymond

et

al.,

2010

).

Moreover,

a

study

compar-216

ing

the

predictive

performance

of

Geno2Pheno

,

PSSM

and

other

217

methods

against

the

first-generation

Trofile

®

_assay

_(validated

_for

218

HIV-1

tropism

determination),

concluded

that

the

concordance

219

being

as

high

as

91%

(

Raymond

et

al.,

2008

).

Similarly,

another

220

work

described

that

HIV-1

tropism

determination

via

plasma

viral

221

V3

RNA

genotyping

coupled

with

Geno2Pheno

interpretation

may

222

represent

a

valid

alternative

to

enhanced

sensitivity

Trofile

®

_assay

223

(

Prosperi

et

al.,

2009

).

224

In

HIV-1

B-subtype,

gp120

mutations

in

the

V3

and

V1/V2

225

domains

are

required

for

co-receptor

switching,

but

in

C-subtype

226

there

is

a

much

stronger

genetic

barrier

to

co-receptor

switching

227

that

involves

the

requirement

for

more

extensive

changes

outside

228

the

V3

region

(

Coetzer

et

al.,

2011

).

Hence,

the

contribution

of

the

229

other

gp120

regions

in

directing

co-receptor

usage

was

excluded

230

in

this

study.

231

3.1.

Physical

and

chemical

V3

properties

and

prevalence

of

V3

232

mutations

233

356

HIV-1

C-subtype

V3-containing

env-sequences

were

col-

234

lected

from

the

Los

Alamos

HIV

Sequence

Database.

Among

them,

235

312

contained

also

gp41

genome

region.

Geno2Pheno

algorithm

236

was

used

to

infer

HIV-1

co-receptor

usage

for

all

the

356

V3-

237

containing

env-sequences.

Among

them,

255

were

CCR5-using

238

(with

FPR

≥80%),

and

101

CXCR4-using

(with

FPR

≤5%).

The

predic-

239

tion

of

co-receptor

usage

was

fully

confirmed

using

both

Fortinbras

240

PSSM

algorithm,

and

the

“net

charge”

and

“11/25”

rules

(

Table

1

)

241

(

Vandekerckhove

et

al.,

2011

).

Thus,

these

3

interpretation

methods

242

for

tropism-prediction

provide

superimposable

results.

243

Previous

studies

have

shown

that

CXCR4-using

viruses

were

244

infrequently

found

in

HIV-1

C-subtype

infection

compared

to

B-

245

subtype

(

Cecilia

et

al.,

2000;

Ndung’u

et

al.,

2006;

Pollakis

et

al.,

246

2004;

Zhang

et

al.,

2006

):

thus,

this

can

explain

the

low

number

247

of

CXCR4-related

env

sequences

retrieved

and

employed

for

the

248

entire

study.

249

By

evaluating

the

V3-loop

sequences,

we

have

identified

11

250

amino

acids

at

specific

V3

positions

whose

prevalence

was

sig-

251

nificantly

higher

in

CCR5-using

than

in

CXCR4-using

viruses

(P

252

values

from

1.40E

−30

to

1.66E

−2)

(

Fig.

1

).

All

of

them

(D25D

253

and

S11S,

and

T2V,

N5N,

N6N,

N7N,

K10ET,

P16P,

G24T,

D29N

254

and

Q32E)

had

a

prevalence

≥10%

(ranging

from

12.2%

to

100%)

255

in

CCR5-using

viruses.

We

also

identified

46

amino

acids

at

spe-

256

cific

V3

positions

whose

prevalence

was

significantly

higher

in

257

CXCR4-using

than

in

CCR5-using

viruses,

suggesting

their

asso-

258

ciation

with

CXCR4-usage

(P

values

from

2.34E

−38

to

4.49E

−2).

259

Among

them,

18

(S11R

and

D25KRQ,

and

N5G,

T8KR,

K10R,

S11G,

260

13–14insIL/IG/VG,

P16RQ,

Q18KR,

T19AV,

F20ILV,

A22TV,

T23A,

261

T23HK,

G24DE,

G24KR,

I26V,

and

Q32KR)

had

a

prevalence

_≥10%

262

(ranging

from

10.9%

to

91.1%)

in

CXCR4-using

viruses,

suggesting

263

(and

mimicking

the

trend

observed

in

B-subtype)

that

within

the

264

V3

region,

much

more

mutations

are

associated

with

CXCR4

usage

265

(

Fig.

1

).

In

fact,

in

a

study

enlarged

to

flanking

V3

regions

that

used

266

samples

with

experimentally

determined

phenotype,

mutations

at

267

23

positions

within

V3

were

significantly

associated

with

HIV-1

B-

268

subtype

X4

viruses,

as

well

as

for

13

positions

in

V2

and

2

in

C4,

269

respectively

(

Thielen

et

al.,

2009

).

270

A

detailed

analysis

of

the

classical

V3

positions

11

and

25

271

showed

that

the

wild-type

amino

acid

at

positions

11

and

272

25

(S11S

and

D25D)

were

significantly

associated

with

CCR5-

273

usage

(P

=

6.77E

−10;

ϕ

=

0.41),

respectively,

while

S11GR

and

274

D25KRQ

mutations

were

significantly

associated

with

CXCR4

usage

275

(P

=

3.36E

−4;

ϕ

=

0.31)

(

Fig.

1

).

Among

the

other

mutations

found

276

at

V3

position

25

of

HIV-1

C-subtype,

the

prevalence

of

E

(wild-

277

type

for

B-subtype)

was

higher

in

CCR5-using

than

CXCR4-using

278

viruses

(15.7%

and

6.9%,

respectively,

P

=

0.071).

Conversely,

the

279

mutations

K,

N,

P,

Q,

R,

T

and

V

at

position

25

were

mainly

found

280

in

CXCR4-using

viruses

(1.2%

in

CCR5

versus

56.4%

in

CXCR4).

Only

281

the

mutations

AGS

at

position

25

had

a

similar

prevalence

in

CCR5-

282

and

CXCR4-using

viruses.

283

The

analysis

of

position

11

showed

the

complete

absence

of

284

the

Lysine

at

this

position

in

HIV-1

C-subtype

(while

S11K

is

com-

285

mon

in

HIV-1

B-subtype

CXCR4-using

viruses)

and

the

presence

of

286

glycine.

This

glycine

is

completely

absent

in

all

V3

sequences

from

287

CCR5-using

viruses,

while

it

was

observed

in

12.8%

of

CXCR4-using

288

viruses

(P

=

4.74E

−8)

(

Fig.

1

).

When

the

position

11

was

mutated

289

(47.5%)

the

corresponding

virus

was

always

CXCR4-using.

290

We

also

analyzed

the

V3

region

encompassing

the

amino

291

acids

5–8

including

the

N-linked

glycosylation

site

(N

6

XT

8

).

This

292

4 S.Dimonteetal./VirusResearchxxx (2012) xxx–xxx

Table1

V3andgp41chemico-physicalpropertiesofCCR5-andCXCR4-tropicviruses.

CCR5-usingviruses,N=255 CXCR4-usingviruses,N=101 P-valueb

Meanaveragehydrophilicitya _{0.06 0.12 <0.001}

MeannetchargeatpH7.0a _{2.85 5.32 <0.001}

NumberofV3sequenceswithoutN-linkedglycosylationsites 4(1.6%) 21(20.8%) <0.001

CCR5-usingviruses,N=255 CXCR4-usingviruses,N=57 P-valueb

Numberofgp41N-linkedglycosylationsites(=3) 7(2.7%) 4(7.0%) >0.05

Numberofgp41N-linkedglycosylationsites(=4) 180(70.6%) 36(63.1%) >0.05

Numberofgp41N-linkedglycosylationsites(≥5) 68(26.7%) 17(29.8%) >0.05

a_{ThemeanhydrophilicityandthemeannetchargewerecalculatedbyusingInnovagen’sPeptidePropertyCalculator(http://www.innovagen.se).} b _{PvalueswerecalculatedbyusingMann–WhitneyUtest(forcontinuousvariables)and}2_{test(forcategoricalvariables).}

mutations

at

position

7

have

been

shown

to

abrogate

the

binding

294

with

CCR5

co-receptor

(

Huang

et

al.,

2007

),

while

the

loss

of

the

295

glycosylated

site

has

been

associated

with

CXCR4-usage

in

both

B-296

and

C-subtypes

(

Back

et

al.,

1994;

Li

et

al.,

2001;

Losman

et

al.,

1999;

297

Malenbaum

et

al.,

2000;

McCaffrey

et

al.,

2004

).

In

our

dataset,

298

N7K

mutation

was

found

only

in

CXCR4-using

viruses

(prevalence

299

7.9%

in

CXCR4-using

versus

0%

in

CCR5-using

viruses;

P

=

5.48E

−6)

300

(

Fig.

1

).

This

suggests

that

N7K

can

be

a

CXCR4

related

marker

also

301

in

C-subtype.

In

addition,

the

loss

of

the

N-linked

glycosylation

site

302

was

observed

in

1.6%

of

CCR5-

and

20.8%

of

CXCR4-using

viruses

303

(P

<

0.001)

(

Table

1

)

(

Nabatov

et

al.,

2004;

Polzer

et

al.,

2002

).

304

Considering

the

physical

and

chemical

properties

of

CCR5-305

versus

CXCR4-using

viruses

(

Table

1

),

the

net

charge

of

CCR5-using

306

viruses

(mean

2.85,

median

3.00,

IQR

2.00–3.00)

was

significantly

307

lower

than

that

observed

in

CXCR4-using

viruses

(mean

5.32,

308

median

5.10,

IQR

4.00–6.09)

(P

<

0.001,

Mann–Whitney

U

tests),

as

309

expected

and

already

known

for

the

group

M

subtypes

(

Clevestig

310

et

al.,

2006

).

This

was

due

to

the

presence

of

increased

numbers

311

of

K

and

R

residues

that

were

scattered

throughout

the

V3

region

312

of

CXCR4-using

viruses,

including

positions

11

and

25.

Moreover,

313

we

observed

an

increase

in

the

V3

hydrophilicity

in

CXCR4-using

314

viruses

compared

to

CCR5-using

viruses

in

both

C-

(median

0.07

315

for

the

V3

sequences

from

CCR5-using

viruses,

and

median

0.13

for

316

V3

sequences

from

CXCR4-using

viruses

[P

<

0.001,

Mann–Whitney

317

U

tests])

and

B-subtype

(median

0.03

for

the

V3

sequences

from

318

CCR5-using

viruses,

and

median

0.13

for

V3

sequences

from

CXCR4-

319

using

viruses

[P

<

0.001,

Mann–Whitney

U

tests]).

The

increased

320

hydrophilicity

of

V3

sequences

from

CXCR4-using

viruses

(for

both

321

B-

and

C-subtypes)

can

be

one

of

the

potential

factors

affecting

322

the

drift

from

CCR5

to

CXCR4

tropism

in

HIV-1

C-subtype

(

Choge

323

et

al.,

2006;

Cilliers

et

al.,

2003;

McCormack

et

al.,

2002;

Ndung’u

324

et

al.,

2006

).

This

could

(at

least

in

part)

explain

tropism

changes

325

observed

in

HIV-1

C-subtype

infected

patients

who

had

progression

326

to

AIDS

during

the

pre-highly

active

antiretroviral

therapy

(HAART)

327

era

(

Connor

et

al.,

1997

).

All

these

results

are

consistent

with

previ-

328

ously

published

papers

showing

correlations

between

an

increased

329

hydrophilicity

and

net

charge

with

syncytium

inducing

ability

and

330

CXCR4

usage

(

Fouchier

et

al.,

1992;

Wang

et

al.,

1998

).

These

two

331

parameters

can

be

markers

of

tropism

changes

acting

on

secondary

332

structure

of

the

V3-loop.

333

Another

V3

region

critical

in

modulation

HIV-1

subtype

co-

334

receptor

usage

is

the

GPGQ

crown

(at

positions

15–18)

(

Coetzer

335

et

al.,

2006;

Lin

et

al.,

2011

).

This

motif

forms

a

proteic

␤-turn

and

336

specific

amino

acid

changes

have

been

shown

to

be

critical

deter-

337

minants

of

co-receptor

usage

(

Cormier

and

Dragic,

2002;

Hartley

338

et

al.,

2005;

Hu

et

al.,

2000;

Pollakis

et

al.,

2004;

Shimizu

et

al.,

339

Fig.1. FrequenciesofHIV-1gp120V3aminoacidchanges.FrequenciesofV3signaturesinHIV-1CCR5-tropicisolateswithFPR≥80%byGeno2Pheno-algorithmprediction

(darkgray)andHIV-1CXCR4-tropicisolateswithFPR≤5%byGeno2Pheno-algorithmprediction(lightgray).Theanalysiswasperformedinsequencesderivedfrom356 patients,255reportedasCCR5-tropicand101reportedasCXCR4-tropicatgenotypictest.Theco-receptorusageofthesequenceswasconfirmedusingFortinbrasPSSM algorithmandthecombinationofcriteriafromthenetchargeand“11/25”rules.Statisticallysignificantdifferenceswereassessedbychi-squaretestsofindependence.P valuesweresignificantatafalse-discoveryrateof0.05followingcorrectionformultipletests.*P<0.05,**P≤0.01,***P≤0.001.Thecodonswithablackdot(22/29)were significantandconfirmedalsousingadatasetofV3sequenceswithphenotypictropismdetermination(423CCR5-and48CXCR4-usingviruses,respectively).

S.Dimonteetal./VirusResearchxxx (2012) xxx–xxx 5

1999;

Suphaphiphat

et

al.,

2003

).

In

this

region,

the

wild-type

340

amino

acid

at

V3

position

18

is

an

R

in

B-subtype

and

a

Q

in

C-341

subtype.

In

B-subtype,

the

position

18

was

found

never

mutated

in

342

114

(90.4%)

of

CXCR4-using

viruses

(one

sequence

per

patient),

sug-343

gesting

and

sustaining

a

functional

role

for

the

wild

type

arginine

344

(

Resch

et

al.,

2001

),

possibly

predisposing

B-subtype

viruses

to

use

345

CXCR4

co-receptor.

In

C-subtype,

Q18R

was

a

frequent

mutation

346

in

CXCR4-using

viruses

(24.7%),

followed

by

Q18H

(6.9%).

Accord-347

ingly,

the

switch

to

CXCR4

usage

may

require

the

acquisition

of

348

Q18RH

in

order

to

increase

the

V3

net

charge

and/or

to

alter

the

V3

349

conformation

(

Hartley

et

al.,

2005

).

350

In

addition,

the

V3

position

18

(along

with

position

20)

resides

351

in

a

domain

shown

to

be

involved

in

the

binding

with

two

specific

352

glycosphingolipids

(GSLs):

galactosylceramide

and

sphingomyelin.

353

This

binding

has

been

shown

to

mediate

the

attachment

of

HIV-1

354

to

plasma

membrane

microdomains

(rafts)

(

Fantini

et

al.,

2002;

355

Rawat

et

al.,

2005;

Hammache

et

al.,

1998

).

Several

works

suggest

356

that

GSLs

are

involved

in

the

entry

of

a

broad

range

of

HIV-1

isolates

357

into

cell

lines

expressing

CD4,

CCR5

and/or

CXCR4,

and

that

a

GSL

358

depletion

blocked

subsequent

viral

fusion

and

infection

(

Hug

et

al.,

359

2000;

Puri

et

al.,

1998

).

Hence,

mutations

at

V3

positions

18

and

20

360

may

have

an

impact

on

HIV-1

ability

to

recognize

these

membrane

361

microdomains.

362

Additionally,

29.7%

of

HIV-1

C-subtype

CXCR4-using

viruses

had

363

an

insertion

of

2

amino

acids

between

V3

positions

13

and

14

364

(

Fig.

1

).

This

signature

has

been

observed

in

other

analysis

on

C-365

subtype

CXCR4-tropic

viruses

(

Cilliers

et

al.,

2003;

Coetzer

et

al.,

366

2006;

Raymond

et

al.,

2010;

Singh

et

al.,

2009

).

Recently,

Zhang

367

et

al.

(2010)

have

shown

that

removal

of

this

insertion

abolished

368

CXCR4

utilization

by

dual-tropic

viruses,

indicating

its

critical

role

369

in

modulating

the

binding

to

CXCR4

co-receptor.

Differently,

this

370

insertion

was

never

found

in

CCR5-using

viruses

(

Fig.

1

).

371

The

high

variability

of

the

V3-loop

is

not

surprising,

since

pos-372

itive

selection

has

been

implicated

in

the

maintenance

of

such

373

diversity

(at

individual-

as

well

as

at

population-level)

It

is

likely

374

that

the

principal

driving

force

in

the

evolution

of

HIV-1

gp120-375

V3

region

is

the

cell

receptor

usage,

the

escape

from

host

immune

376

response,

or

a

combination

of

the

two

(

Leal

et

al.,

2007;

Lemey

377

et

al.,

2007;

Ross

and

Rodrigo,

2002;

Shankarappa

et

al.,

1999;

378

Williamson,

2003;

Yang

et

al.,

2003

).

379

Additionally,

we

selected

from

Los

Alamos

Database

a

new

set

380

of

471

HIV-1

C-subtype

V3-containing

sequences

(one

sequence

381

per

patient),

with

a

phenotypic

characterization

of

HIV-1

tropism

382

(423

CCR5-

and

48

CXCR4-using

viruses,

respectively)

in

order

to

383

confirm

the

correlation

of

V3

mutations

with

different

co-receptor

384

usage.

By

using

this

“phenotypic”

dataset,

despite

the

low

num-385

ber

of

CXCR4-sequences

available,

the

majority

(22/29;

76%)

of

386

V3

signatures

identified

using

genotypic

tropism

prediction

were

387

confirmed

(

Fig.

1

).

Of

note,

in

order

to

assess

the

reliability

of

geno-388

typic

tropism

testing

in

HIV-1

C-subtype,

we

applied

Geno2Pheno

389

and

PSSM

algorithms

to

predict

the

co-receptor

usage

of

the

390

471

V3

sequences

with

phenotypically

determined

viral

tropism.

391

Geno2Pheno

and

PSSM

algorithms

were

96.2%

and

87.5%

concord-392

ant

with

the

phenotypic

determination

of

viral

tropism

and

showed

393

a

sensitivity

of

87.5%

and

87.7%,

and

a

specificity

of

95.2%

and

93.8%,

394

respectively.

These

results

support

that

genotypic

tropism

test-395

ing

can

be

a

valuable

tool

to

predict

co-receptor

usage

in

HIV-1

396

C-subtype

and

is

in

line

also

with

other

studies

(

Raymond

et

al.,

397

2010

).

398

3.2.

Prevalence

of

gp41

amino

acid

changes

399

Among

the

312

env

sequences

containing

the

V3

and

gp41

400

encoding

regions,

both

Geno2Pheno

and

PSSM

algorithms

pre-401

dicted

57

CXCR4-using

and

255

CCR5-using

viruses.

By

analyzing

402

these

C-subtype

gp41

sequences,

we

found

63

out

of

353

gp41

403

positions

significantly

associated

with

different

co-receptor

usage

404

(P

value

from

7.56E

−9

to

4.86E

−2)

(

Fig.

2

).

In

particular,

17

muta-

405

tions,

whose

prevalence

was

significantly

higher

in

CCR5-using

406

than

in

CXCR4-using

viruses,

were

identified:

16

of

them

had

a

407

prevalence

≥10%

(ranging

from

12.2%

to

34.5%)

in

CCR5-predicted

408

viruses

(A14ILV,

A22V,

R133M,

E136G,

N140L,

S154K,

K156Q,

409

N166Q,

N212I,

R221E,

F263L,

A270V,

G293KR,

S294G,

D312N,

410

and

I339LV).

Conversely,

we

identified

51

gp41

mutations

whose

411

prevalence

was

significantly

higher

in

CXCR4-using

than

in

CCR5-

412

using

viruses,

suggesting

their

association

with

the

CXCR4-usage.

413

Among

them,

19

mutations

had

a

prevalence

≥10%

(ranging

from

414

10.5%

to

71.9%)

in

CXCR4-using

viruses

(F8IL,

F11ILV,

T67A,

I84LM,

415

A96N,

S107N,

Q108L,

S125N,

N140T,

K147Q,

D148E,

T165S,

I187TV,

416

F188LMV,

A189G,

N195QR,

G220E,

Q297L,

and

I332AF)

(

Fig.

2

).

The

417

majority

of

statistically

significant

gp41

minor

variants

associated

418

with

different

co-receptor

usage

reside

within

the

Heptad

Repeat

419

1

and

2

(HR1

and

HR2)

(A22,

S23,

A30,

L34,

I37,

T67,

A71,

K77,

420

D78,

I124

S125,

R133,

E136,

N140,

K147,

and

D148),

within

the

421

cluster

I

epitope

(transiently

exposed

during

fusion)

(I84,

L91,

A96,

422

S101,

S107,

and

Q108),

and

within

the

tryptophan-rich

membrane-

423

proximal

external

region

(MPER)

(S154,

K156,

D163,

T165,

and

424

N166).

All

these

positions

are

localized

in

gp41

ectodomains

known

425

to

be

immunodominant

and

to

induce

high-titer

antibodies

in

426

the

majority

of

HIV-1-infected

individuals

(

Cheung

et

al.,

2005;

427

Cleveland

et

al.,

2003;

Hollier

and

Dimmock,

2005;

Hrin

et

al.,

2008;

428

Montero

et

al.,

2008;

Prabakaran

et

al.,

2007;

Xu

et

al.,

1991

).

The

429

fact

that

all

these

polymorphisms

are

localized

in

the

extracellu-

430

lar

domains

is

consistent

with

the

idea

that

the

gp41

may

act

as

a

431

scaffold

in

order

to

maintain

the

stability

of

the

gp120/gp41

com-

432

plex,

and

may

influence

(directly

or

indirectly)

the

viral

tropism

433

and

plausibly

other

functions

such

as

the

envelope

conformation

434

or

the

interaction

with

other

cell-surface

molecules.

Intriguingly,

435

the

localization

of

some

specific

residues

within

recognized

epi-

436

topes

may

support

the

idea

of

potential

role

in

the

modulation

of

437

antibody

recognition

in

this

viral

glycoprotein

(

Blish

et

al.,

2008;

438

Ringe

and

Bhattacharya,

2012

).

439

To

further

support

the

correlation

of

gp41

mutations

with

dif-

440

ferent

co-receptor

usage,

we

collected

from

Los

Alamos

Database

441

125

HIV-1

C-subtype

gp41-containing

env-sequences

with

a

442

phenotypic

determination

of

viral

tropism

(106

CCR5-

and

19

443

CXCR4-using

viruses,

respectively).

Despite

the

very

low

number

of

444

available

phenotypic

CXCR4-tropic

viruses,

the

trend

of

correlation

445

with

different

co-receptor

using

was

confirmed

for

the

majority

of

446

gp41

mutations

identified

using

genotypic

tropism

testing

(44/68;

447

65%).

For

27/68

(40%)

mutations

the

statistical

significance

was

448

also

confirmed

(P

<

0.05).

Probably,

the

low

number

of

CXCR4-using

449

sequences

is

a

limiting

factor

that

objectively

may

produce

a

muta-

450

tional

pattern

that

only

partially

describes

the

signatures

of

the

451

CXCR4-using

viruses.

452

3.3.

Association

among

V3

and

gp41

signatures

453

By

analyzing

the

associations

among

mutations,

statistically

454

significant

correlations

between

V3

and

gp41

signatures

in

HIV-

455

1

clade

C

were

found

(for

the

first

time

in

the

literature).

Some

456

of

these

correlations

involved

the

classical

V3

positions

11

and

457

25

(

Table

2

).

Specifically,

the

D25D

V3

showed

positive

correla-

458

tions

with

several

gp41

substitutions

(A22V,

R133M,

E136G,

N140L,

459

and

N166Q)

(

Table

2

).

All

these

mutations

correlated

with

CCR5-

460

usage

and

localized

in

gp41

ectodomain

(

Cheung

et

al.,

2005;

461

Cleveland

et

al.,

2003;

Hollier

and

Dimmock,

2005;

Hrin

et

al.,

2008;

462

Montero

et

al.,

2008;

Prabakaran

et

al.,

2007;

Xu

et

al.,

1991

).

Con-

463

versely,

S11S

V3

established

a

positive

correlation

with

only

one

464

gp41

mutation

(S154K)

localized

in

HR2

domain

(

Table

2

).

Among

465

positive

correlations

between

V3

and

gp41

signatures

associated

466