CD4+ cell-dependent granuloma formation inhumanized mice infected with mycobacteriaFrank Heutsa, Dolores Gavier-Widénb, Berit Carowa, Julius Juarezc, Hans Wigzella, and Martin E. Rottenberga,1

aDepartment of Microbiology, Tumor and Cell Biology and Center for Infectious Medicine, Karolinska Institutet, 171 77 Stockholm, Sweden; bDepartmentof Pathology and Wildlife Diseases, National Veterinary Institute and Department of Biomedical Sciences and Veterinary Public Health, Swedish Universityof Agricultural Sciences, 75189 Uppsala, Sweden; and cDepartment of Medicine, Karolinska Institutet, 171 77 Stockholm, Sweden

Edited by Tak W. Mak, Campbell Family Institute for Breast Cancer Research, Ontario Cancer Institute at Princess Margaret Hospital, University HealthNetwork, Toronto, ON, Canada, and approved March 8, 2013 (received for review November 16, 2012)

We have used humanized mice, in which human immune cellsdifferentiate de novo from transplanted cord blood progenitorcells, to study the human immune responses to infection withMycobacterium bovis bacillus Calmette–Guérin andMycobacteriumtuberculosis. Granulomas with a core containing giant cells, humanCD68+ macrophages, and high bacilli numbers surrounded by alayer of CD3+ T cells and a fibrotic response encapsulating thelesions were observed in livers and lungs from bacillus Calmette–Guérin-infected humanizedmice but not in nonhumanized infectedcontrols. Paradoxically, humanized mice contained higher myco-bacterial numbers in organs than nonhumanized controls. The en-hancement of bacterial load was mediated by human CD4+ cellsand associated to an increased expression of Programmed Death-1 protein and CD57 on T cells, molecules associated with inhibitionand senescence. The lesions from mice depleted of CD4+ cells werescarcer, minimal, and irregular compared with those from mice de-pleted of CD8+ cells or nondepleted controls. Granulomas of bacil-lus Calmette–Guérin-infected humanized mice administered witha TNF-neutralizing TNF receptor fusion molecule preserved theirstructure, but contained higher levels of intracellular bacilli. Ex-tended necrosis was observed in granulomas fromM. tuberculosis-but not bacillus Calmette–Guérin-infected humanized mice. Ourdata indicate that humanized mice can be used as a model to studythe formation and maintenance of human granuloma in tubercu-losis and other infectious or noninfectious diseases.

BCG | IFN-γ | TNF-α | CD34

Approximately one third of the human population is infected byM. tuberculosis, of whom 5 to 15% develop active tuberculosis

(TB) during their lifetime (1). Mouse models have contributedto our understanding of the pathogenesis and immunology ofTB. However, the relative hierarchy of immune-mediated killingmechanisms in humans is still unclear. Moreover, the mouse modelfails to form granulomas that reproduce those seen in humans (2).Typical mouse lesions lack central necrosis and are less organizedthan TB lesions found in immune-competent humans. On thecontrary, experimental studies of human immunity tomycobacterialinfections are limited by obvious practical and ethical restrictions.The host counters mycobacterial infections primarily via CD4+

T helper1 (Th1) cell-mediated immune responses involving cel-lular effector mechanisms resulting in macrophage activation. Inmurine models of mycobacterial infection, CD4+ T cells have beenshown to be required for the formation of granulomas (3).TNF, a major regulator of macrophage activation, apoptosis,

chemokine and cytokine production, and cellular recruitment viatransendothelialmigration, is critical for host responses to infectionwithM. tuberculosis, including granuloma formation in mice (4–6).Nonobese diabetic NOD/SCID/γcR

−/− (NSG) mice and BALB/cRag2−/−/γcR

−/− (BRG) engrafted with human CD34+ hematopoi-etic stem cells show a de novo differentiation of myeloid andlymphoid cell populations (7–9). These humanized mice have beenshown to be useful tools to examine different microorganisms thatinvolve human hematopoietic cells in their life cycles (10).Here we studied the outcome of infection with attenuated

Mycobacterium bovis bacillus Calmette–Guérin and virulent

M. tuberculosis, pathogens that infect myeloid cells, in humanizedmice. Humanized mice infected with bacillus Calmette–Guérinshowed dysfunctional T-cell responses and lack of bacterialcontrol. Notwithstanding abnormal T-cell responses, mice showorganized granulomas containing human T cells and macro-phages that resemble lesions observed in human TB. TNF con-trolled bacterial load within granulomas as well as the severity ofinflammation, and human CD4+ but not CD8+ T cells were re-quired for granuloma formation in humanized mice.

ResultsOutcome of Infection with M. bovis Bacillus Calmette–Guérin inHumanized Mice. Confirming previous reports (11, 12), bloodfrom NSG mice 10 wk after inoculation of CD34+ cells containeda higher frequency of human CD45+ cells compared with BRGmice treated or not with the myeloablative compound busulfanto increase engraftment efficiency (Fig. S1A). The frequency ofCD3+ within CD45+ cells was also higher in NSG than in BRGmice (Fig. S1B). NSG mice were thus chosen as recipients in ourhumanized model. Spleen and lymph nodes from NSG miceshowed CD4+ and CD8+ T cells and naive mature B cells (Fig.S1 C–F). Splenic T cells from humanized mice proliferated andsecrete cytokines when stimulated with a mitogen or in an allo-geneic mixed lymphocyte reaction (MLR) as previously shown (8).Humanized mice were infected i.v. with 106 M. bovis bacillus

Calmette–Guérin and killed 4 wk after infection. Whereas thefrequency of CD45+ and CD3+ cells in spleens from infected anduninfected mice remained similar (Fig. S2 A and B), the CD4+/CD8+ T-cell ratio was higher (Fig. S2 C and D). The majority ofCD4+ and CD8+ T cells differentiated from a naive (CD45RA+

CCR7+) into an effector-memory (CD45RA−CCR7−) phe-notype after infection (Fig. S2 E–G). The frequency of IFN-γ–and/or TNF-secreting CD4+ and CD8+ spleen T cells in responseto a polyclonal stimulation was higher in infected comparedwith uninfected mice (Fig. S2 H–K).The percentage of human CD45+ cells in lungs and livers and

the frequency of CD3+ within CD45+ human cells sharply in-creased after bacillus Calmette–Guérin infection (Fig. 1 A, B, E,and F). In contrast to observations in spleens, the CD4+/ CD8+

lung T-cell ratio in infected or uninfected mice was similar (Fig.1 C, D, G, and H).Surprisingly, higher bacterial loads were observed in the lung

and liver of humanized mice compared with nontransplantedcontrols (Fig. 1 I and J). We speculated that T-cell dysfunctionscould account for the lack of T-cell control of bacterial growth.The engagement of Programmed Death-1 (PD-1) protein onactivated T-cells down-regulates T-cell function (13). We found

Author contributions: F.H. and M.E.R. designed research; F.H., D.G.-W., and B.C. per-formed research; F.H., D.G.-W., B.C., J.J., and M.E.R. analyzed data; and H.W. and M.E.R.wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. E-mail: martin.rottenberg@ki.se.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1219985110/-/DCSupplemental.

6482–6487 | PNAS | April 16, 2013 | vol. 110 | no. 16 www.pnas.org/cgi/doi/10.1073/pnas.1219985110

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0

that the percentage and expression levels of PD-1–expressingCD4+ and CD8+ lung or spleen cells increased after bacillusCalmette–Guérin infection (Fig. 1 K, L, O, and P and Fig. S3 A–D). Also, CD57 expression in T lymphocytes has been recognizedas a marker of in vitro replicative senescence (14). The percentageof CD57+ within CD4+ (but not in CD8+ T cells) was higher inlungs and spleens from infected humanized mice (Fig. 1 M, N, Q,and R). The levels of expression of CD57 in infected and un-infected T cells were similar (Fig. S3 E–H).Whether a higher bacterial uptake or intracellular growth in

human compared with mouse macrophages associates with thehigher susceptibility of humanized mice to bacillus Calmette–Guérin infection was then studied. Bacillus Calmette–Guérinuptake by human macrophages was slightly higher in human thanin mouse macrophages (Fig. S4A). Human macrophages alsocontained higher bacterial levels than mouse macrophages at4 d after infection. The titers of IFN-γ–regulated human GP91

PHOX and INOS mRNA that code for molecules that mediatemycobacterial control by phagocytes were similar in organs frominfected and control mice (Fig. S4 B–E).

Granuloma Formation in Bacillus Calmette–Guérin-Infected HumanizedMice.The histopathological features of the lungs and livers frombacillus Calmette–Guérin-infected humanized mice were thenstudied. Livers showed rather typical tuberculoid round-shapedgranulomas composed of a core of human CD68+ macro-phages, surrounded by CD3+ T cells (Fig. 2 A, E, and F). Theorganized structure of the granuloma was less obvious in lungs(Fig. 2B). Granulomas were observed in neither uninfectedhumanized mice nor bacillus Calmette–Guérin-infected non-humanized NSG mice (Fig. 2 C and D). A rim of fibroblasts anda collagen mantle surrounded the granulomas (Fig. 2 A and G).

CD57

0

20

40

60

80

100 *

%P

D-1 in

CD

4+

104

105

106

107 ***

CF

U/ lu

ng

104

105

106

107 ***

CF

U/ liver

0

20

40

60

80

100*

% C

D45

0

20

40

60

80

100*

%C

D3 in

CD

45

+

0

20

40

60

80

100

%C

D4 in

CD

3+

0

20

40

60

80

100

%C

D4 in

CD

3+

0

20

40

60

80

100 *

%C

D3 in

CD

45

+

0

20

40

60

80

100 **

% C

D45

0

5

10

15 *

%C

D57 in

CD

4+

0

10

20

30

40 *

%C

D57 in

CD

4+

0

2

4

6

8

10

%C

D57 in

CD

8+

0

5

10

15

%C

D57 in

CD

8+

0

20

40

60

80

100 *

%P

D-1 in

CD

8+

0

20

40

60

80

100 *

%P

D-1 in

CD

4+

0

20

40

60

80

100 *

%P

D-1 in

CD

8+

PD-1

A B C

K L M N

O P Q R

0

20

40

60

80

100

%C

D8 in

CD

3+

0

20

40

60

80

100 *

%C

D8 in

CD

3+

D

E F G H

I J

Uninfected InfectedL

un

gL

iver

Lung LungSpleen Spleen

CD

4C

D8

Non-humanized Humanized

Uninfected Infected

Fig. 1. Defective mycobacterial control in humanized mice. Frequencies ofCD45+ cells (A and E), of CD3+ within CD45+ cells (B and F), and of CD4+ (Cand G) and CD8+ (D and H) within CD3+ cells in lungs and livers of bacillusCalmette–Guérin-infected and control humanized mice were analyzed byFACS. The mean and SEM for each group are depicted. Differences betweengroups were significant (*P < 0.05 and **P < 0.01, Student t test). The cfucounts in lungs (I) and livers (J) from humanized or nonhumanized mice weredetermined 4 w after infection with bacillus Calmette–Guérin. The hori-zontal bar indicates the median value. Differences between groups aresignificant (*P < 0.05, **P < 0.01, and ***P < 0.001, Mann–Whitney U test).The percentage of PD-1+ (K, L, O, and P) and CD57+ (M, N, Q, and R) CD4+ orCD8+ cells in lungs (K, M, O, and Q) or spleens (L, N, P, and R) from bacillusCalmette–Guérin-infected or uninfected controls are depicted. The mean andSEM for each group (n = 5) are depicted. Differences between groups aresignificant (*P < 0.05, Student t test).

Non humanized infected

Humanized infected

A

CD3

Sirius red

E F CD68

I

G

AFB

Non humanized infected

Humanized infected

Humanized uninfected

J AFB

H

B

Non humanized infnn eff ctett dee

Humanized infnn eff ctett de

Lung Liver0

100

200

300

400

500#

infla

mm

ator

y fo

ciNo

Hu

Hu

0

10

20

30

40

50

Affe

cted

are

a %

Humanized uninfected

C D

Fig. 2. Granuloma formation in bacillus Calmette–Guérin-infected human-ized mice. Inflammatory lesions in liver (A) and lung (B) from humanizedmice 4 wk after infection with bacillus Calmette–Guérin observed in sectionsstained with H&E. The number of inflammatory foci in livers and lungs (C)and the percentage of total lung area occupied with lesions in infectedhumanized mice are compared with uninfected humanized mice or infectednonhumanized controls (D). The mean ± SEM in at least five mice per groupare depicted. The distribution of CD3+ T cells (E) and CD68+macrophages (F)in liver granulomas was determined by immunohistochemistry. A colla-genic mantle surrounding the granuloma was visualized by using Siriusred staining (G). (H) Micrograph of a giant cell within a granuloma in anH&E-stained section. Localization of acid-fast bacilli identified by Ziehl–Nielsen staining (arrows) within the granuloma in the liver of humanizedmice at magnifications of 400× (I) and 1,000× (J).

Heuts et al. PNAS | April 16, 2013 | vol. 110 | no. 16 | 6483

IMMUNOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0

Multinucleated giant cells characteristic of human granulomaswere also present (Fig. 2H). Few to moderate numbers of apo-ptotic and necrotic cells were also observed in the granulomas ofhumanized mice. Acid fast staining of bacteria in tissue sectionsindicated that bacillus Calmette–Guérin preferentially localized assingle cells within the granuloma (Fig. 2 I and J and Fig. S4F).Moderate to abundant numbers of mouse CD45+ cells werepresent mainly in the periphery of the granulomas and did notcolocalize with bacillus Calmette–Guérin (Fig. S4 G and H).Transcript levels of human IFN-γ as well as of IFN-γ–regulated

T-cell chemotactic molecules CXCL9 and CXCL10 were increasedin the lungs and livers of bacillus Calmette–Guérin-infected hu-manized mice compared with those from noninfected animals(Fig. 3 A–F). A trend to increased titers of monocyte chemotactic

CCL2 mRNA was observed in livers from infected mice(Fig. 3G).

Role of CD4+ Cells and TNF in Granuloma Formation in BacillusCalmette–Guérin-Infected Humanized Mice. Whether CD4+ andCD8+ cells participated in granuloma formation was then stud-ied. The administration of anti-CD4 and anti-CD8 mAbs beforeand during bacillus Calmette–Guérin infection reduced the re-spective T-cell population (Fig. S5 A–F). Livers from CD4+ cell-depleted mice showed fewer, minimal, and irregular foci, con-sisting of macrophages and some neutrophils, whereas CD3+

cells were occasional or absent (Fig. 3 H–K). On the contrary,CD8+ cell-deficient humanized mice retained the granulomanumbers and their complex structure and size (Fig. 3H). Few

Uninfected

Anti-CD4 Anti-CD8 Control

0

1

2

3

4

****

Seve

rity

scor

e

1

10

100 ***

0.1

1

10

100 **

0.1

1

10

100

**

0.1

1

10

100 **

1

10

100

1000**

0.1

1

10

100

***

B C

D E F

Uninfected Infected

Lu

ng

Liv

er

CD3 CD68

0

1

2

3

4 ** *

Scor

e0

1

2

3

4

Fold

cha

nge ccl2

mR

NA

A

G H

IFN- CXCL9 CXCL10

Fo

ld c

han

ge m

RN

A

I

104

105

106

107 *** **

******

CF

U/ lu

ng

104

105

106

107 ***

**** *

CF

U/ liver

non humanized humanized humanized + CD4 humanized + CD8

L M

Anti-CD4ControlJ K

Control Anti-CD4Infected

Fig. 3. Human CD4+ cells are required for granuloma formation in bacillusCalmette–Guérin-infected humanized mice. The mean levels of IFN-γ (A and D),CXCL9 (B and E), CXCL10 (C and F), and CCL2 (G) mRNAnormalized to human β-2microglobulin housekeeping gene (β2M) transcriptsweremeasuredby real-timePCR in lungs (A–C) and livers (D–G) from bacillus Calmette–Guérin-infected oruninfected humanized mice. Differences between groups are significant (*P <0.05, **P < 0.01, and ***P < 0.001, Student t test). Humanized mice were in-oculated i.p. with 100 μg anti-human CD4 or 50 μg anti-CD8 mAbs for threeconsecutive days starting 3 d before and 14 d after bacillus Calmette–Guérininoculation. The mean histopathological scoring of H&E-stained liver sections(H) and the relative score of CD3+ or CD68+ cells in liver granulomas frommAb-treatedor control humanizedmice are depicted (I). Differences between groupsare significant (*P < 0.05, **P < 0.01 and ***P < 0.001 Student t test). Micro-graphs of H&E-stained liver sections from control (J) or CD4+ cell-depleted (K)humanized mice 4 wk after inoculation. Arrows indicate granulomas. The cfucounts in lungs (L) and livers (M) fromhumanizedmice treatedwith anti-CD4 oranti-CD8mAbs or nonhumanizedNSGmicewere determined 4wkafter bacillusCalmette–Guérin infection. Quartile boxes and 10th to 90th percentile whiskersare depicted. Differences between groups are significant (*P < 0.05, **P < 0.01,and ***P < 0.001, Mann–Whitney U test).

0

1

2

3

4

5 *Se

verit

y sc

ore

0

1

2

3

4

5

Seve

rity

scor

e0

1

2

3 *

AFB

scor

e

0

1

2

3 *

AFB

scor

e

EtanerceptControl

Liver

106

107

108 **

CFU

/ lun

g

106

107

108 *

CFU

/ liv

er

Etanercept Control

A B

E F

G H

I J

0.1

1

10

100

1000 **

Fo

ld c

ha

ng

e m

ou

se

TNF-

mR

NA

0.1

1

10

Fo

ld c

ha

ng

e h

um

an

TNF-

mR

NA

Uninfected Infected

C D

Control Etanercept

104

105

106

107

CFU

/ lun

g

105

106

107

108

CFU

/ liv

er

K LHumanized Non-humanized

LiverLung Lung

Fig. 4. Effect of treatment with the TNF inhibitor etanercept in the out-come of infection of humanized mice with bacillus Calmette–Guérin. The meanlevels of human (A) and mouse (B) TNF-α mRNA normalized to human β2Mtranscripts were measured by real-time PCR in liver from bacillus Calm-ette–Guérin-infected or uninfected humanized mice. Differences betweengroups are significant (**P < 0.01, Student t test). Mice were treated i.p. with0.125 mg/kg etanercept every other day starting 1 d before bacillus Calm-ette–Guérin infection during 4 wk. Themean severity of inflammation score inH&E-stained liver (C) and lung (D) sections from etanercept-treated or controlhumanized mice are depicted (n > 6 mice per group). The mean acid-fast bacillidensity score in Ziehl–Nielsen-stained liver (E) and lung (F) sections from eta-nercept-treated or control humanized mice are depicted (n > 6 mice pergroup). Representative micrographs showing the acid fast bacteria (AFB) den-sity and their localization within the granuloma of from liver sections treatedwith etanercept or not treated are shown (G and H). The cfu counts in livers(I and K) and lungs (J and L) from humanized (I and J) or nonhumanizedNSG (K and L) mice treated with etanercept were determined 4 wk after bacillusCalmette–Guérin infection. Quartile boxes and 10th to 90th percentile whiskersare depicted. Differences between groups are significant (*P < 0.05, **P < 0.01,and ***P < 0.001, Mann–Whitney U test).

6484 | www.pnas.org/cgi/doi/10.1073/pnas.1219985110 Heuts et al.

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0

human CD8+ cells were found within the granulomas of non-depleted humanized mice (Fig. S5G).Bacillus Calmette–Guérin levels in lungs or livers of anti-CD4–

treated (but not anti-CD8–treated) humanized mice were reducedcompared with nontreated humanized mice (Fig. 3 L and M).TNF is required for granuloma formation in mice infected with

mycobacteria (4). Humanized mice showed increased levels ofhuman and mouse TNF-α mRNA after infection with bacillusCalmette–Guérin (Fig. 4 A and B). Therefore, we studied theeffect of administration of the TNF inhibitor etanercept ongranuloma formation in bacillus Calmette–Guérin-infected hu-manized mice. Etanercept, a recombinant TNF p75 receptorcontaining protein that inhibits the signaling via soluble ormembrane bound TNF (15), is used for treatment of differentautoimmune diseases (16). Spleens from etanercept-treatedinfected mice showed a slightly increased CD4+/ CD8+ cell ratio,but similar proportions of CD45+ and CD3+ cells as controls (Fig.S6A–D). The frequencies of CD45RA-, CCR7-, CD57-, or PD-1–expressing CD4+ or CD8+ cells in spleens from etanercept-trea-ted and untreated infected mice were similar (Fig. S6 E–J).Livers (but not lungs) from etanercept-treated humanized mice

displayedmilder inflammatory responses (Fig. 4C andD), whereasthe histological structure of the granuloma was preserved.Bacterial aggregates were observed in granulomas from eta-

nercept-treated mice, whereas granulomas from control miceshowed lower numbers of bacteria usually as single cells orsmaller aggregates (Fig. 4 E–H). Accordingly, livers and lungsfrom etanercept-treated humanized mice contained higher cfulevels than controls (Fig. 4 I and J), whereas bacterial levels inetanercept-treated and control nonhumanized NSG mice weresimilar (Fig. 4 K and L). Levels of TNF-α mRNA in livers fromcontrol and anti-CD4–treated or anti-CD8–treated infected micewere similar (Fig. S6K).

Granuloma Formation in M. tuberculosis-Infected Humanized Mice.The outcome of aerosol infection with 250 M. tuberculosis bac-teria in humanized mice was studied next. Similar to observationsin mice infected with bacillus Calmette–Guérin, IFN-γ, CXCL9,and CXCL10 mRNA levels in lungs and livers from humanizedmice were elevated after M. tuberculosis infection (Fig. 5 A–F).Bacterial levels in humanized or control NSGmice were strikinglyincreased compared with those in bacillus Calmette–Guérin-infectedmice (more than 3 and 1 log in lungs and livers respectively).Higher bacterial titers were quantified in the liver from humanizedmice compared with nonhumanized controls (Fig. 5G).Nonhumanized mice infected with M. tuberculosis (but not

with bacillus Calmette–Guérin) showed granulomatous lesions inlivers (Fig. 5K and Fig. S7A). However, the pathology was sig-nificantly more severe in humanized mice (Fig. 5 H and J andFig. S7 B and C). Small and irregular microscopic granulomatouslesions formed by macrophages and epithelioid cells were ob-served in livers from NSG control mice. Instead, humanized miceshowed small and large macroscopic lesions, the latter irregularor round in shape. The large lesions showed CD3+ cells in theperiphery (Fig. 5L) and were occasionally surrounded by a col-lagen layer (Fig. S7D). Giant cells were observed in the centerof the granulomas. An extensive necrosis was present in liverlesions from humanized but not in nonhumanized mice (Fig. 5 H,J, and K and Fig. S7 A and B).Humanized and control mice showed similar severity of lesions

in the lungs. However, lung granulomas from humanized miceshowed increased necrotic areas compared with nonhumanizedcontrols (Fig. 5I and Fig. S7 E and F).

DiscussionHere we demonstrate that, despite impaired bacterial control,humanized mice showed organized granulomas when infectedwith bacillus Calmette–Guérin or M. tuberculosis. Human CD4+

but not CD8+ cells were required for granuloma formation inmycobacteria-infected humanized mice. The control of myco-bacteria in granulomas and the severity of inflammatory responses

were reduced by inoculation of a TNF inhibitor, but the granu-loma structure was preserved.The majority of CD4+ and CD8+ T cells from mycobacterial-

infected humanized mice differentiated into an effector-memoryphenotype, resembling the generalized immune activation observedin homeostatic peripheral T-cell expansion during lymphopeniccondition, which is accompanied by a decreased threshold forT-cell receptor activation (17). Surprisingly, higher bacteriallevels were detected in humanized mice compared with controls,which could be reverted by depletion of human CD4+, but notCD8+, cells.The majority of T cells from bacillus Calmette–Guérin-infected

humanized mice also expressed high levels of PD-1. PD-1 limitsproliferation, increases apoptosis, and interferes with effectorfunctions of T cells against infection with M. tuberculosis in man(18). Increased expression of CD57, a marker of T-cell clonalexhaustion, was also detected in CD4+ cells of infected human-ized mice. The augmented expression of PD-1 and CD57 prob-ably reflect dysfunctional T-cell responses in infected humanizedmice that might underlie their defective mycobacterial control.The granuloma, which is the classic pathological feature of TB, is

the niche in which the bacillus can grow or persist, and the micro-environment in which immune cells interact to prevent mycobac-terial dissemination (19). The strength of our humanized model liesin the formation of granulomas that resemble those observed inhuman mycobacteriosis. Granulomas showed a core with largenumbers of human CD68+macrophages, giant multinucleated cells,and higher density of bacilli compared with that in surroundingtissues. A layer of lymphocytes and fibroblasts surrounded the core.Particularly in the liver, these lesions were more organized andsphere-like than those formed during mouse M. tuberculosis in-fection (20). The accumulation of fibroblasts and a collagen capsuleoccurs in human benign evolution ofM. tuberculosis infection (21).Consistent with granuloma formation, organs from infected

humanized mice showed a dramatic accumulation of humanCD45+ cells and an augmented expression of IFN-γ, CXCL9,and CXCL10, as well as CCL2, chemokines shown to participatein the formation of granulomas in mouse models (22, 23).CD4+ but not CD8+ cells were required for granuloma forma-

tion in bacillus Calmette–Guérin-infected humanized mice. Simi-larly, inMhcII−/− orCd4−/−mice, CD4+T cells have been shown tocontribute to the organization of granulomas in M. tuberculosis-infected mice (24, 25).Whether CD4+ T cells regulate granuloma formation in hu-

manized mice results from the cognate interactions of antigen-specific cells is uncertain. Low-level T-cell responses against viralinfections were induced in some cases (8, 26), but are absent inother reports in humanized mice (27, 28). In our model, T cells arepositively selected by mouse MHC and might not function wellin a HLA-restricted manner (29). Non–mycobacterial-specificT cells might account for granuloma formation in humanized micebecause (i) the adoptive transfer of monoclonal, nonbacterialspecific CD4+ T cells restores the granulomatous response tobacillus Calmette–Guérin infection of Rag1−/− mice, but not thebacterial control (30); and (ii) mycobacteria-specific T cells arehardly more arrested than T cells of other specificities in thegranuloma (31).The treatment of humanized but not control mice with the TNF

inhibitor etanercept resulted in a larger number of bacilli andreduced severity of inflammation, confirming studies on the roleof TNF in mycobacterial containment in conventional mousemodels (5, 32). However, in contrast to murine models (32), thegranuloma structure was preserved in etanercept-treated hu-manized mice. TNF neutralization in nonhuman primates resul-ted in disseminated TB but a normal granuloma structure (33).Thus, although the human transplant is required for granulomaformation during bacillus Calmette–Guérin infection, TNF con-trols bacterial levels in the granuloma of humanized mice.Of importance, human and mouse TNF and TNF receptors

are functionally cross-reactive (34), and etanercept also inhibitsmouse TNF (35). Human and murine TNF-α mRNA levels were

Heuts et al. PNAS | April 16, 2013 | vol. 110 | no. 16 | 6485

IMMUNOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0

increased in organs from humanized bacillus Calmette–Guérin-infected mice. However, the titers of bacillus Calmette–Guérinin the liver or lung from NSG mice inoculated with etanerceptwere similar, suggesting that etanercept inhibits human TNF inhumanized mice. Accordingly, M. tuberculosis-infected mice trea-ted with higher concentrations of etanercept than described hereshowed similar bacterial load as controls (36).Despite the expression of IFN-γ, levels of human INOS and

GP91PHOX mRNA did not increase after infection. Thus, theexpression of iNOS or gp91phox in humanized mice has differentrequirements than that of CXCL9 or CXCL10, which were up-regulated after infection. In relation, treatment of patients with TBwith aerosolized IFN-γ increased CXCL10 but not iNOS levels inlung cells (37). The failure to induce these microbicidal mechanisms,and the superior bacterial uptake by human compared with mousemacrophages, might also underlie the lack of protection againstbacillus Calmette–Guérin infection in humanized mice.Nonhumanized mice infected with M. tuberculosis showed

granulomatous lesions, whereas these were not observed in

bacillus Calmette–Guérin-infected mice. An accelerated granu-loma formation in response to M. tuberculosis compared withbacillus Calmette–Guérin has been described (38). RD1, a viru-lence region that is absent in bacillus Calmette–Guérin, has beenshown to participate in granuloma formation (39, 40). Aggregatesof epithelioid macrophages have been previously observed inmycobacterial-infected SCIDmice (41, 42). However, humanizedmice showed granulomas with a more organized structure and anincreased severity of the size and area occupied by the lesions.Large necrotic areas were observed in lesions from M. tubercu-

losis-infected humanized mice but not in bacillus Calmette–Guérin-infected humanized mice orM. tuberculosis-infected nonhumanizedmice. The necrosis was coagulative rather than liquefactive, andthe dead tissue architecture was preserved. Supporting our data,M. tuberculosis stimulates necrosis, allowing viable bacilli to es-cape from host cells, after which they can infect new cells andelicit inflammation (43–45). T cells are required for necrosis ingranulomas ofMycobacterium avium-infected mice (46). However,

0.1

1

10

100

1000 **

0.1

1

10

100 ***E F

0.1

1

10

100 ***

0.1

1

10 **

IFN

106

107

108

109

1010

Lung Liver

*

CF

U/o

rg

an

0.01

0.1

1

10

100

1000 *

0.1

1

10

100 ***

CXCL9 CXCL10

Fold

cha

nge

mR

NALu

ngLi

ver

Severity Necrosis Cellularity

0

1

2

3

4

5 * **

Sc

ore

Severity Necrosis Cellularity

0

1

2

3

4

5 * *

Sc

ore

LungLiver

A B C

D

G H I

Humanized Non-humanizedKJ

L CD3

InfectedUninfected

Non-humanized Humanized

Fig. 5. Granuloma formation in M. tuberculosis-infected humanized mice. Total RNA was extractedfrom lungs (A–C) and livers (D–F) from humanizedmice 4 wk after M. tuberculosis infection or fromuninfected controls. The IFN-γ (A and D), CXCL9 (Band E), and CXCL10 (C and F) mRNA levels normal-ized to human β2M were assessed by real-time PCR.The horizontal lines are the mean for each group.Differences between groups are significant (*P <0.05, **P < 0.01, and ***P < 0.001, Student t test).Quartile boxes and 10th to 90th percentile whiskersdepict cfu counts in organs from humanized mice4 wk after infection with M. tuberculosis (G). Dif-ferences between groups (n ≥ 8) are significant(*P < 0.05, Mann–WhitneyU test). Histopathologicalscoring of H&E-stained liver (H) and lung (I) sectionsfrom humanized and control mice infected withM. tuberculosis. The score of severity of lesions, thepresence of necrosis, and macrophage cellularity inthe granulomas in lung and liver are depicted. Dif-ferences between groups are significant (*P < 0.05and **P < 0.01, Student t test). H&E staining showsrepresentative inflammatory lesions in the liverfrom humanized (J) and control NSG (K) mice 4 wkafter infection with M. tuberculosis. (Magnification:400×.) (L) Micrograph of CD3+ stained cells in theperiphery of a liver granuloma fromM. tuberculosis-infected humanized mouse.

6486 | www.pnas.org/cgi/doi/10.1073/pnas.1219985110 Heuts et al.

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0

whether human T cells mediate necrosis in granulomas fromM. tuberculosis-infected humanized mice remains to be studied.Granuloma initiation has been interpreted as a host-protective

event, to provide the microenvironment in which specific T cellsactivate macrophages to contain M. tuberculosis infection (20). Onthe contrary, our results suggest that dysfunctional CD4+ T cellsparticipate in granuloma formation in humanized mice and mayfacilitate bacterial growth. Whether such a conclusion can begeneralized for the natural course of infection requires furtherstudies. Data in a zebrafish model suggest that mycobacteria exploitthe granuloma for local expansion and systemic dissemination (47).Thus, the humanized mouse model can be used to study the

role of cytokines, chemokines, different immune cells, the effectof HIV coinfection, host genetics, and bacterial componentsin mycobacterial granuloma formation. We also propose the useof humanized mice as a model for other infectious and non-infectious granulomatous diseases.

Materials and MethodsAll animal experiments were conducted in accordance with guidelines ofKarolinska Institute, and approved by Stockholm’s District Ethical Committeeof Animal Research. The procedures for generation and infection of hu-manized mice with mycobacteria are described in Supporting Information.The quantification of transcripts in organs by real-time PCR, the staining ofsurface receptors and intracellular cytokines, the histopathological analysis,and the generation of human and mouse macrophages are explained in SIMaterials and Methods.

ACKNOWLEDGMENTS. We thank Dr. Francesca Chiodi, Dr. Antonio Rothfuchs,and Dr. Camille Locht for comments on the manuscript; Ms. Margareta E.Andersson and Mr. Kenth Andersson for excellent technical assistance; and Ms.Ewa Westergren for the preparation of histological slides. This work wassupported by a European Community 200732 HOMITB (Host MycobacterialInteractions in Immunity and Pathogenesis of Tuberculosis) Grant, EuropeanCommunity Marie Curie HIV-TB (HIV and Tuberculosis infections) project, theKarolinska Institutet, and the Swedish Research Council.

1. World Health Organization (2010) Global Tuberculosis Control 2010. Available at http://www.who.int/tb/publications/global_report/en/index.html. Accessed November 1, 2012.

2. North RJ, Jung YJ (2004) Immunity to tuberculosis. Annu Rev Immunol 22:599–623.3. Saunders BM, Frank AA, Orme IM, Cooper AM (2002) CD4 is required for the de-

velopment of a protective granulomatous response to pulmonary tuberculosis. CellImmunol 216(1-2):65–72.

4. Flynn JL, et al. (1995) Tumor necrosis factor-alpha is required in the protective im-mune response against Mycobacterium tuberculosis in mice. Immunity 2(6):561–572.

5. Bean AG, et al. (1999) Structural deficiencies in granuloma formation in TNF gene-targeted mice underlie the heightened susceptibility to aerosol Mycobacterium tu-berculosis infection, which is not compensated for by lymphotoxin. J Immunol 162(6):3504–3511.

6. Flynn JL (2006) Lessons from experimental Mycobacterium tuberculosis infections.Microbes Infect 8(4):1179–1188.

7. Ishikawa F, et al. (2005) Development of functional human blood and immune sys-tems in NOD/SCID/IL2 receptor gamma chain(null) mice. Blood 106(5):1565–1573.

8. Traggiai E, et al. (2004) Development of a human adaptive immune system in cordblood cell-transplanted mice. Science 304(5667):104–107.

9. Hiramatsu H, et al. (2003) Complete reconstitution of human lymphocytes from cordblood CD34+ cells using the NOD/SCID/gammacnull mice model. Blood 102(3):873–880.

10. Legrand N, Weijer K, Spits H (2006) Experimental models to study development andfunction of the human immune system in vivo. J Immunol 176(4):2053–2058.

11. Brehm MA, et al. (2010) Parameters for establishing humanized mouse models tostudy human immunity: Analysis of human hematopoietic stem cell engraftment inthree immunodeficient strains of mice bearing the IL2rgamma(null) mutation. ClinImmunol 135(1):84–98.

12. Lepus CM, et al. (2009) Comparison of human fetal liver, umbilical cord blood, andadult blood hematopoietic stem cell engraftment in NOD-scid/gammac-/-, Balb/c-Rag1-/-gammac-/-, and C.B-17-scid/bg immunodeficient mice. Hum Immunol 70(10):790–802.

13. Bour-Jordan H, et al. (2011) Intrinsic and extrinsic control of peripheral T-celltolerance by costimulatory molecules of the CD28/B7 family. Immunol Rev 241(1):180–205.

14. Focosi D, Bestagno M, Burrone O, Petrini M (2010) CD57+ T lymphocytes and func-tional immune deficiency. J Leukoc Biol 87(1):107–116.

15. Peppel K, Crawford D, Beutler B (1991) A tumor necrosis factor (TNF) receptor-IgGheavy chain chimeric protein as a bivalent antagonist of TNF activity. J Exp Med174(6):1483–1489.

16. Feldmann M, Maini RN (2001) Anti-TNF alpha therapy of rheumatoid arthritis: Whathave we learned? Annu Rev Immunol 19:163–196.

17. King C, Ilic A, Koelsch K, Sarvetnick N (2004) Homeostatic expansion of T cells duringimmune insufficiency generates autoimmunity. Cell 117(2):265–277.

18. Jurado JO, et al. (2008) Programmed death (PD)-1:PD-ligand 1/PD-ligand 2 pathwayinhibits T cell effector functions during human tuberculosis. J Immunol 181(1):116–125.

19. Flynn JL, Chan J, Lin PL (2011) Macrophages and control of granulomatous in-flammation in tuberculosis. Mucosal Immunol 4(3):271–278.

20. Saunders BM, Britton WJ (2007) Life and death in the granuloma: Immunopathologyof tuberculosis. Immunol Cell Biol 85(2):103–111.

21. Canetti G (1955) The Tubercle Bacillus in the Pulmonary Lesion of Man. Histo-bacteriology and Its Bearing on the Therapy of Pulmonary Tuberculosis (Springer,New York).

22. Seiler P, et al. (2003) Early granuloma formation after aerosol Mycobacterium tu-berculosis infection is regulated by neutrophils via CXCR3-signaling chemokines. Eur JImmunol 33(10):2676–2686.

23. Peters W, et al. (2001) Chemokine receptor 2 serves an early and essential role inresistance to Mycobacterium tuberculosis. Proc Natl Acad Sci USA 98(14):7958–7963.

24. Caruso AM, et al. (1999) Mice deficient in CD4 T cells have only transiently diminishedlevels of IFN-gamma, yet succumb to tuberculosis. J Immunol 162(9):5407–5416.

25. Mogues T, Goodrich ME, Ryan L, LaCourse R, North RJ (2001) The relative importanceof T cell subsets in immunity and immunopathology of airborne Mycobacterium tu-berculosis infection in mice. J Exp Med 193(3):271–280.

26. Yajima M, et al. (2008) A new humanized mouse model of Epstein-Barr virus infectionthat reproduces persistent infection, lymphoproliferative disorder, and cell-mediatedand humoral immune responses. J Infect Dis 198(5):673–682.

27. Baenziger S, et al. (2006) Disseminated and sustained HIV infection in CD34+ cordblood cell-transplanted Rag2-/-gamma c-/- mice. Proc Natl Acad Sci USA 103(43):15951–15956.

28. An DS, et al. (2007) Use of a novel chimeric mouse model with a functionally activehuman immune system to study human immunodeficiency virus type 1 infection. ClinVaccine Immunol 14(4):391–396.

29. Shultz LD, et al. (2010) Generation of functional human T-cell subsets with HLA-restricted immune responses in HLA class I expressing NOD/SCID/IL2r gamma(null)humanized mice. Proc Natl Acad Sci USA 107(29):13022–13027.

30. Hogan LH, et al. (2001) Mycobacterium bovis strain bacillus Calmette-Guérin-inducedliver granulomas contain a diverse TCR repertoire, but a monoclonal T cell populationis sufficient for protective granuloma formation. J Immunol 166(10):6367–6375.

31. Egen JG, et al. (2011) Intravital imaging reveals limited antigen presentation andT cell effector function in mycobacterial granulomas. Immunity 34(5):807–819.

32. Mohan VP, et al. (2001) Effects of tumor necrosis factor alpha on host immune re-sponse in chronic persistent tuberculosis: Possible role for limiting pathology. InfectImmun 69(3):1847–1855.

33. Lin PL, et al. (2010) Tumor necrosis factor neutralization results in disseminated dis-ease in acute and latent Mycobacterium tuberculosis infection with normal granu-loma structure in a cynomolgus macaque model. Arthritis Rheum 62(2):340–350.

34. Bossen C, et al. (2006) Interactions of tumor necrosis factor (TNF) and TNF receptorfamily members in the mouse and human. J Biol Chem 281(20):13964–13971.

35. Popivanova BK, et al. (2008) Blocking TNF-alpha in mice reduces colorectal carcino-genesis associated with chronic colitis. J Clin Invest 118(2):560–570.

36. Plessner HL, et al. (2007) Neutralization of tumor necrosis factor (TNF) by antibody butnot TNF receptor fusion molecule exacerbates chronic murine tuberculosis. J Infect Dis195(11):1643–1650.

37. Raju B, et al. (2004) Aerosolized gamma interferon (IFN-gamma) induces expressionof the genes encoding the IFN-gamma-inducible 10-kilodalton protein but not in-ducible nitric oxide synthase in the lung during tuberculosis. Infect Immun 72(3):1275–1283.

38. Adams DO (1975) The structure of mononuclear phagocytes differentiating in vivo. II.The effect of Mycobacterium tuberculosis. Am J Pathol 80(1):101–116.

39. Volkman HE, et al. (2004) Tuberculous granuloma formation is enhanced by a myco-bacterium virulence determinant. PLoS Biol 2(11):e367.

40. Sherman DR, et al. (2004) Mycobacterium tuberculosis H37Rv: Delta RD1 is morevirulent than M. bovis bacille Calmette-Guérin in long-term murine infection. J InfectDis 190(1):123–126.

41. North RJ, Izzo AA (1993) Granuloma formation in severe combined immunodeficient (SCID)mice in response to progressive BCG infection. Tendency not to form granulomas in thelung is associated with faster bacterial growth in this organ. Am J Pathol 142(6):1959–1966.

42. Hänsch HC, et al. (1996) Mechanisms of granuloma formation in murine Mycobac-terium avium infection: the contribution of CD4+ T cells. Int Immunol 8(8):1299–1310.

43. Gan H, et al. (2008) Mycobacterium tuberculosis blocks crosslinking of annexin-1 andapoptotic envelope formation on infected macrophages to maintain virulence. NatImmunol 9(10):1189–1197.

44. Chen M, Gan H, Remold HG (2006) A mechanism of virulence: Virulent Mycobacte-rium tuberculosis strain H37Rv, but not attenuated H37Ra, causes significant mito-chondrial inner membrane disruption in macrophages leading to necrosis. J Immunol176(6):3707–3716.

45. Molloy A, Laochumroonvorapong P, Kaplan G (1994) Apoptosis, but not necrosis, ofinfected monocytes is coupled with killing of intracellular bacillus Calmette-Guérin. JExp Med 180(4):1499–1509.

46. Ehlers S, et al. (2001) Alphabeta T cell receptor-positive cells and interferon-gamma,but not inducible nitric oxide synthase, are critical for granuloma necrosis in a mousemodel of mycobacteria-induced pulmonary immunopathology. J Exp Med 194(12):1847–1859.

47. Davis JM, Ramakrishnan L (2009) The role of the granuloma in expansion and dis-semination of early tuberculous infection. Cell 136(1):37–49.

Heuts et al. PNAS | April 16, 2013 | vol. 110 | no. 16 | 6487

IMMUNOLO

GY

Dow

nloa

ded

by g

uest

on

Apr

il 17

, 202

0