Optimal T-cell receptor affinity forinducing autoimmunitySabrina Koehlia, Dieter Naehera, Virginie Galati-Fourniera, Dietmar Zehnb,c, and Ed Palmera,1

aDepartments of Biomedicine and Nephrology, University Hospital Basel and University of Basel, CH-4031 Basel, Switzerland; bSwiss Vaccine ResearchInstitute, CH-1066 Epalinges, Switzerland; and cDivision of Immunology and Allergy, Department of Medicine, Lausanne University Hospital, 1011Lausanne, Switzerland

Edited* by Philippa Marrack, Howard Hughes Medical Institute, National Jewish Health, Denver, CO, and approved October 14, 2014 (received for reviewFebruary 17, 2014)

T-cell receptor affinity for self-antigen has an important role inestablishing self-tolerance. Three transgenic mouse strains express-ing antigens of variable affinity for the OVA transgenic-I T-cellreceptor were generated to address how TCR affinity affects theefficiency of negative selection, the ability to prime an autoimmuneresponse, and the elimination of the relevant target cell. Miceexpressing antigens with an affinity just above the negativeselection threshold exhibited the highest risk of developing ex-perimental autoimmune diabetes. The data demonstrate thatclose to the affinity threshold for negative selection, sufficientnumbers of self-reactive T cells escape deletion and create anincreased risk for the development of autoimmunity.

T cell | TCR | affinity | tolerance | autoimmunity

Aprotective and self-tolerant T-cell repertoire is generated inthe thymus (1–3), where negative selection reduces the

number of self-reactive thymocytes with autoimmune potential (4–6). However, negative selection is not a perfect process, and a smallnumber of self-reactive T cells are found in the periphery (7).Previous work from our laboratory defined the affinity thresholdwhere negative selection is initiated (1, 8). Thymocytes expressingMHC I restricted T-cell receptors (TCRs) undergoing negativeselection when binding a self-antigen with a Kd ≤ 6 μM and positiveselection when binding a self-antigen with lower affinity (8, 9).The autoimmune regulator transcription factor ensures that

tissue-restricted antigens are also expressed in the thymus withinresident medullary thymic epithelial cells (mTECs) (10); how-ever, individual self-antigens are expressed on only a small per-centage of mTECs, and the presentation of a single antigen islimited on the mTEC or dendritic cell surface (11–14). There-fore, self-reactive T cells might stochastically escape negativeselection (15). A high frequency of self-reactive T cells in theperipheral repertoire correlates with the susceptibility to developautoimmune disease. Studies showed that mice with high fre-quencies of myelin-specific T cells are more susceptible for theinduction of experimental autoimmune encephalomyelitis com-pared with mice with lower frequencies of these cells (16). Insingle transgenic mice expressing lymphocytic choriomeningitis vi-rus (LCMV) glycoprotein (gp33) or nucleoprotein with a polyclonalT-cell repertoire, cytotoxic T lymphocytes (CTLs) generated duringthe immune response are of lower affinity compared with non-transgenic mice, suggesting that the highest affinity T cells for theneoantigen were removed from the T-cell repertoire by centraltolerance (16–18). Nevertheless, the remaining lower affinity T cellswere capable of inducing diabetes upon LCMV infection. Similarfindings have been reported for the rat insulin promoter (RIP)–membrane-bound form of ovalbumin (mOVA) mouse line coex-pressing the OT-I TCR (OVA transgenic-I T-cell receptor) β-chain(17). Self-reactive T cells also play a role in the spontaneous de-velopment of diabetes in nonobese diabetic (NOD) mice (19).Recently, we reported a peripheral correlate of the thymic

negative selection affinity threshold such that above-thresholdligands drive T cells into asymmetrical T-cell division, allowing

them to differentiate into short-lived effector T cells (20). Below-threshold T cells expand less efficiently but can still threaten thehost upon recognition of below-threshold antigen in an in-fectious environment (21). In fact, below-threshold T cells areindeed able to mediate autoimmunity after immunization withrecombinant Listeria monocytogenes (Lm) expressing a below-threshold antigen (22).Antigen affinity affects important parameters related to the

development of autoimmunity: (i) the efficiency of central toler-ance, (ii) the efficiency of T-cell priming, and (iii) the efficiency ofdestroying a target cell expressing the self-antigen. In this study, weexamined these parameters individually and in combination. Weshow a striking threshold effect on central deletion, T-cell priming,and cytolysis of antigen-expressing cells in the target tissue. Theseexperiments demonstrate that T cells expressing TCRs just abovethe affinity threshold have the highest potential to induce anautoimmune disease.

ResultsRIP-Variant Mice Express Variant OVA Proteins in Pancreatic β Cells.In RIP-OVA mice, the transgenic RIP drives OVA expression inpancreatic Langerhans islet β cells, proximal tubular epithelialcells in the kidney, mTECs in the thymus, and testes of male mice.For these studies, two different RIP-OVA lines were used: RIP-sOVA and RIP-mOVA express the soluble and membrane formsof ovalbumin, respectively. We also generated three new strainsof RIP-variant transgenic mice expressing the mOVA variants:Q4H7 (RIP-mQ4H7, below-threshold), T4 (RIP-mT4, thresh-old), and Q4R7 (RIP-mQ4R7, above-threshold). To determinewhether OVA-variant antigens are expressed in pancreatic β cells,OT-I mice were crossed to the various RIP-OVA–variant mice to

Significance

The adaptive immune system has the potential to generate aself-reactive response, which can eventually lead to an auto-immune disease. To avoid this outcome, T lymphocytes withhigh-affinity, self-reactive antigen receptors are blocked fromentering the mature T-cell pool (negative selection). Given thismechanism for removing dangerous high-affinity T cells, wewondered whether autoimmunity is more likely to be caused bychronic stimulation of low-affinity T cells or by stimulation ofa few high-affinity T cells that escaped negative selection. In thispaper, we show that T cells with an affinity just above the se-lection threshold can bypass negative selection and have thehighest potential to cause an experimental autoimmune disease.

Author contributions: S.K., D.N., V.G.-F., and E.P. designed research; S.K., D.N., and V.G.-F.performed research; D.N. and D.Z. contributed new reagents/analytic tools; S.K. and E.P.analyzed data; and S.K. and E.P. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.1To whom correspondence should be addressed. Email: ed.palmer@unibas.ch.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1402724111/-/DCSupplemental.

17248–17253 | PNAS | December 2, 2014 | vol. 111 | no. 48 www.pnas.org/cgi/doi/10.1073/pnas.1402724111

Dow

nloa

ded

by g

uest

on

June

28,

202

0

generate double-transgenic F1 animals (i.e., OT-I TCR × RIP-OVA F1, OT-I TCR × RIP-mQ4H7 F1, OT-I TCR × RIP-mT4F1, and OT-I TCR × RIP-mQ4R7 F1). Surprisingly, double-transgenic F1 mice of all crosses showed high mortality within thefirst 2 wk after birth (89% in OT-I × RIP-mQ4H7, 100% inOT-I × RIP-mT4, 97% in OT-I × RIP-mQ4R7, 72% in OT-I ×RIP-mOVA, and 52% in OT-I × RIP-sOVA) (Fig. 1). Blood glu-cose levels of newborn mice were elevated in all crosses, indicatingsevere hyperglycemia. From these results, we conclude that all RIP-OVA–variant transgenic mice expressed the OVA protein in theβ cells of the corresponding RIP-OVA strains and that centraltolerance mechanisms were overwhelmed in these F1 mice.

Below-Threshold T Cells Comprise a Low Risk of Autoimmunity. Wewondered how target cell antigen affinity influences the de-velopment of diabetes following infection with a pathogen ex-pressing a high-affinity epitope (i.e., molecular mimicry). Accordingly,we transferred OT-I T cells into RIP-variant mice, which were sub-sequently infected with Lm expressing the highest affinity OVAepitope (Lm-OVA). RIP-transgenic mice expressing threshold (RIP-mT4) and above-threshold (RIP-mQ4R7, RIP-OVA) OVA variantsdeveloped diabetes when as few as 3 × 103 OT-I T cells weretransferred (Fig. 2A). The transgenic strain RIP-mQ4H7 expressingthe below-threshold variant Q4H7 did not develop diabetes; adop-tive transfer of as many as 3 × 105 OT-I T cells did not resultin glucosuria.Autoimmune diseases can also be initiated by priming auto-

reactive T cells with self-antigens that are identical to the targetantigen. Therefore, OT-I T cells were transferred into various RIP-transgenic strains, followed by immunization with Lm expressingthe same OVA variant as is expressed in the host pancreas (Fig.2B). Because diabetes induction was similar in RIP-sOVA andRIP-mOVA mice, we combined them into one group (RIP-OVA).A breakdown of the two strains is provided in the legend for Fig. 2.We observed efficient diabetes induction with only 3 × 103 trans-ferred OT-I T cells in mice expressing above-threshold self-anti-gens (RIP-OVA: 100%, nine of nine mice; RIP-mQ4R7: 73%,eight of 11 mice). Strikingly, mice expressing the threshold variant,RIP-mT4, required 100-fold as many OT-I T cells (3 × 105) fordiabetes induction. In contrast, there was no induction of diabetesin RIP-mQ4H7 mice expressing the below-threshold OVA variant.All transgenic lines receiving Lm, but no OT-I T cells, were freeof diabetes (SI Appendix, Fig S6A). These data demonstrate that

when the immunizing and target antigens are identical, only above-threshold antigens efficiently induce autoimmune diabetes.

Negative Selection Affinity Threshold Observed in Vivo. To studynegative selection in these transgenic strains, we generatedmixed bone marrow (BM) chimeras, in which a mixture ofCD45.1+ B6 BM (70%) and CD45.1+/.2+ OT-I Rag KO BM(30%) was injected into previously irradiated CD45.2+ trans-genic RIP-variant hosts (Fig. 3A). All chimeric mice were simi-larly reconstituted with OT-I BM (21–33%) (SI Appendix,Fig. S2). We assessed the frequency (Fig. 3B) and number (Fig.3C) of single positive CD8αβ+ OT-I T cells in the lymph nodes(LNs) of the reconstituted chimeras 12 wk after BM transfer. Theextent of clonal deletion was calculated in each chimeric strain(legend for Fig. 3C). Clonal deletion of OT-I T cells was notobserved in RIP-mQ4H7 (below-threshold) or RIP-mT4 (thresh-old) hosts (Fig. 3 B and C). In contrast, the efficiency of clonaldeletion was 91% in RIP-mQ4R7 (above-threshold) hosts and>99% in RIP-OVA (highest affinity) hosts (Fig. 3D). Similarresults were obtained when analyzing thymocytes from the variouschimeric mice (SI Appendix, Fig. S3). Taken together, these datademonstrate an affinity threshold that dictates the efficiency ofnegative selection in vivo (Fig. 3D). These results are similar towhat we observed in fetal thymic organ culture (8).

Phenotype of T Cells Escaping Negative Selection. We wonderedwhether OT-I T cells escaping negative selection in miceexpressing above-threshold self-antigens (RIP-mQ4R7 and RIP-OVA) are phenotypically different from conventional OT-I cells.Analysis of LN cells revealed that only OT-I/B6 BM → RIP-OVA (high-affinity) chimeras contained elevated percentages ofCD44+ (Fig. 4A) and CD122+ OT-I cells (Fig. 4B). Additionallythese hosts contained increased frequencies of OT-I cells with

Fig. 1. Incidence of overt autoimmune diabetes in RIP-mOVA–variant × OT-Idouble-transgenic mice. Rag-2–deficient OT-I mice were crossed to RIP-sOVA,RIP-mOVA, RIP-mQ4R7, RIP-mT4, or RIP-mQ4H7 mice. Mortality was assessedby comparing the number of pups born with the number surviving at 2 wk.The asterisk indicates the pooled result of RIP-sOVA (52% mortality, 16 of 31mice) and RIP-mOVA (72% mortality, 36 of 50 mice).

Fig. 2. Below-threshold T cells comprise a low risk for autoimmunity. (A)Urine glucose was monitored in RIP-OVA– and RIP-OVA–variant mice injec-ted with the indicated number of OT-I T cells, followed by infection with Lm-OVA. (B) Immunization with self-antigen expressed in Lm. Mice injected withthe indicated number of OT-I T cells were infected with Lm expressing thesame self-antigen 1 d later. Mice were considered diabetic if urine glucoselevels were sustained at >1,000 mg/dL for >2 d. The asterisk indicates pooledresults of RIP-sOVA (100%, n = 2–5 mice per group) and RIP-mOVA (100%,n = 2–5 mice per group).

Koehli et al. PNAS | December 2, 2014 | vol. 111 | no. 48 | 17249

IMMUNOLO

GYAND

INFLAMMATION

Dow

nloa

ded

by g

uest

on

June

28,

202

0

memory phenotypes (Fig. 4). The increase of these two memorypopulations was not correlated with negative selection per se,because they were not increased in negatively selecting RIP-mQ4R7 hosts. Recently, Tsai et al. (23) demonstrated that low-affinity, antigen-specific CD8+CD44+CD122+ T cells from NODmice have suppressive properties. This population might developin response to strong recognition of self-ligand in the thymus.However, because CD8+CD44+CD122+ OT-I (suppressor phe-notype) T cells were present in all chimeric strains (Fig. 4 A andB, Bottom Right), there was no clear correlation of their presencewith negative selection in this experimental model. Chimericmice expressing above-threshold OVA variants also contained

increased frequencies of CD8αα OT-I LN T cells (RIP-mQ4R7vs. B6 host: P = 0.06, RIP-OVA vs. B6 host: P = 0.03) (SI Ap-pendix, Fig. S5). The development of CD8αα+ MHC I-restrictedT cells is driven by high-affinity self-antigen (agonist selection)(24–27). These cells carry a suboptimal coreceptor and havelikely bypassed negative selection.

Highest Potential for Autoimmunity Occurs Just Above the AffinityThreshold. All chimeric mice were tolerant, because there was nosign of spontaneous autoimmune diabetes. To assess their risk ofdeveloping autoimmune diabetes, we immunized each chimericstrain with the corresponding self-peptide + LPS and plotted theincidence of diabetes vs. tetramer affinity (Fig. 5). Responsiveness

CD45.1

CD

45.2

A

CD3+tetramer+CD8 +

B6

OT-I (CD45.1+/.2+) / B6 (CD45.1+) Bone Marrow

102 103 104 1050

102

103

104

105

0

6.3 45.844.9

RIP-mQ4R7

102 103 104 1050

102

103

104

105

0

10.4 1.881.9

RIP-mT4Threshold

Central Tolerance

102 103 104 1050

102

103

104

105

0

2.9 46.946.4

RIP-mQ4H7

12.4 29.849.9

102 103 104 1050

102

103

104

105

0

RIP-OVA

102 103 104 1050

102

103

104

105

0

10.2 0.287.4

Host (CD45.2+):

% O

T-I T

cel

ls

amon

g C

D8

LN T

cel

ls

B

RIP-mQ4H

7B6

RIP-mT4

RIP-mQ4R

7

RIP-OVA*

Host:

C D

% E

ffici

ency

of

Neg

ativ

e Se

lect

ion

Tetramer KD (nM)1000

RIP-OVA*

B6RIP-mQ4H7RIP-mT4 RIP-mQ4R7

Host:

RIP-mQ4H

7B6

RIP-mT4

RIP-mQ4R

7

RIP-OVA*

Host:Affinity

nr O

T-I T

cel

ls in

LN

s

Fig. 3. Efficiency of OT-I deletion in OT-I/B6 mixedBM chimeras expressing OVA-variant proteins is de-pendent on antigen affinity. (A–D) Flow cytometricanalysis of lymphocytes in OT-I/B6 mixed BM chi-meras was performed 12–15 wk after reconstitution.(A) Representative flow cytometry plots of LN cells inOT-I/B6 mixed BM chimeras are shown. Percentage(B) and absolute numbers (C) of CD8αβ OT-I T cellswithin LNs were assessed by flow cytometry. Hori-zontal bars in B and C represent the geometric meanvalue of the individual data points. The asteriskindicates pooled results of RIP-sOVA (0.3%, 9,537, n =4) and RIP-mOVA (0.1%, 5,156, n = 3) mice. Resultsshown were pooled from six independent experi-ments. (D) Efficiency of negative selection for eachBM chimeric strain is plotted vs. antigen affinity (8).Efficiency of negative selection (%) = 100 − [(meannumber of OT-I cells in RIP-OVA–variant host/meannumber of OT-I cells in B6 host) × 100].

CD

44

B6

CD62L

88

6.90.6

4.6

RIP-mQ4H7 RIP-mT4

0.7 6.6

6.0 86.7

RIP-mQ4R7

80.4

9.12.1

8.4

RIP-mOVA*

47.9

377.8

7.4

6.7

85.2

2.5

5.8

Host:A

B

% T

EM O

T-I

% T

CM O

T-I

TCM TEM naive

% s

up. O

T-I

suppressor phenotype

OT-I (CD45.1+/.2+) / B6 (CD45.1+) Bone Marrow

% n

aive

OT-

I

Fig. 4. Phenotypic characterization of OT-I T cells in mixed BM chimeras. (A) LN cells from BM chimeras were stained with mAbs specific for CD44, CD62L, andCD122. Representative flow cytometry plots are shown. Numbers in the plots depict frequency (%) of cells in each quadrant. (B) Analysis of OT-I LN T cells withmemory phenotypes in RIP-OVA–variant chimeras. Fractions of central memory T cells (TCM) (CD44+CD62L+), effector memory T cells (TEM) (CD44+CD62L−),naive (CD44−CD62L+), suppressor phenotype cells (CD44+CD122+), and OT-I CD8 αβ+ cells are shown. The asterisk indicates pooled results of RIP-sOVA (n = 4)and RIP-mOVA (n = 3) mice.

17250 | www.pnas.org/cgi/doi/10.1073/pnas.1402724111 Koehli et al.

Dow

nloa

ded

by g

uest

on

June

28,

202

0

of OT-I T cells to these peptides is shown in SI Appendix, Fig. S4.OT-I/B6 BM → RIP-mQ4H7 chimeras immunized with Q4H7peptide + LPS developed neither glucosuria nor lethal diabetes.On the other hand, 100% (eight of eight) OT-I/B6 BM → RIP-mT4 chimeras immunized with the threshold peptide, T4 + LPS,developed glucosuria (Fig. 5A) that resolved in 50% of theseanimals (compare gray symbols in Fig. 5 A and B). In this model,the threshold ligand induces autoimmune symptoms but not al-ways irreversible disease (50%). Considering above-thresholdself-antigens, 86% of OT-I/B6 BM → RIP-mQ4R7 chimerasimmunized with Q4R7 + LPS developed irreversible diabetes(Fig. 5 A and B). In contrast, not a single (none of 10) OT-I/B6BM→ RIP-OVA chimeras immunized with OVA peptide + LPSdeveloped glucosuria or irreversible diabetes (Fig. 5 A and B).Unlike the RIP-mQ4R7 chimeras, the tolerant state in thesehighest affinity RIP-OVA chimeras is robust and stable.All chimeric strains, regardless of the affinity of their OVA-

variant self-antigen, contained a similar frequency (SI Appendix,Fig. S6B) of CD4+FoxP3+ regulatory T cells (Tregs; derived fromB6 donor BM). This finding was expected because the only dif-ference between the RIP strains is an MHC class I-presentedepitope. Therefore, Treg frequency does not explain the dramaticdifferences between the RIP-mQ4R7 and RIP-OVA chimeras interms of susceptibility to developing autoimmune diabetes.

Self-Antigen Affinity vs. Risk of Developing Autoimmunity. Based onthese results, we plotted self-antigen affinity vs. the number ofperipheral OT-I cells and color-coded the risk to develop auto-immune diabetes upon self-antigen challenge (Fig. 6). In Fig. 6A,the number of OT-I cells from each of the four strains of RIP-variant chimeric mice (from Figs. 3 and 5) is plotted vs. antigenaffinity. In Fig. 6B, the risk of developing lethal diabetes wasextrapolated to cover a wide range of self-antigen affinities andnumbers of self-reactive peripheral T cells. We assumed that achimeric mouse expressing a self-antigen of lower affinity

compared with Q4H7 or containing lower numbers of peripheralOT-I T cells would also carry a very low risk (∼0%) of developingdiabetes. Mice expressing a self-antigen at the affinity thresholdhave an intermediate (∼50%) risk of developing lethal autoim-munity, but only if they contain large numbers (∼2 × 106) of OT-Icells (sectors 2 and 3) (Figs. 2B and 5). In contrast, the risk ofdeveloping diabetes is very high (∼90%) in RIP-mQ4R7 chi-meras. One can assume that chimeric mice expressing an evenhigher affinity self-antigen and containing ≥105 OT-I cells wouldalso have a very high (≥90%) risk of developing this form ofautoimmunity. Finally, RIP-OVA chimeras containing 6 × 103

OT-I T cells remain tolerant to a challenge with OVA + LPS.We assumed that chimeric mice containing as few as 6 × 103 OT-Icells and expressing a lower affinity self-antigen would alsohave a similarly low risk of developing autoimmune diabetes(sector 5).In Fig. 6C, the hypothetical plot in Fig. 6B has been corrected

for the effects of central tolerance. Because above-threshold self-antigens prevent most but not all self-reactive thymocytes fromentering the periphery, the risk of developing autoimmunity issignificantly reduced. As illustrated in Fig. 6C, we propose thatthe “dangerous” fraction of the peripheral T-cell repertoire isconcentrated among T cells with a TCR affinity just above thenegative selection threshold (i.e., Kd of ∼6 μM). At this affinity,negative selection is incomplete and the escaping T cells havesufficient affinity to induce autoimmune diabetes in this modelsystem (20). At higher TCR affinities, negative selection is in-creasingly complete such that the few escaping T cells are in-capable of inducing a lethal autoimmune disease.

DiscussionTo address how self-antigen affinity influences several differentsteps in the establishment of T-cell tolerance, we generated newtransgenic mouse strains expressing antigens with variable af-finity for the OT-I TCR. All RIP-OVA–variant strains werecrossed to OT-I cells to generate double-transgenic mice. Alldouble-transgenic strains develop diabetes, indicating expressionof the neo–self-antigen in the pancreas. The incidence of di-abetes in RIP-mQ4H7, RIP-mT4, and RIP-mQ4R7 double-transgenic mice is 89–100%, suggesting that tolerance mecha-nisms were overwhelmed. Lethal diabetes appears in only 52%of RIP-sOVA and 72% of RIP-mOVA double-transgenic mice,indicating that very high-affinity self-antigens more efficientlydeplete self-reactive thymocytes. These results are consistentwith observations of Hubert et al. (28), who observed that 88%OT-I × RIP-mOVA double-transgenic mice develop severe di-abetes. Similarly, McGargill et al. (29) found that 80% of K14-OVAp/OT-I double-transgenic mice, where OVA is expressedunder control of the keratin promoter 14, succumb to a lethalautoimmune disease due to insufficient deletion of developingOT-I cells. Nevertheless, 20% of these animals survived. Surpris-ingly, double-transgenic animals expressing the below-threshold

Glu

cosu

ria (%

)

Ove

rt d

iabe

tes

(%)

RIP-OVA*

B6

RIP-mT4 RIP-mQ4R7

Host

RIP-mQ4H7

Immunogen

OVA/LPS

OVA/LPS

T4/LPS Q4R7/LPS

Q4H7/LPS

Tetramer KD (nM)Affinity

A B

Tetramer KD (nM)Affinity

Fig. 5. Highest risk for autoimmunity occurs slightly above the affinitythreshold for central tolerance. The percentage of immunized chimeric micewith glucosuria (A) and sustained diabetes (B) is shown. RIP-OVA, RIP-mQ4H7, and C57BL/6 chimeras were free of diabetes for 2 wk followingimmunization. The asterisk indicates pooled results of RIP-sOVA (n = 5) andRIP-mOVA (n = 4) mice.

Self-antigen affinity

107

106

105

104

103Nr.

perip

hera

l T c

ells

Q4H7 T4

Q4R7

OVA

Threshold

2

Self-antigen affinity

107

106

105

104

103Nr.

perip

hera

l T c

ells

1 3

1 3

4

5

ThresholdA B

Experimental Data Extrapolation Effect of Central Tolerance

C

Self-antigen affinity

107

106

105

104

103Nr.

perip

hera

l T c

ells

Threshold

Risk Lethal disease ~ 0% ~ 50 % ~ 90 %

6

Fig. 6. Self-antigen affinity vs. risk of developing auto-immunity. (A) Number (Nr.) of OT-I T cells in each chimericstrain (Fig. 3) is plotted vs. antigen affinity. The negativeselection affinity threshold is indicated by an arrow. Therisk of developing diabetes is indicated by color (blue,∼0%risk; yellow, ∼50% risk; red, ∼90% risk). (B) Data in A areextrapolated for an extended range of self-antigen affin-ities and numbers of self-reactive OT-I T cells. A descriptionof different regions of the plot is provided in Results. (C)Extrapolated plot in B is corrected for the effects of neg-ative selection; central deletion removes most, but not all,OVA-reactive T cells with TCR affinities above the negativeselection threshold (sector 6). In principle, this model couldbe applied to any self-reactive CD8 T cell.

Koehli et al. PNAS | December 2, 2014 | vol. 111 | no. 48 | 17251

IMMUNOLO

GYAND

INFLAMMATION

Dow

nloa

ded

by g

uest

on

June

28,

202

0

antigen Q4H7 develop diabetes (8, 20, 22). A possible explanationis that just after birth, an extremely large number of OT-I T cellslikely enter the periphery before significant numbers of Tregsaccumulate. In this relatively lymphopenic environment, a suffi-cient number of OT-I T cells might expand by homeostatic divisionwithin the first week of life, generating diabetogenic CTLs.In a mimicry model (30–32), where transferred OT-I cells were

fully activated following Lm-OVA (high-affinity) infection (33),RIP-mQ4H7 mice did not develop diabetes, indicating that hosttissues expressing a below-threshold self-antigen are difficult todamage. All other RIP-OVA–variant mice efficiently developedlethal diabetes. Therefore, under conditions of molecular mimicry,host tissues expressing threshold or above-threshold self-antigensare highly susceptible to autoimmune attack. In a different setting,where the host experiences an infection of a peripheral organ,inflammation-induced tissue damage might result in release andpresentation of self-tissue antigens, which subsequently inducea pathogenic response from self-antigen specific T cells that havesurvived negative selection. To address the role of antigen affinityin this context, RIP-OVA– and RIP-OVA–variant mice wereadoptively transferred with OT-I cells and infected with Lm ex-pressing the same OVA variant expressed by the self-antigentransgenic mouse. Mice expressing self-antigen above the affinitythreshold (RIP-mQ4R7 and RIP-mOVA) are at high risk to developdiabetes, even when low numbers of OT-I cells were transferred. Incontrast, mice expressing the threshold antigen T4 required 100-foldmore transferred OT-I T cells to develop diabetes.Many early studies examining tolerance to very high-affinity

neo–self-antigens did not address tolerance to intermediate-and low-affinity antigens (4, 16, 18, 34, 35). More recent studiesindicate that tolerance to intermediate- and low-affinity anti-gens is inefficient (17, 22). To address this issue in a morequantitative manner, we generated mixed OT-I/B6 BM chi-meric mice using RIP-OVA–variant hosts. Chimeras expressingbelow-threshold or at-threshold affinity antigens do not inducenegative selection in vivo, whereas hosts expressing above-threshold self-antigens are highly efficient in preventing OT-IT cells from entering the periphery (RIP-mQ4R7 = 91% effi-cient, RIP-mOVA > 99.5% efficient, RIP-sOVA > 99.4% ef-ficient). We also observed deletion of self-reactive OT-I cellswithin the thymus, indicating that in chimeric mice, a largecomponent of their tolerance is attributable to thymic selectionprocesses (SI Appendix, Fig S3). These results are similar towhat we observed in fetal thymic organ culture (8). It is worthnoting that just above the affinity threshold, negative selectionis incomplete.Because agonist-selected thymocytes enter diverse T-cell line-

ages (36–41), we examined the phenotype of peripheral OT-I T cellsin chimeric RIP-mOVA–variant mice. The frequency of CD8ααOT-I cells increased in chimeras expressing above-threshold anti-gens (42–45). Given their reduced antigen binding (46, 47), CD8ααthymocytes have an increased chance to evade negative selection.The proportion of OT-I T cells with a central (CD44hiCD62Lhi) oreffector (CD44hiCD62Llo) memory phenotype was clearly in-creased, but only in chimeras expressing the highest affinity ligand,OVA. CD8+CD44hiCD122hi cells have been reported to pre-vent diabetes in NOD mice (23); however, the frequency ofCD8+CD44hiCD122hi OT-I cells did not correlate with the stateof tolerance in our chimeric strains. Both OT-I/B6 BM → RIP-mT4 (nontolerant) and OT-I/B6 BM → RIP-mOVA (tolerant)mice contained a similar percentage of CD8+CD44hiCD122hi

OT-I cells. On the other hand, the total OT-I number variesconsiderably between these two strains. For this reason, it wasdifficult to evaluate the role of CD8+CD44hiCD122hi OT-I cellsin maintaining tolerance in this particular diabetes model.The highest potential for autoimmunity occurs near the affinity

threshold for negative selection (RIP-mT4 chimeras: 50% irre-versible diabetes, RIP-mQ4R7 chimeras: 86% irreversible

diabetes). In contrast, below-threshold affinity chimeras (OT-I/B6 BM → RIP-mQ4H7) and the highest affinity chimeras (OT-I/B6 BM→ RIP-mOVA and OT-I/B6 BM→ RIP-sOVA) failed todevelop diabetes. Below-threshold RIP-mQ4H7 chimeras arelikely protected due to inefficient T-cell priming and poortarget cell lysis. High-affinity RIP-mOVA and RIP-sOVAchimeras are likely protected due to extremely efficient neg-ative selection. Interestingly, the highest risk of developingautoimmune diabetes in this model is just above the negativeselection affinity threshold. In this narrow range of TCR af-finity, negative selection is leaky and the escaping peripheralT cells can be sufficiently activated by their self-antigen toinduce autoimmune pathology.Under certain conditions, low-affinity T cells cause diabetes in

an RIP-mOVA model. Enouz et al. (22) have convincingly shownthat below-threshold antigens induce peripheral CD8 T cells todifferentiate into CTLs; furthermore, these below-threshold stim-ulated T cells cause hyperglycemia in mice expressing a thresholdaffinity antigen in the pancreas. The experiments described herewere slightly different; although T cells can be stimulated by abelow-threshold antigen, such T cells do not induce diabetes inmice expressing the same below-threshold antigen on pancreatic βcells. Therefore, our results are not inconsistent with the resultsreported by Enouz et al. (22). Our data make the important pointthat it is quantitatively more difficult to induce autoimmune pa-thology using a below-threshold antigen compared with an above-threshold antigen.The frequency of lethal diabetes is high in this model (50%

for chimeras expressing a threshold antigen, 86% for chimerasexpressing an above threshold antigen). This observation islikely due to the fact that the chimeras were reconstituted witha large fraction (30%) of stem cells derived from OT-I mice. Incontrast, the frequency of self-reactive thymocytes for a singleepitope is much smaller in polyclonal individuals, which likelyimproves the efficiency of thymic deletion. This idea is con-sistent with the relatively low incidence of autoimmunity in thegeneral population; 3% of the human population suffers fromsome form of autoimmunity (48). Nevertheless, an individualcan probably cope with a small number of high-affinity pe-ripheral T cells and still be considered functionally tolerant,allowing negative selection to be less than perfect, which seemsto be the case (49, 50). Recent computational studies (51)propose that an immune or autoimmune response requiresa minimum (quorum) number of antigen-specific T cells. If anindividual carries fewer than the quorum number of above-threshold T cells for a particular epitope, it may be extremelydifficult to induce an autoimmune disease. Our studies suggestthat T cells with an antigen affinity just above the affinity thresholdare more likely to exceed the quorum number in the periphery. Forthis reason, T cells with self-reactivity just above the negative se-lection threshold have the highest autoimmune potential.

Materials and MethodsMice. RIP-sOVA and RIP-mOVA mice (20, 52, 53), OT-I TCR transgenic micerecognizing Kb/Ova257–264, and CD45-1 congenic C57BL/6 mice were allobtained from the Jackson Laboratory. All animal work was done in accor-dance with the federal and cantonal laws of Switzerland. Animal researchprotocols were approved by the Animal Research Commission of the Cantonof Baselstadt, Switzerland.Generation of OVA-variant transgenic mice. Generation of OVA-variant trans-genic mice is described in SI Appendix, Fig S1.Generation of BM chimeric mice. Generation of BM chimeric mice is described inSI Appendix, Fig S2.

Adoptive Cell Transfer and Infections. Mice were injected i.v. with single-cellsuspensions of OT-I cells; on the following day, 5,000 cfu of Lm-OVA or Lm-OVA variant was injected i.v.. Recombinant Lm expressing the full-lengthOVA protein containing the CD8 epitope SIINFEKL (OVA) or altered ligandsQ4R7, T4, or Q4H7 was previously described (21).

17252 | www.pnas.org/cgi/doi/10.1073/pnas.1402724111 Koehli et al.

Dow

nloa

ded

by g

uest

on

June

28,

202

0

Blood and Urine Glucose Measurements. Blood samples from the tail vain wereread on a Contour blood glucose reader (Bayer). Mice with blood glucoselevels >15 mmol/L were considered to be hyperglycemic. Urine glucose wasassessed with test strips (DIABUR-TEST 5000; ACCU-CHEK). Mice with sus-tained urine glucose levels >1,000 mg/dL were considered diabetic.

Tetramer, Surface Antibody Staining, and Flow Cytometry. Lymphocytes werestained with the following antibodies and/or tetramers: CD45.1-FITC, CD45.1-phycoerythrin (PE), CD45.2-allophycocyanin (APC)-Cy7, CD8α-biotin, CD4-Alexa700, and CD3e-APC (all from Becton Dickinson); CD8β-peridininchlorophyll Cy5.5, CD44-APC, CD122-FITC, and CD62L-PE-Cy7 (all fromBioLegend); CD8α-PE-Cy7 (eBioscience); and Kb-OVA (SIINFEKL) tetramers(from V.G.-F. and E.P.).

Data Analyses. Flow cytometry measurements were performed on a FACSCanto II (Becton Dickinson), and data were analyzed with FlowJo software(TreeStar). Graphs were generated with Prism (GraphPad). Unpaired t testsand one-way-ANOVA were performed (Prism) where indicated.

ACKNOWLEDGMENTS. We thank C. G. King, S. Keck, O. Stepanek, T. Rolink,and D. Finke for critical discussions; M. Schmaler for advice on culturing Lm;S. Keck, L. Wyss, R. Lang, and B. Hausmann for excellent experimentalassistance; K. Thienel and E. Dalmas for advice on pancreatic islet isolation;and the team of U. Schneider for animal husbandry. The work wassupported by Research Grants 310030B_133131/1, Synergia (Swiss NationalScience Foundation), Sybilla (European Union Seventh FrameworkProgramme), and TerraIncognita (European Research Council) (all to E.P.)and by Grant 310030_130512 (Swiss National Science Foundation) (to D.Z.).

1. Palmer E (2003) Negative selection—Clearing out the bad apples from the T-cellrepertoire. Nat Rev Immunol 3(5):383–391.

2. Starr TK, Jameson SC, Hogquist KA (2003) Positive and negative selection of T cells.Annu Rev Immunol 21:139–176.

3. Stritesky GL, Jameson SC, Hogquist KA (2012) Selection of self-reactive T cells in thethymus. Annu Rev Immunol 30:95–114.

4. Kisielow P, Blüthmann H, Staerz UD, Steinmetz M, von Boehmer H (1988) Tolerance inT-cell-receptor transgenic mice involves deletion of nonmature CD4+8+ thymocytes.Nature 333(6175):742–746.

5. von Boehmer H, Teh HS, Kisielow P (1989) The thymus selects the useful, neglects theuseless and destroys the harmful. Immunol Today 10(2):57–61.

6. Goldrath AW, Bevan MJ (1999) Selecting and maintaining a diverse T-cell repertoire.Nature 402(6759):255–262.

7. Taniguchi RT, et al. (2012) Detection of an autoreactive T-cell population within thepolyclonal repertoire that undergoes distinct autoimmune regulator (Aire)-mediatedselection. Proc Natl Acad Sci USA 109(20):7847–7852.

8. Daniels MA, et al. (2006) Thymic selection threshold defined by compartmentalizationof Ras/MAPK signalling. Nature 444(7120):724–729.

9. Naeher D, et al. (2007) A constant affinity threshold for T cell tolerance. J Exp Med204(11):2553–2559.

10. Metzger TC, Anderson MS (2011) Control of central and peripheral tolerance by Aire.Immunol Rev 241(1):89–103.

11. Anderson MS, et al. (2002) Projection of an immunological self shadow within thethymus by the aire protein. Science 298(5597):1395–1401.

12. Mathis D, Benoist C (2009) Aire. Annu Rev Immunol 27:287–312.13. Klein L, Hinterberger M, Wirnsberger G, Kyewski B (2009) Antigen presentation in the

thymus for positive selection and central tolerance induction. Nat Rev Immunol 9(12):833–844.

14. Gallegos AM, Bevan MJ (2004) Central tolerance to tissue-specific antigens mediatedby direct and indirect antigen presentation. J Exp Med 200(8):1039–1049.

15. Gotter J, Brors B, Hergenhahn M, Kyewski B (2004) Medullary epithelial cells of thehuman thymus express a highly diverse selection of tissue-specific genes colocalized inchromosomal clusters. J Exp Med 199(2):155–166.

16. Ohashi PS, et al. (1991) Ablation of “tolerance” and induction of diabetes by virusinfection in viral antigen transgenic mice. Cell 65(2):305–317.

17. Zehn D, Bevan MJ (2006) T cells with low avidity for a tissue-restricted antigen rou-tinely evade central and peripheral tolerance and cause autoimmunity. Immunity25(2):261–270.

18. von Herrath MG, Dockter J, Oldstone MB (1994) How virus induces a rapid or slow onsetinsulin-dependent diabetes mellitus in a transgenic model. Immunity 1(3):231–242.

19. Clemente-Casares X, Tsai S, Huang C, Santamaria P (2012) Antigen-specific thera-peutic approaches in Type 1 diabetes. Cold Spring Harb Perspect Med 2(2):a007773.

20. King CG, et al. (2012) T cell affinity regulates asymmetric division, effector cell dif-ferentiation, and tissue pathology. Immunity 37(4):709–720.

21. Zehn D, Lee SY, Bevan MJ (2009) Complete but curtailed T-cell response to very low-affinity antigen. Nature 458(7235):211–214.

22. Enouz S, Carrié L, Merkler D, Bevan MJ, Zehn D (2012) Autoreactive T cells bypassnegative selection and respond to self-antigen stimulation during infection. J ExpMed 209(10):1769–1779.

23. Tsai S, et al. (2010) Reversal of autoimmunity by boosting memory-like autor-egulatory T cells. Immunity 32(4):568–580.

24. Cruz D, et al. (1998) An opposite pattern of selection of a single T cell antigen receptor inthe thymus and among intraepithelial lymphocytes. J Exp Med 188(2):255–265.

25. Poussier P, Ning T, Banerjee D, Julius M (2002) A unique subset of self-specific in-traintestinal T cells maintains gut integrity. J Exp Med 195(11):1491–1497.

26. Levelt CN, et al. (1999) High- and low-affinity single-peptide/MHC ligands have dis-tinct effects on the development of mucosal CD8alphaalpha and CD8alphabeta Tlymphocytes. Proc Natl Acad Sci USA 96(10):5628–5633.

27. Denning TL, et al. (2007) Mouse TCRalphabeta+CD8alphaalpha intraepithelial lym-phocytes express genes that down-regulate their antigen reactivity and suppressimmune responses. J Immunol 178(7):4230–4239.

28. Hubert FX, et al. (2011) Aire regulates the transfer of antigen frommTECs to dendriticcells for induction of thymic tolerance. Blood 118(9):2462–2472.

29. McGargill MA, et al. (2002) A spontaneous CD8 T cell-dependent autoimmune diseaseto an antigen expressed under the human keratin 14 promoter. J Immunol 169(4):2141–2147.

30. Gronski MA, et al. (2004) TCR affinity and negative regulation limit autoimmunity.Nat Med 10(11):1234–1239.

31. Hudrisier D, et al. (2001) Structural and functional identification of major histocom-patibility complex class I-restricted self-peptides as naturally occurring molecularmimics of viral antigens. Possible role in CD8+ T cell-mediated, virus-induced auto-immune disease. J Biol Chem 276(22):19396–19403.

32. Nugent CT, et al. (2000) Characterization of CD8+ T lymphocytes that persist afterperipheral tolerance to a self antigen expressed in the pancreas. J Immunol 164(1):191–200.

33. Pope C, et al. (2001) Organ-specific regulation of the CD8 T cell response to Listeriamonocytogenes infection. J Immunol 166(5):3402–3409.

34. Kurts C, Heath WR, Carbone FR, Kosaka H, Miller JF (1998) Cross-presentation of selfantigens to CD8+ T cells: The balance between tolerance and autoimmunity. NovartisFound Symp 215:172–181, discussion 181–190.

35. Kirberg J, et al. (1994) Thymic selection of CD8+ single positive cells with a class IImajor histocompatibility complex-restricted receptor. J Exp Med 180(1):25–34.

36. Hogquist KA (2001) Signal strength in thymic selection and lineage commitment. CurrOpin Immunol 13(2):225–231.

37. Singer A (2002) New perspectives on a developmental dilemma: The kinetic signalingmodel and the importance of signal duration for the CD4/CD8 lineage decision. CurrOpin Immunol 14(2):207–215.

38. Izon DJ, et al. (2001) Notch1 regulates maturation of CD4+ and CD8+ thymocytes bymodulating TCR signal strength. Immunity 14(3):253–264.

39. Hayes SM, Love PE (2006) Strength of signal: A fundamental mechanism for cell fatespecification. Immunol Rev 209:170–175.

40. Smith-Garvin JE, et al. (2010) T-cell receptor signals direct the composition andfunction of the memory CD8+ T-cell pool. Blood 116(25):5548–5559.

41. Chang JT, et al. (2007) Asymmetric T lymphocyte division in the initiation of adaptiveimmune responses. Science 315(5819):1687–1691.

42. Leishman AJ, et al. (2002) Precursors of functional MHC class I- or class II-restrictedCD8alphaalpha(+) T cells are positively selected in the thymus by agonist self-pep-tides. Immunity 16(3):355–364.

43. Baldwin TA, Hogquist KA, Jameson SC (2004) The fourth way? Harnessing aggressivetendencies in the thymus. J Immunol 173(11):6515–6520.

44. Jordan MS, et al. (2001) Thymic selection of CD4+CD25+ regulatory T cells induced byan agonist self-peptide. Nat Immunol 2(4):301–306.

45. Zhou D, et al. (2004) Lysosomal glycosphingolipid recognition by NKT cells. Science306(5702):1786–1789.

46. Cawthon AG, Lu H, Alexander-Miller MA (2001) Peptide requirement for CTL acti-vation reflects the sensitivity to CD3 engagement: Correlation with CD8alphabetaversus CD8alphaalpha expression. J Immunol 167(5):2577–2584.

47. Arcaro A, et al. (2001) CD8beta endows CD8 with efficient coreceptor function bycoupling T cell receptor/CD3 to raft-associated CD8/p56(lck) complexes. J Exp Med194(10):1485–1495.

48. Cooper GS, Bynum ML, Somers EC (2009) Recent insights in the epidemiology ofautoimmune diseases: Improved prevalence estimates and understanding of cluster-ing of diseases. J Autoimmun 33(3-4):197–207.

49. Moon JJ, et al. (2011) Quantitative impact of thymic selection on Foxp3+ and Foxp3-subsets of self-peptide/MHC class II-specific CD4+ T cells. Proc Natl Acad Sci USA108(35):14602–14607.

50. Trudeau JD, et al. (2003) Prediction of spontaneous autoimmune diabetes in NODmice by quantification of autoreactive T cells in peripheral blood. J Clin Invest 111(2):217–223.

51. Butler TC, Kardar M, Chakraborty AK (2013) Quorum sensing allows T cells to dis-criminate between self and nonself. Proc Natl Acad Sci USA 110(29):11833–11838.

52. Kurts C, Kosaka H, Carbone FR, Miller JF, Heath WR (1997) Class I-restricted cross-presentation of exogenous self-antigens leads to deletion of autoreactive CD8(+) Tcells. J Exp Med 186(2):239–245.

53. Behrens GM, et al. (2004) Helper requirements for generation of effector CTL to isletbeta cell antigens. J Immunol 172(9):5420–5426.

Koehli et al. PNAS | December 2, 2014 | vol. 111 | no. 48 | 17253

IMMUNOLO

GYAND

INFLAMMATION

Dow

nloa

ded

by g

uest

on

June

28,

202

0