Journal of Clinical Immunology

, Volume 28, Issue 4, pp 350–360

Human CD8 Responses to a Complete Epitope Set from Preproinsulin: Implications for Approaches to Epitope Discovery

Authors

  • Caroline Baker
    • Department of Cellular and Molecular Medicine, School of Medical SciencesUniversity of Bristol
  • Liliana G. Petrich de Marquesini
    • Henry Wellcome Laboratories for Integrative Neuroscience and EndocrinologyUniversity of Bristol
  • Amanda J. Bishop
    • Henry Wellcome Laboratories for Integrative Neuroscience and EndocrinologyUniversity of Bristol
  • Alan J. Hedges
    • Department of Cellular and Molecular Medicine, School of Medical SciencesUniversity of Bristol
  • Colin M. Dayan
    • Henry Wellcome Laboratories for Integrative Neuroscience and EndocrinologyUniversity of Bristol
    • Department of Cellular and Molecular Medicine, School of Medical SciencesUniversity of Bristol
Article

DOI: 10.1007/s10875-008-9177-4

Cite this article as:
Baker, C., Petrich de Marquesini, L.G., Bishop, A.J. et al. J Clin Immunol (2008) 28: 350. doi:10.1007/s10875-008-9177-4

Abstract

Purpose

In this study, we explored the breadth of CD8 T cell reactivity to preproinsulin (PPI) in type 1 diabetes.

Materials and Methods

We tested a complete peptide set in pools covering all 406 potential 8–11mer epitopes of PPI and 61 algorithm-predicted human leukocyte antigen (HLA)-A2-specific epitopes (15 pools) from islet-specific glucose-6-phophatase catalytic subunit-related protein (IGRP), using a CD8-specific granzyme B enzyme-linked immunosorbent spot assay.

Results

Responses were seen to 64 of the 102 PPI pools in two or more newly diagnosed patients (63%) compared to 11 pools in the control subjects (11%, p < 0.0001, Fisher’s exact test). We identified five pools containing 20 peptides, which distinguished patients from control subjects, most of which had predicted low-affinity binding to HLA class I molecules. In contrast, fewer (5 of 15 = 33%) IGRP peptide pools, selected by higher binding affinity for HLA-A2 (present in seven of eight patients and five of seven control subjects), stimulated responses in two or more patients, and none stimulated responses in more than two control subjects (p = 0.042, Fisher’s exact test).

Conclusion

Thus, we conclude that CD8 T cell reactivity to PPI in patients with type 1 diabetes can be much broader than shown previously and more diverse than seen in control subjects. Furthermore, responses were often stimulated by peptides with low predicted HLA-binding affinities.

Keywords

Type 1 diabetesautoimmunityhumanCD8 T cells

Introduction

Type 1 diabetes is caused by autoimmune destruction of the insulin-producing β cells in the pancreatic islets of Langerhans. It is now well established that this destruction is caused by islet antigen-specific autoreactive T cells, and there is much evidence that both the CD4 and CD8 T cell subsets are required for disease to occur [1, 2].

Early investigation of T cell responses to islet autoantigens in type 1 diabetes primarily focused on CD4 T cells [3]. However, recently, much attention has been directed toward the CD8 T cell subset. Epitopes have now been identified from multiple islet antigens including proinsulin [47], glutamic acid decarboxylase-65 (GAD65) [79], prepro-islet amyloid polypeptide protein [1012], glial fibrillary acidic protein [11], insulinoma-associated antigen-2 (IA-2) [12], and most recently islet-specific glucose-6-phophatase catalytic subunit-related protein (IGRP), a membrane-spanning protein expressed predominantly in pancreatic β cells [7, 11, 12]. These investigations have selected T cell epitopes using a number of different methods, and it is clear, taken together, that T cells from both patients newly diagnosed with type 1 diabetes and from patients with long-standing disease recognize a variety of epitopes from different autoantigens.

In respect of preproinsulin (PPI), two epitopes from PPI recognized by CD8 T cells have been identified in NOD mice. Insulin B15–23 was shown to stimulate the highly diabetogenic CD8 T cell clone, G9C8, in NOD mice [13], and it is interesting to note that human antigen presenting cell pulsed with this peptide have been targeted by CD8 T cells from patients with type 1 diabetes [14]. A peptide from proinsulin, the precursor to insulin, B25–C34, was able to induce cytotoxic T lymphocytes and administration of this epitope to prediabetic NOD mice delayed disease onset and reduced the incidence of diabetes [15]. Previously, PPI peptides, predicted from proteosome cleavage sites, have been reported to stimulate interferon (IFN)-γ production detected by enzyme-linked immunosorbent spot (ELISPOT) in patients with both newly diagnosed and long-standing type 1 diabetes [5]. PPI epitopes were also identified by a reverse-immunology technique, using proteosome complexes to establish potential HLA class I epitopes combined with epitope prediction algorithms. The peptides were then tested by immunizing HLA-A2 transgenic mice [6]. Some of these peptides stimulated responses in patients with type 1 diabetes [7]. Responses to an insulin B-chain peptide, measured using an HLA-A2–peptide tetramer, were measured in patients who previously received a pancreas transplant but lost their allograft [16]. While these approaches have clearly identified PPI epitopes with respect to HLA-A2 and provide important data for testing patient responses, they may not highlight all responses, as there are limitations to the proteasome digestion technique [6]. Furthermore, not all patients express HLA-A2. While tetramer staining provides important direct confirmation of an epitope, in many cases, the frequency is too low to detect without expanding the cells in culture.

IGRP is expressed predominantly in pancreatic β cells [17]. A peptide from murine IGRP (mIGRP206–214) was found to stimulate the diabetogenic 8.3 CD8 T cells [18]. T cells reactive to this peptide were detectable in both the islets and peripheral blood of NOD mice [19] and are a predominant population in the infiltrate of islets at a late stage of prediabetes (20 weeks) in NOD mice [20]. Three further CD8 T cell epitopes were identified by screening a mIGRP library using NOD β2mnull.HHD mice, one of which was identical to the corresponding region in the human IGRP protein and the other two epitopes differed from the human sequence by just two amino acids [21]. Using a combination of peptide prediction by algorithm and major histocompatibility complex (MHC)-binding studies, two independent investigations have shown human T cell responses to IGRP152–160. This epitope was found to induce both granzyme B and IFN-γ production by peripheral blood mononuclear cells (PBMC) from patients with type 1 diabetes [11, 12], and responses to IGRP215–223 were also detected [12].

Unlike the previously published work, we used a nonselective approach to screen the range of all possible CD8 epitopes of PPI using a peptide library of 406 peptides—8–11mers spanning the PPI molecule, as the protein is small enough from a practical point of view to use this approach. In addition, as a comparison, we tested a broad range of candidate peptides of IGRP. We investigated CD8 T cell responses to PPI and IGRP, using the sensitive CD8-specific granzyme B ELISPOT assay in patients newly diagnosed with type 1 diabetes. We show that CD8 T cell responses to both PPI and to IGRP are extremely diverse and, when tested in this way, far broader than has been previously demonstrated using more selective approaches and that patients may be distinguished from control subjects by the breadth of these responses.

Materials and Methods

Subjects

Patients with newly diagnosed type 1 diabetes were recruited from the South West Newly Diagnosed Diabetes Collection (SWENDIC), a network covering 14 centers throughout the southwest of England and South Wales (n = 8, mean age 26.6, range 22–31 years). Patients were included if they were between 18 and 40 years at diagnosis, not overweight, presented with weight loss, and were ketosis prone. The patients were all positive for autoantibodies to GAD and/or IA-2. All patients were insulin-treated from the time of diagnosis. Healthy control subjects with no personal or family history (first- or second-degree relative) of type 1 diabetes or other autoimmune disorders were recruited from university staff and students (n = 7, mean age 28.9, range 25–36 years). All subjects were Caucasoid. Ethical permission was obtained for the study, and all individuals gave informed consent before taking part.

HLA Analysis and Measurement of Autoantibodies

HLA analysis was performed on deoxyribonucleic acid extracted from peripheral blood by polymerase chain reaction (PCR)-sequence-specific primers in the Histocompatibility and Immunogenetics Laboratory of the National Blood Service, Southmead Hospital, Bristol. Autoantibodies to GAD and IA-2 were measured in serum by radioimmunoassay by A.J.K. Williams and Prof. P.J. Bingley as previously described [22].

Peptides

The PPI and IGRP peptides were synthesized by Mimotopes (Melbourne, Australia) using a unique parallel synthesis technology, and control peptides for each set were checked by the manufacturer for purity, identity, and quantity by reverse-phase high-performance liquid chromatography, electrospray mass spectrometry, and amino acid analysis, respectively. The PPI panel consisted of a truncated PepSet™ of 406 peptides from PPI (Genbank accession CAA23828 PPI 110 amino acids). The PPI peptides were grouped into two pools of three peptides and 100 pools of four peptides (designated as pools 1–102, respectively). Each pool of four consisted of an 11mer peptide and three corresponding shorter peptides, each one further truncated by one amino acid from the amino terminus (8-, 9-, 10-, and 11mer), and pools covered the whole molecule in steps of one amino acid per pool (Fig. 1). The lengths were chosen to cover all possible MHC class I peptides. As the peptides were sequential and unselected for HLA type, the pools are likely to have contained peptides with different binding affinities for HLA-A2 and other HLA types. A further 61 individual peptides were synthesized from IGRP. Fifty-four of the IGRP peptides were chosen based on their predicted binding affinity for HLA-A*0201 using the online computer algorithm SYFPEITHI [23]. Peptides with an intermediate or high predicted binding affinity for HLA-A*0201 (SYFPEITHI score greater than 20) were selected. In addition, IGRP206–214 was chosen for testing, as this peptide has been previously identified as a CD8 epitope in NOD mice [19], and permutations of six IGRP peptides identified in studies in HLA-A2 transgenic mice, originally reported at American Diabetes Association Scientific Sessions (2005), were also synthesized [21]. The IGRP peptides were randomly grouped into 14 pools of four peptides and one pool of five peptides (designated later as pools 103–117, respectively; Table I). As all IGRP peptides had an intermediate to high predicted binding affinity for HLA-A2, peptides within individual pools were likely to bind to the HLA with a similar affinity.
https://static-content.springer.com/image/art%3A10.1007%2Fs10875-008-9177-4/MediaObjects/10875_2008_9177_Fig1_HTML.gif
Fig. 1

A truncated PepSet contains pools of peptides. Illustration shows a section of the B chain of PPI (residues 35–50) and the composition of two peptide pools from this region. Each pool consists of an 11mer, a 10mer, a 9mer, and an 8mer peptide formed from the truncation of the 11mer from the N terminus. The PepSet contains peptide pools covering the whole PPI protein in steps of one amino acid per pool

Table I

Sequences of Peptides in Each IGRP Pool

Pool number

Peptide size (aa)

Peptide sequences

103

9

RLLCALTSL

NLFLFLFAV

YLLLRVLNI

GLVRNLGVL

 

104

LTWSFLWSV

VLFGLGFAI

VIGDWLNLI

LLRVLNIDL

 

105

FIPYSVHML

GMDKFSITL

WLIQISVCI

QISVCISRV

 

106

VILGVIGGM

ILGVIGGML

FLFAVGFYL

LNIDLLWSV

 

107

TILQLYHFL

FLHRNGVLI

AVIGDWLNL

VTAALSHTV

 

108

ALSHTVCGM

AINSEMFLL

LTSLTILQL

SASIPLTVV

 

109

FLWSVFWLI

DLLWSVPIA

CALTSLTIL

TYLKTNLFLb

 

110

10

VLNIDLLWSV

LLRVLNIDLL

ALTSLTILQL

RLTWSFLWSV

 

111

FLHRNGVLII

AMGASCVWYV

LLLRVLNIDL

TLSFRLLCAL

 

112

SIPLTVVAFI

AVIGDWLNLI

VILGVIGGML

VIGGMLVAEA

 

113

FLFAVGFYLL

LLCALTSLTI

FIATHFPHQV

ILGVIGGMLV

 

114

SLGTYLKTNL

YLKTNLFLFL

VLSFCKSASI

WLNLIFKWIL

 

115

LLWSVPIAKK

HIDTTPFAGL

ILQLYHFLQI

AFIPYSVHML

 

116

9 and 10

TLHRLTWSFL

HTPGIQTASL

AVGFYLLLRV

VLNIDLLWSa

 

117

RVLNIDLLWa

GVLFGLGFAIa

VLFGLGFAINa

TVVAFIPYSVa

VVAFIPYSVHa

aa Amino acid

aPeptides selected based on permutations of peptides tested in [21]

bPeptide selected based on [19]

Isolation of PBMC

PBMC were isolated from 130 ml fresh peripheral blood (within 4 h of taking the sample) using Ficoll-Paque Plus according to the manufacturer’s instructions. PBMC were resuspended in complete RPMI medium (RPMI 1640 plus Glutamax plus 25 mM HEPES with 1% antibiotic/antimycotic, supplemented with 5% heat-inactivated human serum) and were transferred to round-bottomed culture tubes (250 μl per tube) at a concentration of 4–6 × 106/ml.

Peptide Stimulation of PBMC

Peptide pools were added to culture tubes at a concentration of 1.25 μg/ml per peptide. As the positive control, 50 ng/ml phorbol myristate acetate (PMA) and 1 μg/ml ionomycin or 0.4 × 106 anti-CD3 Dynabeads® (Dynal Biotech, UK) per 106 PBMC were added to culture tubes. The culture medium alone with PBMC was used as a negative control. To standardize culture conditions, dimethyl sulfoxide (DMSO), used for initial solubilization of peptides (1:1,000), was added to positive and negative controls. PBMC were incubated for 72 h at 37°C, 5% CO2 and washed with, and resuspended in, complete RPMI and peptides were added as previously.

Granzyme B ELISPOT

The granzyme B ELISPOT was performed according to the manufacturer’s instructions with minor modifications (U-cyTech, Utrecht, The Netherlands). Briefly, 96-well immuno plates (Nunc Maxisorp) were coated overnight at 4°C with anti-granzyme B capture antibody, washed ten times with phosphate-buffered saline, and blocked overnight at 4°C. Peptide-stimulated PBMC from one preincubation tube (1–1.5 × 106) were divided equally into three ELISPOT wells and incubated for 18–20 h at 37°C, 5% CO2. PBMC were then discarded, and spots were developed using biotinylated anti-granzyme B antibody, goat anti-biotin antibody, and activator solutions. Spots were analyzed using a BIOREADER® 4000 PRO-X ELISPOT plate reader (Bio-Sys GmbH, Karben, Germany), and spot counts were verified by eye using an inverted light microscope. The numbers of spots in triplicate wells were added and responses to peptide pools were divided by the background response, and a stimulation index (SI) was generated. We considered a response to a peptide pool that was three times the background response to be positive, that is, SI ≥ 3. In addition, we have added the number of spots in each triplicate and subtracted the background number of spots from this for each peptide pool and standardized the result as spot-forming cells (SFC) per million cells.

To set up the methodology and establish experimental conditions, an influenza peptide was used in the granzyme B ELISPOT assay. In those initial experiments, CD4 and CD8 depletion studies were also performed. These results showed that all responses were abolished when CD8 T cells were depleted but enhanced when CD4 T cells were depleted, indicating specificity of the response by CD8 T cells and not natural killer (NK) or NK T cells (data not shown).

Statistical Analysis

The SI values of the patients and control subjects were compared using Fisher’s exact test performed on Graphpad Prism software. We arbitrarily used response in two or more individuals as the minimal criterion for suggesting that an epitope might be of importance and also the more stringent criterion of response in four patients but no response in control subjects. Maximum difference analysis to identify pools of peptides to which responses of SI > 3 in the maximum number of patients but never in control subjects was performed with an algorithm using SAS 9.1 (2002–2003, SAS Institute, Cary, NC, USA). First, a cutoff value was chosen for the analysis, for example responses of SI ≥ 3 are positive. Therefore, the responses of all subjects were scored on the basis of this cutoff, with responses greater than or equal to 3 scored as “1” and all other responses scored as “0.” For each of the peptide pools, these binary values (1 or 0) were added firstly across the patient group and then separately across the control group, giving a total binary value for each peptide pool. Next, the limits for comparing the patients with the control subjects were chosen. For example, to identify those peptide pools to which four patients responded but to which no control subjects responded, the limits were 4 in the patient group and 0 in the control group. Based on those chosen limits, any total binary value for a peptide pool equal to 4 in the patient group was scored “1,” and all other pools were scored “0,” and in the control group, any total binary value for a peptide pool equal to 0 was scored as “1” and all others scored as “0.” Finally, for each peptide pool, the patient score was multiplied by the control score; thus, those pools whose product was “1” (1 × 1) showed the greatest difference between the patients and control subjects, both in the intensity of the response and in the number of responders.

To estimate the frequency of PPI-specific granzyme B-producing CD8 T cells for each patient and control subject, the number of SFC per million cells was calculated, as above, for each of the PPI peptide pools, and the background spots were subtracted. The number of SFC for each PPI pool was then added for each of the patients and control subjects to give an estimate of the total frequency of SFC per 100 million cells. The frequency of PPI-specific cells in patients and control subjects were compared using a Mann–Whitney U test.

Results

HLA Analysis and Autoantibody Measurements

The details of patients and control subjects together with HLA type and autoantibody status is shown in Table II. All patients were positive for autoantibodies to GAD or IA-2 or both. None of the control subjects were positive for either antibody. Seven of eight patients and five of seven control subjects expressed HLA-A2.
Table II

Demographics of Patients and Control Subjects

IDa

Sex

Ageb

Duration of T1D (days)b

Autoantibodyc

HLA

IA-2

GAD

A

B

C

DRB1

DQB1

ND1

F

26

20

1.5

38.5

3/24

18/39

5/7

3/8

2/4

ND2

F

22

82

1,154.3

141.12

2/3

15/37

3/6

1/4

3/5

ND3

F

24

66

3.4

47.39

2/25

8/40

3/7

3/4

2/3

ND4

M

26

11

2.2

478

1/2

8/44

5/7

3/4

2/3

ND5

F

26

69

1,030.4

109.72

2/31

40/40

3/6

4/4

3/3

ND6

M

31

72

10.8

693.42

2/2

40/44

3/5

4/4

3/3

ND7

M

30

207

2

821.3

2/3

35/57

4/6

3/4

2/3

ND8

M

28

134

19.8

2.9

2/2

7/15

3/7

4/13

3/6

C1

M

29

2.59

3.54

2/2

18/44

5/12

12/15

3/6

C2

F

36

1.92

4.63

1/2

8/50

4/7

3/4

2/4

C3

M

25

1.27

2.66

1/3

8/57

6/7

3/13

2/6

C4

M

27

1.94

8.38

1/2

35/40

3/4

1/15

5/6

C5

M

27

2.37

2.72

2/29

7/38

7/12

13/15

6/6

C6

M

32

1.83

4.56

2/11

7/49

4/7

4/15

3/6

C7

M

26

1.73

2.49

1/24

7/15

3/7

1/12

3/5

aND indicates patients, C indicates healthy control subjects

bWhen blood sample was taken. T1D - Type 1 Diabetes

cWorld Health Organization (WHO) units. IA-2 > 6 = positive, GAD > 14 = positive

CD8 T cell Responses to PPI and IGRP Peptides

Fresh PBMC from eight patients with newly diagnosed type 1 diabetes and seven healthy control subjects were stimulated with pools of PPI and IGRP peptides, and responses were measured using the granzyme B ELISPOT assay. An example of positive and negative responses is shown in Fig. 2. The volume of blood available from two patients was limited, and we were therefore only able to test the 102 PPI pools in one patient and the first 81 PPI pools in another patient.
https://static-content.springer.com/image/art%3A10.1007%2Fs10875-008-9177-4/MediaObjects/10875_2008_9177_Fig2_HTML.gif
Fig. 2

Representative positive and negative granzyme B ELISPOT responses. The top four wells show responses of one patient with type 1 diabetes and one control subject to the negative control (PBMC plus media and DMSO at 1:1,000) and the positive control (PBMC plus 50 ng/ml PMA and 1 μg/ml ionomycin or 0.4 × 106 anti-CD3 Dynabeads®/106 PBMC and DMSO at 1:1,000). The bottom four wells show negative and positive responses to peptide pools in a patient and a control subject. Each well shown is representative of the response observed in a triplicate of wells for each condition

All subjects secreted high levels of granzyme B in response to the positive control (PMA and ionomycin or anti-CD3 beads, Fig. 2). It was clear that when analyzed together, there were markedly more responses from the patient group as a whole compared with the control group (Fig. 3). Positive responses (SI ≥ 3) to PPI peptide pools were detected in all control subjects and patients (Fig. 3a,b), and responses to IGRP peptide pools were detected in three of seven control subjects and five of six patients (Fig. 3a,b). However, 63% of the PPI peptide pools stimulated responses in two or more patients compared with only 11% of the pools in the control group (p < 0.0001), and 33% of the IGRP peptide pools stimulated responses in two or more patients, whereas there were no pools that stimulated responses in two or more healthy control subjects (p = 0.042). Responses were markedly higher in patients, with 21% of positive responses to PPI pools reaching an SI ≥ 6 compared with just 7% of positive responses in control subjects reaching the same level (p < 0.0013). Twenty-six percent of patient responses to IGRP pools reached SI ≥ 6, whereas in the control subjects, no responses reached this level (not significant, p = 0.13). It should be noted that 45% of all positive responses to PPI and IGRP seen in the healthy control subjects were contributed by one individual who expressed the classic autoimmune haplotype HLA-A1, B8, DR3 but did not have diabetes associated autoantibodies tested in this study.
https://static-content.springer.com/image/art%3A10.1007%2Fs10875-008-9177-4/MediaObjects/10875_2008_9177_Fig3_HTML.gif
Fig. 3

Responses of patients with type 1 diabetes and healthy control subjects to PPI and IGRP. a Granzyme B ELISPOT responses to PPI peptide pools 1–102 and IGRP peptide pools 103–117 in healthy control subjects. Background responses were 1–5 SFC per million cells. b Granzyme B ELISPOT responses to PPI peptide pools 1–102 and IGRP peptide pools 103–117 patients with type 1 diabetes. Background responses were 1–5 SFC/million cells except for ND7 who had a background of 24 SFC/million cells. Responses are shown as a SI. Line indicates SI = 3, and responses of SI ≥ 3 are considered to be positive. A second line is drawn at SI = 6 to indicate the cutoff used for the separate analysis of the stronger responses observed. Colored shapes represent individual subjects

In comparing our results with those previously published (Table III), we have confirmed some of the existing epitopes as well as identified some potentially new epitopes. The analysis of our data to identify epitopes that are recognized only by patients and not by control subjects, using maximum difference analysis, has shown that pools 31, 41, 60, 79, and 99 (Table IV) elicited a positive response (SI > 3) in four of eight of our patient group but not in any of the control subjects. We have also used the SYFPEITHI database to ascertain the predicted binding affinities of the component peptides for the HLA-A and HLA-B types for the patients where a positive response was obtained (Table V). It can be seen that the predicted binding affinities for the potential epitopes for which data were available were low for all the epitopes (score less than 20) apart from one where the score was 25.
Table III

References of PPI and IGRP Epitopes in Humans

Reference

Referenced name

AA

Length

Peptide pool

Patients

Control subjects

Mallone et al. [7]

PPI 2–10

PPI 2–10

9

2

2

0

Mallone et al. [7]

PPI 6–14

PPI 6–14

9

6

2

0

Hassainya et al. [6]

B9-B18

PPI 33–42

10

34

0

1

Pinkse et al. [16], Hassainya et al. [6]

B10–B18

PPI 34–42

9

34

0

1

Toma et al. [5], Mallone et al. [7]

34–42

Toma et al. [5]

38–46

PPI 38–46

9

38

0

0

Kimura et al. [14]

Insulin 22–30a

PPI 39–47

9

39

2

1

Toma et al. [5]

39–48

PPI 39–48

10

40

3

1

Toma et al. [5]

41–50

PPI 41–50

10

42

2

0

Toma et al. [5], Mallone et al. [7]

42–51

PPI 42–51

10

43

3

1

Toma et al. [5]

44–51

PPI 44–51

8

43

3

1

Toma et al. [5]

45–53

PPI 45–53

9

45

3

1

Toma et al. [5]

49–57

PPI 49–57

10

50

0

0

Toma et al. [5]

51–61

PPI 51–61

11

53

2

0

Hassainya et al. [6], Mallone et al. [7]

C22–C30

PPI 76–84

9

76

1

0

Hassainya et al. [6]

C27–C35

PPI 81–89

9

81

4

1

Hassainya et al. [6], Mallone et al. [7]

C31–A5

PPI 85–94

10

86

2

1

Hassainya et al. [6]

A1–A10

PPI 90–99

10

91

2

1

Hassainya et al. [6], Mallone et al. [7]

A12–A20

PPI 101–109

9

101

0

1

Standifer et al. [11], Ouyang et al. [12]

IGRP 152–160

IGRP 152–160

9

109

0

1

Ouyang et al. [12]

IGRP 215–223

IGRP 215–223

9

106

1

0

Mallone et al. [7], Takaki et al. [21]

IGRP 228–236

IGRP 228–236

9

106

1

0

Mallone et al. [7], Takaki et al. [21]

IGRP 265–273

IGRP 265–273

9

104

1

1

Table shows an overview of the published epitopes of PPI and IGRP in humans. Columns indicate (in order): author, epitope name used in the publication, a standardized epitope name (based on amino acid positions in PPI, 1–110), epitope length, peptide pool in the current study that the epitope is found in, and number of patients and control subjects from the current study that responded to the pool containing the epitope

aLYLVCGERG is referenced here as insulin B chain 22–30; however, the correct designation is insulin B chain 15–23 [13]

Table IV

Sequences of Peptides in a Discrete Group of PPI Pools that Discriminate Between Patients and Control Subjects

Pool number

PPIa

Peptide sequences

11mer

10mer

9mer

8mer

31

29–39

HLCGSHLVEAL

LCGSHLVEAL

CGSHLVEAL

GSHLVEAL

41

39–49

LYLVCGERGFF

YLVCGERGFF

LVCGERGFF

VCGERGFF

60

58–68

AEDLQVGQVEL

EDLQVGQVEL

DLQVGQVEL

LQVGQVEL

79

77–87

LQPLALEGSLQ

QPLALEGSLQ

PLALEGSLQ

LALEGSLQ

99

97–107

TSICSLYQLEN

SICSLYQLEN

ICSLYQLEN

CSLYQLEN

aAmino acids numbered from 1 to 110 starting from the amino terminus of PPI

Table V

MHC–Peptide Binding Scores for Sequences in Five PPI Peptide Pools that Discriminate Between Patients and Control Subjects

Pool

Sequence

Patients

HLA in patients

HLA in database

MHC–peptide binding score

8

9

10

11

31

HLCGSHLVEAL

ND2

A2

A*0201

16

17

ND3

A3

A*03

1

1

ND6

B7

B*0702

14

13

ND8

B8

B*08

12

12

 

B15

B*1501

3

2

 

 

B*1510

13

 

B44

B*4402

14

13

41

LYLVCGERGFF

ND2

A2

A*0201

8

12

ND3

A3

A*03

18

16

ND5

B8

B*08

17

6

ND6

B15

B*1501

16

19

 

 

B*1510

8

 

B44

B*4402

12

11

60

AEDLQVGQVEL

ND2

A1

A*01

0

3

1

12

ND3

A2

A*0201

25

13

ND4

A3

A*03

16

6

ND7

B8

B*08

10

19

 

B15

B*1501

10

4

 

 

B*1510

15

 

B44

B*4402

11

14

79

LQPLALEGSLQ

ND1

A2

A*0201

9

2

ND2

A3

A*03

18

12

ND5

A24

A*2402

2

4

ND8

B7

B*0702

2

11

 

B15

B*1501

10

0

 

 

B*1510

1

 

B18

B*18

3

0

 

B39

B*3901

2

99

TSICSLYQLEN

ND1

A2

A*0201

6

13

ND2

A3

A*03

6

12

ND5

A24

A*2402

0

0

ND6

B15

B*1501

1

0

 

 

B*1510

4

 

B18

B*18

4

2

 

B39

B*3901

6

 

B44

B*4402

3

2

Binding scores were predicted using the SYFPEITHI algorithm [23]

To show the overall performance of the granzyme B ELISPOT assay, Table VI summarizes the responses of each individual to the positive and negative controls as SFC per million cells. In addition, the range of responses detected to all peptide pools and the responses to the five pools identified by maximum difference analysis are also shown for each patient and control subject. Furthermore, we have estimated the frequency of granzyme B-secreting, PPI-specific CD8 T cells in the patients and control subjects and shown the result in Fig. 4. This indicates that, overall, patients have a significantly higher frequency of CD8 T cells responding to PPI than control subjects (p = 0.014).
https://static-content.springer.com/image/art%3A10.1007%2Fs10875-008-9177-4/MediaObjects/10875_2008_9177_Fig4_HTML.gif
Fig. 4

Frequency of PPI-specific granzyme B-producing CD8 T cells in patients and control subjects. PPI-specific responses are shown as SFC per 100 million cells. The bar indicates the median number of SFC in each group of subjects, and the median value and the range of responses in each subject group is shown

Table VI

Performance of the Granzyme B ELISPOT Assay

ID

Negative control

Positive control

Range for all peptide pools

Peptide pool

31

41

60

79

99

ND1

1

104

0–8

1

1

1

4

8

ND2

1

81

0–15

3

3

3

3

6

ND3

2

267

0–21

12

8

9

0

-

ND4

3

23

0–12

2

1

10

3

3

ND5

1

322

0–5

0

3

2

3

4

ND6

2

110

0–37

20

8

3

4

7

ND7

24

113

2–119

42

13

87

4

31

ND8

5

48

0–25

23

7

5

19

0

C1

4

111

0–21

2

2

3

1

1

C2

2

148

0–8

3

1

2

2

3

C3

1

54

0–16

1

1

1

1

1

C4

3

179

0–12

3

1

4

3

1

C5

3

210

0–14

1

0

1

4

6

C6

5

127

0–29

2

11

2

5

4

C7

1

73

0–3

1

0

0

0

1

Data are shown as the number of SFC per million cells

Discussion

In this report, we have, for the first time, used a CD8-specific technique (granzyme B production) to study the complete range of peptide epitope responses at diagnosis of type 1 diabetes to a major autoantigen (PPI). Our findings indicate firstly that the CD8 response at this stage of the disease is very much broader than previously appreciated. Using our approach in eight patients, we saw responses to 11 pools containing 12 of 18 previously reported epitopes of PPI (Table III) but also identified 53 additional epitopic regions that elicited responses in two or more patients. Five of these regions elicited responses in four patients (50%) but none of the control subjects (Table IV), giving them a high likelihood of being of disease relevance. Although there was some sequence overlap between the peptides in these five pools and previously reported epitopes, in four of these, the overlap was insufficient for any of the epitopes contained in the pools to be identical to the previously studied peptides (Table IV). The granzyme B ELISPOT readout in our assays is more specific than the IFN-γ assays used in most of the other studies, and this may account, in part, for the identification of new regions of PPI in this study. It should be noted that we also observed stronger T cell responses in the newly diagnosed patients than in control subjects: 21% of the positive responses in patients had an SI > 6, whereas this was only true of 7% of control responses (p = 0.0013). Overall, the patients had significantly more PPI-specific CD8 T cell responses than the control subjects (Fig. 4).

Second, our findings suggest that many of the CD8 responses seen in newly diagnosed patients are to epitopes with very low predicted peptide–MHC binding affinity such that they would not be detected by conventional epitope prediction techniques. For example, when the 20 possible peptides of the five regions of PPI identified in four of eight patients, discussed above, were analyzed for predicted SYFPEITHI binding scores to the HLA-A and B types of the patients who responded, 66 of the 67 peptide binding scores available were less than 20, and the remaining 9mer from pool 60 had a binding score of 25 for HLA-A*0201 (Table V). Our results concord with those of Ouyang et al. who reported an inverse relationship between MHC–peptide binding and T cell response measured by IFN-γ production. They also showed that peptides with a high predicted binding affinity for HLA-A*0201 did not, in fact, demonstrate high-affinity binding during the binding assays [12]. Furthermore, those peptides that did bind with a high affinity were not the most immunogenic peptides in the T cell assays [12]. Indeed Standifer et al. [11] also reported that the peptide that generated responses in the most patients and autoantibody positive relatives had demonstrated an intermediate to low affinity for HLA-A*0201 in their binding assays.

Mallone et al. [7] have recently reported that a limited panel of epitopes can identify HLA-A*0201 patients with high sensitivity and is sufficient for the prediction of disease. It is well known that epitope spreading occurs during autoimmune disease progression [24, 25], certainly for CD4 T cells [26], and this may be so for CD8 cells. It is postulated that responses to dominant peptides are likely to occur early in pathogenesis, unmasking cryptic epitopes [27]. Could it be that the wide range of responses to low-affinity epitopes we have observed simply represent the results of epitope spreading after disease initiation and are unlikely to be of pathogenic relevance? This is currently impossible to determine for human epitopes, but experience in the NOD mouse suggests otherwise. In our own studies, we found low-affinity epitopes to be powerful activators of CD8 T cells and to include a key pathogenic epitope [28]. This finding has recently been replicated in HLA-A*0201 transgenic mice in which insulin A2–10 was observed to be an immunodominant epitope, but its binding to HLA-A*0201 was undetectable in vitro [29]. Furthermore, cryptic epitopes have been found to be very effective inducers of tolerance in late-stage disease in NOD mice [30]. Hence, it would be premature to dismiss responses seen to low-affinity epitopes in humans as of no importance in either the disease process or immunointervention approaches.

We compared the broad range of responses seen in unselected CD8 analysis of PPI to the results of screening IGRP with peptide epitopes selected by prediction algorithms. We tested 61 IGRP peptides with SYFPEITHI scores of greater than 20 in pools of 4–5 and obtained relatively few responses. Overall, the IGRP binding affinities were mostly high—SYFPEITHI score greater than 20. Therefore, there were no lower-affinity peptides within those pools that could have been “outcompeted” for HLA-A2 binding by higher-affinity peptides. Thus, competition for HLA binding is not likely to be the reason here for the relatively low response observed. In the patients, responses were seen in two pools, which contained three of the four previously reported epitopes [7, 11, 12]; however, overall, the proportion of IGRP pools to which two or more patients responded was far lower than for the PPI pools (33% compared with 63%). Our findings for PPI suggest that responses to more IGRP peptide pools may have been identified had we chosen peptides that bind the MHC with a lower affinity. It is also notable that even if peptides are chosen with reference to one MHC type, binding and presentation may occur through other MHC molecules, as we found in this study, when responses were seen to peptides of IGRP selected in reference to HLA-A*0201 but in a control who did not express HLA-A*0201. As with other selective studies of IGRP epitopes, some responses may have been missed, although it should also be pointed out that PPI and IGRP are autoantigens of a different nature. PPI is a preprohormone, ultimately cleaved to release a circulating hormone, and IGRP is an enzyme within the beta cells. It is possible that the proteins may generate different autoreactive responses that may also vary with the course of disease.

The requirement to test such a broad range of peptides in the current study inevitably leads to limitations. First, we were only able to study adult subjects (because of the volume of blood required—up to 130 ml), and we acknowledge that responses may be different in children with newly diagnosed disease. Second, because of the large number of peptides to be tested, the studies could not have been done without peptide pooling. It is possible that we have not identified some PPI epitopes in the patients, as higher HLA binding affinity peptides in the pools, which do not stimulate responses in the patients, may have competed with lower HLA binding affinity peptides, masking a response. Third, it was not possible to retest the individual peptides in each pool to confirm the exact epitope recognized or to grow T cell lines or clones to the peptides to reconfirm the nature of the responses and their HLA restriction. Finally, we were only able to study the smallest of the major autoantigens, PPI, in this way (110 amino acids) as this has many fewer amino acids than IGRP (335 amino acids), GAD65 (585 amino acids), GAD67 (594 amino acids), and IA-2 (979 amino acids). Nonetheless, there is much evidence from both mouse models [13, 15, 29, 31] and human studies (Table III, [57, 16]) that PPI is a highly relevant autoantigen for CD8 T cells in autoimmune diabetes. It should be noted that we detected a number of responses to epitopes within the signal peptide, the C-peptide, and to those that span the B–C and C–A junctions, in addition to those solely within the B or A chain, making it unlikely that these responses are related to the administration of exogenous insulin (which lacks the C-peptide region and the signal peptide) after diagnosis.

Thus, from our study, our major conclusion is that using a complete overlapping library of peptides to screen PBMC samples from patients with type 1 diabetes can elicit far broader responses than has been shown previously using other approaches, and those responses are more diverse than the responses seen in healthy control subjects. In addition, peptides with low predicted HLA binding affinities can stimulate responses in patients with type 1 diabetes and yet may be overlooked if only those peptides with a higher predicted HLA-binding affinity are selected for study. We believe our findings have important implications for further studies of CD8 responses in type 1 diabetes in humans. They emphasize the lessons learned from NOD mouse studies that selecting candidate epitopes based on binding affinity alone may result in missing a large number of CD8 target epitopes, some of which may be highly relevant to pathogenesis and immunointervention. We acknowledge that comprehensive peptide library screening is not feasible for the larger autoantigens, but alternative approaches exist for peptide selection, such as peptide elution from disease-related HLA molecules [32] or “reverse immunology” examining proteosome processing products [6]. Where selected peptide panels are used to study the CD8 T cell responses, it should be appreciated that although these are hopefully representative of the autoimmune process, there is a broad range of CD8 activity that goes undetected in these assays and may be relevant especially in patients who are “negative” to the “core” panel. The challenge will be to develop a method that will reflect the true range of peptide responses, sufficient to take patient heterogeneity into account and allow for monitoring of immune responses in the natural history of disease as well as in response to therapy.

Acknowledgments

C.B. was supported by a Diabetes UK studentship. F.S.W. was a Wellcome Trust Senior Fellow in Clinical Science. We are very grateful to Mr. L. Keen at the National Blood Service for the HLA typing analysis and to Mr. A.J.K. Williams and Prof. P.J. Bingley for the autoantibody measurements. We would also like to thank Mr. S.J. Chapman for his technical assistance and the physicians and diabetes specialist nurses in all of the collaborating centers for referring newly diagnosed patients to SWENDIC.

Copyright information

© Springer Science+Business Media, LLC 2008