Heparinized peripheral blood was obtained from 34 patients with verified primary adrenocortical insufficiency due to AAD, included in the Norwegian National Registry for Organ-Specific Autoimmune Diseases (ROAS). Patient characteristics are summarized in Supplementary Table 1 . Serving as controls, peripheral blood was obtained from 27 healthy blood donors (C1-C6, C9, and C11-C30) from the Blood Bank at Haukeland University Hospital (Bergen, Norway), and three individuals with other (non-adrenal) autoimmune diseases (C7, C8 and C10) included in ROAS. The study was approved by the Regional Committee for Medical and Health Research Ethics, project number 2018/1417. All patients and controls from ROAS had given informed consent approved by the Health Region West ethics committee (147/96-47·96). Healthy individuals had given informed consent at the Blood Bank.

Autoantibodies against 21OH were measured by routine radioimmunoassays as described by Oftedal et al. (14), except that dithiothreitol was omitted from the sample buffer. Positivity was defined as a threshold index ration based on a pool of 200 blood donor sera and a positive control serum from an AAD patient containing high levels of 21OH autoantibodies. All AAD patients were positive for autoantibodies against 21OH, confirming autoimmune etiology.

The expression of HLA-A2 and HLA-B8 molecules was determined by flow cytometry on lysed whole blood or PBMCs using FITC conjugated anti-human HLA-A2 (Biolegend, clone BB7.2) and APC conjugated anti-human HLA-B8 (Miltenyi Biotec, clone REA145) antibodies, respectively. Genotyping of HLA-C7 alleles was performed by PCR using sequence-specific primers as described by Bunce et al. (15). For the majority of the patients and a subgroup of the controls, the presence of specific HLA alleles could be confirmed by other genotyping methods from other studies, such as sequence-specific oligonucleotide probe system (8) and allelic imputation from genome-wide array genotyping (10).

Custom designed, PE-labeled HLA-A*0201 and HLA-B*0801 dextramers containing the LLNATIAEV (A2*LLNATIAEV) and EPLARLEL (B8*EPLARLEL) peptides, respectively, were purchased from Immudex. Custom designed PE-labeled HLA-C*0701 streptamer containing the ARLELFVVL (C7*ARLELFVVL) peptide was purchased from IBA Lifesciences. Purified LLNATIAEV and EPLARLEL peptides were purchased from Proimmune/thinkpeptides. Peptides used for epitope mapping were purchased from GenScript and included the long C-terminal peptide GEPLARLELFVVLTRLLQ (Pep34, 21OH_430-447) and ten consecutively overlapping 9-mer peptides covering the same sequence (Pep1-Pep10). All peptides were reconstituted in cell culture-grade DMSO at 10-40 mg/mL to obtain stock solutions and diluted in RPMI-1640 (Life Technologies) prior to use.

PBMCs were isolated from heparinized blood by Ficoll-Paque PLUS (GE Healthcare) density gradient centrifugation. Fresh samples were available from five of the patients, while cryopreserved cells (preserved in 90% AB serum/10% DMSO (Sigma-Aldrich) at -150°C) were used for the remaining patients and healthy controls. For some patients and controls, not enough cells were available for all experiments described below. Accompanying tables to Figures 1 – 4 are available in the Supplementary materials, containing details about which patient and control samples were included for each experiment. PBMCs were cultured in RH-10 medium consisting of RPMI-1640, 10% human AB serum (Sigma-Aldrich), 1% non-essential amino acids, 1% Na-Pyruvate, 1% HEPES, 1% L-glutamine, 1% penicillin-streptomycin (all from Lonza) and 55 µM β-mercaptoethanol (Sigma). On day 0, PBMCs were resuspended in RH-10 medium containing IL-7 (Biolegend, 25 ng/mL) to provide survival signals for naïve and memory T cells (16, 17) and stimulated with 21OH peptides LLNATIAEV, EPLARLEL or ARLELFVVL (10 µg/mL) in 24-well plates at 1.5-3.5x10⁶ cells/mL. On day 3, cells were replenished with fresh RH-10 medium containing IL-2 (Thermo Fisher Scientific, 50 U/mL) and this procedure was repeated every three days thereon until day 9. The cell cultures were split when reaching high confluency (>90%) or when acidification of the medium was evident. For patients P3 and P11 only few LLNATIAEV specific T cells could be demonstrated ex vivo. Therefore, an alternative protocol was used where IFNγ-producing cells in response to 6 hour stimulation with LLNATIAEV (day 0) were isolated using an IFNγ Secretion Assay (Miltenyi Biotec), and then expanded with IL-2 as outlined above. On day 13, cells were harvested for dextramer or streptamer staining, and IFNγ ELISPOT was performed for selected samples.

PBMCs were stained with PE-labeled HLA-A2*LLNATIAEV and/or HLA-B8*EPLARLEL dextramers or HLA-C7*ARLELFVVL streptamers on day 0 and following expansion (day 13). One million PBMCs were stained on day 0 and 4x10⁵-1x10⁶ PBMCs on day 13. The amount of dextramer used for staining was adjusted accordingly. Cells were initially centrifuged at 400 x g before being resuspended in 50 µL 5% FBS/PBS (v/v) and stained with 4-10 µL of dextramers for 10 min at room temperature, or with 4 µl of streptamers for 45 min on ice. The cells were then washed, centrifuged and resuspended in 100-200 µL 5% FBS/PBS, and stained with antibodies towards CD4, CD19, CD14 (all FITC), and CD8 (APC) for 20 min on ice. As part of a test setup, cells from P3 were only stained with dextramers and CD8 antibodies (no negative gate with CD4/CD14/CD19). Finally, cells were washed, centrifuged and resuspended in 200 µL 5% FBS/PBS before being analyzed on an Accuri C6 flow cytometer. The gating strategy is shown in Supplementary Figure S1 . To reduce false positives, a strict gate was drawn for dextramer/streptamer-positive cells, based upon fluorescence minus one (FMO) controls for each individual sample.

Quantification of IFNγ-producing cells upon stimulation or restimulation with 21OH peptides was performed with human IFNγ ELISpot^PRO or IFNγ ELISpot^PLUS kits (Mabtech). On day 0 (ex vivo), 6x10⁵ PBMCs were stimulated with LLNATIAEV, EPLARLEL, ARLELFVVL, or Pep34 (10 µg/mL), or with DMSO (0.1% or 0.025%) (blank), in 600 µL RH-10 medium and seeded in triplicates in pre-blocked (RH-10 medium) ELISPOT plates. Plates were incubated for 40h at 37°C before being developed in accordance with the manufacturer’s instructions. Spot-forming cells (SFC) in each experiment were counted manually and independently by two or three individuals.

For LLNATIEV and EPLARLEL, cells from some patients and controls were also tested for IFNγ production on day 13 after expansion. Autologous or syngeneic (HLA-A2 or HLA-B8-matched) PBMCs, depending on the availability of autologous cells, were thawed and subjected to magnetic-activated cell sorting with CD14 and blood dendritic cell antigen-4 (BDCA-4) microbeads (Miltenyi Biotech). The purified population contains monocytes and macrophages (CD14), as well as plasmacytoid dendritic cells (BDCA-4) which possess cross-presenting abilities (18). Sixty thousand CD14⁺/BDCA-4⁺ cells were used as APCs and pulsed with LLNATIAEV or EPLARLEL peptide (2 µg/mL), or with DMSO (blank), for 2 hours at 37°C. Next, 6x10⁵ expanded cells were co-cultured with pulsed APCs in 200 µL RH-10 in pre-blocked ELISPOT plates. For P9, HLA-A2⁺ Priess lymphoblastoid B cells were used as APCs. For P3, 1.5x10⁴ expanded cells were co-cultured with 7.5x10³ peptide or DMSO-pulsed autologous APCs. Plates were incubated for 24 hours at 37°C prior to development and counting as described above.

For mapping of alternative epitopes within 21OH_430-447 sequence, ex vivo PBMCs were stimulated with individual overlapping 9-mer peptides and seeded in ELISPOT wells as described above for day 0. The plates were developed and analyzed accordingly.

Supernatants from selected ELISPOT assays at both day 0 and day 13 were harvested and kept at -80°C for Milliplex analyses and/or ELISAs to screen for other relevant cytokines in response to peptide stimulation. A human CD8⁺ T cell premixed multiplex assay comprising the following 17 cytokines was purchased from Millipore (catalogue no: HCDM8MAG15k17PMX): GM-CSF, sCD137, sFas, sFasL, Granzyme A, Granzyme B, IFNγ, IL-2, IL-4, IL-5, IL-6, IL-10, IL-13, MIP-1α, MIP-1β, Perforin and TNFα. Twenty-five microliter samples from ELISPOT assays were run in duplicates or single wells. The assay was performed as recommended by the manufacturer and Luminex 100 (Luminex Corp) and STarStation Software v.3.0 (Applied Cytometry System) was used for recording and analyzing the data. For IL-6 and TNFα, supernatants were also tested with traditional sandwich ELISA (Thermo Fisher Scientific).

To evaluate other potential HLA-B*0801-restricted epitopes within Pep34, we used a custom epitope discovery service offered by ProImmune (19). The service consisted of the following four modules: custom peptide synthesis, MHC-peptide binding assay, MHC pentamer synthesis, and MHC-peptide complex off-rate (t_1/2) assay. Two sets of overlapping 8-mer and 9-mer peptides covering the Pep34 sequence (GEPLARLELFVVLTRLLQ) were synthesized (module 1) and tested for their ability to bind and stabilize HLA-B*0801 compared to a known T cell epitope (module 2). Candidate binders identified in module 2 were selected for HLA-B*0801 pentamer synthesis (module 3) and the stability (“off-rates”) of the resulting complexes were tested at 6 time points over a 24h period to approximate their half-life (module 4).

The NetMHC 4.0 Server was used to predict binding of 21OH peptides to specific HLA class I molecules (http://www.cbs.dtu.dk/services/NetMHC/) (20, 21). The coordinates for the X-ray solved crystal structure of HLA-C*0702 in complex with the peptide ligand RYRPGTVAL (pdb id: 5VGE) (22), were downloaded from the Protein Data Bank at https://www.rcsb.org. Potential interactions between ARLELFVVL and HLA-C*0701 were modeled by PyMOL Molecular Graphics System v.2.2, with the use of the mutagenesis tool in the software to replace the RYRPGTVAL peptide with ARLELFVVL and to replace Lys90 and Ser123 of the HLA-C molecule with Asn90 and Tyr123, respectively. The APBS electrostatics plugin of PyMOL was used to visualize the electrostatic potential of the peptide binding cleft of HLA-C*0702.

To determine if proteins from human pathogens contained homologous sequences to 21OH-derived T cell epitopes, we performed a protein BLAST search (https://blast.ncbi.nlm.nih.gov/) using the 9-mer sequences LLNATIAEV and ARLELFVVL. The peptide sequences were blasted against a list of the most common microbial pathogens known to colonize humans in Norway, available from the web pages of the Norwegian Institute of Public Health (https://www.fhi.no/sv/smittsomme-sykdommer/sykdommer/). Only microbes containing homologous sequences to both LLNATIEV and ARLELFVVL were considered. Default settings in the protein BLAST search engine were used. Microbial peptide hits containing gaps were not tolerated.

Comparisons of peptide specific SFC frequencies and dextramer/streptamer frequencies between patients and controls were tested by non-parametric Mann-Whitney U-test using GraphPad Prism v5.02. P-values below 0.05 were considered significant. Only significant P-values are shown in figures.

Initial staining (day 0) of PBMCs from AAD patients with highly sensitive MHC dextramers revealed frequencies of LLNATIAEV- and EPLARLEL-specific T cells ranging from 0.002% to 0.183% of total CD8+ T cells ( Figure 1A ). The frequency of LLNATIAEV-specific cells was significantly higher in patients than controls (P = 0.0127). In the patients, LLNATIAEV-specific cells were also more frequent than EPLARLEL-specific cells (P = 0.0018, data not shown). There was no difference in the frequency of EPLARLEL-specific cells between patients and controls ( Figure 1B ). Accordingly, the highest individual frequency observed was for LLNATIAEV-specific cells (P5, 0.183% of CD8+ T cells). Small populations of bright T cells stained with A2*LLNATIAEV were present in five of the patients (P2, P5, P7, P9 and P10), but none of the controls, on day 0. For all individuals expressing both HLA-A2 and HLA-B8, LLNATIAEV-specific cells were more frequent than EPLARLEL-specific cells ( Supplementary Figure 2 ), hence LLNATIAEV was chosen as the stimulating peptide in expansion cultures for these patients (except for P11 where both were used).

Following enumeration of peptide-specific CD8+ T cells on day 0, PBMCs were stimulated with LLNATIAEV and/or EPLARLEL peptide in the presence of IL-7. The cells were allowed to expand for 13 days supported by IL-2 supplementation (supplied on day 3, 6 and 9), and then stained with dextramers again ( Figures 1C – E ). For P9, there was too few cells for analyses on day 13. Distinct bright populations of peptide-specific cells were evident in 7 out of 11 patients on day 13, including four of the patients that harbored such populations at day 0 (P2, P5, P7 and P10) ( Figure 1C ). Only one patient, P8, presented with a distinct cellular population stained by the B8*EPLARLEL dextramers, but only at frequencies 10-100 fold lower than for A2*LLNATIAEV ( Figure 1D ). The strongest expansion was seen for patients P2, P3, P5, P7 and P10, resulting in frequencies up to 16% of total CD8+ T cells. In the remaining four patients without distinct populations, there were negligible or no increase in the number of dextramer-positive cells. Surprisingly, strong expansion was also evident for C1 and to a certain degree C9, giving rise to populations of LLNATIAEV-specific cells that accounted for 2.5% and 0.5% of CD8+ T cells, respectively. Notably, none of the controls tested positive for 21OH autoantibodies.

We followed up previous reports on IFNγ responses to peptides 21OH_337-354 and 21OH_431-450 using the shorter, predicted dominant epitopes LLNATIAEV (13) and EPLARLEL (12), respectively. IFNγ ELISPOT assays were performed both ex vivo and with expanded cells ( Figures 2A, B ). Ex vivo IFNγ SFC counts in response to LLNATIAEV were highest in the patients harboring small populations of A2*LLNATIAEV-staining cells on day 0 (P2, P5, P7, P9 and P10). Although ex vivo responses in general were low, a significant difference between patients and controls were noted for IFNγ SFCs (P <0.05, Figure 2A ). For HLA-B8 carriers, SFC counts rarely exceeded background in response to EPLARLEL stimulation ( Figure 2B ).

After two weeks of expansion with the selected peptides, IFNγ responses seemed confined to patients in which there was clear expansion of LLNATIAEV-specific cells (P2, P3, P5, P7 and P10) ( Figure 2A ). Similarly, LLNATIAEV-specific IFNγ-producing cells were also frequent in control C1 who displayed a distinct population of cells with such specificity on day 13. Again, we could not detect specific responses to EPLARLEL ( Figure 2B ). Thus, IFNγ responses were largely restricted to LLNATIAEV and mainly present in the patients (both ex vivo and upon in vitro expansion).

Beyond IFNγ, there are few data on the cytokines produced by autoreactive T cells in AAD. In order get insight into such cytokine profiles, we performed a preliminary cytokine screen on supernatants from day 0 and day 13 ELISPOT assays, using a panel of 17 cytokines associated with CD8⁺ T cell responses. Unfortunately, due to limited amounts of material, we were only able to include some of the patients exhibiting peptide-induced expansion and IFNγ responses above background, as well as three control subjects. A complete overview of the results from the entire panel is given in Supplementary Tables 2 – 7 . Four cytokines appeared to be secreted by cells from several patients, specifically in response to ex vivo peptide stimulation on day 0: IL-6, TNFα, MIP-1α and MIP-1β. Some of these data were confirmed by performing individual ELISAs for two of the cytokines; IL-6 and TNFα. Supplementary Figure 3 displays the results from ELISA (S3A-D) and multiplex (S3E-H) measurements. However, upon restimulation on day 13, virtually all responses faded or were even repressed compared to unstimulated cells. A similar effect was seen on some of the other cytokines in individual patients. However, due to the relative small number of patients and controls and the lack of appropriate replicates for some of the individuals tested, these results must be interpreted with caution.

Still, the overall impression from measuring cytokine responses to 21OH peptides was therefore that IFNγ production as measured by ELISPOT stood out as the most reliable cytokine biomarker for peptide-specific CD8⁺ T cells, in particular for expanded cells.

Our dextramer and cytokine data ( Figures 1 , 2 and Supplementary Figure S2 ) did not support previous reports suggesting EPLARLEL to be a major CD8⁺ T cell epitope in AAD. Most of the data from these publications were from stimulation experiments with longer peptides (21OH_431-450 and 21OH_430-447). Therefore, we suspected that alternative sequences were targeted, and performed a detailed epitope mapping of 21OH_430-447.

PBMCs from patients and controls were stimulated with a panel of overlapping 9-mer peptides (Pep1-10) covering Pep34 (21OH_430-447), as well as Pep34 itself, and were assayed for IFNγ secretion using ELISPOT ( Figure 3 ). Five out of ten tested patients displayed responses above background/DMSO to Pep34, and among these, four had a clear response to Pep5 (peptide sequence: ARLELFVVL, 21OH_434-442, Figure 3 ). Responses to the other peptides were generally low.

In order to test whether ARLELFVVL could represent an alternative HLA-B8 restricted epitope, we took advantage of the REVEAL epitope discovery system offered by ProImmune, in which 8-mer and 9-mer peptides covering Pep34 were synthesized and tested for their ability to bind HLA-B*0801 ( Supplementary Tables 8 – 11 ). Individual binding scores (REVEAL scores) were calculated as percentage binding compared to a positive control peptide. Based on a threshold of ≥45% of the control, ARLELFVVL did not show any affinity to HLA-B*0801 (REVEAL score = 0.3), in contrast to EPLARLEL that was considered to be a strong ligand (REVEAL score = 78).

As we had established that ARELELVVL could not be a HLA-B*0801 ligand, we started to search for alternative HLA class I molecules able to present this peptide. Initially, we used the NetMHCpan-4.1 server that predicts the binding of peptides to specified HLA molecules. We had a particular emphasis on the ancestral haplotype 8.1 (AH 8.1) as the HLA class I alleles (HLA-A*01 and HLA-C*0701) within AH 8.1 is in strong linkage disequilibrium (LD) with the predisposing HLA-B8, -DR3 and -DQ2 alleles. HLA-A*01 was not predicted to bind ARLELFVVL, or to any possible 8-, 9-, 10- or 11-mer peptide from Pep34 (21OH_430-447). In contrast, HLA-C*0701 was predicted to bind ARLELFVVL, but not any other possible 8-, 9-, 10- or 11-mer peptide from Pep34. We also assessed to what degree other HLA class I molecules not part of the AH 8.1 were predicted to bind ARLELFVVL ( Supplementary Table 12 ). Several other HLA class I molecules were predicted to bind ARLELFVVL, but these were all rare among AAD patients, with allele frequencies ranging from 0.0% to 5.7%. The only exception was HLA-C*0702, present in 14.5% of AAD patients, which only differs from HLA-C*0701 by two amino acids and has a very similar peptide binding pocket. Notably, HLA-C*0701 is the most common HLA-C allele among Norwegian AAD patients, being present in 33.4% of patients but only 14.7% of controls (10).

As the crystal structure of HLA-C*0702 has been solved, we evaluated the electrostatic potential of its peptide binding cleft, and modeled potential interactions of ARLELFVVL and HLA-C*0701 by use of the PyMOL software ( Supplementary Figure S4 ). The electrostatic potential of the B pocket of HLA-C*0701 protein, which probably contains an anchor site of peptide ligands, displayed an excess of negative charges at the surface, indicating a potential to bind positively charged peptide residues (S4A). Indeed, the molecular models indicated that the positively charged arginine residue at position 2 of ARLELFVVL may serve to anchor the peptide to the floor of the antigen-binding cleft of HLA-C*0701 via a salt bridge with aspartic acid at amino acid position 9 (p.Asp9) and a hydrogen bond with serine at amino acid position 24 (p.Ser24) of the HLA molecule (S4B). At the same time, hydrophobic interactions between leucine 9 of the peptide and leucines at positions 81 and 95 of HLA-C*0701 could potentially ensure the stabilization of the peptide-HLA interaction further (S4C).

To confirm the relationship between HLA-C*0701 and ARLELFVVL (21OH_434-442), we purchased custom-designed C7*ARLELFVVL streptamers to determine the frequency of ARLELFVVL-specific CD8⁺ T cells among PBMCs isolated from 13 patients (P18-20, P23, P25-P30, P32-P34). Ex vivo staining of PBMCs with streptamers revealed populations of ARLELFVVL-specific CD8⁺ T cells in >70% of the patients tested ( Figure 4A ). The highest individual frequency observed was 0.32% of total CD8⁺ T cells. The overall frequency of ARLELFVVL-specific cells was significantly higher in patients than controls (P = 0.0021).

Subsequently, we stimulated cells from six of the patients and five of the controls with the ARLELFVVL peptide and attempted to expand the cells for 13 days under similar conditions as for LLNATIAEV and EPLARLEL. A marked expansion was evident in all patients, including some with low levels of ARLELFVVL-specific cells on day 0 ( Figure 4B ), with frequencies of C7*ARLELFVVL streptamer-positive cells reaching as high as 1.5% of CD8⁺ T cells on day 13 ( Figure 4C ). In contrast, no signs of expansion were seen for any the controls.

Having demonstrated the presence of circulating ARLELFVVL-specific CD8⁺ T cells in AAD patients, we also assessed their ability to secrete IFNγ by ELISPOT. PBMCs from 14 patients (P18-P20, P23-P33) and 14 healthy controls were stimulated with ARLELFVVL and Pep34 for 40h. Together, the patients had significantly higher IFNγ responses to ARLELFVVL after background subtractions than the controls (P <0.0001, Figure 4D ). This was also the case for the longer peptide, Pep34 (P <0.0001, Figure 4D ). Notably, the strongest responding patients had comparable reactivity against Pep34. In contrast, none of the controls displayed IFNγ responses above the background (except for one control with just a single IFNγ SFC).

BLAST searches revealed high sequence homology between proteins from several human pathogens and both 21OH-derived T cell epitopes investigated in this study. Peptide hits were predominantly from bacteria (both facultative and obligate intracellular, as well as extracellular species) but also a few parasitic helminths. The microbial species and the corresponding proteins containing highest sequence homology are enlisted in Tables 1 and 2 .