Proinsulin Expression by Hassall’s Corpuscles in the Mouse Thymus

Ins1-KO and Ins2-KO mice were a gift of Dr. J. Jami (Institut Cochin, Paris, France) and were generated as described by Duvillié et al. (27). Ins2-KO mice have their Ins2 gene replaced by a functional LacZ gene that is under the regulation of the endogenous Ins2 gene promoter, whereas Ins1 was disrupted by introducing a neomycin cassette against the same genetic background. All mice were bred at our animal facility under conditions specified by the Canadian Council of Animal Care.

Pancreatic and thymic RNA from 2- to 4-week-old mice was isolated by Trizol (Gibco, Rockville, MD) and treated with DNAse (Gibco). RNA was isolated from enriched thymic cell populations with RNeasy Mini Kit (Qiagen, Mississauga, Ontario, Canada) with typical yields of RNA between 50 and 400 ng. An equal amount of each RNA (2.5 μg of tissue-extracted and 50–100 ng of single cells or more as with the case of DCs) was reverse-transcribed using random primers and Superscript II Reverse-Transcriptase (Gibco). Parallel samples in which reverse transcriptase was omitted were always included to confirm the absence of DNA contamination.

Insulin expression in the thymus- and pancreas-extracted tissues was analyzed by quantitative RT-PCR. An equal amount of each cDNA sample (2 μl from thymic and 2 μl from 1:300 diluted pancreatic cDNA) was then added to serial dilutions of the insulin competitor (56, 28, 14, 7, 3.5, 1.7, or 0 amol/tube) or with the cyclophilin competitor (2.5, 1, 0.5, 0.25, 0.1, or 0 ×10⁹ copies/tube) for loading control. The insulin competitor is an internally deleted cloned insulin sequence amplified with the same primers but 45 bp shorter, constructed as described by Forster (28). For cyclophilin analysis, we used the PCR Kit for cyclophilin (Quantum mRNA; Ambion, Austin, TX) as described by the manufacturer. The PCR products were resolved using PAGE, and bands were quantified using Geldoc Software (Bio-Rad, Hercules, CA). The insulin concentrations were normalized for the amount of starting RNA using cyclophilin measurement as described (8). β-Gal expression in pancreas and thymus was analyzed by semiquantitative RT-PCR. An equal amount of cDNA (2 μl of diluted 1:1, 1:10, and 1:100 cDNA samples from tissues and undiluted amount for cell preparations) was amplified by PCR. The thermal cycler conditions for β-Gal and insulin amplification were 150 and 180 s at 92–95°C followed by 40 cycles of 92–94°C for 30–60 s, 53–55°C for 30–60 s, and 74–72°C for 60–90 s, respectively. The following primer pairs were used (forward and reverse primers respectively): insulin-specific primers: 5′-GGCTTCTTCTACACACCCCA-3′ and 5′-TACCAGCTGGAGAACTACTG-3′; and bacterial-specific β-Gal primers: 5′-TGGCGGACCGCTAT-CAGGAC-3′and 5′-TGTATCATCTGGTCGCTGG-3′.

Organs from mice were harvested, embedded in Tissue-Tek (Sakura, Finetech, Torrance, CA; OCT compound), snap-frozen in isopentane precooled in dry ice, and sectioned in 10-μm-thick sections.

For β-Gal activity, the sections were rinsed in 0.1 mol/l PBS, fixed with 0.2% glutaraldehyde, 5 mmol/l EGTA, and 2 mmol/l MgCl₂ in PBS for 10 min at room temperature, washed three times (5, 10, and 15 min) in PBS, then stained with color solution (1 mg/ml X-gal, 5 mmol/l K₄Fe [CN]₆, 5 mmol/l K₃Fe [CN]₆, and 2 mmol/l MgCl₂ in PBS) for 24–48 h at 37°C. The sections were rinsed in PBS then counterstained with Nuclear Fast Red (DAKO, Mississauga, Ontario, Canada) as described by the manufacturer.

For Immunohistochemistry, the sections stained with X-gal were rinsed and endogenous peroxidase activity was quenched for 5 min in 0.3% H₂O₂, saturated with normal goat or rabbit sera, and then incubated with anti-CD11c (N418; Serotec, Raleigh, NC) and the anti-gp40 antibody (murine homologue of human EpCAM) G8.8 (DSHB, University of Iowa, Iowa City, IA) antibodies for 30 min. After three successive washings in PBS (3 × 5 min), the sections were incubated with affinity-purified biotin-conjugated secondary antibodies goat anti-hamster IgG antibody (Vector Laboratories, Burlingame, CA) and rabbit anti-rat IgG (Vector Laboratories), respectively. Negative control sections were incubated only with secondary antibodies. Slides were then incubated with Vectastatin Elite ABC reagent for 30 min (Vector Laboratories) using the horseradish peroxidase substrate diaminobenzidine tetrahydrochloride (Vector).

Thymic DCs were isolated as described by Crowley et al. (29). In the first digestion, individual thymic lobes were perfused with 100 units/ml collagenase D (Sigma, Oakville, Ontario, Canada) and then incubated for 15 min at 37°C with 400 U/ml collagenase D. The glands were then disrupted, and all released cells were centrifuged at 1,400 rpm for 10 min. The cell pellet was resuspended in BSA (35%, ρ = 1.080; Sigma), overlaid with 2 ml of RPMI medium and centrifuged for 20 min at 7,500g. In an alternative protocol, thymi were minced and digested with two 15-min incubations at 37°C with collagenase IV (Sigma), pelleted, and resuspended in BSA. The low-density floating fraction was collected, washed, and incubated for 90 min in a Petri dish. The nonadherent cells were removed by pipetting. The monolayer was incubated for an additional hour to remove additional small round lymphocytes. Then the monolayer was cultured overnight, at which time most DCs became nonadherent and were harvested. For further purification and enrichment for DCs, the cells were sorted in a Vantage FACS (fluorescent-activated cell sorter) (Becton Dickinson, Franklin Lakes, NJ) using anti-CD11c antibodies. For examining freshly isolated DCs without the phenotypic changes that might occur during overnight culture, the low-density floating fraction was incubated in 25 mmol/l EDTA/PBS at 37°C for 5 min and DCs were then directly FACS sorted with CD11c alone, DEC205 (Serotech, NLDC-145) alone, or CD11c and I-A^b (MHC II; Pharmingen, Mississauga, Ontario, Canada) with essentially identical results.

Thymic epithelial cells (TECs) were isolated according to Klein et al. (30). Briefly, 4–5 thymi of 2- to 4-week-old mice were digested for 30–40 min with Collagenase/Dispase (Sigma; 0.2 mg/ml each) and DNase (Roche, Mannheim, Germany; 25 μg/ml) in PBS-1% FCS. EDTA was added to a final concentration of 10 mmol/l for the last 5 min. Cells were then separated on a discontinuous Percoll (Amersham Biosciences, Uppsala, Sweden) density gradient. The TEC-enriched fraction (the two upper phases) were harvested, washed, and stained with G8.8 followed by phycoerythrin-conjugated rabbit anti-rat IgG antibody (Serotech). After 30 min at 4°C, cells were stained with anti-CD45-FITC (Pharmingen) to distinguish epithelial cells from G8.8-positive marrow-derived cells. TECs were sorted according to the phenotype of CD45^−/lo and G8.8⁺. Blocking with the anti-FcR mAb 2.4G2 (Pharmingen) preceded the stainings for DCs, whereas blocking with 1% horse serum was done for the TEC isolations. The average cell yield per thymus was 30,000 TECs.

Approximately 10⁶ cells from the Percoll gradient isolation (epithelial cells) or from the DC isolation (low-density fraction) were labeled with G8.8 or CD11c antibodies and resuspended in 1 ml of RPMI (10% FCS) with 10 μl of endogenous β-Gal–like activity inhibitor from the C₁₂FDG kit (ImaGene Green Kit; Molecular Probes, Eugene, OR) for 30 min at 37°C. After this, 0.5 ml of prewarmed C₁₂FDG (15 μmol/l) substrate was added to the cells and incubated for 20 min at 4°C, room temperature, or 37°C. Results from the incubations at all three temperatures were very similar, but the background was significantly lower at 4°C and the results shown were obtained at this temperature. This method was used to compare the percentage of cells positive for both G8.8 (or CD11c) and C₁₂FDG between Ins2-KO mice (β-Gal expressed under the Ins2 promoter) and Ins1-KO mice (no β-Gal transgene). The difference in double-positive cells between the two genotypes was taken to represent cells that truly express transgenic β-Gal. The number, as a percentage of all G8.8⁺ cells, was much higher (10- to 30-fold higher) in Ins2-KO mice than in Ins1-KO in all four experiments regardless of whether the background, in the absence of C₁₂FDG, was subtracted from each. In CD11c⁺ cells, no significant difference from background was seen.

In mice, two unlinked, nonallelic, Ins1 and Ins2 genes encode insulin. Ins1 is expressed at two to three times lower than Ins2 in the pancreas and thymus (8,31,32). Figure 1A shows proinsulin expression in the thymus of 2- to 4-week-old mice. Proinsulin is expressed two to three times less in the Ins2-KO mice when compared with the Ins1-KO mice, confirming previous results (8). For both mice, the concentration of proinsulin was normalized with cyclophilin (Fig. 1B). As expected, β-Gal expression was detected in thymus and pancreas of Ins2-KO but not Ins1-KO mice (Fig. 1C). These results show that proinsulin is abundantly expressed in the thymus of Ins1-KO mice and bioactive β-Gal in the thymus of Ins2-KO mice. These mice were used for the analyses of proinsulin/β-Gal expression in fractionated thymic stromal cells.

To examine proinsulin/β-Gal gene expression in thymic stromal cells, we enriched distinct cell populations to high purity by a combination of density fractionation and cell sorting. DCs were isolated from thymi of 2- to 4-week-old mice using the standard protocol (29) that includes enrichment by adherence to culture dishes after overnight incubation before cell sorting with anti-CD11c antibody. Alternatively, for avoiding possible phenotypic changes during the overnight incubation, possibly comparable to DC maturation in vivo (33), the overnight incubation step was omitted, and the carefully removed low-density cell fraction from the BSA gradient was sorted immediately using an anti-CD11c or anti-DEC205 antibody (Fig. 2A). Both techniques yielded CD11c-positive cells within the expected range of 30,000–70,000 DCs per mouse thymus (29). Without overnight incubation, sorting by CD11c and I-A^b (MHC II) antibodies yielded similar results (Fig. 2B). Virtually the entire population of CD11c⁺ cells were also positive for I-A^b.

Epithelial cells were isolated as described by Klein et al. (30) and enriched by cell sorting using G8.8 and CD45 antibodies. The G8.8⁺ population consists of 40–60% TECs, with the remainder of cells being mostly thymocytes (30,34). Marrow-derived cells were eliminated by selecting for G8.8⁺/CD45^−/lo providing exclusively medullary and cortical epithelial cells (19,30) (Fig. 2C and D). Such isolations yielded very similar results to Derbinski et al. (19), as we consistently isolated ∼30,000 epithelial cells/thymus. G8.8⁺/CD45⁻ and G8.8⁺/CD45^hi cell fractions were also isolated (Fig. 2D).

By RT-PCR, strong proinsulin and β-Gal bands were detected in G8.8⁺/CD45^−/lo TECs from Ins1-KO and Ins2-KO mice, respectively, whereas only a very faint band was detected in the CD11c⁺, DEC205⁺, or CD11c⁺/I-A^b+ DCs from the same mice, isolated without overnight incubation (Fig. 3B, lanes 1–5, and C, lanes 1–4). Even with a two- to threefold greater number of DCs, as compared with TECs, and with a substantially larger amount of RNA used for RT-PCR, controlled for by cyclophilin, only very weak proinsulin and β-Gal bands could be detected within the DC extracts (Fig. 3B, lanes 1, 2, and 4, and C, lanes 1 and 2). Reducing the cDNA quantity in the DCs only reduced the intensity of the proinsulin and β-Gal bands, ruling out saturation of cDNA as a possible inhibitor of the PCR (Fig. 3B, lanes 1 and 2). Similarly and as expected, the intensity of the proinsulin band increased with epithelial enrichment of the G8.8⁺ fraction by double sorting (G8.8⁺/CD45^−/lo; Fig. 3A, lanes 2 and 3). β-Gal was also detectable in G8.8⁺ and G8.8⁺/CD45^−/lo cells from Ins2-KO mice (Fig. 3, lanes 6 and 7; note that loading was not normalized for this result). No proinsulin/β-Gal expression was detected in G8.8⁺/CD45^hi cells (Fig. 3D) and from DCs with overnight incubation (Fig. 3A, lanes 1 and 5). Proinsulin was also strongly detected in G8.8⁺CD45⁻ epithelial cells even when poor RNA yields did not allow for the detection of cyclophilin (Fig. 3C, lane 4). All of these data demonstrate that in the thymus, proinsulin mRNA transcripts are predominantly detected in TEC.

To be more specific about which population of cells is responsible for insulin gene transcription in the thymus, we localized β-Gal activity in histologic sections. Figure 4 shows histochemistry results of the β-Gal activity on pancreatic and thymic frozen sections of Ins1-KO and Ins2-KO mice. These results show that β-Gal activity is specifically detected in pancreatic islets of Ins2-KO mice but not in Ins1-KO mice (Fig. 4A and B). In the thymus, β-Gal activity is detected in Ins2-KO mice but not in Ins1-KO and was localized in the mTECs forming the concentric structure that resembles HC. Approximately 1 in every 10 HC were positive for β-Gal activity, some of which contained as many as four X-gal-stained cells. (Fig. 4C–F). Because our sections represent only a slice of a three-dimensional HC, the true proportion must be higher, and it is indeed possible that every HC contains these cells. Immunohistochemistry co-localized all of the β-Gal-positive cells with G8.8 (Fig. 5B–E) but never with CD11c (DC; Fig. 5A). These data demonstrate that proinsulin-producing cells in the mouse thymus are mTECs of the HC.

We further localized β-Gal activity in purified populations of thymic stromal cells by flow cytometry using fluorescent β-Gal substrate (C₁₂FDG) and cell type-specific antibodies. The results consistently revealed a population of between 1 and 3% G8.8⁺/β-Gal-positive cells within the purified G8.8⁺ thymic epithelial cell extract (TEC) from Ins2-KO mice (Fig. 6B). We could not detect a similar β-Gal-positive population in G8.8⁺ cells of control Ins1-KO mice (Fig. 6A). In a separate experiment, we incubated sorted G8.8⁺ cells with C₁₂FDG, under similar conditions as above and observed them by fluorescence microscopy. The results showed a similar proportion of β-Gal-positive cells in the Ins2-KO preparation, consistent with the FACS data (Fig. 6C). The same experiments were done for sorted CD11c⁺ DCs, and no population of β-gal-positive DCs was observed in Ins2-KO mice that was above the background levels obtained with DCs from Ins1-KO mice (data not shown).

Our results show that in mouse thymus, proinsulin expression is limited to a very rare fraction of mTECs forming the HC. Previous studies that attempted to identify which cells are responsible for this proinsulin expression, by RT-PCR, gave contradictory results with expression being detected in DCs (26) or TECs (19) but not both. To resolve this discrepancy and to best characterize the identity of these cells, we used β-Gal as a marker of transcriptional activity of the endogenous Ins2 promoter. Semiquantitative RT-PCR shows that proinsulin/β-Gal is very strongly and easily detected in TECs but only very faintly detected in DCs that are sorted without overnight incubation. Throsby et al. (26) also detected a very faint band in DC extracts, and they required using a radiolabeled probe to see it adequately. This band may be the result of some epithelial cell contamination in the DC extracts, or it might represent expression at much lower levels or much fewer cells among the DCs.

Immunohistolocalization experiments revealed that these thymic cells were confined to the thymic medulla and were part of the HC. The expression of insulin and other ectopic or promiscuous tissue-specific self-antigens by mTECs has been documented extensively by Derbinski et al. (19). The authors, however, did not identify the exact location of the mTECs responsible for the production of insulin or other antigens and their spatial relationship to the rest of the stroma. Our immunohistochemistry of β-Gal activity showed that proinsulin production is a unique property of rare mTECs of the HC.

It has been reported that self-antigen-expressing cells in the thymus, including proinsulin-producing cells, were frequently found in a cluster of two to four positive cells when analyzed ex vivo (35). This may reflect in vivo clustering, as suggested by our observation of clusters of more than one β-gal-positive cell. Rodewald et al. (36) demonstrated that the thymus medulla consists of epithelial cell islets, each derived from a single progenitor. It is not known how this clonality relates to the epigenetic differences that distinguish the rare cells involved in the thymic expression of tissue-restricted antigens from the rest of the TECs. It also is not known whether such expression is the property of a large number of cells, each specializing in a particular antigen, or of a small number of cells that have developed the promiscuous transcriptional machinery to express many diversely regulated antigens. Our ability to isolate the insulin-producing cells by scaling up the flow cytometry depicted in Fig. 6B will enable us, in studies currently under way, to provide an answer to these questions.

Thymocyte-negative selection occurs mainly in the medulla (37,38), and the HC cells have been associated with apoptotic thymocytes (22). DCs and macrophages have been frequently observed around HC (23,38), and the DCs were shown to be responsible for phagocytosis of apoptotic thymocytes (23). In human thymus, Pugliese et al. (39) found that insulin can be detected in a number of thymic and secondary lymphoid DCs by double immunostaining. Our findings are compatible with the presence of a small number of proinsulin-expressing DCs, but it also is conceivable that these DCs do not express insulin at the transcriptional level but merely take it up for presentation. For tolerizing both CD4⁺ and CD8⁺ T-cells, presentation of the self-antigen by major histocompatibility complex class I and II molecules is necessary. It is known that mTECs can present endogenously produced self-antigens by both classes, whereas DCs do not seem to present endogenously produced antigen on class II (40).

If proinsulin expression in thymus medulla is important in negative selection, then how can contact of all or most thymocytes with the few, scattered insulin-producing cells be ensured? This may be the result of chemoattractant properties (23) that drive each thymocyte to scan many HCs. Alternatively or in parallel, other antigen-presenting cells, particularly DCs, may also contribute by taking up secreted antigen for presentation.

We have previously shown that thymic insulin expression levels modulate insulin-specific autoreactive T-cell tolerance even in non-diabetes-prone mice (8). Against the NOD background, the Ins2 knockout mice (with extremely low insulin expression in the thymus but normal levels in the pancreas) develop diabetes with higher incidence and earlier than the wild type (20, confirmed by our own preliminary observations and those of the group of Dr. G. Eisenbarth; Moriyama et al., personal communication).

We believe that the isolation of proinsulin-producing cells, using the techniques that we have developed here, will allow detailed study of the cellular and molecular biology of proinsulin production, central tolerance mechanisms, and the possibility of designing novel forms of gene or cellular therapy.

This work was supported by a grant from the Canadian Diabetes Association in honor of the late Emily C. Haynes.

We thank Dr. Jacques Jami for the generous gift of the knockout mice and Rosemarie Grabs for technical assistance.