OBJECTIVE—Weak major histocompatibility complex (MHC) binding of self-peptides has been proposed as a mechanism that may contribute to autoimmunity by allowing for escape of autoreactive T-cells from the thymus. We examined the relationship between the MHC-binding characteristics of a β-cell antigen epitope and T-cell autoreactivity in a model of autoimmune diabetes.
RESEARCH DESIGN AND METHODS—The binding of a proinsulin epitope, proinsulin-1(47–64) (PI-1[47–64]), to the MHC class II molecules I-Ag7 and I-Ak was measured using purified class II molecules. T-cell reactivity to the proinsulin epitope was examined in I-Ag7+ and I-Ak+ mice.
RESULTS—C-peptide epitopes bound very weakly to I-Ag7 molecules. However, C-peptide–reactive T-cells were induced after immunization in I-Ag7–bearing mice (NOD and B6.g7) but not in I-Ak–bearing mice (B10.BR and NOD.h4). T-cells reactive with the PI-1(47–64) peptide were found spontaneously in the peripancreatic lymph nodes of pre-diabetic NOD mice. These T-cells were activated by freshly isolated β-cells in the presence of antigen-presenting cells and caused diabetes when transferred into NOD.scid mice.
CONCLUSIONS—These data demonstrate an inverse relationship between self-peptide–MHC binding and T-cell autoreactivity for the PI-1(47–64) epitope in autoimmune diabetes.
Defective negative selection caused by weak interactions between self-peptides and major histocompatibility complex (MHC) molecules has been proposed as a mechanism that may contribute to autoimmunity. We have been interested in examining the relationship between peptide-MHC interactions and T-cell autoreactivity for disease-relevant antigens in autoimmune diabetes. In the case of the nonobese diabetic (NOD) mouse, the class II MHC molecule, I-Ag7, has been shown to be a weak peptide binder (1). While examining the CD4 T-cell response to peptides derived from the insulin β-chain in the NOD mouse, we were struck by its low binding affinity to the I-Ag7 class II MHC molecule (2). Many of the spontaneous T-cells that were identified reacted to a segment previously identified as a focus for spontaneous T-cell reactivity (3–6). Such a peptide apparently bound in two registers to I-Ag7, but binding was in the low micromolar range and showed a fast dissociation rate. Many of the amino acids in the peptide affected both binding and T-cell recognition, implying a very loose peptide-MHC complex (2).
The relationship of low affinity of a peptide to an MHC molecule and autoimmune reactivity has also been noted in the encephalitogenic peptide from myelin basic protein (7,8). This has led to the speculation that low-affinity peptides may not be conducive to the normal mechanisms in the thymus that control autoreactivity. A study by McNeil and Evavold noted that peptides with fast dissociation rates and poor binding to MHC molecules were ineffective at deleting thymocytes even though the same ligands were capable of eliciting proliferation of mature T-cells in the periphery (9). Similarly, in a tumor model system, low-affinity epitopes were shown to activate tumor-specific cytotoxic T-lymphocytes and could confer antitumor immunity on immunization (10). Lastly, a study by Roep and colleagues (11) that measured in vitro binding interactions of proinsulin epitopes to various HLA-DR molecules noted that DR alleles associated with protection from diabetes bound various proinsulin epitopes with a higher affinity when compared with DR molecules that predispose to type 1 diabetes.
To further characterize the relationship between peptide-MHC binding strength and immune reactivity, we have examined other regions of the proinsulin molecule targeted by CD4 T-cells. The insulin molecule is derived from the processing of the prohormone proinsulin and is secreted from the pancreatic β-cell with equimolar amounts of the connecting peptide C-peptide (Table 1) (12). We examined the proinsulin-1(47–64) (PI-1[47–64]) (SPGDLQTLALEVARQKRG) segment because reactivity to it had been found in NOD mice (5,13). These data extend our understanding of the role of proinsulin as an autoantigen in type 1 diabetes and demonstrate an inverse relationship between peptide-MHC affinity and T-cell autoreactivity against a defined β-cell antigen epitope.
RESEARCH DESIGN AND METHODS
NOD, NOD.scid, NOD.h4, B6.g7, and B10.BR mice were purchased from The Jackson Laboratories (Bar Harbor, ME). All mice were housed and cared for in accordance with the guidelines of the Washington University Committee for the Humane Care of Laboratory Animals and with National Institutes of Health guidelines on laboratory animal welfare.
Synthetic peptides were purchased from Biosynthesis (Lewisville, TX). All peptides were >95% pure, confirmed by high-performance liquid chromatography and mass spectrometry analysis.
Peripancreatic lymph node assays.
Spleens, axillary and inguinal lymph nodes, and peripancreatic lymph nodes (PPLNs) were isolated from 10- to 16-week-old male and female NOD mice and dispersed into single-cell suspensions by passage through cell strainers. The splenocytes or lymph node cells (3–5 × 104/well) were subsequently cultured with irradiated NOD splenocytes (5 × 105/well) and 1–5 μmol/l peptide for 7 days in round-bottom 96-well plates in a final volume of 200 μl. On day 7, the cultures were re-stimulated with fresh irradiated NOD splenocytes (5 × 105/well), 25 units/ml interleukin (IL)-2, and 1–5 μmol/l peptide. T-cell–positive wells were tested days 14–17 with antigen-presenting cells (APCs) and 10 μmol/l peptide. Only data from cells with >200 counts per minute were considered, the stimulation index was calculated by dividing values for APCs plus peptide by APCs alone. The data presented in Fig. 2A were pooled from seven independent experiments. All tissue culturing was done in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% FCS (DMEM 10% FCS).
Primary T-cell lines.
The anti–C-peptide (ACP) primary T-cell lines were generated by immunization of 8- to 10-week-old NOD mice with 10 nmol PI-1(47–64) in complete Freund's adjuvant (CFA). Immunized lymph nodes were harvested at day 7 and cultured at 5 × 105 cells/well in DMEM with 10% FCS and 1 μmol/l peptide. On day 7, and at 7- to 10-day intervals thereafter, the T-cells (1–2 × 103–4/w) were restimulated with fresh irradiated NOD splenocytes (5 × 105 cells/well), IL-2 (25 units/ml), and peptide (1 μmol/l). Hybridomas were generated from these primary lines by fusion to the BW5147 α-β-thymoma partner cell line following standard techniques (14).
Primary T-cell proliferation assays were performed by incubation of T-cells (2 × 104/well) with irradiated NOD splenocytes (5 × 105/well) in the presence of peptide or dispersed mouse β-cells. Primary T-cell proliferation assays were pulsed with thymidine (1 μCi/well) on day 2 and harvested day 3. T-cell hybridoma assays were performed with NOD splenocytes or the I-Ag7–expressing cell line, C3G7 (2.5 × 104/well), as APCs and supernatants were collected at 24 h for IL-2 quantitation by cytotoxic T-lymphocyte line assay. Fixation of the C3G7 cell line was performed by incubation in 1% paraformaldehyde for 15 min followed by incubation in 0.2 mol/l dl-lysine for 10 min and then washing three times in DMEM 10% FCS.
Islets of Langerhans were isolated from mice of various ages and dispersed using standard techniques (15). The digestion of the pancreatic tissue was performed with type XI collagenase (Sigma, St. Louis, MO) without DNase.
Enzyme-linked immunosorbent spot-forming cell analysis.
Mice were immunized with 10 nmol PI-1(47–64) peptide in CFA and lymph nodes were examined on day 7 by enzyme-linked immunosorbent spot-forming cell (ELISPOT) analysis for IL-2+ cells. Lymph node cells (1 × 106/well) were plated in triplicate on anti–IL-2 capture antibody–coated filter plates in the presence of 10 μmol/l antigen and developed 24 h later with anti–IL-2 detection antibodies (BD Biosciences, San Jose, CA). Plates were read by C.T.L. Cellular Technology and analyzed with ImmunoSpot 3.2 software.
Soluble I-Ag7 or I-Ak was produced using the recombinant baculovirus system as previously described (16). Peptide binding assays were done under acidic (pH 5.5) conditions. Briefly, 0.5–1 μg I-Ag7 or I-Ak/class II–associated invariant chain peptide was treated with 0.1 unit thrombin to cleave both the zipper tails and peptide linker (Novagen, Madison, WI) and simultaneously incubated with 0.125 pmol I125-radiolabeled mimotope reference peptide (GKKVATTVHAGYG) and increasing doses of unlabeled peptides in 200 mmol/l Tris(2-carboxyethyl) phosphine hydrochloride, 20 mmol/l MES, and 150 mmol/l sodium chloride. Binding reactions were incubated overnight at 25°C in 30-μl volumes. Complexes were purified from free peptide by gel filtration Bio-spin columns (Bio-Rad). The percentage of bound peptide was evaluated by gamma counting. Usually, ∼25–35% of input peptide was bound, whereas <0.5% of peptides nonspecifically passed through the Bio-spin columns. The half-maximal inhibitory concentration (IC50) value is very close to the binding equilibrium constant. The IC50 values shown in Table 2 are averages of three to five independent binding experiments.
Biological dissociation rates.
The C3G7 APC line was incubated with peptide (10 μmol/l) for 2 h and washed. ACP T-cells (2 × 104/well) were added to the APC-containing wells (2.5 × 104/well) cultures at time zero and at 4 and 8 h after washing. The supernatants were collected at 24 h, and IL-2 was measured by cytotoxic T-lymphocyte line assay.
Adoptive T-cell transfer.
Two of the three primary ACP T-cell lines were expanded in vitro and transferred into 8- to 12-week-old NOD.scid recipients by tail vein injection. Diabetes was defined by the measurement of blood sugars >250 mg/dl on two separate occasions. All recipients were followed for 30 weeks or up to disease onset.
The PI-1(47–64) peptide bound weakly to I-Ag7.
Binding assays were done with soluble I-Ag7 and I-Ak and synthetic peptides. The IC50 value indicates the concentration of unlabeled peptide required to affect binding of a reference peptide to the class II molecules: the higher the value, the weaker the binding of the peptide to the MHC. The PI-1(47–64) and PI-2(49–66) peptides bound poorly to I-Ag7, with IC50 values of 98.9 and 135.9 μmol/l, respectively (Table 2). In contrast, these peptides bound more strongly to I-Ak, with IC50 values of 4.8 and 2.1 μmol/l, respectively. The PI-1(49–61) peptide bound slightly better to I-Ag7, with an IC50 of 20 μmol/l, perhaps due to the loss of the positively charged P62Lys, which is likely disfavored at the carboxy-end of the peptide-binding groove of I-Ag7. As a point of reference, the mouse insulin-2 β-chain (9–23) peptide bound to I-Ag7 with an IC50 of 3 μmol/l.
Reactivity to PI-1(47–64) is I-Ag7 dependent.
The role played by the MHC class II molecule, I-Ag7, and background genes in autoreactivity to C-peptide was examined by characterizing the T-cell responses in NOD, NOD.h4, B6.g7, and B10.BR mice. Mice were immunized with the PI-1(47–64) peptide, and lymph nodes were examined on day 7 by ELISPOT analysis for IL-2+ cells. Both I-Ag7–containing strains, NOD and B6.g7, had peptide-reactive T-cells (Fig. 1A and C), compared with the I-Ak+ strains that did not (Fig. 1B and D). The frequency of PI-1(47–64)–reactive cells detected under these conditions was low: ∼1/30,000 in the NOD mice and 1/50,000 in the B6.g7 mice. However, PI-1(47–64)–specific T-cells were not detected in the NOD.h4 and B10.BR mice. The T-cell reactivity against PI-1(47–64) present in the NOD and B6.g7 mice was clearly I-Ag7 restricted as evidenced by inhibition with MHC class II–blocking antibody, AG.2.42.7 (17). These data demonstrate that I-Ag7 alone is necessary and sufficient for the generation of PI-1(47–64) T-cells independent of other background genes.
PPLNs harbor C-peptide–reactive T-cells.
Given the central importance of the PPLNs in the priming of diabetogenic T-cells, we began by examining the PPLNs of pre-diabetic mice for the presence of T-cells specific for the PI-1(47–64) peptide. Like others (5,18), we did not detect reactivity to proinsulin peptides in primary proliferation assays with PPLNs or spleens from 10- to 16-week-old NOD mice (data not shown). Given this difficulty in detecting primary T-cell responses, we expanded PPLN T-cells, in addition to T-cells from the spleen and peripheral lymph nodes, for 2 weeks in vitro with the PI-1(47–64) peptide, and then tested all T-cells in the expanded cultures for peptide reactivity. T-cells specific for the PI-1(47–64) peptide were identified in the PPLNs from 10- to 16-week-old mice but were not found in the spleen or axillary and inguinal lymph nodes (Fig. 2A and B). Five of 246 expanded T-cell lines from the PPLNs reacted with the PI-1(47–64) peptide (stimulation index >5). In contrast, no PI-1(47–64)–reactive T-cells were detected from the 454 T-cell lines expanded from the spleen or axillary and inguinal lymph nodes. The presence of C-peptide–reactive T-cells in the PPLNs was in agreement with previous data that indicated these nodes as the site for priming of diabetogenic T-cells (19–21). Moreover, this finding also suggested a possible role for these T-cells in the early pathogenesis of the disease. To further characterize the biology of C-peptide–reactive T-cells, we first generated T-cell lines by immunization of young NOD mice.
NOD mice lack tolerance to the PI-1(47–64) self-peptide.
Mice were immunized with PI-1(47–64) peptide in CFA, and several primary CD4+ T-cell lines were generated, including the ACP 2 line (Fig. 3A). All lines demonstrated similar reactivity in vitro; they were maximally stimulated by the PI(47–64) peptide from proinsulin-1 and also reactive with the PI(49–66) peptide from proinsulin-2. The T-cell lines also recognized the PI-1(49–61) and the PI-2(51–63) segments, which differ only by one residue (Arg for PI-1 at position 60 and Gln for PI-2 at position 62). The T-cell lines were stimulated by the full-length C-peptides from proinsulin 1 and 2, although less than they were by the PI-1(47–64) peptide.
To determine whether the ACP T-cell lines were reactive with the naturally processed and presented forms of the proinsulin epitope, T-cells were tested for reactivity with freshly isolated mouse β-cells. All three primary ACP lines, including ACP 2, reacted to dispersed mouse β-cells from young NOD mice (4–6 weeks of age) and β-cells from other strains such as 129/Svj (Fig. 3B). Furthermore, the ACP T-cells reacted to dispersed islets isolated from 5-month-old NOD mice, demonstrating that APCs from the islet infiltrate present the C-peptide epitope (Fig. 3C).
Minimal epitope mapping and antigen processing requirements for C-peptide–reactive T-cells.
Analysis of ACP T-cell hybridomas showed heterogeneity in their reactivity to truncated peptides (Fig. 4 A; Table 3). The likely minimal T-cell epitope spanned the PI-1(51–61) sequence (LQTLALEVARQ) because loss of the P51Leu resulted in a noticeable decrease in T-cell activation for all hybridomas examined (Fig. 4A; Table 3). Some T-cells, such as ACP2.1, responded well to the PI-1(49–60) peptide lacking the P61Gln, whereas others did not (see ACP1.28), indicating variable requirements for carboxy-terminal residues. Identification of the nonamer canonical core within the PI-1(51–61) segment could not be determined by T-cell reactivity, and all T-cells tested (0 of 8) with the three possible nonamers spanning the PI-1(51–61) peptide were negative (51–59, LQTLALEVA; 52–60, QTLALEVAR; and 53–61, TLALEVARQ). These results may indicate a need for amino- or carboxy-flanking residues that serve as T-cell receptor contacts, as was seen with the β-chain (9–23) epitope (2), loss of MHC binding affinity with the truncated peptides, or both.
To assess the requirements for antigen processing of the C-peptide T-cell epitopes, T-cells were tested for reactivity with synthetic peptides presented by fixed APCs. We found that ACP T-cell hybridomas were stimulated by fixed APCs presenting the PI-1(47–64), PI-1(49–61), and full-length C-peptide (Fig. 4B), indicating that all three peptides, even the full-length C-peptide-1 29-mer, could form recognizable complexes without the need for further processing.
We next examined the stability of the peptide-MHC complexes formed by PI-1(47–64) and I-Ag7 by functional experiments in which T-cells were applied to antigen-pulsed and washed APCs over an 8-h time course (Fig. 4C). The antigen-pulsed APCs quickly lost their ability to stimulate C-peptide–reactive T-cells. The T-cell response, for both PI-1(47–64)–and C-peptide-1–pulsed APCs, was reduced to 50% of its maximal value in 3 h, indicating that the C-peptide–MHC complexes disassociated rapidly.
C-peptide–reactive T-cells induce diabetes.
The pathogenicity of the ACP primary T-cell lines was assessed by adoptive transfer into NOD.scid recipient mice. Three separate transfer experiments were performed: group 1 (n = 8) received 9.4 × 106 ACP no. 1 T-cells; group 2 (n = 4) received 17 × 106 ACP no. 2 T-cells; and group 3 (n = 4) received three infusions of ACP nos. 1 and 2 T-cells. The incidence of diabetes was 50% for all groups; group 1 (4 of 8) by 20 weeks after transfer, group 2 (2 of 4) at 10 and 20 weeks after transfer, and group 3 (2 of 4) at 20 and 22 weeks after transfer (Fig. 5A). The induction of diabetes in 8/16 NOD.scid mice, which never naturally develop diabetes, indicated that the ACP T-cells were pathogenic in the adoptive transfer model. Although the kinetics of disease induction in these experiments is less than some observed with activated T-cells from T-cell receptor transgenic mice, e.g., BDC2.5 the incidence is similar to the spontaneous incidence seen in wild-type NOD mice. Furthermore, all recipient mice, including those that were not overtly hyperglycemic at 30 weeks, had severe destructive insulitis and loss of β-cell mass (Fig. 5B).
The main findings from these studies are 1) binding of the PI-1(47–64) and PI-1(49–61) peptides to I-Ag7 is incredibly poor, and the peptide-MHC complexes are short-lived; 2) I-Ag7 plays a central role in the generation of the ACP T-cell repertoire, in that autoreactivity was found to strains bearing I-Ag7 but not I-Ak molecules; 3) C-peptide–reactive T-cells are found in the PPLNs of pre-diabetic NOD mice; 4) ACP T-cells are reactive with mouse β-cells and respond to full-length C-peptide presented by live and fixed APCs; and 5) despite this weak peptide-MHC interaction, C-peptide–reactive T-cells are capable of causing diabetes in an adoptive transfer model.
Our data confirm and extend the observations of other groups that have identified the C-peptide region of proinsulin as a CD4+ T-cell target in the NOD mouse (5,13,22,23). The recovery of C-peptide–reactive T-cells from the PPLNs and not from other locations suggests that these T-cells may encounter their antigen in this critically important site of diabetogenic T-cell priming (19–21,24). Intra-islet APCs, most likely dendritic cells, must encounter large amounts of C-peptide in the islets and may traffic to the PPLNs, where they could prime and activate naïve C-peptide–reactive T-cells. Alternatively, lymphatic drainage of the pancreas might deliver high concentrations of C-peptide to the PPLNs, allowing uptake and presentation by nodal dwelling APCs. The islets readily provided antigen to APCs (Fig. 3B), and APCs recovered from insulitic islets stimulated the T-cells (Fig. 3C). In either case, it is not difficult to imagine a scenario in which PPLNs are exposed to large amounts of C-peptide, which would be required for activation of the T-cells by these very weak MHC-binding peptides. Although C-peptide circulates, its concentration in blood (2–5 nmol/l) would not be sufficient for either activation or regulation in APCs at peripheral sites.
Why are C-peptide–reactive T-cells not deleted in the thymus? Weak MHC-binding peptides may well be the best at fostering autoimmunity, in great part by avoiding negative selection in the thymus or by failure of regulatory mechanisms, although other factors are likely to play a role. The PI-1(47–64) epitope represents a striking example of a very poor binding MHC ligand, IC50 values ∼100 μmol/l, that gives rise to T-cells capable of causing diabetes in adoptive transfer experiments. The relationship between poor peptide-MHC binding and the fostering of an autoimmune, pathogenic T-cell repertoire has been suggested in studies on experimental allergic encephalomyelitis (7,8) and in our recent work on the insulin β-chain (9–23) epitope (2) and in other model systems (9–11).
The obligatory role for the diabetogenic I-Ag7 molecule in eliciting C-peptide–reactive CD4 T-cells remains a key issue. Our results with the four mouse strains NOD (I-Ag7+), NOD.h4 (I-Ak+), B6.g7 (I-Ag7+), and B10.BR (I-Ak+) demonstrated that expression of I-Ag7, independent of the genetic background, was sufficient to generate C-peptide–reactive CD4 T-cells (Fig. 1). This is in agreement with previous data from Kanagawa et al. (25) and Ridgway et al. (26), indicating that the expression of I-Ag7 in different strains of mice allowed escape of autoreactive T-cells. A more recent study from Stratmann et al. (27) used MHC tetramers to trace islet-reactive CD4+ T-cells in I-Ag7–expressing mice of different genetic backgrounds. Here again, the expression of I-Ag7 in different strains selected for tetramer-reactive CD4+ T-cells. The NOD strain has also been shown to be susceptible to other autoimmune diseases (28–31). Taken together, these data point to a central role for I-Ag7 in selection (or escape) of autoreactive T-cells. We believe that the explanation lies in the structural features of I-Ag7, characterized by an overall low peptide-binding affinity and instability (32), in addition to other susceptibility genes. These findings do not lessen the importance of other factors in NOD mice that play a role in the loss of self tolerance (33–35). The fact that B6.g7 mice do not develop diabetes highlights the need of both MHC- and non-MHC–encoded susceptibility genes.
A second important factor in the lack of T-cell tolerance may be specific to the insulin molecule. It may be that the peptide-MHC complexes formed in the thymus do not recapitulate the complexes that appear in the periphery. The deleting complexes are presumably formed by the traditional processing and presentation of proinsulin peptides in thymic medullary epithelial cells. Here, peptides are selected by the biological and physical forces that govern peptide presentation by I-Ag7 molecules in thymic cells and are not the same as events that might occur in the periphery. Although proinsulin is clearly expressed in the thymus (36,37), perhaps largely under the control of the autoimmune regulator transcription factor (38), there is no evidence of the expression of the prohormone processing machinery, such as the prohormone convertases 1 and 2, which would be required to generate the actual C-peptide fragment that is secreted by the β-cell. The absence of tolerance to C-peptide, as it is presented in the pancreas, may be explained, in a sense, by the lack of a secretory apparatus in the thymus.
The data presented here extend our understanding of the role of proinsulin as an autoantigen and again seem to illustrate an inverse relationship between weak peptide-MHC binding and T-cell autoimmune pathogenicity. The degree to which C-peptide–reactive T-cells contribute to disease pathogenesis in the NOD mouse cannot be determined with certainty. The T-cells were pathogenic in circumstances of adoptive transfer, and they represent one set of the many proinsulin-reactive T-cells that seem to play a key role in the development of autoimmune diabetes. The identification of T-cell reactivity against C-peptide epitopes in type 1 diabetic patients suggests that C-peptide might also be an important autoantigen in the human disease (39–42).
|Mouse I||RREVEDPQVEQLELGGSP GDLQTLALEVARQKR|
|Mouse I||RREVEDPQVEQLELGGSP GDLQTLALEVARQKR|
The sequences of the human, mouse-1, and mouse-2 proinsulin molecules are shown. The amino acid residues in bold are cleaved off during the physiological processing of the prohormone, and C-peptide is secreted in equimolar amount with the mature insulin heterodimer. The amino acid residue numbering refers to the mouse proinsulin-1 sequence, and the PI-1(47–64) segment is underlined.
|Peptide .||Sequence .||Binding|
|.||.||I-Ag7 .||I-Ak .|
|Peptide .||Sequence .||Binding|
|.||.||I-Ag7 .||I-Ak .|
This panel shows the binding analysis of synthetic peptides to soluble I-Ag7 and I-Ak. Binding is expressed by IC50 values (μmol/l). The mime peptide is a synthetic reference peptide that has been used in I-Ag7 binding assays (16). The binding of mouse insulin-2 B:(9–23) (3 μmol/l) peptide is presented as a point of reference. Binding studies were performed at pH 5.5 as described in research design and methods.
|.||1.23 .||1.24 .||1.28 .||1.56 .||2.1 .||2.55 .||2.78 .||2.87 .|
|.||1.23 .||1.24 .||1.28 .||1.56 .||2.1 .||2.55 .||2.78 .||2.87 .|
Values shown are concentrations of peptide required to half-maximally stimulate a given T-cell (μmol/l; —, none detected). Values are averages of two to five independent experiments.
Published ahead of print at http://diabetes.diabetesjournals.org on 8 April 2008.
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/licenses/by-nc-nd/3.0/ for details.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
See accompanying commentary, p. 1788.
This work was supported by grants from the Barnes-Jewish Hospital Foundation, the National Institutes of Health (DK-067199 and DK-020279), and the Kilo Diabetes and Vascular Research Foundation.
We thank Shirley Petzold for technical assistance.