Thymically-derived Foxp3+ regulatory T cells are the primary regulators of type 1 diabetes in the non-obese diabetic mouse model

Regulatory T cells (Tregs) are an immunosuppressive population that are identified based on the stable expression of the fate-determining transcription factor forkhead box P3 (Foxp3). Tregs can be divided into distinct subsets based on whether they developed in the thymus (tTregs) or in the periphery (pTregs). Whether there are unique functional roles that distinguish pTregs and tTregs remains largely unclear. To elucidate these functions, efforts have been made to specifically identify and modify individual Treg subsets. Deletion of the conserved non-coding sequence (CNS)1 in the Foxp3 locus leads to selective impairment of pTreg generation without disrupting tTreg generation in the C57BL/6J background. Using CRISPR-Cas9 genome editing technology, we removed the Foxp3 CNS1 region in the non-obese diabetic (NOD) mouse model of spontaneous type 1 diabetes mellitus (T1D) to determine if pTregs contribute to autoimmune regulation. Deletion of CNS1 impaired in vitro induction of Foxp3 in naïve NOD CD4+ T cells, but it did not alter Tregs in most lymphoid and non-lymphoid tissues analyzed except for the large intestine lamina propria, where a small but significant decrease in RORγt+ Tregs and corresponding increase in Helios+ Tregs was observed in NOD CNS1-/- mice. CNS1 deletion also did not alter the development of T1D or glucose tolerance despite increased pancreatic insulitis in pre-diabetic female NOD CNS1-/- mice. Furthermore, the proportions of autoreactive Tregs and conventional T cells (Tconvs) within pancreatic islets were unchanged. These results suggest that pTregs dependent on the Foxp3 CNS1 region are not the dominant regulatory population controlling T1D in the NOD mouse model.

A number of studies have shown that Tregs can develop both in the thymus (tTreg) and post-thymically in the periphery (pTreg), especially in the gut. During negative selection in the thymus, high affinity, self-reactive cells can escape clonal deletion by differentiating into Tregs, which is facilitated through CD28 costimulation and Foxp3 expression [11,12]. In contrast, naïve conventional T cells (Tconvs) exposed to self-antigens and the microbiome in the periphery can differentiate into Tregs, especially in mucosal tissues and in context of transforming growth factor-beta (TGF-β) and other pro-Treg environmental factors [13][14][15][16]. Direct evidence of pTreg generation can be seen when antigens were administered in the periphery by osmotic pumps [17], oral administration [18], or targeted antigen delivery to dendritic cells [19].
Previous studies have shown that Tregs are defective in both mice and humans that develop T1D. Thymic development and the Treg repertoire have been shown to be altered in NOD mice [20,21], and reducing T-cell receptor (TCR) diversity in NOD mice alleviates T1D, connected to a lack of insulin beta chain (InsB:9-23) autoreactive T cells [22]. In fact, recent studies have identified islet-antigen-specific Tregs that can control disease progression [23]. Modulation of the immune system has also been important in improving treatment of T1D with islet transplantation. Cytotoxic T-lymphocyte associated protein 4 (CTLA4), a coinhibitory molecule expressed on Tregs that inhibits CD28 signaling, has been made into a soluble fusion protein called CTLA4Ig. Combined administration of murine thymoglobulin (mATG) and CTLA4Ig during allogeneic islet transplantation selectively preserves Tregs while depleting autoreactive T cells, reversing T1D in NOD mice and indefinitely preserving graft survival [24]. Additionally, expression of programmed death-ligand 1 (PD-L1) on hematopoietic stem cells (HSPCs) inhibits autoimmunity while increasing the percentage of Tregs in the spleen and islets [25]. However, it remains unclear where these Tregs are generated and which subset of Tregs are defective. Finally, fusion peptides made by post-translational modifications of insulin derivatives and other proteins also create neoantigens that more strongly bind to potentially autoreactive cells in NOD mice [26][27][28] and humans [27,[29][30][31]. Although the role of these neoantigens and their recognition by Tconvs and Tregs remain unclear, their existence suggests that peripherally-derived antigens not typically found in the thymus might drive T1D and potentially pTreg generation. Interestingly, adoptive transfer experiments have shown non-redundant roles for tTregs and pTregs in maintaining tolerance [32].
Identification of phenotypic or molecular markers that distinguish tTregs from pTregs in vivo has proven more challenging. Expression of the surface marker Neuropilin-1 (Nrp1, also known as CD304) [33,34] and the transcription factor Helios [35] have been linked to tTregs; however, Helios and Nrp1 can also be markers of activation [36][37][38], and Nrp1 expression has been observed in some pTregs generated during adoptive transfer experiments and in tissue resident Tregs [39]. As a result, efforts have been made to identify distinct regulatory elements that differentially control Foxp3 expression in the different Treg subsets. Genetic studies by Rudensky and colleagues showed that there are 3 conserved non-coding sequence (CNS) elements, termed CNS1, CNS2, and CNS3, upstream of the Foxp3 promoter that control Foxp3 expression. One element, CNS1, was shown in C57BL/6J (B6) mice to selectively affect pTreg development without perturbing thymic expression of Foxp3 [40]. This region contains Smadbinding motifs downstream of the TGF-β signaling pathway, which has been implicated in the development of induced Tregs (iTregs) differentiated from naïve CD4 + T cells in vitro and gut-derived pTregs through mechanisms such as the activation of latent TGF-β through interactions with integrin α v β 8 [41]. Administration of TGF-β in vivo can also enhance the antigen-specific generation of pTregs [19,42]. B6 CNS1 -/blocked increases in Treg percentages in peripheral tissues with age, resulting in Th2-mediated pathology in the lungs and intestinal tract [43]. Continued characterization of the B6 CNS1 -/mouse revealed defects in pTreg generation and gut microbiome colonization in CNS1 -/mice [44,45] as well as in maternal-fetal tolerance [46].
There have been limited studies on the relative roles of pTregs and tTregs in controlling autoimmunity. Studies of experimental allergic encephalomyelitis (EAE) in a B6 CNS1 -/mouse model suggested that pTreg deficiency did not exacerbate disease [43], and adoptive transfer of TCR transgenic tTregs but not pTregs protected mice from EAE [33]. These results are consistent with another group who showed that a Smad3 binding mutation within Foxp3 CNS1 of NOD mice did not affect susceptibility to colitis, and although not stated explicitly, the authors did not note an increased incidence of T1D [47]. In contrast, other studies have shown that in NOD CD28 -/mice, adoptive transfer of islet antigen-specific pTregs was equally effective as tTregs in controlling T1D [33], and in one study, co-transfer of TCR transgenic BDC2.5tg + pTregs with BDC2.5tg + CD4 + T cells was sufficient to control T1D in NOD RAG1 -/mice [48]. In sum, these results suggest that tTregs are most critical in preventing and reversing autoimmunity and that in most cases pTregs are not effective. However, in the case of unmanipulated, polyclonal Tregs, the results are less clear.
Therefore, the goal of this study was to address the role of polyclonal pTregs in controlling autoimmune diabetes. pTreg development in the autoimmune-prone NOD mouse model was directly impaired through the use of the CRISPR/Cas9 gene editing system to selectively delete the CNS1 region of Foxp3. The results suggest that polyclonal tTregs are sufficient to control T1D. In parallel, another group generated a NOD CNS1 -/model using the CRISPR/Cas9 system. Their results differed, suggesting that pTregs were indeed critical for control of autoimmune NOD diabetes, with NOD CNS1 -/males and females exhibiting higher rates of T1D incidence [49]. Our results suggest that tTregs are the dominant regulators of T1D in the NOD mouse model, but under certain conditions such as targeted antigen-specificities or certain external factors, pTregs may play a role in the progression of T1D.

CNS1 deletion in NOD mice inhibits in vitro iTreg generation
The CNS1 region of Foxp3 was directly deleted in the NOD mouse background using the CRISPR-Cas9 genome editing system ( Fig 1A). Single guide RNAs (sgRNAs) were generated based on the 705 bp region that spans the regulatory element previously described in the B6 Foxp3 locus, namely from +2003 to +2707 bp from the transcriptional start site of Foxp3 [40]. PCR and DNA sequencing confirmed the deletion of a 736 bp sequence (+1976 to +2711), which spans the CNS1 region (Figs 1B and S1). To verify that the genetic deletion corresponded with a functional defect, we performed an in vitro Foxp3 induction assay using TGFβ. This assay demonstrated an impairment in the generation of iTregs (Fig 1C), confirming that genetic deletion of the CNS1 region in the NOD background conferred the previously described functional defect observed in the B6 background.

Treg quantitation and phenotyping in NOD CNS1 -/mice
In order to directly address the specific consequences of Foxp3 CNS1 deletion on pTreg generation, pre-diabetic female NOD mice aged 8-13 weeks were sacrificed and the in vivo Treg compartment was analyzed. Overall Treg percentages were comparable between NOD and NOD CNS1 -/mice in both lymphoid and non-lymphoid tissues, including the thymus, spleen, mesenteric lymph nodes (mLN), pancreatic islets, and large intestine lamina propria (LI LP), with a small but significant decrease in Treg percentages in the pancreatic lymph nodes of NOD CNS1 -/females (Fig 2A).
We also characterized the Tregs present based on markers such as Nrp1 and Helios expression. As expected, Nrp1 and Helios expression within the thymus were comparable between NOD and NOD CNS1 -/females (Fig 2B and 2C). We examined whether a decrease in the Nrp1and Heliospopulations in the periphery would be evident based on previous results in B6 mice suggesting that tTregs are Nrp1 + and Helios + [33][34][35]. Representative staining for Nrp1 and Helios in NOD mouse lymph nodes shows that co-expression of these markers is high among Tregs, and there is also a significant double-negative population that could represent potential pTregs ( Fig 2D). However, there was no significant difference in the percentage of Nrp1 -Helios - (Fig 2E), Nrp1 - (Fig 2F), or Helios - (Fig 2G) Tregs in the mLN and pLN.
Previous studies have suggested that the LI LP is largely comprised of Helios -pTregs, with a majority being positive for the transcription factor retinoic acid-related orphan receptor-γt (RORγt) [50,51]. These RORγt + Tregs are induced by the presence of complex microbiota and provide anti-inflammatory functions [52] and have been shown to have a distinct repertoire [53]. Genotyping PCR showing deletion of the CNS1 region in the primary founder and genetic sequencing after CRISPR/Cas9 deletion verifying a 736 base pair deletion in the Foxp3 locus. (C) CD4 + CD8 -CD25 -CD62L hi naïve T cells were sorted from NOD and NOD CNS1 -/mice and stimulated in 96-well plates with platecoated anti-CD3 (2 μg/mL) and anti-CD28 (0.5 μg/mL) antibodies for 72 hours with hIL-2 (100 IU/mL) in the absence or presence of TGF-β. Cells were than stained for CD4, CD25, and Foxp3 and analyzed with flow cytometry (n = 2 mice per genotype, two technical replicates per mouse used).  For example, RORγt + pTregs responsive to the short-chain fatty acid (SCFA) butyrate are generated in the gut [44] where they function to establish border-dwelling bacteria that protect against microbial exposure and maintain normal metabolite profiles [45]. Therefore, we hypothesized that there would be a reduction in the Heliosand RORγt + pTreg populations in NOD CNS1 -/females. Cells expressing Helios versus RORγt were mutually exclusive in Tregs isolated from the gut of both NOD ( Fig 2H) and NOD CNS1 -/- (Fig 2J) mice. There was a small but significant decrease in the percentage of the Heliosas well as RORγt + Treg populations in NOD CNS1 -/mice (Fig 2I and 2K, respectively). The implications of these differences are unclear and may reflect changes in the ability of the pTregs to develop in response to exposure to microbiota.

Islet antigen-reactive Tconv and Treg cells are unchanged in NOD CNS1 -/mice
Previous studies have shown that defects in Tregs in local tissues result in the expansion of autoreactive Tconv cells. Therefore, we examined the relative number of antigen-specific T cells in the pancreatic islets of NOD and NOD CNS1 -/female mice. To assess antigen-specific T cells, I-A g7 -peptide tetramers was used to stain cells specific for two insulin beta chain (InsB:9-23) variants: p8E, which represents the natural InsB peptide sequence; and p8G, which contains a mutation mimicking a proposed C-terminal InsB:9-20 truncation that shifts the MHC register of presentation for the epitope [54]. Insulin peptide is thought to be produced within the thymus by medullary thymic epithelial cells (mTECs) expressing the autoimmune regulator (Aire) transcription factor, driving thymic selection of T cells [55]. I-A g7peptide tetramers were also used to stain for reactivity against post-transcriptional hybrid insulin peptides (HIPs) generated from the fusion products of proinsulin C-peptide with chromogranin A (ChgA) or islet amyloid polypeptide 2 (IAPP2) [27].
MHC II-islet peptide-specific tetramer staining revealed that Tconvs were more reactive to the HIP peptides than InsB peptides (Fig 3A). This result is consistent with the idea that InsBreactive Tconvs would be deleted within the thymus while the so-called HIPs could represent peripherally-derived neoantigens that escape negative selection. Moreover, when comparing tetramer staining in Tconvs between NOD and NOD CNS1 -/females, no statistical difference was observed between the percentage of cells reactive to either the InsB:9-23 p8E (Fig 3C) or p8G variants (Fig 3D). The same was true for the Ins:ChgA 2.5HIP (Fig 3E) and the Ins:IAPP2 6.9HIP (Fig 3F) tetramers.
Thus, we hypothesized that HIP-reactive Tregs would represent a pTreg population derived from HIP-reactive Tconvs, which would predict that there would be fewer HIP-reactive Tregs in the islets of NOD CNS1 -/mice. However, Tregs isolated from pancreatic islets harvested from wild type NOD mice did not react with either the HIP2.5 or the HIP6.9 tetramers, suggesting that there may not be a significant percentage of pTregs in the islets overall. In contrast, Tregs reactive with the p8E and p8G tetramers were present in higher as cmfd compared to Tconvs (Fig 3B). There was no statistical difference observed between NOD and NOD CNS1 -/females in the percentage of Tregs reactive to the p8E (Fig 3C), p8G (Fig 3D), 2.5HIP (Fig 3E), and 6.9HIP (Fig 3F) tetramers. These data support the idea that CNS1 deletion does not alter the relative percentage of autoreactive Tconvs or Tregs within the pancreatic islets. Furthermore, the Treg compartment within the pancreatic islets seem to be comprised of largely tTregs.

No difference in glucose tolerance, time of T1D onset, or disease incidence in NOD CNS1 -/mice despite differences in insulitis
The ultimate potential consequence of the effects of disruption of pTreg induction was read out by the impact on T1D development in the NOD background. Female NOD and NOD CNS1 -/pancreata were collected at 6, 10, and 14 weeks of age for histology to measure insulitis (Fig 4A). Histological analysis of the isolated islets showed that there was some increase in insulitis in NOD CNS1 -/mice as early as 6 weeks and 10 weeks of age in the CNS1 -/mice. However, by 14 weeks, the time of initial development of clinical diabetes in our colony, there was only a marginal difference in insulitis between NOD and NOD CNS1 -/females, suggesting that the deletion of CNS1 did not ultimately lead to an increase in beta cell destruction. In fact, in younger pre-diabetic NOD CNS1 -/female mice, normal glucose tolerance was maintained (Fic 4B-C). These results show that the small differences in insulitis between NOD and NOD CNS1 -/mice at earlier ages were not consequential for either acute or chronic changes in blood glucose control. Importantly, these early differences in insulitis did not influence the Pooled insulitis scoring of female NOD and NOD CNS1 -/mice at 6, 10, and 14 weeks of age (n = 3 mice per age and genotype). Representative H&E histology images of pancreatic islets illustrating insulitis scoring rubric are shown at the top (0 = no insulitis, 1 = peri-insulitis or <25% infiltration, 2 is 25%-75% infiltration, and 3 is >75% infiltration or full islet destruction). p < 0.0001 for 6 weeks and 10 weeks, and p = 0.0140 at 14 weeks based on a chi-square comparison of islets scored as 0 to islets scored as 1, 2, or 3. onset or incidence of T1D in either female or male NOD CNS1 -/mice, which was not statistically different between the groups (Fig 4D).

Discussion
Our data show that deletion of CNS1 in the NOD background does not impact Treg frequencies or expression of known markers associated with pTregs and tTregs. Furthermore, deletion does not change the frequency of islet antigen-specific Tregs or Tconvs within the pancreatic islets. This includes InsB:9-23, a self-antigen known to be expressed in the thymus, and HIP tetramers, specific for post-translationally fused peptides generated in the periphery. These results are consistent with recent reports that show that InsB:9-23 tetramer binding is enriched in Tregs while HIP tetramer binding is enriched in Tconvs [23,28]. Finally, there is no difference in glucose tolerance, diabetes onset or incidence when comparing the NOD and NOD CNS1 -/mice.
These results are consistent with another group who generated a Smad3 binding mutation within CNS1 of NOD mice, which led to an age-dependent decrease in Treg frequency within the LI LP and other gut associated lymphoid tissues (GALT) but did not affect susceptibility to colitis [47]. However, another group independently generated a NOD CNS1 -/model using different sgRNAs [49]. Their work revealed a similar cellular phenotype where the frequency of Tregs found in most lymphoid and non-lymphoid tissues remained largely the same, but they observed a small but significant decrease in the Treg frequencies within the LI LP instead of the pLN. Interestingly, in this alternate NOD CNS1 -/model, there were significant reductions in Helios expression in Tregs in more lymphoid and non-lymphoid tissues than what we observed. Moreover, the alternate NOD CNS1 -/model showed a much starker difference in insulitis at 6 and 10 weeks of age, and although there is not a significant difference in insulitis by 14 weeks of age, overall insulitis was much higher in both their NOD and NOD CNS1 -/females. These results suggest that there may be a role for pTregs early in disease progression but overtime, the tTregs dominate and control disease progression such that depending on the colony and the degree of early beta cell damage, the resulting ability to control disease and maintain normoglycemia may be different. In the case of our mice, tTregs control disease progression preventing clinical disease, while in the other NOD Foxp3 CNS1 deletion model a clinical disease phenotype is evident in both male and female mice.
Although many of the results we obtained are consistent with the previous NOD CNS1 -/studies, there are a number of factors that could account for both earlier and more severe insulitis in the CNS1 -/mice described by Kissler and colleagues than those in our colony. One difference between the different NOD CNS1 -/mouse strains is that the CNS1 deletion is not identical. Kissler and colleagues used sgRNAs that span a wider region around the DNA than the ones we utilized. Thus, the resulting edit generated a slightly larger 795 bp deletion (+1949 to +2743 from the Foxp3 transcriptional start site) (unpublished information from Kissler) than was generated in the animals herein (+1976 to +2711). Whether these additional nucleotides might alter DNA binding of other enhancer proteins remains unclear but might be impactful.
Perhaps the most likely difference between the two mouse models resides in the gut microbiome or other environmental factors. There is extensive literature documenting that innate microbial sensing and gut microbial constituents can shift T1D incidence, explaining differences observed in diabetes incidence among animal facilities [56,57]. Specific differences in the gut microbial community of NOD males and females post-puberty also contribute to the differences in T1D incidence seen between the sexes [58,59], and microbiome composition can be used to predict development of T1D [60]. The gut microbiome might be linked to T1D due to the fact that the pLN is also a draining site for microbial antigens [61], and autoreactive cells have also shown cross-reactivity between microbial antigens and islet antigens [62,63]. In fact, recent characterization of the B6 CNS1 -/mouse directly revealed that defects in gut pTreg generation alter gut microbiome colonization [44,45]. Since there is a change in gut Tregs in both NOD CNS1 -/models generated, the ability to observe a similar disease phenotype in our NOD CNS1 -/model might be masked by differences in gut microbiome composition. Future studies that sequence and compare the microbiome of NOD and NOD CNS1 -/mice generated in the two facilities would be required to address this issue directly.
Although no differences were observed in the Treg compartment in either NOD CNS1 -/model, it is possible that other CD4 + T cell subsets were impacted by the loss of CNS1-dependent pTregs. For instance, increased Th2-mediated inflammation and responses in the intestines have been observed in both B6 CNS1 -/mice [45] and B6 Foxp3 Cre RORγt fl/fl mice [50]. Intestinal RORγt + Tregs have also shown a role in regulating tissue-resident Th1 and Th17 cells [51]. The impact of CNS1-dependent pTregs on these subsets has yet to be explored in either mouse model, although the other model noted that overall CD4 + T cell proportions did not change in any of the tissues mentioned. Since a reduction of RORγt + Tregs has been observed in the LI LP of both models, future work should explore whether this shift impacts either the proportion or cytokine production of tissue-resident Th1, Th2, or Th17 cells. Additionally, selective shifts in these subsets within the pancreata of one NOD CNS1 -/model versus the other could potentially provide another means for explaining the differences observed in T1D incidence.
In summary, our data suggest that tTregs play the dominant role in the control of islet-specific autoimmunity and subsequent development of diabetes. This conclusion is supported by the observation that within the pancreatic islets, there are not many Tregs that recognize tissue-specific, peripherally-generated autoantigens as detected by HIP I-A g7 tetramers, consistent with the notion that the Treg compartment within the islets is largely comprised of tTregs. Although our work shows that tTregs are sufficient for controlling T1D in NOD mice, the role of CNS1-dependent pTregs might be influenced by environmental factors based on the differences seen between our model and an independently-generated NOD CNS1 deletion. This observation is consistent with what is known about T1D and could provide further insight into the interplay between the immune system and environment that determines both mouse and human disease susceptibility and outcomes.

Mice
NOD CNS1 -/mice were made by the JAX Model Generation services using the CRISPR/Cas9 system (gRNAs 5'-CATTACAAAGCACAATAAAT-3' and 5'-ATACCAGCCATGGGTGTC TC-3'). Potential founders were genotyped and sequenced using the following primers: 5'-CAGCAGTGCTCTTACCCATG-3' and 5'-CAGTGAGAGCAGTTTAGAGG-3'. Founder mouse progeny were crossed to NOD JAX mice for at least three generations before experiments to account for potential off-target CRISPR/Cas9 effects. Hyperglycemia and spontaneous induction of T1D in NOD mice were monitored by weekly blood glucose measurements made with the OneTouch Ultra 2 glucometer and UNISTRIP1 generic blood glucose test strips. Mice with two consecutive blood glucose readings of at least 250 mg/dL were considered diabetic. All experiments were conducted under a protocol approved by the University of California San Francisco's Institutional Animal Care and Use Committee.

Isolation of lymphocytes and cell staining
Lymph nodes were physically dissociated and filtered to generate single-cell suspensions. Spleens were physically dissociated and incubated at room temperature in ACK red blood cell lysis buffer before filtration. Large intestine lamina propria (LI LP) were digested using the Lamina Propria Dissociation Kit (MACS Miltenyi Biotec) per manufacturer's instructions. Islet isolation was done courtesy of the UCSF Mouse Islet Isolation Facility [64] and then dissociated in Cell Dissociation Buffer Enzyme-Free PBS-based (Thermo-Fisher Scientific) at 37˚C in a tissue culture incubator. Antibodies used for flow cytometry staining are listed in S1 Table. Cell viability was measured before flow cytometry using the Vi-Cell XR Cell Viability Analyzer (Beckman Coulter) and tracked during flow cytometry using the LIVE/DEAD Fixable Blue Dead Cell Stain Kit for UV Excitation (ThermoFisher Scientific). After viability staining, tetramer staining was performed at room temperature for 1 hour, and then without washing, concentrated surface stain was added for an additional 15 minutes at room temperature. The following mouse I-A g7 tetramers were used courtesy of the NIH Tetramer Core Facility: InsB p8E variant (HLVERLYLVCGEEG) conjugated to APC; InsB p8G variant (HLVERLYLVCGGEG) conjugated to APC; Ins:ChgA 2.5HIP (LQTLALWSRMD) conjugated to APC; Ins:IAPP2 6.9HIP (LQTLALNAARD) conjugated to PE; and DPB1 � 04:01/DPA1 � 01:03 control human CLIP 87-101 conjugated to APC, BV421, and PE (PVSKMRMATPLLMQA). Fixation and intracellular staining were done using the eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo-Fisher Scientific) per manufacturer's instructions. Stained single-cell suspensions were analyzed using a Fortessa flow cytometer running FACSDiva (BD Biosciences). FSC 3.0 files were analyzed and presented with FLOWJO Software (flowjo.com).

Histology and pancreatic insulitis scoring
Pancreata were fixed in 10% neutral buffered formalin solution overnight and dehydrated in 70% ethanol. Histology was performed by HistoWiz Inc. (histowiz.com) using a Standard Operating Procedure and fully automated workflow. Samples were processed, embedded in paraffin, and sectioned at 6 μm with 175 μm steps between slices. Three sections per mouse pancreas were collected. Haemotoxylin and Eosin (H&E) staining was performed, and then sections were dehydrated and film coverslipped using a TissueTek-Prisma and Coverslipper (Sakura). Whole slide scanning (40x) was performed on an Aperio AT2 (Leica Biosystems). Islets were scored blinded to genotype and age with at least 100 islets scored per mouse. Insulitis was scored as none (0), peri-insulitis or less than 25% infiltration (1), 25-75% infiltration (2), or greater than 75% infiltration (3). Islet scoring was done in the open source software QuPath (qupath.github.io) [65].

Glucose tolerance test
Glucose tolerance was tested in 8-9-week-old NOD and NOD CNS1 -/mice after a 16 hour overnight fast. Mice were challenged with injections of 1M glucose in PBS (2 g glucose/kg of total body weight) intraperitoneally and blood glucose was measured with the OneTouch Ultra 2 glucometer and UNISTRIP1 generic blood glucose test strips at baseline and 15, 30, 60, 90, 120, and 150 minutes post-challenge.

Statistics and methods of analysis
Data analyses were performed using GraphPad Prism 7 software (graphpad.com/scientificsoftware/prism), and values at p < .05 were deemed significant. For comparison of the insulitis scores, the chi-square test was used to calculate if there was a significant difference in the proportion of islets without any peri-insulitis or insulitis (score = 0) to islets with peri-insulitis or full insulitis (score = 1, 2, or 3). All other statistical comparisons were done using a two-tailed, unpaired t test.

S1 Fig. Sequence of CNS1 deletion in NOD mice.
Displayed here is the +1960 to +2722 region from the transcriptional start site of Foxp3. The original CNS1 region identified in B6 mice (+2003 to +2707) is highlighted in yellow, while the sequence of the deleted region generated in the NOD background with CRISPR/Cas9 (+1976 to +2711) is bolded in red. (TIF) S1