Results and Discussion

We chose B6.H-2^g7 (B6g7) mice, congenic for the NOD-derived major histocompatibility complex (MHC), as comparators for mapping the NOD NK-cell defect. The two mouse strains had similar fractions of CD49b⁺CD3⁻ NK cells in the spleen and other lymphoid organs, as previously reported (5, 9), with similar maturation (CD27 vs. CD11b) and activation (CD25 vs. CD69) profiles (Fig. S1). However, largely reflecting the unique structure of the NK complex on chr 6 in NOD mice (10), the two strains had very different patterns of NK-cell receptor expression, e.g., of NK1.1, Ly49I, Ly6c, and NKG2A (Fig. S1).

We quantified two parameters in the genetic mapping studies. “NK numbers” refers to the percentage of CD49b⁺CD3⁻ cells in total splenocytes and hovered around 2% in both strains (Fig. 1A). “NK activity” was measured by an in vivo assay (3, 5) that compared the survival of coinjected MHCI-positive and -negative target cells over an 18-h assay period, yielding a cytotoxicity index (Fig. 1B). Routinely, about 2- to 3-fold fewer of the MHCI-negative targets were killed in NOD than in B6g7 mice (Fig. 1C).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

“NK numbers” and “NK activity” in B6g7 vs. NOD mice. (A) Flow cytometric determination of NK numbers in 7-week-old B6g7 and NOD mice. The percent symbol (%) refers to the fraction of CD49⁺ CD3⁻ cells in the lymphocyte gate. (B) Schema for in vivo assay of NK activity. MHCI-positive and MHCI-negative splenocytes were differentially labeled with CFSE, mixed equally, and injected into B6g7 or NOD mice. Remaining donor cells were counted 18 h later, and a cytoxicity index was calculated. (C) Representative indices of NK activity in 7-week-old B6g7 or NOD mice.

NOD × B6g7 F₁ offspring had NK numbers indistinguishable from those of the two parental strains (Fig. 2A), whereas their NK activity was relatively high, as in B6g7 animals (Fig. 2B), indicating that this phenotype is dominant. Unexpectedly, average NK numbers in a cohort of 187 NOD × B6g7 F₂ offspring were higher than in both of the parental strains or in the F₁’s (Fig. 2A), suggesting that counterbalancing positive and negative influences are encoded in the NOD and B6g7 genomes. Individual F₂’s showed a broad range of NK cytotoxicity indices (Fig. 2B), arguing for polygenic control of this trait. There was only a poor correlation between the NK number and activity values (Fig. 2C), meaning that at least some of the loci controlling these traits operate independently.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Genetic analysis of NK cells in B6g7 vs. NOD mice. (A and B) NK numbers and in vivo activity for B6g7, NOD × B6g7 F₁, NOD, and F₂ mice at 7 weeks of age. (C) NK activity vs. numbers in F₂ mice (n = 187). (D–F) Genetic linkage with NK numbers on activity. (i) LOD score plot (R-QTL) for genomewide linkage (n = 187). The peak LOD score on chr 8 was 7.4. The dotted line delineates significance, as calculated by a permutation test (n = 10,000; P < 0.05). (ii) A higher resolution map of individual chromosomes. (iii) Distribution of NK numbers on activity for each genotype. B/B, both alleles from B6g7; N/B, heterozygotes; N/N, both alleles from the NOD genome. Symbols to the right of each set of values indicate mean ± SEM. (E–G) Epistasic interactions influencing NK numbers and activity. (i) 2D genomewide linkage in F₂ genotypes. Joint LOD scores were plotted as a heat map (12). The significant joint LOD scores of chrs 1 and 6, chrs 1 and 8, and chrs 6 and 8 were 11.88, 17.40, and 23.49, respectively. Greater than 8.4 was a significant value per permutation test (n = 10,000; P ≤ 0.05).

Genomewide single-nucleotide polymorphism (SNP) genotyping was performed on this F₂ cohort and was analyzed relative to both NK numbers and activity. The data on NK numbers yielded a major peak of linkage on chr 8, with a LOD score of 7.4, well above the significance threshold of 3.3 determined by permutation testing (Fig. 2D i). The chr 8 linkage was broad, extending from 30 to 100 Mb and peaking at 65 Mb (Fig. 2D ii). The NOD haplotype in this region was correlated with lower NK-cell numbers (Fig. 2D iii). Joint LOD scores, calculated to assess epistatic influences (12), revealed interactions between the chr 8 interval and loci residing on several other chromosomes (Fig. 2E i, detailed in the legend to this figure). For example, the chr 9 locus had a strong negative impact on the chr 8 effect, the allele at chr 9 largely counteracting the B6 allele at chr 8 (Fig. 2E ii), which can easily explain why average NK numbers end up being very similar in NOD and B6g7 mice.

Intervals on chrs 1, 6, and 8 showed significant linkage to NK activity, i.e., had a LOD score >3.5 (Fig. 2F i and ii). The chr 8 interval had the strongest effect, with a directionality consistent with it being responsible for the low NK-cell activity of the NOD strain (Fig. 2F iii). The analysis of joint LOD scores revealed that the chr 1 and chr 6 regions both had an epistatic interaction with the interval on chr 8, as well as with each other (Fig. 2G i). The dampening effect of the NOD chr 8 allele on NK-cell performance was reduced when the chr 1 interval was NOD/B6 and was almost completely masked when that interval was NOD/NOD (Fig. 2G ii Left). In contrast, the negative impact of the chr 8 allele was most evident when the chr 6 interval was present in NOD-homozygous state (Fig. 2G ii Right). Because the chr 1 and chr 6 genetic influences did not drive NK activity in the relevant direction, i.e., the B6 haplotype was associated with lower activity, we did not further pursue their localization. However, given the unique structure of the NK complex in NOD mice (10), one or more of its genes are likely candidates for the chr 6 effect. One obvious candidate for the chr 1 locus is the gene encoding 2B4 (Cd244), which is active in NK cells, or more generally the SLAM/CD2 gene cluster, of which NOD mice carry an unusual haplotype (13).

Next, we constructed congenic mouse lines carrying B6-derived intervals on chr 1, chr 8, or both on the NOD genetic background (NOD.C1^b/b, NOD.C8^b/b, and NOD.C1^b/b8^bb, respectively). Confirming the F₂ data, NK numbers were higher in the NOD.C8^b/b and NOD.C1^b/b8^b/b congenic lines than in either of the parental strains (Fig. 3A Left). Also anticipated was that the NOD.C8^b/b line would exhibit the relatively high NK activity characteristic of the B6g7 strain, which would be further augmented in the double-congenic line, given the epistatic interaction between the chr 1 and chr 8 intervals mentioned above (Fig. 3A Right). (The increased activity of NOD.C1^b/b vis-à-vis NOD mice probably reflects epistatic interactions between loci on chr1 and other chromosomes.)

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Establishment of congenic mouse lines and bone-marrow chimeras. (A) NK cell number and activity phenotypes of B6g7, NOD, and congenic mice. (Left) NK numbers for B6g7, NOD, NOD.Chr1^b/b, NOD.Chr1^b/bChr8^b/b, and NOD.Chr8^b/b mice. Each dot represents an individual mouse. (Right) A genotype schema for the congenic strains. (B) The different elements to be assayed in the radiation bone-marrow chimeras. Stroma are radiation resistant and hematopoietic cells radiation sensitive. (C) Test of stroma vs. hematopoietic cells. Marrow cells were injected into lethally irradiated NOD.Chr8^b/b (CD45.1) (n = 8) mice or NOD (CD45.1) (n = 8). Donor NK numbers, distinguished by their CD45.2 surface marker, were analyzed. (D) Test of NK cells vs. other hematopoietic cells. Marrow from NOD.CD45.2 donor mice and NOD.Chr8^b/b (CD45.1) donor mice were mixed equally and injected into lethally irradiated NOD-CD45.1/CD45.2 heterozygous recipient mice (n = 9). NK cells were distinguished by their CD45 allele and NK numbers derived from the two donors plotted. Error bars in C and D represent mean + SD.

Exploiting these congenic lines, we addressed whether the NOD NK-cell defect specified by the chr 8 interval partitioned with the radio-resistant stromal-cell compartment, with the radio-sensitive hematopoietic cell compartment, or intrinsically within NK cells (Fig. 3B). First, we transferred bone-marrow cells from NOD mice carrying the CD45.2 allelic marker into lethally irradiated NOD or NOD.C8^b/b hosts, both bearing the CD45.1 allele, and quantified NK-cell numbers 7 weeks later. There was no difference in the fraction of NK cells in the spleens of the two types of host (Fig. 3C), indicating that the dampening effect of the NOD chr 8 segment was not determined by the stromal, but rather by the hematopoietic cell compartment. Secondly, we performed a similar experiment, but transferred an equal mix of NOD.CD45.2 and NOD.C8^b/b (CD45.1) bone-marrow cells into lethally irradiated NOD.CD45.1/CD45.2 heterozygotes. There was no difference in the fraction of donor CD45.1⁺2⁻ and donor CD45.1⁻2⁺ NK cells in the spleen, both being relatively low (Fig. 3D), establishing that the impact of the NOD chr 8 allele was not NK-cell intrinsic and was thus determined by other lymphoid or myeloid cells.

The congenic lines were also the starting point for finer mapping within the chr 8 interval. Crossing the NOD.C1^b/b and NOD.C1^b/b8^b/b mice (following an intercross/backcross strategy to generate mice with a single recombinant chr 8) yielded 147 offspring with recombination events within the 65.7–110.0 Mb interval, while fixing the epistatic B6 segment on chr 1. SNP mapping on this cohort generated almost identical curves for NK numbers and activity, peaking between 76 and 91 Mb (Fig. 4A). From these, we established three subcongenic lines with recombination points within the interval (Fig. 4B). F₈₀ and F₈₃ mice had relatively low NK numbers and activity values similar to those of the NOD.C1^b/b comparator, whereas F₈₅ mice had the higher numbers and activity characteristic of NOD.C1^b/b8^b/b and B6g7 mice (Fig. 4C). Thus, the major NK-cell defect in NOD mice maps between positions 83 and 85 Mb on chr 8. According to the Ensembl database, seven protein-coding genes reside within this interval (Fig. 4D). A microarray comparison of gene expression by splenocytes from B6g7 and NOD mice (Affymetrix M430, triplicate) revealed that the locus encoding the cytokine, IL-15, was expressed most differentially between the two strains—about 2-fold less by NOD splenocytes (Fig. 4E).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Fine mapping of the chr 8 influence. (A) LOD score plots of NK numbers (Upper) and NK activity (Lower) in 7-week-old recombinant mice (n = 147). A significant score for NK numbers and NK activity was estimated as 3.07 and 3.01, respectively, via a permutation test (n = 10,000; P ≤ 0.01). (B) Genotype map of F₈₀, F₈₃, and F₈₅ subcongenic mice. (C) Phenotypes of NK cells from F₈₀ (n = 6), F₈₃ (n = 19), F₈₅ (n = 9), NOD.Chr1^b/b (n = 12), and NOD.Chr1^b/bChr8^b/b (n = 6) mice. Each dot represents an individual mouse. (D) Map of the relevant interval on chr 8 (adapted from the Ensembl browser; build 37) with the various genes known to be located within this interval and the relative expression in B6g7 vs. NOD splenocytes, from our microarray data. Each dot represents the NOD vs. B6 ratio for an individual probe. Probes yielding values with a false discovery rate of <0.01 are highlighted as filled dots. The two-headed arrow indicates the interval of interest, delineated by the F₈₃ and F₈₅ recombinant lines. (E) Probes for and profiles of candidate genes. FC, fold-change NOD vs. B6g7; FDR, false discovery rate for the corresponding FC.

Given its well-known roles in NK-cell maturation, survival, and activity, IL-15 makes an excellent candidate for the molecule responsible for the NK-cell defect in the NOD strain. We sequenced the exons and promoter region of the Il15 genes from B6g7 and NOD mice, uncovering 18 SNPs: 8 in the promoter region, 5 in the 5′-UTR, and 5 in the 3′-UTR (Fig. 5A). As none of the SNPs alter the protein sequence, we focused on differential IL-15 expression in the two strains. RT-PCR analysis confirmed that there were reduced IL-15 transcript levels in the spleens of NOD mice and that this was also true of the two known splice variants (Fig. 5B). Expression-QTL (eQTL) analysis, using spleen RNA from the same F₂ offspring used for mapping the NK numbers/activity QTLs, revealed that the differential in IL-15 expression was cis-determined (Fig. 5C). According to RT-PCR analysis on sorted splenocyte populations, dendritic cells and macrophages showed the highest IL-15 transcript levels and the greatest differential between NOD and B6g7 mice (Fig. 5D), consistent with the above conclusion that the NK-cell defect was dictated by non-NK hematopoetic cells (Fig. 3B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Il15 in NOD and B6g7 mice within Il15 exons. (A) Organization of Il15 on chr 8. The eight standard exons are indicated as crossbars or crossboxes, which are expanded below. Open and filled boxes indicate untranslated and protein-coding regions, respectively. The positions of SNPs identified by sequencing and distinguishing B6 from NOD are given as vertical bars. Pairs of probes used for RT-PCR analysis are shown as a, b, and c. (B) Expression of IL-15 transcripts in the two mouse strains. The pairs of probes in A were used to quantify expression of the various splice forms (mean + SD). (C) eQTL analysis for Il15. Splenic IL-15 transcript levels in F₂ mice (n = 93) were quantified by RT-PCR and linkage analysis showed linkage to chr 8. A P = 0.05 significant eQTL LOD score was estimated as 3.5 by a permutation test (n = 10,000). (D) Il15 gene expression in diverse cell types. Each cell type was sorted from spleen cell suspensions, and IL-15 transcript levels were determined by quantitative RT-PCR (mean + SD).

Soluble IL-15 has little in vivo activity on its own; rather, it is usually bound to the IL-15 receptor α (IL15Rα) chain, mostly on dendritic cells, and “presented” in trans to NK (and other) cells (14). Complexing IL-15 with IL15RαFc stabilizes it, promoting its bioavailability (15–17). To see whether “IL-15 complexes” could boost NOD NK-cell numbers and activity to B6g7 levels, we injected NOD mice with differing amounts in differing ratios every other day for 8 days, and assayed NK-cell properties. Serial injections of 25 ng IL-15 complexed with 100 ng IL-15RαFc produced the desired few-fold increase in NK numbers and activity (Fig. 6A). We also applied this regimen to the simplified (TCR transgenic) T1D model, BDC2.5/NOD, of which 15–25% normally develop diabetes beginning at 15–20 weeks of age. As with the standard NOD strain, IL-15 complexes boosted NK-cell numbers in the spleens of BDC2.5/NOD mice a few-fold and also in the pancreatic lymph nodes and the islet infiltrate (Fig. 6B i). However, according to their profiles of CD69 and CD25 expression, the NK cells in complex-injected mice were not unduly activated (Fig. 6B ii). The increase in cell numbers at these doses was quite specific for the NK-cell compartment, as no augmentation in numbers of CD4⁺ or CD8⁺ cells was observed in the different organs nor of activated (CD4⁺25⁺ or CD8⁺44⁺122⁺) T cells, nor of dendritic cells (CD11b⁻¹¹C⁺) and macrophages (CD11b⁺¹¹C⁻) (Fig. 6B iii). Strikingly, beginning as soon as 2 weeks after the first injection, about 30% of the BDC2.5/NOD mice administered IL-15 complexes every other day starting at 14 days of age developed diabetes, i.e., long before they normally would (Fig. 6B iv). This effect evokes previously described associations between numbers or activity of NK cells and hyperacute diabetes in mouse models (7, 8).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Induction of NK cells and diabetes in NOD mice by IL-15 complexes. (A) Titration of IL-15/IL-15R complexes effect in vivo. Dilutions of Il-15 complexes, starting at 3,500 ng IL-15RαFc and 750 ng IL-15, were injected every 48 h into NOD mice, from 14 days of age. Splenocytes were analyzed after eight injections. Treatment with 100 ng IL-15RαFc and 25 ng IL-15 yielded the desired 2- to 3-fold increase in NK numbers (i), which translated to a similar augmentation in NK activity (ii). (B) Induction of diabetes through induction of NK activity. (i) The fraction of NK cells in spleens (SPL), pancreatic lymph nodes (PLN), or pancreata (Panc.) of the BDC2.5/NOD mice injected with Il-15/IL-15Rα complexes (n = 9) or PBS (n = 9) for 2 weeks starting at 14 days of age. (ii) Expression of CD69 and CD25 activation markers on NK cells from mice after injection of Il-15/IL-15R complexes as per B i. (iii) The effect of Il-15 complexes on diverse cell types. Numbers of CD49b⁺CD3⁻ NK cells (NK), CD3⁺CD4⁺CD8⁻ T cells (CD4⁺), CD3⁺CD4⁺CD25⁺ T cells (CD25⁺), CD3⁺CD4⁻CD8⁺ T cells (CD8⁺CD3⁺), CD8⁺CD44⁺CD122⁺ T cells (CD44⁺CD122⁺), CD11b⁻CD11c⁺ predominantly dendritic cells, DC), and CD11b⁻CD11c⁺ (predominantly macrophages, MF) in spleens of mice administered IL-15/IL-15R complexes as per B i (n = 9). (iv) Diabetes incidence in BDC2.5/NOD mice treated with Il-15/IL-15R complexes every other day from 14 days to 12 weeks of age.

Thus, inadequate expression of the Il15 gene is a major contributor to the NK-cell defect in NOD mice. IL-15 regulates the maturation and survival of NK cells and also controls their activity by up-regulating expression of NK receptors and cytotoxic effector molecules (18). Whereas our genetic dissection revealed Il15 to have an important impact on both the numbers and activity of NK cells, several other loci also influenced these properties. The integration of these different effects, which could be either positive or negative and could alter numbers and activity independently, led to normal NK-cell numbers with 50–70% reduced activity in the NOD compared with the B6g7 strain. Although we did not investigate these aspects, IL-15’s known influence on other cell types (18) has the potential of explaining certain other aberrant characteristics of the NOD strain, e.g., deficiencies in the NK-T or dendritic cell compartments (19, 20).