Adult HCs (>18 years) were admitted to the “Servizio Medicina Preventiva” at the San Raffaele University Hospital, Milan, Italy. Pediatric HCs (≤18 years) were children with no immune system-related diseases undergoing surgery only for congenital disease at the Orthopedic Pediatric Department at the San Raffaele Hospital. Both adult and pediatric HCs had no family members with T1D.

Adult T1D patients were recruited at the Endocrinology Department. For genetic analysis, both new-onset (blood withdrawal within 3 months from diagnosis) and long-standing adult T1D individuals were included in the study. Pediatric T1D subjects were enrolled at the Pediatric and Neonatology Department at the San Raffaele Hospital. Only children with a blood withdrawal within the first 10 days after diagnosis (new-onset T1D) were included in the study. The diagnosis of T1D was made based on sustained hyperglycemia (documented by repeated glucose measurements and HbA1c measurement), fasting C-peptide levels, 1.0 ng/ml, or the presence of at least one islet-specific autoantibody (16, 17).

FDRs of T1D patients were recruited under the umbrella of the Type 1 diabetes TrialNet Pathway to Prevention Trial (TN01) at the TrialNet Clinical Center of the San Raffaele Hospital. These subjects were divided in two groups based on the presence of circulating islet-specific autoantibodies (e.g., GAD65, IA2, mIAA, or ZnT8) at blood withdrawal: FDRs without (Ab^neg) or with at least one (Ab^pos) autoantibody.

Human peripheral blood was collected upon informed consent in accordance with the Declaration of Helsinki and with local ethical committee approvals: IRB#DRI-003 for HCs and T1D patients and IRB#NHPROT32803-TN01 for Ab^neg and Ab^pos subjects. For pediatric subjects, the informed consent has been provided by parents. All the individuals included in the study were European Caucasians, and samples were collected from 2016 to 2017. A full blood count laboratory test was performed for all the samples received. Subjects with ongoing infections were excluded from the analysis.

Characteristics of subjects enrolled in the study are listed in ( Table 1 ) and in ( Table 2 ).

The frequencies of DC-10 and other monocyte/myeloid populations, and the HLA-G, ILT4, CD83, and HLA-DR expression levels were assessed on EDTA peripheral blood within 6 h after blood withdrawal. In brief, 50 µl of antibody (Ab) mix was directly added to 180 µl of whole blood and incubated for 15 min at room temperature in the dark. The samples were then incubated for additional 13 min at room temperature in the dark with lysis buffer (EDTA 1 mM, KHCO₃ 10 mM, and NH₄Cl 150 mM) to perform red blood cell lysis, washed twice, and re-suspended in phosphate-buffered saline (PBS) (Sigma, CA, USA) with 2% fetal bovine serum (FBS) (Lonza, Italy). Cells were identified using a multiparametric approach based on the combination of monoclonal Abs ( Supplementary Table S1 ). Samples were acquired within 12 h after the staining using a FACSCanto II flow cytometer (Becton Dickinson, Mountain View, CA). Samples were processed and stained in batches to reduce technical variability. Moreover, to ensure a standardized sample analysis and a reliable mean fluorescence intensity (MFI) evaluation among samples, flow cytometer was calibrated daily using 8-peak rainbow calibration particles (Spherotech, IL, USA) before sample acquisition. MFI was evaluated on the whole gated populations. Data were analyzed with FCS express v6 (De Novo Software, Glendale, CA), and quadrant markers were set accordingly to the relative fluorescence minus one (FMO) staining performed for all the donors analyzed.

Statistical methods applied for each set of data in Figures 1 , 2 are reported in the related figure legend. For Figures 3 , 4A, B , and Supplementary Figures S3, S4C and S5 , comparisons among groups were performed with linear mixed-effects (LME) models to account for the presence of several subjects within the same family. In each analysis, to meet the assumption of normality of the residuals of the models, an appropriate transformation, indicated in the figure legend, was eventually applied to the response variable; and, when necessary, few outliers identified by the specific model were not included in that analysis. For the sake of completeness, these outliers have been included in the figures as black dots ( Figures 3B, C , Supplementary Figures S3A and S5 ) and considered in the computation of the reported descriptive statistics. For testing difference between all pairs of groups, a post-hoc analysis was performed by using the R package phia and by applying Bonferroni’s correction to account for multiple comparisons. For ( Figure 4C ), the linear model was employed to evaluate the dependency between % of CD83⁺ DC-10 and % of HLA-G⁺ DC-10, within each group. The standard linear regression was estimated for the groups HC, Ab^pos, and T1D, since all subjects were independent, while the LME model was used in case of Ab^neg in order to account for the presence of several subjects within the same family. In each analysis, in order to meet the assumption of normality of the residuals of the model, an appropriate transformation was applied to the response variable (% of HLA-G⁺ DC-10).

For the genetic analyses of HLA-G 3′UTR and 5′URR (PROMO) regions reported in ( Figure 5 , Supplementary Figure S6, Supplementary Tables S5 and S7 ), logistic mixed-effects models were employed for comparing the frequency of the single haplotypes or genotypes between each group HC, Ab^pos, and T1D versus Ab^neg. The models account for the presence of several subjects within the same family and, in case of the haplotypes, also for the presence of two alleles per subject. Comparisons have been performed only for haplotypes or genotypes with at least two groups with absolute frequency of at least 10. P-Values were adjusted by applying Bonferroni’s correction to account for multiple testing.

Multiple mixed-effects regression analyses were performed to identify potential predictive variables of T1D development, since these models can account for the presence of several individuals within the same family. The logistic mixed-effects model was employed for predicting the probability to be Ab^neg versus HCs. The cumulative link mixed model was used for predicting the probability to be: a subject at risk without (Ab^neg) or with autoimmunity (Ab^pos) or to be a person with T1D (considering the groups as ordered). For both analyses, the variables considered in the starting model were Age, % of DC-10, % of cDC2, % of HLA-G⁺ DC-10, % of CD83⁺ DC-10, and 3′UTR genotype (considering the comparisons of all genotypes with respect to DelC/DelC genotype). The final models were obtained with a backward procedure of variable selection.

In all the analyses, p-values less than 0.05 were considered significant. All p-values of the analyses were reported in the figures and/or ( Supplementary Tables S2, S5 and S7 ). Statistical analysis was performed using R 3.5.0 (http://www.R-project.org/).

Using the newly identified set of markers CD14, CD16, CD141, and CD163 (14), and the gating strategy shown in ( Supplementary Figure S1A ), we evaluated the frequency of circulating DC-10 in a cohort of 55 healthy donors (HCs). The distribution of classical (c)DC1 (CD11c⁺CD14⁻BDCA-3⁺), cDC2 (CD11c⁺BDCA-1⁺), and plasmacytoid (p)DCs (CD11c⁻BDCA-2⁺) subsets ( Supplementary Figure S1B ) was analyzed in parallel. The frequency of DC-10 was constant with respect to the age among the donors analyzed and represented on average 0.31% ± 0.15% (range 0.07%–0.70%) of CD45⁺ cells ( Figure 1A ). Similarly, the frequency of cDC1 and cDC2 subsets was stable in young and adult HCs being on average 0.45% ± 0.20% (range 0.05%–1.19%) and 0.23% ± 0.09% (range 0.06%–0.49%) of CD45⁺ cells, respectively. In line with previous studies (18), the frequency of pDC decreased with age (r = −0.3061, p = 0.0231) and represented on average 0.17% ± 0.08% (range 0.05%–0.45%) of CD45⁺ cells ( Figure 1A ).

DC-10 are a subset of CD14⁺CD16⁺ cells (14); thus, we analyzed the frequency of circulating monocyte subsets according to the standard classification (19): classical (CD14^highCD16⁻), CD14⁺CD16⁺, and non-classical (CD14^lowCD16⁺) monocytes. The proportion of non-classical monocytes decreased with age (r = −0.2872, p = 0.0335) with an average of 0.44% ± 0.21% (range 0.16%–0.95%) in CD45⁺ cells, while the frequencies of classical and CD14⁺CD16⁺ monocytes were constant in young and adult HCs, with an average of 2.55% ± 1.65% (range 0.31%–6.74%) and of 1.36% ± 1.02% (range 0.25%–4.72%) in CD45⁺ cells ( Figure 1B ).

Overall, in healthy conditions, the distribution of DC-10 in the peripheral blood is constant in young and adults.

In vitro differentiated DC-10 express HLA-G (7, 13); thus, we investigated whether this feature is shared by their in vivo counterpart. Circulating DC-10 from HCs expressed variable levels of HLA-G ranging from 0.03% to 24.37%, with the percentage of HLA-G-positive DC-10 correlating with HLA-G expression levels, indicated by the MFI (r = 0.4724, p = 0.0005) ( Figure 2A ).

CD14⁺ monocytes constitute the main subset of immune cells expressing HLA-G, although at low and variable levels [e.g., 0.12% ± 0.07% of HLA-G⁺CD14⁺ cells (20–24)]. In our cohort of HCs, circulating DC-10 expressed significantly higher HLA-G levels compared to those expressed by total CD14⁺ cells in terms of both percentage of HLA-G⁺ cells (2.19% ± 3.84% and 0.66% ± 0.93%, respectively, p < 0.0001) and levels of protein expression (MFI of HLA-G, 25.95% ± 10.82% and 26.23% ± 18.96%, respectively, p < 0.0001) ( Figure 2B and Supplementary Figure S2A ). The frequency of HLA-G⁺ DC-10 was also significantly higher compared with the percentage of HLA-G⁺CD11c⁺ (0.30% ± 0.37%, p < 0.0001) and of HLA-G⁺cDC1 cells (0.78% ± 1.17%, p = 0.0004) ( Figure 2C and Supplementary Figure S2B ). Moreover, HLA-G MFI on DC-10 was significantly higher compared with that on CD11c⁺ cells (32.55% ± 47.25%, p = 0.0015), but similar to that detected on cDC1 (31.84% ± 19.08%, p = 1.0000) ( Figure 2C and Supplementary Figure S2B ).

Altogether, these results indicate that DC-10 contain a higher percentage of HLA-G-expressing cells among the myeloid cell subsets analyzed.

Hypothesizing that DC-10 frequency could be altered in diseases characterized by breakage of tolerance, we investigated possible variations of DC-10 frequency in subjects with different stages of T1D. Glucose variability in long-standing T1D patients can affect myeloid and stem and progenitor cell compartments (25, 26); thus, we performed the analysis in pediatric new-onset T1D patients, in comparison with age-matched FDRs of T1D patients without or with autoantibodies (Ab^neg and Ab^pos, respectively), and age-matched HCs ( Table 1 ). Circulating DC-10 were significantly lower as frequency and as absolute cell count in T1D patients (0.20% ± 0.13%, p = 0.0176; and 12.20% ± 9.44 cells/ml, p = 0.0005, respectively), Ab^pos FDRs (0.18% ± 0.10%, p = 0.0044; and 10.53 ± 6.96 cells/ml, p = 0.0002, respectively), and Ab^neg FDRs (0.24% ± 0.19%, p = 0.0499; and 18.25 ± 18.87 cells/ml, p = 0.0118, respectively) compared with HCs (0.31% ± 0.15%, and 22.02 ± 12.22 cells/ml) ( Figure 3A and Supplementary Table S2 ). The proportion, but not the total cell number, of classical cDC2 was significantly higher in T1D patients compared with HCs (0.32% ± 0.13% and 0.22% ± 0.07%, respectively; p = 0.0025) ( Figure 3B and Supplementary Table S2 ). Notably, the DC-10/cDC2 absolute cell number ratio was significantly higher in HCs (1.31 ± 0.92) compared with Ab^neg FDRs (0.89 ± 1.59, p < 0.0001), Ab^pos FDRs (0.62 ± 0.45, p = 0.0017), and T1D patients (0.73 ± 0.60, p = 0.0005) ( Figure 3C and Supplementary Table S2 ). No differences in the frequency of classical cDC1 and pDC were observed among the different cohorts ( Supplementary Figure S3A and Supplementary Table S2 ). According to previous data (27, 28), the frequency of CD11c⁺HLA-DR⁺ myeloid DCs was reduced in T1D patients (36.60% ± 10.15%) compared with that present in HCs (47.06% ± 13.45%, p = 0.0107) and in Ab^neg FDRs (46.86% ± 14.00%, p = 0.0424) but was similar to that observed in Ab^pos FDRs (38.13% ± 14%) ( Supplementary Figure S3B and Supplementary Table S2 ). Finally, in T1D patients, and in Ab^neg FDRs and Ab^pos FDR subjects, the frequency of the different monocyte subsets was comparable with that of HCs, except for classical monocytes being significantly higher in Ab^neg FDRs compared with HCs (3.57% ± 1.45% and 2.42% ± 1.65%, respectively; p = 0.0008) ( Supplementary Figure S3C and Supplementary Table S2 ).

Overall, these findings indicate a specific reduction in the tolerogenic DC-10 compartment that is parallel to the increase of the pro-inflammatory cDC2 in T1D patients, Ab^neg FDRs, and Ab^pos FDRs.

We further characterized circulating DC-10 focusing on the expression of HLA-G and of ILT4, tolerogenic molecules associated with DC-10 regulatory activity (13, 14), and the activation marker CD83. A higher percentage of HLA-G⁺ DC-10 was identified in Ab^neg FDRs (2.96% ± 3.28%, p = 0.0145) and in T1D patients (3.24% ± 3.51%, p = 0.0029) and, as a tendency, in Ab^pos FDRs (2.56% ± 2.23%, p = 0.0593) compared with HCs (1.11% ± 0.78%) ( Figure 4A and Supplementary Table S2 ). Notably, in Ab^neg FDRs, DC-10 expressed significantly higher levels of HLA-G (MFI of HLA-G 37.74% ± 14.16) than did HCs (MFI of HLA-G, 21.89 ± 6.41, p < 0.0001) and T1D patients (MFI of HLA-G 30.07 ± 17.90, p = 0.0005) ( Figure 4A and Supplementary Table S2 ). Moreover, despite the high variability, Ab^neg FDRs and HCs showed on average the highest percentage of CD83 ⁺ DC-10, which was significantly higher compared with that of T1D patients (42.17% ± 19.16% and 48.19% ± 20.59% vs. 26.61% ± 15.43%, p = 0.0063 and p < 0.0001, respectively) ( Figure 4B , Supplementary Figure S4A and Supplementary Table S2 ). The percentage of CD83⁺CD11c⁺ and of CD83⁺ cDC1 cells was comparable among the groups analyzed ( Supplementary Figure S5 and Supplementary Table S2 ). Interestingly, the percentage of HLA-G⁺ DC-10 correlated with the percentage of CD83⁺ DC-10 in Ab^neg FDRs (p = 0.0073). The dependency between the percentage of HLA-G⁺ DC-10 and the percentage of CD83⁺ DC-10 was also detected in HCs (p = 0.0069) but not in Ab^pos FDRs or T1D patients (subjects with autoimmunity) ( Figure 4C ). No differences in the proportion of ILT4 ⁺ DC-10 were detected among the different cohorts ( Supplementary Figures S4B, C and Supplementary Table S2 ).

These findings indicate that in T1D patients and Ab^neg and Ab^pos FDRs, despite the lower proportion, DC-10 expressed HLA-G at higher levels compared with DC-10 from HCs. Moreover, in subjects with autoimmunity (e.g., T1D patients and Ab^pos FDRs), DC-10, but not other myeloid cell subsets, are less activated compared with those present in subjects without autoimmunity (e.g., Ab^neg FDRs and HCs).

Our data indicate that in T1D patients at onset and in subjects at risk to develop the disease (FDRs), DC-10 number and phenotype are different from those in HCs. Moreover, in Ab^neg FDRs, despite that DC-10 are present at lower frequency compared with those in HCs, these cells expressed high levels of HLA-G ( Figure 4A ). An association between specific HLA-G genotypes/haplotypes and T1D has been previously demonstrated (10–12); thus, we analyzed nine specific variations of the 3′UTR HLA-G locus and inferred haplotypes and genotypes in the different cohorts of subjects ( Supplementary Tables S3 and S4 ).

To further define differences between Ab^neg FDRs and HCs, we performed multiple regression analysis considering the following variables: age of the persons, % of DC-10, % of cDC2, % of HLA-G⁺DC-10, % of CD83⁺DC-10, and 3′UTR HLA-G genotype (considering the comparisons InsG/InsG, DelC/InsG, and UTR-3/X genotypes with respect to the DelC/DelC genotype) ( Table 3 ). Multiple regression analysis identified all variables investigated to significantly predict the probability of being Ab^neg FDRs, with the only exception of the DelC/InsG vs. DelC/DelC comparison ( Table 3 ), indicating that overall Ab^neg FDR subjects differ from HCs with respect to these parameters. These findings confirm and further support previous evidence that Ab^neg FDRs cannot be considered healthy, as they already have signs of metabolic and immunological alterations (29–31).

To define whether the same set of variables that distinguish Ab^neg FDRs from HCs can predict the probability to be an individual at risk without (Ab^neg) or with autoimmunity (Ab^pos) or a T1D patient, considering the groups as ordered, we employed a cumulative link mixed model. The analysis showed that only the percentage of CD83⁺ DC-10 can be considered a significant risk factor to develop the disease (coefficient of the model = −0.0881, p = 0.0373). Indeed, a progressive decrease of CD83⁺ DC-10 percentage is associated with disease progression (e.g., from Ab^neg FDRs, to Ab^pos FDRs, to overt disease). The observation that none of the other variables, including the % of DC-10 or the % of HLA-G⁺ DC-10, were retained in the final cumulative link mixed model while they were retained in the previous final model predicting Ab^neg FDRs vs. HCs indicates that Ab^neg FDRs share several similarities with Ab^pos FDRs and with T1D patients, while they differ from HCs with respect to these variables.

The observations that specific 3′UTR HLA-G genotypes can predict the probability of being Ab^neg FDRs, and that Ab^neg FDRs are more similar to subjects with autoimmunity, prompted us to perform an additional HLA-G genetic analysis to explore if HLA-G genetic variations represent risk factors for developing T1D. To this aim, we enlarged the analysis by including adult subjects ( Table 2 ), and by genotyping both the 3′UTR and promoter (PROMO) regions of HLA-G. The frequency of UTR-3 allele was significantly higher in Ab^neg FDRs (20.1%) compared with HCs (8.3%, p = 0.0283; Figures 5A, B , Supplementary Table S5 ) and comparable with the frequency assessed in Ab^pos FDRs (10.6%, p = 0.2603) and in T1D patients (11.6%, p = 0.5331). We also analyzed 28 different single-nucleotide polymorphisms in the PROMO region of HLA-G gene ( Supplementary Table S6 ). We inferred almost a hundred different haplotypes, of which 16 was already described, six with a frequency >0.5%, and 78 with a frequency <0.5%, which clustered in four major PROMO families (data not shown). Among them, we observed a significantly higher frequency of the PROMO-0104 allele in Ab^neg FDRs compared with HCs (17.5% vs. 7.1%, p = 0.0390) ( Supplementary Figure S6 and Supplementary Table S7 ). These results, in conjunction with the high frequency of UTR-3 in Ab^neg FDRs, confirm previous data on the linkage disequilibrium between the 3′UTR-3 and the PROMO-0104 alleles (9). However, probably the smaller sample size of Ab^pos FDRs and T1D groups did not allow to define the protective role of PROMO-0104 alleles versus disease progression, although it was present with a lower percentage in these two groups with respect to Ab^neg FDRs (for Ab^pos FDRs: 6.7%, p = 0.0664; for T1D: 8.1%, p = 0.2073).