To determine whether autoreactive responses were modulated after tolDC treatment, we evaluated ex vivo T-cell proliferation and cytokine production upon antigen stimulation. Seven out of nine patients showed changes in islet autoimmunity upon injection with C19-A3 pulsed tolDC ( Figure 1 ). Patients were coded according to the received tolDC dosage per injection and order of administration (p1-1, p1-2 and p1-3 with 0.5x10⁷ cells; p2-1, p2-2 and p2-3 with 1x10⁷ cells; and p3-1, p3-2 and p3-3 with 2x10⁷ cells). T-cell autoimmunity to the vaccine peptide pre-existed before tolDC injection in three out of nine patients (p1-2, p1-3 and p3-3). Strikingly, all three responded to the tolDC therapy by reducing the proliferation (lymphocyte stimulation test (LST), Figure 1B ) and cytokine production (ELISPOT, Figure 1A ) to loss of responsiveness (cases p1-2 and p1-3) or IL-10 production alone (case p3-3) at the end of the 6-month study period. Of the six patients not reacting to the vaccine peptide before tolDC therapy, four showed a transient change in cytokine production (p1-1, p2-1, p2-3 and p3-2), with one patient (p2-3) attaining a moderate IL-10 producing phenotype at 6-month post-tolDC.

Next to the vaccine peptide, we measured autoreactive responses to preproinsulin protein (PPI) containing the vaccine peptide, and to other beta-cell autoantigens (GAD65 and IA-2; Figure 2A ; Figure S1 ), and enumerated beta-cell autoreactive CD8⁺ T cells ( Figure 2B ; Figure S2 ) and auto-antibody titers to GAD, IA-2 and ZnT8 ( Figure 2C ). Before tolDC administration, inter-individual differences in proliferative response to autoantigens were noted in the frequencies of autoreactive PPI-specific CD8⁺ T-cells and autoantibody titers ( Figure 2 ). Patients p1-2 p1-3 and p3-3 showed strong proliferative responses to PPI antigen (SI>10), while patients p1-1, p2-3 and p3-3 had PPI-peptide specific CD8⁺ T-cells before tolDC. Following tolDC administration, all patients presented with reduced values in both proliferative autoimmune responses and frequencies of autoreactive CD8⁺ T-cells, while autoantibody titers remained unaffected ( Figure 2 ). The responses to IA-2 and GAD65 prior to therapy did not change significantly by 6 months after tolDC therapy ( Figure S1 ) in 8 out of 9 patients, whereas patient p1-1 showed minimal proliferation to control antigen (tetanus toxoid; TetTox) and beta-cell autoantigens prior to tolDC injection, which could reflect a refractory period of lymphocytes shortly after leukapheresis. Indeed, this patient subsequently showed increasing proliferation to all tested antigens including TetTox control. Of note, CD8 ⁺ autoreactivity and autoantibody titers in this patient were similar before leukapheresis and after tolDC injection.

To investigate long-term effects of the tolDC therapy, patients were invited to participate in a single re-call visit (17-32 months after receiving the tolDC injections), at which point the responses to the vaccination peptide, PPI antigen and diabetes control were again assessed in 8 patients available. Cytokine responses and proliferation to the vaccine peptide remained low or undetectable more than two years after the tolDC injection in all eight patients ( Figures 3A, B ). Moreover, IFNg production to PPI was long-term decreased in all patients as compared to the levels before tolDC injection (p=0.0016; Figure 3C ). IL-10 showed more individual variation with a decrease in most patients except for p1-3 and p2-1. The strong proliferation to PPI (SI > 10) before tolDC in the case of p1-2, p1-3 and p3-3 ( Figure 3D ), decreased by 6 months after tolDC and persistently remained lower than before therapy. T-cell proliferation to other beta-cell antigens was also lower on the long term than at start in most patients (not shown).

Regarding glycemic control, all participants had long standing but well controlled T1D with HbA1c < 8% at start (range 6.4 – 7.6%), which was maintained after tolDC therapy and showed a clear decrease on the long term by an average of 0.34% (range 6.2 – 7.3%, p=0.0029, ANOVA), in spite of constant insulin needs ( Figures 3E, F ). Intriguingly, the HbA1c decline in some cases (p1-1, p2-1, p3-2, p3-3) resulted in personal ‘all-time low’ values long term, which was not always paired with detectable maintained endogenous insulin production (p2-2, p2-3, p3-2) (24).

To assess the general immune competence after tolDC treatment, total white blood cell counts and frequencies of leukocyte subsets measured in full blood by flow cytometry showed no systematic changes in total cell number or the frequency of leukocytes (B-cells, monocytes, DCs, NK cells, NKT cells, CD4 ⁺ and CD8 ⁺ T-cells; Figure 4 ; Figure S3 ). Further subdivision into subsets of B cells, monocytes, DC and NK cells did not show significant changes post tolDC treatment either (not shown). Antigen-specific immune competence unrelated to the tolDC-vaccine peptide or general islet autoimmunity (i.e. proliferation to TetTox and virus-specific CD8⁺ T-cell count) did not change for patients either in the placebo period or post tolDC treatment ( Figures S1 and 2 , respectively).

Peep-diving into more subtle changes in minor subsets of immune cells, frozen PBMC sampled at all monitoring timepoints were stained simultaneously with a panel consisting of 39 lanthanide-conjugated antibodies ( Table S1 ) and analyzed using Hierarchical Stochastic Neighbor Embedding (HSNE). Confirming the results obtained from fresh blood, no significant changes were observed in major leukocyte subsets or total CD4 ⁺ , CD8 ⁺ and TCRγδ ⁺ T-cell populations prior to or after tolDC treatment ( Figure S4 ). Subsequent unsupervised HSNE analysis of the total CD4 ⁺ T-cell population on the first level (level 1) allowed separation into seven major CD4-clusters based on the landmarks in the HSNE map ( Figure 5A ), with designations based on differential phenotypes ( Figure 5B ). While the cumulated frequencies of the clusters were similar between patients ( Figure 5C ), three out of seven clusters changed frequencies after tolDC administration compared to baseline (cluster 1; p=0.012, cluster 4; p=0.007 and cluster 7; p=0.010, ANOVA). Cluster 1 featuring a phenotype of tissue resident memory T cells (CD103⁺ Trm) showed an average increase of 0.13% (p=0.009) 1 month after the first tolDC injection and an increase of 0.10% 1 month after the second tolDC injection (p=0.047), compared to baseline value (0.56% of total CD4 ⁺ T cells), after which the frequencies normalized to baseline ( Figure 5D , top graph). Cluster 4, consisting of naïve CD4 ⁺ T-cells (CD45RA⁺CCR7⁺) dropped from baseline (46.2% of total CD4 ⁺ ) by 4.5% at 1 month after the first tolDC injection (p=0.012). After the second tolDC injection, the frequencies of cluster 4 returned to baseline ( Figure 5D middle graph). Cluster 7, comprised of effector CD4 ⁺ T-cells with a mixed central and effector memory (CM/EM) phenotype (CD45RA^-CCR7^dim), showed an increase in most patients with an average of 3.9% at 1 month after the first tolDC injection, which declined after the second tolDC injection in many patients below the baseline level (40.5% of total CD4 ⁺ ) and remained low during the follow-up of 6 months after tolDC injection ( Figure 5D ).

To further explore the tolDC-induced changes, CD4 ⁺ T-cell clusters 1, 4, 6 and 7 were analyzed in subclusters (level 2) ( Figure 5E ). Within the Trm cluster 1, one subcluster 1.3 demonstrated significant change (p=0.0003), though all Trm subclusters showed a peak in frequency after the first tolDC injection ( Figure S5A ). Within the CM/EM cluster 7, subclusters 7.1-7.4 shared expression of CCR6 and showed significant changes after tolDC injection (p=0.0003, p=0.003, p=0.0006 and p=0.001, respectively; Figure 5E and Figure S5B ). Cluster 6 consisted of cells with a CD25^highCD127^- phenotype and thus likely included both thymic derived Tregs and peripherally induced Tregs (26, 27). Although this Treg cluster showed no change in time overall, subdivision into four Treg subclusters (subcluster 6.1-4; Figure 5E ) on basis of phenotype and pseudo-time analysis revealed changes corresponding with tolDC vaccination ( Figure 6 ). These subclusters likely represent naïve Tregs expressing CD45RA (subcluster 6-1 and 6-2), an intermediate cluster previously described as Fraction III (28) and activated effector Tregs (eTregs; subcluster 6-4) expressing TIGIT, ICOS, CD39 and CCR4 ( Figures 6A, B and Figure S6 ). Pseudo-time analysis showed a shift from naïve and Fraction III Tregs towards eTregs (subcluster 6-4) starting 1 month after the first tolDC injection with 2.7% eTregs at baseline increasing 0.5% on average by (p=0.035) at cost of the other Treg subclusters ( Figure 6C ).

A similar HSNE analysis of the CD8 ⁺ T-cell compartment revealed eight major clusters ( Figure S7 ). A significant change was detected in cluster 7 (p=0.042), representing CD8 ⁺ T-cells with a memory phenotype (CD45RA^-CCR7^dim) expressing CD25^dim and the skin-tropic receptor CCR4 ( Figure S7D ). One month after the first tolDC injection, a 0.8% increase of CD8 ⁺ was observed (p=0.0029), which normalized to baseline values (4.5% of total CD8 ⁺ ) after the second tolDC injection. Within cluster 7, subcluster 7.3 expressing KLRG1 (associates with T-cell senescence) increased significantly after tolDC injection (p=0.022) (29). Subclusters 7.1, 7.4 and 7.7 sharing low CD25 expression but lacking KLRG1 also showed a peak in frequency 1 month after the first injection ( Figure S7F ).

To study for relationships between clinical outcome with immune responses and T-cell subsets that associated significantly with tolerogenic vaccination, multi-variable correlations between treatment dose, HbA1c, phenotypic and immunological parameters were performed on all available data using Kendall’s rank correlation coefficient ( Figure 7 ).

Before therapy, proliferative responses to PPI and vaccine peptide PPI_C19-A3 correlated with cytokine responses to these autoantigens, as well as proliferative responses to other islet autoantigens (GAD65 and IA-2), but not with immune response to TetTox ( Figure 7 ). Treg cluster 6.4 was inverse correlated with IFNg production in response to PPI as expected if this Treg subset suppresses autoimmunity. Likewise, the CD4 T effector clusters 1 (Trm) and 7 (CM/EM) inversely correlated with the production of cytokine IL-10 in response to PPI.

Three months after the first vaccination, correlations between cytokine responses to vaccine-specific and other autoantigens disappeared, only leaving correlations between proliferative response to PPI with IA-2 and GAD65. Instead, inverse correlation appeared between proliferation to the vaccine peptide and CD4 effector T-cell cluster 7, and to a lesser extent with HbA1c. The correlations further changed at 6 months when the correlation of effector T-cell subsets with proliferation to the PPI-vaccine peptide at 3 months transited into an inverse correlation with IL-10 production to PPI protein, supporting the induction of PPI-specific immune regulation following immunization with the PPI peptide. CD4 effector T-cell cluster 7 correlated inversely with autoimmunity to IA-2, while higher IL-10 response to the vaccine peptide associated with lower (i.e., improved) HbA1c values.

At time of the revisit (17-32 months after first vaccination and 11-26 months after completion of the clinical trial), the changes in correlations between immune responses to autoantigens compared to baseline persisted, with additional high IL-10 responses to PPI now correlating with reduced frequencies of CD8 T-cells against PPI.

The clinical study protocol was approved by the Dutch Central Committee on Research Involving Human Subjects and the Medical Ethical Committee of Leiden University Medical Center (LUMC; Leiden, The Netherlands), EudraCT number 2013-005476-18 (accessible through the national trial registry). Patients were selected for screening by their own physician at Diabeter clinic (The Netherlands). Nine HLA-DR4 positive patients eligible for the study and willing to participate were allocated at random to a dose treatment group: 0.5x10⁷ (p1-1, p1-2 and p1-3), 1x10⁷ (p2-1, p2-2 and p2-3) or 2x10⁷ (p3-1, p3-2 and p3-3) tolDCs per injection round (24). A randomly selected patient in each dose-cohort first received 5, 10 or 20 intradermal injections of saline (the tolDC vehicle), respectively. In this ‘placebo’ period, participants were monitored for three months (12 weeks), after which they received tolDC injections and were monitored as a third participant in their respective dose-group.

Detailed description of clinical tolDC quality assessment has been reported (24). TolDCs were administered twice (prime/boost) by intradermal injection in the upper-left abdominal quadrant, using a MicronJet600 microneedle (NanoPass Technologies Ltd, Nes Ziona, Israel), at the Clinical Cellular Research Unit of the Hemapheresis Unit (LUMC) (24).

Antigen-specific immunomonitoring was performed using techniques previously validated for clinical trials: fresh PBMCs were cultured with GMP-grade C19-A3 peptide, or PPI protein to detect antigen-specific IFN-g and IL-10 producing cells using ELISPOT assay (18); antigen-specific proliferation was assessed in the lymphocyte stimulation test (LST) (38, 39). In short, fresh PBMC were incubated with GMP-grade C19-A3 peptide, recombinant human proteins PPI, GAD65 and IA-2 (all at final conc. 10 µg/mL), synthesized at the Protein Facility (LUMC). For control, cells were incubated with medium alone (med), tetanus toxoid (1%, 7.23 Lf/mL; Statens Serum Institut) or recombinant IL-2 (Proleukin^®, 35 U/mL; Novartis, Bazel, Switzerland). Stimulation index (SI) is calculated as cpm in the presence of antigen/stimulus divided by cpm with medium alone. To quantify PPI-specific CD8⁺ autoreactive T-cells, the Q-dot assay was performed using frozen PBMC as described and validated previously (7, 40). PBMC subsets were measured using whole blood flow cytometry. Cryopreserved PBMC were used for the CyTOF analysis. Samples were stained with a panel consisting of 39 metal-conjugated antibodies ( Table S1 ) as described previously (41). To minimize inter-assay variability, samples of different timepoints from the same patient were stained and acquired on the same day and a reference sample was included as quality control. The reference sample consisted of 1,5% PHA stimulated PBMC from a healthy donor. To check the consistency between staining and measurement days, reference samples were analyzed in a tSNE using Cytosplore and Jensen-Shannon plots were generated in Matlab (version R2016a) ( Figure S8C ).

Proliferation in LST was tested using 2-way ANOVA and Dunnet’s multiple comparison test. Changes in ELISPOT were tested using Wilcoxons matched-pairs signed rank test. Flow cytometric analyses were performed using FlowJo software (Ashland, OR, USA). Principal component analysis was performed using R (package 3.5.2, ‘ggplot2’). Graphs and statistical calculations were performed using GraphPad software (San Diego, CA, USA). Values at p<0.05 after multiple testing correction were deemed significant.

Data obtained by CyTOF were analyzed as follows: Beads were excluded from the dataset and live, CD45⁺ single cells were selected with DNA stains and Gaussian parameters (width and residual) ( Figure S8A ) in FlowJo. Expression values were arcsine transformed and Hierarchical Stochastic Neighbor Embedding (HSNE) implemented in Cytosplore (version 2.2.1) (42) was used for dimensionality reduction analysis. The dataset was explored in different levels of hierarchies as described in the results section. Subsequent clusters were generated using the Gaussian-mean-shift method ( Figure S8B ). FCS files from generated clusters were exported and loaded in R (version 3.6.2) using the “Cytofast” package (43) for further downstream analysis. A mixed linear model was used to evaluate changes in cell frequency of specific clusters after tolDC treatment. Cell frequencies were compared with the baseline frequency at timepoint 0, before the first tolDC injection. The cell frequency of the cluster is selected as dependent variable, timepoints after tolDC injections are the fixed effects and the random effects are the different patients. An ANOVA test was used to assess whether overall changes are observed in a cluster after tolDC treatment and p-values below 0.05 were considered statistically significant. Estimates of fixed effects from the mixed linear model are reported in the results as change in percentage of CD4⁺ or CD8⁺ T-cells. Statistical tests were performed using R package “lmerTest”. Diffusion maps were generated in the OMIQ Data Science Platform using the Wanderlust function.