CO-STIMULATION BLOCKADE DISRUPTS CD4⁺ T-CELL MEMORY PATHWAYS AND UNCOUPLES THEIR LINK TO DECLINE IN β-CELL FUNCTION IN TYPE 1 DIABETES

Abstract

We previously reported that co-stimulation blockade by abatacept limits the decline of β-cell function and the frequency of circulating CD4⁺ central memory (CD45RO⁺CD62L⁺) T (T_CM) cells in new-onset type 1 diabetes. In human subjects receiving placebo we found a significant association between increase in CD4⁺ T_CM cells and decline of β-cell function. To extend and refine these findings, we examined changes in human CD4⁺ and CD8⁺ naïve and memory T-cell subsets at greater resolution using polychromatic flow and mass cytometry. In the placebo group we successfully reproduced the original finding of a significant association between T_CM and β-cell function, and extended this to other T-cell subsets. Furthermore, we show that abatacept treatment significantly alters the frequencies of a majority of CD4⁺ conventional and regulatory T-cell subsets; in general, antigen-naïve subsets increase and antigen-experienced decrease, while CD8⁺ T-cell subsets are relatively resistant to drug effects, indicating a lesser reliance on CD28-mediated co-stimulation. Importantly, abatacept uncouples the relationship between changes in T-cell subsets and β-cell function that is a component of the natural history of the disease. Although these data suggest immunological markers for predicting change in β-cell function in type 1 diabetes, the finding that abatacept blunts this relationship renders the biomarkers non-predictive for this type of therapy. In sum, our findings point to a novel mechanism of action for this successful immunotherapy that may guide other disease-modifying approaches for type 1 diabetes.

Introduction

Type 1 diabetes is an autoimmune disease characterised by gradual loss of pancreatic islet β-cell function/mass mediated by immune cells, especially autoreactive CD4⁺ and CD8⁺ T cells (1, 2).

During a T-cell mediated immune response, naïve (antigen-inexperienced) T cells encounter their cognate target and under conditions in which co-stimulatory signals are provided (e.g. via CD80/CD86 interaction with CD28) become activated, antigen-experienced cells. There follows a sequential differentiation process that may involve further exposure to autoantigens, resulting in T-cell subsets with memory properties (rapid response to the same antigen), effector phenotypes (e.g. cytokine secretion, tissue homing properties) and distinct functional characteristics and longevity (3, 4).

In particular, the critical role of the co-stimulation process in T-cell activation and acquisition of memory/effector function led to the development of therapeutic strategies aimed at blocking key molecular interactions (5). The immunomodulatory drug cytotoxic T-lymphocyte-associated protein 4 (CTLA-4)Ig (abatacept) prevents full activation of subsets of T cells requiring co-stimulation, by blocking the binding of CD80/CD86 expressed on antigen-presenting cells with the co-stimulatory molecule CD28 on the T cell (6).

In a clinical trial (“Effects of CTLA-4 Ig (abatacept) on the Progression of Type 1 Diabetes in New Onset Subjects (TN-09)”, (NCT00505375)) in patients with new-onset type 1 diabetes, abatacept treatment for two years significantly reduced the rate of loss of β-cell function, an effect sustained one year after completion of treatment, indicating that co-stimulation dependent immune pathways play a role in disease progression after diagnosis (7, 8). In the same study, we reported that in the placebo arm a significant association existed between an increase in the frequency of CD4⁺ central memory T (T_CM) cells (CD45RO⁺CD62L⁺) from baseline and loss of β-cell function (9). Importantly, during abatacept treatment this relationship was no longer observed. These findings suggested that co-stimulation blockade has the effect of limiting CD4⁺ T-cell differentiation pathways that are a key pathobiological component of disease progression. These original findings were made using a fresh whole blood flow cytometry analysis with a limited panel of surface phenotypic markers (n=5), including those (e.g. CD62L) that are difficult to stain on cryopreserved cells. The objective of the current study is to characterise, in-depth, the effect of co-stimulation modulation, by abatacept, on T-cell subsets to expand and refine the previous finding of a link between immunological change and metabolic change. This may be important in the interpretation of outcomes of ongoing clinical trials such as the TrialNet study TN-18 examining the effects of abatacept in delaying disease progression in high-risk subjects with multiple autoantibodies (NCT01773707).

Materials and Methods

Samples

Cryopreserved peripheral blood mononuclear cells (PBMC) were provided by Type 1 Diabetes TrialNet from the clinical trial NCT00505375 (7). Samples from 40 abatacept- and 19 placebo-treated individuals were analysed at baseline, year 1 and year 2. Samples were provided blinded to treatment and time of sampling; each subject’s samples from the three collection times were stained and analysed on the same day. TrialNet Coordinating Centre provided clinical and demographic metadata following submission of raw data files. PBMCs were thawed at 37°C, washed in pre-warmed basal media (RPMI1640 Glutamax with HEPES, containing 100U/mL penicillin and 100μg/mL streptomycin (all Gibco)) containing 5% human heat-inactivated AB serum (Gibco), 20mM tumour necrosis factor (TNF)-α Protease Inhibitor-2 (TAPI-2) (Calbiochem) and 50U/mL Benzonase (Millipore). Cells were rested for 1 hour at 37°C/5% CO₂ in basal media containing 10% human AB serum and 20mM TAPI-2 before washing in 1X basal media and 1X DPBS (Gibco) and staining of 2×10⁶ PBMC each for flow and mass cytometry.

Sample staining and acquisition

For flow cytometry PBMCs were stained with LIVE/DEAD™ Fixable Aqua Dead Cell Stain Kit (Invitrogen) and subsequently with antibodies against CD197 (clone: G043H7, BV421, Biolegend) for 10 minutes at 37°C/5% CO₂ followed by staining for CD14 (clone: M5E2, V500, BD), CD16 (clone: 3G8, V500, BD), CD19 (clone: HIB 19, V500, BD), CD3 (clone: SK7, APC-H7, BD), CD4 (clone: OKT4, BV785, BL), CD8 (clone: RPA-T8, FITC, BL), CD27 (clone: O323, BV605, BL), CD45RA (clone: HI100, APC, BD), CD95 (clone: DX2, PE, BL), CD45R0 (clone: UCHL1, PerCPCy5.5, BL), CD62L (clone: DREG-55, PE-Cy7, BL) and CD31 (clone: WM59, PE-CF594, BD) for 25 minutes on ice. A BD LSRFortessa set up with CS&T Beads (BD Biosciences) and application settings was used for acquisition. Where available conjugated CyTOF-ready antibodies were used; unconjugated Maxpar-ready antibodies were conjugated in-house using the Maxpar antibody labelling kit (Fluidigm). PBMCs were stained in 1X Rh-Intercalator (Fluidigm) for 15 minutes at room temperature (RT), before blocking FcR with TruStain FcX (Biolegend) for 10 minutes at RT. Cells were stained with antibodies against CD196 [CCR6] (clone: G034E3, 141Pr, Fluidigm), CD11a (clone: HI111, 142Nd, Fluidigm), CD45RA (clone: HI100, 143Tm, Fluidigm), CD195 [CCR5] (clone: NP-6G4, 144Nd, Fluidigm), CD4 (clone: RPA-T4, 145Nd, Fluidigm), IgD (clone: IA6-2, 147Sm, BD), CD14 (clone: RMO52, 148Nd, Fluidigm), CD16 (clone: 3G8, 148Nd, Fluidigm), CD25 (clone: 2A3, 149Sm, Fluidigm; clone: M-A251, Biolegend), CD31 (clone: WM59, 150Nd, eBioscience), CD49d (clone: 9F10, 151Eu, eBioscience), CD62L (clone: DREG-56, 153Eu, Fluidigm), CD45 (clone: HI30, 154Sm, Fluidigm), CD183[CXCR3] (clone: G025H7, 156Gd, Fluidigm), CD194 [CCR4] (clone: 205410, 158Gd, Fluidigm), CD197 (CCR7) (clone: G043H7, 159Tb, Fluidigm), CD28 (clone: CD28.2, 160Gd, Fluidigm), CD39 (clone: eBioA1, 161Dy, eBiosience), CD69 (clone: FN50, 162Dy, Fluidigm), CD161 (clone: HP-3G10, 163Dy, Biolegend), CD95 [FAS] (clone: DX2, 164Dy, Fluidigm), CD45RO (clone: UCHL1, 165Ho, Fluidigm), CD44 (clone: BJ18, 166Er, Fluidigm), CD27 (clone: L128, 167Er, FLUIDIGM), CD8a (clone: RPA-T8, 168Er, Fluidigm), Integrin beta7 (clone: FIB504, 169Tm, BD), CD3 (clone: UCHT1), 170Er, FLUIDIGM), CD19 (clone: HIB19, 171Yb, Biolegend), CD57 (clone: HCD57, 172Yb, Fluidigm), CD38 (clone: HIT2, 173Yb, Biolegend), HLA-DR (clone: L243, 174Yb, Fluidigm), CD279/PD-1 (clone: EH12.2H7, 175Lu, Fluidigm) and CD127 (clone: A019D5, 176Yb, Fluidigm) for 30 minutes at RT. After washing cells were fixed in 2% paraformaldehyde (Electron Microscopy Sciences) overnight at 4°C. Cells were washed and stained with 1X lr-intercalator for 20 minutes at RT, washed extensively in Milli-Q water (Millipore) and analysed by CyTOF (Fluidigm). CyTOF data were normalised using EQ-Four-Element-Calibration Beads (Fluidigm). T-cell subsets in flow and mass cytometry were gated in FlowJo (version 10, FlowJo). T cell subsets were defined accordingly; naïve (T_NV), recent thymic emigrant (T_RTE), stem-cell-memory (T_SCM), central memory (T_CM), transitional memory (T_TM), effector memory (T_EM), late-effector (T_LE); by two markers: pre-T_NV/RTE/SCM # (CD45RO⁻/CD27⁺), pre-T_CM/TM # (CD45RO⁺/CD27⁺), pre-T_EM # (CD45RO⁺/CD27⁻), pre-T_LE # (CD45RO⁻/CD27⁻); by four markers: pre-T_NV/RTE #.# (CD45RO⁻/CD27⁺/CD95⁻/CCR7⁺), pre-T_SCM #.# (CD45RO⁻/CD27⁺/CD95⁺/CCR7⁺), pre-T_CM #.# (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁺), pre-T_TM #.# (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁻), pre-T_EM #.# (CD45RO⁺/CD27⁻/CD95⁺/CCR7⁻), pre-T_LE #.# (CD45RO⁻/CD27⁻/CD95⁺/CCR7⁻); by six markers: T_NV (CD45RO⁻/CD27⁺/CD95⁻/CCR7⁺/CD45RA⁺/CD31⁻), T_RTE (CD45RO⁻/CD27⁺/CD95⁻/CCR7⁺/CD45RA⁺/CD31⁺), T_SCM-CD31 (CD45RO⁻/CD27⁺/CD95⁺/CCR7⁺/CD45RA⁺/CD31⁻), T_SCM+CD31 (CD45RO⁻/CD27⁺/CD95⁺/CCR7⁺/CD45RA⁺/CD31⁺), T_CM-CD31 (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁺/CD45RA⁺/CD31⁻), T_CM+CD31 (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁺/CD45RA⁺/CD31⁺), T_TM-CD31 (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁻/CD45RA⁺/CD31⁻), T_TM+CD31 (CD45RO⁺/CD27⁺/CD95⁺/CCR7⁻ /CD45RA⁺ /CD31⁺), T_EM-CD31 (CD45RO⁺ /CD27⁻ /CD95⁺ /CCR7⁻/CD45RA⁺/CD31⁻), T_EM+CD31 (CD45RO⁺/CD27⁻/CD95⁺/CCR7⁻/CD45RA⁺/CD31⁺), T_LE-CD31 (CD45RO⁻/CD27⁻/CD95⁺ /CCR7⁻/CD45RA⁺/CD31⁻) T_LE+CD31 (CD45RO⁻/CD27⁻/CD95⁺/CCR7⁻/CD45RA⁺/CD31⁺). Unsupervised clustering algorithm FlowSOM is part of Cytobank Enterprise platform (Cytobank) (10). For FlowSOM analysis, which relies on consistent inter-sample staining profiles for clustering, we analysed a subcohort of 29 individuals (20 abatacept- and 9 placebo-treated) that exhibited consistent staining profile between individuals. FlowSOM settings used were: 50 meta-cluster derived from 256 clusters and 10 iterations.

Statistical analysis

All statistical analysis, except for mixed linear modelling, were performed in Prism (v8, GraphPad) or in R Studio using packages (all available via CRAN; cran.r-project.org) FactoMineR (v1.42) (11) and Factoextra (v1.0.5) (12) for PCA, ComplexHeatmap (v2.1.0) (13) and Circlize (v0.4.7) (14) for Heatmaps, and ggplot2 (v3.2.0) (15) for visualisation. Generalized linear mixed modelling was conducted with SAS (V.9.4, Cary, NC). Statistical significance in T-cell subset frequencies was calculated using Mann-Whitney U-test or Wilcoxon signed-rank test as indicated; adjustment for multiple comparison testing by Benjamini-Hochberg and considered significant at p<0.05. β-cell function and mass was measured with stimulated C-peptide from a mixed meal tolerance test (MMTT). The area under the stimulated C-peptide curve (AUC) over the 2-hour MMTT was computed using the trapezoidal rule and divided by the duration of the test to estimate the average C-peptide level during stimulation (“average C-peptide”). For analysis of the relationship between percent change from baseline in C-peptide and in T-cell subsets the log (natural) of the percents were analysed with mixed-effects general linear models (GLMs) having subject as a random effect and adjusting for baseline T-cell level. Treatment-related alterations in the slope of the relationship between T-cell subset change and C-peptide change were declared statistically significant whenever the type 3 p-value for testing differences between treatment-specific slopes was significant after False Discovery Rate (FDR) adjustment set to yield a 5% FDR. A False Discovery Rate of 10% was used to select T-cell subsets as having a significant difference between treatments. Parameter estimates from the GLM analysis were used to estimate the retention of C-peptide level per unit of change in T-cell subset. The 2 year change models also considered the “lagged” T-cell value (i.e. the value of the T cell at the first visit after baseline) versions of the T-cell variables. Pearson correlations between change from baseline in T-cell subsets at time points (year 1 and 2) were computed and considered statistically significant at P=0.01.

Study approval

Protocol and consent documents were approved by appropriate Independent Ethics Committees or Institutional Review Boards for participating TrialNet centres. All subjects or their parents gave written informed consent and assent as appropriate before study participation.

Data and Resource Availability

The datasets generated during the current study are available from the corresponding author upon reasonable request. Clinical data and patient characteristics can be requested from the NIDDK public repository at https://repository.niddk.nih.gov/studies/trialnet/.

Results

Study cohort

Flow and mass cytometry were performed on a randomly sampled, representative subset of the original study cohort comprising 59 individuals (placebo n=19, abatacept n=40) (7, 9). The year 2 abatacept-induced reduction in C-peptide loss was comparable to that observed in the original study cohort. Mean baseline-adjusted C-peptide level was 75% higher in the treatment arm of our study (0.323 nmol/L compared to 0.185 nmol/L for abatacept and placebo, respectively) compared to 59% higher in the original study (0.378 nmol/L and 0.238 nmol/L) (data not shown).

Defining T_CM cells in cryopreserved peripheral blood mononuclear cells

In the original study, flow cytometry was performed on fresh whole blood and CD4⁺ T_CM were defined as CD45RO⁺CD62L⁺. CD62L is known to be shed in cryopreserved samples. We added the metalloprotease tumour necrosis factor alpha protease inhibitor 2 (TAPI-2) during thawing and cell recovery to limit CD62L shedding (16) but find that prevention of CD62L shedding is inconsistent. This leads to a marked underrepresentation of CD4⁺ T_CM (CD45RO⁺CD62L⁺) in cryopreserved samples and a modest correlation with the data from fresh staining (Figure 1). However, we observed a stronger and highly significant correlation between CD4⁺ T_CM (CD45RO⁺CD62L⁺) in fresh staining and on cryopreserved samples using other memory markers (Figure 1), with CD45RO⁺CD27⁺ being close to equivalence (slope=1.1, R=0.82). Subsequently, therefore, CD27 was used as a surrogate for CD62L when defining central memory subsets.

Figure 1: Relationship between fresh and cryopreserved sample staining to identify T_CM.

The frequency of CD4⁺ T_CM (CD45RO+CD62L+) in fresh whole blood (on the x-axis) is compared to the frequency of CD4⁺ T_CM in the cryopreserved sample defined using the markers indicated. Frequency is calculated as proportion of total CD4⁺ T cells; each data point is a single subject at one timepoint. Linear regression analysis = black line; 95% CI = grey shaded area; equality = grey line.

Abatacept treatment decreases memory and increases naïve CD4 T-cell subsets

We next examined abatacept-induced changes in memory and naïve CD4⁺ T-cell subsets with this refined flow cytometry panel to replicate and extend our original findings (gating strategy in Supplemental Figure 1) (17). We identified 12 traditionally defined T-cell subsets by sequential quadrant gating and performed principal component analysis (PCA) on the change of T-cell subsets from baseline to year 1 and year 2 (defined as “% of baseline”; change in frequency of a specific T-cell subset, over time (i.e. from baseline to year 1 or year 2), represented as proportion of baseline frequency”).

PCA separate individuals by treatment arm at year 1 and year 2, indicating that changes to CD4⁺ T-cell subsets are induced by abatacept (Figure 2a, b). Dimension 1 (accounting for ~47.2% of the variance in year 1 and 43% in year 2) is mainly driven by changes to CD4⁺ T cells with phenotypes corresponding to naïve (T_NV), transitional memory (T_TM), effector memory (T_EM) and late-stage effector (T_LE) cells; whereas dimension 2 (accounting for ~19% in year 1/ 21% in year 2 of the variance) is driven by changes to CD4⁺ T cells with stem-cell-memory-like (T_SCM) and central memory (T_CM) phenotypes (Supplementary Figure 2a, b).

Figure 2: Change in CD4⁺ T-cell subsets in treatment groups over time.

Change in CD4⁺ T-cell subsets from baseline to year 1 and to year 2 in treatment groups is analysed. Principal component analysis of individuals from treatment groups are shown for baseline to year 1 (a) and to year to 2 (b). Change (as percentage from baseline) for all CD4⁺ T-cell subsets is shown (c). Mann-Whitney U-test with Benjamini-Hochberg correction; * p<0.05, ** p≤0.01, *** p≤0.001, **** p≤0.0001; The horizontal line in the middle of each box indicates the median; top and bottom borders mark the 75^th and 25^th percentiles, respectively; whiskers according to Tukey. See Material & Methods for definition of T-cell subsets. % of baseline; change in frequency of a specific T-cell subset, over time (i.e. from baseline to year 1 or year 2), represented as proportion of baseline frequency.

Within these subsets we observed a treatment-associated increase from baseline for antigen-inexperienced (i.e. CD45RO⁻CD95⁻) T_NV and recent thymic emigrant (T_RTE) and a decrease for several antigen-experienced (CD95⁺) T-cell subsets (i.e. T_SCM, T_CM, T_TM, T_EM, T_LE,) (Figure 2c). For the majority of subsets the change from baseline in the treatment group is significantly different from the placebo group, including after correcting for multiple comparison testing at both timepoints (T_NV, T_RTE, T_CM, T_TM; for CD31^+/−) or year 2 only (T_EM, CD31⁻ only); some subsets (i.e. T_SCM, T_LE; for CD31^+/−) are not significantly different between the treatment groups at either timepoint. We also noted that, in addition to changes in T-cell subset frequency, abatacept treatment is associated with reduced cell surface expression of markers within the first hierarchy gate (CD45RO/CD27). We observed a significant reduction in the median fluorescence intensity (MFI) of CD45RO and CD95 in the pre-CD4⁺ T_CM/EM (CD45RO⁺CD27⁺) T-cell subset (Figure 3).

Figure 3: Reduction in expression levels of phenotypic markers during treatment.

The change of expression level (as median fluorescence intensity) of 6 phenotypic markers was assessed on pre-T-cell subsets (#; defined by CD45RO and CD27) from baseline to year 1 and to year 2. Mann-Whitney U-test with Benjamini-Hochberg correction; * p<0.05, ** p≤0.01; The horizontal line in the middle of each box indicates the median; top and bottom borders mark the 75th and 25th percentiles, respectively; whiskers according to Tukey; see Material & Methods for definition of T-cell subsets.

In contrast, treatment effects of abatacept on CD8⁺ T-cell subsets were minor. Individuals from treatment groups do not separate in PCA and a significantly different change in frequency between treatment groups was found only in T_EM and T_LE at baseline to year 2 (Supplementary Figure 3).

Treatment with abatacept uncouples the correlation between changes in T-cell subsets and C-peptide

In the original mechanistic study we reported that, as part of the natural history of type 1 diabetes post-diagnosis, a relationship between changes in naïve/memory T-cell subsets and changes in β-cell function exists. Specifically, in the placebo arm, the decline from baseline value of average secreted C-peptide during a mixed meal tolerance test (MMTT) was correlated with an increase from baseline in CD4⁺ T_CM (CD45RO⁺CD62L⁺) when measured at a preceding timepoint. Here, we extend this analysis to other T-cell subsets and identify, within the placebo arm, several immune changes that correlate with C-peptide change over time (Figure 4). As examples, higher rates of C-peptide retention from baseline to year 1 are associated with an increased frequency of CD4⁺ pre-T_NV/SCM (CD45RO⁻CD27⁺) (slope of linear association = 5.75) (Figure 4a) and conversely a decreased frequency of CD4⁺ pre-T_CM/TM (CD45RO⁺CD27⁺) (slope of linear association = −5.89) (Figure 4b).

Figure 4: Relationship of change in T-cell subsets and β-cell function.

Panels a and b: Analysis of the linear association between a change in a T-cell subset and β-cell function from baseline for both treatment groups. In the placebo group, CD4⁺ pre-T_NV/SCM # have a positive (a) and CD4⁺ pre-T_CM/TM # have a negative (b) linear association with β-cell function as indicated by treatment-specific regression lines. Panels c and d: C-peptide retention (95% CI) estimated by the linear association models. Only T-cell subsets having significantly different rates of C-peptide retention between treatment groups are shown. Linear models that estimate the retention of baseline C-peptide for a 10% increase of specified for CD4⁺ and CD8⁺ T-cell subset. See Material & Methods for definition of T-cell subsets.

We used the linear model to estimate the retention of baseline C-peptide at year 1 for a 10% increase of a specified T-cell subset from baseline to year 1 (as by our prior analysis (9)). For example, a 10% increase in CD4⁺ T_NV (CD45RO⁻CD27⁺CD95⁻CCR7⁺CD45RA⁺CD31⁻) is associated with a C-peptide retention of 141% (95% CI, 112%−176%). In contrast, a 10% increase in CD4⁺ pre-T_CM/TM (CD45RO⁺CD27⁺) is associated with C-peptide retention of 57% (CI 41%−80%) (Figure 4c).

We next examined the effect of abatacept treatment on this relationship, since its therapeutic effect is retention of C-peptide (7), in association with changes to several T-cell subsets (Figure 2 and Supplementary Figure 3). Using the same model, all linear associations (slope of the relationship) between T-cell subsets and C-peptide attenuate to zero (Figure 4c). For example, in the treatment arm, a 10% increase of CD4⁺ T_NV (CD45RO⁻CD27⁺CD95⁻CCR7⁺CD45RA⁺CD31⁻) is associated with C-peptide retention of 97% (CI 69%−136%); similarly, a 10% increase in CD4⁺ pre-T_CM/TM (CD45RO⁺CD27⁺) is associated with C-peptide retention of 105% (CI 64%−172%). Thus abatacept significantly decreases, i.e. uncouples, the relationship between specific T-cell subsets and β-cell function observed in the placebo arm. Similarly, the uncoupling effect was also observed for CD8⁺ T-cell subsets (Figure 4d).

Abatacept-treatment induced changes to T_CONV and T_REG subsets

We analysed the samples by mass cytometry to examine activation, differentiation and homing in memory/naïve subsets in CD4⁺ T_CONV (i.e. non-T_REG) and T_REG (Supplementary Figure 1 for gating strategy). There was a good correlation between identical subsets (See Material & Methods for definition of T-cell subsets.) identified by flow and mass cytometry (data not shown), yet some attenuation is observed for large flow values.

For T_CONV, PCA separates the treatment groups at year 1 and year 2 (Figure 5a, b and Supplementary Figure 2 e, f), similar to that seen by flow cytometry. In year 1, dimension 1 (accounting for 42.9% of the variance) is mainly driven by changes to T_CM and T_TM (Supplementary Figure 2e) whereas dimension 2 (accounting for 15.8% of the variance) is mainly driven by changes to T_EM. In year 2 (Supplementary Figure 2f) dimension 1 (accounting for 44.3% of the variance) is mainly driven by changes to T_CM, whereas dimension 2 (accounting for ~17.4% of the variance) is mainly driven by changes to T_EM, T_TM and T_NV (Figure 5c).

Figure 5: Change in T_CONV subsets in treatment groups.

Change in T_CONV subsets from baseline to year 1 and to year 2 in treatment groups is analysed (a, b and c). Principal component analysis of individuals from treatment groups are shown for baseline to year 1 (a) and to year to 2 (b). Change (as percentage from baseline) for T_CONV subsets is shown (c). T_CONV subsets (in depth phenotyping) with a significant abatacept treatment effect are shown (d). Mann-Whitney U-test with Benjamini-Hochberg correction; * p<0.05, ** p≤0.01, *** p≤0.001, **** p≤0.0001; The horizontal line in the middle of each box indicates the median; top and bottom borders mark the 75^th and 25^th percentiles, respectively; whiskers according to Tukey; See Material & Methods for definition of T-cell subsets. % of baseline; change in frequency of a specific T-cell subset, over time (i.e. from baseline to year 1 or year 2), represented as proportion of baseline frequency.

We then further defined the “standard” T-cell subsets (defined by CD45RO and CD27) with additional markers (i.e. CCR4, CCR5, CCR6, CXCR3, CD69, PD-1, CD39) and tested for abatacept-induced change. We found that abatacept treatment reduces T_CONV expressing CCR4, CCR6, CD39 or PD-1, early-stage memory (pre-T_CM/TM) T_CONV expressing CCR5, and early and late-stage memory (pre-T_CM/TM and pre-T_EM) T_CONV expressing CCR4, CD39, CD69 or PD-1 (Figure 5d).

For T_REG, PCA again separates the treatment groups at year 1 and year 2, in a similar fashion to that observed for the T_CONV, indicating that changes to T_REG subsets are likewise induced by abatacept treatment (Figure 6a, b).

Figure 6: Change in T_REG subsets in treatment groups.

Change in T_REG subsets from baseline to year 1 and to year 2 in treatment groups is analysed (a, b and c). Principal component analysis of individuals from treatment groups are shown for baseline to year 1 (a) and to year to 2 (b). Change (as percentage from baseline) for T_CONV subsets is shown (c). T_CONV subsets (in depth phenotyping) with a significant abatacept treatment effect are shown (d). Abatacept treatment effect on total T_REG (e) Mann-Whitney U-test with Benjamini-Hochberg correction; * p<0.05, ** p≤0.01, *** p≤0.001, **** p≤0.0001; The horizontal line in the middle of each box indicates the median; top and bottom borders mark the 75^th and 25^th percentiles, respectively; whiskers according to Tukey; See Material & Methods for definition of T-cell subsets. % of baseline; change in frequency of a specific T-cell subset, over time (i.e. from baseline to year 1 or year 2), represented as proportion of baseline frequency.

In year 1, dimension 1 (accounting for 31.5% of the variance) is mainly driven by changes to T_REG with phenotypes corresponding to T_NV, T_CM, T_TM whereas dimension 2 (accounting for 16.8% of the variance) is mainly driven by changes to T_SCM, T_EM. In year 2 dimension 1 (accounting for 26% of the variance) is mainly driven by changes to T_NV and T_CM whereas dimension 2 (accounting for 24.6% of the variance) is mainly driven by changes to T_EM and T_NV (Figure 6c and Supplementary Figure 2g, h).

We further defined the activation, differentiation and homing state of T_REG with additional markers (i.e. CCR4, CCR5, CCR6, CXCR3, CD69, PD-1, and CD39) within the T-cell subsets (defined by CD45RO and CD27) and tested for abatacept-induced change. Abatacept treatment reduces T_REG expressing CCR4, CD39 or PD-1, naïve (pre-T_NV/SCM) T_REG expressing CD39, early and late-stage memory (pre-T_CM/TM and pre-T_EM) T_REG expressing CD39, CCR4, CCR6, CD69 or PD-1 and late-stage memory (pre-T_EM) T_REG expressing CD39 (Figure 6d). Additionally, abatacept treatment leads to a significant reduction of the frequency of T_REG within CD4⁺ T cells over time (Figure 6e).

Abatacept treatment selectively reduces a diverse subset of antigen-experienced CD4⁺ T_CONV

To study changes in CD4⁺ T_CONV at greater depth, we used the unsupervised clustering algorithm FLowSOM. Performing clustering on 18 phenotypic markers, FlowSOM identifies 50 subsets of T_CONV providing an in-depth characterisation of the effect of abatacept treatment on the phenotypic landscape. The median baseline frequency of subsets ranged from 0.02% (range 0%−0.28%) to 55% (range 26%−81%). In 28 unique subsets abatacept treatment significantly alters the frequency from baseline to year 1 (22 subsets) or to year 2 (23 subsets) (Supplementary Figure 4).

PCA of the change in frequency from baseline to year 1 and to year 2 of the 50 subsets clearly separates treatment groups at both timepoints (Figure 7a, b), and achieves better visual separation than is seen analysing CD4⁺ T cells, T_REG or T_CONV (Figures 2, 5, 6). A total of 17 subsets are significantly affected in the treatment group compared to the placebo group (one increases and 16 decrease) at both timepoints and these subsets drive separation in the PCA (Figure 7c, d and Supplementary Figure 2i, j). Hierarchical clustering of the 17 significantly affected subsets identifies the major groups, namely T_NV, T_CM, T_TM and T_EM (Figure 7c). Within the 17 significantly affected subsets, absent or low levels of CD127 (IL-7 receptor) expression on antigen experienced cells marks a subset of T_CONV subsets that actually decrease, particularly those expressing PD-1 (subsets 5, 6, 34, 44), CD38 (5, 6, 10), CD39 (35, 36) and homing makers beta-7 (2), CXCR3 (4), CCR4 (36, 38) CCR6 (7) (Figure 7c, d). A decrease in cell frequency is also observed for subsets expressing medium to high levels of CD127 in conjunction with expression of CD25 (8, 18, 23, 33), CD161 (8, 18, 19, 23) or homing makers CCR4 (33) and CCR6 (8, 10). This suggests that abatacept treatment specifically targets antigen experienced T-cell subsets with a distinct and diverse differentiation, homing and activation status.

Figure 7: Change in FlowSOM-identified T_CONV subsets in treatment groups.

Principal component analysis of changes to the 50 FlowSOM-identified T_CONV subsets are shown for baseline to year 1 (a) and to year 2 (b) (see also Supplementary Figure 3i, j). For the 17 FlowSOM-identified T_CONV subsets that are significantly different between the treatment groups for change from baseline to year 1 and year 2 hierarchical clustering (c) and change in frequency of subset from baseline (d) are shown. Mann-Whitney U-test with Benjamini-Hochberg correction; * p<0.05, ** p≤0.01; The horizontal line in the middle of each box indicates the median; top and bottom borders mark the 75^th and 25^th percentiles, respectively; whiskers according to Tukey.

Discussion

Taking advantage of the setting of a phase 2 therapeutic intervention trial in patients with type 1 diabetes, our study was able to address two major aspects, namely the impact of co-stimulation blockade on peripheral T-cell compartments, and its effect on immune changes that are associated with the natural history of the disease. We define these two aspects, at unprecedented depth and resolution.

Defining antigen experience in T-cell compartments we see pronounced abatacept-induced changes for CD4⁺ T cells (both CD4⁺ T_CONV and T_REG), but only limited changes in CD8⁺ T cells, in line with these being less dependent on CD28-mediated co-stimulation for proliferation and activation (5, 18). This apparent difference in effects on distinct T-cell differentiation pathways (CD4⁺ and CD8⁺) may be an important consideration for building on the deployment of abatacept beyond its use as a monotherapy in type 1 diabetes intervention. An attractive approach, for example, might be to use abatacept in combination with a CD8-targeted therapy.

Our data demonstrate that abatacept affects CD4⁺ T cells according to their state of antigen experience: in general antigen-naïve subsets increase, whereas antigen-experienced subsets decrease. In particular, the frequency of T_NV and T_RTE increases significantly, while early memory T_CM, T_TM and to a lesser extent late memory T_EM decrease; we observe little or no change in the frequency for T_SCM. Several subsets with immunological memory are affected, indicating that treatment is acting on the generation and/or maintenance of specific cell subsets. According to the two main models of phenotypic maturation of T cells: (i) following activation and effector T-cell generation, there is a contraction phase after which T cells die or transition into a memory phenotype; or (ii) T cells progress gradually through the memory phenotypes upon every consecutive round of antigen stimulation (3, 19). Regardless of the model that may apply, our data showing increased frequencies in T_NV after abatacept provide evidence that treatment is limiting the activation and recruitment of naïve CD4⁺ T cells into memory phenotypes. Both models suggest that we should observe a reduction in the frequency of other memory T-cell subsets (downstream of T_CM in the case of model (ii)), which is indeed what we observe at year 2. The frequency of T_SCM, a T-cell subset defined by its longevity, to provide long-term memory (20) hardly changes, indicating that abatacept has a limited effect on cells with high capacity for homeostatic proliferation.

Besides interfering with memory pathways abatacept directly targets T-cell subset. Within the pre-T_CM/_TM (CD45RO⁺CD27⁺), we find that the expression level of CD95 is significantly reduced following treatment. CD95 (Fas receptor) is a death receptor from the tumour necrosis factor receptor superfamily associated with programmed cell death, cell growth and proliferation (21). The data suggest that pre-T_CM/TM expressing high levels of CD95 are preferentially deleted during treatment, potentially through increased apoptotic signalling (22).

Our deep-phenotyping approach supports the notion of selective loss of T-cell subsets. We find significant treatment-induced reduction in frequency in subsets expressing homing markers CCR4, CCR6, and alpha-4 beta-7 and activation markers HLA-DR, CD38, CD39, and PD-1. Abatacept predominantly affects T-cell subsets with low levels of CD127 (IL-7 receptor). Affected subsets expressing higher levels of CD127 typically also express CD25 (IL-2 receptor alpha chain) indicating a reliance on IL-2 signalling. Abatacept prevents T-cell activation which in turn leads to lower IL-2 secretion(23). Given that T_REG, a lineage reliant on IL-2 for optimal function and survival, are significantly reduced by abatacept, also previously observed by us and others (9, 24), strengthens the notion of reduced IL-2 levels during therapy. Again, in T_REG we observe significant reduction during therapy in subsets associated with activation and homing. The potentially negative effects of reduced T_REG frequency during treatment seem to be off-set by the beneficial effects of modulating T_CONV activation and proliferation. In addition, a recent study shows that CTLA-4 Ig can replace T_REG function in promoting memory CD8⁺ T cells in a quiescent state (25).

Overall our findings therefore suggest that co-stimulation blockade targets some T-cell subsets more effectively than others, and it appears that the least affected cells possess a higher capacity for homeostatic proliferation and longevity. It is thus likely that not all diabetogenic cells (e.g. T_SCM and late effectors) are targeted by treatment.

Co-stimulation blockade with abatacept profoundly affects the phenotypic landscape of peripheral T cells by operating largely in relation to the antigen-experience and co-stimulation dependence of a cell. A notable outcome of this treatment effect in terms of the natural history of type 1 diabetes is that it uncouples the relationship between acquisition of CD4⁺ T-cell memory and loss of β-cell function. Abatacept treatment uncoupled this link in all instances where we observed a significant relationship between change in a specific cell subset and change in β-cell function Thus, natural history markers such as change in T-cell memory will have only limited utility as tools for monitoring disease progression in the setting of studies on co-stimulation blockade since such a therapy renders the biomarker non-predictive in the group receiving treatment.

We have not fully explored the explanation of attenuation in our analysis and this remains for the future. Nevertheless, abatacept-induced changes to the subset landscape are vast and of variable magnitude, and mostly uniform in direction within the study cohort. We could not detect a correlation between the magnitude of change in any T-cell subset and β-cell function, suggesting that a beneficial outcome is achieved by crossing a certain threshold and/or that we do not asses the relevant T-cell subset. Given the depth of analysis it is unlikely that such a subset was missed albeit it is conceivable that the vast changes induced by the treatment hide such a subset. However, it is more likely that a relationship between change in T-cell subsets and β-cell function persists when focusing on type 1 diabetes-associated antigen specific T cells or specific phenotypes thereof (26).

Due to the difficulty to access pancreatic tissue for sampling during clinical trials, we sample peripheral blood to assess the effect of co-stimulation modulation on T-cell subset. At present, there is no study either confirming or challenging the notion that monitoring peripheral blood allows insight into the immunological situation in the pancreatic islets of patients with type 1 diabetes. Nevertheless, several recent reports correlate changes in peripheral immune-cell subset to changes in β-cell function and/or type 1 diabetes progression (26–28). These data suggest that one can utilise the immunological activity in peripheral blood as a proxy for pathological activity within the pancreatic islets. It remains for future studies to establish whether the changes in immune subset in peripheral blood (or draining lymph nodes (29), as a closer proxy) are identical to those in in the pancreatic islets.

We have confirmed and extended our initial finding of a relationship between changes in T-cell subsets and β-cell function as part of the natural course of type 1 diabetes pathology. Abatacept treatment uncoupled this link in all instances where we observed a significant relationship between change in a specific cell subset and change in β-cell function. Thus, natural history markers such as change in T-cell memory will have only limited utility as tools for monitoring disease progression in the setting of studies on co-stimulation blockade since such therapies render the biomarker non-predictive in the group receiving treatment. In a larger context, an important lesson from our analysis is that every biomarker needs to be validated in treated populations as well as control populations.

Key points.

• Co-stimulation blockade reduces the loss of β-cell function in type 1 diabetes

• CB effects peripheral regulatory and conventional CD4⁺, but not CD8⁺ T-cell subsets

• CB blunts relationship between disease progression and CD4⁺ memory T-cell subsets

Abbreviations used in this article:

AUC

area under the curve

CI

confidence interval

CTLA-4

cytotoxic T-lymphocyte-associated protein 4

FDR

false discovery rate

GLMs

general linear models

MFI

median fluorescence intensity

MMTT

mixed meal tolerance test

PBMC

peripheral blood mononuclear cell

PCA

principal component analysis

TAPI-2

tumour necrosis factor (TNF)-α Protease Inhibitor-2

T_CM

central memory T cell

T_CONV

conventional T cell

T_EM

effector memory T cell

T_LE

late-effector T cell

T_NV

naïve T cell

T_REG

regulatory T cell

T_RTE

recent thymic emigrant T cell

T_SCM

stem-cell-memory T cell

T_TM

transitional memory T cell