Type 1 diabetes (T1D) is an autoimmune disorder causing pancreatic β-cell loss that results in progressive failure of insulin secretion to control blood glucose levels. There are around 537 million diabetic patients worldwide, T1D representing 5-10% of these cases (1). T1D incidence has drastically increased, especially in young children, with an annual increase of 3-4% over the last 3 decades. T1D is characterized by an autoimmune attack on β-cells. Abnormal interactions between β-cells and immune cells include autoantibody (AAb) production by B-cells and the engagement of cytotoxic CD8⁺ T-cells as the final actors of β-cell destruction (2–6).

The genetics of T1D is complex. It is a polygenic disease, with more than 60 risk loci identified that account for ~80% of T1D heritability (7). Many of these genes are expressed in human islet cells and immune cells with known functional roles, suggesting that T1D pathogenesis involves both cell types (8–10). While T1D prevalence in the general population is ~0.1%, family and twin studies suggested concordance rates of around 50% and 7% for identical twins and siblings, respectively (11–13). Around 50% of T1D genetic risk is related to the human leukocyte antigen (HLA) region, especially in the loci of HLA class II DR and DQ (14). These loci and the other genetic variants associated with T1D, whether coding or non-coding, require experimental validation of their impact on β-cell destruction to recapitulate and understand the pathophysiology of T1D. Since obtaining pancreatic tissue from live T1D patients is a rare occurrence (15), several β-cell models have been developed to recapitulate and investigate the progression of T1D. The most common in vivo autoimmune models for studying T1D are the non-obese diabetic (NOD) mouse and the BioBreeding (BB) rat (16, 17). Many T1D pathogenetic elements and immune pathways were identified using these valuable animal models. However, the physiology of β-cell development and function as well as the pathophysiology of T1D in humans are fundamentally different compared to rodents (18–20). Therefore, several human β-cell models were established to understand and recapitulate the pathophysiology of diabetes ( Figure 1 ). Furthermore, understanding the crosstalk between the β-cells and the immune cells is crucial to generate immune-protected β-cells for cell replacement therapy. In this review, we summarize recent progress in T1D disease modeling and future opportunities using in vitro systems of β-cell and T-cell crosstalk.

Abnormal activation of many types of antigen-presenting cells (APCs) plays a role in the pathophysiology of T1D (77). However, T-cells, particularly CD8⁺ T-cells, are the final effectors of pancreatic β-cell destruction. They are also the most abundant immune subset infiltrating the islets of T1D patients (78). Different strategies are available to generate islet-reactive T-cells ( Figure 4 ), and their strengths and weaknesses are summarized in Figure 5 .

The most obvious way to study these lymphocytes and their interaction with β-cells would be to isolate them from peripheral blood mononuclear cells (PBMCs). For instance, total CD8⁺ T-cells can be isolated from PBMCs by magnetic selection, and those reactive to islet antigens can be sorted after labelling using HLA Class I multimers (79). When directly studied ex vivo, these lymphocytes have a high clinical relevance and reliable recognition of their cognate peptide. However, they are found at low frequencies (1-50/10⁶ CD8⁺ T-cells for a given antigenic peptide) in the circulation of T1D and healthy donors, which corresponds to the expected range for any naïve CD8⁺ T-cell specificity (80). Although higher frequencies have been reported by other teams (3, 81–86), in our hands this range displays limited variation across antigenic specificities and individuals, in line with the largely naïve phenotype of these circulating islet-reactive CD8⁺ T-cells. In comparison, CD8⁺ T-cells directed against peptides derived from commonly encountered viruses, e.g., Influenza virus, EBV, CMV are 10- to 1,000-fold more frequent (3, 82, 86). Furthermore, we have recently shown that these islet-reactive T-cells do not display higher frequencies in the peripheral blood of T1D patients than in healthy subjects (3, 27, 82). Because of this low frequency and, subsequently, low isolation efficiency, they cannot be used directly ex vivo for high-throughput β-cell/CD8⁺ T-cell interaction models.

The first solution to reach higher cell numbers is to generate T-cell lines (i.e., polyclonal, comprising multiple different T-cell clonotypes recognizing the same antigens, often admixed with irrelevant T-cells) or clones (i.e. monoclonal, comprising a single T-cell clonotype). To obtain T-cell lines, islet-reactive lymphocytes are labelled with HLA Class I multimers or activation markers following a short in vitro antigen stimulation and single-sorted in culture plates for further expansion. Single-cell sorting can be done directly ex vivo or following stimulation with peptide-loaded APCs (e.g., monocyte-derived dendritic cells or unfractionated PBMCs) in order to obtain a preliminary antigen-driven expansion, usually boosted through the addition of cytokines such as IL-2, IL-7, IL-15. In both cases, single cells are plated in individual wells and expanded, so that all the progeny represents a single clonotype. The most studied islet-reactive CD8⁺ T-cell clone is 1E6, which recognizes a preproinsulin (PPI)_15-24 peptide (87, 88). The generation of both T-cell lines and clones is labour-intensive, while their length of in vitro expansion is limited, as these cells need to be periodically (every 10-14 days) restimulated by peptide-pulsed APCs or mitogens and eventually undergo exhaustion and/or functional anergy. Moreover, this expansion requires the addition of third-party PBMCs that act as feeders to sustain growth. Although these feeder cells are routinely irradiated to avoid their expansion, they can occasionally survive in culture and dilute the peptide-reactive T-cell fraction of interest. Moreover, only a small fraction of clones (<10% in our hands) can be proficiently expanded in vitro and confirm the antigen reactivity against which they were originally selected. Another limitation of clones obtained from primary CD8⁺ T-cells is their variable behaviour across different stimulation cycles and limited in vitro expansion, limiting the number of single-batch cells available for high-throughput assay platforms.

In order to overcome the limited expansion and variable reproducibility of clones, immortalized “avatars” of autoreactive T-cells have been developed by transducing T-cell receptors (TCRs) directed against β-cell peptides into different host cells. Hybridomas provide a first type of recipient cells. They are obtained by immortalizing a T-cell through fusion with a tumour cell line. The widely used 5KC line is a murine hybridoma obtained from a mouse CD4⁺ T-cell fused with thymoma cells (89–91). It constitutively expresses murine CD4, but can be transduced with human CD4 or CD8, and CD3 molecules in order to increase their responsiveness to cognate peptide following introduction of a human TCR (92). A modified version elaborated in 1993 does not express an endogenous TCR, reducing the risk of mispairing when an exogenous TCR is transduced (91).

Another interesting model is the Jurkat cell line. It is an immortalized human CD4⁺ T-cell line obtained from a non-Hodgkin lymphoma patient (93). A derived cell line developed by the Steinberger Lab, called Jurkat Triple Parameter TCR signalling reporter (TPR), presents several advantages. First, it is knocked-out for endogenous TCRα/β subunits, thus decreasing the risk of mispairing between the endogenous and exogenous TCR. Second, it expresses the CD8αβ co-receptor to stabilise TCR-HLA interactions, as well as three fluorescent reporter genes under the transcriptional control of nuclear factor of activated T-cells (NFAT), NF-κB and activator protein (AP)-1. These three transcription factors are activated upon TCR triggering, thus leading to fluorescent protein expression when peptide-HLA recognition takes place. These fluorescent reporters offer a simple read-out of T-cell activation (94). Furthermore, Jurkat cells present a TCR transduction efficiency close to 100% (94).

Both 5KC and Jurkat have the advantage to be immortalized cell lines, easy to grow in a highly reproducible fashion. However, they are derived from CD4⁺ T-cells and therefore lack cytotoxic machinery. Moreover, being derived from highly mutated tumour cells, they show abnormalities in TCR signalling pathways (95), endowing them with a lower responsiveness to their cognate peptide compared to primary CD8⁺ T-cell lines and clones. If we consider the fact that autoimmune TCRs have generally lower affinity than those recognizing foreign antigens, this impaired signalling can occasionally mask responsiveness and yield false negative results (92, 96).

To overcome these limitations, primary human CD8⁺ T-cells are increasingly used as recipients to transduce islet-reactive TCRs (97–100). They present the key advantage to carry the cytotoxic equipment needed for β-cell killing experiments and are transferable to in vivo rodent models. In 2015, Babad et al. successfully transferred primary human CD8⁺ T-cells transduced with an IGRP-reactive TCR into immunodeficient HLA-A2-transgenic NSG-A2 mice (97). Although one drawback is the lower (25-50%) TCR transduction efficiency compared to immortalized T-cell lines (97, 98, 101), this can be easily bypassed by sorting the TCR-transduced fraction prior to expansion.

However, as sophisticated as these cellular models can be, co-culturing β-cells with CD8⁺ T-cells remains a reductionist model and does not account for all the complexity of the immune system crosstalk. These models may be fine-tuned by adding other components. For instance, iPS cells could be used to develop isogenic systems by generating other immune cell types such as monocytes, macrophages (102), dendritic cells but also β-cells, all derived from the same donor and co-cultured with TCR-transduced CD8⁺ T-cells in a completely autologous cell system bypassing the problem of alloreactivity (99).

New approaches are currently being developed to derive T-cells from hPSCs, as reviewed in (103). T-cells can be generated from hPSCs either directly or through differentiation to bona fide hematopoietic stem cells, which give rise to T lymphocytes and other immune cells. However, the current differentiation protocols need further optimization as the generated T-cells are immature. To allow the maturation of the generated T-cells, others have generated thymic epithelial cells (TECs) from hPSCs, which are important for the positive selection of T-cells. Although the generated TECs were not functional in vitro, they were able to differentiate and function when implanted in athymic mice in vivo (104). Creating a multi-tissue platform combining SC-islets, SC-T-cells and SC-TECs is a promising avenue, however, considerable challenges need to be addressed before this platform can be used to model T1D in vitro.

The β-cell and T-cell crosstalk can be studied using readouts that focus on one side of this interaction, alone or in combination. These readouts are schematically depicted in Figure 6 , with their strengths and weaknesses summarized in Figure 7 .

New systems are emerging to study the autoimmune interactions between human β-cells and immune cells. This is facilitated by the recent development of human β-cell lines and the possibility to generate SC-islets and TCR-transduced primary T-cells from healthy and T1D subjects. The choice of the model depends on the parameters under study, the chosen readout and the affordability of the platform. While SC-islets provide high β-cell functionality and easy genetic engineering, their 3D cluster morphology can make some readouts (such as measuring surviving target cell numbers) more challenging, making β-cell lines more convenient for some experiments.

The role of genetic variants involved in β-cell/T-cell crosstalk can be investigated using genetically engineered SC-islets to elucidate the vulnerability or protection potential of the genetic modification under study. Another challenge that can be tackled using β-cell/T-cell models is how to hide β-cells from immune surveillance to avoid autoimmune or allogenic immune attack for β-cell replacement therapies. This is relevant also for autologous iPSC derivatives, since graft (auto)immune rejection also occurs in this fully HLA-matched setting (138, 139).

Human β-cells have the ability to upregulate the expression of multiple factors to interact with the immune system in response to inflammatory environment. HLA class I upregulation is a major driver of T1D pathogenesis (27, 140), as these molecules present self-antigens to autoreactive CD8⁺ T-cells. To counteract the inflammatory response, β-cells upregulate the Programmed Death-Ligand 1 (PD-L1) which acts as an immune checkpoint inhibitor by binding to its receptor PD-1 on T-cells, thus inhibiting TCR signalling and T-cell activation (141, 142). It is considered an important protective immunomodulatory factor when expressed in β-cells in response to pro-inflammatory cytokines. This protection mechanism is similar to that mounted by cancer cells, which upregulate PD-L1 to evade immune surveillance (143). This provided the basis of blocking anti-PD-1/PD-L1 antibody therapeutics (144), with agonist anti-PD-1 antibodies under development to achieve an opposite tolerogenic effect for autoimmune diseases. β-cell/T-cell crosstalk models may provide valuable platforms to screen for such therapeutics. They may also elucidate the dynamics of upregulation and interaction between PD-1/PD-L1 and other immunomodulatory receptor/ligand pairs. A similar application is envisaged for candidate β-cell-protective agents, e.g. verapamil (145), JAK, STAT and TYK2 inhibitors, as exemplified in our proof-of-concept study (74). Using a TYK2 inhibitor and an in vitro β-cell/T-cell model, we observed only a partial protection from β-cell destruction, due to an inhibition of the IFNα-induced upregulation on both HLA Class I and PD-L1.

Although islet-reactive T-cells exist in healthy individuals, they are not residing in islets, as is the case for T1D patients (3, 27, 146). Interestingly, T-cells cultured with autologous SC-islets from non-diabetic and T1D donors showed neither activation nor β-cell target cytotoxicity unless SC-islets were preliminarily exposed to endoplasmic reticulum stressors (147). This emphasizes the increasing appreciation of the active role that β-cells play in shaping their autoimmune vulnerability (148–151), which will be another paramount aspect to address using β-cell/T-cell crosstalk models.

Similar to primary human islets, SC-islets have the ability to upregulate HLA Class I molecules and PD-L1 upon pro-inflammatory cytokine exposure (152, 153). They can also present epitopes such as preproinsulin (PPI) peptides to CD8⁺ cytotoxic T-cells (154). Disruption of the β2-microglobulin (B2M) gene diminishes surface HLA class I expression, thus preventing CD8⁺ T-cell-mediated cytotoxicity. However, since non-classical HLA class I molecules HLA-E and HLA-G inhibit the activation of the innate immune natural killer (NK) cells, B2M knockout cells display increased vulnerability to NK-mediated lysis (155). Different approaches are being exploited to create hypoimmunogenic stem cells that can avoid both T-cells and NK-cells. While overexpressing HLA-E fused to B2M reduced NKG2A⁺ NK-cells-mediated lysis of B2M knockout cells (156), it was not adequate to suppress KIR2DL1–4⁺ NK-cell populations, which are suppressed by HLA-C or HLA-G (157, 158). Indeed, another approach disrupting HLA-A and HLA-B, but not HLA-C, improved NK- cell evasion (159). Other studies used a multiplexed CRISPR/Cas9 system to delete HLA-A/-B/-C genes and HLA class II genes by targeting CIITA. In addition to these deletions, overexpressing the immunomodulatory factor CD47, alone (160) or together with PD-L1 and HLA-G (161), inhibited the activation of T-cells, NK-cells, and macrophages. Although upregulating immunomodulatory factors in β-cells enhanced their survival against immune cells, introducing overexpression cassettes in stem cells is prone to silencing in their differentiated progeny (162–164). Therefore, an alternative approach is to knockout the NK-activating ligands expressed in SC-islets. Profiling these ligands showed high gene expression levels of B7-H3 and CD155 throughout SC-islets differentiation. Generating a triple knockout of B2M, B7-H3 and CD155 demonstrated a significant reduction in T-cell- and NK-cell-mediated lysis in vitro and in vivo (165). However, evaluating long-term survival and functionality requires further assessment.

These examples of double-edged genetic manipulation exemplify the need for more complex in vitro models recapitulating the action of different immune cell mediators. It will also be important to test the possibility of restoring immune surveillance in the case of β-cell infection or malignant transformation, or whether harnessing these transplanted β-cells with kill-switch suicide genes may trigger autoimmunity by releasing self-antigens. In this perspective, the evolution of these in vitro platforms toward in vivo systems using humanized mouse models is highly desirable to thoroughly address these concerns.

Development of human β-cell lines and of genetically engineered SC-islets from T1D or healthy donors combined with T-cell co-culturing platforms allows a quantitative measurement of β-cell killing using different readouts. Recent technological advancements in developing these assays open new venues to model a complex disease like T1D. Investigating the interaction between β-cells and cytotoxic immune cells in autologous or allogeneic settings will shed light on the pathogenic mechanisms and help identifying novel therapeutic targets for T1D and β-cell replacement strategies. Genetic and drug screening using these assays may identify new targets to decrease β-cell vulnerability and/or T-cell aggressiveness and to allow transplanted β-cells to escape immune recognition.

CH and HI contributed equally to writing the review and share first authorship. TO acquired funding and participated in manuscript writing. RM conceived and supervised the review, acquired funding and participated in manuscript writing. All authors contributed to the article and approved the submitted version.