Past studies have clearly demonstrated the central role of anti-gluten CD4 ⁺ T-cell immunity in CeD pathogenesis, thereby establishing the link between the main environmental trigger, dietary gluten, and the main genetic risk factor, the major histocompatibility complex (MHC) class II molecules. Thus, 90% of the patients express HLA-DQ2.5 (DQA1*05-DQB1*02) and most of the remaining patients express HLA-DQ8 (DQA1*03-DQB1*03:02) or HLA-DQ2.2 (DQA1*02:01-DQB1*02) (reviewed in 34, 35) ( Figure 2).

Gluten is the viscoelastic blend obtained by mixing flours from wheat, barley, or rye with water, which is used to make bread and pasta. It comprises hundreds of proteins displaying repeated sequences rich in proline and glutamine ³⁷ . These proteins are incompletely digested in the gut lumen and release peptides, which can reach the subepithelial tissue and bind HLA-DQ2.5/8 molecules at the surface of intestinal antigen-presenting cells. As a consequence, gluten peptides are presented to and activate specific CD4 ⁺ T cells (reviewed in 34, 35) ( Figure 2). Most patients with CeD respond to a limited and shared set of peptides thereby defined as public or immunodominant epitopes ^{38, 39} . A larger number of these epitopes are recognized in the context of HLA-DQ2.5 than in that of HLA-DQ8, likely accounting for the preferential association of CeD with HLA-DQ2.5. Immune recognition of the gluten epitopes is highly dependent on their post-translational modification by TG2, which converts neutral glutamine into negatively charged glutamic acid residues within the intestinal mucosa. This modification enhances gluten epitope avidity for HLA-DQ2/8, thus allowing the formation of stable HLA-DQ–gluten complexes that are critical for efficient T-cell engagement and activation (reviewed in 34, 40).

Recent technical developments have enabled further characterization of the T- and B-cell immune responses elicited by gluten. A very high proportion of the gut plasma cells, which expand massively in the lamina propria during active CeD, were shown to produce IgA specific for gluten or TG2 or both. HLA-DQ molecules complexed gluten T cell epitopes as well as co-stimulatory molecules were detected at their surface, leading to suggest their role in gluten presentation to T cells ^{36, 41, 42} . Observation of IgA ⁺ DQ2.5-glia-α1a presenting cells among TG2-specific plasma cells also strengthens the so-called hapten-carrier hypothesis as a mechanism by which TG2 specific B cells get help from gluten reactive T cells. This hypothesis was proposed by Sollid to explain why anti-TG2 auto-antibodies provide a serum signature specific for active CeD and disappear after GFD ⁴⁰ . Gluten-specific CD4 ⁺ T cells have been extensively characterized. They produce large amounts of cytokines, notably interferon gamma (IFNγ) and IL-21 (reviewed in 34, 35). They possess a polyclonal TCR repertoire but preferentially use some variable (V)-gene segments and frequently display a non-germline, positively charged arginine residue in the highly variable CDR3 region of the Vβ chain ^{38, 39, 43} . The biased use of TCR-Vα chain segments is thought to reflect their preferential interaction with HLA-DQ. The arginine in the CDR3β loop might act as a lynchpin in the peptide–HLA-DQ interaction (discussed in 38). Fluorescent tetramers made of HLA-DQ2.5 molecules bound to immunodominant gluten epitopes have been designed to track gluten-specific CD4 ⁺ T cells in vivo ⁴⁴ , and single-cell sequencing was used to follow-up changes in their repertoire during the patients’ life ⁴⁵ . This elegant combination of approaches showed that gluten-specific CD4 ⁺ T cells clonally expand in the gut of untreated patients, where they represent about 2% of CD4 ⁺ T cells. A very small number of the latter cells circulate in blood, where they decrease after GFD. Yet memory gluten-specific CD4 ⁺ T cells persist in blood for decades despite GFD and can re-expand rapidly after oral gluten challenge, explaining the need for definitive GFD to prevent relapse ⁴⁵ . Overall, the gluten-specific T-cell response is now well characterized and its role in CeD pathogenesis undisputable. Therefore, many efforts are made for developing alternative treatments to GFD which may prevent this response, notably using oral enzymes ⁴⁶ , TG2 blockers ( https://zedira.com), or peptide-based immunotherapy to desensitize patients ⁴⁷ .

If gluten-specific CD4 ⁺ T cells are indispensable to trigger CeD, it is now clear that tissue damage also requires the activation of cytotoxic IELs in the presence of interleukin-15 (IL-15). A massive expansion of IELs and notably of CD8 ⁺ IELs with an αβ TCR is one of the cardinal features of active CeD ( Figure 2). IL-15 has emerged as a key player in CeD. This cytokine, notably produced by intestinal epithelial and dendritic cells, stimulates IEL expansion and cytotoxic activity and impairs their negative regulation by regulatory T cells and transforming growth factor beta (TGF-β) (reviewed in 34, 35). Studies in mouse models further indicate that upregulation of IL-15 and activation of intestinal CD4 ⁺ T cells by the dietary antigen are both necessary to drive the cytotoxic activation of CD8 ⁺TcRαβ ⁺ IELs and epithelial damage (reviewed in 48). The mechanism of cooperation is not yet definitively demonstrated but was ascribed in one mouse model to the synergistic effects of IL-15 and of IL-2 produced by antigen-specific CD4 ⁺ T cells ⁴⁹ . In this model, the two cytokines were sufficient to drive the expansion of IELs and to enhance their expression of granzyme B and of NK receptors independently of any direct antigen recognition. These results are reminiscent of findings in CeD. Thus, in active CeD, CD8 ⁺ T-IELs display enhanced expression of granzyme B and of the activating NK receptors NKG2D and CD94-NKG2C. In the presence of IL-15 in vitro, these cells can kill targets expressing MICA and HLA-E, the respective ligands of these receptors (reviewed in 34, 35). This scenario may operate during active CeD, when expression of MICA, HLA-E, and IL-15 is upregulated in the duodenal epithelium ^{50, 51} . In CeD, it is likely that not only IL-2 but also IL-21 that is produced by gluten-specific CD4 ⁺ T cells can cooperate with IL-15 to drive IEL activation ( Figure 2).

A second subset of T-IELs characterized by the expression of a γδ TCR has attracted much attention in CeD. Indeed, TCRγδ ⁺ IELs expand in patients with potential CeD before the development of epithelial lesions and their number remains increased long after initiating GFD (reviewed in 35). Recent work indicates that the repertoire of resident TCRγδ ⁺ cells in various tissues is shaped by butyrophilin-like molecules (BTNLs) expressed by epithelial cells ⁵² . In the human gut, expression of BTNL3 and BTNL8 drives the selective expansion of resident Vγ4 ⁺/Vδ1 ⁺ IELs ⁵² . Strikingly, in active CeD, the gut epithelium loses BTNL8 expression and this loss is associated with a profound depletion of resident Vγ4 ⁺/Vδ1 ⁺ IELs that are replaced by a distinct subset of Vδ1 ⁺ IELs. The latter cells failed to recognize BTNL3/BTNL8 and displayed a distinct functional program dominated by the production of IFNγ. Avoidance of dietary gluten restored BTNL8 expression but was insufficient to reconstitute the physiological resident Vγ4 ⁺/Vδ1 ⁺ IELs subset among TCRγδ ⁺ IELs ⁵³ . Overall, these data show that chronic inflammation permanently reconfigures the tissue-resident TCRγδ ⁺ IELs compartment in CeD. However, the exact role of the latter cells in CeD pathogenesis remains elusive.

Several questions remain to be answered. It is still unclear why only 5 to 10% of patients carrying predisposing HLA will develop gluten-specific CD4 ⁺ T cells and tissue damage. Besides HLA-DQ2.5 homozygosity, which increases the risk of CeD (reviewed in 40), about 50 CeD-associated common genetic variants, most of which are shared with other autoimmune diseases, have been identified by genome-wide studies. Yet their influence is modest overall and they bear no predictive value ^{54, 55} . A possible role of the gut microbiota remains elusive. Yet changes in duodenal and fecal microbiota in active CeD and observations in mouse models suggest its influence on gluten-induced pathology, possibly via modulation of gluten degradation ^{56, 57} . The long-lasting hypothesis of a viral trigger was recently examined in humanized mice carrying HLA-DQ8, which develop gluten-specific CD4 ⁺ T cells ⁵⁸ . Intestinal infection by a reovirus that induced type I interferon resulted in loss of systemic tolerance to gluten, although the mice did not develop any intestinal lesions ⁵⁸ . Recent epidemiological data from the TEDDY cohort of children at risk of CeD support the predisposing role of infections by gut-tropic viruses. Thus, risk of seroconversion increased within the three months following an episode of gastrointestinal infection whereas risk of CeD decreased in children vaccinated against rotavirus before three months of age ⁵⁹ . Finally, one unanswered question concerns the mechanism(s) of IL-15 upregulation in the intestine of patients with CeD. The inducing role of epithelial stress has been suggested ⁶⁰ perhaps triggered by microbiota-derived innate signals or by peptides present in wheat or gluten ^{27, 61} . Despite these interrogations, the mechanisms described above provide a plausible scenario and explain the efficacy of a strict GFD in the vast majority of patients with CeD. Yet primary or secondary resistance to GFD can develop in a small fraction of patients with RCD.

As indicated above, differential diagnosis of RCDI and RCDII is based on the presence or absence of a clonal population of IELs with an unusual phenotype. Accordingly, the intestine of patients with RCDI contains polyclonal T cells and immunophenotyping does not reveal any significant difference with uncomplicated CeD except for a moderate and inconstant increase in the percentage of CD4 ⁺ IELs ( 13 and personal data). Extra-intestinal autoimmunity is more frequent than in uncomplicated CeD, and disease is improved by immunosuppressive therapies. Therefore, we have suggested that autoimmune cells develop in the intestine of patients with RCDI and drive the persistence or relapse of intestinal lesions. However, this mechanism remains hypothetical. Long-lasting inflammation may promote the frequent development of collagenous sprue and predispose patients to the onset of overt lymphomas. Yet this severe complication is much less frequent than in RDII (reviewed in 13).

In contrast to RCDI, RCDII is now a well-characterized entity that can be defined as a low-grade clonal intraepithelial lymphoproliferation with a high risk of transformation into overt enteropathy-associated T-cell lymphoma (EATL) (reviewed in 13). In keeping with their intraepithelial origin, malignant IELs express the αE integrin (CD103), the expression of which may disappear with disease progression, notably in EATL (reviewed in 13 and personal observation). As indicated above, in most patients, malignant IELs lack expression of surface CD3-TCR complexes and CD8 but they contain intracellular CD3 and display clonal rearrangement of TCR genes ¹³ . Conversely, the abnormal IELs express NK receptors, notably NKP46, and, in the presence of IL-15, they can kill enterocyte lines in vitro ^{14, 50, 62} , a property which may explain the severe ulcerative jejunitis often observed in patients with RCDII ¹³ . These characteristics of the malignant IELs are instrumental for diagnosis. They have also raised many speculations on their cellular origin. We have recently shown, contrary to many expectations, that they do not derive from the transformation of T-IELs but from a small subset of unusual innate-like T-IELs present in the normal intestine. Like their malignant counterpart, the latter cells do not express surface CD3 but contain intracellular CD3 and DNA rearrangements of the TCR, and they also possess NK receptors and NK functions ¹⁴ . This unusual phenotype is imprinted by a combination of IL-15 and NOTCH signals during their differentiation in the gut epithelium from bone marrow precursors. Innate-like T-IELs form only a very small fraction of IELs in the normal adult intestine as well as in uncomplicated CeD, where most IELs are T lymphocytes ¹⁴ . If almost all RCDII cases arise from a transformed clone of innate-like T-IELs, there are some exceptions. Thus, in rare RCDII cases, the clonal intraepithelial lymphoproliferation develops from TCRγδ ⁺ IELs or even from TCRαβ ⁺ IELs expressing (or not) CD8 (unpublished observations). Immunohistochemical detection of NKP46 can help diagnosis as this NK marker is expressed by malignant IELs in most cases of RCDII but by only a minority of normal T-IELs in CeD and RCDI ¹⁶ .

Why may a clone of innate-like T-IELs (RCDII IELs) progressively expand at the expense of the normal polyclonal T-IELs, which massively infiltrate the gut epithelium of patients with CeD? Following work showing that IL-15 provides signals to RCDII IELs which promote their expansion and cytotoxic activation (reviewed in 13, 35), we recently showed that RCDII IELs frequently contain somatic JAK1 or STAT3 gain-of-function mutations (or both), which confer hyper-responsiveness to IL-15 ¹⁴ . These mutations may also promote response to other cytokines present in the intestine of patients with CeD, notably IL-2 and IL-21, which are produced by gluten-activated CD4 ⁺ T cells ⁶³ . Thus, JAK1 and STAT3 mutations may enable transformed innate-like T-IEL to outcompete normal resident T lymphocytes in the cytokine-rich environment of the intestine of patients with CeD. Our ongoing work further suggests that RCDII IELs can acquire additional mutations that may promote their dissemination in and beyond intestine and ultimately lead to their transformation into aggressive EATL ( 13, 21 and unpublished observations).

Overall, these data provide much better insight into the mechanisms that drive lymphomagenesis in CeD. They also suggest possible therapeutic strategies, such as IL-15 blockade as tested in a recent international clinical trial ⁶⁴ or alternatively the use of JAK inhibitors. However, possible risks are to impair a putative anti-tumoral response or to promote the clonal escape of malignant cells carrying additional mutations conferring a growth advantage or both. More work is necessary to delineate benefits and risks of these therapies. Another question concerns factors, which predispose patients with CeD to develop RCDII and EATL. The higher frequency of HLA-DQ2.5 homozygosity in RCDII (about 65%) than in uncomplicated CeD (about 30%) ¹³ points to a key role of the adaptive anti-gluten T-cell response. Accordingly, RCDII and EATL develop preferentially in patients with poor adherence to the diet or undiagnosed until late in life ¹³ . Moreover, strict GFD, even if insufficient to treat RCDII, is indispensable to control malignant cell expansion and to reduce the epithelial damage that is induced by malignant cells ¹³ . A recent genome-wide analysis in two small cohorts of patients with RCDII of Dutch or French origin also points to a predisposing genetic locus in 7p14.3 ( P = 2.37 Å ~ 10 ⁻⁸, odds ratio = 2.36), which may control expression of FAM188B ⁶⁵ . The exact contribution of this variant to CeD-associated lymphomagenesis remains elusive but interestingly FAM188B was recently implicated in the regulation of P53, which plays an important role in tumor surveillance ⁶⁶ .