Differential expression of tissue-restricted antigens among mTEC is associated with distinct autoreactive T cell fates

Medullary thymic epithelial cells (mTEC) contribute to the development of T cell tolerance by expressing and presenting tissue-restricted antigens (TRA), so that developing T cells can assess the self-reactivity of their antigen receptors prior to leaving the thymus. mTEC are a heterogeneous population of cells that differentially express TRA. Whether mTEC subsets induce distinct autoreactive T cell fates remains unclear. Here, we establish bacterial artificial chromosome (BAC)-transgenic mouse lines with biased mTEClo or mTEChi expression of model antigens. The transgenic lines support negative selection of antigen-specific thymocytes depending on antigen dose. However, model antigen expression predominantly by mTEClo supports TCRαβ+ CD8αα intraepithelial lymphocyte development; meanwhile, mTEChi-restricted expression preferentially induces Treg differentiation of antigen-specific cells in these models to impact control of infectious agents and tumor growth. In summary, our data suggest that mTEC subsets may have a function in directing distinct mechanisms of T cell tolerance.

T cell tolerance to self-antigen is established, in part, concurrently with T cell development in the thymus. Developing T cells bearing antigen receptors with high reactivity to self-antigen encountered in the thymus can be deleted via negative selection, diverted towards immunoregulatory lineages, or rendered unresponsive [1][2][3] . A significant proportion of negative selection occurs in response to ubiquitously expressed selfantigen in the thymic cortex, while the medullary microenvironment plays an important role in the generation of T cell tolerance to tissue restricted antigens (TRA) [4][5][6] . The promiscuous nature of gene expression within the medullary thymic epithelial cell (mTEC) population provides an opportunity for developing T cells to test the self-reactivity of their antigen receptor against peptides derived from~90% of all genes 7 .
Various T cell-extrinsic parameters that affect the strength of T cell receptor (TCR) signals play an important role in tolerance induction in the thymus. For instance, the frequency of major histocompatibility complex (MHC) class II + cells expressing a given TRA in the thymus correlates with the mechanism of CD4 + T cell tolerance; it has been suggested that high numbers of cells expressing self-antigen drive clonal deletion whereas lower levels of thymic antigen presentation predominantly leads to nondeletional tolerance mechanisms and regulatory T cell (T reg ) generation 8 . Modulation of co-stimulatory signals also affects tolerance induction. Co-stimulation is required for negative selection of MHC class II-restricted autoreactive thymocytes as well as T reg development 9,10 . Further, the presence or absence of co-stimulation discriminates between clonal deletion and TCRαβ + CD8α + CD8β − (CD8αα) intraepithelial lymphocyte (IEL) lineage diversion in thymocytes bearing antigen receptors with high-affinity for self-peptides 11 .
It is becoming increasingly clear that there is significant heterogeneity among mTEC and that TRA are differentially expressed among these subsets [12][13][14][15][16] . mTEC subsets can be distinguished by differences in MHC and co-stimulatory molecule expression. The mTEC lo (CD80/CD86 lo MHC class II lo ) compartment includes "immature" progenitor cells, as well as a functionally mature subset and a population of terminally differentiated epithelial cells 12,13,17 . As they differentiate, immature mTEC lo upregulate co-stimulatory molecule expression and MHC class II levels to become mTEC hi18, 19 . mTEC hi include a subset that expresses autoimmune regulator (Aire), a well-known driver of TRA expression that also plays pleiotropic roles in T cell tolerance 20 . While Aire-driven TRA expression is restricted to the mature mTEC hi compartment, Aire-independent TRA are also expressed by the mTEC lo subset 21 . Therefore, different mTEC populations likely support tolerance to distinct TRA pools, but whether these subsets play unique roles in the quality or mechanism of T cell tolerance is not yet clear.
To decipher whether self-antigen expression by different mTEC subsets contributes to distinct mechanisms of T cell tolerance, we describe new transgenic (Tg) mice in which Airedependent and Aire-independent TRA promoters drive distinct patterns of model antigen expression among mTEC. We observe that autoreactive T cell diversion into immunoregulatory T cell lineages is influenced by the mTEC subset expressing the model antigen in these models with functional consequences. This is particularly relevant to the design of therapeutic approaches that aim to modulate the function of self-reactive T cells.

Distinct patterns of model antigen expression among mTEC.
To characterize T cell tolerance to TRA differentially expressed among mTEC subsets, we generated new Tg mouse models. In these mice, expression of a model antigen is directed from either the C-reactive protein (Crp) locus, whose Aire-independent expression is preferentially detected among immature mTEC lo , or the Insulin2 (Ins2) locus, where expression is restricted to Aire + mTEC hi ( Supplementary Fig. 1a, b 12,21 ). To maintain many of the physiological elements that regulate expression from these loci without interfering with gene function, we engineered 200 kb bacterial artificial chromosomes containing the locus of interest to include a modified ovalbumin (OVA) gene at the start codon of the Crp and Ins2 genes (Fig. 1a). The modified OVA gene contains additional MHC class I and II epitopes (LCMV gp33, LCMV gp66, and 2W), referred to as a "Universal Self-Antigen" (USA), is flanked by the Kb leader sequence and Db transmembrane region 22 , and is followed by a sequence encoding for a self-cleaving peptide and GFP; the construct is designed such that OVA and the additional epitopes are in the extracellular domain.
We assessed the fidelity of USA expression in sorted mTEC populations from Ins2-and Crp-promoter driven Tg mice (INS2 and CRP Tg mice, respectively) (Fig. 1b, c and Supplementary  Fig. 1a). In INS2 Tg mice, USA mRNA is restricted to mTEC hi and is dependent on Aire, whereas USA mRNA is more abundantly expressed in the mTEC lo subset in CRP Tg mice on both wild-type (WT) and Aire-deficient backgrounds (Fig. 1b). We screened USA expression in multiple Tg founder lines and maintained two independent lines of each for further analysis; the CRP lo and INS2 lo Tg founder lines have lower levels of USA mRNA expression (Fig. 1c). Next, because USA expression was difficult to detect in total mTEC in some Tg lines, we estimated total expression based on qRT-PCR analysis of the mTEC lo and mTEC hi subsets. More specifically, the relative expression values in mTEC lo and mTEC hi were multiplied by the proportion of each population to obtain an estimation of the expression in total mTEC. There is limited variability in expression among INS2 Tg founder lines; INS2 lo Tg express~2-fold lower levels of USA mRNA as compared to INS2 Tg mice (Fig. 1d). Overall, the INS2 lo and CRP lo Tg founder lines have similar levels of USA expression in the total mTEC compartment, although USA mRNA is preferentially expressed among mTEC lo in the CRP lo Tg mice, while it is restricted to mTEC hi in the INS2 lo Tg mice. In addition, USA is not detected in B cells, macrophages nor dendritic cells isolated from the thymus of the INS2 or CRP Tg mice ( Supplementary Fig. 1c, d). In the periphery, model antigen is expressed in the liver of the CRP Tg mice and in the isletenriched fraction of the pancreas of INS2 Tg mice (Supplementary Fig. 1e).
Apart from the amount of co-stimulatory molecules expressed by model antigen positive cells, the number of TRA expressing cells as well as TRA expression level on a per cell basis could also impact the signal received by developing thymocytes and, thus, their fate. While it is well known that Aire-dependent TRAs such as INS2 are usually expressed by only a small proportion of mTEC hi cells, the expression pattern of Aire-independent TRAs genes such as Crp is less well characterized. To further identify differences in Crp and Ins2 expression and characterize transgene expression, we performed single-cell RNA sequencing analysis oñ 10,000 total sorted thymic epithelial cells (TEC) from CRP, CRP lo , and INS2 Tg mice (Fig. 1e-i and Supplementary Fig. 1f-h). We detect USA expression in TEC from CRP Tg mice. However, transgene expression is not detected in TEC from CRP lo Tg mice and is very low in TEC isolated from INS2 Tg mice reflecting the differences we observe in the bulk mTEC population from these Tg mice. Although the number of cells expressing the transgene (USA) in CRP Tg mice is lower than the number of endogenous Crp-positive cells (3.2% vs 36.6%), their distribution across cell types is similar, with a marked predominance in mTEC lo cells (85.9% and 84.3%, respectively) (Fig. 1g). Endogenous Crp and Ins2 expression at a single-cell level in TEC isolated from the three Tg mice reveals differences between these two TRA in the mTEC hi subset; Crp is expressed in a greater proportion of mTEC hi , but at lower levels on a per cell basis while endogenous Ins2 is expressed in very few mTEC hi cells, but at a relatively higher level in individual cells (Fig. 1h, i). In addition, the higher endogenous Crp and CRP transgene expression in bulk mTEC lo vs mTEC hi as determined by qPCR ( Supplementary Fig. 1b and Fig. 1b, c), is further explained by the single-cell RNA sequencing data. A higher proportion of mTEC lo express Crp, but Crp expression in individual cells is similar between these two mTEC subtypes (Fig. 1h, i); the same is true for USA expression among   mTEC lo and mTEC hi in the CRP Tg mouse (3.2% of mTEC lo and 1.2% of mTEC hi express the USA transgene with an average expression of 0.7 and 0.6, respectively). These analyses demonstrate differences in the expression patterns of two TRAs in TECs and suggest that, although CRP and INS2 transgene expression is detected in fewer cells, the broad pattern of expression is similar to that of their endogenous genes counterparts.
As a read-out of the quantity and quality of thymic model antigen presentation in the CRP and INS2 Tg mice, mature, model-antigen specific OT-I TCR Tg T cells isolated from lymph nodes were overlaid on thymic slices from the TRA Tg mouse lines, and their proliferation and activation state were measured 72 h later. We observe differences in the proportion of activated, proliferating OT-I TCR Tg T cells dependent on USA expression levels in the thymus (Fig. 1j-m and Supplementary Fig. 1i). INS2driven USA mRNA expression in mTEC hi leads to a reduced proportion of OT-I TCR Tg cells that proliferate and a lower level of CD44 and PD-1 expression as compared to the CRP-driven USA expression (Fig. 1j-l). However, the cells that proliferate do so to a similar extent as indicated by the proliferation index (Fig. 1m). Despite similar overall levels of USA mRNA expression in the CRP lo and INS2 lo Tg strains (Fig. 1d), CRP lo Tg thymic slices fail to induce significant proliferation or activation of the antigen-specific CD8 + T cells which may be due to the low levels of co-stimulatory molecules in the mTEC lo population and the low level of model antigen expression in the mTEC hi compartment. Altogether, these data show that the four TRA Tg mouse models exhibit biased mTEC lo or mTEC hi expression of a model antigen.
Negative selection of CD8 lineage T cells in Tg mice. To analyze the extent of deletion of model antigen-specific MHC class Irestricted cells, we first tracked the development of OVA-specific OT-I TCR Tg Rag-deficient (OT-I) cells in USA expressing Tg mice. To limit competition for cognate antigen in these models, we used two different approaches. We first assessed negative selection using organotypic culture of thymic slices from CRP and INS2 Tg mice overlaid with a 1:1 mix of total CFSE-labeled OT-I and B6.SJL (CD45.1) control thymocytes. Despite significantly higher expression of the model antigen in the CRP Tg mice, the extent of deletion of high-affinity thymocytes after 24 h is similar in both CRP and INS2 Tg thymic slices as evidenced by a reduction in the proportion of OT-I:control thymocytes as compared to littermate controls (LMC) (Fig. 2a, b and Supplementary Fig. 2a) and the lower percentage of CD8 + single positive (SP) thymocytes within the remaining OT-I TCRβ hi population (Fig. 2c). Deletion of OT-I thymocytes is inefficient on thymic slices from CRP lo and INS2 lo Tg mice. In line with its Aire-dependency (Fig. 1b), the deletion of the OT-I TCR Tg thymocytes is completely abrogated in Aire −/− INS2 thymic slices, whereas deletion of antigen-specific thymocytes on CRP Tg thymic slices is unaffected by the absence of Aire (Fig. 2d). To confirm these results in vivo, we generated low-density bone marrow chimeras (BMC) in which LMC, CRP and INS2 Tg recipient mice were reconstituted with 1% OVA-specific OT-I or gp33-specific P14 TCR Tg TCRα knock-out (P14) and 99% WT bone marrow. We observe significant deletion of high-affinity OVA-specific OT-I cells in both the thymus and lymph nodes of CRP and INS2 Tg BMC as compared to LMC (Fig. 2e, f and Supplementary Fig. 2b, c). In addition, we detect partial deletion of P14 Tg thymocytes in the low-density BMC possibly reflecting either the lower affinity of the P14 TCR for its cognate peptide and/or less efficient processing and presentation of gp33 in this context ( Supplementary Fig. 2d). These results are also consistent in CRP and INS2 Tg mice crossed to OT-I TCR Tg mice ( Supplementary Fig. 2e). Together, these results suggest that negative selection of MHC class I-restricted thymocytes depends on thymic antigen level in these models. mTEC lo support TCRαβ + CD8αα IEL differentiation. Given the important role of co-stimulation in directing clonal deletion versus diversion toward the TCRαβ + CD8αα IEL pathway, we assessed the phenotype of the surviving OT-I thymocytes when cognate antigen is preferentially expressed in the mTEC lo (CD80/ 86 lo ) versus mTEC hi (CD80/86 hi ) compartments. We find a higher proportion of TCRβ hi OT-I DN cells in CRP but not INS2 Tg thymic slices (Fig. 3a, b and Supplementary Fig. 2a), and a large portion of TCRβ hi OT-I DN thymocytes express both PD-1 and CD122 in the CRP Tg thymic slices (Fig. 3c, d), markers of IEL precursors (IELp). In addition, a greater proportion of the TCRβ hi OT-I thymocytes are DN in the low-density CRP Tg BMC as compared to INS2 Tg and LMC chimeric mice (Fig. 3e, f and Supplementary Fig. 2b). In fact, a higher absolute number of CD8αα IELp is observed in the CRP Tg mice despite the significant depletion of OT-I cells in this model (Fig. 3f). While the proportion of OT-I T cells is greatly reduced in the thymus and peripheral lymph nodes from CRP and INS2 Tg BMC (Fig. 2e, f), a significant accumulation of these cells is observed in the IEL from the small intestine of the CRP Tg mice, and they are almost exclusively TCRαβ + CD8αα IELs ( Fig. 3g-i). To rule out the possibility that generation of this population was due to the significantly higher level of USA mRNA expression in the CRP Tg mice as compared to the INS2 Tg mice, we repeated these experiments in the CRP lo Tg founder line. Interestingly, the lower USA transgene expression also induces an accumulation of TCRαβ + CD8αα OT-I IEL as compared to LMC (Fig. 3j, k) though OT-I thymocytes are not deleted (Fig. 3l). Notably, an increase in TCRβ hi DN thymocytes and TCRαβ + CD8αα IEL is also observed in low-density P14 Tg BMC in CRP, but not LMC or INS2 Tg mice (Supplementary Fig. 2f-j). In addition, we tested whether USA expression by radiation resistant cells is sufficient to generate TCRαβ + CD8αα IELs in this model by generating BMC where CRP Tg mice were irradiated and reconstituted with 99% β2 M −/− and 1% OT-I β2 M −/− bone marrow. Indeed, in CRP Tg mice in which MHC class I expression is restricted to radiation resistant cell populations, OT-I T cells differentiate into TCRαβ + CD8αα IELs and accumulate in the small intestine ( Supplementary Fig. 2k, l). Furthermore, no increase in TCRαβ + CD8αα IEL differentiation is observed 6 weeks after an adoptive transfer of mature OT-I T cells into CRP Tg mice as compared to LMC suggesting a thymic origin of these cells ( Supplementary  Fig. 2m, n). Again, similar results are obtained regarding preferential CD8αα IEL development in CRP versus INS2 Tg mice when they are crossed onto the OT-I TCR Tg background ( Supplementary Fig. 2o, p).
It has been suggested that the IELp transit through a CD4 + CD8 + double positive (DP) intermediate in the thymus. However, it is not clear to what extent IELp selection occurs in the medulla. Both CCR7 + post-selection DP thymocytes and more mature CD8 + SP thymocytes could encounter self-antigen on mTEC lo cells. Using reporter mice in which developmentally distinct thymocyte subsets express YFP after CD4, distal Lck, or E8i promoter driven Cre mediated excision of a 5′ stop codon (CD4-YFP, distal Lck-YFP, and E8i-YFP, respectively), we sought to determine the developmental intermediates that give rise to thymic IELp. In CD4-YFP mice in which YFP expression is induced at the DP stage 23 , almost all IELp are YFP + , suggesting that they progress through a DP stage before differentiating into IELp ( Supplementary Fig. 3a-c). In the distal Lck-YFP and the E8i-YFP mice, in which YFP expression is induced after positive selection (late DP) 24 or at the CD8 + SP stage 25 , respectively, a significant proportion of IELp are YFP + suggesting that IELp diversion may also occur at the post-selection DP, and, perhaps to a lesser extent, the SP stage.
Model antigen expression in CRP Tg mice is preferentially detected in the mTEC lo population, but a significant proportion of mTEC hi express USA in these mice, raising the question of which cells are responsible for negative selection and differentiation of autoreactive thymocytes into CD8αα IELp. In order to better understand the role of each subset, mTEC lo and mTEC hi from WT and CRP Tg mice were sorted and co-cultured with OT-I thymocytes. Interestingly, mTEC lo and mTEC hi cells isolated from CRP Tg mice both induce a reduction in the proportion of OT-I CD8 + SP cells suggesting that both are able to induce negative selection ( Supplementary Fig. 3d, e). However, the differentiation of CD4 − CD8 − DN cells is higher when OT-I thymocytes encounter cognate antigen presented by mTEC lo than by mTEC hi (Supplementary Fig. 3d, f) suggesting that reduced levels of costimulatory molecules may favor CD8αα IELp differentiation in this model. mTEC hi preferentially induce T reg differentiation. Airedependent expression of TRA and co-stimulation are known to be important for thymic T reg differentiation 9,10,26-28 . Thus, we sought to determine if Aire-dependent expression of the model antigen in the INS2 Tg mice preferentially supports antigenspecific T reg differentiation in vivo. To do so, CRP and INS2 Tg mice were irradiated and reconstituted with 1% MHC class IIrestricted OVA-specific OT-II TCR Tg Rag-deficient (OT-II) and 99% WT bone marrow, and the proportion and phenotype of the OT-II cells were analyzed after reconstitution. The proportion and number of mature OT-II cells in the thymus and the peripheral lymph nodes is not significantly changed in the CRP and INS2 Tg BMC as compared to the LMC (Fig. 4a, b and Supplementary Fig. 4a-d). However, a significant increase in the proportion and absolute number of CD25 + FOXP3 + OT-II cells is observed in the thymus and the lymph nodes from the lowdensity INS2 Tg BMC specifically ( Fig. 4c-e). To determine the role of hematopoietically-derived antigen presenting cells (APC) in T reg induction in this model, CRP and INS2 Tg mice were irradiated and reconstituted with 1% OT-II and 99% MHC class II −/− bone marrow. A large proportion of OT-II cells still differentiate into CD25 + Foxp3 + T reg in INS2 Tg mice as compared to LMC when the majority of hematopoietic cells do not express MHC class II ( Supplementary Fig. 4e, f), suggesting that antigen presentation by mTEC hi is sufficient to induce OT-II diversion into CD25 + Foxp3 + T reg in this model.
TRA expression profiles impact polyclonal T cell fate. We expanded our study to assess the fate of T cells in the polyclonal population specific for model antigen expressed by different mTEC populations. We observe differences in the number of tetramer positive cells between the LMC of the different USAexpressing mouse strains, so the results are normalized to the LMC in each line in order to be able to compare the different Tg mice. Consistent with our results using MHC class I-restricted TCR Tg models, a lower number of H-2K b -OVA-specific CD8 + T cells are found in the peripheral lymphoid organs from naïve CRP and INS2 Tg mice as compared to their respective LMC, suggesting partial clonal deletion of OVA-specific MHC class I  Supplementary  Fig. 5a). Furthermore, naïve CRP Tg mice have a significant accumulation of H-2K b -OVA-specific TCRαβ IEL in the small intestine as well as a higher proportion of TCRαβ + CD8αα IEL (Fig. 5c-e). For MHC class II-restricted model antigen-specific cells, we observe that the relative number of I-A b -gp66-specific CD4 + T cells is lower in both naïve CRP and INS2 Tg mice as compared to their respective LMC counterparts (Fig. 5f, g and Supplementary Fig. 5b), while only the INS2 Tg mice have an increased proportion of model antigen-specific CD25 + Foxp3 + T reg relative to LMC (Fig. 5h, i). However, the relative absolute number of gp66-specific CD25 + Foxp3 + T reg is not increased in INS2 Tg mice as compared to LMC (Fig. 5i).
To determine the consequence of the thymic selection bias in immune responses, CRP and INS2 Tg mice were infected with lymphocytic choriomeningitis virus engineered to express OVA (LCMV-OVA 29 ). Subsequently, the impact of model antigen expression in different mTEC compartments on USA-specific  (Fig. 6a, b and Supplementary Fig. 5c). Moreover, a significant decrease in the proportion of LCMV gp33-specific CD8 + T cells is observed in the CRP and INS2 mice (Fig. 6c, d). No reduction in the LCMV NP396specific response is observed in any of the USA-expressing mice as compared to LMC, confirming that the decreased immune response is specific to the epitopes present in the USA construct ( Supplementary Fig. 5d, e). In addition, a lower proportion and absolute number of LCMV gp66-specific CD4 + T cells is observed in both CRP and INS2 Tg mice after LCMV infection (Fig. 6e, f), but only the INS2 Tg mice show an increase in gp66-specific CD25 + Foxp3 + T reg (Fig. 6g, h). Surprisingly, while the LCMV-OVA infected CRP lo Tg mice do not display any reduction in the gp66-specific CD4 + T cells response, the INS2 lo Tg mice have reduced gp66-specific CD4 + T cells of which a high proportion are CD25 + Foxp3 + T reg (Fig. 6e-h). Interestingly, when Aire −/− INS2 Tg mice, in which the transgene expression is lost in the thymus (Fig. 1b), are infected with LCMV-OVA, we still observe a significant reduction in the proportion of OVA-specific CD8 + T cells at day 8 postinfection, suggesting that Aire-independent peripheral expression of OVA is sufficient to dampen the OVA-specific effector CD8 + T cells response (Fig. 6i). However, analysis of the gp66specific CD4 + T cell response suggests that peripheral transgene expression is not sufficient to decrease the effector response nor to increase the proportion of gp66-specific T reg , suggesting that a central tolerance mechanism requiring expression of the TRA in the thymus is dominant (Fig. 6j, k). In sum, the data revealed by the tetramer staining of polyclonal T cells from naïve and infected mice confirm that mTEC lo biased and mTEC hi restricted expression of model antigen induce distinct autoreactive T cell fates in the context of a polyclonal TCR repertoire in these models.

CD25-APCCy7
FoxP3-A488  TCRαβ + CD8αα IELs play context-dependent immunoprotective and pathologic roles in intestinal homeostasis 30,31 . To determine if the higher proportion of OVA-specific TCRαβ + CD8αα IEL in CRP Tg mice have functional consequences, low-density OT-I BMC were infected by oral gavage with Listeria monocytogenes engineered to express OVA (Lm-OVA), and the number of bacteria in the small intestine and mesenteric lymph nodes were enumerated three days later, prior to significant influence from conventional TCRαβ + populations 32,33 . Interestingly, CRP OT-I BMC have a higher number of bacteria in the mesenteric lymph nodes and the small intestine following Lm-OVA infection as compared to LMC and INS2 OT-I BMC (Fig. 7a). In addition, the increased number of bacteria is observed with Lm-OVA, but not Lm-WT, bacterial infections (Fig. 7b) consistent with the idea that TCRαβ + CD8αα IELs require antigen-specific TCR stimulation for activation 34 . T reg are known to influence tumor control 35 . Thus, we evaluated growth of B16-OVA tumors in LMC, CRP, and INS2 Tg mice. The kinetics of tumor growth is similar in LMC and CRP Tg mice inoculated with B16-OVA cells (Fig. 7c). However, B16-OVA tumor growth is accelerated in INS2 mice as compared to LMC. The exacerbated tumor growth in INS2 Tg mice is both Aire and OVA antigen dependent (Fig. 7c, d). In addition, depletion of CD25 + T reg in INS2 mice restores control of tumor growth to similar levels as in LMC (Fig. 7e). Together, these data suggest an important role of thymic derived OVA-specific T reg in the loss of control of tumor growth observed in the INS2 Tg mice.

Discussion
Recent single-cell RNA sequencing studies of murine TECs have bolstered our understanding of the diversity of mTEC populations 12,13 . These studies reinforce the idea that distinct mTEC populations express different subsets of TRA, but whether they play unique roles directing autoreactive T cell fates is less clear [12][13][14]21,36,37 . In the present study, we generate new Tg mouse models with mTEC lo -biased or mTEC hi -restricted expression of model antigens that lead to differences in the quality of tolerance induction. These models not only allow us to confirm an important role of Aire-dependent TRA expression by mTEC hi in T reg differentiation, but also uncover a potential role for mTEC lo biased antigen expression in the diversion of autoreactive T cells into TCRαβ + CD8αα IELs with physiologically important differences in modulating the immune response. Thus, we show that the mechanism of tolerance may not only be based on TCR affinity for self-peptide and the number of antigen expressing cells in the thymus, but may also be influenced by the mTEC subset expressing the self-antigen.
In our model, antigen expression predominantly in the mTEC lo compartment supports antigen-specific thymocyte diversion toward the TCRαβ + CD8αα IEL lineage. TCRαβ + CD8αα IELs are a heterogenous cell population that arise from distinct developmental progenitors 31,38 . Two TCRαβ + CD8αα IEL populations develop in the thymus and differ in their selfreactivity 39 . The IELp population that expresses PD-1 is enriched in self-reactive cells that are induced after agonist selection 31,39 . Increases in IELp have been observed in the absence of costimulatory molecule expression by thymocytes 11 suggesting that autoreactive thymocyte interactions with APCs bearing low expression of costimulatory molecules are required to induce diversion toward this lineage. Indeed, diversion of autoreactive thymocytes into TCRαβ + CD8αα IELp is generally thought to occur at the DP stage or during the DN-to-DP transition after agonist signal 4,[39][40][41][42][43][44][45][46] , and DP thymocytes are largely located within the cortex limiting their interaction with mTEC. Nevertheless, in our study, mTEC lo biased expression of model antigens induces differentiation of antigen-specific thymocytes into IELp and generates a high proportion of antigen-specific TCRαβ + CD8αα IEL. In addition, the use of reporter mice in which developmentally distinct thymocyte subsets express YFP support the idea that the majority of IELp arise from pre-selection DP thymocytes. Although we cannot rule out that reporter expression is upregulated after differentiation, we also provide evidence that a portion of IELp possibly differentiate in response to agonist signals at the post-selection DP or even CD8 + SP stage supporting the idea that differentiation signals may also be provided in the medulla following the recognition of TRA expressed by mTEC lo . In further support of this notion, induction of a high proportion of antigen-specific TCRβ + DN thymocytes has been observed in the RIP-mOVA model in which OVA is highly expressed in both mTEC lo and mTEC hi cells 28,44,47 . Moreover, in vitro culture of OT-I thymocytes with sorted mTEC lo isolated from CRP Tg mice leads to OT-I differentiation into TCRβ hi CD4 − CD8 − DN cells confirming that these cells are able to divert autoreactive thymocytes into CD8αα IELp. In fact, while both mTEC lo and mTEC hi induce similar deletion of OT-I thymocytes, mTEC lo seem to be more efficient at inducing OT-I differentiation into TCRβ hi CD4 − CD8 − DN cells supporting the idea that absence of co-stimulation favors differentiation into CD8αα IELp. mTEC lo and mTEC hi cells from CRP Tg mice express similar levels of USA mRNA on a per cell basis suggesting that the level of TRA expression by individual cells is not a Fig. 6 TRA expression by distinct mTEC compartments impacts the mode of tolerance induction in the polyclonal T cell repertoire in LCMV infected mice. Mice were infected with LCMV-OVA and the antigen-specific response was measured in the spleen 8 days post-infection (NI, not infected mice) a, c Representative CD44 and H-2K b -OVA or H-2D b -gp33 tetramer staining on CD8 + T cells from littermate controls (LMC), CRP, and INS2 Tg mice. Relative response and absolute number of OVA-(b) and gp33-(d) specific CD8 + T cell response in USA expressing mice as compared to LMC. e Representative CD44 and I-A b -gp66 tetramer staining on CD4 + T cells from LMC, CRP, and INS2 Tg mice. f Relative and absolute gp66-specific response in USA expressing mice. g Representative flow plot of CD25 and FoxP3, gated on live I-A b -gp66-specific CD4 + T cells. h Relative proportion of CD25 + FoxP3 + T reg among gp66-specific CD4 + T cells and absolute number of gp66-specific CD4 + CD25 + FoxP3 + T reg ( Tg). Proportion of OVA-specific CD8 + T cells (i), gp66-specific CD4 + T cells (j) and gp66-specific T reg (k) in Aire +/+ and Aire −/− LMC and INS2 Tg mice (mean ± SEM, n = 8 Aire +/+ LMC, 5 Aire +/+ INS2, 8 Aire −/− LMC, 5 Aire −/− INS2). The relative number of tetramer + cells was obtained by dividing the number of tetramer + cells of each mouse by the mean number of the corresponding LMC mice in each experiment. Data from three independent experiments. The p-values for the left panels of b, d, f, h were calculated by Kruskal-Wallis test followed by Dunn's post-hoc comparisons to LMC, two-sided. For the right panels in b, d, f, h and panels i-k a two-tailed Mann-Whitney U test was performed. *p < 0.05, **p < 0.01, ***p < 0.001, relative to LMC. Source data are provided as a Source Data file.
determining factor for CD8αα IEL differentiation. Moreover, lower global model antigen expression in CRP lo Tg mice also induces significant differentiation of autoreactive cells into CD8αα IEL pointing to a role for the type of cells expressing the antigen rather than antigen level in these models.
TCRαβ + CD8αα IEL play an important role in intestinal homeostasis. In this study, we observe that mice with a higher number of OVA-specific TCRαβ + CD8αα IEL are less able to control model antigen-expressing bacteria growth and spread after oral challenge. Following antigen recognition, model antigen-specific TCRαβ + CD8αα IELs could produce immunomodulatory cytokines leading to a decreased defense against pathogens. It is also conceivable that the role of these cells is to reduce inflammation at a barrier site or to specifically inhibit an autoimmune response in order to maintain tolerance against self and commensal microorganisms 34,48,49 . However, the mechanism by which the higher number of OVA-specific TCRαβ + CD8αα IELs leads to a greater Lm-OVA burden remains to be determined in this context. In addition, the role of Aire-independent TRA-specific TCRαβ + CD8αα IEL in maintaining intestinal homeostasis requires further study. While a significant number of intestinal TRA are expressed in the thymus, they are not significantly enriched in the mTEC lo compartment 16 . It is possible that a subset of TCRαβ + CD8αα IEL selected on TRA expressed by mTEC lo simply serves to expand the TCR repertoire of this population or that this subset plays a unique role in intestinal homeostasis.
Considering that autoreactive T cell fate is determined by TCR signal strength, which is, itself, influenced by TCR affinity and avidity, it is not surprising that, in addition to TCR sequence itself 26,50 , the number of MHC and costimulatory molecules expressed by the antigen presenting cell influences the outcome of this interaction. Only CRP Tg mice express USA in mTEC lo at detectable levels. Thus, if we assume that autoreactive T cell diversion into IELp is induced in the absence of co-stimulation, it makes sense that CRP, and not INS2 Tg mice can support their differentiation, and one might predict that CD80/86-deficient INS2 Tg mice would also support CD8αα IEL development. In contrast, it is well accepted that T reg differentiation requires costimulation and that Aire-dependent expression of model antigen in mTEC hi supports T reg differentiation 3,51 . However, both INS2 and CRP Tg mice express similar levels of USA in the mTEC hi subpopulation raising the question as to why autoreactive T cell diversion in T reg is enhanced only in INS2 Tg mice. It is possible that, in CRP Tg mice, model antigens are mainly expressed by a transitional mTEC hi population expressing lower levels of costimulatory molecules thus decreasing their capacity to induce T reg differentiation. In addition, it was been shown in some models that antigen transfer to or direct antigen expression by BM-APC was important for negative selection and T reg differentiation 28,52,53 . As these processes are dependent on Aire 52,54,55 , it is possible to speculate that these mechanisms are less efficient for Aire-independent TRA, reducing their capacity to induce T reg differentiation. However, MHCII deletion on BM-APC did not abrogate T reg differentiation in INS2 Tg BMC suggesting that TRA presentation by BM-APC is not the main mechanism triggering T reg differentiation in our model and is consistent with other studies showing that MHCII deletion on BM-APC have only a mild impact on T reg generation 51,56 . It is also conceivable that although CRP Tg mice express model antigen in mTEC hi , TRA expression in mTEC lo reduces the chance of interaction between autoreactive T cells and mature mTEC therefore limiting the diversion of T cells into T reg .
Separately, it has been demonstrated that the mechanism of CD4 + T cell tolerance is influenced by the number of self-antigen expressing cells in the thymus 8,57 . Using mice with more subtle differences in model antigen expression level (CRP vs CRP lo and INS2 vs INS2 lo Tg mice), we suggest that overall levels of model antigen in the thymus has a large impact on the extent, but not necessarily the type, of the tolerance mechanisms induced in these Tg models. Instead, we observed, by single-cell RNA sequencing, that Ins2 is expressed in very few mTEC hi cells, but its expression level on a per cell basis is higher than that of Crp, suggesting that T reg differentiation might require, in addition to high co-stimulatory molecule expression, relatively higher TRA expression on a per cell basis. Thus, according to our study, the pattern of antigen expression across different mTEC populations, and, potentially, the expression level of self-antigen on a per cell basis, may impact T reg differentiation.
In conclusion, mTEC hi restricted and mTEC lo biased model antigen expression induces distinct autoreactive T cells fates in these Tg models. As such, we suggest that differential expression of TRA among mTEC subsets may contribute to the type of T cell tolerance generated; it will be important to determine the extent to which these observations are generalizable to endogenous Airedependent and -independent TRA.  70,71 . BAC clones RP23-51J21 and RP23-160P13 containing Ins2 and Crp, respectively, near the middle of the cloned genomic region were procured from the BACPAC Resources Center. USA-T2A-eGFP was recombineered into each clone at the start codon of the gene of interest. The USA-T2A-eGFP knock-in was validated by sequencing. BAC DNA was microinjected in C57BL/6 embryos to generate founders by the transgenic core facility at the Institute for Research in Immunology and Cancer (Université de Montréal). Mice were maintained in a specific pathogen-free environment at the Maisonneuve-Rosemont Hospital Research Center under a 12 h/12 h light/dark cycle. Temperature was maintained at 22°C with a relative humidity of 40%. Experimental control mice were co-housed. Euthanasia was performed by cervical dislocation or carbon dioxide inhalation. All animal protocols were approved by the local Animal Care Committee at the Maisonneuve-Rosemont Hospital in accordance with the Canadian Council on Animal Care guidelines. Immune cell isolation. Spleens and lymph nodes were dissociated using frosted glass slides, while single-cell suspensions of the thymus were prepared using a glass tissue homogenizer. Red blood cells in spleens and thymus were lysed before analysis with Ack lysis buffer (0.15 M NH 4 Cl, 10 mM KHCO 3 , 0.1 mM Na 2 EDTA). To isolate IEL from the small intestine, the connective tissue, fat, Peyer's patches and feces were removed. One cm intestine pieces were incubated in RPMI media supplemented with 3% FBS, 5 mM EDTA and 0.145 mg ml −1 DL-Dithiotreitol (Sigma-Aldrich) for 20 min at 37°C with agitation. The IEL were then isolated by shaking the small intestine pieces in RPMI media with 2 mM EDTA followed by a 30% Percoll (GE Healthcare) gradient of the released cells. For isolation of the thymic epithelial cells, the thymus was first cut in small pieces and pipetted up and down with a wide bore 1000 μl tip until cell release was no longer detectable. The remaining tissue was digested in RPMI media supplemented with 10% FBS, 10 mM HEPES, 0.25 mg ml −1 papain, 0.25 mg ml −1 collagenase D, 0.1 mg ml −1 DNase for 15 min at 37°C. The TEC were then enriched using EpCAM magnetic microbeads (Miltenyi Biotec) and sorted using a BD FACS Aria III with an 85 μM nozzel.

Methods
Infections. Six to twelve week-old female mice were infected with 2 × 10 5 focus forming units (FFU) of LCMV-OVA Armstrong intraperitoneally (i.p.) and spleens were harvested 8 days post-infection for flow cytometry analysis. Listeria monocytogenes (Lm) infections were performed by gavage of female mice with 1 × 10 9 colony forming units (CFU) of Lm-WT or Lm-OVA bacteria. The mesenteric lymph nodes and the small intestine were harvested 3 days post-infection and homogenized in distilled water plus 0.5% Nonidet P-40 (Bio Basic). Fold serial dilutions were then plated onto brain heart infusion (VWR) agar plates containing 200 mg ml −1 streptomycin (Bio Basic). Plates were incubated at 37°C for 24 h, and colonies were enumerated.
Bone marrow chimeras. Six to twelve week-old male or female donor mice were injected i.p. with 100 μg InVivoMAb anti-mouse Thy1 (BioXcell) at day −2 and −1 prior to harvesting the bone marrow cells. Six to twelve week-old sex matched male or female recipient mice were irradiated at 12 Gy and reconstituted by injecting 5 × 10 6 T cell depleted bone marrow cells from the donor mice intravenously (i.v.). The recipient mice were then injected i.p. with 100 μg InVivoMAb anti-mouse Thy1 (BioXcell) 1 and 7 days after bone marrow cells injection. BMC were analyzed 6-8 weeks post-irradiation.
Adoptive cell transfer. Lymph nodes of six to twelve week-old male or female OT-I and B6.SJL mice were harvested, cells were mixed at a 1:1 ratio, and 4 × 10 6 cells were injected i.v. in six to twelve week-old sex matched male or female USAexpressing or LMC mice. Six to eight weeks later, the proportion of TCRαβ + CD8αα IEL among the OT-I T cells isolated from the small intestine was analyzed by flow cytometry.
qPCR. For mRNA quantification, thymic B cells (CD19 + ), DC (CD11c + ), macrophages (F4/80 + ), mTEC lo (EpCAM + CD45 -UEA-1 + Ly51 − CD80 lo MHC-II lo ) and mTEC hi (EpCAM + CD45 − UEA-1 + Ly51 − CD80 hi MHC-II hi ) were sorted directly into TRIzol LS (ThermoFisher Scientific) from five to ten week-old male or female mice. Spleens were dissociated using frosted glass slides and a red blood cell lysis was performed. Livers were cut in small pieces and digested for 15 min at 37°C in RPMI media with 1 mg ml −1 collagenase D. The liver pieces were then passed through a 100 μm cell strainer and a red blood cell lysis was performed. For spleen and liver samples, 1 × 10 6 cells were pelleted and resuspended in 1 ml TRIzol. Pancreas were injected with HBSS containing 10% FBS, 0,1 mg ml −1 DNAse I and 1 mg ml −1 collagenase P and digested for 20 min at 37°C. Following digestion, the pancreas was triturated and centrifuged for 1 min at 200 × g. The pellet was resuspended in 10 ml HBSS 10% FBS, triturated and centrifuged again. This washing/dispersion step was repeated for four times and the remaining pellet was resuspended in 1 ml TRIzol. Total RNA was extracted using Phase Lock Gel (VWR) following the manufacturer's recommendations. Two micrograms of total RNA were then digested with DNAse I and reverse-transcribed with Superscript II using oligoDT primer (ThermoFisher Scientific) according to the supplier's instructions. Quantitative PCR was performed with the following primers: OVA mRNA forward 5′-CTTGAGCAGCTTGAGAGTATAA-3′, OVA mRNA reverse 5′-CCATCTTCATGCGAGGTAAG-3′, HPRT mRNA forward 5′-CTCCTCAGAC CGCTTTTTGC-3′, HPRT mRNA reverse 5′-TAACCTGGTTCATCATCGCTAA TC-3′, CRP mRNA forward 5′-TCAGCTTCTCTCGGACTT-3′, CRP mRNA reverse 5′-CTGCTTCCAGAGACACATAG-3′, INS2 mRNA forward 5′-GTGGC TTCTTCTACACACC-3′, INS2 mRNA reverse 5′-TACAATGCCACGCTTCT G-3′. Each reaction was performed in triplicate using a real-time cycler ABI Prism 7500 (Life Technologies) and Power SYBR Green (ThermoFisher Scientific). No RT controls were performed in parallel to confirm the absence of genomic DNA contamination. Results are reported as relative expression as compared to the WT LMC negative control (negative values were set to a Ct value of 40) and were normalized to HPRT.
Tetramer-based enrichment of antigen-specific T cells. The spleen and lymph nodes of six to twelve week-old naïve mice were harvested and dissociated using frosted glass slides, and the single-cell suspension was passed through nylon mesh. The cells were washed with cold sort buffer (PBS 2% FBS), and the cell pellet was resuspended in Fc block. The cells were then stained with PE-and APC-labeled tetramers. Finally, the tetramer positive cells were enriched using Miltenyi anti-PE microbeads and LS magnetic columns (Miltenyi Biotec).
In vitro culture of sorted mTEC with thymocytes. mTEC lo (EpCAM + CD45 − UEA-1 + Ly51 − CD80 lo MHC-II lo ) and mTEC hi (EpCAM + CD45 − UEA-1 + Ly51 − CD80 hi MHC-II hi ) cells were sorted from five to ten week-old male or female WT and CRP Tg mice. mTEC, WT thymocytes, and OT-I thymocytes (mixed 1:1:1 for a total of 10 5 -10 6 cells) were combined, and the cells were pelleted. The supernatant was carefully removed, and the pellet was vortexed to form a slurry which was deposited in a standing drop on a cell culture insert over RPMI 10% FBS. The cells were incubated at 37°C, 5% CO 2 for 24 h.
Single-cell RNA sequencing. Thymic epithelial cells (CD45 − EpCAM + ) were sorted from 9 week-old female CRP, CRP lo and INS2 Tg mice, and a 10X Genomics Single-Cell 3′ transcriptome library was prepared (Génome Québec). The libraries were sequenced using the Illumina HiSeq4000 PE100. Cell Ranger version 3.0.1 (10X Genomics) was used with default parameters to demultiplex sequencing reads, align them to a custom reference genome, distinguish cells from background, and obtain gene counts per cell. Reads were aligned to the mm10 reference genome provided by 10X modified to include the sequence of the transgene (Fig. 1a). The Ensembl transcriptome (v93) was used for counting, and, for each sample, cells were filtered on the following quality control (QC) metrics: mitochondrial content (indicative of cell damage), number of genes, and number of unique molecular identifiers (UMIs), using the Seurat package (v2.3) 74 . Thresholds for each sample are provided in Supplementary Table 1 and were selected according to the distribution of each metric within the sample, which varies with sequencing coverage and the number of cells captured. QC metrics statistics before and after filtering are provided in Supplementary Table 1, as well as in Supplementary Fig. 1g.
Samples were then processed using Seurat and cytobox 75 . Libraries were scaled to 10,000 UMIs per cell and log-normalized. UMI counts and mitochondrial content were regressed from normalized gene counts and the residuals z-scored gene-wise. Dimensionality reduction was performed using principal component analysis (PCA) applied to the most variant genes, and PCA was used as input for projection to two dimensions using uniform manifold approximation and projection (UMAP) and clustering. Clusters were identified using the shared nearest neighbor (SNN) modularity optimization-based clustering algorithm provided by Seurat, on the first 30 principal components, with a resolution of 0.6.
Tumor cell inoculation. 5 × 10 5 B16 or B16-OVA melanoma cells were injected subcutaneously in the right flank of six to twelve week-old male or female mice, and tumor area was measured every 2 days starting 7 days post-inoculation. Tumor growth was followed by measuring tumor diameter using a caliper and calculation of the tumor area. Mice were euthanized by cervical dislocation if tumors reached a diameter of 17 mm or in case of ulceration as requested by the institutional ethical board. For some experiments, the mice were injected with 200 μg α-CD25 antibody at day −4 and −1 in order to deplete CD25 + T reg prior to tumor inoculation.
Statistical analysis. Each experiment was repeated independently with similar results at least two times except for single cell RNA sequencing. For comparison between two groups, data with a normal distribution were analyzed for statistical significance using the unpaired Student t-test while data without a normal distribution were analyzed with the two-tailed Mann-Whitney U test. Data from more than two groups were analyzed with Kruskal-Wallis test followed by Dunn's post-hoc comparisons to LMC, two-sided. Tumor growth was subjected to a linear mixed effect modeling applied to log pre-processed tumor surfaces. All values where p < 0.05 were considered statistically significant. Analyses were performed with GraphPad Prism version 6. *p < 0.05, **p < 0.01, ***p < 0.001.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The single-cell RNA-sequencing data have been deposited in NCBI Gene Expression Omnibus (GEO) and is accessible through GEO Series accession number GSE153288 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE153288]. The source data underlying Fig. 1b-d, g-i, k-m, Fig. 2b-f, Fig. 3b, d, f, g, i-l, Fig. 4a, b, d, e, Fig. 5b, c, e, g, i, Fig. 6b, d, f, h-k, Fig. 7a-e, Supplementary Fig. 1c, d, e, Supplementary Fig. 2 d-g, i-p, Supplementary Fig. 3c, e, f, Supplementary Fig. 4e, f, and Supplementary Fig. 5e are provided as a Source Data file. All other data supporting the findings of this study are available within the article and its supplementary information files, or are available upon request to the authors. Source data are provided with this paper.