To identify QFL T cells in tissues of naïve WT mice, we generated Qa-1^b-FL9 dextramers (QFL-Dex) based on the ‘dextran-doping’ technique (Bethune et al., 2017). Prior work of our lab showed that unlike the Qa-1^b-Qdm tetramer which stains NK cells, the Qa-1^b-FL9 tetramer (QFL-Tet) does not bind to the NKG2A receptor but specifically stains CD8⁺ T cells (Nagarajan et al., 2012). Because the same QFL monomers were used for generating the QFL-multimers in this study, we reasoned that it is unlikely that the QFL-Dex will exhibit shifted specificity. Moreover, in comparison with QFL-Tet, QFL-Dex showed improved sensitivity and specificity for QFL T cell detection (Supplementary Fig. 1), and further allowed enrichment of QFL T cells using magnetic beads. The average number of QFL T cells – defined as CD45⁺CD19⁻ TCRβ⁺CD4⁻QFL-Dex-PE⁺APC⁺ (Fig.1 a) – detected in the siIEL compartment of naïve WT mice was comparable to the population in the spleen. Additionally, QFL T cells were present at a frequency of ~1 in 1,000 of the siIEL CD8⁺ T cells, which was 10 times more frequent than in the spleen (Fig.1 b, c). Both the splenic and siIEL QFL T cells were essentially all CD8⁺ T cells, as these cells were barely detectable within the CD4⁺ T population (Supplementary Fig. 2). Consistent with prior studies, ~80% of the splenic QFL T cells expressed the ‘invariant’ TCR Vα segment Vα3.2 which was significantly higher than the proportion of Vα3.2-expressing cells within the total CD8⁺ T population in the spleen (<5%) (Fig.1 d, e) (Guan et al., 2017). Notably, QFL T cells in the siIEL compartment also contained a relatively larger proportion of Vα3.2⁺ cells (20.1%) compared to the total CD8⁺ TCRαβ siIELs (<10%) (Fig.1 f, g). The average percentage of Vα3.2⁺ cells within the QFL T population (20.1%) was lower in the siIEL compartment than in the spleen (63.3%). Vα3.2⁺ QFL T cells were nevertheless abundantly present in both the spleen and gut of naïve WT mice.

We have previously observed a high percentage of CD44^hi cells within the bulk splenic QFL T population of naïve WT mice(Guan et al., 2017; Nagarajan et al., 2012). Here we further investigated the association between preferred Vα3.2 usage by QFL T cells and their phenotype. We observed that Vα3.2⁺QFL T cells contained a significantly larger proportion of CD44^hi cells (~90%) than the Vα3.2⁻QFL T cells, suggesting that the antigen experienced phenotype of the QFL T population was mainly contributed by the Vα3.2-expressing subpopulation (Fig.2 a, b). In contrast, only around 20% of the total Vα3.2⁺CD8⁺ T population expressed high CD44. It is worth noting that the percentage of CD44^hi cells within the total Vα3.2⁺CD8⁺ T population was slightly higher compared to the Vα3.2⁻ population (Supplementary Fig. 3 a), which was in line with a previous report (Prasad et al., 2021). We thus conclude that although both Vα3.2⁺ and Vα3.2⁻ cells were detected within the splenic QFL T population, a memory phenotype was specifically enriched in Vα3.2⁺QFL T cells.

Given the high frequency of QFL T cells in the gut, we further investigated their phenotype. We assessed CD8αα expression, a hallmark of the unconventional phenotype of natIELs, on Vα3.2⁺ and Vα3.2⁻ QFL T cells in the siIEL compartment of naïve WT mice. A relatively larger fraction of Vα3.2⁺QFL T cells lacked CD4 and CD8αβ expression but expressed CD8αα compared to the Vα3.2⁻ subpopulation suggesting that the majority of Vα3.2⁺ QFL T cells were CD8αα⁺ natIELs (Fig.2 c, d). No such difference was observed between the total Vα3.2⁺ and Vα3.2⁻ CD8⁺ IEL populations (Supplementary Fig. 3 b). These observations indicate that the Vα3.2⁺ QFL T cells in both the spleen and siIEL compartment are unconventional CD8⁺ T cells which likely are derived from the same precursors in thymus.

We further investigated what drives the unconventional phenotype of Vα3.2⁺ QFL T cells by analyzing mice deficient of molecules that impact Qa-1^b-FL9 antigen presentation including Qa-1^b, ERAP1 or TAP (Qa-1^b-KO, ERAP1-KO, TAP-KO). While we consistently observed loss of both the splenic and siIEL Vα3.2⁺QFL T cells in TAP-KO mice, the populations were detectable in both tissues of Qa-1^b-KO or ERAP1-KO mice with a slightly reduced population size in the spleen of Qa-1^b-KO mice (Fig.3 a). Vα3.2⁺ QFL T populations showed reduced percentages of CD44^hi cells in the spleen of Qa-1^b and ERAP1 deficient mice with the reduction being more significant in Qa-1^b-KO mice. In contrast, the phenotype of splenic Vα3.2⁻ QFL T cells was not significantly affected (Fig.3 b, c). Similar pattern of phenotype change was observed for the siIEL Vα3.2⁺ QFL T cells which showed almost complete loss of the unconventional natIEL phenotype including CD8αα expression in Qa-1^b-KO mice (Fig.3 d, e, Supplementary Fig. 4). These results demonstrate that TAP and Qa-1^b both play a role in the establishment of the unconventional Vα3.2⁺ QFL T population, with TAP being required for the presence of the population, and Qa-1^b being required for the unconventional phenotype imprinting of these cells. Paradoxically, loss of ERAP1, which would be expected to increase the presentation of FL9, also led to a reduction of memory and unconventional phenotypes in QFL T cells, perhaps due to the loss of certain QFL T clones as a result of tolerance mechanisms, such as negative selection. It is also worth noting that total splenic Vα3.2⁺CD8αβ⁺ T cells showed a similar loss of memory phenotype in Qa-1^b-KO and ERAP1-KO mice, whereas Vα3.2⁻CD8αβ⁺T cells did not (Supplementary Fig. 3 c). Thus, the unconventional Vα3.2⁺QFL T cells specifically, and a substantial proportion of total Vα3.2⁺ splenocytes, are impacted by Qa-1^b and ERAP1.

To further investigate the functional features of Vα3.2⁺QFL T cells, we utilized transgenic mice (QFLTg) expressing the predominant invariant TCR found on QFL T cells (Vα3.2Jα21, Vβ1Dβ1Jβ2-7), referred to hereafter as QFLTg cells. Abundant QFLTg cells or control OT-1 cells were detected in both the spleen and siIEL compartment of the respective TCR transgenic mice (Supplementary Fig. 5 a). Phenotypically, the QFLTg cells recapitulated the unconventional phenotype of polyclonal Vα3.2⁺QFL T cells with high CD44 expression in the spleen and CD8αα expression in the siIEL compartment. In contrast, OT-1 cells were phenotypically similar to conventional naïve CD8⁺ T cells, as they expressed low to intermediate levels of CD44 in the spleen and lacked CD8αα expression in gut (Supplementary Fig. 5 b, c).

Cytokines including IL-15 and the combination of IL-7/IL-18 have been implicated in the maintenance and survival of both ‘virtual’ memory T cells and IELs (Klose et al., 2014; Okazawa et al., 2004; Prasad et al., 2021; Schluns et al., 2003). We thus compared the proliferation of QFLTg and OT-1 cells in response to these cytokines. In keeping with their virtual memory surface marker phenotype, splenic QFLTg cells were significantly more responsive to IL-15 or IL-7/IL-18 stimulation than OT-1 cells (Fig.4 a). At 72hs, ~80% of QFLTg cells had proliferated in response to the cytokines, compared with only ~50% of OT-1 cells (Fig.4 b). In contrast, T cells isolated from the siIEL compartment of QFLTg or OT-1 mice showed a distinct pattern of cytokine responsiveness from the splenic T cells. First, both QFLTg and OT-1 cells from siIEL were less responsive to IL-15 than the splenic population. Second, siIEL QFLTg cells showed significantly reduced proliferation in response to the cytokines, especially to the combination of IL-7/IL-18, as compared with OT-1 cells (Fig.4 c, d).

To ensure that altered effector function was not skewed by the high specific TCR frequency in TCR transgenics, we generated partial hematopoietic chimeric mice with physiological precursor frequencies (QFLTg_WT) by transferring bone marrow cells from GFP⁺QFLTg mice into nonirradiated naïve WT neonates (Ladi et al., 2008). OT-1_WT chimera mice were generated in parallel as representative of conventional CD8⁺ T cells (Fig.4 e). This experimental design allowed us to track the GFP⁺ donor-derived T cells (GFP⁺CD45⁺CD19⁻TCRβ⁺CD4⁻Vα3.2⁺) separately from the recipient’s endogenous T cells (Fig.4 f). Based on the hyperresponsiveness of splenic QFLTg cells to cytokines, we hypothesized that these cells would exert effector functions more rapidly than conventional CD8 T cells, similar to the known innate-like T cells such as iNKT. We measured IFN-ɣ production by QFLTg or OT-1 cells from splenocytes or siIEL cells of naïve QFLTg_WT or OT-1_WT chimera mice stimulated the with 2µM FL9 or SIINFEKL(SL8) peptide respectively for 4.5 hours directly ex vivo. A large fraction (~60%) of splenic QFLTg cells expressed high levels of IFN-ɣ under these conditions, while only 6% of OT-1 cells did so (Fig.4 g, h). Notably, despite the low responsiveness to epitopes, siIEL QFLTg cells showed relatively higher basal levels of IFN-ɣ expression than OT-1 cells (Fig.4 i, j). In contrast to the hyperresponsiveness of splenic QFLTg cells to their ligand, neither QFLTg nor OT-1 cells from the siIEL compartment showed significant IFN-ɣ production at 4.5 hours. The low IFN-ɣ production of IEL QFLTg cells was in keeping with their natIEL phenotype - antigen-experienced yet functionally quiescent (Cheroutre et al., 2011; Denning et al., 2007). We conclude that Vα3.2⁺QFL T cells in the spleen are functionally innate-like, whereas the population with the same TCR specificity in the small intestine shows functional features of natIEL.

The establishment and function of the intestinal immune system is intimately associated with homeostasis of the gut microbial community, as evidenced by an altered gut immune cell composition in germ-free (GF) mice (Belkaid et al., 2014; Macpherson et al., 2004; Round and Mazmanian, 2009). To investigate if an association exists between gut microbiota and gut QFL T cells, we compared the numbers and frequencies of QFL T cells in the spleen and siIEL compartment of naïve specific-pathogen-free (SPF) and GF WT mice of various ages. First, we found the absolute number of QFL T cells in the spleen was unaffected by the absence of gut microbiota in either young or old mice. The frequency of QFL T cells was relatively higher in GF than in SPF mice, as a result of a decreased number of non-QFL CD8⁺ T cells in the spleen of GF mice (Fig.5 a, b). However, the percentage of CD44^hi cells within the total CD8⁺ T population was proportionally higher in the spleen of GF WT than in SPF WT mice regardless of their Vα3.2 expression (Supplementary Fig. 6 a). This observation suggested that gut microbiota might be more closely associated with the presence of conventional naïve CD8⁺ T cells rather than the memory phenotype CD8⁺ T population in the spleen. Second, while the number and frequency of QFL T cells in siIEL compartment showed no significant difference between GF and SPF mice of relatively young age (8~17-week), the population was gradually lost in old GF mice (18~22-week) (Fig.5 c, d), with the decrease in GF mice starting from around 16 weeks of age (data not shown). In addition, the QFL T cells detected in the spleen or siIEL compartment of young SPF or GF mice showed similar expressions of Vα3.2, CD44 and CD8αα (Supplementary Fig. 7). We conclude that although QFL T cells could home to gut in the absence of microbiota, microbe-derived antigens or signals are required for their maintenance there. Additionally, we observed that the average percentage of CD8αα⁺ cells among the total TCRβ⁺CD4⁻ T population was proportionally larger in GF mice than SPF mice regardless of Vα3.2 expression, although the Vα3.2⁻ population showed a more significant increase (Supplementary Fig. 6 b). This was mainly a result of a decreased number of CD8αβ⁺ T cells in the siIEL of GF mice (data not shown). Nevertheless, here we showed that Qa-1^b restricted T cells could home to the small intestine epithelium and display a natIEL phenotype independent of gut microbiota.

We next investigated how the QFL T population was retained in the gut of SPF WT mice, hypothesizing that the presence of specific antigen might play a role. QFL T cells specifically recognize the Qa-1^b-FL9 complex which is uniquely presented on ERAP1 deficient cells. Because it is unlikely this ligand is constantly presented in gut epithelium under homeostatic conditions (Supplementary Fig. 8 a), we reasoned that QFL T cells might be retained in the gut through exposure to FL9 homologue peptide(s) expressed by commensal bacteria that colonize small intestine.

To identify potentially cross-reactive peptides, we first determined the key residues of the FYAEATPML(FL9) peptide that could potentially affect activation of QFL T cells. Due to the lack of structural information on QFL-TCR:FL9:Qa-1^b ternary complex, we started by predicting the structure using AlphaFold2 (Mirdita et al., 2022). The sequences of the α1 and α2 domain of Qa-1^b α-chain, FL9 and Vα3.2Jα21, Vβ1Dβ1Jβ2-7 were submitted as input. Most of the residues in the predicted structure have high to intermediate confidence scores as was measured by the predicted local distance difference test (pLDDT). The output structure model of the top ranked pLDDT score (‘rank_1’ = 90.8) was selected for further visualization (Supplementary Fig. 8 b). The simulated QFL-TCR:FL9:Qa-1^b complex model suggests a similar TCR docking as observed in TCRαβ complexed with classical MHC I (Fig.6 a). While we fully appreciated the limitations in accurately predicting TCR docking, prediction of the pMHC complex is nevertheless relatively reliable based on the well-defined principles (Nielsen et al., 2020). Superimposition of simulated FL9:Qa-1^b model onto the crystallographic structure of Qdm:Qa-1^b (PDB 3VJ6) suggested similar main chain conformations between FL9 and Qdm (Fig.6 b). We thus refer to the reported structure of Qdm:Qa-1^b whose anchoring residues were P2, P3, P6, P7 and P9 of AMAPRTLLL(Qdm) for determination of putative key residues on FL9 (Zeng et al., 2012). Among the rest of the non-anchoring residues, we hypothesized that the glutamic acid at P4 of FL9 may be critical for interacting with the QFL TCR. Notably, the importance of P4 for TCR interaction was also highlighted in recently published studies where crystal structures of several disease-related TCR:pMHC I complexes were reported including an αβTCR complexed with an HIV peptide loaded in HLA-B5301 (Li et al., 2023) and a pre-TCR complexed with vesicular stomatitis virus octapeptide(VSV8) loaded in H-2K^b (Li et al., 2021). To test the hypothesis, we used the QFL T hybridoma cell line BEko8Z which expresses TCR-induced β-galactosidase (LacZ) to test whether the mutant FL9 peptide with the P4 residue substituted by alanine altered T cell responses. The P6 threonine whose side chain basically bends towards to the binding groove with relatively limited access to TCR engagement was tested in parallel as a putative anchoring residue. By measuring the LacZ production of BEko8Z cells in response to Qa-1^b-expressing Lmtk⁻ (L-Qa-1^b) cells pulsed with FYAAATPML (FL9-P4A) or FYAEAAPML (FL9-P6A), we found that in agreement with the prediction, replacement of the P6 threonine reduced TCR recognition, while replacement of the P4 glutamic acid led to the complete loss of the BEko8Z response (Fig. 6c). Additionally, we performed Qa-1^b binding predictions for various FL9 mutant peptides using NetMHCpan4.1 (Reynisson et al., 2020). The predicted binding scores suggested that substitution of the P9 Leu with Ala should decrease binding of FL9 peptide to Qa-1^b, whereas mutation of the rest of the residues were less likely to have such effect. Hence, Leu at P9 position may serve as a key anchoring residue (Supplementary Fig. 8 c). Thus, P4, P6 and C-terminus positions were considered the key residues with the P4 Glu being particularly critical for QFL T recognition. Based on these data, the presence of glutamic acid at P4 was considered to be important, while the C-terminus leucine and P6 threonine were considered secondary. Although it is possible that the residues that were not tested in the T cell assays are involved in peptide binding or TCR interaction, data on the P4, P6 and P9 residues of FL9 peptide were sufficient for determining peptide homology and establishing a testable library of bacterial FL9 homologs.

To identify potential bacterial FL9 homologous peptides, we aligned the FL9 peptide sequence with proteomes of the five dominant gut commensal bacterial phyla, including Actinobacteria, Bacteroidetes, Firmicutes, Proteobacteria, Verrucomicrobia from the NCBI Swissprot Non-redundant UniProtKB database (Donaldson et al., 2016). The alignment result was further curated to 30 candidate peptides based on their homology with FL9 peptide as described above (Fig. 8 d, Supplementary Table.1). By measuring LacZ production of BEko8Z cells in response to L-Qa-1^b cells pulsed with 50nM of each candidate peptide, we identified a single stimulatory peptide FYAEDGTPIL (FL10) that is expressed in P. pentosaceus, a gram-positive lactic acid bacteria (LAB) that colonizes in the small intestine (Fig.6 e). While sensitivity to FL10 was~1,000 folds less than FL9 peptide, BEko8Z cells could recognize FL10 at concentrations as low as 1.5nM. In contrast, FYAEDDTPIV(FV10), a FL10 homologous peptide expressed in Lactobacillus johnsonii which is a common LAB that colonize GI tract, failed to activate BEko8Z cells (Fig.6 f, Supplementary Fig. 8 d).

To further investigate the association between P. pentosaceus and Vα3.2⁺QFL T cells in gut, we first generated QFLTg_WT partial hematopoietic chimeras using germ-free WT mice (GF QFLTg_WT) and monocolonized the GF chimera mice with P. pentosaceus at 8 weeks of age. The mice were analyzed 8 weeks later (Fig.7 a). The donor-derived QFLTg cells and the recipient-derived Vα3.2⁺QFL T cells were distinguished by GFP expression and further defined as the CD45⁺CD19⁻TCRβ⁺CD4⁻Vα3.2⁺QFL-Dex-PE⁺ population on flow cytometry as was described earlier. While the donor-derived QFLTg populations were present and phenotypically unaltered in the spleen and siIEL compartment of GF QFLTg_WT chimera mice at 8 weeks of age, they were barely detectable in the GF chimera mice at 16 weeks of age (Supplementary Fig. 9, Fig.7 b). Notably, monocolonization of the GF QFLTg_WT chimera mice with P. pentosaceus completely restored the donor-derived QFLTg population in both tissues in 16-week GF chimera mice (Fig.7 b). We then further tested whether restoration of the gut QFL T population was specifically associated with P. pentosaceus by monocolonizing GF WT mice with P. pentosaceus or L. johnsonii (Fig.7 a). Analysis of the microbiome composition of fecal pallets showed that both species efficiently colonized the GF mice (Fig.7 c). While the endogenous Vα3.2⁺QFL T population in the siIEL compartment was retained by P. pentosaceus colonization of GF mice at 16 weeks of age, L. johnsonii colonization failed to restore the population to level comparable with SPF or P. pentosaceus colonized mice. On the other hand, maintenance of the endogenous splenic QFL T population was independent of microbes, in line with our earlier observations (Fig.7 d). Donor-derived QFLTg and endogenous Vα3.2⁺QFL T cells in the spleen likely showed discrepant results due to different sources of precursors for the two populations. Unlike the endogenous populations which were constantly replenished, the donor-derived QFLTg cells rose from a fixed number of bone marrow cells, and gradually decreased in number as the animal age (Supplementary Fig. 10). Thus, presence of the ligand which was supplied by peptide derived from P. pentosaceus might be required for the donor-derived QFLTg cells to be retained in the spleen, whereas the endogenous splenic population was unaffected by microbes. Phenotype analysis revealed that both the donor-derived QFLTg cells and the endogenous Vα3.2⁺QFL T cells from the colonized GF chimera mice showed a memory phenotype identical to that seen in the SPF chimera mice (Supplementary Fig. 11). We thus conclude that the late-life maintenance of Vα3.2⁺QFL T cells in the small intestine was dependent on the commensal bacterium P. pentosaceus which expresses a FL9 homologue peptide. While the presence of the bacterial FL10 peptide likely plays an important role in retention of gut QFL T cells, we do not exclude the possible contributions of other commensal bacterial factors. In addition, the unconventional phenotype of QFL T cells was likely acquired through microbe-independent mechanisms.

SPF wild-type C57BL/6J, H-2T23-deficient Qa-1^b-KO, C57BL/6-Tg (TcraTcrb)1100Mjb/J OT-1, or C57BL/6-Tg (UBC-GFP)30Scha/J B6.GFP mice were obtained from the Jackson Laboratory. QFL T TCR transgenic mice QFLTg were generated in the laboratory of E. Robey (University of California Berkeley, Berkeley, CA) and housed in our facility. GFP⁺QFLTg mice were generated in our facility by crossing B6.GFP mice with QFLTg mice. SPF QFLTg_WT, SPF OT-1_WT and GF QFLTg_WT partial hematopoietic chimera mice were generated as previously described(Ladi et al., 2008). GF wild-type C57BL/6J and GF QFLTg_WT chimera mice were generated and maintained in the Johns Hopkins Germ-free Mouse Core Facility. Mice were housed and all procedures were done in accordance with protocols approved by Animal Care and Use Committee of the Johns Hopkins University School of Medicine.

The QFLTg TCR α-and β-chain sequences were cloned and amplified from the genomic DNA of BEko8Z hybridoma(Guan et al., 2017; Nagarajan et al., 2012). The TCR α-chain was amplified with the forward primer 5’-AAAACCCGGGCCAAGGCTCAGCCATGCTCCTGG-3’ and the reverse primer 5’-AAAAGCGGCCGCATACAACATTGGACAAGGATCCAAGCTAAAGAGAACTC-3’. The TCR β-chain was cloned with the forward primer 5’-AAAACTCGAGCCCGTCTGGAGCCTGATTCCA-3’ and the reverse primer 5’-AAAACCGCGGGGGACCCAGGAATTTGGGTGGA-3’. The TCR α-chain DNA fragment was cloned into pTα cassette vector by inserting it between the Xmal and NotI sites, while the TCR β-chain DNA fragment was cloned into pTβ cassette vector in between the XhoI and SacII sites(Kouskoff et al., 1995). The ampicillin resistance gene was removed from pTα and pTβ cassette by EarI enzyme digest. The QFLTg mice were generated on C57BL/6J background in the Cancer Research Laboratory Gene Targeting Facility at UC Berkeley under standard procedures. Founder mice were identified by flow cytometry and PCR genotyping of tail genomic DNA using primers mentioned above.

Antibody for flow cytometry were from BioLegend (anti-CD45(30-F11), anti-CD19(1D3/CD19), anti-TCRβ(H57-597), anti-CD8β(53-5.8), anti-Vα3.2(RR3-16), anti-Vα2(B20.1), anti-CD90.2(53-2.1), anti-CD5(53-7.3), anti-IFN-ɣ(XMG1.2), anti-CD62L(MEL-14) and BD Biosciences (anti-CD4(RM4-5), anti-CD8α (53-6.7), anti-CD44(IM7)). BEko8Z, L-Qa-1^b or Lmtk⁻ cells were maintained as previously described(Nagarajan et al., 2012). Peptides were obtained from GenScript. The purity of SL8, FL9 and FL10 peptides were ⩾98%.

Qa-1^b-FL9 (QFL) monomers were synthesized by the Tetramer Core Facility of the US National Institutes of Health. Phycoerythrin (PE)- or allophycocyanin (APC)-conjugated streptavidin were obtained from Agilent. The QFL dextramers were generated following the ‘dextran doping’ technique(Bethune et al., 2017). The QFL monomers were incubated with PE or APC conjugated streptavidin at a molar ratio of 3:1 at 4℃ for 10min followed by addition of biotinylated dextran molecular weight 500kDa at a molar ratio of 1:20 with respect to streptavidin. The mixture was further incubated at 4℃ for ≧1h before being used in experiments.

Small intestinal IELs were isolated following an established protocol with minor modifications (Qiu et al., 2018). In brief, the small intestine with Payer’s patches removed were cut into appropriate length. The fecal content and mucus were removed by expelling with the flat side of forceps followed by flushing with PBS. The intestine was cut open longitudinally to reveal the epithelium and further cut laterally into ~2cm pieces. The intestine pieces were then placed in 25ml warmed dithioerythritol (DTE) solution (Ca²⁺- and Mg²⁺-free Hanks balanced salt solution, HEPES-bicarbonate buffer, 10% FCS) in 50ml conical tube and were shaken at 75rpm 37℃ for 20min. The tube was vortexed for 10s before the supernatant was transferred into a new 50ml conical tube through a 70µM cell strainer. The cells were palleted and resuspend in 44% Percoll solution (44% Percoll in RPMI 1640). Density gradient was generated by underlaying the cell suspension with 67% Percoll solution (67% Percoll in RPMI 1640). The gradient cell suspension was further centrifuged at 1600 × g for 20 min at RT without using the brake. The layer of cells at the 44% and 67% interphase were thus collected as IELs.

Splenic dextramer-positive cells were enriched as previously described(Nagarajan et al., 2012). IELs were resuspend in 100µl sorter buffer (0.1% sodium azide and 5% FCS in PBS). PE or APC labeled QFL dextramers were added at a final dilution of 1:100. Cells were incubated at room temperature for 45 min, then were washed twice with 3ml of sorter buffer. 20µl of anti-PE and 20µl of anti-APC microbeads (Miltenyi Biotec) were added into cells resuspended in 450µl of sorters buffer, followed by incubation of 20 min at 4℃. Cells were washed twice with 3ml sorter buffer. The PE- and APC-labeled cells were positively selected by passing through LS magnetic columns (Miltenyi Biotec). The entire isolated population was stained with anti-CD45, anti-CD19, anti-TCRβ, anti-CD4, anti-CD8α and anti-CD8β. QFL T cells were gated as CD45⁺CD19⁻TCRβ⁺CD4⁻QFL-Dex-PE⁺QFL-Dex-APC⁺ population. The absolute numbers and frequencies of the cells were calculated based on a fixed number of CountBright Beads (Thermo Fisher) added into each sample.

Splenocytes or small intestinal IELs isolated from QFLTg or OT-1 mice were depleted of non-T cells using Pan T isolation Kit II (Miltenyi Biotec). The negatively selected T cells were then labeled with 5µM Cell Tracing Violet (CTV) (Thermo Fisher) and adjusted to the concentration of 10⁶/ml. 10⁶ of the CTV labeled T cells were stimulated in vitro for 3 or 5 days in cultures containing 100ng/ml IL-15 (BioLegend) or a combination of 100ng/ml IL-7 (BioLegend) and 100ng/ml IL-18 (BioLegend) in 24-well plate at 37℃. Proliferation of cells were measured by dilution of CTV on flow cytometer.

Lymphocytes isolated from the spleen or siIEL compartment of naïve QFLTg_WT or OT-1_WT chimera mice were cultured in vitro with 2µM of FL9 or SL8 peptide respectively together with GolgiPlug (BD Biosciences) for 4.5h. WT splenocytes were supplemented into siIEL cultures as APCs. Cells were then washed and stained for surface markers. For measurement of IFN-ɣ production, cells were fixed and permeabilized using Fixation/Permeabilization Kit (BD Biosciences) and stained for intracellular IFN-ɣ.

10⁵ of BEko8Z hybridoma cells were cocultured with 10⁵ of L-Qa-1^b or Lmtk-cells in medium containing peptides of indicated concentrations in 96-well plates at 37℃ for 12 hours. T cell response was determined by cleavage of the chromogenic LacZ substrate chlorophenol red-β-D-galactopyranoside (CPRG, Millipore Sigma) by TCR-induced LacZ and measure by spectrophotometry as presented by absorbance at 595 nm (A595) (Sanderson et al., 1994). Background signals were subtracted by measurement of absorbance at 655nm.

Pediococcus pentosaceus Mees (33314) and Lactobacillus johnsonii (33200) were obtained from ATCC. The bacteria were grown overnight at 37℃ in 416 Lactobacilli MRS Broth (BD Biosciences). GF QFLTg_WT chimera mice were gavaged with 200µl of 2×10⁸/ml CFU of P. pentosaceus Mees at 8 weeks of age. GF WT mice were gavaged with 200µl of 2×10⁸/ml CFU of P. pentosaceus Mees or L. johnsonii at 8 and 10 weeks of age. Fecal pallets collected from the colonized mice were homogenized in 416 Lactobacilli MRS Broth, and the associated mice were determined by plating the appropriate dilution of homogenate on 416 Lactobacilli MRS agar plates. Bacterial colonization efficiency was determined by whole shotgun metagenomic sequencing of fecal pallets collected from the colonized mice at 16 weeks of age (TransnetYX Microbiome).

The predicted structure of QFL TCR:FL9:Qa-1^b complex was obtained by submitting the sequences of the α1 and α2 domain of Qa-1^b α-chain, FL9 and Vα3.2Jα21, Vβ1Dβ1Jβ2-7 as ‘hetero-oligomer’ to ColabFold v1.5.2-patch: AlphaFold2 using MMseqs2 at https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb. The prediction was conducted using the alphafold2_multimer_v3 model on default settings. Visualization and alignment of the protein structures was conducted on the PyMOL software.

The Qa-1^b binding prediction was conducted by submitting the sequences of Qdm, FL9 or mutant FL9 peptides and Qa-1^b (UniProt: P06339) to NetMHCpan4.1 at https://services.healthtech.dtu.dk/services/NetMHCpan-4.1/.