Butyrophilin-like proteins display combinatorial diversity in selecting and maintaining signature intraepithelial γδ T cell compartments

Butyrophilin-like (Btnl) genes are emerging as major epithelial determinants of tissue-associated γδ T cell compartments. Thus, the development of signature, murine TCRγδ+ intraepithelial lymphocytes (IEL) in gut and skin depends on Btnl family members, Btnl1 and Skint1, respectively. In seeking mechanisms underlying these profound effects, we now show that normal gut and skin γδ IEL development additionally requires Btnl6 and Skint2, respectively, and furthermore that different Btnl heteromers can seemingly shape different intestinal γδ+ IEL repertoires. This formal genetic evidence for the importance of Btnl heteromers also applied to the steady-state, since sustained Btnl expression is required to maintain the signature TCR.Vγ7+ IEL phenotype, including specific responsiveness to Btnl proteins. In sum, Btnl proteins are required to select and to maintain the phenotypes of tissue-protective γδ IEL compartments, with combinatorially diverse heteromers having differential impacts on different IEL subsets.

F rom jawless vertebrates through to humans, many extralymphoid tissues harbour distinct immune cell populations. Those include tissue-resident memory (T RM ) cells that infiltrate tissues after antigen priming in lymphoid tissues, and remain well-placed to respond to local antigen recurrence 1 . In addition, various myeloid and lymphoid cells, including macrophages 2 , T-regulatory cells 3 and γδ T cells become associated with tissues developmentally, remaining in situ lifelong [4][5][6][7][8] . Such cells are implicated in protecting tissue integrity, and γδ T-cell deficiency is causally linked to cancer, tissue inflammation and defective wound healing 9-12 . In addition, the molecular phenotypes of local T cells commonly emphasise their relationships to specific anatomical sites 3 . Thus, murine γδ T cells, which were revealed some 30 years ago to be prototypic tissue-associated T cells, display tissue-restricted T-cell receptor (TCR) repertoires, including Vγ5Vδ1 in the epidermis, Vγ6Vδ1 in the uterus, dermis and lung, and Vγ7 + cells expressing a variety of Vδ chains in the small intestine 13 . Nonetheless, how such TCRs contributed to tissue protection remained enigmatic, particularly given that Vγ5Vδ1 + dendritic epidermal T cells (DETC) were shown to use innate receptors, specifically NKG2D, to respond rapidly to epithelial cell dysregulation 14,15 .
Recently however, signature TCRs were shown to mediate the tissue-specific selection of γδ T cells by members of the heretofore enigmatic butyrophilin-like (Btnl) subfamily of B7 genes. Thus, Btnl1 −/− mice mostly lack intestinal Vγ7 + cells 16 , while mice deficient in Skint1 (a Btnl-related gene) specifically lack Vγ5Vδ1 + DETC 17,18 . The conservation of this biology became evident when human colonic Vγ4 + cells were shown to be specifically regulated by BTNL3 16,19 , while Butyrophilin 3A1 (BTN3A1) and BTN2A1 20 were found to be critical for signature responses of human peripheral blood Vγ9Vδ2 + cells to low molecular mass phosphoantigens such as isopentenyl pyrophosphate [21][22][23] . Moreover, dysregulation of the BTNL3-Vγ4 + axis has been implicated in celiac disease 24 . Thus, there is considerable interest in the mechanisms by which Skint/Btnl/BTN genes exert their effects.
Consistent with their regulation of γδ T-cell subsets defined by their TCRs, Btnl/BTNL proteins have emerged as bona fide T-cell selecting ligands akin to MHC or CD1. In addition, evidence from cell culture and biochemical experiments argues that Btnl/BTNL/ BTN proteins may exert their impacts as heterodimers of Btnl1 + 6, BTNL3 + 8, and BTN3A1 + 2A1, respectively 16,19,20,25 . Nonetheless, the functional significance of heteromers has not been universally accepted 26 , with one concern being that the most compelling evidence is based on cellular over-expression systems 27,28 .
This study has addressed this important issue by use of genetics. By showing that Vγ5Vδ1 + DETC development depends on Skint2 as well as on Skint1, and that Vγ7 + intestinal IEL development depends on Btnl6 as well as on Btnl1, we now provide formal genetic evidence that single Btnls are not sufficient for IEL selection. Most unexpectedly, however, different Btnl pairings had differential effects on IEL with different TCRs, revealing a potential for combinatorial diversity that could finely tune IEL repertoire composition. The major impacts of Skint1 and Btnl1 on IEL maturation occur during narrow time-windows in early life. Beyond this, the sustained expression of Btnl genes is herewith shown to be required to maintain signature intestinal IEL phenotypes. In sum, epithelial Btnl proteins mediate a sustained and complex regulation of local γδ T-cell compartments.
The cells with a TCR most closely related to Vγ5Vδ1 + DETC are uterine and lung γδ T cells expressing Vγ6 paired with a Vδ1 chain identical to that in DETC. In the absence of a generally available Vγ6-specific antibody, such cells were identified as TCRγδ + Vγ1 − Vγ4 − Vγ5 − , and in Skint2 Δ/Δ mice such cells were largely unaffected ( Supplementary Fig. 1d,e), again phenocopying Skint1 hypomorphs 30 . Collectively, these genetic data show that Skint2 as well as Skint1 is critically required for the specific maturation of Vγ5Vδ1 + DETC progenitors, supporting the hypothesis that discrete γδ T-cell compartments are naturally regulated by Btnl heteromers.
Indeed, the capacity of Skint1 and Skint2 to form physical complexes in vitro and in vivo was validated when anti-Skint1 immunoprecipitates from 293T cells transfected with Nterminal Flag-tagged Skint1 and HA-tagged Skint2 were shown to contain both Skint1 and Skint2, as detected by western blot (Supplementary Fig. 1f). Moreover, anti-Skint1 and anti-Skint2 antibodies could detect Skint1 and Skint2, respectively, in western blots of anti-Flag immunoprecipitates from thymi of transgenic mice expressing an N-terminal Flag-tagged Skint1 construct (NF-Skint1 Tg ) 32 , but not from non-transgenic FVB mice ( Fig. 1g; long arrows). Note that the detection of anti-Flag antibody chains in the FVB lysates ( Fig. 1g; asterisks) validated protein loading. Moreover, the specificity of Skint1 and Skint2 detection in the immunoprecipitates was verified by the failure to detect actin in anti-Flag immunoprecipitates, despite its detection in total input protein (Fig. 1g, lowest panel). The failure to detect Skint1 or Skint2 in total input protein is consistent with their very low levels of protein expression 32 . The inefficiency of Skint1/Skint2 elution from the beads seemingly reflects a greater affinity of the anti-Flag antibody for Flag-tagged Skint proteins versus Flag peptide. This notwithstanding, the data show an evident capacity of Skint1 and Skint2 to associate in cell lines and in primary mouse tissue.     Fig. 2a,b). The intestinal IEL compartment of adult Btnl1 Δgut mice strikingly phenocopied the complete Btnl1 knockout, displaying substantial and significant reductions of Vγ7 + cells, and of Vγ7Vδ4 + cells that we previously showed to be particularly affected by Btnl1 deficiency 16 : hence, the profound γδ-regulatory impact of Btnl1 seems attributable exclusively to IEC (Fig. 2b, c). Most unexpectedly, however, Btnl6 Δ/Δ and Btnl6 Δgut mice showed an overtly intermediate phenotype, with Vγ7 + IEL and Vγ7Vδ4 + IEL significantly reduced relative to either C57BL/6 or Btnl6 fl/fl mice, as measured by a 33-50% decrease in their percentage representation among gut γδ cells or by a 2-fold drop in absolute numbers of Vγ7 + IEL, by contrast to the cells' almost complete loss from Btnl1 Δgut mice ( Fig. 2b-d).
The few residual IEL in Btnl1 Δgut mice showed major dysregulation of the signature Vγ7 + IEL phenotype, with many cells displaying low expression of CD122 (the IL-15R chain) and high Thy1 (CD90) expression (Fig. 2d, e; Supplementary Fig. 2c). Strikingly, such dysregulation was not true for Btnl6 Δgut or Btnl6 Δ/Δ mice, in which residual IEL showed comparable phenotypes to controls (Fig. 2d, e; Supplementary Fig. 2c). Furthermore, whereas the Vγ7Vδ4 TCR mean fluoresecence intensity (MFI) was somewhat lower in Btnl1-deficient mice, consistent with defective selection 16 , it was unaltered in Btnl6 Δgut and Btnl6 Δ/Δ mice (see Fig. 2c). Hence, Btnl1 and Btnl6 differentially affected Vγ7 + IEL development, with Btnl6 required for the normal size of the IEL compartment, but not for the acquisition of the signature phenotype by residual Vγ7 + IEL. Moreover, the unique phenotype of Btnl6 Δ/Δ mice was specific in that myriad immune subsets in the spleen were unaltered relative to controls (Supplementary Fig. 2d; Supplementary Table 1), as was reported for Btnl1 −/− mice 16 .
To test whether those Vγ7 + IEL that developed seemingly normally in Btnl6-deficient mice were Btnl-dependent, we generated mice lacking all three intestinal epithelial Btnls, by targeting sites upstream of the initiator ATG codons in Btnl1 and Btnl6, respectively (note that those genes are transcribed in head-to-head orientation), thereby deleting over 130 kb in between, including the Btnl4 gene (Fig. 2f). The resultant (Btnl146 INDEL ) mice expressed no detectable Btnl1, Btnl4 or Btnl6 transcripts, and there was also reduced expression of the Btnl2 gene that immediately flanks the deletion. Conversely, Psmb9, which is more distal to the recombination point was unaffected ( Supplementary Fig. 2e).
When Btnl146 INDEL mice were contemporaneously compared with Btnl1 Δ/Δ , Btnl6 Δgut and Btnl4-deleted mice (Btnl4 KOMP ) that we previously characterised 16 , it was clear that Btnl146 INDEL mice (colour-coded green in Fig. 2g-i) largely phenocopied the near-ablation of Vγ7 + IEL in Btnl1deficient mice (Fig. 2g, h). In both cases, residual Vγ7 + cells showed reduced Vγ7 and Vδ4 TCR MFI and failed to upregulate CD122 or downregulate Thy1, by comparison to IEL in control, Btnl4 KOMP or Btnl6-deficient strains (Fig. 2g, i). When we integrated data from large numbers of mice of the different strains described, it became clear that the strains' respective Vγ7 + IEL compartments were consistent and stable over time for >120 days ( Supplementary Fig. 2f).
In sum, further support for the heteromer hypothesis was provided by the developmental dependence of approximately one-third to one-half of intestinal Vγ7 + IEL on Btnl6 as well as Btnl1. Nonetheless, an unanticipated nuance was introduced in that there was a hierarchy of Btnl regulation, with Vγ7 IEL numbers depending almost completely on Btnl1, partially on Btnl6, and not on Btnl4, while the signature phenotypes of Vγ7 + IEL present in the different strains were largely dependent on Btnl1, but independent of either Btnl6 or Btnl4.
Vγ7 + IEL respond to different Btnl combinations. Biochemical and molecular evidence has shown that Btnl1 + Btnl6 function is mediated by Btnl6 engaging Vγ7, while Btnl1 acts as a critical chaperone 19,37 . Btnl4 has near-identity to Btnl6 in the region (CFG) that engages Vγ7 and both are diverged from Btnl1 (Fig. 4a, colour-coded orange, blue and red). To interrogate whether Btnl4 might substitute for Btnl6, we subjected primary IEL to co-culture with the MODE-K enterocyte cell line expressing either Btnl1 + Btnl6 (L1L6) or Btnl1 + Btnl4 (L1L4). The former pairing induced TCR downregulation and CD25 upregulation in Vγ7 + IEL from wt mice 16,19 (Supplementary Fig. 3b; top row, right panel), relative to which a significant albeit reduced effect was also induced by Btnl1 + Btnl4 (Supplementary Fig. 3b; top row, centre panel).
This response was further investigated by co-expressing Btnl1 in 293T cells with Btnl4 alleles mutated in each of three regions whose counterparts in Btnl6 are implicated in Vγ7 engagement (Fig. 4a) 19 . Specifically, the Btnl4 sequences were replaced by counterparts from Btnl1, in one case via the substitution of a single amino acid. Those mutations essentially ablated the TCR downregulation and CD69 upregulation ordinarily induced by Btnl1 + Btnl4 in a human T-cell line, J76, expressing a monoclonal Vγ7Vδ4 TCR (Fig. 4b). Thus, Btnl4 can phenocopy Btnl6 in co-operating with Btnl1 to regulate Vγ7 + IEL, albeit that no IEL depend on Btnl4 for their maturation, as shown above.
When the responding Vγ7 + IEL were further scrutinised, it became clear that in wt and Btnl4 −/− mice, the striking discrimination between Btnl1 + 6 and Btnl1 To further investigate Vγ7 + IEL regulation by Btnl1, Btnl4 and Btnl6, we expressed each separately and in combination in 293T cells. Surface display of Btnl6 was highly inefficient, but was rescued by co-expression of Btnl1 ( Supplementary Fig. 4a). Conversely, Btnl4 alone could travel to the cell surface ( Supplementary Fig. 4a): hence, Btnl6 and Btnl4 are not strictly comparable. Vγ7 + J76 transductants (see above) were strongly stimulated by cells co-expressing Btnl1 + 6 but not by cells expressing Btnl6 alone, nor by an admixture (Btnl1/Btnl6 sep) of cells expressing Btnl6 with cells expressing Btnl1 ( Supplementary  Fig. 4b, red). Conversely, 293T cells transduced with Btnl4 alone showed some capacity to stimulate Vγ7 + J76 cells, although this was clearly increased by co-expressing Btnl1, but not by stimulating with an admixture of cells separately expressing Btnl4 and Btnl1 ( Supplementary Fig. 4b, blue). These data evoke the activity of human BTNL3, a human Vγ4 + -TCR ligand, which alone can provoke human Vγ4 + -TCR downregulation, but whose effects are greatly amplified by BTNL8 co-expression 19 .
In sum, Btnl4 is evidently not required for Vγ7 + IEL selection, but its capacity to stimulate Vγ7 + IEL in vitro likely explains its capacity to select IEL, primarily Vγ7 + Vδ6.3 + , in Btnl6-deficient mice. The starkly different phenotypes of Btnl6-deficient and Btnl1-deficient mice, argues that any Btnl4-intrinsic capacity to select Vγ7 + IEL relies in vivo on co-expression with Btnl1. Added to this, our data show that the signature preferential responses of Vγ7Vδ4 + cells to Btnl1 + 6 was seen only in cells from mice in which Btnl6 was expressed. This conditioning might be enforced during developmental selection, and/or be maintained in the steady-state by Btnl heteromers expressed in epithelial cells that juxtapose mature IEL. However, there has not hithtero been formal evidence of a maintenance function for Btnl genes, beyond their roles in selection. We therefore investigated this by use of conditional knockout mice.
Phenotypic maintenance by Btnl1 and Btnl6. We crossed floxed Btnl1 and Btnl6 strains to tamoxifen-regulated Villin-Cre mice, in order to generate mice in which Btnl1 and Btnl6 were inducibly deleted in IEC. Indeed, there was sustained loss of Btnl1 and Btnl6 expression, as assessed by RNAscope at 8 days and 22 days Fig. 2 The intestinal IEL compartment is shaped by expression of distinct Btnl molecules. a Targeting strategy to generate conditional and constitutive Btnl1 and Btnl6 knockout mice. Depending on the Cre-transgenic strain used (i/ii) resulting animals are either ubiquitously deleted for the respective Btnl gene (Δ/Δ), or harbour a tissue-specific deletion (Δgut). Black: translated exons, grey: untranslated regions, blue triangles: loxP sites. b Quantification of Vγ7 + IEL (black gate in c) in Btnl1 and Btnl6-deficient strains. Data are mean ± SEM of ≥2 independent experiments. n ctrl : 32, n Btnl1-KOMP : 4, n Btnl1Δgut : 8, n Btnl6Δgut : 8, n Btnl6ΔΔ : 9. Statistical analysis: Kruskal-Wallace and Dunn's multiple comparison analysis. c FACS profiles of IEL preparations from animals of indicated genotypes, gated on TCRγδ + cells. The label Btnl1 Komp indicates germline Btnl1 KO animals generated by the IMPC that have been described 16 . Black gate: all Vγ7 + cells, green gate: Vγ7 + Vδ4 + cells. d Quantification of IEL: Vγ7 + IEL numbers (top), %Vγ7 + IEL (middle) and %CD122 + cells (bottom) in animals of indicated genotypes. Data are mean ± SD of a representative experiment. Top panel: n ctrl : 12, n Btnl1 : 2, n Btnl6Δgut : 2, n Btnl6Δ/Δ : 3. Statistical analysis: Kruskal-Wallace & Dunn's multiple comparison. Middle and bottom panel: n ctrl : 12, n Btnl1Δgut : 3, n Btnl6Δgut : 2, n Btnl6Δ/Δ : 3. Statistical analysis: Kruskal-Wallace & Dunn's multiple comparison. e Histogram for surface expression of CD122 in Vγ7 + IEL from animals of the indicated genotypes. f Scheme depicting the strategy to generate Btnl146 Indel mice. Short guide RNAs flanking the 5′ region of Btnl1 and 5′ region of Btnl6 were injected with HDR templates. Due to the nature of CRISPR/Cas9 the intervening region was excised and a Btnl146 INDEL mouse lacking the Btnl1-4-6 locus was created. Blue triangle: loxP site that was inserted due to the nature of the HDR template (see Methods). g FACS analysis of TCRβ − CD3 + IEL in Btnl1/4/6-KO and Btnl146 INDEL mice. Colours correspond to coloured bar graphs in (h) and (i). h Quantification of Vγ7 + IEL depicted in (g). Data are mean ± SEM of three independent experiments. n Ctrl : 12, n Btnll1Δ/Δ : 7, n Btnl4KOMP : 7, n Btnl6Δgut : 11, n Btnl146Indel : 10. Statistical analysis: Kruskal-Wallace & Dunn's multiple comparison. i Surface expression of CD122 (left) and Thy1.2 (right) in Vγ7 + IEL from animals of indicated genotypes.
following the start of 5 days' of tamoxifen administration (Fig. 5a), whereas there was no effect of tamoxifen treatment on mice lacking the relevant Cre allele (Fig. 5a; middle column). Durable loss of Btnl1 and Btnl6 expression suggested that gene deletion had occurred in enterocyte stem cells, as reported 36 . Over a 15-day period post tamoxifen-mediated Btnl1/Btnl6 deletion, no significant reduction was apparent in the representation of Vγ7 + intestinal IEL, particularly by comparison to the reduced numbers seen in constitutively deleted Btnl6 Δ/Δ mice and Btnl1 Δ/Δ mice (Fig. 5b). Thus, signature IEL could be maintained at steady-state for at least two weeks in the absence of either Btnl1 or Btnl6.
Nonetheless, to investigate whether there might be more immediate effects of Btnl1/Btnl6 deletion, we applied tamoxifen daily for 3 days, and examined IEL 3 days later (Fig. 5c). Within this short time-frame, CD122 expression was markedly reduced on a large percentage of Vγ7 + IEL in both Btnl1 fl/fl Vcre ERT2+ and Btnl6 fl/fl Vcre ERT2+ mice (Fig. 5d). While this echoed the limited expression of CD122 by residual Vγ7 + IEL in constitutive Btnl1 −/− mice, it seemed a priori to conflict with sustained CD122 expression in constitutive Btnl6 −/− mice (above). This conflict, however, was resolved by the finding that in mice acutely depleted of Btnl6, rapid CD122 downregulation was mostly limited to approximately half of Vγ7Vδ4 + IEL (Fig. 5d) (Fig. 5d), providing another example of the differential effects of Btnl proteins on different Vγ7 + IEL subsets.
In this regard, we hypothesise that Vγ7Vδ6.3 + IEL may have been selected on Btnl1 + 4 even in wt mice, with their CD122 expression likewise maintained by Btnl1 + 4; hence, they were essentially insensitive to acute Btnl6 depletion, phenocopying Vγ7Vδ6.3 + IEL and some Vγ7Vδ4 + IEL in constitutive Btnl6 −/− mice. Evidence in support of this hypothesis was provided by further analysis of Vδ usage by Vγ7 + IEL, which was essentially unaffected at 3 days following Btnl6 depletion, but which was significantly skewed toward Vδ6.3 + cells by day 56 (Fig. 5e, f). This would be consistent with natural IEL turnover favouring newly-maturing Vγ7 Vδ6.3 + IEL versus Vγ7Vδ4 + IEL, since following Btnl6 deletion, the former could more efficiently engage Btnl1 + 4.
The differential impacts of Btnl1, 4 and 6 on different IEL subsets might reflect their different spatio-temporal regulation. We therefore analysed single-cell RNA data available from studies in which distinct small intestinal populations were investigated. Consistent with our and others' studies 34,38 , all three Btnls were restricted to enterocytes and enterocyte progenitors ( Supplementary Fig. 5a) Fig. 5b), aligning with the distribution of γδ + IEL 39 . In sum, there was no obvious difference in spatiotemporal expression that might explain the proteins' differential effects, although there was an apparent hierarchy of RNA expression levels-Btnl1 >> Btnl6 > Btnl4-that evoked the hierarchy of the genes' effects on Vγ7 + IEL.
Response maintenance by Btnl1 and Btnl6. To further investigate the requirement for sustained expression of Btnl genes, we examined IEL at 54 days after gut epithelium-specific deletion of the whole Btnl1,4,6 locus, making comparisons with wt mice and constitutive Btnl146 INDEL mice (Fig. 6a). (Note, acute loss of Btnl1,4,6 could not be examined because of variable penetrance of locus deletion until 1-month post tamoxifen treatment.) Locus loss for~8 weeks again failed to diminish Vγ7 + IEL numbers (Fig. 6b, middle panel, light green bars), supporting the conclusion that steady-state maintenance of Btnl-selected Vγ7 + IEL numbers does not require sustained Btnl expression. Moreover, there was no significant increase in Vδ6.3 + cells consistent with there being no Btnl1 + 4 heteromers to promote their selective advantage (Fig. 6c). Interestingly, however, induced Btnl1,4,6 locus deletion also phenocopied constitutive Btnl146 INDEL mice in that the capacity of co-cultured Vγ7 + IEL to respond preferentially to Btnl1 + 6 versus Btnl1 + 4 was lost over time (Fig. 6d). Diminished responses to Btnl1 + 6 were seen for Vγ7Vδ4 + IEL and particularly for Vγ7Vδ6.3 + cells (Fig. 6e). This provides further support for the hypothesis that Btnl6 needs to be sustained to establish and to maintain the phenotype of cells that preferentially respond to Btnl1 + 6.
Discussion γδ T cells, particularly those residing within extralymphoid tissues, have been increasingly implicated in the regulation of tissue maintenance and protection against cancer [39][40][41][42][43][44] . Nonetheless, the cells' biologies remain poorly elucidated. Germane to this, a substantive advance was made by the discovery that different compartments of mouse and human γδ T cells are critically and specifically regulated by butyrophilin and butyrophilin-like (Btnl) proteins. Moreover, recent cell biological, molecular and biochemical data fuelled the hypothesis that the active forms of Btnl proteins may be heterodimers, although there was heretofore no formal evidence supporting this in vivo. This study now provides genetic evidence for the importance of Skint/Btnl proteins functioning collaboratively, as would be the case for heteromers. In addition, our approach has revealed some surprising findings that emphasise the importance of genetics in understanding cell regulatory mechanisms. Thus, signature murine skin γδ IEL are shown to depend upon Skint1 + Skint2 and the normal intestinal γδ IEL compartment shown to depend upon Btnl1 + Btnl6. However, whereas Btnl1 Δ/Δ mice lacked the great majority of Vγ7 + IEL,~50% were retained in different strains of Btnl6-deficient mice. In seeking to understand this unanticipated hierarchy of Btnl proteins, we identified a potential of Btnl4 to substitute for Btnl6. However, whereas Vγ7 + IEL from wt mice ordinarily responded better in vitro to Btnl1 + Btnl6 versus Btnl1 + Btnl4, this was not so in Btnl6-deficient mice wherein the compartment of mature Vγ7Vδ6.3 + and Vγ7Vδ4 + IEL, that was presumably selected by Btnl1 + Btnl4, responded comparably well to Btnl1 + Btnl4.
The CD122 hi phenotype of most Vγ7 + IEL was reduced when Btnl1 was acutely depleted, providing formal evidence that sustained expression of a Btnl gene-product is required to maintain the signature status of the wt γδ IEL compartment. By contrast, only a fraction of Vγ7 + IEL showed CD122 downregulation when Btnl6 was acutely depleted. Moroever, the unaffected cells were enriched in Vδ6.3 + cells, phenocopying the repertoire composition in constitutive Btnl6-deficient mice. These data are consistent with the hypothesis that whereas Btnl4 is not required for the selection and/or maintenance of any Vγ7 + IEL 16 , some IEL in wt mice have naturally selected on Btnl1 + Btnl4 while others selected on Btnl1 + Btnl6. Indeed, we propose that discrete Btnl heteromers ordinarily select those cells that respond most strongly to them and/or that they condition the responses of the cells they select. Thereafter, the Btnl heteromer on which the cells are selected is required to maintain the cells' signature phenotype. In sum, Btnl proteins operate in different combinations (i.e. show combinatorial diversity) in refining and regulating the composition of IEL compartments.
The biophysical basis for the preference of some cells for Btnl1 + Btnl6 versus Btnl1 + Btnl4 is unresolved. Btnl6 and Btnl1 physically associate, either directly or via an intermediate, and in this complex Btnl6 seemingly interacts directly with Vγ7 37 . Although there is currently less evidence available for the direct interaction of Btnl4 with Btnl1, it is likely that the two associate given the Btnl1-dependence of essentially all Vγ7 + IEL, and the capacity of Btnl1 co-expression to greatly increase the impact of Btnl4 on Vγ7 + IEL. In this regard, the association of BTN2A1 and BTN3A1, which are jointly required to regulate human Vγ9Vδ2 cells, only became evident after chemical cross-linking 45 . Nonetheless, some capacity of some mouse or human Btnl , derived with 3D-JIGSAW from the crystal structure of BTN3A1 (PDB accession code 4F80), with the same annotation as in (a). Side chains are displayed for the two residues that differ in the CFG face motifs of Btnl6 versus Btnl4 (Ala versus Glu, Glu versus Asp). b TCR downregulation (left) and CD69 upregulation (right) by J76 cells expressing a Vγ7Vδ4 TCR and co-cultured with 293T transiently transfected with Btnl1 in combination with Btnl4 wild-type (L4 WT ) or mutated in the CFG region as indicated on the X-axis. Results are normalised to 293T transfected with empty vector (EV). Data are represented as mean ± SD of duplicate co-cultures, representative of n = 2 independent experiments. FC, fold change. c Experimental setup to analyse IEL from various KO strains in co-cultures with MODE-K cells overexpressing either Btnl1 and Btnl4 (L1L4) or Btnl1 and Btnl6 (L1L6). IELs are isolated from indicated mouse strains which can display distinct combinations of Btnl molecules on the epithelial surface during development. Following isolation, IEL were co-cultured o.n. with MODE-K cells displaying either Btnl1+4 or Btnl1+6 on their surface. MODE-K cells transduced with empty vector (EV) were used as control. d IEL response to MODE-K cells expressing different Btnl dimers (L1L4 or L1L6) was measured by analysing CD25 + cells gated on Vγ7 + cells in animals of indicated genotypes. Data are mean ± SEM of five independent experiments, n ctrl : 13, n Btnl1KO : 8, n Btnl4KO : 9, n Btnl6KO : 10, n Btnl146Indel : 15. Statistical analysis: two-way ANOVA & Tukey's multiple comparison post test. e IEL response to MODE-K cells expressing different Btnl dimers (L1L4 or L1L6) was measured as %CD25 + cells and further gated on Vγ7 + Vδ4 + (left) or Vγ7 + Vδ6.3 + (right) cells in animals of indicated genotypes. Data are mean ± SEM of five independent experiments, n ctrl : 13, n Btnl4KO : 9, n Btnl6KO : 10. Statistical analysis: two-way ANOVA & Tukey's multiple comparison post test.
proteins (e.g. Btnl4; BTNL3), when over-expressed, to traffic to the cell surface and to regulate γδ IEL, albeit suboptimally, leaves open the possibility that non-heteromeric complexes might be active, e.g., in disease settings in which BTNL proteins might be dysregulated.
Btnl1-dependence has to date been attributed solely to Vγ7, and so an influence of Vδ chains on the Btnl response was a priori surprising. Possibly pairings with particular Vδ chains might affect the response by altering the quarternary structures of TCRs. Alternatively, Btnl1 + 4-responsive IEL may comprise qualitatively distinct cells whose responsiveness might reflect their development along a distinct pathway: indeed, Vδ6.3 expression has been associated with PLZF-expressing innate-like lymphocytes 46 . This issue will be addressed by single-cell transcriptomics.  Because the murine gut epithelium expresses Btnl1, Btnl4 and Btnl6, it is also not obvious why Btnl1 + Btnl6 is the dominant selecting combination, although this might reflect expression levels (considered above), a prospect which cannot be investigated at the protein level until appropriate reagents are available.
Intriguingly, Vγ7 + IEL numbers did not decline over many weeks following acute Btnl gene locus ablation. This was surprising given that the IEL showed reduced expression of CD122, the receptor for IL-15 which is an important IEL growth factor 47,48 . Possibly, Vγ7 + IEL were still able to compete for IL-15 because of the reduction in receptor expression by most such cells. Alternatively, the impact of reduced IL-15R expression on IEL might become evident by assessing the cells' replenishment in mice following infection or injury. The Btnl-dependence of sustained CD122 expression is also interesting in the light of reports that IL-15 regulates mucosal T-cell mobility within the gut, as part of immune surveillance 47,49,50 .
In this regard, an unanticipated observation was that although Vγ7 + IEL numbers were maintained in mice acutely depleted of Btnl6, the TCRδ repertoire changed toward that seen in constitutive Btnl6 −/− mice. This presumably reflects ongoing replenishment of the gut IEL the half-lives of which have been reported to range from 2 to 14 weeks 51 . We hypothesise that during a developmental window in early life, Btnl proteins are required to drive the selective differentiation and proliferation of IEL progenitors so that mature, expansive repertoires form. Thereafter, local self-renewal occurs from a mature progenitor pool, akin to that recently identified for memory CD8 T cells 52 , that does not require sustained Btnl expression but that is nonetheless influenced by it. Hence, somatic changes in Btnl expression patterns have the potential to change the IEL repertoire and status, as occurred in this study. This scenario may model human disease settings where BTNL protein becomes altered, e.g. by inflammation or other gut pathophysiology 24 . However, the consequences may be greater than in mouse, because to date the potential to make only one type of heteromer (BTNL3 + 8) has been identified in the human colon. Hence, the reduced expression of either BTNL protein, as has been reported in colon cancer (www.oncomine.org), might undermine the capacity to sustain the normal IEL repertoire and its functions in tissue maintenance, that have seemingly been conserved from agnathans to Homo sapiens. Finally, we note that future studies should investigate whether Skint/Btnl/BTNL heteromers exert cellautonomous effects on the epithelial cells that express them, outside of the impacts on their local lymphocyte compartments.
Mice. Wild-type (WT) C57BL/6J and FVB mice were obtained from Jackson Laboratories. NF-Skint1 Tg , Btnl1-KOMP, Btnl4-KOMP and Skint1 Δ/Δ mice have been previously described 16,29,32 . Genetically engineered mice were generated at the Francis Crick Institute's Transgenic Facility. The sg RNAs & PAM sequences (see Table 1) were cloned into the g-RNA basic vector, translated in vitro, purified and co-injected with Cas9 into day 1 zygotes and transferred into pseudopregnant foster mice by the Francis Crick Institute's Transgenic Facility. Targeted animals were identified and validated by PCR and later genotyped using the Transnetyx platform. All animals were maintained at The Francis Crick Institute's Biological resource facilities with a 12 h light/dark cycle and access to food and water ad libitum, temperature 19-23°C, 55 ± 10% humidity. Animal experiments were undertaken in full compliance with UK Home Office regulations and under a project license to A.C.H. (7009056).
Generation of Skint1, Skint2 rat monoclonal antibodies. Rat monoclonal antibodies against Skint1 and Skint2 were generated by immunization of Lou/c rats with purified GST-tagged human Skint1 or Skint2 extracellular domain, respectively. Hybridoma cells were generated and binding to Skint1 or Skint2 protein was analysed by enzyme-linked immunosorbent assay (ELISA). Positive hybridoma supernatants were further assayed for their potential in immunoblotting. Hybridoma clones Skint1 2G2 and Skint2 3G8 (both IgG2a/k) were recloned by limiting dilution to obtain stable monoclonal cell lines.
Quantitative RT-PCR. Samples were stored in RNAlater (Ambion) or directly frozen in RLT buffer prior to RNA purification with DNAse digest (QIAGEN RNeasy kit). cDNA was generated using Superscript-II (Invitrogen) and analysed using Sybr-green assay (Invitrogen) using a Quant-studio 5 or Viaa7 Real-time PCR machine (Applied Biosystems) and qPCR primers indicated in Table 2.
Isolation of mouse intestinal IEL. Mouse IEL were isolated from small intestine 18 . Briefly, small intestine was opened, washed in PBS, cut into 0.5-cm long pieces and    incubated at RT on awheel in complete RPMI supplemented with 1 mM DTT. Tissues were then washed, vortexed in complete RPMI and filtered through 70 nm nylon cell strainers. Vortexing and filtration steps were repeated twice. IEL were then purified by Percoll density centrifugation and stained by flow cytometry (for antibodies, see Table 3).
Plasmids and transfection. Overlap-extension PCR (OE-PCR) was used to replace the GFG regions of Btnl4 with those of Btnl1 on plasmids encoding Btnl1/ 4/6 16 . HEK293T cells were transfected with the indicated combinations of FLAG-Btnl1, HA-Btnl6, HIS-Btnl4 and empty vector (EV) encoding plasmids. Medium was replaced 16 h after transfection and cells were harvested at 48 h and used for the co-culture assay. For antibodies see Table 3.
Preparation of epidermal sheets. Ear epidermis was separated from dermis following incubation in 0.5 M ammonium thiocyanate for 35 min at 37°C. Isolated epidermal sheets were fixed with ice cold acetone at −20°C. The samples were blocked in 5% FCS for 1 h at room temperature and stained for 1 h at 37°C using Vγ 3 TCR-FITC (clone 536, BD), MHC I-A/I-E-AF647 (clone M5/114.15.2, Bio-Legend) and CD45-eFluor450 (clone 30-F11, eBioscience) antibodies. Tissue samples were mounted on microscope glass in Prolong Gold mounting medium under a stereomicroscope to ensure flat epidermal mounting. Confocal images were recorded using Leica SP5 confocal microscope with 40× 1.25 NA HCX PL APO CS lens. Three confocal records 387.5 × 387.5 µm size were acquired from each epidermal sheet. Image quantification was performed using Definiens Developer software (version XD2.7). Each channel in a record was processed with Gaussian filter followed by application of automated multi-threshold segmentation. Individual cells (CD45+, Langerhans cells, and T cells) were detected based on their relative intensity in CD45, MHC II and TCR channels, respectively. Cell number and morphology were measured for each cell type.
Preparation of lung and uterus γδ cells. Lungs and uteri from experimental mice were collected in medium and minced with razor blades. Samples were digested using Miltenyi Multi Tissue Dissociator kit 1, according to the manufacturer's instructions. Briefly, samples were transferred to GentleMACS C tubes containing 2.5 mL digestion mix (100 µL Enzyme D, 50 µL Enzyme R and 12.5 µL Enzyme A) and incubated at 37°C for 40 min with shaking. Following incubation, single-cell suspensions were prepared by homogenisation using GentleMACS program C and filtering through 70 µm cell strainers. Single-cell suspensions were stained with Live/Dead Aqua for dead cell exclusion, followed by Fc-block and surface stain with specific antibodies.
Biochemistry. Cells were lysed for 30 min in ice cold RIPA buffer with protease inhibitors (Roche) and phosphatase inhibitors (Phosphatase inhibitor cocktails 2 & 3, Sigma) and spun at 20,000 × g for 15 min at 4°C. Protein concentrations in supernatants was determined using a BCA kit (Pierce).
Immunoprecipitation from cell lysates. Lysates were precleared on Protein G Sepharose (Millipore-Sigma) for 1 h, incubated with antibodies for 1 h followed by incubation with 1% BSA blocked Protein G beads for a further hour. Following three washes in RIPA buffer immunoprecipates or samples were mixed with NuPAGE LDS Sample Buffer supplemented with 1x NuPAGE reducing agent and separated by electrophoresis on NuPage 4-12% Bis-Tris protein gels (Thermo Fisher) and then transferred onto PVDF membranes. Membranes were blocked in PBS-0.1% Tween20 and 5% BSA for 60 min at room temperature and incubated with primary antibodies overnight. Membranes were washed three times with PBS-0.1% Tween20 and incubated for 60 min with HRP conjugated secondary antibodies, washed again and developed using ECL detection reagents (Merck).
Immunoprecipitation from thymus. 12 Thymi of E17/E18 FVB and 22 Thymi of NF-Skint1 Tg pups were lysed as described above and incubated with Flag-M2 coated beads overnight at 4°C. Following three washes in RIPA buffer immunoprecipates were eluted with 3xFlag peptide to obtain the eluate or beads were directly boiled for analysis. Beads, Eluate or samples were mixed with NuPAGE LDS Sample Buffer supplemented with 1x NuPAGE reducing agent and separated by electrophoresis on NuPage 4-12% Bis-Tris protein gels (Thermo Fisher) and then transferred onto PVDF membranes. Membranes were blocked in PBS-0.1% Tween20 and 5% BSA for 60 min at room temperature and incubated with primary antibodies overnight. Membranes were washed three times with PBS-0.1% Tween20 and incubated for 60 min with HRP conjugated secondary antibodies, washed again and developed using ECL detection reagents (Merck).
Bioinformatics analysis. Raw gene counts were obtained from GSE109413 (Moor et al.) 54 and GSE92332 (Haber et al.) 55 and each cell-set was preprocessed using Seurat 56 (version 3.1.1.9023). In the case of data from Moor et al. data, cells with <200 and >3000 detectable genes and cells with percentage mitochondrial expression greater 5% were removed. For all cell-sets, the total counts were scaled to 1e4 counts, a log transformation applied and genes were z-score across all cells. In the case of data from Moor et al., both replicate cell-sets were merged using Seurat's IntegrateData function after which the gene-wise vectors were rescaled. The dot plots were produced using the DotPlot function from Seurat and profile plots were produced across the villus regions using the z-score scaled data.  Modelling software. Figures for all modelling data were generated in PyMOL v2.0.7 (Schrodinger LLC). 3D-JIGSAW was used to generate 3D models of proteins and perform docking simulations, respectively.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
This work did not include any data which mandated deposition in public databases. Associated raw data are provided in the main and/or supplementary figures. Relations to summary data charts are indicated and a full list of figures with associated raw data is provided in the reporting summary linked to this article. Raw gene counts were obtained from GSE109413 (Moor et al.) 54 and GSE92332 (Haber et al.) 55 . For bioinformatics single-cell analysis scripts are available on github: https://github.com/ajandke/ Jandke_etal_naturecomms. Immunophenotyping data for pipeline procedure can be found under https://www.mousephenotype.org/data/secondaryproject/3i.