The P5-ATPase ATP13A1 modulates MR1-mediated antigen presentation

The monomorphic antigen presenting molecule MHC-I-related protein 1 (MR1) presents small molecule metabolites to mucosal-associated invariant T (MAIT) cells. The MR1-MAIT cell axis has been implicated in a variety of infectious and non-communicable diseases and recent studies have begun to develop an understanding of the molecular mechanisms underlying this specialised antigen presentation pathway. Yet, the proteins regulating MR1 folding, loading, stability, and surface expression remain to be identified. Here, we performed a gene trap screen to discover novel modulators of MR1 surface expression through insertional mutagenesis of an MR1-overexpressing clone derived from the near-haploid human cell line HAP1 (HAP1.MR1). The most significant positive regulators identified included β2-microglobulin, a known regulator of MR1 surface expression, and ATP13A1, a P5-ATPase in the endoplasmic reticulum (ER) with putative transporter function not previously associated with MR1-mediated antigen presentation. CRISPR/Cas9-mediated knock-out of ATP13A1 in both HAP1.MR1 and THP-1 cell lines revealed a profound reduction in MR1 protein levels and a concomitant functional defect specific to MR1-mediated antigen presentation. Collectively, these data are consistent with the ER-resident ATP13A1 as a key post-transcriptional determinant of MR1 surface expression.

While the full spectrum of MR1 ligands is still being elucidated (2,(4)(5)(6)(7)(8)(9)(10), the best characterized MAIT-activating MR1 ligands are derivatives of pyrimidine intermediates of riboflavin biosynthesis, a pathway specific to certain fungi and bacteria and, thus, intrinsically non-self for humans (1,11). Accordingly, MAIT cells are activated upon recognition of MR1 ligands derived from a variety of microbes capable of riboflavin synthesis such as Escherichia coli (E.coli), Salmonella typhimurium, and Mycobacterium tuberculosis (12)(13)(14)(15). Molecularly, MAIT cells are characterized by high expression of the C-type lectin CD161, the interleukin (IL)-18 receptor, CD26 (16)(17)(18) and the canonical Vα7.2-Jα33/12/20 T cell receptor (TCR) α chain in humans (19)(20)(21). Owing to their restricted TCR repertoire, their relative abundance, and their effector-memory phenotype, MAIT cells are counted as an innate-like lymphocyte subset (22). Their high numbers and poised phenotype allow them to rapidly carry out effector functions in response to their cognate microbial antigens and place them at the interface between the innate and the adaptive immune system (22,23). Furthermore, the MR1 antigen presentation pathway is an attractive target for both therapeutic and vaccination strategies as the monomorphic nature of the presenting molecule renders it independent of individual variations in human leukocyte antigen (HLA) expression (24)(25)(26)(27)(28). A better understanding of this non-classical antigen presentation pathway is, thus, crucial for harnessing the protective potential of MAIT cells in such therapeutic approaches. MR1 transcript is detectable in almost all human cell lines and tissues but endogenous surface expression of MR1 protein is difficult to detect in most human cells even upon exposure to microbial MR1 ligand (29,30). Nevertheless, such low MR1 surface levels are capable of inducing a potent interferon-γ (IFNγ) response in MAIT cells (29,30). Since endogenous MR1 is difficult to detect, most of the current knowledge has been obtained in over-expression systems (30)(31)(32)(33)(34)(35). In one widely used model of MR1 overexpression, the majority of cellular MR1 resides in the ER in a partially folded, ligand-receptive state in the absence of exogenous or microbial ligands (34,36). Upon ligand binding, the MR1 heavy chain-β 2 -microglobulin (β 2 m)-complex translocates to the cell surface (34,37). ER egress is thought to require neutralisation of the positive charge on the lysine 43 (K43) residue within the MR1 ligand binding groove (34,37). In other studies, MR1 has been found to partially co-localise with LAMP1 + in endolysosomal compartments (35,38) and the relative contributions of these two intracellular pools to antigen presentation under physiological conditions remain to be determined (39). Importantly, a growing body of evidence suggests that the loading and trafficking of MR1 in the context of intracellular bacterial infection differs from the presentation of exogenous ligand (30,35,(39)(40)(41)(42).
In light of the ubiquitous expression of MR1 transcript and the observation that activating MR1 ligands can be derived from both pathogenic and commensal microorganisms, tight regulation of this highly sensitive antigen presentation pathway is required (31,41). Using a previously described functional genetic screen utilising the near-haploid human cell line HAP1 (43,44), we identified the P 5 -ATPase ATP13A1 as one of the proteins involved in the modulation of MR1 surface expression.

Results
A gene trap screen identifies the P 5 -ATPase ATP13A1 as a putative modulator of

MR1 surface expression
The near-haploid human cell line HAP1 is a powerful tool for genetic loss-of-function screens as only one copy of a gene needs to be mutated to achieve a functional knock-out (43). Since HAP1 WT cells did not express appreciable amounts of MR1 at the protein or transcript level (Supporting Information Figure 1), we transduced HAP1 cells with lentiviral particles encoding the MR1 complementary DNA (cDNA) sequence. To ensure that any difference in MR1 surface levels observed in the screen was due to gene editing rather than varying copy numbers of the MR1 cassette in the polyclonal HAP1.MR1 population, we single cell sorted MR1 + cells (Supporting Information Figure 1). The resulting HAP1.MR1 parent clones were analysed for ploidy and responsiveness to the MR1-stabilising ligand Acetyl-6-formylpterin (45) (Ac6FP) (Supporting Information Figure 1). The HAP1.MR1 parent clone D9 was expanded and transduced with a gene trap virus as previously described (44) (see Experimental Procedures for details). Subsequently, the polyclonal mutant population was pulsed with Ac6FP to induce MR1 surface translocation and surface stained for MR1. We then sorted the MR1 hi and MR1 low tails of the distribution by flow cytometry (Supporting Information Figure 2) and sequenced the DNA to identify the retroviral insertion sites (44). In total, >1.75x10 6 unique insertion sites were recovered for each sorted population with approximately 48% being sense insertions (Supporting Information Table 1). The sense insertions across both sorted populations mapped to 16,952 genes, covering about 85% of the currently predicted protein-coding genes in the human genome (46). Out of these, 199 genes were scored as putative regulators of MR1 surface expression (false discovery rate (FDR)-corrected p-value (fcpv) < 0.01), including the MR1 transgene itself and β 2 m (Figure 1, Supporting Information Table 2). Intriguingly, putative negative regulators included HLA-A as well as components of the peptide loading complex (PLC; TAP1, TAP2, and TAPBP, the gene encoding tapasin (47)) but also genes involved in the regulation of protein transport (e.g. TMEM131 (48) and RALGAPB (49)), endolysosomal trafficking and homeostasis (e.g. LAMTOR2 (50,51) and VAC14 (52,53)), N-terminal protein acetylation (e.g. NAA30 (54)), and ubiquitination (e.g. CUL3 (55,56)) (all highlighted in Figure 1). Similarly, the putative positive regulators comprised genes with diverse functions ranging from implication in immune responses (e.g. B2M (57,58) and IRF2 (59,60)) to endosomal recycling (e.g. VPS29 (61) and VPS53 (62)), cytoskeletal organisation (e.g. NHLRC2 (63,64)), glycosylation (e.g. GANAB (65) and SPPL3 (66)), mRNA processing (e.g. PARN (67) and ZCCHC14 (68)), and ion homeostasis in the ER (ATP13A1 (69)) (all highlighted in Figure 1). . Genes significantly (fcpv < 0.01) enriched in the high or low fraction are highlighted in yellow and blue, respectively, and represent putative negative and positive regulators of MR1 surface expression. MR1, its known regulator β2m (encoded by gene B2M), and selected genes based on functional annotations are highlighted (see text). Raw data are available in Supporting Information Table 2. The difference in the number of positive and negative candidate genes identified reflects the fact that the screen setup was more powerful in detecting putative positive regulators than negative ones. The screen was carried out in the presence of Ac6FP which results in the translocation of most MR1 molecules to the surface. Thus, a further increase in MR1 surface expression, as would be expected upon knock-out of a negative regulator, may be masked. Loss of a positive regulator, on the other hand, leads to reduced MR1 surface levels and was more easily detected in our system. Therefore, we focused our analysis on the putative positive regulators of MR1 surface expression.
One of the most prominent and functionally interesting putative positive regulators of MR1 surface expression was the P 5 -ATPase ATP13A1 ( Figure 1, Supporting Information Table   2). While information on mammalian ATP13A1 has been sparse, recent evidence points towards a dislocase function (73). The yeast homologue Spf1p (also known as Cod1p) localises to the ER and is involved in ion homeostasis (69,74). Consequently, loss of spf1 causes hypoglycosylation of proteins and results in ER stress (69,(74)(75)(76)(77)(78). In the context of MR1 antigen presentation, ATP13A1 was a particular promising target because of its localisation to the ER and its putative transporter function.

ATP13A1 modulates the size of the cellular MR1 pool
To investigate the role of ATP13A1 in MR1-mediated antigen presentation we generated HAP1.MR1 ATP13A1 knock-out (KO) clones. The same HAP1.MR1 parent clone used in the original screen (clone D9) was transiently transfected with plasmids encoding the Cas9 protein, the fluorescent protein mRuby, and one of three different sgRNAs targeting different regions of the ATP13A1 gene (Supporting Information Figure 4). Single mRuby + cells were sorted and individual clones screened for loss of ATP13A1 expression by Western blot (Figure 2A). Strikingly, total cellular MR1 levels were reduced in all but one HAP1.MR1 ATP13A1 KO clones regardless of the sgRNA used ( Figure 2A and Supporting Information Figure 5). A notable exception was clone 3-15 which displayed reduced cellular MR1 levels despite expressing ATP13A1 (see below and Supporting Information Figure 5).
Sequencing revealed that a six base pair deletion in clone 3-15 left the reading frame of ATP13A1 intact but altered a highly conserved region within the amino acid sequence (79) (Supporting Information Figure 6). The importance of this GxPF sequence is not only underscored by its conservation across yeast, murine and human ATP13A1 homologues (75,79,80) (NCBI Reference Sequences NP_010883.3, NP_573487.2, and NP_065143.2, respectively), but also by the observation that a construct lacking part of this motif was unable to rescue the spf1 KO phenotype in yeast (75). Alterations to this region appear to have resulted in the production of a non-or dys-functional protein, which could explain the "KO-like" phenotype of clone 3-15 (Supporting Information Figure 5). This makes clone 3-15 a very interesting control as it suggests that loss of ATP13A1 function and not merely expression of the protein is responsible for the MR1 phenotype.
When ATP13A1 expression was restored in a subset of clones comprising one KO clone for each sgRNA and clone 3-15, total MR1 levels were rescued, confirming that this phenotype is indeed attributable to loss of ATP13A1 and that the mutation found in clone 3-15 is not dominant negative ( Figure 2B). Testing the same three HAP1.MR1 ATP13A1 KO clones in T cell stimulation assays, we observed a reduction in MAIT cell activation compared to the parental clone ( Figure 2C). Importantly, ATP13A1 over-expression increased the antigen presentation capacity of all three HAP1.MR1 ATP13A1 KO clones and this trend was statistically significant at low antigen concentrations ( Figure 2C and Supporting Information Figure 7). Although statistical significance varied across experimental replicates, the trends reported here were consistent across sgRNAs used for knock-out, MAIT cell donors, and assays, together strengthening the general conclusion that the MR1 phenotype can be rescued (at least partially) by reconstitution of ATP13A1 expression.  Figure 5. B, ATP13A1 expression was reconstituted in a subset of clones using a lentiviral expression system and MR1 expression in the polyclonal transduced cell lines was analysed by Western Blot. C, the HAP1.MR1 parent clone, these KO clones and the reconstituted cell lines were pulsed with 5-A-RU+50 µM MG and incubated with sorted human MAIT cells for at least 36 h. IFNγ in the supernatants was quantified by ELISA. Each symbol represents the mean of technical duplicates for one clone per cell line and median is shown for the groups. Statistical significance of differences between ATP13A1 KO clones and reconstituted lines was analysed using two-tailed, paired t tests. The HAP1.MR1 parent clone D9 is shown for reference but was not included in statistical analyses. D + E, ATP13A1 sufficient (solid lines), KO (dashed lines), and reconstituted (filled) cells were incubated with 5 µg/ml of Acetyl-6-formylpterin (Ac6FP, blue) or DMSO (black) for 5 h before staining for MR1 (D) and MHC class I (E) surface expression. Isotype/fluorescence minus one (FMO) controls are from one sufficient, one KO, and one reconstituted clone each. F + G, B2M and MR1 transcript levels were compared in three ATP13A1 sufficient clones and three ATP13A1 KO clones. Each data point represents the mean of technical triplicates for one clone and mean and SD are shown for each group. Statistical significance of differences in transcript levels were analysed using an unpaired t test. Data in C + D are representative of three experiments for all clones except clone 3-1 and its reconstituted cell line although statistical significance varied across repeats (see Supporting Information Figure 7). Data in F + G are representative of two experiments. For HAP1 clones shown in C -G see Supporting Information Table 3. KO clones are highlighted in red. ns = p > 0.05. LoD = limit of detection.
Next, we compared one clone per sgRNA that was deficient for ATP13A1 ("ATP13A1deficient clones") to one clone per sgRNA that was mRuby + upon sorting but still expressed the protein ("ATP13A1-sufficient clones"). Details on editing of the clones used for the individual experiments can be found in Supporting Information Table 3 and Supporting Information Figure 6. Consistent with the results from the gene trap screen, surface levels of folded MR1 in HAP1.MR1 clones deficient for ATP13A1 were lower than those in their ATP13A1-sufficient counterparts both in the presence and absence of Ac6FP and independent of the sgRNA used to knock out ATP13A1 ( Figure 2D). As with the total protein content, the reduced MR1 surface expression could be reversed by over-expression of ATP13A1 ( Figure 2D). Importantly, surface expression of classical MHC class I molecules was not similarly impacted by loss of ATP13A1, suggesting that the phenotype is specific to MR1 ( Figure 2E). Indeed, MHC class I surface levels appeared to be elevated in the ATP13A1 KO clones, possibly because of increased availability of β 2 m. HAP1 cell cultures are prone to becoming polyploid over time, due to a competitive growth disadvantage of haploid cells (81). The flow cytometry and the functional data in Figure 2 are representative of two experiments, but one HAP1.MR1 ATP13A1 KO clone (3-1) lost its MR1 phenotype in a third repeat (Supporting Information Figure 7). This correlated with morphological changes indicative of a loss of the haploid state and so for further validation experiments we used THP-1 cells as a model system (see below). Importantly, β 2 m was not differentially expressed in the KO compared to the sufficient clones at the transcript ( Figure 2F) or protein level (Supporting Information Figure 8), indicating that the difference in MR1 surface expression was likely not due to a lack of its smaller subunit. This conclusion is corroborated by the rescue of both total and surface levels of MR1 in the reconstituted HAP1.MR1 ATP13A1 KO clones ( Figure 2B and D) which demonstrates that β 2 m is not a limiting factor for MR1 expression in these cells. However, there was a trend towards lower MR1 transcript levels in the ATP13A1 KO clones ( Figure   2G). Thus, we cannot exclude an effect of ATP13A1 on MR1 mRNA stability in the HAP1 system.
To confirm that the MR1 phenotype was not specific to the HAP1.MR1 parent clone used in the screen, we used sgRNA2 to knock out ATP13A1 in two other HAP1.MR1 parent clones (D3 and F2, compare Supporting Information Figure 1). These newly derived clones replicated the loss of total MR1 by Western blot and largely also the MR1 surface phenotype except for one of the three tested F2 ATP13A1 KO clones, which had likely become polyploid (Supporting Information Figures 9 and 10). Overall, these data indicate that the effect of ATP13A1 is not related to the number or site of insertions of the lentiviral MR1 expression cassette.

ATP13A1 specifically affects MR1-mediated antigen presentation but not classical peptide presentation
Next, we validated the role of ATP13A1 in the monocytic cell line THP-1, which is frequently used as a model system for MR1-mediated antigen presentation (2,30,82,83). We generated ATP13A1 KO clones from a THP-1 parent clone derived by limiting dilution ( Figure 3A) and confirmed disruption of ATP13A1 on the genomic level by Next Generation Sequencing. This allowed us to investigate whether ATP13A1 deficiency impacted endogenous MR1. While MR1 mRNA levels were unaffected ( Figure 3B), MR1 surface expression was reduced in five THP-1 ATP13A1 KO clones compared to ATP13A1sufficient controls ( Figure 3C). This trend did not reach statistical significance in one repeat due to one sample having high isotype staining (Supporting Information Figure 11). Importantly, endogenous MR1 surface levels were lower in ATP13A1 KO clones not only at baseline but also in the presence of the MR1-stabilising ligand Ac6FP ( Figure 3C). Indeed, the fold-change of MR1 geometric mean fluorescence intensity (GeoMean) induced by Ac6FP was smaller in these clones ( Figure 3E, top panel and Supporting Information Figure   11C, top panel), suggesting that the observed differences in MR1 surface expression are not simply due to the smaller pool of MR1 molecules available. While surface expression of classical MHC class I molecules was more heterogeneous than MR1 in ATP13A1 KO clones, there was no statistically significant difference compared to ATP13A1-sufficent clones ( Figure 3D). As expected, MHC class I surface levels were unaffected by incubation with Ac6FP ( Figure 3E, bottom panel).

Figure 3: MR1 surface expression is reduced in THP-1 ATP13A1 KO clones.
A THP-1 WT clone was transiently transfected with a CRISPR/Cas9 plasmid encoding sgRNA 2 targeting the first exon of ATP13A1 and cells were sorted into single wells. A, clones were screened by Western Blot. Membranes were probed with anti-ATP13A1 (green) and anti-β actin (ACTB; red) antibodies. Molecular weight of marker bands is shown in kDa to the left of the blot. B, MR1 transcript levels were compared in five ATP13A1 sufficient clones and five ATP13A1 KO clones. C + D, those ten clones were incubated with Acetyl-6-formylpterin (Ac6FP, blue) or DMSO (black) for 5 h before staining for MR1 (C) and MHC class I (D) surface expression. Histograms for two representative clones are shown on the left and cumulative data from all 10 clones are shown on the right. The same parent sample is shown in both histograms for comparison. E, the same data as in C + D shown as fold-change of Geometric Mean Fluorescence Intensity (GeoMean) with Ac6FP compared to DMSO control. Data in A are representative of at least three experiments for each clone except for WT THP-1 which was not always included. Results in B -E are supported by another experiment each although the repeat for C differed in statistical significance (see Supplementary Information Figure 11). Data are shown as mean with standard deviation in B and median in C -E. Statistical significance was calculated using a two-tailed t-test in B and the Mann-Whitney test in C -E. L/D = live/dead stain. ATP13A1 KO clones are highlighted in red. ns = p > 0.05. When ATP13A1 expression was reconstituted in the THP-1 ATP13A1 KO clone 2, MR1 surface expression was restored (Supporting Information Figure 12). Total MR1 protein expression could not be investigated in these clones as endogenous MR1 levels are undetectable by Western blot (see e.g. Supporting Information Figure 12B).
Consistent with reduced levels of MR1 at the cell surface, the THP-1 ATP13A1 KO clones presented the MR1 ligand 5-(2-oxopropylideneamino)-6-D-ribitylaminouracil (5-OP-RU; added as 5-amino-ribityl uracil (5-A-RU) + methylglyoxal (MG)) less efficiently to sorted human MAIT cells than their ATP13A1 sufficient counterparts ( Figure 4A). This trend was observed across multiple experiments using different MAIT cell donors and 5-A-RU concentrations, although it did not always reach statistical significance (Supporting Information Figure 13A). Since the antigen presentation pathways for exogenous ligand and intracellular bacteria differ mechanistically (31,39,42), we also tested the effect of ATP13A1 deficiency on MAIT cell activation in the context of bacterial infection. As with exogenous synthetic ligand, THP-1 ATP13A1 KO clones infected with E.coli showed a trend of eliciting a lower IFNγ response from sorted human MAIT cells as compared to ATP13A1 sufficient clones ( Figure 4B and Supporting Information Figure 13B). Importantly, this functional defect appeared to be specific to MR1-mediated antigen presentation since the response of a peptide-specific T cell line to peptide-pulsed ATP13A1 KO clones was comparable to that elicited by ATP13A1 sufficient clones ( Figure 4C and Supporting Information Figure 13C). Similarly, presentation of the CD1d ligand α-Galactosylceramide (αGalCer) to sorted human invariant natural killer T (iNKT) cells was not significantly affected by loss of ATP13A1 ( Figure 4D and Supporting Information Figure 13D). Of note, ATP13A1 was not a hit in a recent whole-genome siRNA screen for modulators of antigen presentation by CD1d (84). Importantly, when ATP13A1 expression was restored in ATP13A1 KO clone 2, its capacity to present 5-OP-RU was partially restored ( Figure 4E and Supporting Information Figure   13E). Over-expression of ATP13A1 in the THP-1 parent clone, however, did not increase antigen presentation. Thus, it appears plausible that the transporter operates within a tightly controlled regulatory network optimised to prevent aberrant MAIT cell activation.  Figure 13). Statistical significance of differences between indicated groups was analysed using unpaired t tests. ns = p > 0.05.

ATP13A1-deficient HAP1.MR1 clones do not suffer from elevated ER stress
Having established that ATP13A1 KO clones display an antigen presentation phenotype largely specific to MR1 and attributable to ATP13A1 deficiency, we next investigated the mechanism by which ATP13A1 expression modulates MR1 protein levels. Since ATP13A1 has been implicated in the transport of cations (69,74), we hypothesized that it could indirectly influence MR1 protein folding, loading, or stability in the ER by modulating ion concentrations in this compartment. Consistent with this, a large body of literature has linked deficiency for Spf1p and ATP13A1 to ER stress in yeast and human cells, respectively (69,74,(76)(77)(78). We, thus, tested whether HAP1.MR1 clones deficient for ATP13A1 differentially up-regulated components of the unfolded protein response (UPR).
Although all tested clones responded with up-regulation of the ER stress markers HERPUD1 (85) and ATF4 (86) when treated with the ER stress-inducing inhibitor thapsigargin (TG) (87), we found no evidence for differential UPR induction in the ATP13A1 KO clones ( Figure 5A and B). The same pattern was evident when analysing the splice variants of XBP1, a transcription factor that controls the expression of UPR proteins in response to ER stress and is itself activated by alternative splicing of its mRNA (88) ( Figure   5C). The data also indicate that the HAP1.MR1 ATP13A1 KO clones do not suffer from elevated ER stress at baseline, suggesting that the lower MR1 protein levels are not due to a general defect in protein folding.  Table 3). ATP13A1 KO clones are highlighted in red and the ATP13A1 mutant clone 3-15 is highlighted in bold and starred. noRT = no reverse transcriptase control. Divalent cations such as calcium and manganese are important co-factors for many enzymes and disturbing their homeostasis leads to defects in glycoprotein processing (69,(74)(75)(76). Since the MR1 protein sequence features one known N-linked glycosylation site at Asn86 (32), we next determined whether MR1 in the HAP1.MR1 KO clones might be hypoglycosylated. To investigate the glycosylation state of MR1 in the HAP1.MR1 ATP13A1 KO clones, whole cell lysates were digested with Endoglycosidase H (EndoH).
Since EndoH cannot cleave the highly complex polysaccharide residues generated during post-translational modification in the Golgi apparatus, sensitivity to EndoH digestion can also be used as a proxy of ER-residency (89,90). It has previously been shown that in many cell lines MR1 primarily resides in a pre-Golgi compartment at steady state as reflected by its sensitivity to digestion with EndoH (32,34,70). Accordingly, the cellular MR1 pool in the HAP1.MR1 parent clone D9 was also sensitive to digestion with EndoH ( Figure 5D). Interestingly, the much smaller pool of MR1 molecules in the HAP1.MR1 ATP13A1 KO clones was equally sensitive to the endoglycosidase, indicating that these molecules are glycosylated normally within the ER. This makes it seem unlikely that a general dysfunction of the ER glycosylation machinery is the cause for the reduced MR1 protein levels at baseline. Importantly, protein levels of classical MHC class I molecules were unaffected in the HAP1.MR1 ATP13A1 KO clones (Supporting Information Figure 14), further corroborating this conclusion. Of note, the antibody used to detect MHC class I by Western Blot does not react with HLA-A and HLA-G alleles (91). Hence, potential allelespecific effects on MHC class I molecules would not be detected in this assay and can, thus, not be excluded.

ATP13A1 deficiency affects MR1 levels at or immediately after protein synthesis
Since MR1 transcript levels did not significantly differ between ATP13A1 KO and sufficient HAP1.MR1 ( Figure 2G) or THP-1 ( Figure 3B) clones and we found no evidence for a differential global induction of the unfolded protein response in the HAP1.MR1 ATP13A1 KO clones (Figure 5), we hypothesized that MR1 may be specifically targeted for ER-associated degradation (ERAD) in these cells. To test this hypothesis, we inhibited proteasomal degradation with the reversible peptide aldehyde inhibitor MG-132 (92,93). In parallel, we inhibited ERAD at a different step with the irreversible p97 inhibitor NMS-873 (94). p97 aids in the extraction of ubiquitinated target proteins from the ER lumen into the cytosol and, thus, encourages misfolded proteins towards proteolysis (94,95). Even though proteasomal degradation was successfully inhibited in this experiment as shown by the stabilisation of HERPUD1, this inhibition did not lead to an increase in full-length MR1 protein levels ( Figure 6A). This suggests that MR1 is not differentially degraded by ERAD upon ATP13A1 KO under homeostatic conditions. However, proteasome inhibition may have led to the accumulation of ubiquitinated, high molecular weight MR1 species which could not be detected with the experimental setup used. To further characterize the effect of ATP13A1 on the kinetics of MR1 expression, we performed pulse-chase experiments after metabolic labeling of nascent proteins with radioactive amino acids. In order to be able to immunoprecipitate nascent MR1 molecules regardless of their folding state, we overexpressed human influenza haemagglutinin (HA)tagged MR1 in one THP-1 ATP13A1 KO clone and one THP-1 ATP13A1-sufficient clone as well as the THP-1 parent clone (Supporting Information Figure 12). Confirming the phenotype seen in the HAP1.MR1 ATP13A1 KO clones, total protein levels of recombinantly overexpressed MR1-HA were lower in the absence of ATP13A1 compared to the THP-1 WT parent clone and the ATP13A1-sufficient clone (Supporting Information Figure 12B).
Immuno-precipitation of a radiolabelled cohort of proteins revealed that the levels of MR1-HA were reduced in the ATP13A1 KO cells as early as two minutes after the pulse ( Figure   6B). Indeed, in the absence of ATP13A1 the HA-reactive population of molecules incorporated the radioactive amino acids less efficiently compared to the parent or the ATP13A1-sufficient cells as evident from the lower signal at time = 0 min ( Figure 6B). This indicates that ATP13A1 modulates either the rate of MR1 protein synthesis or the stability of the nascent MR1 polypeptide immediately after its co-translational translocation into the ER. Importantly, the abundance and kinetics of folded MHC class I molecules were comparable between all cell lines ( Figure 6C).
To exclude the possibility that the reduced MR1-HA expression in the ATP13A1 KO background was caused by differences in transduction efficiency, we transduced the clones with an MR1-Emerald construct featuring an internal ribosomal entry site (IRES) (96). In these cells, MR1 and Emerald are co-transcribed but not fused on the protein level. Thus, expression of the green fluorescent protein can be used as an indicator of transduction efficiency without affecting MR1 protein function. Although overall transduction efficiency was indeed slightly lower in the THP-1 ATP13A1 KO clone 2 ( Figure 7A+B), MR1 surface levels were reduced in these cells compared to the THP-1 parent and the ATP13A1sufficient clone 15 across a range of transduction levels ( Figure 7C+E). Importantly, MHC class I surface expression was not affected ( Figure 7D+F). In conclusion, in the absence of ATP13A1, we observed reduced expression of endogenous MR1 as well as of three exogenously expressed constructs, corroborating the conclusion that MR1 protein levels are stabilized by the P 5 -type ATPase.

Discussion
MAIT cells are one of the most recent additions to the family of innate-like, unconventional T cells (97). Unlike classical, MHC class I and II-restricted T cells, they are characterised by a limited TCR repertoire, reactivity to conserved, non-peptidic antigens, and an effectormemory phenotype, which allows them to rapidly exert effector functions (15,22). MAIT cells have been implicated in the control of numerous infectious diseases (recently reviewed in (15)) and emerging data extend their activity to protective as well as pathogenic roles in sterile inflammation, autoimmunity, tissue repair, and cancer (98)(99)(100)(101)(102)(103)(104). Their unusually high abundance in both blood and tissues, combined with their restriction by a monomorphic antigen presenting molecule makes them a particularly attractive therapeutic target (24,25,28,103,104). However, to effectively leverage the potential of these cells in Due to its localisation in the ER, its predicted transporter function, and its high enrichment in the MR1 low population in this study, we focused on the P 5 -ATPase ATP13A1. The data presented here support the conclusion that this transmembrane helix dislocase modulates the size of the intracellular pool of MR1 and consequently, MR1 surface expression and antigen presentation. ATP13A1 is the mammalian orthologue of the yeast ATPase Spf1p (also known as Cod1p) and is unique among the five known mammalian P 5 -ATPases (80).
The only member of the P 5 A subgroup in humans, it differs from its P 5 B paralogues ATP13A2-5 in its intron/exon structure and is predicted to have resulted from an earlier gene duplication than the other isoforms (80,106). In addition, ATP13A1 features a stretch of 67 amino acids which is not present in the other mammalian isoforms in either human or mouse but in the yeast, Caenorhabditis elegans, and Drosophila melanogaster orthologues (79,80). Like all P-type ATPases, ATP13A1 features a central DKTGTLT motif containing the catalytic aspartate (77,79,80,107) and it also contains conserved nucleotide binding domains (79,80). Topology predictions have identified at least 10 putative transmembrane domains for mammalian ATP13A1 (80). Both ATP13A1 and ATP13A4 localise to the ER and have been reported to be involved in calcium homeostasis (69,74,108). However, direct calcium transport activity has not been shown for ATP13A1 and ATPase activity of purified Spf1p was not dependent on or enhanced in the presence of calcium in vitro (74).
The concentration of manganese in microsomes isolated from Δspf1 cells was decreased compared to that in wild type microsomes, though other divalent cations such as calcium and magnesium were not tested in this study (69).
In addition to their implication in ER ion homeostasis, P 5 A-ATPases are associated with a wide variety of cellular functions including ER stress and the UPR, protein processing and turnover, membrane insertion and targeting of secretory proteins, intracellular sterol distribution and vesicular transport, and many more (109). Despite, or perhaps because of, these pleiotropic phenotypes, the substrate of this group of transporters remained enigmatic until very recently (109). In a landmark publication, McKenna et al. demonstrated that human and yeast P 5 A-ATPases act as protein dislocases, removing mis-targeted tailanchored mitochondrial proteins wrongly inserted into the ER membrane (73). In this manuscript we show that loss of ATP13A1 corresponded to a drastic decrease in total cellular MR1 levels in two different cell lines and examined possible mechanisms by which this transporter may impact the amount of MR1 available in the cell.
Since spf1 mutants suffer severe ER stress (69,74,(76)(77)(78), we hypothesized that the MR1 phenotype we observed in the ATP13A1-deficient cells may be caused by ER-associated degradation due to misfolding of either MR1 itself or a crucial chaperone. However, we did not find any evidence for elevated ER stress in ATP13A1-deficient HAP1.MR1 cells and inhibition of proteasomal degradation did not rescue MR1 levels in these cells at steady state. This indicates that MR1 molecules do not accumulate in an unfolded state in the ER, which would necessitate clearance by the UPR and is in stark contrast to the yeast system (74,77,78,110). Of note, the metazoan ER stress response comprises three overlapping branches while the equivalent system in yeast cells relies on IRE1 sensing only (111). With less in-built redundancy, this simpler system may be less robust, providing a potential explanation as to why yeast cells appear to be more dependent on Spf1p for maintaining ER homeostasis. Alternatively, ER stress may be a downstream effect of Spf1p deficiency rather than a direct consequence of it and mammalian cells may have more sophisticated mechanisms to counteract such secondary effects.
Yet, the small population of MR1 molecules detectable in HAP1.MR1 ATP13A1 KO clones was still sensitive to endoglycosidase digestion, confirming that glycoprotein processing in the ER was not affected upon loss of the transporter. This observation is consistent with reports that EndoH-sensitive core glycosylation is intact in spf1 KO yeast strains and only higher order polysaccharide residues could not be processed in these mutants (76).
Since the reduction of total MR1 protein was not accompanied by a statistically significant difference on the mRNA level, we focussed our efforts on post-transcriptional events.
Pulse-chase experiments revealed that the difference in total MR1 protein levels was detectable within minutes of translation and affected MR1 molecules regardless of their folding state. This is consistent with a model where ATP13A1 function is important for the stabilisation of nascent MR1 polypeptides immediately after translation and suggests a very early bifurcation into a cohort of MR1 molecules that successfully complete protein synthesis and initial folding and a cohort of molecules that is immediately degraded.
Alternatively, the rate of MR1 protein synthesis may be affected in these cells. The observed trend that radioactive labelling of MR1 was less efficient in an ATP13A1 KO clone at early time points compared to THP-1 WT cells or an ATP13A1-sufficient clone derived from the same parent supports this hypothesis. Intriguingly, protein translocation into the ER through Sec61 is impaired in an spf1 yeast mutant (112), providing a potential mechanistic basis for this model. However, such an effect on protein translocation would be expected to have a global effect, yet antigen presentation by classical MHC class I and CD1d was unaffected in our ATP13A1 KO cells. Pertinent to this, a recent siRNA screen to identify modulators of CD1d-mediated antigen presentation did not find ATP13A1 (84). A recent CRISPR screen for MR1 chaperones, however, also found sgRNAs targeting ATP13A1 to be enriched in cells with low MR1 surface expression (36), further supporting the role for this ATPase described here.
Although Spf1p has been shown to modulate ion concentrations in the ER, it has also been postulated that its substrates may be more diverse, potentially even including macromolecules such as aminophopholipids based on sequence similarities to P 4 -type ATPases (74,109). Since it is still unclear how soluble MR1 ligands access the ER where they encounter partially folded, ligand-receptive MR1 molecules (37,41) it is tempting to speculate that ATP13A1 may also transport vitamin metabolites, implying a role analogous to TAP which transports antigenic peptides into the ER for loading onto MHC class I molecules (113). Pertinent to this, Sørensen and colleagues recently identified phopsphatidyl-inositol 4-phosphate as a potent stimulator of Spf1p ATPase activity and hypothesised that sterol flippase activity of P 5 A-ATPases may underlie the diverse phenotypes observed upon mutation of the transporters (112). Interestingly, two putative negative regulators of MR1 surface expression identified in our screen (VAC14; log2MI = 2.087; fcpv = 7.39x10 -6 ; and FIG4; log2MI = 1.879; fcpv = 1.05x10 -3 ) are components of a complex that regulates the interconversion of phosphatidylinositol-derived signalling lipids (52), further strengthening a potential link between MR1-mediated antigen presentation and sterol homeostasis.

Recent work by McKenna et al. further extended the list of P 5 A-ATPase substrates to short
transmembrane helices of mis-localised proteins (73). In this study, the authors obtained the first structural information on Spf1p, which provides an explanation for the unusual substrate selectivity of P 5 A-ATPases: compared to other P-type transporters, Spf1p features an unusually large binding pocket with a lateral opening facing the lipid phase of the ER membrane. This topology is ideally suited for its interaction with short, helical transmembrane segments which the ATPase removes from the lipid bilayer. In this way, tail-anchored mitochondrial proteins that are incorrectly inserted into the ER membrane can be removed and released into the cytosol for re-targeting to the correct organelle. Using quantitative proteomics, the authors show that loss of ATP13A1 leads to decreased protein abundance of a variety of membrane-anchored proteins (73), consistent with our findings on reduced MR1 protein levels in ATP13A1 KO cells. Unfortunately, the proteomics analysis does not include data on MR1 (73). Of note, HLA-A protein levels were not affected (73), consistent with our results. Although mis-oriented N-terminal signal sequences can also be substrates for ATP13A1 (73) and MR1 features such a peptide at is N-terminus, ATP13A1 did not coimmunoprecipitate with MR1 in our hands (data not shown). Similarly, a recent coimmunoprecipitation screen provided no evidence that MR1 directly interacts with the transporter (36).
It remains a likely possibility that the defect in MR1 antigen presentation is an indirect rather than a direct consequence of loss of ATP13A1. Since inhibition of proteasomal degradation did not rescue the MR1 phenotype, it further remains to be determined whether MR1 molecules in the ATP13A1 KO cells are degraded by a different means or whether the rate MR1 protein synthesis is reduced in these mutants. If protein synthesis was affected, further questions would arise as to the mechanism by which the ER transporter regulates translation and, most importantly, how this is specific to MR1.
In summary, we provide evidence that the P 5 -ATPase ATP13A1 is an important cellular factor impacting MR1 surface expression and antigen presentation through a role in MR1 biogenesis in the ER. Our findings suggest that the mechanism underlying this effect does not involve differential glycosylation or general ER stress but appears to act on nascent MR1 polypeptides upon or immediately after protein synthesis. Further investigations into the transcriptional and translational effects of ATP13A1 deficiency as well as the extent to which these effects can be generalised to in vivo models are needed to confirm our data and determine the molecular mechanism responsible for the MR1 phenotype.

Experimental Procedures
Cell culture: THP-1, C1R, and HEK293T cells were purchased from ATCC. The generation of the Melan A-reactive T cell line was described previously (114). HAP1 WT 1524 was obtained from the Nijman lab (Nuffield Department of Medicine, University of Oxford, UK).
MAIT cells and iNKT cells were sorted from human peripheral blood mononucleocytes (PBMCs) isolated from leukocyte cones obtained from the NHS Blood and Transplant Unit as described below. This work was covered by HTA licence number 12433. Cells were kept at 37° C with 5% CO 2 . In general, THP-1 cells and their derivatives were cultured in RPMI- Cells were stored at 4° C o/n before surface staining for CD3 (eBioscience, clone SK7; not always included), CD161 (eBioscience, clone HP-3G10) and the Vα7.2 TCR (BioLegend, clone 3C10) to enrich for Vα7.2 + CD161 high MAIT cells on a SONY MA900, a FACSAria Fusion, or a FACSAria III sorter (both BD Biosciences) in the WIMM Flow Cytometry Facility. Sort purity was routinely >90%. Invariant natural killer T (iNKT) cells were enriched with the antibody clone 6B11 (BioLegend) and maintained as described (116).

Generation of over-expressing cell lines:
To generate cell lines over-expressing human MR1, the pHR-SIN-MR1-IRES-Emerald lentiviral expression vector previously made in our lab (96) was digested with XhoI and NotI endonucleases and the Emerald coding sequence was replaced with a 21 base pair adapter designed to include a BamHI restriction site to enable screening by PCR. The undigested plasmid was used for the IRES-Emerald transductions. The plasmid encoding the HA-tagged MR1 has been described previously (5). Briefly, the MR1 coding region was amplified from the pHR-SIN-MR1-IRES-Emerald vector (96) with a forward primer having a BglII overhang and a reverse primer introducing an HA sequence and a NotI restriction site. This was then cloned into the pHR-SIN-IRES-Emerald vector using the BamHI and NotI restriction sites upstream and downstream of the IRES-Emerald cassette, respectively. This vector was derived from the pHR-SIN-CSGW plasmid (117,118). Of note, the MR1 coding sequence used carries two silent mutations: A138G and C939T. All oligonucleotide sequences can be found in Supporting Information  pulsing was performed in serum-free medium. α-Galactosylceramide (αGalCer) was purchased from Enzo Life Sciences. 5-A-RU was synthesised as described previously (120) and combined with methyl-glyoxal (Sigma-Aldrich) immediately before addition to the culture. Stimuli were washed off before co-incubation with either 15 000 or 20 000 T cells per well for at least 36 h. Interferon γ (IFNγ) in the culture supernatants was measured by ELISA using commercially available antibody pairs (BD Biosciences, clones NIB42 for the capture antibody and 4S.B3 for the biotinylated detection antibody). Antigen binding was visualized with avidin-peroxidase (Sigma-Aldrich) and developing solution containing phenylenediamine (Sigma-Aldrich). The reaction was stopped with 2 M H 2 SO 4 (Fluka) when the standard curve had fully developed. Colour intensity was measured at 490 nm in an iMark plate reader (Bio-Rad) and quantified using a 4-parameter standard curve. qRT-PCR: RNA was extracted using either the QIAGEN RNeasy kit or the Ambion RNAqueous kit according to the manufacturer's instructions including the respective DNase digestion protocol. RNA was quantified by Nanodrop (Thermo Scientific) and equal amounts of RNA were reverse transcribed using Ambion or Takara Bio kits. Only cDNA transcribed with the same kit was directly compared in qRT-PCR reactions. HAP1 WT or D9 parent control RNA was not always extracted on the same day as the clone RNA.
Pooled "no RT" and "no RNA" controls were routinely included and occasionally gave a signal but always at a much higher C T value than the samples. qRT-PCRs were run as technical duplicates or triplicates on a QuantStudio7 qRT-PCR machine (Life Technologies) and expression was normalised to a house keeping gene (HKG). Taqman probes used can be found in Supporting Information Table 4.
ER stress assay: HAP1.MR1-derived clones were treated with 0.1 µM thapsigargin (Sigma-Aldrich) or solvent control DMSO for 3 h. Protein lysates were obtained and Western Blots performed as above. In the same experiments, RNA was extracted and reverse transcribed as above. XBP1 cDNA was PCR amplified with previously published primers (121) (Supporting Information Table 4) and analysed for the XBP1 splice variant on a 2.5% agarose gel. D9 cells were thawed and approximately 3x10 9 cells were pulsed with 5 µg/ml Ac6FP for 5 h before cell surface staining with anti-MR1 primary antibody (clone 26.5) and goat antimouse secondary antibody coupled to Alexa488 (Life Technologies). Stained cells were fixed, permeabilised, and stained with Propidium Iodide before sorting on a LE-SH800 sorter (Sony) fitted with a 100 µm chip over four days gating on haploid cells based on PI staining. MR1 hi and MR1 low populations comprising approximately 1% of total, respectively, were sorted (Supporting Information Figure 2).

Recovery of insertion sites and bioinformatics analysis:
Gene trap insertion sites were recovered using linear amplification-mediated polymerase chain reaction (LAM-PCR) similarly to the protocol described by Blomen et al. (122) and Brockmann et al. (44). Briefly, sorted MR1 hi and MR1 low cells were de-crosslinked and DNA was extracted from the two populations separately. Long terminal repeat (LTR)-proximal gene regions were amplified from 4x1 µg genomic DNA in four separate reactions per MR1 population using a doublebiotinylated primer. Biotinylated single-stranded DNA (ssDNA) was isolated and Illumina sequencing adapters were added. Libraries were PCR-purified and sequenced on a HiSeq2500 machine (Illumina) generating 50 base pair (bp) single end reads with the 5' sequencing primer. All primer sequences can be found in Supporting Information Table 4.
Recovered sequences were analysed with the pipeline described by Brockmann et al. (44).
In short, reads were mapped onto the human reference genome hg19 by running the aligner Bowtie (123) twice, allowing for 0 and 1 bp mismatches to avoid omitting reads that do not map uniquely when tolerating 1 bp mismatch. Reads were mapped using hg19 protein-coding gene coordinates (Refseq) and intersectBED (124). A customised BED file was used for both gene mapping and determining the orientation of the integration relative to the gene (sense or antisense). Only sense insertions in non-overlapping gene regions outside the 3' untranslated region (UTR) were considered to have the potential to disrupt gene function and taken into account for further analysis. The number of unique disruptive insertions for a gene in the MR1 hi or MR1 low population was normalised to the total number of insertions in the respective populations and compared to the number of normalised insertions for that gene in the opposite population using a two-sided Fisher's exact test.
Significance of enrichment in either of the two tails was determined by applying the Benjamini-Hochberg false discovery rate correction to the calculated p-values. In addition, enrichment in either tail was expressed as the Mutation Index (MI), calculated for each gene as shown below. If a gene was sequenced in only one of the populations, one insertion was added in order to enable calculation of the mutational index.  Table   4, Supporting Information Figure 4). sgRNA sequences were tested for specificity using the UCSC-genome browser BLAT tool (https://genome.ucsc.edu/cgi-bin/hgBlat) and cloned into the pX458_Ruby backbone (Addgene #110164

Supporting Information
This article contains supporting information. The SI references Wang et al. (125) and Hollien et al. (121).

Conflicts of Interest
M.S. holds consultancies with Nucleome Therapeutics and Enarabio. All other authors declare that they have no conflicts of interest with the contents of this article.