Author response: A forward genetic screen reveals novel independent regulators of ULBP1, an activating ligand for natural killer cells

Article Figures and data Abstract eLife digest Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract Recognition and elimination of tumor cells by the immune system is crucial for limiting tumor growth. Natural killer (NK) cells become activated when the receptor NKG2D is engaged by ligands that are frequently upregulated in primary tumors and on cancer cell lines. However, the molecular mechanisms driving NKG2D ligand expression on tumor cells are not well defined. Using a forward genetic screen in a tumor-derived human cell line, we identified several novel factors supporting expression of the NKG2D ligand ULBP1. Our results show stepwise contributions of independent pathways working at multiple stages of ULBP1 biogenesis. Deeper investigation of selected hits from the screen showed that the transcription factor ATF4 drives ULBP1 gene expression in cancer cell lines, while the RNA-binding protein RBM4 supports ULBP1 expression by suppressing a novel alternatively spliced isoform of ULBP1 mRNA. These findings offer insight into the stress pathways that alert the immune system to danger. https://doi.org/10.7554/eLife.08474.001 eLife digest Cancer is caused by a series of mutations that result in uncontrolled cell growth and division. Yet, the body's immune system can often detect and destroy abnormal cells before they cause tumors and disease. Natural killer cells are part of the immune system and have receptors on their surface that allow them to tell the difference between healthy host cells and host cells that are stressed or abnormal. Some of these receptors activate the natural killer cells when they bind to their target molecules. Other receptors have the opposite effect and inhibit the natural killer cells. Activation occurs when the signaling from the activating receptors is stronger than the signals from the inhibitory receptors. One of the well-studied activating receptors recognizes a number of proteins and molecules that are produced by abnormal or tumor cells, including a protein called ULBP1. This protein is absent from the surface of healthy cells but is found in abundance on tumor cells. However, it is still not clear what drives tumor cells to produce ULBP1 (or other molecules) that are recognized by natural killer cell receptors. Now, Gowen et al. report on a genetic screen that has revealed numerous genes that regulate the levels of ULBP1 in human cells. Many of these genes had independent effects that when added together accounted for most of the ULBP1 present on the cell surface. Gowen et al. then explored some of the ‘regulators’ encoded by these genes in more detail. One called ATF4, which had previously been linked to stress responses, was shown to increase the expression of the gene for ULBP1 in cancer cells. Another regulator called RBM4 instead acted in a different way and at a later stage in ULBP1 production. All together, these findings offer insight into the stress pathways that alert the immune system to abnormal cells. The next challenge will be investigating how these pathways might be exploited for cancer immunotherapy. https://doi.org/10.7554/eLife.08474.002 Introduction Natural killer (NK) cells are lymphocytes of the innate immune system that play a critical role in limiting tumor growth (Vivier et al., 2011; Marcus et al., 2014; Mittal et al., 2014). NK cell activation is controlled by a balance of signals from activating and inhibitory receptors, which recognize cognate ligands expressed by potential target cells (Vivier et al., 2011; Shifrin et al., 2014). One of the best-studied NK-activating receptors is NKG2D, which is also expressed on certain subsets of T cells (Raulet, 2003). Engagement of NKG2D by its ligands displayed on a target cell membrane leads to NK cell activation, cytokine secretion, and lysis of target cells, such as tumor cells. NKG2D recognizes a family of ligands that are structurally similar to MHC Class I proteins. Humans express up to eight NKG2D ligands (ULBP1-6, MICA, and MICB), and mice express 5–6 different ligands, depending on the strain (RAE-1α-ε, H60a-c, and MULT1) (Raulet et al., 2013). Healthy cells typically do not display NKG2D ligands on their surface and are thus poor targets for NKG2D-mediated lysis by NK cells. However, cellular stresses associated with transformation, viral infection, or other danger to the host cause the upregulation of NKG2D ligand expression (Raulet et al., 2013). Primary tumors and cancer cell lines frequently express one or more NKG2D ligand, and NKG2D expression is important for the control of tumors in vivo in models of spontaneous cancer (Guerra et al., 2008). Tumors arise despite the tumor-suppressive effects of the immune system, and some tumors show evidence of adaptation to escape immune control (Schreiber et al., 2011). In the case of NKG2D-mediated tumor recognition, published results suggest that one mechanism of tumor immune evasion is the loss or decreased expression of NKG2D ligands (Guerra et al., 2008; McGilvray et al., 2009). In other cases, tumors progress despite sustained expression of NKG2D ligands (Vetter et al., 2002; Guerra et al., 2008; McGilvray et al., 2009; Hilpert et al., 2012). The evidence as a whole suggests that upregulation of NKG2D ligands on early stage tumor cells is part of a host defense mechanism, but that the immune response subsequently applies selective pressure for tumors that have either extinguished expression of NKG2D ligands or have activated immune suppressive mechanisms (Raulet and Guerra, 2009). Therefore, identifying factors and pathways that regulate NKG2D ligands will improve our understanding of the cellular properties used by the immune system to define unwanted cells and will also help reveal how tumors evade the corresponding immune responses. Prior investigations have identified regulators of NKG2D ligands using a candidate approach based on the roles of these regulators in known stress pathways. Such approaches have implicated the DNA damage response pathway (Gasser et al., 2005), heat shock (Venkataraman et al., 2007; Nice et al., 2009), hyperproliferation (Jung et al., 2012), and pattern recognition receptors (Hamerman et al., 2004), among others, in the regulation of one or more NKG2D ligands. However, these pathways do not account fully for expression of ligands in tumor cells, since inhibiting them may decrease ligand expression but typically does not abrogate it. For example, the DNA damage response is active in many cancer cells and tumor cell lines, but inhibiting that pathway only partially inhibits ligand expression (Gasser et al., 2005; Gasser and Raulet, 2006; Soriani et al., 2014). Similarly, hyperproliferation can drive NKG2D ligand expression, but blocking proliferation does not completely eliminate NKG2D ligand expression by tumor cell lines (Jung et al., 2012). These findings suggest that unidentified molecular cues in tumor cells also initiate the expression of NKG2D ligands, allowing potentially dangerous tumor cells to be distinguished from normal cells. Identifying those cues, especially for human NKG2D ligands, is important for understanding the biological regulation of NKG2D ligands and devising approaches for immunotherapy based on that knowledge. To identify novel drivers of NKG2D ligand expression, we performed a genome-wide loss-of-expression mutant screen in the tumor-derived human cell line HAP1 (Carette et al., 2009; Carette et al., 2011) and used CRISPR/Cas9 gene targeting methodology for confirmation of the hits and extension of the results. The results reveal previously unknown regulators for NKG2D ligands, provide evidence for selectivity of the regulators for specific ligands, and support the cooperation of different stress pathways in the regulation of one such ligand. Results A genome-wide screen to identify novel drivers of ULBP1 expression Many tumors and cancer cell lines express multiple NKG2D ligands, possibly due to ongoing stress responses associated with the transformed state (Raulet et al., 2013). To identify novel drivers of human NKG2D ligand expression in transformed cells, we employed a retroviral gene-trap mutagenesis screen using the near-haploid human cell line HAP1 (Figure 1) (Carette et al., 2009; Carette et al., 2011). Like many cell lines, HAP1 cells express multiple NKG2D ligands (Figure 1—figure supplement 1). We chose to screen for drivers of ULBP1 expression because it showed the brightest staining on HAP1 cells, making it particularly amenable to our loss-of-expression screen. Following mutagenesis, we selected for mutants with decreased expression of ULBP1 but intact expression of the unrelated GPI-anchored protein CD55 (Figure 1A). Selection of CD55+ cells was used to reduce the fraction of selected cells that had lost ULBP1 expression due to mutations that alter cell surface expression of all proteins or of all GPI-linked proteins. In the first round of selection, we depleted ULBP1high cells from the mutant cell population using magnetic bead-based depletion of cells labeled with a ULBP1 antibody. After briefly expanding the selected cells, we used flow cytometry to further select for ULBP1lowCD55+ cells. Figure 1B shows ULBP1 and CD55 expression on WT and post-selection HAP1 cells. Figure 1 with 1 supplement see all Download asset Open asset A genetic screen of a haploid human cell line to identify regulators of ULBP1 expression. (A) HAP1 cells (∼108 cells) were transduced with a retroviral gene-trap vector. To enrich for mutant cells with decreased ULBP1 expression, we initially depleted ULBP1high cells by labeling cells with an anti-ULBP1 antibody followed by magnetic bead-based cell depletion. Following a brief recovery and expansion of the cells, we used FACS to further enrich ULBP1lowCD55+ cells. Deep-sequencing of genomic DNA from pre- and post-selection cells was used to map sites of gene-trap insertions, and mutations enriched in ULBP1low cells were identified. (B) Flow cytometric analysis of WT and post-selection HAP1 cells. Cells were stained for ULBP1 and CD55, an irrelevant GPI-linked protein. https://doi.org/10.7554/eLife.08474.003 We employed deep-sequencing to map and quantify the frequencies of independent insertion sites of the retroviral gene-trap in selected cells and compared this with the landscape of insertions in unselected control cells. Table 1 shows a selected ‘hit list’ of genes that were targeted significantly more frequently in selected ULBP1lowCD55+ cells than in unselected cells. The ULBP1 gene itself was a highly significant hit, providing a validation of this approach. Many genes encoding enzymes involved in GPI synthesis were also represented despite the selection for CD55 expression; many of these were removed from Table 1, for simplicity. The complete list of hits (p < 0.05) is shown in Supplementary file 1, along with the analysis of all independent insertions mapped in the selected data set. Raw sequencing data for the screen are available under NCBI Bioproject PRJNA284536, containing the datasets for HAP1 gene trap control cells (Accession number SAMN03703230) and cells from the ULBP1 screen (Accession number SAMN03703231). We chose hits for validation and follow-up experiments based on their statistical ranking and expectations that the corresponding proteins play roles in stress responses, protein biogenesis, or gene/mRNA regulation. The genes chosen encode ATF4 (a stress-associated transcription factor), RBM4 (an RNA-binding protein), HSPA13 (a protein chaperone), and SPCS1 and SPCS2, which are both non-catalytic subunits of the signal peptidase complex. Table 1 Selected list of genes enriched for gene-trap insertions after selection of ULBP1lowCD55+ cells https://doi.org/10.7554/eLife.08474.005 Gene symbolFunction/Processp-valuePIGWGPI synthesis/anchoring1.12E-196PIGQGPI synthesis/anchoring3.26E-155PIGBGPI synthesis/anchoring2.38E-103ULBP1NKG2D ligand2.65E-76PIGOGPI synthesis/anchoring1.49E-65RBM4RNA-binding protein1.29E-24PIGVGPI synthesis/anchoring4.43E-23SPCS1Non-catalytic subunit of signal peptidase complex1.25E-15PIGMGPI synthesis/anchoring6.06E-14C1GALT1C1Protein O-linked glycosylation2.22E-13SLC35A1Golgi-localized CMP-sialic acid transporter1.77E-12ST3GAL2Sialyltransferase1.64E-11SPCS2Non-catalytic subunit of signal peptidase complex6.20E-09HSPA13Microsome-associated protein with ATPase activity1.24E-05FLJ37453Non-coding RNA0.00115SLC17A9*Vesicular nucleotide transporter0.00715RPS25Ribosomal protein0.00715ATF4Stress-induced transcription factor0.0206PMM2Oligosaccharide synthesis, protein glycosylation0.0234NCRNA00167Non-coding RNA0.0363CRNKL1*Pre-mRNA splicing0.0363ICKIntestinal cell kinase, MAPK-related0.0363TBC1D19TBC domain-containing protein0.0416ZNF236Zinc-finger protein0.0496 The gene symbols of hits (p < 0.05) are shown with a brief description of known or predicted gene functions. A p-value of enrichment was determined using Fisher's exact test, followed by correction for the false discovery rate. The list was manually curated to remove known genes that have occurred in several unrelated screens using the same cells, perhaps indicating pleiotropic effects. For simplicity, a number of genes related to GPI biosynthesis and anchoring were removed. Bold text indicates genes confirmed in this study to impact ULBP1 expression. Blue text indicates genes involved in GPI biosynthesis and anchoring. Red text indicates genes involved in protein glycosylation. Asterisks indicate two genes (SLC17A9 and CRNKL1) that, when targeted with CRISPR/Cas9, failed to result in decreased ULBP1 expression. Validation of hits using the CRISPR/Cas9 system and gene restoration To confirm that selected genes from the screen regulate ULBP1, we employed the CRISPR/Cas9 mutagenesis system, targeting sites in the 5′ coding regions of each candidate gene in HAP1 cells (Jinek et al., 2013; Mali et al., 2013). The ULBP1 gene was targeted for comparison. After HAP1 cells were transiently transfected with plasmids encoding Cas9 and guide RNAs (sgRNAs) for each candidate gene, a population of ULBP1low cells appeared that was absent in control transfected cells (Figure 2—figure supplement 1). In each case, individual ULBP1low cells were sorted into 96-well plates, and expanded clones were screened for mutations by PCR and sequencing. For further analysis, we selected clones with insertions or deletions that resulted in frameshift mutations in each targeted gene (Figure 2—figure supplement 2). Since the sites targeted were near the beginning of each coding region and the cells are haploid, the frameshift mutations are expected to result in complete loss-of-function of the corresponding proteins. Analysis of HAP1 cell lysates by Western blot confirmed the loss-of-protein expression in ATF4, RBM4, and SPCS2 mutant cell lines (data not shown). Cells with a ULBP1 mutation lacked cell surface ULBP1 staining altogether, as expected, whereas the other mutations analyzed resulted in a partial (twofold to threefold) decrease in cell surface expression of ULBP1 (Figure 2A). The effect of each mutation was specific to ULBP1, as we found no change in cell surface expression of other proteins, including four other NKG2D ligands (ULBP2, ULBP3, MICA, and MICB), HLA Class I, the unrelated GPI-anchored protein CD59, or PVR and Nectin-2, the ligands for DNAM-1, another NK cell-activating receptor (Figure 2B,C, Figure 2—figure supplement 3). The minor changes in ULBP3 staining seen in Figure 2—figure supplement 3B were not consistently observed across experiments. The finding that the mutations each affect only ULBP1 among the NKG2D ligands tested supports the hypothesis that different NKG2D ligands are subject to distinct regulatory processes. It was surprising that SPCS1 and SPCS2 mutations only impacted cell surface staining of ULBP1 and not the six other membrane proteins tested, as we had expected that mutating components of the signal peptidase complex would cause a more generalized defect in cell surface protein expression (see ‘Discussion’). In all cases, ULBP1 expression on mutant lines could be restored by re-expressing the gene of interest with a doxycycline-inducible lentiviral vector (Figure 2D). These findings established that ATF4, RBM4, HSPA13, SPCS1, and SPCS2 each contribute partially to cell surface display of ULBP1 in HAP1 cells in steady-state culture conditions. Figure 2 with 3 supplements see all Download asset Open asset Decreased ULBP1 expression upon targeted mutation of screen hits. (A–C) Flow cytometric analysis of cell surface expression of ULBP1 (A), the NKG2D ligand MICA (B), or pan-HLA Class I (C) on WT and mutant HAP1 cells. WT and mutant (KO) cells are represented as black and red histograms, respectively. The shaded gray histogram represents isotype control staining. The blue trace in panel A shows staining of ULBP1 KO HAP1 cells and matches isotype control staining. Data are representative of at least three independent experiments. (D) To restore expression of ULBP1 drivers, mutant cell lines were transduced with a doxycycline-inducible lentiviral vector containing the gene of interest. Cells were treated for 24 hr with doxycycline (Dox) at a final concentration of 100 ng/ml for ATF4 and 1000 ng/ml for all other genes. After treatment, cells were analyzed by flow cytometry. Black histograms: WT cells transduced with control vector, +Dox. Red histograms: mutant cells transduced with Dox-inducible gene of interest, −Dox. Blue histograms: mutant cells transduced with Dox-inducible gene of interest, +Dox. The shaded gray histogram represents isotype control staining. Data are representative of three independent experiments. (E) RT-qPCR analysis of canonically spliced ULBP1 mRNA expression levels in WT and mutant HAP1 cells. Expression levels were normalized to ACTB, GAPDH, and HPRT1 and are shown as mean ±SE. The data were analyzed by 1-way ANOVA with Dunnet's multiple comparisons test comparing all samples to WT. ***p < 0.001. Data are representative of three independent experiments. In one out of three total experiments performed, the level of ULBP1 mRNA was significantly increased in the SPCS2 mutant compared to WT. https://doi.org/10.7554/eLife.08474.006 Consistent with their presumptive roles in gene expression and/or splicing, ATF4 and RBM4 KO cells each showed decreased amounts of canonically spliced ULBP1 mRNA commensurate with the change in cell surface expression (Figure 2E). Consistent with their roles as chaperones and protein-processing components, HSPA13, SPCS1, and SPCS2 KO cells all showed WT amounts of ULBP1 mRNA. Stepwise regulation of ULBP1 expression To assess whether the genes implicated in ULBP1 expression work independently or in common pathways, we generated and analyzed double and triple-mutant cell lines. We generated ATF4/RBM4 double-mutant cells by targeting ATF4 in the RBM4 KO line, and triple-mutant cells by further targeting either HSPA13 or SPCS2 in the double-mutant line. Notably, we observed stepwise decreases in ULBP1 expression with each additional mutation, with triple-mutant cells showing up to a 20-fold reduction in ULBP1 expression compared to WT cells (Figure 3A). These data suggested that the genes tested (with the likely exception of SPCS1 vs SPCS2) contribute largely independently to steady-state ULBP1 expression in HAP1 cells and support a model in which constitutive NKG2D ligand expression in cell lines is due, at least in some instances, to the contribution of several pathways that act cumulatively. ATF4/RBM4 double mutants showed a larger decrease in canonically spliced ULBP1 mRNA than either single mutant, suggesting that ATF4 and RBM4 may act independently in regulating the amounts of ULBP1 mRNA (Figure 3B). There were no greater decreases in ULBP1 mRNAs when mutations in either HSPA13 or SPCS2 were added to the double-mutant cells, as expected if they act co-translationally or post-translationally. Figure 3 Download asset Open asset Double and triple-mutant cell lines show stepwise decreases in ULBP1 expression. (A) Flow cytometric analysis of ULBP1 expression on single, double, and triple-mutant HAP1 cells. ATF4−/RBM4− double-mutant cells were generated by mutagenesis of ATF4 in RBM4− cells. Triple-mutant cells were generated by mutagenesis of HSPA13 or SPCS2 in ATF4−/RBM4− double-mutant cells. Shaded gray histograms represent isotype control staining. Data are representative of three independent experiments. (B) RT-qPCR analysis of canonically spliced ULBP1 mRNA expression levels in the cells described in (A). Expression levels were normalized to ACTB, GAPDH, and HPRT1 and are shown as mean ±SE. The data were analyzed by 1-way ANOVA with Bonferroni's multiple comparisons test and substantive significant differences are shown. Data are representative of three independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, n.s.: not significant. https://doi.org/10.7554/eLife.08474.010 ATF4 supports ULBP1 expression on multiple tumor-derived cell lines Having validated several hits from our we the roles of ATF4 and RBM4 in regulating ULBP1 expression. To whether ATF4 ULBP1 in other cell we ATF4 in the cell line and the cell line. ATF4 KO and cells were generated using Cas9 and a of that the ATF4 The ATF4 gene was in the mutant lines, the of the mutant cells a ATF4 protein HAP1 cells a similar of ATF4 had an to the ATF4 frameshift mutant line described (data not shown). WT cells had ULBP1 expression, and mutation of ATF4 resulted in the complete of ULBP1 from the cell as well as an reduction in ULBP1 mRNA (Figure In in cells resulted in a reduction in ULBP1 mRNA and a reduction in ULBP1 cell surface staining. HAP1 cells showed effects of The results suggest that ATF4 drives ULBP1 expression in multiple tumor cell lines, perhaps constitutive activation of stress pathways. Figure Download asset Open asset ATF4 drives ULBP1 expression in multiple cell lines. (A) Flow cytometric analysis of ULBP1 expression on WT ATF4 KO of and cells. Data are representative of three independent experiments. (B) RT-qPCR analysis of canonically spliced ULBP1 mRNA expression levels in the cells described in (A). Expression levels were normalized to ACTB, GAPDH, and HPRT1 and are shown as mean ±SE. Expression in WT cells was to for each cell the different cell are not in this The data were analyzed by ANOVA with Bonferroni's multiple comparisons Data are representative of three independent one of the three not show a significant *p < 0.05, ***p < 0.001. ATF4 drives ULBP1 transcription in response to cell stress ATF4 is in a of stress that arise in and transformed these stress to NKG2D ligand expression may of cells. Such including acid the protein response and the of ATF4 of the factor and et al., 2012). of protein but the of ATF4 et al., The targets of ATF4 acid and protein which in with the reduction in protein synthesis help the cellular stress et al., 2003). However, ATF4 expression also drives the expression of the transcription factor suggesting that ATF4 not only drives adaptation to stress but also cell if the stress be and 2011; et al., 2013). ATF4 expression has been in some tumors et al., 2005), providing a mechanism for to expression of ULBP1, and to tumor To stress known to ATF4 expression, we treated cells with the which inhibits thus acid et al., et al., or which the and 2011). or each significantly ULBP1 mRNA in WT and cells, but was or when cells lacked ATF4 (Figure of ULBP1 transcription by the et al., was not in the cells (data not indicating that certain cellular activate ULBP1 independently of analysis of HAP1 cells confirmed upregulation of ULBP1 mRNA in response to (Figure supplement 1). RT-qPCR analysis of other NKG2D ligands expressed by HAP1 cells showed that caused an increase in mRNA that was partially while MICA expression was decreased in an (Figure supplement 2). ULBP3 and were not by or loss of ATF4 (Figure supplement 2). and mRNAs were absent or by analysis of HAP1 cells, and those genes showed no of by (data not shown). These data suggest that the stress response expression of ULBP1, and to a minor but not most other NKG2D ligands. Figure with 2 supplements see all Download asset Open asset ATF4 drives increased expression of ULBP1 mRNA and surface protein in response to cell (A) Cells were treated for 24 hr with 2 to acid or to the protein was from treated and control cells, and canonically spliced ULBP1 mRNA levels were determined by Expression levels were normalized to ACTB, GAPDH, and HPRT1 and are shown as mean ±SE. Expression in WT cells was to for each cell the different cell are not in this For the for ULBP1 in WT cells were for cells, for HAP1 cells, and for cells. The data were analyzed by ANOVA with Bonferroni's multiple comparisons test and are representative of three independent in one of the three of ATF4 KO cells, cells than cells but not *p < 0.05, ***p < 0.001, n.s.: not significant. Flow cytometric analysis of ULBP1 (B) and HLA Class I expression (C) on cells treated with as in (A). of surface staining shown in (B) and Data are as the mean of the specific the of the isotype control Data are representative of three independent experiments. cell surface ULBP1 protein expression was also and transcription of they also inhibit protein synthesis, both by limiting the availability of and by These effects it to whether stress will result in increased amounts of specific protein encoded by an in cells, we observed a of cell surface ULBP1 by (Figure In HAP1 and cells, in of ULBP1 cell surface expression in WT cells, but caused decreased ULBP1 expression in cells (Figure with the of protein synthesis from resulted in decreased cell surface expression of HLA proteins, which are not by ATF4 (Figure in Figure the effect of ATF4 by was to and/or ULBP1 cell surface expression in the of protein synthesis associated with this stress To whether ATF4 ULBP1 we used analysis to whether it to ULBP1 regulatory in HAP1 cells. with three independent ATF4 showed a of ATF4 as