Massively parallel interrogation of human functional variants modulating cancer immunosurveillance

Genome-wide mapping of critical S/T/Y residues modulating PD-L1 expressions by ABE-based screening

The interaction between PD-L1 on tumor cells and PD-1 on T cells impedes activation, proliferation, and effector functions of antigen-specific CD8 + T cells, thus promoting cancer immune evasion.⁶ To systematically explore the functional residues modulating PD-L1, the core factor involved in immunotherapy, we leveraged ABEmax to generate site-directed mutagenesis for achieving large-scale screens. Our recent work has established an ABE-based sgRNA library targeting all feasible protein-coding regions containing S/T/Y residues within the editing window leading to missense mutations. This library encompasses a staggering 818,619 sgRNAs, which collectively target 277,051 S, 165,599 T, and 141,687 Y residues.²⁵ The de novo synthesized S/T/Y library consists of two sub-libraries–one targeting the sense strand (465,554 sgRNAs) and the other one targeting the antisense strand (354,595 sgRNAs). Both sub-libraries were supplemented with the same negative controls targeting the AAVS1 locus. To better handling such an extensive library effectively, the sgRNA library was constructed with three internal barcodes (iBARs) (hereinafter referred to as sgRNA^iBAR library), as previously described.²⁴ This system ensures a high-quality screening even at a high multiplicity of infection (MOI) while significantly reducing the number of cells required for the screening process.

PD-L1 expression can be driven by tumor-intrinsic mechanisms or induced by inflammatory cytokines, such as IFNγ, which is secreted by immune cells within the TME.³ To probe functional residues affecting cell surface PD-L1 expression in both constitutive and induced contexts, we performed screens using the S/T/Y sgRNA^iBAR library in a human melanoma cell line, A375, which was engineered to stably express ABEmax. This cell line exhibits low level of endogenous PD-L1 but shows substantial upregulation of PD-L1 upon exposure to IFNγ (Supplementary Fig. 1a). The two S/T/Y sub-libraries were separately transduced into A375-ABEmax cells at an MOI of 3. Subsequently, following ten days of sgRNA transduction, the library cells were subjected to both IFNγ-stimulated and non-stimulated conditions. Through two rounds of fluorescence-activated cell sorting (FACS) enrichment, we collected cell populations with either lower or higher level of surface PD-L1 expression in each condition (Fig. 1a; Supplementary Fig. 1e–h). We also maintained a control group of library cells without FACS selection throughout the positive screening process (Supplementary Fig. 1b–d). The library cells from the control group and FACS-selected experimental groups were subjected to next-generation sequencing (NGS), and the NGS data was subsequently analyzed using the MAGeCK-iBAR algorithm.²⁴ This analysis involved evaluating the change in sgRNA abundance and calculating the P value for each sgRNA, considering the significance and consistency of three iBARs per sgRNA in each screen. The screen score was then generated as −log₁₀ of the P value after Benjamini-Hochberg (BH) adjustment (Fig. 1a).

ABE-based screens identify functional S/T/Y residues modulating PD-L1 expression at the genome-wide level. a Schematic overview of the ABE screens for identifying S/T/Y residues that regulate PD-L1 expression with and without IFNγ stimulation in A375 cells. b, c Significant S/T/Y residues enriched from the sense library (sense lib, left) and antisense library (antisense lib, right) that upregulate or downregulate PD-L1 expression in the absence of IFNγ (b) and upon IFNγ treatment (c). Positively or negatively enriched sites were selected based on a screen score > 1 or < −1. d, e Gene ontology (GO) enrichment analysis of related genes haboring identified mutations that lead to PD-L1 upregulation (upper panels) and downregulation (lower panels) in the absence of IFNγ (d) and upon IFNγ treatment (e)

We selected sgRNAs with a screen score >1 for further investigations. In each screen, numerous novel sites were identified in both the high and low directions of regulating PD-L1 expression (Fig. 1b, c; Supplementary Tables 1–4). To obtain a holistic understanding of the functional residues identified, we initially performed a gene ortholog (GO) analysis for all the related genes enriched in the screens, focusing on biological process. In the PD-L1 screen without IFNγ stimulation, the dominate terms in the PD-L1^high group were associated with histone modification, covalent chromatin modification, and the regulation of DNA-binding transcription factor activity. In contrast, the representative terms in PD-L1^low group included positive regulation of cytokine production and chromatin silencing (Fig. 1d). In the PD-L1 screen with IFNγ exposure, the enriched terms were significantly correlated with interferon stimulation, encompassing processes such as the JAK-STAT cascade, transforming growth factor signaling pathway, cellular response to IFNγ, and the regulation of phosphatase activity. Moreover, some terms overlapped between the IFNγ-treated and IFNγ-absent conditions, particularly in PD-L1^high group, where terms such as peptidyl-lysine modification and covalent chromatin modification indicated the presence of conserved factors involved in tumor-intrinsic PD-L1 regulation, regardless of IFNγ treatment (Fig. 1e).

Massively parallel validation of regulatory variants affecting PD-L1 in A375 cells

For a deeper insight into the top-ranked hits from the screens, we integrated representative genes from both high and low directions in each screen. Subsequently, we built protein-protein interaction (PPI) networks using STRING followed by GO analyses. In IFNγ-absent PD-L1 screen, the network prominently showcased multiple genes enriched in processes related to histone modification, regulation of protein stability, chromatin remodeling, and heme biosynthesis process (Fig. 2a). In the IFNγ-treated group, a large portion of genes were enriched in terms such as IFNγ-mediated signaling pathway, regulation of phosphorylation, and immune response (Fig. 2b).

Validation of regulatory residues of PD-L1 enriched in various pathways. a, b STRING analysis of related genes harboring top-ranked mutations identified in the PD-L1 screens in the absence of IFNγ (a) and upon IFNγ treatment (b). c, d Individual validations of negative and postive regulators of cell surface PD-L1 in A375 cells in the absence of IFNγ (c) and upon IFNγ treatment (d) by flow cytometry analysis. Cell surface PD-L1 expression was analyzed following incubation without or with 100 ng/mL IFNγ for 48 h. The relative median fluorescence intensity (MFI) of surface PD-L1 for each mutant represents the ratio normalized to the MFI of AAVS1-targeting control cells. Data are presented as the mean ± SD (n = 3). P values were calculated using two-tailed Student’s t test, *P < 0.05, ** P < 0.01, ***P < 0.001, ****P < 0.0001; n.s., not significant

To verify the regulatory roles of the identified variants, we selected candidate sites involved in different pathways and individually transduced each targeting sgRNA into A375-ABEmax cells via lentiviral infection. Subsequently, we conducted flow cytometry analysis to assess surface PD-L1 levels without or with IFNγ stimulation. Compared with the negative control sgRNA targeting the AAVS1 locus, most of the sites showed significant regulation of PD-L1 expression.

In the absence of IFNγ, a standout performer was the UROD_Y164 site, alongside other confirmed residues within the UROD protein, including T163, T298 and Y311 (Fig. 2c). UROD is involved in the heme synthesis pathway, whose disruption has been recognized to lead to an increase in PD-L1 expression.⁸ Besides UROD, we also successfully verified the functionality of several mutations in FECH and CPOX, the other two core factors participating in heme synthesis but with no reported roles in regulating PD-L1 expression. Additionally, a series of novel sites enriched on genes associated with chromatin remolding, especially TAF5L and TAF6L, the integral components of the PCAF histone acetylase complex, were prominently ranked in the validation process (Fig. 2c). Further analysis indicated that most of these variants showing a noteworthy phenotype influenced the expression of the target genes, ultimately resulting in an upregulation of overall and surface PD-L1 levels (Supplementary Fig. 2a). In the PD-L1^low group, due to its low baseline PD-L1 expression, a relatively smaller number of sites were identified and subjected to validation. Notably, PD-L1_Y118 and Y81 displayed the most significant impact, with Y118 being a previously recognized phosphorylation site. We also verified their association with PD-L1 expression for several additional sites, which are linked to genes known to be involved in immune response or ICB, such as WWOX_S259 and KMT2D_Y1407^26,27 (Fig. 2c).

Regarding the IFNγ-stimulated condition, several mutants reducing PD-L1 expression in IFNγ-absent condition were also validated under IFNγ treatment, including CD274_Y118, which exerted the strongest effect on downregulating surface PD-L1, consistent with the screening results (Fig. 2d). Meanwhile, with IFNγ stimulation, more sites were identified and verified within these functional genes, such as WWOX and PCYT2 (Supplementary Fig. 2b). A systematic analysis revealed that numerous mutations reduced the protein levels of their respective coding genes, as observed in the PD-L1^high group with STUB1, and in the PD-L1^low group with WWOX, TBRG1, and IKBKB (Supplementary Fig. 2c, d). Moreover, there were variants that did not significantly affect their protein expression, including HNRNPK_Y449, EED_Y308, and EED_Y365, suggesting that they may induce PD-L1 expression through other mechanisms (Supplementary Fig. 2c). Remarkably, a substantial number of sites were enriched on genes linked to the IFNγ-mediated signaling pathway and regulation of phosphorylation (Fig. 2d; Supplementary Fig. 2b), we thus delved into investigating the regulatory mechanisms of these candidate sites.

Systematic combing of functional residues within the IFNγ signaling pathway

We observed plenty of novel sites emerged on well-established genes linked to the IFNγ-mediated signaling pathway, including IFNγ receptors, Janus kinases, among others. Most of these mutations negatively regulated PD-L1 expression by affecting their respective coding genes, such as IFNGR1, IFNGR2, JAK1, JAK2, and IRF1. All investigated mutations on IFNGR1 and IFNGR2 were found to simultaneously decrease the total and membrane PD-L1 protein levels (Fig. 2d; Supplementary Fig. 3a). Notably, IFNGR1_Y457, a known phosphorylation site responsible for mediating the interaction between IFNGR1 and STAT1 proteins,²⁸ was found to significantly downregulate PD-L1 expression. This suggests that the IFNGR1_Y457H mutation might affect its binding with STAT1, blocking the transmission of IFNγ signals and resulting in a substantial reduction in PD-L1 expression. Similarly, all tested mutations in the downstream non-receptor tyrosine kinases JAK1 and JAK2 led to a reduction in both the total and membrane protein levels of PD-L1 (Supplementary Fig. 3b, c). Among these, two mutations on JAK2 consistently reduced both the mRNA and protein levels of JAK2, while most verified sites on JAK1 did not affect its own expression at both the mRNA and protein levels (Supplementary Fig. 3c, d). In addition, multiple sites on JAK1 and JAK2 were closely associated with phosphorylation, as exemplified by four known phosphorylation sites and two predicted phosphorylation sites on JAK1, and two conserved phosphorylation sites, Y1007 and Y1008, on JAK2, which are critical for JAK2 function²⁹ (Supplementary Fig. 3b).

Intriguingly, some genes related to the IFNγ signaling pathway contained residues with both negative and positive regulatory roles in PD-L1 expression. Notably, STAT1 and STAT3 were identified in this context (Fig. 2d), which could not be detected in canonical screens at the gene level. STAT1, an important transcription factor connecting cytokine receptors with downstream target genes, is involved in the signaling of many cytokines, including IFNγ. In our screening, numerous functional S/T/Y sites were identified on the STAT1 protein, distributed across its four domains as well as the coiled-coil region (Fig. 3a). Among them, five mutations were confirmed to upregulate PD-L1 expression, with two in the coiled-coil region and three in the DNA binding domain. The majority of mutations appeared to inhibit PD-L1 expression and were dispersed across functional regions, including the N-terminal domain, DNA-binding domain, SH2 domain, phosphorylated tail segment, and the transcriptional activation domain. One of the well-known sites was STAT1_Y701, located in the phosphorylated tail segment, where phosphorylation is required for the dimerization and nucleation of STAT1.³⁰ Besides Y701, we also identified another confirmed phosphorylation site, STAT1_Y106, and 9 predicted phosphorylation sites that resulted in decreased PD-L1 expression following mutation, among which 7 sites were in the SH2 domain, indicating a close relationship between the SH2 domain and phosphorylation-mediated signaling transmission.

Novel residues on canonical and non-canonical regulatory proteins involved in IFNγ signal transduction affect PD-L1 expression. a Distribution of identified S/T/Y residues on STAT1 protein. b Protein expression levels of STAT1 and PD-L1 in the indicated A375 mutant cells treated with IFNγ. The upper heatmap shows the relative surface PD-L1 level of A375 cells with each mutation, based on the results of flow cytometric analysis from Fig. 2d. The lower IB analysis shows the total protein level of STAT1 and PD-L1 for each corresponding mutant. c Pie chart of STAT1 residues that are classfied based on their differential regulation of STAT1, total PD-L1, and surface PD-L1 expression. d Schematic of the molecular structure and intramolecular interactions around the STAT1_S462 residue within tyrosine phosphorylated STAT1 and DNA complex (PDB: 1BF5). The WT S462 or the mutated G462 residue is labeled in yellow (left). The table shows the interaction between STAT1_S462/G462 and DNA chain A/B, indicated by the parameters of interfacing residue (IR) and buried surface area (BSA) (right). e Distribution of identified S/T/Y residues on SH2B2 and SH2B3 proteins. f IB analysis of typical JAK/STAT signaling components, SH2B3, and PD-L1 in A375 cells infected with sgRNAs targeting AAVS1 and each mutation, respectively. g IB analysis of anti-Flag immunoprecipitates (IPs) and whole-cell lysates (WCLs) of HEK293T cells co-transfected with the indicated plasmids expressing HA-tagged JAK2 and Flag-tagged SH2B3 WT or variants. h IB analysis of anti-HA IPs and WCLs of HEK293T cells co-transfected with the indicated plasmids expressing Flag-tagged CBL and HA-tagged SH2B3 WT or variants. i Distribution of identified S/T/Y residues on PTPN1 and PTPN2 proteins. j IB analysis of typical JAK/STAT signaling components, PTPN1, and PD-L1 in A375 cells infected with sgRNAs targeting AAVS1 and PTPN1_S270, respectively. k Schematic diagram of SH2-B family proteins and PTP family proteins regulating the IFNγ-induced JAK/STAT signaling pathway. The S/T/Y residues identified in the screens are labeled on the corresponding proteins. For (a, e, and i), the regulatory residues are marked in two directions on the protein structure: mutations above the line indicate negative regulators and mutations below the line indicate positive regulators. The relative length of each vertical line reflects the regulatory effect of the indicated residue, based on the results of flow cytometric analysis from Fig. 2d. All cell samples were treated with 100 ng/mL IFNγ for 48 h

We further performed immunoblot (IB) verification for all the selected sites within STAT1. Five PD-L1^high variants consistently upregulated PD-L1 expression in both total and membrane protein levels, while leaving the STAT1 protein level unchanged (Fig. 3b). We hypothesized that these variants represent gain-of-function (GOF) mutations that promote the shuttle of STAT1 into the nucleus, facilitating its binding to DNA. Conversely, the majority of PD-L1^low mutations, distributed across various domains of STAT1, had an inhibitory effect on STAT1 expression, resulting in a reduction in both total and membrane protein levels of PD-L1. Interestingly, nearly half of these variants had no discernible impact on STAT1 expression. Some of them only affected the membrane PD-L1 levels, leaving the total PD-L1 level unchanged. This category includes mutations such as Y651 and S715. Others induced PD-L1 reduction in both total and membrane protein levels (Fig. 3c). One notable example in this category is STAT1_S462G, a variant located in the DNA binding domain of STAT1, which was speculated to destroy the interaction between STAT1 and DNA. To verify this conjecture, we investigated the interaction between STAT1 and DNA before and after S462 mutation using the PDBePISA website (https://www.ebi.ac.uk/pdbe/pisa/). The analysis revealed that STAT1_S462 has interface contact with both strands of DNA, indicating that S462 is located at the interaction interface. However, S462G mutation completely abolished the ability of STAT1 to interact with one strand of DNA and decreased the contact area with the other DNA strand (Fig. 3d). The analysis indicated that S462G mutation is likely to reduce the binding capacity of STAT1 to DNA, weakening its transcriptional effect and ultimately affecting the expression level of PD-L1.

Critical S/T/Y residues within adaptor proteins and tyrosine phosphatases participate in the regulation of PD-L1 expression

In addition to the proteins directly involved in the IFNγ signaling pathway mentioned above, a series of mutations were associated with the regulation of phosphorylation. Among them, multiple residues on two types of proteins, which belong to the Src homology 2-B (SH2-B) protein family and the protein tyrosine phosphatase (PTP) protein family, were significantly enriched in the PD-L1^high screen. It’s worth noting that, for most of these proteins, their relevance in PD-L1 regulation especially at the residue level has not been intensively investigated in previous studies.

The SH2-B family, comprising SH2B1, SH2B2 (APS), and SH2B3 (Lnk),³¹ is a conserved family of adaptor proteins with similar structural characteristics. They possess a Pleckstrin homology domain (PH) that recognizes phosphatidylinositol lipids, enabling protein transfer to the cell membrane, as well as an SH2 domain that recognizes phosphorylated tyrosine residues (Fig. 3e). In human cells, SH2-B proteins recognize and bind to phosphorylated Y813 of JAK2 via their SH2 domains,^32,33 and an active region at the C-terminal of these proteins gets phosphorylated, interacting with the tyrosine kinase binding (TKB) domain of the intracellular E3 ubiquitin ligase CBL (c-cbl). This interaction recruits CBL to the vicinity of JAK2, leading to the degradation of JAK2 through ubiquitination modification, thereby negatively regulating the IFNγ signaling pathway.³⁴

We found that the residues with the most significant effects were clustered in the SH2 domain of these proteins, one for SH2B2 and four for SH2B3 (Fig. 3e). For SH2B3, all four mutations did not alter the protein level of SH2B3 but significantly increased the total abundance of PD-L1 protein. These mutations were found to activate the IFNγ-induced JAK-STAT signaling pathway, with JAK2 showing increased abundance and pSTAT1 levels significantly elevated (Fig. 3f). The overall pattern of SH2B2_S513 closely resembled that of SH2B3 (Supplementary Fig. 3e), suggesting that both proteins influence JAK-STAT signaling by regulating the protein abundance of JAK2. Additionally, CBL_Y274 was identified and verified in the study, which located within the TKB domain and closely related to the recognition of SH2-B family.³⁴ Its regulation on downstream JAK-STAT signaling was consistent with that of SH2B2 and SH2B3 (Supplementary Fig. 3f), further highlighting the critical role of the “JAK2-adaptor-CBL” loop in regulating IFNγ-mediated JAK/STAT signaling pathway and PD-L1 expression.

To further understand the regulatory mechanisms of these mutations, we focused on the SH2 domain and selected representative sites, namely SH2B3_S417, S444, and SH2B2_S513, for further investigation. Genomic sequencing indicated that targeting SH2B3_S417 generated L416P mutation, SH2B3_S444 generated the expected S444P mutation, and SH2B2_S513 targeting generated S513P and the bystander mutation L512P (Supplementary Material). Consequently, we overexpressed both the wild-type (WT) cDNAs and all the corresponding mutant sequences of these two genes to perform co-immunoprecipitation (Co-IP) experiments with JAK2 and CBL, respectively. Both the L416P and S444P mutations in SH2B3 simultaneously disrupted the interaction between SH2B3 and JAK2, as well as CBL, with a particularly notable impact on the interaction with JAK2 (Fig. 3g, h). This severe destruction in the interactions with both JAK2 and CBL resulted in a weakened ubiquitination degradation of JAK2, leading to JAK2 upregulation and enhanced IFNγ signal, ultimately promoting PD-L1 expression. Intriguingly, distinct from the residues in SH2B3, neither the SH2B2_L512P nor the L513P single mutant, as well as the L512P/S513P double mutant, affected the interaction between SH2B2 and JAK2. However, these mutations significantly reduced the interaction between SH2B2 and CBL (Supplementary Fig. 3g, h). We speculated that SH2B3_L416 and S444 are located close to the interaction center where the SH2 domain binds to JAK2_Y813,³⁴ while not for SH2B2_S512 or S513, thereby leading to a clear disruption in the interaction between SH2B3 and JAK2 after mutation. The analysis above suggests that mutations, especially those occurring within the SH2 domain of these two adaptor proteins, can dramatically affect the IFNγ signaling pathway, albeit through different regulatory patterns.

The screens also identified functional residues within PTPN1 and PTPN2, two members of the PTP family known to negatively regulate the cytokine signaling pathway through dephosphorylation of phosphorylated tyrosine residues on targeted proteins.^35,36 Most of the identified residues were enriched within the phosphatase domain of each protein (Fig. 3i). As such, we speculated that these mutations might affect their phosphatase activity, thereby enhancing the transmission of IFNγ signal.

We noticed that PTPN1_S270/Y271 and PTPN2_S268/Y269 are homologues residues, implying that they might exert their regulatory effects through similar mechanisms. We selected PTPN1_S270 and PTPN2_S268 as representatives and confirmed that their respective targeting sgRNAs generated the expected mutations with minimal bystander editing (Supplementary Material). To verify the function of these dominant mutations, we separately overexpressed the WT cDNA and the corresponding mutants in A375 cells. Introduction of PTPN1_S270P or PTPN2_S268P variant decreased the expression of PTPN1 or PTPN2, respectively. This, in turn, resulted in an upregulation of PD-L1 in both total and membrane protein levels (Supplementary Fig. 3i). To comprehensively investigate the regulation of these two endogenous mutations in A375 cells, we focused on examining typical proteins involved in IFNγ signaling pathway. Both mutations increased the protein levels of JAK2, subsequently leading to a significant upregulation in pSTAT1 levels without changing the overall abundance of STAT1 protein (Fig. 3j; Supplementary Fig. 3j). These results suggest that PTPN1_S270P and PTPN2_S268P activate the IFNγ pathway by reducing the abundance of each respective protein, ultimately resulting in an increased pSTAT1 level and, consequently, an upregulation of PD-L1 expression. Similarly, we found that multiple mutations identified in PTPN1 and PTPN2 also led to a reduction in their own protein levels and an increase in the total and membrane PD-L1 abundance (Supplementary Fig. 3k; Fig. 2d). For these loss-of-function (LOF) mutations, their subsequent effects were consistent with the outcomes of knocking out PTPN1 or PTPN2 using the CRISPR/Cas9 system (Supplementary Fig. 3l).

We summarized the critical S/T/Y residues within the SH2-B and PTP family proteins to illustrate their regulatory effects on the canonical IFNγ pathway (Fig. 3k). Additionally, we evaluated the impact of representative residues involved in the IFNγ pathway on surface PD-L1 expression in additional human cancer cell lines, including the melanoma cell line A875, the sarcoma cell line HT1080, and the breast cancer cell line MCF-7, all of which were engineered to stably express ABEmax protein. In A875 cells, all selected variants showed the same phenotypes as those observed in A375 cells, suggesting the broad relevance of these functional sites across melanoma cell lines. In HT1080 and MCF-7 cells, most variants on STAT1, SH2B3, SH2B2 and PTPN2 were confirmed as functional, with the exception of SH2B3 and SH2B2 variants, which showed little or no effects in MCF-7 cells (Supplementary Fig. 4). These findings suggest that most variants retain functionality across different tumor backgrounds.

Based on the screening and validation results, in combination with prior related studies, we have created a gene network diagram outlining PD-L1 modulation at the single amino acid level (Supplementary Fig. 5). The rich information of functional residues contributes to a better understanding of the roles played by these corresponding proteins and provides initial insights for refining the PD-L1 regulatory network from a single residue perspective.

Genome-wide mapping of critical residues modulating HLA-I expression using S/T/Y library

Tumor cells can employ various strategies to evade immune surveillance. In addition to increase the expression of immune checkpoint ligands, defects in MHC-I-mediated antigen processing and presentation (APP) can directly hinder the tumor recognition of CD8 + T cells and restrain their activation and proliferation.³⁷ Genetic mutations in essential genes involved in MHC-I APP have been implicated in tumor progression and the development of resistance to ICB therapy.^13,38 Therefore, beyond interpreting the regulation of PD-L1 pathway, it is also crucial to understand the regulatory mechanisms of MHC-I in tumor cells.

We thus performed an additional S/T/Y library screen to investigate the functional residues that modulate HLA-I expression in A375 cells. Using the pan-human HLA-I-specific antibody W6/32 for protein staining, we observed a relatively high level of surface HLA-I expression in A375 cells without IFNγ stimulation, which enables to identify both positive and negative regulators of HLA-I expression in this context (Supplementary Fig. 6a). Consequently, we conducted the library screen for HLA-I regulators in A375-ABEmax cells in the absence of IFNγ. Through the same procedure of FACS enrichment and data analysis as described for the PD-L1 screen (Fig. 1a; Supplementary Fig. 6b, c), we identified numerous sites within genes related to APP that were prominent in HLA^low cells (Fig. 4a). These regulators included multiple allelic variants of HLA, the TAP binding protein TAPBP (tapasin), antigen transporters TAP1 and TAP2, and the component of MHC-I complex, B2M. In the HLA^high group, we also observed novel sites enriched on several negative regulators of HLA-I, including SUSD6_Y177 and WWP2_Y704, whose coding genes were recently reported to form an HLA-I inhibitory axis (SUSD6/TMEM127/WWP2) for cancer immune evasion.³⁹

ABE-based screens identify functional S/T/Y residues modulating HLA-I expression at the genome-wide level. a Significant S/T/Y residues enriched from the sense library (left) and antisense library (right) that upregulate or downregulate HLA-I expression in A375 cells without IFNγ stimulation. The screening procedure is the same as that in Fig. 1a. Positively or negatively enriched sites were selected based on a screen score > 1 or < −1. b, c GO enrichment analysis of related genes haboring identified mutations that lead to HLA-I upregulation (b) and downregulation (c) in the absence of IFNγ. d Individual validations of representative sites related to APP and protein glycosylation in A375 cells in the absence of IFNγ by flow cytometry analysis. The method to generate relative MFI of HLA-ABC are the same as those shown in Fig. 2c, d. e Editing outcomes of sgRNA targeting SLC35A1_Y98 by NGS analysis. f Protein expression levels of SLC35A1 and HLA-ABC in A375 cells infected with sgRNAs targeting AAVS1 and SLC35A1_Y98, respectively. g Relative mRNA expression levels of SLC35A1 in A375 cells infected with respective sgRNAs targeting AAVS1 and SLC35A1_Y98, respectively. The mRNA level of each sample was quantified by real-time qPCR and normalized to GAPDH. The indicated relative mRNA level of each sample was normalized to that of AAVS1-targeting control cells. h Relative MFI of surface sialic acid in A375 cells infected with sgRNAs targeting AAVS1 and SLC35A1_Y98 by flow cytometry analysis. Data were normalized to that of the isotype. i Killing resistance and sensitivity of A375 cells infected with sgRNAs targeting residues on glycosylation-related genes to expanded anti-NY-ESO-1 CD8 + T cells. j, Schematic diagram of the HLA-I regulatory network focused on identified residues on representative APP and glycosylation-related genes. For (d, g, h, and i), data are presented as the mean ± SD (n = 3). P values were calculated using two-tailed Student’s t test, **P < 0.01, ***P < 0.001, ****P < 0.0001; n.s., not significant

Upon integrating all the relevant genes identified through the screens, a GO analysis of biological process revealed that several terms similar to those from the PD-L1 screens were among the top-ranked in the HLA^high group. These terms included processes related to histone modification, covalent chromatin modification, and peptidyl-lysine modification, highlighting the general regulatory influence of genes on both PD-L1 and HLA-I (Fig. 4b). In contrast, in the HLA^low group, multiple terms related to antigen processing and presentation were highly enriched (Fig. 4c). STRING PPI network analysis of top-ranked regulators further revealed genes involved in antigen processing and presentation, immune response, and cellular protein modification process (Supplementary Fig. 6d). Of note, we identified several nodes connecting multiple networks, such as HLA-A, B2M, indicating their central roles in regulating the expression of each respective protein.

Interpretation of novel residues regulating antigen recognition and presentation

Compared with the candidate sites identified in the PD-L1 screens, the HLA-I screens revealed a multitude of unique variants with unknown functions that were enriched in both high and low directions, not limited to sites within APP-related genes. To further assess their impact on surface HLA-I expression, we conducted a large-scale verification of candidate sites enriched on various regulatory pathways (Fig. 4d; Supplementary Fig. 7a).

In the HLA-I screens, a dominant category of functional sites enriched in APP-related genes. Nearly all relevant residues were subjected to individual validation, all of which were verified to significantly reduce surface HLA-I expression upon targeted mutation (Fig. 4d). For the gene B2M, two noteworthy sites, Y30 and S31 located in its Ig-like C1-type domain (Supplementary Fig. 7b), showed significant phenotypic effects. Targeting each of these two sites led to a double mutation, Y30H and S31P, which led to reduced levels of both membrane and total HLA, while B2M expression remained unaltered (Supplementary Fig. 7c, d). As for the TAP1 and TAP2 genes, multiple hits within TAP1 were mainly localized in its N-terminal domain and ABC transmembrane type-1 domain, and functional residues of TAP2 converged on its ABC transmembrane type-1 domain (Supplementary Fig. 7e). Unlike the mutants of B2M, several top-ranked mutations of TAP1 not only disrupted its own expression but also further reduced HLA expression at both total and membrane protein levels (Supplementary Fig. 7f). As for TAPBP (Supplementary Fig. 6e), we investigated eight mutants with significant phenotype of reduced surface HLA levels, among which five mutations slightly reduced the overall HLA expression and three had no significant effect on total HLA levels (Supplementary Fig. 7g).

In particular, we identified a significant number of mutations in glycosylation-related genes that were enriched in both HLA^high and HLA^low groups. Among these, targeting SLC35A1_Y98 with ABEmax led to a dramatic increase in surface HLA expression (Fig. 4d), which was further verified to result in the generation of the Y98C mutation (Fig. 4e). SLC35A1 is a membrane-bound transporter located in the Golgi apparatus responsible for transferring CMP-sialic acid from the cytosol into the Golgi apparatus. This process facilitates the sialylation of proteins by various sialyltransferases. Importantly, the Y98C mutation did not interfere with the expression of SLC35A1 at both the protein and mRNA levels (Fig. 4f, g). The residue Y98 is involved in the binding of CMP and CMP-sialic acid and is essential for optimal transport competence, as confirmed by previous in vitro studies.⁴⁰ We thus examined the sialic acid level on the cell membrane and found that SLC35A1_Y98C mutation significantly impaired the transport of sialic acid (Fig. 4h). To further explain the relevance between SLC35A1 mutation and HLA abundance, we assessed the total HLA level in SLC35A1_Y98 mutant cells. Intriguingly, we found that the mutation increased the median fluorescence intensity (MFI) of surface HLA without changing overall HLA expression (Fig. 4d, e). Prior study has reported that sialic acid residues on glycosphingolipid (GSL), synthesized by B3GNT5, is involved in shielding critical epitopes of HLA-I molecules on the cell surface, thus diminishing their interactions with several immune cell receptors and decreasing CD8 + T cell responses.⁴¹ We thus infer that the change in sialic acid modification generated by SLC35A1_Y98 mutation may affect the accessibility of epitope recognition of HLA-I specific antibodies.

Similarly, several sites were identified in additional glycosylation-related genes. These genes included negative regulators, such as another nucleotide sugar transporter called SLC35A2 responsible for transporting UDP-galactose, as well as the glycosyltransferase C1GALT1 and its chaperone C1GALT1C1. On the positive regulatory side, there were genes like SPPL3, which negatively regulates B3GNT5 expression, thereby controlling GSL synthesis.⁴¹ Additionally, MOGS (α-glucosidase I), which encodes the first enzyme responsible for trimming N-glycans in the endoplasmic reticulum (ER),⁴² and PDIA3 (ERp57), which is involved in the general glycoprotein folding process within the ER and is required for optimal tapasin activity,⁴³ also had identified sites. While most of the residues identified in these glycosylation-related genes exhibited no effect on the overall HLA-I expression, they significantly influenced surface HLA-I levels (Supplementary Fig. 7h). Therefore, it is likely that these residues can affect the glycosylation of various regulators or perturb the structural stability of glycoproteins involved in APP process.

To evaluate the role of these residues in immunosurveillance, we investigated the susceptibility of representative mutants, particularly those related to glycosylation, to CD8 + T cell-mediated cytotoxicity. We generated each mutation in A375-ABEmax cells, which endogenously express HLA-A2 and NY-ESO-1 antigen. Human CD8 + T cells transduced with an HLA-A2-restricted T cell receptor (TCR) specific for the NY-ESO-1 antigen⁴⁴ were co-cultured with the mutant cells. We found that the HLA^high variants were more sensitive to T cell-driven killing, with SLC35A1_Y98 being an example. In contrast, the HLA^low mutations conferred significant resistance to T cell killing, thus subverting T-cell-mediated immunosurveillance (Fig. 4i; Supplementary Fig. 7i). These novel residues have been summarized for their functional roles in different glycosylation processes (Fig. 4j), which can alter the glycosylation of related regulatory genes or impact the quality control machinery of glycoprotein involved in the assembly of HLA-I molecules, consequently affecting the recognition of tumor cells by T cells.

Integrated analysis for potential co-regulators of surface PD-L1 and HLA-I

The above analyses drew a comprehensive map of regulators for surface PD-L1 and HLA-I at the residue level. However, in the in vivo tumor microenvironment, various factors collectively influence the fate of tumor cells. To gain a deep understanding of the co-regulators of PD-L1 and HLA-I, two of the principal factors for immunotherapy, we conducted a comparison of candidates from the HLA-I and PD-L1 screens in the presence and absence of IFNγ stimulation (Fig. 5a). This analysis identified five mutants that upregulated HLA-I expression while downregulating PD-L1 expression in the presence of IFNγ, including Y137 and Y138 of the N-terminal acetyltransferase NAA20 (Supplementary Fig. 2b; Supplementary Fig. 7a). These mutants were validated to significantly sensitize A375 cells to CD8 + T cell-mediated cytotoxicity, suggesting their potential roles as positive regulators of antitumor immunity (Supplementary Fig. 7i). Conversely, one mutant, MAPK3_Y333, was found to downregulate HLA-I expression and upregulate PD-L1 expression and render A375 cells to be more resistant to T cell killing (Fig. 2c; Fig. 4d; Supplementary Fig. 7i), indicating its potential role in promoting tumor evasion. Additionally, we identified 13 mutants that concurrently upregulated HLA-I and PD-L1 expressions, including six hits that increased PD-L1 levels upon IFNγ treatment, such as EZH2_Y153, EED_Y308, and SETD2_Y1666, all of which are involved in epigenetic modulations.

Interpretation of functional residues that co-regulate surface PD-L1 and HLA-I. a Comparison of S/T/Y residues identified in PD-L1 screens and HLA-I screens using venn diagram. b General information of SETD2_Y1666. The upper structure schematic indicates the location of Y1666 residue on SETD2 protein. The lower figures (left) indicate the editing outcomes of sgRNA targeting SETD2_Y1666 by NGS analysis. The lower table (right) indicates the information of clinical relevance of SETD2_Y1666. c Relative MFI of surface PD-L1 and HLA-I in A375 cells infected with sgRNAs targeting AAVS1 and SETD2_Y1666 with different IFNγ treatment. The method to generate relative MFI of PD-L1 or HLA-I and the statistics are the same as those shown in Fig. 2c, d. d Protein expression levels of SETD2, PD-L1, HLA-ABC and H3K36me3 in A375 cells infected with sgRNAs targeting AAVS1 and SETD2_Y1666, respectively. e Volcano plots showing the DEGs between SETD2_Y1666-targeted A375 mutant cells and AAVS1-targeted A375 control cells. The represented genes are listed. f Representative KEGG pathway analysis of upregulated DEGs in SETD2_Y1666-targeted A375 mutant cells compared with the AAVS1-targeted control. The DEGs were selected using the threshold of FC > 1.5 and P value < 0.1 according to the RNA-seq data. g ChIP-seq tracks for H3K36me3 at SH2B3 gene locus between SETD2_Y1666-targeted A375 mutant cells and AAVS1-targeted A375 control cells. h IB analysis of SH2B3 and typical JAK/STAT signaling components in A375 cells infected with sgRNAs targeting AAVS1 and SETD2_Y1666, respectively

To explore the regulatory mechanisms of these novel co-regulators, we focused on the functional investigation of the category with the largest number of mutants, which increased the expression level of both PD-L1 and HLA-I. Among them, SETD2_Y1666, as well as the corresponding coding gene, stood out as a novel regulator, whose relevance with PD-L1 or HLA-I has not yet been reported. SETD2 is the primary histone methyltransferase responsible for catalyzing H3K36me3, representing a marker of transcriptional activation. SETD2 is associated with diverse biological functions, such as maintenance of genomic stability,⁴⁵ antiviral immune response,⁴⁶ and restriction of tumor metastasis.⁴⁷ SETD2 mutations are prevalent in various human tumors and are reported to be associated with tumor progression, including glioma, clear cell renal cell carcinoma, leukemia, and prostate cancer.^48,49,50,51 We then investigated the impact of SETD2 gene knockout and observed that its disruption significantly upregulates both the cell surface and total PD-L1 and HLA-I expression levels with IFNγ treatment (Supplementary Fig. 8a, b).

For the SETD2_Y1666 mutation, Y1666 is in the SET domain of SETD2, which is the catalytic domain mediating H3K36me3-specific methyltransferase activity.⁵² SETD2_Y1666 targeted by ABEmax could generate the Y1666C mutation, a reported mutation from both COSMIC (Catalogue of Somatic Mutations in Cancer) and ICGC database (Fig. 5b). We found that Y1666C didn’t change the expression of SETD2 at both the mRNA and protein levels, but it significantly increased the total and membrane protein levels of PD-L1 and HLA-I upon IFNγ exposure (Fig. 5c, d). Meanwhile, the expression level of H3K36me3 was markedly decreased, suggesting that the Y1666C mutation disrupted the catalytic activity of SETD2 without affecting its own protein expression (Fig. 5d).

To elucidate the mechanisms of PD-L1 and HLA-I regulation by SETD2_Y1666, we performed RNA-seq and H3K27me3 ChIP-seq analysis for SETD2_Y1666C mutant cells and control cells with IFNγ stimulation, gaining insight into the potential targets of SETD2. We analyzed the differential expressed genes (DEGs) from the RNA-seq data and identified numerous representative upregulated DEGs in the mutant cells, as exemplified by CD274, IRF1, TAP1, B2M, HLA-A, HLA-B and HLA-C, all of which are directly associated with PD-L1 and HLA-I expressions (Fig. 5e). By analyzing the enriched KEGG pathways of upregulated genes, we found dominant terms, including cytokine-cytokine receptor interactions, transcriptional misregulation in cancer, NF-κB signaling pathway, JAK-STAT signaling pathway, and antigen processing and presentation (Fig. 5f). We further referred to the ChIP-seq data to search for the methylated targets of SETD2 and found genes with a significant reduction in H3K36me3 signal, such as RCC1 (Supplementary Fig. 8c), which was reported to enhance PD-L1 expression and improve the ICB sensitivity after gene knockdown.⁵³ RCC1 was also downregulated in the RNA-seq analysis (Fig. 5e), indicating that SETD2_Y1666 mutation could decrease the H3K36me3 modification of RCC1, thus upregulating PD-L1 expression. Interestingly, multiple gene body regions of SH2B3 exhibited a remarkable lower H3K36me3 signal (Fig. 5g), and SH2B3 appeared to be downregulated upon SETD2_Y1666 mutation (Fig. 5e). Considering the negative regulation of SH2B3 on the JAK-STAT signaling pathway (Fig. 3f, g) and the detected enrichment of JAK-STAT signaling in SETD2_Y1666C mutant cells (Fig. 5f), we further investigated the effects on this pathway when Y1666 was mutated. We found that SETD2_Y1666C conferred a significant reduction in SH2B3 expression, along with higher expression of JAK2 and pSTAT1 (Fig. 5h; Supplementary Fig. 8d), which correlated with the effects of SH2B3 mutants. We also detected upregulation of IFNγ responsive genes such as IRF1, some interferon stimulated genes including ISG15, ISG20, and MX1 (Fig. 5e, Supplementary Fig. 8d), which are associated with the upregulation of PD-L1 and HLA-I.^54,55 The above analysis revealed that the Y1666 mutation in the SET domain of SETD2 could boost JAK-STAT signaling pathway, thus increasing PD-L1 expression and antigen processing and presentation.

Functional clinical mutations promote cancer immunotherapy in different tumor models

Considering the regulatory impact of these clinically relevant mutations on both PD-L1 and HLA-I, we intended to dissect their potential effects on tumor progenesis and response to ICB treatment in vivo. Therefore, we generated the homogenous Setd2_Y1640 mutation in the B16F10 mouse melanoma cell line using the ABEmax system, corresponding to the human SETD2_Y1666C mutation. The sgRNA targeting Setd2_Y1640 were infected into B16F10-ABEmax cell line, which resulted in a similar editing pattern and consistent phenotype as observed in A375 cells (Supplementary Fig. 9a, b). Subsequently, we separately injected the Setd2_Y1640-targeted B16F10 cells into the immune-competent C57BL/6 mice, as well as negative control samples infected with an sgRNA targeting the safe-harbor locus, to establish B16F10 melanoma tumors (Fig. 6a). We observed a reduction in tumor growth in Setd2_Y1640-targeted mice, and the combination of anti-PD-1 treatment further inhibited tumor progression (Fig. 6b). Meanwhile, we observed a consistent tumor growth pattern between the mutant group and the control in the immune-deficient BALB/C nude mice (Supplementary Fig. 9c), indicating that the mutation contributes to tumor suppression only through reshaping the immune microenvironment.

Clinically relevant mutation SETD2_Y1666/Setd2_Y1640 contributes to an improved response to ICB therapy in different tumor models. a A schematic view of implanting B16F10 mutant cells and control cells into C57BL/6 mice and the following treatment of PD-1 mAb or IgG isotype control (IgG2a). b Longitudinal tumor size of the indicated B16F10 tumors in C57BL/6 mice treated by control IgG or ICB. c Quantification of GzmB represented as percentages on CD8+ TILs in B16F10 tumors harvested from C57BL/6 mice after the indicated treatments. d Longitudinal tumor size of the indicated CT26 tumors in BALB/c mice treated by control IgG or ICB. e Quantification of GzmB represented as percentages on CD8+ TILs in CT26 tumors harvested from BALB/c mice after the indicated treatments. f Assessment of the impact of representative regulators on cell surface PD-L1 or HLA-I expression in A875, HT1080 and MCF-7 cells by flow cytometry analysis. Cell surface PD-L1 was analyzed following incubation with 100 ng/mL IFNγ for 48 h. Correlation between the SETD2_Y1666-mutation signature and PD-L1 expression, MHC-I expression, intratumoral CTL infiltration (g), ICB response (h), and overall survival and progression-free survival (i) in patients treated with anti-PD-1 in the Gide et al. study⁵⁶ in melanoma. r: Pearson correlation coefficient, PD: progressive disease, SD: stable disease, PR: partial response; CR: complete response. P value was respectively calculated by two-tailed Student’s t test (g) and log-rank test (h). j Schematic of representative residues identified in PD-L1 and HLA-I screens with clinical relevance according to ICGC database. X axis indicates functional residues regulating PD-L1 or HLA-I from the ABE screens. Y axis indicates different cancer types defined in ICGC database. The dot size represents the detected missense mutation rate of each indicated residue. For (b, d), data are presented as the mean ± S.E.M. (n = 5–6 mice/group) for each group at each time point. P values were calculated using Two-way ANOVA with Benjamini-Hochberg adjustment for multiple testing, *P < 0.05, **P < 0.01, ****P < 0.0001; n.s., not significant. For (c, e), data are presented as the mean ± SD (n = 5–6 mice/group) for each group at each time point. P values were calculated using two-tailed Student’s t test, *P < 0.05, **P < 0.01, ****P < 0.0001; n.s., not significant

To further investigate the impact of Setd2_Y1640 mutation and its combination with ICB treatment on TME, we analyzed infiltrated immune cells in B16F10 tumor-bearing C57BL/6 mice. In Setd2_Y1640 mutant group, we observed increased expression of the T cell activation marker Granzyme B (GzmB) in infiltrated CD8 + T cells compared to the control, and the combination of ICB treatment further strikingly elevated the ratio (Fig. 6c). These results indicated that the mutation might reshape the TME through the activation of representative signaling pathways, including NF-κB and JAK-STAT, which in turn upregulated PD-L1 and HLA-I expression. This enhances the cytotoxicity of tumor infiltrating CD8 + T cells and improves the efficacy of anti-PD-1/PD-L1 blockade therapy in vivo.

We also validated the Setd2_Y1640 mutation in an additional colorectal cancer mouse model, the CT26 tumor-bearing model. CT26 mutant cells and control cells were respectively implanted into immunocompetent BALB/c mice, followed by anti-PD-1 or IgG treatment at the specified time point. Consistent with results from the B16F10 tumor model, the combination of anti-PD-1 treatment significantly inhibited tumor progression in the Setd2_Y1640 mutant group. Additionally, elevated GzmB expression levels were also observed in infiltrated CD8 + T cells in the ICB combination group compared to the control (Fig. 6d, e). Furthermore, we investigated the effects of the SETD2_Y1666 mutation in other human cell lines and found that it exhibited consistent phenotypes in A875, HT1080 and MCF-7 cell lines, further demonstrating its broad applicability across different tumor backgrounds (Fig. 6f).

After confirming the effects on immune response in the mouse model, we next attempted to analyze the correlation between genetic mutation-derived functional deficiency and the response to immunotherapy in published ICB treatment cohorts. We first derived the gene expression signature of the SETD2_Y1666C mutation based on its RNA-seq results, as described in a previous study.¹¹ Referring to 91 RNA-seq samples from 54 patients in a melanoma cohort treated with anti-PD-1,⁵⁶ we confirmed that the SETD2_Y1666C-mutation signature was positively correlated with tumor PD-L1, MHC-I, and cytotoxic T-cell infiltration (Fig. 6g). Further analysis revealed that patients responding to ICB therapy (partial response and complete response: PR/CR) exhibited higher SETD2_Y1666C-mutation signature compared with non-response groups (progressive disease and stable disease: PD/SD) (Fig. 6h), and the mutation signature also showed a positive correlation with progression-free survival (Fig. 6i). Interestingly, recent studies also found that patients with different cancer types that harboring SETD2 deleterious mutations showed improved response to ICB therapy.^57,58 Collectively, these findings firstly demonstrated the mechanisms of SETD2_Y1666C mutation in modulating immune surveillance and further supported the notion that the mutation is relevant to a better response to ICB treatment in clinical trials. Due to the high mutation rate of SETD2 in various cancer types, SETD2 may serve as a biomarker for ICB treatment, and a large population of patients may benefit from immunotherapy.

In addition to the clinical mutations SETD2_Y1666C described above, we sought to investigate the clinical relevance of all selected mutants identified in the screens. Referring to different sequencing data from cancer patients, including ICGC and COSMIC, we found 168 sites with detected mutations across 35 tumor types in ICGC (Fig. 6j; Supplementary Fig. 10a), and more than 300 sites recorded in COSMIC (Supplementary Fig. 10b–d). Overall, nearly 40% (416/1083) of the identified residues from the three screens were clinically observed in these databases, providing a rich resource of potential pathogenic mutations, especially those linked to cancer. Furthermore, this information offers guidance on the efficacy of ICB for patients harboring these mutations.