Inhibition of METTL3 Results in a Cell-Intrinsic Interferon Response That Enhances Antitumor Immunity

N6-methyladenosine (m6A) is the most abundant mRNA modification described and is typically implicated in mRNA stability (1), splicing (2), and protein translation (3). This dynamic modification is laid down cotranscriptionally by METTL3 (4), the enzymatically active component of the m6A writer complex. A testament to their importance in cell biology is the fact that METTL3 and its heterodimeric partner, METTL14, are common essential genes necessary for cell survival (5). Recently, the genetic modulation of METTL3 has been used to highlight the role of m6A in regulating normal and malignant hematopoiesis (6, 7), where many of the phenotypes have been linked to an effect on specific transcripts. In turn, these findings have fueled drug-discovery efforts to catalytically inhibit METTL3 in hematologic cancers such as acute myeloid leukemia in which a METTL3 inhibitor has been shown to improve survival in preclinical models (8).

The interplay between METTL3 and the immune system has attracted significant attention over recent years. Loss of METTL3 has been associated with activation of the innate immune system by way of a direct effect on the double-stranded RNA (dsRNA) sensor RIG-I (9) through the derepression of endogenous retroviral elements (ERV; ref. 10), direct stabilization of IFNb (11, 12), and increased formation of dsRNA resulting in a cell-intrinsic inflammatory response (13). Additionally, viruses have also developed mechanisms to hijack the m6A writer complex, methylating their own RNA in order to evade immune recognition (14). Although genetic manipulation of METTL3 has been associated with immune activation, it remains to be determined whether these effects are primarily due to the loss of the m6A RNA modification or if these findings are independent of the catalytic activity of METTL3 and instead attributable to the destruction of the METTL3/METTL14 protein complex.

The advent of immunotherapy has dramatically improved the natural history of a range of cancers, including melanoma and lung cancer. The most successful anticancer immunotherapies to date have all targeted key immune-checkpoint molecules on either tumor cells or T cells (15). Despite the success of these agents in many individuals, the majority of patients treated with these advanced immunotherapeutics fail to be cured and exhibit either primary resistance, a partial response, or late relapse with resistant disease (16, 17). As such, leveraging nonoverlapping mechanisms to further increase the immunogenicity of tumor cells remains a priority. Here, we disclose the structure of a second-generation and structurally distinct METTL3 inhibitor and provide the molecular and preclinical rationale to illustrate that METTL3 inhibition can be leveraged to activate the immune system and enhance antitumor immune responses.

To investigate the therapeutic potential of targeting METTL3's catalytic activity to increase antitumor immunity, we developed a highly potent, selective, and cell-permeable second-generation METTL3 inhibitor, STM3006. This compound is structurally distinct from the previously described inhibitor STM2457 (8) and has improved biochemical and cellular potency (Fig. 1A). Switching the pyrido[1,2-a]pyrimidin-4-one head-group of STM2457 to a 6-bromo-1H-indazole combined with bioisosteric replacement of the pendant 2-carboxamide with a 4-linked 1,2,3 triazole significantly improved binding affinity and enzymatic inhibition. STM3006 inhibited recombinant METTL3 enzyme activity with an IC₅₀ of 5 nmol/L, a value restricted by the lower detection limit of the assay (Fig. 1B). Using a more sensitive assay system, surface plasmon resonance (SPR) analysis demonstrated direct high-affinity binding of STM3006 to METTL3 with a Kd of 55 pmol/L (Fig. 1C). Moreover, STM3006 achieved 20-fold increased cellular potency compared with STM2457, demonstrated by its ability to reduce m6A on polyA⁺-enriched RNA with an IC₅₀ of 25 nmol/L (Fig. 1D). Mass spectrometry analysis confirmed reduction of m6A in the mRNA-rich “elution” fraction compared with the remaining RNA species (“flow-through” fraction; Fig. 1E). In addition, STM3006 potently inhibited proliferation of multiple cell lines (Supplementary Fig. S1A–S1C) and induced dose- and time-dependent activation of apoptosis (Supplementary Fig. S1D and S1E). Importantly, STM3006 showed greater than 1,000-fold selectivity for METTL3 when tested against a broad panel of 45 RNA, DNA, and protein methyltransferases (Fig. 1F). X-ray crystallography of the protein–ligand complex (Fig. 1G) confirmed a competitive binding mode within the S-adenosylmethionine (SAM) pocket (Supplementary Fig. S1F), and overlaying the structure of the STM3006-bound conformation with that of STM2457 [Protein Data Bank (PDB) 7O2I] revealed a largely conserved binding pose (Supplementary Fig. S1G). The retention of key binding interactions observed with the highly selective STM2457, in part, explains the exquisite selectivity of STM3006 for METTL3 over the other methyltransferases tested here. Specifically, as for STM2457, the amine tail of STM3006 is shown to form a salt bridge with Asp395, resulting in a structural reorganization of Lys513 away from Asp395 and toward Glu532. This movement reveals a hydrophobic cavity, adjacent to the homocysteine binding pocket used by SAM, that is occupied by the lipophilic 4,4-dimethyl piperidyl group of STM3006. Additionally, the 1H-indazole of STM3006 makes an H-bond interaction with Asp377 that is not observed with STM2457 (Fig. 1G). Together, these data demonstrate that STM3006 is a novel, chemically distinct inhibitor of METTL3 with increased potency compared with STM2457, providing the field with an additional powerful pharmacologic tool to interrogate this key enzyme central to RNA biology.

We initially sought to understand the transcriptional impact of targeting the catalytic domain of METTL3. To address this issue, we performed whole transcriptome [RNA sequencing (RNA-seq)] analysis on a human ovarian cancer cell line, CaOV3, after treatment with STM3006. Differential gene expression and gene ontology analyses revealed that the dominant transcriptional programs induced in these cells following METTL3 inhibition were signatures associated with interferon signaling and antiviral responses (Fig. 2A and B; Supplementary Fig. S2A). To validate the response of cells treated with METTL3 inhibitors, we performed a Western blot analysis of interferon-stimulated genes (ISG) for both STM2457 and STM3006. These data confirmed a dose-dependent increase of interferon activation following METTL3 inhibition and augmentation of ISGs such as IFIH1 (MDA-5), IFIT1, OAS2, and ISG15 (Fig. 2C). More­over, we could demonstrate a dose-dependent increase in secretion of IFNβ and the chemokine CXCL10 (IP-10) by cells treated with METTL3 inhibitors (Fig. 2D and E). Importantly, analysis of other “m6A regulators” did not show any significant difference in the presence of STM2457 or STM3006, with a possible exception of YTHDF2, which showed a modest decrease after STM3006 treatment (Supplementary Fig. S2B). To illustrate the broad conservation of the ISG cellular response to METTL3 inhibition, we noted a similar transcriptional response in the mouse AT3 triple-negative breast cancer (TNBC) cell line following treatment with STM2457 and STM3006 (Supplementary Fig. S2C–S2G).

Genetic ablation of METTL3 has been associated with the formation of dsRNA (13, 18). We therefore sought to determine whether the ISG signature noted through catalytic inhibition of METTL3 may be accounted for by the formation of dsRNA. Using the anti-J2 antibody, we assessed dsRNA formation in both mouse AT3 TNBC cells (Fig. 2F) and B16 melanoma cells (Supplementary Fig. S2H) treated with the inactive negative control compound STM2120 (8) and the two chemically distinct METTL3 inhibitors, STM2457 and STM3006. Confocal microscopy analysis confirmed increased formation of dsRNA following catalytic inhibition of METTL3, which notably is more prominent with the more potent compound STM3006 (Fig. 2F). Concurrently, we also confirmed increased dsRNA formation and an ISG response in the B16 melanoma cells when METTL3 was knocked down (Supplementary Fig. S3A–S3C), which is in keeping with previously published findings (13). As genes associated with antigen processing and presentation are direct ISGs (19), we next examined MHC-I expression on the surface of cancer cells following treatment with increasing concentrations of STM3006. Our data confirmed a dose-dependent increase of cell-surface MHC-I (Fig. 2G; Supplementary Fig. S3D). Taken together, these data illustrate that catalytic inhibition of METTL3 results in the formation of dsRNA and consequent activation of ISGs, including those associated with antigen presentation.

The precise location and quantitative assessment of m6A is an evolving area of research. One emerging technology that provides insights into the regulation of m6A on mRNA is long-read direct RNA nanopore sequencing (NPS) of polyA-enriched transcripts (20). Despite the inherent lack of sensitivity of this approach, NPS analysis of RNA derived from AT3 cells following treatment with STM2457 and STM3006 revealed several transcripts containing pronounced differences in m6A (Fig. 3A and B; Supplementary Fig. S4A). Interestingly, among the mRNAs with the most significant changes in the m6A-specific DRACH motif were several transcripts associated with antigen presentation, including H2-K1, Calr, and Pdia3 and the ISG Isg15 (Fig. 3A and B). Consistent with our findings, gene ontology analysis of the most significantly modified transcripts in our assays [< −log₁₀(50)] demonstrated most significant enrichment of “MHC class I peptide loading complex” and “antigen processing and presentation” pathways (Fig. 3C). Of note, ISGs were not significantly altered in this pathway analysis, suggesting that the ISG expression signatures occurring in the presence of METTL3 inhibition are unlikely to be due to a preferential alteration of m6A levels on interferon-induced transcripts.

As METTL3 has been implicated in a diverse range of cotranscriptional and posttranscriptional regulatory processes, we next wanted to address the influence of catalytic inhibition on transcription and mRNA stability. We performed a SLAM-seq (Thiol-linked alkylation for the metabolic sequencing of RNA) uridine chase experiment (21) on AT3 cells with and without exogenous IFNγ (Fig. 3D). Our assessment of nascent transcripts suggested that METTL3 inhibition does not substantially affect transcriptional output either at steady state or when transcription is induced through the addition of IFNγ (Fig. 3E). Furthermore, we noted that the predominant consequence of catalytic inhibition of METTL3 was stabilization of newly synthesized transcripts (Fig. 3F). Pathway analysis of the nascent transcripts stabilized by METTL3 inhibition revealed these to be known targets of modulating m6A or its function as an adapter site for protein binding (Fig. 3G; Supplementary Fig. S4B). To validate our findings using an independent method, we performed an actinomycin-D transcription inhibition assay in AT3 cells. These data confirmed that genes involved in antigen presentation are highly induced and these induced transcripts are retained longer in the presence of METTL3 inhibition (Fig. 3H).

In light of the cell-intrinsic activation of ISGs, including those involved in antigen presentation, we next wanted to determine if METTL3 inhibition might affect antitumor immunity. To specifically interrogate the impact of METTL3 inhibition on tumor cell killing by activated cytotoxic CD8⁺ T cells via recognition of tumor-specific antigen presented on MHC-I, we used the OT-I ovalbumin coculture system (ref. 22; Fig. 4A). In these assays, we noted augmented antitumor killing when either STM2457 or STM3006 was added during the coculture of tumor and T cells (Fig. 4B). Consistent with the more potent inhibition of METTL3, we observed greater potentiation of antitumor immunity with STM3006 in a dose-dependent manner (Fig. 4B). To determine if this result is due to a cell-intrinsic effect in the tumor cells, potentiation of T-cell activation, or a combination effect, we first pretreated ovalbumin-expressing B16 tumor cells with a METTL3 inhibitor, washed it out, and then incubated the tumor cells with activated OT-I T cells, which had not been exposed to a METTL3 inhibitor. Similarly, we did the converse and pretreated activated OT-I T cells with the METTL3 inhibitors prior to coculture with ovalbumin-expressing tumor cells that had not been exposed to a METTL3 inhibitor. In keeping with a tumor-specific effect, we noted enhanced killing only when the tumor cells were pretreated with the METTL3 inhibitors (Fig. 4C; Supplementary Fig. S5A), and saw no evidence of augmented tumor killing in the case of T-cell pretreatment (Fig. 4D). These findings in B16 melanoma cells were independently validated in AT3-ovalbumin–expressing TNBC cells (Supplementary Fig. S5B). To provide additional support for the role of antigen presentation in augmented tumor killing to METTL3 inhibitors, we generated β2M knockout B16-ovalbumin–expressing cells and demonstrated that the improved T cell–mediated killing is abrogated (Fig. 4E). Extending the relevance of these findings to human cancer, human SKOV3 tumor cells cocultured with peripheral blood mononuclear cells (PBMC) also confirmed augmented killing in the presence of STM3006, along with a dose-dependent increase in levels of proinflammatory cytokines, including IFNγ, IL1β, and TNFα, in the culture media (Supplementary Fig. S5C–S5E).

We next wanted to obtain an unbiased insight into the main regulators of the enhanced antitumor immunity mediated by METTL3 inhibition. To address this, we coupled whole-genome CRISPR screening with our coculture assays using our B16-ovalbumin tumor model (Fig. 5A). When directly comparing the factors that influence T cell–mediated killing in the presence or absence of METTL3 inhibition, we clearly observed that the antitumor effect of METTL3 inhibition is critically dependent on the ability to sense dsRNA, as well as a competent JAK/STAT signaling pathway coupled with intact antigen presentation (Fig. 5B). These data using a global unbiased approach serve to reinforce the importance of the dsRNA-sensing machinery and the subsequent tumor cell–intrinsic interferon response induced by METTL3 inhibition to drive antitumor immunity. Related to this, we validated that Prkra, a component of the dsRNA-sensing machinery (23), which was the most prominent resistance hit at high doses (Supplementary Fig. S5F and S5G), was specifically required to potentiate T cell–mediated killing only in the context of METTL3 inhibition (Fig. 5B and C). To further illustrate the importance of the dsRNA-sensing machinery in mediating the cell-intrinsic interferon response to increase MHC-I antigen presentation, we created an isogenic series of cell lines in which the four major dsRNA-sensing components were repressed either individually or in combination using CRISPR interference (CRISPRi; Fig. 5D). These data showed that the repression of each of the four dsRNA sensors [PKR, RIG-I, RNAse-L, and IFIH1 (MDA-5)] reduced the higher cell-surface MHC-I induced by METTL3 inhibition (Fig. 5E). Concomitant repression of all four sensors had one of the strongest phenotypes. In summary, these findings confirm that METTL3 inhibition results in the formation and sensing of endogenous dsRNA, resulting in a cell-intrinsic interferon response and enhanced antigen-dependent tumor killing by cytotoxic CD8⁺ T cells.

Although components of the antigen presentation machinery are established interferon target genes, it has also been recognized that key immune-checkpoint molecules such as PD-L1 are simultaneously induced by this inflammatory cytokine to restrain T-cell activation (24). In keeping with its role as an ISG, we saw a dose-dependent increase in PD-L1 expression levels following METTL3 inhibition (Supplementary Fig. S6A and S6B). These data raised the possibility that the antitumor immunity effects of METTL3 inhibition may be further potentiated with the addition of therapies that target the PD-1/PD-L1 checkpoint. We were also interested in understanding if effective antitumor immunity required sustained treatment with a METTL3 inhibitor or if a short duration of treatment was sufficient to initiate and maintain an effective immune response. Although STM3006 is a more potent inhibitor of METTL3, its pharmacokinetic properties, particularly the rapid metabolism of the compound, precluded its use in vivo (Supplementary Fig. S6C); therefore, to examine the effects of METTL3 inhibition on antitumor immunity in vivo, we used our AT3 TNBC model and treated mice with STM2457.

After an appropriate latency to allow for an established and measurable tumor burden, we commenced treatment with vehicle, anti–PD-1 treatment alone, a short 5-day course of STM2457, and a long course of STM2457 with or without anti–PD-1 treatment. These data showed improved survival for the combination of STM2457 with anti–PD-1 therapy over and above that provided by single-agent therapy alone (Fig. 6A). While these experiments examined the capa­city for endogenous T cells to mount an effective antitumor response, we also wanted to study the effects of enhancing the activity of T cells specifically capable of recognizing tumor-intrinsic neoantigens. To assess this issue, we repeated the same experimental approach; however, this time we adoptively transferred OT-I cells to target ovalbumin peptide expressed via MHC-I on AT3 cells in vivo (Fig. 6B). Here again, we noted excellent disease control that was comparable for mice treated with either single-agent STM2457 or anti–PD-1. Notably, we found that when these treatments were combined, the disease control was substantially improved, including tumor eradication in a subset of mice (Fig. 6C). Interestingly, we found that the improved survival associated with METTL3 inhibition was apparent even with short-term METTL3 inhibition, particularly in combination with anti–PD-1 therapy (Supplementary Fig. S7A and S7B). Notably, the efficacy of combined STM2457 and anti–PD-1 therapy was not confined to solid cancers, as even aggressive hematologic malignancies such as the A20 lymphoma model showed significantly longer overall survival (Supplementary Fig. S7C and S7D). Together, these findings support the conclusion that while METTL3 inhibition alone can improve antitumor immunity, the cell-intrinsic interferon response also increases immune-checkpoint molecules, and consequently METTL3 inhibition alongside anti–PD-1 immunotherapy synergizes to improve disease control.

As the overall tumor burden is comprised of a mix of malignant clones with distinct genetic and epigenetic properties (25), we next wanted to understand if anti–PD-1 therapy and METTL3 inhibition target the same or distinct clones. To study this, we used our SPLINTR lineage-tracing methodology (26), which enables the temporal monitoring of malignant clonal dynamics throughout the treatment course. Of 1,190 AT3 clones that were transplanted into immunocompetent animals, only a fraction (12%, 140 clones) established disease in more than one immunocompetent mouse. Moreover, just 3% (41 clones) of these clones evaded immune surveillance by OT-I T cells (Fig. 6D). Together, these data illustrate that only a minor population of malignant clones can initiate and evade effective immune surveillance directed toward a cancer-specific neoantigen. Next, we specifically wanted to understand which of these immune evasive clones were targeted by the addition of anti–PD-1 and STM2457 as single agents or in combination. Here, we observed a subset of immune evasive clones that responded only to either anti–PD-1 (pink clones) or STM2457 (yellow clones, Fig. 6D). More notably, several clones that were insensitive to these single-agent therapies were effectively eliminated with the combination therapy (Fig. 6D, blue clones). Finally, we also identified malignant clones that remained insensitive to all the treatment arms (orange clones), explaining why, although the combination therapy resulted in the most impactful survival advantage, not all the mice treated were cured (Fig. 6C). Together, these data provide compelling evidence at clonal resolution that anti–PD-1 and METTL3 inhibitors target distinct subsets of malignant clones and, importantly, that the combination of these therapies can augment antitumor immunity to eliminate malignant clones insensitive to these agents alone.

We next wanted to understand what transcriptional features were associated with resistance to METTL3 inhibition, PD1 inhibition, or combination therapy. Single-cell RNA-seq (scRNA-seq) of the resistant cells showed a preponderance of cells from each treatment arm within certain Uniform Manifold Approximation and Projection (UMAP) clusters (Fig. 6E). This was confirmed through unbiased hypergeometric testing, which revealed anti–PD-1–resistant cells were most enriched in cluster 3, STM2457-resistant cells were most enriched in cluster 5, and cells resistant to combination therapy were most enriched in cluster 2 (Fig. 6F). While increased expression of pathways was unrewarding, gene ontology enrichment analysis of these major clusters showed those cells resistant to anti–PD-1 therapy downregulated genes associated with leukocyte migration and those resistant to STM2457 downregulated genes associated with antigen processing and presentation. Of note, the cells resistant to combination therapy showed a composite signature encompassing the major pathways associated with resistance to anti–PD-1 therapy and METTL3 inhibition (Fig. 6G). Together, these data provide key insights into the distinct molecular programs that underpin response and resistance to these therapies, which act to enhance antitumor immunity against different malignant clonal populations.

Immune-checkpoint inhibitors have revolutionized the treatment of many cancers; however, it is increasingly clear that to further improve clinical outcomes, innovative appro­aches to enhance host anticancer immune surveillance are required (16, 17). Here, we demonstrate that catalytic inhibition of METTL3 results in dsRNA formation and a potent cell-intrinsic interferon response that can stimulate antitumor immunity. This represents a complementary mechanism of action that is distinct from current immunotherapeutic agents that target immune checkpoints such as the PD-1/PD-L1 axis, CTLA4, or LAG3. The interferon response following METTL3 inhibition is associated with increased expression of components associated with antigen presentation as well as increased PD-L1. In this regard, our assessment of the consequence of METTL3 inhibition on newly synthesized transcripts illustrates that the catalytic inhibition of METTL3 has no discernible effect on transcription either at steady state or following induction with inflammatory cytokines. Instead, our findings highlight that one of the primary consequences of METTL3 inhibition is the stabilization of newly synthesized transcripts. Although an accurate and quantitative assessment of changes in m6A on individual transcripts is an evolving area of research, our results using NPS also demonstrated that transcripts associated with the MHC-I pathway exhibit some of the most dramatic changes in m6A levels after METTL3 inhibition. Taken together, these data suggest that as components of the MHC-I pathway are both interferon target genes and heavily m6A-modified, they are disproportionately affected by METTL3 inhibition, favoring increased antigen presentation and enhanced anticancer immunity.

Our study discloses a novel, more potent second-generation METTL3 inhibitor. Although the rapid metabolism of STM3006 precluded efficacy experiments in vivo, the compound enabled detailed mechanistic studies of METTL3 inhibition and provides the field with an orthogonal chemically distinct drug to investigate the biological function of this essential RNA methyltransferase. Using this approach, we illustrated that catalytic inhibition of METTL3 results in the extensive formation of dsRNA (13). The importance of dsRNA formation is exemplified by our unbiased global CRISPR screen, which emphasized the fact that the sensing of dsRNA and subsequent activation of the interferon signaling pathway are the critical components underpinning T cell–mediated antitumor immunity. Although our findings suggest the primary influence of METTL3 inhibition is via a tumor-intrinsic activation of interferon signaling rather than a direct potentiation of T-cell function, our findings do not exclude a potential role for other immune subsets, particularly in the in vivo setting. Additionally, the current body of work leaves open the possibility that METTL3 inhibition might induce immunogenicity through complementary mechanisms that are independent of the formation of dsRNA (27, 28).

Together, the results of our study provided the foundation to combine METTL3 inhibition with anti–PD-1 therapy, which showed improved survival outcomes in clinically relevant disease models of hematologic and solid cancers. In this regard, STC-15, a METTL3 inhibitor with comparable potency to STM3006 and improved oral bioavailability and metabolic stability, is currently being evaluated in a phase I clinical trial in solid cancers (NCT05584111). Of further clinical relevance, our results indicate that even a short course of METTL3 inhibition might be sufficient to engender superior disease control. This has important implications for therapeutic translation with respect to how these inhibitors might be administered in the clinic, as they suggest that even a short period of exposure may provide enduring clinical benefit, which is entirely consistent with the effects being driven through effective immune activation.

Furthermore, using SPLINTR (26), our lineage-tracing technology, we illustrated at single-clone resolution the molecular rationale for combining METTL3 inhibition with anti–PD-1 therapy. Specifically, the combination overcomes immune-evasive cancer clones insensitive to the single agents alone, supporting our hypothesis that anti–PD-1 therapy and METTL3 inhibition work through distinct but complementary mechanisms. Although our study provides the molecular and preclinical evidence to recommend catalytic inhibition of METTL3 as a complementary therapy alongside anti–PD-1 treatment, it also raises the prospect that targeting this RNA-modifying enzyme may potentiate other immunothe­rapies, including cellular therapies such as chimeric antigen receptor T cells.

RapidFire assay has been previously described (8).

Kinetic characterization of STM3006 was carried out on a Bia­core 8K instrument (GE HealthCare) using Series S NTA chips (GE HealthCare) at 20°C. Approximately 2 to 3 kRU (1,000 relative units) his-tagged METTL3/14 complex was immobilized to flow cell 2 of the sensor chip, preactivated with EDC/NHS according to the manufacturer's protocol in a running buffer of 20 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, 1 mmol/L TCEP, and 0.05% Tween 20. After protein capture, surfaces were washed with 60-second pulses of ethanolamine and 10 mmol/L EDTA, respectively. Serial dilutions of the compound in 20 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, 1 mmol/L TCEP, 0.05% Tween 20, and 2% DMSO were injected using single-cycle kinetics mode with 90 seconds association time at a flow rate of 80 μL/minute. Dissociation of the samples was monitored for 1,800 seconds. Data processing was performed using Bia­core Insight Evaluation software v3.0.12 (Biacore, GE HealthCare). Sensorgrams recorded on the reference flow cell 1 of the same channel without protein were subtracted from sensorgrams recorded on the METTL3/14 surface. For double-referencing, injections of the running buffer without STM3006 were run. Solvent correction was applied to all sample sensorgrams to correct for buffer mismatches. To determine kinetic parameters, data were fitted to a simple 1:1 Langmuir interaction model.

THP1 cell lines were grown in the presence of DMSO or different concentrations of STM3006, with each datapoint corresponding to an individual culture of 2 × 10⁷ cells. Sixteen hours after dosing, RNA was prepared using Qiazol (Qiagen) according to the manufacturer's instructions, and polyA⁺ mRNA was selected using the Oligotex mRNA Mini Kit (Qiagen) as described by the manufacturer. RNA concentrations, determined using the Qubit RNA BR Assay Kit on the Qubit 3 fluorometer (Thermo Fisher), were kept consistent between samples.

The entire polyA⁺ mRNA eluate and a 10 μL portion of the non-polyA⁺ RNA flow-through were concentrated using a Savant SpeedVac DNA-120 instrument for 2 hours at a high heat setting. Both of these RNA fractions were digested to their component nucleosides using a cocktail of benzonase, antarctic phosphatase, and phosphodiesterase I (all from Merck) according to the manufacturer's protocol. Once digestion was complete, an equal volume of isotopically labeled uridine (Cambridge isotopes) was added as an internal standard.

Nucleosides were resolved using reverse-phase liquid chromatography (Agilent 1290 Infinity II). For the mobile phase, eluent A consisted of water with 0.1% formic acid and eluent B consisted of acetonitrile with 0.1% formic acid. The flow of the mobile phase was held at 300 μL/minute. The solid phase consisted of a T3 HSS C18 100 mm × 2.1 × 1.8 ACQUITY UPLC column held at 15°C throughout the experiment. A nonlinear gradient of eluent B of 1% to 15% over 11 minutes wholly resolved the nucleosides. The eluent was sprayed into a 4500 triple quadrupole mass spectrometer and characterized using a multiple reaction monitoring approach. Injection volumes were kept to a constant 2 μL for both calibration curve and samples, and injections were made in technical triplicate in all cases. Injection amounts were validated by internal calibration with isotopically labeled uridine and quantification interpolated from external calibration of a range of nucleoside standards using Multiquant (Sciex) and Prism (GraphPad) software.

The selectivity profile of STM3006 was determined by testing the level of inhibition across a panel of RNA, DNA, and histone methyltransferases as previously described (8).

All incubation steps were performed on a plate shaker at room temperature (RT). (i) For plate saturation, a 96-well streptavidin-coated plate (Meso Scale Discovery) was first saturated for 1 hour with 5% Blocker A (Meso Scale Discovery) in buffer 1 (1× TBS, 0.05% Tween-20). The plate was washed twice with buffer 1 and once with buffer 2 (20 mmol/L Tris-HCl pH 7.5, 300 mmol/L NaCl, and 1 mmol/L EDTA). (ii) For biotinylated oligo-dT capture, the plate was incubated for 30 minutes with buffer 2 containing biotinylated oligo-dT (Promega) and then washed twice with buffer 2 and once with buffer 2 supplemented with RNasin ribonuclease inhibitors (Promega). (iii) For RNA capture, the plate was primed with buffer 2 supplemented with 1,000 U/mL of RNasin and then was incubated for 30 minutes with buffer 2 containing 600 ng of total RNA (preheated 5 minutes at 65°C). The plate was washed twice with buffer 2 and once with buffer 1, both supplemented with 2 U/mL of RNasin. (iv) For primary antibody capture, the plate was then incubated for 2 hours with 2 μg/mL of Rabbit anti-m6A antibody (Active Motif) in buffer 1 supplemented with 1% blocker A and 50 U/mL of RNasin. The plate was washed 3 times with buffer 1 supplemented with 2 U/mL of RNasin. (v) For secondary antibody capture, the plate was incubated for 1 hour with 1 μg/mL Goat anti-rabbit Sulfo-Tag (Meso Scale Discovery) in buffer 1 supplemented with 1% blocker A and 50 U/mL of RNasin. The plate was washed 3 times with buffer 1 supplemented with 2 U/mL of RNasin. (vi) For MSD signal detection, 1× MSD read buffer (Meso Scale Discovery) was then added and MSD signal intensity was immediately measured on a Sector 600 Meso Scale Reader.

Cloning, protein extraction, and purification, as well as crystallization and soaking, of STM3006 were performed as described (8).

Data were collected to 2.21 Å at the Swiss Light Source PXI beamline and processed using autoPROC (29). PDB 5L6D was used as the Phaser (30). Iterative cycles of refinement using REFMAC5 (31) and model building using Coot (v10.11.4; ref. 32; RRID: SCR_014222) were performed. STM3006 was fitted, and restraints were created using SMILES string and AFITT (33). Final structure statistics can be found in Supplementary Table S1. The structure was deposited with PDB access code 8BN8.

Murine B16 (cat. # CRL-6475, RRID:CVCL_0159) and B16-ovalbumin–expressing cells and the human ovarian cell line CaOV3 (cat. # HTB-75, RRID:CVCL_0201) were cultured in DMEM (Gibco) supplemented with 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin, and 2 mmol/L glutamax. Murine AT3 (cat. # SCC178, RRID:CVCL_VR89) and AT3-ovalbumin cells were cultured in DMEM (Gibco) supplemented with 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L glutamax, and 15 mmol/L HEPES and MEM Nonessential Amino Acids (Gibco). SKOV3 cells (cat. # HTB-77, RRID:CVCL_0532) were grown in McCoy's 5A (Gibco) supplemented with 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin, and 2 mmol/L glutamax. SKOV3 cells stably expressing NucLight Red tag (SKOV3-NLR) were generated by lentiviral induction of NLR-expressing plasmid (Sartorius). SKOV3-NLR cells were maintained as above, with the addition of 4 μg/mL of puromycin. Puromycin was removed prior to coculture. Murine A20 cells (cat. # TIB-208 RRID:CVCL_1940) were obtained from ATCC and cultured in RPMI 1640 supplemented with 10% non–heat-inactivated FBS + 0.05 mmol/L 2-mercapto­ethanol. All cell lines were incubated at 37°C and 5% CO₂ and were regularly tested and verified to be negative for Mycoplasma contamination by PCR analysis through in-house genotyping. B16-F10–constitutive Mettl3 knockdown was generated by retroviral induction of LMP blue fluorescent protein (BFP) plasmid and maintained as above. K562 (dsRNA-sensing genes (RIG-I, PKR, RNAseL, and IFIH) were knocked out together (4KD) or individually in K562 cells (cat. # CCL-243, RRID:CVCL_0004) through lentiviral transduction and maintained in RPMI (Gibco) supplemented with 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin, and 2 mmol/L glutamax. Flow cytometry was performed on the BD Fortessa X20, BD LSR II, or BD Symphony A3. Flow cytometry antibodies used are shown in Supplementary Table S2.

Cells were treated in triplicate with a series of STM3006 or STM2457 concentrations for 5 days. Media were then removed, and equal volumes of PBS and CellTiter-Glo reagent (Promega) were added to each well. Contents were mixed for 2 minutes on an orbital shaker to assist cell lysis. Plates were incubated for 10 minutes at RT, and luminescence was measured on the SpectraMax i3x microplate reader (Molecular Devices). IC₅₀ was calculated by nonlinear regression using GraphPad Prism 9.

CaOV3 and SKOV3 cells were treated in triplicate with a series of STM3006 or STM2457 concentrations. To measure apoptosis, Caspase-3/7 Green Dye (Sartorius) was added at a final concentration of 5 μmol/L. Images were taken every 4 hours at 10× magnification in the IncuCyte Live-Cell Analysis System for 60 hours (CaOV3) or 120 hours (SKOV3). Analyses were performed using the integrated software, and data were presented as green integrated intensity per area, normalized to cell confluence.

CaOV3 cells were treated with 500 nmol/L STM3006 or 0.3% DMSO in triplicate for 48 hours. For total RNA extraction, cells were homogenized in 700 μL TRIzol (Thermo Fisher Scientific), followed by chloroform addition and centrifugation. The aqueous phase was transferred to new tubes, and RNA was precipitated using equal volumes of 100% isopropanol followed by 2 washes with 80% ethanol. RNA pellets were air-dried and resuspended in 25 μL nuclease-free water. RNA concentration was determined by a Qubit RNA broad-range assay kit (Thermo Fisher Scientific, Q10211). QuantSeq 3′ mRNA-Seq Library preparation was performed by Lexogen GmbH according to the manufacturer's instructions. Sequencing was performed using the Illumina NextSeq single-end 75 bp High-Output V2 chemistry. Analysis of CaOV3 samples (STM3006 n = 3, DMSO n = 3) was performed using the nf-core/rnaseq bioinformatics pipeline (workflow container 3.0). Reads were mapped to the human genome GRCh38 using STAR (RRID: SCR_004463) aligner under default parameters (34). Differential gene expression analysis of treated versus nontreated samples was conducted using the R/Bioconductor package DESeq2 1.34.0 (35). Genes showing an absolute fold change greater than two at an adjusted P value for FDR less than 0.05 were selected as being differentially expressed (DE) in STM3006-treated cells. Functional annotation of DE genes was performed using gene ontology enrichment analyses as implemented in the R/Bioconductor package clusterProfiler 4.2.2 (ref. 36; RRID: SCR_016884).

AT3 cells were treated with DMSO, 20 μmol/L STM2457, or 2 μmol/L STM3006 in triplicate for 24 and 48 hours. Total RNA was then isolated using the Qiagen RNeasy Kit. RNA quantification, library preparation, and sequencing were performed as per the above descriptions. Bcl2fastq (Illumina) was used to perform sample demultiplexing and to convert BCL files generated from the sequencing instrument into FastQ files. Reads were aligned to the mouse genome (Mm10) using HiSAT2 (37), and reads were assigned to genes using htseq-count (38). Analysis was performed as described above.

CaOV3 cells were treated with STM2457 (at 1 μmol/L or 5 μmol/L) or STM3006 (at 0.05 μmol/L or 0.25 μmol/L with and without IFNγ stimulation at 0.25 ng/mL). After 30 hours of treatment, cells were harvested by trypsinization and lysed using 40 μL RIPA buffer supplemented by protease and phosphatase inhibitors. Total protein (25 μg) was loaded on a 4%–12% Bis-Tris gel. Resolved proteins were then transferred to a nitrocellulose membrane and blocked for 1 hour with 5% milk/TBST. Antibodies, as detailed in Supplementary Table S3, were diluted in 2.5% milk/TBST and incubated with the blocked membranes overnight at 4°C. Following 3 washes with TBST, the secondary antibody was incubated with the membrane for 1 hour at RT. After 3 additional washes with TBST, signal was developed using electroluminescence. Mettl3 knockdown in B16 cells and the knockout of dsRNA-sensing genes (RIG-I, PKR, RNaseL, and IFIH) in K562 cells either together or individually were confirmed using antibodies detailed in Supplementary Table S3. RIG-I and IFIH knockdown cells were treated with IFNγ at 10 ng/mL for 48 hours to detect protein.

The K562 cell line was transduced with lentivirus TRE_KRAB-dCas9-IRES-BFP (RRID:Addgene_85449) and pLVX_EF1a-TET3GT (TakaraBio, cat. #631359). Cells were induced with doxycycline at 100 μg/mL for 48 hours and sorted on the Fusion 5 cell sorter for BFP⁺ cells. Subsequently, the cells were transduced in single or combination for the dsRNA-sensing genes RIG-I [U6-sgRNA(BsmBI)-SFFV-eGFP, RRID: Addgene_169938], PKR [U6-sgRNA(BsmBI)-SFFV-tRFP, RRID: Addgene_169941], RNaseL [U6-sgRNA(BsmBI)-SFFV-RFP657, RRID: Addgene_169939], and IFIH [U6-sgRNA(BsmBI)-SFFV-PURO] as well as their equivalent nontargeting controls. Cells were induced with doxycycline at 100 ng/mL with or without IFNγ at 10 ng/mL and STM2457 at 10 μmol/L or 20 μmol/L or STM3006 at 1 or 2 μmol/L for 48 hours and analyzed by flow cytometry on the Fortessa X20.

CaOV3 cells were treated with STM2457 or STM3006 for 48 hours. Tumor-conditioned media were collected, spun down at 800 × g for 2 minutes to remove cell debris, and stored at −20°C. IFNβ levels were assessed in duplicates using the S-PLEX Human IFNβ Kit (Meso Scale, K151ADRS) following the manufacturer's instructions, starting from 25 μL of undiluted culture medium. CXCL10 levels were assessed in duplicates using Quantikine ELISA, Human CXCL10/IP-10 Immunoassay (R&D Systems, DIP100) according to the manufacturer's protocol, starting from 100 μL of undiluted culture medium.

Cells grown on 13-mm coverslips were fixed with 4% parafor­maldehyde for 10 minutes at RT, and then permeabilized with 0.5% Triton X for 10 minutes and stained with the dsRNA-specific antibody J2 (Exalpha Biologicals and Nordic-M) using the MOM (mouse on mouse; Vector Labs) staining kit, according to the supplier's recommendations, with streptavidin-AF546 (Thermo Fisher). Stained cells were mounted to glass slides with Vecta­shield containing DAPI (Vector Labs). J2 antibody staining was analyzed via confocal imaging using Nikon C1 under identical settings. For analysis, the J2 signal was quantified using ImageJ (RRID: SCR_003070).

Total RNA was isolated from AT3 cells and CaOV3 cells as described above. To generate cDNA, total RNA was reverse-transcribed with the SuperScript VILO cDNA synthesis kit (Thermo Fisher) according to the manufacturer's instructions. For AT3, quantitative real-time PCR was performed on the Applied Biosystems StepOnePlus using Fast SYBR green reagents (Thermo Fisher). mRNA levels were normalized to Gapdh. For CaOV3, cells were treated for 24 hours. Quantitative real-time PCR was performed on the Applied Biosystems QuantStudio 5 using TaqMan Fast Advanced Master Mix and TaqMan gene expression probes, using the manufacturer's protocol (Thermo Fisher). mRNA levels were normalized to the housekeeping gene PPIA. All samples were assayed in triplicate with relative quantification of target gene expression performed using the comparative cycle threshold (CT) method. Primer sequences and TaqMan probes are listed in Supplementary Tables S4 and S5, respectively.

NPS was undertaken on AT3 cells treated with DMSO, 20 μmol/L STM2457, or 2 μmol/L STM3006 for 48 hours. All samples were processed in duplicate. Total RNA was isolated using TRIzol LS reagent (Invitrogen) according to the manufacturer's instructions and ethanol precipitation. RNA concentration was quantified with a Qubit Fluorometer (Thermo Fisher Scientific). Extracted total RNA was processed to obtain polyA-tailed transcripts with Dynabeads (Invitrogen) according to the manufacturer's instructions. A target amount of 200 ng was used to create a readable transcript library for a single flowcell (type FLO-MIN106) using a protocol based on Oxford Nanopore Technology's DirectRNA SQK-RNA002 kit protocol. Each sample was sequenced on a single flowcell with version R9.4.1 chemistry. A GridION instrument controlled by MINKNOW software (v20.10.6) was loaded with four flowcells, and four channel libraries were run separately and simultaneously. Readfish (v3.0.0) selective sequencing (39) was used during the GridION run to eject abundant and irrelevant transcripts. Briefly, the accessions of 81 mouse transcripts comprising mitochondrial transcripts and cytoplasmic ribsomal protein–encoding transcripts were used to generate an exclude list. Reads were basecalled in real time using Guppy (v4.2.3). FAST5 files were recorded for later processing by Nanopolish (v0.11.2; ref. 40). Basecalled reads and Fast5 signal files were processed to prepare for Nanocompore (v1.0.3) comparison analysis according to the author's instructions (20). Briefly, fastq files were aligned to a mouse transcriptome version M20 with minimap2 (v2.2.0; ref. 41) to obtain bam alignment files, and then Nanopolish's eventalign module was used to estimate kmer pore transition (“dwell”) times from the alignment and the Fast5 signal files. The Nanopolish compress tool was used to average Nanopolish output by kmer. Nanocompore was used to discern significant differences in signal current intensity and dwell time between the control and treatment replicate pairs. Using the GMM-logit option, Gaussian mixture modeling was used to discern the presence of two different nucleotide modification states, and logit regression analysis was used to calculate the difference in occupancy between the samples.

To visualize the differences in m6A occupancy between samples, the transcript isoforms were condensed to the gene level, and only biophysical alterations occurring at DRACH motifs (± 2 nucleotide tolerance) were included. The most significant DRACH alterations per transcript were used to plot the coordinate space of the modified transcripts, and only transcripts with a modified site scoring <1e−50 were forwarded for functional annotation and enrichment.

AT3 cells were treated with DMSO, 20 μmol/L STM2457, or 2 μmol/L STM3006 for 30 minutes (± 2 ng/mL murine IFNγ for up to 10 minutes) before the addition of 100 μmol/L 4-thiouridine (4sU, Sigma) as per the schematic in Fig. 3D. Samples were collected at time 0 and at 30 and 60 minutes after addition of 4sU for analysis of nascent transcription. Six hours after addition of 4sU, media were washed off and a uridine chase was commenced for a further 6 hours. Cells were harvested at the end of each 6-hour incubation period. All samples were generated and collected in triplicate. S2 cells were spiked in at 10% of total cells, and RNA was extracted using the Qiagen RNeasy Mini Prep Kit. Total RNA was subjected to carboxyamidomethylation using freshly prepared 10 mmol/L iodoacetamide (Sigma) for 15 minutes, and RNA was repurified by the Qiagen RNeasy MinElute Cleanup Kit. Libra­ries were prepared with the QuantaSeq 3′ mRNA Library Prep kit (Lexogen), and samples were sequenced on the NovaSeq S1 with 75 bp single-end reads. Reads were quality checked using FastQC v0.11.6 (RRID: SCR_014583) and then trimmed with Trimmomatic v0.39 (ref. 42; RRID: SCR_011848), and the resulting trimmed reads were also quality checked with FastQC. The SLAM-DUNK pipeline v0.2.4 (43) was used with default parameters with the exception of -n 100 -m -mv 0.2 to align reads to a joint Mm10-Dm6 genome and bed file of 3′UTR sites as extracted from a gtf file of the respective genomes in Python v3.5.2. MultiQC v1.8 was used to gene­rate a report of all SLAM-DUNK QC output and trimmed reads FastQC reports. The edgeR package (44) in R v4.2.0 was used to perform Trimmed Mean of M-values (TMM) normalization of reads and differential expression analysis. The annotatr (45) R package was used to generate a 3′UTR Mm10 annotation. Plots were gene­rated using base R, ggplot2, and ggpubr. Gene set enrichment was conducted using the enrichR package (46) for enrichment in the RNA-seq Automatic Gene Expression Omnibus (GEO) Signatures Mouse Up database.

Ovalbumin-expressing cell lines were generated using MSCV-ovalbumin-mCherry or MSCV-ovalbumin-GFP and subsequently transduced with FUCas9Cherry (Addgene 70182, deposited by M. Herold, Walter and Eliza Hall Institute of Medical Research, Olivia Newton-John Cancer Research Institute) or pHRSIN-PSFFV-Cas9-PPGK-Blasticidin (47). sgRNAs for b2M, STAT1, and PRKRA were additionally introduced using sgRNA.SFFV.RFP657 (Addgene 169939, deposited by B. Ebert, Dana-Farber Cancer Institute) and enriched through flow-based sorting. Cells were treated with DMSO, 10 and 20 μmol/L of STM2457, or 1 μmol/L and 2 μmol/L STM3006 up to 48 hours prior to coculture. OT-I T cells were generated from OT-I mice (gifted from the Trapani Lab, Peter MacCallum Cancer Centre). In brief, spleens were harvested and cells were maintained in T-cell media (RPMI 1640, 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L glutamax, 10 mmol/L HEPES and 1× MEM Nonessential Amino Acids, 1 mmol/L Na Pyruvate, and 2-mercaptoethanol) supplemented with 100 U/mL IL2 (Roche) and 20 ng/mL SIINFEKL (ovalbumin peptide; Sigma-Aldrich) for 48 hours and then expanded in IL2 alone for a further 48 hours. After red cell lysis, tumor cells and T cells were cocultured at various effector:target ratios for a period of 18 to 48 hours. Cell death was assayed by flow-cytometric propidium iodide (BD Pharmingen) analysis. All analyses were run with three technical replicates.

B16 cells were transduced with a lentiviral pHRSIN-PSFFV-Cas9-PPGK-Blasticidin encoding Cas9 and selected with blasticidin 10 μg/mL. For the coculture screen, 4.7 × 10⁸ cells were infected with a whole-genome pooled lentiviral sgRNA Brie library (Addgene 73633, deposited by D. Root and J. Doench, Broad Institute) at a multipli­city of infection of 0.3 and subsequently selected with 1 μg/mL of puromycin for 72 hours commencing 48 hours after transduction. Cells were then treated with DMSO or STM2457 at 5 and 20 μmol/L for 48 hours before being cocultured with OT-I cells for ∼12 hours (prepared as per cocultures above) at effector:target ratios of 1:1, 1:5, and 1:20. T cells were washed off, and cells were rested for 24 hours before being exposed to drug and coculture for a second round. At the end of the screen, genomic DNA was extracted using the NucleoSpin BloodXL (Machery-Nagel) kit. sgRNA sequences were amplified using a single round of PCR with adapters added for Illumina sequencing. Samples were sequenced with single-end 75-bp reads on an Illumina NextSeq. The sequence reads were trimmed to remove the constant portion of the sgRNA sequences with cutadapt (48) and then mapped to the reference sgRNA library with bowtie2 (49). After filtering to remove multialigning reads, the read counts were computed for each sgRNA. The RSA algorithm or MAGECK (50) was used to rank the genes for which targeting sgRNAs were significantly enriched in the sorted populations compared with the control unsorted populations grown in parallel.

SKOV3-NLR cells were seeded at 7,500 cells per well into 96-well plates in assay media of RPMI 1640 + 10% Human AB serum, 20 mmol/L HEPES, and 4 mmol/L L-glutamine + 100 U/mL penicillin/streptomycin. Prior to coculture, SKOV3-NLR cells were treated with 10 ng/mL of IFNγ (Biotechne, #285-IF-100). PBMCs were isolated from healthy donor buffy coats by density gradient separation, prepared with a suboptimal dose of αCD3 (BioLegend HIT3a Clone, #300332) for a final concentration of 5 μg/mL, and seeded at 50,000 cells per well. STM2457, STM3006, and vehicle control treatments were prepared in assay media and added at the point of coculture. Caspase-3/7 Green Reagent (Essen Bioscience, #4440) was added at a final concentration of 5 μmol/L. Assay plates were transferred into an IncuCyte S3 System, and images were collected every 2 hours for 72 hours. Following experimental completion, culture supernatants were harvested and frozen for cytokine analysis. Image analysis was performed within the IncuCyte S3 system. Target cell growth was detected from SKOV-NLR images in the red fluorescence channel, quantified as red object count per image, and normalized as a percentage to T = 0 hours. Target cell apoptosis was determined by detecting caspase-3/7 signal in the green fluorescence channel and colocalizing it to SKOV-NLR cells detected within the red fluorescence channel.

Supernatant samples from coculture experiments were frozen at −80°C at the time of sampling and thawed at RT prior to analysis by Luminex Multiplex. Bio-Plex Pro Human Cytokine IL1β (Bio-Rad, #171B5001M), IFNγ (Bio-Rad, #171B5019M), and TNFα (Bio-Rad, #171B5026M) analyte bead sets were prepared in Bio-Plex assay buffer (Bio-Plex Pro Reagent Kit III, #171304090M) and washed 2 times using wash buffer (Bio-Plex Pro Reagent Kit III, #171304090M) on the Bio-Plex Pro plate washer. Cell supernatant samples were incubated with analyte beads for 30 minutes at RT on a shaker at 850 rpm. Following this, 3 washes were performed and a detection antibody was applied. Samples were incubated on the shaker at 850 rpm for 30 minutes at RT. A further 3 washes were performed, and streptavidin-PE (Bio-Plex Pro Reagent Kit III, #171304090M) was added for 10 minutes at RT. A final 3 washes were performed, and beads were resuspended in assay buffer. Data were acquired on the Bio-Plex 200 System, which had been calibrated with a Bio-Rad Calibration Kit (Bio-Rad, #171203060).

Human, mouse, and rat microsomes were obtained from BioIVT. Microsomes were removed from storage at −80°C and thawed at 37°C. Microsomes were diluted in assay buffer (potassium phosphate solutions 1 and 2 were prepared and combined to form a pH 7.42 solution at 37°C). Solution 1: 13.93 g potassium phosphate dibasic anhydrous (K2HPO4, 80 mmol/L) dissolved in 1 L deionized water. Solution 2: 2.72 g potassium phosphate monobasic anhydrous (KH2PO4, 20 mmol/L) dissolved in 1 L deionized water. pH 7.4 with 2 mmol/L magnesium chloride, to achieve a final protein concentration of 0.5 mg/mL and 1 mmol/L NADPH. The microsomal incubation plate was transferred to a heater shaker at 300 rpm, and the solution was prewarmed to 37°C for 10 minutes. A no-cofactor control at 0 and 45 minutes and one replicate of each batch of test compound (final incubation concentration of 1 μmol/L) were included. Microsomes were incubated at 37°C on a shaker set at 300 rpm throughout the assay. At each time point (0, 5, 15, 30, 45 minutes), 50 μL of sample was removed from the 96-well plate and added to 200 μL of quench solution (acetonitrile containing 0.1% formic acid and sulfisoxazole/tolbutamide/imipramine 200 nmol/L). Samples were diluted 1:1 with water using the Janus Robot and analyzed by LC-MS/MS.

Three catheterized male Sprague Dawley rats were given STM3006 by intravenous route at a target dose of 1 mg/kg with a nominal formulation concentration at 0.5 mg/mL in NMP/Solutol HS15/phosphate buffer pH 7.4 50 mmol/L (10%/5%/85%;v/w/v). Drug formulation was administered by intravenous route at the tail vein using a stainless steel needle fitted on a plastic syringe. Blood (200 μL) was collected via catheter using a serial sampling design at 0.083, 0.25, 0.5, 1, 2, 3, 5, 7, and 24 hours after administration, and transferred to 0.5 mL EDTA-coated tubes. Collected blood was centrifuged at 2,500 rpm for 10 minutes at ca. +4°C. Plasma samples (20 μL each) were collected in duplicate in a 96-well plate and then stored at ca. −20°C until analysis. Plasma samples (20 μL) were mixed with 120 μL of precipitant solution (acetonitrile + 0.02 μmol/L trifluperidol as internal standard). Half of the supernatant (70 μL) was then diluted 1/1 into water, and 2.5 μL was analyzed by LC-MS/MS.

All mouse experiments were performed according to animal ethics approved by our institutional review boards at Peter MacCallum Cancer Institute or Crown Bioscience UK Ltd testing facilities and in line with associated ethical regulations. Six- to 12-week-old BL6 or BALB/cN female mice were injected subcutaneously with AT3-ovalbumin–expressing or A20 cells, respectively. In cases in which tumor cells were barcoded, SPLINTR_V2_EF1a_mCh (26) was introduced in vitro. After 48 hours, barcode-positive cells were sorted and allowed to expand ex vivo for ∼12 to 14 cell doublings, allowing each barcode to be represented with 50 to 100× coverage per mouse. Each mouse received 1 × 10⁶ barcoded cells. The remaining cells from the culture were cryopreserved for baseline (T0) scRNA-seq and 5 × 10⁵ cells were lysed in 40 μL of Viagen lysis buffer containing 0.5 mg/mL proteinase K (Invitrogen) according to the manufacturer's guidelines for population-based DNA barcoding. Tumor growth was monitored through serial measurement. Once tumors reached an average of 100 to 150 mm³, drug treatment was commenced. For the AT3-ovalbumin model, STM2457 “short” comprised 5 days of treatment at a dose of 100 mg/kg i.p., whereas “long” treatment comprised 5 days of STM2457 at 100 mg/kg followed by ongoing treatment at 50 mg/kg for up to 6 weeks. Mouse anti–PD-1 (Bio X Cell) treatment comprised a total of five intraperitoneal injections given at a frequency of two injections/week at 200 μg per dose. In experiments in which adoptive transfer was used, OT-I T cells were prepared ex vivo as described above. OT-I cells (2 × 10⁶) were injected intravenously. All mice receiving AT3 cells received sublethal irradiation of 2.5 Gy after the initial “short” course of treatment. Tumor was harvested once the endpoint of 1,200 mm³ was reached or earlier if humanely required and dissociated with 1 mg/mL Type IV collagenase, 1 mg/mL Dnase in DMEM with 100 IU/mL penicillin, and 100 μg/mL streptomycin. Tumor cells were viably frozen in Viagen again as per the manufacturer's guidelines. Peripheral blood was collected concurrently in EDTA-coated tubes. A complete blood count was run on the Sysmex XP-300. All flow-cytometric analysis was performed on the BD LSRFortessa, and resulting data were analyzed using FlowJo (v10; RRID: SCR_008520). For barcode sequencing, 5 × 10⁵ cells were lysed in 40 μL of Viagen lysis buffer containing 0.5 mg/mL proteinase K (Invitrogen) according to the manufacturer's guidelines. For the A20 model, STM2457 was administered orally at a 100 mg/kg daily dose for 29 consecutive days. Mouse anti–PD-1 (Crown Bioscience) treatment was given by intraperitoneal injection twice a week at a dose of 10 mg/kg for the duration of STM2457 treatment.

scRNA-seq analysis of all samples was conducted on the 10X Chromium system using the 10X Genomics NextGEM 3′ v3 ­Single-Cell Gene Expression Solution (10X Genomics). Cryopreserved samples were rapidly thawed at 37°C, and mCherry-positive viable cells were bulk-sorted into Eppendorf tubes. The cells were then washed twice with PBS + 0.04% BSA. Cell number was determined using an average of two cell counts on the Countess II automated cell counter (Thermo Fisher Scientific). Cells were then centrifuged and resuspended in an appropriate volume of PBS + 0.04% BSA + 0.4 U/μL RNase Inhibitor (Roche) so that the final cell concentration was suitable for superloading onto the 10X Chromium Single-Cell Chip (>1,500 cells per μL). The final cell suspension was put through a cell strainer, and a final cell count was conducted for accuracy. A total of 40,000 to 45,000 cells from the baseline time point were superloaded onto a single-cell lane of the 10X Chromium Chip to maximize clonal recovery. Different samples from the disease time points were individually labeled using MULTI-seq Lipid Modified Oligos (LMO) as previously described (51), pooled together and superloaded as above onto two lanes of the 10X Chromium Single-Cell Chip. Details of the LMO sequences and corresponding sample pools are provided in Supplementary Table S6. Following droplet emulsion and reverse transcription, cDNA amplification and library preparation were performed according to the manufacturer's protocol using the Chromium Single-Cell 3′ Library and Gel Bead Kit v3 (10X Genomics). Where necessary, an additional primer specific to the oligo sequenced conjugated to the MULTI-seq LMOs was added during the reverse transcription stage. Next, cDNA was amplified and the fraction containing the amplified LMOs was separated from the full-length mRNA from each cell using SPRI beads (10X Genomics). These two fractions were then processed as separate cDNA libraries. The quality of final libraries was checked using the 2100 Bioanalyser (Agilent) and stored at −20°C until they were sequenced. scRNA-seq libraries were pooled at equimolar ratios and sequenced on the Illumina NovaSeq 6000 system (paired-end 150-bp reads, ∼30,000–50,000 reads per cell). MULTI-seq LMO libraries were sequenced on the Illumina NextSeq or the Illumina MiSeq (paired-end 75-bp reads, 5,000 reads per cell) system.

Count matrices were generated from demultiplexed scRNA-seq fastq files using the 10X Genomics Cell Ranger (v.7) count pipeline against the mm10/GRCm38 genome (v.3.0.0). scRNA-seq quality control was performed using Seurat v.4 in R (52, 53). Low-quality cells were removed by filtering out cells with fewer than 200 genes or 4,000 unique molecular identifiers. Cells with greater than 10% mitochondrial RNA content were also removed. MULTI-seq LMO libraries were demultiplexed using the CITE-seq-Count v.1.4.3 and HTODemux methods in Seurat v.4 using the default settings (54).

Where available, 10X scRNA-seq droplets containing doublet cells were identified based on LMO oligo combinations with the HTODemux function in Seurat. Normalization of scRNA-seq datasets was performed using the SCTransform method (55) with percentage mitochondrial reads as well as S phase cell-cycle score and G₂–M cell-cycle phase score variables included in the regression model as sources of technical variation to remove. Dimensional reduction, k-nearest neighbor graph construction, and clustering were performed using Seurat v.4. In general, the default settings were used unless specified otherwise. The first 30 principal components were used to compute a nonlinear dimensional reduction using the UMAP method (56). Louvain clustering was performed for a resolution of 0.5. Marker gene identification and differential gene expression analysis for clusters and groups of cells were performed using MAST (57). Hypergeometric enrichment testing per cluster for each treatment group was performed in R using the phyper function.

GraphPad Prism version 9 (GraphPad Software Inc., RRID: SCR_002798) was used to generate figures and perform statistical analyses. Data are reported as mean ± SD, SEM, or independent replicates shown as individual datapoints, as indicated in the figure legends. A P value of < 0.05 was considered statistically significant.

The STM3006 structure has been deposited as PDB 8BN8.

All sequencing data have been deposited in the GEO repository (GSE217923; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE217923).

A.A. Guirguis reports grants from mRNA Victoria during the conduct of the study. Y. Ofir-Rosenfeld reports personal fees and other support from Storm Therapeutics Ltd during the conduct of the study, as well as a patent for WO 2022/254216 A1 pending. Y.-C. Chan reports grants from the National Health and Medical Research Council (NHMRC) during the conduct of the study, as well as grants from NHMRC outside the submitted work. B. Andrews reports other support from Storm Therapeutics Ltd during the conduct of the study, as well as other support from Storm Therapeutics Ltd outside the submitted work. L. Vasiliauskaite reports personal fees and other support from Storm Therapeutics Ltd during the conduct of the study, as well as a patent for WO 2022/254216 A1 pending. A. Azevedo reports personal fees and other support from Storm Therapeutics Ltd during the conduct of the study. J. Obacz reports personal fees and other support from Storm Therapeutics Ltd during the conduct of the study. M. Carkill reports other support from Storm Therapeutics Ltd and Charles River Laboratories during the conduct of the study, as well as other support from Charles River Laboratories outside the submitted work. M.R. Albertella reports other support from Storm Therapeutics Ltd outside the submitted work, as well as a patent for WO2022/254216 issued to Storm Therapeutics Ltd. O. Rausch reports personal fees and other support from Storm Therapeutics Ltd during the conduct of the study, as well as a patent for WO 2022/254216 A1 pending. M.A. Dawson reports grants and personal fees from Storm Therapeutics Ltd during the conduct of the study, as well as personal fees from GSK and Cambridge Epigenetix/Biomodal, and grants from Pfizer outside the submitted work. No disclosures were reported by the other authors.

A.A. Guirguis: Conceptualization, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing. Y. Ofir-Rosenfeld: Conceptualization, validation, investigation, methodology, writing–original draft, writing–review and editing. K. Knezevic: Conceptualization, validation, investigation, methodology, writing–original draft, writing–review and editing. W. Blackaby: Formal analysis, validation, investigation, writing–review and editing. D. Hardick: Formal analysis, investigation, writing–review and editing. Y.-C. Chan: Formal analysis, investigation, writing–review and editing. A. Motazedian: Formal analysis, investigation, writing–review and editing. A. Gillespie: Formal analysis, investigation, writing–review and editing. D. Vassiliadis: Formal analysis, investigation, writing–review and editing. E.Y.N. Lam: Formal analysis, investigation, writing–review and editing. K. Tran: Formal analysis, investigation, writing–review and editing. B. Andrews: Formal analysis, investigation, writing–review and editing. M.E. Harbour: Formal analysis, investigation, writing–review and editing. L. Vasiliauskaite: Formal analysis, investigation. C.J. Saunders: Investigation. G. Tsagkogeorga: Formal analysis, investigation. A. Azevedo: Formal analysis, investigation. J. Obacz: Investigation. E.S. Pilka: Investigation, writing–review and editing. M. Carkill: Investigation, writing–review and editing. L. MacPherson: Investigation. E.N. Wainwright: Investigation. B. Liddicoat: Investigation, methodology, writing–review and editing. B.J. Blyth: Investigation, methodology, writing–review and editing. M.R. Albertella: Conceptualization, investigation, methodology, writing–review and editing. O. Rausch: Conceptualization, methodology, writing–original draft, writing–review and editing. M.A. Dawson: Conceptualization, funding acquisition, methodology, writing–original draft, writing–review and editing.

We thank the Flow Cytometry core facility and the Molecular Genomics Core at the Peter MacCallum Cancer Centre as well as S. Jackson, J. Schreuders, R. Walker, K. Warren, K. Jhuang, T. Gulati, K. Simpson, and T. Semple for their technical contributions to this project. We thank Elliott Bayle and Beth Thomas from Storm Therapeutics Ltd for editing the medicinal and structural chemistry sections of the manuscript. We thank the following funders for fellowship, scholarship, and grant support: Cancer Council Victoria, the Sir Edward Dunlop Research Fellowship, NHMRC Investigator Grant 1196749, the mRNA Victoria Research Acceleration Fund, Howard Hughes Medical Institute International Research Scholarship 55008729 (M.A. Dawson), a VCA Mid-Career Research Fellowship (E.Y.N. Lam), and The Leukemia & Lymphoma Society, fellowship #3411-22 (D. Vassiliadis). This research was also funded in part by the NHMRC grants 1085015/1106444 (M.A. Dawson) and 1128984 (M.A. Dawson).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.