Our model for the action of WINi in MLLr leukemia cells is based on analysis of two early-generation compounds (Aho et al., 2019a): C3 (K_d = 1.3 nM) and C6 (K_d = 100 pM). Subsequently (Tian et al., 2020), we discovered more potent molecules such as C16 (K_d < 20 pM) that have not been extensively profiled. To determine if improvements in the potency of WINi have resulted in divergent activities, we first compared C6 with C16 (Figure 1A). Both molecules bind the WIN site of WDR5 (Figure 1B, Figure 1—figure supplement 1A), but differ in affinity due to a bicyclic dihydroisoquinolinone core that locks C16 into a favorable binding conformation (Figure 1C). Consistent with its higher affinity, C16 is ~20 times more potent than C6 in inhibiting MV4;11 (MLL–AF4) and MOLM13 (MLL–AF9) leukemia lines (Figure 1—figure supplement 1B and C): in MV4;11 cells, for example, the GI₅₀ for C6 is 1 μM compared to 46 nM for C16. These differences in potency are reflected at the level of RPG suppression. Using a target engagement assay (Florian et al., 2022) that measures transcript levels from seven RPGs—five (RPS14, RPS24, RPL26, RPL32, and RPL35) that are always bound by WDR5 and two (RPS11 and RPS14) that are never bound—we observe that maximal suppression of RPG transcripts occurs at ~2 μM for C6 and ~100 nM for C16 in MV4;11 (Figure 1D) and MOLM13 (Figure 1—figure supplement 1D) cells. To functionally compare these two inhibitors of different potencies, we used these RPG-normalized doses in all our subsequent studies.

We performed RNA-sequencing (RNA-seq) on MV4;11 cells treated for 48 hr with DMSO, C6, or C16 (Figure 1—figure supplement 1E, Figure 1—source data 1). Spike-in controls were not included. Both compounds elicit thousands of gene expression changes (Figure 1E), a majority of which are less than twofold in magnitude (Figure 1—figure supplement 1F). We had previously performed RNA-seq on MV4;11 cells treated with 2 μM C6 and observed just ~75 induced and ~460 reduced transcripts (Aho et al., 2019a). In this earlier work, however, increased variance among replicates made it difficult for as many small gene expression changes to reach statistical significance as in the current study (Figure 1—figure supplement 1G). Comparing the new RNA-seq datasets, we observe similar effects of C6 and C16 on the MV4;11 cell transcriptome (Figure 1F), with more than 80% of the transcripts altered by C16 altered in the same direction by C6 (Figure 1G). In both cases, suppressed genes are enriched in those bound by WDR5 in MV4;11 cells (Figure 1H), although a majority of transcriptional changes occur at loci bereft of detectable WDR5 binding (Figure 1F). For both compounds, expression of genes connected to protein synthesis, the cell cycle, DNA replication, mTORC signaling, and MYC are reduced, while expression of those connected to chemokine signaling, apoptosis, and p53 is induced (Figure 1I, Figure 1—source data 2). Indeed, ~90 ‘consensus’ p53 target genes (Fischer, 2017) are induced by C6/C16 (Figure 1—figure supplement 2A and B), including the tumor suppressor ZMAT3 (Figure 1F; Bieging-Rolett et al., 2020). Finally, we note that for both compounds the transcriptional effects on RPG expression are almost entirely confined to those RPGs bound by WDR5 (Figure 1J). The conspicuous exception to this trend is RPL22L1—a paralog of RPL22—mRNA levels which are strongly reduced by C6/C16 (Figure 1F and J). This feature of the response is not confined to MLLr cells as RPL22L1 expression is also decreased by WINi in sensitive rhabdoid tumor cell lines (Florian et al., 2022), but not in the insensitive K562 leukemia line (Bryan et al., 2020; Figure 1—figure supplement 2C). Despite not being a direct WDR5 target gene, therefore, RPL22L1 expression is recurringly suppressed by WINi in responsive cancer cell lines.

Together, these data reveal that improvements in the potency of WINi have not resulted in substantive changes in their impact on the transcriptome of MLLr cells and reinforce the concept that inhibition of select RPG expression—and induction of a p53-related transcriptional program—defines the response of the transcriptome to WIN site blockade in this setting.

We previously showed that treatment of MV4;11 cells with C6 results in a time-dependent decrease in translation as measured by bulk labeling of nascent polypeptide chains with O-propargyl-puromycin (OPP) (Aho et al., 2019a). We confirmed this finding with C6 and extended it to C16, showing that while there is no significant effect on protein synthesis capacity after 24 hr of WINi treatment, a progressive decline begins at 48 hr, reaching an ~40% reduction at the 96 hr treatment point (Figure 2—figure supplement 1A and B). To determine more precisely the effects of WIN site inhibition on translational processes, we performed ribosome profiling (Ribo-seq; McGlincy and Ingolia, 2017) in parallel with the RNA-seq analyses described above. Spike-in controls were not included. By sequencing ribosome-protected fragments (RPFs) in 48-hr-treated and control cells, and normalizing to transcript levels from RNA-seq, we calculated the translation efficiency (TE) of each transcript and used this to determine how C6/C16 influence translation, independent of effects on mRNA abundance.

In these experiments, RPFs have the characteristic length of ribosome-protected mRNA fragments (28–32 nucleotides; Figure 2—figure supplement 2A), are enriched in coding sequences (Figure 2—figure supplement 2B), and map to the expected reading frame (Figure 2—figure supplement 2C), all of which indicate successful profiling. In contrast to RNA-seq, where we see equal numbers of transcript increases and decreases in response to C6/C16, the overwhelming effect of these compounds on translation efficiency is inhibitory (Figure 2A). Of the ~10,000 transcripts profiled, between ~4,500 (C16) and ~5,900 (C6) transcripts show decreased TE compared to less than 10 transcripts with increased translation (Figure 2B, Figure 2—source data 1). As we observed in the RNA-seq, changes in TE are generally less than twofold (Figure 2A) and there is extensive overlap between the two inhibitors, with ~90% of the transcripts decreased in TE by C16 also decreased by C6 (Figure 2C). In general, C6/C16 reduce translation of mRNAs in a manner independent of basal translation efficiencies (Figure 2D), although if we bin transcripts according to basal TE we observe that the number of highly translated transcripts (fourth quartile) impacted by C6/C16 is greater than for those transcripts with lower basal TE (Figure 2E). Within these quartiles, however, the magnitude of reduction in TE is equivalent (Figure 2—figure supplement 2D). Interestingly, mRNAs carrying better matches to the 5'TOP motif—which links translation to mTORC1 signaling (Philippe et al., 2020)—show less decrease in TE compared to those with poorer matches (Figure 2—figure supplement 2E), suggesting that mTORC1-regulated mRNAs may be spared from the full translational effects of WIN site inhibition.

Interrogating specific changes in TE induced by C6/C16, we see that the biological categories of transcripts with decreased TE echo many of those observed with decreased mRNA levels, but include more genes in each category. For example, manually curating each list for transcripts encoding the ~60 validated substrates of the protein arginine methyltransferase PRMT5 (Radzisheuskaya et al., 2019) reveals that 42 are translationally suppressed by C6/C16 compared to just 10 that are suppressed at the mRNA level (Figure 2—figure supplement 2F). Probing for Hallmark categories in the Human Molecular Signatures Database (MSigDB; Liberzon et al., 2015) uncovers the extent of this phenomenon (Figure 2—source data 2), with categories linked to MYC, E2F, mTORC1 signaling, and the G2M checkpoint (Figure 2F) all represented by more genes in the ribosome profiling than in the RNA-seq experiments. Within these categories, a majority of genes suppressed by C6/C16 at the mRNA level are also further suppressed translationally (Figure 2—figure supplement 2G). This is not a general trend in the response, however, as fewer than half of the total transcripts with reduced mRNA abundance experience this additional translational inhibition (Figure 2—figure supplement 2H). The finding that TE changes induced by C6/C16 extend the biological characteristics of changes in mRNA abundance may indicate a role for impaired translation in contributing to at least some of the mRNA level differences triggered by WINi. Indeed, comparing transcripts with decreased TE but no mRNA decrease with transcripts with decreased TE and mRNA levels reveals the latter are enriched in so-called ‘optimal codons’ (Wu et al., 2019) that normally promote mRNA stability but are linked to mRNA instability when translation is inhibited (Figure 2—figure supplement 2I).

Implicit in the previous discussion, a majority of the translational decreases triggered by C6/C16 occur at transcripts for which there are no significant changes in mRNA abundance (Figure 2—figure supplement 2H). Expectedly, genes with decreased TE but no mRNA level changes are enriched in several of the major Hallmark categories described above (Figure 2—source data 3). But we also observe enrichment in genes connected to the proteasome, spliceosome, mRNA surveillance, and translation (Figure 2G). The latter category includes subunits of the mitochondrial ribosome (Figure 2—figure supplement 3A), translation and ribosome biogenesis factors, and an expanded cohort of transcripts from RPGs (Figure 2H). Indeed, compared to mRNA levels, where C6/C16-induced changes are confined (with the exception of RPL22L1) to a decrease in expression of WDR5-bound RPGs, translational effects are not, and some of the most pronounced TE changes occur at non-WDR5 ribosomal protein target genes. Thus, beyond what we have been able to infer from previous studies, WINi causes a widespread reduction in the ability of MLLr cells to efficiently translate mRNAs connected to almost every aspect of protein synthesis and homeostasis.

Based on the finding that WDR5 controls expression of half the RPGs, we speculated that WINi induce a ribosome subunit imbalance that leads to induction of p53 (Aho et al., 2019a). It is also possible, however, that quality control mechanisms deplete the entire inventory of ribosomal proteins (RP) during prolonged WIN site blockade. To distinguish between these possibilities, we tracked changes in ribosomal protein levels at two timepoints: 24 hr, when there is no overt cellular response to WIN site inhibition, and 72 hr, when cell proliferation begins to be inhibited (Aho et al., 2019a). The abundance of ribosomal proteins allows for the use of label-free quantitative mass spectrometry (LFQMS; Cox et al., 2014) in whole-cell lysates to feasibly track ribosome protein levels, while also providing insight into other changes in protein levels promoted by WINi. Spike-in controls were not included.

In this analysis, we tracked ~3,200 proteins at each timepoint (Figure 3A, Figure 3—source data 1), ~850 of which are significantly altered by C16 treatment. Consistent with the subtle effects of WINi on mRNA abundance and TE, most differences in protein levels triggered by C16 are less than twofold in magnitude (Figure 3B and C). At 24 hr, ~90% of the proteins that change in response to C16 score as increased, whereas by 72 hr this number drops to ~60% (Figure 3A). Although the induction of proteins in response to C16 is unexpected for agents that decrease translational capacity, we note that protein synthesis is largely unaffected after 24 hr of C16 treatment (Figure 2—figure supplement 1), and that this phenomenon is unlikely to be an artifact of normalization. Indeed, the distribution of peptide intensities in the LFQMS data is unaffected by normalization (Figure 3—figure supplement 1A), and we see almost as many highly abundant proteins increasing as decreasing with C16 treatment (Figure 3—figure supplement 1B–D), arguing against the idea that large decreases in highly expressed proteins (such as the RPs; Figure 3—figure supplement 1B) are creating the appearance of less abundant proteins being induced. Instead, we suggest that this may be a transient compensatory mechanism, or an early part of the response to WINi. Regardless, most instances of increased protein levels are transient—fewer than one-third of these proteins are still induced at day 3 (Figure 3D)—whereas a majority of the proteins decreased at day 1 are also decreased at day 3. Enrichment analysis (Figure 3—figure supplement 2A, Figure 3—source data 2) reveals that proteins induced at 24 hr are modestly enriched in those connected to exocytosis and leukocyte activation, as well as mTORC1 signaling and MYC. By 72 hr, we see induction of proteins linked to glycolysis and fatty acid metabolism, as well as apoptosis. Additionally, manual curation reveals that the number of induced p53 target proteins increases over time: 13 are induced at 24 hr compared to 24 at the 3-day point (Figure 3E). Commensurate with the onset of a functional response to WINi, therefore, is a modest expansion in the apparent impact of p53 on the proteome, as well as the emergence of apoptotic response indicators.

Not surprisingly, proteins that are reduced in abundance at 24 hr are significantly enriched in those linked to the ribosome (Figure 3—figure supplement 2B). This enrichment becomes stronger at 72 hr. We also observe, at 72 hr, suppression of proteins linked to MYC and E2F targets, as well as mTORC1 signaling. In terms of ribosome components, this analysis reveals a progressive decline in the ribosomal protein inventory. Going from 24 to 72 hr, there is an increase in the number of impacted ribosomal subunits as well as in the magnitude of their suppression (Figure 3F), and eventually almost all ribosomal subunits are in deficit, regardless of whether or not they are encoded by a WDR5-bound gene (Figure 3G). RPL22L1 is the most strongly suppressed protein at 72 hr (Figure 3C), with its levels reduced by an order of magnitude after 3 d of C16 treatment. Consistent with the highly coordinated nature of ribosome biogenesis (Dörner et al., 2023), decreases in the abundance of RP are accompanied by a progressive decline in rRNA expression—as revealed by metabolic labeling of RNAs with 2'-azido-2'-cytidine (AzCyd) (Figure 3—figure supplement 3, Figure 3—figure supplement 3—source data 1). Together, these experiments reveal that changes in ribosomal protein levels predicted from our transcriptomic studies manifest in reduced expression of ribosome components. Contrary to our earlier idea that WINi promote ribosome subunit imbalance, however, these data support a simpler model in which these inhibitors ultimately induce attrition of the majority of ribosomal proteins—as well as mature rRNAs.

Finally, we asked if the decline in ribosome inventory triggered by C16 is associated with nucleolar stress, as we have shown with C6 (Aho et al., 2019a). We used immunofluorescence to measure the redistribution of nucleophosmin (NPM1) from the nucleolus to the nucleoplasm; a characteristic of this phenomenon (Russo and Russo, 2017). Because inhibition of ribosome biogenesis via some rRNA inhibitors can induce a DNA damage response (Sanij et al., 2020), we also probed for the DNA damage marker ɣ-H2AX. We observed no obvious change in nucleolar morphology with up to 72 hr of C16 treatment (Figure 3—figure supplement 4A). We did, however, see a significant decrease in the nucleolar enrichment of NPM1 at the 72 hr treatment timepoint (Figure 3—figure supplement 4B), indicative of a nucleolar stress response. Notably, we did not observe induction of ɣ-H2AX foci in either the nucleolus or nucleoplasm, detecting it only in cells that were morphologically apoptotic (Figure 3—figure supplement 4A), consistent with studies showing that ɣ-H2AX is induced via the DNA fragmentation that occurs during apoptosis (Rogakou et al., 2000). Based on this analysis, we conclude that activation of nucleolar stress occurs in response to prolonged exposure to C16 and that, compared to some rRNA inhibitors, widespread induction of DNA damage is not a specific consequence of the action of WINi.

Next, we conducted a two-tier loss-of-function screen to identify genes that modulate the response of MLLr cells to WINi (Figure 4A). Our objective was to compare C6 and C16, and to identify high-confidence hits that are disconnected from cell viability. In tier 1, we carried out a screen using the GeCKOv.2 sgRNA library (Joung et al., 2017), which targets ~19,000 genes with six sgRNAs each, as well as ~1,200 miRNAs (four sgRNAs each). After transducing the library into MV4;11 cells expressing Cas9, we treated for 2 wk with 2 μM C6, during which time rapidly growing cells emerged within the transduced population (Figure 4—figure supplement 1A). We harvested genomic DNA, performed next-generation sequencing, and compared sgRNA representation before and after C6 treatment. We then inventoried genes with significant enrichment/depletion in corresponding sgRNAs in the treated population, removed pan-essential genes (Tsherniak et al., 2017), and created a custom library in which non-essential protein-coding ‘hits’ are targeted by four different sgRNAs (Doench et al., 2016). The smaller tier 2 library was then screened against C6 or C16, this time against a parallel DMSO-treated control population. This two-tiered approach allowed us to efficiently screen two different WINi and identify hits that are validated (for C6 at least) with up to 10 unique sgRNAs.

Although the first tier did not discriminate between genes that modulate fitness and those that modulate WINi response, several interesting observations emerged. Guide RNAs corresponding to ~70 genes were enriched and ~675 were depleted (Figure 4B), most of the latter of which are pan-essential (Figure 4—source data 1). Satisfyingly, TP53 is the most highly enriched gene in the screen (Figure 4B, Figure 4—figure supplement 1B). CDKN2A also scored as highly enriched in the initial screen (Figure 4B)—specifically those sgRNAs targeting p14^ARF(Figure 4—figure supplement 1C and D), an inhibitor of the p53 ubiquitin ligase MDM2 (Sherr, 2001). Further support for the importance of p53 is evidenced by network analysis (Chang and Xia, 2023) of the 27 miRNAs flagged as enriched (Figure 4—source data 1), which display connections to p53 (Figure 4—figure supplement 1E). Finally, we note that the second most highly enriched gene in the first tier encodes a ribosomal protein: RPL22 (Figure 4B, Figure 4—figure supplement 1F). Because of the strong enrichment of sgRNAs against RPL22 and TP53, we removed both genes from the second tier screen.

The second tier screen (Figure 4—figure supplement 1G, Figure 4—source data 2) revealed that the response of MV4;11 cells to C6 and C16 is very similar, both in terms of the enriched/depleted genes and their rankings (Figure 4C). A majority of genes that modulate the response to C6 similarly modulate the response to C16 (Figure 4D). For most of the depleted genes that appear specific to one WINi, similar depletion is observed with the other WINi, but is generally just over the FDR cutoff (Figure 4—figure supplement 1H). But for C6-specific enriched genes we see that most have high FDR values in the C16 samples, arguing that the earlier generation compound has expanded, off-target, activities. Gene Ontology (GO) enrichment analysis of the 57 common genes emerging from the screen revealed enrichment in four overlapping categories connected to p53 signaling, apoptosis, the DNA damage response (DDR), and histone modifications (Figure 4E, Figure 4—figure supplement 1I). The representation of genes connected to p53 and apoptosis reinforces the importance of p53-mediated cell death to the response of MLLr cells to WINi. We observe, for example, that loss of function of the p53 antagonist and ubiquitin ligase MUL1 (Jung et al., 2011), increases sensitivity to C6/C16, whereas loss of canonical p53 effectors NOXA, BAX, and ZMAT3 is associated with a decrease in response. The DNA damage response category overlaps with that of p53 but is nonetheless distinct and includes genes encoding the ATM and CHK2 kinases (Blackford and Jackson, 2017) and the FOXM1 transcription factor that activates DDR gene expression networks (Zona et al., 2014). This category also includes two depleted genes, encoding FBXL5—which antagonizes ATM signaling (Chen et al., 2014)—and DYRK1A—a kinase involved in the DDR (Laham et al., 2021), DREAM complex activation (Litovchick et al., 2011), and RPG transcription (Di Vona et al., 2015). The involvement of this category of enriched genes is intriguing, given the lack of ɣ-H2AX accumulation in non-apoptotic cells (Figure 3—figure supplement 4), and warrants further investigation in the future.

A majority of genes in the histone modification category, when disrupted, blunt the response to both WINi (Figure 4E). These genes include those encoding the H3K27 demethylase KDM6A (Lan et al., 2007), the MLL/SET protein KMT2D (MLL2; Shinsky et al., 2015), and ARID4A—a component of the mSin3/HDAC1 co-repressor complex (Lai et al., 2001). The most conspicuous sensitizing gene in this group is BRD3, a member of the BET family of proteins that includes BRD2 and BRD4 (Eischer et al., 2023). Interestingly, although BRD4 was not included in the second screen tier as it is pan essential, BRD2 was not included because it was not significantly enriched/depleted in the first tier (Figure 4—source data 1), revealing that the actions of BRD3 in modulating response to WINi are not shared with all family members. Further supporting the importance of BRD3 to the response, we note that SPOP, which targets BET family proteins for proteasomal destruction (Janouskova et al., 2017), is one of the most significantly enriched hits from the screen (Figure 4E).

Collectively, these findings demonstrate functional involvement of the ribosomal protein RPL22 in the response to WINi and confirm the importance of the p53 network to robust inhibition of MLLr cell growth by these agents. These findings also identify a number of candidate predictive biomarkers of response that can be further interrogated for their value as patient selection criteria.

Given the ways in which resistance to WINi can arise (Figure 4), the most efficacious application of these agents will likely be in combination with other therapies. We therefore asked whether C16 synergizes with 11 approved or targeted agents. Several of the agents were chosen based on the results of our screen. Harmine—an inhibitor of the DYRK1A kinase (Göckler et al., 2009)—and the BET family inhibitor mivebresib (Lin et al., 2017) each target the product of sensitizing genes, whereas venetoclax inhibits BCL-2 (Souers et al., 2013)—an inhibitor of BAX, which scored as a resistance gene. We also tested agents connected to the DDR (etoposide, olaparib, and the ATR inhibitor VE821; Charrier et al., 2011), protein synthesis and homeostasis (alvespimycin and rapamycin), and p53 (nutlin-3a). Due to the enrichment of PRMT5 substrates in our translational profiling (Figure 2—figure supplement 2F), we queried the PRMT5 inhibitor pemrametostat (Chan Penebre et al., 2015). Because DOT1L inhibitors suppress not only classic MLL fusion target genes (Bernt and Armstrong, 2011; Daigle et al., 2011) but also RPGs (Lenard et al., 2020), we tested for synergy with the DOT1L inhibitor pinometostat (Daigle et al., 2011). We treated MV4;11 cells with a dose matrix spanning 49 unique dose combinations and quantified synergy δ-scores using the zero interaction potency (ZIP) model (Yadav et al., 2015; Figure 5A and B, Figure 5—figure supplement 1A and B, and Figure 5—source data 1).

In MV4;11 cells, we observe synergy with mivebresib, pemrametostat, pinometostat, etoposide, harmine, and venetoclax. Within this group are three agents selected based on sensitizing targets from the CRISPR screen, providing additional support for the role of DRYK1A, BCL-2/BAX, and BRD3 in the responsiveness to WINi. Of the three agents connected to the DDR, only etoposide displays significant synergy. Agents that target protein synthesis and homeostasis yield mixed results—we observe potent antagonism with the mTOR inhibitor (Raught et al., 2001) rapamycin (peak δ-score –16), while the HSP90 inhibitor alvespimycin (Schnur et al., 1995) is either antagonistic or synergistic, depending on dose (Figure 5A). Finally, we note that—of the agents displaying synergy—four are particularly strong (peak δ-scores > 10) and observed at agent doses consistent with on-target activity (Figure 5—figure supplement 1B), suggesting that mivebresib, pemrametostat, pinometostat, and venetoclax should be prioritized for in vivo testing. Focusing on these agents is further justified by our finding that all four are synergistic with C16 in MOLM13 cells (Figure 5C and D, Figure 5—figure supplement 2A and B).

To understand how combination with another agent impacts the response to WINi, we transcriptionally profiled MV4;11 cells treated for 48 hr with C16 and mivebresib, either as single agents or in combination, at concentrations that yield peak synergy between them (100 nM C16 and 2.5 nM mivebresib). Spike-in controls were not included. By RNA-seq (Figure 5—source data 2), it is clear that the functional synergy between C16 and mivebresib is apparent at the transcript level, with more than 6,200 gene expression changes in the combination treatment, compared to less than 1,800 for C16 and 2700 for mivebresib (Figure 5E). Notable are the very distinct transcriptional profiles induced by each agent alone, with fewer than 200 shared gene expression changes in each direction (Figure 5F). The impact on RPG expression of both agents is additive (Figure 5—figure supplement 3A), but in general we find that the combination of C16 and mivebresib dysregulates similar categories of genes for each agent alone, but with substantially more genes in each category (Figure 5—source data 3). This is clear for genes linked to translation (Figure 5G), p53 (Figure 5—figure supplement 3B), and the induction of apoptosis (Figure 5H). Thus, although further investigation is needed, this analysis is consistent with the idea that synergy between C16 and mivebresib results from alterations in the expression of distinct but complementary sets of genes that ultimately conspire to augment induction of p53.

Despite the importance of p53 in the response to WINi, WINi cause only a slight increase, if any, in p53 levels (Figure 6—figure supplement 1A, Figure 6—source data 1; Aho et al., 2019a). Interestingly, inactivation or loss of RPL22 in cancer is associated with increased expression of RPL22L1 and inclusion of exon 6 in MDM4 (Ghandi et al., 2019), an event that promotes MDM4 expression by preventing formation of a ‘short’ MDM4 mRNA isoform (MDM4s) that is destroyed by nonsense-mediated decay (Rallapalli et al., 1999). MDM4 is intriguing because it can suppress p53 without altering its stability (Francoz et al., 2006). It is also intriguing because skipping of exon 6 in the MDM4 mRNA is stimulated by ZMAT3 (Bieging-Rolett et al., 2020; Muys et al., 2021) and antagonized by RPL22L1 (Larionova et al., 2022)—two genes that are oppositely regulated by WINi. We therefore asked if WINi induce changes in the levels of mRNA splice isoforms and if this includes MDM4.

RNA-seq data (Figure 1) were interrogated for alternative splicing events (Shen et al., 2014). At an FDR < 0.05 and a threshold of ≥5% change in exon inclusion (∆ψ), C6 and C16 each result in changes in ~1,000 differentially spliced mRNAs (Figure 6—figure supplement 1B), ~250 of which are shared between the two inhibitors (Figure 6A). Many of these changes reflect events with low read counts or at minor splice sites (Figure 6—source data 2), representative examples of which are presented in Figure 6—figure supplement 1C and D. That said, WINi clearly promote accumulation of MDM4 transcripts in which exon 6 is skipped (Figure 6B). We also observe splicing changes at RPL22L1 itself (Figure 6—figure supplement 1E), where WINi leads to the depletion of transcripts in which exon 2 is spliced to a distal 3′ acceptor site in exon 3. This splicing event encodes the RPL22L1a isoform that is incorporated into ribosomes (Larionova et al., 2022). Splicing to the proximal 3′ acceptor site, which generates a non-ribosomal RPL22L1b isoform that modulates splicing, is insensitive to WINi. We confirmed the impact of C16 on MDM4 and RPL22L1 splice isoforms by semi-quantitative RT-PCR (Figure 6—figure supplement 1F, Figure 6—source data 3) and quantitative RT-PCR (Figure 6—figure supplement 1G). Based on these observations, we conclude that treatment of MLLr cells with WINi promotes the selective loss of transcripts encoding RPL22L1a and MDM4.

The association of RPL22 loss with increased expression of RPL22L1 and inclusion of exon 6 in MDM4 (Ghandi et al., 2019) prompted us to ask how RPL22 contributes to the response of MLLr cells to WINi. Knockout (KO) of RPL22 (Figure 6—figure supplement 2A, Figure 6—source data 4) decreases the sensitivity of MV4;11 and MOLM13 cells to C16 by three- to fivefold compared to non-targeted (NT) control cells (Figure 6C, Figure 6—figure supplement 2B), as well as attenuating the modest induction of p53 protein observed in the MV4;11 line (Figure 6D, Figure 6—source data 5). The response of relatively insensitive (p53-null) K562 cells, in contrast, is unaffected by RPL22 disruption (Figure 6C). RNA-seq analysis, performed without spike-in controls, (Figure 6—figure supplement 2C and D, Figure 6—source data 6) reveals that disruption of RPL22 does not impact the effect of WINi on WDR5-bound RPGs (Figure 6—figure supplement 2E), but it does block the effects of C16 on expression of RPL22L1 and ZMAT3 (Figure 6—figure supplement 2D), as well as tempering its ability to suppress genes connected to the cell cycle, mTORC1 signaling, and MYC (Figure 6—figure supplement 2F, Figure 6—source data 7). Notably, RPL22 loss also impairs induction of genes involved in p53 signaling (Figure 6E, Figure 6—figure supplement 2G). We also observe that mitochondrial RPGs are induced by WINi uniquely in RPL22-null cells (Figure 6—figure supplement 2H). We conclude that RPL22 is needed for a majority of the characteristic responses of MLLr cells to WINi, including activation of p53.

Finally, we asked if RPL22 knockout alters patterns of alternative splicing induced by WINi. Thousands of differences were detected in splice isoforms between the various pairwise comparisons (Figure 6—source data 8). In general, RPL22KO cells show fewer C16-induced changes in alternative splicing patterns than NT cells (Figure 6—figure supplement 3A). As we observed above, a majority of the changes reflect events with low read counts or at minor splice sites, with two notable exceptions: MDM4 and RPL22L1. In the absence of WINi, disruption of RPL22 promotes exon 6 retention in MDM4 (Figure 6—figure supplement 3B) and induces expression of the mRNA splice isoform encoding RPL22La (Figure 6—figure supplement 3C). In the presence of C16, disruption of RPL22 mitigates skipping of exon 6 in MDM4 (Figure 6F) and largely blocks suppression of the RPL22L1a-encoding mRNA isoform (Figure 6—figure supplement 3C). Importantly, these changes in mRNA isoforms manifest at the protein level (Figure 6G, Figure 6—source data 9) as we observe that both MDM4 and RPL22L1 protein expression is suppressed by C16, that RPL22L1 is induced by RPL22 disruption, and that loss of RPL22 blocks a majority of the ability of C16 to reduce expression of the MDM4 and RPL22L1 proteins. We also note that loss of RPL22 blocks induction of the p53 target, p21. Taken together, these data demonstrate that RPL22 loss can cause the induction of RPL22L1 and inclusion of exon 6 in MDM4 observed in cancer (Ghandi et al., 2019) and reveal that splicing-driven suppression of MDM4 is an important mechanism through which WINi activate p53 in MLLr cells.

All key resources are provided in Appendix 1—key resources table.

Plasmids and cell lines generated in this study are available upon request from the corresponding author (william.p.tansey@vanderbilt.edu).

MV4;11 (RRID:CVCL_0064), MOLM13 (RRID:CVCL_2119), and K562 (RRID:CVCL_0004) cell lines and their derivatives were cultured in RPMI-1640 media with 10% FBS, 10 U/mL penicillin, and 10 µg/mL streptomycin at 37°C and 5% CO₂. HEK293T (RRID:CVCL_1926) cells were cultured in DMEM media with 10% FBS, 10 U/mL penicillin, and 10 µg/mL streptomycin at 37°C and 5% CO₂. MV4;11 and MOLM13 cell lines are male. K562 and HEK293T cell lines are female. Cell lines were split every 2–4 d and suspension cells maintained between 1 × 10⁵ and 1 × 10⁶ cells/mL. Cell line identity was authenticated by STR profiling. All cell lines tested negative for mycoplasma.

MV4;11, MOLM13, and K562 control (NT) and RPL22 knockout (KO) cell lines were generated by CRISPR using the multi-guide Synthego Gene Knockout System. Briefly, ribonucleoprotein (RNP) complexes containing Cas9-2NLS (Synthego) and either non-targeting (NT) control sgRNA#1 (Synthego) or RPL22 sgRNAs (Synthego Gene Knockout Kit v2 – human – RPL22) were formed by incubating 90 pmol sgRNA and 10 pmol Cas9-2NLS in Buffer R (Component of Neon Transfection System Kit; Thermo Scientific) at room temperature (RT) for 10 min. MV4;11, MOLM13, or K562 cells were electroporated (2 × 10⁵ cells per reaction) with RNP complexes using the Neon Transfection System (Thermo Fisher Scientific) with the following parameters using Buffer R in 10 µL reactions: MV4;11 cells: 1175 V pulse, 40 ms pulse width, one pulse; MOLM13 cells: 1075 V pulse, 30 ms pulse width, two pulses; K562 cells: 1450 V pulse, 10 ms pulse width, three pulses. Cells recovered undisturbed in media absent of antibiotics for 48 hr before expansion and screening for loss of RPL22 expression by western blot analysis.

Cells were treated with 0.1% DMSO or varying concentrations of C6 or C16 for 24 hr. A custom QuantiGene Plex panel (Thermo Fisher Scientific) was used in conjunction with the QuantiGene Sample Processing Kit for cultured cells (Thermo Fisher Scientific), and QuantiGene Plex Assay kit (Thermo Fisher Scientific) to quantify transcripts following the manufacturer’s instructions. Probe regions and accession numbers are as follows: RPS24 (NM_001026, region 5-334), RPL35 (NM_007209, region 2-430), RPL26 (NM_000987, region 37-445), RPS14 (NM_005617, region 61-552), RPL32 (NM_000994, region 95-677), RPS11 (NM_001015, region 139-634), RPL14 (NM_003973, region 108-530), and GAPDH (NM_002046, region 2-407). The average net mean fluorescence intensity was read on a Luminex FLEXMAP 3D System (Invitrogen). Signals from RPGs were normalized internally to those from GAPDH, and then to the DMSO control. Dose–response curves from the mean of biological replicates were calculated with the R package drc (Ritz et al., 2015).

Cells were collected by centrifugation and washed once with ice-cold PBS. Cells were lysed in either RIPA buffer (50 mM Tris, pH 8.0; 150 mM NaCl; 5 mM EDTA; 1.0% NP-40; 0.5% sodium deoxycholate; 0.1% SDS) or Triton-X buffer (50 mM Tris, pH 8.0; 150 mM NaCl; 5 mM EDTA; 1% Triton X-100), each supplemented with protease and phosphatase inhibitors (2× cOmplete, EDTA-free, Protease Inhibitor Cocktail [Roche]; 1× PhosSTOP Phosphatase Inhibitor [Roche]; 100 µg/mL Pefabloc SC [Roche]), while incubating on ice for 10 min. Chromatin was sheared by brief sonication at 25% on ice, insoluble material cleared by centrifugation, and protein quantified by Pierce BCA Protein Assay (Thermo Scientific). Protein samples were diluted to equal concentrations in lysis buffer and boiled for 5 min in 1× Laemmli Sample Buffer. Samples were run on 4–20% TGX Precast Polyacrylamide Gels (Bio-Rad) or hand-cast single percentage polyacrylamide gels, wet transferred to Amersham Protran Western Blotting Nitrocellulose Membrane (Cytiva) for 1 hr at 100 V in Towbin Buffer (25 mM Tris; 192 mM glycine; 10% methanol), and blocked in 5% milk in TBS-T before incubation overnight with one of the following primary antibodies: anti-p53 (Santa Cruz Biotechnology, Cat# sc-126), anti-RPL22 (Santa Cruz Biotechnology, Cat# sc136413), anti-RPL22L1 (Thermo Fisher Scientific, Cat# PA5-63266), anti-MDM4 (Sigma-Aldrich, Cat# M0445), anti-p21 (Cell Signaling Technology, Cat# 2947), anti-⍺-actinin (Cell Signaling Technology, Cat# 12413), or anti-GAPDH (Cell Signaling Technology, Cat# 8884). Membranes were washed three times with TBS-T and, if required, incubated with anti-mouse-HRP secondary antibody (Jackson ImmunoResearch Laboratories, Inc, Cat# 115-035-174) or anti-rabbit-HRP (Cell Signaling Technology, Cat# 7074) for 1 hr. Blots were developed with Clarity ECL Western Blotting Substrate (Bio-Rad) and imaged on a ChemiDoc Imaging System (Bio-Rad).

MV4;11 cells were treated with 0.1% DMSO (vehicle) or C16 (100 nM) for up to 72 hr or actinomycin D (5 nM) for 6 hr. Cells were fixed in 4% paraformaldehyde (PFA) for 10 min, washed three times with PBS, then cytospun onto slides. Cells were permeabilized with 0.5% Triton X-100 in PBS (PBSTx) for 15 min then blocked with 1% bovine serum albumin in PBSTx (blocking buffer) and immunostained with antibodies against NPM1 (Abcam, ab10530) and gH2A.X pSer139 (Cell Signaling Technologies, 9718). Cells were washed with PBSTx then stained with secondary antibodies (Thermo Fisher, A11001 and A11037). Following PBSTx washes, cells were counterstained with Hoechst (Thermo Fisher, H3570), washed with PBS, then mounted with ProLong Antifade Gold (Thermo Fisher). Images were acquired on using a Plan Fluor ×40 Oil DIC H N2 (NA 1.3, WD 240 mm) objective on a Nikon Ti-2 microscope with a Nikon D-LEDI light source and a Prime BSI Express Scientific sCMOS camera in the Vanderbilt University Cell Imaging Shared Resource. Images were processed and analyzed using NIS-Elements (version 5.42.03) and FIJI (version 2.3.0/1.53q). Images presented are single z-sections of representative cells. To quantify nucleolar localization of NPM1, masks of nuclei were generated from Hoechst channel and nucleolar NPM1 was manually thresholded. The integrated fluorescence intensity of nucleolar NPM1 was then divided by total nuclear NPM1. p-Values were calculated by Student’s t-tests comparing treatment samples to DMSO samples within each timepoint.

Bulk protein synthesis was measured using the OP-PURO labeling method (Liu et al., 2012). MV4;11 cells were treated with either 0.1% DMSO, 2 µM C6, or 100 nM C16 for 24, 48, or 96 hr. For a positive control for inhibition of protein synthesis, MV4;11 cells were treated with 100 µg/mL cycloheximide (Research Products International) for 30 min. Following treatments, 2 × 10⁶ cells were pulsed with 50 µM O-propargyl-puromycin (Invitrogen, Cat# C10459), or 0.1% DMSO for the ‘No OPP’ unlabeled control, for 1 hr at 37°C. Cells were collected, washed with ice-cold PBS, and cross-linked in 500 µL Cross-Linking Buffer (1× PBS, 1% formaldehyde) for 15 min on ice. Cross-linked cells were washed with ice-cold PBS and permeabilized in 500 µL Permeabilization Buffer (1× PBS, 3% FBS, 10% saponin) for 5 min at RT. Click-iT reactions containing cells in 500 µL Click-iT Reaction Cocktail (Invitrogen) with 5 µM Alexa Fluor 647 Azide (Invitrogen) were performed for 30 min at RT while protected from light. Cells were washed with PBS + 3% FBS and resuspended in PBS. Fluorescence intensities from Alexa Fluor 647 were measured using a BD LSRFortessa Cell Analyzer and geometric means calculated from 10,000 cells per sample with FlowJo software (BD Bioscience). Experiments were repeated in biological triplicate. Normalized fluorescence values were calculated by setting fluorescence from cycloheximide-treated samples as the baseline. p-Values were calculated by Student’s t-tests comparing treatment samples to DMSO samples within each timepoint. Flow Cytometry experiments were performed in the VUMC Flow Cytometry Shared Resource.

Metabolic labeling of rRNA was performed as previously described, with some modifications (Wang et al., 2020). MV4;11 cells were treated with either 0.1% DMSO, 2 µM C6, or 100 nM C16 for 24, 48, or 96 hr. For a positive control treatment for inhibited rRNA transcription, MV4;11 cells were treated with 5 nM actinomycin D (Cayman Chemical Company) for 60 min. Following treatment, cultures were pulsed with 1 mM 2’-azido-2’-deoxycytidine (Biosynth), or 0.1% DMSO for an unlabeled control, for 12 hr and total RNA isolated using TRIzol Reagent (Invitrogen) as per the manufacturer’s instructions. Pelleted RNA was resuspended in 20 µL nuclease-free water and treated with DNase I (New England BioLabs) for 10 min at 37°C in the presence of RNasin Ribonuclease Inhibitor (Promega). DNase-treated RNA was purified with the RNA Clean and Concentrator-25 Kit (Zymo Research) as per the manufacturer’s instructions. SPAAC reactions containing 12.5 µg RNA, 100 µM MB 680R DBCO (Vector Laboratories), 20 U RNasin Ribonuclease Inhibitor (Promega), and 1× PBS were incubated for 2 hr at 37°C, and then RNA purified with the RNA Clean and Concentrator-25 Kit (Zymo Research) as per the manufacturer’s instructions. RNA was subjected to electrophoretic separation on 1% TAE-agarose gels and MB 680R-labeled RNA imaged on an Odyssey CLx Imager (LI-COR). Total RNA was stained with SYBR Safe DNA Gel Stain (Invitrogen) and imaged on a ChemiDoc Imaging System (Bio-Rad). Fluorescence signals from 28S and 18S bands were quantified using Empiria Studio (LI-COR).

Ribo-seq was performed as previously described with some modifications (McGlincy and Ingolia, 2017). MV4;11 cells treated for 48 hr with either 0.1% DMSO, 2 µM C6, or 100 nM C16 were washed with ice-cold PBS and lysed in 400 µL Lysis Buffer (20 mM Tris, pH 7.4; 150 mM NaCl; 15 mM MgCl₂; 1 mM DTT; 100 µg/mL cycloheximide; 1% Triton X-100; 25 U/mL Turbo DNase I) by incubation on ice for 10 min followed by homogenization by syringe. Lysates were cleared by centrifugation at 4°C, and RNA quantified by Qubit RNA HS Assay (Invitrogen) following the manufacturer’s instructions. 30 µg RNA was diluted in 200 µL Polysome Buffer (20 mM Tris, pH 7.4; 150 mM NaCl; 5 mM MgCl₂; 1 mM DTT; 100 µg/mL cycloheximide) and incubated with 15 U RNase I (Lucigen) for 45 min while rotating at RT. RNA digestion was quenched with 10 µL SUPERaseIn RNase Inhibitor (Invitrogen) and samples transferred to 13 mm × 51 mm ultracentrifuge tubes (Beckman-Coulter), underlaid with 900 µL 1 M sucrose in polysome buffer supplemented with 20 U/mL SUPERaseIn RNase Inhibitor, and centrifuged at 540,628 × g 1 hr at 4°C. Ribosome pellets were suspended in TRIzol Reagent (Invitrogen), and RNA extracted from ribosome pellets by Direct-zol RNA MiniPrep Kit (Zymo Research). RNA and carrier glycogen were precipitated by adding 1.5 volumes 100% isopropanol supplemented with 0.12 M NaOAc, pH 5.5, followed by incubation on dry ice for 30 min and centrifugation at 16,800 × g for 30 min at 4°C. RNA was resuspended in 5 µL 10 mM Tris, pH 8.0, and 1× Denaturing Sample Loading Buffer (98% formamide; 10 mM EDTA; 300 µg/mL bromophenol blue) and subjected to electrophoresis on 15% polyacrylamide TBE-Urea gels (Invitrogen). Gels were stained briefly with 1× SYBR Gold (Invitrogen), 17–34 nucleotide fragments excised, and RNA fragments extracted by mechanical disruption, suspension in 500 µL RNA Gel Extraction Buffer (300 mM NaOAc, pH 5.5; 1 mM EDTA; 0.25% SDS), freezing on dry ice for 30 min, and rotating overnight at RT. Polyacrylamide was removed by centrifugation through Costar Spin-X columns (Corning) and RNA precipitated with isopropanol as described above. RNA was dephosphorylated by incubation with 5 U T4 Polynucleotide Kinase in 1× T4 PNK Buffer (New England BioLabs) supplemented with SUPERaseIn RNase Inhibitor for 1 hr at 37°C and ligated to bar-coded linkers (NI-810: /5Phos/NNNNNATCGTAGATCGGAAGAGCACACGTCTGAA/3ddC/; NI-811: /5Phos/NNNNNAGCTAAGATCGGAAGAGCACACGTCTGAA/3ddC/; NI-812: /5Phos/NNNNNCGTAAAGATCGGAAGAGCACACGTCTGAA/3ddC/) pre-adenylated with 100 U of T4 RNA Ligase 2, truncated K227Q, in 1X T4 RNA Ligase Buffer (New England BioLabs) supplemented with 35% w/v PEG-8000 and incubated at 37°C for 3 hr. Ligation was verified by electrophoresis, samples combined, and linker-ligated RNA precipitated with isopropanol as described above. Ribosomal RNA was depleted from samples using the RiboCop rRNA Depletion Kit (Lexogen) and RNA precipitated with isopropanol as described above. Linker-ligated RNA was reverse transcribed with 200 U SuperScript III in 1× First Strand Buffer (Invitrogen), dNTPs, DTT, 10 U SUPERaseIn RNase Inhibitor and primer NI-802 (/5Phos/NNAGATCGGAAGAGCGTCGTGTAGGGAAAGAG/iSp18/GTGACTGGAGTTCAGACGTGTGCTC). RNA template was hydrolyzed for 15 min at 98°C in the presence of 0.1 M NaOH and cDNA precipitated with isopropanol as described previously. Reverse-transcribed DNA was subjected to electrophoresis on 15% polyacrylamide Novex TBE-Urea gels (Invitrogen) and the 105-nucleotide reverse-transcription product excised from polyacrylamide as described above, except with DNA Gel Extraction Buffer (300 mM NaCl; 10 mM Tris, pH 8.0; 1 mM EDTA). cDNA was circularized with 100 U CircLigase ssDNA Ligase (Lucigen) in the presence of 1× CircLigase Buffer, ATP, and MnCl₂ at 60°C for 1 hr followed by 80°C for 10 min. Circularized cDNA was quantified by qPCR, amplified using Phusion Polymerase (New England BioLabs) with unique dual-indexed primers (UDI0050_i5: AATGATACGGCGACCACCGAGATCTACACGCTCCGACACACTCTTTCCCTACACGACGCTCTTCCGATCT; UDI0050_i7: CAAGCAGAAGACGGCATACGAGATTAGAGCGCGTGACTGGAGTTCAGACGTGT; UDI0051_i5: AATGATACGGCGACCACCGAGATCTACACATACCAAGACACTCTTTCCCTACACGACGCTCTTCCGATCT; UDI0051_i7: CAAGCAGAAGACGGCATACGAGATAACCTGTTGTGACTGGAGTTCAGACGTGT; UDI0052_i5: AATGATACGGCGACCACCGAGATCTACACGCGTTGGAACACTCTTTCCCTACACGACGCTCTTCCGATCT; UDI0052_i7: CAAGCAGAAGACGGCATACGAGATGGTTCACCGTGACTGGAGTTCAGACGTGT), amplicons subjected to electrophoresis on 8% polyacrylamide Novex gels (Invitrogen), and products >160 bp excised as described. Libraries were submitted to VANTAGE (Vanderbilt Technologies for Advanced Genomics) for sequencing on a NovaSeq 6000.

Adapters were trimmed from reads using cutadapt (Martin, 2011), and UMIs removed from reads and attached to read IDs using UMI-tools (Smith et al., 2017). Reads were demultiplexed using sabre and aligned against ribosomal RNA using bowtie2 (Langmead and Salzberg, 2012). Reads not mapping to rRNA were mapped to the hg19 transcriptome using STAR (Dobin et al., 2013) and deduplicated using UMI-tools (Smith et al., 2017). Count tables for reads mapping to central ORFs were generated using the coverage command from bedtools (Quinlan and Hall, 2010). After batch removal, Ribo-seq read counts were normalized to mRNA read counts using Xtail (Xiao et al., 2016) to calculate translation efficiencies and statistics. FDR values were calculated using the Cochran–Mantel–Haenszel test (CMH). Genes with significantly altered translation efficiencies were those with FDR < 0.05 and absolute log2FC > 0.25. Identification of optimal RPF P-site offsets, RPF triplet periodicity, and RPF localization to CDS and UTR regions was performed with the R package riboWaltz (Lauria et al., 2018).

For RNA-seq performed in parallel with Ribo-seq, RNA was isolated by Direct-zol RNA MiniPrep Kit (Zymo Research) from 100 µL of cell lysates after homogenization by syringe and clearing by centrifugation. For combination WINi/BETi treatment, MV4;11 cells were treated for 48 hr with either 0.2% DMSO, 100 nM C16, 2.5 nM mivebresib, or combined 100 nM C16 and 2.5 nM mivebresib, and RNA isolated by Direct-zol RNA MiniPrep Kit (Zymo Research) with on-column DNAse-treatment. For RNA-seq in MV4;11 NT and RPL22 KO cells, cultures were treated for 48 hr with either 0.1% DMSO or 100 nM C16 before RNA isolation as described above for WINi/BETi RNA-seq. For all RNA-seq experiments, RNA was submitted to the Vanderbilt Technologies for Advanced Genomics (VANTAGE) core facility for library preparation with rRNA-depletion using standard Illumina protocols and sequencing on an Illumina NovaSeq 6000.

Adapters were trimmed from RNA-seq reads using cutadapt (Martin, 2011) and reads aligned to the hg19 genome using STAR (Dobin et al., 2013). Gene expression was quantified using featureCounts (Liao et al., 2014) and differential analysis performed using DESeq2 (Love et al., 2014) which calculates p-values through the Wald test and adjusts p-values by the Benjamini–Hochberg procedure to calculate FDR. Changes in levels of alternative splicing events were quantified using rMATS, which calculates changes in exon inclusion levels (∆ψ), and p-values through a likelihood-ratio test. Genes with significantly altered transcript levels are those with FDR < 0.05. Significant changes in alternative splicing events are those with FDR < 0.05 and ∆ψ > 5%.

To generate Cas9 expression lentivirus, HEK293T cells were transfected with the viral transfer plasmid lentiCas9-Blast (Sanjana et al., 2014) (gift from Feng Zhang; Addgene plasmid # 52962), the viral packaging plasmid psPAX2 (gift from Didier Trono; Addgene plasmid # 12260), and the viral envelope plasmid pMD2.G (gift from Didier Trono; Addgene plasmid # 12259) using Lipofectamine 3000 Transfection Reagent (Invitrogen). After 48 hr, virus-containing media was collected and used to transduce MV4;11 cells by spinfection (2 hr; 1000 × g; RT; 8 µg/mL hexadimethrine bromide). Following spinfection, virus-containing media was replaced with fresh media and cells allowed to recover for 48 hr before selection with 10 µg/mL blasticidin (Research Products International). A clonal MV4;11 Cas9 cell line was established by serial dilution of the population and screening for retention of WINi sensitivity.

Tier 1 CRISPR screens were performed essentially as described (Joung et al., 2017). Briefly, the Human GeCKOv2 CRISPR Knockout Pooled Library (A+B) in the lentiGuide-Puro vector backbone (gift from Feng Zhang; Addgene plasmid # 1000000048) was amplified and purified as directed by Addgene. Lentiviral particles were generated by transfecting HEK293T cells with the GeCKOv2 CRISPR Knockout Pooled Plasmid Library, psPAX2 (gift from Didier Trono; Addgene plasmid # 12260), and pMD2.G (gift from Didier Trono; Addgene plasmid # 12259) using Lipofectamine 3000 Transfection Reagent (Invitrogen). After 48 hr, viral media was collected, aliquoted, and stored at –80°C. In duplicate, clonal Cas9-expressing MV4;11 cells were transduced by spinfection (2 hr; 1000 × g; RT; 8 µg/mL hexadimethrine bromide) with a volume of virus-containing media sufficient to infect 30% of cells and at a scale to generate >200 transduced cells per sgRNA in the library. Cells recovered in fresh media overnight, were split 1:2, and selected with 1 µg/mL puromycin for 48 hr to generate the MV4;11 Cas9+GeCKOv2 population.

MV4;11 Cas9 and MV4;11 Cas9+GeCKOv2 cells were treated with either 0.1% DMSO or 2 µM C6, replenished every 3 d with fresh media and C6, and counted daily by trypan blue exclusion. DMSO-treated populations were grown until cultures reached >8 × 10⁵ cells/mL to verify C6-treatment efficacy. C6-treated MV4;11 Cas9+GeCKOv2 populations were maintained below 8 × 10⁵ cells/mL and grown until a resistant population emerged relative to C6-treated MV4;11 Cas9 cells. Genomic DNA was isolated from MV4;11 Cas9+GeCKOv2 cells collected before and following sustained C6 treatment using the Quick-gDNA MidiPrep Kit (Zymo Research) as per the manufacturer’s directions. Sequencing libraries were generated by amplifying sgRNA sequences from genomic DNA using bar-coded Illumina-compatible adapter-containing primers and NEBNext High-Fidelity 2× PCR Master Mix (New England BioLabs). PCR products were pooled and purified with a ZymoSpin V column with Reservoir (Zymo Research). Libraries were sequenced on an Illumina NextSeq 500 in the Vanderbilt Technologies for Advanced Genomics (VANTAGE) core facility.

The tier 2 sgRNA plasmid library was generated as previously described with some modifications (Joung et al., 2017). Briefly, sgRNA sequences against a curated collection of genes and 200 non-targeting control sgRNA sequences were extracted from the Brunello sgRNA Library (Doench et al., 2016). For genes of interest not included in the Brunello library, four sgRNAs targeting each gene were designed with the CHOPCHOP sgRNA design tool (Labun et al., 2019). sgRNA sequences were appended with 5′ and 3′ flanking sequences and synthesized as an Oligo Pool (Figure 4—source data 2; Twist Bioscience) followed by PCR amplification using NEBNext HiFidelity 2× Master Mix (New England BioLabs) with Fwd primer (GTAACTTGAAAGTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACC) and KO Rev primer (ACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC). PCR amplicons were subjected to agarose gel size selection using the NucleoSpin Gel and PCR Clean-up Kit (Macherey-Nagel). Amplicons were cloned into BsmBIv2-digested (New England BioLabs) lentiGuide-PURO plasmid (gift from Feng Zhang; Addgene plasmid # 52963) via Gibson Assembly (New England BioLabs). Gibson Assembly products were precipitated with isopropanol and electroporated into Endura ElectroCompetent Escherichia coli (Lexogen). Amplified plasmids were isolated from E. coli using the Nucleobond Xtra Maxi EF Kit (Macherey-Nagel) and adequate representation of sgRNAs in the library was verified by next-generation sequencing and analysis with the Python script count_spacers.py (Joung et al., 2017).

Tier 2 sgRNA Library lentiviral particles were generated and MV4;11 Cas9 cells transduced as described above for the tier 1 screen at a scale to achieve >500 cells per sgRNA in the library. MV4;11 Cas9+Targeted sgRNA Library populations were treated with either 0.1% DMSO, 2 µM C6, or 100 nM C16 for 15 d. Cultures were maintained below 8 × 10⁵ cells/mL and cultures replenished every 3 d with media and fresh DMSO, C6, or C16. Genomic DNA was isolated, and Illumina-compatible next-generation sequencing libraries generated as described above for the tier 1 screen. Libraries were sequenced on an Illumina NovaSeq 6000 in the Vanderbilt Technologies for Advanced Genomics (VANTAGE) core facility.

Adapters were trimmed from reads using cutadapt (Martin, 2011). Generation of sgRNA count tables and determination of significant gene-level alterations in sgRNA representation were performed using MAGeCK (Li et al., 2014), which utilizes a negative binomial model to determine p-values of sgRNA changes and ranks sgRNAs by significance. Gene-level alterations and p-values were calculated from the ranked list of sgRNAs using the modified robust ranking aggregation (α-RRA) algorithm and FDR values calculated by the Benjamini–Hochberg procedure. Tier 1 screen analysis compared populations before and after C6 treatment. Tier 2 screen analysis compared DMSO-treated populations to C6- or C16-treated populations. Significantly enriched or depleted genes were those with FDR <0.05.

Opaque 384-well plates were seeded with 250 cells per well in 25 µL media supplemented with either 0.1% DMSO or a threefold dilution series of C6 or C16, all in technical quadruplicate wells. Cells were grown for 72 hr before equilibrating to RT and addition of 12.5 µL CellTiter-Glo Luminescent Cell Viability Assay reagent (Promega). At RT and protected from light, plates were rocked for 5 min, incubated for 20 min, and luminescence measured on a GloMax Explorer Multimode Microplate Reader (Promega). To calculate relative cell viability, mean fluorescence from quadruplicate treatment wells was divided by mean fluorescence from quadruplicate DMSO wells. Dose–response curves, GI₅₀ concentrations, and standard error values were calculated from at least three biological replicates with the R package drc (Ritz et al., 2015).

Opaque 384-well plates were seeded with 250 cells per well in 25 µL media supplemented with either 0.2% DMSO, a threefold dilution series of either C16 or compound 2, or a combination of threefold dilutions of both C16 and compound 2 covering a 7 × 7 dose matrix, all in quadruplicate wells. Compound 2 consisted of either nutlin-3a (Cayman Chemical Company), rapamycin (MedChem Express), pinometostat (Cayman Chemical Company), harmine (Sigma-Aldrich), mivebresib (Cayman Chemical Company), venetoclax (Cayman Chemical Company), etoposide (Cayman Chemical Company), olaparib (Cayman Chemical Company), VE-821 (Cayman Chemical Company), pemrametostat (Selleck Chemicals), or alvespimycin (Cayman Chemical Company). Following 72 hr, plates were equilibrated to room temperature and 12.5 µL CellTiter-Glo Cell Viability Assay (Promega) reagent added to each well. While protected from light, plates were rocked for 5 min, incubated for 20 min, and luminescence measured on a GloMax Explorer Multimode Microplate Reader (Promega).

Technical replicate wells were averaged and resulting means used to calculate relative cell viability by dividing drug treatment by DMSO treatment. Mean δ-scores and standard deviations were calculated from three biological replicates via SynergyFinder Plus (Zheng et al., 2022) using the ZIP model (Yadav et al., 2015). ZIP δ-scores represent the percent of growth inhibition beyond that expected if the agents do not potentiate one another. δ-scores greater than zero are synergistic, δ-scores of zero are additive, and δ-scores less than zero are antagonistic. Statistical significances of peak synergistic and antagonistic δ-scores were calculated by one-sample t-tests using the tsum.test function from the R package PASWR. Significant synergy and antagonism δ-scores were those with p<0.05.

In quadruplicate, MV4;11 cells were seeded at 2 × 10⁵ cells/mL and treated with 0.1% DMSO or 250 nM C16 for either 24 or 72 hr. Cells were collected by centrifugation and washed three times with ice-cold 1× PBS before lysis on ice in SDS Lysis Buffer (5% SDS; 50 mM ammonium bicarbonate). Chromatin was sheared by brief sonication at 25% on ice and insoluble material cleared by centrifugation. Soluble proteins were quantified by Pierce BCA Protein Assay (Thermo Scientific). Of note, protein was isolated from equivalent cell numbers at 24 hr as changes in proliferation are not observed until beyond 24 hr WINi treatment in MV4;11 cells. At 24 hr, DMSO-treated cultures yielded 344.75 ± 21.7 µg total soluble protein and C16-treated cultures yielded 366.50 ± 15.8 µg total soluble protein (mean ± SEM).

Protein samples for LC-MS/MS analyses were prepared by S-Trap (ProtiFi) digestion. Protein samples (50 µg) were reduced with DTT (MilliporeSigma) at a final concentration of 20 mM at 95°C for 10 min and alkylated with iodoacetamide (MilliporeSigma) at a final concentration of 40 mM at RT for 30 min in the dark. Aqueous phosphoric acid (Fisher Scientific) was added to the samples at a final concentration of 1.2% followed by 90% methanol containing 100 mM TEAB at 6.6 times the volume of the sample. The samples were loaded on the S-Trap micro columns and centrifuged at 4000 × g until all the volume was passed through the column. The columns were washed four times with 150 µL 90% methanol containing 100 mM TEAB, pH 7.1. Proteins were digested with trypsin gold (Promega) at 1:50 enzyme to protein ratio in 50 mM TEAB, pH 8.0, for 1 hr at 47°C. Peptides were eluted by serial addition of 40 µL each of 50 mM TEAB, 0.2% formic acid, and 35 µL of 0.2% formic acid in 50% acetonitrile. Eluted peptides were dried in a speed-vac concentrator, resuspended in aqueous 0.1% formic acid, and analyzed by LC-coupled tandem mass spectrometry (LC-MS/MS).

An analytical column (360 µm O.D. × 100 µm I.D.) was packed with 25 cm of C18 reverse-phase material (Jupiter, 3 µm beads, 300 Å; Phenomenex) directly into a laser-pulled emitter tip. Peptides were loaded on the reverse phase column using a Dionex Ultimate 3000 nanoLC and autosampler. The mobile phase solvents consisted of 0.1% formic acid, 99.9% water (solvent A) and 0.1% formic acid, 99.9% acetonitrile (solvent B). Peptides were gradient eluted at a flow rate of 350 nL/min using a 120 min gradient. The gradient consisted of the following: 1–100 min, 2–38% B; 100–108 min, 38–90% B; 108–110 min, 90% B; 110–111 min, 90–2% B; 111–120 min (column re-equilibration), 2% B. Upon gradient elution, peptides were analyzed using a data-dependent method on an Orbitrap Exploris 480 mass spectrometer (Thermo Scientific), equipped with a nanoelectrospray ionization source. The instrument method consisted of MS1 using an MS AGC target value of 3 × 10⁶, followed by up to 15 MS/MS scans of the most abundant ions detected in the preceding MS scan. The intensity threshold for triggering data-dependent scans was set to 1 × 10⁴, the MS2 AGC target was set to 1 × 10⁵, dynamic exclusion was set to 20 s, and HCD collision energy was set to 30 nce.

For identification of peptides, LC-MS/MS data were searched with Maxquant, version 2.0.1.0 (Cox and Mann, 2008). MS/MS spectra were searched with the Andromeda search engine (Cox et al., 2011) against a human database created from the UniprotKB protein database (Bateman et al., 2021) and the default Maxquant contaminants. Default parameters were used for Maxquant, with the addition of selecting LFQ and match between runs as a global parameter. Maxquant parameters included first and main search mass tolerances of 20 ppm and 4.5 ppm, respectively. Variable modifications included methionine oxidation and N-terminal acetylation, and carbamidomethyl cysteine was selected as a fixed modification. A maximum of two missed cleavages was allowed. The false discovery rate (FDR) was set to 0.01 for peptide and protein identifications. Label-free quantitative (LFQ) analysis of identified proteins was performed with the MSstats R package (Choi et al., 2014), version 4.0.1, using default parameters, which include the following: equalize medians for the normalization method, log2 transformation, Tukey’s median polish as the summary method, and model-based imputation. Protein fold changes were considered as significant with adjusted p-values ≤0.05.

Cell pellets were suspended in TRIzol Reagent (Invitrogen), rotated for 15 min at RT, and insoluble cellular debris pelleted by centrifugation. The soluble fraction was mixed with equal volume of 100% ethanol and RNA was isolated using the Direct-zol RNA Miniprep Kit (Zymo Research) according to the manufacturer’s instructions, including on-column DNA digestion. Complementary DNA (cDNA) was synthesized in 20 μL cDNA reactions containing 1 μg RNA, Random Hexamers (Invitrogen), and SuperScript III Reverse Transcriptase (Invitrogen) as per the manufacturer’s instructions. Final cDNA products were diluted fivefold with nuclease-free water before use in semi-quantitative RT-PCR or quantitative RT-PCR.

PCR reactions were performed with 2 μL cDNA template using primers amplifying splicing variants of RPL22L1 (RPL22L1_RTPCR_F: ATGGCGCCGCAGAAAGAC; RPL22L1_RTPCR_R: CTAGTCCTCCGACTCTGATT) or MDM4 (MDM4_RTPCR_F: GAAAGACCCAAGCCCTCTCT; MDM4_RT_PCR_F: GCAGTGTGGGGATATCGTCT), or within GAPDH (GAPDH_RTPCR_F: TCACCAGGGCTGCTTTTAAC; GAPDH_RTPCR_R: ATCGCCCCACTTGATTTTGG) using Taq DNA Polymerase (New England BioLabs) with primer-specific annealing temperatures and cycle numbers (RPL22L1: 50°C, 30 cycles; MDM4: 54°C, 33 cycles; GAPDH: 52°C, 27 cycles). PCR products were electrophoretically separated on 2% agarose gels in TBE buffer, gels incubated 30 min in TBE buffer containing 1× SYBR Safe DNA Stain (Invitrogen) with agitation, and imaged on a ChemiDoc Imaging System (Bio-Rad).

qPCR reactions containing 1× KAPA SYBR Fast qPCR Master Mix (Roche), transcript-specific primers, and 2 μL cDNA template were performed in technical duplicate wells on a C1000 Touch Thermal Cycler (Bio-Rad) with a CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Primer pairs targeted total RPL22L1 (RPL22L1ab_qPCR_F: tcgagtggttgcatctgaca; RPL22L1ab_qPCR_R: tcctccgactctgattcatct), RPL22L1a (RPL22L1a_qPCR_F: cgccgcagaaagacaggaa; RPL22L1a_qPCR_R: ctcccgtagaaattgctcaaaat), RPL22L1b (RPL22L1b_qPCR_F: cgcagaaagacaggaagcc; RPL22L1b_qPCR_R: tgcaaaactagggaagagaacc), MDM4 exon 5–6 junction (MDM4_Jnct_5_6_qPCR_F: AGAATCTTGTCACTTTAGCCACT; MDM4_Jnct_5_6_qPCR_R: CGAGAGTCTGAGCAGCATCT), MDM4 exon 6–7 junction (MDM4_Jnct_6_7_qPCR_F: TCAAGACCAACTGAAGCAAAGT; MDM4_Jnct_6_7_qPCR_R: TAGGCAGTGTGGGGATATCG), MDM4 exon 4 (MDM4_Ex_4_qPCR_F: AGCAACTTTATGATCAGCAGGAG; MDM4_Ex_4_qPCR_R: GACGTCCCAGTAGTTCTCCC), MDM4 exon 7 (MDM4_Ex_7_qPCR_F: AGAGGAAAGTTCCACTTCCAGA; MDM4_Ex_7_qPCR_R: ATGCTCTGAGGTAGGCAGTG), or GAPDH (GAPDH_qPCR_F: AAGGTGAAGGTCGGAGTCAAC; GAPDH_qPCR_R: GTTGAGGTCAATGAAGGGGTC). Ct values for each well were determined by the Bio-Rad CFX Manager Software v3.1 using the regression model, and mean Ct values from technical replicate wells used for subsequent calculations. Relative isoform levels were calculated via the 2^(-ΔΔCt) algorithm by internally normalizing isoform-specific Ct values to GAPDH Ct values, then relative to DMSO-treatment.

The statistical test, threshold for statistical significance, and n for each experiment, representing biological replicates, can be found in the figure legends.

Images of C6 and C16 bound to the WIN-Site of WDR5 and overlaid structures in WDR5-binding conformations were generated with PyMOL using published X-ray crystal structures (C6, PDB: 6E23 [Aho et al., 2019a]; C16, PDB: 6UCS [Tian et al., 2020]).

Gene set enrichment analyses (GSEA) and over-representation analyses (ORA) were performed with the R package fgsea (Korotkevich et al., 2021) using the Molecular Signatures Database v7.4 (Subramanian et al., 2005; Liberzon et al., 2011; Liberzon et al., 2015). Significantly enriched or depleted gene sets were those with FDR < 0.05.