Abstract
CRISPR–Cas9-based genetic screens have successfully identified cell type–dependent liabilities in cancer, including acute myeloid leukemia (AML), a devastating hematologic malignancy with poor overall survival. Because most of these screens have been performed in vitro using established cell lines, evaluating the physiologic relevance of these targets is critical. We have established a CRISPR screening approach using orthotopic xenograft models to validate and prioritize AML-enriched dependencies in vivo, including in CRISPR-competent AML patient-derived xenograft (PDX) models tractable for genome editing. Our integrated pipeline has revealed several targets with translational value, including SLC5A3 as a metabolic vulnerability for AML addicted to exogenous myo-inositol and MARCH5 as a critical guardian to prevent apoptosis in AML. MARCH5 repression enhanced the efficacy of BCL2 inhibitors such as venetoclax, further highlighting the clinical potential of targeting MARCH5 in AML. Our study provides a valuable strategy for discovery and prioritization of new candidate AML therapeutic targets.
There is an unmet need to improve the clinical outcome of AML. We developed an integrated in vivo screening approach to prioritize and validate AML dependencies with high translational potential. We identified SLC5A3 as a metabolic vulnerability and MARCH5 as a critical apoptosis regulator in AML, both of which represent novel therapeutic opportunities.
This article is highlighted in the In This Issue feature, p. 275
Introduction
Acute myeloid leukemia (AML) is a heterogeneous hematologic malignancy characterized by the accumulation of abnormal myeloblasts. Despite the efficacy of chemotherapy and stem cell transplantation for some patients, cure rates for AML remain between 35% and 40% overall and less than 15% for older adults (1). Continued efforts are needed to identify new therapeutic strategies for these patients.
The successful adaptation of CRISPR–Cas9 approaches for genetic screens has become a powerful tool for the unbiased discovery of essential genes in mammalian cells (2, 3). First-generation, large-scale functional genomic screens to identify the critical genes involved in cancer cell maintenance have been completed, such as the Broad Institute's and Sanger Center's Cancer Dependency Maps (DepMap; https://depmap.org/; refs. 4, 5). These efforts have revealed hundreds of potential genetic vulnerabilities in AML cells in vitro. How to distinguish candidates with the highest translational potential, however, remains a challenge. Therefore, a secondary functional validation approach is necessary to prioritize those gene targets for therapeutic targeting to guide the development of new antileukemia treatments.
These large-scale genetic screens were primarily performed in vitro. Thus, one important consideration in prioritizing candidate genes is to evaluate their essentiality in a proper in vivo microenvironment because the niche may influence the physiologic behavior of cancer cells. Human AML orthotopic disease modeling is highly physiologically relevant, as AML cells will engraft in the bone marrow microenvironment in the mouse. In vivo CRISPR screening has been performed to identify tumor growth modulators in several genetically engineered mouse models of hematologic malignancies (6–8). However, the feasibility of such an application in human AML orthotopic xenograft models has not been demonstrated. Therefore, we optimized a protocol for CRISPR screening in orthotopic xenograft models to enable the systematic evaluation of the physiologic relevance of top AML dependencies emerging from genome-scale CRISPR–Cas9 in vitro screens.
Established AML cell lines are amenable to genome-scale screens; however, they cannot fully recapitulate all pathophysiologic aspects of the disease. Target validation directly in primary patient samples is desirable yet not readily accessible. Instead, the AML patient-derived xenograft (PDX) has emerged as a valuable preclinical model that largely reflects the molecular and phenotypic characteristics of the primary disease (9, 10). To interrogate the translational relevance of the targets identified from established cell lines, we developed AML PDX models amenable to genome editing. By combining in vivo screening and CRISPR-competent PDX models, we devised an integrated pipeline to prioritize AML dependencies and investigated the top novel targets emerging from this approach.
Results
In Vivo CRISPR Screens Using Xenograft Models of Human AML
To identify AML-enriched dependency genes, we explored the DepMap Avana CRISPR–Cas9 screen dataset and selected the genes that AML cells are more dependent on for growth compared with the other cancer types included in the screen. This gene set was intersected with additional AML in vitro screen datasets, including the combined Broad Institute and Novartis short hairpin RNA (shRNA) screens and a focused in vitro CRISPR screen in AML cell lines (11, 12). These 200 top-ranked AML-enriched gene dependencies were involved in various biological pathways, such as chromatin and transcriptional regulation, metabolism, and mitochondria organization (Supplementary Fig. S1A and S1B). To distinguish the on-target from off-target antiproliferative effects caused by CRISPR-mediated DNA cutting in amplified regions (13, 14), we designed three targeting single-guide RNAs (sgRNA) and three intronic control sgRNAs for each gene. With 120 additional negative control sgRNAs, a focused library with 1,320 sgRNAs was constructed (Fig. 1A).
To better evaluate the therapeutic potential of candidate genes, we sought to investigate their in vivo essentiality not only in established AML cell lines but also in PDX models, argued to be the most faithful to primary human disease (11). CRISPR-mediated genetic studies in PDXs have been challenging due to the poor transduction efficiency and limited growth in vitro. To develop PDX models that are tractable for CRISPR editing, we screened a cohort of PDX samples and identified those transducible and suitable for short-term in vitro culture (Supplementary Table S1). These PDX cells were transduced with lentivirus coexpressing Cas9 and a fluorescent protein (GFP or mCherry). Cas9-expressing PDX cells were purified based on fluorescence and expanded via serial transplantation into immunodeficient NSGS (NOD scid gamma SGM3) mice (Supplementary Fig. S2A and S2B). Cas9 activity was assessed using a fluorescent protein (mAmetrine)–linked sgRNA targeting CD33. More than 80% of PDX cells receiving the sgRNA became CD33 negative, indicating that a high Cas9 activity can be achieved in these models (Supplementary Fig. S2C).
Next, to ensure sufficient library representation in vivo, we optimized the screening conditions through barcoding experiments using a library of 3,152 barcodes. Five to 10 million barcoded MV4-11 cells were injected via tail vein into NSGS mice. Sublethal irradiation was necessary for improved barcode representation in bone marrow and reduced mouse-to-mouse variation (Supplementary Fig. S2D and S2E). Although the barcode distribution was skewed in individual mice, even with irradiation, a complete and balanced library representation could be recovered by combining readouts from multiple mice (Supplementary Fig. S2F and S2G).
With all conditions optimized, we then performed parallel in vivo and in vitro screens using Cas9-expressing MV4-11 and U937 cell lines, as well as PDX16-01(CALM–AF10 fusion, NF1, PHF6, and TP53 mutations; Fig. 1B). AML cells were transduced with the screen library in duplicate and selected for 2 days with puromycin when an aliquot of cells was collected as the input reference. We then injected 10 million cells per mouse by tail vein into four to five irradiated mice per replicate and in parallel cultured an aliquot of cells from each replicate in vitro. In vitro cultures were harvested 2 (for MV4-11) or 3 (for U937 and PDX16-01) weeks later, and the bone marrow and spleens were collected when the mice displayed signs of overt disease, such as hindlimb paralysis and dyspnea, with high leukemic engraftment (Supplementary Fig. S3A and S3B). There was strong replicate reproducibility for both the in vitro and in vivo results, although the data generated from U937 displayed a smaller dynamic range likely due to weaker Cas9 activity (Supplementary Fig. S3C). The average relative abundance of each sgRNA in the output compared with the input samples was determined. Both the abundance and depletion of individual sgRNA in the bone marrow versus the spleen were strongly correlated (Fig. 1C; Supplementary Fig. S3D); therefore, we focused on the bone marrow data for the downstream analysis.
We calculated a normalized depletion score for each sgRNA (see Supplementary Methods and Supplementary Table S2). The median value of each set of three sgRNAs was used to represent the score of the corresponding gene. Using the intronic guide population as a null distribution, we defined hits for each model (Supplementary Table S3). In vitro and in vivo hits were generally well correlated. However, a modest number of targets did not score well in vivo, with a few targets displaying in vitro versus in vivo discrepancies in multiple models (Supplementary Fig. S3E and S3F). These results underscore the importance of an in vivo validation strategy for refining the hits emerging from a primary in vitro screen. Notably, many genes were confirmed as hits in PDX16-01 in vivo and overlapped with those validating in the MV4-11 and U937 models (Fig. 1D), lending support to the relevance of using AML cell lines for dependency identification. Gene Ontology analysis showed an enrichment of metabolism- and mitochondria-associated pathways in all three models (Supplementary Fig. S4A and S4B), consistent with recent findings that AML cells rely on unique metabolic and mitochondrial properties for survival (15, 16). In addition, several hematopoietic lineage–related transcription factors appeared to be strong in vivo dependencies (Supplementary Fig. S4C), corroborating recent studies targeting transcriptional vulnerabilities in AML (17, 18). Further supporting the validity of our screen, the transcription factors KMT2A (also called MLL, mixed lineage leukemia gene) and ZFP64 specifically scored in MV4-11, which is driven by an MLL fusion oncogene. ZFP64 is required to sustain expression of this fusion (Supplementary Fig. S4C; ref. 19). Altogether, our screen provided an informative list of AML targets with high physiologic relevance (Supplementary Table S3).
Next, we focused on targets that previously had not been described as AML dependencies and ranked highly as in vivo hits in all three models in the bone marrow: the sodium/myo-inositol cotransporter SLC5A3 and the mitochondria-localized RING-type ubiquitin E3 ligase MARCH5 (Fig. 1E; Supplementary Fig. S5A; refs. 20, 21). SLC5A3 and MARCH5 also displayed strong depletion scores in the data from spleen in the MV4-11 and PDX16-01 models and a modest depletion score in the U937 model (Supplementary Fig. S5B). We re-mined the latest edition of the DepMap CRISPR screening datasets, which continue to be expanded, and confirmed that SLC5A3 and MARCH5 were indeed strong dependencies in most AML cell lines, displaying a more essential role in AML compared with other cancer types (Supplementary Fig. S5C; Supplementary Table S4). We therefore selected these two targets for further validation as novel therapeutic opportunities for AML.
SLC5A3 Is Required for Sustaining AML Growth
SLC5A3 belongs to the solute carrier family, and among all five solute carrier family members included in our screen library, SLC5A3 was the top scoring (Supplementary Fig. S6A). We validated this dependency in several AML cells lines and PDX models using two independent SLC5A3-targeting sgRNAs. SLC5A3 depletion suppressed the growth of AML cells as demonstrated by an in vitro competition assay (Fig. 2A and B). As a validated SLC5A3 antibody was unavailable, we confirmed the genomic editing of the SLC5A3 locus by Sanger sequencing and the Inference of CRISPR Edits (ICE) analysis, with a high editing efficiency achieved (Supplementary Fig. S6B–S6D). The on-target effect was further supported by using a CRISPR-resistant SLC5A3 cDNA to rescue the growth defect (Fig. 2C). The cellular alterations after SLC5A3 deletion were examined. Interestingly, SLC5A3 inhibition disturbed cell-cycle distribution in a cell context–dependent manner, including reduced S phase and increased sub-G1, G1/G0, or G2 phase (Supplementary Fig. S6E). SLC5A3 depletion did not promote obvious differentiation, as assessed by CD11b staining, except for PDX17-14 (MLL–AF10 fusion), which displayed an upregulation of CD11b expression associated with morphology changes consistent with differentiation (Supplementary Fig. S6F and S6G). By contrast, an upregulation of Annexin V and cleaved caspase 3 levels was consistently observed in all AML models tested, indicating that apoptotic cell death is a common consequence of SLC5A3 disruption (Fig. 2D and E).
We next asked whether disruption of SLC5A3 posttransplantation can repress AML progression in vivo. PDX16-01 cells were transduced with a doxycycline-inducible CRISPR vector coexpressing GFP (22). Purified GFP+ cells were transplanted into NSGS mice, and the nontargeting control or SLC5A3-targeting sgRNA was induced 1 week later by serving a doxycycline-containing diet (Fig. 2F). Compared with the control group, mice receiving SLC5A3-knockout cells displayed significantly reduced leukemic burden as evaluated by bone marrow aspiration, as well as prolonged survival (Fig. 2G and H). Notably, a subset of the leukemia cells from the SLC5A3-knockout group was GFP negative at the time of disease progression, which was not observed in the control group (Supplementary Fig. S6H and S6I). In accordance, ICE analysis showed that only a minor population of these cells retained SLC5A3 locus editing (Supplementary Fig. S6J). Therefore, it is likely that AML cells escaping from SLC5A3 deletion outgrew and contributed to leukemia progression, emphasizing the essential role of SLC5A3 for AML progression in vivo.
SLC5A3 Transports Myo-inositol to Support AML Cell Proliferation
Myo-inositol and its derivatives are involved in several cellular processes. Because SLC5A3 is one of the major myo-inositol transporters, we investigated whether the growth defect caused by SLC5A3 inactivation resulted from the myo-inositol deficiency. A previous study in Slc5a3-knockout mice has shown that myo-inositol provided at supraphysiologic concentrations can bypass SLC5A3 to enter the cells, possibly via other low-affinity transporters (23). Standard culture medium contains around 0.3 mmol/L myo-inositol, similar to the level detected in human serum (24). Strikingly, with the addition of supplementary myo-inositol in the culture medium, the proliferation of SLC5A3-knockout cells was completely rescued (Fig. 3A). Of note, extra myo-inositol did not promote the growth of parental AML cells (Supplementary Fig. S7A). In accordance, depletion of the basal myo-inositol from the culture medium largely impeded the growth of parental AML cells, causing similar phenotypes as SLC5A3 deletion, with cell context–dependent alterations of cell cycle and induction of apoptotic cell death (Fig. 3B; Supplementary Fig. S7B and S7C). Together, these data reveal that myo-inositol is a critical metabolite for AML.
Because a subset of AML cell lines was not dependent on SLC5A3 based on the DepMap dataset (Supplementary Fig. S5C), we explored the potential biomarkers associated with SLC5A3 essentiality in AML. SLC5A3 was ubiquitously expressed in AML cell lines, and its expression was not correlated with its dependency. Intriguingly, however, the low expression of inositol-3-phosphate synthase 1 (ISYNA1) predicted a strong SLC5A3 dependency in AML cell lines (Fig. 3C). In addition to importing myo-inositol from the extracellular fluid, cells can also synthesize myo-inositol de novo from glucose 6-phosphate, and ISYNA1 encodes the rate-limiting enzyme in this myo-inositol biosynthesis pathway (25). Thus, we postulated that SLC5A3 becomes essential in AML cells with insufficient myo-inositol biosynthesis capacity. We confirmed the low expression of the ISYNA1 protein in AML cells sensitive to SLC5A3 deletion, and importantly, overexpression of ISYNA1 can completely relieve the SLC5A3 dependency (Fig. 3D–F). Moreover, knockout of ISYNA1 in the ISYNA1-high cell line M07e exacerbated the growth defect associated with SLC5A3 depletion (Fig. 3G and H). Altogether, these results strongly demonstrate that SLC5A3 is required for maintaining sufficient myo-inositol levels to support AML proliferation.
MARCH5 Loss Represses AML Cell Growth In Vitro and In Vivo
We next sought to validate the dependency of AML cells on MARCH5. Inactivating MARCH5 via either doxycycline-inducible CRISPR or shRNA systems induced a severe growth defect in various AML cell lines and PDX models (Fig. 4A and B; Supplementary Fig. S8A and S8B). The growth defect could be reversed by a CRISPR-resistant cDNA encoding wild-type MARCH5, proving the on-target effect. By contrast, MARCH5 mutations (H43W and C68S) that disrupt its RING domain and thus ubiquitinase function ablated the rescuing ability (26, 27), indicating the requirement for the catalytic function of MARCH5 in AML (Fig. 4C; Supplementary Fig. S8C and S8D). In addition, for MARCH5 validation, we deployed a dTAG system, which uses a hetero-bifunctional small molecule that binds the FKBP12F36V-fused target protein (i.e., MARCH5) and an E3 ligase complex (i.e., VHL), bringing the two in close proximity and leading to the ubiquitination and proteasome-mediated degradation of the target protein (Supplementary Fig. S8E; ref. 28). We were able to establish the MARCH5 dTAG degradation system in both AML cell line and PDX models in which we deleted endogenous MARCH5 by CRISPR and expressed exogenous FKBP12F36V-hemagglutinin (HA)–tagged MARCH5 protein at a physiologic level comparable to the endogenous one (Supplementary Fig. S8F). dTAG-MARCH5 cells displayed a similar basal proliferation rate and apoptosis compared with control cells expressing Cas9 only (Supplementary Fig. S8F–S8H). Similar to CRISPR deletion of MARCH5, MARCH5 degradation with the dTAG molecule dTAGV-1 markedly impaired cell growth (Fig. 4D and E).
MARCH5 dependency was further confirmed using an in vivo competition assay in a third PDX model (PDX68555 with an MLL–AF9 fusion and a FLT3 mutation). PDX cells expressing a GFP-linked MARCH5 sgRNA were depleted in NSGS mice, as evidenced by a dramatic reduction of the GFP+ fraction in engrafted cells. In contrast, the PDX cells expressing a nontargeting sgRNA were maintained (Fig. 4F–H). To confirm that the in vivo growth disadvantage of MARCH5-depleted cells is not caused by homing defects, we used MV4-11 cells expressing luciferase and doxycycline-inducible CRISPR directed against MARCH5. Doxycycline-mediated deletion of MARCH5 posttransplantation led to a marked attenuation of AML progression in NSGS mice as monitored by bioluminescence imaging, which translated to prolonged survival (Fig. 4I–K). We examined the MARCH5 sgRNA-expressing cells collected from the leukemic mice and found that the MARCH5 expression was partially restored as compared with the cells with MARCH5 knockout induced in vitro, supporting that loss of MARCH5 is incompatible with AML maintenance (Supplementary Fig. S8I). Collectively, these results demonstrate that targeting MARCH5 can suppress the progression of AML cells both in vitro and in vivo.
Differential MARCH5 Dependency in Healthy Human Hematopoietic Stem and Progenitor Cells Compared with AML Blasts
We next attempted to determine whether MARCH5 is required by healthy human CD34+ hematopoietic stem and progenitor cells (HSPC). Because of the low efficiency of lentiviral transduction of Cas9 into CD34+ HSPCs, we used the nucleofection of Cas9–sgRNA ribonucleoprotein complexes (RNP) to enable genome editing in these cells and evaluated their progenitor activity via colony formation assays (Fig. 5A). We first validated this approach using AML cells. RNPs containing MARCH5 sgRNA were introduced into NB4 and AML PDX cells; RNPs with sgRNA targeting a gene desert region in chromosome 2 (sgCHR2) or the common essential gene RPA3 were also included as negative and positive controls, respectively. A high genome editing efficiency was achieved for all RNPs (Supplementary Fig. S9A). As expected, MARCH5 deletion dramatically impaired the colony-forming capacity of AML cells, causing a more than 80% reduction of colony number and largely decreasing colony size (Fig. 5B; Supplementary Fig. S9B). We then evaluated CD34+ HSPCs derived from either bone marrow or umbilical cord blood and obtained comparable genome editing efficiency (Supplementary Fig. S9C). In contrast to AML cells, sgMARCH5-nucleofected CD34+ cells displayed only a 10% to 30% and 30% to 50% reduction of colony number in the erythroid and myeloid lineages, respectively (Fig. 5C). Colonies of the myeloid lineage were relatively more affected, with a more frequent appearance of smaller and/or less compacted colonies (Supplementary Fig. S9D). Nonetheless, these results highlight that healthy human HSPCs are less dependent on MARCH5 compared with AML cells, supporting the potential therapeutic window for MARCH5-targeted treatment.
MARCH5 Prevents Apoptosis in AML
MARCH5 inactivation in AML cells resulted in a slight reduction of S phase and increase in sub-G1 and G1/G0 phases (Supplementary Fig. S10A and S10B). Induction of the sub-G1 phase suggested the occurrence of cell death. Indeed, apoptosis was consistently observed and strongly induced in some models, as indicated by upregulated cleaved caspase 3 and Annexin V (Fig. 6A and B; Supplementary Fig. S10C). In PDX17-14, which was highly sensitive to MARCH5 inhibition, 2-hour treatment with dTAGV-1 was sufficient to prime cells for apoptosis, as indicated by BH3 profiling (Supplementary Fig. S10D; ref. 29). Importantly, knockout of the mitochondrial apoptosis effectors BAX or BAK1 reversed the apoptosis induction and growth defect of MARCH5-null cells (Fig. 6C; Supplementary Fig. S10E–S10G). AML cell lines displayed differential reliance on BAX and BAK1 for the execution of MARCH5 depletion–mediated apoptosis, and in some models, such as NB4 and PDX17-14, double knockout of BAX and BAK1 was required to rescue MARCH5 inactivation (Fig. 6D; Supplementary Fig. S10H and S10I). These findings demonstrate that apoptosis induction is an essential cellular mechanism accounting for the inhibitory effect of MARCH5 depletion in AML.
The activation of the mitochondrial apoptotic pathway is determined by the counterbalance between proapoptotic and antiapoptotic BCL2 family proteins (30). Strikingly, multiple genome-wide screens revealed that the dependency scores of MARCH5 and MCL1, but not other antiapoptotic BCL2 family genes, were significantly correlated across the AML cell lines as well as other cancer models (Fig. 6E and F; Supplementary Fig. S11A), suggesting a functional connection between MARCH5 and MCL1. This compelled us to test whether overexpression of antiapoptotic BCL2 family proteins can reverse the effect of MARCH5 inhibition. Overexpression of MCL1 did not robustly rescue MARCH5 deletion, particularly in the models most sensitive to loss of MARCH5, but overexpression of BCL2 or BCLXL invariably rescued the growth impairment caused by MARCH5 depletion in AML (Fig. 6G and H; Supplementary Fig. S11B). This inconsistent rescuing ability of overexpressed MCL1 is in accordance with altered MCL1 function upon loss of MARCH5.
We then investigated whether MARCH5 depletion can modulate the expression of BCL2 family members. Although most BCL2 proteins remained unaltered, an upregulation of MCL1 and NOXA, a BH3-only proapoptotic protein known to interact with and inhibit MCL1, was observed, which is consistent with the previous findings that MARCH5 was responsible for degrading the MCL1–NOXA complex (Fig. 6I; Supplementary Fig. S10G; refs. 31, 32). Although total levels of MCL1 increased, its localization primarily to the mitochondria did not change (Supplementary Fig. S11C). Recent studies suggest that increased NOXA mediates several downstream events of MARCH5 inhibition, such as sensitizing cells to stress stimuli (32–34). To investigate whether increased NOXA levels also account for the apoptosis induction in AML cells, NOXA-knockout cells were generated. Strikingly, MARCH5 depletion caused defective growth and induced cell death in NOXA-null cells to the same degree as in control cells (Fig. 6J and K). Given the possible redundant roles among BH3 protein members, we also examined BIM-knockout cells, as well as BIM/NOXA double-knockout cells. However, neither of these displayed improved resistance to MARCH5 inhibition compared with control cells (Supplementary Fig. S11D and S11E). In fact, knockout of BID or BIK, two other BH3 members capable of neutralizing MCL1, was not able to rescue MARCH5 knockout (Supplementary Fig. S11F and S11G). Together, these results show that the diminished MCL1 antiapoptotic activity upon MARCH5 depletion is not caused by a loss of MCL1 expression or by enhanced activity of other proapoptotic BCL2 members but rather likely the alteration of other downstream mediators of MARCH5.
MARCH5 Depletion Sensitizes AML Cells to Venetoclax
Apoptosis sensitivity can determine the response to various therapies in cancer (29). Because loss of MARCH5 primed AML cells for apoptosis, we investigated whether MARCH5 inhibition can sensitize AML cells to anticancer therapies. Control and MARCH5-knockdown OCI-AML2 cells were subjected to a chemical screen across a library of 3,247 anticancer compounds. Fifty-eight compounds displayed an enhanced inhibitory effect on MARCH5-knockdown cells compared with control cells. Notably, the BH3 mimetics, a class of small molecules that mimic BH3 proteins to bind and inhibit antiapoptotic BCL2 proteins, were enriched (Fig. 7A–C; Supplementary Fig. S12A). These data prompted us to examine whether MARCH5 inactivation in AML cells enhances their sensitivity to venetoclax, a BH3 mimetic that specifically blocks BCL2 and is FDA approved for the treatment of older adults with AML in combination with a hypomethylating agent (35). MARCH5 deletion through inducible CRISPR indeed sensitized AML cells to venetoclax (Fig. 7D). Similarly, dTAGV-1 treatment elicited a dramatic venetoclax sensitizing effect in dTAG-MARCH5 cells but not in control cells (Fig. 7E; Supplementary Fig. S12B). Upregulation of MCL1 or BCLXL activity has been identified as a mechanism causing venetoclax resistance (36). Consistent with the loss of MARCH5 impairing MCL1 function, while MARCH5 depletion did not alleviate venetoclax resistance caused by BCLXL overexpression, it can diminish MCL1-induced venetoclax resistance (Fig. 7F and G; Supplementary Fig. S12C), even in the AML models whose growth impairment with MARCH5 knockout is rescued by MCL1 (Fig. 6H; MV4-11 and PDX16-01), highlighting a synergistic cooperation between MARCH5 ablation and venetoclax. NOXA levels are another determinant of venetoclax response consistent with its antagonistic function on MCL1 (37). Accordingly, MARCH5 inhibition also attenuated the venetoclax resistance resulting from NOXA knockout (Supplementary Fig. S12D). Overall, the results further support the notion that MARCH5 can regulate MCL1 function through a NOXA-independent manner in AML cells.
The cooperativity between MARCH5 inhibition and venetoclax was also demonstrated in vivo using the xenograft model of luciferase-expressing MV4-11 cells. We used a venetoclax dose that is below the maximum tolerated dose in mice, and although this venetoclax regimen showed minimal efficacy in control cells, it enhanced the antileukemic activity of MARCH5 depletion in vivo (Fig. 7H and I). There are currently no small-molecule inhibitors directed against MARCH5. We thus deployed the dTAG system as an approximation of pharmacologic inhibition of MARCH5 and investigated its impact in combination with venetoclax in an orthotopic PDX model of AML. To this end, we first characterized the dose and schedule of dTAGV-1 to achieve maximum in vivo degradation. NSGS mice injected with PDX17-14 dTAG-MARCH5 cells were treated with a series of doses of dTAGV-1 via intravenous administration once daily or intraperitoneal administration twice daily for 4 days, and then FKBP-HA-MARCH5 levels were evaluated by intracellular staining with antibodies against HA-tag (Fig. 7J). Concordant results were obtained with two independent HA antibodies, showing that dTAGV-1 treatment led to dose-dependent MARCH5 degradation in vivo. Maximal achievable degradation was observed at 40 mg/kg—the highest dose that can be achieved given the solubility limitation of the compound (Supplementary Fig. S12E and S12F). Intraperitoneal administration twice daily at 40 mg/kg resulted in similar degradation efficacy compared with intravenous administration once daily (Supplementary Fig. S12G). The intraperitoneal regimen was subsequently chosen for efficacy studies given its improved tolerability to daily intravenous administrations. An independent cohort of mice receiving PDX17-14 dTAG-MARCH5 cells was assigned to four groups and treated with vehicle, venetoclax, dTAGV-1, or both venetoclax and dTAGV-1. The leukemic burden in bone marrow was evaluated after 2 weeks of treatment (Fig. 7J). Venetoclax did not elicit a clear antileukemic effect. Although dTAGV-1 induced a modest reduction in leukemic burden, the combination treatment markedly decreased leukemic burden (Fig. 7K). All told, these results highlight the combinational targeting of MARCH5 and BCL2 as a potential therapeutic approach for AML.
Discussion
Large-scale genome-wide CRISPR screening has provided numerous candidate AML targets, yet uncertainty remains about their physiologic relevance and translational potential. In vivo CRISPR screens offer a strategy for identifying leukemia dependencies within the context of the physiologic microenvironment. Our study thus aimed to refine the candidate list by providing in vivo functional references to better inform future studies. We have developed an in vivo CRISPR screening pipeline in orthotopic xenograft models of human AML and defined experimental conditions necessary for an optimal in vivo screen. One major consideration is the requirement for sufficient in vivo library representation to avoid false-positive hits. Because AML is established by leukemia-initiating cells, only a subset of the AML cell population retains this leukemia-initiating activity in mouse recipients. As evidenced by our barcoding experiments, random sets of barcodes prevailed in different individual recipients, reflecting the selective leukemia-initiating activity and clonal expansion. Therefore, it is critical to use irradiation to maximize the engraftment capacity of AML cells and include multiple mice to achieve complete in vivo library representation. The number of mouse recipients must be adjusted according to the library size and the leukemia-initiating capacity of the model.
PDXs provide useful preclinical models for therapy evaluation. These models largely retain the histologic and genetic characteristics of the primary disease and have also been shown to be predictive of clinical outcome (9). Although PDX models have been widely used for drug evaluation, we established CRISPR-competent PDX models to determine whether the dependency landscape obtained from the cell lines is reflected in PDXs. Quite strikingly, dependencies identified across AML cell line models were strongly recapitulated in the PDX model that we evaluated. We also observed a strong correlation between the in vitro and in vivo dependencies, likely related to the reduced off-target effects of CRISPR compared with other approaches, such as RNAi, and the robust library representation. However, a few genes reproducibly appeared as in vitro–only dependencies, demonstrating that the microenvironment can influence the essentiality of a target. Although our current study was not positioned to discover in vivo–specific dependencies, our pipeline is adaptable for de novo dependency discovery in human AML when coupled with other focused libraries. In general, PDX models with reproducible engrafting capacity and relatively short disease latency are preferred for in vivo CRISPR screening. Reports from other groups and our own experience indicate that AML PDX models can be more readily established from disease subtypes with intermediate and poor prognosis (38). Although PDX models with MLL fusions and a CALM–AF10 fusion were used in this study, it should be feasible to establish CRISPR-competent PDX models with different genetic backgrounds and thus extend our dependency discovery capacity to other high-risk subtypes of AML.
Metabolic reprogramming contributes to tumor development and sustains cancer cell proliferation. Like other cancers, AML has altered metabolic features and is addicted to certain metabolites and metabolic pathways for survival, providing new possibilities for AML treatment. For instance, the one-carbon folate pathway is reported to be critical for AML growth, and AML stem cells selectively depend on amino acid metabolism to fuel oxidative phosphorylation. Accordingly, blocking these pathways has exhibited antileukemic activity in preclinical studies (39, 40). Here, we revealed myo-inositol as a metabolic dependency in AML. Our data suggest that AML cells rely on either SLC5A3-mediated extracellular transportation or ISYNA1-mediated intracellular de novo synthesis for their myo-inositol supply, and AML cells with low ISYNA1 levels depend on SLC5A3 exclusively for fueling myo-inositol metabolism. Interestingly, the SLC5A3–ISYNA1 correlation is less evident in other cancer types; most ISYNA1 low–expressing cancer cell lines are not dependent on SLC5A3. The underlying biology rendering SLC5A3 as a selective dependency in ISYNA1-low AML but not in other ISYNA1-low cancers is not clear. It is possible that the myo-inositol addiction is endowed by the cell lineage in AML or that alternative myo-inositol transporters are used in ISYNA1 low–expressing cancer cells that are not SLC5A3 dependent. Although SLC5A3-specific inhibitors have not been reported, small-molecule inhibitors against SLC5A1 and SLC5A2—two solute carrier family 5 members with a similar structure to SLC5A3—have been developed (41, 42). Therefore, SLC5A3 is potentially actionable, and our study supports the development of chemical probes against this strong AML dependency.
Mitochondrial physiology constitutes another important axis in AML disease maintenance and drug response. It is therefore particularly relevant that MARCH5, a target located in the outer mitochondrial membrane, scored as a top hit in our screens. MARCH5 has been reported to serve multiple context-dependent functions in cells, including regulation of mitochondrial dynamics through its modulation of fission or fusion effector proteins, such as DRP1 and MFN2 (21); protection against stress stimuli through facilitation of mitochondria homeostasis or regulation of stress-responding proteins such as inositol-requiring kinase 1 and NOXA (27, 32, 33, 43, 44); and prevention of persistent innate immune response through the reduction of mitochondrial antiviral signaling aggregates (45). In contrast, in the case of AML, we have revealed that MARCH5 is essential for cell survival under physiologic conditions; inhibiting MARCH5 by itself is sufficient to activate the canonical mitochondrial apoptosis pathway in a BAX/BAK1-dependent manner. The dependency correlation analysis and our functional studies strongly suggest that MARCH5 regulates apoptosis by interfering with MCL1 function. Although the mechanism by which MARCH5 loss represses MCL1 activity is not fully delineated, we have ruled out several possibilities, including reduction of MCL1 expression, relocation of MCL1 from the mitochondria, or induction of NOXA. Prior studies of a MARCH5–NOXA connection have focused on cellular response to stress inducers where NOXA loss is reported to attenuate the impact of MARCH5 repression. In contrast, our study evaluated the impact of MARCH5 directly in apoptosis regulation. In this latter context, NOXA knockout is insufficient to rescue MARCH5 repression. All told, our results are consistent with MARCH5 playing a regulatory role in MCL1 activity, possibly through posttranslational modification or through another mediator of MCL1 function—mechanisms that will be the study of future investigations.
Venetoclax has received FDA approval for the treatment of newly diagnosed older adult patients with AML in combination with hypomethylating agents. This promising combination has resulted in a remission rate of approximately 70%. However, intrinsic and acquired resistance still emerged in a significant percentage of patients, and the efficacy of this combination drops precipitously in patients with relapsed and refractory AML (35). Thus, additional combination approaches are desired to enhance the clinical benefits of venetoclax. Overcoming venetoclax resistance via targeting mitochondrial components is an emerging theme. Blockage of oxidative phosphorylation activity, inhibition of mitochondrial translation, and disruption of mitochondrial cristae structures can all result in venetoclax sensitization in AML cells (40, 46, 47). Consistent with previous reports showing that loss of MARCH5 sensitizes cells to BH3 mimetics (27, 33), our data have demonstrated that MARCH5 is another promising synergistic mitochondrial target for enhancing the efficacy of venetoclax in AML. Given that MARCH5 inhibition can reverse the venetoclax resistance caused by MCL1 overexpression and NOXA deletion, but not that caused by BCLXL overexpression, it is likely that the MCL1 repression also contributes to the venetoclax sensitization conferred by MARCH5 depletion. Collectively, MARCH5 is positioned as a strong AML dependency, as well as a synergistic target for anti-BCL2 therapy. With the success of targeting other E3 ligases (e.g., MDM2 and XIAP) by small molecules already in clinical testing in humans (48, 49), and given that the enzymatic activity of MARCH5 is critical, MARCH5 targeting holds promise for patients with AML and potentially the treatment of other malignancies.
Taken together, our in vivo screening approach, coupled with CRISPR-competent PDX models, constitutes a platform for prioritizing AML targets emerging from in vitro screens for therapy development, with SLC5A3 and MARCH5 nominated as two top targets for further consideration in AML. According to the DepMap dataset, strong MARCH5 dependency is observed in AML cell lines with various French–American–British (FAB) subtypes and genetic contexts, suggesting broad applicability. By contrast, it is intriguing that strong SLC5A3 dependency is enriched in AML cell lines of the FAB M5 subtype, which are also enriched for MLL fusions (Supplementary Table S4). Further investigation is required to confirm whether MLL fusions or the M5 subtype will be robust biomarkers of response to SLC5A3 repression in AML. Finally, we demonstrated the utility of dTAG-directed protein degradation for mimicking pharmacologic inhibition of the target, which is a valuable addition to AML target validation, especially for studying targets without a tool compound inhibitor available, such as SLC5A3 and MARCH5.
Methods
Cell Culture and PDX Samples
All commercially available cell lines were obtained from ATCC or DSMZ. AML cell lines (MV4-11, U937, MOLM14, NB4, and P31FUJ) were cultured in RPMI supplemented with 10% FBS and 1% penicillin–streptomycin (PS). HEK293T cells were cultured in DMEM with 10% FBS and 1% PS. During the preparation of this study, Mycoplasma negativity was confirmed using a LookOut Mycoplasma PCR Detection Kit (MP0035; Sigma-Aldrich) and the identity of all cell lines was validated through short tandem repeat profiling by the Molecular Diagnostics Laboratory at the Dana-Farber Cancer Institute (DFCI). Details on myo-inositol depletion culture can be found in the Supplementary Methods.
Primary patient samples were acquired following written informed consent in accordance with the Declaration of Helsinki, and PDXs were established under protocols approved by DFCI and Cincinnati Children's Hospital Medical Center (CCHMC) Institutional Review Boards. The detailed information on PDXs is provided in Supplementary Table S1. For short-term in vitro culture, PDX cells were maintained in Iscove's modified Dulbecco's medium containing 20% FBS and 1% PS and supplemented with 10 ng/mL human SCF, TPO, FLT3L, IL3, and IL6 (300-07, 300-18, 300-19, 200-03, and 200-06; PeproTech).
Frozen CD34+ HSPCs derived from bone marrow or umbilical cord blood were purchased from Lonza (#2M-101) and the Cell Processing Core at CCHMC, respectively. Details on nucleofection, genomic editing efficiency analysis, and colony formation assay can be found in the Supplementary Methods.
Lentivirus Production and Transduction
Details on lentiviral vectors used in this study can be found in the Supplementary Methods and Supplementary Table S5. Virus was produced using HEK293T cells transfected with lentiviral expression vectors, together with envelope VSVG and the gag-pol psPAX2 constructs. For transduction, AML cells were mixed with viral supernatant and 4 to 8 μg/mL polybrene. In some experiments, cells were centrifuged in viral supernatant at 1,000 × g for 1 hour at 33°C to enhance the transduction efficiency.
Competition Assay
Cells were transduced with lentivirus vectors coexpressing an sgRNA and a fluorescent protein, such as GFP, mCherry, or mAmetrine, at an efficiency of approximately 50%, or the transduced cells were mixed with nontransduced cells at approximately a 1:1 ratio. The cell growth was evaluated by the change in the fraction of cells expressing the fluorescent protein, which was monitored by flow cytometry. For the dTAG experiment, 500 nmol/L dTAGV-1 was added into culture and replenished every 3 to 4 days.
Venetoclax Treatment
Venetoclax was acquired from Selleck (S8048) or MedChemExpress (HY-15531). Cells were plated in 384-well plates at 1,000 to 1,500 cells per well and mixed with serially diluted concentrations of venetoclax or 0.1% DMSO as a control. The viability of cells was measured after a 3-day incubation using a CellTiter-Glo Luminescent Cell Viability Assay kit (Promega) following the manufacturer's protocol. Data were analyzed using GraphPad Prism software (GraphPad Software).
Xenograft Transplantation
All animal experiments were approved by the DFCI and CCHMC Institutional Animal Care and Use Committees with adherence to all appropriate guidelines. Transplantation was performed on 6- to 8-week-old NOD/SCID/IL2RG−/− immunodeficient mice with transgenic expression of human SCF, GM-CSF, and IL3 (NSGS; The Jackson Laboratory). When necessary, mice were conditioned with sublethal irradiation at least 6 hours before transplantation. For doxycycline-inducible sgMARCH5 experiments, 2 × 105 luciferase-expressing MV4-11 cells were transplanted into each mouse via tail vein. Doxycycline-containing food was initiated on day 4 posttransplantation. Leukemia progression was serially assessed using bioluminescence imaging. Mice were injected with 75 mg/kg i.p. d-Luciferin (Promega), anesthetized with 2% to 3% isoflurane, and imaged on an IVIS Spectrum (Caliper Life Sciences). A standardized region of interest encompassing the entire mouse was used to determine total bioluminescence flux. For the experiment evaluating venetoclax treatment in combination with MARCH5 deletion, doxycycline-containing food was initiated on day 7 posttransplantation; 10 days postinjection, mice were treated with vehicle (60% Phosal 50 propylene glycol, 30% polyethyleneglycol 400, and 10% ethanol) or venetoclax (75 mg/kg body weight) daily by oral gavage for 1 week.
To establish the Cas9-PDX models, 0.5 to 2 × 106 AML PDX cells were transplanted. Engrafted cells were collected when mice displayed overt disease, and cells were sorted and retransplanted when needed. For doxycycline-inducible sgSLC5A3 experiments, 5 × 105 PDX16-01 cells were transplanted into each mouse via tail vein, and doxycycline-containing food was initiated on day 7 posttransplantation. Leukemia progression was assessed by bone marrow aspiration 2 weeks afterward. For dTAG-MARCH5 experiments, 2 × 106 dTAG-MARCH5 PDX17-14 cells were transplanted into each mouse via tail vein. The treatment was initiated at 2 weeks (for the efficacy study) or 3 weeks (for the pharmacodynamic study) posttransplantation. dTAGV-1 stock was prepared in DMSO, then a 1:20 dilution was made in saline solution containing 5% solutol, and the working solution was administrated via intraperitoneal or intravenous injection as indicated. Venetoclax was prepared as above. Details on flow cytometry analysis of leukemic engraftment and FKBP-HA-MARCH5 protein levels can be found in the Supplementary Methods.
In Vivo and In Vitro Screens
Screens were performed in duplicate. AML cells stably expressing Cas9 were transduced with the screen library at an efficiency of 30% to 50%, so that most cells received only one sgRNA. Puromycin selection was initiated on the day after transduction for 2 days, and the selected cells were recovered in fresh medium for 1 more day. On day 4 posttransduction, 1.5 million selected cells—sufficient for a representation of more than 1,000 cells per guide—were collected as the input reference. The remaining cells were divided into the in vivo and in vitro screens. For the in vivo screen, 10 million cells were transplanted into each sublethally irradiated NSGS mouse via tail vein, with four to five mice per replicate. Cells from mouse bone marrow and spleen were collected at around week 3 posttransplantation, when overt disease was observed. For the in vitro screens, at least 1.5 million cells were maintained throughout the 14- to 21-day culture period and collected at the end of the screen. Genomic DNA was extracted from collected cell pellets using a Qiagen DNeasy Blood and Tissue Kit (#69506) or a NucleoSpin Blood L kit (Takara #740954.20). The sgRNA barcodes were PCR amplified and submitted for standard Illumina sequencing, as previously described (4). The barcoding experiments were performed in a similar manner except using a barcoding library. Details on library construction and screen data analysis can be found in the Supplementary Methods.
Authors' Disclosures
S. Lin reports grants from the Leukemia & Lymphoma Society and NCI during the conduct of the study. B.K.A. Seong reports other support from Department of Defense (CA181249) during the conduct of the study. N.V. Dharia reports grants from St. Baldrick's Foundation during the conduct of the study, as well as other support from Genentech, Inc., a member of the Roche Group, outside the submitted work. J.A. Ryan reports personal fees from Zentalis outside the submitted work, as well as a patent for BH3 profiling issued to the Dana-Farber Cancer Institute and a patent for BH3 profiling licensed to Zentalis. F. Piccioni is a current employee of Merck Research Laboratories. L.H. Lee reports grants from St. Baldrick's Foundation during the conduct of the study. A. Letai reports personal fees from Zentalis Pharmaceuticals, Dialectic Therapeutitics, and Flash Therapeutics outside the submitted work, as well as patents related to BH3 profiling pending, issued, licensed, and with royalties paid to the Dana-Farber Cancer Institute. Dr. Letai and his lab get a fraction of the licensing fees. K. Stegmaier reports grants from the NCI, St. Baldrick's Foundation, and Children's Leukemia Research Association and other support from Cuban's Curing Childhood Cancer (4C's) during the conduct of the study, as well as personal fees from AstraZeneca, Bristol Myers Squibb, and Kronos Bio, personal fees and other support from Auron Therapeutics, and grants from Novartis outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
S. Lin: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration. C. Larrue: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration. N.K. Scheidegger: Data curation, formal analysis, validation, investigation, writing–review and editing. B.K.A. Seong: Data curation, formal analysis, validation, investigation, methodology, writing–review and editing. N.V. Dharia: Software, formal analysis, investigation, visualization, methodology. M. Kuljanin: Data curation, formal analysis, investigation, visualization, methodology. C.S. Wechsler: Data curation, formal analysis, validation, investigation, visualization. G. Kugener: Software, investigation, methodology. A.L. Robichaud: Data curation, validation, investigation, methodology. A.S. Conway: Data curation, investigation, methodology. T. Mashaka: Data curation, formal analysis, investigation, visualization. S. Mouche: Data curation, validation, investigation. B. Adane: Data curation, validation, investigation, visualization. J.A. Ryan: Data curation, formal analysis, validation, investigation, visualization, methodology. J.D. Mancias: Resources, supervision, investigation, writing–review and editing. S.T. Younger: Software, formal analysis, investigation, visualization, methodology. F. Piccioni: Software, formal analysis, investigation, visualization, methodology, writing–review and editing. L.H. Lee: Conceptualization, resources, data curation, formal analysis, investigation, methodology, writing–review and editing. M. Wunderlich: Conceptualization, resources, data curation, formal analysis, investigation, methodology, writing–review and editing. A. Letai: Resources, supervision, methodology. J. Tamburini: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration. K. Stegmaier: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration.
Acknowledgments
This work was supported by NCI R35 CA210030 (K. Stegmaier), P50 CA206963 (K. Stegmaier), R50 CA211404 (M. Wunderlich), and K99 CA263161 (S. Lin); the Children's Leukemia Research Association (K. Stegmaier); a St. Baldrick's Foundation Robert J. Arceci Innovation Award (K. Stegmaier); and the 4C's (K. Stegmaier). S. Lin is a Fellow of the Leukemia & Lymphoma Society. N.K. Scheidegger is supported by the Swiss Cancer League. B.K.A. Seong is supported by a Department of Defense PRCRP Horizon Award (CA181249). N.V. Dharia was supported by a St. Baldrick's Foundation Fellowship. L.H. Lee is a St. Baldrick's Foundation Scholar. The Flow Cytometry Core at the Cincinnati Children's Medical Center is supported by NIH S10OD023410.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).