Cell fate determinant Llgl1 is required for propagation of acute myeloid leukemia

Scribble complex proteins can influence cell fate decisions and self-renewal capacity of hematopoietic cells. While specific cellular functions of Scribble complex members are conserved in mammalian hematopoiesis, they appear to be highly context dependent. Using CRISPR/Cas9-based genetic screening, we have identified Scribble complex-related liabilities in AML including LLGL1. Despite its reported suppressive function in HSC self-renewal, inactivation of LLGL1 in AML confirms its relevant role for proliferative capacity and development of AML. Its function was conserved in human and murine models of AML and across various genetic backgrounds. Inactivation of LLGL1 results in loss of stemness-associated gene-expression including HoxA-genes and induces a GMP-like phenotype in the leukemia stem cell compartment. Re-expression of HoxA9 facilitates functional and phenotypic rescue. Collectively, these data establish LLGL1 as a specific dependency and putative target in AML and emphasizes its cell-type specific functions.


INTRODUCTION
The Scribble protein complex has been discovered and characterized in drosophila [1].Complex members serve as scaffold proteins and regulate cell polarity, motility and growth mainly through protein-protein interactions.In drosophila models, genetic inactivation of either Scribble complex member (Lgl, Scrib, Dlg) led to neoplastic tissue overgrowth supporting their role as one of the first described tumor suppressors [2].Recent data suggest involvement of Scribble polarity complex proteins in regulation of HSC biology [3,4], immune cell function [5] and potential implications in development of hematopoietic cancers [4,6,7].
Genetic inactivation of Llgl1 was associated with an increase of long-term (LT-) HSC numbers and these cells showed competitive advantage when transplanted serially into recipient mice [4].Llgl1 deletion by itself did not cause leukemia, however, its expression was correlated with decreased survival in AML.Recently, mutations of Llgl2, a close human homolog of Llgl1, have been described as an early genetic event in progression from severe congenital neutropenia to AML [8].In contrast, tumor suppressor function of Llgl1 was not conserved in murine models of lymphoid (B-and T-cell) leukemia [9].Deletion of Scrib resulted in impairment of long-term HSC function [7] and also affected proliferative capacity of AML.So far, data on the function of Dlg and its human homologs in hematopoiesis and leukemic transformation is lacking.
While specific cellular functions of Scribble complex members appear to be conserved in mammalian hematopoiesis, they seem to be highly context dependent.Differentiation stage, lineage commitment, underlying genetic background, the mechanism of genetic inactivation and the underlying cellular state may contribute to a highly variable phenotype.In this study, we screen for the requirement of Scribble complex members in AML.Unexpectedly, we identify Llgl1 as a functional vulnerability.We address the issue of context dependent effects of Llgl1 inactivation in models of acute myeloid leukemia and provide evidence

Blood analysis and bone marrow cytospins
Blood was collected into EDTA-coated tubes and investigated using a BC-5000Vet (Mindray, China).To analyze cell morphology, 1 × 10 5 bone marrow cells were centrifuged onto glass slides.Peripheral blood smears and bone marrow cytospins were stained with Wright-Giemsa (BioScientific).

Histological imaging of mouse organs
Spleen, liver and lung were fixed and embedded according to standard protocols.Slides were automatically processed for hematoxylin and eosin staining (Leica AutoStainer XL, Leica Biosystems, Wetzlar, Germany).Images were acquired at 10× magnification on an AxioImager A.2 (Carl Zeiss Microscopy, Jena, Germany).Images were processed and analyzed using the ZEN software (blue edition, version 2.3, Carl Zeiss Microscopy GmbH, Jena, Germany).

Vectors
For RNAi, shRNAs were cloned into a lentiviral pLKO.1_purovector system for puromycin selection.For HoxA9 overexpression (rescue) experiments, an MSCV-IRES-GFP backbone was used.Lentiviral and retroviral infections were performed as previously described (Schnoeder et al. Blood 2022).Detailed information on vectors and sequences are provided in Supplementary Tables 1 and 2.

Genome editing by CRISPR/Cas9
Genetic editing by CRISPR/Cas9 was performed as previously described [10,16] unless otherwise stated.Guide RNAs were designed using the Broad GPP tool [17].For cloning of sgRNA sequences, the improvedscaffold-pU6-sgRNA-EF1Alpha-PURO-T2A-RFP (ipUSEPR) vector system [18], with puromycin resistance and RFP selection marker was used.Genetic inactivation by CRISPR/Cas9 was performed as published before [15].HEL cells were transduced with the screen library and selected for 2 days with puromycin following collection of an aliquot as the input reference.Cells were cultured in vitro and samples for sequencing were collected 2 and 3 weeks later (Fig. 1A).The average relative abundance of each sgRNA in the output compared to the input samples was determined.We calculated a depletion score for each sgRNA.The median of 3-4 sgRNAs per gene was used to represent the score of the corresponding gene.Knockdown efficiency was assessed by quantitative real-time PCR (RT-qPCR) 5-7 days post-infection as published before [10].qPCR primer sequences are listed in the Supplementary Table 3. sgRNA sequences are provided in the Supplementary Tables 4 and 5.

CRISPR/Cas9 in vitro screen
HEL cells were transduced with the Scribble complex member library at a multiplicity of infection (MOI) of 20%, selected for 2 days with puromycin and an input reference (baseline) was taken at day 4 post-infection.The cells were cultured in vitro for 21 days and replicates 1-4 were taken at day 14 and day 21, respectively.Genomic DNA was isolated using the QIAmp DNA Blood Mini Kit (Qiagen, Hilden Germany), and amplification was performed using specific Illumina primer compatible sequences (Supplementary Table 6).Sequencing was performed at Genewiz (HiSseq, 150 bp, paired end) (Illumina, South Plainfield, NJ, USA).A depletion score for day 14 and day 21 were analyzed compared to the input reference using the MAGeCKFlute [19].

RNA-sequencing
RNA was isolated from cultured cells using the Qiagen RNeasy Mini kit or TRIZOL as previously described [15,20].Subsequently, mRNAs were purified using the "NEBNext® Poly(A) mRNA Magnetic Isolation Module" followed by RNAseq library preparation using the "NEBNext® Ultra™ RNA Library Prep Kit for Illumina®" according to the manufacturer's instruction.Sequencing was performed on an Illumina NextSeq500 or an Illumina HiSeq2000 (75 bp, single end) (Illumina, South Plainfield, NJ, USA).

SC-flow cytometry
For the single-cell flow cytometry analysis, bone marrow cells were resuspended in PBS/2%FBS.The staining was performed for 20 minutes at 4 °C with the antibodies mentioned in Supplementary Table 7.After washing, samples were acquired on an Aurora spectral flow cytometer (Cytek Bioscinces, Fremont, CA) equipped with 5 lasers (355 nm, 405 nm, 561 nm, 640 nm) using SpectroFlo version 3.1.0(Cytek Bioscinces, Fremont, CA).All flow cytometry data were analyzed and exported as compensated channel values with FlowJo TM software (Treestar, Ashland, OR; version 10.8.1).

Dimensionality reduction of flow cytometry data
Each sample was subset down to 30,000 cells LogNormalization (Seurat function; v4.3.0) was applied before downstream analysis [21].The compensated channel value for each cell were normalized by total channel values for that cell, multiplied by 10,000 (TP10K), and then logtransformed by log10 (TP10k + 1).After scaling, the dimensionality of the flow data was set to 33 principal components that were used as input for UMAP representation.Projection (UMAP) was applied to visualize the cell population.Subsequently, the cells were clustered using the Louvain algorithm [22] based on a Shared Nearest Neighbor (SNN) graph with a resolution parameter set to 0.29.However, one particular cell cluster was excluded from further analysis as it displayed no detectable expression of any marker genes.Ultimately, these identified clusters were annotated according to their respective cluster markers.The plots were generated using the packages ggplot2 (v3.4.1) and virdis (v0.6.3) in R 4.2.3.

Statistical analysis
Kaplan-Meier curves were plotted using GraphPad Prism version 9.0 (GraphPad Software, San Diego, CA) using the log-rank test (Mantel-Cox test).Statistical analyses were performed using ANOVA with FDR p-value correction for comparing more than two groups or t-test for comparing two groups, unless stated otherwise.Significance of p-values in figures are indicated using the following ranges: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.Each dot represents an individual biological replicate.

Inactivation of LLGL1 results in impaired proliferative capacity of human AML
We sought to investigate the context specific effects and functional dependencies of all human Scribble complex homologs in human AML.We performed an arrayed CRISPR/Cas9-based negative selection screen (Fig. 1A and Supplementary Fig. 1A) using the human AML cell line HEL.All positive controls resulted in dropout of the respective clones on days 14 and 21, while negative controls left the infected HEL cells unaffected.Out of the Scribble complex members, depletion of LLGL1 and DLG4 (days 14 and 21) and DLG1 (only day 21) could be detected (Fig. 1B).These findings were rather unexpected as deletion of LLGL1 had resulted in expansion of normal HSCs and its expression had been correlated with decreased survival in AML [4].To confirm the CRISPR/Cas9 screening results, we inactivated LLGL1 by CRISPR/ Cas9 induced knockout in 8 different human AML cell lines.All cell lines showed decreased proliferative capacity upon depletion of LLGL1 (Fig. 1C).This loss of proliferative capacity could not be attributed to induction of apoptosis (Fig. 1D) and was variable between cell lines harboring different driver mutations (Fig. 1C, D, Supplementary Fig. 1B, C).To validate the functional impact of LLGL1 deletion in human AML cells in vivo, we performed RNAimediated deletion of LLGL1 in MOLM-13 cells and assessed for disease dynamics after transplantation into humanized NSGS mice (Fig. 1E, F, Supplementary Fig. 1D).Inactivation of LLGL1 reduced disease activity as indicated by reduced spleen size of recipient mice (p = 0.0355, Fig. 1F).Moreover, deletion of LLGL1 delayed disease progression in vivo.When injecting 1 × 10 5 transduced cells, overall survival was significantly improved (median survival of shSCR: 36 days; shLLGL1: 90 days; p = 0.0002) (Fig. 1E).Consistently, inactivation of LLGL1 in leukemic cells derived from different AML patient-derived xenograft models (PDX) resulted in reduced colony forming capacity in methylcellulose (Fig. 1G, H, Supplementary Fig. 1E).attributed to reduced cell cycle activity (Fig. 2B, Supplementary Fig. 2B) rather than induction of apoptosis (Fig. 2C, Supplementary Fig. 2C).
As genetic deletion of Llgl1 may result in disturbed cell adhesion and division, we sought to validate its function in a model system that allows assessment of functional consequences following conditional Llgl1 deletion in adult hematopoietic cells [23] and also in established leukemia.In this model, exon 2 is flanked by loxP-sites and genetic deletion of this region results in loss of a functional protein [4,13].These conditional Llgl1 knockout animals were intercrossed with Mx1-Cre + animals and a conventional (straight) MLL-AF9 knock-in model that develops myeloid leukemia with a median latency of 5-6 months [12].We compared MLL-AF9 negative animals with Llgl1 inactivation (MA9 − ; Llgl1 F/F ; Mx + ) to littermate MLL-AF9 positive controls.MLL-AF9 positive animals had either unexcised/wildtype Llgl1 (MA9 + ; Llgl1 F/F ; Cre − or MA9 + ; Llgl1 +/+ ; Mx + ), a heterozygous state (MA9 + ; Llgl1 +/F ; Mx + ) or homozygosity of the conditional allele (MA9 + ; Llgl1 F/F ; Mx + ) (Fig. 2D).Genetic inactivation of Llgl1 was induced by administration of pIpC at 4 weeks of age.We monitored peripheral blood (PB) counts and distribution of immune cell subsets as well as clinical signs of disease development over time.MLL-AF9 negative animals with conditionally inactivated Llgl1 did not develop any signs of disease.In contrast, deletion of Llgl1 in MLL-AF9 positive mice resulted in delayed disease development of heterozygous (median survival: 174 days; p = 0.0943) and homozygous (median survival: 203 days, p = 0.0361) animals as compared to Llgl1 wildtype littermate controls (median survival: 147 days) (Fig. 2E).While disease development appeared prolonged in a gene dose dependent manner, we found neither changes in disease penetrance nor significant differences in blood counts or spleen size at timepoints between 64 and 300 days after pIpC injection, when individual animals were sacrificed due to clinically apparent signs of disease.
In order to confirm our findings in a second oncogenic model of acute myeloid leukemia, we crossed FLT3-ITD knock-in mice [11] with transgenic animals expressing the interferon inducible Mx1-Cre-recombinase and the conditional Llgl1 knockout (Fig. 3A).FLT3-ITD is one of the most common recurrent genetic mutations found in patients with AML [24] and FLT3-ITD knockin mice have been a valuable research model for studying the effect of cooperative gene mutations [25].Most recently, we have shown that -in this model -Cre-expression aggravates the AML disease phenotype [26].Following induction of Mx1-Cre recombinase by repeated pIpC injections at 4 weeks of age, FLT3 ITD/ITD ; Llgl1 +/+ animals developed rapid onset of leukemia with a median survival of 37 days (Fig. 3B).Consistent with our findings in the MLL-AF9 driven model, homozygous deletion of Llgl1 resulted in significant delay of disease progression (median survival FLT3 ITD/ITD ; Llgl1 −/− ; Mx + : 51 days; p = 0.0005).Although leukemic cells of FLT3 ITD/ITD ; Llgl1 −/− ; Mx + animals showed blast-like appearance with strong expression of myeloperoxidase (Fig. 3C), peripheral white blood counts at disease onset were significantly lower compared to Llgl1 +/+ controls (p = 0.0012; Fig. 3D).Histopathologic analysis of hematopoietic organs showed decreased leukemia infiltration in FLT3 ITD/ITD ; Llgl1 −/− animals and a rescued organ architecture (Fig. 3E).
These findings indicate that Llgl1 is required for proliferative capacity of murine AML irrespective of the underlying driver mutation and its inactivation leads to prolonged disease development in vivo.
Taken together, the phenotype induced by inactivation of Llgl1 is characterized by a decrease of stemness-associated gene expression signature, including loss of HoxA gene expression.Conversely, re-expression of HoxA9 can rescue this phenotype and results in re-establishment of an immature and aggressive leukemia.

DISCUSSION
Previously, genetic inactivation of Llgl1 had been associated with a significant increase in long-term (LT-) HSC numbers and these cells show a competitive advantage when transplanted serially into secondary recipient mice [4].Moreover, loss of Llgl1 expression had been associated with inferior survival in datasets of AML and the Llgl1 knockout gene signature in HSCs correlated with AML gene sets predicting dismal outcome.On the other hand, the tumor suppressor function of Llgl1 was not conserved in murine models of lymphoid (B-and T-cell) leukemia [9] suggesting that Llgl1 may act in a highly cell context specific manner.Here, we report on an unbiased targeted CRISPR/Cas9 screening approach to investigate the functional impact of Scribble polarity complex members SCRIB, LLGL1&2 and DLG1-5 on cellular function of AML.Loss of LLGL1 led to depletion of transduced cells.Impairment of proliferative capacity could be confirmed in human AML in vitro and in vivo as well as in two different mouse models of AML.These findings provide first evidence that tumor-suppressor function of Llgl1 as indicated by the gain of function phenotype in Llgl1-deficient HSCs is not conserved in models of acute myeloid leukemia.Conversely, inactivation of Llgl1 reveals inhibitory effects on disease development and propagation.Likewise, loss of Llgl1 had not altered the course of murine lymphoid neoplasms induced by constitutive Notch, c-Myc or Jak2 expression [9].These results suggest that the role of Llgl1 in hematopoietic cells may depend on lineage specificity, type of underlying oncogenic mutations and specific cellular contexts.
Interestingly, inactivation of Llgl1 resulted in loss of stemnessassociated gene expression consistent with acquisition of a more differentiated GMP-like immunophenotype.GMP-like leukemia stem cells have revealed a less aggressive phenotype in MLLfusion induced models when compared to HSC-derived leukemias [27].These findings are consistent with the observed delay in Llgl1-deficient leukemia development.Although FLT3-ITD driven AML is not among genetic subtypes with the highest levels of HOX-gene expression (such as NPM1-mutated AML) we found reduced expression of the HoxA-gene cluster along with other stemness associated genes.Loss of a stemness-associated gene expression program by deletion of Llgl1 may indicate its involvement in regulation of cell polarity, which has been previously described in other model systems [28].Here, disruption of polarity may alter asymmetric cell division (ACD), shift ACD of leukemia stem cells towards symmetric cell division and result in loss of self-renewal capacity.Conversely, gain of a self-renewal associated gene expression signature in normal HSCs [4] may rather be attributed to increased symmetric renewalin a context specific manner.Impressive work on normal HSCs has recently shown that HSC fitness response to stress depends on signaling molecules Yap1 and Taz, and that deletion of Yap1 and Taz induces loss of HSC quiescence and symmetric self-renewal ability [3].Moreover, this work provided evidence that Scrib-complex member Scribble and Yap1 coordinate to control Cdc42 activity and HSC fate determination.While the mechanistic role of Llgl1 deletion in cell fate decisions of LSCs and its potential interaction with Yap-Taz-signaling is clearly beyond the scope of this manuscript, it is tempting to speculate on a cell-context specific role of Llgl1 in maintenance of an immature cell state through interaction with the described pathways.Taken together, cell-type specific functions of Scribble complex member Llgl1 could be confirmed in acute leukemia, which may suggest a possible disease specific modulation of cell polarity complexes.

FollowingAFig. 1
Fig. 1 Targeted Scribble polarity complex screen identifies LLGL1 as a dependency in human AML.A Schematic depicting CRISPR/Cas9 negative selection screen strategy.B Dependency scores for days 14 and 21 compared to baseline.C, D Genetic inactivation of LLGL1 by CRISPR/ Cas9-induced knockout of LLGL1 compared to non-targeting control (sgLUC) in human AML cell lines expressing Cas9.C Proliferation assayed by cell counting after trypan blue exclusion.n = 3-5 independent experiments; mean ± SD; paired Student's t test.D Apoptosis assays using SytoxBlue and Annexin V staining.n = 3-5 independent experiments; mean ± SD; paired t-test.E, F Xenografting of human AML cells in NSGS mice.E Survival curves of NSGS recipient mice following transplantation of 1 × 10 5 MOLM-13 cells following inactivation of LLGL1 (shLLGL1_1/2; n = 12) by RNAi compared with non-targeting control (shSCR; n = 8); Mantel-Cox test.F Spleen size of recipient animals; mean ± SEM; unpaired t-test.G, H Colony formation (CFU-) assays of primary human AML following inactivation of LLGL1 (shLLGL1_1/2) by RNAi compared with nontargeting control (shSCR) in methylcellulose; n = 4 independent replicates, paired t-test.G Colony count, total cell numbers and H Representative pictures of colony morphology.Cytogenetic and molecular genetic aberrations are indicated above.Scale bars = 200 µm; magnification 4×.
These findings indicate the requirement for LLGL1 to maintain proliferative capacity in human AML across various genetic subtypes.
Llgl1 is required for disease maintenance in AML driven by different oncogenes Murine models of AML allow for a more detailed and controlled assessment of functional consequences after inactivation of cell fate determinants such as Llgl1 as they lack the inter-individual heterogeneity of human samples.Inactivation of Llgl1 by RNAi with two different shRNAs in murine leukemia induced by retroviral infection with MLL-AF9 resulted in significant decrease of proliferative capacity compared to non-targeting control (p < 0.0001; Fig.2A, Supplementary Fig.2A).This effect could be