The LMO2-LDB1-TAL1 complex regulates transcription networks in acute myeloid leukemia

Key Points

• •Patients with relAML have distinct transcriptomes potentially regulated by a core group of transcriptional coregulators (LMO2-LDB1-TAL1).
• •These transcriptional coregulators can regulate cell cycle, proliferation, and potentially response to chemotherapeutic agents in patients.

Visual Abstract

Abstract

Relapsed acute myeloid leukemia (relAML) remains a clinical challenge. We have shown that epigenetic heterogeneity may contribute to transcriptional dysregulation and disease progression in AML, but the specific aberrant transcriptional programs have not been identified. We analyzed molecular profiles from patient-matched diagnostic and relapse AML specimens. A subset of differentially expressed genes (DEG) that were disparate in direction of expression change identified 2 patient subtypes. We predicted that transcriptional regulators (TR) might regulate the expression patterns observed. The expression patterns of the top TR predicted for the disparate genes associated with clinical outcomes. The top TR predicted for the disparate DEG and DEG identified in a patient-derived xenograft model of relAML included members of the LIM domain only 2 - LIM domain binding 1 - TAL BHLH TF1, erythroid differentiation factor (LMO2-LDB1-TAL1) multisubunit complex (LTMC). Analysis of DepMap data identified LMO2-dependent cells with a subset highly expressing TAL1, suggesting coordinated regulation. TAL1 copurified in immunoprecipitation for LMO2 and LDB1 followed by tandem mass spectrometry analysis in HEL and K562 cells, and results from chromatin immunoprecipitation experiments suggest significant co-occupancy of TAL1 and LDB1. Loss-of-function experiments targeting LMO2, LDB1, and TAL1 in AML cell lines associated with reduced cell growth, downregulation of cell cycle genes, and a negative association with gene expression patterns observed in relapsed patients with increased TAL1 expression. Our results from primary AML specimens and functional analyses of AML cell lines supports an essential role for the LTMC in AML. Targeting the complex or downstream effectors could provide novel therapeutic considerations for a subset of patients with AML.

Introduction

Disease relapse remains a clinical challenge for adult patients with acute myeloid leukemia (AML).¹ Few studies have assessed molecular patterns acquired in adult relapsed AML (relAML).^2,3 Recurrent somatic events specific to relAML have not been identified.^4,5 Epigenetic heterogeneity may contribute to transcriptional dysregulation patterns and disease progression in AML.^6,7 However, differences in transcriptional programs between diagnosis and relapse are not well understood.

Transcriptional regulators (TR; including transcription factors [TF], chromatin, and epigenetic regulators) play critical roles in hematopoiesis and leukemogenesis⁸ and associate with clinical outcomes.⁹ Identifying functional TR in a disease system based solely on gene expression profiles is a nontrivial task. TR control target gene transcription either by directly interacting with DNA elements at promoters or enhancers (in the case of TF) or by interacting with chromatin features such as nucleosomes and histone modifications (for chromatin regulators). A single TR can regulate hundreds to thousands of genes in a given cell type, whereas any singular gene may be regulated by different TR under varying biological contexts. Without direct measurement of TR binding in each sample, TR activity can only be inferred computationally.

Research into the role of TR in AML has focused on mutations and chromosomal abnormalities altering the function of TR, with less focus on transcriptional changes brought on by nongenetic events.^10,11 Furthermore, patient-derived AML specimens used for molecular analysis are predominately diagnosis samples and are lacking in patient-matched specimens at the time of relapsed disease.^10,11 Because this area of research is understudied, we sought to investigate the role of TR in patient-matched diagnosis and relapse specimens. We hypothesized that TR may play a role in AML relapse pathogenesis.

Methods

Patient specimens

Results of the study include somatic events, DNA methylation, and gene expression data that have been published previously^6,12,13 (supplemental Table 1). New study specimens included peripheral blood and bone marrow collections from serial specimens from 2 patients with AML during their disease time course (AML_124 and AML_130), specimens collected from a patient with AML used in a xenograft experiment (AML_141), and CD34⁺ bone marrow cells isolated from a healthy donor (NBM_15). All patients signed informed consent according to the Declaration of Helsinki for collection and use of sample materials in research protocols. Study protocols were approved by the institutional review boards at Weill Cornell Medicine (protocol number 0805009783), the University of Virginia (protocol number IRB-HSR 19796), the University of Pennsylvania (protocol number 703185), and London King’s College (approved by London Westminster research ethics committee; reference 06/Q0702/140). Specimens were processed as previously described⁶; mononuclear cell selection was followed by lymphocyte depletion before RNA extraction for downstream assays. A full description of the methods are in the supplemental Methods.

Cytogenetics, somatic mutations, and epigenetic and genetic evolution data from study patients

Diagnostic cytogenetics and somatic mutation data, as well as epigenetic and genetic evolution results for 59 patients were obtained from previously published reports.^6,12 A custom targeted panel was used for relative gene expression validation (see supplemental Information).

RNA-seq

RNA-sequencing (RNA-seq) raw data generated from the patient-matched diagnosis and relapsed specimens were obtained from dbGaP accession phs001027.v4.p1. As previously reported,^6,13 specimens were prepared in 2 batches (2 patient cohorts) in which all paired specimens were prepared within the same batch, and new bone marrow specimens were prepared with each batch and used as analysis references accordingly (supplemental Table 1).

RNA-seq libraries for AML_141 specimens were prepared using TruSeq RNA-seq by poly(A) enrichment (Illumina) and sequenced on HiSeq2000 (Illumina) using a 50–base pair paired-end approach per the manufacturer’s recommendations.

RNA-seq libraries for the serial specimens from AML_124 and AML_130 were prepared using the TruSeq Stranded Total RNA H kit (Illumina) and sequenced on a HiSeq2500 (Illumina) using a 150–base pair paired-end approach per manufacturer’s recommendations.

Data alignment, differential expression calculations, and comparison statistics for the study cohorts, xenograft experiment, and serial specimens are described in the supplemental Information.

Unsupervised clustering of DEG and patients

K-means clustering was applied with k values of 2, 3, 5, 7, and 9, and repeated 20 times for differentially expressed genes (DEG; identified between diagnosis and relapse) and patient-derived AML samples, respectively. The adjusted rand index (ARI) was calculated between each pair of clustering results. The clustering results for genes and patients with the highest average ARI were used for the downstream analysis (k = 3 for DEG, with average ARI of 0.999; and k = 2 for patient samples with average ARI of 1).

TR inference using BART

We used binding analysis for regulation of transcription (BART),^14,15 a computational method for identifying candidate TR that potentially regulate DEG between diagnosis and relapse, and the DEG in the mouse xenograft specimens. BART predicts the functional TR that most likely regulate a given gene set as input, by leveraging a large compendium of public chromatin immunoprecipitation sequencing (ChIP-seq) data sets that capture the genome-wide binding patterns of >900 human TR across many cell types. BART is ranked among the top-performing tools according to a benchmark study.¹⁵ TR were considered significant if the BART-associated Irwin Hall P value was <.05.

RP analysis

Regulatory potential (RP)¹⁶ assesses the level of regulatory effect of a TR binding profile (usually from a ChIP-seq data set) on each gene, based on the number of binding sites for that TR in a 200-kilobase (kb) genomic region surrounding the gene’s transcription start site (TSS) and weighted by their distance to the TSS. The RP of a TR from a ChIP-seq data set on each gene was calculated as the sum of ChIP-seq peak signals weighted by the genomic distance from the TSS. Specifically, ChIP-seq peaks surrounding gene i (TSS ± 100 kb) were collected and weighted by an exponential decay function for the total RP effect RPi on this gene:

$${RP}_i=\sum _j\frac{{2e}^{{-ux}_{ij}}}{1+e^{{-ux}_{ij}}}{S}_j\text{,}$$

where x_ij is the distance between TSS of gene i and peak j, and S_j is the peak signal (MACS2 q value). The parameter u determines the decay rate and is set so that the half-life of the decay function is 10 kb.

To address whether the TR were specific to the gene groups from which the TR were predicted, the RP scores from ChIP-seq data of PU.1, TAL basic helix-loop-helix TF 1, erythroid differentiation factor (TAL1), GATA binding protein 1 (GATA1), and RB binding protein 5, histone lysine methyltransferase complex subunit (RBBP5) were compared among the 3 groups of genes identified. The ChIP-seq peak files were collected from Cistrome Data Browser (DB)^17,18 with the data from human hematopoietic cells with high area under the curve (AUC) scores by BART prediction. The Student t test was used to compare the RPs between the gene groups identified.

GSEA

Functional gene annotation analysis was performed using gene set enrichment analysis (GSEA; https://www.gsea-msigdb.org/gsea/index.jsp; version 4.2.3) and Hallmark and C2 MSigDB gene sets.19, 20, 21 Enrichr (https://maayanlab.cloud/Enrichr/)^22,23 was used with the pathway database of KEGG_2021_Human²³ and MSigDB gene sets.19, 20, 21 Corrected P values of <.05 and normalized enrichment score of ≥|1.5| were used for significance determination in the GSEA analysis, and adjusted P value <.05 was used for significance in the Enrichr analyses.

Survival analysis

We performed gene set survival analysis using the advanced expression survival analysis method.²⁴ Results are represented in Kaplan-Meier curves and computed using Cox proportional hazard regression. Relative gene expression values (z score) for the analyses were the Beat AML data set (obtained from: https://biodev.github.io/BeatAML2/) and for the TCGA data set (obtained from: https://portal.gdc.cancer.gov/).

Tandem mass spectrometry

FLAG-tagged LIM domain binding 1 (LDB1)²⁵ and FLAG-tagged LIM domain only 2 (LMO2) were used for these experiments. Chromatin extracts from K562 and HEL cell lines were independently mixed with 125 mL of packed FLAG resin (Sigma) and incubated with mixing at 4°C for 4 hours.²⁶ Immune complexes were isolated by brief centrifugation, and washed rapidly 3 times with 4 mL wash buffer (20mM HEPES [N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid; pH 7.6], 300mM NaCl, 0.1% Triton X-100, 10% glycerol), 625 μL wash buffer supplemented with 100 pmol/l3FLAG peptide (catalog no. F4799; Sigma) was then added to the resin, and the mixture was incubated for 30 minutes to elute F-LDB1– and LMO2-F–bound proteins. Eluents were combined with equal volume of 10% sodium dodecyl sulfate and then precipitated and digested using S-Trap Micro Spin Columns (ProtiFi) according to manufacturer’s recommended protocol. Eluted peptides were dried with a vacuum concentrator, resuspended in 0.2% formic acid, and autosampled onto a 100 μm × 20 cm microbore column containing 3 μm C18 particles (Jupiter, Phenomenex), and subsequently resolved using a 70-minute aqueous-to-organic gradient ultrahigh-performance liquid chromatography delivered via a Dionex Ultimate 3000 system. Both full-scan spectra and data-dependent tandem mass spectra were acquired over the course of the gradient using an Exploris 480 orbitrap instrument (Thermo Scientific). Resulting tandem mass spectra were searched against a human protein database (UniProt canonical) using Sequest²⁷ and resulting peptide spectral matches were filtered and collated to the protein level using Scaffold 5 (Proteome software: https://www.proteomesoftware.com/products/scaffold-5). A table detailing the specimens profiled by mass spectrometry is located in the supplemental Information.

ChIP for LDB1

ChIP-seq was performed using an anti-LDB1 antibody as described in the supplemental Information. Genomic annotation to genes was performed by BedTools using GRCh38 v109 GTF downloaded from Ensembl. Overlap of gene-annotated peaks between the ChIP-seq data generated and publicly available ChIP-seq data from TAL1 (ENCODE) was performed using a hypergeometric test.

Loss-of-function in vitro functional experiments

Briefly, HEL and K562 cell lines were used for experiments. Knockout and knockdown (KD) approaches as described in the supplemental Information were performed to target LMO2, LDB1, and TAL1. Perturbations of expression were validated using western blotting. Cell proliferation assays were performed using cell counting to assess comparative cell counts between experimental groups. Cells from experimental and control specimens were harvested, and RNA was isolated and profiled by RNA-seq. Differential expression and functional annotation analyses were performed. Full methods describing the experiments are included in the supplemental Information.

Beat AML treatment sensitivity

Raw cell viability data were obtained from the Beat AML repository (beataml_wv1to4_raw_inhibitor_v4_dbgap.txt). Measurements were filtered to include only those recorded between 18 and 30 hours of drug treatment. For each drug, normalized viability values were restricted to a range of 0 to 100, and replicate measurements were averaged at each concentration. Half-maximal inhibitory concentrations were calculated using the compute 50% inhibitory concentration function from the R package PharmacoGx (version 3.10.0), and AUC values were computed using the computeAUC function from the same package. Patients with AML in Beat AML study who had RNA-seq and drug treatment data were filtered for high (top 10%) and low (bottom 10%) expression of TAL1 expression. The significance of difference between the 2 groups identified was assessed using a Wilcoxon rank-sum test.

Results

To address our hypothesis, we used data from published patient-matched diagnosis and relapsed specimens.^6,12,13 All patients included in the study had newly diagnosed AML, which entered remission after receiving induction treatment and experienced disease relapse within 5 years from diagnosis time. Gene expression (RNA-seq) data were generated from 59 patients divided into 2 data batches (cohort 1, n = 29; cohort 2, n = 30; supplemental Table 1). Patients in both cohorts were female and male adults aged ≥18 years, received combination intensive chemotherapy for induction treatment, and had no differences in sex distribution and time to relapse (Fisher exact test, P > .05).⁶

Relapsed AML is associated with convergent differential expression that defines 2 patient clusters not associated with specific somatic or clinical features

We integrated the gene expression with genomics data from matched diagnosis and relapse specimens.^6,12 We identified 5416 DEG between diagnosis and relapse in cohort 1 (supplemental Table 2A-B). Unsupervised k-means clustering on the DEG identified 3 distinct gene groups: group A (88% downregulated), group B (variable expression pattern), and group C (65% upregulated) genes (Figure 1A; supplemental Table 2C-D). The expression patterns of the group B genes separated the patients into 2 clusters using k-means clustering: C1 with upregulation, and C2 with downregulation of group B genes (Figure 1A; supplemental Table 2E). These patient clusters were consistent with those generated using independent clustering generated on the DEG (supplemental Figure 1A). We validated the 3 identified gene groups in cohort 2 samples, in which we observed similar differential expression patterns and segregation into 2 clusters (supplemental Figure 1B; supplemental Table 3). Select DEG changes were validated in a subset of patients^6,12 (supplemental Table 1; supplemental Figure 1C). The frequency of recurrent diagnostic somatic events and evolution patterns between diagnosis and relapse were not significantly different between the 2 patient clusters (Figure 1A; supplemental Figure 1B; supplemental Table 4A-F). Furthermore, clinical features (sex; age at diagnosis; time to relapse; French, American, and British classification; and 2017 European LeukemiaNet classification²⁸) were not significantly different between the 2 patient clusters (Fisher exact tests or Student t tests, P > .05; supplemental Table 4G).

Figure 1.

Distinct patterns of differential gene expression in relAML are associated with predicted divergent activities of TR. (A) Differential gene expression patterns of 5416 significantly DEG between diagnosis and relapse in cohort 1 of study patients with AML. K-means clustering was applied to genes (k = 3) and patient samples (k = 2). Log₂FC >0 represents genes with higher expression at relapse than at diagnosis. Diagnostic cytogenetic events in cohort 1 are displayed under each patient column (black/gray). Epigenetic (dark blue/yellow/green) and genetic (gray/yellow/pink) evolution models, mutation, and CNV alteration patterns (red/blue/gray) are displayed under each patient column. (B) BART-inferred TR from group B DEG. The TR highlighted in red were differentially expressed between diagnosis and relapse in either patient cluster. P value scores were generated using BART analysis. (C) BART-inferred TR from downregulated genes in a mouse xenograft model of relAML. Specimens collected 4 weeks after exposure to cytarabine were compared with engrafted specimens. The top TR that were shared with BART results from patient data (P < .01) are listed. (D) Hypergeometric analysis of group B genes (left) and group B TR (right) overlap with Hallmark gene sets. The x-axis indicates the number of genes overlapping with each gene set, and color represents statistical significance of the overlap. CNV, copy number variation; FC, fold change; WT, wild-type.

We next aimed to determine what mechanisms could underlie the gene expression changes observed upon relapse. First, we assessed whether promoter differential DNA methylation was associated with the DEG identified. We observed that relapse was associated with the acquisition of both hypomethylation and hypermethylation within proximal gene promoters (supplemental Figure 2A-D; supplemental Table 5), although most DEG were not associated with differential methylation (supplemental Figure 2E-H; supplemental Table 6A-C). These results suggested that gene expression patterns acquired upon relapse were not associated with specific clinical features and were largely independent of somatic events and promoter DNA methylation.

Differential expression patterns between diagnosis and relapse are driven by TR activities

We hypothesized that the observed differential expression patterns are driven by TR activities. We used BART¹⁴ to identify TR that potentially regulate each of the gene groups A, B, and C (Figure 1B; supplemental Table 6D-F). BART identified¹⁴ several hematopoietic TF (Figure 1B; supplemental Figure 3A-B) that were also differentially expressed between the 2 patient clusters (supplemental Figure 3C-E; supplemental Table 7). To address whether the TR were specific to the gene groups from which the TR was predicted, we assessed the RP¹⁶ of select TR in their own gene group compared with other gene groups. We found that TR inferred from gene groups had significantly higher RP¹⁶ in their own gene group than other gene groups (supplemental Figure 4B-E), suggesting that TR were specific to the group from which they were predicted.

We focused our analysis on the group B genes due to their ability to segregate patients (Figure 1A). For this gene group we identified 184 TR (Figure 1B; supplemental Table 6E). Interestingly, the TR identified included several members of a protein complex known to function in hematopoiesis²⁹ that also has a pathogenic role in T-cell acute lymphoblastic leukemia²⁵ (T-ALL; TAL1, GATA1, and LDB1; LMO2-LDB1-TAL1 leukemogenic multisubunit complex [LTMC]). TAL1 and GATA1 also had increased expression in C1 and decreased expression in C2 (supplemental Figure 3D). To validate this prediction, we transplanted an independent diagnostic AML specimen into mice and collected cells at engraftment and 4 weeks after cytarabine treatment and performed RNA-seq on these samples. We identified DEG and TR between pretreatment and posttreatment specimens (supplemental Table 8A-C). TAL1 and GATA1 were downregulated in posttreatment specimens; and TAL1, GATA1, and LDB1 were predicted TR for downregulated genes (Figure 1C; supplemental Table 8A-B), consistent with previous observations from C2 (Figure 1B). Additionally, LMO2, a member of the LTMC, was also identified as a predicted TR.

We further validated TR predictions by comparing the relapsed AML transcriptome with gene expression changes observed after TR KD using publicly available knockTF data sets.³⁰ We calculated the correlation between KD DEG and each patient, then assessed whether the correlation was associated with the differential expression of the respective TR across patients (supplemental Figure 5A). As expected, genes predicted to be regulated by TAL1 and GATA1 in patients were differentially expressed after KD in K562 cells. Expression changes co-occurring with an increase in TAL1 or GATA1 expression at relapse were inversely correlated with expression changes after TAL1 or GATA1 KD in K562 cell lines (supplemental Figure 5B-E; supplemental Table 9A-B).

To determine what biological processes are regulated by the predicted TR, we performed pathway enrichment analysis on group B genes, which were enriched for cell cycle gene sets (supplemental Table 10A-D). Group B–predicted TR were enriched for transcriptional regulation, cell cycle, and inflammatory signaling gene sets as expected (Figure 1D; supplemental Table 10). Importantly, diagnostic expression levels of the top 30 group B–predicted TR were associated with patient survival (Figure 2).

Figure 2.

The expression patterns of top predicted TR associate with survival in independent AML cohorts. (A) Survival outcome of the TCGA AML cohort separated by the expression pattern of the top 30 predicted group B genes’ TR from the BART results. (B) Survival outcome of the Beat AML cohort separated by the expression pattern of the top 30 predicted group B genes’ TR from the BART results. Expression patterns were analyzed using the composite gene expression score separated by 50% of the cohort. Permutation P value was corrected using 100 permutations. CI, confidence interval; HR, hazard ratio.

LTMC regulates transcriptional programs and AML cell growth

Group B–predicted TR demonstrated enrichment for known hematopoietic TF that interact and share common targets (including TAL1 and LDB1⁹), which have been suggested to be essential in AML,^9,31 and are members of the LTMC.²⁵ To assess the LTMC in AML, we performed coimmunoprecipitation followed by mass spectroscopy in AML cell lines. TAL1 and other TR copurified with LMO2 and LDB1 (Figure 3A), suggesting the complex assembles in AML. We next assessed the expression of the complex members in AML. We assessed for essentiality scores in the DepMap KD experimental results³¹ in leukemia cell lines and identified LMO2-essential and -nonessential AML cell lines,³¹ suggesting that the LTMC is functional in a subset of patients (Figure 3B; supplemental Figure 6A-C). To further assess the potential for LTMC function in AML, we performed LDB1 ChIP-seq in K562 cells and compared the results with genomic annotations from TAL1 ChIP-seq (ENCODE) in the same cell line. There was overlap between the peaks identified for the 2 TR (Figure 3C). Furthermore, we observed significant overlap at gene-annotated peaks (P < .01; Figure 3D-E), notably proximal to transcriptional start sites (supplemental Figure 7A). Gene targets included the genes of the LTMC themselves, implying autoregulation of LTMC subunits (supplemental Figure 7B-C). Additionally, erythroid-specific gene targets, many of which cluster with LMO2-dependent AML lines, showed LDB1 and TAL1 co-occupancy (supplemental Figure 7D-F).

Figure 3.

LTMC subunits interact and associate with transcriptional patterns and target genes in AML cell models. (A) Interacting proteins identified by coimmunoprecipitation followed by tandem mass spectrometry using FLAG-tagged LDB1 and LMO2 in K562 and HEL cells. Darker color background indicates a higher number of peptides that were annotated to specific proteins. (B) Heat map of relative expression levels (z score) of DEG between LMO2-essential (dark blue) and -nonessential (light blue) cell lines defined using DepMap Essentiality scores. Distances were computed using the Canberra method and were clustered using Euclidean (columns) and Complete (rows) methods. (C) Tornado plots of LDB1 peaks and TAL1 peaks annoated to the same loci identified in ChIP-seq experiments in K562 cells. (D-E) Overlap of LDB1 and TAL1 target genes annotated from ChIP-seq experiments in K562 cells; TAL1 ChIP-seq data: ENCFF804FGY (D) and ENCFF207AUR (E). ∗∗∗P < .0001.

Given these results, we hypothesized that the complex regulates gene expression in AML cells. To assess this, we performed loss-of-function experiments in human AML cell lines (supplemental Figure 6B-C). Previous reports indicate that TAL1 KD in K562 cells associates with a loss in cell growth phenotype.³² Downregulation of LMO2 and LDB1 in K562 was also associated with reduced cell growth (Figure 4A-B, supplemental Figure 8A-F). To assess the effects on leukemogenic potential at relapse, we performed loss of function experiments in a relapsed AML cell line (HEL). Functional loss of TAL1 and LDB1 in HEL also associated with reduced cell growth (Figure 4C; supplemental Figure 8G-K). To determine whether the effects on the leukemogenic phenotype was mediated by shared transcriptional programs, we generated gene expression data from these in vitro experiments. Transcriptional patterns acquired after loss of function with all 3 TR identified high correlation between DEG, suggesting common targets among these TR (supplemental Table 11A-D). As expected, DEG acquired after LMO2-LDB1-TAL1 loss of function were negatively enriched for upregulated genes associated with high TAL1 expression at relapse, including cell cycle and heme metabolism gene sets^33,34 (Figure 4D; supplemental Table 11E-H).

Figure 4.

LTMC subunit KD reduces cell growth and regulates convergent transcriptional patterns in AML cells. (A-C) Growth curves for LMO2-degron (A), shLDB1 (B), and siTAL1 (C) relative to starting cell numbers. Student t test was used for significance testing (day 4). (D) GSEA of ranked LMO2-degron (left) and siTAL1 (right) compared with select gene sets. Size of dots are scaled to NES. and all P values <.05. NES, normalized enrichment score; shLDB1, shRNA targeting LDB1; siTAL1, siRNA targeting TAL1.

TAL1 expression level is associated with predicted drug sensitivity

Finally, we aimed to determine whether the LTMC regulated expression patterns could predict drug sensitivity.¹¹ Loss of LMO2, TAL1, or LDB1 associated with the acquisition of expression patterns predicting sensitivity to kinase inhibitors and other drugs (Figure 5A; supplemental Table 12). We next used results from ex vivo functional drug screens in the same study. As predicted, TAL1-high specimens harbored resistance to kinase inhibitors when compared with TAL1-low specimens (Figure 5B-C; supplemental Figure 9A-C). This suggested that the patient clusters (C1 and C2; Figure 1A) may be differentially sensitive to such inhibitors.¹¹

Figure 5.

LTMC regulates genes associated with sensitivity to chemotherapeutic agents in AML cells. (A) GSEA of ranked LMO2-degron (left) and siTAL1 (right) compared with gene sets associated with chemotherapy response used as the reference gene sets. Size of dots scales with NES and color scales with false discovery rate q values. (B-C) AUC of treatment response to erlotinib (B) and sorafenib (C) in patients in the Beat AML cohort separated by the top (red) and bottom (blue) 10% of patients with TAL1 expression. Wilcoxon rank-sum test was used for significance assessment.

Discussion

In summary, we analyzed 2 independent cohorts of patients with AML with paired diagnosis and relapse samples to generate transcriptomic signatures defining relAML. These signatures were not associated with any specific AML subtype, cytogenetic abnormality, or somatic gene mutation in our study cohort. By using the robust BART algorithm¹⁴ to elucidate the molecular drivers behind these transcriptional signatures, we identified multiple TR of the LTMC protein complex, which we could also purify from AML cells. Additional functional studies showed consistency in effects on leukemic cell growth and acquired gene expression changes associated with disruption of LTMC members’ expression, further validating its role as a leukemic driver.

The implication of the LTMC in leukemia pathogenesis and relAML expands our understanding of this protein complex in human disease. LMO2 and TAL1 are frequent drivers of human T-ALL, in which they can be deregulated by diverse chromosomal gene rearrangements.³⁵ As a TF, TAL1 expression in AML is relatively low and small changes in protein abundance could have significant effects. Overexpression of LMO2 and TAL1 results in altered expression programs,^36,37 and certain AML cell lines may require LMO2 and LDB1 for survival (Figures 3B and 4A-C; supplemental Figure 8) suggesting an expanded role in multiple leukemia types. Unlike in T-ALL, the LTMC is normally expressed in hematopoietic stem and progenitor cells and LTMC components (TAL1, LMO2, and LDB1) are subject to somatic events in <1% of patients of all ages with AML.³⁸ Previous studies have shown that relAML contains a higher proportion of leukemia-initiating stem cells than AML at initial diagnosis.³⁹ Thus, it is conceivable that the relAML leukemia-initiating stem cell population is enriched with an LTMC-driven transcriptome. Because the LTMC has such a prominent role in the maintenance of normal hematopoietic stem and progenitor cells,²⁹ it may similarly sustain the cancer stem cells that drive relapse.

Additionally, LTMC may directly control genes involved in targeted chemotherapy sensitivity (Figure 5; supplemental Figure 9; supplemental Table 12). Receptor tyrosine kinase signaling pathways contribute to AML pathogenesis⁴⁰ and may be an effective treatment modality to consider in AML with low TAL1 expression, including relAML. Along these lines, our approach supports the idea that patient subgroups can be defined based on transcriptional heterogeneity, regardless of the molecular background or association with outcomes. Furthermore, our results identified consistency in transcriptional programming past the first relapse (supplemental Figure 10; supplemental Table 13), suggesting that relapsed transcriptional programs may be sustained and contribute to further disease progression. This finding warrants further investigation due to limited patient specimen availability in our study. Future studies deciphering the molecular and phenotypic events facilitated by the LTMC will yield insight into disease biology and potential effective therapeutic targeting strategies of relAML and AMLs in which the complex is predicted to be functional.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

The current affiliation for T.L. is Molecular Cell Biology Program at SUNY Downstate, New York, NY.

The current affiliation for I.P. is Macon & Joan Brock Virginia Health Sciences at Old Dominion University, Norfolk, VA.

The current affiliation for F.R. is Sanofi, Bridgewater, NJ.

The current affiliation for C.S. is Immunai, New York, NY.

The current affiliation for M.W.B. is OhioHealth, Columbus, OH.

The current affiliation for A.M.M. is Josep Carreras Leukaemia Research Institute, Barcelona, Spain.