TP53-mutant AML with ribosomal gene loss exhibits impaired protein translation and sensitivity to HSP90 inhibition

Abstract

TP53-mutated acute myeloid leukemia (AML) represents a particularly aggressive and therapeutically refractory subtype of the disease. While recurrent chromosomal abnormalities such as -5/del(5q), -7/del(7q), and del(17p) are well studied in this context, additional co-occurring events remain less well defined. Using the multidimensional Leucegene dataset (~700 primary AML specimens), we identified and comprehensively characterized a distinct subset of TP53-altered AML marked by recurrent deletions on the short arm of chromosome 3 [del(3p), >20% TP53-mutated cases]. These deletions frequently co-occur with del(5q) and encompass several ribosomal protein genes (RPGs), leading to a global down-regulation of the ribosomal network and reduced protein synthesis. We show that this ribosomopathy-like phenotype is most pronounced in TP53-mutated cases with combined RPG deletions on chromosomes 3p and 5q, suggesting a cooperative oncogenic mechanism. Chemical screening identified HSP90 inhibition as a selective vulnerability in AML with low RPG expression. These findings highlight a previously unappreciated TP53-altered AML subset characterized by converging genomic and translational defects and suggest that ribosomal stress may serve as a therapeutic entry point for targeted intervention of this patient subgroup.

Ribosomal gene loss on 3p/5q induces translational defects and HSP90 inhibitor sensitivity in TP53-mutant AML.

INTRODUCTION

Acute myeloid leukemia (AML) is a genetically heterogeneous malignancy characterized by recurrent chromosomal aberrations and gene mutations. Its diagnosis is based on complementary laboratory tests including morphology, flow cytometry, cytogenetics, and targeted mutation analysis that enable classification of this disease into distinct biological entities and allow risk stratification (1).

Despite regular advances, the clinical outcome of AML remains poor, with a 5-year survival of 32.9% (2). Despite the development of newer therapeutic agents, long-lasting remissions remain uncommon for high-risk disease (1, 3). Among the most aggressive subtypes is TP53-altered AML, which is found in ~5 to 10% of de novo cases and up to 30% of therapy-related or elderly AML (4, 5). AML with mutated TP53 was classified as a distinct diagnostic entity in the 2022 International Consensus Classification (ICC). Under this definition, any AML case with ≥20% blasts and a pathogenic TP53 mutation exhibiting a variant allele frequency (VAF) of ≥10% qualifies for inclusion in this category (6). TP53-mutated AML shows a unique biology with extremely poor prognosis with high relapse rate post–hematopoietic stem cell (HSC) transplantation (7, 8).

Although TP53-mutated AML frequently harbors large chromosomal deletions, most notably of 5q, 7q, and 17p (4, 9), the landscape of other, potentially cooperating genomic lesions is poorly defined. Given the central role of TP53 in DNA repair, apoptosis, and cellular stress responses (4), delineating the genomic context in which TP53 loss of function drives leukemogenesis is crucial to the identification of novel disease mechanisms and vulnerabilities.

In this study, we used the Leucegene dataset (https://leucegene.ca/), a comprehensive, multidimensional resource comprising 691 AML specimens from 656 patients selected to capture the full genetic diversity of the disease. Through integrative transcriptomic and genomic analyses, we identified a subset of TP53-altered AMLs characterized by recurrent deletions on chromosome 3p that co-occur with del(5q). These deletions span several ribosomal protein genes (RPGs), resulting in network-wide down-regulation of the ribosome and impaired protein synthesis, a phenotype evoking somatic ribosomopathies. This ribosomal dysfunction defines a biologically distinct subgroup of TP53-mutated AML, for which we demonstrate increased sensitivity to heat shock protein 90 (HSP90) inhibition, highlighting a potential therapeutic vulnerability.

RESULTS

Accompanying this study, we report the complete Leucegene data collection (https://leucegene.ca/) including genomic and transcriptomic data, as well as clinical annotations (see Data, code, and materials availability), for 691 primary AML specimens. These samples are distributed into 15 cytogenetically defined molecular subgroups, with frequencies ranging from 0.3% [-17/del(17p) not complex and hyperdiploid with numerical abnormalities only] to 40.8% (normal karyotype) (Fig. 1A, left, and table S1; see Fig. 1A, right, for the distribution of TP53-mutated cases among molecular subgroups; see Fig. 1B for the distribution of general clinical characteristics).

Fig. 1. Cytogenetic distribution and CNA landscape of the Leucegene cohort.

(A) Frequency of cytogenetic subgroups in the Leucegene cohort (n = 691, left) and distribution of TP53-mutated cases in these subgroups (right). Cytogenetic subgroups are mutually exclusive, and each specimen is assigned to a single subgroup based. (B) Distribution of general clinical characteristics of AML specimens in the Leucegene cohort. FAB, French-American-British (FAB) classification; WBC, white blood cell count; UND, undetermined; L, liter. (C) Copy number alteration (CNA) landscape of the whole Leucegene cohort (yellow and dark blue curves for gains and losses, respectively) or limited to TP53-mutated AML (brown and light blue shaded area for gains and losses, respectively). Reported mean frequencies of gain (top) and loss events (bottom) are calculated per window of 3 Mb. Main haploinsufficient candidate genes are indicated under the peaks formed by well-characterized TP53m-associated deletions (5q, 7q, and 17p). Vertical solid black and gray lines separate chromosomes and depict centromeres, respectively. Horizontal dotted lines indicate 25% of frequency.

We analyzed nonredundant specimens of the collection (n = 656), prioritizing primary diagnostic samples. All specimens were subjected to RNA sequencing (RNA-seq; Materials and Methods), enabling comprehensive mutational (fig. S1) and transcriptomic profiling. In parallel, we used low-pass whole-genome sequencing (WGS; Materials and Methods) to assess copy number alterations (CNAs) profiles across the entire cohort and within TP53-mutated cases (n = 64), allowing the identification of events specifically associated with the mutation (Fig. 1C). Notably, losses of chromosome 3p appeared as one of the most frequent large-scale deletions in TP53-mutated cases (16 of 64, 25% of samples with available WGS), alongside known associated hotspots on 5q, 7q, and 17p.

Recurrent deletions of chromosome 3p co-occur with -5/del(5q) and TP53 alterations

The -3/del(3p) subgroup was composed of 18 AML specimens comprising 14 cases with partial deletions of chromosome 3p and 4 cases with complete loss of chromosome 3 (included based on shared cytogenetic, clinical, and transcriptomic characteristics). Among these specimens, 15 harbored TP53 mutations, of which 11 showed complete loss of wild-type TP53 expression and 3 presented compound heterozygous TP53 mutations.

To specifically characterize -3/del(3p) AML, all Leucegene specimens presenting deletions of the chromosome (whole chromosome 3, n = 4; or limited to the small arm, n = 14) were subjected to further analysis (Materials and Methods and table S2 for detailed information about the cohort). The median log₂ copy number profile revealed a core commonly deleted region (CDR) spanning 3p22.1-3p13, corresponding to the most consistently shared segment across del(3p) specimens (Fig. 2A). While deletion boundaries were heterogeneous, many cases also exhibited an extended deletion footprint reaching 3p24.3 and encompassing RPL15 (3p24.2).

Fig. 2. Genomic architecture and mutational landscape of -3/del(3p) AML.

(A) Chromosome 3p deletion footprint of -3/del(3p) specimens obtained by plotting median log₂ copy ratio calculated for windows of 10 kb (depicted by gray dot) along chromosome 3. Genomic positions are indicated along the x-axis and the schematic representation of chromosome 3 is drawn at a log₂ copy ratio value of 0 corresponding to a normal diploid state. Blue solid segments represent the median log₂ copy ratio calculated for overlapping windows of 10 Mb. (B) Left: Upset plot representation of the intersect between specimens presenting large 3p, 5q, 7q, or 17p deletions. Orange lines and bars depict intersects implying the del(3p) group. Right: Bubble plot representation of associations between specimen 3p, 5q, 7q, or 17p deletions and small mutations. -3/del(3p) specimens are excluded from other tested groups; for each tested group, the remaining AML specimens of the Leucegene cohort are used as control group. Bubble sizes and colors are representative of association odds and P values, respectively. Only variables presenting a significant association (positive or negative) with one of the groups are represented. (C) Mutation heatmap of -3/del(3p) AML. Genes (y axis) composing the heatmap were either mutated in one or more -3/del(3p) specimens (x axis). Genes were ordered (from top to bottom) based on their mutation frequencies (indicated by the gray bar graph on the right). Specimens were grouped according to their mutation status (from left to right). Co-occurring large deletions of chromosomes 5, 7, and 17; TP53 copy status; and expression quartiles are indicated in two panels at the top of the figure. aneup., aneuploidy, comp; het.: compound heterozygous; cnLOH, copy neutral loss of heterozygosity; perc., percentile. See the legend at the bottom for the color code.

The group was essentially composed of AML with myelodysplasia-related changes (n = 16/18, P < 1 × 10⁻⁴; Table 1), lower white blood cell counts than other AML (P < 5 × 10⁻³; Table 1) and an association with erythroid leukemia M6 morphology under the French-American-British (FAB) classification (P < 5 × 10⁻³; Table 1). All -3/del(3p) specimens harbored a complex karyotype (CK) and were associated with adverse cytogenetic risk (n = 18/18, P << 1 × 10⁻⁴; Table 1). Most presented -5/del(5q) alterations (n = 16/18; versus other AML, P << 1 × 10⁻⁴; versus CK, P < 0.05), suggesting cooperative losses, and often carried additional -17p and -7/del(7q) (n = 9 and 8 of 18, respectively) (Table 1 and Fig. 2, B and C).

Table 1. Characteristics of the del(3p) AML cohort.

AML-MRC, acute myeloid leukemia with myelodysplasia-related changes; CK, complex karyotype; FAB, French-American-British classification; WBC, white blood cell count; WHO, World Health Organization; Pos, positive; Neg, negative; NC, not classified.

		Del(3p) (n = 18)	Other (n = 638)	Assoc. type with del(3p)	P values
Age (median, range)		58.5 (39–82)	57 (17–87)	–	–
Gender	Female	5	269	–	–
	Male	13	369	–	–
WBC (median, range)		8.9 (1.5–128.1)	28.5 (0.7–447.0)	–	<5 × 10−3
Cytogenetic risk	Undetermined	0	2	–	–
	Adverse	18	154	Pos	<<1 × 10−4
	Intermediate	0	402	Neg	<1 × 10−4
	Favorable	0	80	–	–
FAB	M0	2	40	–	–
	M1	3	157	–	–
	M2	2	96	–	–
	M3	0	30	–	–
	M4	1	104	–	–
	M5	0	87	–	–
	M6	3	7	Pos	<5 × 10−3
	M7	0	3	–	–
	NC/other	7	114	–	–
WHO classification 2016	AML-MRC	16	175	Pos	<1 × 10−4
Subgroup	-5/del(5q)	16	52	Pos	<<1 × 10−4

Expectedly, TP53 was the most frequently mutated gene (n = 15 of 18; Fig. 2, B and C, and table S2) and presented a strong enrichment in this subgroup when compared to other AML (P << 1 × 10⁻⁴; Fig. 2C). Odds to identify TP53 mutations were higher for AML with -3/del(3p) alterations than for specimens of the -5/del(5q) group (71x versus 36x), suggesting a strong cooperative effect between p53 loss of function and deletions of chromosome 3p (Fig. 2B). As previously observed (10, 11), the functional consequence of TP53 alteration in -3/del(3p) AML was most often biallelic inactivation, driven by co-occurring hemizygous deletions or copy-neutral loss of heterozygosity, resulting in homozygous expression of the mutant allele in 11 of the 18 cases (Fig. 2C and table S2). In addition, three cases harbored compound heterozygous mutations (Fig. 2C). Overall, only 1 of the 18 -3/del(3p) cases showed no detectable TP53 alteration at the DNA or RNA level.

While available data did not allow concluding on the acquisition timeline of main co-occurring alterations, copy number ratios indicated that 3p deletions were clonal, displaying values consistent with those of 5q events and VAFs of TP53 mutations (fig. S2). Of note, patients in the -3/del(3p) had very poor overall survival (OS) with no significant difference compared to patients with -5/del(5q) without alteration of chromosome 3p (fig. S3 and table 3).

-3/del(3p) alteration is associated with a global reduction in RPG expression

A differential expression analysis comparing the del(3p) subgroup (n = 18) to other AML (n = 638) identified a general down-regulation of cytosolic RPGs [n = 99, log fold change (logFC) < 0] with 65 RPGs showing a significant expression drop [n = 65, logFC < −0.5, false discovery rate (FDR) < 0.05; Fig. 3A and table S4]. As expected, RPGs located on chromosome 3p (Fig. 2A) were among the most consistently differentially expressed genes (DEGs) in -3/(del3p) AML. These included RPL15 (3p24.2, top 1 downexpressed gene) and RPL14 (3p22.1, top 3), both associated with the MDS (myelodysplastic syndrome)- and AML-predisposing syndrome Diamond-Blackfan anemia (DBA) when mutated (12–14), a blood disorder characterized by a selective reduction of erythroid precursors and progenitors provoked by ribosomopathies. Additionally, the 5q-encoded MDS-related haploinsufficient driver RPS14 (5q33.1) (15) and RPS23 (5q14.2) were also among the top DEGs, which is consistent with the frequent co-occurrence of del(5q) in -3/del(3p) cases. The respective contributions of del(3p) and del(5q) to ribosomal gene deregulation are further dissected below using integrative analyses.

Fig. 3. Transcriptomic analysis of 3p deletion and coordinated down-regulation of ribosomal genes in AML.

(A) Volcano plot representation of the differential expression analysis conducted on RNA-seq data comparing del(3p) AML versus other AML of the Leucegene cohort. The horizontal dashed line indicates an adjusted P value of 0.01 and vertical dashed lines indicate log fold change (logFC) of 0.5 and −0.5. Orange and blue dots correspond to genes significantly over- and underexpressed (logFC > |0.5|, FDR < 0.01), respectively. Triangles depict differentially expressed RPGs (color code indicates the location: blue, chromosome 3; turquoise, chromosome 5; green, chromosome 17; and dark blue, other chromosomes). (B) Density histogram representation of the pairwise correlation of cytosolic RPGs. (C) Volcano plot representation of enrichments obtained from the gene set enrichment analysis (GSEA) comparing -3/del(3p) AML to other AML and using WikiPathways as gene set. NES, normalized enrichment score. Significant hits with positive or negative NES are depicted as orange and blue dots, respectively.

To test for an eventual perturbation of the co-regulation network of RPGs in this context, we conducted a pairwise correlation analysis of cytosolic ribosomes (Fig. 3B and fig. S4). No major discoordination of expression in -3/del(3p) AML was observed compared to other AML and most genes remained highly correlated. Furthermore, genes carried by chromosome 3p such as RPL15, presenting a consistent expression drop in -3/del(3p) AML, maintained a comparable correlation level than observed in other AMLs (fig. S5).

A gene set enrichment analysis (GSEA) expectedly confirmed the significant down-regulation of genes involved in ribosomal production and translation pathways and, notably, revealed a strong positive enrichment of genes associated with DNA replication and cell cycle progression (Fig. 3C). This apparent paradox, coordinated suppression of ribosome biogenesis and translation alongside activation of proliferative programs, is consistent with a ribosomal stress-adapted proliferative state, in which impaired protein synthesis capacity is uncoupled from cell-cycle control, a feature described in ribosomopathies and TP53-altered malignancies. Overall, these results support the presence of a somatic ribosomopathy combined with aberrantly active proliferation in the -3/del(3p) subgroup.

Combination of chromosome 3p and 5q RPG-CNAs with mutated TP53 is associated with the weakest overall expression of cytosolic RPGs in AML

To obtain a unique and representative expression value per specimen for all the cytosolic RPGs together, we computed a ribosomal eigengene vector [first principal component (PC) of a principal components analysis of the RPGs; Materials and Methods and table S2]. Total protein synthesis was assessed by flow cytometry using puromycin labeling (Materials and Methods) and confirmed that the calculated eigengene expression was representative of the translational levels in tested cells (Fig. 4A and fig. S6). This value was therefore used as the reference variable to test associations and correlations throughout subsequent analyses (see table S5 for associations with clinical variables). Low ribosomal eigengene values were notably associated with an adverse cytogenetic risk and enriched in FAB M5 and M6 morphologic subtypes, whereas high ribosomal expressions were associated with M0 and M1 leukemias. These associations likely reflect the strong enrichment of low ribosomal eigengene expression in TP53-mutated and CK AML, rather than an independent prognostic effect of ribosomal dysfunction. Testing the distribution of this eigengene vector in groups defined by recurrent genetic anomalies identified in the Leucegene cohort (including structural variants and small mutations; n = 70 variables identified in ≥5 specimens), distinguished -3/del(3p) and, to a lesser extent, -5/del(5q) alterations from other anomalies (Fig. 4B). Consistent with this interpretation, del(5q) AML showed much less significance when del(3p) and/or TP53m co-occurrences were excluded [see del(5q)* and del(5q)^‡ in Fig. 4B]. The ribosomal eigengene did not rely on RPGs commonly deleted in the -3/del(3p) subgroup, indicating that the observed scores reflect a global transcriptional effect rather than a direct consequence of these copy number losses (fig. S7).

Fig. 4. Ribosomal gene copy number loss associates with reduced translation and TP53-linked 5q deletions in AML.

(A) Translational activity measured by puromycin incorporation [mean fluorescence intensity (MFI); blue, bottom x axis] in selected -3/del(3p) AML (n = 6) and other AML (n = 6) with low and high ribosomal eigengene values, respectively (gray, top x axis). (B) Volcano plot comparing ribosomal eigengene distributions between AML subgroups defined by recurrent genetic anomalies in the cohort (structural variants and small mutations; n = 70 variables identified in ≥5 cases) and control AML lacking the tested alteration. The x and y axes indicate mean eigengene differences [test-control (test-ctrl)] and Wilcoxon P values, respectively. Each circle represents one recurrent genetic anomaly, with circle size proportional to the frequency of specimens in the lower quartile of eigengene expression. IAK, intermediate abnormal karyotype (excluding isolated trisomy/tetrasomy 8). (C) Ribosomal eigengene expression (top) and number of RPG-CNAs (bottom, inverted axis) across -3/del(3p) and other AML groups. P values correspond to comparisons with -3/del(3p). CBF, core-binding factor AML [RUNX1::RUNX1T1, inv(16)]; NK, normal karyotype. (D) Ribosomal eigengene values stratified by RPG-CNA number (mutually exclusive groups). The orange line shows least-squares regression. The P value for extreme groups (<4 versus >8 RPG-CNAs) is indicated. (E) Volcano plots of log₂ copy ratio differences for RPGs comparing -3/del(3p), -5/del(5q) TP53-mutated (TP53m), and -5/del(5q) TP53 wild-type (TP53wt) AML versus other AML (mutually exclusive groups). Dashed lines indicate FDR = 0.01 and ±0.5 log₂ thresholds. Blue symbols denote significantly deleted RPGs (log₂ < −0.5, FDR < 0.01), highlighting chromosomes 3, 5, and 17. (F) Chromosome 5q deletion profiles in TP53m and TP53wt -5/del(5q) AML, showing median log₂ copy ratios across 10-kb windows. The common deleted region is indicated. (G) Percentage of hemizygous TP53m and wild-type -5/del(5q) cases per 5q cytoband. The black curve represents Fisher’s exact test P values (dashed line, P = 0.05).

Overall, the eigengene followed a linear trend with lowest values associated with the combined low expression of RPGs located on both chromosome 3 and 5 (fig. S8), such as RPL15 (3p24.2) and RPS14 (5q33.1), followed by (in ascending order) RPL15^low/RPS14^high, RPL15^high/RPS14^low, and RPL15^high/RPS14^high. Expectedly, -3/del(3p) cases presented significantly lower eigengene values compared to other AML subgroups (Fig. 4C, top) and were directly followed by -5/del(5q) AML mutated for TP53 with lower ribosomal gene expression than nonmutated specimens, consistent with prior evidence that ribosomal stress can favor the emergence of TP53 loss of function.

In the Leucegene cohort, the number of deleted cytosolic RPGs per specimen was found to correlate with the general ribosomal eigengene (Fig. 4D). This correlation indicates a quantitative relationship between the RPG-CNA burden and the global expression of the RPG network.

In line with these results, -3/del(3p) and -5/del(5q) AML mutated for TP53 presented the highest number of RPG-CNAs in the cohort with a median of 9 and 6 events per specimen, respectively, while nonmutated -5/del(5q) cases only presented a median value of 2 deletion events (Fig. 4C, bottom). Overall, despite few discrepancies such as PML::RARA-positive leukemias presenting a higher expression than suggested by the number of RPG losses, most subgroups showed a correlated relationship between RPG-CNAs and RPG eigengene. When analyzed together with sorted normal hematopoietic populations, -3/del(3p) AML specimens localize within the erythroid-biased progenitor range of ribosomal eigengene expression, positioning between Ery-III and Ery-II populations (fig. S9 and Materials and Methods).

Recalling the co-occurrence of copy losses in the -3/del(3p) cohort, chromosomes 3, 5, and 17 had the higher frequencies of RPG-CNAs in the subgroup, with recurrently and significantly deleted RPGs restricted to chromosomes 3p and 5q (Fig. 4E). Although chromosome 19 was the first contributor at the scale of the whole Leucegene cohort given its high density in RPGs (~0.25 RPGs/Mb versus <0.1 for other chromosomes), it showed no enrichment in the -3/del(3p) specimens (fig. S10). As for -5/del(5q) AML, both TP53-mutated and wild-type cases presented chromosomes 5 and 17 as top contributing chromosomes but with significantly stronger enrichments when mutated (fig. S10). Although additional chromosomal losses frequently observed in del(5q) AML—including chromosomes 12, 13, 16, 18, and 20—were present in TP53-mutated cases, these regions did not show enrichment for RPG-CNAs.

We and others previously identified a relationship between longer 5q deletions and alterations of TP53 (11, 16, 17), suggesting a cooperation between the copy loss of genes located in the area of chromosome 5 specifically associated with inactivation of p53. Accordingly, the comparison of 5q deletion footprints depending on TP53 status revealed that although sharing the CDR, wild-type -5/del(5q) specimens usually present events with boundaries limited from 5q14.3 to 5q33.3. In contrast, deletions in mutated cases more often extend from 5q12.1 to the chromosome end (Fig. 4, F and G) and, hence, frequently include RPS23 (5q14.2) and RPL26L1 (5q35.1)—65 and 22% of mutated and wild-type cases, respectively, are hemizygous for 5q14.2 and 5q35.1 (P < 5 × 10⁻³; Fig. 4G), which explains the specific association of their copy losses with mutated cases (Fig. 4E). Furthermore, as reported here and in our recent study on -5/del(5q) (11), most of these mutated specimens present a homozygous expression of the mutated allele due to large or focal deletions of chromosome 17p, which translated here as a significant association of RPL26 deletion located at 17p13.1 close to TP53 (Fig. 4E). Note that no significant difference in 5q deletion footprint was identified between TP53-mutated -3/del(3p) and -5/del(5q) specimens (fig. S11). Overall, these data demonstrated the importance of the combination of events on chromosomes 3p, 5q (including TP53-dependent differentially deleted regions), and 17p (including TP53 mutations) to obtain the stronger ribosomopathy-like phenotype, as observed in most -3/del(3p) specimens.

TP53-dependent events could contribute to the destabilization of ribosomal network in AML

To evaluate the coregulation of key biological processes and RPGs in AML, we assessed the correlation between the ribosomal eigengene vector and genes constituting the 50 Hallmark sets from the Human Molecular Signatures Database (MSigDB) (fig. S12A and table S6). This analysis revealed that low RPG expression is rather associated with a proinflammatory state, sets such as “complement,” “inflammatory response,” “TNF-α signaling via NF-κB,” and “interferon-γ response” being among the top negatively correlated pathways. Although this may result from the general TP53-mutated context of low eigengene expressors, cellular stress and impairment of the ribosome-dependent inhibition of the inflammatory response are also likely at play (18, 19). On the other hand, gene sets dependent on Myc (“MYC targets” V1 and V2), one of the major modulators of ribosome biogenesis (19), presented the strongest positive correlation with the eigengene (Fig. 5A and fig. S12A), while mammalian target of rapamycin (mTOR) signaling, the other master regulator (19), showed no evidence of co-modulation (fig. S12A).

Fig. 5. Association of ribosomal eigengene expression with MYC targets and ribosome-associated genes in AML.

(A) Single-gene correlation versus ribosomal eigengene (Pearson’s r, left) and corresponding density plot (right). MYC targets [V1 and V2 combined from the Hallmark set of the Human Molecular Signatures Database (MSigDB)] are indicated by blue dots (left) or bars (right). Median (Med), maximum (Max), and minimum (Min) correlation values for MYC targets are directly indicated on the left panel. The solid black curve in the right panel corresponds to the density obtained from the 48 other Hallmark sets together. (B) Pairwise correlation plot comparing a selection of MYC targets and the ribosomal eigengene expression in the Leucegene cohort (n = 656 AML). Area and blue shades of colored surfaces in pie charts (color scale indicated at the right of the panel) are representative of Pearson’s r (also indicated in each chart). (C) RACK1 and NPM1 expression in the -3/del(3p), -5/del(5q) TP53m, or TP53wt and other AML of the Leucegene cohort (left of each plot) or distributed into expression quartile subgroups (right of each plot). Median values are indicated by black lines on each dot plot. As indicated in the legend (right of the figure), the color code is representative of the ribosomal eigengene expression. P values resulting from the comparison between groups are indicated on the plot (NS, not significant). Transversal dotted lines delimit quartiles. Median ribosomal eigengene expression values for each expression quartile are directly indicated on the plots.

Running a GSEA comparing the -3/del(3p) subgroup to other AML specimens also confirmed the conjoined down-regulation of ribosomes and MYC targets (fig. S12B). More specifically, apart from several RPGs expectedly coregulated (table S6), other genes potentially contributing to the ribosomopathic phenotype were found among the top correlated MYC targets (Fig. 5B): the two elongation factors EEF1B2 and EEF2, the fibrillarin (FBL) that promotes early pre–ribosomal RNA (rRNA) processing (20) [correlation coefficient (r) = 0.90, 0.77, and 0.76, respectively; table S6 and fig. S12A], and, particularly interesting given their role and their location in the TP53-related differentially deleted region at the end of chromosome 5 (5q35.3 and 5q35.1; Fig. 4, F and G), the receptor for activated C kinase 1 (RACK1) and the nucleophosmin 1 (NPM1) (r = 0.84 and 0.71, respectively; tables S4 and S6 and fig. S12A). RACK1 is involved in translation and ribosome quality control (21–28); as for NPM1, besides its importance in AML, the dysregulation represents a particular interest here given its role in different phases of ribosome synthesis and nucleolar stress response (29–33). Detailing RACK1 and NPM1 expression profile in the whole Leucegene cohort confirmed that only low expressors reached low levels of ribosomal eigengene (Fig. 5C), with -3/del(3p) and TP53-mutated -5/del(5q) AML, which present highest frequencies of RACK1 and NPM1 deletion (Fig. 4G), showing the lowest expression levels (Fig. 5C and table S4). Expectedly, specimen 09H045 with the highest ribosomal eigengene expression of the -3/del(3p) subgroup and no alteration of chromosome 5 (table S2) also presented the highest expression of both genes in the group and was the only -3/del(3p) specimen in the top expression quartiles of the Leucegene cohort (Fig. 5C). Note that, while it has been hypothesized that a reduction of NPM1 dosage, as provoked by commonly encountered small insertions in the last exon of the genes (NPM1c), could alter ribosome synthesis (34), NPM1c alone was not associated with a variation of ribosomal expression (n_NPM1wt and n_NPM1c > 400 and 200, respectively; fig. S12C).

Reduction of ribosomal protein levels provoked by HSP90 inhibitors lead to the sensitization of low RPG expressors

Based on these findings, we hypothesized that the transcriptional and translational specificities associated with ribosomopathy-like profiles could represent a therapeutic vulnerability. To explore this, we used data from a preexisting single-dose high-throughput screening (HTS) assay, which tested over 10,000 structurally diverse compounds across 56 primary AML specimens from the Leucegene cohort (fig. S13). Notably, there was no overlap between the samples included in this screen and the -3/del(3p) cohort. Compounds exhibiting similar inhibitory profiles across specimens, potentially targeting the same cellular pathways, were grouped into compound correlation clusters (CCCs) using hierarchical clustering based on a minimum spanning tree approach (Fig. 6A and Methods).

Fig. 6. Identification of HSP90 inhibitors as ribosomal eigengene–associated vulnerabilities in AML.

(A) Icicle representation of selective hit compounds of the initial single-dose HTS assay depicting peaks of highly correlating inhibitory profiles constituting compound correlation clusters (CCCs). Concentric circles depict levels of correlation as indicated on the figure (from 0 at the center to 1 at the edge). CCC_HSP90i composed of the 17-AAG and 33 of its analogs is depicted by a blue peak. (B) Inhibition footprint (%) obtained for compounds composing the CCC_HSP90i. Orange and blue lines represent data corresponding to the 17-AAG and its analogs, respectively. (C) Volcano plot representation of the correlation level between compounds constituting the CCCs and the ribosomal eigengene expression values obtained for tested specimens. The horizontal dashed line indicates a P value of 0.05 and vertical dashed lines indicate correlation of 0.2 and −0.2. Orange and blue dots correspond to compounds significantly correlated and inversely correlated (correlation > |0.2|, P < 0.05), respectively. Dark blue diamonds depict compounds from the CCC_HSP90i composed of the 17-AAG (indicated by a larger blue diamond) and 33 of its analogs. (D) Ribosomal eigengene expression values according to tier 1 (“sensitive” tier) and tier 3 (“resistant” tier) sensitivity groups for 17-AAG (left) and the best analog (right), i.e., presenting the strongest difference between the two groups. Median values are indicated by black lines on each dot plot. P values resulting from the comparison between groups are indicated on the plot. (E) Ribosomal eigengene expression values according to tier 1 (sensitive tier, n = 114 cell lines) and tier 3 (resistant tier, n = 98) 17-AAG sensitivity groups determined using data obtained from The Genomics of Drug Sensitivity in Cancer (GDSC) database.

We then evaluated the correlation between RPG eigengene values, calculated for each tested specimen, and the inhibitory profiles of each identified CCC. This analysis highlighted a chemical cluster centered on 17-AAG/tanespimycin, a HSP90 chaperone inhibitor (HSP90i) that has been investigated in multiple clinical trials for various cancer types (35–39), along with 33 of its analogs. Notably, this cluster showed the strongest and most consistent anticorrelation with the RPG eigengene, with 31 of the 34 compounds in the CCC displaying significant anticorrelation (Fig. 6, A to D). These findings suggest an association between low ribosomal expression and sensitivity to HSP90 inhibition.

To validate this observation, we leveraged The Genomics of Drug Sensitivity in Cancer (GDSC) database (40) as an independent dataset. Despite the substantial genetic variability among the >200 cancer cell lines tested, the association between ribosomal expression and HSP90 inhibitor sensitivity was confirmed (Fig. 6E).

We further conducted a targeted chemical screen for validation, testing three HSP90 inhibitors—17-AAG, geldanamycin, and alvespimycin—on a cohort of -3/del(3p) AML samples (n = 13) and control specimens (n = 25) with varying ribosomal expression profiles (Fig. 7A and table S7). The three compounds exhibited similar response profiles, with the strongest correlation observed between 17-AAG and alvespimycin (Pearson’s r = 0.79; Fig. 7B).

Fig. 7. Enhanced sensitivity of -3/del(3p) AML to HSP90 inhibition and ribosomal protein depletion.

(A) Heatmap of responses to HSP90 inhibitors (17-AAG, geldanamycin, and alvespimycin). Colors represent z-scores derived from the median inhibitory concentration (IC₅₀) values (scale shown). The bottom annotation indicates specimen subgroup [-3/del(3p) or control AML]. Columns are ordered by unsupervised hierarchical clustering of IC₅₀ values. (B) Correlation between responses to 17-AAG and alvespimycin (Pearson’s r = 0.79). The dashed line indicates least-squares regression. (C) Average response to HSP90 inhibitors [avg(HSP90i); mean of rescaled IC₅₀ values from −1 to 1] according to subgroup [-3/del(3p) versus control AML], TP53 status, and ribosomal eigengene expression (tier 1, low; and tier 3, high). (D) Representative Western blots of RPS14, RPL14, RPL29, and α-tubulin (TUBA; loading control) in U937 cells treated for 24 hours with DMSO or HSP90 inhibitors. (E) Representative Western blots of RPL29 and TUBA in primary AML cells [(A) to (D): control AML; (E) to (H): -3/del(3p)] treated for 24 hours with DMSO or HSP90 inhibitors. (F) Correlations between RPL29 protein levels (normalized to TUBA) and IC₅₀ values for each HSP90 inhibitor. Pearson’s r values are indicated; dashed lines represent least-squares regression. (G) Ex vivo proliferation of primary -3/del(3p) (gray) and control AML (white) following exposure to DMSO, geldanamycin, or alvespimycin. Cell counts were normalized to Fresh (D0) input and expressed as fold change at days 1 and 4 (NS, not significant). (H) Ex vivo viability of primary -3/del(3p) (gray) and control AML (white) after treatment, normalized to Fresh (D0), assessed at days 1 and 4.

-3/del(3p) specimens showed a significantly increased sensitivity to HSP90 inhibition compared to controls, independently of p53 status alone (Fig. 7, A and C). Overall, although this smaller screen did not yield a statistically significant association between ribosomal expression and sensitivity, these results remained consistent with the initial findings (Fig. 7C).

Last, drawing on prior reports indicating that HSP90 inhibition or knockout leads to ribosome degradation or reduced ribosomal protein levels (41, 42), we investigated whether this mechanism could explain the observed sensitization of AML with low RPG expression. Western blot analysis of three ribosomal proteins (RPS14, RPL14, and RPL29) in U937 cells treated with 17-AAG, geldanamycin, and alvespimycin revealed a marked reduction in ribosomal protein levels across all three HSP90 inhibitors (Fig. 7D). Complementarily, Western blotting of RPL29 in primary AML cells with (n = 4) and without (ctrl AML, n = 4) del(3p) alterations was performed following treatment with the three HSP90 inhibitors (Fig. 7E and fig. S14). In control AML specimens, a reduction in RPL29 levels was observed following treatment. Furthermore, these levels were found to correlate with the median inhibitory concentration (IC₅₀) values (r ≥ 0.9; Fig. 7F), indicating that the sensitivity to these inhibitors is likely dependent on RPG abundances. In del(3p) AML cells, which already exhibited markedly low baseline RPL29 levels, treatment with HSP90 inhibitors resulted in a further reduction, decreasing RPL29 to undetectable or virtually undetectable levels (Fig. 7E). Although HSP90 inhibition reduced RPL29 levels in both control and -3/del(3p) AML specimens, only samples with low baseline ribosomal protein abundance exhibited pronounced functional vulnerability, indicating that HSP90 inhibition exacerbates a preexisting ribosomal protein deficiency rather than inducing a lethal ribosomal defect de novo.

To functionally validate this vulnerability, we assessed the effects of alvespimycin and geldanamycin on ex vivo proliferation and viability of primary AML specimens (Materials and Methods). At 24 hours, proliferation differences between -3/del(3p) and control AML were modest and variable across conditions, whereas after 4 days of exposure, -3/del(3p) AML samples displayed a marked reduction in proliferative capacity under both alvespimycin and geldanamycin treatment (Fig. 7G). This effect was not attributable to vehicle exposure alone and was consistently observed across samples. In parallel, viability assays revealed a progressive and pronounced loss of viable cells in -3/del(3p) AML after 4 days of HSP90 inhibition, while control AML samples retained substantially higher viability (Fig. 7H). Note that the -3/del(3p) sample displaying the highest proliferative response corresponds to the case with the highest ribosomal eigengene expression within this subgroup [−0.09 compared to −0.17 on average for the other -3/del(3p) cases], consistent with reduced sensitivity in samples with relatively preserved ribosomal protein expression. Together, these results indicate that sustained HSP90 inhibition selectively compromises both proliferation and survival of -3/del(3p) AML cells in a time-dependent manner.

To determine whether this vulnerability extends in vivo, we evaluated the efficacy of the HSP90 inhibitor 17-AAG in patient-derived xenograft (PDX) models established from independent -3/del(3p) AML specimens (Fig. 8A and Materials and Methods). Across models, treatment with 17-AAG induced a dose-dependent reduction of human CD45⁺ leukemic cells in total bone marrow (BM) and peripheral blood compared to vehicle-treated controls (Fig. 8 and fig. S15). Treatment was generally well tolerated, with no major toxicity observed. These results demonstrate that HSP90 inhibition with 17-AAG is sufficient to impair -3/del(3p) AML progression in vivo as a single agent.

Fig. 8. In vivo evaluation of 17-AAG efficacy in a -3/del(3p) AML PDX model.

(A) Experimental design of the 17-AAG efficacy study in mice engrafted with -3/del(3p) #1 PDX cells (n = 5 mice per group; lines indicate treatment time points; BM asp, BM aspiration). (B) Body weight follow-up during the 17-AAG efficacy study (n = 5 mice, means ± SD). (C) Dot plot representation of the percentage of human CD45⁺ cells in BM aspirates performed 3 days before the treatment initiation, after 2 weeks of treatment as well as in total BM and peripheral blood (PB) at sacrifice (dots represent individual mice, means ± SD). (D) Representative FACS profiles of human CD45⁺ engrafted cells in total BM and PB at sacrifice.

Overall, these findings highlight the significant ribosomal protein deficiency in -3/del(3p) AML and provide a mechanistic basis for their heightened sensitivity to HSP90 inhibition. More generally, they support the therapeutic potential of HSP90 inhibition in ribosomopathy-like AML.

DISCUSSION

Although loss of chromosome 3p has been previously noted as a recurrent coaberration with 5q deletions in AML (43), we provide a detailed characterization of a TP53-altered AML subset defined by hemizygous deletions of RPGs on 3p and 5q. These include RPSA (essential for 40S ribosomal subunit assembly and stability), RPL29 (a 60S subunit component), RPL14 and RPL15 (60S subunit components and DBA-associated genes), as well as the 5q-syndrome-associated haploinsufficient gene RPS14 (15, 44). These deletions were associated with a general reduction in cytosolic RPG expression, decreased translation, creating an overall ribosomopathy-like phenotype. Consistent with this, our study extends beyond RNA-based inference by directly assessing global protein synthesis through puromycin incorporation, thereby capturing a functional consequence of ribosomal perturbation rather than relying on steady-state protein abundance measurements. More broadly, our findings contribute to defining the somatic ribosomopathy in AML, suggesting that it represents an underestimated driving mechanism in leukemogenesis, with the extent of ribosomal regulatory network perturbations proportional to the RPG-CNA burden. This quantitative association produces a continuum of phenotypes, with -3/del(3p) TP53-altered AML representing an extreme manifestation. Results also suggest that, although TP53 mutations are linked to the strongest phenotype, they are not drivers of ribosomopathy. Hence, this work highlights that TP53-mutated AML can be categorized as ribosomopathic or not with a full spectrum of severity that depends on the associated deletions. These observations provide a framework for stratifying AML cases according to ribosomal dysfunction, with potential implications for therapeutic vulnerabilities. In this context, it is important to consider the clinical experience with HSP90 inhibitors. Several HSP90 inhibitors, including the first-generation compound 17-AAG (tanespimycin), have been evaluated in early-phase clinical trials across multiple cancer types, including hematological malignancies (35–39). While these studies demonstrated target engagement and biological activity, their clinical development has been limited by modest single-agent efficacy and the lack of robust biomarkers to guide patient selection. Recent work has further suggested that sensitivity to HSP90 inhibition is highly context-dependent and may reflect underlying proteostasis stress and epichaperome dependency rather than uniform HSP90 addiction, notably in TP53-mutant AML (45). In this context, our findings indicate that AML cases characterized by RPG haploinsufficiency and impaired translational capacity may represent a biologically defined subgroup with heightened vulnerability to HSP90 inhibition, providing a rationale for biomarker-guided therapeutic stratification.

From a mechanistic perspective, the consequences of disrupted ribosome biogenesis and function in cancer cells have only recently been recognized. However, emerging evidence suggests that these disruptions may drive the oncogenic process through multiple mechanisms (18, 46), potentially enabling a paradoxical shift from an initially hypoproliferative state to a proliferative state (47), matching the expression signature of del(3p) AML cells. This likely implies the selective translation of mRNAs encoding genes crucial for tumor survival and proliferation, such as oncogenes and cell cycle regulators, compounded by genomic instability as metabolic alterations associated with ribosomal dysfunction impairs the DNA damage response, increasing mutation rates and driving the acquisition of additional oncogenic alterations (48). Furthermore, in a leukemic context, stress resistance associated with low biosynthetic activity may confer microenvironmental advantages to preleukemic and leukemic clones, allowing to outcompete normal hematopoietic cells for limited resources (49).

Recent large-scale studies (n > 10,000 tumor genomes) revealed that more than 40% of analyzed specimens harbored hemizygous deletions of RPGs (50), including frequent deletions of RPSA, RPL14, and RPL29 on chromosome 3. These results suggest that ribosomal dysregulation could play an underestimated role in cancer onset and progression. More specifically, given that tightly regulated protein synthesis is necessary for maintaining the HSC pool and allowing lineage commitment, alterations in protein synthesis rates can lead to HSCs depletion and leukemogenesis (51, 52). DBA, a syndrome caused by mutations in RPGs or ribosome biogenesis factors that critically impair erythropoiesis by compromising the high protein supply needed during erythroid maturation, illustrates how a reduction in ribosome availability leads to hematopoiesis disruption (53, 54). We show that similar mechanisms appear to operate in AML, where multiple RPG deletions seem to contribute to global ribosomal defects and are comparable to the effects of inactivating mutations (50). Moreover, AML cases with low ribosomal protein expression showing enrichment in M6 morphologic subtype particularly align with observations derived from studies on DBA.

As with most ribosomopathy-like tumors (50), -3/del(3p) AML exhibited p53 loss of function, which allows cells to evade p53-dependent nucleolar stress responses triggered by ribosomal defects, thereby bypassing growth arrest (55, 56). Beyond this, the interaction between ribosomal defects and p53 loss of function takes on particular relevance in -5/del(5q) AML, as our previous work demonstrated that TP53-mutated AML specimens tend to harbor longer 5q deletions (11). We now demonstrate that these extended deletions include RPG-CNAs located in differentially altered regions of the chromosome arm, likely intensifying the ribosomopathy phenotype. Together, these observations uncover insight into one possible cooperative oncogenic mechanism linking 5q deletions and TP53 mutations in AML. In addition to their quantitative impact on ribosomal protein depletion, 5q deletions in TP53-mutated cases more frequently affect genes such as NPM1 and RACK1 (located at 5q35), whose down-regulation could further amplify ribosomal defects. Depletion of RACK1, a core ribosomal protein of the small ribosomal subunit, has notably been associated with alteration of translational capabilities and ribosome-associated quality control (21–28). Meanwhile, the ubiquitous role of NPM1 in rRNA transcription, maturation, and transport (29, 30) suggests its dysregulation in TP53-mutated specimens, showing enrichments of copy alterations, as a possible part of the cooperative set of driving events.

Beyond providing mechanistic insight, these results also reveal a potential clinical perspective for a subset of TP53-altered AML. While ribosomopathies in AML can modulate the expression of proapoptotic or drug-target proteins, potentially contributing to resistance to chemotherapy, they may also represent a lever for tumor sensitization. Through integrative chemical-genetic screening, we identified HSP90 inhibition, which exacerbates ribosomal defects by limiting ribosomal protein availability, as selectively toxic to cells with low RPG expression, including TP53-altered AML with -3/del(3p).

In this context, HSP90 inhibition does not induce an acute cytotoxic response but instead progressively exacerbates an underlying ribosomopathy-like state, leading to delayed yet pronounced impairment of proliferation and viability in -3/del(3p) AML cells. The time-dependent nature of this response is consistent with a model in which sustained disruption of ribosomal protein homeostasis gradually compromises cellular fitness rather than triggering immediate cell death. The observation that single-agent HSP90 inhibition reduces leukemic burden in vivo supports the notion that this dependency is not restricted to ex vivo conditions and can be therapeutically exploited.

Together, our data support the existence of a TP53-altered, ribosomopathy-like AML subtype. This not only refines our understanding of leukemogenesis in this high-risk group but also provides a rationale for exploring tailored proteostasis-targeting strategies in this otherwise refractory disease.

MATERIALS AND METHODS

Primary AML specimens

The Leucegene project (https://leucegene.ca/) is an initiative approved by the Research Ethics Boards of Université de Montréal and Maisonneuve-Rosemont Hospital. Leucegene AML samples (n = 691, BM or blood samples) were collected from 656 patients between 2001 and 2019 and characterized by the Quebec Leukemia Cell Bank after obtaining an Institutional Research Ethics Board–approved protocol with informed consent according to the Declaration of Helsinki. The Quebec Leukemia Cell Bank is a biobank certified by the Canadian Tissue Repository Network. Cytogenetic aberrations and composite karyotypes were described according to the International System for Human Cytogenomic Nomenclature 2020 guidelines (57). Specimens were classified on the basis of the latest recommendations from the World Health Organization and the ICC of Myeloid Neoplasms and Acute Leukemias (unless otherwise specified) (6, 58). Using the analytic approach on WGS data described below, a recurrent large deletion of chromosome 3p was identified as shared by 14 AML cases. The -3/del(3p) cohort was completed to a total of 18 specimens (table S2) by adding four cases with a complete loss of the chromosome 3, including 2 specimens for which cytogenetics information only was available, presenting similar clinical and transcriptomic characteristics. The -5/del(5q) cohort was composed of 68 specimens [n = 52, -3/del(3p) excluded], of which 49 presented a TP53 mutation.

Survival analysis and Leucegene AML prognostic cohort

Survival analysis was conducted using the Leucegene AML prognostic cohort (n = 470; table S3), a subset of the Leucegene AML cohort including patients with newly diagnosed AML, excluding acute promyelocytic leukemia (APL) and patients who were treated with an intensive chemotherapy regimen with a curative intent (mostly 7+3 regimen backbone). OS was evaluated using the Kaplan-Meier method. Cox proportional hazards models were used to calculate hazard ratios between subgroups with 95% confidence intervals. OS was defined as time from diagnosis to time of death or last follow-up.

Low-pass WGS and analysis

Tumor genomic DNAs (gDNAs; n = 611) were sequenced on NovaSeq6000 S4 (paired-end 150 base pairs). Alignment to GRCh38 was done using the BWA aligner (v0.7.12) (59), polymerase chain reaction duplicates were removed using Picard (MarkDuplicates) (60), and a GATK (v4.1.0, BaseRecalibrator) (61) base quality score recalibration was applied. A mean depth coverage ~5× was reached for each sample. Identification of regions of genomic gains and losses was done using FREE-Copy number caller (v11.5) (62). The optimization of algorithm parameters (breakPointThreshold = 1.4; window = 100,000; step = 13,000; readCountThreshold = 20; contaminationAdjustment = “TRUE”; minMappabilityPerWindow = 0.95; breakPointType = 4; minCNAlength = 1) was conducted using known alterations as reference. The concatenation of adjacent CNAs was done using the “merge” option of BedTools (v2.25.0) (63). For each gene located in identified CNA regions, a median log₂ copy ratio (log₂R) was then calculated by considering small windows at least partly overlapping a segment centered on the gene and extended by 25 kb on each side.

RNA-seq and analysis

RNA-seq libraries were constructed according to TruSeq Protocols (Illumina), and sequencing was performed using an Illumina HiSeq 2000/4000 instrument. Trimming of sequencing adapters and low-quality bases was done using the Trimmomatic (v0.38) tool (64). Resulting reads were mapped to the reference using the RNA-seq aligner STAR (v2.7.1) (65), and quantification of gene and isoform expression was performed using RSEM (v1.3.2) (66). The limma package and its Voom method (67) were used to conduct differential expression analysis. Point mutations were identified from RNA-seq data as previously described (10). The GSEA software (68) was used to conduct the GSEA. Genes were preranked according to the signed P values obtained from the differential expression analysis [sign(logFC)*−log10(P value)]. The WikiPathways and Hallmark sets from the MSigDB were used for the analysis.

Ribosomal eigengene expression calculation

The PCs of log₂-transformed, normalized expression data (TPM) for RPGs (Kyoto Encyclopedia of Genes and Genomes gene sets: www.gsea-msigdb.org/gsea/msigdb/cards/KEGG_RIBOSOME) were calculated using the prcomp function in R. The ribosomal eigengene was defined as the value of the PC1 for each sample. Eigengene expression values were then rescaled to generate a vector ranging from −1 to 1. Note that ribosomal eigengene values did not rely on the expression of RPGs that were found as commonly deleted in the -3/del(3p) group (fig. S7). Normal populations data were obtained from Maiga et al. (69).

Translational level assay

Puromycin (10 μg/ml) was incorporated in primary specimens for 30 min. Cells were then fixed, and permeabilized, and intracellular staining of puromycin was achieved using Alexa Fluor 647–labeled anti-puromycin antibody (Merck Millipore, no. MABE343). Relative rates of protein synthesis were assessed by flow cytometry by calculating the geometric mean fluorescence intensity of puromycin signals after subtracting background fluorescence for each specimen.

HTS assay

HTS assay of more than 10,000 structurally diverse compounds, including molecules obtained from a pharmaceutical partner (Bristol Myers Squibb) as well as selected groups of molecules proprietary to Institute for Research in Immunology and Cancer (IRIC), was conducted in 56 primary AML specimens (fig. S13) and 2 normal control samples (expanded CD34⁺ cells from cord blood samples). Due to the inclusion of proprietary compounds from an industrial partner, the full chemical identities of all screened molecules are not available; compounds were provided as internal identifiers, with annotation disclosed only for selected hit compounds or clusters of interest. Cell culture and viability assays were conducted as previously described (70). Briefly, for CCC determination, percentage of inhibition data for selective hit compounds was rank transformed and clustered by minimum spanning tree. Groups of molecules in icicle peaks with σ > 0.9 were selected and further filtered for elimination of outlier compounds by selection of profiles correlating with r > 0.9 to the median of the group. The remaining compounds were selected as part of CCCs.

Validation chemical screen

Freshly thawed primary AML specimens were used for chemical screens following the procedure previously described by our group (71). Compounds were dissolved in dimethyl sulfoxide (DMSO) at 10 mM, diluted in medium immediately before use, and added to seeded cells in serial dilution to determine IC₅₀ values (8 dilutions, 1:3, 10 μM down to 4.5 nM) or at the unique concentration of 1 μM, in duplicate wells. Cell viability was evaluated after 6 days in culture using the CellTiterGlo assay (Promega) and compared to DMSO-treated cells.

Western blots

Alvespimicin (Selleckchem), geldanamycin (MedChemExpress), and 17-AAG (MedChemExpress) were used at 1 M for U937 cells (RRID:CVCL_0007) and 500 nM for primary specimens. U937 cells were treated with indicated compounds for 24 hours. For Western blot analysis, U937 cells were harvested and lysed in total protein extraction lysis buffer [25 mM tris (pH 7.5), 150 mM NaCl, 1% NP-40, and protease inhibitors]. Antihuman antibodies were used to detect RPS14 (Thermo Fisher Scientific, no. PA5-77004; RRID: AB_2720731), RPL14 (Thermo Fisher Scientific, no. A305-051A; RRID: AB_2621245), RPL29 (Thermo Fisher Scientific, no. PA5-118248; RRID:AB_2902848), and tubulin (TUBA; Cell Signaling Technology, no. 2144; RRID:AB_2210548).

AML specimen treatment

AML cells were thawed at 37°C in Iscove’s modified Dulbecco’s medium (IMDM) containing 20% fetal bovine serum and deoxyribonuclease I (100 μg/ml). Cells were cultured for 4 hours in the presence or the absence of HSP90 inhibitors (alvespimycin, geldanamycin, and 17-AAG; 500 nM) in IMDM, 15% BIT (bovine serum albumin, insulin, and transferrin; STEMCELL Technologies, 09500), stem cell factor (SCF; 100 ng/ml; Shenandoah, 100-04), Fms-like tyrosine kinase 3 ligand (FLT3L; 50 ng/ml; Shenandoah, 100-21), IL-3 (20 ng/ml; Shenandoah, 100-80), granulocyte colony-stimulating factor (20 ng/ml; Shenandoah, 100-72), 10⁻⁴ M β-mercaptoethanol, gentamicin (50 μg/ml), and ciprofloxacin (10 μg/ml). Cells were then subjected to Western blot analysis.

Ex vivo viability and proliferation

AML cells were seeded at a density of 5 × 10⁵ cells/ml and cultured in the presence of the HSP90 inhibitors alvespimycin or geldanamycin (200 nM) for 1 or 4 days. Absolute cell counts were determined at each time point, and cell viability was assessed by 7-aminoactinomycin D (BD Biosciences, 559925; RRID:AB_2869266) exclusion using flow cytometry. Proliferation was expressed as fold change relative to day 0.

In vivo experiments on PDX models

All animal procedures complied with recommendations of the Canadian Council on Animal Care and were approved by the Deontology Committee on Animal Experimentation at the University of Montreal. NSG mice were purchased from The Jackson Laboratory and maintained in a pathogen-free animal facility. Eight- to 12-week-old sublethally irradiated female mice were randomly assigned to experimental groups and transplanted via tail vein injection with 0.5 million of case #1 or 1.1 million of case #2 PDX cells. Three to 4 weeks after transplantation, treatments were initiated with vehicle or 17-AAG at 50 or 80 mg/kg by oral gavage (10% DMSO in 0.2% carboxymethylcellulose, 10 ml/kg), administered 5 days per week for several weeks. Each treatment group included five mice. Mice were daily monitored, including body weight follow, to assess tolerability of the treatment. Humane endpoints included >15 to 20% body weight loss, signs of distress, or impaired mobility. BM aspiration was performed after 2 weeks of treatment to monitor in vivo leukemia progression. At sacrifice, total BM cells and PB (only for case #1 PDX model) were collected and analyzed by flow cytometry using a BD FACSCanto II cytometer. Leukemic engraftment was quantified using the following antibodies: antihuman CD45 Pacific Blue (BioLegend, no. 304029; RRID:AB_2174123), antihuman CD33 PE (phycoerythrin; BD Biosciences, no. 555450; RRID:AB_395843), and anti-mouse CD45.1 APC (allophycocyanin)-eFluor780 (eBioscience, no. 47-0453-82; RRID:AB_1582228). Flow cytometry data were analyzed using FlowJo software.

Statistical analysis

All statistical analyses were performed using R (version 4.3.2). Depending on the data structure and distribution, Fisher’s exact test, Student’s t test, or the Mann-Whitney U test was applied, as appropriate. A two-sided P < 0.05 was considered statistically significant. For graphical representation, P values lower than 1 × 10⁻⁴ were reported as “P < 1 × 10⁻⁴,” and the notation “P << 1 × 10⁻⁴” was used to denote particularly strong associations.

Supplementary Materials

The PDF file includes:

Figs. S1 to S15

Legends for tables S1 to S7

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S7