Integrated genetic,epigenetic,and immune landscape of TP53 mutant AML and higher risk MDS treated with azacitidine

Abstract

Background:

TP53 mutations are associated with an adverse prognosis in acute myeloid leukemia (AML) and higher-risk myelodysplastic syndromes (HR-MDS). However, the integrated genetic, epigenetic, and immunologic landscape of TP53-mutated AML/HR-MDS is not well defined.

Objectives:

To define the genetic, epigenetic, and immunologic landscape of TP53-mutant and TP53 wild-type AML and HR-MDS patients.

Design:

Post hoc analysis of TP53-mutant and TP53 wild-type patients treated on the randomized FUSION trial with azacitidine ± the anti-PD-L1 antibody durvalumab.

Methods:

We performed extensive molecular, epigenetic, and immunologic assays on a well-annotated clinical trial dataset of 61 patients with TP53-mutated disease (37 AML, 24 MDS) and 144 TP53 wild-type (89 AML, 55 MDS) patients, all of whom received azacitidine-based therapy. A 38 gene-targeted myeloid mutation analysis from screening bone marrow (BM) was performed. DNA methylation arrays, immunophenotyping and immune checkpoint expression by flow cytometry, and gene expression profiles by bulk RNA sequencing were assessed at baseline and serially during the trial.

Results:

Global DNA methylation from peripheral blood was independent of TP53 mutation and allelic status. AZA therapy led to a statistically significant decrease in global DNA methylation scores independent of TP53 mutation status. In BM from TP53-mutant patients, we found both a higher T-cell population and upregulation of inhibitory immune checkpoint proteins such as PD-L1 compared to TP53 wild-type. RNA sequencing analyses revealed higher expression of the myeloid immune checkpoint gene LILRB3 in TP53-mutant samples suggesting a novel therapeutic target.

Conclusion:

This integrated analysis of the genetic, epigenetic, and immunophenotypic landscape of TP53 mutant AML/HR-MDS suggests that differences in the immune landscape resulting in an immunosuppressive microenvironment rather than epigenetic differences contribute to the poor prognosis of TP53-mutant AML/HR-MDS with mono- or multihit TP53 mutation status.

Trial registration:

FUSION trial (NCT02775903).

Keywords

AML gene expression immune phenotype MDS

Introduction

Mutations in TP53 are found in 10–15% of patients with acute myeloid leukemia (AML) and myelodysplastic syndromes (MDS).^1,2 TP53 mutations have been associated with increased resistance to conventional cytotoxic chemotherapy and venetoclax-based therapy, and poor survival even after allogeneic hematopoietic cell transplantation.^1,3–5 Although frequently associated with complex karyotypes and dismal outcomes, recent studies have suggested that TP53 mutations in AML and higher risk (HR)-MDS are heterogeneous with a variable impact on prognosis.^6–9 The data are conflicting regarding the impact of molecular characteristics of TP53 mutations such as monoallelic versus biallelic mutation status on prognosis, which might be partially due to the differences in patient populations and treatment characteristics of the individual studies.^6,10–13 Recent data also suggest that the prognosis of AML and HR-MDS patients with TP53 mutations is generally comparable and both could be considered as a single molecularly defined disease entity rather than being classified separately based on blast count.^10,11

Limited data suggest that TP53 mutations confer an immunosuppressive phenotype that is characterized by upregulation of programmaed death ligand-1 (PD-L1) expression on hematopoietic stem cells and a higher percentage of immunosuppressive regulatory T-cells, which have been identified as an independent predictor of adverse survival.¹⁴ However, to our knowledge, no studies have assessed the impact of TP53 allelic status on the immune environment in MDS/AML, and there are limited data correlating TP53 mutations, but not allelic status, with the epigenetic disease phenotype.¹⁵ A distinct analysis by TP53 mutant allelic status and other molecular characteristics could help delineate mechanisms contributing to immune escape and ultimately disease relapse.

We have previously reported results of the FUSION AML-001 trial, a large, randomized phase II trial in which 205 patients received azacitidine (AZA) or AZA with the anti-PD-L1 antibody durvalumab in two separate cohorts of older patients with AML or HR-MDS.^16,17 Given the unusually high number of patients with TP53 mutations in our study (n = 61; 29.8%), and the fact that durvalumab did not improve response rates or survival outcomes in either cohort, we have combined patients treated in either treatment arm to assemble one of the largest, highly annotated datasets of TP53 mutated patients, and compared them to 144 patients with TP53 wild-type status enrolled in the same study. As clinical outcomes of this patient cohort have been reported separately,¹⁸ this manuscript focuses on the correlative studies describing the molecular, epigenetic, and immunologic characteristics of TP53-mutant AML and MDS.

Methods

Study population

The FUSION trial (NCT02775903) was a randomized, multicenter, open-label, phase II study comparing AZA monotherapy with AZA + durvalumab in previously untreated patients with AML or HR-MDS, who were unfit for intensive chemotherapy. The results of the primary analyses of this trial have previously been published.^16,17 Briefly, adult (⩾18 years), previously untreated patients with HR-MDS [intermediate, high, or very high risk based on the revised International Prognostic Scoring System (IPSS-R)] or AML (age ⩾65 years and unfit for intensive chemotherapy) were randomized in a 1:1 ratio to either AZA + durvalumab or AZA monotherapy. Patients received either AZA 75 mg/m² subcutaneously during days 1–7 of each 28-day cycle or AZA 75 mg/m² subcutaneously during days 1–7 + durvalumab 1500 mg intravenously on day 1 of each 28-day cycle. Treatment was continued until disease progression or occurrence of unacceptable toxicity. As there was no difference in response rates or survival outcomes by treatment arm neither in the overall study population nor among TP53-mutated patients and given recent studies reporting similar survival in AML and HR-MDS patients with TP53 mutations,^10,11 we combined AML and MDS patients independent of treatment assignment for a comparison of TP53-mutant and TP53 wild-type patients. In the trial, overall response rate was defined as a composite of complete remission (CR) and CR with incomplete count recovery among AML patients and as a composite of CR, marrow CR, partial remission, and hematologic improvement among MDS patients in line with International Working Group 2003 and 2006 response criteria, respectively.^19,20 Overall survival (OS) was assessed from the time of randomization in the FUSION trial. Outcomes were assessed every 3 months.

Translational analyses

Mutation profiling

The Munich Leukemia Laboratory (MLL) used a next-generation targeted sequencing assay to assess and characterize gene mutations of samples collected at screening. Thirty-eight genes were assessed and included those frequently mutated in myeloid malignancies (Supplemental Table 1). The mean sequencing coverage across the panel and samples was approximately 3000x. Genetic alterations not matching the reference sequence were noted as mutated, common single nucleotide polymorphism (SNP), or as a variant of unknown significance (VUS). Mutational load and coverage were assessed at each nonreference location. Only pathogenic TP53 variants with a variant allele frequency (VAF) ⩾2% were included. Patients with a common SNP or VUS in the TP53 gene were classified as TP53 wild-type in the analyses comparing TP53 mutant and TP53 wild-type patients.

Similar to prior publications,^11,12 we defined multihit TP53 mutations as any of the following: (I) two or more TP53 gene variants irrespective of the VAF, (II) at least one TP53 gene variant co-occurring with a cytogenetic aberration involving chromosome 17p (e.g., abnormality of 17p or monosomy 17), or (III) a single TP53 mutation with a VAF >55%. For comparisons by allelic state, patients with no mutations or a VUS in TP53 who had a chromosome 17p abnormality were classified as monohit TP53 mutations. Based on prior publications, we defined R175H, G245S, R248Q, R248W, R249S, R273C, R273H, and R282W as gain-of-function (GOF) mutations and all other TP53 mutations as non-GOF.²¹

RNA sequencing

Gene expression profiles of bone marrow (BM) aspirates from screening and cycle 3 day 22 (C3D22) were studied by bulk RNA sequencing.^16,17 EA Genomics (Q2 Solutions) used the Qiagen Micro RNeasy kit (Hilden, Germany) to extract RNA and made libraries using TruSeq SBS v4 chemistry (Illumina, San Diego, USA). Strand specific libraries were prepared by using polyA enrichment and included barcodes. Sequencing was performed on an Illumina HiSeq 2500, with 2 × 50 bp read lengths. Alignment was performed using a two-pass mode with STAR (v2.5.2b) on the full hg38 human genome, and gene-level counts were obtained using the quantmode GeneCounts option. Gene expression was then normalized with the function voom in the R package limma. Where gene expression is presented in the figures, the data were normalized with voom and log2 transformed, and raw Wilcoxon p values shown in plots comparing two groups. The T-cell genes and CD34 were chosen based on a prior hypothesis and so the p values were not adjusted for multiple testing. Global differential gene expression analysis was performed with limma using a model with the TP53 mutation status as the term, and the volcano plots show the negative of the log10 of the multiple testing adjusted p values (also called the false discovery rate) on the y-axis and the log2 of the fold change on the x-axis. For gene set enrichment analysis (GSEA), we used the clusterProfiler R package (version 3.14), which uses the R package fGSEA internally and used 1000 permutations to estimate the significance.^22,23 We searched the MSigDB and H (Hallmark) gene sets as provided by the msigdbr R package (version 7.0.1).^22,24 Gene co-expression modules were generated using the baseline samples by taking 6000 genes with the highest standard deviation and calculating all pairwise Pearson correlations followed by module identification using affinity propagation via the R package apcluster version 1.4.9.^25,26 The central member of each module was defined as the gene with the highest mean correlation to all the other genes in the module. The central members of each module were treated as nodes, with edges between nodes with a Pearson correlation greater than 0.7 and grouped into communities using a Louvain clustering algorithm and visualized using CanvasXpress.²⁷ All co-expression module analysis steps were performed interactively using the Gene Atlas software (Needle Genomics, LLC, Seattle, USA). The module expression in van Galen et al.²⁸ (GSE116256) was found by taking the mean expression of all the genes in the module across all the cells of a particular cell type, as labeled in GSE116256.²⁸

BM flow cytometry

BM aspirates were collected at screening, C3D22, and C6D22 and sent to MLL for processing and flow cytometry. Two panels of flow cytometry antibodies were used to detect granulocytes, lymphocytes, monocytes, and T-cells and reported as a percent of parent population. Tumor blasts were gated based on CD34 and CD117 variant expression. Surface expression of PD-L1, Programmed death-1 (PD-1), and T-cell immunoglobulin and mucin domain 3 (TIM-3) was assessed on BM cells. PD-L1 (clone 29E.2A3) surface expression was quantified using QuantiBrite Beads (BD Biosciences, New Jersey, USA) and was reported as molecules of equivalent soluble fluorochrome. Data was analyzed by MLL into a collection of 170 reportable subpopulations in the immune cell and immune checkpoint panel and 68 reportables in the tumor/blast panel.

DNA methylation

DNA methylation of peripheral blood samples was assessed using Illumina’s Infinium Human Methylation EPIC methylation array at cycle 1 day 1 (C1D1; pretreatment) and on treatment (C2D1). Functional normalization was used to normalize for interarray technical variation using control probes.²⁹ ß-values were calculated for each CpG site using the equation β = M/(M + U), wherein M and U are the number of methylated and unmethylated probes, respectively. Global DNA methylation scores (GDMSs) were calculated for each sample by counting the number of methylation probes with ß-values greater than 0.7. Changes in median GDMS were calculated by subtracting a patient’s C2D1 GDMS from their C1D1 GDMS. Focal DNA demethylation ß-values were analyzed in PD-1, PD-L1, and PD-L2 regulatory regions. A full list of analyzed loci is shown provided in Supplemental Table 2.

Statistical analysis

Survival analysis was performed using the R package ‘survival’ and ‘survminer’, with the log-rank p values and median survival shown in the plots. Patient characteristics are compared between patients with mutated versus wild-type TP53 using the ‘prop.test’ method in R when comparing the proportion of patients in each subgroup (i.e. each row of Table 1), or a Wilcoxon rank sum test for numeric characteristics. All boxplots (box-and-whisker plots) were generated using ggplot2 where the central line shows the median, the ends of the boxes show the interquartile range (IQR) and the whiskers show the most distant point ⩽1.5 times the IQR.

Table 1.

Baseline patient characteristics.

Variable	AML			HR-MDS
Variable	TP53 mutated (n = 37)	TP53 wild-type (n = 89)	p Value	TP53 mutated (n = 24)	TP53 wild-type (n = 55)	p Value
Median patient age (years; IQR)	75 (72–78)	76 (71.5–81.5)	0.315	74 (69.5–77)	74 (68.5–78)	0.733
Male sex (N; %)	20 (54.1%)	48 (53.9%)	1.000	17 (70.8%)	39 (70.9%)	1.000
ECOG (N; %)
0	11 (29.7%)	39 (43.8%)	0.228	7 (29.2%)	25 (45.5%)	0.268
1	22 (59.5%)	46 (51.7%)	0.589	15 (62.5%)	27 (49.1%)	0.394
2	3 (8.1%)	4 (4.5%)	0.715	2 (8.3%)	2 (3.6%)	0.751
Median hemoglobin (g/L; IQR)	89 (84.5–100)	95.5 (86–108)	0.106	90 (81.5–97)	91 (80–99.5)	0.554
Median platelet count ×10⁹/L (IQR)	37 (26.5–60.5)	62.5 (34–97)	0.011	49.5 (25–93)	56 (32.5–99.5)	0.554
Median leukocyte count ×10⁹/L (IQR)	1.77 (1.11–3.32)	2.28 (1.29–3.71)	0.434	3.03 (1.98–4.67)	2.37 (1.50–3.77)	0.131
Median ANC ×10⁹/L (IQR)	0.63 (0.28–1.07)	0.52 (0.26–1.25)	0.880	1.14 (0.79–2.04)	0.96 (0.48–1.81)	0.248
Median bone marrow blasts (%; IQR)	27.5 (21–32)	35.5 (24–56.5)	0.001	7 (5–10)	8.8 (6–15)	0.114
Median peripheral blood blasts (%; IQR)	12 (6–16.5)	12 (4–27)	0.536	3 (2–9.5)	2 (1–4)	0.197
AML WHO classification (N; %)				N/A	N/A	N/A
AML-MRC	23 (62.2%)	39 (43.8%)	0.093
AML, not otherwise categorized	8 (21.6%)	34 (38.2%)	0.112
AML with gene mutations	1 (2.7%)	5 (5.6%)	0.810
Therapy-related myeloid neoplasms	4 (10.8%)	1 (1.1%)	0.042
AML with recurrent genetic abnormalities	0	2 (2.3%)	1.000
MDS WHO classification (N; %)	N/A	N/A	N/A
RCMD				6 (25.0%)	5 (9.1%)	0.127
RAEB-I				8 (33.3%)	21 (38.2%)	0.875
RAEB-II				6 (25.0%)	23 (41.8%)	0.241
MDS unclassified				0	2 (3.6%)	1.000
MDS/MPN-CMML				0	2 (3.6%)	1.000
Others				4 (16.7%)	2 (3.6%)	1.000
MDS type
Primary				19 (79.2%)	54 (98.2%)	0.0134
Secondary				5 (20.8%)	1 (18%)	0.0134
IPSS-R cytogenetic riskclassification (N, %)	N/A	N/A	N/A
Very good				0	2 (3.6%)	1.000
Good				1 (4.2%)	17 (30.9%)	0.021
Intermediate				1 (4.2%)	15 (27.3%)	0.041
Poor				4 (16.7%)	13 (23.6%)	0.692
Very poor				18 (75.0%)	8 (14.6%)	<0.001
IPSS-R (%)	N/A	N/A	N/A
Intermediate				1 (4.2%)	9 (16.4%)	0.258
High				8 (33.3%)	27 (49.1%)	0.294
Very high				15 (62.5%)	19 (34.6%)	0.039

AML, acute myeloid leukemia; AML-MRC, acute myeloid leukemia -myelodysplasia-related changes; ANC, absolute neutrophil count; CMML – chronic myelomonocytic leukemia; ECOG, Eastern Cooperative Oncology Group; HR-MDS, higher-risk myelodysplastic syndromes; IPSS-R, revised International Prognostic Scoring System; IQR, interquartile range; MDS, myelodysplastic syndrome; MPN, myeloproliferative neoplasm; RAEB, refractory anemia with excess blasts; RCMD, refractory cytopenia with multilineage dysplasia.

Results

Outcomes by TP53 mutation status and treatment assignment

Among the patients with known TP53 mutation status enrolled in the FUSION trial, 37 had TP53-mutated AML, 89 were TP53 wild-type AML, 24 had TP53-mutated HR-MDS, and 55 were TP53 wild-type HR-MDS patients. We first analyzed OS by treatment assignment (AZA + durvalumab versus AZA monotherapy). There was no OS difference between treatment arms neither among AML (median OS 13.0 months for AZA + durvalumab versus 14.4 months for AZA monotherapy; p = 0.20) nor MDS patients (median OS 11.6 months for AZA + durvalumab versus 16.7 months for AZA monotherapy; p = 0.74). We next compared OS for patients with TP53 mutations and those with TP53 wild-type disease. In the AML cohort, patients with TP53 mutations had a significantly shorter median OS compared to patients who had TP53 wild-type disease (median OS 8.1 months versus 16.6 months; p < 0.001). Similarly, median OS was inferior for TP53-mutant MDS patients compared with TP53 wild-type patients (median OS: 9.8 months versus 23.5 months; p = 0.002; Supplemental Figure 1). These findings support the rationale for combining patients independent of treatment assignment and disease type (AML versus MDS) for further analyses.

Genetic landscape of TP53 mutations

The median VAF for TP53 mutations was 40%, and 90% had a VAF of ⩾10%. The majority (93%) of TP53 mutations mapped to the p53 DNA-binding domain [Figure 1(a)]. Twelve patients (14.3%) had GOF TP53 mutations. In terms of TP53 allelic status, 22 (11.0% of total study population; 36.1% of TP53-mutant patients) and 41 (20.5%; 67.2%) patients were classified as monohit and multihit, respectively. Patients with TP53 mutations had a median of 2 (IQR of 1–2) co-occurring mutations, which was fewer than TP53 wild-type patients who had a median of 3 (IQR of 2–4) co-occurring mutations [Wilcoxon p < 0.001; Figure 1(b) and (c)] with distinct patterns in the distribution of co-occurring mutations [Figure 1(d) and Supplemental Figure 2]. Among patients with TP53 mutations, no other somatic mutations were detected in 25 patients (42%). Mutations in SRSF2, RUNX1, and ASXL1 were enriched in TP53 wild-type patients compared with TP53-mutant patients.

Figure 1.

TP53 and co-occurring mutations. Panel (a) Illustrates the distribution and type of TP53 mutations as a lollipop plot. The majority of TP53 mutations were missense mutations localized in the p53 DNA-binding domain. Patients with TP53 mutations had a lower number of co-occurring mutations compared to TP53 wild-type patients (b; the central line shows the median, the ends of the boxes show the interquartile range (IQR) and the whiskers show the most distant point ⩽1.5 times the IQR). TP53 mutations occurred in isolation in 41.7% of patients, while TP53 wild-type patients frequently had several co-occurring mutations (c). The mutational and cytogenetic patterns by TP53 mutational status are shown in panel (d).

In the AML cohort, patients with TP53 mutations were more likely to have poor-risk cytogenetics (59.5% versus 11.2%; p < 0.001), therapy-related myeloid neoplasm (10.8% versus 1.1%; p = 0.042), and lower BM blast percentage (27.5% versus 35.5%; p = 0.001) compared to TP53 wild-type. However, there were no statistically significant differences in terms of other baseline demographics and clinical characteristics between TP53 mutant and TP53 wild-type patients. Table 1 provides an overview of baseline patient characteristics.

Comparing baseline characteristics of HR-MDS patients (Table 1), patients with TP53 mutations were statistically significantly more likely to have secondary MDS (20.8% versus 1.8%; p = 0.013), very poor risk cytogenetics by IPSS-R (75% versus 14.6%; p < 0.001), and very high risk IPSS-R score (62.5% versus 34.6%; p = 0.039).

Epigenetic, gene expression, and immunophenotypic landscape of TP53-mutant and TP53 wild-type AML and MDS at baseline

We analyzed DNA methylation status in peripheral blood samples of AML and HR-MDS patients using the global methylation scores as published previously.^16,17 Global DNA methylation was independent of TP53 mutation status at baseline when comparing monohit, multihit, and wild-type patients [p value = 0.98 wild-type versus monohit; p = 0.21 wild-type versus multihit; p = 0.46 monohit versus multihit; Figure 2(a)]. Assessing methylation of PD-1, PD-L1, and PD-L2 gene loci at baseline, we found that especially in PD-L2, various loci showed differences in methylation status depending on TP53 mutation status (Supplemental Figure 3).

Figure 2.

Baseline global DNA methylation status and immune phenotype by TP53 mutation status. Panel (a) shows the global DNA methylation score at baseline without statistically significant differences between TP53 wild-type, monohit or multihit TP53-mutant patients. Global DNA methylation score is shown as box-and-whisker plots with median and interquartile range. Distribution by TP53 mutation status was compared using Wilcoxon rank sum test. Figure panels (b–d) show flow cytometry from baseline bone marrow samples for AML patients (b), HR-MDS patients (c), and the combined patient cohort (d) by TP53 mutation status, respectively. Box-and-whisker plots show the median and interquartile range. Distribution by TP53 mutation status was compared using Wilcoxon rank sum test.

In the BM of AML patients at baseline [Figure 2(b)], CD3⁺ T-cell and PD-L1-positive CD34⁺ progenitor cell percentages were statistically significantly higher in TP53-mutated patients, while the myeloid progenitor cell percentage (all as a percent of total cells) was higher in TP53 wild-type patients. However, these differences were not present in MDS patients [Figure 2(c)]. Additionally, AML patients with TP53 mutations had a higher abundance of exhausted T-cells [CD3⁺CD8⁺PD1⁺TIM3⁻ T-cells (p = 0.033) and CD3⁺CD8⁺PD1⁻TIM3⁺ T-cells (p = 0.008); each as percentage of total cells] compared to TP53 wild-type AML patients (plot not shown; and p > 0.05 in MDS). When analyzing the pooled cohort of AML and HR-MDS by TP53 mutation status, patients with TP53 mutations had higher median levels of PD-L1-positive, CD34⁺ progenitors (5.6% versus 2.4%; p = 0.006) and fewer myeloid progenitor cells (58% versus 75%; p = 0.005) than TP53 wild-type patients [Figure 2(d)].

To evaluate differences in gene expression profile by TP53 mutation status, we conducted bulk RNA sequencing of baseline BM aspirates (24 TP53 mutant and 84 TP53 wild-type). In the baseline samples, the number of differentially expressed genes (DEGs) between TP53-mutant samples and TP53 wild-type samples was higher in the AML cohort (1887 genes) compared to the HR-MDS cohort (78 genes) (Supplemental Figure 4 and Supplemental Table 3). DEGs from our cohort overlapped significantly with an independent dataset of TP53-mutant versus TP53 wild-type AML samples from the Beat AML cohort (Supplemental Figure 5), supporting the validity of our findings. Patterns of DEGs were similar for multihit TP53mutations versus wild-type as they were for all TP53 mutations versus wild-type. The gene which had the largest difference in expression in TP53-mutant samples relative to TP53 wild-type samples was ZNF560. Additionally, we found that TP53-mutant patients had higher expression of T-cell genes (e.g. IL7R) and markers of proliferation (MKI67) compared to TP53 wild-type patients (Figure 3). PD-L1 (CD274) expression was also higher in TP53-mutant patients compared to wild-type among both AML and HR-MDS patient cohorts.

Figure 3.

Gene expression pattern by TP53 mutation status in screening samples. Figure shows genes with statistically significant RNA expression levels with higher expression in TP53-mutant patients. TP53-mutant patients had higher expression of T-cell genes (e.g. IL7R), markers of proliferation (MKI-67), PD-L1 (CD274) expression compared to wild-type. Differences in gene expression between TP53-mutant and TP53 wild-type samples applied to both AML and HR-MDS patient cohorts. Box-and-whisker plots show the median and interquartile range. Distribution by TP53 mutation status was compared using Wilcoxon rank sum test.

To infer which cell subsets were underlying these differences in gene expression, we generated co-expression modules of the 6000 genes with the highest standard deviation and calculated the fold change of the central gene of each module by TP53 mutation status (TP53 mutant versus wild-type). Using previously published single-cell RNA sequencing (scRNA-seq) data from BM cells of AML patients and healthy subjects as a reference,²⁸ we mapped the co-expression modules to individual cell compartments. We found that modules with higher expression in T-cells (central genes: BCL11B, TBX21, SPOCK2, SIT1, SLFN12L, PLEKHA1) and myeloid cells (central genes: LILRB3, CKAP4, ANXA3, COL17A1, NFAM1, SNX18P8) demonstrated higher expression in TP53-mutant compared to TP53 wild-type samples (Figure 4), corroborating the T-cell specific single genes selected and shown as examples in Figure 3. Additionally, another community of modules was higher in TP53-mutated samples and were more highly expressed in the early and late erythroid cell clusters from van Galen et al.²⁸ Some of those erythroid related modules had genes involved in cell cycle regulation (e.g. SPTB, TOP2A, SLC22A23, DNAJC6). Interestingly, the differential expression of cell cycle genes mapped to progenitor cell types and late erythroid precursors likely reflecting the tumor cell compartment.

Figure 4.

Co-expression modules colored by differential expression in TP53 mutant versus wild-type and projected onto scRNA-seq data from van Galen et al.²⁸ Figure shows the largest 53 co-expression modules from the bulk RNA-seq baseline samples and their logFC in TP53-mutant versus TP53 wild-type and expression in clusters of single cells. In panel (a), each node represents 25–57 genes which are co-expressed and labeled with the most central member of the module. Edges in the network are drawn if the central member has a Pearson correlation greater than or equal to 0.7 to any other node, and the thickness of the line is scaled to the correlation value. The color of each node is the log2 fold change of TP53 mutant versus TP53 wild-type with red showing modules which are higher in TP53-mutated samples and blue showing the opposite. Communities are shown by circles and defined by the graph structure. The circle labeled ‘singles’ is a collection of singleton modules, which are not connected to any other modules above a correlation of 0.7. The labels of the remaining communities are from (b), which is a heatmap showing the expression of each co-expression module from (a) in the cell types from the AML and healthy BM scRNA-seq data from van Galen et al.²⁸ (GSE116256). The color of each element in the heatmap is based on the scaled mean of the expression of the genes within each module from (a), and the community names are based on the top cell type labels.

Finally, we conducted GSEA of the combined AML and HR-MDS DEGs and found enrichment of genes in the hallmark gene sets of heme metabolism, inflammatory response, cell cycle regulation (e.g. G2M checkpoint, mitotic spindle, and E2F targets), and transforming growth factor-β signaling pathways in TP53-mutant patients.²⁴ Conversely, oxidative phosphorylation, interferon (IFN)-α, and MYC target genes were enriched in TP53 wild-type patients (Supplemental Figure 6).

Epigenetic, gene expression, and immunophenotypic landscape of TP53-mutant and TP53 wild-type AML and HR-MDS following treatment with AZA and durvalumab

To evaluate for any differences in epigenetic, gene expression, or immunophenotypic profiles by TP53 mutation status in response to treatment with AZA ± durvalumab, we analyzed on-treatment samples after one cycle (DNA methylation profile and flow cytometry) or three cycles of treatment (gene expression profile).

In bulk analysis from peripheral blood, AZA led to statistically significant changes in GDMS after one cycle of treatment when compared to baseline across all cohorts but did not differ by TP53 mutation status (monohit versus multihit versus wild-type) (Figure 5). When comparing methylation status of specific loci in the PD-1, PD-L1, and PD-L2 genes, all differences in baseline methylation status between multihit TP53 mutant and TP53 wild-type samples were reversed on treatment except for two loci of the analyzed PD-L2 gene (cg07211259 and cg11299543) with lower methylation scores in the TP53 wild-type samples (Supplemental Figure 3).

Figure 5.

Serial assessment of global DNA methylation status by TP53 mutation status. Figure shows the global DNA methylation score at baseline and after one cycle of treatment with azacitidine ± durvalumab for all patients (a). Patients experienced a decline in global DNA methylation scores independent of TP53 mutation status (b) without statistically significant differences between TP53 wild-type, monohit or multihit TP53-mutant patients neither at baseline nor after one cycle of treatment (c). This supports that HMAs exert an objective pharmacodynamic effect in patients independent of TP53 mutation status. Global DNA methylation score is shown as box-and-whisker plots with median and interquartile range. Distribution by TP53 mutation status was compared using Wilcoxon rank sum test.

To evaluate differences in gene expression profiles following treatment with AZA by TP53 mutation status, we conducted RNA sequencing after three cycles of therapy (27 TP53 mutant and 97 TP53 wild-type). When comparing TP53-mutant and TP53 wild-type patients, AML samples had 116 genes, which were differentially expressed while HR-MDS samples had zero DEGs after three cycles of therapy (Supplemental Table 3).

Discussion

In this study, we provide an assessment of the molecular, epigenetic, and immunologic landscape of TP53-mutant AML/HR-MDS before and after AZA-based therapy in very well annotated samples from 61 patients with TP53 mutations and 144 TP53 wild-type patients who were enrolled in a randomized clinical trial.

One potential mechanism underlying the poor prognosis associated with TP53 mutations is the induction of an immunosuppressive microenvironment that permits immune evasion of tumor cells.^14,30,31 Supporting this hypothesis, we found higher expression of T-cell-related genes and PD-L1 by RNA sequencing and flow cytometry in pretreatment TP53-mutated AML samples compared to pretreatment TP53 wild-type samples. Of note, these differences were not present in HR-MDS patients. This is in line with prior work showing that PD-L1 expression is increased in MDS and secondary AML BM specimens via activation of MYC.¹⁴ Recently, AML patients with TP53 mutations were demonstrated to have higher cytotoxic T-cell and NK-cell infiltration and IFN-γ gene signature.³² However, it has been previously suggested that immune activation might be balanced by the presence of exhausted cytotoxic T-cells and NK-cells that are characterized by expression of Lymphocyte Activation Gene (LAG-3) and TIM-3.³² As such, treatment with antibodies targeting LAG-3 and TIM-3 in addition to anti-PD-1 might be more effective in overcoming this immune-exhausted state.^33–35

The single gene and co-expression module analysis of the RNA sequencing studies showed that genes involved with inflammatory response and cell cycle regulation were higher in TP53-mutant patients compared to wild-type patients. Although we found higher Ki-67 expression in TP53-mutant compared with wild-type samples, the reverse was found for BM blast percentage. As our bulk RNA sequencing is limited in terms of distinguishing gene expression on an individual cell level (e.g. tumor versus immune cells), it is unclear whether the higher Ki-67 expression was due to differences in blast proliferation or if it reflects a higher rate of turn-over in another cell compartment. Since the blast count in the BM was slightly lower in TP53-mutant samples and the gene expression implies that the T-cells and myeloid cells were slightly higher in TP53-mutant samples, it seems more likely that the difference in MKI67 gene expression is due to the non-blast cells. Although single-cell analyses were not available in our study to confirm this hypothesis, our observations are also supported by a recent study using single-cell multiomics of hematopoietic stem/progenitor cells from patients with TP53-mutant post-MPN AML.³⁶ Another study using data from the Cancer Genome Atlas (TCGA) and BeatAML cohorts showing an increase in CD8⁺ T-cells and upregulation of immune response signaling pathways in TP53-mutant AML patients.³⁷ The role of other DEGs in TP53-mutant AML and HR-MDS warrants additional studies although some of the genes identified in our analysis such as ZNF560 have been implicated in ferroptosis and potentially confer an adverse prognosis in AML.³⁸

Treatment with hypomethylating agents (HMAs) has been shown to lead to an upregulation of PD-1 and PD-L1 in preclinical models of AML, which suggested a potential synergy of combined treatment with HMA and anti-PD-L1 antibodies.³⁹ With data demonstrating an higher expression of PD-1/PD-L1 in patients with TP53 mutations compared to wild-type, this combination might be especially effective in such patients.^14,40 However, the negative findings of the FUSION trial and post hoc analyses of TP53-mutant patients suggest that targeting PD-L1 might not be clinically useful in AML and HR-MDS.^16–18 As a pharmacodynamic surrogate, we found that changes in the GDMSs with AZA treatment were similar in TP53-mutant and TP53 wild-type patients supporting that AZA may have equal pharmacodynamic effects in TP53-mutant patients, and that these changes are insufficient to overcome the adverse prognosis of TP53 mutations. While the specific epigenetic changes induced by HMA are poorly characterized, and the tumor cells were not a majority cell type in the bulk cells of peripheral blood, there were several differences in methylation of various loci in the genes for PD-1, PD-L1, and PD-L2 in patients with TP53 mutations versus TP53 wild-type patients. Although the functional implications of most of the variably methylated loci are largely unknown, changes in methylation patterns that lead to upregulation of immune checkpoint receptors could potentially contribute to resistance to HMA ± immune checkpoint inhibitors.

The role of myeloid immune checkpoints in myeloid malignancies has received increasing attention recently.⁴¹ Leukocyte immunoglobulin-like receptor B-3 (LILRB3) is exclusively expressed on myeloid cells and its activation has recently been associated with enhanced leukemic cell survival and inhibition of T cell-mediated anti-tumor activity.⁴² In our analyses, we found that LILRB3 expression was statistically significantly higher among TP53-mutant patients and mapped to the myeloid compartment, which includes both normal myeloid cells and potential tumor cells, suggesting that the immunosuppressive effects of TP53 mutations are not limited to the anti-tumor T-cell immune function and that inhibitory myeloid immune checkpoints such as LILRB3 might constitute another therapeutic target.⁴³ Preclinical studies have already demonstrated the anti-leukemic effect of LILRB3-blocking antibodies in several AML xenograft models.⁴² Although different members of the LILRB family have distinct biologic functions, LILRB4 has also been shown to mediate T-cell suppression in AML models and clinical trials targeting LILRB4 are already ongoing (NCT04372433).^44,45

Potential limitations of our study include the smaller sample size compared to other studies and the pooling of treatment and disease entities. However, as the outcomes, baseline characteristics, and most immunophenotypic and molecular analyses of AML and HR-MDS patients as well as treatment group (AZA + durvalumab versus AZA monotherapy) were comparable, the effect of this pooled analysis is unlikely to affect the main conclusions of our study. While our power to detect small differences was limited by sample size, the potential benefits of using samples collected as part of a randomized clinical trial reduce the risk of confounding by treatment, patient, and disease characteristics that might have affected prior studies. As our analyses were performed on samples from clinical trial patients, the extent of the analyses was limited by sample availability and research methods available. Future studies using single-cell sequencing technologies instead of bulk RNA sequencing will be important to decipher the impact of specific immune cell subsets. Additionally, integration with other research methodologies to evaluate other aspects of the underlying disease biology such as ATAC-seq to evaluate chromatin accessibility and proteomic studies can be informative to derive a more comprehensive understanding of TP53-mutant AML/MDS. However, these studies were beyond the scope of this manuscript.

In summary, we present the largest comprehensive molecular, epigenetic, and immune-phenotypical characterization of AML and HR-MDS patients with TP53 mutations treated with HMA-based therapy in a clinical trial. Upregulation of inhibitory immune checkpoints such as PD-1 and notably the myeloid immune checkpoint LILRB3 could be contributing to the poor prognosis and frequent disease relapse via immune escape.

Supplemental Material

sj-docx-1-tah-10.1177_20406207241257904 – Supplemental material for Integrated genetic, epigenetic, and immune landscape of TP53 mutant AML and higher risk MDS treated with azacitidine

Supplemental material, sj-docx-1-tah-10.1177_20406207241257904 for Integrated genetic, epigenetic, and immune landscape of TP53 mutant AML and higher risk MDS treated with azacitidine by Amer M. Zeidan, Jan Philipp Bewersdorf, Vanessa Hasle, Rory M. Shallis, Ethan Thompson, Daniel Lopes de Menezes, Shelonidta Rose, Isaac Boss, Stephanie Halene, Torsten Haferlach and Brian A. Fox in Therapeutic Advances in Hematology

Footnotes

Acknowledgements

AMZ is a Leukemia and Lymphoma Society Scholar in Clinical Research.

Declarations

ORCID iDs

Amer M. Zeidan

Jan Philipp Bewersdorf

Rory M. Shallis

Supplemental material

Supplemental material for this article is available online.

References

Haferlach

Nagata

Grossmann

, et al. Landscape of genetic lesions in 944 patients with myelodysplastic syndromes. Leukemia 2014; 28: 241–247.

Cancer Genome Atlas Research Network, Ley

Miller

, et al. Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N Engl J Med 2013; 368: 2059–2074.

Jadersten

Saft

Smith

, et al. TP53 mutations in low-risk myelodysplastic syndromes with del(5q) predict disease progression. J Clin Oncol 2011; 29: 1971–1979.

Bejar

Stevenson

Abdel-Wahab

, et al. Clinical effect of point mutations in myelodysplastic syndromes. N Engl J Med 2011; 364: 2496–2506.

Lindsley

Saber

Mar

, et al. Prognostic mutations in myelodysplastic syndrome after stem-cell transplantation. N Engl J Med 2017; 376: 536–547.

Bernard

Nannya

Hasserjian

, et al. Implications of TP53 allelic state for genome stability, clinical presentation and outcomes in myelodysplastic syndromes. Nat Med 2020; 26: 1549–1556.

Haase

Stevenson

Neuberg

, et al. TP53 mutation status divides myelodysplastic syndromes with complex karyotypes into distinct prognostic subgroups. Leukemia 2019; 33: 1747–1758.

Boettcher

Miller

Sharma

, et al. A dominant-negative effect drives selection of TP53 missense mutations in myeloid malignancies. Science 2019; 365: 599–604.

Kanagal-Shamanna

Montalban-Bravo

Katsonis

, et al. Evolutionary action score identifies a subset of TP53 mutated myelodysplastic syndrome with favorable prognosis. Blood Cancer J 2021; 11: 52.

10.

Weinberg

Siddon

Madanat

, et al. TP53 mutation defines a unique subgroup within complex karyotype de novo and therapy-related MDS/AML. Blood Adv 2022; 6: 2847–2853.

11.

Grob

Al Hinai

ASA

Sanders

, et al. Molecular characterization of mutant TP53 acute myeloid leukemia and high-risk myelodysplastic syndrome. Blood. 2022; 139: 2347–2354.

12.

Zeidan

Bewersdorf

Hasle

, et al. Prognostic Implications of Mono-Hit and Multi-Hit TP53 Alterations in Patients with Acute Myeloid Leukemia and Higher Risk Myelodysplastic Syndromes Treated with Azacitidine-based Therapy. Leukemia. 2022; In press.

13.

Short

Montalban-Bravo

Hwang

, et al. Prognostic and therapeutic impacts of mutant TP53 variant allelic frequency in newly diagnosed acute myeloid leukemia. Blood Adv 2020; 4: 5681–5689.

14.

Sallman

McLemore

Aldrich

, et al. TP53 mutations in myelodysplastic syndromes and secondary AML confer an immunosuppressive phenotype. Blood. 2020; 136: 2812–2823.

15.

Wang

Auer

Zhang

, et al. Impact of epigenomic hypermethylation at TP53 on allogeneic hematopoietic cell transplantation outcomes for myelodysplastic syndromes. Transplant Cell Ther 2021; 27: 659.e1–659.e6.

16.

Zeidan

Boss

Beach

, et al. A randomized phase 2 trial of azacitidine with or without durvalumab as first-line therapy for higher-risk myelodysplastic syndromes. Blood Adv 2022; 6: 2207–2218.

17.

Zeidan

Boss

Beach

, et al. A randomized phase 2 trial of azacitidine with or without durvalumab as first-line therapy for older patients with AML. Blood Adv 2022; 6: 2219–2229.

18.

Zeidan

Bewersdorf

Hasle

, et al. Prognostic implications of mono-hit and multi-hit TP53 alterations in patients with acute myeloid leukemia and higher risk myelodysplastic syndromes treated with azacitidine-based therapy. Leukemia. 2023; 37: 240–243.

19.

Cheson

Bennett

Kopecky

, et al. Revised recommendations of the International Working Group for Diagnosis, Standardization of Response Criteria, Treatment Outcomes, and Reporting Standards for Therapeutic Trials in Acute Myeloid Leukemia. J Clin Oncol 2003; 21: 4642–4649.

20.

Cheson

Greenberg

Bennett

, et al. Clinical application and proposal for modification of the International Working Group (IWG) response criteria in myelodysplasia. Blood 2006; 108: 419–425.

21.

Roszkowska

Piecuch

Sady

, et al. Gain of function (GOF) mutant p53 in cancer – current therapeutic approaches. Int J Mol Sci. 2022; 23: 13287.

22.

Subramanian

Tamayo

Mootha

, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 2005; 102: 15545–15550.

23.

Korotkevich

Sukhov

Sergushichev

. Fast gene set enrichment analysis. bioRxiv. 2019.

24.

Liberzon

Birger

Thorvaldsdóttir

, et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 2015; 1: 417–425.

25.

Bodenhofer

Kothmeier

Hochreiter

. APCluster: an R package for affinity propagation clustering. Bioinformatics 2011; 27: 2463–2464.

26.

Frey

Dueck

. Clustering by passing messages between data points. Science 2007; 315: 972–976.

27.

Neuhaus

CanvasXpress: a JavaScript library for data analytics with full audit trail capabilities. https://canvasxpress.org/

28.

van Galen

Hovestadt

Wadsworth Ii

, et al. Single-cell RNA-Seq reveals AML hierarchies relevant to disease progression and immunity. Cell. 2019; 176: 1265–1281.e24.

29.

Fortin

Labbe

Lemire

, et al. Functional normalization of 450k methylation array data improves replication in large cancer studies. Genome Biol 2014; 15: 503.

30.

Vadakekolathu

Lai

Reeder

, et al. TP53 abnormalities correlate with immune infiltration and associate with response to flotetuzumab immunotherapy in AML. Blood Adv 2020; 4: 5011–5024.

31.

Williams

Basu

Garcia-Manero

, et al. The distribution of T-cell subsets and the expression of immune checkpoint receptors and ligands in patients with newly diagnosed and relapsed acute myeloid leukemia. Cancer 2019; 125: 1470–1481.

32.

Dufva

Pölönen

Brück

, et al. Immunogenomic landscape of hematological malignancies. Cancer Cell. 2020; 38: 380–399.e13.

33.

Bewersdorf

Shallis

Zeidan

. Immune checkpoint inhibition in myeloid malignancies: moving beyond the PD-1/PD-L1 and CTLA-4 pathways. Blood Rev 2020; 45: 100709.

34.

Wolf

Anderson

Kuchroo

. TIM3 comes of age as an inhibitory receptor. Nat Rev Immunol 2020; 20: 173–185.

35.

Abdelhakim

Cortez

, et al. LAG3 Inhibition decreases AML-induced immunosuppression and improves T cell-mediated killing. Blood 2019; 134: 3605-.

36.

Rodriguez-Meira

Norfo

Wen

, et al. Single-cell multi-omics identifies chronic inflammation as a driver of TP53-mutant leukemic evolution. Nat Genet 2023; 55: 1531–1541.

37.

Wen

X-M

Z-J

Jin

, et al. Association analyses of TP53 mutation with prognosis, tumor mutational burden, and immunological features in acute myeloid leukemia. Front Immunol 2021; 12: 717527.

38.

Wang

Zhuo

Wang

, et al. Identification and validation of a prognostic risk-scoring model based on ferroptosis-associated cluster in acute myeloid leukemia. Front Cell Dev Biol 2022; 9: 800267-.

39.

Yang

Bueso-Ramos

DiNardo

, et al. Expression of PD-L1, PD-L2, PD-1 and CTLA4 in myelodysplastic syndromes is enhanced by treatment with hypomethylating agents. Leukemia 2014; 28: 1280–1288.

40.

Zajac

Zaleska

Dolnik

, et al. Expression of CD274 (PD-L1) is associated with unfavourable recurrent mutations in AML. Br J Haematol 2018; 183: 822–825.

41.

Abaza

Zeidan

. Immune checkpoint inhibition in acute myeloid leukemia and myelodysplastic syndromes. Cells 2022; 11: 2249.

42.

Schultz

, et al. LILRB3 supports acute myeloid leukemia development and regulates T-cell antitumor immune responses through the TRAF2-cFLIP-NF-κB signaling axis. Nat Cancer 2021; 2: 1170–1184.

43.

Yeboah

Papagregoriou

Jones

, et al. LILRB3 (ILT5) is a myeloid cell checkpoint that elicits profound immunomodulation. JCI Insight 2020; 5: e141593.

44.

Deng

Gui

Kim

, et al. LILRB4 signalling in leukaemia cells mediates T cell suppression and tumour infiltration. Nature 2018; 562: 605–609.

45.

Deng

Huang

, et al. LILRB4 ITIMs mediate the T cell suppression and infiltration of acute myeloid leukemia cells. Cell Mol Immunol 2020; 17: 272–282.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.95 MB