PARP10 is highly expressed and associated with inferior outcomes in acute myeloid leukemia

Patients’ data

The RNA expression data, clinical and laboratory parameters data, gene mutation data, and survival data of 173 newly diagnosed AML patients in the TCGA dataset were downloaded from the cBioPortal dataset. The expression differences in PARP family members between AML patients and normal patients were analyzed based on three datasets, i.e., TCGA versus GTEx (UCSC Xena project) [16], GSE15061 [17], and Bloodspot [18].

Gene differential expression analysis

The gene expression data of AML patients and normal patients downloaded from the GSE15061 dataset [17] was normalized before the differential expression analysis of AML patients versus normal patients was performed using an unpaired t test. The gene expression data of the TCGA and GTEx datasets downloaded from the UCSC Xena project were already computed by a standard pipeline; therefore, differential expression analysis was conducted using an unpaired t test [16]. Bloodspot included gene expression data, whereby human normal hematopoiesis cells were from GSE42519 [19] and human AML cells were from GSE13159 [20].

Bioinformatics analysis

PARP10 co-expression analysis was conducted using cBioPortal based on 173 RNA-seq datasets from TCGA AML patients [21]. KEGG pathway analysis of the significant co-expressed genes was conducted using Enrichr [22]. Gene set enrichment analysis (GSEA) of the PARP10 high expression versus low expression groups was conducted using GESA software (Broad) [23].

Cell culture, plasmids, virus package, and cell proliferation assay

Lenti-X 293T cells (Clontech, USA) were cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, USA) with 1 mM sodium pyruvate, 4500 mg/L glucose, and 10% fetal bovine serum (FBS). Cas9-expressing MOLM13 (MOLM13-Cas9) cells were cultured in Roswell Park Memorial Institute (RPMI, USA) 1640 (Thermo Fisher Scientific, USA) with 10% FBS. Cells used in this study were cultured with a medium containing 1/100 Penicillin–Streptomycin (Gibco).

LentiCas9-blast (#52962) and lentiGuide-RFP-blasticidin (#167930) were purchased from Addgene (USA). Two sgRNAs [24] were synthesized, annealed, and inserted into a lentiGuide-RFP-blasticidin plasmid. Lentiviral particles containing sgRNAs were produced by transient transfection of Lenti-X 293T cells using the lipofection method. MOLM13-Cas9 cells were incubated with lentiGuide-RFP-blasticidin lentiviral supernatants for 48 hours, and then, subjected to selection with blasticidin (30 ug/mL) for 3 days. A flow cytometry-based growth competition assay was performed by using lentivirus-infected cells without blasticidin selection. Pooled cells were subjected to flow cytometry to detect the percentage of RFP-positive cells at the indicated days, and the percentage of RFP-positive cells was normalized to day 0. The cell proliferation assay was performed by plating 0.1 × 10⁶ cells per well in a 12 well-plate in triplicate, and monitoring the cell density every two days.

Statistical analyses

Median expression of PARP10 was used as threshold to define PARP10 high or low group in the datasets. The relationship between PARP10 expression and the clinical and laboratory parameters was analyzed using the Chi-square test. Moreover, the Kaplan-Meier method and log-rank test were used to generate survival curves and to analyze the survival differences between the different groups. Univariate and multivariate analyses were performed using Cox proportional hazards model. All of the statistical analyses were performed using SPSS 24.0 software (SPSS Inc., Chicago, IL, USA), R software, and Prism GraphPad. All tests were two-sided and statistical significance was P < 0.05.

Differentially expressed PARPs in AML versus normal bone marrow cells

To explore the transcriptional expression profiles of PARP genes in AML, we first analyzed the mRNA levels of the TCGA versus GTEX databases. The results showed that PARP1, PARP2, PARP3, PARP4, PARP5b, PARP6, PARP8, PARP9, PARP10, PARP11, PARP12, PARP13, PARP14, PARP15, and PARP16 were differentially expressed in AML patients compared to normal bone marrow cells (P < 0.05, Figure 1 and Supplementary Figure 1). Next, we conducted a validation analysis of the GEO database (GSE15061) and found that PARP3, PARP6, PARP10, and PARP11 genes exhibited a similar differential expression pattern, whereby these genes showed significantly higher expressions in AML patients compared to normal bone marrow cells, in both databases (Figure 1 and Supplementary Figure 2). The other validation was performed in the Bloodspot dataset, where only PARP6 and PARP10 exhibited significantly higher expressions in AML compared to equivalent normal cells (Figure 1 and Supplementary Figure 3). Therefore, PARP6 and PARP10 genes might play potential roles in AML progression.

High PARP10 expression was associated with poorer survival in AML

To better understand the significance of these two genes (PARP6 and PARP10)—in the prognosis of AML patients, we generated Kaplan-Meier survival curves for the overall survival (OS) and event-free survival (EFS) of patients with high and low expression of these genes (cutoff: median expression). Results showed that patients with higher expressions of PARP10 were significantly associated with poorer OS and EFS compared to patients with lower expressions of PARP10 (OS: median: 0.88 vs. 2.19 years; P = 0.001; EFS: median: 0.65 vs. 1.12 years; P = 0.041), while the expression of PARP6 showed no influence on the prognosis of AML patients (Figure 2 and Supplementary Figure 4). Therefore, we identified PAPR10 as the only gene among the PARP family that was highly expressed and associated with poorer outcomes in AML.

Association of PARP10 expression with clinical characteristics from the TCGA database

We further analyzed the association between PARP10 expression and the clinical characteristics in AML patients taken from the TCGA database (Table 1). We found that the patients present in the high expression PARP10 subgroup were older (≥60 years of age, P = 0.012), possessed more frequent TP53 mutations (P = 0.024), and had shorter overall (P = 0.001) and event-free survivals (P = 0.01). Moreover, the patients who possessed the mutant TP53 showed relatively higher expressions of PARP10 compared to patients carrying the wildtype TP53 (P < 0.05, Figure 3A). PARP10 expression was also significantly associated with FAB subtypes (P = 0.013), ELN risk classification (P = 0.001), and cytogenetic risk classification (P < 0.001). According to ELN risk classification, patients with poorer risks exhibited the highest expression of PARP10, while patients with good risk exhibited the lowest expression of PARP10 (P = 0.001, Figure 3B); cytogenetic risk classification showed a similar pattern (P = 0.006, Figure 3C). No significant differences in sex, white blood cells, bone marrow blasts, peripheral blood blasts, and allo-SCT status were observed in the PARP10 high expression versus low expression groups (P > 0.05) (Table 1). Overall, these data indicated that the high expression of PARP10 was a feature of higher-risk AML, which is more frequently seen in older AML patients, TP53 mutant patients, and those with high-risk karyotypes, suggesting its clinical significance in predicting poorer clinical outcomes of AML patients.

Univariate and multivariate Cox regression analyses identified PARP10 as an independent factor of AML prognosis

To evaluate the prognostic value of PARP10 expression in the presence of other clinical and molecular factors, we included the following dichotomous variables in univariate and multivariate Cox regression analyses: PARP10 expression levels (high vs. low), age (<60 vs. >60), sex (male vs. female), WBC count (<30 vs. ≥30 × 10⁹/L), eight common gene mutations (NPM1, FLT3, CEBPA, IDH1/IDH2, TP53, and NRAS/KRAS; wildtype vs. mutant), and transplant status (no vs. yes) (Table 2).

Univariate Cox regression analysis showed that higher PARP10 expression had an adverse effect on both OS (HR = 1.840, 95% CI 1.007–2.171; P = 0.001) and EFS (HR = 1.567, 95% CI 1.015–2.420; P = 0.043). Lower WBC (<30 × 10⁹/L) and being younger in age (<60 years old) were associated with favorable EFS (HR = 0.485, 95% CI 0.314–0.749; P = 0.001) and OS (HR = 0.319, 95% CI 0.219–0.466; P < 0.001), respectively. Wildtype FLT3 and TP53 were also associated with a more favorable EFS (HR = 0.585 95% CI 0.368–0.930; P = 0.023) and OS (HR = 0.373 95% CI 0.205–0.679; P < 0.001), respectively. Furthermore, not having received transplant contributed to an inferior OS (HR = 1.925, 95% CI 1.314–2.821; P = 0.001), while the other clinical and molecular factors had no effect on either OS or EFS (P > 0.05) (Tables 2 and 3).

Multivariate Cox regression analysis indicated that high PAPR10 expression and TP53 status were independent risk factors for both OS and EFS, even in the presence of the other covariates (P < 0.05, Tables 2 and 3). Age group was also an independent risk factor for OS (P < 0.001) after adjusting for the PARP10 group, TP53, and transplant status. The WBC group and FLT3 status were independent risk factors for EFS after adjusting for the PARP10 group, TP53, and transplant status (P < 0.05). Therefore, our Cox regression analyses identified high PARP10 expression as an independent risk factor for both OS and EFS in AML patients.

Allo-SCT improved the prognosis of AML patients with high PARP10 expression but not patients with low PARP10 expression

As allo-SCT has a profound positive influence on the prognosis of AML [25, 26], we further analyzed the influence of transplant status on OS and EFS in PARP10 high and low AML patients (Figure 4). The results showed that among patients with a high expression of PARP10, allo-SCT significantly increased their OS compared to the group without transplants (P < 0.001), while there was no difference in EFS between these two groups (P = 0.898) (Figure 4A and 4B). Among patients with a low expression of PARP10, allo-SCT did not increase their OS compared to the group without transplants (P = 0.273), whereas allo-SCT significantly decreased the EFS compared to the group without transplants (P = 0.007) (Figure 4C and 4D). These results suggest that allo-SCT therapy can benefit patients with a higher expression of PARP10 while having an adverse effect on those with a lower expression of PARP10. Thus, our results suggest that allo-SCT therapy is highly recommended for patients with higher expressions of PARP10, while it is inappropriate for patients with low expressions of PARP10.

Potential molecular mechanism mediated by PARP10 in AML

To further explore the mechanism of PARP10 in AML, we performed co-expression network analysis using RNA sequencing data from the 173 AML patients taken from the cBioPortal database [21] and found 558 positively co-expressed genes (r > 0.5; P < 0.05) and 94 negatively co-expressed genes (r < −0.5; P < 0.05) (Supplementary Table 1). KEGG pathway analysis revealed that the 558 positively co-expressed genes were significantly enriched in the oncogenic pathways (e.g., MAPK signaling pathway, AMPK signaling pathway, NF-kappa B signaling pathway, mTOR signaling pathway, and pancreatic cancer), and intriguingly, the chronic myeloid leukemia and acute myeloid leukemia pathways were also among the top enriched pathways (adjusted P < 0.05), indicating that PARP10 is highly relevant to AML, potentially through these functional pathways (Figure 5A and Supplementary Table 2). However, KEGG pathway analysis of the 94 negatively co-expressed genes (r < −0.5; P < 0.05) did not illustrate any significantly enriched pathways (adjusted P > 0.05) (Supplementary Table 3), although this could be due to the limited number of genes.

We next performed gene set enrichment analysis (GSEA) to further explore the involved biological pathways and cofactors of PARP10 in AML. Median expression of PARP10 was used as threshold to define PARP10 high or low group in GSE15061 dataset. For the PARP10 high expression group, the gene sets were significantly enriched in the MYC oncogenic signature (NES = 1.8; P = 0.006), MYC binding (NES = 1.67; P = 0.000), NFKB signaling (NES = 1.76; P = 0.008), and EZH2 targets (NES = 1.87; P = 0.004) (Figure 5B–5E). STRING protein–protein interaction network analysis also demonstrated interaction of PARP10 and MYC (Supplementary Figure 5). As MYC, NFKB, and EZH2 are key transcriptional factors and epigenetic regulators in tumorigenesis [27–29], our data suggest that PARP10 could be involved in transcription and epigenomic regulation in AML.

PARP10 knockout impaired AML cell proliferation in vitro

To further gain insight into the function of PARP10 in AML proliferation, we performed PARP10 knockout in the MOLM13 cell line using CRISPR/Cas-mediated gene editing. Genotyping of the genomic DNA regions that encompass targeting sites by two guide RNAs [24] successfully validated the random insertions and/or deletions introduced by CRISPR/Cas9-mediated PARP10 editing in MOLM13 cells (Figure 6A and Supplementary Table 4). In growth competition assays, which tracked the percentage of sgPARP10–RFP cells in MOLM13-Cas9 cells relative to the non-transduced, the RFP negative cells at days 0, 1, 3, 5, and 7 indicated that the expression of sgPARP10 reduced the growth of MOLM13-Cas9 cells compared to the sgEV control cells (Figure 6B). We also established proliferation curves for the control and PARP10 knockout cells and found PARP10 knockout, by guide RNA1, impaired the proliferation of MOLM13 cells, although it’s not significant at day 8 (P = 0.059), whereas PARP10 knockout by guide RNA2 significantly impaired the proliferation of MOLM13 cells (P < 0.05) (Figure 6C and 6D). Thus, our bioinformatic analysis and functional experiments highlighted the important role of PARP10 in AML cell proliferation, indicating that targeting PARP10 might provide a novel strategy for treating AML.