Abstract

Background. Acute myeloid leukemia (AML) is a heterogeneous disease characterized by diverse genetic abnormalities. The NPM1 is the most commonly mutated gene in newly diagnosed patients. Optimizing risk stratification in this population could facilitate more rational clinical decision-making.

Objectives. To identify biomarkers that optimize risk stratification in AML patients with NPM1 mutations.

Materials and methods. Acute myeloid leukemia patients from multiple centers were included in this study. Univariate, multivariate and Kaplan–Meier survival analyses were used to assess risk factors and clinical outcomes. The gene set enrichment analysis (GSEA) was conducted to identify the related enrichment of biological function.

Results. TG-interacting factor 1 (TGIF1) is a good prognostic indicator of disease progression in AML patients. It is closely related to NPM1 mutation, in which age and TGIF1 expression are independent prognostic factors. Multicenter data sources have shown that high expression of TGIF1 is beneficial for AML, regardless of whether patients received bone marrow transplantation. In the NPM1-mutated AML group, age, FLT3-ITD and TGIF1 were independent prognostic factors. Moreover, the NPM1-mutated subgroup could be well dichotomized into 2 groups with distinct prognoses through TGIF1 combined with European LeukemiaNet (ELN) 2017 risk stratification.

Conclusions. The TGIF1 has an important value in the prognosis of AML. The NPM1-mutated patients were further subdivided into risk stratification groups based on TGIF1 expression, which could optimize the ELN 2017 to achieve individualized treatment.

Key words: risk stratification, NPM1, acute myeloid leukemia, TG-interacting factor 1

Background

Acute myeloid leukemia (AML) is a highly aggressive hematologic malignancy of the myeloid lineage characterized by extramedullary malignant infiltration of various organs, portending a poor prognosis.¹ For patients with AML who underwent conventional chemotherapy, the 5-year overall survival (OS) rate was 40–45% compared with 10–20% OS rate of the elder ones.^{2, 3} Although the efficacy of the intensive frontline therapy is encouraging, 30% of those patients still relapse after treatment.⁴ Furthermore, the pathophysiology of AML is tightly linked with distinct genetic and molecular abnormalities, and clinicians obtain prognostic information and classification due to the molecular changes occurring in AML. Despite the recent advances in the understanding of pathogenesis and therapeutic methods of AML, it is still intractable and its risk stratification and targeted therapy are complicated due to the cytogenetical, molecular and clinical heterogeneities. Since more and more abnormalities in gene expression have been reported in the occurrence and development of AML, it is apparent that finding more specific biomarkers for AML patients is conducive to patient classification, and serves as an essential prerequisite for individualized treatment.^{5, 6}

The NPM1-mutated AML is one of the monocytic subtypes of AML, which acts by translocating the NPM1 mutation gene to partner genes such as anaplastic lymphoma kinase gene (NPM-ALK) and normal karyotype gene (NPM-NK), rendering heterodimers, producing fusion proteins and promoting cytoplasmic delocalization and cellular transformation.^{7, 8, 9} About 30% of heterozygous NPM1-mutated AML cases occur in adults and 50–60% of them involve normal karyotype AML (NK-AML).¹⁰ However, the complex interactions of this mutation with some prognosis-related accompanying mutations including FLT3-ITD, NPM1/N-RAS and NPM1-RAD21, as well as their joint effect on the clinical outcome remain unclear.^{11, 12, 13, 14} Novel molecular information is needed to identify subsets of patients and improve the diagnosis and outcome prediction.

The TG-interacting factor 1 (TGIF1) is a member of the three-amino-acid loop extension (TALE) family of atypical homeodomain-containing transcription factors.¹⁵ It is known as a transcriptional repressor, which participates in various biological signaling pathways including TGF-β, retinoic acid signaling and others, in order to exert its suppressive effects by recruiting and interacting with other co-repressor complexes, such as mSin3/HDAC and CtBP.^{16, 17} As a critical suppressive factor of the important signaling pathway, the function of TGIF1 in tumorigenesis and tumor development has attracted considerable research attention. Therefore, in our study, we evaluated the expression of TGIF1 in patients with an initial diagnosis of AML, as well as explored the clinical and potential therapeutic value of TGIF1 in AML.

Objectives

Although several preclinical studies have confirmed that TGIF1 plays a key role in regulating hematopoietic stem cell self-renewal, and that a loss of TGIF1 can lead to the occurrence of AML, there is little evidence of clinical application of TGIF1 in guiding healthcare decisions. Therefore, in our study, we evaluated the expression of TGIF1 in patients with an initial AML diagnosis and demonstrated that TGIF1 plays an independent role in AML and NPM1-mutated AML in terms of the prediction of clinical outcomes, as well as provides useful insight for risk stratification and clinical individualized diagnosis and treatment.

Materials and methods

Data acquisition

The expression of TGIF1 was analyzed in the publicly available mRNA sequencing cohort from the Oregon Health and Science University (OHSU)-AML project (405 patients) that included detailed clinical data.¹⁸ Another source of clinical data and gene expression for validation was The Cancer Genome Atlas (TCGA)-AML (151 patients).¹⁹ Validation data analyses for survival outcomes were also obtained from the Therapeutically Applicable Research to Generate Effective Treatment (TARGET)-AML project (296 patients), and Gene Expression Omnibus (GEO) data that are accessible under accession No. GSE37642. We have downloaded the published data until April 28, 2022.

The patients were classified according to the French–American–British classification systems and the risk group stratification was as per the National Comprehensive Cancer Network (NCCN) guidelines. Patients included in the study were evaluated for somatic mutations as well as fusion genes that are common in AML. The datasets generated in this study are available as the Supplementary material. The patient data were obtained from publicly available datasets and did not require further ethical approval.

Gene set enrichment analysis

Gene set enrichment analysis (GSEA, v. 4.0.1; http://www.gsea-msigdb.org/) is a widely used bioinformatics analysis tool that determines concordant differences between biological processes. The GSEA was used to verify the differences in molecular pathways between high-expression and low-expression TGIF1 groups. The input data matrix was derived from the OHSU cohort. The ‘c2.cp.kegg.v7.1.symbols.gmt’ were examined and 1000 permutations were run.

Statistical analyses

The univariate and multivariate Cox regression analyses were performed using the ‘survival’ R package, and the main code was provided in Supplementary material. The Kaplan–Meier survival analysis with log-rank test was used to evaluate the prognostic role of TGIF1 cutoff value, and AML patients were grouped according to the optimal difference in OS. Results are expressed as hazard ratio (HR) and 95% confidence interval (95% CI). Statistical analysis and graphing were performed using GraphPad Prism v. 8.02 (GraphPad Software, San Diego, USA) and R software v. 3.5.2 (R Foundation for Statistical Computing, Vienna, Austria). The Mann–Whitney test and χ² test were used to clarify the association between gene expression and clinicopathological characteristics, and when expected counts produced by IBM SPSS v. 26.0 software (IBM Corp., Armonk, USA) were less than 1, the Fisher’s exact test was performed (Table 1).²⁰ The value of p < 0.05 was considered statistically significant.

Results

Patient characteristics

A total of 405 AML cases were taken from the OHSU dataset and divided into 2 groups according to the expression level of TGIF1, for a further determination of the relevance between TGIF1 expression and other various clinical and molecular characteristics. The baseline characteristics of the patients are listed in Table 1. A considerable difference between TGIF1 high- and low-expression groups could be observed in cytogenetic risk stratification: the results showed that patients with high expression of TGIF1 tended to be allocated to favorable cytogenetic risk groups (p < 0.0001), whereas low TGIF1 expression was associated with adverse cytogenetic risk stratification (p < 0.0001). In addition, the TGIF1 high-expression group had a higher frequency of gene fusions CBFB-MYH11 (p = 0.0009) and PML-RARA (p = 0.0108) when compared with the TGIF1 low-expression group. There was no statistical association between TGIF1 and age, gender, white blood cell (WBC) count, platelet (PLT) count, serum creatinine (SCr), peripheral blast (PB) count, complex karyotype, DEK-NUP214, GATA2-MECOM, MLLT3-KMT2A, RUNX1-RUNX1T1, FLT3-ITD, or NPM1 mutations.

TGIF1 is a good indicator
of AML disease progression

Due to the baseline association between TGIF1 and established prognosis-related cytogenetic risk stratification, it was important to analyze the clinical outcomes of patients with AML with different TGIF1 expression levels. Four databases were analyzed including OHSU, TCGA, TARGET, and GSE37642. The Kaplan–Meier survival analysis of OHSU cohort demonstrated that patients with high TGIF1 levels had longer OS than those with low TGIF1 expression (p < 0.0001, Figure 1A). Similar results were obtained from mutual validations over 3 other independent datasets (p < 0.0001 compared to p < 0.0001 compared to p = 0.0115; Figure 1B–D). On the other hand, more favorable event-free survival (EFS) could be observed in the high TGIF1 expression group when compared to the patients with low TGIF1 expression (TCGA-AML cohort: HR = 0.5015, 95% CI: 0.2994–0.8399, p = 0.0035; TARGET-AML cohort: HR = 0.6997, 95% CI: 0.5304–0.9231, p = 0.0115; Figure 1E,F). Figure 1G presents a slightly lower cumulative incidence of relapse in the high TGIF1 expression group (TARGET cohort: HR = 0.7576, 95% CI: 0.5551–0.1034, p = 0.0759; Figure 1G).

Univariate and multivariate analyses for prognostic factors

Since the described results suggested the expression level of TGIF1 is closely correlated to a favorable AML prognosis, univariate and multivariate analyses were performed for further analysis. We evaluated clinical and molecular genetic factors of AML patients from the OHSU cohort, including age (≥60 years compared to <60 years), gender (male compared to female), FLT3-ITD (positive compared to negative), NPM1 (mutated compared to wild-type), CBFB-MYH11 (positive compared to negative), RUNX1-RUNX1T1 (positive compared to negative), MLLT3-KMT2A (positive compared to negative), PML-RARA (positive compared to negative), cytogenetic risk stratification (adverse compared to intermediate/favorable), and expression levels of TGIF1 (high compared to low). After multivariable analyses, age <60 years (HR = 1.94, 95% CI: 1.48–2.55, p < 0.0001) and high expression level of TGIF1 (HR = 0.65, 95% CI: 0.50–0.86, p = 0.0020) remained significant independent predictors of a good prognosis in AML patients (Table 2).

Effect of TGIF1 expression on disease progression in AML patients with and without bone marrow transplantation

Currently, the standard treatment protocol for most AML patients is intensive induction chemotherapy, and once a complete remission is achieved, appropriate postremission therapies are necessary, such as conventional chemotherapy and hematopoietic cell transplantation.²¹ To verify the role of TGIF1 in different treatment groups, we compared Kaplan–Meier curves for the survival of the patients with and without bone marrow transplantation. The results presented in Figure 2A,D (data from OHSU and TCGA cohort, respectively) show that bone marrow transplants can improve the OS rate by mutual validation. However, survival distribution curves of bone transplantation demonstrated a trend for shorter EFS than those of the standard chemotherapy group (p = 0.0825; Figure 2G). Considering post-transplant complications such as graft-versus-host disease, the results were acceptable. In the chemotherapy group, the high-TGIF1-expression subgroup showed better OS (both p < 0.0001 in OHSU and TCGA cohorts) and EFS (p = 0.0044) than the low-expression subgroup (Figure 2B,E,H). In the transplantation subgroup, high TGIF1 expression level was associated with better OS (p = 0.0357; Figure 2C, and p = 0.0051; Figure 2F), but no difference was found regarding EFS (p = 0.2394; Figure 2I). The findings indicate that high TGIF1 might be a beneficial factor in AML patients undergoing conventional chemotherapy or bone marrow transplantation.

Risk stratification optimization
based on AML cohort analysis

Patients from the OHSU cohort were classified according to the European LeukemiaNet (ELN) 2017 risk stratification systems²² into favorable (n = 117), intermediate (n = 135) and adverse (n = 152) risk groups (Figure 3A). Previous studies have suggested that there was a significant interaction effect between mutant NPM1 and FLT3-ITD, leading to different treatment responses and survival prognoses.^{23, 24} Figure 3B demonstrates the effect on OS according to the interrelations between NPM1 and FLT3-ITD mutations as per the survival curve, which was consistent with the current guidelines.^{25, 26} However, mutant/wild-type NPM1 alone cannot independently predict OS in AML patients (p = 0.9253; Figure 3C). Interestingly, TGIF1 was proved to be independently associated with favorable clinical outcomes in the NPM1-mutated subgroup of AML (Table 3). Therefore, attempts were made to reanalyze the role of TGIF1 played in the NPM1-mutated subgroup from the OHSU cohort. The NPM1-mutated group with low TGIF1 levels had a shorter OS when compared with the NPM1 wild-type group (p < 0.05; Figure 3D). Besides, patients with high TGIF1 level had a longer OS than those with low TGIF1 in the NPM1-mutated subgroup (p < 0.01; Figure 3D). These results prove the prognostic role TGIF1 plays in AML, especially in the NPM1-mutated subgroup. Furthermore, by integrating the information on AML prognosis relevant to NPM1 mutation and TGIF1 expression level, improved risk stratification may be revealed (p < 0.0001; Figure 3E), providing a theoretical basis and new ideas for treatment. For further validation, the same analyses were conducted with another TCGA dataset, and the results confirmed that TGIF1 expression significantly affected the prognosis of the NPM1-mutated group and supported risk stratification optimization (Figure 4). Likewise, the TGIF1 expression remained a significant predictor of EFS in NPM1-mutated AML patients, which is consistent with our previous findings (Supplementary material).

GSEA delineates biological pathways correlated with TGIF1 expression

The GSEA is an important approach to identifying gene expression-related pathways. We performed GSEA between high and low TGIF1 levels data sets to elucidate biological pathways correlated with TGIF1. Some enrichment results showed that there was a significant correlation between high- and low-expression groups: multiple pathways, including galactose metabolism, biosynthesis of unsaturated fatty acids, glycolysis gluconeogenesis, peroxisome proliferator-activated receptor (PPAR) signaling pathway, and several others displayed the enrichment of low expression of TGIF1 and ribosome in the high-TGIF1-expression group (Figure 5). The gene set pathways were listed in order of significance in Table 4. These results may provide a mechanistic explanation for the scientific value and clinical significance of TGIF1.

Discussion

Besides commonly reported cytogenetic changes and recurrent gene mutations mentioned in ELN risk classification,²² numerous novel biomarkers also are related to AML leukemogenesis and were used to predict clinical outcomes,^{26, 27, 28} providing insights to further disease mechanisms and therapeutic directions. The TGIF1 belongs to the family of TALE homeodomain, which regulates diverse cellular processes including proliferation, apoptosis and differentiation as a transcriptional repressor. According to recent studies, TGIF1 functions as a tumor suppressor in pancreatic ductal adenocarcinoma by inhibiting Twist1 expression and activity.^{29, 30} The deletion of TGIF1 can induce the activation of the HAS2-CD44 signaling pathway and the upregulation of PD-L1, resulting in promoting the development of pancreatic ductal adenocarcinoma (PDAC). Given the essential role TGF-β and retinoic acid (RA) signaling play in hematopoiesis, some studies have been extended in recent years, attempting to find out the connections between TGIF1 and hematological diseases.³¹ In myeloid lineage leukemias, TGIF1 seems to function as a vital regulator in normal and malignant hematopoiesis.^{15, 30} Hamid and Brandt³¹ proposed that TGIF1 may regulate the balance between proliferation and differentiation during the myelopoiesis of human myeloid leukemia cells. Yan et al. showed that TGIF1 acts as a novel regulator of normal hematopoietic stem cell (HSC) function by suppressing HSC maintenance, self-renewal and quiescence in mice, resulting in the reduction of malignant transformation risk and leukemic stem cell or progenitor cell survival with chemotherapy.¹⁵ Moreover, further exploration analyzed the specific role TGIF1 plays in leukemia initiation and progression.^{32, 33} For example, TGIF1 exerts antileukemic contributions by affecting TGF-β- and RA-driven functions.³¹ Willer et al. showed that TGIF1 interferes with MLL-rearranged AML maintenance by competing with MEIS1 for chromatin-binding regions.³⁰ Collectively, these findings demonstrated that TGIF1 plays an important role in stem cell function regulation by maintaining the balance between proliferation and differentiation, and might possess potential prognostic value.

In the present study, clinical and molecular characteristics data of AML patients from the OHSU study cohort were analyzed to explore the relevance between TGIF1 expression and AML clinical features. A strong relationship between TGIF1 and cytogenetic risk stratification was confirmed. Considering the tight association between cytogenetics and the prognosis of the disease,⁵ we examined TGIF1 expression compared to AML outcome, including OS, EFS and cumulative incidence of relapse (CIR) to assess the prognostic effects of TGIF1. Following the identification of the relevance between higher TGIF1 expression and better prognosis, the same analyses were expanded on other independent datasets for confirmation, while univariate and multivariate analyses were performed for prognostic independence. Therefore, TGIF1 expression was regarded as the independent factor predicting better survival in patients with AML. Further analyses on TGIF1 should be performed to ensure the association between its expression and AML subtypes. For the OHSU cohort, Figure 3D shows that AML patients with high TGIF1 level had a longer OS than those with low TGIF1 in the NPM1-mutated subgroup. We performed the Kaplan–Meier curve survival analysis, and the curve revealed a significant difference in survival between different expression levels of TGIF1 in the NPM1-mutated subgroup. This conclusion was congruent to TCGA cohort, which indicated that the favorable impact on prognosis that TGIF1 provided also applied for AML with NPM1 mutations.

The NPM1 and FLT3-ITD collectively determined AML prognosis when referring to the risk stratification systems by ELN 2017.²² Therefore, based on the information above, we successfully constructed a more practical stratification system by integrating prognostic information of different TGIF1 expressions and NPM1-mutant AML into the original one. The reliability of this improvement was corroborated by a comprehensive analysis of a separate study cohort.

Limitations

The study was based on information obtained from mutual validation of publicly available data. The strengths of the trial include its strong eligibility criteria and a uniform treatment regimen according to standard guidelines. Despite the fact that our results provide a novel risk stratification option for NPM1-mutant AML, there are still certain limitations. Our analysis has a retrospective study design, and thus the accuracy rate may be lower in small sample cases. This work is based on the results of an RNA-sequencing dataset, and morphological insights remain to be explored. Further verification of the reliability is needed for the multivariate analysis to identify predictors after univariate analysis in statistical descriptions, similarly to previous studies.^{33, 34}

Conclusions

Our study proved that the upregulation of TGIF1 is closely associated with favorable prognosis in AML, and adding the expression level of TGIF1 examination can optimize risk stratification with NPM1-mutant AML, enhancing sensitivity and specificity in patient classification as well as providing reliable evidence for clinical decision-making. Further research is needed for new biomarkers and their combinations for personalized treatment.

Supplementary material

The supplementary files are available at https://doi.org/
10.5281/zenodo.7414416. The package consists of the following files:

Supplementary Information 1. The difference analysis results of 2 groups of continuous variables involved in this study, including age, WBC, PLT, bone marrow (BM) blasts count, PB, and SCr.

Supplementary Information 2. Cox regression based on Schoenfeld residuals for testing proportional hazards assumptions.

Supplementary Information 3. The R code used in the proportional hazards assumption in this study.

Supplementary Information 4. The R code used in the univariate analysis in this study.

Supplementary Information 5. The R code used in the multivariate analysis in this study.

Supplementary Information 6. The R code used in the OS analysis in this study.

Supplementary Information 7. Event-free survival of AML patients was analyzed using TCGA-AML data.