Aging-related genes related to the prognosis and the immune microenvironment of acute myeloid leukemia

Acute myeloid leukemia (AML), one of the common malignancies of the hematologic system, has progressively increased in incidence. Aging is present in both normal tissues and the tumor microenvironment. However, the relationship between senescence and AML prognosis is still not elucidated. In this study, RNA sequencing data of AML were obtained from TCGA, and prognostic prediction models were established by LASSO-Cox analysis. Differences in immune infiltration between the different risk groups were calculated using the CIBERSORT and ESTIMATE scoring methods. The KEGG and GO gene enrichment and GSEA enrichment were also used to enrich for differential pathways between the two groups. Subsequently, this study collected bone marrow samples from patients and healthy individuals to verify the differential expression of uncoupling protein 2 (UCP2) in different populations. Genipin, a UCP2 protein inhibitor, was also used to examine its effects on proliferation, cell cycle, and apoptosis in AML cell lines in vitro. It showed that aging-related genes (ARGs) expression was correlated with prognosis. And there was a significant difference in the abundance of immune microenvironment cells between the two groups of patients at high risk and low risk. Subsequently, UCP2 expression was found to be elevated in AML patients. Genipin inhibits UCP2 protein and suppresses the proliferation of AML cell lines in vitro. ARGs can be used as a predictor of prognosis in AML patients. Moreover, suppressing UCP2 can reduce the proliferation of AML cell lines, alter their cell cycle, and promote apoptosis in vitro.


Introduction
AML, one of the common hematologic malignancies, is usually characterized by the accumulation of myeloid progenitor cells in the bone marrow and peripheral blood and has a very poor prognosis. Although many different treatments are available, studies have shown that the 5-year survival rate for AML is only 24%. The main treatment remains chemotherapy, but resistance mechanisms are very common in AML, and the transition to drug resistance in patients after chemotherapy is a major focus and difficulty in the treatment of AML, and there is now a lot of literature on the different resistance mechanisms [1]. However, the prognostic guidelines for AML are not uniform, and this calls for research to explore more prognostic signatures. With the development of bioinformatics, we can use new methods to provide more clinical guidance for survival prediction and therapeutic target selection [2,3].
Indeed, cells undergo senescence regardless of their age. Although cellular senescence in tumors can also play a role in tumor suppression and tissue repair, studies have also demonstrated that this process can promote tumor proliferation, invasion, etc. [4,5]. Senescent cells can secrete more Dongxu Gang and Yinyan Jiang have contributed equally to this work. 1 3 cytokines, chemokines, growth regulators, and other factors [6]. Aging-related genes (ARGs) remain unstudied in AML, although they have been used to predict disease prognosis [7]. The prognostic role of ARGs and their mechanism of action remain unclear.
UCP2 is a mitochondrial protein that regulates mitochondrial function [8]. It is commonly studied in non-tumor diseases such as diabetes and obesity. Recently, it has been shown to have anti-tumor effects in a variety of cancers [10,11]. However, there is still a gap in the field of research on AML.
In this study, we investigated the prognostic value of ARG in acute myeloid leukemia and selected one of the UCP2 genes for the next study.

Acquisition of ARGs
ARGs were collected from the GeneCards database (https:// www. genec ards. org/), which provides comprehensive information on human genes. The term "aging" was used as a keyword search, and genes with a correlation score > 8 were filtered as ARGs in the results.

Collection of data sets
We collected 200 available samples from The Cancer Genome Atlas (TCGA), including level-3 RNA-seq expression data from 132 patients. All samples enrolled in the cohort ensured appropriate prognostic data and other clinical information. This data from the TCGA is used as a training cohort to build the risk model. The microarray data and corresponding survival information for the remaining 510 samples were obtained from 2 different Gene Expression Omnibus (GEO) datasets. Five hundred and ten samples from GSE12417 and GSE71014 were included and again ensured that the corresponding clinical information and prognostic data were included for all samples before use.

DEG analysis
TCGA data identified 9,325 genes associated with prognosis in AML by univariate COX regression. These prognosisassociated genes were then intersected with the collected aging genes and this resulted in 75 aging genes associated with AML prognosis. Differential expression analysis of patients at high and low risk in the TCGA cohort was performed by the limma package and visualized as a volcano plot. Adjusted P value (adj. P) < 0.05 and |fold change (FC)|> 0.5 were considered statistically significant for identifying DEGs.

Building and validating the risk model
We used the LASSO regression method to construct the obtained prognosis-related ARGs as a multivariate model of ARGs. Median values were used to classify patients into high-or low-risk groups. Meanwhile, Kaplan-Meier survival analysis was constructed and the log-rank test was used to assess overall survival (OS) between groups. The sensitivity and specificity of prognostic performance were viewed by receiver operating characteristic (ROC) curves. The area under the curve (AUC) values indicated discrimination.

Tumor immune microenvironment landscape and its potential impact on immunotherapy
We assessed the tumor immune microenvironment using different methods on data of AML patients from the TCGA. CIBERSORT was used to calculate the infiltration abundance of 22 immune cell types in AML patients, grouped according to high-and low-risk populations. ESTIMATE scores of immune cells, stromal scores, immune scores, and tumor purity in AML patients were estimated using the ESTIMATE algorithm. Six immune cell abundances were also calculated using the EPIC score. Samples with P < 0.05 were selected for further analysis.

Functional enrichment analysis
Gene enrichment analysis is an important tool to integrate genes with function, and Gene Set Enrichment Analysis (GSEA) annotation was used to find potential mechanisms of aging-related genes in AML. We used both Gene Ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) to identify which molecular mechanisms differ between high-risk and low-risk patients.

UCP2 pan-cancer analysis
We downloaded the uniformly normalized pan-cancer dataset: TCGA TARGET GTEx (PANCAN, N = 19,131, G = 60,499) from the UCSC (https:// xenab rowser. net/) database, from which we further extracted the expression data of ENSG00000175567 (UCP2) gene in each sample. We then analyzed the relationship between UCP2 and the prognosis of various tumors and the correlation with immunomodulatory genes, respectively. Then we calculated the relationship between gene expression and tumor stemness [12].

Clinical samples and qRT-PCR analysis
Bone marrow samples were obtained from 19 AML patients and 19 healthy donors from the First Hospital of Wenzhou Medical University. Our study was approved by the ethics committee of the First Hospital of Wenzhou Medical University. Quantitative real-time PCR (qRT-PCR) total RNA was extracted using TRIZOL reagent (Life Technologies). Reverse transcription was then performed using the HiScript Q RT SuperMix kit (Vazyme, Nanjing, China). Then qRT-PCR was performed to assess mRNA expression using SYBR Green Master Mix (CWBIO, Jiangsu, China) in an Applied Biosystems QuantStudio 3 Real-Time PCR System (Thermo Fisher Scientific, MA, USA) as previously described.

Cell culture
Cell culture of human leukemia cells HL-60, Kasumi-1 cell line was purchased from BNCC (Henan, China) and maintained according to the manufacturer's instructions. We have confirmed the cell lines used in the experiments by specialized STR analysis and tested them for mycoplasma contamination.

Cell viability assay
Cell proliferation was determined by Cell Counting Kit-8 (meilunbio, Dalian, China) assay. Cells were inoculated in 96-well plates at a density of 1 × 10 4 cells per well. Cells were treated with Genipin for 48 h, followed by CCK-8 solution at 37 °C for 3 h.

Colony formation assay
Cells were inoculated at a low density of approximately 1 × 10 3 cells per well into 6-well plates lined with methylcellulose. Cells were cultured for 7 days. Photographs were taken under an inverted microscope lens.

Apoptosis analysis (flow cytometry)
Cells were untreated or treated with Genipin for 48 h, with 5 μL Annexin V-fluorescein isothiocyanate reagent and 10 μL 7-Aminoactinomycin D (7-AAD) reagent for 30 min at room temperature and protected from light. Cells were analyzed by flow cytometry (Beckman Coulter, Brea, CA, USA) immediately after termination of staining.

Cell cycle analysis
For cell cycle assays, cells were treated with Genipin for 48 h. Cells were then stained with propidium iodide at a final concentration of 0.05 mg/mL and incubated at 4 °C for 20 min in the dark. Data were collected and analyzed using flow cytometry.

Incorporation of mitochondrial reactive oxygen species (ROS)
HL-60, Kasumi-1 cells were inoculated in 6-well plates and treated with Genipin for 6 h. After incubation with MitoSOX (Thermo Fisher Scientific) at 37 °C for 30 min, cells were collected by centrifugation and analyzed for mitochondrial ROS using flow cytometry.

Statistical analysis
All RNA-seq expression data were log 2 normalized for further analysis. At the same time, some of the figures are made by the SangerBox Website (http:// sange rbox. com) [13]. Results with p values less than 0.05 were considered significantly different. R (version 3.6.1) and GraphPad Prism (version 8.0.1) were used for statistical analysis.

Identification analysis of OS-related ARGs in AML
The expression of 75 of AGRs in AML patients was considered to be meaningfully correlated with OS (Fig. 1A). Then we integrated the survival time and survival status of AML patients, together with the ARGs expression data, and performed regression analysis using the Fig. 1 Identification analysis of OS-related ARGs in AML and validation of ARG-related prognostic features in the training set. A Prognosis-related genes that overlap between TCGA and ARGs. B Selection of optimal parameters (lambda) in the least absolute shrinkage and selection operator (LASSO) model; dashed vertical lines are drawn at the optimal values using the minimum criterion. C LASSO coefficient curves for 75 prognosis-related ARGs with non-zero coef-ficients determined by the optimal lambda. D Overall survival curves stratified by the low-and high-risk group. E Distribution of risk scores calculated from risk scores and the distribution of patients in low-and high-risk fractional groups based on survival status. A heat map of the ARGs expression chart is shown below. F ROC curves for ARG-based overall survival prediction LASSO-Cox method. We finally took the lambda minimum value:0.136228890858806 and screened out 12 genes (Fig. 1B,C). The

Validation of ARG-related prognostic features in the training set
AML patients were divided into two groups, high and low risk, by median risk score. Kaplan-Meier survival analysis demonstrated significant differences in OS between the two groups (Fig. 1D). Next, we used time-dependent ROC curves to assess the predictive efficacy for different time points in the training set (Fig. 1F). The AUC values at 1, 3, and 5 years were 0.82, 0.83, and 0.90 (Fig. 1F). Respectively, they indicated that the predictive efficacy of the feature was high. As the risk score increased in the high-risk versus low-risk group, the survival time decreased gradually in both groups. The expression profile heat map of the 12 ARGs is shown in Fig. 1E. In the high-risk group, AIFM1, PTPN1, SOCS2, TERF2, TGFB1, UCP2, and GH1 were highly expressed, while DLL3, GDF11, HBP1, INSR, and TPP2 in the low-risk group expression was higher (Fig. 1E).

Patients at different risks showed different immune status
Next, this study further explored the differences in immune status between patients in different risk groups. First, the CIBERSORT algorithm was used to assess the percentage of immune cell types in each patient. The results show the percentage of immune cells in AML patients in the low-risk group versus the high-risk group ( Fig. 2A). In particular, patients in the high-risk group had elevated proportions of T-cells-regulatory-(Tregs), NK-cells-activated, and Monocytes, while the low-risk group showed higher proportions of T-cells-CD4-memory-resting, macrophages-M0 dendritic-cells-activated, and Mast-cells-resting (Fig. 2B). The figure below also shows the correlation between different types of immune cells (Fig. 2C). Subsequently, the ESTI-MATE algorithm was used to assess the immune differences between the two groups of patients with different risks. The results showed that patients in the high-risk group had higher immune scores, stromal scores as well as assessment scores, and lower tumor purity compared to the low-risk group (Fig. 2D). In addition, the EPIC score was also used to assess immune differences, and the results showed a higher proportion of B-cells, endothelial, and macrophages, and a lower proportion of other cells in the high-risk group compared to the low-risk group, with no significant differences in other types of immune cells (Figs. 2E,F).

DEG and functional analyses
AML patients in the TCGA dataset were grouped according to high and low risk, and this was used to identify differential genes for high and low risk in the TCGA dataset. A total of 906 differential genes were screened, with 578 genes upregulated and 328 genes downregulated in the high-risk group compared to the low-risk group (Fig. 3A). Figure 3B shows the protein-protein interaction (PPI) networks between differential genes. GO gene enrichment analysis showed that DEGs were mostly enriched in biological processes such as the immune system process, immune response, and cell activation (Fig. 3C). In terms of cellular components, DEGs were mainly enriched in the vesicle, cytoplasmic vesicle part, cytoplasmic vesicle, and intracellular vesicle (Fig. 3D). Meanwhile, in molecular functions, DEGs mainly showed cytokine binding, cargo receptor activity, and peptide binding (Fig. 3E). In the KEGG enrichment analysis, DEGs were mainly enriched in the pathways of phagosome, hematopoietic cell lineage, tuberculosis, Staphylococcus aureus infection, and rheumatoid arthritis (Fig. 3F).
On this basis, GSEA was used to obtain the enrichment of DEGs in different pathways in this study, and the results showed that the high risk was mainly associated with ALZ-  (Fig. 3G). These results demonstrate that are mainly associated with immune-related pathways.

Validation of ARG-related prognostic features in an external dataset
To further validate the predictive efficacy of ARG-related prognostic features in the external dataset, the risk scores of different datasets (GSE12417, GSE71014) were calculated according to the feature formula. To calculate the optimal cutoff value of the risk score, patients were divided into two cohorts. In the GSE12417 cohort, OS showed differences A Immune cell type percentages in the low-and high-risk groups. B Differences in the abundance of immune cells between the highand low-risk groups. C Correlation matrix of immune infiltrating cells. D Stromal score, immune score, ESTIMATE score, and tumor purity calculated by ESTIMATE algorithm. E The abundance of six immune filtrating cells evaluated by EPIC. F EPIC assesses immune cell type differences in low-and high-risk groups between the high-and low-risk groups, with an AUC value of 0.61 for 1-year OS and 0.53 for 3-year AUC in this cohort (Fig. 4A,B); meanwhile, the high-risk group in the GSE71014 set had worse OS than the low-risk group, with a 1-year AUC of 0.58 and a 3-year AUC of 0.55 (Fig. 4C,D). The above suggests that this feature can be used to predict the prognosis of AML patients.

UCP2 expression levels are elevated in AML patients
We selected UCP2, a less studied gene in leukemia, as a target for follow-up research. The experiment showed that AML patients expressed higher UCP2 in bone marrow compared to healthy individuals (Fig. 4E), and the expression levels of UCP2 also differed in different cell lines. (Fig. 4F).

UCP2 pan-cancer immunoassay
Subsequently, this study calculated the difference in expression between normal and tumor samples in each tumor in the TCGA database and performed differential significance analysis; 24 tumors were observed to be significantly upregulated as shown in Fig. 5A and significant downregulation was observed in 3 tumors. Subsequently, Cox, proportional hazards regression mode was used to  (Fig. 5B). We next calculated the Pearson correlation between ENSG00000175567 (UCP2) and marker genes of the five types of immune pathways as shown in Fig. 5C. Obtaining DNAss tumor stemness scores calculated by methylation profiles in each tumor and calculating their Pearson correlation, we observed    (Fig. 5D).

Genipin can effectively inhibit cellular activity in vitro
To further investigate the role of UCP2 in AML, we selected Genipin as a protein inhibitor. It inhibits UCP2 in cells. It is now often used in studies of type 2 diabetes and has been reported to act in breast cancer cells in oncology. To clarify whether Genipin kills AML cells, two common AML cell lines (Kasumi-1, HL-60) were used in this study, and cell proliferation activity was assayed after 48 h of treatment with Genipin. The results showed that Genipin inhibited the proliferative activity of AML cells in a dose-dependent manner (Fig. 6A). At lower concentrations of Genipin, the experimental results showed that Genipin did not exhibit a killing effect on AML cell lines (Fig. 7A). It was only after the action of higher concentrations of Genipin (25-100 µM) that it showed toxic effects against HL-60, Kasumi-1. To see whether Genipin has leukemia-specific effects, we extracted CD34 + cells from the bone marrow of healthy donors and treated them with different concentrations of Genipin in vitro in culture. The results showed that the activity was not affected by the action of Genipin at low concentrations (Fig. 7B,D). When the concentration of Genipin was increased to 100 µM, it was able to inhibit its proliferation, but it did not inhibit healthy human CD34 + cells to the same extent as it did for AML cell lines (Fig. 7C,E). This is similar to the results of other studies in tumors, where the effective concentration of Genipin is currently large, and its application still needs to be further investigated. Afterward, the proliferative passaging ability of Genipin-treated cells was assessed by colony formation assay, as shown in Figs. 6B,C. Genipin treatment at 50 μM significantly reduced the number of colony formation of AML cells on day 7.

Genipin blocks the AML cell cycle in vitro and promotes apoptosis
The cell cycle distribution of each group was examined after treatment with different concentrations of Genipin for 48 h. As shown in Fig. 6D, in the HL-60 and Kasumi-1 cell line, Genipin significantly increased the proportion of G2 phase cells compared to the control, while the proportion of G1 phase cells was not reduced or slightly reduced, and the proportion of S-phase cells also decreased gradually with Genipin dose (Fig. 6E,F). These data suggest that Genipin may stunt the proliferation of AML cells by altering their cycle. We then examined the effects of low concentrations of Genipin on the cell cycle and apoptosis of AML cell lines. The results showed that lower concentrations of Genipin did not alter the cell cycle or the level of apoptosis in AML cell lines (Fig. 7F, G) (Fig. 8A). Next, this study also used flow cytometry to examine the number of Annexin V/7AAD-stained cells. The results showed that in both cell lines, apoptotic cells gradually increased with increasing doses of Genipin after 48 h of treatment. It indicates that Genipin can increase AML cell apoptosis in a dose-dependent manner (Fig. 6G).

Genipin affects the level of reactive oxygen species in AML cells
In this study, when AML cells were treated with low concentrations of Genipin for 6 h, ROS levels in both cell lines did not show significant changes at most concentrations of Genipin, but were significantly increased when HL-60 cell lines were treated with a concentration of 1 nM Genipin, which may be related to the specificity between cell lines. (Fig. 7H,I) In contrast, ROS levels in HL-60 cells decreased with increasing drug concentrations in response to higher concentrations of Genipin. In contrast, the Kasumi-1 cell line did not show a clear relationship between drug dose and ROS levels. This suggests that UCP2 may not affect cellular activity by influencing cellular ROS levels and, thus, cellular activity.
In this study, a dose-dependent decrease in ROS levels was found in the HL-60 cell line 6 h after Genipin treatment (Fig. 6H). In contrast, the Kasumi-1 cell line did not show significant changes in ROS levels, suggesting that UCP2 may not affect cellular activity by altering its ROS levels.

Discussion
AML is the common neoplastic disease of the hematologic system, and despite the many studies and results available, the prognosis for patients with AML remains poor. Worse, although most patients improve with treatment, a significant number of patients eventually relapse, which contributes to the low survival rate of AML. In recent years, there has been a lot of interest in predictive models for predicting the prognosis of AML [14,15]. While the role of ARGs in AML and their potential mechanisms are still not elucidated, this study constructed and validated ARG risk models and predicted their potential impact on the prognosis of AML patients, and we discovered that Genipin, a drug that inhibits UCP2 protein, can kill AML cells in vitro.
Cellular senescence is frequently mentioned in AML, and it has been shown that senescence-related factors can regulate the transition from acquired aplastic anemia (AA) and paroxysmal nocturnal hemoglobinuria (PNH) to secondary myelodysplastic syndromes (MDS) and AML [16]. Genetic targets commonly used to treat leukemia, such as BCL-2 and P53, can also act by regulating cellular senescence [17,18].

3
The immune microenvironment is present in various tumors and is strongly associated with tumor prognosis [19], while cellular senescence is also present in the bone marrow immune microenvironment, and it has been shown that T cell senescence and prognosis can be used to predict the prognosis of patients with AML [20], which has important roles in inducing drug resistance and mediating immune escape [21,22]. Another study showed that a higher proportion of Tregs in AML patients compared to healthy donors can also affect AML prognosis [23,24]. Activated NK cells kill cancer cells, and various related products are currently used to adjunctly target cancer therapy [25][26][27]. Resting CD4 + T cells, and resting mast cells are associated with tumor cell killing. In contrast, activated dendritic cells activate T cells in the tumor microenvironment and modulate the immune response. It has been shown that in cancer, senescent cancer cells elicit anti-tumor protection mediated by CD8 + T cells and that this protection is highly preferential. Induction of senescence in cancer cells enhances their ability to activate CD8 + T. This study suggests that senescent cancer cells can be used to develop a CD8 + T cell-dependent anti-tumor immune response [28]. SASP is the production by senescent cells of a variety of secreted proteins, including inflammatory cytokines, chemokines, stromal remodeling factors, and growth factors, which play critical but distinct roles in the tumor microenvironment. Senescence induces growth arrest of damaged cells and SASP recruits different immune cells to clear them. SASP can also lead to anti-tumor immune responses by recruiting immune cells. Macrophages interact with senescent cells and this role is crucial for mammals. Recent studies have shown that the recruitment of macrophages by senescent cells not only induces their senescence, but also affects their polarization [29]. Another study showed that senescent cells are able to disrupt the circadian gene regulation and function of macrophages [30]. These suggest that senescent cells have a significant impact on the immune microenvironment, which may account for the differences in their immune microenvironment. Our study shows that the tumor microenvironment differs significantly between high-and low-risk groups, suggesting that immunotherapy remains largely unexplored and may become a new therapeutic tool [31,32].
UCP2 belongs to the family of mitochondrial anion carrier proteins located on the inner mitochondrial membrane. Many reports have shown that it can induce adaptive responses to prevent oxidative stress [34]. ROS are mainly produced by mitochondria to maintain redox homeostasis in the body, and ROS are not only essential for the normal organism but also play a key role in tumors. A strong association between ROS and AML cell proliferation has been reported in the literature [35]. And either too high or too low ROS affects AML cell proliferation [36,37]. In this study, UCP2 levels were found to be closely associated with the prognosis of AML. Genipin is a specific UCP2 inhibitor that is derived from plants [38]. Its mechanism of action has been recently explained [39,40]. We used Genipin to act on AML cell lines in vitro. In this study, in vitro experiments showed that Genipin induces anti-tumor activity against AML by inhibiting cell proliferation and cell cycle, while also inducing apoptosis and scavenging high ROS levels in AML. However, there are still some shortcomings in our study; we have only discussed the possible effects of UCP2 on AML cells at the level of in vitro cell lines, but not the role that UCP2 plays in the microenvironment and the possible mechanisms. This suggests that drugs targeting UCP2 protein, especially Genipin, may become new targets for the treatment of AML.