Cysteine- and glycine-rich protein 1 predicts prognosis and therapy response in patients with acute myeloid leukemia

Bone marrow samples from 224 newly diagnosed AML patients and 23 healthy individuals were obtained at the First Affiliated Hospital of Zhengzhou University between February 2017 and July 2020. We named our cohort as ZZU cohort. The inclusion criteria included: 1) patients were diagnosed as AML in our hospital; 2) patients were ≥ 14 years old; and 3) patients accepted at least one course of chemotherapy in our hospital. The exclusion criteria were patients who didn’t treat or treat elsewhere and patients with acute promyelocytic leukemia. Patient clinical information, such as age, gender, risk stratification, FAB classification, gene mutations, and other basic information, was gathered from electronic medical records. All data were collected independently by two researchers who received training to reliably complete all assessments. Details of treatment regimens are reported [15, 16]. Induction chemotherapy regimens include IA and DA regimens: standard-dose cytarabine (Ara-C) 100–200 mg·m⁻²·d⁻¹ × 7 d combined with idarubicin 10–12 mg·m⁻²·d⁻¹ × 3d or daunorubicin 60 mg·m⁻²·d⁻¹ × 3d. The treatment plan after remission is high-dose Ara-C (2–3 g/m², once every 12 h, 3d). Those who have not undergone HSCT share four courses; those who receive HSCT share two courses, followed by HSCT. For elderly patients over 60 years old or patients who cannot tolerate intensive chemotherapy, they received chemotherapy with demethylating drugs ± CAG ± venetoclax regimen until progression. In total, 38 subjects received an allogeneic HSCT. Subjects were followed up until death, loss to follow-up or December 2022. The follow‐up was conducted through in‐clinic follow‐up and telephone calls. Patients were classified into low CSRP1 group and high CSRP1 group according to the expression of CSRP1 at diagnosis. The clinical significance of CSRP1 in the AML cohort was analyzed retrospectively. Complete remission, relapse, risk stratification, overall survival (OS), and relapse-free survival (RFS) were defined as described according to the 2017 ELN recommendations from an international expert panel [17]. The study was approved by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University, and written informed consent was exempted by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University.

Bone marrow mononuclear cells were obtained via density gradient centrifugation. Total RNA was extracted using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). The cDNA was synthesized using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA) (18). CSRP1 transcript levels were detected by the Taqman method using real-time quantitative polymerase chain reaction (RT-qPCR) as previously described [18]. Serial dilutions of plasmids expressing CSRP1 and ABL1 were amplified to construct standard quantification curves [19]. The CSRP1 and ABL1 copy numbers were calculated from standard curves. As previously described, the CSRP1 transcript levels were calculated as the ratio of the CSRP1 copy number/ABL1 copy number [20]. The primers and probe sequences are shown in Table S1.

We obtained clinical information and gene expression profiles for AML patients from three independent sources. The first source is Tyner's study, which included 405 AML patients (hereafter referred to as the Beat-AML dataset) [21]. The second source is the TCGA database, which provides RNA-seq expression data and detailed clinical data on 173 AML patients (hereafter referred to as the TCGA-LAML dataset) [22]. The third source is the GSE12417 from the GEO database (https://​www.​ncbi.​nlm.​nih.​gov/​geo/​; hereafter referred to as the GSE12417 dataset) [23]. The AML samples' Level 3 HTSeq-FPKM and HTSeq-Count data were downloaded from the TCGA website (https://​portal.​gdc.​cancer.​gov/​repository) for further investigation.

We utilized the pROC package to determine the optimal cutoff value for each gene and divided AML patients into two subgroups: the “high-expression” and “low-expression” groups according to the cutoff value of the gene. Then we performed Kaplan–Meier survival analysis to investigate the prognostic value of gene expression using the survival package.

The DESeq2 R package was used to compare the expression data of low and high expression of CSRP1 (cutoff value of 50%) in AML samples to identify DEGs [24]. Heatmap analysis of the top 20 DEGs was performed.

Functional enrichment analysis was performed on DEGs with |logFC| > 0.5 and padj < 0.05. The ClusteProfiler package [25] in R was used to implement Gene Ontology (GO) functional analysis, which included cellular component (CC), molecular function (MF), and biological process (BP) analysis, as well as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis.

CSRP1 immune infiltration analysis was carried out using ssGSEA utilizing the GSVA package [26] in R. As previously stated [27], 24 kinds of infiltrating immune cells were obtained. The Spearman correction was used to examine the relationship between CSRP1 and the enrichment scores of 24 different kinds of immune cells. The Wilcoxon rank-sum test was used to examine the enrichment scores of the groups with high and low CSRP1 expression.

The Search Tool for the Retrieval of Interacting Genes (STRING) database (http://​string-db.​org) was used to predict the DEG PPI network [28]. The cutoff criterion was set at an interaction score level of 0.4. The PPI network was mapped with Cytoscape (version 3.6.1) [29], and the most critical modules in the PPI network were identified with MCODE (version 1.5.1) [30]. The following were the selection criteria: MCODE scores greater than 5, degree cutoff = 2, node score cutoff = 0.2, Max depth = 100, and k-score = 2.

A nomogram was developed using the rms R package (version 6.2–0) to individualize the prediction of overall survival (OS) in AML patients, which comprised key clinical features and calibration plots. The calibration curves were visually assessed by plotting the nomogram-predicted probabilities against the observed rates, with the 45 line representing the best predictive values. All statistical tests were two-tailed, and the statistical significance threshold was set at 0.05.

R (3.6.3) was used to perform and show all statistical analyses and figures. Wilcoxon rank-sum test was used to evaluate CSRP1 expression in unpaired samples, while Wilcoxon signed-rank test was utilized in paired samples. Quantitative data are presented as means ± SD (standard deviations) or medians with interquartile ranges. To assess the connection between clinical/cytogenetic features and CSRP1 expression, the Kruskal–Wallis test, Wilcoxon signed-rank test, and logistic regression analysis were performed. The prognostic variables were evaluated using Cox regression analysis and the Kaplan–Meier technique. The cutoff of CSRP1 was calculated by X-tile. There were few missing data on the key variables in our main analysis. Participants who were lost to follow-up were deemed alive after loss to follow-up. Multivariate analyses were conducted, controlling for confounders.

RNA-seq data were obtained in the TCGA and GTEx databases. By comparing the expression of CSRP1 in normal samples in the GTEX database and corresponding tumor samples in the TCGA database, CSRP1 was found to be significantly overexpressed in 7 cancers, including acute myeloid leukemia (LAML), and underexpressed in 18 cancers (Fig. 1A). For the CSRP gene family, CSRP1 was significantly overexpressed in AML, while CSRP3 was significantly reduced in AML compared to healthy controls (Fig. 1B). Furthermore, by comparing the expression of CSRP1 in 23 normal bone marrow (NBM) and 224 newly diagnosed AML patients in the ZZU cohort, we confirmed that CSRP1 was overexpressed in AML (Fig. 1C).

The main clinical characteristics of AML in the ZZU cohort are shown in Table 1. A total of 224 AML patients were included in the prognostic analysis. The median follow-up time was 529 days. Up to the follow-up time, 138 patients died. The AML patients were divided into the low CSRP1 group (109 cases) and the high CSRP1 group (115 cases) according to the cut-off CSRP1 expression (225%). The high CSRP1 group had higher BM-blasts and more DNMT3A mutations than the low CSRP1 group (P < 0.05, Table 1 and Fig. 5G). We also compared the expression of CSRP1 in AML patients by the other characteristics. The expression levels of CSRP1 showed no significant difference in patients with different gender, age, WBC, cytogenetic risk group, BM-blasts, and PB-blasts (P > 0.05; Fig. 2A–F).

Patients with high CSRP1 expression had a significantly worse prognosis than those with low CSRP1 expression (hazard ratio [HR], 2.36 (1.53–3.64); P < 0.001; Fig. 3A) in the TCGA-LAML dataset. The predictive significance of elevated CSRP1 was verified in the Beat-AML dataset (Fig. 3B), the ZZU cohort (Fig. 3C), and the GSE12417 dataset (Fig. 3D–E). The time-dependent ROC curve from the TCGA-LAML dataset demonstrated that CSRP1 was an excellent predictor of AML patient survival.

Based on the TCGA-LAML dataset, independent prognostic factors for worse OS included CSRP1 (high vs. low, P < 0.001), cytogenetic risk (adverse vs. favorable, P < 0.001; intermediate vs. favorable, P = 0.002), and age (> 60 vs. ≤ 60, P < 0.001) (Table 2 and Fig. 4A). The independent prognostic value of these three factors was further confirmed in the ZZU cohort (Fig. 4B).

A nomogram was developed based on the Cox regression analyses in the TCGA dataset (Fig. 4C) and the ZZU cohort (Fig. 4D) to better predict AML patients' prognoses. Three independent prognostic variables, age, cytogenetic risk, and CSRP1 expression, were included in the model at a statistical significance level of 0.05. A point scale was utilized to allocate points to these factors based on multivariate Cox analysis. Results from the nomogram calibration curve of OS prediction were consistent with observations of all patients in both the TCGA dataset (Fig. 4E) and the ZZU cohort (Fig. 4F).

The gene expression profiles of the high and low CSRP1 groups were analyzed for differences in median mRNA expression. A total of 2758 DEGs from RNA-seq-HTSeq counts were found to be statistically significant between the high and low CSRP1 groups (|log fold change (logFC)|> 0.5, P adj < 0.05). (Fig. 5A). The heat map depicted the top ten up-regulated DEGs and top ten down-regulated DEGs between the high and low CSRP1 groups (Fig. 5B).

Biological process (BP) associated with high CSRP1 included neutrophil activation, neutrophil activation involved in immune response, neutrophil degranulation, and neutrophil mediated immunity (Fig. 5C). Cellular components (CC) associated with high CSRP1 included secretory granule membrane, ficolin-1-rich granule, tertiary granule, and cell leading edge (Fig. 5C). Molecular function (MF) associated with high CSRP1 included actin binding, Rho GTPase binding, cytokine binding, and cytokine receptor activity (Fig. 5C). KEGG included phagosome, cytokine-cytokine receptor interaction, lysosome, oxytocin signaling pathway, cell adhesion molecules, rap1 signaling pathway, chemokine signaling pathway, gap junction, HIF-1 signaling pathway, proteoglycans in cancer, Fc gamma R-mediated phagocytosis, natural killer cell mediated cytotoxicity, JAK-STAT signaling pathway, and apoptosis (Fig. 5D).

The prognostic value of the top ten up-regulated DEGs and top ten down-regulated DEGs was further investigated. Six of the top ten DEGs that were up-regulated (ONECUT2, CES1, CD163, KCNN2, THBS1, and B3GALT1) and six of the top ten DEGs that were down-regulated (MYT1L, ASTN1, DLL3, MYO18B, POU4F1, and PCDH10) also had vital prognostic significance in the TCGA AML dataset (Fig. 6).

CSRP1 expression in the AML microenvironment was associated with the level of immune cell infiltration as measured by ssGSEA. Mainly, CSRP1 was negatively related to T cells and T helper cells and positively associated with macrophages (Fig.S1C), neutrophils (Fig.S1D), NK CD56 dim cells, NK CD56 bright cells, Tem, and iDC, TFH, Th17 cells, Th1 cells, eosinophils, Treg, and aDC (Fig.S1A). Patients with high CSRP1 had higher enrichment scores of aDC, Eosinophils, iDC, macrophages, neutrophils, NK CD56 bright cells, NK CD56 dim cells, Tem, Th1 cells, Treg, and lower enrichment scores of T helper cells.

The PPI network was constructed using the String website and the data were generated by Cytoscape (MCODE plug-in). A total of 321 DEGs were imported into the PPI network. Complete module including all genes is presented in Fig.S2A. We obtained 235 nodes and 900 edges. Nineteen genes were present in the most significant module (MCODE score: 7.667; Fig.S2B). The TCGA-LAML dataset was then used for OS analysis on the hub genes. Nine of the 19 hub genes, including CD163, CX3CR1, C5AR1, THSD7A, ADMATS18, IL10, THBS1, ADAMTS15, and LILRB2, were correlated with OS in AML (P < 0.05, Fig.S3).

We utilized the “oncoPredict” tool to estimate sensitivity to frequently used chemotherapy agents to better correlate the CSRP1 expression with clinical practice. Accordingly, drug sensitivity of patients in high and low CSRP1 groups to multiple chemotherapy agents including 5-fluorouracil, gemcitabine, rapamycin, cisplatin, and fludarabine was predicted. Based on the findings, the high CSRP1 groups of patients in the TCGA datasets showed higher sensitivity to 5-fluorouracil, gemcitabine, rapamycin, and cisplatin and lower sensitivity to fludarabine (Fig.S4). The relationship of sensitivity to other chemotherapy agents with CSRP1 expression is listed in the supplementary data.