High ROBO3 expression predicts poor survival in non-M3 acute myeloid leukemia

Abstract

Roundabout guidance receptor proteins are crucial components of the SLIT/ROBO signaling pathway. This pathway is important for the nervous system and in embryonic development. Recently, increasing evidence has shown that roundabout guidance receptor proteins and the SLIT/ROBO signaling pathway also participate in tumorigenesis. Here, by analyzing transcriptome data from the TCGA and GEO databases, we found that ROBO3 is highly expressed in non-M3 acute myeloid leukemia. High ROBO3 expression was associated with increased age at diagnosis and poorer risk classification (both P < 0.01). Patients with high ROBO3 expression had higher rates of TP53 and RUNX1 mutations (both P < 0.05). Significantly worse overall survival and event-free survival were observed in high ROBO3 expression patients compared with low ROBO3 expression patients (OS: P = 0.004; EFS: P= 0.012). High ROBO3 expression was also associated with poorer overall survival and event-free survival in a subgroup of patients who received intensive chemotherapy (OS: P = 0.024; EFS: P = 0.040). Moreover, multivariate analysis indicated that high ROBO3 expression was an independent risk factor for poor overall survival in non-M3 acute myeloid leukemia patients who are younger than 60 and received intensive chemotherapy during remission induction. Bioinformatics analysis by Kyoto Encyclopedia of Genes and Genomes and Gene Ontology revealed that high ROBO3 expression significantly altered cell adhesion and extracellular matrix-related pathways (adjusted P < 0.05). Taken together, the data demonstrate that ROBO3 is upregulated in non-M3 acute myeloid leukemia and may be a potent biomarker of inferior prognosis.

Keywords

ROBO3 expression acute myeloid leukemia survival clinical

Impact statement

Our data for the first time revealed that ROBO3 is aberrantly expressed in non-M3 AML. High ROBO3 expression is associated with increasing diagnosis age, poorer risk classification, and inferior gene mutations. Significantly, worse overall survival and event-free survival were observed in ROBO3 high expression patients. Multivariate analysis indicated that high ROBO3 expression was an independent risk factor for poor overall survival in non-M3 AML patients who are younger than 60 and received intensive chemotherapy during remission induction. Functional pathways analysis revealed that cell adhesion and extracellular matrix-related pathways may take part in high ROBO3 expression associated non-M3 AML. Future mechanistic studies will further illuminate the roles of ROBO3 in non-M3 AML.

Introduction

Acute myeloid leukemia (AML) is the most common acute leukemia type that originates from myeloid precursors.^1,2 Standard anthracyclines and a cytarabine-based 7 + 3 regimen followed by hematopoietic stem cell transplantation (HSCT) is recommended for younger patients who can tolerate side effects.³ The use of hypomethylating agents, such as decitabine and azacitidine, also improves the outcome of older patients or those who are unsuitable for intensive chemotherapy.^4–6 However, the prognosis of AML patients is still not ideal¹ and conventional treatment methods often yield quite different therapeutic effects owing to the huge heterogeneity in AML. In the last decade, the precision medicine based on deep sequencing technology has uncovered kinds of important genetic alterations in human cancers including AML.^2,7 Drugs targeting specific genetic alterations have achieved remarkable success in the treatment of AML.^8–11

Roundabout guidance receptor (ROBO) family members consist of four conserved transmembrane receptor proteins (ROBO1, ROBO2, ROBO3, and ROBO4), which are mainly expressed in nerve cells.^12–14 Slit guidance ligand (SLIT) and the ROBO-mediated SLIT/ROBO signaling pathway have been studied for their critical roles in embryonic and nervous system development.^2,9,15 Mutations in ROBO proteins have been identified in some human neurological diseases.^16,17 Recently, increasing evidence has shown that the SLIT/ROBO signaling pathway also participates in various human cancers.^18–23

The SLIT/ROBO signaling pathway is involved in tumorigenesis mainly by promoting cell proliferation and migration. For example, high expressions of SLIT1, ROBO2, and ROBO4 are reportedly important in the tumorigenesis of colorectal cancer.^20,23 In nasopharyngeal carcinoma, high expression of ROBO1 is linked with increased tumor growth and poor survival.²² On the other hand, the SLIT/ROBO pathway also acts as a tumor suppressor in some conditions. For instance, upregulation of SLIT2 and SLIT3 can inhibit tumor cell invasion in pancreatic cancer and lung carcinoma, respectively.^24,25 Interestingly, some studies have reported that even the same member in SLIT/ROBO families may have exactly the opposite roles in different cancers. An example is the oncogenic role of ROBO2 in colorectal cancer and hepatocellular cancer, and its tumor suppressor role in ovarian cancer and pancreatic cancer.^20,26,27 Such opposing roles indicate the complex biological functions of the SLIT/ROBO signaling pathway in different human cancers.

Compared with other ROBO family members, the role of ROBO3 in human cancers is relatively less studied. Overexpression of ROBO3 plays important roles in the tumor development of pancreatic carcinoma by increasing Wnt/β-catenin pathway activity.¹⁸ Inactivation of ROBO3 by promoter hypermethylation is associated with tumor progression in cervical cancer.²⁸ However, the role of ROBO3 in hematological malignancies is unknown. Here, combining data from TCGA database and GEO database, we illustrate the expression features, clinical significance, and potential biological functions of ROBO3 in AML. We found that ROBO3 is increased in non-M3 AML patients. Its high expression is associated with poor clinical parameters as well as inferior outcomes. Upregulation of ROBO3 and dysregulated SLIT/ROBO signaling pathway activity may drive the leukemogenesis of non-M3 AML.

Materials and methods

GEPIA and UALCAN databases

The expression difference of ROBO3 between AML patients and normal controls was first examined using the GEPIA database.²⁹ The cutoff values of |log2FC| and P were set as 1 and 0.01, respectively.

The expression feature of ROBO3 based on the French–American–British (FAB) classification was visualized using the UALCAN database.³⁰ Both the GEPIA and UALCAN databases provide convenient access to publicly available cancer OMICS data for further bioinformatics analysis.

Acquisition of detailed transcriptome and clinical data from the GEO and TCGA databases

The validation of ROBO3 expression between AML patients and normal controls was performed using transcriptome data acquired from the GEO database. The index words were set as acute myeloid leukemia, Homo sapiens, and expression. Finally, dataset GSE13159 was selected, as it enrolled both AML patients (n = 542) and normal controls (n = 74). GSE37642 was selected to validate the prognostic significance of ROBO3 expression in non-M3 AML patients.

Transcripts per million (TPM) and clinical data of 173 newly diagnosed AML patients who underwent RNA-seq in the TCGA AML cohort were acquired from the cBioPortal database.³¹ Four patients without specific FAB classification or ROBO3 expression data were excluded from further analysis. The median TPM of ROBO3 was set as the cutoff and the patients were divided into high and low ROBO3 expression groups. The mutation conditions of DNMT3A, FLT3, NPM1, TET2, TP53, EZH2, IDH1, IDH2, RUNX1, and U2AF1 in 173 AML patients were also downloaded from the cBioPortal database.

Identification of differentially expressed genes, and Kyoto encyclopedia of genes and genomes and gene ontology pathway enrichment analyses

TCGA transcriptome data were used to identify the DEGs between high and low ROBO3 expression patients. In order to reveal the underlying mechanism more effectively, we used a stricter grouping method. We defined high ROBO3 expression patients as ROBO3 expression level in the top 25% and low ROBO3 expression patients as ROBO3 expression level in the bottom 25%. The DEseq2' R language package was used to analyze the DEGs. The threshold was set as adjusted P < 0.05, and |log2 FC|>1.

KEGG and GO analyses were performed using DAVID Bioinformatics Resources 6.8 based on the DEGs between patients with high and low ROBO3 expression.³²

Gene function network prediction of ROBO3

Gene function networks including co-expression, genetic interaction, co-localization, shared protein domains, and pathway of ROBO3 were predicted by GeneMANIA (https://genemania.org/) and STRING database (https://string-db.org/).^33,34 GeneMANIA works best under the condition that most of the input genes are functionally related. Under the condition that only a single gene is inputted, GeneMANIA can make gene function predictions based on GO annotation patterns.

Statistical analyses

This paper used the Cox proportional hazard model for multivariate analysis to search independent risk factors. The difference in ROBO3 expression between different groups (AML patients vs. normal controls; non-M3 patients vs. M3 patients) was analyzed by Mann–Whitney U-test. The Kruskal–Wallis test was carried out to analyze the expression difference of ROBO3 between different age groups. Chi-squared test was conducted to reveal the relationship between ROBO3 expression and patients’ clinical chrematistics. The differences in prognosis between high and low ROBO3 expression patients were analyzed using the Kaplan–Meier method and log-rank test. P-values <0.05 were considered statistically significant. All statistical analysis was performed with SPSS 24.0 software.

Results

ROBO3 is increased in non-M3 AML patients

Since the data source of GEPIA was based on the TCGA AML cohort, we first investigated the expression feature of ROBO3 in AML patients using the GEPIA database. The expression of ROBO3 was significantly higher in AML patients compared with normal controls (P < 0.01, Figure 1(a)). To validate this finding, we analyzed the microarray data of GSE13159 from the GEO database. Coincident with its high expression in the TCGA AML cohort, the expression of ROBO3 was also significantly upregulated in AML patients in the validation cohort GSE13159 (P < 0.001, Figure 1(b)).

Figure 1.

ROBO3 is increasedly expressed in AML patients. (a) ROBO3 expression feature in AML patients and normal controls in TCAG AML cohort (Analyzed by GEPAI database). (b) ROBO3 expression feature in AML patients and normal controls in GSE13159 AML cohort. (c) Expression prolife of ROBO3 in TCAG AML cohort based on FAB classification (Analyzed by UALCAN database). (d, e) Expression difference of ROBO3 between M3 and non-M3 AML patients in TCAG AML cohort (d) and GSE13159 (e). (f, g) Expression difference of ROBO3 in non-M3 AML patients and normal controls in TCAG AML cohort (f) and GSE13159 (g). (A color version of this figure is available in the online journal.)

Next, we investigated the detailed expression profile of ROBO3 in AML patients based on the FAB classification. Results from the UALCAN database indicated that M3 patients had an obviously lower ROBO3 expression level than patients in other classifications (Figure 1(c)). Further statistical analysis showed that ROBO3 expression was significantly lower in M3 patients in both the TCGA AML cohort and GSE13159 (both P < 0.001, Figure 1(d) and (e)). Since M3 is a very special subtype of AML due to its unique genetic alteration (PML-RARα fusion gene) and remarkably better prognosis than other AML subtypes, we further tested the expression feature of ROBO3 after excluding M3 patients. As shown in Figure 1(f) and (g), non-M3 AML patients still exhibited a high ROBO3 expression profile in both TCGA AML and GSE13159 cohort (P < 0.001).

High ROBO3 expression is associated with increasing age at diagnosis and poorer risk classification

To illustrate the clinical relevance of high ROBO3 expression, we studied the association between ROBO3 expression and clinical parameters in non-M3 AML patients. ROBO3 high expression patients were significantly older at diagnosis (median age 64 years) compared with patients with low expression of ROBO3 (median age 54 years) (P < 0.001, Table 1, Figure 2(a)). Next, we regraded the patients into four groups with respect to age at diagnosis (<30, 30–60, 60–80, and ≥80 years) to explore the detailed relationship between ROBO3 expression and patient age at diagnosis. The expression of ROBO3 increased significantly with increasing age at diagnosis (P = 0.002, Figure 2(b)). The relationships between ROBO3 expression, age at diagnosis, and FAB classification in AML patients are shown in Figure 2(c).

Table 1.

The association of ROBO3 expression with clinical features in non-M3 AML patients.

	ROBO3 ^High(TPM≥median, n = 77)	ROBO3 ^Low(TPM<median, n = 76)	P
Sex, n (%)			0.690
Female	34 (44.2)	36 (47.4)
Male	43 (55.8)	40 (52.6)
Age, years			0.000
Median(range)	64 (21–88)	54 (18–81)
BM blasts, (%)			0.455
Median(range)	72 (30–100)	71 (30–99)
PB blasts, (%)			0.532
Median(range)	44.0 (0–98)	41.0 (0–97)
WBC (×10⁹/L)			0.002
Median(range)	11.0 (0.6–117.9)	30.1 (1.2–297.4)
Cytogenetic risk classification, n (%)^a			0.007
Favorable/ intermediate	50 (66.7)	64 (85.3)
Poor	25 (33.3)	11 (14.7)
ELN risk stratification, n (%)^a			0.001
Favorable/ intermediate	36 (48.0)	55 (73.3)
Adverse	39(52.0)	20 (26.7)

^aEvaluable patients’ number is 150 for three patients who do not have cytogenetic information.

BM: bone marrow; PB: peripheral blood; WBC: white blood cell; ELN: European Leukemia Net.

Figure 2.

High ROBO3 expression is associated with increasing diagnosis age in non-M3 AML patients. (a) Difference of diagnosis age between ROBO3 high expression patients and ROBO3 low expression patients. (b) Expression prolife of ROBO3 based on patients’ diagnosis age. (c) Sankey diagram of the relationship between ROBO3 expression, patients’ diagnosis age, and FAB classification in AML patients. (A color version of this figure is available in the online journal.)

In contrast, we observed that poor cytogenetic risk classification accounted for a much higher percentage in ROBO3 high expression patients compared with ROBO3 low expression patients (33.3% vs. 14.7%, P = 0.007; Table 1). Likewise, more than half of ROBO3 high expression patients (52.0%) had adverse European Leukemia Net (ELN) risk stratification. In contrast, this ratio was only 26.7% in ROBO3 low expression patients (P = 0.001). Interestingly, ROBO3 high expression patients had a significantly lower white blood cell (WBC) level than ROBO3 low expression patients (median: 11.0 vs. 30.1, P = 0.002). The differences in sex, bone marrow blasts, and peripheral blood blasts between ROBO3 high expression patients and ROBO3 low expression patients were not significant (all P > 0.05). Taken together, these data indicated that high ROBO3 expression was linked with high-risk clinical features, suggesting its potential clinical relevance in non-M3 AML patients.

High ROBO3 expression is associated with inferior gene mutations in non-M3 AML patients

With the recent development of sequencing technology and precision medicine, the importance of genetic alterations in human cancers has received extensive attention. In AML, besides the classical gene mutations like TP53 and FLT3-ITD, a growing number of novel mutations have been identified and shown to play important roles in leukemogenesis.^5,35 We acquired DNA sequencing data and explored the relationship between ROBO3 expression and common recurrent gene mutations in non-M3 AML patients. TP53 and RUNX1 mutations were more frequently detected in patients with high ROBO3 expression (TP53: 13.0% vs. 3.95%, P = 0.045; RUNX1: 24.7% vs. 6.58%, P = 0.002; NPM1: 19.5% vs. 42.1%, P = 0.002) (Table 2). In contrast, ROBO3 low expression patients had a significantly higher NPM1 mutation rate compared with ROBO3 high expression patients. No statistical differences were observed in DNMT3A, FLT3-ITD, TET2, EZH2, IDH1, IDH2, and U2AF1 mutations between the two groups of patients (all P > 0.05). TP53 and RUNX1 mutations have been widely recognized as inferior genetic alterations in AML patients. Moreover, patients harboring NPM1 mutations have a better prognosis than patients lacking NPM1 mutations in AML. These results further suggest that high ROBO3 expression may have significant clinical relevance in non-M3 AML patients.

Table 2.

The association of ROBO3 expression with common gene mutations in non-M3 AML patients.

	ROBO3 ^High(TPM≥median, n = 77)	ROBO3 ^Low(TPM<median, n = 76)	P
DNMT3A, n (%)			0.551
Wild type	58 (75.3)	54 (71.1)
Mutated	19 (24.7)	22 (28.9)
FLT3, n (%)			0.873
Wild type	61 (79.2)	61 (80.3)
FLT3-ITD	16 (20.8)	15 (19.7)
NPM1, n (%)
Wild type	62 (80.5)	44 (57.9)	0.002
Mutated	15 (19.5)	32 (42.1)
TET2, n (%)
Wild type	67 (87.0)	71 (93.40)	0.183
Mutated	10 (13.0)	5 (6.58)
TP53, n (%)			0.045
Wild type	67 (87.0)	73 (96.05)
Mutated	10 (13.0)	3 (3.95)
EZH2, n (%)			0.082
Wild type	74 (96.10)	76 (100)
Mutated	3 (3.90)	0 (0)
IDH1, n (%)			0.578
Wild type	70 (90.92)	68 (88.2)
Mutated	7 (9.08)	8 (11.8)
IDH2, n (%)
Wild type	66 (85.7)	71 (93.42)	0.119
Mutated	11 (14.3)	5 (6.58)
RUNX1, n (%)
Wild type	58 (75.3)	71 (93.42)	0.002
Mutated	19 (24.7)	5 (6.58)
U2AF1, n (%)			0.424
Wild type	73 (94.81)	74 (97.37)
Mutated	4 (5.19)	2 (2.63)

High ROBO3 expression predicts poor survival in non-M3 AML patients

The above results provided evidence of the association of ROBO3 with some important clinical and laboratory parameters in non-M3 AML. Considering that older age, poor cytogenetic/ELN risk classification, and TP53 and RUNX1 mutations are all high-risk factors for an inferior outcome, we wondered whether high ROBO3 expression was associated with poor survival in non-M3 AML patients. To unravel this, we performed survival analysis to directly investigate the prognostic significances of ROBO3 expression in non-M3 AML patients. Patients with PML-RARα fusion gene (M3) were excluded because their remarkably favorable prognosis may have influenced the analysis. Results showed that there was a distinct difference in survival between high and low ROBO3 expression non-M3 patients. A prominently worse overall survival (OS) and event-free survival (EFS) was observed in ROBO3 high expression non-M3 patients (Figure 3(a) and (b)) (median OS: 11.2 vs. 24.6 months, P = 0.004; median EFS: 7.7 vs. 10.2 months, P = 0.012). To further validate the relationship between high ROBO3 expression and inferior prognosis, we analyzed two additional data sets: GSE37642-GPL96 (n = 383, non-M3 AML patients) and GSE37642-GPL570 (n = 126, non-M3 AML patients). We observed that high ROBO3 expression was still related with poor overall survival in both GSE37642-GPL96 (Figure 4(a)) (median OS: 273 vs. 521 days, P = 0.012) and GSE37642-GPL570 (Figure 4(b)) (median OS: 352 vs. 548 days, P = 0.028).

Figure 3.

High ROBO3 expression is associated with inferior outcome in non-M3 AML patients. (a, b) Survival analysis results of non-M3 AML patients. Differences in OS (a) and EFS (b) between ROBO3 high expression patients and ROBO3 low expression patients. (c, d) Survival analysis results of non-M3 AML patients who received intensive chemotherapy during remission induction. Differences in OS (c) and EFS (d) between ROBO3 high expression patients and ROBO3 low expression patients. (e, f) Kaplan–Meier analysis results of non-M3 AML patients are younger than 60 and received intensive therapy during remission induction. Differences in OS (e) and EFS (f) between ROBO3 high expression patients and ROBO3 low expression patients. (A color version of this figure is available in the online journal.)

Figure 4.

The association of ROBO3 expression and OS in non-M3 AML patients in validation cohort. OS curve of non-M3 patients between ROBO3 high patients and ROBO3 low in GSE37642-GPL96 (a) and GSE37642-GPL570 (b). (A color version of this figure is available in the online journal.)

Treatment choice in remission induction also significantly affected patient outcomes. For non-M3 patients, anthracyclines and cytarabine-based 7 + 3 regimen is the standard treatment option for remission induction. However, not all patients can tolerate the side effects of intensive chemotherapy. Low-intensity therapy (low-dose cytarabine or hypomethylation agents) is recommended for patients who cannot withstand intensive remission induction therapy, such as patients with significant comorbid conditions. To better confirm the prognostic significance of ROBO3 expression in non-M3 patients, we further excluded patients who did not receive intensive therapy during remission induction. ROBO3 high expression was linked with poorer OS and EFS in non-M3 patients who received intensive chemotherapy (Figure 3(c) and (d)) (median OS: 17.1 vs. 27.0 months, P = 0.024; median EFS: 8.3 vs. 13.4 months, P = 0.040). Since intensive remission induction therapy is highly recommended for non-M3 AML patients who are younger than 60, survival analysis showed that ROBO3 high expression was still related with poorer OS and EFS in this subgroup of patients although the difference in EFS was not statistically significant (Figure 3(e) and (f)) (median OS: 19.0 vs. 56.3 months, P = 0.023; median EFS: 8.5 vs. 16.4 months, P = 0.075).

In multivariate analysis, the relationship between ROBO3 high expression and poorer OS and EFS was no longer detected after adjusted for age, DNMT3A, FLT3, NPM1, TET2, TP53, and RUNX1 mutation (P > 0.05) (Figures 5(a) and (b), 6(a) to (c)). However, subgroup analyses revealed that ROBO3 high expression was an independent risk factor of worse OS in non-M3 AML patients who are younger than 60 and received intensive therapy during remission induction (Figure 5(c)) (HR = 2.187, 95% CI: 1.110–4.307, P = 0.024). In conclusion, above results revealed that ROBO3 high expression was associated with high-risk clinical and laboratory parameters, and also predicted poor survival in non-M3 AML patients.

Figure 5.

Forest plots of the association between ROBO3 high expression and OS in non-M3 AML patients. Cox proportional hazards model for OS in non-M3 AML patients (a), non-M3 AML patients who received intensive therapy during remission induction (b), and non-M3 AML patients are younger than 60 and received intensive therapy during remission induction (c). (A color version of this figure is available in the online journal.)

Figure 6.

Forest plots of the association between ROBO3 high expression and EFS in non-M3 AML patients in validation cohort. Cox proportional hazards model for EFS in non-M3 AML patients (a), non-M3 AML patients who received intensive therapy during remission induction (b), and non-M3 AML patients who are younger than 60 and received intensive therapy during remission induction (c). (A color version of this figure is available in the online journal.)

Functional pathway enrichment analysis between ROBO3 high expression and ROBO3 low expression non-M3 AML patients

Previous studies have shown that ROBO family members play important roles in kinds of human cancers through various functional pathways.^18,19,27 However, the potential biological function of ROBO3 in non-M3 AML patients remains largely unknown. To investigate the underlying biological significance of high ROBO3 expression in non-M3 AML, we performed bioinformatic analysis to explore the significantly dysregulated genes and pathways between ROBO3 high expression patients and ROBO3 low expression patients. A total of 1209 DEGs were identified between ROBO3 high and low expression patients (Figure 7(a)). Among these 1209 DEGs, the number of upregulated genes and downregulated genes in ROBO3 high expression patients was 965 and 244, respectively. Detailed information of the top 20 upregulated and downregulated genes is shown in Table 3.

Figure 7.

Pathways enrichment analysis between ROBO3 high expression and ROBO3 low expression non-M3 AML patients. (a) Heatmap of 1209 differential expression genes between two groups of non-M3 AML patients based on ROBO3 expression. The thresholds are P adjust<0.05 and |log2 FC|>1. (b) Top 10 most enriched KEGG pathways between two groups of non-M3 AML patients based on ROBO3 expression. (c) Top 10 most enriched GO pathways between two groups of non-M3 AML patients based on ROBO3 expression. (d) Circle plot of 16 differential expression genes which participated in greater than or equal to three GO pathways and their relationship with the significantly enriched GO pathways. (A color version of this figure is available in the online journal.)

Table 3.

Top 20 most up-regulated and down-regulated genes in ROBO3 high expression non-M3 AML patients.

	log2 FC	P adj		log2 FC	P adj
FAM155B	5.48	6.56E-16	C2CD4B	−3.26	4.47E-06
C6orf141	4.44	1.26E-07	FOXF2	−3.04	2.36E-03
DLK1	4.32	1.58E-09	RUNX1T1	−3.00	1.16E-05
NBPF4	4.32	5.67E-09	CD163	−2.85	3.59E-11
SPON1	4.12	1.79E-30	LMX1B	−2.50	2.80E-05
UCHL1	4.06	6.88E-15	POU1F1	−2.71	2.23E-03
KCNIP1	4.00	3.46E-08	EMX2	−2.69	1.04E-04
MFAP2	3.91	1.06E-15	IL22RA2	−2.59	3.42E-04
ELN	3.89	5.25E-17	IDO1	−2.31	2.57E-03
NOVA1	3.89	1.20E-04	EPB41L3	−2.24	1.33E-06
GREM1	3.80	2.31E-06	FOXL1	−2.21	3.25E-02
BEND7	3.75	9.53E-12	LOC401463	−2.12	2.45E-03
COBL	3.66	7.31E-09	C1orf141	−2.07	2.05E-02
HMX3	3.65	3.90E-02	ARHGAP29	−2.02	1.71E-07
FIGN	3.64	3.83E-06	MSR1	−1.97	2.19E-05
CREG2	3.60	8.36E-08	PDK4	−1.96	9.08E-07
ANKFN1	3.53	2.80E-05	C7orf16	−1.96	7.77E-03
RIMBP2	3.43	2.14E-06	PRR16	−1.95	3.02E-03
MEG3	3.39	3.26E-11	ROBO2	−1.94	4.57E-02
CECR7	3.19	1.59E-14	BHLHE22	−1.88	2.69E-03

Next, KEGG and GO pathway enrichment analyses were performed based on the 1209 DEGs. The KEGG analysis revealed that the nervous system-related pathways (serotonergic synapse, neuroactive ligand-receptor interaction, and axon guidance) were significantly enriched between ROBO3 high expression patients and ROBO3 low expression patients (Figure 7(b)). This result was consistent with the classical biological function of ROBO family members and the SILT/ROBO signaling pathway in the nervous system. In addition, we also identified some pathways involved in human cancers (protein digestion and absorption, extracellular matrix (ECM) receptor interaction), although the degree of enrichment was not as significant as the pathways involved in the nervous system (Figure 7(b)).

GO analysis further provided more information about the potential functions of ROBO3 in non-M3 AML patients based on the three aspects of molecular function, cellular component, and biological process. Similar to the KEGG enrichment results, a nervous system-related pathway in biological processes (nervous system development) was significantly enriched in GO analysis (Figure 7(c)). Since ROBO family members are transmembrane receptor proteins, we found that three cell membrane-related terms (plasma membrane, integral component of plasma membrane, and integral component of membrane) were in the top 10 most enriched GO results (Figure 7(c)). In addition to the canonical functions of ROBO family members, cell adhesion and ECM organization pathways, which have long been recognized as critically important factors in tumor progression and invasion were also significantly enriched between ROBO3 high expression patients and ROBO3 low expression patients (Figure 7(c)). Intriguingly, we observed that both KEGG and GO results demonstrated that pathways associated with the ECM exhibited a high degree of enrichment, suggesting their possible roles in ROBO3 highly expressed AML. Finally, the DEGs that participated in multiple GO pathways (≥3) and their relationship with the significantly enriched GO pathways are shown in Figure 7(d).

Prediction of protein–protein interaction network of ROBO3

To explore the potential functional partners or regulators of ROBO3, we performed PPI network prediction using the STRING and GeneMANIA databases. Since ROBO3 is an important member of the SLIT/ROBO signaling pathway, we identified that ROBO3 interacts with other proteins in the SLIT/ROBO signaling pathway (SLIT1, SLIT2, SLIT3, and ROBO1) in both STRING prediction (Figure 8(a)) and GeneMANIA prediction (Figure 8(b)). Besides SLIT/ROBO signaling pathway members, we observed that some other proteins that play vital roles in the nervous system (NELL2, CHN1, NTN1, and HOXA1) interacted with ROBO3 (Figure 8(a)). More importantly, we found that PAK3, APBA2, and SALL4 were also in the PPI prediction network. Considering previous work has shown that PAK3, APBA2, and SALL4 take part in tumor formation and development in different ways, their interaction with ROBO3 indicates their potential cooperative relationships in human cancers. Taken together, these results indicate the underlying PPI network of ROBO3, suggesting its complex roles in the nervous system and human cancers.

Figure 8.

PPI network predication by STRING database (a) and GeneMANIA database (b). (A color version of this figure is available in the online journal.)

Discussion

In this study, we performed a series of bioinformatic analyses using data from public databases. The analyses demonstrated that ROBO3, an important member of the SILT/ROBO signaling pathway, was highly expressed in non-M3 AML patients. Moreover, high expression of ROBO3 was related with increasing age at diagnosis, poor risk classification, and inferior outcome. Our data provide evidence that high ROBO3 expression may be a potential prognostic biomarker for non-M3 AML patients.

The SILT/ROBO signaling pathway was originally identified in the nervous system, particularly in axon guidance and axon repulsion.^12,15 More recently, the SILT/ROBO signaling pathway has been shown to play vital roles in other biological processes, such as angiogenesis, heart development, and inflammatory cell chemotaxis.^36–38 Likewise, the function of the SILT/ROBO signaling pathway in cancer progression has also aroused widespread concern. Dysregulated ROBO family members have been reported to have diverse clinical significance in different human cancers. For example, ROBO1 overexpression was significantly associated with poorer OS in pancreatic ductal adenocarcinoma and nasopharyngeal cancer.^22,27 In pancreatic carcinoma, ROBO3 expression was positively correlated with tumor stage, and high ROBO3 expression was associated with inferior outcome.¹⁸ On the other hand, ROBO2 was downregulated and was implicated as a tumor suppressor in prostate cancer.³⁹ These results indicated that the SILT/ROBO signaling pathway produced oncogenic or anti-cancer activity that was largely dependent on the specific cancer types.

The underlying roles of ROBO family members in hematological malignancies have not been intensively investigated. Mutations in ROBO1 and ROBO2 have been detected in myelodysplastic syndromes (MDS).²⁵ Moreover, ROBO1 and ROBO2 mutations were independent factors of inferior outcome in MDS patients.²⁵ In vitro experiments demonstrated that upregulation of ROBO1 or ROBO2 suppressed proliferation and promoted apoptosis in K562 chronic myelogenous leukemia cells and HEL acute erythrocytic leukemia cells.²⁵ Considering the oncogenic role of ROBO1 in pancreatic carcinoma, its antineoplastic activity in leukemia cell lines further supports the functional complexity of the SILT/ROBO signaling pathway in human cancers.

Our data revealed that ROBO3 was increased in non-M3 AML patients. Patients with PML-RARα fusion gene exhibited a relatively lower ROBO3 expression level compared with other AML subtypes. High expression of ROBO3 was associated with increasing age at diagnosis and poorer risk classification in non-M3 AML patients. These results were similar to the finding that higher ROBO3 expression correlated with progressive tumor stages in pancreatic carcinoma. Interestingly, we observed that ROBO3 high expression patients had lower WBC counts than ROBO3 low expression patients. This may be partly explained by the significantly older age and accompanying hematopoietic failure of ROBO3 high expression patients. We further analyzed the relationship between ROBO3 expression and common recurrent gene mutations in non-M3 AML patients. TP53 and RUNX1 mutations were more frequently detected in ROBO3 high expression patients. TP53 is broadly recognized as the most important tumor suppressor gene in human cancers. Mutations of TP53 have been linked with extremely poor prognosis in all studied cancer types.⁴⁰ On the other hand, the RUNX1 mutation is a newly established marker of inferior outcome in AML patients, although its specific clinical significance is still contentious.³⁵ All these results strongly indicated that high ROBO3 expression may predict poor survival in non-M3 AML patients. Further survival analysis directly corroborated these results. In contrast to ROBO3 low expression patients, the OS and EFS were obviously shorter in ROBO3 high expression patients. This relationship between ROBO3 expression status and survival still existed in a subgroup of patients who received intensive remission induction therapy. For patients who are younger than 60 and received intensive therapy during remission induction, the differences in EFS were no longer statistically significant. Since patients who can tolerate the side effects of intensive chemotherapy commonly are in better physical condition, it is understandable that these patients exhibit greater consistency in clinical characteristics and have obviously better outcomes compared with other patients. We also observed that high ROBO3 was not an independent risk factor of poorer OS and EFS in non-M3 AML patients if age and remission induction treatment choice were not taken into consideration. Since older age is strongly associated with inferior outcome in AML patients, the significant relationship between ROBO3 expression and patients’ age may interference the effect of high ROBO3 expression as an independent survival factor. This can also explain why high ROBO3 expression was an independent risk factor of poorer survival in patients who are younger than 60 and received intensive therapy during remission induction.

Previous studies have exposed that aberrant SILT/ROBO signaling pathway takes part in tumor formation, angiogenesis, and metastasis via a variety of mechanisms. For instance, overexpression of SLIT2 and ROBO1 can promote epithelial-mesenchymal transition, tumor cell proliferation, and liver metastasis during colorectal cancer development by mediating E-cadherin ubiquitination and lysosomal degradation.²³ In pancreatic cancer, overexpression of ROBO3 activates the Wnt-β-catenin pathway by enhancing the expression of β-catenin, phosphorylated GSK3β, and Wnt signaling target C-myc.¹⁸ Presently, the analysis of the DEGs and the potentially altered pathways between ROBO3 high expression patients and ROBO3 low expression patients revealed significant enrichment of a series of cancer-related pathways, including cell adhesion and ECM organization pathways. Particularly, the results showed that ECM-related pathways were significantly enriched in both KEGG and GO analyses. Although cell adhesion and ECM are generally considered to contribute to the tumor invasion and metastasis of solid tumors, increasing evidence indicates that they also play important roles in hematological malignancies.^41–43 Leukemic cells adherent to the ECM are responsible for the chemotherapy resistance and inferior outcome.⁴³ Our results provide bioinformatic evidence that cell adhesion and ECM-related pathways may act downstream of aberrant high ROBO3 expression in non-M3 AML. On the other hand, the PPI prediction indicated a complex interaction network and suggested that ROBO3 may have a close relationship with some oncogenes, such as PAK3, APBA2, and SALL4. Further studies illustrating the detailed underlying mechanisms of how ROBO3 regulates or cooperates with these oncogenic pathways and genes will lead to a better understanding of its roles in non-M3 AML.

In conclusion, our data revealed the high expression of ROBO3 in non-M3 AML patients. Upregulated ROBO3 expression was associated with increased age at diagnosis, poor risk classification, high TP53 and RUNX1 mutation rates, and shorter survival. These results indicated that ROBO3 may be a potent biomarker of inferior outcome in non-M3 AML. Dysregulated cell adhesion and ECM-related pathways may take part in high ROBO3-mediated leukemogenesis. Exploring the detailed biological functions of ROBO3 or the SILT/ROBO signaling pathway in AML may be an illuminating direction for fundamental and clinical research in the future.

Footnotes

AUTHORS’ CONTRIBUTIONS

JFW and ZC collected and analyzed the data, respectively. ZMC was responsible for drafting and writing this manuscript. HQW edited this manuscript.

DECLARATION OF CONFLICTING INTERESTS

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

FUNDING

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The paper was supported by The Health and Family Planning Technology Project in Lianyungang under Award Number 201711.

ORCID iD

Zhimei Cai

References

Dohner

Weisdorf

Bloomfield

CD.

Acute myeloid leukemia. N Engl J Med 2015; 373:1136–52

Bullinger

Dohner

Genomics of acute myeloid leukemia diagnosis and pathways. J Clin Oncol 2017; 35:934–46

Tamamyan

Kadia

Ravandi

Borthakur

Cortes

Jabbour

Daver

Ohanian

Kantarjian

Konopleva

Frontline treatment of acute myeloid leukemia in adults. Crit Rev Oncol Hematol 2017; 110:20–34

Dombret

Gardin

An update of current treatments for adult acute myeloid leukemia. Blood 2016; 127:53–61

Welch

Petti

Miller

Fronick

O'Laughlin

Fulton

Wilson

Baty

Duncavage

Tandon

Lee

Wartman

Ghobadi

Tomasson

Pusic

Romee

Fehniger

Stockerl-Goldstein

Vij

Abboud

Cashen

Schroeder

Jacoby

Heath

Luber

Janke

Hantel

Khan

Sukhanova

Knoebel

Stock

Graubert

Walter

Westervelt

Link

DiPersio

Ley

TJ.

TP53 and decitabine in acute myeloid leukemia and myelodysplastic syndromes. N Engl J Med 2016; 375:2023–36

Craddock

Slade

De Santo

Wheat

Ferguson

Hodgkinson

Brock

Cavenagh

Ingram

Dennis

Malladi

Siddique

Mussai

Yap

Combination lenalidomide and azacitidine: a novel salvage therapy in patients who relapse after allogeneic stem-cell transplantation for acute myeloid leukemia. J Clin Oncol 2019; 37:580–8

Prada-Arismendy

Arroyave

Rothlisberger

Molecular biomarkers in acute myeloid leukemia. Blood Rev 2017; 31:63–76

Stone

Mandrekar

Sanford

Laumann

Geyer

Bloomfield

Thiede

Prior

Dohner

Marcucci

Lo-Coco

Klisovic

Wei

Sierra

Sanz

Brandwein

de Witte

Niederwieser

Appelbaum

Medeiros

Tallman

Krauter

Schlenk

Ganser

Serve

Ehninger

Amadori

Larson

Dohner

Midostaurin plus chemotherapy for acute myeloid leukemia with a FLT3 mutation. N Engl J Med 2017; 377:454–64

DiNardo

Stein

de Botton

Roboz

Altman

Mims

Swords

Collins

Mannis

Pollyea

Donnellan

Fathi

Pigneux

Erba

Prince

Stein

Foran

Traer

Stuart

Arellano

Slack

Sekeres

Willekens

Choe

Wang

Zhang

Yen

Kapsalis

Yang

Dai

Fan

Goldwasser

Liu

Agresta

Attar

Tallman

Stone

Kantarjian

HM.

Durable remissions with ivosidenib in IDH1-mutated relapsed or refractory AML. N Engl J Med 2018; 378:2386–98

10.

Eide

Kurtz

Kaempf

Long

Agarwal

Tognon

Mori

Druker

Chang

Danilov

Tyner

JW.

Simultaneous kinase inhibition with ibrutinib and BCL2 inhibition with venetoclax offers a therapeutic strategy for acute myeloid leukemia. Leukemia 2020; 34:2342–53 Sep

11.

Stein

Fathi

DiNardo

Pollyea

Roboz

Collins

Sekeres

Stone

Attar

Frattini

Tosolini

See

MacBeth

de Botton

Tallman

Kantarjian

HM.

Enasidenib in patients with mutant IDH2 myelodysplastic syndromes: a phase 1 subgroup analysis of the multicentre, AG221-C-001 trial. Lancet Haematol 2020; 7:e309–19

12.

Blockus

Chedotal

Slit-robo signaling. Development 2016; 143:3037–44

13.

Cornide-Petronio

Barreiro-Iglesias

Role of slit and robo proteins in the development of dopaminergic neurons. Dev Neurosci 2013; 35:285–92

14.

Morlot

Thielens

Ravelli

Hemrika

Romijn

Gros

Cusack

McCarthy

AA.

Structural insights into the slit-robo complex. Proc Natl Acad Sci U S A 2007; 104:14923–8

15.

Brose

Bland

Wang

Arnott

Henzel

Goodman

Tessier-Lavigne

Kidd

Slit proteins bind robo receptors and have an evolutionarily conserved role in repulsive axon guidance. Cell 1999; 96:795–806

16.

Wang

Hou

Guo

van der Voet

Boxem

Dixon

Chisholm

Jin

The EBAX-type Cullin-RING E3 ligase and Hsp90 guard the protein quality of the SAX-3/robo receptor in developing neurons. Neuron 2013; 79:903–16

17.

Yamada

Okita

Shofuda

Yoshioka

Nonaka

Mori

Nakajima

Kanemura

Ipsilateral hemiparesis caused by putaminal hemorrhage in a patient with horizontal gaze palsy with progressive scoliosis: a case report. BMC Neurol 2015; 15:25

18.

Chaturvedi

Robinson

LA.

Slit2-robo signaling in inflammation and kidney injury. Pediatr Nephrol 2015; 30:561–6

19.

Huang

Wen

Kong

Cheng

Zhang

Quan

Bian

Chen

Zhang

Chen

Liu

Zhu

Fushimi

Hua

JY.

USP33 mediates slit-robo signaling in inhibiting colorectal cancer cell migration. Int J Cancer 2015; 136:1792–802

20.

Sanz-Pamplona

Berenguer

Cordero

Mollevi

Crous-Bou

Sole

Pare-Brunet

Guino

Salazar

Santos

de Oca

Sanjuan

Rodriguez-Moranta

Moreno

Aberrant gene expression in mucosa adjacent to tumor reveals a molecular crosstalk in colon cancer. Mol Cancer 2014; 13:46–64

21.

Dickinson

Fegan

Ren

Hillier

Duncan

WC.

Glucocorticoid regulation of SLIT/ROBO tumour suppressor genes in the ovarian surface epithelium and ovarian cancer cells. PLoS One 2011; 6:e27792

22.

Alajez

Lenarduzzi

Ito

Hui

Shi

Bruce

Yue

Huang

Waldron

O'Sullivan

Liu

FF.

MiR-218 suppresses nasopharyngeal cancer progression through downregulation of survivin and the SLIT2-ROBO1 pathway. Cancer Res 2011; 71:2381–91

23.

Zhou

Geng

Chi

Zhang

Niu

Lan

Yang

Wang

Ding

Geng

JG.

Slit-robo signaling induces malignant transformation through hakai-mediated E-cadherin degradation during colorectal epithelial cell carcinogenesis. Cell Res 2011; 21:609–26

24.

Gohrig

Detjen

Hilfenhaus

Korner

Welzel

Arsenic

Schmuck

Bahra

Wiedenmann

Fischer

Axon guidance factor SLIT2 inhibits neural invasion and metastasis in pancreatic cancer. Cancer Res 2014; 74:1529–40

25.

Han

Cao

Tang

Wang

Xue

Zhang

ROBO3 promotes growth and metastasis of pancreatic carcinoma. Cancer Lett 2015; 366:61–70

26.

Avci

Konu

Yagci

Quantification of SLIT-ROBO transcripts in hepatocellular carcinoma reveals two groups of genes with coordinate expression. BMC Cancer 2008; 8:392–402

27.

Pinho

Van Bulck

Chantrill

Arshi

Sklyarova

Herrmann

Vennin

Gallego-Ortega

Mawson

Giry-Laterriere

Magenau

Leuckx

Baeyens

Gill

Phillips

Timpson

Biankin

Rooman

ROBO2 is a stroma suppressor gene in the pancreas and acts via TGF-beta signalling. Nat Commun 2018; 9:5083–96

28.

Narayan

Goparaju

Arias-Pulido

Kaufmann

Schneider

Durst

Mansukhani

Pothuri

Murty

VV.

Promoter hypermethylation-mediated inactivation of multiple slit-robo pathway genes in cervical cancer progression. Mol Cancer 2006; 5:16–27

29.

Tang

Kang

Gao

Zhang

GEPIA: a web server for cancer and normal gene expression profiling and interactive analyses. Nucleic Acids Res 2017; 45:W98–W102

30.

Chandrashekar

Bashel

Balasubramanya

SAH

Creighton

Ponce-Rodriguez

Chakravarthi

Varambally

UALCAN: a portal for facilitating tumor subgroup gene expression and survival analyses. Neoplasia 2017; 19:649–58

31.

Cerami

Gao

Dogrusoz

Gross

Sumer

Aksoy

Jacobsen

Byrne

Heuer

Larsson

Antipin

Reva

Goldberg

Sander

Schultz

The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov 2012; 2:401–4

32.

Huang da

Sherman

Lempicki

RA.

Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res 2009; 37:1–13

33.

Warde-Farley

Donaldson

Comes

Zuberi

Badrawi

Chao

Franz

Grouios

Kazi

Lopes

Maitland

Mostafavi

Montojo

Shao

Wright

Bader

Morris

The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res 2010; 38:W214–20

34.

Szklarczyk

Gable

Lyon

Junge

Wyder

Huerta-Cepas

Simonovic

Doncheva

Morris

Bork

Jensen

Mering

CV.

STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res 2019; 47:D607–D13

35.

Gaidzik

Teleanu

Papaemmanuil

Weber

Paschka

Hahn

Wallrabenstein

Kolbinger

Kohne

Horst

Brossart

Held

Kundgen

Ringhoffer

Gotze

Rummel

Gerstung

Campbell

Kraus

Kestler

Thol

Heuser

Schlegelberger

Ganser

Bullinger

Schlenk

Dohner

RUNX1 mutations in acute myeloid leukemia are associated with distinct clinico-pathologic and genetic features. Leukemia 2016; 30:2160–8

36.

Klagsbrun

Eichmann

A role for axon guidance receptors and ligands in blood vessel development and tumor angiogenesis. Cytokine Growth Factor Rev 2005; 16:535–48

37.

Zhao

Mommersteeg

MTM.

Slit-robo signalling in heart development. Cardiovasc Res 2018; 114:794–804

38.

Feng

Park

Havlioglu

Wen

Tang

Bacon

Jiang

Zhang

Rao

The neuronal repellent slit inhibits leukocyte chemotaxis induced by chemotactic factors. Nature 2001; 410:948–52

39.

Choi

Yoo

Lee

SH.

Down-regulation of ROBO2 expression in prostate cancers. Pathol Oncol Res 2014; 20:517–9

40.

Sabapathy

Lane

DP.

Therapeutic targeting of p53: all mutants are equal, but some mutants are more equal than others. Nat Rev Clin Oncol 2018; 15:13–30

41.

Verma

Zanetti

Godavarthy

Kumar

Minciacchi

Pfeiffer

Metzler

Lefort

Maguer-Satta

Nicolini

Burroni

Fontenay

Krause

DS.

Bone marrow niche-derived extracellular matrix-degrading enzymes influence the progression of B-cell acute lymphoblastic leukemia. Leukemia 2020; 34:1540–52

42.

Izzi

Heljasvaara

Pihlajaniemi

Understanding the extracellular matrix in acute myeloid leukemia. Haematologica 2017; 102:1807–9

43.

Berrazouane

Boisvert

Salti

Mourad

Al-Daccak

Barabe

Aoudjit

Beta1 integrin blockade overcomes doxorubicin resistance in human T-cell acute lymphoblastic leukemia. Cell Death Dis 2019; 10:357–69