Patients
In total, 343 participants comprising 299 HCs and 44 ESCCs were enrolled for this urinary biomarker study of ESCC. After case matching by age and sex and exclusion of 20 urine samples with poor miRNA quality, 193 patients (150 HCs and 43 ESCCs) were ultimately selected as the whole cohort. This entire cohort was randomly divided into two: 9 patients (6 HCs and 3 ESCCs) in the discovery cohort and 184 patients (144 HCs and 40 ESCCs) in the training/test cohort. In the discovery cohort, potential urinary miRNA biomarkers for ESCC diagnosis were identified by miRNA microarray analysis; then, in the training/test cohort, qRT-PCR analysis confirmed, and established the urinary miRNA biomarkers. The established biomarkers were then validated by internal cross-validation analysis for all stage ESCC and stage I ESCC in training and test sets, respectively. Additionally, serum (26 HCs and 21 ESCCs) and tissue (20 ESCC tissues) miRNAs were examined by in vitro and in silico analyses. Finally, using 152 urine samples (144 HCs and 8 EACs) and tissue samples from seven patients with EAC, we validated the urinary miRNA biomarkers for EAC diagnosis (Fig. 1).
Table 1 summarizes the overall patient characteristics. Age, sex, and serum creatinine were not significantly different between the HC and ESCC groups. Stage I and T1 stage were found in 37.2% and 48.8% of ESCCs.
Table 1
Characteristics of the whole cohort
|
HC
(n = 150)
|
ESCC
(n = 43)
|
P value
|
Median age [IQR] (years)
|
69 [63–74]
|
70 [65–75]
|
0.072†
|
Sex
|
|
|
1.000‡
|
Female
|
15 (10.0%)
|
4 (9.3%)
|
|
Male
|
135 (90.0%)
|
39 (90.7%)
|
|
Median serum creatinine [IQR] (mg/dl)
|
0.84 [0.76–0.94]
|
0.81 [0.70–0.99]
|
0.592†
|
Tumor location
|
|
|
|
Cervical
|
|
2 (4.7%)
|
|
Thoracic
|
|
39 (90.7%)
|
|
Abdominal
|
|
2 (4.7%)
|
|
TNM Stage
|
|
|
|
I
|
|
16 (37.2%)
|
|
II
|
|
4 (9.3%)
|
|
III
|
|
12 (27.9%)
|
|
IV
|
|
11 (25.6%)
|
|
Differentiation (squamous cell carcinoma)
|
|
|
|
Well to moderately
|
|
39 (90.7%)
|
|
Poorly
|
|
3 (7.0%)
|
|
Basaloid
|
|
1 (2.3%)
|
|
T stage
|
|
|
|
T1
|
|
21 (48.8%)
|
|
T2
|
|
1 (2.3%)
|
|
T3
|
|
10 (23.3%)
|
|
T4
|
|
11 (25.6%)
|
|
†, Mann-Whitney U test |
‡, Fisher’s exact probability test |
HC, healthy control; ESCC, esophageal squamous cell carcinoma; IQR, interquartile range |
Comprehensive analysis with miRNA microarray
To identify urinary miRNA biomarker candidates, we first interrogated a microarray-based miRNA expression profiling comparing the urinary miRNA expression levels of the HC group with those of the ESCC group, in the discovery cohort. Seventy miRNAs (22 up-regulated and 48 down-regulated miRNAs) were differentially expressed between the two groups (absolute fold change > 2.0, P < 0.05) (Additional file 2: Figure S1). Among these miRNAs, we selected eight miRNAs for the succeeding step.
Establishment and validation of urinary miRNA biomarkers using qRT-PCR
Next, using the qRT-PCR assay, we evaluated the urinary expression levels of eight miRNAs (miR-1273f, miR-619-5p, miR-150-3p, miR-4327, miR-3135b, miR-5585-3p, miR-6875-5p, and miR-345-3p) in the training/test cohort. The urinary levels of miR-1273f, miR-619-5p, miR-150-3p, miR-4327, and miR-3135b were significantly higher in the ESCC group than in the HC group (miR-1273f, P < 0.001; miR-619-5p, P < 0.001; miR-150-3p, P < 0.001; miR-4327, P = 0.007; miR-3135b, P = 0.021; Table 2). Subsequently, we conducted internal cross-validation analysis in the training and test sets according to the k-fold cross-validation method (k = 5). These five urinary miRNAs were consistently significantly different between HC and all stage ESCC (AUC: miR-1273f: 0.792, 0.788; miR-619-5p: 0.786, 0.786; miR-150-3p: 0.692, 0.696; miR-4327: 0.639, 0.625; miR-3135b: 0.620, 0.626 in the training and test sets, respectively) (Table 3). In total, these urinary miRNAs were also significantly different between the HC and ESCC groups in ROC analysis (Fig. 2A). Urinary miR-1273f and miR-619-5p were the top two qualified biomarkers for ESCC diagnosis, with 82.5% and 80.0% sensitivity and 59.7% and 65.3% specificity, respectively.
Table 2
|
Median 2−ΔCT (x10− 2)
[IQR]
|
P value
|
Median 2−ΔCT (x10− 2)
[IQR]
|
P value
|
|
HC
|
All stage ESCC
|
HC vs. All Stages
|
Stage I ESCC
|
HC vs. Stage I
|
miR-1273f
|
3.313
[1.694–7.899]
|
14.014
[5.482–45.629]
|
< 0.001
|
15.257
[6.725–45.494]
|
< 0.001
|
miR-619-5p
|
23.183
[12.112–42.318]
|
73.644
[34.289-188.674]
|
< 0.001
|
73.183
[44.617-165.866]
|
< 0.001
|
miR-150-3p
|
0.594
[0.307–1.731]
|
1.517
[0.654–3.116]
|
< 0.001
|
1.633
[0.692–4.037]
|
0.009
|
miR-4327
|
29.680
[14.314–84.168]
|
64.618
[30.404-121.001]
|
0.007
|
89.331
[62.138-172.789]
|
0.001
|
miR-3135b
|
99.363
[35.556-235.372]
|
263.782
[42.933-589.915]
|
0.021
|
146.608
[9.591-613.165]
|
0.821
|
miR-5585-3p
|
1.416
[0.563–3.674]
|
2.319
[0.785–5.535]
|
0.183
|
2.426
[0.377–10.734]
|
0.337
|
miR-6875-5p
|
6.538
[3.759–17.432]
|
7.785
[3.969–11.871]
|
0.629
|
9.517
[3.962–17.756]
|
0.807
|
miR-345-3p
|
0.081
[0.017–0.263]
|
0.105
[0.026–0.216]
|
0.562
|
0.096
[0.019–0.220]
|
0.883
|
HC, healthy control; ESCC, esophageal squamous cell carcinoma; IQR, interquartile range |
Table 3
Internal cross-validation of urinary miRNA biomarker
All stage ESCC
|
Training set
AUC [95% CI]
|
Test set
AUC [95% CI]
|
miR-1273f
|
0.792 [0.777–0.811]
|
0.788 [0.713–0.850]
|
miR-619-5p
|
0.786 [0.771–0.801]
|
0.786 [0.727–0.846]
|
miR-150-3p
|
0.692 [0.678–0.709]
|
0.696 [0.628–0.760]
|
miR-4327
|
0.639 [0.627–0.653]
|
0.625 [0.589–0.664]
|
miR-3135b
|
0.620 [0.602–0.635]
|
0.626 [0.556–0.703]
|
Stage I ESCC
|
Training set
AUC [95% CI]
|
Test set
AUC [95% CI]
|
miR-1273f
|
0.826 [0.803–0.848]
|
0.819 [0.727–0.910]
|
miR-619-5p
|
0.817 [0.794–0.852]
|
0.823 [0.696–0.921]
|
miR-150-3p
|
0.713 [0.680–0.753]
|
0.705 [0.563–0.822]
|
miR-4327
|
0.749 [0.729–0.766]
|
0.733 [0.661–0.796]
|
miR-3135b
|
0.520 [0.490–0.553]
|
0.558 [0.399–0.714]
|
ESCC, esophageal squamous cell carcinoma; AUC, area under the curve; 95% CI, 95% confidence interval |
Importantly, the urinary levels of miR-1273f, miR-619-5p, miR-150-3p, and miR-4327 were significantly higher in the stage I ESCC group than in the HC group (miR-1273f, P < 0.001; miR-619-5p, P < 0.001; miR-150-3p, P = 0.009; miR-4327, P = 0.001; Table 2). These urinary miRNAs consistently exhibited excellent diagnostic performance for stage I ESCC in both training and test sets (AUC: miR-1273f: 0.826, 0.819; miR-619-5p: 0.817, 0.823; miR-150-3p: 0.713, 0.705; miR-4327: 0.749, 0.733 in the training and test sets, respectively) (Table 3). In total, these four urinary miRNAs were significantly different between HC and stage I ESCC (Fig. 2B). Specifically, urinary miR-1273f and miR-619-5p were 92.9% and 92.8% sensitive and 59.0% and 65.3% specific, respectively, for detecting stage I ESCC.
The urinary levels of miR-1273f, miR-619-5p, miR-150-3p, and miR-3135b had no significant correlation with disease stage, although miR-4327 negatively correlated with disease stage (r = − 0.435) (Additional file 2: Figure S2).
Additional validation analysis using serum, tissue samples, and cell lines
To further validate the established urinary miRNA biomarkers, we conducted additional analyses using serum and tissue samples and in vitro analysis. The serum levels of the five miRNAs (miR-1273f, miR-619-5p, miR-150-3p, miR-4327, and miR-3135b) were compared between 26 HCs and 21 ESCCs. Patient characteristics were not significantly different between the two groups. Despite significant differences in urine samples, the serum levels of all five miRNAs had no significant differences between the ESCC and HC groups (data not shown). Regarding tissue miRNA analysis, 20 ESCC samples were available for miRNA extraction, of which 11 patients underwent surgery and nine patients underwent ESD, and 13 patients (65.0%) were still in stage I. The expression levels of miR-1273f, miR-619-5p, miR-150-3p, and miR-4327 tended to be higher in the ESCC tissues than those in the adjacent normal tissues, but not significant because of small sample size (Additional file 2: Figure S3).
In addition, using the CM from normal esophageal and ESCC cell lines, we investigated whether these miRNAs were released from ESCC. As shown in Fig. 3, these miRNAs were significantly more abundant in the CM from the ESCC cell lines than those in the CM from the normal esophageal cell line.
Esophageal adenocarcinoma analysis
Moreover, we investigated whether the established urinary miRNA biomarkers for ESCC can be applied to EAC. In the EAC biomarker study, 152 participants consisting of 144 HCs and 8 EACs were analyzed. Patient characteristics were similar between two groups, and 37.5% of patients with EAC were in stage I (Additional file 1: Table S2). Consistent with ESCC, the EAC group had significantly higher urinary levels of all the five miRNAs than the HC group (miR-1273f: median 2−ΔCT [× 10− 2], 24.834 vs. 3.313, P = 0.002; miR-619-5p: median, 132.526 vs. 23.183, P = 0.007; miR-150-3p: median, 3.073 vs. 0.594, P = 0.004; miR-4327: median, 94.140 vs. 29.680, P = 0.038; miR-3135b: median, 651.224 vs. 99.363, P = 0.004; Fig. 4A–E). Moreover, these five urinary miRNAs were significantly different between HC and EAC, with excellent AUCs (Fig. 4F).
We also tested the tissue expression levels of miRNA for 7 EAC tissues (5: surgery, 2: ESD) that were available for miRNA extraction. Despite having no significance because of the small sample size, the expression levels of all the five miRNAs tended to be higher in the EAC tissues than in the adjacent normal tissues (Additional file 2: Figure S4).
In silico analysis
Finally, we conducted GO analysis to explore the functions of miR-1273f and miR-619-5p, which are the leading miRNA biomarker candidates. In miRTarBase, 295 and 425 genes were listed as the targets of miR-1273f and miR-619-5p, respectively. According to the GO analysis, the target genes showed similar profiles between miR-1273f and miR-619-5p. Although the CC terms of these two miRNAs were different, miR-1273f and miR-619-5p shared the top two enriched BP terms (gene expression and cellular macromolecule biosynthetic process) and the top four enriched MF terms (organic cyclic compound binding, heterocyclic compound binding, ion binding, and nucleic acid binding) (Additional file 2: Figure S5).