Genomic characterization and risk stratification of esophageal squamous dysplasia

Samples characteristics and cohort design

In total, 66 healthy donors, 141 ESD patients, 88 ESCC patients and 13 NMBzA rats were used in this study. The flow chart of the present study and the number of samples and/or cases for each analysis are shown in Figure S1. Briefly, we collected 72 ESD samples and corresponding 49 normal samples from 49 esophageal squamous cell carcinoma-free patients (ESD cohort) for whole-exome sequencing (WES) and genomic analyses, then compared with 88 ESCC dataset (ESCC cohort) and 13 NMBzA-induced rat models. Samples were collected from 2007 to 2016 and no patients had received chemotherapy or radiotherapy (Table S1). 1) For somatic single nucleotide variants (SNVs) and insertions and deletions (InDels) analysis, 72 ESD samples with matched normal samples, including 3 SH, 27 LD and 42 HD samples (Figure S1), were used, and then extended analyses with 88 ESCC samples from our published dataset [11]; 2) For CNA analysis, 72 ESD samples and 88 ESCC samples were used; 3) The major findings were confirmed sequentially in our independent ESD cohort (66 patients) by Sanger sequencing, and a cohort of NMBzA model of 13 rats by WGS; 4) For the clinical application studies, we used sera of 26 HD patients as well as 66 healthy controls to explore the potential of genomic alterations in ESD non-invasive diagnosis. All samples were obtained with the approval of the ethics committee of Chinese People’s Liberation Army General Hospital, Peking University Cancer Hospital and The Fourth Hospital of Hebei Medical University. All patients provided written informed consent before enrollment in this study.

Establishment of esophageal cancer model in rat using NMBzA

NMBzA (purity>98 %) was generously provided by Professor Shih-Hsin Lu. Dimethyl sulfoxide (DMSO) was purchased from Sigma-Aldrich Inc (St. Louis, MO, USA). All procedures involving rats were carried out in accordance with the standards approved by the Ethical Committee of National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences. Male SD rats (3–4 weeks old) were purchased from Vital River Laboratory Animal Technology Co., Ltd (Beijing, China). Rats were housed two animals per cage under standard conditions (24 ± 2 °C, 40 %–70 % relative humidity, and 12 h light/dark cycles) and with unlimited access to standard rodent maintenance feed (KEAO XIELI FEED Co., Ltd, Beijing, China) and water. Male SD rats were administrated with NMBzA (0.30 mg/kg) three times/week by subcutaneous injections, and lasted for 12 weeks to induce the development of ESD. From week 16 to 44, rats were euthanized at a four-week interval. Each esophagus was opened longitudinally, epithelium lesion and tumors were collected.

Cell lines and cell culture

Immortalized esophageal epithelium cell line HET-1A was cultured in high glucose DMEM (Gibco) with 10 % fetal bovine serum (FBS). All of these cells were maintained at 37 °C with 5 % CO₂.

Whole-exome and whole-genome sequencing

For human whole exome sequencing, capture libraries were prepared from 200 ng genomic DNA (gDNA) using Agilent SureSelect Human All Exon V6 kit (Agilent Technologies, CA, USA) following manufacturer’s recommendations. Fragmentation was carried out by hydrodynamic shearing system (Covaris, Massachusetts, USA) to generate 280 bp fragments. DNA fragments with ligated adapters on both ends were selectively enriched in a PCR reaction which hybridized with liquid phase with biotin-labeled probes. All of the constructed libraries were sequenced on HiSeqX platform with 2 × 150 bp paired-end reads.

For whole genome sequencing (WGS) of rat, paired-end DNA library were prepared according to manufacturer’s instructions (Illumina Truseq Library Construction). Two microgram genomic DNA was sheared into 350 bp fragments, followed by end-polished, A-tailed, and ligation with the adaptors. The concentration and the size distribution of the libraries were determined on an Agilent Bioanalyzer DNA 1000 chip. DNA libraries were sequenced on Illumina Hiseq X according to manufacturer’s instructions and paired-end reads were generated.

Somatic mutations detection

All data were rigorously filtered to remove any reads containing adapter sequence and low-quality reads. High quality paired-end reads were aligned to GRCh38 (for human samples) or rn6 (for rat samples) reference genome using BWA (v0.7.17) [12]. Picard (v1.119) was used to mark duplicate reads caused by PCR. Somatic mutations were identified using consensus calling pipeline from Pan-Cancer Analysis of Whole Genomes (PCAWG) [13], which consisted of multiple calling approaches, including GATK4 (v4.1.0.0, MuTect2) [14, 15], CaVEMan (v1.15.3) [16], Bcftools (v1.16) [17], MuSE (v2.0) [18] for calling somatic SNVs, and MuTect2, cgpPindel (v3.10.0) [19], Platypus (v0.8.1) [20] for calling somatic InDels. All mutations were annotated using ANNOVAR (v2020Jun7) [21]. Somatic mutations were subjected to a rigorous filtering, including coverage of mutated sites ≥10 reads, at least four mutated reads, and variant allele frequency of somatic mutations ≥0.02 were retained. To filter possible polymorphic sites of rat, we removed any variants that present in rat dbSNP (build 149) or control samples and those with at least two occupies in different rat individuals. Only variants on Cancer Gene Census (CGC) genes or rat orthologues of human genes were selected for further analysis. Rat:human ortholog pair assignment was retrieved from Ensembl compara [22] through BioMart, and rat orthologues with high confidence were used for screening. Somatic CNAs were identified by CNVkit (v0.9.6) while adjusting for the tumor cell fraction or purity (ABSOLUTE v1.0.6) [23, 24]. All control samples were used to construct the copy number reference.

We used weighted-genomic integrity index (wGII) method to assess CIN dynamic during ESD progression [25]. Briefly, we calculated integer ploidy number of each ESD and ESCC sample, then evaluated the percentage of genomic alteration (gain or loss) for every autosomal chromosome. The mean percentage of genomic alteration was then calculated across all 22 autosomal chromosomes, and determined as wGII score.

Identification of significantly mutated genes (SMGs)

Significantly mutated genes were identified with SMG module of the MuSiC2 (v0.2) tools suite [26]. Firstly, MuSiC2 calculated the background mutation rate (BMR) of seven mutational categories, including AT transitions, AT transversions, CpG transitions, CpG transversions, CG (non-CpG) transitions and transversions, and indel category. After comparison between those respective mutation rate and BMR, seven ps were generated for each gene. Three SMG test, convolution test (CT), Fisher’s combined p test (FCPT) and likelihood ratio test (LRT), were applied to summarize all ps. We also compared identified SMGs with previously published driver genes in ESCC studies.

Mutational signature analysis

SigProfilerExtractor [27] was applied to extract de novo mutational signatures, to decompose de novo mutational signatures to known COSMIC signatures (v3.3) [28], then assigned de novo mutational signatures to each sample by YAPSA [29]. For ESCC cohort, only somatic mutations overlapped with exonic regions were used for signature analysis. For ESD cases with multiple samples, only higher grade sample was selected for signature analysis. Possible sequencing artefacts were removed according to COSMIC signatures. Clustering of mutational signatures and clinical data was performed by R package pheatmap (v1.0.12) [30].

Validation of somatic mutations

Somatic mutations were validated by PCR amplification and Sanger sequencing. The PCR primers for putative somatic mutations were designed by primer 5 in silicon. PCR was performed on a Dual 96-well GeneAmp PCR System 9700 (Applied Biosystems), 20 ng template DNA from each sample was used per reaction. The products were sequenced by a 3730xl DNA Analyzer (Applied Biosystems). All sequences were analyzed by the Sequencing Analysis Software Version 5.2 (Applied Biosystems). If the mutations were successfully confirmed in the tumors but not identified in the matched normal DNA, the somatic statuses would be determined as successful validation.

Identifying significant frequent copy number alterations

GISTIC (v2.0.23) [31] algorithm was used to infer recurrently amplified or deleted genomic regions. G-scores were calculated for genomic and gene-coding regions on the basis of the frequency and amplitude of amplification or deletion affecting each gene. A significant CNA region was defined as having amplification or deletion with a G-score>0.1, corresponding to a p threshold of 0.05 from the permutation-derived null distribution.

Colony formation assay and MTS assay

Cells transfected with ACSM5 siRNAs or stable expressed wild-type or mutated ACSM5 were seeded into 6-well culture plates and incubated at 37 °C with 5 % CO₂ for 10 days. Culture plates were performed in triplicate. After washing with pre-cooled PBS, cultures were fixed with pre-cooled methanol for 20 min and stained with crystal violet for 15 min. Colonies were examined and automatically calculated by G:box (Syngene). MTS assay was performed according to the manufacturer’s instructions.

RNA isolation and quantitative real-time PCR

RNA isolation was performed as described in our previous study [32]. Briefly, total RNA of cultured cells was extracted with TRIzol reagent (Invitrogen, Carlsbad, California, USA) and 100 μL blood plasma was applied to RNA isolation by acid phenol-chloroform. The cDNA was synthesized with the PrimeScript RT reagent Kit (Promega, Madison, WI, USA). Real-time PCR was carried out using an ABI 7500 real time PCR system (Applied Biosystems, Foster City, California, USA). Bulge-Loop hsa-miR-4292 qRT–PCR Primer Set (Ribobio, miRQ0016919-1-2) was used for the measurement of the relative quantity of hsa-miR-4292.

Western blot

Western blot was performed as described in our previous study [32]. The antibodies included antibodies to ACSM5 (Sigma, St Louis, MO, USA), Flag tag (Abcam, USA), GFP (Cell Signaling, Boston, MA, USA), β-actin (Sigma, St Louis, MO, USA). Secondary antibodies such as goat-anti-mouse IgG (1:2,000) and goat-anti-rabbit IgG (1:3,000) conjugated with horseradish peroxidase (HRP) were used to probe membrane for 1 h. The membrane was rinsed in 1 × PBS with 0.1 % Tween. After incubation with the Chemiluminescence substrate, photographs were taken by Image Reader LAS-4000 (Fujifilm) and analyzed by the Multi Gauge V3.2 software.

Transwell migration assays

Migration assay was performed as described in our previous study [32]. In brief, migration of cells was assayed in Transwell cell culture chambers with 6.5 mm diameter polycarbonate membrane filters containing 8 μm pore size (Neuro Probe, Gaithersburg, MD, United States). After transfected with has-miR-4292 mimic or control, 1 × 10⁵ HET-1A cells in 100 μL of serum-free medium were added to the upper chamber of the device, and the lower chamber was filled with 600 μL fresh media (added 20 % FBS). After 12 h of incubation at 37 °C, the non-migration cells were removed from the upper surface of the membrane with a cotton swab. The filters were then fixed in methanol for 10 min, stained with crystal violet solution for 1 h, and counted. Five random microscopic fields (×100) were counted per well and the mean was determined.

Immunohistochemistry

Immunohistochemistry (IHC) analysis was performed and diagnosed by two pathologists blindly on the paraffin-embedded ESCC tissue sections (Tissue Microarray) and the xenograft tumor tissues. In brief, the sections were deparaffinized with xylenes and rehydrated in graded ethanol. Sections were submerged into EDTA antigenic retrieval buffer (pH=8.0) and microwaved for antigenic retrieval. The sections were then treated with 3 % hydrogen peroxide in methanol to quench the endogenous peroxidase activity, followed by incubation with 1 % goat serum albumin to block nonspecific binding. The tissue sections were incubated with rabbit anti-ACSM5 (1:200; Sigma) overnight at 4 °C. After washing, the tissue sections were treated with goat anti-mouse/rabbit IgG HRP-polymer (ZSGB-BIO, Beijing, China) for 20 min. 3,3′-Diaminobenzidine was used as the chromogen. The scores were determined by combining the intensity of staining and the proportion of positively stained tumor cells as described previously [33].

Xenograft tumor formation in nude mice model

For xenograft tumor formation study, 5 × 10⁶ HET-1A cells with stable expressed wild-type ACSM5, mutated ACSM5 (p.C135delinsCMR) and control cells were injected subcutaneously into the right dorsal flank of BALB/c nude mice (10 mice per group). The tumor volume and weight of mice were measured twice per week. Tumors were then excised and embedded in paraffin for hematoxylin and eosin staining (H&E) and immunohistochemistry (IHC) analysis.

Statistical analysis

Two tailed Student’s t-test was used to analyze the results expressed as Mean ± S.E.M. The survival curves were plotted by using Kaplan-Meier analysis and compared by log-rank test. Survival data were evaluated by multivariate Cox regression analysis. Differences were considered significant when the p-value was less than 0.05. Fisher’s exact test was used to analyze the significance of the correlation between clinical data and mutated events. For mutational burden and signature, Wilcoxon rank sum test was applied. All statistical analyses were performed using the R (version 3.6.1) environment.

Comparison of exonic mutational burden and signatures between ESD and ESCC

As part of the International Cancer Genome Consortium-Accelerating Research in Genomic Oncology (ICGC-ARGO) project, we recruited 141 ESD patients without invasive cancer by endoscopic submucosal dissection (Figure 1A, Figure S1). Slides stained with H&E were reviewed independently by three experienced pathologists to identify the consensus areas of morphological classification according to WHO fifth edition, and grouped into one of the following histological categories: simple hyperplasia (SH), low-grade dysplasia (LD) and high-grade dysplasia (HD) (Figure 1B). Seventy two ESD samples of different histological grades from 49 patients were subjected to whole-exome sequencing (WES) (Figure S1), with a median sequencing depth of 182× (range 106×–317×) (Table S2). The somatic mutations per megabase (Mb) ranged from 0.03 to 11.26, with a median of 0.70 mutations/Mb (Table S3). The mutational burden of LDs (0.34 mutations/Mb) was lower than that of HDs (1.21 mutations/Mb; Wilcoxon rank sum test, p=0.01), all of which were lower than previously reported LDs (3.3 mutations/Mb) and HDs (4.1 mutations/Mb) from invasive tumor-bearing patients [8]. Although SHs represented a lower mutational burden (0.18 mutations/Mb), we excluded SHs for mutation burden comparisons due to small sample size (n=3). Interestingly, when compared with our published dataset of 88 ESCC patients (2.84 mutation/Mb) [11], we observed both LDs and HDs harbored lower mutational burden than ESCC (Figure 2A; Wilcoxon rank sum test, p=1.1 × 10⁻⁸ and 2.1 × 10⁻⁹, respectively), and this difference was not driven by the purity levels of ESD and tumor samples, since there was no significant correlation between the number of single nucleotide variants (SNVs) and the purity of those samples (Spearman’s rho, r=−0.159, p=0.18).

Figure 1:

Overview of the study and the representative hematoxylin and eosin staining of esophageal epithelium and lesions. (A) Schematic representation of sample collection of patients and NMBzA rat model. (B) The representative hematoxylin and eosin staining of esophageal epithelium and lesions.

Figure 2:

Comparison of mutational burdens and signatures of ESD and ESCC. (A) Mutational burden of simple hyperplasia (SH, n=3), low-grade dysplasia (LD, n=27), high-grade dysplasia (HD, n=42) and ESCC (n=88). The data were analyzed by the Wilcoxon rank-sum test, two-sided. *p<0.05, **p<0.01, ***p<0.001. (B) Base substitutions of three de novo mutational signatures. Base substitutions of each signature are shown as color bars. The height of each bar represents the proportion of substitution pattern. (C) Comparison of relative contributions of three de novo mutational signatures between dysplasia and ESCC samples. Wilcoxon rank-sum test. (D) Clustering of three de novo mutational signatures. Top, hierarchical clustering analysis of mutational signatures of ESD and ESCC patients; middle, clinical features indicated by color blocks, combined means the combination of drinking and smoking; bottom, relative contribution of three de novo mutational signatures in each individual.

We then used SigProfilerExtractor to extract de novo mutational signatures [27], to decompose de novo mutational signatures to known COSMIC signatures (v3.3) [28], and to assign de novo mutational signatures to each sample by YAPSA [29]. After excluding possible sequencing artefacts, we identified three de novo mutational signatures (Figure 2B). Each de novo signatures was matched to a combination of COSMIC signatures (Figure S2). Signature A was characterized by clock-like signatures (SBS5 and SBS1) and SBS18 signature (Damage by reactive oxygen species). Activity of APOBEC family of cytidine deaminases (SBS13 and SBS2) was found as the predominant signature in Signature B. Signature C matched a combination of COSMIC signatures SBS5, SBS37 (unknown aetiology), SBS22 (Aristolochic acid exposure), and SBS1. In addition, significantly different contributions of Signatures B and C were observed between ESD and ESCC, suggesting that various mutational processes contribute to tumor progression at different stages (Figure 2C). Clustering analysis based on mutational signatures displayed four distinct clusters across ESD and ESCC samples (Figure 2D). Interestingly, we observed that patients with a history of drinking or combined with smoking, were significantly enriched in cluster 4 (Fisher’s exact test, p=0.002) which was dominated by Signature C. Collectively, these results revealed that ESD had distinct mutational burden and signatures compared with ESCC, and exogenous and endogenous factors contributed to mutagenesis during tumor progression.

Mutational landscape associated with ESD progression across species

We next investigated significantly mutated genes (SMGs) that facilitated ESD progression. Several well-known ESCC SMGs were identified in both ESD and ESCC patients (Figure 3A), including TP53, NOTCH1, ZNF750, CDKN2A, FAT1, and PIK3CA [11, 34], [35], [36], [37], [38], implying that these genes may play an important role in tumor development from initiation to invasive tumor. Subsequently, we analyzed ESD samples of two patients, each had multiple histological grades, including one SH, two LDs and one HD from Patient P040, one SH and one HD from Patient P023. Multiple mutations were found at SH or LD stage, and persisted until HD stage (Figure S3), for example, TP53 non-sense mutation (p.R64X, Patient P023) and NOTCH1 splice-site mutation (c.2969+2T>C, Patient P040) (Table S4). Furthermore, three shared TP53 mutations were maintained and present in all stages during tumor progression (Figure 3B, Table S5). These results suggested that TP53 and NOTCH1 mutations may occur at early stages of tumorigenesis even at SH stage, and may confer a selective advantage on the cells with those mutations. Notably, TP53 showed lower mutation frequencies in ESD compared with ESCC (odd ratio=0.1, p-value<0.0001, fisher’s exact test), while NOTCH1 showed a reverse pattern (odd ratio=2.83, p=0.011, fisher’s exact test), indicating their differential contributions to tumor progression.

Figure 3:

Significantly mutated genes in ESD and ESCC samples of patients and NMBzA rat model. (A) Mutational landscape of SH, LD, HD and ESCC patients. Top, mutations per megabase (Mb) of each sample; middle, significantly mutated genes colored by mutational types; right, the percentage of samples with specific mutated genes according to different stages. (B) Venn plot shows the shared and unique somatic mutations among LDs, HDs and ESCCs. (C) Mutational landscape of NMBzA rat model. Top, the number of non-synonymous mutations in each sample; bottom, significantly mutated genes colored by mutational types.

About 75 % of HDs progress to ESCC without proper treatment [5], which inspired us to figure out the mutations that occurred in HDs and ESCCs concurrently. We identified ZNF750, EYS, MUC17 and SMPD1 which emerged in HDs and persisted in ESCCs, but were absent in LDs. A previous study reported that ZNF750 was a tumor suppressor, and the mutant ZNF750 promoted cell proliferation, foci formation, migration, and xenograft tumor growth in multiple squamous cell carcinomas [39]. MUC17 mutations was associated with poor prognosis in glioma cohorts [40]. It is possible that those mutations occurring at key transitional points will promote extensive mutagenesis and the development of invasive carcinoma [41].

In addition, new SMGs were identified across ESD samples, including EYS, RERE, CDH9, CACNA2D1, PITPNB, RBP3, ACSM5. We then employed a well-established rat ESCC model [42, 43] to investigate whether these driver events were conserved across species in the progression of ESD to ESCC. In brief, male SD rats were treated with NMBzA to induce the development of esophageal lesions (Figure 1A). At week 16–44, rats were euthanized at a four-week interval, and epithelium lesion and tumors were collected. In total, 13 samples of NMBzA induced rat esophagus lesions (n=8 for LD; n=3 for HD; n=2 for ESCC) and four normal samples were conducted with whole-genome sequencing (Tables S6 and S7). The mutation spectrum was dominated by C>T/G>A transitions (Figure S4 and Table S8). As expected, well-known ESCC SMGs, such as Tp53, Notch1, Fat1, were present in NMBzA model, and new SMGs were also identified, including Net1, Bmp4 and Acsm5 (Figure 3C). None of the two ESCC samples of NMBzA model borne Tp53 mutations, the possible reason may be the small sample size (n=2) of ESCC samples of NMBzA model. Strikingly, about 70 % (9/13) of samples harbored mutations in Wnt signaling pathway, including Fat1 and Fat3. This suggested the essential role of Wnt signaling pathway in the initiation of dysplasia. Of the new SMGs, Acsm5 was detected in both ESD and ESCC samples of the NMBzA model. Driver genes shared between ESD patients and NMBzA model highlighted the importance of these mutated genes in the malignant transformation of esophageal epithelial cells.

ACSM5 is a novel driver gene for ESD progression

Our cross-species genomic profiles identified a novel SMG, ACSM5, which was not been linked to tumorigenesis previously. ACSM5 (acyl-CoA synthetase medium chain family member 5), catalyzes the first step of fatty acid oxidation with CoA to produce acyl-CoA, the activation status of fatty acid [44]. We observed that ACSM5 was mutated in three of 72 ESD samples (2/49 patients, 4.08 %). Among the three ACSM5 somatic mutations, p.C135delinsCMR reoccurred in our independent ESD cohort (66 patients) (Table S9), which prompted us to decipher the function of this mutation. Interestingly, knockdown of ACSM5 in immortalized normal esophageal cell lines, HET-1A, promoted cell proliferation and colony formation (Figure 4A, C and D). Additionally, mutated ACSM5 (p.C135delinsCMR) remarkably promoted the malignant phenotypes of immortalized normal esophageal cells in vitro and in vivo (Figure 4B–G).

Figure 4:

Altered expression of ACSM5 or its mutant contributes to the malignant transformation of immortalized normal esophageal epithelial cell. (A) Knockdown of ACSM5 by siRNAs promoted HET-1A cell proliferation (MTS assay). Three independent experiments of each siRNAs (siACSM5-1 and siACSM5-2) were performed. (B) ACSM5 mutation (p.c135delinsCMR) increased HET-1A cell proliferation (MTS assay). (C and D) Knockdown of ACSM5 by siRNAs enhanced the colony formation ability of HET-1A (up), ACSM5 mutation (p.c135delinsCMR) enhanced the colony formation ability of HET-1A (bottom). (C) Representative results of colony formation. (D) Quantitative data of colony formation. (E-G) Mutated ACSM5 (p.c135delinsCMR) significantly promoted tumor growth in vivo. (E) Representative tumor xenografts. (F) Tumor weight. (G) Tumor volume. (H) Expression level of ACSM5 in normal, ESD and ESCC tissues (IHC assay). (I) Kaplan–Meier survival analysis of ACSM5 expression in ESCCs (Log-rank test). (J) Multivariate Cox regression survival analysis. *p<0.05, **p<0.01, ***p<0.001, ns, not significant. ESD, Esophageal squamous dysplasia; ESCC, esophageal squamous cell carcinoma.

Subsequently, when interrogating the whole exonic region of ACSM5 in our independent ESD cohort (66 patients), we identified another four coding mutations (Table S9). Additionally, we found six coding mutations of ACSM5 in our previously published WGS cohort of ESCC patients (Table S10) [38]. We next examined protein expression of ACSM5 in normal, ESD and tumor tissues. ACSM5 was highly expressed in normal tissues compared to ESDs and ESCCs (Figure 4H). Statistical analysis revealed that the expression level of ACSM5 was negatively correlated with American Joint Committee on Cancer (AJCC) stage (p=0.041), lymph node metastasis (LNM; p=0.027) and gender (p=0.044) (Tables S11 and S12). Notably, low expression of ACSM5 was strongly associated with poor survival of ESCC patients (p=0.001, Kaplan–Meier survival analysis and log-rank test, Figure 4I), and the five-year survival rate in the low expression group of ACSM5 (7.69 %) was substantially lower than that in the high expression group (31.25 %). The median survival in the low expression group of ACSM5 was 15 months (95 % CI 10.048–19.952), while 27 months (95 % CI 17.309–36.691) in the high expression group. Multivariate Cox regression survival analysis adjusting for age, tumor stage, LNM, pathological grade and gender also reported the strong correlation between low expression of ACSM5 and shorter survival (p=0.033, HR=0.580, 95 % CI 0.351–0.958; Figure 4J), indicating that the expression of ACSM5 was an independent prognostic factor for clinical outcome in ESCC. Collectively, our findings suggested that ACSM5 may act as a tumor suppressor, and mutated ACSM5 or altered expression of ACSM5 probably exerted tumor-promoting potential during tumor progression.

Profiling of copy number alterations and clinical application of miR-4292 as a diagnostic marker

Somatic CNA analysis revealed that the CNA burden of LDs and HDs were significantly lower than that of ESCCs (Wilcoxon rank sum test, p=0.025 and 2.4e-06, respectively) (Figure 5A). And this trend was also observed in chromosome instability (CIN) (Figure S5). LDs and HDs accumulated more CNAs during tumor development (Figure 5A and B). For well-defined CNAs in ESCC, we noticed that 29.63 % of LDs and 47.62 % of HDs, while 65.91 % of ESCCs had the most prevalent amplification region 11q13.3, which harbored CCND1, FGF4, FGF19, MYEOV, TPCN2, and our previously identified microRNA (MIR548K) (Figure 5B) [11, 32]. Other CNAs identified in our study included amplifications in 3q, 5p, 9q, and 8q, and deletions in 3p and 9p (Figure 5B).

Figure 5:

The landscape of CNAs. (A) Comparison of CNA burden between ESCC (n=88) and LD (n=27), HD (n=42), respectively. (B) CNAs profiling in SH, LD, HD and ESCC samples. Amplifications are denoted by yellow, and deletions are denoted by green. Y axis indicates CNA frequency. (C) Representative fluorescence in situ hybridization (FISH) of miR-4292 and NOTCH1 status. (D) miR-4292 mimic overexpression increased colony formation of HET-1A cells. (E) miR-4292 mimic overexpression promoted cell proliferation of HET-1A cells (MTS assay). (F) miR-4292 mimic overexpression enhanced cell mobility and invasiveness of HET-1A cells (Transwell assay). (G) The abundance of miR-4292 in serum across ESD patients and health donors were detected by real time PCR. Relative expression level was normalized to U6. (H) ROC curve evaluated the diagnostic accuracy of miR-4292 abundance. ROC, receiver operator characteristic. AUC, area under the curve. Error bars indicate SEM of three independent experiments. *p<0.05, **p<0.01, ***p<0.001.

We next analyzed the focal amplified region of 9q34.3, which harbored NOTCH1, and several miRNAs including miR-4292. Fluorescence in situ hybridization (FISH) assay validated that miR-4292 and NOTCH1 were amplified in HD regions of patient (P025) and patient P030 (Figure 5C). Furthermore, overexpression of miR-4292 mimic significantly promoted cell proliferation, colony formation, cell mobility and invasiveness (Figure 5D–F). Cell-free circulating miRNAs are promising diagnostic biomarkers for early detection of cancer [45]. Thus, we sought to evaluate the diagnostic value of miR-4292 in serum samples of 26 ESD patients and 66 health donors. Interestingly, the serum abundance of miR-4292 was significantly higher in ESD patients than that in health controls (Figure 5G, Student’s t-test, p<0.001), and also higher in publicly independent ESCC serum cohort and ESCC tissue cohort (Figure S6, Wilcoxon rank sum test, p<0.001) [46, 47]. Receiver operator characteristic (ROC) curve was used to determine the diagnostic accuracy of miR-4292 in this cohort, and the AUC was 0.774 with 95 % CI (0.678–0.870) (Figure 5H). Collectively, these findings illustrate the potential significance of miR-4292 in the progression of ESD and highlight its utility as a non-invasive diagnostic marker for early detection of ESD.