Identification of serum 5’tRF-Val as a promising non-invasive biomarker for early detection of esophageal cancer

Abstract

Esophageal cancer (EC) is frequently detected at an advanced stage because of the lack of effective noninvasive screening tools, leading to poor prognosis. This study aimed to identify serum transfer RNA-derived small RNA (tsRNA) biomarkers for early detection of EC. Candidate tsRNAs were screened using the tsRFun database, and a quantitative real-time PCR (qRT‒PCR) method for serum 5’tRF-Val was subsequently established and validated. A total of 289 participants were enrolled, including 124 EC patients (71 with early-stage, 53 with advanced-stage disease), 56 individuals with benign lesions, 77 healthy subjects, and 32 postoperative EC patients. Detection performance was evaluated by receiver operating characteristic (ROC) curve analysis, and correlations with clinicopathological parameters were examined. 5’tRF-Val levels were significantly higher in the serum of EC patients than in that of healthy subjects. Its expression correlated positively with aggressive pathological features, including advanced TNM stage and the presence of metastasis. Notably, 5’tRF-Val demonstrated high detection accuracy for early-stage EC, with an AUC of 0.857. Importantly, it effectively distinguished early EC lesions from benign lesions (AUC = 0.824). Furthermore, longitudinal analysis revealed a significant decrease in 5’tRF-Val levels following radical surgery, suggesting its potential for monitoring treatment response. Serum 5’tRF-Val was identified in this study as a novel and promising noninvasive biomarker for the early detection of EC. Owing to its high detection accuracy, association with disease progression, and postoperative dynamic changes, 5’tRF-Val may help to fill the current gap in noninvasive detection of early EC, but this needs to be further verified in multicentre studies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54913-2.

Introduction

Esophageal cancer (EC) is a major public health problem worldwide. As it is the seventh most frequently diagnosed malignant tumour worldwide, approximately 470,000 new cases are reported per year¹, with an overall five-year survival rate of approximately 20%². This poor prognosis is largely attributable to late detection. EC is typically detected at an advanced stage, often because early symptoms are absent, which limits the feasibility of curative treatment. As a result, even recent surgical and immunotherapeutic advances have resulted in limited improvements in survival^3,4. Improving early detection is therefore critical. However, although endoscopic biopsy remains the gold standard for diagnosis, its invasiveness, high cost, and reliance on specialized personnel severely limit its applicability in large-scale screening⁵, creating an urgent need for novel, noninvasive biomarkers.

Currently, pan-cancer analytical strategies capable of integrating multi-omics data to systematically uncover molecular signatures shared across different cancer types have been widely adopted, for instance, have identified broadly dysregulated biomarkers such as CORO1A, CLIC6, BEND3, and PPME1^6–9 and hold clear value for population-wide screening; however, the design of these signatures favours universal signals over tissue-specific traits and therefore cannot provide specific detection of EC. This inherent lack of specificity substantially also limits their ability to distinguish early-stage malignancies from benign lesions, which is a limitation, especially for EC. In this context, transfer RNA-derived small RNAs (tsRNAs) have recently emerged as a promising alternative class of liquid biopsy biomarkers. Specifically, generated by site-specific tRNA cleavage under cellular stress, tsRNAs actively participate in the regulation of gene expression, translation, and cell signalling pathways, thereby playing important roles in tumour development and progression^10–12. Compared with other noncoding RNAs, tsRNAs offer distinct advantages: they are highly stable and readily detectable in serum, plasma, and exosomes, and they typically exhibit strong tissue specificity and disease correlation^13–15. These properties, combined with their demonstrated clinical potential in colorectal cancer, breast cancer, and hepatocellular carcinoma^16–19, position tsRNAs as attractive candidates for noninvasive early cancer detection. However, the exploration of tsRNAs for the early detection of EC remains in its infancy, and no study has systematically evaluated the performance of a specific serum tsRNA for distinguishing early EC patients from both healthy subjects and patients with benign esophageal lesions.

In this study, we first screened the differentially expressed tsRNAs in EC tissues through the tsRFun database, and by validating their expression levels in EC cells and EC serum samples, we ultimately selected the significantly highly expressed 5’tRF-Val for study. By analysing its differential expression in serum, evaluating its detection performance for early diseases, and exploring its correlation with clinical pathological characteristics, this work aimed to identify a promising and verified biomarker for the noninvasive early detection of EC.

Methods

Study design, participants, and sample collection

In this study, 289 participants from the Nantong area were enrolled between September 2024 and December 2025, comprising 124 patients with EC (EC; 71 early-stage, 53 advanced-stage), 56 patients with benign esophageal lesions, 77 healthy subjects, and 32 postoperative patients with EC. All participants provided written informed consent. The study was approved by the Ethics Committee of Nantong University Affiliated Hospital (Approval No. 2023-L166) and conducted in accordance with the Declaration of Helsinki. The allocation of serum and tissue samples across the screening, methodological evaluation, tissue validation, and detection performance assessment stages is summarized in Fig. 1A.

Fig. 1.

This study screened and validated the expression of tsRNAs in EC. (A) Participant enrollment and sample allocation across the study stages. (B) This panel illustrates the workflow for screening 5’tRF-Val. (C) The expression levels of three tsRNAs (tsRNA-Val-5-0092, tsRNA-Gln-5-0033, and tsRNA-Ile-5-0044) were measured in serum samples from 20 EC patients and 20 healthy subjects (n = 20 per group). Two-way ANOVA with Šídák’s multiple comparisons test was used. (D) The expression of 5’tRF-Val was analyzed in 20 pairs of EC tissues and adjacent non-cancerous tissues. Statistical analysis used two-tailed paired t-test (E) The expression of 5’tRF-Val was examined in normal esophageal epithelial cells (SHEE) and three EC cell lines (ECA-109, KYSE-150, KYSE-30) (n = 3 independent experiments per group). One-way ANOVA with Dunnett’s multiple comparisons test was applied. Data are relative expression (2^−ΔΔCt) and are shown as mean ± SD; exact P values are shown in the figure.

All participants were aged ≥ 18 years. EC and benign lesions were confirmed by histopathology; healthy subjects showed no evidence of esophageal disease on physical examination or routine tests. The exclusion criteria included pregnancy; lactation; a history of any other malignancy within the past five years; and the use of radiotherapy, chemotherapy, or endoscopic treatment within 3 months before blood collection. Eligible participants were enrolled consecutively without additional selection.

Peripheral venous blood (5 mL) was collected into serum separation tubes (BD Vacutainer) before invasive procedures or treatments were performed. After coagulation (30 min, room temperature), serum was isolated by centrifugation (3000 rpm, 10 min, 4 °C), aliquoted into RNase-free tubes, and stored at − 80 °C. For postoperative patients, blood was drawn 7 days after radical resection²⁰. Tumour and adjacent normal mucosal tissues (approximately 5 cm from the tumour margin)²¹ were obtained during surgical resection or endoscopic submucosal dissection, rinsed with ice-cold phosphate-buffered saline, snap-frozen in liquid nitrogen within 10 min, and stored at − 80 °C. With the exception of the intentional freeze‒thaw experiments, no sample underwent more than one freeze‒thaw cycle. Sample quality was monitored by visual inspection for haemolysis (no samples were excluded for this reason) and by U6 Ct values (< 35 in all analysed samples), confirming acceptable RNA integrity and controlled preanalytical conditions.

Cell culture

Perpetual human esophageal epithelial cells (SHEEs) were used as normal controls. Three human EC cell lines—ECA-109, KYSE-150 and KYSE-30—represent malignant phenotypes. All cell lines were obtained from the certified cell library at the Shanghai Institute of Biological Sciences of the Chinese Academy of Sciences. Before the experiments were performed, each cell line was identified by short tandem repeated sequence (STR) analysis. The cells were maintained in a humidified incubator at 37 °C and 5% CO₂ (Thermo Fisher Scientific, USA). The culture medium consisted of RPMI-1640 (KeygenBioTECH, China) supplemented with 10% foetal bovine serum (Vazyme, China) and 1% penicillin‒streptomycin.

RNA extraction, cDNA synthesis, and quantitative PCR

A special RNA extraction kit (BioTeke Corporation, China) was used to extract total RNA from patient serum samples. TRIzol reagent (Vazyme, China) was used to separate total RNA from the EC tissue specimens. Total RNA was subsequently reverse‑transcribed into cDNA using a stem‑loop reverse transcription primer specific for 5’tRF‑Val and U6 in a 10-µL reaction, which was incubated at 42 °C for 60 min, followed by 70 °C for 5 min. The obtained cDNA was stored immediately for a short time at 4 °C or for a long time at -20 °C. Quantitative reverse transcription‒PCR (qRT‒PCR) with a reaction volume of 20 µL was carried out using an ABI QuantStudio 5 real-time fluorescence quantitative PCR system (Thermo Fisher Scientific, USA). U6 was used as an internal reference. The specificity of amplification was confirmed by melting curve analysis, and the efficiency of the primers used was validated prior to the study. The inter- and intra-assay reproducibility of the entire workflow from RNA extraction to qRT‒PCR was evaluated using pooled serum samples. Stem‑loop reverse transcription primers and specific PCR primers for 5’tRF-Val were designed and synthesized (RiboBio, Guangzhou, China; order number: PA20250721007).

Methodological evaluation

To evaluate the stability of serum RNA under different storage conditions and to analyse its detection sensitivity, we mixed blood samples collected from 20 random individuals, which were then treated differently.

Room-temperature stability

The mixed serum was incubated at 25 °C for different durations. Total RNA was extracted from each aliquot at 0, 6, 12, 18, and 24 h, and the expression levels of 5’tRF-Val were quantified by qRT‒PCR.

Freeze‒thaw stability

The pooled serum was subjected to 0, 2, 4, 6, and 8 complete freeze‒thaw cycles. Each cycle consisted of complete solidification at − 80 °C followed by complete dissolution at room temperature. Total RNA was extracted from each sample, and the expression level of 5’tRF-Val was measured by qRT‒PCR.

Gradient dilution experiment

Total RNA from mixed serum was extracted from the mixture and reverse transcribed into cDNA. This cDNA reserve solution was diluted 10 times in series, ranging from 10 times to 100,000 times. 5’tRF-Val expression at all dilution points was measured by qRT‒PCR.

Assay for traditional biomarkers

For traditional biomarkers, including CEA and CA19-9, the clinical reference threshold used was the threshold established by the Affiliated Hospital of Nantong University. The Beckman DxI 800 platform (Beckman Coulter, USA), which uses a chemiluminescence enzyme immunoassay and corresponding proprietary reagents, was used to measure serum CEA and CA19-9 levels.

Statistical analysis

IBM SPSS Statistics 20 and GraphPad Prism 10.1.2 were used for statistical analysis. Continuous variables with a normal distribution are expressed as the average ± standard deviation (SD). The level of 5’tRF-Val in the serum of EC patients, healthy subjects and benign lesion patients was analysed using single-factor analysis (ANOVA), and an independent sample t test was subsequently used for intergroup comparisons. The correlation with clinical pathological characteristics was evaluated by the χ² test, while the correlation between continuous variables was evaluated by the Pearson correlation coefficient. The detection efficiency was evaluated by analysing the receiver operating characteristic (ROC) curve, the best cut-off value was determined by maximizing the Youden index, and the area under the corresponding curve (AUC) was calculated. A two-tailed statistical approach was adopted, with significance defined as P < 0.05. For the combined biomarker panels, binary logistic regression (stepwise) was used to generate prediction probabilities on the basis of serum levels of 5’tRF-Val, CEA, and CA19-9. These predicted probabilities were subsequently assessed by ROC analysis. To explore the potential clinical utility beyond statistical discrimination, a multivariable logistic regression model incorporating age, sex, 5’tRF-Val, CEA, and CA19-9 was constructed. Decision curve analysis (DCA) was subsequently performed across a range of clinically plausible threshold probabilities to assess the net benefit in three predefined scenarios: early EC versus healthy subjects, early EC versus benign lesions, and advanced EC versus healthy subjects.

Results

Screening of the 5’tRF-Val database

Using the tsRFun database with strict criteria (P < 0.05, |log₂FC| > 2) for preliminary screening, we identified 3 differentially expressed tsRNAs (tsRNA-Gln-5-0033, tsRNA-Val-5-0092, and tsRNA-Ile-5-0044). To validate these findings in clinical samples, qRT‒PCR was used to detect the expression levels of these 3 tsRNAs in serum from EC patients and healthy subjects. Among them, tsRNA-Val-5-0092 (which was subsequently named 5’tRF-Val according to MINTbase v2.0) showed the most significant difference and thus was selected as the target for the subsequent study (Fig. 1B-C). In EC tissues and their corresponding adjacent normal mucosa, the expression of 5’tRF-Val was also significantly upregulated (Fig. 1D). This upregulation trend was also confirmed in vitro: compared with those in normal esophageal epithelial cells, the levels of 5’tRF-Val in the three EC cell lines were significantly increased (Fig. 1E). The consistent upregulation of 5’tRF-Val at the serum, tissue and cellular levels suggests that 5’tRF-Val may serve as an early detection biomarker for EC.

The structure and origin of 5’tRF-Val

Following the initial screening, 5’tRF-Val was chosen as the lead candidate for further studies. In accordance with the MINTbase v2.0 nomenclature guidelines, this tsRNA was named 5’tRF-Val, which is annotated in the database as a 30-nt 5’tRF (Supplementary Fig. 1A). Analysis using the UCSC Genome Browser database revealed that 5’tRF-Val was localized to chromosome 11 (Supplementary Fig. 1B). Agarose gel electrophoresis of its qRT‒PCR product confirmed that it was approximately 80 bp in size (Supplementary Fig. 1C). Subsequent Sanger sequencing confirmed that the qRT‑PCR product matched the expected full‑length sequence of 5’tRF-Val (5′-GGTTCCATAGTGTAGCGGTTATCACGTCTG-3′), which is identical to the sequence annotated in the MINTbase v2.0 database (Supplementary Fig. 1D).

Methodological evaluation of serum 5’tRF-Val

We evaluated the precision, stability and linearity of the serum detection of 5’tRF-Val. The method was consistent across room temperature storage, freeze‒thaw cycles and gradient dilution experiments. The repeatability of the 5’tRF-Val assay was evaluated by the coefficient of variation (CV) of the Ct values. The interassay repeatability revealed that the average Ct value of 5’tRF-Val was 26.151 ± 0.348 (CV = 1.331%), and that of the reference U6 was 23.938 ± 0.441 (CV = 1.839%). The results of the intra-assay reproducibility showed that the average Ct value of 5’tRF-Val was 26.301 ± 0.373 (CV = 1.416%) and that of U6 was 23.529 ± 0.798 (CV = 3.406%). The expression level of 5’tRF-Val remained stable under all conditions (Fig. 2A, B), and the melting curve showed a single peak, confirming the stability and specificity of the method. The linear range of the assay was assessed using a gradient dilution method. The linear regression equation for 5’tRF-Val was Y = -2.921 *X + 9.082 (R² = 0.997) and that for the internal reference gene U6 was Y = -2.739 * X + 11.96 (R² = 0.995) (Fig. 2C, D). Both exhibited excellent linear relationships across five orders of magnitude (10⁻¹ to 10⁻⁵ dilution). The 5’tRF-Val signal remained robustly detectable up to a 100,000-fold dilution of the cDNA pool (Ct = 23.974), confirming that the method provides accurate and reliable detection over a wide concentration range with high analytical sensitivity.

Fig. 2.

Validation of the detection method for 5’tRF-Val. (A) Stability of serum 5’tRF-Val levels across 0, 2, 4, 6, and 8 freeze-thaw cycles. No significant differences were observed among groups (P > 0.05, one-way ANOVA). (B) Stability of serum 5’tRF-Val levels after incubation at 25 °C for 0, 6, 12, 18, and 24 h. No significant differences were observed among time points (P > 0.05, one-way ANOVA). (C) Standard curve for 5’tRF-Val generated using a 10-fold serial dilution of cDNA (10^-1 to 10^-5). The linear regression equation is Y = -2.921 X + 9.082, with R² = 0.997. (D) Standard curve for the internal control U6, with a linear regression equation of Y = -2.739 X + 11.960 and R² = 0.995. Each point represents the mean of triplicate qRT-PCR reactions. Exact P values are shown in the figure.

Serum 5’tRF-Val expression and its correlation with clinical pathological parameters

To evaluate its detection potential, we analysed serum 5’tRF-Val expression across the EC, benign lesion, and healthy subject cohorts. qRT‒PCR analysis revealed that serum 5’tRF-Val levels were significantly higher in patients with EC than in healthy subjects and patients with benign lesions (both P < 0.0001). In contrast, no significant difference was found between the benign lesions and healthy subjects (P = 0.956) (Fig. 3A). Using the median expression level (4.87) as the cutoff, we divided the EC patients into high- (n = 62) and low-expression (n = 62) groups. Chi-square analysis revealed that elevated 5’tRF-Val was positively correlated with advanced TNM stage (P = 0.046), deep tumour invasion (T stage, P = 0.017), lymphatic metastasis (P = 0.024), perineural invasion (P = 0.002), and lymphovascular invasion (P = 0.036) but negatively correlated with tumour differentiation (P = 0.033). No significant association was observed with tumour size, sex, or age (Table 1). Further analysis confirmed these trends, revealing that 5’tRF-Val levels increased significantly with advancing TNM stage, depth of invasion, and extent of lymph node metastasis (Fig. 3B–D), as well as in poorly differentiated tumours and samples with perineural or lymphovascular invasion (Fig. 3E–G). Finally, analysis of paired pre- and postoperative serum samples from EC patients revealed a significant decrease in 5’tRF-Val levels following radical surgery, which returned to levels comparable to those in healthy subjects (Fig. 3H, I), indicating its potential for monitoring postoperative disease dynamics.

Fig. 3.

Clinical value of serum 5’tRF-Val in EC. (A) Expression levels of serum 5’tRF-Val in EC patients (n = 124), benign lesion patients (n = 56), and healthy subjects (n = 77). (B) Expression levels of serum 5’tRF-Val in EC patients categorized into early (Stage I–II, n = 71) and advanced (Stage III–IV, n = 53) TNM stages, as well as in healthy subjects (n = 77). (C) Expression levels of serum 5’tRF-Val in EC patients and healthy subjects at different stages of tumor invasion depth (T1–T2: n = 75; T3–T4: n = 49; healthy subjects: n = 77). (D) Expression levels of serum 5’tRF-Val in EC patients with (n = 44) or without lymphatic metastasis (n = 80). (E) Expression levels of serum 5’tRF-Val in well-differentiated (including well and moderately differentiated, n = 85) and poorly differentiated EC patients (n = 39). (F) Expression levels of serum 5’tRF-Val in EC patients with (n = 40) or without Lymph blood vessel invasion (n = 84). (G) Expression levels of serum 5’tRF-Val in EC patients with (n = 41) or without Perineural invasion (n = 83). (H) Expression levels of serum 5’tRF-Valin postoperative EC patients (n = 32) and healthy subjects (n = 77). (I) Expression levels of serum 5’tRF-Val in 32 EC patients before and after surgery. Each line connects the pre- and post-operative values from the same patient. Data were analyzed by paired t-test. Data are presented as relative expression (2^−ΔΔCt) and shown as individual values with mean ± SD. Panels A–C were analyzed by one-way ANOVA with Tukey’s multiple comparisons test; panels D–H by unpaired t-test. Exact P values are shown in the figures.

Table 1.

Association between 5’tRF-Val levels and clinicopathological features in esophageal cancer .

Characteristic	n	5’tRF-Val		P value
		Low	High
Total	124	62	62
Gender				0.186
Male	98	46	52
Female	26	16	10
Age(years)				0.358
<70	49	27	22
≥ 70	75	35	40
Tumor size(cm)				0.186
< 5	98	52	46
≥ 5	26	10	16
Differentiation grade				0.033*
Well differentiation	85	48	37
Poorly differentiation	39	14	25
Lymphatic metastasis (Yes/No)				0.024*
Yes	44	16	28
No	80	46	34
Perineural invasion(Yes/No)				0.002**
Yes	40	12	28
No	84	50	34
Lymph blood vessel invasion(Yes/No)				0.036*
Yes	41	15	26
No	83	47	36
TNM stage				0.046*
I-II	71	41	30
III-IV	53	21	32
T stage				0.017*
T1-T2	75	44	31
T3-T4	49	18	31

(The cutoff value for high/low expression was the median serum 5’tRF-Val level (4.87))

Statistical annotations: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; NS, no significan.

Detection performance of 5’tRF-Val for esophageal cancer

Given the association between elevated 5’tRF-Val and aggressive tumour characteristics, we hypothesized that changes in its expression occur early in tumorigenesis. We therefore evaluated its detection performance across different clinical scenarios.

We first evaluated the detection performance of 5’tRF-Val for early EC (Fig. 4A). 5’tRF-Val achieved an AUC of 0.857 (95% CI, 0.798–0.916), significantly outperforming CEA (AUC = 0.794) and CA19-9 (AUC = 0.629) (Fig. 4B). At the optimal cut-off value of 1.7959 (determined by maximizing Youden’s index, 0.557), 5’tRF-Val yielded a sensitivity of 81.69% and specificity of 74.03% (Table 2). Combining 5’tRF-Val with conventional markers via logistic regression-derived composite indices further improved the detection accuracy over any single marker alone (Fig. 4C). Calibration analysis revealed that while 5’tRF-Val alone as an independent predictor exhibited some deviation (Hosmer–Lemeshow test, P = 0.005), the joint detection model integrating 5’tRF-Val with CEA or with both CEA and CA19-9 demonstrated good consistency between the predicted probabilities and observed frequencies (P = 0.142; Supplementary Fig. 2A–D). Importantly, DCA of this scenario revealed that 5’tRF-Val alone, as well as the full model, consistently provided a greater net benefit than CEA or CA19-9 across a wide range of threshold probabilities, suggesting that using 5’tRF-Val to guide clinical decisions in early-stage screening may offer greater practical utility than relying on traditional markers alone (Supplementary Fig. 3A).

Fig. 4.

Expression and detection value of serum 5’tRF-Val in different cohorts. (A–C) Comparison between early EC patients and healthy subjects. (A) Serum levels of 5’tRF-Val in early EC patients (n = 71) and healthy subjects (n = 77). (B–C) Detection performance of 5’tRF-Val, CEA, and CA19-9 for distinguishing early EC from healthy subjects. (D–F) Comparison between early EC patients and benign lesion patients. (D) Serum levels of 5’tRF-Val in early EC patients (n = 71) and benign lesion patients (n = 56). (E–F) Detection performance of 5’tRF-Val, CEA, and CA19-9 for distinguishing early EC from benign lesion patients. (G) Serum levels of 5’tRF-Val in advanced EC patients (n = 53) and Early EC patients (n = 71). (H–I) Detection performance of 5’tRF-Val, CEA, and CA19-9 for distinguishing advanced EC (n = 53) from healthy subjects(n = 77). Data are relative expression (2^−ΔΔCt) and are shown as mean ± SD. Panels A, D, G were analyzed by unpaired t-test; panels B, C, E, F, H, I by ROC curve analysis. Exact P values are shown in the figures.

Table 2.

Comparative performance of 5’tRF-Val, CEA, and CA19-9 for detecting EC across different comparison groups.

Marker panel		Early EC vs. Healthy	Early EC vs. Benign Lesions	Advanced EC vs. Healthy
5’tRF-Val	SEN(%)	81.69 (58/71)	81.69 (58/71)	77.36 (41/53)
	SPE(%)	74.03 (57/77)	67.86 (38/56)	94.81 (73/77)
	ACCU(%)	77.70 (115/148)	75.59 (96/127)	87.69 (114/130)
	PPV(%)	74.36 (58/78)	76.32 (58/76)	91.11 (41/45)
	NPV(%)	81.43 (57/70)	74.51 (38/51)	85.88 (73/85)
CEA	SEN(%)	22.54 (16/71)	22.54 (16/71)	11.32 (6/53)
	SPE(%)	93.51 (72/77)	85.71 (48/56)	93.51 (72/77)
	ACCU(%)	59.46 (88/148)	50.39 (64/127)	60.00 (78/130)
	PPV(%)	76.19 (16/21)	66.67 (16/24)	54.55 (6/11)
	NPV(%)	56.69 (72/127)	46.60 (48/103)	60.50 (72/119)
CA19-9	SEN(%)	11.27 (8/71)	11.27 (8/71)	5.66 (3/53)
	SPE(%)	94.81 (73/77)	89.29 (50/56)	94.81 (73/77)
	ACCU(%)	54.73 (81/148)	45.67 (58/127)	58.46 (76/130)
	PPV(%)	66.67 (8/12)	57.14 (8/14)	42.86 (3/7)
	NPV(%)	53.68 (73/136)	44.25 (50/113)	59.35 (73/123)
5’tRF-Val + CEA	SEN(%)	85.92 (61/71)	85.92 (61/71)	79.25 (42/53)
	SPE(%)	70.13 (54/77)	60.71 (34/56)	89.61 (69/77)
	ACCU(%)	77.70 (115/148)	74.80 (95/127)	85.38 (111/130)
	PPV(%)	72.62 (61/84)	73.49 (61/83)	84.00 (42/50)
	NPV(%)	84.38 (54/64)	77.27 (34/44)	86.25 (69/80)
5’tRF-Val + CA19-9	SEN(%)	81.69 (58/71)	81.69 (58/71)	77.36 (41/53)
	SPE(%)	70.13 (54/77)	62.5 (35/56)	90.91 (70/77)
	ACCU(%)	75.68 (112/148)	73.23 (93/127)	85.38 (111/130)
	PPV(%)	71.60 (58/81)	73.41 (58/79)	85.42 (41/48)
	NPV(%)	80.60 (54/67)	72.92 (35/48)	85.37 (70/82)
5’tRF-Val + CEA+CA19-9	SEN(%)	85.92 (61/71)	85.92 (61/71)	79.25 (42/53)
	SPE(%)	66.23 (51/77)	55.36 (31/56)	85.71 (66/77)
	ACCU(%)	75.68 (112/148)	72.44 (92/127)	83.08 (108/130)
	PPV(%)	70.11 (61/87)	70.93 (61/86)	79.25 (42/53)
	NPV(%)	83.61 (51/61)	75.61 (31/41)	85.71 (66/77)

ACCU accuracy; NPV negative predictive value; PPV positive predictive value; SEN sensitivity; SPE specificity; EC esophageal cancer.undefined

We next assessed whether 5’tRF-Val could be used to distinguish early EC lesions from benign lesions, which is a clinically challenging differential diagnosis. 5’tRF-Val maintained reliable detection performance, with an AUC of 0.824 (95% CI, 0.755–0.894; sensitivity 81.69%, specificity 67.86%) (Fig. 4D, E). Notably, CEA and CA19-9 performed poorly in this setting (AUC = 0.586 and 0.571, respectively). The combination of 5’tRF-Val with CEA and CA19-9 increased the AUC to 0.834 (95% CI, 0.765–0.903), with satisfactory sensitivity and specificity (Fig. 4F; Table 2), and the calibration curve demonstrated good fit (Supplementary Fig. 2E–H). Crucially, DCA of this comparison revealed that 5’tRF-Val alone maintained a clear positive net benefit across a variety of clinically relevant threshold probabilities, while the net benefit of CEA and CA19-9 hovered near zero (Supplementary Fig. 3B). This finding indicates that 5’tRF-Val could add genuine decision-support value precisely in the most challenging clinical scenario of distinguishing early malignancy from benign disease, where conventional markers offer little advantage.

Having established its detection value in early EC, we further asked whether 5’tRF-Val could also be used to differentiate between stages and detect advanced EC. Notably, 5’tRF-Val successfully distinguished early from advanced EC and demonstrated excellent detection performance for advanced EC patients versus healthy subjects, with an AUC of 0.923 (95% CI, 0.879–0.968), a sensitivity of 77.36%, and a specificity of 94.81%. In contrast, both CEA and CA19-9 had markedly lower AUCs (0.690 and 0.594, respectively) and sensitivities (11.32% and 5.66%, respectively) for advanced-stage detection (Fig. 4G–I; Table 2). The calibration remained stable (Supplementary Fig. 2I–L). DCA of this scenario further confirmed that the models incorporating 5’tRF-Val yielded the greatest net benefit across a broad range of threshold probabilities, underscoring its potential to support clinical decision-making not only for early detection but also for monitoring disease progression (Supplementary Fig. 3C).

Collectively, these results suggest that serum 5’tRF-Val is a promising noninvasive biomarker for the early detection of EC. It has high accuracy in the detection of early EC and can reliably distinguish early EC from benign lesions. DCA further suggested that the incorporation of 5’tRF-Val into clinical decision-making may result in significant net benefits in the early detection scenario. Taken together, these findings indicate that 5’tRF-Val is a candidate to help fill the current detection gap in early EC.

Preliminary bioinformatic prediction of 5’tRF‑Val target genes and associated pathways

To explore the potential biological mechanisms underlying the clinical association between 5’tRF-Val and EC progression, we conducted a preliminary bioinformatics investigation.We first examined the distribution of 5’tRF-Val in EC cells by nucleocytoplasmic separation. The tsRNA was predominantly enriched in the cytoplasm (Fig. 5A). Then, to identify its potential target genes, we performed a computational prediction using four prediction tools (MiRanda, RNAhybrid, TargetScan, and Pita). The intersection of the predictions yielded 422 candidate target genes (Fig. 5B). Gene Ontology (GO) enrichment analysis of these candidates revealed significant associations with biological processes such as transcriptional regulation, RNA biosynthesis, and macromolecule biosynthesis and with molecular functions related mainly to DNA binding and transcriptional regulation (Fig. 5C). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis further revealed that the candidate targets were notably enriched in several signalling cascades, including the Rap1, p53, and PI3K‑Akt pathways (Fig. 5D).

Fig. 5.

Cytoplasmic enrichment and bioinformatic prediction of 5’tRF-Val target genes and associated pathways. (A) Nucleocytoplasmic separation assay in ECA 109 cells shows that 5’tRF-Val is predominantly localized in the cytoplasm. U6 and 18s served as nuclear and cytoplasmic controls, respectively. Data are presented as mean ± SD of relative expression (2^−ΔΔCt) from three independent experiments. Cytoplasm versus nucleus for 5’tRF-Val (two-way ANOVA with Sidak’s multiple comparison test). (B) Venn diagram illustrating the 422 common candidate target genes predicted by four algorithms (MiRanda, RNAhybrid, TargetScan, and Pita). (C) Gene Ontology (GO) enrichment analysis of the candidate targets, revealing significant associations with transcriptional regulation, RNA biosynthesis, and DNA binding. (D) KEGG pathway analysis indicating enrichment of the candidate targets in Rap1, p53, and PI3K-Akt signaling cascades.

Discussion

The poor prognosis of patients with EC is primarily attributed to the challenges associated with early detection²², highlighting the urgent need for noninvasive screening tools, among which liquid biopsy has emerged as a leading modality. With respect to liquid biopsy, cfDNA-based assays currently dominate clinical applications and have demonstrated notable success. However, their sensitivity can be fundamentally constrained by limited tumour DNA shedding, and their specificity can be confounded by clonal haematopoiesis-derived variants²³. These limitations have motivated the exploration of alternative classes of circulating analytes. Among these, tsRNAs have attracted increasing attention. Unlike cfDNA, which relies on passive release from dying cells, tsRNAs are actively generated through stress-responsive enzymatic cleavage of tRNAs—a mechanism that may be more intimately coupled to early malignant transformation and less dependent on tumour bulk. This inherent stability and cancer-specific dysregulation have already enabled serum tsRNAs to demonstrate diagnostic value in hepatocellular carcinoma and gastric cancer^24,25. However, the potential of specific serum tsRNAs for the early detection of EC remains largely unexplored²⁶. To address this gap, our study adopted a stepwise, data-driven screening strategy that combines database mining with clinical cohort validation. This approach reflects the broader paradigm of systematic biomarker discovery, in which multiomics integration has become a cornerstone for identifying and prioritizing clinically relevant molecules across tumour types²⁷. Through this strategy, we identified serum 5’tRF-Val as a promising candidate biomarker and provided initial evidence for its utility in the noninvasive detection of early EC.

To ensure reliable quantification, we first validated the serum qRT‒PCR assay for 5’tRF-Val. The results of the qRT‒PCR assay demonstrated a wide linear dynamic range (R² = 0.997 across a 10⁵-fold dilution), satisfactory precision (intra- and inter assay CV < 3.5%), and high resilience against preanalytical variables, with serum 5’tRF-Val levels remaining stable after 24 h at room temperature and after up to eight freeze‒thaw cycles. These analytical validations confirm that our measurement method is reproducible and resilient, forming a reliable basis for subsequent case‒control comparisons.

Having established the reliability of serum 5’tRF-Val quantification, we next examined whether its levels correlated with the clinicopathological characteristics of patients with EC. Elevated pre‑operative 5’tRF-Val was associated with features of more advanced disease, such as perineural invasion and lymphatic metastasis, suggesting that its circulating levels increase in parallel with disease progression. Consistent with these findings, the 5’tRF-Val values decreased significantly after radical surgery. This bidirectional change, which increases with disease progression and decreases following tumour removal, reinforces the link between the 5’tRF-Val and tumour burden and highlights its potential utility in monitoring disease progression and therapeutic response.

We next evaluated the detection potential of serum 5’tRF-Val. With respect to early EC, the 5’tRF-Val discriminated patients from healthy subjects, with an AUC of 0.857, and exhibited relatively high sensitivity and specificity. In the more clinically challenging scenario of differentiating early EC from benign lesions, the AUC was 0.824, outperforming both CEA (AUC = 0.586) and CA19‑9 (AUC = 0.571) in our cohort. Because a higher AUC alone does not guarantee clinical usefulness, we performed DCA, which preliminarily suggested a positive net benefit for 5’tRF-Val in this differential detection setting. Importantly, serum 5’tRF-Val levels did not differ between patients with benign lesions and healthy subjects (P = 0.956), indicating that its elevation is malignancy-specific—a property not consistently shared by conventional markers such as CEA and CA19‑9, which can be elevated in benign lesions²⁸. With respect to other emerging analytes, a recently reported serum protein, DSG2, had an AUC of 0.72 for early esophageal squamous cell carcinoma²⁹, and a meta‑analysis of circulating microRNAs reported a pooled AUC of 0.89³⁰; however, these studies did not specifically assess discriminative power in early-stage or benign-malignant settings. Thus, the present data offer initial evidence that 5’tRF‑Val may help address the current gap in noninvasive early EC detection and benign–malignant discrimination. To further explore whether the detection accuracy could be enhanced, we combined 5’tRF-Val with established markers. A model integrating 5’tRF-Val with CEA improved the detection accuracy for early EC, whereas the addition of CA19‑9 contributed only marginally, suggesting that a focused two‑marker panel (5’tRF-Val + CEA) may offer a parsimonious yet effective approach. Taken together, these findings identify serum 5’tRF-Val as a promising candidate biomarker for the early detection of EC.

Building on these clinical findings, we explored the potential underlying biological mechanisms. The predominantly cytoplasmic localization of 5’tRF-Val suggests a role in post‑transcriptional gene regulation. To identify candidate targets, we focused on genes commonly predicted by the four tools. These overlapping targets were significantly enriched in signalling pathways frequently dysregulated in EC, including the Rap1, p53, and PI3K‑Akt cascades^31–40, all of which are well‑established drivers of tumour invasion, metabolic reprogramming, and immune microenvironment remodelling^41–51. This convergence raises the hypothesis that the 5’tRF-Val may be functionally involved in EC progression rather than being a passive bystander. Dedicated functional studies are now needed to test this hypothesis and clarify its biological role in EC pathogenesis.

Several limitations should be considered. (1) The single-centre, retrospective design, the absence of prospective and independent external validation, and the restriction to a single geographic region (Nantong) limit the generalizability of the findings. (2) The sample sizes for benign lesions (n = 56) and postoperative cases (n = 32) were relatively modest, and no prospective screening data from asymptomatic high-risk individuals were collected. (3) Quantification relies on relative qRT-PCR without reporting absolute limits of detection or quantification; technical biases inherent to small-RNA measurement cannot be fully excluded⁵². Future studies should focus on multicentre prospective validation in diverse populations, evaluation in high-risk cohorts, comparison or integration with other liquid biopsy technologies, and functional experiments to elucidate the mechanistic role of 5’tRF-Val in EC.

Conclusions

In conclusion, this study provides the first evidence that serum 5’tRF-Val is significantly elevated in patients with EC, and its significant postoperative decline further supports a direct link to tumour burden. To our knowledge, this is the first report of serum 5’tRF-Val as a noninvasive biomarker specifically for the early detection of EC. As a detection biomarker, it demonstrated high accuracy not only in detecting early EC but also, more importantly, in distinguishing early EC from benign lesions. The results of DCA further suggest its potential to provide clinical net benefit in these detection settings. Collectively, these findings identify serum 5’tRF-Val as a promising noninvasive candidate that may help address the current gap in early EC detection.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

LZ, RJ, and CC conceived and supervised the study. LZ and BZ performed the experiments. LZ, YC, MC and CC collected patients’ serum samples and clinical information. LZ, BZ, and JH carried out the statistical analysis. LZ drafted the manuscript. SJ, RJ and MC reviewed and edited the manuscript. All authors contributed to the work and approved the final version of the manuscript.

Funding

This research was funded by Special Research Fund for Clinical Medicine of Nantong University (2025LZ014).

Data availability

The datasets generated and analyzed during the present study are available from the corresponding author on request. The Sanger sequencing data for 5’tRF-Val have been deposited in GenBank under accession number PZ267099.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

This study received ethical approval from the Affiliated Hospital of Nantong University’s Ethics Committee (Approval No. 2023-L166). Informed consent was obtained from all individual participants or their legal guardians, including for the use of any tissue samples.