Evaluation of Salivary Exosomal Chimeric GOLM1-NAA35 RNA as a Potential Biomarker in Esophageal Carcinoma

This article is featured in Highlights of This Issue, p. 2941

Cancer-specific macromolecules, such as mutated or chimeric nucleic acids and proteins, often find their way into the blood and various bodily fluids, fostering the development of a “liquid biopsy” for cancer (1). Exosomes are small (30–100 nm) membrane-bound vesicles containing RNA and protein cargo, and secreted by eukaryotic cells into circulation (2). The contents of cancer cell–derived exosomes may potentially serve as a source of tumor biomarkers. The saliva is a readily accessible bodily fluid, and salivary exosomal contents have recently been investigated for diagnosis and prognosis (3, 4).

Golgi membrane protein 1 (GOLM1, also known as Golgi phosphoprotein 2 and Golgi membrane protein GP73) is a type II Golgi membrane protein of unclear function, but its involvement in hepatocellular carcinoma and ability to promote cancer metastasis have been amply documented (5). The GOLM1 gene is located on human chromosome 9q21.33, along with its closest neighbor, the Nα-acetyltransferase 35 gene (NAA35, also known as MAK10). NAA35 is an auxiliary subunit of the N-terminal acetyltransferase C complex. We previously screened a set of 32 chimeric RNAs known to be differentially expressed in cancer (6). In esophageal squamous cell carcinoma (ESCC), GOLM1-NAA35 chimeric RNA (G-NchiRNA) is highly expressed relative to matched adjacent nonmalignant esophagus as well as to normal esophagus from subjects without ESCC (7). Mechanistic studies have shown that G-NchiRNA encodes a secreted chimeric protein, and is generated by transcription read-through/splicing or trans-splicing, but not by mutational events that produce a fusion gene (7).

ESCC is the third most common cancer of the gastrointestinal tract and the sixth leading cause of cancer-related death worldwide (8–11). Currently, there are no widely accepted biomarkers for ESCC screening, treatment response, or recurrence. Endoscopic examination/biopsy and imaging studies are widely used diagnostic and monitoring approaches; these approaches are either invasive (endoscopy) or nonsensitive (imaging) as screening modalities. Recently, minimally invasive technologies such as cytosponge or transnasal endoscopy have become available. However, these tests have not been popularized because of cost and discomfort. Hence, specific and convenient biomarkers are badly needed to optimize diagnosis and prognosis of ESCC. Here, we evaluated the diagnostic potential of exosomal G-NchiRNA in tissues and saliva from mice and patients with ESCC. Because saliva collection is noninvasive and painless, we further evaluated the clinical utility of salivary exosomal G-NchiRNA (seG-NchiRNA) as a candidate convenient, robust biomarker assay for therapeutic response, recurrence, and early detection in two prospective ESCC patient cohorts.

Shantou, Guangdong, China is located in a coastal region with a high incidence of ESCC. Anyang, Henan, is located in another high-incidence region in central China. This study involves a retrospective tissue archive, and a prospective observational clinical study including three cohorts from two institutions: (i) Cancer Hospital of Shantou University Medical College (CHSUMC, Shantou, Guangdong, China) and (ii) Anyang Tumor Hospital (ATH, Anyang, Henan, China).

The retrospective tissue archive consisted of 120 pairs of frozen human ESCC specimens and adjacent tissues that were banked from June 1, 2009 to August 31, 2011 at CHSUMC. The demographics and clinicopathologic characteristics of this cohort are summarized in Supplementary Table S1. The prospective study included a discovery cohort, a training cohort, and a validation cohort. In the discovery phase, 10 patients with ESCC who were to receive surgery were recruited at CHSUMC, and 8 healthy volunteers were recruited in Shantou from May 1, 2015 to June 30, 2015. From July 1, 2015 to September 30, 2016, a training cohort of 275 patients with ESCC and 65 healthy volunteers were similarly recruited in Shantou. Concurrently with the recruitment for the training cohort, a validation cohort of 124 patients with ESCC were recruited at ATH and 42 healthy volunteers were recruited in Anyang. From July 1, 2018 to November 10, 2018, another cohort including 10 patients with ESCC who were to receive surgery were recruited at CHSUMC, and 10 healthy volunteers were recruited in Shantou to evaluate the distribution pattern of seG-NchiRNA in salivary fractions (cell pellet, exosomes, and exosome-depleted supernatant). All healthy subjects were approached for participation in the study at public places (e.g., parks, senior activity centers, and shopping areas); they were matched to at least one ESCC case for gender, age, and tobacco usage and were not eligible if they had any history of malignancy, severe oral disease, diabetes, lung disease, renal/hepatic dysfunction, severe immune alterations, and cardiovascular event in the past 6 months. Inclusion and exclusion criteria of training and validation cohorts are shown in Supplementary Fig. S8. The cases were selected on the basis of new pathologic diagnosis of ESCC, without anticancer treatment. The median follow-up time was 15 months (range: 8–22).

The pathologic stage was assessed according to the Union for International Cancer Control (UICC) Tumor-Node-Metastasis (TNM) staging system (7th edition)(12). Patients with stage I/IIa were classified as early-stage patients and patients with stage IIb/III/IV were classified as late-stage ones. At CHSUMC and ATH, 87 and 59 patients, respectively, underwent surgical resection of ESCC, and were further analyzed as the surgical subcohorts. Forty patients underwent chemoradiation at CHSUMC, and were further analyzed as the chemoradiation subcohort. There were only 10 chemoradiation patients at ATH, which were too few to be analyzed as a subcohort (Supplementary Fig. S8B). The chemoradiation regimen was standard-of-care treatment that consisted of paclitaxel (45 mg/m²) and cisplatin (25 mg/m²) on days 1–3 and intensity-modulated conformal radiotherapy (60 Gy in 30 fractions). Responses to chemoradiation were assessed according to Response Evaluation Criteria in Solid Tumors (RECIST). Written informed consents were obtained from all participants in accordance with the principles established by the Helsinki Declaration. The clinical study was approved by the Institutional Ethics Committees and conducted under Institutional Review Board–approved protocols of CHSUMC (IRB serial number: #04-070) and ATH (AZLL022016008161201), which included patients from the respective hospitals and healthy subjects from public places.

All statistical analyses were performed using the SPSS 19.0 statistical software package (SPSS Inc.) and R V.3.33 (The R Project for Statistical Computing, http://www.r-project.org). Summary statistics reporting means, SE, and 95% confidence intervals were stated as appropriate. Statistical methods used included t test, one-way ANOVA, Shapiro–Wilk test, Brown–Forsythe test, Wilcoxon matched-pair signed-rank test, Pearson correlation, logistic regression, ROC analysis, Kaplan–Meier survival analysis, and Cox proportional hazard modeling. Details about statistical analysis are provided in online Supplementary Materials and Methods.

Details for cells, animals, exosomes, and biochemical assays are included in online Supplementary Materials and Methods.

We previously showed that G-NchiRNA, which contains a discernable 5′ splice donor site from GOLM1 fused to a 3′ splice acceptor site from NAA35 at its RNA–RNA junction (capitalized bases in Fig. 1A), was enriched in human ESCC versus nonneoplastic esophageal tissues, suggesting this chimeric RNA as a promising biomarker for ESCC (7). Therefore, we performed a retrospective study to investigate the association between tissue G-NchiRNA levels and ESCC prognosis in 120 patients. An optimal discriminative cut-off value according to Youden index (i.e., relative tissue G-NchiRNA expression = 0.019) was chosen to classify patients into either high expression (n = 57) or low expression (n = 63) group (Supplementary Table S1). High G-NchiRNA expression was associated with poor differentiation and lymph node metastasis (P = 0.001 and P = 0.017, respectively; χ² test; Supplementary Table S1). Moreover, Kaplan–Meier analysis demonstrated that high G-NchiRNA expression was significantly associated with shorter overall survival (OS) than low G-NchiRNA expression (P < 0.001, log-rank test; Fig. 1B). Multivariable Cox regression analysis showed that G-NchiRNA expression in ESCC tissues was a significant independent predictor of OS (HR = 2.49, 95% CI: 1.51–4.11; P < 0.001; Supplementary Table S2).

We hypothesized that G-NchiRNA is exported from ESCC cells by exosomes. To validate this hypothesis, G-NchiRNA was first silenced in two ESCC cell lines (HKESC-2 and TE1) (7) by two specific shRNAs (shG-N #1 and #2; Supplementary Fig. S1A and S1B), while parental genes were not significantly affected (Supplementary Fig. S1A, S1C, and S1D). As shown by qRT-PCR, both shRNAs efficiently knocked down G-NchiRNA levels in two cell lines (P < 0.001 for both, one-way ANOVA with post hoc Dunnett test; Supplementary Fig. S1B); shG-N #2 (more efficient, hereafter designated shG-N) was chosen for subsequent experiments. Next, we isolated exosomes from conditioned media of either ESCC cell line or immortalized human esophageal epithelial cell line (NE2). Isolation of exosomes was confirmed by immunoblotting of exosomal markers (ALIX, CD9, TSG101, and CD63) (13) (Fig. 1C) and transmission electron microscopy (TEM; i.e., spherical membrane-bound particles with diameters between 30 and 100 nm) (14, 15) (Fig. 1D). G-NchiRNA was significantly more abundant in both ESCC cells versus nonneoplastic NE2 cells, and shG-N decreased G-NchiRNA levels versus a nonspecific control shRNA (shCtrl) in both cell lines (P < 0.001 for indicated comparisons; Fig. 1E and F, left). G-NchiRNA was also significantly more abundant in exosomes from conditioned media from both ESCC cell lines than in exosomes from NE2 cell–conditioned medium, and its levels were significantly lowered by shG-N versus shCtrl (P < 0.001 for the indicated comparisons; Fig. 1E and 1F, middle). Of note, ratio of exosomal to cellular G-NchiRNA in NE2 cells [0.290 ± 0.011 (SD)] was comparable with that in HK2-shCtrl cells (0.274 ± 0.019) or that in HK2-shG-N cells (0.322 ± 0.034) and none of intergroup comparisons were statistically significantly different (Fig. 1E, right). Similar results were obtained by similar experiments replacing HK2 cells with TE1 cells (NE2: 0.317 ± 0.018, TE1-shCtrl: 0.290 ± 0.011, TE1-shG-N: 0.272 ± 0.013; Fig. 1F, right). These results confirmed that intracellular G-NchiRNA closely correlated with exosomal G-NchiRNA. Therefore, we reasoned that exosomal G-NchiRNA should reflect intracellular content of G-NchiRNA in ESCC.

Exosomes can be secreted into the circulation by tumors (16), but saliva has also been put forth as a convenient, noninvasive source of biomarkers of systemic diseases and cancer (4, 17). We therefore ascertained whether exosomes containing G-NchiRNA could also be detected in exosomes from saliva in ESCC-bearing animals. HK2-shG-N cells (i.e., HKESC-2 cells in which G-NchiRNA had been stably knocked down by transfection with shG-N) and HK2-shCtrl cells (i.e., HKESC-2 cells stably transfected with a control vector) were xenografted into nude mice. The day on which tumors became palpable was designated as day 0, and saliva was collected at 4-day intervals thereafter (Fig. 2A). Knockdown of G-NchiRNA decreased tumor growth [P < 0.01 for all timepoints beyond 4 days (one-way ANOVA with post hoc Tukey test; Fig. 2B, left)] as well as tumor weight (P < 0.001, t test; Fig. 2B, right), suggesting that G-NchiRNA promoted ESCC progression (Fig. 1B; Supplementary Tables S1 and S2). Isolation of exosomes from mouse saliva was confirmed by immunoblotting and TEM (Fig. 2C and D). G-NchiRNA from saliva, blood, and tumor lysate of each mouse was measured by qRT-PCR. SeG-NchiRNA increased with tumor growth only in HK2-shCtrl mice, but not in HK2-shG-N mice (P < 0.05 for all timepoints; one-way ANOVA with post hoc Tukey test; Fig. 2E). These results demonstrated that seG-NchiRNA increased in parallel with tumor growth, and that xenografts were a specific source of seG-NchiRNA. At the end of each experiment, we collected saliva, blood, and tumor tissues from each mouse; the amount of G-NchiRNA in tumor lysate significantly correlated with its amount in salivary exosomes and in serum exosomes (r = 0.691 and P = 0.003; r = 0.610 and P = 0.008, respectively, Pearson correlation test; Fig. 2F, left and middle). In cells treated with shG-N, the ratio of G-NchiRNA in salivary exosomes relative to tumor tissue [0.805 ± 0.147 (SD)] was comparable with the ratio of serum exosomes relative to tumor tissue (0.649 ± 0.261) and not statistically different from each other; similar results were observed in cells treated with shCtrl (0.547 ± 0.227 vs. 0.429 ± 0.232, respectively; Fig. 2F, right). Taken together, these in vivo data suggest that seG-NchiRNA correlates with tumor extent and can be used to monitor tumor burden.

To translate our findings from cells and mice to patients, an investigation was carried out in a discovery cohort of 10 patients with ESCC and 8 healthy subjects (Fig. 3A–F). Isolated salivary exosomes were first characterized by immunoblotting and TEM (Fig. 3A and B). Nanoparticle tracking analysis further confirmed that human exosomes had an average diameter of 99 nm (Fig. 3C) (18, 19). SeG-NchiRNA levels were 2.1-fold higher in patients with ESCC than in healthy subjects (P = 0.005, t test; Fig. 3D). SeG-NchiRNA levels also significantly correlated with tumoral G-NchiRNA levels (r = 0.881 and P < 0.001, Pearson correlation test; Fig. 3E). We next evaluated the pattern of G-NchiRNA distribution in fractions of saliva. qRT-PCR using RNA from different fractions of saliva showed that the levels of exosomal G-NchiRNA were significantly higher in patients ESCC than healthy controls (P < 0.001, t test; Supplementary Fig. S2A). There was no significant difference in G-NchiRNA in the salivary cell pellet (presumably mostly cells from the buccal mucosa), and no significant difference in exosomes-depleted salivary supernatant. Furthermore, we found that the seG-NchiRNA was resistant to RNase digestion but susceptible to RNase when exosomal membranes were broken by Triton X-100 (i.e., RNase and Triton X-100 simultaneously; P < 0.001, one-way ANOVA with post hoc Dunnett test; Supplementary Fig. S2B), supporting that exosomal G-NchiRNA was stable and suitable for measurement by qRT-PCR. To determine the minimum amount of exosomal RNA required for reliable qPCR, different exo-RNA inputs were used to detect seG-NchiRNA. SeG-NchiRNA was reliably detected using 50 ng or more, but not detectable using only 20 ng of exosomal RNA (Fig. 3F; Supplementary Figs. S3–S5). Four of 105 reactions showed nonspecific amplification in reactions using 50 ng exo-RNA, while only 1 of 105 reactions showed specific amplification in reactions using 20 ng exo-RNA (Supplementary Figs. S3–S5). On the basis of the distribution of the C_t value of seG-NchiRNA among all subjects, the median C_t value for cancer cases was about 30, and the median C_t value for healthy controls was about 33 (Supplementary Fig. S6). Therefore, the C_t values of all subjects were in a reliable range. Circadian variability among different times of day was low (Supplementary Fig. S7A). The ICV was 7.03%–10.85% among 5 patients. SeG-NchiRNA levels at different times of day significantly correlated with tumoral G-NchiRNA levels, with excellent correlation coefficients (r = 0.827 and P = 0.032 for 9 A.M., r = 0.958 and P = 0.003 for 3 P.M., r = 0.983 and P = 0.001 for 9 P.M., Pearson correlation test; Supplementary Fig. S7B). These results suggested that seG-NchiRNA represents a potential cancer biomarker for human ESCC that lacks significant circadian variability, adding to its suitability.

We then evaluated the diagnostic performance of seG-NchiRNA in two cohorts from two independent institutions: (i) CHSUMC (Shantou, Guangdong, China) and (ii) ATH (Anyang, Henan, China; Supplementary Fig. S8). A total of 399 patients were enrolled and 581 saliva samples were collected from 322 eligible patients, including 220 of 275 (80%) eligible patients from CHSUMC as a training cohort and 102 of 124 (82.3%) eligible patients from ATH as a validation cohort (Supplementary Fig. S8). The qRT-PCR data from the patient cohorts with the amplification profile data (C_t values and amplification plots) were provided in the separate Supplementary “zip” archive files available for download. The cut-off value for seG-NchiRNA as a diagnostic test was determined in the training cohort (n = 220), and then its diagnostic performance was confirmed in the validation cohort (n = 102). Among demographic and clinicopathologic characteristics, age, gender, tobacco use, location of tumors, lymph node metastasis, distant metastasis, and stage did not show any significant differences between the two cohorts (Supplementary Table S3). However, tumor depth (T criterion; P = 0.010, χ² test) was significantly different (Supplementary Table S3).

In the training cohort from CHSUMC, seG-NchiRNA in early-stage patients (0.037 ± 0.016, n = 61) was higher in healthy subjects (0.019 ± 0.010, n = 55), but lower in advanced-stage patients (0.071 ± 0.024, n = 159; P < 0.001 for both, one-way ANOVA with post hoc Dunnett test; Fig. 3G, left). On the basis of an ROC curve, which had an AUC of 0.912, an optimal cut-off value (i.e., 0.035) as a binary classifier was chosen by Youden index to discriminate patients (i.e., early- and advanced-stage) from healthy subjects; seG-NchiRNA expression was significantly different between these two groups (P < 0.001, Mann–Whitney U test; Fig. 3G, middle). Using the cut-off value of 0.035, sensitivity for identifying ESCC was 89.1%, while specificity was 89.1% (Table 1). In the validation cohort, seG-NchiRNA levels in early-stage patients (0.048 ± 0.021, n = 38) were significantly different from healthy subjects (0.018 ± 0.012, n = 35) and advanced-stage patients (0.070 ± 0.028, n = 64; P < 0.001 for both, one-way ANOVA with post hoc Dunnett test; Fig. 3G, right). The performance of the 0.035 cut-off value was then tested in the validation cohort from ATH, where sensitivity for identifying ESCC was 85.3% and specificity was 91.4% (Table 1). Diagnostic ORs (DOR) for the training and validation cohorts, respectively, were 66.69 (95% CI: 24.68–85.98) and 61.87 (95% CI: 16.76–99.36). Thus, seG-NchiRNA could effectively identify patients with ESCC from healthy subjects.

To further examine the potentials of seG-NchiRNA for early detection of ESCC, ROC analysis was performed to differentiate early-stage patients (stage I/IIa) and healthy subjects. The ROC had an AUC of 0.790; an optimal cut-off value (i.e., 0.030) was chosen by Youden index to discriminate early-stage patients from healthy subjects; seG-NchiRNA expression was significantly different between early-stage patients and healthy subjects (P < 0.001, Mann–Whitney U test; Supplementary Fig. S9). In the training cohort, according to this screening cut-off value, the sensitivity for identifying early-stage ESCC was 73.8%, while the specificity was 85.5% (Supplementary Table S4). In the validation cohort, the performance of this screening cut-off value was validated, and the sensitivity was 81.6% and the specificity was 85.7% (Supplementary Table S4). DORs for the training and validation cohorts were 16.52 (95% CI: 5.42–36.26) and 26.57 (95% CI: 6.83–60.83), respectively. These analyses revealed the abilities of seG-NchiRNA to effectively distinguish early-stage patients from healthy subjects. Collectively, our data demonstrate that seG-NchiRNA accurately detects early- and advanced-stage ESCC, supporting the strong clinical potential of seG-NchiRNA as a convenient, noninvasive diagnostic test.

Given that seG-NchiRNA level correlated with tumor burden in our mouse model, we examined whether its level reflected dynamic changes in ESCC before versus after surgery. In two subcohorts of patients who underwent surgical resection at CHSUMC (n = 87, Supplementary Table S5) or ATH (n = 59, Supplementary Table S6), salivary specimens were collected 7 days before surgery and on the seventh day after surgery (Supplementary Fig. S10). First, in the CHSUMC surgical subcohort, preoperative seG-NchiRNA levels in tumors with length ≥ 5 cm (20) were significantly higher than those in tumors with length < 5 cm (P = 0.004, χ² test; Supplementary Table S5); a similar result was observed in the ATH surgical subcohort (P = 0.036, χ² test; Supplementary Table S6). Second, a paired comparison of pre- versus postoperative levels showed that seG-NchiRNA levels decreased significantly after surgery in both subcohorts (CHSUMC surgical subcohort: median preoperative level = 0.064, median postoperative level = 0.023, P < 0.001; ATH surgical subcohort: median preoperative levels = 0.044, median postoperative level = 0.017, P < 0.001; Wilcoxon matched-pair signed-rank test for both). Intriguingly, seG-NchiRNA levels increased in 8 of 87 cases and 5 of 59 cases in the CHSUMC and ATH subcohorts, respectively (Supplementary Fig. S10). Among these 13 outlier cases, 3 of the 8 CHSUMC cases with a postoperative increase in seG-NchiRNA were later diagnosed with ESCC recurrences at their 15-month follow-up appointment after surgery. Similarly, 2 of the 5 ATH cases with a postoperative increase in seG-NchiRNA were diagnosed with ESCC recurrence at their 11-month follow-up. Furthermore, Fisher exact test showed that the patients with increased seG-NchiRNA expression after surgery were associated with postsurgical recurrence in both the training (P = 0.001; Supplementary Table S7) and validation cohorts (P = 0.006; Supplementary Table S8). Therefore, we concluded that seG-NchiRNA levels reflect ESCC tumor burden in the majority of cases, with substantial potential implications for postsurgical recurrence monitoring.

In a subcohort of 40 patients who underwent chemoradiation at CHSUMC (Supplementary Table S9), seG-NchiRNA levels were measured longitudinally and correlated with clinical response to chemoradiation according to RECIST criteria. Comparing saliva samples collected pretherapy and at one half-month posttherapy (Fig. 4A, top), 7 of 9 (77.8%) patients without a fall in seG-NchiRNA after therapy were chemoradiation nonresponders [stable disease (SD) or progressive disease (PD)], whereas 80.6% (25/31) of those with a fall in seG-NchiRNA after therapy seG-NchiRNA were responders [complete response (CR) or partial response (PR); P < 0.001, χ² test; Fig. 4A; Supplementary Table S10]. Paired comparison of pre- and postchemoradiation levels showed that seG-NchiRNA significantly decreased after chemoradiation (P = 0.015, Wilcoxon matched-pair signed-rank test; Fig. 4A). Logistic regression analysis revealed that change in seG-NchiRNA levels pre- and postchemoradiation [i.e., ΔseG-NchiRNA = (seG-NchiRNA after chemoradiation) – (seG-NchiRNA before chemoradiation)] was a predictor of clinical response (OR = 17.67, 95% CI: 2.39–130.74; P = 0.005) while adjusting for demographic and clinicopathologic characteristics (Supplementary Table S11). Thus, we conclude that seG-NchiRNA levels are potentially promising for monitoring the therapeutic response of patients with ESCC.

Kaplan–Meier analysis demonstrated that a positive ΔseG-NchiRNA (i.e., a rise in seG-NchiRNA after therapy) was associated with a significantly shorter (P < 0.001, log-rank test; Fig. 4B) PFS than a negative ΔseG-NchiRNA (i.e., a fall in seG-NchiRNA after therapy). Multivariable Cox regression analysis revealed that ΔseG-NchiRNA was an independent predictor of PFS of patients with ESCC undergoing chemoradiation (HR = 3.97; 95% CI: 1.29–12.20; P = 0.016; Supplementary Table S12). Therefore, ΔseG-NchiRNA is predictive of prognosis in patients with ESCC undergoing chemoradiation.

Longitudinal measurement of seG-NchiRNA levels was carried out according to the scheme outlined (Fig. 4C). Fourteen patients with ESCC who had both evaluable longitudinal clinical data and seG-NchiRNA data were analyzed. Seven patients were nonresponders (PD), and 7 others were responders (CR or PR). Dynamics of seG-NchiRNA levels paralleled, or in some cases preceded, the trends of the longest tumor dimension as measured on contrast esophagram (Fig. 4D and E; Supplementary Figs. S11 and S12). Case 01 (Fig. 4D) was a patient who had a remarkable drop in seG-NchiRNA levels two months ahead of noticeable tumor shrinkage, as judged by radiography. Case 07 (Fig. 4E) was a patient who initially did not show disease progression, as examined by X-ray and CT scan until 4 months after initiation of therapy. However, this patient had a significant rise of seG-NchiRNA levels 2 months after initiation of therapy, and continued to increase, suggesting that seG-NchiRNA levels could detect disease progression earlier than radiological studies. Additional examples, including case 16 with CR (Supplementary Fig. S11A), case 04 with PR (Supplementary Fig. S12A), case 17 with PD (Supplementary Fig. S11B), and case 14 with PD (Supplementary Fig. S12B) demonstrated that the dynamics of seG-NchiRNA levels paralleled the clinical response to chemoradiation over time. Wilcoxon matched-pair signed-rank test revealed that the seG-NchiRNA assay might predict PD earlier than radiologic imaging by a median of 35 days in nonresponders for chemoradiation (P = 0.018; Supplementary Table S13). Thus, seG-NchiRNA level offers substantial potential as a biomarker to longitudinally evaluate the clinical effectiveness of chemoradiation in ESCC.

Tumor-secreted exosomes may contain transcriptional, translational, or epigenetic information about the cancer and have emerged as a liquid biopsy substrate for cancer diagnosis or prognosis (1). While circulating tumor cells and nucleic acids are mainly analyzed using blood samples, exosomes are present in a variety of biological fluids that may be more accessible than blood. Unlike miRNAs, which are inherently stable in biological fluids, mRNAs are easily degraded unless they are inside exosomes (21, 22). In support of this point, circulating exosomal mRNAs have been reported as a biomarker for the diagnosis and prognosis of pancreatic cancer (16). In this study, we demonstrated that a chimeric mRNA was present in salivary exosomes of patients with ESCC and in animal saliva, and that levels of this chimeric RNA in salivary exosomes served as a noninvasive biomarker for early and advanced stage ESCC detection, as well as for postoperative surveillance, therapeutic response, and tumor recurrence. To our knowledge, this is the first report of a salivary exosomal chimeric RNA as a disease biomarker.

Fusion transcripts (e.g., BCR-ABL) and their products have been employed as cancer biomarkers for many years (23, 24). With recent advances in next-generation RNA sequencing and bioinformatics, mechanisms producing chimeric RNAs besides gene fusion have been discovered, such as transcriptional read-through and alternative RNA splicing (25–28). Here, we showed that the aberrant splicing–induced G-NchiRNA found in ESCC tissues was associated with malignant progression (Supplementary Table S1) and short overall survival (Fig. 1B). Furthermore, we demonstrated the direct relevance of G-NchiRNA expression to ESCC progression by experimentally inhibiting its expression, which significantly slowed tumor growth in vivo (Fig. 2). Finally, salivary chimeric RNA (seG-NchiRNA) levels accurately reflected the dynamic change in tumor burden or growth in mice and human patients with ESCC.

In this study, >90% of patients with ESCC showed a considerable drop in seG-NchiRNA after surgery in cohorts from two high ESCC incidence regions in China. These findings imply that seG-NchiRNA can serve as a biomarker of tumor burden, with substantial potential in postsurgical recurrence monitoring. In addition, 13 patients from these cohorts exhibited increased levels after surgery (Supplementary Fig. S10): 5 of these 13 developed recurrence soon after follow-up. Whether the remaining 8 will develop recurrence remains to be observed. Thus, seG-NchiRNA levels may even presage clinical detection of postoperative tumor recurrences.

Like most solid tumors, ESCC detection currently involves invasive (e.g., endoscopy/biopsy), minimally invasive (e.g., transnasal endoscopy/biopsy or cytosponge), or expensive clinical procedures (e.g., CT scan with contrast and endoscopy/biopsy) (29, 30). Compared with the existing options, our salivary test is preferred because of economy, comfort, convenience, and patient acceptance. Our study shows that seG-NchiRNA levels permit repeated testing for primary and secondary detection over the entire clinical course of ESCC. Moreover, our results suggest that seG-NchiRNA may predict ESCC recurrence or disease progression earlier than standard radiological methods. We demonstrate that ΔseG-NchiRNA acts as an independent predictor of PFS after chemoradiation. Although our data showed that increased seG-NchiRNA after surgery was associated with postsurgical recurrence and that increased seG-NchiRNA after chemoradiation was associated with disease progression, larger cohorts are needed for confirmation. Thus, seG-NchiRNA levels show promise as a molecular assay for close monitoring of ESCC, with minimal patient discomfort and avoidance of radiation exposure or other complications (e.g., anesthesia complications during endoscopy, intravenous contrast allergy, and intravenous contrast nephropathy). SeG-NchiRNA levels may be useful for evaluating efficacy of cancer therapies in conjunction with standard methods, while simultaneously decreasing the frequency of invasive monitoring procedures or costly tests.

Even after surgery with curative intent, the 5-year survival of ESCC is only 26.2%–49.4% (31). Studies have suggested that survival will be dramatically improved if ESCC is diagnosed at an early stage or chemoradiation response is monitored closely (32, 33). Unlike endoscopy or imaging methods, seG-NchiRNA measurement is based on exosome purification and qRT-PCR, which constitute a noninvasive, low-cost, convenient platform for ESCC screening. Although several serum biomarkers, including squamous cell cancer antigen (SCC-Ag), carcinoembryonic antigen (CEA), and cytokeratin 21–1 fragment (CYFRA21-1) have been proposed for ESCC diagnosis (34, 35), their implementation in clinical settings have been challenged by insufficient sensitivity and limited specificity (34, 35). Recently, serum exosomal miRNA-21 has been investigated in a study involving 51 ESCC cases and 41 benign controls, revealing an association of circulatory exosomal miRNA-21 with ESCC progression (36). On the basis of the training and validation cohorts in our prospective study (322 cases and 90 controls in total), sensitivity of seG-NchiRNA is 85.3%–89.1% and specificity is 89.1%–91.4% for ESCC diagnosis, while DOR is 61.87–66.69. Notably, seG-NchiRNA distinguished early-stage patients from healthy subjects in both training and validation cohorts. As a screening test to distinguish early-stage (stage I/IIa) patients from healthy volunteers, a cut-off value of 0.030 can achieve a sensitivity of 73.8%–81.6% and specificity of 85.5%–85.7% (DOR: 16.52–26.57). These results supported seG-NchiRNA as a potential biomarker for early disease detection. Given a high pretest probability for ESCC in high-incidence regions, multiple-cohort clinical trials are now warranted to determine whether seG-NchiRNA can serve as a cost-effective screening biomarker for ESCC.

One limitation of our study was the relatively small sizes of our patient cohorts, which limited the number of covariates that could be included in multivariate statistical models. In addition, there was no nonsurgical comparison group for the surgical subcohorts to show whether changes in seG-NchiRNA levels were due to debulking the tumor: in this context, it would have been unethical to randomize patients with ESCC to a no-treatment group. Moreover, patients who choose supportive care only are usually end-stage patients with poor performance status, who do not follow up in the oncology or surgery clinics. A large, multicenter, double-blinded (i.e., clinical evaluators blinded to the seG-NchiRNA results, and performers of the seG-NchiRNA assay blinded to the clinical status) prospective study should now be performed to provide a definitive validation of this biomarker in several clinical contexts: for primary cancer detection, for prognosis, and early secondary detection marker of recurrence/progression. Because exosomes are present in bodily fluids other than blood and saliva, that is, most importantly urine, the performance of exosomal G-NchiRNA in spot urine or 24-hour urine samples for ESCC diagnosis and prognosis should be compared with seG-NchiRNA. To further expand the translation of use of this biomarker to other types of cancer, it should be evaluated in other cancer types.

In summary, seG-NchiRNA levels, which do not require a blood draw or other invasive testing, constitute a convenient and reliable biomarker of human ESCC. Our prospective investigations in training and validation cohorts demonstrated that seG-NchiRNA assays can serve as a noninvasive approach for cancer molecular detection even at early stages, for monitoring of tumor burden, and for surveillance of treatment response to tailor therapeutic decisions. Further analysis with regard to clinical outcomes of ESCC will be possible as outcome data for the two clinical cohorts become available in the next few years. Further large independent prospective studies from different regions of China and other cohorts in other countries are needed to validate these results.

S.J. Meltzer is a consultant/advisory board member for TwoXAR, Inc. S.-C.J. Yeung is a consultant/advisory board member for Celgene, and reports receiving commercial research grants from DepMed and Bristol-Myers Squibb ARISTA. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Zhang

Development of methodology: H. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Lin, H. Dong, W. Deng, K. Li, X. Xiong, Y. Guo, F. Zhou, C. Ma, Y. Chen, H. Ren, H. Yang, N. Dai, H. Zhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Lin, H. Dong, W. Deng, W. Lin, L. Ma, S.J. Meltzer, S.-C.J. Yeung, H. Zhang

Writing, review, and/or revision of the manuscript: Y. Lin, H. Dong, W. Deng, S.J. Meltzer, S.-C.J. Yeung, H. Zhang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Lin, H. Dong, W. Deng, W. Lin, K. Li, X. Xiong, Y. Guo, F. Zhou, C. Ma, Y. Chen, H. Yang, H. Zhang

Study supervision: H. Zhang

The authors thank the surgeons, radiotherapists, physicians, and patients who participated in these studies. This work was supported in part by National Natural Science Foundation of China (81572876, 81773087, 81071736 and 30973508 to H. Zhang).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.