Microbial and epidemiological factors in early detection of esophageal squamous cell carcinoma and precancerous lesions

To the Editor: Esophageal cancer (EC) ranks ninth and fifth among the leading cause of global cancer-related morbidity and mortality, respectively.[1] Esophageal squamous cell carcinoma (ESCC) is the predominant histologic subtype of EC in China. Population screening effectively decreases the morbidity and mortality of ESCC, highlighting the necessity of early detection and early diagnosis.[2] However, it is hard to generalize the endoscopy screening from high-risk populations in high-risk areas to the natural populations in larger areas. With the development of high-throughput sequencing technologies, the microbiota is an emerging field to provide new clues about the primary screening of ESCC. Consequently, this study summarized indicative genera associated with ESCC progression in paired esophageal biopsy and swab specimens, and developed risk stratification models of high-risk populations based on microbial factors, epidemiological factors, and actual detection requirements. Based on the national EC screening project in China, 234 participants from Linzhou (Henan province), including 70 healthy participants, 69 participants with esophagitis, 70 participants with low-grade intraepithelial neoplasia (LGIN), 18 participants with high-grade intraepithelial neoplasia (HGIN), and seven participants with ESCC were enrolled in the present study. Trained epidemiological investigators collected their baseline information such as dietary habits, lifestyle, and oral health. All participants were informed and signed informed consent. This study was approved by the Institutional Review Board of the Cancer Hospital of the Chinese Academy of Medical Sciences. (No. 19/176-1960) Paired esophageal swab and biopsy specimens were collected from each participant. Swab specimens were collected using a sterile brush (Puritan, sterile polyester tipped applicators). If there was a lesion, five loops were taken at the lesion or else at the middle esophagus. The brush head was cut into a sterile tube containing 1.5 mL of cell-preserving fluid (Hologic, ThinPrep, PreservCyt Solution, San Diego, USA). The biopsy specimens at the brushing site were macro-dissected with sterile forceps and placed into a sterile tube. All specimens were stored at -80 °C immediately after sampling and transported to the laboratory on dry ice. Bacterial DNA was extracted using PowerSoil DNA Isolation Kit (12888100, Qiagen, Dusseldorf, Germany) and stored in Tris-Ethylene diamine tetra acetic acid buffer solution at -80 °C before other processes. The V4 region of the 16S ribosomal RNA (rRNA) gene was amplified using the universal bacterial primer (515F:5΄- GT GYCAG-CMGCCGCGGTAA-3΄ and 806R:5΄- GGACTACNVGG-GTWTCTAAT-3΄). Polymerase chain reaction (PCR) mixtures contained 1 μL of forward and reverse primer (10 μmol/L), 1 μL of template DNA, 4 μL of deoxy-ribonucleoside triphosphate (dNTPs) (2.5 mmol/L), 5 μL of 10×EasyPfu buffer, 1 μL of EasyPfu DNA polymerase (2.5 U/μL), and 1 μL of double-distilled water in 50 μL total reaction volume. The PCR thermal cycling involved steps as follows: denaturation at 95°C for 5 min, 30 cycles of denaturation at 94°C for 30 s, annealing at 60°C for 30 s, extension at 72°C for 40 s, and a final extension step at 72°C for 4 min. Amplicons were quantified using a Qubit dsDNA HS assay Kit (Thermo Fisher Scientific/Invitrogen catalog No.Q32854, Waltham, USA). Then, the amplicon libraries were pooled at an equal mass of 100 ng per sample and were sequenced on the Illumina MiniSeq platform (Illumina, San Diego, USA). Raw sequences were performed for quality control and to feature table construction using the DADA2 algorithm. Subsequent analyses were based on quantitative insights into microbial ecology (QIIME2; https://view.qiime2.org). Taxonomic assignment was determined using a pre-trained Naive Bayes classifier via the q2-feature-classifier plugin. To avoid sampling depth bias, 1000 reads were randomly selected from each sample to calculate the relative abundance of taxa. To meet the inclusion criteria, the indicative genus had to either demonstrate a statistical difference "between" or "among" normal and esophagitis, LGIN, HGIN, and/or ESCC groups. Four detection requirements were included in this study based on their capacity to distinguish: (A) normal, esophagitis, LGIN, HGIN, and ESCC; (B) normal, esophagitis, LGIN, and HGIN and above (including HGIN and ESCC); (C) normal/esophagitis, LGIN, and HGIN and above groups; (D) between normal/esophagitis and LGIN and above (including LGIN, HGIN, and ESCC) groups. The chi-squared test or Fisher's exact test was used to compare participants' data. We calculated the average relative abundance (ARA%) for each genus in each participant group. Based on the slope in the linear regression equation, we divided genera into increasing group (slope was greater than zero) and decreasing group (slope was less than zero) respectively. The tendency changed from normal, esophagitis, LGIN, and HGIN to ESCC. Then, we compared the ARA% between normal and other groups using the Wilcoxon test and among five individual groups using the Kruskal–Wallis test within and between biopsy and swab specimens, respectively. Multiple testing with the Bonferroni correction was also performed. Finally, we developed risk stratification models of ESCC and precancerous lesions based on multinomial logistic regression and plotted the receiver operator characteristic (ROC) curve with the area under the curve (AUC). Ten-fold cross-validation was used as an internal validation method and the normalized mean square error (NMSE) was calculated. All statistical analyses were performed in R studio (version 1.1.456). Statistical significance was set at P < 0.05. According to the results of the present study, significant differences were observed in age, education level, oral health (gingival bleeding), and dietary habits (drinking water, meat, fried food and scallion, ginger, or garlic) among the normal, esophagitis, LGIN, HGIN, and ESCC groups. In the esophagitis group and above, most participants were over 55 years old (P <0.05). All healthy participants or those with esophagitis never or seldom ate fried food (P <0.05). Alpha diversity of the esophageal microbiota was statistically influenced by education level, numbers of relatives with cancer, tooth loss, gingival bleeding, drinking water, vegetable, spicy food, and salty-tasting food. In the biopsy specimens, 37 indicative genera associated with ESCC progression were identified. The ARA% of Fusobacterium and Bergeyella were significantly different among the participant groups and were significantly higher in the ESCC group than in the normal group. As for precancerous lesions, Neisseria and Oribacterium were significantly different between the precancerous group (LGIN and HGIN group, respectively) and the normal group. Sphingomonas was the only genus with significant differences between the normal and other groups, and among the five groups. In swab specimens, there were more indicative genera (16 genera) with an increasing tendency from normal, esophagitis, LGIN, HGIN to ESCC than esophageal biopsy (11 genera). In addition to significance among five groups, Capnocytophaga, Aggregatibacter, Bergeyella, Streptococcus, and Megasphaera were statistically different between the normal and ESCC group, Fretibacterium, Filifactor, and Solobacterium were notable between the normal and LGIN and above. The multiple testing results suggested that the ARA% of Bergeyella was statistically higher in the ESCC group than in the normal group based on both swab (P <0.01) and biopsy (P <0.05) specimens. Ralstonia was dominant in the normal group. Fourteen identical and indicative genera (including unidentified genera) were observed between biopsy and swab specimens revealing that the ARA% of Haemophilus, Neisseria, Fusobacterium, Aggregatibacter, Bergeyella, and Alysiella increased from normal, esophagitis, LGIN, HGIN to ESCC. In contrast, the ARA% of Streptococcus, Actinomyces, Rikenellaceae RC9 gut group, Oribacterium, Filifactor, and Novosphingobium decreased from normal, esophagitis, LGIN, HGIN to ESCC. Among them, Neisseria, Rikenellaceae RC9 gut group, Oribacterium, Novosphingobium, and Alysiella were significantly different between the esophageal biopsy and swab specimens in at least one participant group. At the beginning of the risk stratification model, we used normal, esophagitis, LGIN, HGIN, and ESCC groups as the dependent variables. Based on the classification of the five groups, the accuracy of the model enrolling seven significant epidemiological factors among normal, esophagitis, LGIN, HGIN, and ESCC groups was 54.26%, which was higher than that of the model (37.23%) enrolling eight significant epidemiological factors relevant to alpha diversity. Regarding microbial factors, eight indicative genera were enrolled in the risk stratification model of the five groups with an accuracy of 41.49%. The accuracy of each model when each indicative genus was combined with seven significant epidemiological factors among five groups was: Filifactor (60.64%), Haemophilus (58.52%), Bergeyella (58.52%), Fusobacterium (58.52%), Streptococcus (58.52%), Actinomyces (57.45%), Selenomonas (56.38%), and Aggregatibacter (54.26%). Based on the results, we flexibly combined the above eight indicative genera and seven significant epidemiological factors with four detection requirements to develop risk stratification models [Supplementary Table 1, https://links.lww.com/CM9/B265]. Comparatively better models for each detection requirement (Supplementary Table 1, https://links.lww.com/CM9/B265) were model_1 (Group A: normal, esophagitis, LGIN, HGIN, and ESCC), model_6 (Group B: normal, esophagitis, LGIN, and HGIN and above), model_7 (Group C: normal/esophagitis, LGIN, and HGIN and above), and model_12 (Group D: normal/esophagitis and LGIN and above). We optimized the model by including two non-significant but clinically epidemiological factors (tooth loss and number of relatives with cancer). By adding tooth loss, the accuracies of model_1, model_6, and model_7 were higher than those of the no-tooth loss enrolled in the risk stratification models [Supplementary Table 1, https://links.lww.com/CM9/B265]. Consequently, the highest accuracy of the risk stratification models for each detection requirement were 68.09% (model_1-adding tooth loss, AUC = 0.87), 68.09% (model_6-adding tooth loss, AUC = 0.84), 73.40% (model_7-adding tooth loss, AUC = 0.88), and 84.04% (model_12-adding tooth loss, AUC = 0.90). Subsequently, we utilized a 10-fold cross-validation as an internal validation method for the four models, and the NMSE values were 0.77 (model_1-adding tooth loss), 0.89 (model_6-adding tooth loss), 0.83 (model_7-adding tooth loss) and 0.92 (model_12-adding tooth loss). Previous studies have revealed that different genera were associated with ESCC.[3] Additionally, epidemiological evidence has indicated that poor oral health is a crucial factor associated with microbiota and ESCC progression,[4] providing clues for early detection via oral specimens combined with poor oral health. Tooth loss and periodontal disease are associated with poor oral health, with tooth loss being a quantitative factor and an easier quality to control in practice.[5] In this study, we also simulated various detection requirements during primary screening for ESCC by developing risk stratification models. "Group A" was the ideal model, although it was limited by the unbalanced sample size. The screening project for ESCC in China revealed that individuals with LGIN or HGIN and above required follow-up and treatment, respectively, which explains the rationality of "Group D". More flexibly, the present study presented "Group B" and "Group C" as compromise methods to meet actual screening costs, medical personnel, and medical equipment. Furthermore, identifying an easily collected microbial specimen for early ESCC detection based on microbial and epidemiological factors was necessary. Dong et al[6] discussed the microbial characteristics associated with esophageal and oral specimens. Microbial studies on dental plaque and saliva have indicated that oral infectious bacteria in oral specimens are associated with EC. Therefore, oral specimens may be preferable alternatives. However, further studies using large sample sizes and participants from multiple centers are required for optimizing and improving the current findings. Overall, this exploratory study has some limitations. First, this study was a natural population-based study based on a national screening project in China. The detection rate of esophagitis, LGIN, HGIN, and ESCC decreased in the natural population,[7] leading to unbalanced but reasonable sample size for each participant group. Second, all participants were from the same county and had a similar lifestyle. Third, data for oral and esophageal specimens obtained from the same participants could not be validated without collecting the oral specimen. In conclusion, the vast human microbiota resources could aid in understanding their role in human health and disease, elucidate the disease etiology, and further explore relevant microbial biomarkers. This exploratory study indicated that it is feasible to combine microbial factors and epidemiological factors to differentiate ESCC and precancerous lesions from normal and esophagitis. However, collecting esophageal specimens is challenging. Therefore, an in-depth understanding of the microbial signatures and correlation between the oral cavity and esophagus is required, as oral specimens are much easier to obtain. The primary screening of ESCC and precancerous lesions using microbial biomarkers should be further promoted. Funding This study was supported by grants from the National Natural Science Foundation of China (No.81974493), the National Science & Technology Fundamental Resources Investigation Program of China (No. 2019FY101101), and the National Key Research and Development Program of Precision Medicine (No. 2016YFC091404). Conflicts of interest None.