A comparison of biopsy and mucosal swab specimens for examining the microbiota of upper gastrointestinal carcinoma

Abstract

Background

There is currently no optimal sampling method for upper gastrointestinal (UGI) tract microbiota. We compared biopsies and mucosal swab specimens for microbial sampling from UGI carcinoma patients.

Methods

A total of 67 esophageal squamous cell carcinoma (ESCC) and 36 gastric cardia adenocarcinoma (GCA) patients were recruited in the Linxian Cancer Hospital, Henan, China. Sterile biopsies and swabs were used to collect paired samples from the resection specimens from carcinoma and adjacent normal tissue. Data from 16S rRNA gene sequencing was processed using QIIME2 to evaluate differences in alpha- and beta-diversity and taxonomic relative abundances between specimen types.

Results

Alpha diversity was not significantly different between swab specimens and biopsies, both for ESCC and GCA. Paired specimens were correlated for both sample types from ESCC (rho>0.6, p<0.001) but not GCA (rho<0.4, p>0.05). For beta diversity, distinct clustering by sampling method was not observed for adjacent normal or tumor tissue from ESCC or GCA. There was a high correlation for weighted UniFrac and Bray-Curtis distance only in ESCC paired specimens (rho>0.6, p=0.001). The ten dominant bacterial genera were similar between swab and biopsy specimens. However, higher levels of Veillonella (p=0.0002) and Streptococcus (p=0.0002) were detected in ESCC adjacent normal and GCA carcinoma swabs, respectively, compared to the biopsies.

Conclusions

Mucosal swab specimens and biopsies could yield similar microbial profiles from ESCC but not GCA. Both can be used to characterize UGI microbiota, one sampling method should be selected for future studies.

Impact

This study provides insight for planning microbiota collections from the UGI tract.

INTRODUCTION

Upper gastrointestinal (UGI) carcinoma significantly contributes to the global cancer burden (1), with gastric and esophageal cancer ranking as the 3th and 6th leading causes of global cancer death, respectively (2). In China, gastric cardia adenocarcinoma (GCA), (cancer occurring in the gastroesophageal junction) and esophageal squamous cell carcinoma (ESCC) are the most common types of UGI carcinoma in some regions (3,4). Both ESCC and GCA have a poor prognosis, with overall 5-year survival rates of less than 30%, which is mainly due to late stage diagnoses and distant metastases (1,5-8). Although ESCC and GCA have similar geographic distributions in China (9) and share some risk factors, the etiology of ESCC and GCA are still poorly understood.

Increasing evidence indicates a key role for bacterial microbiota in carcinogenesis (10,11). There are millions of microorganisms colonizing the gastrointestinal tract, which may interact with genetic and environmental factors to metabolize dietary constituents and xenobiotics, among other functions (12). When the microbial balance is disturbed, the microbiota could alter host cell proliferation and death, manipulate the immune system, and influence host metabolism, giving rise to carcinoma (13). Several studies have reported an important role of the human microbiota in UGI carcinoma (14,15) and found associations between the microbiota and some diseases of the UGI tract, such as esophagitis and Barrett’s esophagus (16,17), and with squamous dysplasia of the esophagus and gastric atrophy (18).

The method of specimen collection may affect the microbial composition obtained from the UGI tract. Tissue biopsy (19,20) and mucosal specimens (21-23) are the primary sampling methods used for tissues. A tissue biopsy appears to currently be the “gold standard” method for sample collection, since most studies used upper endoscopy to collect biopsy specimens from the UGI tract for microbiota detection (19,20,24-26). Some researchers have used esophageal brushes, which might have improved performance, especially on glandular tissue, to collect microbiota samples from the UGI tract (27,28). However, a systematic comparison of mucosal specimens and tissue biopsies has not been conducted.

In order to compare tissue biopsy and mucosal swab specimens for assessment of the UGI microbiota, paired tissue biopsy and mucosal swab specimens were collected from patients diagnosed with incident ESCC or GCA from the Linxian Cancer Hospital, Henan Province, a high risk area for ESCC in China (29), who were scheduled for surgery with organ removal. Without a true gold-standard method, we rigorously compared the two methods to assess differences in order to provide insight for planning future microbiota collections from the UGI tract.

MATERIALS AND METHODS

Study participants

A total of 103 inpatients with newly diagnosed ESCC or GCA were enrolled at the Linxian Cancer Hospital in Henan Province, China, during the period of October 2015 to January 2016. All the patients underwent surgical resection and had a histological diagnosis of primary ESCC or GCA. Ethical approval was obtained from Cancer Hospital, Chinese Academy of Medical Sciences. Written informed consent was obtained from all participants before specimen collection.

Specimen collection

All participants fasted for 12 hours prior to surgical resection. A mucosal specimen and tissue biopsy from both the carcinoma and the adjacent normal mucosa from each participant were collected for a total of 412 specimens from 103 participants immediately after resection. The mucosal specimen from the carcinoma and the adjacent normal mucosa were collected using sterile swabs (Puritan®, sterile polyester tipped applicators) prior to the biopsies to prevent contamination of the mucosal specimens with blood. The adjacent normal specimens were collected from a location more than four centimeters from the carcinoma border. After mucosal sampling, the head of the swabs were broken off into sterile tubes (Cryovial®, cryogenic tube 3.0mL) including 1.5mL of cell preserving fluid (Hologic®, ThinPrep®, PreservCyt Solution). Tissue biopsies measuring 5*5*5mm³ were collected from carcinoma tissue and adjacent normal tissue by using sterile biopsy forceps and placed into a sterile tube without a preservative fluid. All specimens were stored at −80°C immediately after sampling and shipped to the laboratory at Promegene (Shenzhen, China) on dry ice.

DNA extraction, amplification and sequencing

DNA was isolated from UGI tract biopsies and mucosal specimens by using the MOBIO PowerSoil® DNA Isolation Kit 12888-100, and the extracted DNA was stored at −80°C in Tris-EDTA buffer solution prior to additional processing. To control for reagent contamination, we included water as a negative control without DNA template during specimen processing.

The V4 region of the 16S rRNA gene was amplified using the universal bacterial primer set of 515-Forward (5’-GTGYCAGCMGCCGCGGTAA-3’) and 806-Reverse (5’-GGACTACNVGGGTWTCTAAT-3’) (30,31). PCR mixtures contained 1 μL of each forward and reverse primer (10μM), 1 μL of template DNA, 4 μL of dNTPs (2.5mM), 5 μL of 10 × EasyPfu Buffer, 1 μL of Easy Pfu DNA Polymerase (2.5 U/μL), and 1 μL of double distilled water in a 50 μL total reaction volume. The PCR thermal cycling consisted of an initial denaturation step at 95°C for 5 minutes, followed by 30 cycles of denaturation at 94°C for 30 seconds, annealing at 60°C for 30 seconds, and extension at 72°C for 40 seconds, with a final extension step at 72°C for 4 minutes. Amplicons from each specimen were run on an agarose gel to ensure consistent sequencing length. Expected band size for 515-Forward to 806-Reverse is 300-350 base pairs (bp). Amplicon quantification was performed using the Qubit dsDNA HS Assay Kit (Thermo Fisher/Invitrogen cat. no. Q32854, follow manufacturer’s instructions). The negative controls showed no 16S rRNA gene amplification. The amplicon library for high-throughput sequencing was combined at equal concentrations and volumes and subsequently quantified (KAPA Library Quantification Kit KK4824) according to the manufacturer's instructions.

Using Illumina V4 chemistry and paired-end 2×150 bp reads, sequencing was performed on the Illumina MiniSeq platform (Illumina, San Diego, CA, USA). All sequencing was performed in a single MiniSeq run. Original sequence data processing was performed by the Illumina MiniSeq Reporter to remove adapter and primer sequences and then sequence data was exported in the FASTQ format.

Bioinformatics data processing and statistical analysis

Out of the total 412 specimens collected, 396 (96.1%) specimens were successfully amplified and sequenced. The 16S rRNA gene sequencing data of 380 (92.2%) paired specimens was processed using the Quantitative Insights into Microbial Ecology (QIIME2, http://qiime2.org/) platform (32). All raw 16S rRNA gene sequences went through quality control and feature table construction using the DADA2 algorithm (33). Possible phiX reads and chimeric sequences were removed, and the remaining reads were truncated from 0 to 140 base pairs (for both forward and reverse reads) to avoid including sequencing errors at the ends of the reads. Paired-end reads were matched at a maximum mismatch parameter of 6 bases, which indicates a minimum similarity threshold of 90% for the overlap of the forward and reverse reads. The representative sequences (named “features” in QIIME2 nomenclature) were then generated by removing the redundant and low occurrence (n<5 within all samples) sequences. We checked whether using only the forward reads generated different genus-level taxa, but observed generally consistent taxa and relative abundances as the paired-end data. The rarefaction curve was constructed for the Shannon index (Supplemental Fig.1) and data from all samples was rarefied to 1000 reads for both diversity and relative abundance to avoid bias due to differing sampling depths. We included 244 (59.2%) successfully paired specimens (swab and biopsy from the same participant) with at least 1000 reads. The taxonomic assignment of the sequence variants (99% similarity) was assigned using the trained Naive Bayes classifier (trained on the Greengenes 13_8 (34)) through the q2-feature-classifier plugin, and the taxonomic composition at the phylum and genus level were generated based on operational taxonomic units (OTUs) annotation. A total of 6,227,379 16S rRNA sequence reads were generated with identification of 1739 OTUs which could be classified into 350 unique genera. Alpha diversity estimates were calculated including observed OTUs and the Shannon diversity index. The Wilcoxon signed-rank test was used to test the difference between the paired groups and the Spearman correlation analysis for alpha diversity was calculated for the paired specimens. Similarly, both taxonomic (Bray-Curtis distance) and phylogenetic (unweighted and weighted UniFrac distance) beta diversity matrices were calculated in addition to principal coordinates analysis (PCoA) of these matrices. The adonis statistical method was used (R package vegan 2.5.4) to determine differences between the independent beta diversity matrices, and the Mantel test (35) was used to compare the similarity of distance matrices from paired groups (R package vegan 2.5.4). The relative abundances were calculated at the phylum and genus level for each type of specimen and the microbiota was compared at the genus level in the two specimen types from both carcinoma and adjacent normal tissue by using the Wilcoxon signed-rank test. For comparing the top ten genera, Bonferroni correction was used to adjust the significance level due to multiple testing (α=0.05/10=0.005). All analyses were conducted using R (version 3.5.1).

RESULTS

Participants overview

A total of 244 biopsies and mucosal specimens from 103 participants who had a pathological diagnosis of ESCC (N=50, 48.5%) and GCA (N=27, 26.2%) were included. The average age for the ESCC and GCA participants was 62 years and 63 years, respectively. Males represented 69% of the ESCC participants and 93% of the GCA participants (Table 1).

TABLE 1.

Numbers of successfully sequenced esophageal and gastric cardia specimens from both the carcinoma and adjacent normal tissue and participant characteristics.

	Total	ESCC	GCA
Number of participants (N)	103	67	36
Participants with paired specimens (N)	77	50	27
Matched specimens (N)	244*	176	68
Average age (SD)	62 (6.6)	62 (6.7)	63 (6.5)
Male (%)	77	69	92

^*contains 88 paired specimens from ESCC and 34 paired specimens from GCA.

Alpha and beta diversity analysis

For the 50 ESCC participants, the analysis included 47 paired specimens from the carcinoma and 41 from the adjacent normal tissue (Table 1). For both observed OTUs (swab=67, biopsy=66, p=0.46) and the Shannon diversity index (swab=4.4, biopsy=4.3, p=0.41) of the carcinoma tissue, there were no significant differences between mucosal specimens and tissue biopsy specimens. Similar findings were observed in the adjacent normal tissue, such that neither measure of alpha diversity (observed OTUs: swab=61, biopsy=63, p=0.46; Shannon index: swab=3.8, biopsy=3.9, p=0.56) was statistically different (Fig. 1A and B). In addition, there was a high correlation (> 0.60) in alpha diversity for the paired specimens from both carcinoma and adjacent normal tissue (Table 2).

FIGURE 1. Microbial diversity in ESCC specimens.

Alpha diversity of observed OTUs (A) and the Shannon Index (B). Paired specimen principal coordinate analysis plot based on unweighted UniFrac (C), weighted UniFrac (D) and the Bray–Curtis distance (E) from all ESCC specimens. Samples from the same participant and tissue location are connected with a line.

TABLE 2.

Spearman correlation test for alpha diversity of paired specimens from esophageal and gastric cardia carcinoma.

Sample Type	Observed OTUs		Shannon Index
	rho	p-value	rho	p-value
ESCC Carcinoma	0.61	5.2E-06	0.59	1.6E-05
ESCC Adjacent Normal	0.66	2.8E-06	0.84	2.2E-16
GCA Carcinoma	0.23	0.28	0.20	0.33
GCA Adjacent Normal	0.12	0.76	0.40	0.29

For the 27 GCA participants, the analysis included 25 paired specimens from the carcinoma and 9 from the adjacent normal tissue (Table 1). Similar to ESCC, observed OTUs (carcinoma: swab=62, biopsy=66, p=0.72; adjacent normal: swab=43, biopsy=41, p=0.95) and the Shannon diversity index (carcinoma: swab=4.2, biopsy=4.0, p=0.98; adjacent normal: swab=2.4, biopsy=2.0, p=0.36) showed no statistical differences in alpha diversity between the swabs and the biopsies (Fig. 2A and B). However, we did not find a significant Spearman correlation between the paired specimens (Table 2).

FIGURE 2. Microbial diversity in GCA specimens.

Alpha diversity of observed OTUs (A) and the Shannon Index (B). Paired specimen principal coordinate analysis plot based on unweighted UniFrac (C), weighted UniFrac (D) and the Bray–Curtis distance (E) from all GCA specimens. Samples from the same participant and tissue location are connected with a line.

For ESCC, there was no distinct clustering in beta diversity by sample collection type, except for unweighted UniFrac, from the carcinoma (unweighted UniFrac: R²=0.033, p=0.001, weighted UniFrac: R²=0.010, p=0.444, Bray-Curtis: R²=0.005, p=0.998, adonis test) and adjacent normal tissue (unweighted UniFrac: R²=0.094, p=0.001, weighted UniFrac: R²=0.014, p=0.308, Bray-Curtis: R²=0.008, p=0.910, adonis test) (Fig. 1C, D, and E). For GCA, the same trend was observed for both the carcinoma tissue (unweighted UniFrac: R²=0.106, p=0.001, weighted UniFrac: R²=0.027, p=0.216, Bray-Curtis: R²=0.025, p=0.174, adonis test) and the adjacent normal tissue (unweighted UniFrac: R²=0.146, p=0.024, weighted UniFrac: R²=0.018, p=0.762, Bray-Curtis: R²=0.021, p=0.872, adonis test) (Fig. 2C, D, and E). The Mantel test showed a high correlation for both weighted UniFrac (> 0.40) and Bray-Curtis (> 0.50), particularly for the ESCC specimens (> 0.60). The Mantel statistic rho of unweighted UniFrac was low (< 0.40) for both specimen types from both ESCC and GCA participants (Table 3).

TABLE 3.

Mantel test for beta diversity of paired specimens from esophageal and gastric cardia carcinoma.

Sample Type	Unweighted UniFrac		Weighted UniFrac		Bray-Curtis
	rho	p-value	rho	p-value	rho	p-value
ESCC Carcinoma	0.35	0.001	0.61	0.001	0.67	0.001
ESCC Adjacent Normal	0.25	0.001	0.73	0.001	0.67	0.001
GCA Carcinoma	0.29	0.008	0.42	0.002	0.51	0.003
GCA Adjacent Normal	0.26	0.128	0.55	0.017	0.61	0.011

Taxonomic relative abundance

For the ESCC participants, at the phylum level, all specimens contained Bacteroidetes, Firmicutes, Proteobacteria, Fusobacteria and Actinobacteria. Similarly, all specimens from the carcinoma and adjacent normal tissue had the same top ten genera (Supplemental Fig. 2 and 3). In pairwise comparisons, there were three genera enriched in the swab sample compared to the biopsy specimen in the carcinoma: Prevotella, Streptococcus and Veillonella. The mean relative abundances of these three genera were 35.2%, 12.6% and 11.4%, respectively, in the swab specimen and 32.4%, 10.0% and 8.9%, respectively, in the biopsy specimen. These three genera were also observed to be different in adjacent normal swabs and biopsies, with the relative abundances of 30.7%, 24.6% and 11.5% in swabs and 27.1%, 26.9% and 8.4% in biopsies, respectively. After adjusting for multiple comparisons for testing the top ten genera, only the difference in Veillonella detected in the adjacent normal specimens was statistically significant between swabs and biopsies (p=0.0002) (Fig. 3A).

FIGURE 3. Comparison of the relative abundance of microbial taxa between the two specimen collection types for ESCC and GCA.

(A) Mean relative abundance of the top ten prevalent genera in swab and biopsy specimens of ESCC. (B) Mean relative abundance of the top ten prevalent genera in swab and biopsy specimens of GCA. Statistical differences were calculated between two specimen types using the Wilcoxon signed-rank test, and error bars are the standard error of the mean. An unclassified genus similar to Prevotella is indicated as [Prevotella]. An unclassified bacterium at the genus level was shown as f_Enterobacteriaceae;.

For the GCA participants, all specimens contained Bacteroidetes, Firmicutes, Proteobacteria, Fusobacteria and Actinobacteria and had the same top ten genera (Supplemental Fig. 4 and 5). For the carcinoma specimens, four bacteria at the genus level Prevotella, Streptococcus, Veillonella and Helicobacter varied between the sample types with a mean relative abundance of 25.6%, 18.3%, 12.4% and 4.0% in the swabs, respectively, and 23.2%, 8.9%, 8.6% and 13.4% in the biopsies, respectively. For the adjacent normal specimens, these four genera were observed with the relative abundances of 19.3%, 4.9%, 7.3%, 48.9%, respectively, in the swabs and 15.6%, 2.8%, 5.1%, 60.1%, respectively, in the biopsies. After adjusting for multiple comparisons for testing the top ten genera, only Streptococcus detected in the carcinoma was significantly greater in the carcinoma swab samples compared to the biopsy samples (p=0.0002) (Fig. 3B).

DISCUSSION

Using the paired tissue biopsies and mucosal swab specimens from UGI carcinomas and adjacent normal tissues, overall, alpha diversity was no significantly different in the swab specimens compared to the biopsy specimens from both esophageal squamous tissue and gastric cardia glandular tissue. Paired specimens were correlated for both sample types from ESCC but not GCA. Additionally, the community structure, as assessed in the beta diversity analyses, suggested that the swab and biopsy specimens had generally similar bacterial communities, except for unweighted UniFrac. The relative abundances of the top phyla and genera from the two types of specimens were generally similar for ESCC or GCA tissues except that the relative abundance of Veillonella was observed to be decreased in ESCC adjacent normal biopsies and Streptococcus was observed to be decreased in GCA carcinoma biopsies compared to the swab specimens.

Our results generally showed a similar alpha diversity for the swab specimens compared to the biopsies, which is not consistent with other studies (27,36). Gall and colleagues (27) collected brush mucosal specimens from the esophagus and found higher species diversity in mucosal specimens compared to biopsies as measured by quadratic entropy analysis, but they did not compare other diversity metrics between sample types. Watt and colleagues (36) observed no statistical differences in alpha diversity as measured by the Shannon and inverse Simpson’s indices between colonic lavage mucosal specimens and biopsies from the sigmoid colon but Watt and colleagues (36) did observe higher numbers of OTUs in colonic lavage than those in biopsy, however, esophageal sampling may differ from that in the colon. Our calculation of alpha diversity by QIIME2 is different from other studies, which could influence the similarity to some extent. Moreover, the high correlation indicated good consistency between ESCC paired samples. Likely because of our limited sample size, we did not see the same correlation trend for GCA. Our beta diversity analysis showed similar community structures by specimen collection types from carcinoma and adjacent normal tissue for the beta diversity measures which take relative abundance into account (i.e., weighted UniFrac and Bray-Curtis). Elliott and colleagues found similar microbial communities in biopsies and mucosal brushes from esophageal adenocarcinoma (23), similar to our findings. In addition, the weighted UniFrac and Bray-Curtis were correlated between two sample types. The differences in specimen types for unweighted UniFrac suggests that these two methods may detect different rare taxa, since unweighted UniFrac takes into account the presence or absence of taxa and not their evenness.

The microbial communities in mucosal specimens from UGI carcinoma participants in our study were mainly composed of the phyla Bacteroidetes, Firmicutes and Proteobacteria. The dominant genera of Prevotella and Veillonella in the UGI carcinomas were observed in our study, which were also detected in participants with precancerous lesions for esophageal adenocarcinoma of esophagitis or Barrett’s esophagus (25). The composition of the microbiota in mucosal specimens was similar to that in biopsies, except that there was a higher relative abundance of Veillonella in swab specimens from ESCC adjacent normal and Streptococcus in swab specimens from GCA carcinoma compared to the biopsy. We stratified the analysis by ESCC and GCA tissue since the histopathological structure of GCA is columnar epithelium (37), which is distinct from the normal stratified squamous epithelium of the esophagus (38), and we hypothesized that the sampling method could vary by tissue type. In particular, we hypothesized that swabs might be a better sampling method for the esophagus and biopsy might be a better method for the gastric cardia, and we did observe similar microbial diversity trends for both esophageal and gastric cardia tissue. However, the high correlation was only detected in ESCC paired specimens, we did not observe the same trend in GCA which could be influenced by the limited sample size.

Few studies have evaluated the microbiota from the UGI tract likely due to difficulty in sampling. Unlike the relatively easy sampling of oral and fecal specimens, human studies of the UGI tract rely principally on specimens collected through mucosal biopsies using endoscopy, which is invasive and requires a skilled endoscopist. Several studies used the string test (22,39) or a Cytological device (18,23) as a less invasive method to collect the esophageal microbiota and have been observed to collect high levels of microbial DNA. But this type of sampling includes microbial communities from the entire length of the esophagus and the oral cavity (22,23). For that reason, we compared a potentially lower contamination approach to collect the microbiota of the UGI tract and found that the mucosal swab and biopsy had relatively similar microbial communities. However, we did not evaluate these methods using endoscopic sampling and therefore, future studies should evaluate any contamination by oral microbial communities or other endoscope-related contamination. Recent studies have shown that specimens with relatively small amounts of microbial biomass can produce inaccurate results on sequencing, due in part to DNA contamination of the reagents used (40,41), but our reagent controls did not show any contamination.

There are limitations to this study. We had a limited sample size for all of the comparisons that were included in this study, so we may be underpowered to detect significant differences between the sampling methods. In addition, only individuals with cancer were included, and although similar findings were seen in the adjacent normal tissue, it is unclear if the sampling methods would perform as well in non-diseased individuals. We also did not extract DNA from the cell preserving fluid to examine contamination. Since this solution was used for only the mucosal swab specimens, some of the differences between the mucosal swab and the biopsy may be due to the cell preserving fluid. However, we did not find any non-human associated bacterial taxa with a relative abundance of more than 1% in the samples. Also, the majority of specimens that had to be dropped due to insufficient read counts were from swabs of the GCA adjacent normal tissue, which may indicate that swabs were not appropriate for GCA sampling. Finally, we did not test mucosal brushes and instead used a swab to simulate mucosal sampling. Future studies may wish to evaluate mucosal brushes and swabs to evaluate comparability.

In conclusion, mucosal specimens from ESCC participants appear to yield similar microbial profiles as tissue biopsies. Since a strong correlation between the two types of specimens in GCA was not observed, it is even more important to use a consistent collection method in any study of the gastric cardia. The collection method for a new study should be determined based on feasibility and sampling invasiveness, but all study comparisons should be made within one sample type due to the differences between the two collection methods. Additional studies of samples collected during an upper endoscopy are needed to confirm our findings.