Esophageal microbiome correlates with post-esophagectomy anastomotic leak in cancer patients

Abstract

Background

Despite continuous improvement in long-term survival after esophagectomy, potential serious post-operative complications, such as anastomotic leaks (ALs), still occur. Several risk factors for ALs have been proposed, including environmental factors. Our main objective was to examine the correlation of esophageal tumor microbiome composition and functional profile with ALs. Additionally, we analyzed the microbiome of esophageal tumors and their potential correlation with clinical features of the patients.

Materials and methods

Surgical specimens of esophageal tumors and adjacent normal tissues were collected from consecutive patients who underwent an esophagectomy. Formalin-fixed paraffin-embedded (FFPE) tissue samples were processed using 16S ribosomal DNA multiple fragments amplicon sequencing to characterize bacterial microbiome composition. The tumor and normal tissue microbiome and bacterial functional profile were analyzed based on the clinical outcome of ALs.

Results

Out of 60 patients who met the inclusion criteria, 52 (86.7%) patients had both normal adjacent tissue (NAT) and tumor (T) FFPE samples included with sufficient bacterial DNA extracted for analysis. A total of 28% of participants had esophageal ALs. Proportion tests [P < 0.05, false discovery rate (FDR) < 0.25] revealed operational taxonomic units (OTUs) significantly present in T samples as opposed to NAT samples, as well as significantly present OTUs in patients with AL as opposed to patients without AL complication.

Conclusions

In this study, we provide a profile of the understudied esophageal microbiome and its connection to ALs. Our results can provide potential clues on how to avoid ALs by considering a patient’s personal microbiome when providing perioperative care.

Highlights

• •We identified bacteria with a higher prevalence in esophageal T tissue compared with NAT.
• •Specific bacterial OTUs are significantly associated with the incidence of post-esophagectomy ALs.
• •We characterized a connection between predicted esophageal microbiome functional pathways and AL incidence.

Introduction

Esophageal cancer represents a global medical challenge due to increased incidence, especially in younger populations,¹ and an adverse prognosis. Esophageal cancer is cited as the sixth most common cause of cancer-related deaths worldwide and has two main subtypes: esophageal squamous-cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC).² Esophageal cancer is a multifactorial disease associated with environmental factors (e.g. smoking and dietary habits), other comorbidities (obesity and gastric reflux), and a precursor state for EAC termed Barrett esophagus.^3,4 The role of environmental and host factors in the pathogenesis of esophageal cancer remains unclear. Here, we focus on the microbiome and its connection to esophageal cancer and the development of anastomotic leaks (ALs) after an esophagectomy.

Standard treatment of locally advanced esophageal cancer consists of neoadjuvant chemoradiotherapy⁵ or perioperative chemotherapy,⁶ followed by surgery. Despite advances in diagnosis and treatment strategies for esophageal cancer in the past decade, disease outcomes have remained poor. Post-operative complications remain an Achilles’ heel of surgical treatment,⁷ which could benefit from the inclusion of personalized therapy.8, 9, 10 ALs are considered one of the most prominent and severe complications, with incidence ranging between 10% and 35% across studies.8, 9, 10 Multiple risk factors for ALs have been proposed, including host factors (e.g. comorbidities and prior medications), tumor location, surgical technique, and preoperative treatment.^9,11 Given the complex mechanism of host and tumor factors in the pathogenesis of esophageal ALs, we aimed to explore this phenomenon’s underlying biology and investigate the microbiome’s role.

The human microbiome composition has been associated with carcinogenesis, cancer progression, and response to treatment, including immunotherapy and fecal microbial transplants,12, 13, 14, 15 and is included as one of the hallmarks of cancer.¹⁶ Recent studies demonstrated that the gut microbiota is altered in cancer patients, and the tumor itself harbors a unique microbiome composition.17, 18, 19 Intriguingly, the intratumoral microbiota was shown to be correlated with tumor type, genetic milieu, phenotypes, and patients’ clinical data (e.g. smoking history and response to immunotherapy).^20,21 Furthermore, emerging evidence alludes to specific functional attributes of the intratumoral microbiota, such as exacerbating chemotherapy resistance22, 23, 24, 25 and eliciting tumoral immune reactivity.²⁶ The role of the microbiome should be considered in cancer research, specifically in the trajectory of cancer treatment and surgical management of cancer patients, which may majorly impact clinical outcomes.

Analysis of the esophageal tumor microbiome remains relatively undetermined. Yamamura et al.²⁷ demonstrated that elevated levels of Fusobacterium nucleatum in esophageal tumors are associated with an advanced tumor stage and poor survival and could be an indicator for patients at risk for poorer prognosis.²⁷ It has also been shown that a higher burden of F. nucleatum correlates with a poor response to neoadjuvant treatment in ESCC patients.^22,27 Similar findings have been described in other tumor types, such as colorectal cancer (CRC) and breast cancer.28, 29, 30 A common complication of surgical esophageal cancer treatment is an AL, and it has been shown that ALs in CRC are correlated with distinct microbiota profiles.^20,31

Additionally, it was shown that post-esophagectomy AL incidence is connected to higher variance in oral and gastric microbiome composition.³² Given these findings and the potential environmental effects on ESCC and EAC, we studied the microbiome of esophageal tumors and their potential correlation with post-esophagectomy ALs. To this end, we characterized the microbiome composition of tumor (T) and normal adjacent tissue (NAT) of formalin-fixed paraffin-embedded (FFPE) esophageal cancer samples retrieved from patients who underwent an esophagectomy. We hypothesized that several classes of microbiota would be represented in the tumor tissue based on previous studies.^17,33 The previously described 5R analysis¹⁷ was applied, enabling the identification of microbial operational taxonomic units (OTUs) and providing a relatively high-resolution phylogenetic analysis. Our analysis yielded a correlation between specific bacterial OTUs and post-operative ALs.

Materials and methods

Ethics statement

Informed consent was obtained from all patients who participated in this study. This retrospective study was held to a high level of ethical standards and was approved by the Rabin Medical Center Institutional Review Board (IRB #0803-19RMC) and Rambam Medical Center (IRB #0672-19RMB).

Patients

Consecutive patients with esophageal cancer who underwent surgical resection between 2017 and 2020 were identified from a prospectively maintained surgical database at the Rabin Medical Center and Rambam Medical Center in Israel. Patients for whom FFPE surgical samples were available and of sufficient quality for analysis were included in the study. An AL was defined according to the Esophagectomy Complications Consensus Group (ECCG) guidelines,³⁴ and the severity of the leak was graded by Clavien–Dindo grade. Clinical staging and tumor location were reported according to the eighth American Joint Committee on Cancer TNM (tumor–node–metastasis) classification. All patients included in the study were divided into ‘events’ (i.e. AL following surgery) and ‘control’ (i.e. no leak). Clinical data were obtained from available electronic records, including patients’ demographics, tumor staging, neoadjuvant treatment, other perioperative complications, pathological characteristics, and oncological outcomes. Patients were excluded if they underwent definitive chemoradiation or had metastatic spread during laparoscopy. Patients with a complete response were excluded due to the lack of tumor tissue in the specimen for analysis. Patient confidentiality was maintained throughout data collection and analysis by replacing protected personally identifiable information with research identification codes. Two FFPE samples were retrieved from each patient—T and NAT, both integral parts of the surgical specimen.

DNA extraction

DNA was extracted using the DNeasy Blood & Tissue Kit (#69504, Qiagen, Hilden, Germany) according to the manufacturer’s protocol with the following modifications: paraffin slices (3-5, 10-μm thick) were mixed with ATL buffer (#69504, 170 μl, Qiagen) and heated for 10 min at 90°C, followed by incubation at room temperature for 5 min. Samples were transferred to Bead Tubes (#A29158, Invitrogen, Carlsbad, CA) containing ATL buffer (#69504, 100 μl, Qiagen), DX buffer (#19088, 0.1 μl, Qiagen, Germany), and proteinase K (#69504, 30 μl, Qiagen). Samples were incubated at 56°C for 1 h followed by incubation at 90°C for 45 min. After another incubation at room temperature for 5 min, the samples were homogenized using the Bioprep-24 Homogenizer (Allsheng, Hangzhou, China) for five cycles of 30 s at 7 m/s with 10-s intervals. Samples were centrifuged for 1 min at 10 000 g. AL buffer (#69504, 200 μl, Qiagen) and ethanol (#1085430250, 200 μl, Merck, Jerusalem, Israel) were added to samples and mixed. Next, samples (liquid and most beads) were transferred to a DNeasy mini spin column and centrifuged for 1 min at 12 000 g. Columns were washed with AW1 buffer (#69504, 500 μl, Qiagen), centrifuged for 1 min at 12 000 g, and then washed with AW2 buffer (#69504, 500 μl, Qiagen), followed by centrifugation for 3 min at 16 000 g. Next, columns were centrifuged 1 min at 16 000 g. Finally, DNA was eluted with pre-warmed (at 37°C) Molecular Biology Grade Water (#01-869-1A, Sartorius, Beit Haemek, Israel) and incubated for 10 min at 37°C, and then centrifuged in a new sterile collecting tube for 1 min at 16 000 g. Margins of paraffin were included as non-tissue controls; the number of controls amounted to 10% of all samples.

16S rRNA amplicon sequencing

Amplicon sequencing of five short regions along the 16S rRNA (5R)

The quality of DNA extracted from the FFPE samples was evaluated using a NanoDrop spectrophotometer (Thermo Scientific, Waltham, MA). We set our minimum DNA concentration at 10 ng/μl (except in non-tissue controls which could be lower) and set the A260/A280 ratio to be between 1.5 and 2.1. Amplicon sequencing for five different regions of the 16S ribosomal RNA (rRNA) gene was done as previously described.^17,35 For the first amplification step, five sets of primers were used to amplify the FFPE extracted DNA (1 ng from each sample, to ensure bias analysis), for 30 cycles, using Phusion polymerase (#F565, Thermo Scientific). The amplicon products of the first step were further amplified for six cycles with primers containing barcodes and Illumina adapters (same polymerase as in step 1). All amplification products were pooled and purified with the QIAQUICK PCR purification kit (#28106, Qiagen) and Agencourt AMPure XP beads (#A63881, Beckman Coulter, Brea, CA).

5R 16S data analysis

Taxonomic classification was done using the Short MUltiple Regions Framework (SMURF) software package using the 5R primers.^17,35 Relative abundances were converted to read counts by multiplying by the total number of reads. Samples were normalized by applying a factor of the ratio of the average number of reads/samples in each library to the overall reads average across all samples. Samples with <1000 normalized reads were removed. To assess the presence or absence of species within samples, we floored to zero species with <10⁻⁴ relative abundances, while assigning 1 to species with >10⁻⁴ relative abundances. The basic database used was the Greengenes (GG) 16S rRNA gene database (May 2013 version), as previously applied.³⁵ All species detected in the controls from DNA extraction, PCR amplification, and paraffin were removed.

The α-diversity within samples was measured by counting the observed species in each sample. The β-diversity between groups was measured using the principal coordinate analysis (PCoA) based on Bray–Curtis distances between samples. Comparisons of taxa and pathways between groups were assessed using the two-proportion z-test, comparing the proportion of samples positive for a given bacterial taxon in a group with its proportion in a second group. The two-tailed P value is reported for each taxon, and its inverse log version is used as the y-axis of the volcano plot figures. The comparison P values were corrected for false discovery rate (FDR) by applying the Benjamini–Hochberg correction for multiple hypotheses testing.

PICRUSt2 analysis

Our 5R 16S sequencing results were filtered, and all the OTUs found in the controls were removed. Subsequently, PICRUSt2³⁶ was used to generate MetaCyc22³⁷ probable pathways present in the samples of patients with and without ALs. To calculate the prevalence of each pathway, we compared the pathway abundance to the (total mean)/4 and (sample mean)/4 and if the pathway was greater than either value it was assigned 1 and if not, 0. All pathways present or absent in all samples were removed to include only varied pathway. A two-proportion z-test was carried out on all samples comparing MetaCyc22 pathway prevalence in tissues from patients with and without ALs.

Data availability

Sequencing results have been deposited in the NCBI sequence read archive (SRA) under the BioProject accession number: PRJNA979957 (https://dataview.ncbi.nlm.nih.gov/object/PRJNA979957?reviewer=51hi234imsiptffj4sctu7hsej)

Results

Patient population characteristics

Overall, 60 patients met the inclusion criteria and were included in the study. Fifty-two (86.7%) patients had T and NAT FFPE samples included, and eight patients had a single sample, either T or NAT, due to insufficient bacterial DNA production for analysis. Sixty-one percent of the patients were males, the median age was 67 years, the median weight loss before surgery was 8 kg, the median body mass index (BMI) was 23.8 kg/m², and 55% were cigarette smokers. Most patients had EAC (75%), and the most common tumor location was within the criteria of Siewert 2 (46.7%) (Table 1). The two most common postsurgical complications were ALs (28%) and pulmonary complications (26.7%) (Table 1).

Table 1.

Demographic, preoperative, intraoperative, and perioperative characteristics

Variables		Cohort (n = 60)
Sex	Female	23 (38.3)
	Male	37 (61.7)
Age, median (years)		67
BMI, median (kg/m2)		23.8
Weight loss, median (kg)		8
ASA	1	8 (13.3)
	2	35 (58.3)
	3	17 (28.3)
	4	0
Comorbidity	DM	11 (18.3)
	HTN	37 (61.7)
	IHD	15 (25)
	COPD	7 (11.7)
Habits	Smoking	33 (55)
	Alcohol	2 (3.3)
Histology	SCC	15 (25)
	AC	45 (75)
Location	Mid	10 (16.7)
	Siewert 1	22 (36.7)
	Siewert 2	28 (46.7)
cTNM	Tis	1 (1.7)
	I	7 (11.7)
	II	9 (15)
	III	41 (68.3)
	IV	2 (3.3)
Type of surgery	McKeown (MI)	39 (65)
	Ivor Lewis (hybrid)	21 (35)
Post-operative complications	Anastomotic leak	17 (28.3)
	Pulmonary	16 (26.7)
	Cardiovascular	11 (18.3)
	Post-operative bleeding	1 (1.7)
	Thromboembolic event	1 (1.7)
	Chylothorax	7 (11.7)
	Infective/sepsis	8 (13.3)
	Diaphragmatic hernia	1 (1.7)
	Vocal cord injury	1 (1.7)
Reoperation		11 (18.3)
Mortality		1 (1.7)
Length of stay, median (days)		15

All values are n (%) unless otherwise specified.

AC, adenocarcinoma; ASA, American Society of Anesthesiology; BMI, body mass index; cTNM, clinical tumor–node–metastasis; DM, diabetes mellitus; HTN, hypertension; IHD, ischemic heart disease; MI, minimally invasive; SCC, squamous-cell carcinoma.

Microbiome analysis

We studied the intratumoral microbiome of esophageal cancer patients by analyzing FFPE samples from the T and NAT and comparing them (Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2025.100172). We applied the previously described 5R analysis, sequencing five short regions along the 16S rRNA gene and combining the results into a coherent solution using the SMURF algorithm.³⁵ Using multiple regions of the 16S rRNA gene allowed us to characterize the microbiome compositions at the species level. To assess differences in microbial richness in T compared with NAT, we first carried out α-diversity analysis on all the samples (Figure 1A), which revealed no significant differences. Next, to assess microbial similarities and dissimilarities between the samples, we carried out β-diversity on all NAT and T tissues individually (Figure 1B). The Bray–Curtis PCoA of microbiome β-diversity demonstrated no significant differences. Microbial phylogeny analysis on the family level (Figure 1C) showed no significant differences between T and NAT. However, proportion tests (P < 0.05, FDR < 0.25) on the OTU levels, which classify groups of related bacteria, did indicate specific OTUs significantly present in T and NAT samples (Figure 1D). The genera Rickettsia (P = 0.004) and Sphingomonas (P = 0.02) and the latter’s species Sphingomonas yabuuchiae (P = 0.01) were significantly higher in NAT compared with T tissue (Figure 1D). The genus Sphingomonas belongs to the Sphingomonadaceae family, which is represented in the 30 most prevalent families in tumor and normal tissue (Figure 1C). In T tissue, the genus Ralstonia (P = 0.006) was significantly more represented than in NAT, as well as the family Rhizobiaceae (P = 0.05) (Figure 1D). Another noticeable difference is the amount of significantly prevalent OTUs in each of the two tissues; in NAT, there are considerably more OTUs compared with T tissue (Figure 1D).

Figure 1.

The microbiome composition of esophageal T tissues is mostly similar to that of NAT; however, specific OTUs show specificity to either T or NAT. (A) Microbiome α-diversity represented by the number of OTUs detected in each sample, in NAT versus T tissues. (B) Bray–Curtis PCoA analysis of the microbiome β-diversity in NAT versus T tissues. (C) Bar graphs of the microbiome composition on the family level in NAT versus T tissues. (D) Volcano plot of the OTU levels in NAT versus T tissues, the size of the dots is proportional to their taxonomic levels, and colored dots represent significance (proportion test P < 0.05, FDR < 0.25). FDR, false discovery rate; NAT, normal adjacent tissue; OTU, operational taxonomic unit; T, tumor.

We further analyzed the esophageal tissue microbiome prevalence in T and NAT combined regarding additional patient parameters (Figure 2). Our analyses included age (over or under 67 years of age, the median age of our cohort), BMI (over or under 24.9 kg/m², the upper value of normal BMI), cigarette smoking (yes or no), and weight loss before surgery (over or under 8 kg, the median weight loss in our cohort). All these analyses revealed distinct bacterial taxa in each comparison (Figure 2A-D). We also compared the esophageal tissue microbiome (T and NAT, combined and separately) with regard to tumor location (middle, Siewert 1, and Siewert 2), histology assessment and participants’ sex (ESCC or EAC), see Supplementary Figures S2A and S3, available at https://doi.org/10.1016/j.esmogo.2025.100172. In each comparison, distinct bacterial taxa were identified.

Figure 2.

Microbiome analysis of T tissue and NAT combined on the OTU level shows the differentiation of patients based on their clinical parameters. (A-D) Volcano plots of microbiome proportion test analysis comparing patients based on their clinical parameters, the size of the dots is proportional to their taxonomic levels, and colored dots represent significance (P < 0.05, FDR < 0.25). (A) Age (years). (B) BMI (kg/m²). (C) Weight loss (kg). (D) Cigarette smoking status. BMI, body mass index; FDR, false discovery rate; NAT, normal adjacent tissue; OTU, operational taxonomic unit; T, tumor.

Microbial profile in ALs

To assess the differences between patients with and without ALs, we carried out microbiome α-diversity, β-diversity, and phylogenetic analyses (Figure 3A-C). No significant differences were detected in microbiome richness or diversity, or in the microbial phylogenetic family-level analyses between the cohorts (Figure 3A-C).

Figure 3.

α, β, and family-level analyses of the tissue microbiome do not distinguish between patients with ALs. (A) Microbiome α-diversity represented by the number of OTUs detected in each sample, in patients with ALs versus patients without ALs, analyzed from both T and NAT. (B) Bray–Curtis PCoA analysis of the microbiome β-diversity in patients with ALs versus patients without ALs, analyzed from both T and NAT. (C) Bar graphs of the microbiome composition of the 30 most abundant bacterial families in ALs versus patients without ALs. AL, anastomotic leak; NAT, normal adjacent tissue; PCoA, principal coordinate analysis; T, tumor.

The bacterial OTU levels were analyzed to study the samples in more detail, using proportion tests within all T and NAT samples and combined. These high-resolution analyses revealed specific bacteria significantly more prevalent in patients with and without ALs (Figure 4A-C and E). The species Staphylococcus pasteuri and the order Rickettsiales were significantly increased in patients with ALs when analyzing NAT (P = 0.04 and P = 0.01, respectively) and the two tissues together (P = 0.01 and P = 0.03, respectively) (Figure 4A, C, and E). In the combined tissues and T tissue of patients with ALs, the presence of the genus Massilia (P = 0.02 and P = 0.02, respectively) and Corynebacterium (P = 0.02 and P = 0.01, respectively) was significantly higher compared with patients without ALs (Figure 4A, B, and E). In the T tissue of patients without ALs, there was significantly more Lactobacillus (P = 0.03) and its corresponding family Lactobacillaceae (P = 0.03) (Figure 4B and E). Figure 4D presents the overall overlapping and non-overlapping significantly differentiated OTUs in all three comparative analyses.

Figure 4.

Tissue microbiome analysis on the OTU level enables differentiation of patients with ALs compared with patients without ALs. (A-C) Volcano plots of microbiome proportion test analysis comparing patients with ALs (left) with patients without ALs (right), the size of the dots is proportional to their taxonomic levels, and colored dots represent significance (P < 0.05, FDR < 0.25) (A) in all tissue samples, (B) in T tissue samples, and (C) in NAT samples. (D) Venn diagram presenting similarities in OTUs in all comparative analyses described in A-C. (E) Heatmap presenting prevalence of bacterial OTUs comparing patients with ALs with patients without ALs, in all tissue samples (left), in NAT samples, (middle), and in T tissue samples (right); data presented from proportion test analysis of significantly (P < 0.05, FDR < 0.25) more prevalent bacterial OTUs (∗) of known taxa in the comparisons. (F) Volcano plot of enriched bacterial MetaCyc functions proportion test analysis comparing patients with ALs (left) with patients without ALs (right); colored dots represent significance (P < 0.05, FDR < 0.25).

To compare the putative functional potential of bacteria from patients based on their AL incidence, we used the PICRUSt2 tool³⁶ to predict the genes and pathways harbored by the identified bacteria. Using proportion tests, we observed differences in putative pathway prevalence between the groups (Figure 4F, Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2025.100172). The mycothiol biosynthesis pathway was significantly more prevalent (P = 0.01) in samples from patients without ALs (Figure 4F).

Discussion

Esophagectomy has a profound long-term impact on patients’ health-related quality of life (HRQoL) and combined treatment modalities can exacerbate the effects on the HRQoL.³⁸ Post-operative complications, including ALs, can hamper clinical course, affect HRQoL, and influence survival and morbidity outcomes. Delineating the potential contributors to the pathogenesis of an AL after an esophagectomy can provide insight into adequate measures to minimize this harmful complication. Increasing data indicate that gut microbes are pivotal in integrating environmental cues with host physiology and metabolism and may affect the balance of intestinal cell proliferation and death, systemic and local immune homeostasis, and alterations of the host metabolic activities, which are key processes in carcinogenesis.³⁸ Specific gut bacteria correlate with responses to immunotherapy in multiple cancer types.^15,39

The patient characteristics we included in our analyses, age, BMI, smoking, and weight loss before surgery, have not been connected to a higher risk of ALs.⁴⁰ However, age, smoking, and BMI are risk factors contributing to the development of esophageal cancer.⁴¹ To examine the relationship between these risk factors and the microbiome profile of esophageal cancer patients, we analyzed the prevalence of OTUs for each characteristic. We compared the profiles of the patients over and under the median age of the study (67 years), over and under the healthy BMI (24.9 kg/m²), smoking and non-smoking, and over and under the median weight loss before surgery (8 kg). Each comparison showed diverse bacterial taxa, implying distinct microbial patterns, which could provide further insights upon broader and more detailed analysis of the esophageal microbiome.

Analysis of the esophageal microbiome remains a relatively understudied yet promising field. Increasing our understanding could help predict complications as has been seen for post-esophagectomy pneumonia⁴² and AL complications in CRC.^20,31 In this study, we hypothesized that ALs might be affected by a differential microbial content in the tumor or the normal adjacent esophageal tissue. This hypothesis carries implications for clinical practice since standard perioperative prophylactic antibiotic treatment is empirical, focused on surgical site infection prevention, and is not tailored to the patient’s microbial profile. Other protocols like selective decontamination of the digestive tract were formerly shown to improve outcomes but are not widely accepted or implemented into routine clinical practice.^43,44 Our results indicate that patients who developed a post-operative AL displayed a differential microbial profile compared with those without a clinical leak. These results shed light on the potential role of the esophageal microbiome in the pathogenesis of ALs. We aimed to characterize the microbial milieu of the tumor and compare it to the NAT. We hypothesized that several classes of microbiota would be represented in the tumor, based on former studies,¹⁷ though not shown in esophageal cancer.

Our study population comprised two cohorts, which showed a similar abundance pattern and microbial representation (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2025.100172). The comparison of T and NAT revealed no significant differences in overall microbiome diversity, in agreement with T and NAT comparisons in other tumor types mentioned in Nejman et al.¹⁷ However, higher-resolution analysis on the prevalence of microbial OTUs identified specific OTUs significantly associated with T or NAT. There are multiple OTUs significantly relevant in the NAT, while in the T tissue substantially fewer OTUs are significant. This circumstance can be attributed to the OTUs prevalent in the NAT being represented in multiple samples. In contrast, T tissue carries few OTUs found in multiple samples, and therefore, they are less statistically significant. From this, it can be inferred that the divergence between species in the T tissue is greater than in the NAT.

Associated with T tissue were bacteria from the Rhizobiaceae family and the Ralstonia genus, both gram-negative bacterial taxa. The Rhizobiaceae family was previously linked to esophageal reflux disease,⁴⁵ a risk factor for EAC, and the Ralstonia genus is affiliated with esophageal cancer patients and patients predisposed to esophageal cancer (i.e. patients with Barrett esophagus).⁴⁶ NAT-associated bacteria were the Rickettsia and Sphingomonas genera, and S. yabuuchiae (Figure 1D). Sphingomonas yabuuchiae has been linked to normal stomach tissue when compared with T tissue in gastric cancer.⁴⁷ In a study comparing the esophageal microbiome, collected by brushing the esophageal tissue, of patients with Barrett esophagus with that of controls,⁴⁸ Sphingomonas was more prevalent. However, Sphingomonas has also been described as a contaminant in other studies using FFPE samples.^49,50

Next, we compared the tissue microbiome of patients with and without ALs. We analyzed the NAT and T tissues independently and combined. These analyses revealed specific bacterial taxa significantly associated with the prevalence of ALs. Among these, S. pasteuri, a gram-positive species, was significantly more prevalent in NAT and combined tissue analyses. A study comparing the endoscopic esophageal washes of patients with primary esophageal carcinoma and controls found lower amounts of Staphylococcus in the esophageal microbiome of patients with primary esophageal carcinoma. Several digesting enzymes were identified in this species, raising the question of whether it can degrade tissue and contribute to the AL complication.^51,52 The gram-negative Massilia genus was significantly more prevalent in tumoral and combined tissue samples of patients with ALs. This genus is elevated in duodenal tissue samples of pancreatic cancer patients⁵³ and was more abundant in T tissue than in normal tissue in a lung tumor microbiome study.⁵⁴ Additionally, it was found in greater amounts in tumors with TP53 mutations and in smokers.⁵⁴ Interestingly, Massilia can degrade polycyclic aromatic hydrocarbons, a carcinogenic substance from tobacco.^54,55 The order Rickettsiales was significantly more prevalent in combined and NAT samples of patients with ALs. Rickettsiales includes the Rickettsia genus, which was more prevalent in NAT when compared with T tissue, as mentioned earlier. Rickettsiales is enriched in the oral microbiome in throat cancer patients⁵⁵ and Rickettsia can invade endothelial cells and includes human pathogenic species.⁵⁶ The Corynebacterium genus⁵⁷ was significantly more prevalent in the T and combined tissues of patients with ALs. While Corynebacterium is mainly known as part of the normal skin microbiome,⁵⁸ it has been found in multiple studies of the esophageal microbiome.^57,59 The presence of Corynebacterium decreased the longer Barrett column measured in biopsies from patients with Barrett esophagus.⁶⁰ It was also recently found in breast tumor tissues, associated with estrogen receptor-positive breast tumors.¹⁷ In the T tissue of patients without ALs, Lactobacillus was more prevalent. Lactobacillus was also found in higher abundance in esophageal biopsies from patients with EAC in a study by Elliott et al. (2017).⁶¹ This genus is resilient to low pH conditions⁶² and it is hypothesized that their production of lactic acid could further acidify the microenvironment, creating an acidic tumor niche^57,61; it is also more abundant in gastric cancer.⁶³

Some of the bacteria identified as associated with ALs in this study (e.g. Corynebacterium⁶⁴ and S. pasteuri⁶⁵) have been described as susceptible to antibiotics, specifically tetracyclines, which may be used as a treatment in preparation for an esophagectomy.

We used the PICRUSt2³⁶ tool to predict the functional compositions in our bacterial data and found pathways that were increased in patients without ALs, such as the mycothiol biosynthesis pathway. Mycothiol is an intracellular reductive agent that protects bacterial cells from oxidative stress and alkylating agents.⁶⁵ Oxidative stress is an indicator for ALs after surgery.⁶⁶ Understanding this pathway and its relationship to oxidative stress environments requires further functional and mechanistic studies.

While this work sets a precedent in the analysis of esophageal microbiome in the context of esophageal cancer, especially in the relevance to post-esophagectomy ALs, we recognize limitations in our approach. The size of our cohort is relatively small, yet we had a 28.3% prevalence of ALs, which is within expected rates.¹⁰ However, the size of the cohort prevented us from carrying out sub-analysis on the incidence of ALs in relevance to neoadjuvant therapy since the participants received five different chemotherapy protocols, which prevented reliable conclusions due to the sub-sample size. Additionally, the cohort size limited sub-analysis of our data also based on the histological type of the tumors. Data regarding dietary habits and nutritional supplements of the participants were not collected as part of the clinical data of this study. Due to the geographical location of the cohort, a trend to the Mediterranean diet can be assumed; however, no concrete conclusions can be made. Furthermore, our experimental approach to carrying out this analysis allowed us to make conclusions only based on differences between the groups and not draw causal conclusions based on these results alone. One future approach to establish a causal connection would be to introduce esophageal cancer murine models colonized with specific bacteria of interest identified in this work. However, this was out of the scope of the current work.

Only FFPE patient samples were available for our analysis, which come with limitations. DNA within FFPE samples is usually fragmented due to sample processing and since the samples are from patient tissue, most of the DNA is human. Additionally, these samples were only processed in sterile conditions once our team handled them. To address these limitations, we included controls of margins (i.e. without any tissue) of FFPE blocks and included template controls for each step of sample and DNA handling. Each step included several controls that amounted to 10% of all samples handled. The 5R method¹⁷ (to sequence and analyze DNA) applied in this work allowed us to analyze fragmented DNA since the method depends on sequencing five short fragments of the bacterial 16s ribosomal DNA gene. We used the PICRUSt2 tool³⁶ to predict the functional profiles of bacteria in our samples. This method is based on reference genomes that might not represent the true diversity of the microbiome and cannot capture the full functional potential of microbial communities in our samples. Future studies should consider incorporating metagenomic and metatranscriptomic approaches to overcome these limitations.

This study is retrospective with consecutive patients from a specific population. Our results show the need for a prospective study in which microbiome samples retrieved from saliva, esophageal secretions, stool, and tumor tissue are collected longitudinally to correlate over time with the occurrence of ALs. Results from such a project could also strengthen the research suggesting that the oral microbiome is similar to the esophageal microbiome, enabling the simplification of sampling procedures.67, 68, 69 Today, the perioperative standard of care is empiric and even, which, given our results, may warrant distinct evaluation for tailoring the perioperative antibiotic regimen upon host and tumor microbial signatures.

Future studies on larger cohorts and across different geographical locations will help to provide mechanistic insight into AL etiology and potentially personalized treatment to target critical bacteria underlying this post-operative complication.