Microbiome typing in uveal melanoma is associated with plaque radiotherapy

Background: Microbiomes have been identied in various tumor types and could affect tumor progression and treatment. As the most prevalent primary malignant eye tumor in adults, uveal melanoma (UM) has not been explored regarding its endogenous microbiome. Plaque radiotherapy (PRT) is the gold standard for the treatment of UM. Hereby, we recruited 71 UM patients, sequenced the 16S rRNA gene of their tumor tissues, and analyzed the association between UM microbiome and disease phenotypes. Results: Clear bacterial signals were observed in UM tissues by using in situ uorescence hybridization. 523 bacterial species passed strict decontamination against 58 environmental control samples in 16S rRNA gene analysis, and these species formed three distinct types by unsupervised clustering. The UM microbiome types were signicantly associated with PRT. A biomarker analysis showed that Pseudomonas and Sphingomonas were signicantly enriched in the radiation group (RG) compared to the non-radiation group (NRG). A kind of radiation-resistant bacteria had a signicantly higher positive rate in tumor tissues that underwent radiotherapy. We found that radioresistant bacteria Deinococcus was associated with larger tumor and later tumor stage, while Pedobacter, Bradyrhizobium, Variovorax, Pseudomonas and Stenotrophomonas were associated with later metastasis. Conclusion: There was an endogenous microbiome in UM tissues, and the distribution of microbiota was correlated with PRT and clinical features.


Introduction
Uveal melanoma (UM) is the most common primary intraocular malignancy in adults, with an annual incidence in Asian countries of 0.6/1000000 [1,2]. Due to the lack of a lymphatic system in the eye, UM cells enter the systemic circulation mainly through blood vessels in the eye, resulting in distant metastasis. Studies have shown that nearly 50% of patients develop metastases, with liver involvement more common, and nearly 80% of patients die within one year after metastasis [3]. At present, the treatment of UM is mainly divided into enucleation and eye-preserving therapy, in which plaque brachytherapy, a kind of plaque radiotherapy (PRT) is the main treatment [4].
The human microbiome is made up of microbes that live inside and on the surface of the human body.
All humans share a core microbiome [5] in which the gut microbiome is in uenced by differences in diet and host phenotype, thus diversifying the individual microbial ecosystem [6]. The gut microbiome has been shown to in uence tumor biology in many ways, such as regulating in ammatory responses, inducing DNA damage, producing metabolites associated with tumorigenesis or tumor suppression, and responding to cancer treatment, including immunotherapy [7]. Previous studies have con rmed that the diversity of intestinal ora in melanoma patients receiving checkpoint blockade immunotherapy is signi cantly correlated with the therapeutic effect[8], while UM patients do not respond well to checkpoint blockade immunotherapy.
In addition to intestinal ora, bacteria have been found in breast cancer, lung cancer, cutaneous melanoma and other tumor tissues, and the composition of different types of tumor microbiome is different. The characteristics of tumor ora and metabolic pathway are related to tumor type and subtype, clinical characteristics of patients, tumor development and therapeutic effect [9,10]. It is speculated that the bacteria selectively colonize tumors with abnormal vascular systems with abundant blood supply and relative leakage, thus growing and multiplying in the relatively hypoxic tumor microenvironment (especially anaerobe or facultative anaerobe). They can regulate enzyme activity, activate speci c receptors on cancer cells to make them resistant to chemotherapy, and alter the tumor microenvironment through immune regulation [11]. A recent study found that oral vancomycin was associated with enhanced radiotherapy in cutaneous melanoma xenotransplantation models [12].
Based on the above research background, we speculate that there may be a microbiome in UM. Therefore, 71 cases of UM tissues were sequenced by 16S rRNA and analyzed in the present study. We want to explore the composition of the UM microbiota and the relationship between bacteria and PRT.

Results
Clinical Data of 71 UM samples A total of seventy-one UM tissues were included in this study and 16S rRNA sequencing was performed, including 36 cases in radiation group (RG) and 35 cases in non-radiation group (NRG) (Fig. 1). Our cohort's average age of onset was younger than western patients, NRG was 50.7 ± 11.6 years and RG was 46.7 ± 11.5 years. 34 males and 37 females were included. There were 41 tumors in the right eye and 30 in the left eye. In this study, 7.2% of the tumors were in Stage I, 52.1% were in Stage II, and 40.6% were in Stage III (The American Joint Committee on Cancer classi cation, 8th ). According to Collaborative Ocular Melanoma Study's recommendation [13], the tumor thickness was no more than 10.0mm could consider accepting PRT, so there was a signi cant difference in average tumor thickness between NRG (10.7 ± 3.8mm) and RG (6.2 ± 2.6mm) (p < 0.001), but there was no signi cant difference in the average largest basal diameter between the two groups (14.7 ± 4.4 vs 13.4 ± 3.6mm).

Tyramide signal ampli cation-uorescence in situ hybridization detected bacteria in UM tissues
We performed tyramide signal ampli cation-uorescence in situ hybridization (TSA-FISH) in 5 tissues (2 of NRG and 3 of RG) using the probe EUB338 I-III targeting bacterial 16S rRNA. At the same time, we also conducted a negative control experiment and found that the uorescence signal of speci c staining could not be observed without adding EUB probe. Staining for EUB338-probed hybridized sections were detected both in RG and NRG tissues, as shown in Fig. 2. Most bacteria are distributed in the cytoplasm and close to the nucleus. The positive bacterial signals were constantly observed in hybridized sections using the probe EUB338 I-III in all ve patients.

Characterizations of UM microbiome in NRG and RG
A total of 71 tissue, including UM samples (35 of NRG and 36 of RG) and 58 environmental samples (operating room and laboratory environment) have been sent to undergo 16S rRNA gene sequencing (V3-V4 regions). We set four ltering criteria for algorithmic decontamination (threshold score of 0.01, 0.1, 0.5, 0.9, the lower the score, the stricter the decontamination degree) to remove potential contaminations from UM tissue using the forementioned environmental samples as contamination sources. After decontamination, we obtained the OTU number of the corresponding threshold score, respectively ( Fig. 3A). To ensure the authenticity of the existence of bacteria, we used the OTU table with a decontamination threshold score of 0.01, removed the samples with reads less than 2000 (NR34, R21 and R26) and standardized analysis according to the minimum sequencing quantity of the remaining samples, which OTU number was 523 and had 160956 reads as the basis for further analysis.
We hope to observe the OTU number and relative abundance of all kinds of bacteria in NRG and RG at different species classi cation levels. The results showed that the UM tissue was dominated by Proteobacteria at the phylum level, while in some RG samples, the abundance of Firmicutes and Bacteroidetes was relatively high (Fig. 3B). At the family and genus level, the abundance of different species in each sample was relatively stable (Figs. 3C and 3D). The relative abundance of Stenotrophonomoas in 11 RG samples was more than 50%, while only 6 in NRG. We found that there are only 17 species whose relative abundance was greater than 0.01 and can be classi ed at the genus level.

Microbiota types in UM tissues
We referred to the enterotype classi cation method of the gut microbiota [14] and performed it in UM samples. The UM samples formed three clusters that we designate them as UM 'Types' (Fig. 4, Supplement Fig. 1). Type2 markers were Stenotrophomonas, Methylobacterium, Pseudomonas, Elizabethkingia, Pedobacter, Variovorax, Sphingomonas and Comamonas (Supplement Fig. 1A-H), the enrichment of these bacteria in Type2 was signi cantly higher than that of Type1 and Type3. Although we did not nd any bacterial taxonomy that is speci cally enriched in Type1 or Type3, we compared the relative abundance of 17 species and 'others' in RG and NRG, and found that the relative abundance of Pseudomonas and Sphingomonas, which were enriched in Type2, was also higher in RG (Supplement Fig. 2A and 2B), and the number of samples in RG in Type2 is more than that in NRG (Fig. 4A).

Differential expression and correlation network of bacteria in NRG and RG
Under the strictest threshold (0.01) of decontamination, comparison of the alpha and beta diversity indices showed no signi cance between NRG and RG. Additionally, linear discriminant analysis effect size (LEfSe) analysis showed that Pseudomonas and Sphingomonas were signi cantly higher in RG than that in NRG (linear discriminant analysis (LDA) score > 2.0, p < 0.05) (Fig. 4C). We further investigated the correlation of bacteria in RG and NRG at the genus level. The inclusion criteria with correlation coe cient over 0.4 were selected and presented in the form of the network diagram in Fig. 4D. Although there was only positive correlation among all species except 'others' in the two groups, it was still found that the bacteria were divided into two clusters in the RG. In the correlation network, Stenotrophomonas, with the highest abundance, has a positive correlation with only one cluster (left).
A. Composition of RG and NRG samples in each type. B. Types of UM samples which were visualized by PCoA and clustering at the genus level. C. Signi cant different bacteria enrichment between NRG and RG.
The results could be biomarkers in RG. LDA score > 2.0, p < 0.05. D. Co-occurrence networks of NRG and RG at the genus level. The size of the circle represents the abundance of bacteria. The thickness of the connected lines represents the strength of the relationship between bacteria.

Correlating the UM microbiota composition with clinical characteristics
In order to investigate possible correlations between the microbiota composition of UM tissues and clinical characteristics, we included clinical features that may be related to the severity of the disease, calculated their associations with different bacteria, and inferred whether a species is related to the progression of the disease (Fig. 5, Table 1). We observed that Micrococcus, Methylobacterium, Sphingomonas, Variovorax, and Pseudomonas were correlated with younger patients (adjust p < 0.05). For tumor characteristics, tumor thickness was found to be negatively correlated with Sphingomonas, Variovorax, Pseudomonas and Deinococcus (adjust p < 0.05).
In analyses related to clinical outcomes, Pedobacter, Bradyrhizobium, Variovorax, Pseudomonas and Stenotrophomonas were associated with later metastasis, while Variovorax was related to a shorter metastasis-death time (adjust p < 0.05). More basic experiments and clinical trials will be needed to verify our results.

Discussion
With the improvement of microbial detection technology, scientists have paid more and more attention to the research of ora. In recent years, variates studies have found that the existence of bacteria in tumor and it is related to the occurrence and development of disease [9,15]. In this study, we found the existence of bacteria in UM for the rst time, explained the characteristics of UM microbiota and its relationship with PRT and clinical features. It remains controversial whether bacteria are present in tissues previously thought sterile. The rst conceptual evidence for bacterial translocation was reported in 2011 when Amar et al showed that bacterial 16S rDNA gene was detectable in blood [16]. After tumorigenesis, the destruction of physical and molecular barriers, coupled with relative immunosuppression in tumor microenvironment, may increase the likelihood of bacterial translocation to normally aseptic sites. Then various studies supported the hypothesis of translocation bacteria [17,18]. A recent study found that bacteria are widespread in different types of tumors (including exposed and protected), of which melanoma has the least bacterial content [9]. We rst presented the composition of the human UM microbiota and found that it could be divided into three types in the present study. We also observed differences in UM microbiota between NRG and RG, and evidenced that bacterial signatures were signi cantly associated with UM clinical characteristics.
Even though we use strict aseptic procedures during the operation, we still collect a large number of environmental control samples for algorithmic decontamination. After using the most stringent decontamination threshold of 0.01, we removed about 95% of the species, but there were still real bacterial signals, so the conclusion that bacteria were found in UM tissues in this research is credible. This study detected positive bacterial signals in both NRG and RG UM samples using TSA-FISH, distributing in cell plasm but not in nuclear.
After 16s rRNA sequencing and OTU annotation, we found that the UM microbiota was mainly Proteobacteria, and at the genus level, it was mainly Stenotrophomonas, Variovorax and Pedobacter. Stenotrophonomoas has the highest relative abundance in UM microbiome, which is consistent with recent report [15]. Previous studies have shown that as a new opportunistic pathogen, patients with longterm neutropenia and exposure to broad-spectrum antibiotics have a higher risk of Stenotrophonomoas infected, leading to pneumonia, bacteremia, catheter-related infection and other diseases [19]. There were more annotated species in RG with relative abundance > 0.01 than in NRG. While the proportion of 'other' in some samples was very high, which may be due to the following reasons: First, the species diversity was high, but the relative abundance was less than 0.01. Second, some species cannot be classi ed to the level of family or genus. Third, there may be new species that have not been discovered before.
A very interesting phenomenon is that all the UM samples formed three distinct clusters, which was in accordance with studies in gut enterotypes [14]. Among them, RG accounted for more in Type 2. To explore the microbiome characteristics of Type 2, we compared the relative abundance of all detected bacteria in the three types, and found that Stenotrophomonas, Pseudomonas and Variovorax were enriched in Type 2 and had statistical signi cance (Supplement Fig. 1, p < 0.0001). A study had shown that Stenotrophomonas can secrete outer membrane vesicles and elicit a potent in ammatory respons [20]. Lipopolysaccharide from Pseudomonas solanacearum possessed the expressed antimetastatic effect in melanoma B-16 experimental animal [21]. The abundance of Variovorax paradoxus was correlated with histological gastritis [22]. Although Bradyrhizobium, Enhydrobacter, Leptothrix and Acinetobacter are abundant in Type3, they are different among the three types.
To explore the effect of PRT on UM microbiota, we analyzed RG and NRG by LEfSe to nd out the difference between the two groups. The results showed that the abundances of Pseudomonas and Sphingomonas were higher in RG. Previous research showed that Pseudomonas aeruginosa ExoT induces G1 cell cycle arrest in melanoma cells, which signi cantly inhibits the growth of melanoma [23]. Sphingomonas was found to be enriched in tumor tissues in both thyroid cancer [24] and bladder cancer [25], while Sphingomonas was relatively enriched in normal breast tissues compared with breast cancer tissues[26].We found that there was a signi cant difference in the positive rate of Deinococcus with strong radiation resistance [27] between RG (17 of 36 cases) and NRG (8 of 35 cases) (p = 0.032). Therefore, we thought that PRT in uences UM microbiome.
We analyzed the correlation between 17 species and clinical information to explore the in uence of UM microbiota on clinical characteristics and disease prognosis. Deinococcus radiodurans can survive in various environments with extreme radioactive and ultraviolet radiation[28], its extract Deinoxanthin can induce apoptosis of cancer cells [29]. This study found that Deinococcus was associated with small and early tumors, while the positive rate of Deinococcus in RG was high, and the operation-metastasis time and operation-death time in patients with RG was signi cantly longer than that in NRG ( Table 1), suggesting that Deinococcus may play a role in inhibiting tumor development. Pedobacter [30] and Bradyrhizobium [10] have been con rmed to have anti-tumor effects in previous studies, which is consistent with our ndings that they are positively correlated with late metastasis of UM. Although Variovorax was associated with better clinical phenotype before tumor metastasis, it was negatively correlated with metastasis-death time (the higher relative abundance of Variovorax, the faster the metastatic patients die). Stenotrophomonas was found to be associated with a better prognosis in this study, but it was thought in previous studies that it could elicit a strong in ammatory response [20] and enrich in tumor tissue [31], which is inconsistent with our conclusion and needs further experimental veri cation.
There were still some limitations in this study. First, the tumor tissue is small and lacks adjacent tissue samples. Due to the small size of the eyeball, the limitation of local excision, and the improvement of the eye preservation rate of UM, we can rarely obtain tumor and adjacent samples, especially the normal control samples. Second, lack of detection of living bacteria and metabolic function. Although we have detected bacteria in 16s rRNA and TSA-FISH, we cannot directly observe bacteria. We can add electron microscope detection in the future study, supplemented with living bacterial biomarkers and bacterial culture to verify the existence of living bacteria, which is a great challenge, because it requires very strict experimental conditions and techniques. Third, we cannot determine the source of the bacterial DNA detected in UM. Whether the bacteria were derived from the gut or adjacent sites, we need more types of samples and experiments for further exploration. Last, the analysis of UM tissues was based on the strictest threshold of decontamination, which actually may lose real bacterial signal. However, how to set the decontamination threshold to maintain the authenticity of the sequencing results best and eliminate the interference of environmental contamination is still controversial. Nevertheless, previous study has shown that the number of bacteria in cutaneous melanoma is also the lowest compared to other types of tumors [9]. Therefore, more large-scale basic research and bioinformatics analysis are needed to explore whether the UM microbiome plays a causal role in UM occurrence, treatment and progression.

Conclusions
In conclusion, our study provides evidence for the existence and composition of bacteria in UM, exploring the relationship between UM's microbiome and the clinical characteristics. There are differences in tumor microbiota in UM patients treated with PRT or not, and UM could be divided into three types according to the microbiome, which is associated with PRT.

Methods
We chose the tumor tissues of 71 patients treated in Beijing Tongren Hospital. All samples were obtained by local resection of the tumor or enucleation. We divided 71 tumor samples into two groups according to whether they received PRT (RG) or not (NRG), and 16s rRNA sequencing and analysis were carried out. In addition, we selected UM tissues from 5 patients for tyramide signal ampli cation-uorescence in situ hybridization (TSA-FISH), including 3 cases of RG and 2 cases of NRG. All samples were collected under aseptic conditions during the operation. At the same time, 58 environmental negative control samples were collected, including operating room air, sterile clothes, dry gauzes, wet gauzes, instrument washing solution, sterilized skin swabs, physiological saline, sterile towel swabs, sampling tubes, aseptic buffer solutions, laboratory air, sterile water and other environmental samples. The specimens were transferred directly from the operating room to the biosafety cabinet in the aseptic container in the 20 minutes. All samples were stored at -80℃ until undergoing TSA-FISH or 16S rRNA sequencing. The main diagnostic and treatment criteria are consistent with the Collaborative Ocular Melanoma Study [32] . We recorded demographic information (age, sex), eye and tumor features (best-corrected visual acuity, intraocular pressure, laterality, tumor size, shape, pigmentation, location, optic disc involvement, association with subretinal uid, intraocular hemorrhage, ciliary body involvement and pathological types). Whether or not to receive PRT depends on the characteristics of the tumor [4] and the patient's will. Otherwise, we recorded the time of tumor metastasis and death.

TSA-FISH
TSA-FISH was performed following a modi ed protocol described by Choi et al [33]. The basic steps included para n section dewaxing and tissue rehydrating, pretreatments of tissue sections, hybridizating probe and detection with anti-digoxigenin/horseradish peroxidase antibodies (anti-DIG HRP-conjugated antibody). EUB338 I-III were speci c bacterial probes that could bind to the 16S rRNA region of most bacteria. Digoxin labeled EUB probe was selected to label bacteria in UM tissues. Antidigoxin/horseradish peroxidase antibody was used as a secondary antibody to bind to EUB probe and was observed under the same uorescence microscope after ampli cation of Tyramine signal. All steps were performed under sterile conditions.

16S rRNA gene ampli cation and sequencing
Microbial DNA was extracted from UM samples (36 of RG and 35 of NRG) and negative samples (58 of environmental samples) using a DNA extraction kit according to the manufacturer's protocols. The concentration and purity were measured using the NanoDrop One (Thermo Fisher Scienti c, MA, USA). Next, 16S rRNA genes of V3-V4 hypervariable regions were ampli ed with the primers 338F (5'-ACTCCTACGGGAGGCAGCA-3') and 806R (5'-GGACTACHVGGGTWTCTAAT-3') using the thermocycler PCR instrument BioRad S1000 (Bio-Rad Laboratory, CA, USA). The length and concentration of the PCR product were detected by 1% agarose gel electrophoresis. Samples with a bright main strip between could be used for further experiments. PCR products were mixed in equidensity ratios according to the GeneTools Analysis Software (V4.03.05.0, SynGene). Then, the mixture of PCR products was puri ed with E.Z.N.A. Gel Extraction Kit (Omega, USA). After puri cation of PCR products, Sequencing libraries were generated using NEBNext® Ultra™ DNA Library Prep Kit for Illumina® (New England Biolabs, MA, USA).
The library quality was assessed on the Qubit@ 2.0 Fluorometer (Thermo Fisher Scienti c, MA, USA) and sequenced on Illumina Novaseq6000 System with Mode PE250 and 250 bp paired-end reads were generated.
Decontamination and data analysis All 16S rRNA gene datasets were processed using a standardized pipeline in QIIME 2 [34]. For each dataset, demultiplexed sequencing reads from the UM tissues and environmental control samples were denoised to generate amplicon sequence variants (ASVs) using the software DADA2 (V1.6) [35]. , an open-source R package that implements a statistical classi cation procedure that identi es contaminants in 16S data, was used to lter out features on genus level with a high chance of being contaminants using sequenced environmental control samples [37]. Prevalence based contaminant identi cation was used with four different threshold scores of 0.01, 0.1, 0.5, 0.9, respectively. We removed mitochondria, chloroplasts, unclassi ed species and the samples with reads less than 2000, then standardized analysis according to the minimum sequencing quantity of the remaining samples. After combining the above-mentioned methods, we obtained four OTU tables with different decontaminative levels.
Clustering and classi cation were performed using Jensen-Shannon distance and partitioning around medoid (PAM) clustering. The optimal number of clusters was estimated using the Calinski-Harabasz index [14].
Alpha diversity was applied in analyzing the complexity of species diversity for a sample through four indices, including Observed species, Chao1, Shannon, Ace, based on nal OTU tables. All these indices in our samples were calculated with QIIME 2 and displayed with R software(V2.15.3).
Beta diversity analysis was used to evaluate differences of samples in species complexity. Principal Coordinate Analysis (PCoA) was performed to get principal coordinates and visualize from complex, multidimensional data. PCoA analysis was displayed by QIIME 2 and ggplot2 package in R software. NMDS analysis was performed by the vegan package of R software based on the weighted and unweighted unifrac distance matrix.
LEfSe analysis was used to nd the biomarker of NRG and RG based on the nal OTU table with threshold score of 0.01. Non-parametric factorial Kruskal-Wallis (KW) sum-rank test detected species with signi cant differences in abundance between different groups, and Wilcoxon rank sum test was used to judge the difference between the two groups. Finally, LDA was used to achieve dimensionality reduction and evaluate the impact of signi cantly different species (LDA Score) by setting LDA score ≥ 2 and obtained the biomarkers in different groups.
Correlation analysis and variation partitioning analysis were performed by R software based on the key genus (relative abundance ≥ 1%), and canonical correspondence analysis was analyzed based on the nal OTU table with threshold score of 0.01. Spearman correlation index between species or samples based on the OTU table was calculated by R software and visualized by cytoscape (http://cytoscape.org/).
Correlations of the taxonomy counts (phylum and genus relative abundances) and disease features, such as demographic information, clinical features, and follow-up results were estimated using the Spearman correlation coe cient and represented in a correlation plot, using the R Corrplot package.

Statistical analysis
The statistical analyses were performed using the Statistical Product and Service Solutions (SPSS) software version 25 (International Business Machines Corporation, Armonk, New York, United States). Students' t-test was used to determine the difference between two groups. The positive rate of Deinococcus in RG and NRG was analyzed by chi-square test. Statistical signi cance was considered when p-value was less than 0.05.

Availability of data and materials
Raw data of all samples were successfully uploaded on NCBI and the data will be released at https://www.ncbi.nlm.nih.gov/sra/PRJNA807130 to the public on June. 30th, 2023.

Competing of interests
The authors declare that they have no competing interests.    Different UM types and differential expression and correlation network of bacteria in two groups.
A. Composition of RG and NRG samples in each type. B. Types of UM samples which were visualized by PCoA and clustering at the genus level. C. Signi cant different bacteria enrichment between NRG and RG.
The results could be biomarkers in RG. LDA score >2.0, p <0.05. D. Co-occurrence networks of NRG and RG at the genus level. The size of the circle represents the abundance of bacteria. The thickness of the connected lines represents the strength of the relationship between bacteria. Figure 5