Microbiome proling reveals distinct microbial compositions of nasal polyps in chronic rhinosinusitis

Background: Chronic rhinosinusitis (CRS) is characterized according to the presence or absence of nasal polyps (NPs) and displays nasal microbiota dysbiosis. However, optimal sampling methods of the nasal microbiome in CRS have not been identied. We therefore aimed to assess the microbial composition in patients with CRS, comparing different sampling methods (swab and tissue biopsy), tissue types (uncinate tissue and NP), and disease subtypes. Methods: Samples were obtained by swabbing the middle meatus and taking a biopsy of uncinate tissue (UT) in 8 patients with CRS with NP (CRSwNP), 6 patients with CRS without NP (CRSsNP), and 8 controls. NPs were also harvested in CRSwNP. Extracted DNA samples were analyzed by 16S rRNA gene amplicon sequencing. Results: The microbiome diversity did not signicantly differ between disease subtypes in nasal swabs or UT samples. However, a principal coordinates analysis based on weighted UniFrac distances revealed a greater interpersonal variance for nasal swabs than for UT. UT samples presented with a distinct pattern and composition in CRSwNP compared to CRSsNP or control. Compared to UT, NP revealed a unique microbiome prole with signicantly less bacterial diversity. Prevotella was the most abundant genus in the UT regardless of the disease subtype and its prevalence was signicantly reduced in NP. Prevotella abundance was negatively correlated with disease severity as measured by Lund-Mackay scoring. Conclusion: Tissue samples are better specimens than nasal swabs for assessing the microbiomes of CRS patients. Microbiomes of NP tissues revealed an association with clinical severity of CRSwNP.


Background
Chronic rhinosinusitis (CRS) is a common in ammatory upper airway disease that affects approximately 12.1% of the US population (1) and 8.4% of the Korean population (2). CRS is clinically de ned based on the presence of symptoms for over three months, which include positive endoscopic and/or computed tomography (CT) ndings. CRS can be further classi ed according to the presence of nasal polyps (NPs) as CRS with NP (CRSwNP) and CRS without NP (CRSsNP) (3).
Clinically, it is important to differentiate between CRSwNP and CRSsNP, as patients with the former have a greater disease burden than those suffering from the latter in terms of disease severity and poor treatment outcomes (4,5). Additionally, CRSwNP and CRSsNP are known to have distinct biological characteristics, indicating the diverse pathogenesis of CRS as depicted by different in ammatory pro les (6).
Numerous studies have examined the role of commensal bacterial communities in CRS development.
Interactions between the host immune system and sinonasal bacteria implicate microbiota in disease pathogenesis. For example, a well-known association exists between the Staphylococcus aureus superantigen IgE and Th2 in ammation in NPs (7,8). Additionally, Pseudomonas aeruginosa is known to be associated with a poorer quality of life and unfavorable outcomes in patients with CRS (9,10).
However, the roles of other diverse bacteria in the pathogenesis of CRS remain unclear.
Although recent studies of the nasal microbiome have described the overall composition and dysbiosis of the microbiome in CRS (11)(12)(13)(14), individual microbial data vary between studies. This discrepancy could be explained by diverse technical approaches and sampling methods utilized in each study, different anatomical sampling sites, and heterogenous patient distribution, such as different proportions of CRSwNP and CRSsNP.
In this study, nasal microbiota were analyzed in patients with CRS according to the sampling method (swab vs. tissue), disease subtype (with or without NPs), and tissue type [uncinate tissue (UT) vs. NP].
Considering the potential role of bacteria in the pathogenesis of CRS and the distinct characteristics of CRSwNP compared to CRSsNP, the microbiome pro les in the NP are likely to have distinct characteristics.

Study participants
This cross-sectional study was based on the nasal microbiome pro les of patients who underwent primary endoscopic sinus surgery at Seoul National University Hospital. CRSwNP and CRSsNP were diagnosed based on the European Position Paper on Rhinosinusitis with NPs (3). All patients were initially managed medically with three weeks of systemic antibiotics and intranasal corticosteroid treatment. Endoscopic sinus surgery was performed if no improvement was observed after initial therapy.
Subjects who underwent endoscopic septoplasty for nasal-septal deviation without evidence of CRS were recruited as controls. Systemic steroids and antibiotics were avoided for at least four weeks before surgery.
Demographic variables including age, sex, smoking history, atopy, asthma, antibiotic use, and intranasal steroid administration were evaluated for all patients. Atopy was assessed by either a skin prick test to detect reactions to common inhalant allergens or a multiple allergen simultaneous test (Immunosystems, Mountain View, CA, USA). Patients presenting with a positive response to methacholine inhalation or short-acting β2 agonists with a history of wheezing, shortness of breath, or chest tightness were regarded as having asthma (15). Quanti cation of disease severity was analyzed by CT using the Lund-Mackay (LM) scoring system (16).

Sample collection
Three different types of samples were acquired: 1) nasal swabs from the middle meatus, 2) tissue biopsied from the uncinate process (UT), and 3) NP tissue biopsied from the ethmoid sinus. All samples were collected intra-operatively. To eliminate the possibility of contamination, surgical draping with a solution composed of 2% chlorhexidine gluconate and 70% isopropyl alcohol was used for all patients.
Swabs were taken from the middle meatus under endoscopic guidance with cotton ball swabs (Transystem™, COPAN, Brescia, Italy). After swabbing, the UT or NP was harvested surgically. All samples were collected by a single surgeon. Samples were immediately separated into sterile containers, placed on ice, and transported to the laboratory for storage at -80 °C. Any swabs that may have contacted the nasal vestibule were discarded. DNA extraction, PCR ampli cation, and 16S rRNA gene sequencing Swab heads were thawed, cut into small pieces, and then placed in 180 µL of enzymatic lysis buffer (Qiagen, Hilden, Germany) overnight at room temperature. Bead homogenization was performed with 5mm steel beads agitated for 20 s at 15 Hz, followed by 0.1-mm glass beads for 5 min at 30 Hz. The same extraction procedure was carried out for the tissue biopsy samples. The remainder of the extraction protocol was performed using the FastDNA® SPIN Kit for Soil DNA extraction according to the manufacturer's instructions (MP Biomedicals, Irvine, CA, USA). Extracted DNA was stored at -80 °C until sequencing. PCR ampli cation of the V3-V4 hypervariable region of the bacterial 16S rRNA gene was performed using 3 µL of the extracted DNA template and 25 µL 2X KAPA HiFi HotStart ReadyMix (KAPA Biosystems, Wilmington, MA, USA). The primers 318F (5′ TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′) and 806R (5′ GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC-3′) were used for 16S rRNA gene ampli cation. The PCR cycling conditions were 3 min at 95 °C; 25 cycles for 30 s at 95 °C, 30 s at 55 °C, and 30 s at 72 °C; 5 min at 72 °C; holding at 4 °C. The amplicons were puri ed using Agencourt AMPure XP magnetic beads (Beckman Coulter, Brea, CA, USA). The puri ed products were subjected to index PCR using a Nextera Index Kit (Illumina, San Diego, CA, USA) following the manufacturer's instructions. The products were further puri ed using AMPure XP beads. The resulting bacterial amplicon library was quanti ed, mixed with multiple libraries, and sequenced using the MiSeq v3 platform (Illumina). Sequence data processing The raw sequence data les were preprocessed using several bioinformatics tools before downstream data analysis. To remove sequences with low quality scores, the raw reads were pre-ltered using PRINSEQ (17). Adapter sequences were removed using CUTADAPT (18) and ambiguous bases were further trimmed using the FASTX-toolkit (19). Pair-end reads were merged using PEAR (20) and further ltered with PRINSEQ. Chimeric sequences and singletons were screened and reduced using USEARCH (21). A random sampling of 10000 reads was performed to equalize the sequencing depth between samples. Downstream data analysis was completed using QIIME (22) with the Ezbiocloud 16S rRNA gene sequence database (21). Operational taxonomic units were de ned as clusters of sequences with ≥ 97% identity. Alpha-and beta-diversity indices were calculated using the core_diversity_analyses option. The weighted normalized UniFrac distance or Bray-Curtis distance was plotted by principal coordinates analysis (PCoA).

Statistical analysis
Clinical parameters among each group were compared by Kruskal-Wallis tests followed by the Mann-Whitney U test for post hoc comparison. Alpha diversity metrics among each group were compared using the Kruskal-Wallis test, and a Wilcoxon signed-rank test was used for paired analyses. P values were adjusted for multiple comparisons using the Benjamini-Hochberg method. The associations between the microbiome compositions and the study groups were tested using permutational multivariate analysis of variance (PERMANOVA). Linear discriminant analysis effect size (LEfSe) (23), a method for biomarker discovery, was used to identify genera that best characterized each study group. Taxa with linear discriminant analysis scores > 2 and P < 0.05 were considered to be signi cant. Simple linear regression was performed to identify the most relevant genera related to disease severity according to the LM score in CT scans. Spearman correlation coe cients were determined to assess the relationships between the relative abundance of genera and disease severity. Statistical analyses were performed using SPSS ver. 22.0 software (SPSS, Inc., Chicago, IL, USA). Simple linear regression was performed using Weka software. 24 Signi cance was accepted at P < 0.05 for multiple analyses.

Clinical characteristics of study participants
Samples from 22 adults in 3 different groups (8 CRSwNP patients, 6 CRSsNP patients, and 8 control subjects) were collected. Among the three groups, clinical parameters including age, male to female ratio, presence of atopy or asthma, usage of intranasal steroid or oral antibiotics, and smoking status did not differ, but pre-operative CT scores showed signi cant differences (P = 0.001, Table 1). CRS groups had signi cantly higher CT scores than controls (9.9 ± 5.5 vs. 0.1 ± 0.4, P < 0.001); however, there was no signi cant difference between the average CT scores of CRSwNP and CRSsNP groups (10.9 ± 5.1 vs. 8.5 ± 6.2, P = 0.491). Microbial composition varies based on sample sites and disease subtypes The microbial composition from swabs and UT were compared among the three groups (control, CRSsNP, and CRSwNP; Fig. 1). There was no signi cant difference in α-diversity (using the Shannon index) as a function of disease status for the swab and UT samples (Supplemental Figure S1, Additional File 1).
At the phylum level, 11 bacterial phyla were detected in the nasal microbiome (Fig. 1A). Bacteroidetes, Firmicutes, Proteobacteria, Actinobacteria, and Fusobacteria were the dominant phyla, comprising more than > 99% in both swabs and UT regardless of the disease status. The 30 most abundant genera in nasal microbiomes across sampling sites and disease subtypes are presented in Fig. 1B. In general, Prevotella was the most common in both swabs and UT, but the other major abundant taxa differed signi cantly between sample types; Staphylococcus, Propionibacterium, and Kluyvera were more abundant in swabs and Ralstonia, Ruminococcus and Barnesiella were more abundant in UT (Supplemental Figure S2, Additional File 1).
UT sampling provides homogenous microbial pro ling associated with disease subtype PCoA was performed based on the weighted UniFrac distances, which considers genus-level operational taxonomic unit data ( Fig. 2A). Nasal swab samples showed substantial in-group heterogeneity and microbial characteristics were not associated with speci c disease subtypes. In contrast, UT formed a tight PCoA cluster, indicating the relative homogeneity of the microbiome in UT as compared to swabs.
When the UT was analyzed, a distinctive microbiome structure was observed depending on the disease subtype (Fig. 2B). PERMANOVA using distance matrices revealed a signi cantly different microbial composition between the UT from patients with CRSwNP (UT_CRwNP) and those from both control subjects (UT_control, P = 0.006) and patients with CRSsNP (UT_CRSsNP, P = 0.003, Table 2). However, the microbiome of UT_CRSsNP did not signi cantly differ from that of UT_control according to PERMANOVA (P = 0.335). To identify bacterial genera associated with disease subtypes, LEfSe was used to compare bacterial abundances at the genus level. In the UT_CRSwNP, Barnesiella, Terrisporobacter, and Leptotrichia were abundant, whereas the UT_CRSsNP showed higher relative abundances of Sporobacter, Fusobacterium, and Oscillibacter. In the UT_control, Ruminococcus and Paraburkholderia were abundant compared to the CRS group (Fig. 3).

Characteristic dysbiosis in NP: depletion of Prevotella
The microbiomes of NP tissue from patients with CRSwNP were compared to those from the UT_CRSwNP and UT_control groups. The Shannon index was signi cantly decreased in NP compared to UT_ CRSwNP, indicating decreased diversity in the NP (Supplementary Fig S1, Additional File 1). PCoA further revealed a discrete microbiome structure in the NP as compared to the UT_ CRSwNP or UT_control microbiomes ( Fig. 2A). The PERMANOVA values demonstrated a signi cantly different microbial composition in the NP than in the UT_CRSwNP (P = 0.002) and UT_control (P = 0.002) groups (Table 2).
LEfSe identi ed several bacterial genera whose abundances clearly differed between the NP and UT (UT_CRSwNP and UT_control) (Fig. 4). The overall microbial taxa distribution of NP is depicted in Fig. 1. The phylum Firmicutes was remarkably reduced in NP, whereas Proteobacteria were more abundant in NP than in UT. In contrast, Cyanobacteria accounted for less than 23.1% of bacteria in the NP but was less than 0.5% in the UT_CRSwNP and UT_control. Prevotella was the most predominant genus in the UT_CRSwNP (33.0%) and UT_control (32.1%) and was severely depleted to 0.3% in the NP. In addition, Ralstonia, the second most common genus (16.1%), was reduced to 5.9% in the NP. Instead, Sediminibacterium (26.8%), Cyanobacterium_JQ650114 (22.7%), Labilithrix (11.8%), Caulobacter (10.3%), and Sphingomonas (6.5%) were the most abundant genera in the NP tissue. In all, this decrease in microbiome diversity is indicative of a clear dysbiosis in NP.
Microbiome composition is associated with disease severity A simple linear regression was performed to identify the genera most associated with disease severity using the LM score (Fig. 5). The genera most associated with disease severity differed between tissue and disease subtypes. Prevotella was the genus most strongly inversely correlated to disease severity in NP tissue (Rho = -0.771, P = 0.025). However, Lachnospira (Rho = 0.745, P = 0.034) and Roseburia (Rho = -0.971, P = 0.001) were most associated with disease severity in the UT_CRSwNP and UT_CRSsNP, respectively. Several other genera in NP and UT associated with disease severity are summarized in Supplementary

Discussion
This study aimed to characterize the microbiome signature of sinonasal tissues from healthy controls, subjects with CRSsNP, and those with CRSwNP using different collection methods (i.e. nasal swab vs. tissue biopsy) and tissue types (UT vs. NP). We demonstrated that tissue samples may be more optimal specimens for assessing the microbiome of CRS than are nasal swabs. Tissues provided less variation between samples; moreover, in UTs different microbial compositions were associated with disease subtype. In addition, NP tissues had a unique microbiome that was clearly differentiated from UT and revealed a strong association with clinical severity of CRSwNP. To our knowledge, this is the rst study to use UT to assess the microbiome of patients with CRS.
An advantage of using UT is the relative simplicity of sample collection without causing signi cant morbidity. Therefore UT can be easily acquired regardless of disease status. The immunological pro le of UT has been demonstrated to differ according to disease status. For instance, the eosinophil cationic protein (ECP) level in UT is strongly correlated with overall disease severity, comorbid asthma, and the risk of polyp recurrence (25,26). Furthermore, speci c immunological characteristics of UT in different disease subtypes (6, 27) seem to be associated with the distinguishing microbiome pro le.
Recent studies of CRS have found dysbiosis of nasal microbiota. Despite an overall decrease of microbial richness, evenness, and diversity in CRS, inconsistencies or biases in previous studies limited the identi cation of clinically relevant genera or species (14,28,29). Furthermore, an optimal sampling method of the nasal microbiome was not yet determined. To date, most studies have utilized nasal swabs, primarily collecting from the middle meatus. The middle meatus is an intersectional area that shares a common drainage pathway with other paranasal sinuses. Therefore, samples from this area may re ect multiple sites, which could produce increased heterogeneity. However, another study comparing microbial compositions from swab samples collected at different anatomical sites from patients with CRS still demonstrated a high interpersonal variation, which outweighed location-speci c differences. 11 Our study found similar results, where a high interpersonal variability of swab samples may obscure the differences among disease subtypes.
Tissue samples from the UT and NP were superior to swabs and demonstrated decreased interpersonal differences within subgroups. This allowed for the identi cation of discrete tissue-associated microbiomes. The high variation in swab samples could be a function of the vulnerability of the nasal surface microbiome to environmental changes.
Our results are further supported by a previous study that found signi cantly different bacterial compositions in tissue specimens as compared to swab samples (30). Tissue specimens are more inclusive than swabs, as they incorporate bacterial bio lms that grow on the surface and bacteria that penetrate the mucosal epithelium. This is supported by the observation that tissue samples have a greater biomass than swabs, although non-microbial DNA is present (31). Thus, tissue samples should be given greater attention in the future as a collection method.
In CRS, the airway microbiome composition is associated with in ammatory pro les. For example, CRSwNP accompanied by asthma presented with signi cantly higher levels of Th2-related cytokines.
Simultaneously, there was a higher abundance of Proteobacteria than in patients with CRSwNP without asthma. Furthermore, Corynebacterium, which are members of the phylum Actinobacteria, were more abundant in CRSwNP without asthma (32). There is an upregulation of several important cytokines, such as IL-5, in NP compared to UT regardless of disease subtype (6, 27). Therefore, the distinct microbial composition of NP compared to that of UT in our study could be associated with differences in in ammatory pro les.
In the NP, the relative abundance of the phylum Firmicutes was remarkably lower than in the UT, whereas the abundance of the phylum Proteobacteria was higher. Despite the use of different sampling methods, the higher abundance of Proteobacteria than that in controls has been consistently reported. At the genus level, Haemophilus, Escherichia, and Moraxella were highly enhanced in CRSwNP as compared to controls (33)(34)(35). Furthermore, in our study, Sphingomonas, a genus in the phylum Proteobacteria, was signi cantly enriched in the NP as compared to in the UT. In a murine asthma model, glycoproteins from Sphingomonas induced type 2 in ammation via natural killer T cells in an IL-4-, IL-13-, and IL-33dependent manner (36, 37). Interestingly, Sphingomonas was found to be more abundant in bronchoalveolar lavage uid from patients with eosinophil-high asthma than in those with eosinophil-low asthma; this is also related to increased airway hyperresponsiveness to methacholine (38, 39).
We also evaluated the severity of CRS using LM scores, a widely used radiological parameter assessed by CT. The score increases with certain markers of disease severity, such as the increasing grade of polyposis, nature of surgery offered (i.e. more extensive surgery), and treatment outcome (16). Correlation between genera and disease severity differed by both tissue and disease subtype. This indicates unique interactions of microbiome according to the different microenvironmental conditions.
In the NP microbiome, the genus Prevotella was signi cantly and inversely correlated with disease severity. Similar ndings have been observed in other in ammatory diseases including multiple sclerosis (MS). In that study, the abundance of gut Prevotella was reduced in untreated patients with MS, and treatment with disease-modifying therapy was associated with an increased relative abundance of Prevotella (40). This protective effect of Prevotella appears to be mediated through the induction of CD4 + FoxP3 + regulatory T cells (41).
There were several limitations of our study. First, the number of patients was rather small, considering the heterogeneity of CRS. Second, we did not analyze the differences between eosinophilic NP and noneosinophilic NP, which are known to exhibit immunological differences (27). Third, 16S rRNA analysis only reveals the presence of bacteria, but it does not analyze their metabolic activity. Fourth, tissue samples dominated by non-microbial DNA can introduce contamination during amplicon sequencing. Therefore, further validation studies with larger sample sizes are necessary. Future studies should also identify interactions between the tissue microenvironment and microbes in association with immunological pro les.

Conclusion
Tissue microbiomes in CRS reveal highly clustered pro les with low inter-personal differences, with changes are associated with disease status. Most importantly, we identi ed a distinct NP microbiome composition as compared to UT. Therefore, tissue samples are encouraged to be used for future evaluation of CRS microbiomes. In all, the altered microbial habitat and strong association with CRS disease severity suggests a distinct tissue microenvironment that requires further study. Availability of data and materials

List Of Abbreviations
The datasets generated and/or analysed during the current study are not publicly available due to legal reason in Korea, but are available from the corresponding author on reasonable request.    Bacterial genera abundances in uncinate tissue differs with disease subtype. Linear discriminant analysis effect size (LEfSe) was used to determine the most abundant genera in CRSwNP, CRSsNP, and controls. Only genera with P < 0.05 and LDA score > 2 are shown. Abbreviations: CRSwNP, chronic rhinosinusitis with nasal polyp; CRSsNP, chronic rhinosinusitis without nasal polyp.

Figure 4
Linear discriminant analysis (LDA) demonstrates distinct bacterial genera between the nasal polyp and uncinate tissue. A, Nasal polyp vs. uncinate tissue from patients with chronic rhinosinusitis with nasal polyp (UT_CRSwNP) and B, nasal polyp vs. uncinate tissue from controls (UT_control).