Alterations in skin microbiome mediated by radiotherapy and their potential roles in the prognosis of radiotherapy-induced dermatitis: a pilot study

Radiotherapy-induced dermatitis (RID) is an inflammatory cutaneous disorder that is acquired as an adverse effect of undergoing radiotherapy. Skin microbiome dysbiosis has been linked to the outcomes of several dermatological diseases. To explore the skin microbiota of RID and deduce their underlying impact on the outcome of RID, cutaneous microbiomes of 78 RID patients and 20 healthy subjects were characterized by sequencing V1-V3 regions of 16S rRNA gene. In total, a significantly apparent reduction in bacterial diversity was detected in microbiomes of RID in comparison to controls. Overall, the raised Proteobacteria/ Firmicutes ratio was significantly linked to delayed recovery or tendency toward the permanence of RID (Kruskal Wallis: P = 2.66 × 10–4). Moreover, applying enterotyping on our samples stratified microbiomes into A, B, and C dermotypes. Dermotype C included overrepresentation of Pseudomonas, Staphylococcus and Stenotrophomonas and was markedly associated with delayed healing of RID. Strikingly, coexistence of diabetes mellitus and RID was remarkably correlated with a significant overrepresentation of Klebsiella or Pseudomonas and Staphylococcus. Metabolic abilities of skin microbiome could support their potential roles in the pathogenesis of RID. Cutaneous microbiome profiling at the early stages of RID could be indicative of prospective clinical outcomes and maybe a helpful guide for personalized therapy.

Bioinformatic analyses of row reads of 16S rRNA. The analysis and classification of 16S rRNA reads were mainly based on ribosomal sequence variants (RSV). QIIME2 platform with DADA2 plugin was used for sequence preprocessing 20 , which included: trimming, filtering and denoising of row reads as previously described (maxEE = 2; truncation length for forward and reverse reads 270 and 210, respectively; Phred quality ≥ 25) 21 . Finally, high-resolution RSVs were generated by removing the potential chimeric sequences.
RDP's naive Bayesian classifier was applied on representative RSVs for taxonomy assignment against SILVA SSU Ref NR dataset v.132 at 99% sequence similarity 22,23 .
Bacterial diversity was estimated using QIIME2 on the entire dataset without subsampling. Alpha diversity was determined using two approaches; regarding richness (number of observed species and Chao1 index) and evenness of communities (Shannon diversity index). The statistical significance of pairwise comparisons was assessed using unpaired Wilcoxon rank sum test, while multiple group comparisons were estimated using Kruskal-Wallis rank sum test and all P values were corrected using false discovery rate method (FDR) 24 . The core genus in our dataset was defined as the genus that was detected in 80% of all samples, while the core genus for each study group was defined as the genus that was present in all samples of this group.
Phylogenetic relatedness of bacterial communities was identified using both unweighted and weighted Uni-Frac matrices 25 . Similarity distances between cutaneous microbiomes were visualized by principal coordinates analysis (PCoA) based on weighted UniFrac distance matrix. Permutational Multivariate Analysis of Variance (Adonis R, package Vegan) was applied to define the significance of weighted UniFrac distance-based clustering of samples 26 .
To determine the dermotypes of cutaneous microbiomes, a widely used enterotyping pipeline was applied to all genera in our dataset. Clustering of samples and the optimum number of clusters were achieved using partitioning around medoids (PAM) algorithm, Jensen-Shannon divergence (JSD) and Calinski-Harabasz (CH) index 27 .
For predicting the functional profile of cutaneous microbiota, representative RSVs were employed for further taxonomic classification using UCLUST 28 against Greengenes database (V 13.8) 29 at a 97% identity using closed-reference script for OTU picking in QIIME, v 1.9.1 29 . Functional potential of cutaneous microbiomes was predicted using Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PIC-RUSt) for mapping the Kyoto Encyclopedia of Genes and Genome (KEGG) Orthology (KO) Database at level 3 [30][31][32] . LEfSe was applied to the metagenomic data to define the taxon or metabolic pathway (logarithmic LDA scores > 3.0 and α = 0.05) that were differentially abundant among healthy subjects and patients with RID (https ://hutte nhowe r.sph.harva rd.edu/lefse /) 33  www.nature.com/scientificreports/ Correlations between either communities' members or structure of microbiota and patient's clinical data were defined by Spearman correlation coefficient (r ≥ ± 0.6, P ≤ 0. 01) using R package, Hamsic 34 . R packages; Phyloseq and ggpubr, were applied for calculation and plotting of diversity indices (https ://rpkgs .datan ovia.com/ ggpub r/) 35

Results
Patient's characteristics. Ninety-eight subjects were included in this study (20 controls and 78 patients with RID). Patients' data regarding clinical characteristics, demography and outcome were listed in Table 1 (Supplemental Table S1).
Preprocessing and analyses of 16S rRNA sequences. In total, 11,276,243 row sequences (average reads per sample = 11,564) were inputted in Qiime2 for merging, quality checking, removal of low quality reads (1,405,020 sequences, 12.46% of all dataset) and potential chimeric sequences (587,492 sequences, 5.21% of all dataset) that were resulted in 9,283,731high quality sequences that were used for downstream analyses.
Distinct taxonomic profile of cutaneous microbiomes related to RID. 16S rRNA reads were taxonomically assigned to 19 phyla, 58 classes, 109 orders, 237 families, 658 genera and 4643 OTUs. In total, cutaneous microbiomes associated to RID showed a remarkable representation of certain bacterial taxa at different taxonomic levels that was in contrast to healthy subjects (Fig. 1). Phylum level analyses of RID microbiomes exhibited alternately significant predominance of Firmicutes and Proteobacteria (Mean relative abundance ± SD: 55.39 ± 25.07% and 32.45 ± 19.79%, respectively) in addition to relatively constant proportions of Actinobacteria (Mean relative abundance ± SD: 7.15 ± 5.18%). Classification of RID samples to seven groups, regarding the time required for recovery of RID, revealed significant representation and coexistence of Proteobacteria and Firmicutes (Kruskal Wallis; P = 2.58 × 10 -7 : Spearman; r = -0.68, P = 0.001) (Supplemental Table S2). Microbiomes associated to rapidly healed RID (2, 3, 4 weeks) were significantly accompanied by the predominance of Firmicutes over Proteobacteria (Mean relative abundance: 52.2% and 29.42%, respectively; Kruskal Wallis: P = 5.71 × 10 -4 ). In contrast, microbiomes of chronic ulcers were significantly predominated by Proteobacteria Age(mean ± SD) 49 Reduced bacterial diversity characterize the shifts in cutaneous microbiomes of RID. Identification of bacterial diversity associated with RID was performed using both evenness and richness indices. Skin microbiomes of healthy subjects were significantly more diverse than those with RID (Wilcoxon test: P = 0011.47 × 10-5) (Fig. 2). Surprisingly, cutaneous microbiomes of RID showed a significant reduction in bacterial diversity in conjugation with a prolonged duration of RID (Kruskal Wallis, P = 0.00047). Cutaneous microbiomes of RID that turned to chronic ulcer had significantly reduced bacterial diversity in comparison to either healthy individuals or rapidly healed RID (Kruskal Wallis: P = 0.00093). Both structure and composition of bacterial communities significantly drove the classification of cutaneous microbiomes to health-state based clustering (Fig. 3a). PCoA based on weighted UniFrac metrics have sorted RID associated samples to 7 particular clusters that were mainly based on the healing time of RID (Fig. 3b).
Relative proportions of core genera of healthy cutaneous microbiomes could potentially define the outcome of RID. Studying the structure of RID microbiomes at the family level unable to provide a deep understanding of the potential causative agent of RID. For instance, either Enterobacteriaceae or Staphylococcaceae were the most predominant families in several RID specimens, while genus level analysis was provided a clear separation of genera belonging to the previous families (Fig. 4a). In general, Klebsiella, Cutibacterium, Corynebacterium, Bacillus and Paracoccus were the dominant genera in healthy subjects. On the other hand, microbiomes of RID were generally inhabited with Klebsiella, Staphylococcus or Pseudomonas. Interestingly, microbiomes of RID that turned to chronic ulcers showed significant overrepresentation of Klebsiella or Pseudomonas, Cutibacterium and Stenotrophomonas. Also, the coexistence of Pseudomonas, Staphylococcus and Stenotrophomonas was significantly correlated with shifting of RID toward chronicity (Spearman; r = 0.87, P = 0.001). www.nature.com/scientificreports/ Identification of the core genera of the entire dataset revealed that Klebsiella, Pseudomonas, Cutibacterium, Corynebacterium, Bacillus and uncultured Enterobacteriaceae were detected in all samples with variable abundance (Fig. 4b).
Tendency toward the development of radiotherapy-induced chronic ulcers was significantly linked to Dermotype C. Our genera were stratified into three dermotypes (Table 2). Dermotype C was characterized by the predominance of Pseudomonas, Staphylococcus and Stenotrophomonas over Klebsiella. Of note, although dermotype A in healthy individuals was composed of slightly dominant Klebsiella in addition to Bacillus, Cutibacterium and Staphylococcus, it was characterized by an overrepresentation of Klebsiella in RID. Moreover, both dermotype A in RID and dermotype C were significantly linked to delayed recovery and chronic ulcer formation in RID patients (Table 3). Study groups were plotted on the X-axis and alpha indices were plotted on the Y-axis. The median is defined by the line in each box, the interquartile range (IQR) between the 25th and 75th percentile were delimited by the box, and the range was represented by the whisker. Statistical significance of pairwise comparisons was defined using the nonparametric Wilcoxon rank-sum test. Only significant differences were displayed with either * (p < 0.05), ** (p < 0.01) or *** (p < 0.001) 36  Overall functional potential of the cutaneous microbiome was likely to contribute to the pathogenesis of RID. The collective metabolic functional profile of RID microbiome was significantly associated with the overrepresentation of metabolic pathways that could be attributed to impairments in epidermal integrity, alteration in pH and supporting microbial survival such as lipid metabolism, histidine metabolism, enhanced bacterial invasion of epithelial cells, signaling molecules of bacterial toxins, hedgehog signaling pathway, nitrogen metabolism and peptidases (Supplemental Fig. S1).

Discussion
RID represents a challenge for clinicians, especially it is nearly acquired by all cancer patients who undergo radiotherapy and is regarded as an extra section in healthcare expenditure. Till now, a detailed characterization of microbial contribution to the bioburden of skin disorders is exclusively concerned with studying atopic dermatitis, psoriasis, diabetic foot and burns. Furthermore, the major concern is mainly directed to the culturebased approaches that are restricted to a scanty number of microorganisms, and consequentially override the integral role of the entire community 38,39 .
In accordance with the well-established concepts, concerning alpha diversity, about the evident reduction in the microbial diversity of dermatological disorders such as atopic dermatitis and psoriasis, the microbiome www.nature.com/scientificreports/ of RID showed minimal bacterial diversity in comparison to controls 21,[40][41][42][43] . Also radiotherapy was accompanied by significant reduction in microbial diversity of salivary microbiome 44 . The pathophysiology of RID may include epidermal disturbances, cutaneous injuries, reduced blood supply, impairment in cellular repair and misdirected cutaneous immunity. These factors could be explaining the lowered diversity of RID as a result of predominance of certain opportunistic pathogens over the beneficial ones that subsequently interfere with the synthesis of natural antimicrobial peptides and cutaneous immunomodulation 45,46 . Another remarkable observation is the recovery rates of RID that were positively correlated to bacterial diversity at different taxonomy levels. These observations were matched with the protective roles of cutaneous microbiota in the maintenance of epidermal integrity 45 . The impact of restoring skin microbiome on wound healing has been revealed by a study on the animal model 47 .
It is noteworthy that the UniFrac-based similarity distances between RID microbiomes significantly assorted the specimens to particularly separate patterns depending on the recovery time. A recent study revealed the Colored boxes indicate the core microbiome of healthy and patients with RID, and strikes represent the statistically different genera between study groups. Statistical significance of pairwise comparisons was defined using the nonparametric Wilcoxon rank-sum test. Significant differences were displayed with either ns, p-value > 1; *p < 0.05, **p < 0.01 or ***p < 0.001. (d) Potential biomarkers for each group were inferred using LEfSe and the numbers denote LDA scores. www.nature.com/scientificreports/ significant stable temporal shifts of nasopharyngeal microbiota during the period of radiotherapy treatment 48 . Thus, the composition and structure of microbiomes at beginning of RID could reflect the prospective outcome. Taxonomy profiling of cutaneous microbiota associated to RID highlighted the massive prevalence of Proteobacteria or Firmicutes that is in agreement with the overrepresentation of taxa belonging to them in common skin disorders such acne, atopic dermatitis and psoriasis 21,43,49 .
One of the interesting findings is the significant association between diabetes mellitus and developing a chronic ulcer in RID patients (Spearman: r = 0.86; P = 0.001). Diabetes mellitus is a metabolic disorder that is significantly linked to developing chronic wounds due to the impairments in wound healing phases that provides favorable conditions for either opportunistic residents or pathogenic bacteria 50 . Also, colonization of wounds with certain microorganisms was strongly correlated to the persistence of wound via organism induced body changes that include epidermal barrier integrity, alteration in pH and dysregulation of the immune system [51][52][53] . Overall, developing chronic ulcer in diabetic patients with RID could potentially depend on the colonizing microorganisms through intensifying the epidermal disturbances produced by diabetes mellitus.
The main interesting finding of the present study is the collective correlations between community members. While Klebsiella genus was inversely correlated to other predominant genera in all datasets, Pseudomonas, Staphylococcus and Stenotrophomonas have exhibited a significant coexistence especially in prolonged RID and chronic ulcers (Supplemental Table S3). Although the impact of coinfection with S. aureus and P. aeruginosa on the clinical outcome of cystic fibrosis is controversial, the comparisons with mono-infection of S. aureus or P. aeruginosa showed that the pair coexisting was significantly associated with prolonged hospitalization and exacerbation of the disease 54 . Also, the bidirectional interplay between this pair was manifested in the inhibition of phagocytosis, in addition to the promotion of staphylococcal biofilm formation by P. aeruginosa 54,55 . Additionally, Stenotrophomonas maltophilia was detected in our samples with significant prevalence especially in RID patients. S. maltophilia has emerged as an opportunistic pathogen with global concerns, due to its significant association with the high crude mortality in pneumonia and bacteremia of immunosuppressed patients 56 . S. maltophilia was also reported to cause wound infections, soft tissue infections, bone and joint infections, and cellulitis 57,58 . Interestingly, in our study, the coexistence of S. maltophilia and P. aeruginosa was strongly correlated to develop chronic ulcers in RID. This observation was in agreement with the previously reported cooperative pathogenicity of this pair in cystic fibrosis 59 .
As expected in the cutaneous microbiome, Staphylococcus genus was defined as a core genus in healthy and diseased groups. Although S. aureus was significantly enriched in RID microbiomes, it was overrepresented in the rapidly healed RID. These findings were matched with the previous study which reported that the S. aureus infection was conjugated with limited exacerbations and a short duration of hospitalization 54 .
Notably, K. pneumoniae showed fluctuated representation in our specimens. In microbiomes of RID, K. pneumoniae was significantly overrepresented especially in samples that turned to a chronic ulcer, and its abundance was strongly linked to diabetes mellitus. As it is known that K. pneumoniae have been implicated in liver abscess, meningitis and psoas abscess 60 . Furthermore, K. pneumoniae may harbor diverse virulence characteristics including capsules, multidrug resistance and virulence factors that enhance invasion and promote epidermal deteriorations 61,62 . The coexistence of these characteristics and diabetes mellitus could boost epidermal damage, aggravate the inflammatory processes and retard the repair of the epidermal barrier.
The limitations of this study are the slightly small sample size of each group, dependence on unculturable approach, lack of microbiome characterization at different sampling time and isolation of causative agents in order to define their virulence characteristics including antimicrobial resistance patterns.

Conclusions
We have studied the microbiome of patients with RID in order to accentuate the underlying bacterial structure and to provide new insight about the potential role of microbiota as an influential factor in pathogenesis and clinical outcome of RID as well. A notable positive correlation has existed between bacterial diversity and recovery rates of RID. The overwhelming predominance of Klebsiella or coexistence of Pseudomonas, Staphylococcus and Stenotrophomonas potentially prompt the chronic ulcer formation in RID patients. Cutaneous microbiome profiling at the early stages of RID could be indicative of the prospective clinical outcomes and might be a helpful guide for personalized therapy and management of RID. Finally, shifts in the structure and prevalence of Table 3. Summary of case number in each dermotype.

Dermotype
Control (n = 20) RID (Total) (n = 78) 2 weeks (n = 12) 3 weeks (n = 16) 4 weeks (n = 5) 5 weeks (n = 7) 6 weeks (n = 8) 7 weeks (n = 9) www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.