M﻿etagenomic insights into the microbial communities of inert and oligotrophic outdoor pier surfaces of a coastal city

Background Studies of the microbiomes on surfaces in built environment have largely focused on indoor spaces, while outdoor spaces have received far less attention. Piers are engineered infrastructures commonly found in coastal areas, and due to their unique locations at the interface between terrestrial and aquatic ecosystems, pier surfaces are likely to harbor interesting microbiology. In this study, the microbiomes on the metal and concrete surfaces at nine piers located along the coastline of Hong Kong were investigated by metagenomic sequencing. The roles played by different physical attributes and environmental factors in shaping the taxonomic composition and functional traits of the pier surface microbiomes were determined. Metagenome-assembled genomes were reconstructed and their putative biosynthetic gene clusters were characterized in detail. Results Surface material was found to be the strongest factor in structuring the taxonomic and functional compositions of the pier surface microbiomes. Corrosion-related bacteria were significantly enriched on metal surfaces, consistent with the pitting corrosion observed. The differential enrichment of taxa mediating biodegradation suggests differences between the metal and concrete surfaces in terms of specific xenobiotics being potentially degraded. Genome-centric analysis detected the presence of many novel species, with the majority of them belonging to the phylum Proteobacteria. Genomic characterization showed that the potential metabolic functions and secondary biosynthetic capacity were largely correlated with taxonomy, rather than surface attributes and geography. Conclusions Pier surfaces are a rich reservoir of abundant novel bacterial species. Members of the surface microbial communities use different mechanisms to counter the stresses under oligotrophic conditions. A better understanding of the outdoor surface microbiomes located in different environments should enhance the ability to maintain outdoor surfaces of infrastructures. Video Abstract Supplementary Information The online version contains supplementary material available at 10.1186/s40168-021-01166-y.


Introduction 48
The indoor and outdoor surfaces of the built environment are reservoirs of microbial 49 assemblages. The microbial communities on indoor surfaces are influenced by geographical 50 location [1], building function [2], building design [3], cleaning practices [4], human 51 occupancy [5], and occupant activities [6]. Indoor surfaces not only passively receive microbes, 52 but also facilitate microbial growth when moisture is available [7]. In an occupied indoor space, 53 different surface types harbor distinct microbial communities [8], which is largely due to 54 contact by occupants and the subsequent transfer of microbes [9]. The interactions between 55 microbes and surfaces are affected by many factors such as surface hydrophobicity, charge, 56 topography, and other physicochemical attributes [10][11][12][13]. Consequently, the abundances of 57 specific taxa differ depending on the type of surfaces and materials [14,15]. Similarly, the 58 metabolic functions of surface microbial communities and the synthesized metabolites also 59 vary by surface type [8,14] and materials [7]. 60 Unlike indoor surfaces, outdoor surfaces are often exposed to uncontrolled and harsh 61 environmental conditions, such as intense ultraviolet light, fluctuating temperature, desiccation, 62 and poor nutrient supply. These conditions and stressors not only induce esthetic deterioration 63 of the surfaces (e.g., corrosion), but also threaten the survival of microbial residents [16]. 64 However, some of the microbes on outdoor surfaces can adapt to and survive such stresses by the most important parameter driving the compositional differences between surface 148 microbiomes ( Fig. 1f), followed by surface type (pseudo-F = 21.55, R 2 = 0.06, p = 0.001) and 149 sampling location (pseudo-F = 18.21, R 2 = 0.21, p = 0.001). Overall, these results highlighted 150 the importance of surface attributes in structuring the diversity and composition of pier surface 151 microbiomes. 152 153 Diverse species contributed to functional shifts of the surface microbiomes 154 The functional composition of surface microbiomes was found to be significantly 155 correlated with the species-level taxonomic composition (Procrustes test, p = 0.001, correlation: 156 0.7464), suggesting that samples with a similar taxonomic composition tended to have a similar 157 functional composition. Because material was the strongest factor in structuring the 158 composition of surface microbiomes, the species-level contributions to the functional shifts of 159 surface microbiomes were further quantified for the respective concrete and metal surface 160 microbiomes. Twenty-eight and 69 metabolic pathways were found to be significantly enriched 161 on metal and concrete surfaces, respectively, with the majority of them encoding housekeeping 162 functions. Interestingly, pathways related to energy metabolism and xenobiotics 163 biodegradation and metabolism were also enriched. 164 Notably, for metal surfaces, Micrococcus luteus, Bacillus cereus, and a few 165 Bradyrhizobium species were the major drivers of enrichment of xenobiotics biodegradation 166 pathways (Additional file 5: Figure S2). For concrete surfaces, the enrichment of xenobiotics 167 biodegradation pathways was driven predominantly by species including Deinococcus sp. 168 Strain NW-56 and three stone-dwelling actinobacteria including Blastococcus saxobsidens, 169 Modestobacter marinus, and Geodermatophilus obscurus [33]. In addition, diverse species on 170 concrete surfaces were associated with energy metabolism pathways, with a few cyanobacterial 171 species involved in photosynthetic carbon fixation pathways, consistent with their 172 photosynthetic physiology [34]. 173 174 Surface material determined the functional variations of pier surface microbiomes 175 To understand how the microbiomes adapted to the oligotrophic conditions of pier 176 surfaces, the functional profiles of contigs in each sample were characterized following read 177 assembly. Functional gene annotation based on the cluster of orthologous group (COG) 178 categories revealed differential enrichment of genes in metagenomes from different surface 179 materials (Additional file 6: Figure S3). Furthermore, we hypothesized that microbiomes from the same material possessed 189 similar gene repertoires. To test this hypothesis, a two-way hierarchical clustering analysis was 190 performed on the Jaccard distance index between the samples regardless of surface material 191 and type (Fig. 2a). The clustering resulted in four distinct gene clusters (Fig. 2b). Cluster A 192 was dominated by genes sourced from concrete surfaces (66 out of 77), cluster B mostly 193 contained genes from metal surfaces (63 out of 69), cluster C contained genes derived from 194 three bollard samples at a single location, and cluster D comprised genes from 26 metal samples 195 from six locations. The hierarchical clustering results were further supported by the supervised 196 random forest classifier, which yielded an overall out-of-bag error score of 3.4%. These results 197 suggest that surface material regulated the functional traits of surface microbiomes, with 198 microbes from the same material possessing similar metabolic functions. 199 The clear separation of cluster D from the other clusters suggested the presence of a 200 unique gene repertoire (Fig. 2b) Figure S4). Previous studies have shown that bacterial genomes 206 containing specific functional gene inventories enable their survival in particular ecological 207 niches [35]. Therefore, the group of genes in cluster D may confer beneficial adaptive functions 208 on the microbes residing on those particular metal surfaces. 209 Because the biodegradation pathways for a few xenobiotics were differentially enriched 210 between concrete and metal surfaces (Additional file 5: Figure S2), we further queried whether 211 KEGG Orthology (KO) identifiers associated with xenobiotics biodegradation and metabolism 212 pathways differed between surface materials. Two hundred and fifty-one KOs encoding 213 xenobiotics metabolism were identified in the pier metagenomes, with significantly more KOs 214 identified on concrete than on metal surfaces (average 86 vs. 30, MW test, p = 4.2 × 10 -17 ). In 215 addition, KOs encoding xenobiotic metabolism tended to be compositionally more similar in 216 microbiomes from the same than from different surface materials (Additional file 8: Figure S5). 217 Overall, these results further highlight the role of surface material in determining microbial 218

functions. 219
Trace metals such as iron are crucial for the survival of microorganisms [36]. As the 220 metal surfaces sampled contained iron, and given the result that the abundance of genes 221 involved in inorganic ion transport and mechanism [P] differed significantly between the 222 different surface materials and types (Additional file 6: Figure S3), we investigated whether 223 such differences could be reflected in genes related to iron metabolism. Metal surfaces 224 contained a higher relative abundance of genes involved in iron acquisition (MW test, p = 0.03), 225 while genes involved in iron regulation and storage were significantly more abundant on 226 concrete surfaces (p = 0.004 and 1.66 × 10 -6 , respectively) (Additional file 9: Figure S6a). 227 Differences in the composition of iron-related protein families between the concrete and metal 228 surface metagenomes were also found (pseudo-F = 39.71, R 2 = 0.12, p = 0.001, Additional file 229 9: Figure S6b). The iron metabolism analysis further reinforced the notion that differences in 230 microbial functions are based on surface material. However, a relatively low abundance of 231 iron-related genes was found in all of the surface metagenomes, with 15 samples even 232 completely lacking these genes, suggesting that iron metabolism is not a major function in the 233 communities, even on metal surfaces. The low relative abundance of genes encoding iron 234 oxidation on both materials is consistent with the low relative abundance of iron-oxidizing 235 bacteria (IOB) (Fig. 1c). Functional differences of all of the MAGs between either surface type or geographical 255 location were significant in only half of the COG categories (Additional file 11: Table S4). In 256 contrast, significant differences between phyla were detected in all COG categories except for 257 [Z] (KW test, p < 0.05 for all significant comparisons) (Additional file 12: Figure S8 Because the biosynthetic potentials differed between phyla, we further queried whether 282 such differences could be extended to the strain level of a species. To address this question, 283 MAGs classified as the same species were further dereplicated at 99% ANI to differentiate 284 strains [38]. MAGs that could not be classified to the species level were also dereplicated based 285 on the lowest taxonomic rank established. At an ANI threshold of 99%, MAGs regarded as the 286 same strain tended to be detected on the same surface type and material across geographically 287 separated locations (Additional file 14: Figure S9a). Only minor differences in the number and 288 type of putative BGCs were found between genetically highly similar strains (Additional file 289 14: Figure S9b). These results suggest that biosynthetic potentials are largely dictated by 290 taxonomy and to a lesser extent by surface type, surface material, and geography. 291

Discussion 292
The microbiomes of outdoor surfaces are largely unexplored compared with those of 293 indoor surfaces [6,8,40,41]. As an open system at the interface between marine and terrestrial 294 ecosystems, coastal surfaces are a unique habitat that harbors interesting microbiology. In this 295 study, we characterized the taxonomic profile and functional traits of coastal pier surface 296 microbiomes at both the community and genome levels. The results have shed light on the 297 microbes that are present, the metabolic functions that may facilitate adaptation of these 298 members to the harsh environmental conditions, and the parameters that are associated with 299 taxonomic and functional variations. 300 301

Surface materials drive taxonomic variations and functional shifts in pier microbiomes 302
Although the surfaces studied here are inert and oligotrophic, surfaces made of different 303 materials inherently vary in their micro-environmental characteristics, such as pH, structure 304 (e.g., cracks for protection against predation), and nutritional availability [42]. From an 305 ecological perspective, the differences between concrete and metal surfaces may impose 306 different stresses on the microbial colonists, resulting in different microbes colonizing different 307 surfaces depending on their ability to adapt to the surface materials. Therefore, the taxonomic 308 composition of the pier surface microbiomes is largely governed by material, and microbial 309 communities that are functionally more similar tend to originate from the same surface material. The compositional differences between metal and concrete surface microbiomes imply 318 the differential enrichment of taxa between the two materials. In this study, preferential 319 enrichment of microbial taxa across surface type and material was observed. Members of 320 Cyanobacteria were particularly more abundant on the floor than on other surface types, which 321 is consistent with the floor being most susceptible to be imprinted by aquatic taxa from 322 seawater spray. In addition, Cyanobacteria species are known to be prevalent in aerosols above Reads in the surface samples that could be mapped to the contigs in the negative controls were 467 removed using an in-house script, and any unpaired reads were further removed from the 468 paired-end fastq files using fastq-pair (https://github.com/linsalrob/fastq-pair). The algorithm 469 decontam (https://github.com/benjjneb/decontam) executed using the default mode was further 470 applied to evaluate contamination after the read removal procedures. The contaminating 471 species identified were manually curated; however, it was not necessary to remove all of them 472 because their relative abundance was low or they were genuine taxa in the surface samples.

Alpha-and beta-diversity analyses 496
For the taxonomic alpha-diversity analysis, clean paired-end sequences were rarefied 497 to 1.0 million reads per sample using the "seqtk" (v1.3-r106) [75] tool, reducing the dataset 498 from 175 to 155 samples. Although the applied rarefaction depth was not sufficient to capture 499 the richness of surface metagenomes for most samples (Additional file 16: Figure S11), 500 principal coordinate analysis indicated that the adopted depth could still recapitulate the 501 compositional differences between microbial communities (Additional file 17: Figure S12). At 502 the species level, the abundance-based Shannon diversity index was calculated using the 503 function "diversity" in the R (v3.6.1) package "vegan" (v2.5-6) [76]. The Bray-Curtis 504 dissimilarity metric was calculated for the species-level taxonomic composition (unrarefied 505 dataset) using the function "vegdist" in the R package "vegan. The Mann-Whitney and Kruskal-Wallis tests were performed to test the statistical 563 significance involving two and more than two groups using the "wilcox.test" and "kruskal.test" 564 functions of the R package "stats" (v3.6.1), respectively. The post-hoc Kruskal-Wallis test was 565 performed using the "kruskalmc" function of the R package "pgirmess" (v1.6.9) [93]. Because 566 only a single MAG was reconstructed from the phyla Acidobacteria, Firmicutes, and 567 Gemmatimonadetes, they were excluded from the Kruskal-Wallis test when testing the 568 statistical significance of the antiSMASH results. The Procrustes test was preformed using the 569 "protest" function in the R package "vegan" with 999 permutations. 570 A stepwise model selection scheme based on the Shannon diversity index was applied 571 to identify factors that significantly affected the within-sample diversity of surface 572 microbiomes. In the analysis, each parameter in the metadata and all possible two-way 573 interactions of these parameters were set as predictors of the Shannon diversity index score in 574 the linear mode using the "lm" function in the R package "stats." A stepwise model selection 575 was applied with the "stepAIC" function in the R package "MASS" (v7.5-51.5) [94] and the 576 model with the lowest Akaike information criterion (AIC) value was considered to be optimal. 577 The significance of each parameter in the optimal linear model was calculated using the "anova" 578 function in the R package "stats." 579 The Bray-Curtis dissimilarity of species-level compositional differences between 580 samples was analyzed by applying the permutational multivariate analysis of variance 581 (PERMANOVA) test using the "adonis2" function in the R package "vegan." All of the 582 parameters in the metadata and all two-way interactions of the parameters were set as predictors 583 of the Bray-Curtis dissimilarity matrix in the linear model for PERMANOVA. The AIC value 584 of the model was calculated by manually removing one variable at a time until the next removal 585 resulted in no increase in the AIC value. 586 587 Acknowledgments 588 We appreciate the assistance provided by all those who assisted with the sample collection. The innermost ring shows the lowest assigned taxonomic rank of the MAGs. The prefix "s" 879 indicates a known species and the prefixes "g" and "f" indicate the lowest possible assigned 880 Figure 1 Composition and diversity of pier surface microbiomes. (a) Top 10 phyla across the four surface types.

Figures
Other phyla were grouped into "Minor/Unclassi ed." (b) The three phyla that were differentially enriched between different surface types and materials. (c) The mean relative abundances of the corrosion-related bacteria identi ed on the metal and concrete surfaces. The full names of the microbial corrosion mechanism abbreviations are indicated in the Materials and Methods section. The Mann-Whitney test was applied to determine the differential enrichment of corrosion-related bacteria between concrete and metal surfaces (***p < 0.001, **0.001 < p < 0.01, *0.01 < p < 0.05). (d) Contributions by local marine and human skin sources to pier surface microbiomes. (e) Shannon diversity of microbiomes across different surface types. (f) Principal coordinate analysis of surface microbiomes based on the species-level abundance matrix ordinated by the Bray-Curtis dissimilarity metric. The normal con dence ellipses indicate the con dence level at 95%.  Phylogenetic tree of the 150 MAGs and the putative BGCs found in each MAG. The innermost ring shows the lowest assigned taxonomic rank of the MAGs. The pre x "s" indicates a known species and the pre xes "g" and "f" indicate the lowest possible assigned taxonomic rank at the genus and family levels, respectively. 881 The MAGs that could not be assigned to a known species are indicated by a red dot. The heatmap shows the number of putative BGCs of each of the top 12 known types detected in each MAG.
All known types of putative BGCs that were present in < 1% of all of the BGCs were grouped into the "Other" category. The total number of putative BGCs in each MAG is indicated by the green bars in the outer ring. BGC types across phylum (left), location (middle), and surface type (right). The total number of putative BGCs in each category is indicated in the brackets. All known types of putative BGCs that were present in < 1% of all of the putative BGCs were grouped into the "Other" category.