High Taxonomic Diversity in Ship Bilges Presents Challenges for Monitoring Microbial Corrosion and Opportunity To Utilize Community Functional Profiling

ABSTRACT

One of the key areas in which microbially influenced corrosion (MIC) has been found to be a problem is in the bilges of maritime vessels. To establish effective biological monitoring protocols, baseline knowledge of the temporal and spatial biological variation within bilges, as well as the effectiveness of different sampling methodologies, is critical. We used 16S rRNA gene metabarcoding of pelagic and sessile bacterial communities from ship bilges to assess the variation in bilge bacterial communities to determine how the inherent bilge diversity could guide or constrain biological monitoring. Bilge communities exhibited high levels of spatial and temporal variation, with >80% of the community able to be turned over in the space of 3 months, likely due to disturbance events such as cleaning and maintenance. Sessile and pelagic communities within a given bilge were also inherently distinct, with dominant exact sequence variants (ESVs) rarely shared between the two. Taxa containing KEGG orthologies (KOs) associated with dissimilatory sulfate reduction and biofilm production, functions typically associated with MIC, were generally more prevalent in sessile communities. Collectively, our findings indicate that neither bilge water nor an unaffected bilge from within the same vessel would constitute an appropriate reference community for MIC diagnosis. Optimal sampling locations and strategies that could be incorporated into a standardized method for monitoring bilge biology in relation to MIC were identified. Finally, taxonomic and functional comparisons of bilge diversity highlight the potential of functional approaches in future biological monitoring of MIC and MIC mitigation strategies in general.

IMPORTANCE Microbially influenced corrosion (MIC) has been estimated to contribute 20 to 50% of the costs associated with corrosion globally. Diagnosis and monitoring of MIC are complex problems requiring knowledge of corrosion rates, corrosion morphology, and the associated microbiology to distinguish MIC from abiotic corrosion processes. Historically, biological monitoring of MIC utilized a priori knowledge to monitor sulfate-reducing bacteria; however, it is becoming widely accepted that a holistic or community-level understanding of corrosion-associated microbiology is needed for MIC diagnosis and monitoring. Before biology associated with MIC attack can be identified, standardized protocols for sampling and monitoring must be developed. The significance of our research is in contributing to the development of robust and repeatable sampling strategies of bilges, which are required for the development of standardized biological monitoring methods for MIC. We achieve this via a biodiversity survey of bilge communities and by comparing taxonomic and functional variation.

INTRODUCTION

Microbially influenced corrosion (MIC) has been estimated to contribute between 20 and 50% of the costs associated with corrosion globally (1, 2). In maritime vessels, MIC contributes to economic losses through costs associated with asset repair, maintenance, and downtime (3, 4). Additionally, the potential for MIC-causing bacteria to produce toxic H₂S gasses within confined areas on maritime vessels presents a significant risk to crew safety (5).

One of the key areas for potential MIC attack is the ship’s bilge (6). Bilges collect water that is often contaminated with hydrocarbons (HCs) (e.g., petroleum and hydraulic fluid) and other pollutants in high concentrations. These complex mixtures may provide environments in which the high carbon load or physiological conditions (e.g., anaerobic layers, altered pH) act as a catalyst for microbiological activity and MIC (6). Historically, biological monitoring of MIC-associated bacteria incorporated tests that target sulfate-reducing bacteria (SRB) (7). However, SRB are not the only bacteria involved in MIC, and a single-species focus overlooks the complexity and synergies within bacterial communities, and between microbial communities and the environment, which may contribute to MIC (8–10). In line with this realization, the use of whole-community, next-generation sequencing (NGS) approaches to monitoring MIC-associated ecology is growing in popularity (1, 7, 8).

Recently, strategies that couple NGS of microbial communities with quantitative PCR (qPCR) to develop environment-specific MIC detection packages have been recommended (8). In this approach, NGS is used to identify novel biological markers of MIC that may be specific to a particular system or environment and to design qPCR primers to facilitate quantification of these markers. The approach draws on the strength of both technologies. NGS is a nontargeted approach that can capture information about the total community; however, it only assesses relative changes in microbial abundance and so is not diagnostic. Conversely, qPCR can quantitatively determine bacterial abundance but requires a priori knowledge of which microorganism to target.

Because microorganisms generally respond rapidly to environmental change, biological MIC indicators could provide the first indication as to when and how the environment is changing, which can be harnessed in monitoring and prevention efforts (11, 12). This strategy is in line with the notion of monitoring the environment for conditions that are conducive to MIC, rather than for MIC itself, which has been proposed as a better MIC management strategy, as it would provide early warning of the need to take preventative action before MIC has initiated (3).

The amount of biological variation within and between bilge environments is currently unknown but is a critical requirement for developing standardized protocols for sampling the biology and identifying biological indicators of MIC. For example, understanding biological variation within bilges will help disentangle the trade-off between sampling the most biologically relevant proportion of the community and ease of community sampling. While microorganisms that mediate corrosion typically exhibit a sessile lifestyle, sampling planktonic communities forgoes many technical challenges (13). The ease of planktonic sampling is reflected by the fact that many traditional detection techniques in corrosion prevention target planktonic cells (14, 15). Research in oil and gas pipeline corrosion has already indicated that analysis of the planktonic proportion of the community cannot be used to infer the state or composition of sessile communities (16). Indeed, following global sequencing projects, it is clear that marine biofilms harbor biodiversity that is niche specific and distinct from ocean planktonic communities in general (17). Additional understanding of biological variation within bilges is a critical first step in developing strategies to identify biological markers (e.g., indicator species, genes, or functions) of MIC across bilge environments.

To this end, we used Illumina 16S rRNA gene metabarcoding to examine the biological variation in bacterial communities associated with the bilges of Australian maritime vessels at multiple spatiotemporal levels. We hypothesized that bilge water communities would be inherently different from sessile communities due to planktonic and sessile habitats representing different ecological niches, which likely vary in nutrient and oxygen availability. Conversely, we hypothesized that bilges from within the same vessel would exhibit similarities due to shared inputs and temperature fluctuations. We compared two sessile community sampling strategies (swabbing and brushing) at multiple locations within individual bilges to determine where and how communities should be sampled in order to yield biological information most relevant to the study of MIC. We hypothesized that there would be a detectable stratification of the sessile community in wet bilges and that functions pertaining to MIC would be most prevalent at the waterline (WL) or below it. This research contributes essential basic knowledge regarding the amount of natural variation within and between bilge microbial communities, which can be used to identify the most appropriate sampling locations and methods for standardized MIC field sampling protocols.

RESULTS

After quality filtering and denoising, bilge communities (pelagic and sessile) contained 42,848 ± 1,831 reads on average. Four samples containing far fewer reads (ship A1_port_bilge water-1, ship A1_starboard-brushed_B-3, ship B_starboard_swabbed A-1, ship A2_starboard_swabbed_A-2) were excluded from downstream analysis. DNA extracted from brush and swab no-template controls was below detection limits.

There is high biological variation between bilges and within the same bilge over time.

Alpha diversity analyses of pelagic and sessile communities indicated that there was variation in the gross community structure between bilges and within the same bilge sampled over time (port A1 and port A2) with significant differences for Shannon-Wiener diversity as measured by the Kruskal-Wallis H statistic and ANOVA F statistic where appropriate (H_planktonic = 12.2, df = 4, P = 0.016; H_sessile = 20.3, df = 4, P < 0.001), exact sequence variant (ESV) richness (H_planktonic = 10.5, df = 4, P = 0.031; H_sessile = 17.7, df = 4, P = 0.001), and community evenness (F_4,9 = 9.591, P = 0.003; H_sessile = 21.6, df = 4, P = 0.002) among bilges (Fig. 1). In general, communities from the water of ship C, which was of a different class of ship compared to all others, were the least diverse. Despite being sampled from the same vessel at the same time, sessile communities from dry bilges (port A1 and starboard A1) had significantly different diversity and community evenness. Compared to pelagic communities, the ESV richness among sessile communities was generally lower and more consistent among bilges.

FIG 1.

Shannon-Wiener diversity (A and D), Chao1 ESV richness estimates (B and E), and Simpson community evenness (C and F) for bilge-water-associated (A to C) and sessile (D to F) bacterial communities. Sessile community data are derived from brushed samples. Boxes with different letters are significantly different (P < 0.05); n = 3.

Clustering analysis demonstrated that both bilge water communities and sessile communities were distinct (Fig. 2). Despite being sampled at the same time and from the same vessel, the bilge water communities collected from port and starboard bilges of ship A2 had fewer than 25% of ESVs in common, while associated sessile communities had fewer than 25% of ESVs in common. Planktonic communities from the same bilge sampled 3 months apart (port A1 and port A2) were close to 100% dissimilar, suggesting a complete turnover of the community composition within the space of a few months. Except for bilge port A1, a dry bilge, the within-bilge sampling locations formed distinct clusters within every bilge. These data demonstrate the assemblage of distinct communities above, on, and below the bilge water interface.

FIG 2.

Clustering analysis for bilge-water-associated (A) and sessile (B) bacterial communities using Bray-Curtis distances. For sessile communities, data are derived from brushed sampling and within-bilge locations are denoted in tip labels. A, above waterline; WL, on waterline; B, below waterline.

We observed a high variation in the dominant taxa present in each bilge community, with pelagic and sessile communities within a given bilge typically dominated by different genera (Fig. 3). Dominant pelagic ESVs were distinct among ships and over time: Arcobacter was dominant in ship A at the first sampling time (A1) while Alteromonas was dominant at the second (A2), and pelagic communities from ship B were dominated by Novosphingobium. Oleibacter was highly prevalent within the sessile communities from ship A at the second sampling time point (A2), and Marinobacter dominated the sessile communities of ship B. Ship C was excluded from these analyses as no sessile community data could be collected from its bilge.

FIG 3.

Descriptive bar charts of dominant genera within each bilge. Where ESVs could not be assigned a genre, family level classifications are indicated. For each bilge, communities from the pelagic phase (WATER) and sessile phase sampled from either above the waterline (ABOVE), on the waterline (WATERLINE), or below the waterline (BELOW) are displayed. Sessile community data are derived from samples collected via the brush method. A total of 822 genera, each with a relative abundance of <0.3%, have been excluded for clarity. Blank columns represent data missing due to the port and starboard A1 bilges being dry (see Table 1). Note that the pelagic WATER community from port A1 was collected from a distinct but connected bilge section, which still contained water.

Sampling technique and sampling location impact the community detected.

Differences between within-bilge community structure were statistically tested in conjunction with the effect of sampling strategy (swabbed versus brushed) via permutational multivariate analysis of variance (PERMANOVA). For the two dry bilges (port A1 and starboard A1), there was no significant impact of sampling strategy on abundance weighted or unweighted community structure. For one of the dry bilges (starboard A1), community structure differed significantly between waterline and below-waterline locations for both abundance weighted (pseudo-F_(1,7) = 24.35, P = 0.003) and unweighted (pseudo-F_(1,7) = 2.76, P = 0.002) analyses, suggesting a change in dominant species as well as the suites of ESVs present despite the absence of bilge water. In this bilge, Thalassolituus accounted for ∼50% of the sessile community sampled on the waterline, which was detectable as a tidal mark in the absence of bilge water, while Sulfurospirillum accounted for ∼35% of communities sampled below the waterline (Fig. 3).

A significant interaction between sampling location and sampling strategy was detected for abundance weighted community structure for all three wet bilges (port A2, pseudo-F_(2,12) = 2.88, P = 0.02; starboard, pseudo-F_(2,11) = 5.13, P = 0.002; starboard B, pseudo-F_(2,10) = 8.90, P = 0.001). Additionally, both sampling location and sampling strategy were significant main effects.

Clustering analysis revealed that across all three wet bilges, swabbed and brushed samples from above the waterline were more similar to each other than they were to swabbed and brushed samples collected from below the waterline (Fig. 4). For ship B (Fig. 4C), there was a clear clustering of swabbed samples from the waterline and below-waterline locations with bilge water. This clustering was not present for brushed samples, which remained in distinct location clusters. With the exception of this trend, clustering analysis demonstrated that pelagic communities within each bilge were inherently different from their respective sessile communities.

FIG 4.

Clustering analysis (weighted UniFrac metric) of bacterial communities sampled within wet bilges from ship A2 port (A) and starboard (B) sides and ship B (C). Tip labels are indicative of sampling method and location and includes planktonic communities (WATER) as well as sessile communities sampled by brushing or swabbing from above the waterline (BRUSH_A, SWAB_A), on the waterline (BRUSH_WL, SWAB_WL), or below the waterline (BRUSH_B, SWAB_B).

Bilge water communities are not predictive of sessile corrosion-associated taxon abundances.

The distribution of taxa that were predicted to have corrosion-related functionality varied across the bilge surface and between sessile and planktonic communities (Fig. 5). Although there are exceptions, taxa containing KEGG orthologies (KOs) associated with dissimilatory sulfate reduction were generally more prevalent in sessile communities. Based on our taxonomic analysis, taxa canonically identified as SRB, including Sulfurospirillum and Desulfobacter, were only appreciably abundant in bilges from ship A1, both dry bilges, with significant abundances in the starboard bilge (Fig. 3). However, analyses of functional gene distribution reveal that potential SRB capable taxa were present in considerable abundance in every bilge. A total of 715 ESVs representing 106 genera were predicted to contain dissimilatory sulfate-reducing genes. Only 30% of these (214 ESVs) belonged to genera or higher classifications that are routinely described as containing SRBs (see Table S1 in the supplemental material).

FIG 5.

Bar charts describing the distribution of ESVs identified as having sulfate reduction, biofilm formation, or hydrocarbon (HC) degradation functional potential. Blank columns represent data missing due to the port and starboard A1 bilges being dry (see Table 1). Note that the pelagic WATER community from port A1 was collected from a distinct but connected bilge section, which still contained water.

In general, biofilm-forming taxa were also more readily detected in sessile communities and, within a given bilge, had their highest abundances recorded in the waterline communities. Interestingly, there was no clear correlation between the abundance of biofilm-forming taxa and potential sulfate-reducing taxa. In contrast to biofilm production and sulfate reduction, hydrocarbon degradation potential was a prevalent feature of all bilges and bilge locations (Fig. 5).

DISCUSSION

Understanding levels of inherent microbial community variation is critical for developing biological monitoring protocols in environments in which microorganisms cause economic losses or occupational hazards (5). Using a 16S rRNA gene metabarcoding approach, this study investigated the biological variation between and within ship bilges with the aim of establishing sampling protocols to be used in MIC monitoring and diagnosis.

High levels of inherent biological variation among bilges impedes the monitoring of MIC-related taxonomic changes.

Our combined alpha- and beta-diversity analyses revealed high levels of biological variation among the individual bilges for both pelagic and sessile communities. This finding sits in contrast to our hypothesis that bilges from the same ship will harbor similar communities due to a shared operational history and has implications for MIC monitoring. Variation in the bilge communities is likely driven by intrinsic differences between bilge structure, creating a diversity of environmental niches as well as extrinsic differences in the seeding microbial community or nutrient landscape, which will be dependent upon the port and open waters to which a given ship has been exposed (18). Irrespective of the factors driving the observed biological variation, these findings indicate that bilges showing no signs of MIC attack are unlikely to constitute appropriate reference communities for identifying an enrichment of MIC-associated taxa, as, due to the high levels of inherent variability, changes in taxon abundances cannot be easily attributed to MIC-related changes.

Strikingly, some of the largest differences in community structure were observed between communities sampled from the same bilge 3 months apart. In this bilge, community dissimilarity was close to 100%, indicating a near complete turnover of ESVs present in the bilge in the space of a few months. High levels of ESV turnover, such as those observed in ship A, are likely due to routine bilge maintenance in which bilges are emptied and cleaned. Such events facilitate community turnover by disrupting existing communities and allowing the establishment of new primary colonizers, the identity of which will likely be dependent on the ships global position.

Pelagic communities are not representative of sessile bilge communities.

Dominant taxa present in pelagic and sessile communities were generally distinct among bilges, and dominant sessile and pelagic taxa were seldom shared within a bilge (Fig. 3). In addition to changes in dominant taxa, we found that pelagic communities within each bilge were inherently different from their respective sessile communities (Fig. 4). It has been repeatedly demonstrated that pelagic communities are not representative of sessile counterparts (16, 19–21), and the results of this study are no exception. Because planktonic microbial communities are usually not directly connected to MIC (22), these data support that biological monitoring of bilge MIC requires sampling and monitoring of sessile communities.

Stratification of the sessile community.

In line with our initial hypotheses, we observed significant within-bilge variation of sessile communities from all three wet bilges due to the presence of bilge water (Fig. 2 and 4). The formation of distinct communities in the vertical plane of the bilge is most likely due to the inundation of bilges with seawater as well as additional contaminants, such as hydraulic fluid, creating O₂, contaminant, and carbon source gradients that select for distinct communities above and below the waterline. Surprisingly, community stratification was also observed in one dry bilge (starboard A1), suggesting that the selective impact of bilge inundation can still be detected after the bilge has been emptied.

Distribution of MIC-associated functional groups reveals sampling locations for standardized protocols.

Canonical MIC-related taxa, the SRB, were most readily detected in dry bilges despite generally being accepted as being anaerobes (23). This finding underlines the importance of sampling sessile communities over pelagic. Interestingly, the bilge starboard A1, which had the highest SRB abundance, had very little detectable biofilm-forming taxa. It has been proposed that biofilm formers facilitate SRB growth and persistence in aerobic settings by creating hypoxic or anoxic niches that enable SRBs to persist in oxic environments (20, 24). While there is a lack of biofilm formation potential in the starboard A1 community, the high abundance of hydrocarbon degraders, including Thalassolituus (31% of total ESVs in sessile starboard A1 communities), suggests a mechanism whereby high levels of hydrocarbon contamination may abiotically create anaerobic microenvironments that facilitate SRB proliferation (25). However, further research is needed to determine whether SRBs in the dry bilge were metabolically active or whether they constitute a remnant expired community from when the bilge was full.

We hypothesized that MIC-associated taxa would be enriched at or below the bilge waterline. While this hypothesis held true for biofilm-forming taxa, it was not true for taxa with sulfate-reducing potential, which were relatively evenly distributed across bilge surfaces. It is notable that SRB abundances were not correlated with biofilm-forming taxa given their synergies described in the paragraph above. We observed a functional redundancy between bilges with different taxa contributing to the same function in different bilge communities. The observation of functional redundancy is not surprising, as many functional cohorts associated with MIC, including the SRB, sulfur-oxidizing bacteria, and other metal-reducing bacteria are polyphyletic (20). However, this observation underlines the potential of functionally based approaches such as qPCR of functional genes or predictive functional profiling over taxonomic approaches in MIC detection and monitoring (20, 26).

Full details of ESVs with predicted sulfate-reducing function are presented in Table S1 in the supplemental material, but briefly, the previously documented sulfate-reducing taxa detected in this study included the following: Desulfosporosinus spp., Desulfobacteraceae (3 genera), Desulfobulbaceae (4 genera), Desulfomicrobiaceae (1 genus), Desulfovibrionaceae (2 genera), and Sulfurospirillum spp. (27–31). Notably, these canonical SRB taxa only accounted for 30% of ESVs identified as having sulfate reduction potential. The sulfate reduction activities of the remaining 500 ESVs requires validation, but these data highlight that functionally driven approaches are free from a priori assumptions about which taxa are key for monitoring in MIC.

Nonabsorbent brush sampling is a superior sampling strategy for submerged sessile communities.

While it has been recognized that the study of biofilms requires a concerted cross-disciplinary effort to generate standardized methods for the study of biofilms, as yet there is no universal standard for in situ biofilm/sessile surface sampling (32). Our findings indicate that due to the abrasiveness and nonabsorbent qualities of nylon brushes, the technique of brushing surfaces may be an appropriate method to repeatably and reliably sample biofilm/sessile communities. We found that the presence of bilge water caused the communities detected by standard swabbing and by surface brushing to diverge, most likely due to swabbed samples becoming an amalgam of sessile and planktonic communities in the presence of water. This is readily demonstrated in bilges port A2 and port B, where descriptive bar charts show planktonic ESVs (Rhodobacteraceae and Sphingomonadaceae, respectively) present at a high abundance in swabbed but not brushed waterline and below-waterline communities (see Fig. S2 in the supplemental material). This hypothesis is supported by the observation that when Rhodobacteraceae were not dominant in the bilge water, they were detected at similar abundances by brushing and swabbing (e.g., bilges port A1 and star B) (Fig. S2). Other common biofilm sampling strategies include biofilm scraping with a sterile scalpel. However, Celikkol-Aydin et al. (19) demonstrated that this technique is not sufficient for sampling microbiota that comprise the inner layer of the biofilm where MIC-associated processes take place.

Recommendations, limitations, and conclusions.

The absence of standardized practices has already been highlighted as a key issue in laboratory-based biofilm and MIC studies (32, 33). For field-based MIC research, standard sampling protocols are similarly needed to facilitate direct comparison of research findings between groups and to accumulate a larger body of data from which important global trends can be identified. The importance of this latter point is being actively demonstrated in other environmental microbiomes, where global trends in soil microbiology and the human microbiome are being elucidated from large databases, such as the Human Microbiome Project (34), BASE (35), and Earth Microbiome (36), which contain data contributed by many researchers all following standardized sampling protocols. Our data contribute fundamental knowledge for the development of standardized field-sampling protocols for MIC monitoring but also indicate that biological monitoring of bilge MIC is not without ongoing challenges due to the high amount of biological variation observed within and between bilges.

This study did not capture information associated with an MIC failure; however, to identify novel targets for MIC monitoring in the future, evidence of the enrichment of MIC-related bacteria relative to an appropriate reference community is necessary. Without an appropriate reference community, biological diagnosis of MIC using a single time point is impractical. Previous studies have utilized seawater as a reference community to which corrosion-associated sessile communities are compared (8); however, as already discussed, pelagic communities are not representative of sessile communities, with pelagic and sessile biofilm communities occupying distinct ecological niches (17). Another natural choice of reference community for studying bilge MIC are communities from bilges with no signs of MIC, particularly unaffected bilges from within the same vessel as an affected bilge. However, due to the taxonomic variation observed in this study, neither bilge water nor an unaffected bilge from within the same vessel would constitute an appropriate reference community for MIC diagnosis.

Temporal monitoring is a promising strategy for identifying patterns of enrichment that may occur in the lead up to a MIC event. This approach effectively uses pre- and post-MIC event communities from the same bilge as reference and affected communities, respectively. An advantage of temporal monitoring would be the ability to correlate MIC-affiliated taxa with environmental factors, such as high nutrient load, the depletion of dissolved O₂, or shifts in pH and temperature, parameters that might precede MIC. However, due to the (currently) unpredictable nature of MIC events and challenges associated with regular vessel accessibility to civilian researchers, the likelihood of capturing the appropriate time points is low. For this reason, the establishment of a repeatable user-friendly protocol for capturing microbial diversity, which could be applied by nonspecialist crew members as part of routine maintenance is paramount. Additionally, the high temporal and between-bilge taxonomic variation limits the scope for monitoring methods that aim to identify a set of environment-specific MIC biomarkers, as suggested by Geurkink et al. (8).

Despite the taxonomic variation, we observed functional redundancy between bilges, whereby different taxa with the potential to carry out the same MIC-associated functions (i.e., sulfate reduction and biofilm formation) were present in different bilges. As such, monitoring functional traits, as opposed to specific taxa may present a way forward in biological monitoring of bilge MIC. The functions used for monitoring MIC should not be limited to SRB activity, which may only occur during MIC, but rather emerging MIC-affiliated functions, such as extracellular electron transport, as well as functions thought to proceed MIC, such as iron oxidization, sulfur oxidization, and biofilm formation, all potential targets for monitoring (10, 37). A functional approach would overcome the challenges faced regarding reference community choice and the selection of key taxa in MIC monitoring. Similar functionally based strategies have been suggested and are being developed in other microbiology fields (38–41). Our study used predicted functions derived from metabarcoding data and was thus limited to investigating the functional potential of microbial communities. Future work is needed to directly link the activity of key microbial functions with MIC events. Such studies will be needed to disentangle which environmental scenarios activate MIC-associated metabolic processes and which scenarios result in MIC events.

Our collective findings indicate that, in the study of sessile marine communities associated with MIC, a sampling protocol based on brushing defined (size and location) areas of bilge surfaces could be developed into a standardized practice. For bilges, our data indicate that the most relevant location to sample MIC-associated taxa is within sessile communities at and below the bilge waterline. Furthermore, this study and the accompanying discussion highlight how the incorporation of approaches that are starting to permeate other microbiological disciplines, such as the sharing of data through standardized sampling strategies and the movement toward functional understanding of microbial communities, present a clear route for improving the use of biological information in MIC detection, monitoring, and prevention.

MATERIALS AND METHODS

Microbiological sampling.

Microbial communities were collected from the bilges of three Royal Australian Navy vessels between September and December in 2018. Ship A and B were of the same class, while ship C was of a different class. Ship A was sampled twice, 3 months apart. Where possible, the pelagic community was sampled (in duplicate) by collecting 200 to 400 ml of bilge water into sterile Schott bottles. It is worth noting that there was no clear evidence of problems with microbial corrosion in any of the specific bilges sampled in the current work.

Pelagic communities were concentrated onto a Millipore Express Plus polyethersulfone (PES) membrane (0.22 μm), through which 200 ml of bilge water was passed. The membrane was stored at −20°C until DNA could be extracted. Due to the filter size, filters were cut in half, and community DNA was extracted from and sequenced for each half separately. Community profiles from the filter technical replicates were averaged post sequencing.

Sessile communities were collected using sterile cotton swabs (Steriswab MW720; MWE, Wiltshire UK) and individually wrapped nylon toothbrushes from 10-cm square sections of bilge by swabbing/brushing the surface for 60 s. Communities were collected in triplicate from the bilge waterline (WL), 5 cm below the waterline (B), and 5 cm above the waterline (A). If no water was present, only WL and B communities were sampled. A “leap-frog” sampling design was employed, whereby swab sampling and brush sampling was conducted interchangeably along each given bilge sampling height (A, WL, or B) to avoid location effects impacting swab-brush comparisons (see Fig. S1 in the supplemental material). All samples were stored in a cooler and transferred to 4°C storage as soon as possible. All samples were processed within 72 h.

Sampling was necessarily opportunistic, and not all bilge areas could be sampled on each sampling trip. Table 1 lists the samples that were able to be collected on a given sampling expedition. Where possible, bilge water was collected from within the same bilge location that was accessible for surface sampling. Clean swabs and toothbrushes were included as no-template controls (Table 1).

TABLE 1.

Summary of bilge community sampling

Sample parameter	Data for ship:
	A1	A2	B	C
Bilge(s) sampled	Port and starboard	Port and starboard	Starboarda	Port
Pelagic community sampled	Yes	Yes	Yes	Yes
Sessile community sampledb	WL and B	A, WL, and B	A, WL, and B	No
Pelagic and sessile community from same bilge section	Noc	Yes	Yes	NA

^aPelagic communities sampled from port and starboard bilges; n = 3 for port and n = 1 for starboard bilge.

^bA, above waterline; WL, on waterline; B, below waterline.

^cBilge accessible for sessile sampling was dry, bilge water collected from adjacent bilge section using peristaltic pump.

Sequencing and bioinformatics.

Community DNA extraction and 16S rRNA gene metabarcoding was conducted by the Australian Genome Research Facility (AGRF). Libraries were spiked with 25% PhiX and sequenced using an Illumina MiSeq platform (300 PE). Metabarcoding analysis targeted the V3-V4 bacterial hypervariable region using primers 341F (5′-CCTAYGGGRBGCASCAG) and 806R (5′-GGACTACNNGGGTATCTAAT) (42).

Bioinformatic analyses were performed using QIIME 2 (43). Primers were removed, and forward and reverse reads were truncated based on quality scores to 280 and 250 bp, respectively, prior to denoising. Denoising was performed using the QIIME 2 DADA2 plugin, which implements joining, quality filtered, and chimera detection (43, 44). Taxonomic assignment was performed using a naive Bayes classifier trained on V3-V4 regions extracted from the SILVA_132_release 99% operational taxonomic unit (OTU) data set. Phylogenetic trees were created using MUSCLE (45). To generate a table of gene abundances, annotated as KEGG orthology (KO) abundances, PICRUSt2 was implemented (46).

Statistical analysis.

All statistical analyses were conducted in R, version 4.0.3 (47). Prior to downstream analysis, exact sequence variant (ESV) tables were rarefied to 20,000 reads using the rarefy_even_depth function in the package phyloseq (48). Normality and homogeny of variance for alpha diversity data were tested using Shapiro and Bartlett tests, respectively. Where these assumptions were met, one- and two-way analysis of variance (ANOVA) was used with Tukey’s honestly significant difference (HSD) post hoc testing. Where assumptions were not met, Kruskal-Wallis and two-way permuted ANOVAs, implemented via the function aovp from the package lmperm, were used for one- and two-way testing, respectively, with Dunn’s multiple range post hoc tests (49). Beta diversity analysis was performed using the adonis function from the vegan package on Bray-Curtis and UniFrac distance matrices (50, 51). Dendrograms were generated using the packages vegan and dendextend (52).

Functional annotations of ESVs were based on the predicted presence of KOs of interest in individual ESVs. Lists of KOs associated with dissimilatory sulfate reduction and biofilm formation pathways were retrieved from the KEGG database. Given the prevalence of hydrocarbon contamination in these systems, the communities’ functional potential for hydrocarbon degradation was also investigated. To represent hydrocarbon degradation, KOs associated with benzoyl, naphthalene, and toluene degradation pathways were retrieved. ESVs predicted to contain for each KO of interest were pulled from the per sequence contribution metagenome PICRUSt output, and custom R scripts were used to create a table whereby rows represented ESVs containing KOs of interest and columns represented the functional classes represented by KOs (i.e., SRB, biofilm formation, or hydrocarbon [HC] degradation). Functional allocations were appended to phyloseq taxonomy tables to generate bar charts showing the distribution of ESVs containing the functions of interest.

Data availability.

Raw fastq files for this project have been deposited with the NCBI SRA database and can be accessed using BioProject ID number PRJNA639944.