The Trichodesmium microbiome can modulate host N2 fixation

Trichodesmium is a marine, diazotrophic cyanobacterium that plays a central role in the biogeochemical cycling of carbon and nitrogen. Colonies ubiquitously co‐occur with a diverse microbiome of heterotrophic bacteria. We show that manipulation of the microbiome with quorum sensing acyl homoserine lactones (AHLs) significantly modulated rates of N2 fixation by Trichodesmium collected from the western North Atlantic, with positive and negative effects of varied magnitude. Changes in Trichodesmium N2 fixation did not correlate with changes in microbiome composition or geochemistry. With AHL addition, a subset of the significantly differentially expressed genes was related to known quorum sensing responses in model bacteria. However, there was little overlap in specific microbiome transcriptional responses to AHL addition between stations. Overall, these host‐microbiome interactions reflect a complex interplay of biotic and environmental factors that together form an overlooked mechanism modulating Trichodesmium N2 fixation.

Trichodesmium is a keystone member of marine environments because of its ability to provide fixed N 2 that fuels primary productivity in otherwise nutrient poor regions (Capone et al. 1997). Some estimates predict Trichodesmium accounts for approximately half of the total oceanic fixed N 2 (Bergman et al. 2013). In oligotrophic regions, Trichodesmium N 2 fixation is strongly affected by the availability of nutrients in the water column (Sohm et al. 2011). With the high-iron quotas associated with N 2 fixation and photosynthesis, iron limitation is the canonical constraint on diazotrophy in Trichodesmium (Berman-Frank et al. 2001). In the oligotrophic western North Atlantic however, high-iron concentrations relative to phosphorus lead to phosphorus depletion, and Trichodesmium distribution and N 2 fixation is thought to be more strongly influenced by phosphorus availability, relative to other systems (Sañudo-Wilhelmy et al. 2001;Moore et al. 2013;Rouco et al. 2014Rouco et al. , 2018. Despite the fact that resource controls on Trichodesmium eco-physiology are well established, using geochemistry to predict and model the distribution and activities of this organism remains challenging (Capone et al. 2005;McGillicuddy 2014;Snow et al. 2015). Recent studies have suggested that some of the challenges associated with modeling Trichodesmium dynamics are in part due to the fact that its physiology is tightly linked to that of its microbiome (Frischkorn et al. 2017;Lee et al. 2017).
Trichodesmium ubiquitously co-occurs with a microbiome of epibiotic microorganisms (Paerl et al. 1989;Hewson et al. 2009;Hmelo et al. 2012;Rouco et al. 2016;Frischkorn et al. 2017;Lee et al. 2017), yet the role of this microbiome in modulating Trichodesmium physiological ecology is still poorly understood. These epibionts are tightly associated with colonies of Trichodesmium filaments, and together make up a conserved community that is unique from planktonic microbes in the surrounding water column (Hmelo et al. 2012;Rouco et al. 2016). In other systems, such communities of bacteria are known to regulate physiological activities on a population-wide scale through cell-cell signaling called quorum sensing (Miller and Bassler 2001). Quorum sensing molecules in the acyl homoserine lactone (AHL) family modulate a range of activities in bacteria by altering gene expression and subsequently behavior (Waters and Bassler 2005), thus altering physiology through a mechanism that is decoupled from external environmental stimuli. These molecules have been detected in Trichodesmium colonies collected from the environment , however, Trichodesmium does not possess the capacity to produce or respond to quorum sensing molecules (Vannini et al. 2002;Patankar and Gonzalez 2009), suggesting that AHL-driven cell-cell communication takes place solely between microbiome members.
Artificial manipulation of quorum sensing circuits provides a unique opportunity to selectively alter microbiome activity within Trichodesmium colonies. In previous field studies, when Trichodesmium colonies were artificially amended with a cocktail of AHLs, phosphorus acquisition was altered by the induction of an organic phosphorus hydrolyzing enzyme, providing evidence that the phosphorus physiology of the entire consortia could be influenced by solely modulating the microbiome . The microbiome of Trichodesmium contains diverse and abundant metabolic potential (Frischkorn et al. 2017;Lee et al. 2017) that likely influences interactions between the Trichodesmium host and the environment, and could modulate other aspects of host physiology like N 2 fixation. Here, we assess if microbiome activities can modulate Trichodesmium N 2 fixation in the western North Atlantic by amending freshly collected colonies with AHLs and monitoring changes in Trichodesmium N 2 fixation and microbiome gene expression.

Methods
We collected Trichodesmium colonies from surface water along a cruise transect in the western North Atlantic (Fig. 1a) aboard the R/V Atlantic Explorer (AE1409) during May 2014, as previously described (Frischkorn et al. 2017). We focused on field studies with freshly isolated colonies because the clade of the cultured strain (Trichodesmium erythraeum IMS101) is not dominant in the environment (Rouco et al. 2014), and it likely has a derived microbiome. Briefly, we conducted six hauls of a net tow (mesh size of 130 μm) at~07:00 h in the upper 20 m to collect samples for experiments. We isolated Trichodesmium colonies from the net tow and washed them three times with fresh 0.2 μm sterile-filtered local surface seawater, a method that prevents contamination by planktonic microbe carryover and has reproducibly yielded the stable Trichodesmium microbiome (Rouco et al. 2016;Frischkorn et al. 2017Frischkorn et al. , 2018. After washing, we transferred approximately 30 cleaned colonies of similar sizes and morphologies into acid-clean, polycarbonate bottles filled with 30 mL of sterile-filtered seawater. Efforts were made to place colonies in bottles to mimic colony morphology distribution found in net tow samples. To minimize handling effects, time from initial sampling to incubation was less than 15 min. We spiked the six experimental bottles (+AHL) with a cocktail of three di-deuterated AHLs (N-(decanoyl)homoserinelactone, N-(dodecanoyl)homoserinelactone, and N-(tetradecanoyl)homeserinelacton) in dimethylsulfoxide (DMSO) to a final concentration of 500 nmol L −1 , and the six control bottles with only DMSO. Trichodesmium consortia members are known to produce and metabolize dimethylsulfide products (Lee et al. 2017), and additional negative controls may be warranted for future work focused on quorum sensing. Previous field incubations showed microbiome activity was specifically altered by the three AHL compounds used here and not by N-3-oxooctanoyl homoserine lactone . Selective manipulation of the microbiome was possible because natural, nondeuterated forms of these molecules have been observed in Trichodesmium colonies, but Trichodesmium itself cannot produce or respond to AHLs (Van Mooy et al. 2012). We incubated the control and +AHL bottles for 4 h in on-deck flow-through incubators shaded with blue film to mimic in situ conditions. A 4h incubation was chosen as it was short enough to capture +AHL induced transcriptional changes in the microbiome, and long enough to have induced N 2 fixation changes in Trichodesmium. Biomass limitations and destructive sampling precluded time series analysis. The experiments were designed to selectively alter microbiome activity, rather than isolate quorum sensing pathways, and we discuss details regarding the experimental approach and operational negative controls in prior studies Krupke et al. 2016).
After 4 h of incubation, we assayed three of the bottles in each condition for N 2 fixation rate using the acetylene reduction technique as previously described (Capone 1993). Briefly, approximately 30 Trichodesmium colonies were placed in each 60 mL polycarbonate bottle containing 30 mL of filtered seawater as described above. We injected a 1 mL aliquot of acetylene into the bottle through a septum cap, gently inverted the bottle, and incubated in an on-deck incubator as previously described. We then analyzed bottle headspace for ethylene at time-points approximately every 30 min by gas chromatography (Capone 1993). Triplicate daily analysis of a 17 ppm ethylene standard was used to calibrate the gas chromatograph. The average daily coefficient of variation for analysis of the ethylene headspace standard was 4.9%. Linear regressions of headspace ethylene concentration vs. time and appropriate Bunsen coefficients were used to determine the total ethylene production rate (Breitbarth et al. 2004). The limit of detection for ethylene production, defined as the accumulation of ethylene through the course of the incubation to a level three times greater than detected at the first time-point, was 8.1 pmol colony −1 h −1 . The estimated N 2 fixation rate was calculated by dividing the total ethylene production rate by 4 (Capone 1993). Average coefficient of variation of N 2 fixation rates in replicate incubations was 10.2%. Chauvenet's criterion (95% confidence interval) was used to remove occasional outlying rates among replicate incubations.
Student's t-tests were used to identify significant differences in N 2 fixation rates between replicate control incubations and replicate incubations receiving an AHL amendment.
For gene expression analysis, we filtered each of the remaining three bottles of each treatment onto 5 μm pore size polycarbonate filters after the 4 h incubation period and immediately stored them in liquid nitrogen until sample processing. Three bottles set up as controls as previously described were used to measure in situ N 2 fixation rates at stations where metatranscriptome sequencing experiments were not carried out. Total dissolved phosphorus (TDP) was determined on 0.2 μm filtrates of surface water (~5 m depth) samples collected via CTD into acid-clean polycarbonate bottles. Samples were processed at the SOEST Laboratory for Analytical Biogeochemistry at the University of Hawaii, according to facility protocols.
We extracted prokaryotic RNA from triplicate control and +AHL samples, pooling together triplicate samples, sequencing 60 million paired end reads, and trimming as previously described . To obtain read counts for each sample, we mapped trimmed forward and reverse reads to metagenome assemblies from the same sampling locations that were previously characterized and clustered into orthologous groups (OGs) (Frischkorn et al. 2017). OGs were annotated (Frischkorn et al. 2017) and sorted into metabolic functional groups of both known quorum sensing responses and other functions (available as a supplemental dataset hosted on FigShare at 10.6084/m9.figshare.7051589). We carried out mapping using RSEM with the paired-end and Trichodesmium colony N 2 fixation rate measurements were performed; size is proportional to rate. Stations where +AHL amendment experiments were carried out are denoted with a star. TDP was contoured using Ocean Data View and the data intensive visualization and analysis grid method. (b) N 2 fixation rates in triplicate control and +AHL experiments. * indicates significant difference between control and +AHL after Student's t-test (< 0.05). (c) Percent change in N 2 fixation between control and +AHL treatments.
Bowtie2 parameters (Li and Dewey 2011) and tabulated counts across only OGs from the eight core epibiont genome bins (Table 1, supplemental dataset 10.6084/m9.figshare. 7051589). We determined significant changes in OG expression between control and +AHL samples using a stringent empirical Bayes approach called "analysis of sequence counts (ASC)" (Wu et al. 2010). This approach evaluates the posterior probability associated with a given fold change across the pooled triplicates, and performs similarly, but conservatively, on replicated and unreplicated sample datasets (Wu et al. 2010). OGs were considered significantly higher or lower if they had a 95% or higher posterior probability of a fold change greater than 2 between treatment and control. Taxonomic relative abundance estimates for metagenome samples were previously calculated (Frischkorn et al. 2017).

Results and discussion
Biological interactions are a driver of Trichodesmium N 2 fixation Trichodesmium in situ N 2 fixation rates increased from north to south, with stations north of 20 latitude having significantly lower rates of N 2 fixation than those in the south (p = 0.02, one-way ANOVA; Fig. 1a; Supporting Information Table S1). Phosphorus is known to be a limiting nutrient for N 2 fixation for Trichodesmium in this region (Sañudo-Wilhelmy et al. 2001;Sohm et al. 2011), and surface TDP concentration in the stations north of 20 latitude differed significantly from those collected to the south (p < 0.03, one-way ANOVA; Fig. 1a; Supporting Information Table S1). Although the changes in N 2 fixation and TDP between the northern and southern stations are consistent with phosphorus being a strong driver of N 2 fixation across the transect, the in situ rates of N 2 fixation were not significantly correlated with TDP (R 2 < 0.319, p = 0.186, one-way ANOVA; Fig. 1a). This suggests that factors other than phosphorus concentration might also influence Trichodesmium N 2 fixation.
At all stations where we performed experiments (Fig. 1a), N 2 fixation was significantly changed in response to microbiome manipulation with the +AHL amendment, with significant decreases in rate at Sta. 2 and 10 (p = 0.04 and 0.03, respectively), and a significant increase at Sta. 17 (p = 0.01) (Fig. 1b). Although Trichodesmium has co-occurred with a N 2 fixing epibiont in the North Pacific Ocean (Momper et al. 2014;Gradoville et al. 2017), herein Trichodesmium colonies did not show evidence of non-Trichodesmium nif genes (Frischkorn et al. 2017), indicating that Trichodesmium was the only diazotrophic organism within these samples. The magnitude and direction of microbiome-induced changes in host N 2 fixation ranged from significantly decreased at the northernmost station to significantly increased at the southernmost station, despite similarities in TDP between proximal stations (Fig. 1a,c). This suggests that biotic interactions within colonies can act independently of geochemistry to influence Trichodesmium physiology.
Extrapolating the range of changes we observed in +AHL treatments to in situ N 2 fixation rates, we illustrate the potential influence of the microbiome on host physiology (Fig. 2). This theoretical N 2 fixation range was determined from the maximum (+20%) and minimum (−41%) percent change after +AHL amendment to contextualize the potential for biological interactions to alter observed N 2 fixation rates (Fig. 2). Although this visualization should be interpreted with caution, the hypothetical 61% range of variation reflects a scenario where biological interactions drive subsequent N 2 fixation higher or lower than would otherwise be expected, even given noted uncertainty in these measurements (coefficient of variation = 10.2%). Such changes are modulated on Table 1. Number of total OGs from individual microbiome members that had significant differential expression (DE) after +AHL amendment. Counts were tabulated for each gene in a microbiome genome bin that belongs to an OG that was found to be significantly higher or lower in response to +AHL amendment. Bin number and total OGs reflect findings in Frischkorn et al. (2017). OGs total refers to the total number of OGs found in each genome bin.  , as well as that observed across an Atlantic meridional phosphorus gradient (Moore et al. 2009). The variability we observed suggests that interactions in the microbiome should be considered drivers of N 2 fixation in addition to well-studied constraints like iron, phosphorus, and CO 2 concentration (Paerl et al. 1994;Sañudo-Wilhelmy et al. 2001;Hutchins et al. 2015). Overall, it may be important to consider these biotic factors in addition to light and surface ocean nutrients when predicting Trichodesmium physiological ecology.

Microbiome transcriptional patterns varied with AHL amendment
Of the eight core epibiont genomes (Frischkorn et al. 2017), all had OGs which were significantly differentially expressed (1.2-2.5%) in +AHL treatments compared to the controls (Table 1). In model organisms, approximately 5-10% of total genes are differentially expressed as a result of quorum sensing (DeLisa et al. 2001;Schuster and Greenberg 2006). Slightly lower values in our study could reflect the diverse nature of the Trichodesmium microbiome community, where quorum quenching can be active ) and complex signaling responses are likely at play between organisms after 4 h. It could also reflect the conservative nature of ASC (Wu et al. 2010) relative to other approaches. Regardless, these results suggest all members of the microbiome can respond to AHLs, which is consistent with the fact that all genome bins were found to possess putative luxR genes from canonical quorum sensing operons (Frischkorn et al. 2017). The Gammaproteobacterium (bin 7) contributed to 60% of the significantly responsive metabolic functional groups across all three stations tested (Supporting Information Fig. S1). Notably, similar Gammaproteobacteria are ubiquitously found in association with colonies across multiple environments and in culture (Hmelo et al. 2012;Rouco et al. 2016;Frischkorn et al. 2017;Lee et al. 2017) and biologically relevant concentrations of AHLs have been found in Trichodesmium colonies . Taken together, it is likely that the microbiome could ubiquitously influence Trichodesmium N 2 fixation via quorum sensing pathways.
Little is known about the quorum sensing and quenching pathways in the Trichodesmium microbiome, but in model systems these pathways are complex, operating in circuits that can influence each other in distinct ways and lead to a cascade of unique transcriptional responses and resulting shifts in activity (e.g., Wagner et al. 2004;Schuster and Greenberg 2006). Amendment with AHLs induced significant changes in OG expression at each station (Fig. 3), and some of these OGs encoded metabolic functions that were the same as quorum sensing-induced shifts in model organisms such as Pseudomonas aeruginosa (Wagner et al. 2003(Wagner et al. , 2004Schuster and Greenberg 2006). OGs in the energy metabolism, chemotaxis, motility and attachment, and nitrogen and amino acid metabolism functional groups were significantly differentially expressed across all three stations (Fig. 3). Similar to model organisms (Wagner et al. 2003), changes in these functional groups highlight how basic metabolic processes and lifestyle can be modulated in response to signaling molecules like AHLs. Transposon-related OGs were also differentially expressed at all stations (Fig. 3). AHLs are known to stimulate gene transfer agents and increase transposon mobility (Schaefer et al. 2002;Auchtung et al. 2005), and many quorum sensing genes are adjacent to transposons or encoded within them (Thomson et al. 2000;Wei et al. 2006). Although there were similarities in the metabolic functional groupings of differentially expressed OGs (Fig. 3), at the gene family level there was little overlap between specific OGs that contributed to those categories across the three stations (Supporting Information Fig. S2). The five OGs making up the conserved AHL response were all annotated as putative, uncharacterized proteins (Supporting Information Fig. S2). Sampling a time course in AHL addition experiments would help identify more quorum sensing pathways and any conserved responses that occurred earlier than the 4 h time point. In summary, microbiome transcriptional responses to AHLs varied inconsistently between stations and additional experiments are required to evaluate how different microbiome activities drive the observed variation in Trichodesmium N 2 fixation.
Previous work found that community composition of the microbiome varied significantly between northern stations (e.g., Sta. 2) and the southern stations (e.g., Sta. 10 and 17),

Fig. 2.
Illustration of potential microbiome influence on in situ Trichodesmium N 2 fixation across the western North Atlantic transect. Black circles represent in situ N 2 fixation rates observed at each station. To highlight the potential magnitude of the effects that biological interactions can have on Trichodesmium N 2 fixation, the maximum and minimum percent changes we observed after +AHL amendment of colonies in this study were extrapolated to in situ colony N 2 fixation rates. The gray bars reflect this theoretical range (61%) of variability in N 2 fixation that could occur on rapid (< 4 h) time scales as a result of changes in microbiome activity. Average coefficient of variation of N 2 fixation rates in replicate incubations was 10.2%. Stations are organized in order from north to south.
but that the Trichodesmium host did not (Frischkorn et al. 2017). However, in the southern stations where community composition and geochemistry were similar, transcriptional responses in the microbiome were dissimilar and yielded different effects on N 2 fixation. For example, although relative abundances of the core microbiome members between Sta. 10 and 17 were not significantly different (Frischkorn et al. 2017), and TDP was similar, variable epibiont transcription forced by quorum sensing elicited different Trichodesmium N 2 fixation responses. In sum, N 2 fixation responses are not predictable from microbiome community composition alone. Variability in AHL responses in model systems can vary as much as 100-fold due to nutrient status, oxygen availability, and whether cultures were planktonic or growing as a biofilm (Schuster and Greenberg 2006;Duan and Surette 2007). Furthermore, the transcriptional regulators that are activated by quorum sensing molecules do not exist in isolation, but rather interact with a web of regulators and quorum quenching molecules that affect physiology on a genome-wide scale-a finding that has been used as an explanation for the rapid adaptability of bacteria to fluctuating environments (Schuster and Greenberg 2006). The concentration of the AHLs experienced by the epibionts could also affect the direction, magnitude, and characteristics of gene expression, as hydrolytic enzyme activity in marine particulate matter is strongly affected by the concentration of AHLs added (Krupke et al. 2016). Similarly, in P. aeruginosa, different concentrations of the same quorum sensing signaling molecule result in different responses that subsequently elicit opposite host physiological responses (Williams and Cámara 2009). In the Trichodesmium holobiont, the effects of the microbiome on N 2 fixation likely reflect a complex interplay of environment, community composition, chemical signaling, and metabolic functional response, and a more mechanistic understanding of activities in the microbiome is needed to model how biological interactions modulate Trichodesmium N 2 fixation.

Conclusions
Here, we show that selective manipulation of microbiome activities can alter the N 2 fixation rate of the Trichodesmium host over short time scales, expanding the suite of factors that are known drivers of marine N 2 fixation. If the observed interplay between host and microbiome holds true across the full range of oligotrophic environments Trichodesmium inhabits, then these interactions are likely an overlooked factor that influences Trichodesmium N 2 fixation, and future ecological studies of Trichodesmium should take into account the activities of the microbiome. Fig. 3. Shifts in microbiome OG expression in response to +AHL amendment. Positive log 2 fold change values represent OGs with increased expression in the +AHL treatments relative to the control treatments. Triangular symbols denote significantly increased or decreased expression (95% or higher posterior probability of a fold change greater than 2). Colored triangles call out OGs with annotations that align with the suite of metabolic functions known to regulated by quorum sensing in the model bacterium Pseudomonas aeruginosa (Wagner et al. 2003(Wagner et al. , 2004Schuster and Greenberg 2006). D.E., differentially expressed; Q.S., quorum sensing. Annotations and count data for all experiments can be found in the supplemental dataset hosted on FigShare (10.6084/m9. figshare.7051589).