Host-emitted amino acid cues regulate bacterial chemokinesis to enhance colonization

Summary

Animal microbiomes are assembled predominantly from environmental microbes, yet the mechanisms by which individual symbionts regulate their transmission into hosts remain underexplored. By tracking the experimental evolution of Aeromonas veronii in gnotobiotic zebrafish, we identify bacterial traits promoting host colonization. Multiple independently evolved isolates with increased immigration harbored mutations in a gene we named Sensor of Proline Diguanylate cyclase Enzyme (SpdE) based on structural, biochemical, and phenotypic evidence that SpdE encodes an amino acid-sensing diguanylate cyclase. SpdE detects free proline and to a lesser extent valine and isoleucine, resulting in reduced production of intracellular c-di-GMP, a second messenger controlling bacterial motility. Indeed, SpdE binding to amino acids increased bacterial motility and host colonization. Hosts serve as sources of SpdE-detected amino acids, with levels varying based on microbial colonization status. Our work demonstrates that bacteria use chemically regulated motility, or chemokinesis, to sense host-emitted cues that trigger active immigration into hosts.

Graphical Abstract

eTOC

The strategies bacteria use to detect and colonize animal hosts are underexplored. Robinson et al evolved a zebrafish symbiont, Aeromonas, to become a better colonizer. Their study revealed that Aeromonas senses host-emitted amino acid cues to modulate motility, via a process called chemokinesis, and rapidly immigrate into the zebrafish intestine.

Introduction

Host-associated microbial communities, especially those of the vertebrate digestive tract, are compositionally diverse and dynamic (Lloyd-Price et al., 2017), and their membership and activities are intimately tied to host health and development (Gentile and Weir, 2018; Routy et al., 2018). Intensive research has investigated the ecological and evolutionary processes that influence microbiome composition (Koskella et al., 2017; Scanlan, 2019; Walter and Ley, 2011). Existing models for microbiome assembly often focus on sampling of microbes from environmental sources, host selection, and microbe-microbe interactions (Adair and Douglas, 2017; Levy and Borenstein, 2013; Näpflin and Schmid-Hempel, 2018; Zha et al.). However, the role of microbial traits in enhancing the likelihood of sampling by hosts is not well understood.

Microbiome constituents must migrate into the host from external sources to become part of the resident microbial community (Asnicar et al., 2017; Combellick et al., 2018; Korpela et al., 2018). Microbes are continually introduced into hosts’ microbiomes via processes such as ecological succession early in host development (Stewart et al., 2018), after perturbations such as antibiotic treatment (Suez et al., 2018), and through continual flux of strains over time (Caporaso et al., 2011; Faith et al., 2013; Priya and Blekhman, 2019; Schmidt et al., 2019). The origins of these microbes are not entirely understood but include the environment outside of hosts, food, other body sites within the same individual, and transmission from other individuals (Robinson et al., 2019). For microbial immigration (i.e. initial entry into the host), most studies have focused on the role of the host in facilitating this process, thus the role of microbial traits in initial entry into the host is understudied, especially for non-pathogens.

We developed a tractable experimental evolution model in gnotobiotic zebrafish to elucidate selective pressures within host-microbe systems and investigate how gut bacteria adapt to optimize host colonization (Robinson et al., 2018). This gnotobiotic system allows for simplification of a highly complex system to facilitate study and tracking of a bacterial symbiont within a population of hosts. Zebrafish (Danio rerio) enable the study of both ecological processes of microbiome assembly and mechanisms of host-microbe interaction (Burns and Guillemin, 2017; Kanther and Rawls, 2010; Stagaman et al., 2020). Zebrafish larvae can be raised in flasks containing many (>10) fish and inoculated with bacteria via the flask medium, thereby simulating bacterial transmission in natural populations of hosts. By tracking bacterial growth and persistence in hosts and the extra-host environment (i.e. flask medium) we can capture migration dynamics and measure adaptations due to improved intra-host or extra-host fitness, or improved transmission into, out of, or between hosts (Figure 1A).

Figure 1. Rapid and reproducible adaptation of Aer01 serially passaged in populations of larval zebrafish.

A) Schematic of the Aer01 colonization cycle within a host population and potential adaptive strategies. B) Evolution experiment passaging scheme (dpf, days post fertilization). C) Competitive indices (CI) of Aer01 ancestor and evolved isolates, competed with an ancestral reference strain (see also Figure S1A). Each data point represents the CI from an individual fish. Lines 1–3 (gray filled symbols) were evolved in wild-type hosts; Lines 4 and 5 (grey open symbols) were evolved in myd88^−/− hosts. Points outlined in red represent a lower bound for the actual CI (see Methods). Black bar = median. n ≥ 4 fish/condition, one experimental replicate. Dotted line indicates CI of 1 (i.e. no competitive advantage). Inset graph presents CI data for isolates in the flask medium (a single point for each flask); y-axis and symbols match the main graph. See also Figure S1.

In a previous study, we found that a zebrafish gut isolate, Aeromonas veronii (ZOR0001, hereafter referred to as Aer01), engineered to have a high mutation rate, evolved to out-compete the ancestral strain in the intestine by more rapidly immigrating into the host (Robinson et al., 2018). Here, we replicated our experimental evolution with a non-mutator strain to identify genetic changes underlying these adaptations. We describe our discovery of SpdE, an amino-acid sensing diguanylate cyclase, that promotes host colonization by increasing Aer01 motility. Combined, these data demonstrate how bacteria can use chemically regulated motility, or chemokinesis, to achieve host colonization and discriminate between hosts in a microbiome-dependent manner.

Results:

Serial passage of Aer01 in populations of larval zebrafish resulted in rapid and reproducible adaptation

Based on the design of our experimental evolution with a rapidly evolving mutator strain of Aer01 (Robinson et al., 2018), we passaged replicate lines of non-mutator Aer01 populations through germ-free (GF) larval zebrafish (Figure 1B). To explore the impact of the innate immune system on symbiont adaptation, we used two genotypes of hosts—wild type (three replicate lines), and myd88^−/− immunodeficient (two replicate lines). For each passage, Aer01 populations were added to flasks of 10–15 GF zebrafish larvae at 4 days post fertilization (dpf), and at 7 dpf the gut-associated Aer01 populations were pooled as inoculum for the subsequent passage of that line, for a total of 20 passages. Cryopreserved inoculum samples served as the source for randomly selected isolates from passages 5, 10, 15, and 20.

We assayed for adaptation by competing these isolates against a differentially-tagged Aer01 reference strain for larval colonization from 4–7 dpf (Figure S1A). A competitive index (CI) of the competing strain relative to the reference strain was calculated for each fish by dividing the strain ratio (competitor:reference) in the intestine by the strain ratio in the inoculum. Isolates from two lines reached CI’s significantly higher than the ancestor by passage 5; isolates from all replicate lines outcompeted the ancestor by passage 10 (Figure 1C). Of note, none of the isolates with increased CI’s in the gut exhibited increased fitness in the flask medium (Figure 1C, inset), and we observed no growth rate differences between our isolates and the ancestor in rich medium (Figure S1B). . Although the rate of adaptation was slower than for mutator Aer01, the CI’s were comparable to those of the evolved isolates from our previous study (Robinson et al., 2018).

Mutations in a single gene arose in all evolved lines

We sequenced the genomes of all isolates represented in Figure 1C and identified a small set of mutations (Table S1). Only isolates predicted to have adapted, as indicated by increased CI (see Figure 1C), had mutations. The majority of isolates accumulated only one or two mutations, with a maximum of five. All replicate lines carried mutations in the same gene, named herein spdE. Across all lines there were 7 different spdE mutations identified, with no overlap between lines. The majority of the mutations were predicted loss-of-function mutations including 2 nonsense, 2 frame-shift, and 3 missense mutations. The genome resequencing analysis software (breseq; see Methods) also predicted potential genomic rearrangements in a subset of isolates (6/17); however, since none of the rearranged genomic loci were shared across lines, we did not pursue them further.

We observed no difference in the rate and nature of mutations in isolates recovered from the two host genotypes, WT and myd88^-/−. This finding is consistent with results from our previous study showing that early-evolved isolates had the same competitive fitness in both the WT host genotype in which they were evolved and when tested in the myd88^−/− genotype. In our previous study, we only observed host genotype-specific competitive advantages at later points in the evolution. Given the kinetics of adaptation in our current experiment with the non-mutator Aer01, we would not have anticipated reaching this equivalent stage of strain adaptation.

Loss of spdE enables rapid immigration into the host and fish-to-fish transmission

To confirm that loss of spdE conferred the improved colonization phenotype, we deleted the gene in the ancestral genome. We found that deletion of spdE resulted in an average CI similar to that of an evolved isolate with the sole mutation of a 1 bp deletion in spdE (isolate L3_P10) (Figure 2A). Next we overexpressed spdE in the ancestor and measured a five-log decrease in the average CI (Figure 2A; WT^anc_OE). We also engineered the ΔspdE mutant to express either a WT copy of spdE or the evolved 1 bp deletion allele. Expression of the WT spdE allele, under control of the native promotor, reversed the ΔspdE mutant phenotype, with decreased CI similar to overexpression of spdE (Figure 2A; ΔspdE_comp), whereas the evolved allele did not complement the ΔspdE mutant phenotype (Figure 2A; ΔspdE_comp^evol).

Figure 2. Loss of spdE enables rapid immigration into hosts and fish-to-fish transmission.

A) Competitive indices (CI) of ancestral Aer01 (WT^anc), an evolved isolate (L3_P10), ΔspdE, the WT over-expressing spdE (WT^anc_OE), ΔspdE complemented with WT copy of spdE (ΔspdE_comp), and ΔspdE complemented with an evolved mutant allele (ΔspdE_comp^evol). All strains were competed against a WT reference strain (see Figure S1A). Points outlined in red represent a lower or upper bound for the actual CI (see Methods). Dotted line indicates CI of 1 (i.e. no competitive advantage). n ≥ 8 fish/condition, one experimental replicate. Inset graph presents CI data for Aer01 mutants in the flask medium (a single point for each flask); y-axis and symbols match the main graph. B) Competitive indices of ΔspdE and L3_P10 evolved isolate are dependent on mode of inoculation (flask medium (FM inoc.) versus gavage). Dotted line indicates CI of 1 (i.e. no competitive advantage). n ≥ 6 fish/condition, one experimental replicate. C) Immigration rates of ancestral Aer01, ΔspdE, and the L3_P10 mono-associated into larval fish (see Methods and Figure S3A) (n = 3 experimental replicates per time point). Means (± SEM) are plotted. D) Fish-to-fish transmission of ancestral Aer01 (WT^anc) and ΔspdE. Mono-associated “donor” fish were added to a flasks of GF “recipient” fish for 15 hours, then abundances of Aer01 in all fish were determined. n ≥ 27 fish/condition, two experimental replicates. p-value < 0.0001, two-tailed Mann-Whitney test. CFU/gut, colony-forming units per gut. Median and interquartile ranges are plotted in A, B, and D. See also Figure S2.

In our previous study, the first adaptations that evolved enabled rapid immigration into the host from the aqueous environment (Robinson et al., 2018). To test whether loss of spdE conferred this trait, we inoculated competing strains directly into the fish intestine via microgavaging, effectively bypassing the immigration step of colonization (Figure S2A). Fish were dissected and gut ratios of the strains were determined immediately after inoculation (~0.5 hrs), and at 4 hrs. Whereas ΔspdE and the evolved isolate outcompeted the ancestor (WT^anc) when added to the flask water, their CIs when gavaged into the intestine were close to 1 (Figure 2B), meaning their competitive advantage arose outside the host. Importantly, the fish were gavaged at an initial intestinal abundance well below the average carrying capacity of Aer01, and both strains grew within the intestine during the 4 hr competition (Figure S2B).

We next performed direct measurements of migration rates of the evolved and ΔspdE strains by inoculating flasks of GF fish with individual strains, dissecting guts from 10 fish per flask every 60 minutes, and calculating the fraction of fish colonized over time (Figure S2C). Compared to WT^anc, both the evolved isolate and ΔspdE immigrated into fish more quickly (Figure 2C). This elevated immigration rate was not due to differences in abundance of the strains in the flask medium (Figure S2D). To test if loss of spdE impacted Aer01 migration from one fish to another, we mono-associated fish with individual strains for 24 hrs and added two donor fish to flasks of 10–15 GF recipient fish. After 15 hrs co-habitation, we found that the median ΔspdE abundance (765) in fish guts was nearly 10-fold higher than the median for WT^anc (85), with Aer01 below the limit of detection in four fish in the WT^anc group and none in the ΔspdE group, demonstrating that deletion of spdE enhances the rate of Aer01 migration into fish both from the aqueous environment and also from other fish (Figure 2D).

SpdE’s distal PAS/Cache domain binds a small set of amino acids

Sequence annotation of the spdE gene product predicted it to be a 544 amino acid protein containing two transmembrane regions and two functional domains—an N-terminal tandem PAS/double Cache domain (tPAS/dCache), and a C-terminal GGDEF (diguanylate cyclase) domain (Figure 3A). The mutations identified in the evolved genomes map to sites spread across the protein, suggesting that corresponding proteins are truncated, degraded or non-functional (Figure 3A). Based on this predicted architecture, it is presumed that this protein embeds into the inner membrane of Gram negative Aer01, with the GGDEF domain situated in the cytoplasm and the tPAS/dCache domain in the periplasm (Figure 3B). GGDEF domain-containing proteins are ubiquitous in bacteria, with the majority of bacterial genomes encoding multiple GGDEF-containing proteins in addition to other proteins important for production of c-di-GMP, an intracellular bacterial second messenger (Jenal et al., 2017). The Aer01 genome, for example, encodes 48 predicted GGDEF-containing proteins. PAS/Cache domains are well characterized as small molecule receptors that regulate concomitant effector domains such as kinases and nucleotide cyclases (Möglich et al., 2009; Upadhyay et al., 2016a), however, the ligand identity of most PAS/Cache-containing proteins is unknown (Henry and Crosson, 2011; Lacal et al., 2010; Wuichet and Zhulin, 2003). Proteins with similar protein architecture to SpdE (i.e. PAS/Cache domain linked to GGDEF domain) are abundant in bacteria, but only a few have been well characterized (Dahlstrom and O’Toole, 2017; Giacalone et al., 2018). Although SpdE contains a noncanonical GGDEF motif, GGEEL, in our analysis of a structural homology model, important GGDEF sequence motifs and structural elements show high preservation in SpdE, likely allowing for catalysis (see Figure S3).

Figure 3. SpdE protein architecture and ligand binding.

A) SpdE protein architecture. Aer01 SpdE residue numbers are below the bar. Specific mutations in the evolved isolates are labeled at their approximate locations (see Table S1). B) Schematic of SpdE integrated into the inner cell membrane. C) Representative thermal shift assay curves of the tPAS/dCache region of Aer01 SpdE (SpdE^pp). Circles show approximate melting temperatures (T_m) for control and proline. D) Thermal shift results (L-amino acids relative to control) of Aer01 SpdE^pp WT and three point mutants. Amino acid side chains (R group are drawn above the bars. n ≥ 6 technical replicates, 2–3 experimental replicates. E) Alignment of Aer01 and Aer02 SpdE amino acid sequences (tPAS/dCache domains only). Gray-highlighted residues, similar; black, identical; white, dissimilar. Pink and yellow boxes indicate putative binding site residues in the distal PAS/Cache domain and pink and purple boxes are residues mutated to assess ligand binding location, see panel D. F) Thermal shift assay results of Aer02 SpdE^pp. n ≥ 6 technical replicates, 2–3 experimental replicates. G) Competitive index data showing that Aer02 spdE can functionally complement Aer01 spdE in Aer01. See also Figures S3 and S4.

To identify SpdE ligands, we conducted a screen for thermal stabilization of the SpdE tPAS/dCache by candidate ligands, using Biolog Phenotype MicroArray plates as a source of compounds. Using a recombinantly expressed and purified periplasmic (pp) region of the protein (residues 38–286, SpdE^pp), we screened 194 compounds (see Methods and Table S2) and identified one polar and six hydrophobic amino acid ligand candidates. Thermal stabilization was confirmed using 10 mM L-amino acids (Figure 3C) and showed the greatest thermal shift with proline, followed by valine and isoleucine and to a smaller degree leucine, alanine, methionine, and threonine (Figure 3D; solid bars). Glycine was an example amino acid that confers no stabilization. D-isomers of proline and valine, which can serve as bacterial signaling molecules, stabilized much less than L-isomers (Figure S4A).

Searching the RCSB Protein Data Bank ((Berman et al., 2000), rcsb.org) for SpdE homologues identified the tPAS/dCache domain from McpN, a Vibrio cholerae chemoreceptor with alanine bound (PDB ID 3C8C). Sequence alignment of the proteins revealed residue conservation within the membrane-distal alanine binding site of McpN. To test if ligand binding occurs in the SpdE distal PAS/Cache domain, as is true in many tPAS/dCache proteins (Machuca et al., 2017), we generated protein variants with single amino acid changes in residues conserved between SpdE and McpN, two in the distal PAS/Cache (W140A and Y157A) and one in the proximal PAS/Cache (Y211A) (Figure 3A, D). Thermal stabilization by ligands was reduced by mutations in the distal, but not the proximal PAS/Cache (Figure 3D, E).

We next looked for related spdE genes in other Aeromonas species within our zebrafish gut isolate collection. The A. caviae (isolate ZOR0002, hereafter referred to as Aer02) genome contained a spdE homologue in a gene neighborhood with a high level of gene synteny to Aer01 (Figure S4B) and with amino acid sequence identity of 52% across the entire protein. Conservation across the SpdE tPAS/dCache region included key ligand binding residues (Figure 3E). Thermal shift analysis of the Aer02 SpdE^pp recapitulated the same pattern of ligand stabilization as for Aer01 SpdE^pp (Figure 3F). We further genetically complimented the Aer01 ΔspdE strain with spdE from Aer02 (ΔspdE_comp^Aer02), and found that it reduced the CI relative to WT Aer01 to the same extent as the Aer01 spdE gene (Figure 3G; Figure 2A).

SpdE tPAS/dCache crystal structure reveals proline binding specificity

We performed crystallization trials with both Aer01 and Aer02 SpdE periplasmic portions, but only Aer02 SpdE^pp produced high quality crystals sufficient for structure determination. Using a combined Rosetta/molecular replacement approach, we solved the structure of Aer02 SpdE^pp to 1.8 Å resolution (Figure 4A, PDB ID 7K5N, Video S1). Using the Aer02 structure and I-TASSER software (Zhang, 2008), we generated a homology model of Aer01 SpdE^pp which produced a similar fold to Aer02 with a RMSD of 1.41 Å over 232 Cα atoms (Figure 4B, gray). Aer02 SpdE^pp is found as a monomer in the crystal, however, we expect SpdE can form a homodimer in vivo, as is common for tPAS/dCache and GGDEF domains (Sweeney et al., 2018). A molecule of proline, seen with clear electron density, is bound in the distal PAS/Cache domain at the location predicted by our biochemical analysis (Figure 4A, C, D). Analysis of the Aer02 SpdE^pp structure B-factors showed the PAS/Cache domains to be well ordered, with interactions that facilitate ligand binding clearly visible (Figure 4A). Eight key residues and one backbone NH participate in direct interaction with the proline ligand (Figure 4D and Video S1). Hydrogen bonds are donated by the R144 guanidium group, W146 side chain indole and the NH backbone of Y166 to the proline carboxylate, and D193 and D165 accept hydrogen bonds from the proline NH₂. Hydrophobic interactions provided by I129 and pi stacking interactions from the aromatic side chains of Y119 and W133 stabilize the proline ring (Zondlo, 2013). Consistent with our mutational analysis (Figure 3D), the more essential W140/146 (Aer01/Aer02 sequence numbering) shows direct hydrogen bonding to the carboxylate of proline, whereas the less important Y157/163 plays a supporting role by stabilizing D187/193.

Figure 4. SpdE tPAS/dCache crystal structure provides insight into ligand binding.

A) Ribbon diagram of Aer02 SpdE^pp structure with proline bound (pink). Secondary elements are labeled (h, helix; s, beta strand). Shading based on atomic mobility (B-factors, black, more flexible; lighter teal, less flexible). PDB ID 7K5N. B) I-TASSER model of Aer01 SpdE^pp structure (gray) superimposed onto Aer02 SpdE^pp crystal structure (teal). C) Detail of the distal PAS/Cache proline binding pocket showing Fo-Fc “omit” electron density for proline as green mesh (contour level, 2.0 sigma). D) The eight distal SpdE residues and one main chain NH ligating bound proline are shown as teal sticks, bound proline in pink. Aer01/02 sequence numbering shows identically conserved binding site residues. Underlined residues indicate targets of site-directed mutagenesis, see Figure 3D. Hydrogen bonds to proline are shown, distances in Angstroms. For panels A, C and D: red, oxygen; blue, nitrogen.

Modeling D-Pro into the L-Pro binding site in Aer02 SpdE revealed possible reasons for L-isoform discrimination; the D-Pro ring is no longer parallel to the adjacent stabilizing rings of Y119 and W133 and the angles of the D-Pro carboxylate hydrogen bonds to R144 and W146 are no longer appropriate (Figure S4C, D). Additionally, modeling in L-Val revealed that it cannot participate in pi stacking interactions with Y119 and W133 as Pro does (Figures 3D & S4E), explaining the latter’s greater thermal stabilization. L-Ile also binds SpdE and is slightly larger than Val and Pro, but is predicted to fit in the pocket as well. In addition to the hydrophobic and ringed residues lining the ligand side chain region, cavity analysis of the binding pocket suggests only small hydrophobic amino acids could fit (Figure S4 F). All residues revealed to be components of the Aer02 SpdE ligand binding site are 100% conserved between Aer02 and Aer01 and predicted to be in similar locations in the folded proteins, strongly suggesting that Aer01 SpdE binds ligands similarly to Aer02 SpdE. Based on our sequence analysis and biochemical characterization, we have named this protein Sensor of proline diguanylate cyclase Enzyme, SpdE.

SpdE regulates Aer01 chemokinesis and biofilm formation via modulation of c-di-GMP

Because most of our evolved isolates contained predicted loss-of-function mutations in spdE, we hypothesized ligand-binding would inhibit SpdE activity, leading to decreased c-di-GMP levels and increased motility. To test this, we incubated WT and ΔspdE Aer01 in the presence or absence of amino acid ligands, then quantified c-di-GMP levels using mass spectrometry. We found that in the absence of ligand, WT Aer01 c-di-GMP levels were relatively high (15.81 c-di-GMP/OD₆₀₀), but in the presence of proline and valine, c-di-GMP was 2.7- and 1.7-fold lower, respectively (Figure 5A). We saw no reduction in c-di-GMP levels with the addition of the non-ligand amino acid, glycine. In comparison, c-di-GMP levels in ΔspdE were low, similar to the levels for the WT in the presence of proline, irrespective of the presence of ligand (Figure 5A). Complementation of the ΔspdE mutant with a WT copy of spdE (ΔspdE_comp) rescued some of the c-di-GMP levels (p=0.015), while complementation with a loss-of-function (1 bp deletion, isolate L3_P10) allele (ΔspdE_comp^evol) did not rescue c-di-GMP levels (Figure 5A). The ΔspdE_comp strain was highly aggregated, confounding cell enumeration and possibly underrepresenting per capita c-di-GMP levels (see Methods).

Figure 5. SpdE regulates Aer01 chemokinesis and biofilm formation via modulation of c-di-GMP.

A) Intracellular quantification of c-di-GMP in the absence or presence (1 mM) of SpdE ligands. WT^anc, ancestral Aer01; ΔsdE, knock-out strain; ΔspdE_comp and ΔspdE_comp^evol, ΔspdE complemented with a WT copy of spdE or an evolved mutant allele, respectively. n = 3 independent cultures. (B) Population level Aer01 motility for WT^anc and ΔspdE determined by exploration assay (see Figure S7); “reference” strain/condition specified at bottom. n = 2–6 experimental replicates. Bars are mean +/− SD. Dotted line represents fold change of 1 (i.e. no difference in motility). Blue outlined bars, ΔspdE; black outlined bars, WT^anc. C) Representative mass projection plots of movies of fluorescently-tagged WT^anc to visualize motility. White trails are “swimming tracks” of motile cells. Mass projections of all movies shown in Figure S8. D) Quantification of number of motile cells from all movies (see Figure S8) of WT^anc across a range of ligand. Mean +/− SEM is plotted. n = 3 movies/condition. E) Swim velocities of individual Aer01 motile cells (WT^anc and ΔspdE) incubated +/− 1 mM proline. Each box represents data from a single movie (n = 3–4 movies per strain/condition; number of motile cells/movie indicated below each plot); black outline, WT^anc; blue outline, ΔspdE; red fill, proline condition. Statistical groups (each set of box plots for a strain/condition) were compared by combining averages for each group (ANOVA, Tukey’s range test). F) Quantification of biofilm formation for WT^anc and ΔspdE Aer01 +/− ligand. Black outline, WT^anc; blue outline, ΔspdE. ANOVA, Tukey’s range test. n = 9, 3 experimental replicates. G) SpdE ligand impact on biofilm dispersal. Statistical significance (compared to buffer) determined by unpaired, two-tailed t test, *p<0.05. Pro, proline; Val, valine; Gly, glycine. n = 9–12; 3–4 experimental replicates. H) Chemotaxis response of Aer01 strains to amino acids. All amino acid responses compared to buffer control. Each data point represents an independent experiment. Bars represent mean +/− SD. Dotted line represents fold change of 1 (i.e. no difference in chemotaxis); n = 3–4 experimental replicates. I) Competitive indices of WT^anc, ΔspdE, ΔcheA, and ΔspdE/ΔcheA strains when competed in larval fish against a WT reference strain (see Figure S1A). Dotted line indicates CI of 1 (i.e. no competitive advantage). n ≥ 8 fish/condition, one experimental replicate. Data for WT^anc and ΔspdE is the same as those plotted in Figure 2A. *p<.05, two-tailed t test performed on log-transformed CI data. For box and whisker plots in E, F, and G, boxes represent median and interquartile ranges; whiskers represent the min and max. See also Figures S5 and S6.

Decreases in c-di-GMP levels increase bacterial motility (Dahlstrom and O’Toole, 2017), a trait enhanced in the evolved isolates in our previous study (Robinson et al., 2018). To confirm that decreased c-di-GMP levels observed in the presence of SpdE ligand increase Aer01 motility, we performed motility assays at both population and cellular levels. We first used an “exploration assay” to measure motility-facilitated bacterial population expansion into new regions (pipette tips) of identical media (Figure S5A). We quantified differences in the number of cells in the pipette tips after a brief exploration time by monitoring growth curves of the tip populations, and comparing those to a 5-log dilution series of the inoculum (Figure S5B). This allowed us to calculate “fold change in exploration” values of the bacterial populations (Figure S5D, E). Upon addition of ligand, WT increased exploration, with a stronger response to proline (mean fold change = 2.36) compared to valine (mean fold change = 1.47), consistent with thermal shift data and c-di-GMP quantification (Figure 3D, Figure 5A). In contrast, ΔspdE did not have a motility response to ligand (Figure 5B) and was more motile than WT regardless of the presence of ligand (Figure 5B; mean fold change = 3.43). These results were not explained by differences in bacterial abundances in the cultures (Figure S5F). Additionally, we found that Aer01’s motility was highly sensitive to proline as low as 100 nM, and was dose-dependent between 100 nM and 1 mM (Figure S5G).

We next imaged Aer01 populations on glass slides incubated in the presence of different concentrations of proline and valine. Mass projection of the video recordings shows tracks of individual cells (representative plots, Figure 5C; all plots, Figure S6). In the absence of ligand, cells were primarily non-motile, with many clumped in small aggregates, consistent with elevated c-di-GMP. At 1 µM and 1 mM of both proline and valine, the number of motile cells increased markedly, with the largest motile response for 1 mM proline (Figure 5D).

This imaging allowed us to visualize motile Aer01 population responses to SpdE ligands but was not ideal for measuring individual cellular velocities because the cells tended to stick or stall at the glass surface. We next imaged the cells with light sheet microscopy in a cuvette where the cells could move unimpeded by surfaces. Both the WT and ΔspdE were imaged in buffer without ligand and in the presence of 1 mM proline. Custom tracking software measured the velocity of motile cells within replicate movies for each strain and condition (Parthasarathy, 2012). In the absence of ligand, the average of the WT median cell velocity was 12.6 µm/sec, which doubled to 24.5 µm/sec in the presence of proline. The ΔspdE was faster than WT in the absence of ligand (mean= 28.9 µm/sec), and did not increase with the addition of proline (Figure 5E).

Biofilm formation and cellular motility are often inversely regulated, with decreases in c-di-GMP levels leading to less biofilm formation (Guttenplan and Kearns, 2013). To assay if SpdE ligands decrease Aer01’s propensity to form biofilms, we quantified biofilm formation using a crystal violet assay in the presence or absence of 1mM ligand. As anticipated, the WT formed more robust biofilms in the absence of ligand, and proline was more potent than valine in inhibiting biofilm formation (Figure 5F). In contrast, the ΔspdE strain formed less robust biofilms than WT in all conditions. We also tested the ability of SpdE ligands to disperse preformed biofilms. After just 1.5 hrs in the presence of proline we observed significant dispersal of Aer01 biofilms (Figure 5G). Combined, these results show how SpdE interprets the presence and concentrations of specific amino acids to regulate sessile versus motile states.

SpdE ligands mediate chemotaxis, in addition to chemokinesis

Many bacteria will chemotax toward or away from amino acids, which can serve as valuable nutrients or as general environmental cues (Yang et al., 2015). The Aer01 genome encodes >40 predicted chemoreceptors but their specific ligands are unknown. We tested if SpdE ligands are chemoeffectors for Aer01, and if these responses were spdE-dependent. Using an assay similar to the exploration assay, we tested if Aer01 cells suspended in a buffer would chemotax toward or away from pipette tips in which SpdE ligand was present. The data suggested that WT Aer01 perceives proline and valine, and to a lesser degree isoleucine and leucine as chemoattractants (Figure 5H; serine serves as a positive control). However, since the WT is able to undergo both chemotaxis and SpdE-dependent chemokinesis, these two responses conflate interpretation of the results. To disentangle chemotaxis and chemokinesis, we used the chemotaxis defective strain ΔcheA. Whereas ΔspdE showed a chemoattraction response to all ligands, ΔcheA had little to no response to any of the amino acids, including the control amino acid, serine, and only a small response to proline, likely due to chemokinesis. This interpretation was confirmed by the ΔspdE/ΔcheA double mutant which showed no response to any of the ligands. Combined, these results demonstrate that SpdE ligands are chemoattractants for Aer01 (Figure 5H).

To investigate the contribution of enhanced motility (chemokinesis) to host colonization in the absence of chemotaxis, we next assayed the competitive fitness of the ΔcheA and ΔcheA/ΔspdE strains in larval zebrafish. We have shown previously that bacterial symbionts deficient in chemotaxis are attenuated in host colonization, including a Vibrio and a different Aeromonas strain (Stephens et al., 2015; Wiles et al., 2020). We confirmed this to be true for the Aer01 strain, showing that ΔcheA had significantly reduced competitive fitness against the WT strain (Figure 5I). However, even in the absence of chemotaxis signaling ability, deletion of spdE provides a significant advantage for host colonization (Figure 5I). These data support that increased motility conferred by SpdE-dependent chemokinesis, independent of chemotaxis, bolsters host colonization.

SpdE ligands modulate Aer01 host colonization

Since WT Aer01 motility in the presence of SpdE ligands is comparable to ΔspdE motility (Figure 5), we hypothesized that we could alter competitions for GF larval zebrafish colonization between these strains by supplementing the inoculum and flask with SpdE ligands. Indeed, addition of 1 mM proline reduced the competitive index of ΔspdE from a median CI of 723 to a median CI of 2 (Figure 6A). As in previous assays, proline was more potent than an equivalent concentration of valine (median CI= 7), and proline concentrations as low as 10 nM elicited a measurable decreased response. We also directly measured immigration of the WT strain into GF zebrafish larvae in the presence of a SpdE ligand (1 mM valine) and observed significantly faster entry than in unsupplemented flasks and comparable to the ΔspdE mutant (Figure 6B; flask medium abundance control data in Figure S7).

Figure 6. Aer01 host colonization is modulated by SpdE ligands and mediated by the microbiota.

A) Competitive indices of ΔspdE (competed against the WT reference strain) +/− added amino acids in the system. Dotted line indicates CI of 1 (i.e. no competitive advantage). n ≥ 8 fish/condition, 1–3 experimental replicates. Median and interquartile ranges are plotted. Statistical significance compared to no amino acid control group determined by Kruskal-Wallis with Dunn’s multiple comparisons, ****p<.0001; **p<.01; *p<.05; ns, not significant. B) Determination of the effect of SpdE ligand (Val, valine; 1 mM) on immigration rate of WT^anc (see Figure S3A). The WT^anc (no Val) and ΔspdE data are the same as plotted in Figure 2C; included for reference. n = 2–3 experimental replicates per condition. Means (± SEM) are plotted. C) Competitive indices of ΔspdE (competed against the WT reference strain) in GF and CV larval zebrafish. Median and interquartile ranges are plotted. ****p<.0001, two-tailed Mann-Whitney test. n ≥ 28 (GF) or 52 (CV) fish combined from 3 (GF) or 6 (CV) independent experiments. D) Motility of WT^anc determined by exploration assay (see Figure S7) comparing motility in CV fish-conditioned flask media (FC-FM) to GF FC-FM. Each data point represents an independent experiment. n = 8 independent experiments using media collected from different flasks of GF and CV larval zebrafish. Bar = mean (+/− SD). Dotted line represents fold change of 1 (i.e. no difference). *p<0.05; one-sample t test, statistically different than 1. E) Quantification of biofilm formation for WT^anc in GF and CV FC-FM. Dotted lines connect data comparing GF and CV FC-FM collected on the same day from fish from the same egg clutch. n = 6 experimental replicates; each data point represents an independent experiment using CV or GF FC-FM collected from a different fish flask. *p<0.05, paired t test. F) Histogram of the number of predicted proteins in the zebrafish genome, binned according to % proline. Pro, proline. G) Quantification of WT^anc (black outlined bars) and ΔspdE (blue outlined bars) motility (exploration assay) in supernatant from collagenase-digested larval zebrafish (Col+F). Supernatant from untreated fish (Fish), untreated fish spiked with proline (F+Pro), and a collagenase only (Col) are included as references. Each data point represents an independent experiment; “reference” strain/condition indicated on bottom. n = 2 experimental replicates. Bar = mean. Dotted line represents fold change of 1 (i.e. no difference). See also Figure S7.

The microbiota mediates spdE-dependent Aer01 motility

The presence of a functional allele of spdE in the Aer01 genome suggests that this gene provides an advantage to the strain in certain environments, and that the adaptive advantage of spdE mutation was specific to the selective conditions of the simplified experimental evolution in GF zebrafish. To explore requirements for spdE in other conditions, we asked if ΔspdE has the same competitive advantage over WT in conventionally-reared (CV) zebrafish. In the presence of a complex microbiota, ΔspdE was significantly less competitive against WT (median CI = 19), compared to competitions in GF fish (Figure 6C), recapitulating competition with supplementation of SpdE ligands (Figure 6A).

To investigate differences in concentrations of SpdE ligands between GF and CV zebrafish conditions, we collected and filter-sterilized the flask medium of 4–6 dpf GF and CV zebrafish. Attempts to quantify amino acids via analytical techniques using either mass spectrometry or enzymatic methods were inconclusive due to the analytes being below or near the µM range detection limit for those methods (data not shown). As an alternative, we asked if we could detect differences based on Aer01’s phenotypic response, which was sensitive to ligand concentrations in the nM range. We first conducted exploration assays to measure WT Aer01 motility in GF fish-conditioned flask media (FC-FM) as compared to CV FC-FM. We saw a trend of increased motility of WT Aer01 in CV FC-FM, with a median fold change in exploration of 2.07 (Figure 6D; p=0.02). We next compared biofilm formation in these two conditions and observed that WT Aer01 biofilm formation was reduced in CV FC-FM relative to GF FC-FM (Figure 6E). Previous work in our group has shown high inter-individual variation in the composition of the CV microbiota in zebrafish larvae (Stephens et al., 2016). Consistent with this, we observed a high amount of variation in Aer01 motility and biofilm formation in CV FC-FM from different cohorts of larvae (Figure 6D, E). We interpret this variation in Aer01 behavior to reflect variation in the pools of free amino acids associated with different CV hosts, dependent on the functional capacities of their associated microbiota. Consistent with this interpretation, the median CI of ΔspdE relative to WT varied across independent competition experiments with CV fish generated on different weeks, suggestive of different SpdE ligand availabilities (Figure S7). Combined, these results suggest that the CV microbiota augment pools of SpdE ligands, thereby modulating Aer01 motility and facilitating host colonization.

We next investigated a potential mechanism by which the microbiota could augment the availability of SpdE ligands and alter Aer01 colonization. Given the potency of proline as a SpdE ligand, we considered possible sources of this amino acid in our system. Extracellular collagen, which is the most abundant protein in vertebrates (Duarte et al., 2014), is rich in proline residues which generate the polypeptide twists that stabilize the triple helix folds and underlie the structural stability of collagen fibrils (Chow et al., 2018; Karna et al., 2019). Zebrafish collagens, for example, are 13% proline (Figure 6F), a portion of which will be converted to hydroxyproline. Many bacteria, including those of the human gut, secrete a variety of collagenolytic enzymes that can collectively degrade collagen to smaller peptides and free amino acids (Duarte et al., 2014; Shogan et al., 2015; Zhang et al., 2015). We hypothesized that collagenolytic activity of the conventional fish microbiome releases proline and other SpdE ligands that stimulate Aer01 motility. To test this, we treated euthanized GF larval zebrafish with a commercially-available preparation of bacterial collagenase (see Methods). We then used the exploration assay to compare WT Aer01 and ΔspdE motility in the presence of supernatant from digested versus untreated GF fish. As controls, we also measured Aer01 motility in supernatant from the collagenase preparation alone or GF fish spiked with 1mM proline. Collagenase-dependent liberation of free amino acids from fish was confirmed by mass spec analysis, with a fold-fenrichment of 4.6, 2.8, and 3.14 for proline, valine, and isoleucine, respectively (3.2-fold total average; Figure S7C, D). Importantly, abundances of strains did not differ in the supernatants after incubation in the different conditions (Figure S7E). As expected, WT Aer01 exploration was elevated in the presence versus absence of 1 mM proline in the context of untreated GF fish supernatant (Figure 6G). The collagenase preparation alone, which contains some free amino acids, elicited a comparable exploration response to media from GF fish. Notably, the collagenase-treated GF fish supernatant increased exploration of WT Aer01 an average of 9.5-fold (Figure 6G). In comparison, ΔspdE exhibited little differences in exploration between any of the conditions. There was a small increase in exploration in the collagenase-treated GF fish supernatant compared to untreated fish, although much less than the WT (mean= 3.2). This is likely due to the presence of signals in this complex fish digest which impact Aer01 physiology independent of spdE. These data demonstrate that bacterial collagenolytic activity is sufficient to liberate SpdE ligands from zebrafish hosts and stimulate Aer01 motility, a trait that enhances host immigration.

Discussion

Host-associated microbiomes are shaped by transmission of microbial symbionts within host populations, driven by both stochastic and deterministic mechanisms (Burns et al., 2017; Miller et al., 2018; Robinson et al., 2019; Sarkar et al., 2020). Little is known about the extent to which microbes, especially non-pathogens, can influence transmission processes or discriminate between potential new hosts. Our findings reveal how chemokinesis in the presence of host-derived signals can allow bacteria to achieve colonization of motile hosts. Furthermore, by sensing signals that are differentially liberated by different microbiota, bacterial colonization decisions can be informed by hosts’ microbial associations_.

SpdE regulates bacterial colonization of hosts

From this work, we propose a model for SpdE-mediated regulation of Aer01 motility and host colonization (Figure 7). SpdE spans the inner cell membrane and contains a sensing domain that binds specific amino acid ligands in the periplasmic space and regulates activity of the cytoplasmic diguanylate cyclase (GGDEF) domain. In the absence of ligand, unbound SpdE is in a default “ON” state, increasing c-di-GMP, which inhibits motility and promotes biofilm formation (panel A). Upon ligand binding, SpdE turns “OFF,” reducing c-di-GMP levels, increasing motility, and decreasing biofilm formation (panel B). Loss of function mutation of spdE, as selected in the evolution experiment, functionally recapitulates the “OFF” state, and cells are locked in a motile lifestyle (panel C). In our system, hosts serve as sources of SpdE ligands (represented by the red “cloud” gradients in panel D). Aeromonas in the vicinity of the host senses and responds to these amino acid ligands by increasing motility (i.e. chemokinesis) and migrating along the chemical gradient toward the host (i.e. chemotaxis), thereby promoting immigration into the host. The concentrations of these amino acids emitted by a host will depend on the composition and activities of its resident microbiota. In proximity to a host generating relatively high amounts of ligand (e.g. “CV”), WT Aer01 cells (black outline, green fill) will become equivalently motile as the constitutively moving ΔspdE cells (blue outline, green fill) and both will be found in the intestinal population at appreciable amounts. In contrast, near a host emitting relatively low amounts of ligand (e.g. “GF”), the WT Aer01 cells remain more sessile (black outline, red fill), and are readily outcompeted by the highly motile ΔspdE cells for host colonization. The latter scenario exemplifies the conditions in our evolution experiment under which spdE loss-of-function mutants had a significant colonization advantage compared to WT cells.

Figure 7. A model for Aer01’s SpdE-dependent motility and host colonization.

A) unbound SpdE stimulates c-di-GMP production (“ON”). High intracellular c-di-GMP levels promote a sessile lifestyle and inhibit motility. B) Bound SpdE signals to turn “OFF” c-di-GMP production, switching Aer01 from sessile to highly motile. C) Loss of SpdE (e.g. ΔspdE) functionally recapitulates the “OFF” state, with low levels of c-di-GMP and a highly motile lifestyle. D) Hosts can be sources of varying amounts of SpdE ligands (red gradients) depending on microbiome composition. Aer01 can sense and respond to these amino acid gradients, via chemokinesis and chemotaxis, to modulate immigration into hosts emitting high amounts of ligand (CV fish) versus lower amounts of ligand (GF fish).

SpdE regulates c-di-GMP signaling through its tPAS/dCache domain

SpdE’s protein architecture reveals its function as an environmental sensor and signal transducer. Its tPAS/dCache portion belongs to a superfamily of sensor domains found in a variety of bacterial signal transduction proteins, including histidine kinases, diguanylate cyclases (DGCs) and phosphodiesterases (PDEs), other dinucleotide cyclases, and chemoreceptors (Upadhyay et al., 2016b). Proteins that contain tPAS/dCache-DGC architectures like SpdE are common and have broad phylogenetic distribution in bacteria (Henry and Crosson, 2011). Identifying functional SpdE homologs that have similar ligand preference or play similar roles in chemokinesis motility will require detailed mechanistic studies. In a few cases, ligands for PAS/Cache domains have been identified and include amino acids and other organic acids (Liu et al., 2018; Ud-Din and Roujeinikova, 2017; Zhang and Hendrickson, 2010). TM1987 of Salmonella Typhimurium is an example DGC with a PAS/Cache domain that senses L-arginine (Mills et al., 2015). Our high-resolution structure of SpdE^PP reveals how a PAS/Cache domain can discriminate between highly related ligands such as the L versus D stereoisomers of proline.

Intracellular c-di-GMP controls many facets of bacterial physiology (Conner et al., 2017; Jenal et al., 2017). Levels of c-di-GMP are regulated by complex networks of DGCs and PDEs that synthesize and degrade c-di-GMP, respectively, in response to intra- and extracellular inputs. The Aer01 genome encodes approximately 48 predicted DGCs and 17 predicted PDEs, but only mutations in spdE were selected in our experimental evolution, suggesting that SpdE transduction of information about proline, valine, and isoleucine is distinctly relevant for host colonization. We do not yet know how SpdE integrates into the cell’s c-di-GMP network to regulate Aer01 motility and biofilm formation. SpdE may act directly on the flagellar motor complex similarly to the c-di-GMP-regulated molecular brake YcgR in Escherichia coli (Boehm et al., 2010; Paul et al., 2010). Alternatively, SpdE modulation of c-di-GMP pools could impact motility less directly, for example via transcriptional regulation of flagellar genes. Notably, SpdE has a noncanonical GGEEL sequence in its enzymatic domain, although, like another noncanonical DGC (Hunter et al., 2014), it retains a predicted fold and key residues consistent with it being enzymatically active (Figure S4) (Schirmer, 2016). Alternatively, SpdE may regulate cellular c-di-GMP pools independent of enzymatic activity, as has been described for other PAS/Cache GGDEF proteins (Giacalone et al., 2018; Jenal et al., 2017).

Chemokinesis expands the spatial scales across which bacteria can colonize hosts

SpdE functions to regulate chemokinesis, a process distinct from the better-studied phenomenon of chemotaxis by which cells navigate along chemical gradients. Here, we refer to chemokinesis as the regulation of motility in response to changes in ligand concentration, regardless of a gradient. Chemically induced modulation of swim speed is used by several marine bacteria (Barbara and Mitchell, 2003; Garren et al., 2013; Son et al., 2016) and human-associated bacteria (Karmakar et al., 2016) for fine-tuning of environmental navigation and exploitation of resources (Hein et al., 2016; Packer and Armitage, 1994). In our system, the advantage of chemokinetic hypermotility is demonstrated by the difference in competitive fitness of the non-chemotactic, hypermotile ΔcheA/ΔspdE mutant compared to the non-chemotactic ΔcheA mutant (~47-fold). Combining hypermotility with chemotaxis confers an even greater advantage as seen in the competitive fitness difference between ΔspdE and the WT (~250-fold; Figure 5I). In this way, chemokinesis enhances the chemotaxis response and optimizes the bacteria’s response to chemical cues. For a bacterium to colonize a host traveling at speeds and along trajectories that far exceed those of chemotaxing bacteria, such as a swimming fish, chemokinesis can increase colonization success. It allows the bacterium to initiate energy-costly motility only in the vicinity of a fast-moving target, so that chemotaxis-enabled pursuit can be an effective colonization strategy.

SpdE ligands provide information about hosts’ microbial status

SpdE senses a small, selective repertoire of amino acids, which as the building blocks of the biological world, serve as useful spatial cues for heterotrophic microbes seeking out sources of nutrients. In a natural system there is dynamic heterogeneity in the landscape of amino acids from many sources, including hosts, environmental microbial populations and nutrient sources. SpdE’s selectivity for proline and to a lesser extent the branched chained amino acids (BCAA) valine, isoleucine, and leucine, suggests an ecological significance to Aer01’s chemokinesis beyond simple nutrient seeking. Our finding that Aer01’s SpdE-dependent chemokinesis response is enhanced in the presence of conventionally reared as compared to germ-free hosts, demonstrates that SpdE ligand generation is an emergent property of hosts and their resident microbes in our system. Proline and other SpdE ligands are ubiquitous and thus spdE could have evolved for any number of non-host specific reasons. Nonetheless, animal hosts could serve as important sources of SpdD ligands, such as those liberated from the proline-rich and abundant animal protein, collagen (Duarte et al., 2014), which can be degraded by the collective action of different microbial collagenases and peptidases (Zhang et al., 2015). Other potentially relevant sources of host-derived SpdE ligands are mucins, which also contain disproportionally high levels of proline, and are subject to breakdown by bacterially encoded mucinases (Faure et al., 2002). Additionally, BCAAs are abundant in animals and play important roles in tissue structure, metabolism and signaling (Neinast et al., 2019). Moreover, various host disease states, such as aberrant inflammation, can alter the physiological levels of BCAAs (Holeček, 2018). The association of SpdE ligands with both microbial activities and host physiology suggests that bacteria could use chemokinesis to make colonization choices based on a potential host’s microbial colonization history or health status.

STAR Methods

Resource Availability

Lead contact

Further information and requests for resources should be directed to the corresponding author, Karen Guillemin (kguillem@uoregon.edu).

Materials availability

Bacterial strains and plasmids generated by this study are available through the lead contact.

Data and code availability

The raw sequence reads are available through BioProject, accession PRJNA699275 (https://www.ncbi.nlm.nih.gov/sra/PRJNA699275). The A02 SpdE^pp protein structure is available through the RCSB Protein Data Bank, PDB ID 7K5N (https://www.rcsb.org/). The particle tracking software is deposited at GitHub (https://github.com/rplab/TrackingGUI_and_Localization_Public).

Experimental model and subject details

Ethics statement

All experiments using zebrafish were conducted in compliance with protocols approved by the University of Oregon Institutional Animal Care and Use Committee (IACUC); the animal protocol number for this work is 20–16 (42).

Zebrafish (Danio rerio) studies

All zebrafish experiments were conducted following standard protocols and procedures approved by the University of Oregon Institutional Care and Use Committee. Evolution passaging and bacterial competitions were performed using wild type (AB x Tu strain) or myd88 mutant zebrafish. The myd88 mutant zebrafish line was previously generated via CRISPR-Cas9 system and verified to have the expected phenotype of an myd88 KO mutant (Burns et al., 2017). Fish were maintained as previously described (Westerfield), and not fed in any of the experiments described here. Germ free derivation of fish embryos followed protocols previously described (Melancon et al., 2017). Generally, fish were inoculated with bacterial cultures at 4 days post fertilization (dpf). At 7 dpf, fish were euthanized with tricaine (Western Chemical, Inc.) following approved procedures, mounted in sterile 4% methylcellulose solution (Fisher), and the intestines removed by dissection (described in (Milligan-Myhre et al., 2011)) and used for enumeration of colonizing bacteria or as inoculum for GF fish. Conventional (CV) fish were generated by collecting unwashed embryos into crossing-tank water, sorting into flasks containing sterile flask medium (FM), then inoculating each flask with about 500ul of parental tank water. Flask medium contains the following formulation: NaCl (0.8 g/L), KCl (40 mg/L), Na₂HPO₄ (4 mg/L), KH₂PO₄ (6 mg/L), CaCl₂●2H₂O (1.2 mM), MgSO₄●7H₂O (1 mM), NaHCO₃ (4.2 mM).

Bacterial Strains

This study uses the bacterial zebrafish isolates Aer01 (Aeromonas veronii ZOR0001; BioProject Accession PRJNA205571) and Aer02 (Aeromonas caviae ZOR0002; BioProject Accession PRJNA205572), which were previously described (Stephens et al., 2016). Variants of Aer01 were generated using previously described genetic tools (Wiles et al., 2016; 2018). Briefly, markerless, in-frame deletion of spdE (IMG gene ID 2705589917) (ΔspdE), or cheA (IMG gene ID 2705588968) was generated by allelic exchange method, using the pAX1 allelic exchange vector (Wiles et al., 2018). Chromosomal insertions were used to generate fluorescently-tagged (dTomato or superfolder GFP) Aer01 strains, compliment ΔspdE with wild-type spdE (ΔspdE_comp), spdE evolved allele (1 bp deletion at nt 485; ΔspdE_comp^evol), or Aer02 spdE (IMG gene ID 2705595183) (ΔspdE_comp^Aer02), and over-express wild-type spdE in wild type Aer01 (WT^anc_OE). For insertions, a cassette containing the insertion gene under the control of the native promotor (complimented strains) or CP25 constitutive promotor (WT^anc_OE, and fluorescent-tagged strains) and a gentamycin resistance gene were integrated in the chromosome at a specific target location (attTn7). Detailed procedures and protocols are available online (see Additional Resources section). All genetically-modified strains were assayed for in vitro growth deficiencies in TSB broth; no overt fitness defects were detected. Aeromonas veronii (Aer01) and Aeromonas caviae (Aer02) strains were handled under traditional sterile technique and grown in tryptic soy broth shaking or on tryptic soy agar plates at 30°C, unless otherwise noted. Outside of the experimental evolution assay, minimal passaging was performed to prevent unwanted mutations and strains were maintained in 25% glycerol at −80°C. Antibiotics were only included when necessary. The E. coli strains were treated in a similar fashion, except they were grown in lysogeny broth or plates at 37°C, unless otherwise noted.

Method details

Evolution Experiment

Evolution passaging was initiated using an equal mixture of dTomato- and sfGFP-tagged ancestral strains, as a means of detecting gain-of-fitness lineages throughout the experiment, an experimental evolution approach previously described by Reyes et al. (Reyes et al., 2012). Three replicate lines were passaged in WT (AB x Tu) zebrafish and two replicate lines were passaged in myd88 mutant zebrafish. In all replicate lines, the dTomato-tagged lineages became dominant in the evolving populations by the end of the experiment and all of the evolved isolates described here-in descended from this ancestral strain. This could be suggestive of either a small fitness advantage in this genome, a small fitness defect in the sfGFP-tagged ancestral strain, or that the emergence of gain-of-fitness mutants arose in the dTomato lineage in all lines by chance. The first passage was inoculated by pelleting 1 ml TSB overnight cultures of dTomato- and sfGFP-tagged strains, resuspending them in 1 ml sterile flask medium (FM), mixing them 1:1, then adding them to replicate flasks of 4 dpf GF larval zebrafish (10–15 larval fish, 15 ml FM) to a final concentration of about 10⁶ CFU/ml. Inoculated fish were incubated according to IACUC protocol. At 7 dpf, fish were euthanized with tricane and the intestines removed by dissection (described in (Milligan-Myhre et al., 2011)). Whole intestines from all fish in a flask were combined into a single 1.6 ml tube containing 500 μl sterile FM and ~100 μl 0.5 mm zirconium oxide beads (Next Advance, Averill Park, NY), then homogenized using a bullet blender tissue homogenizer (Next Advance, Averill Park, NY; 30 seconds, power 4). Evolving Aeromonas populations were monitored by dilution plating a small aliquot (20 μl) of the combined gut sample, and an aliquot of the FM, on TSA plates. These were incubated at 30°C for 24 hrs, then the colonies counted and recorded. In order to monitor for contamination, colonies were screened for anomalous morphologies and also by fluorescence microscopy; none was detected. Half (~250) μl of each homogenate was mixed with 250 μl of sterile 50% glycerol, then stored at −80°C. The remaining homogenates were stored at 4°C for 4 days, then used as inocula for the subsequent flasks of 4 pdf GF fish. For subsequent inoculations, all ~200 μl of the 4°C sample was added to the next flask of fish (resulting in ~10⁴ CFU/ml at the beginning of the passage). This passaging protocol was repeated for 20 passages total for all five lines. Contamination was detected via genomic sequencing in line 4 after passage 15, therefore isolate and sequencing data for line 4, passage 20 (L4_P20) are not presented here.

Purification of evolved isolates

Cryopreserved stocks of whole populations from selected evolution passages (5, 10, 15, and 20) were streaked for isolation on TSA plates, then incubated at 30°C for 1 day. Isolated colonies were randomly picked from the plates into 5 ml TSB cultures, allowed to grow shaking at 30°C for ~6 hrs, then cryopreserved in 25% glycerol and stored at −80°C.

Preparation of genomic DNA sequencing libraries

Genomic DNA was extracted from overnight cultures (TSB) of evolved isolates and the ancestral strain using a Promega Wizard genomic DNA purification kit (Promega, Madison, WI). DNA samples were quantified using the Quant-iT dsDNA HS kit (Thermo Fisher) and normalized to 0.2 ng/µl Sequencing libraries were prepped using a Nextera XT Library Prep Kit (Illumina; FC-131–1096), according to the manufacturer’s protocol. Pooled sample was made by combining 80 ng of prepped library for each sample, and sequenced on an Illumina HiSeq 4000 (single end, 150-bp) at the Genomics and Cell Characterization Core Facility (University of Oregon).

Bacterial competitions in vivo

For in vivo bacterial competitions, strains (purified evolved isolates, or Aer01 genetic variants) were grown overnight in TSB from freezer stocks. One milliliter of the overnight cultures was pelleted (8,700 rcf, 2 min), then resuspended in 1 ml sterile FM. Competing strains were mixed, then added to flasks of 4 dpf GF (or CV fish; Figure 6) WT fish at ~10⁶ CFU/ml. For all competitions, sfGFP-tagged Aer01 was used as the “reference” strain. For competitions where SpdE ligands (amino acids) were added, a 100 mM stock solution of amino acid (proline, valine, glycine (BioUltra, Sigma-Aldrich)) was prepared in water (filter-sterilized; store at 4°C for up to a month), then added to the inoculum and fish flask to the desired final concentration (1 mM, 100 nM, 10 nM, as indicated in figure legends), and allowed to incubate at room temperature for 30 min to allow Aer01 to respond to ligand before inoculation of the flasks. At 7 dpf, fish intestines were dissected as described above, and each intestine transferred into a 1.6 ml tube containing 500 μl sterile FM and ~100 μl bullet beads, then bullet blended as described above. Homogenized samples were diluted appropriately in sterile FM, spread plated on TSA plates, incubated at 30°C for 1–2 days, and the colonies counted. Strains were differentiated by fluorescence microscopy. Competitive index (CI) was calculated by dividing the strain ratio (test strain divided by reference strain) in the gut at the end of the competition by the strain ratio in the inoculum (Figure S1A): $\frac{{\hspace{0.17em}}^{\left(\raisebox{1ex}{$test$}\!\left/ \!\raisebox{-1ex}{$reference$}\right.\right)}gut}{{\hspace{0.17em}}^{\left(\raisebox{1ex}{$test$}\!\left/ \!\raisebox{-1ex}{$reference$}\right.\right)}inoculum}$. The limit of detection is 5 CFU/gut; for samples where one strain is below the limit of detection, the abundance was set to 5 CFU in order to calculate a CI; therefore those points are either an over- or under-estimate of the actual CI, depending on which strain was undetected, and these data points are outlined in red (Figures 1C and 2A).

Gavage experiments

We followed a previously described gavage protocol (Cocchiaro and Rawls, 2013), with the following modifications (Figure S2A). Gavage needles were produced by pulling 3.5” capillaries (Drummond #3–000-203 GIX), then microforging them to an internal diameter of ~30 μm (DMF1000, World Precision Instruments), and polishing the ends. To prepare the gavaging inoculum 1 ml of TSB overnight cultures was pelleted (8,700 rcf, 2 min), then resuspended in 1 ml sterile flask medium (FM), and competing strains were mixed ~1:4 (competitor:reference); Ancestor (dTomato-tagged) or evolved isolates were competed against the differentially-tagged non-mutator ancestral strain, sfGFP-tagged. Culture mixes were then diluted 1:10 in sterile FM for gavage. Prepared inocula were incubated at room temperature until gavaging and flask inoculation (~30–60 minutes), allowing time for acclimation to the FM. Anesthetized fish (GF, 5 or 6 dpf) were transferred to 3% methylcellulose-coated gavage mold (4 % agar). Gavage needles were loaded with culture mix and 4.6 nl was gavaged directly into the lumen of the gut of individual fish using a Nanonject II (Drummond Scientific Company). Fish were rinsed post-gavage in sterile FM, then transferred into a flask containing sterile FM. Immediately after gavaging, flasks of GF fish were inoculated at 10⁶ CFU/ml with the same inocula used for gavaging. At ~5 hrs post-gavage fish were euthanized with tricaine, dissected, and the guts plated as described above to enumerate Aer01 competing strains.

Immigration rate experiments

Strains were grown, shaking, overnight in TSB, at 30°C. Overnight cultures were pelleted and washed in sterile FM, then diluted to 1:100 in sterile FM, and incubated at room temperature for ~2 hrs. If amino acid SpdE ligands were tested, amino acids were added to the prepared inoculum and fish flasks before the 2 hrs incubation. GF zebrafish (5 or 6 dpf) inoculated with the cultures to yield ~10⁵ CFU/ml. The fish were then split into replicate flasks containing 10 fish and 10 ml inoculated FM. An FM sample was taken and plated immediately to quantify CFU/ml of the inoculating strain. Subsequently, a replicate flask of fish (10 each) for each condition was dissected every ~60 min (for four time points) and the guts individually homogenized as described above in 200 μl sterile FM, and all 200 μl was spread plated on TSA plates (Figure S3A). FM samples were also plated to enumerate bacterial CFU/ml at each time point. Colonies on plates were counted after 24–48 hrs of incubation. To determine the proportion of fish colonized at each time point, a background of 2 colonies per sample, for all samples, was subtracted to account for bacterial cells carried over during the mounting and dissecting procedure that may not have originated from the gut. For each time point, the number of gut samples containing at least one colony (after background subtraction) was then divided by the total number of fish (= % fish colonized). This experiment was repeated three independent times for each strain.

Fish-to-fish transmission assay

At 4 dpf, flasks of GF larval zebrafish were mono-associated with either WT or ΔspdE Aer01. At 5 dpf, two mono-associated fish (‘donors’) were washed 6 times with sterile FM, then transferred into flasks containing 10–15 5 dpf GF larval zebrafish (‘recipients’). Fifteen hours later, all fish in the flasks were dissected and the guts homogenized and plated to enumerate Aer01 in the fish. Donor fish could not be distinguished from recipient fish, so these are included in the data presented in Figure 2D.

SpdE cloning and protein purification

Aer01 (residues 38–287) and Aer02 (residues 43–291) spdE periplasmic portions (tPAS/dCache region, between transmembrane 1 and 2) were subcloned into plasmid pBH using BamH1 and Xho1 by GenScript (Piscataway, NJ). pBH includes an N-terminal 6x His tag and linker which result in fusions to each protein. Aer01 SpdE periplasmic portion (Aer01 SpdE^pp) codes for 272 amino acids (including His tag and linker) and is approximately 32110 g/mol. Aer02 SpdE^pp codes for 271 amino acids (including His tag and linker) and is approximately 31420 g/mol. The three Aer01 SpdE point mutations (W140A, Y157A, Y211A) were created from the wild type Aer01 spdE pBH plasmid and mutagenized by GenScript. All plasmids were transformed individually into BL21 DE3 Escherichia coli for protein expression and purification. For protein expression and purification, the E. coli strains carrying the plasmids were grown at 37° C until OD₆₀₀ 0.4–0.6, then moved to 30°C and induced with 1 mM IPTG for 3–4 hrs. All subsequent steps were performed at 4°C. 1 – 2 L of pelleted E. coli culture were lysed in lysis buffer (50 mM Tris pH 7.5, 300 mM NaCl, and 10 mM imidazole) via sonication, and debris pelleted by centrifugation. The supernatant containing the protein of interest was washed over 5 ml bed volume of Ni-NTA resin (Qiagen, Hilden, Germany) in a gravity column that was pre-washed with lysis buffer. The resin was washed with 15X bed volume of lysis buffer, then 10X bed volume of lysis buffer with 30 mM imidazole, 10X bed volume of lysis buffer with 50 mM imidazole and finally eluted with 3x - 5x bed volume of lysis buffer plus 300 mM imidazole. The high absorbance (280 nm wavelength) fractions, generally 5–20 ml, were pooled and dialyzed overnight into 150 mM NaCl and 50 mM Tris pH 7.5, concentrated in centrifugal protein concentrators (Thermo Fisher Scientific, Waltham, MA), flash frozen in liquid nitrogen, and stored at −80°C as aliquots. SDS PAGE analysis was performed to determine a purity of > 90% for all proteins. Protein concentrations were determined by Abs₂₈₀ using the Beer-Lambert law. Proteins were thawed one time before each use.

Amino acid sequence alignments performed in Clustal Omega (McWilliam et al., 2013; Sievers et al., 2011) and colored using BoxShade. Gray-highlighted residues indicate similar side chains, black indicates identical and white are dissimilar.

Thermal shift Assay

Thermal shift assays were performed using a Thermo Fisher Scientific StepOnePlus Real-time PCR instrument. Each 20 μl sample contained 0.1 mg/ml SpdE protein, 4x SYPRO Orange dye, 165 mM NaCl, 80 mM Tris pH 7.5 and 10 mM ligand (free amino acid dissolved in water). All samples were done in triplicate and repeated 2–3 times on different days. All reagents and steps were performed at 4°C and placed in 96 well PCR plates. The PCR plate was slowly heated (~0.03 °C/sec) from 4°C to 80°C in the StepOnePlus instrument and fluorescent measurements were taken every 8.5 sec. The resultant thermal shift curves were analyzed to find the inflection points, which correspond to the melting temperature (T_m) of the protein (with or without ligand). Inflection points were determined in Microsoft Excel by calculating the halfway point between the minimum fluorescence value prior to fluorescence increase and the maximum fluorescence value. The T_m difference between protein with and without ligand was calculated and plotted to determine the thermal stability shift as a proxy for ligand binding.

Aer02 SpdE crystallization and x-ray collection

A02 SpdE^pp was crystallized at 22°C using the hanging drop method in 80 mM sodium acetate trihydrate pH 4.5 and 1.4 – 1.5 M sodium formate. The reservoir solution was mixed 1:1 with 1 µl of 5 mg/ml SpdE protein in 50 mM Tris pH 7.5 and 300 mM NaCl and suspended above 1 ml of reservoir solution. Crystals were briefly transferred to a cryoprotection solution including the reservoir solution, 10 mM proline and 20 % glycerol and flash frozen in liquid nitrogen. Several native diffraction data sets of A02 SpdE^pp were collected using the Advanced Light Source at the Berkeley Center for Structural Biology. We use beamline 5.0.2. at 1 Å wavelength on a Pilatus detector.

Aer02 SpdE^pp structure determination

Diffraction data for three frozen crystals (137_10, 137_13 and 137_14) were indexed and processed using the HKL2000 suite (Otwinowski and Minor, 1997). The resulting evaluation statistics are reported in Table S3. For crystals 137_10 and 137_13, systematic absence patterns were identified, leading to identification of the crystallographic space group as P4₁22 or P4₃22 with 97% confidence, but were less clear for 137_14. However, the merging statistics for crystal 137_14 were better, and intensities along the h and k axes were weak, so we could not be entirely confident in the identification. Data from crystal 137_14 were processed to 2.0 Å resolution in space group P422 and data were judged to be of reasonably high quality at a nominal resolution of 2.4 Å. The space group ambiguity was left for molecular replacement searches to resolve.

Amino acid homology searches (Söding, 2005; Zimmermann et al., 2018) identified 19 atomic models as structural homologs to the ligand binding domain of Aer02 SpdE in the Protein Data Bank, albeit with weak (15–20%) sequence identity, so we elected to use molecular replacement in the structure solution. We used the Phenix crystallographic package (Adams et al., 2010) to conduct molecular replacement (MR) searches. Search models included the top six hits identified using the HHPred server (Zimmermann et al., 2018) with PDB ID codes (3C8C, 3LIF, 4XMR, 5LTV, 5LTX and 6IOU). Numerous initial experiments using unmodified models, and models modified in various ways (polyalanine backbone, loops truncated, etc.) failed to yield interpretable MR solutions. So, we elected to use the Rosetta program package in concert with Phenix (mr_rosetta (DiMaio et al., 2011)) to build a large number of hypothetical models based on the six homologs identified above. In a typical step of this procedure, 5,000 to 10,000 different models are constructed by the Rosetta package, based on the Aer02 SpdE target amino acid sequence and the provided atomic model, and each resulting model is subjected to a fast MR search by Phenix. The results are sorted and surviving models can be optionally rebuilt using an automated model building procedure. In total, approximately 70,000 models (20,000 CPU hours, using six nodes in parallel with 32 CPUs each) were constructed on the University of Oregon High Performance Talapas cluster.

The effort resulted in a single viable solution, based on a single Rosetta derivative of the 3C8C model, which ultimately proved to be correct. This solution verified the space group assignment to be P4₁22, with Rwork/Rfee 0.44/0.55 at 2.0 Å resolution. Eleven other candidates were identified by the search procedure, but all had incorrect space group assignments and all had Rwork/Rfree of 0.48/0.55 or higher. Although the mr_rosetta search procedure did identify the correct MR solution, it is computationally extremely inefficient. It proved difficult to verify that the 3C8C solution was in fact correct, as initial automated model building and refinement met a dead end. Automated model building failed to lower the R-factor below 0.43, or to add missing segments to the model. The electron density maps were generally uninterpretable in terms of new features, however manual inspection of the model and map revealed that the Rosetta/Autobuild step had incorrectly reinterpreted several segments of the electron density map.

To correct the error, the Rosetta/Autobuild model was discarded in favor of the Rosetta-derived MR solution. That starting model was subjected to the Phenix morph_model procedure, then the map and model were inspected and the model truncated to remove incorrectly placed loops. Another round of morph_model, followed by a conventional autobuild cycle resulted in Rwork/Rfree of 0.417/0.529 and was clearly correct. Several additional cycles of model alterations by hand, using Coot (Emsley et al., 2010; Emsley and Cowtan, 2004), and conventional crystallographic refinement with Phenix led to a reasonably complete model with satisfactory R values at nominal resolution 2.4 Å. For the final steps of crystallographic model building and refinement, a better quality diffraction data set, 99.4% complete to 1.8 Å resolution, was obtained by merging the data from crystals 137_10 and 137_13 (Table S3). Final crystallographic model statistics are satisfactory and are summarized in Table S3. The final model of Aer02 SpdE comprises one continuous polypeptide chain, complete for residues 44–280. Weak electron density is apparent for addition residues at either end, but could not satisfactorily be interpreted either by hand or by automated model building, so these segments are assumed to be flexible. A single proline residue (not included in the crystallization mixture, but present in the cryoprotectant) was very clearly identified as bound to a tight pocket within the distal PAS/Cache domain, and a bound glycerol molecule was identified at the interface between the proximal and distal PAS/Cache domains. The PyMOL command “spectrum b” was used to color structure B-factors in Figure 4B (PyMOL Molecular Graphics System, Version 2.4.0, Schrödinger, LLC.).

Live imaging of Aer01 motility

To perform slide-based, wet mount live imaging of Aer01, sfGFP-tagged Aer01 was grown overnight in 5 ml of TB at 30°C. Cells were centrifuged at 1500 rcf and washed twice with sterile FM, which serves as ligand-free buffer. Washed cells were diluted to OD 0.2 and 100 µl of cells were dispensed into a sterile 96 well tray. Buffer (i.e. sterile FM) containing proline or valine (pH adjusted to 7.8 with HCl/NaOH) was added to the wells to generate the desired final concentrations of amino acid (0, 1 nM, 10 nM, 100 nM, 1 µM, or 1 mM) and cell densities of OD 0.1, each performed in triplicate. The tray was covered by parafilm and let incubate at 30°C for 5 hours. Cells were imaged between hour 5–6. Prior to imaging, bacterial samples were moved to a temperature-controlled box housing the microscope, heated to 30°C. 2 ul of bacteria from treatments were applied to a well slide (MP Biomedicals 10-well multitest slides #096041805), covered with a coverslip, and imaged immediately. Videos of bacteria were captured using a Nikon Eclipse Ti inverted microscope on the GFP channel using 20x magnification at 39 frames per second.

For light-sheet microscope-based imaging and determination of cellular swim speeds, overnight cultures (TSB) of strains (dTomato-tagged Aer01 and dTomato-tagged ΔspdE) were washed twice with buffer (i.e. sterile FM), as above. To 2 ml buffer +/− 1 mM proline, 50 µL of the cultures was added and incubated at 30°C for ~ 4 hours to recover and acclimate. A glass cuvette was filled with culture and imaged on a custom-built light sheet fluorescence microscope, as previously described (Taormina et al., 2012). The light sheet optically sections bulk samples in the center of the cuvette, so as not to constrain the motility of the imaged bacteria by surfaces. Movies in a single optical plane were captured for a duration of 20 seconds (frame rate of 30 frames/sec) with excitation light provided by a 561 nm solid state laser (Coherent Sapphire 20 mW; all strains expressed dTomato fluorescent protein). For each strain and condition, four movies were recorded from randomly selected regions throughout the cuvette (three for WT + proline).

Cyclic di-GMP quantification

Overnight cultures (TSB) of Aer01 Strains were pelleted (2 mL), washed once with sterile FM, and resuspended in sterile FM (+/− 1 mM proline, valine, or glycine) to a final volume of 8 ml. Cultures were incubated, shaking, at 30°C for 4 hours. Sample OD₆₀₀ was measured and the entire volume pelleted. Sample pellets were resuspended in 300 µl ice cold extraction buffer (40/40/20- acetonitrile/methanol/water + 0.1N formic acid), iced for 30 min, then pelleted at maximum speed in tabletop microcentrifuge for 10 min. The supernatant was transferred to a new tube and stored at −80° until being vacuum concentrated until completely evaporated. Samples were shipped (dry ice) to the Waters laboratory at Michigan State University for mass spectrophotometric quantification. Dried samples were resuspended in 100 µl Ultra Performance liquid chromatography-grade water and centrifuged for two minutes at 18,000 x g (Eppendorf® Centrifuge 5242R). The debris-free supernatants were moved to LCMS Certified Clear Glass 12 × 32 mm screw neck max recovery vials (Waters®). Ten microliters of each sample were analyzed using LC-MS/MS on a Quattro Premier XE mass spectrometer coupled with an Acquity Ultra Performance LC system (Waters®). Cyclic di-GMP was detected with electrospray ionization using multiple reaction monitoring in negative-ion mode at m/z 689.16→344.31. The mass spectrometer parameters were: capillary voltage, 3.5 kV; cone voltage, 50 V; collision energy, 34 V; source temperature, 110 °C; desolvation temperature, 350 °C; cone gas flow (nitrogen), 50 l/h; desolvation gas flow (nitrogen), 800 L/h; collision gas flow (nitrogen), 0.15 ml/min; and multiplier voltage, 650 V. Chromatography separation was done using a reverse phase Waters BEH C18 2.1 × 50 mm column with a flow rate of 0.3 ml/min with the following gradient of solvent A (10 mM tributylamine plus 15 mM acetic acid in 97:3 water:methanol) to solvent B (methanol): t = 0 min; A-99%:B- 1%, t = 2.5 min; A-80%:B-20%, t = 7.0 min; A-35%:B-65%, t = 7.5 min; A-5%:B-95%, t = 9.01 min; A-99%:B-1%, t = 10 min (end of gradient). Chemically synthesized c-di-GMP (Axxora) was dissolved in UPLC-grade water at concentrations of 250, 125, 62.5, 31.25, 15.62, and 7.81 nM to generate a standard curve for calculating the c-di-GMP concentration in each extract. For normalization across samples, the c-di-GMP concentration (μM) for each sample was divided by the OD₆₀₀ of the cultures. Of note, the ΔspdE_comp strain is highly aggregated in liquid culture, resulting in elevated OD₆₀₀ values as compared to the WT. This depressed the final per capita c-di-GMP values plotted in Figure 5A, likely underrepresenting the extent of complementation.

Exploration Assay

This assay was developed as a modification of a previously reported high-throughput capillary assay (Bainer et al., 2003). An overview of the protocol and calculation are presented in Figure S7. Overnight cultures (TSB) of Aer01 strains were pelleted (1 ml), and resuspended in 1 ml sterile FM. Two ml of the appropriate media (sterile FM (+/− 1 mM proline, valine, or glycine), or FC-FM (collected from either GF or CV fish) was added to glass culture tubes and 100 µl of the washed culture was added. FC-FM was generated by collecting the FM of GF or CV, 5 or 6 dpf zebrafish larvae, filter-sterilizing and storing at 4°C. The cultures were incubated, shaking, at 30°C for 3 hours. An aliquot of each culture was passed through a 0.2 µm filter (Corning® Costar® Spin-X® centrifuge tube filters) to generate cell-free supernatant. To a sterile 96-well plate round-bottom plate, 80 µl of unfiltered culture was added to each well to include five replicates for each condition. To another sterile 96-well plate round-bottom plate, 80 µl of the cell-free supernatant was added to each well to replicate the layout of the culture plate. Using a Rainin Liquidator™ 96-channel benchtop pipettor, 5 µl of the cell-free supernatant was pulled up into 20 µl Rainin pipette tips. These tips then were lowered down into the wells of the culture plate so the tips were submerged to half the depth of the culture volume in the wells. This set-up was incubated at room temperature for 30 min (“exploration time”), during which time the cells in the culture can swim up into the supernatant in the pipette tips. After incubation, the contents of the tips were ejected into the wells of a sterile 96-well plate (Corning, flat-bottom, #3595) to which 195 µl of sterile TSB medium was added to each well. This plate was immediately placed in a FLUOstar Omega microplate reader (BMG Labtech, Offenburg, Germany) and growth curves monitored by measuring absorbance (600 nm) every 10 min for 12–16 hrs, with 30°C incubation and constant shaking. From the growth curve data, using the mean of the replicates (5 for each strain/condition) for each time point, the time at which the absorbance passed 0.5 (T) was determined for each strain/condition. We first verified that T correlates with starting inoculum concentration by measuring growth curves across a 5-log dilution series of the inoculum (Figure S7B). We then used the slope of the linear regression line calculated from plotting the log-transformed CFU data as a function of T (Figure S7C), and the T values for each condition to determine “fold change in exploration” with the following equation (Figure S7D): ${10}^{\left(T_{test}-T_{ref}\right)/-1.34}$. A minimum of two independent experiments were conducted for each strain/condition.

Biofilm Assay

Overnight cultures (TSB) of Aer01 Strains were pelleted (1 ml), and resuspended in 1 mL (equal volume) of the appropriate media (sterile FM +/− 1 mM proline or valine, or FC-FM (collected from either GF or CV fish)). To the wells of a sterile 96-well plate (Corning, flat-bottom, #3595), 150 µl of each resuspended culture was added, in triplicate. A blank control well was added for each condition using the appropriate uninoculated media. This plate was incubated at 30°C, stationary, for 48 hours. For biofilm dispersal assays, 20 µl of 10 mM amino acid solutions (proline, valine, or glycine) were added to replicate wells (triplicate) containing 48 hr WT Are01 biofilm in FM, then incubated at 30°C for 1.5 hrs. All biofilms were quantified using a standard crystal violet biofilm staining procedure (O’Toole, 2011). For Aer01, the biofilm forms at the air-liquid interface (pellicle biofilm). Briefly, the supernatant was removed from each well, then the wells gently rinsed three times with 150 µl sterile FM. Then, 150 µl 0.1% crystal violet was added to each well and the plate was incubated at room temperature for 10 minutes, followed by five rinses with 150 µl sterile FM. To destain, 150 µl 95% ethanol was added to each well and the plate was incubated at room temperature for 10 minutes. Each well was mixed well and the contents transferred to a clean 96-well plate and the absorbance (570 nm) was read on a FLUOstar Omega microplate reader (BMG Labtech, Offenburg, Germany).

Chemotaxis Assay

This assay was based on a previously reported high-throughput capillary assay of quantifying bacterial chemotaxis responses (Bainer et al., 2003). Overnight cultures (TSB) of Aer01 strains were pelleted (1 ml), and resuspended in 1 ml sterile FM, and 100 µl of the washed culture was added to 2 mL of sterile FM in glass culture tubes. The cultures were incubated, shaking, at 30°C for 3 hours. To a sterile 96-well round-bottom plate, 80 µl of culture was added to each well to include three replicates for each condition. To another sterile 96-well round-bottom plate, 80 µl of 1 mM amino acid solution (proline, valine, isoleucine, or serine; prepped in sterile FM) was added to each well. Serine was included as a positive control. Using a Rainin Liquidator™ 96-channel benchtop pipettor, 5 µl of the amino acid was pulled up into 20 µl Rainin pipette tips. These tips then were lowered down into the wells of the culture plate so the tips were submerged to half the depth of the culture volume in the wells. This set-up was incubated at room temperature for 60 min, during which time the cells in the culture can swim up into the pipette tips. After incubation, the contents of the tips were ejected into the wells of a sterile 96-well plate (Corning, flat-bottom, #3595) to which 195 µl of sterile TSB medium was added to each well. This plate was immediately placed in a FLUOstar Omega microplate reader (BMG Labtech, Offenburg, Germany) and growth curves monitored by measuring absorbance (600 nm) every 10 min for 12–16 hrs, with 30°C incubation and constant shaking. From the growth curve data, using the mean of three replicates for each time point, the time at which the absorbance passed 0.5 (T) was determined for each strain/condition. A “chemotaxis response” was calculated for each amino acid using the same approach and equation $\left({10}^{\left(T_{test}-T_{ref}\right)/-1.34}\right)$ as for “fold change in exploration” (Figure S7D), comparing the time to OD 0.5 for each amino acid (=T_test) to that for the buffer (i.e. FM) control (=T_ref). A minimum of three independent experiments were conducted for each strain/condition.

Collagenase assays

GF larval zebrafish (6 dpf) were euthanized with tricaine, washed once with sterile FM, and resuspended in sterile FM. Fish were combined into Eppendorf tubes (13 fish/tube) in 800 µl sterile FM. A solution of 10 mg/ml Collagenase P (Sigma 11213857001) was freshly prepared in Hanks’ Balanced Salt Solution (HBSS). To one tube of fish, 200 µl of collagenase solution or HBSS was added. A control tube containing 800 µl FM (no fish) and 200 µl collagenase solution was also prepared. All three tubes were incubated at 37°C for 4 hrs. After incubation the tubes were spun at max speed in a microcentrifuge for 2 min, and the supernatant passed through a 0.2 µm filter (Corning® Costar® Spin-X® centrifuge tube filters) to generate cell-free supernatant. To quantify Aer01 motility in media containing these supernatants, the exploration assay was performed as described above with the following modifications. Media was prepared by mixing 450 µl of the supernatants with 2 ml sterile FM. To one tube of media prepared with undigested (no collagenase) fish supernatant, 5 µl of 100 mM proline solution as added to serve as a positive control for increased motility. One milliliter of each of these medias was added to a glass culture tube, 50 µl of washed overnight media was added, and the cultures were incubated at 30°C for 3 hrs. A 60 min “exploration time” was used for these samples. To quantify amino acid concentrations in the supernatants, sample aliquots were sent to the Proteomics & Mass Spectrometry Facility at the Donald Danforth Plant Science Center (St. Louis, MO). For this analysis, 50 µl of sample was mixed with 10 µl of 250 µM of ₁₃C and ₁₅N labeled amino acid internal standards. The samples were dried in a speed vacuum centrifuge, then resuspended in 500 µl of 0.1 N HCl. After filtration, the samples were transferred to LC vials for LC-MS/MS analysis. Additional dilutions were made when necessary to bring the signal of the injected sample within the linear range of calibration.

Quantification and Statistical Analysis

Genomic DNA sequencing analysis

Each sample averaged 6.2M reads. Processed reads were aligned and analyzed against the Aer01 reference genome (Aeromonas veronii ZOR0001; BioProject Accession PRJNA205571) using breseq (Deatherage and Barrick, 2014) with default settings; mean coverage of 102x per genome.

Quantification of Aer01 motility

Videos of treated bacteria were analyzed with particle tracking software, which uses a radial-symmetry-based algorithm, from the lab of Dr. Raghu Partharasarathy (publicly available on GitHub: https://github.com/rplab/TrackingGUI_and_Localization_Public) using these settings: “objects” were identified using bpfilter 3, nsize 7, and gradobjsize 0. A standard threshold of 3.99–6 was applied, depending on the degree of background. Motile bacteria were defined as objects tracked over at least 1 s (39 frames) with a standard deviation in position of pixels/frame of at least 1. For light-sheet based imaging analysis, the same particle tracking software (above) was used with the following parameters: bpfiltsize=7, nsize=7, gradobjsize=7, 1/nhood=true. Tracks were culled using very stringent criteria in an effort to capture accurate cellular swim speeds. Only tracks with minimum length of 30 frames (tracked for 1 sec) were included, resulting in the number of included tracks in the range of 50–308 (labeled in Figure 5E).

Statistical Analysis

Statistical analyses were performed using Prism 6 (GraphPad Software), with each statistical test used specified in the corresponding figure legend. Two-tailed unpaired Student’s t test was used for statistical analysis to determine significant differences when a pair of conditions was compared. Asterisks denote statistical significance (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). The exact value of n representing number of animals or samples, and number of experimental replicates, are included and described in each figure legend. Group statistics (e.g. mean, median), if plotted, are indicated in the figure legends.

Additional Resources

Genetic modification of Aeromonas

Genetic modification of Aer01 was conducted using previously described genetic tools developed for the study of wild and diverse proteobacterial lineages (Wiles et al., 2016; 2018). Detailed procedures and protocols are available online (https://doi.org/10.6084/m9.figshare.7040258.v1).

Supplementary Material

2

Video S1. SpdE’s 1.8 Å resolution crystal structure and proline binding site, related to Figure 4.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
Aeromonas veronii ZOR0001	Karen Guillemin lab; Stephens et al., 2016	BioProject Accession PRJNA205572
Aeromonas caviae ZOR0002	Karen Guillemin lab; Stephens et al., 2016	BioProject Accession PRJNA205572
Aeromonas veronii ZOR0001, ΔspdE	This study	N/A
Aeromonas veronii ZOR0001, ΔcheA	This study	N/A
Aeromonas veronii ZOR0001, ΔspdE_comp	This study	N/A
Aeromonas veronii ZOR0001, ΔspdE_compevol	This study	N/A
Aeromonas veronii ZOR0001, ΔspdE_compAer02	This study	N/A
Aeromonas veronii ZOR0001, WTanc_OE	This study	N/A
Aeromonas veronii ZOR0001, dTomato	This study	N/A
Aeromonas veronii ZOR0001, superfolder GFP	This study	N/A
Escherichia coli BL21 DE3	New England Biolabs	Cat# C2527H
Escherichia coli SM10	Karen Guillemin lab	N/A
Chemicals, Peptides, and Recombinant Proteins
Ni-NTA resin	Qiagen	Cat# 30210
SYPRO Orange dye	Thermo Fisher	Cat# S6650
Collagenase P	Sigma	Cat# 11213857001
Critical Commercial Assays
Nextera XT Library Prep Kit	Illumina	Cat# FC-131–1096
Promega Wizard genomic DNA purification kit	Thermo Fisher	Cat# PR-A1120
Quant-iT dsDNA HS kit	Thermo Fisher	Cat# Q32851
Deposited Data
Raw sequence reads	This study	BioProject Accession PRJNA699275 (https://www.ncbi.nlm.nih.gov/sra/PRJNA699275)
A02 SpdEpp protein structure	This study	PDB ID: 7K5N
Experimental Models: Organisms/Strains
AB x Tu wild type zebrafish, Danio rerio	Univ. or Oregon Zebrafish facility	N/A
Myd88 mutant zebrafish, myd88−/−, Danio rerio	Karen Guillemin lab; Burns et al., 2017	N/A
Recombinant DNA
Plasmid: Aer01 spdEpp pBH	This study	N/A
Plasmid: Aer02 spdEpp pBH	This study	N/A
Plasmid: Aer01 spdEpp W140A pBH	This study	N/A
Plasmid: Aer01 spdEpp Y157A pBH	This study	N/A
Plasmid: Aer01 spdEpp Y211A pBH	This study	N/A
Plasmid: pAX1 allelic exchange vector	Karen Guillemin lab; Wiles et al., 2018	N/A
Software and Algorithms
breseq	Deatherage et al., 2014	https://barricklab.org/twiki/bin/view/Lab/ToolsBacterialGenomeResequencing
Phenix mr_rosetta	DiMaio et al., 2011	https://www.phenixonline.org/documentation/reference/mr_rosetta.html
particle tracking software	Parthasarathy lab Univ. of Oregon; GitHub	https://github.com/rplab/TrackingGUI_and_Localization_Public
Clustal Omega	McWilliam et al., 2013; Sievers et al., 2011	https://www.ebi.ac.uk/Tools/msa/clustalo/
BoxShade	N/A	https://embnet.vitalit.ch/software/BOX_form.html
HKL2000 suite	Otwinowski and Minor, 1997	https://hklxray.com/hkl-2000
Phenix 1.16.3549	Adams et al., 2010	https://www.phenixonline.org/
PyMOL 2.4.0	Schrödinger, LLC	https://pymol.org/2/
Other
Detailed protocol for genetically modifying Aeromonas	figshare.com	https://doi.org/10.6084/m9.figshare.7040258.v1

Highlights.

• Bacteria can regulate motility, via chemokinesis, to trigger immigration into hosts.

• In Aeromonas, SpdE controls chemokinesis in response to host-emitted amino acid cues.

• SpdE’s tPAS/dCache crystal structure reveals proline binding specificity.

• The host microbiome mediates spdE-dependent Aeromonas host colonization.