In 2014, crustal fluids were collected from a subseafloor borehole fitted with a Circulation Obviation Retrofit Kit (CORK), designated as ‘U1383C’ and located at the North Pond site (22°45′N, 46°05′W) along the western flank of the Mid-Atlantic Ridge (Edwards et al., 2012a). Samples were collected at two depth horizons beneath the seafloor: a ‘shallow’ (58–142 m) horizon and ‘deep’ (200–330 m) horizon, as described in (Meyer et al., 2016).

Fluid samples were plated on 10 cm plastic petri dishes containing ~20 mL of a medium modified from DSMZ Medium-113¹ (Kelly and Wood, 2000). A 2X modified media solution (herein referred to as modified DSMZ-113) was made containing (per 500 mL DI-H2O): 2.0 g KH2PO₄, 2.0 g KNO₃, 1.0 g NH₄Cl, 0.8 g MgSO₄ × 7 H₂O, 5.0 g Na₂S₂O₃ × 5 H₂O, 2.0 mg FeSO₄ × 7 H₂O (solubilized in 0.1 N 102 H₂SO₄), 1.0 g NaHCO₃, and 2 mL MC-TMS trace element solution (ATCC, Manassas, VA, USA). The medium was adjusted to pH 7.0 with NaOH and filter sterilized into carbon-free glassware (combusted at 400°C for 5 h). Prior to filtration, each 0.2 μm filter was washed twice with sterile water to prevent potential carbon-source carry over from filter paper “wetting agents” (believed to be glycerol or other utilizable organic compounds). To solidify the medium for plates, 15 g of agar (Fisher Scientific, Fair Lawn, NJ, USA) was autoclaved in 500 mL DI-H₂O and combined with 500 mL of 2X modified DSMZ-113 when cooled below ~50°C. After plating, replicate dishes were incubated at 4°C and 20°C aerobically and at 20°C anaerobically until individual colonies were visible. Distinct colonies that grew on the plates were transferred to combusted carbon-free glass culture tubes containing 5 mL of 1X modified DSMZ-113 and incubated at 20°C while shaking at 180 rpm. A sample from each culture that grew turbid was stored in 15% glycerol at −80°C. The isolation details including the specific fluid sample origin for each of the 46 strains can be found at doi: 10.26300/6f5k-za64.

DNA was extracted from each isolated culture using the Biostic bacteremia DNA Isolation Kit (MoBio, Carlsbad, CA, USA) and stored at −20°C. Extracted DNA was PCR amplified with universal 16S rRNA primers 8F (5’-AGAGTTTGATCCTGGCTCAG) and 1492R (5’-GGTTACCTTGTTACGACTT) [3 min at 94°C, 35 cycles of, 40 s at 94°C, 1.5 min at 55°C, 2 min at 72°C and a final extension for 10 min at 72C]. PCR products were purified with MinElute PCR purification kit (Qiagen, Valencia, CA, USA). Bidirectional Sanger sequencing was performed at the Marine Biological Laboratory (Woods Hole, MA, USA) on an AB 3730XL Genetic Analyzer (Thermo Fisher Scientific, Waltham, MA, USA) using AB BigDye3.1 chemistry. Quality scoring and merging into full-length 16S rRNA gene sequences was performed with Phred and Phrap (Ewing et al., 1998). Taxonomy at the genus level was determined with NCBI BLASTn. Sequence alignment was performed using mothur (Schloss et al., 2009) and a neighbor-joining phylogenetic tree was generated with ClustalX (Larkin et al., 2007) using 1,000 bootstrap trials.

Forty-six strains identified as Halomonas were outgrown from glycerol stocks in modified DMSZ medium 113 at 30°C in 5 mL cultures containing Luria Bertani (LB) broth (Lennox; 10 g Tryptone, 5 g Yeast Extract, 5 g NaCl, components from BD) until turbid, and stored in LB with 20% glycerol at −80°C.

For testing growth and survival dynamics at high-nutrient conditions, strains were incubated in 5 mL LB broth in 18 × 150 mm borosilicate tubes, at 30°C rolling in a TC-7 roller drum (New Brunswick Scientific). For testing survival under low-nutrient conditions, strains were incubated in the modified DSMZ-113 medium in 18 × 150 mm borosilicate tubes, at room temperature (18°C) rolling in a TC-7 roller drum (New Brunswick Scientific). All starter cultures were first inoculated into modified DSMZ-113 medium for 5 days to allow carry-over carbon to be depleted. These ‘carbon-depleted’ cultures were then used to initiate all low-nutrient growth and survival experiments in fresh, no-carbon-added, modified DSMZ-113 medium. All viable cell counts were measured using the spot titering assay plated on LB agar (Kraigsley and Finkel, 2009) with a limit of detection of <1,000 CFU/mL.

Nine representative strains were incubated in triplicate in 5 mL LB broth, as described above. Every 2 days, 5 μL of these 27 cultures (nine representative strains in triplicate) were re-inoculated into a fresh 5 mL LB culture and propagated for a total of 30 passages (Kram et al., 2017). After 30 passages, evolved populations were stored in LB with 20% glycerol in −80°C.

The growth and survival patterns of each of the strains were compared to their respective parental strains when incubated in DSMZ-113 medium, with no addition of carbon or energy sources. However, it is clear that some form of bioavailable organic compounds exist that support low levels of microbial growth under these conditions. Evolved strains that grew significantly worse overall compared to their parental strain, as reflected by either demonstrating a reduced relative cell yield after 5 days, a slower growth rate, and/or entering death phase earlier, were chosen for further analysis. Each candidate population was plated on LB agar. Twelve clones were then picked from individual colonies on each plate, grown overnight in LB medium, and stored in LB with 20% glycerol at −80°C.

DNA from each parental strain and the corresponding 12 evolved clones from each strain was extracted from ~10⁹ cells using the ZymoBIOMICS DNA Miniprep Kit. To obtain the reference genome of the parental strains, a combination of long reads using Nanopore (Wang et al., 2021) and short reads using NextSeq (Illumina) were used to assemble the whole genome. Whole-genome sequencing-library preparation and short-read sequencing of the clones were performed using the NextSeq2000 platform. All sequencing, genome assemblies, and gene annotations were performed by the Microbial Genome Sequencing Center (MiGS), Pittsburgh, PA. Briefly, post sequencing, quality control and adapter trimming was performed with bcl2fastq (Illumina, Inc., n.d.) and porechop (GitHub, Inc., n.d.) for Illumina and ONT sequencing, respectively. Hybrid assembly with Illumina and ONT reads was performed with Unicycler (Wick et al., 2017). Assembly annotation was performed with Prokka (Seemann, 2014).

Genomic sequences of evolved clones were aligned to each respective parental genome using BreSeq version 0.36.0 (Deatherage and Barrick, 2014) in consensus mode to identify SNPs, small indels, deletions, and mobile genetic elements. The comparison of presence or absence of genes was analyzed using Roary (Page et al., 2015). Each genome was visualized through Geneious R8.1.9 software and genomic rearrangement analysis was done using the progressiveMauve algorithm (Darling et al., 2004).

Fluids from the deep and shallow horizons of CORK observatory U1383C that were plated on autotrophic minimal media (modified DSMZ-113) generated distinct, uniform colonies. These colonies were small (~1–2 mm), whitish tan in appearance, and relatively slow growing. The plates incubated at 20°C aerobically exhibited barely visible colony growth after 7 days and distinguishable colonies at ~10 days. Those incubated at the same temperature anaerobically did not show growth, nor did those inoculated aerobically at 4°C. Therefore, all further cultured strains originated from the 20°C aerobic colonies and were given a North Pond diversity identification number (NPDiv#). NPDiv1-34 (n = 32) were from the 1383C deep horizon and NPDiv35-52 (n = 14) were from 1383C Shallow (Supplementary Table 1).

Sanger sequencing of cultures revealed that the majority (46/52) belonged to the genus Halomonas and were closely related to one another. Isolates NPDiv37, 43, and 50 were identified as Pseudomonas and not used in further experiments. NPDiv8, 10, and 49 failed to sequence well enough to determine taxonomy and were not used in further experiments. Phylogenetic analysis of North Pond Halomonas sequences indicated that all the North Pond isolates grouped with other isolates from cold deep seawater samples (‘Ecotype 2B’) (Kaye et al., 2011; Supplementary Figure 3).

All 46 isolated Halomonas strains were incubated in LB medium, and their growth and survival patterns were determined. Nine different phenotypic groups were observed after incubating in batch culture at 30°C in LB for 4 days (Figure 1). The features used to distinguish each group are described in detail below. They include: (i) initial growth yield at the end of log phase, (ii) duration of stationary phase, (iii) time of entry into and duration of death phase, (iv) severity of loss of cell viability during death phase, and (v) the post-death phase dynamics of each strain.

The overnight growth yields were determined for each culture. Groups 1 through 6 (Figures 1A–F) had an average overnight yield of ~2.5×10⁹ CFU/mL while Groups 7 through 9 (Figures 1G–I) displayed yields that were ~ 10-fold lower, at ~5.8 × 10⁸ CFU/mL. Comparing the lengths of stationary phase, strains in Groups 1 through 6 (Figures 1A–F) exhibited a 1-day stationary phase, compared to Groups 7 through 9 (Figures 1G–I) whose stationary phase was twice as long lasting for 2 days.

The timing of entry and duration of death phase also varied considerably between each strain. Of the strains that reached a maximum cell yield of ~10⁹ CFU/mL upon entry into stationary phase, Group 1 (Figure 1A), Group 3 (Figure 1C), and Group 4 (Figure 1D) strains have death phases that last for 2 days. However, among these strains, there were differences in the degree to which cells died: strains in Groups 1 and 4 showed a reduction in viability of ~100-fold, while Group 3 strains showed a more modest 10-fold loss in viability. While Group 2 (Figure 1B), Group 5 (Figure 1E), and Group 6 (Figure 1F) strains also reached ~10⁹ CFU/ml on day 1, their death phases continued through day 4 of the experiment, with populations never entering Long-Term Stationary Phase(). However, the magnitude of the extent of death phase also varied with these three groups, where Group 2 and 6 strains showed a 10-fold decrease in viability, compared to Group 5 strains that suffered up to 1,000-fold decreases in cell viability. Among the strains that reached ~10⁸ CFU/mL at the end of log phase, Group 7 strains (Figure 1G) had a death phase that lasted 2 days, while Group 8 and 9 strains (Figures 1H,I) had a 1-day death phase. Group 7 and 9 strains exhibited ~100-fold decreases in cell yield, while Group 8 strains showed 10-fold losses in viability. Lastly, the post-death-phase dynamics of these groups also differed. Specifically, Group 1 exhibited noticeable re-growth after death phase where cell counts increased ~8-fold, while Group 2, 5, 6, and 7 strains were still declining in cell yield by the end of the experiment. This is in contrast to Groups 3, 4, 8 and 9 strains, which maintained viability at a constant cell density after death phase.

We selected 9 strains and serially passaged them in triplicate, creating a total of 27 individual cultures, for 30 passages in LB (~300 generations) (see Materials and Methods; Supplementary Figure 1). Following the opportunity for adaptive evolution in rich medium, we compared the survivability of each of the 27 cultures in low-nutrient DSMZ-113 medium to its original parental strain and ultimately selected 8 cultures, described below, that exhibited significant growth differences for further study (Supplementary Figure 1). We refer to these 8 cultures as populations A through H (Figure 2). Population A originated from a single parental strain within Group 1. Population B originated from a single parental strain within Group 2. Population C originated from a single parental strain within Group 3. Populations D and E originated from different replicates of the same parental strain within Group 4. Populations F and G originated from different parental strains within Group 7. Lastly, population H originated from a parental strain within Group 8. Therefore, the eight populations, A through H, originated from seven parental strains. In the following sections of this study, parental strains are referred to using their Group number.

For each of the populations that exhibited changes in growth or survival patterns when incubated in minimal medium without additional carbon, cultures were streaked to single colonies and 12 individual clones from each population were selected at random. The growth and survival of each of these “evolved” clones were then compared to their respective parental strains. Overall, the majority of the clones isolated from the evolved populations were less fit than their parent without any addition of carbon (Figure 2). In population A (Figure 2A), 4 clones had reduced Day-5 yields, ranging from ~7×10³ CFU/ml to ~3×10⁵ CFU/mL. The yields of the 8 remaining population A clones were below the limit of detection by day 5, compared to the parental strain with a final yield of ~8×10⁵ CFU/mL (Figure 2A). For population B, only two clones were detectable by day 5 (Figure 2B). For population C, all clones were overall less fit in comparison to the parental strain (Figure 2C). For population D, all 12 clones were less fit, with final yields above the limit of detection by day 5 (Figure 2D), ranging from 5×10³ CFU/ml to 1×10⁶ CFU/mL. However, population E, though derived from the same parental strain as population D, produced clones with significantly worse growth yields, resulting in only one clone able to survive at ~4x10⁴CFU/mL yield by day 5 (Figure 2E). For population F, the yields of 3 clones were below the limit of detection and 9 clones had measurable yields, ranging from 4×10³ CFU/ml to 1×10⁶ CFU/mL, by the end of the experiment on day 5, compared to parental average yields of ~9×10⁵ CFU/mL (Figure 2F). The clones taken from population G also grew poorly, with no growth detected throughout the experiment (Figure 2G). Finally, population H yielded no clones that exhibited yields above the detection limit, except for a single clone on day 1, compared to its parental strain that was able to grow in our experimental condition (Figure 2H).

Genomic sequencing revealed that all seven parental strains (one each from Groups 1, 2, 3, 4, and 8, and two from Group 7) are closely related to each other, with genomes ranging in size from 5,411,111 to 5,411,303 bp, and G + C content of 54.8% (Table 1). Each strain contains from 4,947–4,950 predicted protein coding genes, with a coding density of 89.9%. Though highly isogenic, there are notable differences between strains. First, several major chromosomal inversions exist between strains. Four parental strains from Groups 1, 2, 3, and 8 (referred to as Arrangement I), are fully syntenic with the same arrangement of genes (Figure 3). The parental strain from Group 4 and one of the strains from Group 7, henceforth 7–1, has a different arrangement from Arrangement I (referred to as Arrangement II) due to an inversion between two homologous copies of an IS91 family transposase gene, ISSod25, located at positions 846,036 and 3,511,426 (Figure 3). The other strain from Group 7, henceforth 7–2, has a third arrangement (referred to as Arrangement III), with an inversion between the 23S rRNA genes located at positions 556,910 and 4,214,392 compared to Arrangement I. Another difference is in the genomic sequences that primarily differ from one another through insertions of a repeated sequence in the intergenic regions. This causes the difference in genome sizes while the protein-coding regions of the genome are almost entirely identical. The most notable differences within the protein coding genes are differences in length and sequence of a hypothetical protein that is predicted to be a homolog of tctB, a tricarboxylate transporter (Rosa et al., 2018); a SNP in another hypothetical protein that is predicted to be a quinoprotein dehydrogenase-associated SoxYZ-like carrier, which is a carrier complex involved in sulfur oxidation; and a SNP in the gene lgrB, which codes for gramicidin synthase, a protein involved in the biosynthesis of a pentadecapeptide antibiotic (Table 1). Of these mutations, tctB had the highest variability, where the majority of the strains contain different SNPs in this gene. Finally, there are differences in the 23S rRNA sequences of these strains. While the sizes of the 23S rRNA genes are almost identical, parental strains from Group 3 and Group 7 (both 7–1, and 7–2), have two copies of their 23S rRNA genes that differ from the rest of the strains. This difference is found from position 1,464 bp to 1,510 bp, which is located in the V3 region (Helix 58) of the Escherichia coli 23S rRNA gene.

To further understand the possible mechanisms that enable Halomonas to scavenge scarce nutrients in unsupplemented minimal medium in the laboratory, and possibly the natural world, we sequenced the aforementioned 12 clones from each of the 8 evolved populations (A through H), giving a total of 96 independent clones. In total, there were 32 loci with unique mutations across the 96 clones (Tables 2, 3). Four of those loci had been mutated in more than one independently evolved population. It is important to note that, where the same locus has been mutated, the molecular basis of the mutations differs in each of the cultures, supporting an independent origin for each of these mutations. Each of the four genes with multiple mutations encode hypothetical proteins, and using blastx,² were identified as: (i) a putative tripartite tricarboxylate transporter (tctB) family protein, (ii) a DUF4214 domain-containing protein (DUF stands for Domain of Unassigned Function), (iii) a putative metallopeptidase, and (iv) a HAD-IB family hydrolase. Surprisingly, the DUF4214 domain-containing protein was mutated in 7 out of the 8 evolved populations (A, B, C, D, E, G, and H). Each of the mutations in this coding region were unique and include nonsynonymous mutations, frame-shift mutations, small and large indels, nonsense mutations, and mutations in putative regulatory regions upstream of the gene (Tables 2, 3). In addition, the 3 other protein coding genes were mutated in 2 different strains. As shown in Table 2, populations B and E had mutations in a gene that encodes a putative HAD-IB family hydrolase; in this case, the same insertion mutation occurred in the same location in both strains. Populations C and D had mutations in the gene that codes for a hypothetical tripartite tricarboxylate transporter. In both strains, the codons for amino acids 49 and 52 of the protein were mutated. While both strains have the identical mutation of A49A, a synonymous mutation of GCC to GCG, amino acid 52 in population D had the mutation I52N, while population C had the mutation D52E (Table 2). Lastly, populations F and G had mutations in the gene that encodes a hypothetical metallopeptidase. In addition to these mutations, all evolved populations contained unique mutations. Among these genes were those involved in the biosynthesis of sugars and amino acids, transcriptional regulation, transport of metals across membranes, purine metabolism, and the starvation and SOS stress responses. In particular, there were 7 unique mutations that were found in genes encoding known or putative transport functions, two of which are Zinc binding proteins. The detailed descriptions of all mutations identified are listed in Table 3.

One particular gene, encoding a DUF4214 domain, was mutated in all but one evolved population, suggesting that its function was under negative selection during adaptation to rich medium. To begin determining the unknown function of the DUF4214 domain-containing-protein (referred to as Halo4214 henceforth), we used Phyre2, SWISS-MODEL, and PredictProtein to identify potential functional regions. The predicted gene product consists of 1,038 amino acids (and is likely to contain multiple domains). Phyre2 identified 19 different protein alignment templates and mapped the hypothetical protein onto these templates with >90% confidence. Out of the 19 proteins identified, 6 indicated similarity with a hydrolase. Two templates matched this protein with the highest confidence (99.2%), and both encode a lipase: the extracellular lipase, lipA, from Serratia marcescens and a lipase from a Pseudomonas sp. organism. For the extracellular lipase, 32% coverage was aligned from residues 598–936 and, for the Pseudomonas lipase, 38% coverage was aligned from residues 565–966, significantly overlapping the enzyme from Serratia. Further, the PredictProtein algorithm associated the hypothetical protein to Gene Ontology terms related to S-layer surface proteins, extracellular protein regions, cell wall components, and calcium ion binding domain. The SWISS-MODEL software aligned 88 amino acids (S16-V104) to the S-layer protein from Caulobacter crescentus. From these results, we hypothesize that this large protein is likely to have multiple domains, where one domain anchors itself to the outer membrane of the cell, while another domain functions as an extracellular lipase.