Free-Living, Psychrotrophic Bacteria of the Genus Psychrobacter Are Descendants of Pathobionts

ABSTRACT Host-adapted microorganisms are generally assumed to have evolved from free-living, environmental microorganisms, as examples of the reverse process are rare. In the phylum Gammaproteobacteria, family Moraxellaceae, the genus Psychrobacter includes strains from a broad ecological distribution including animal bodies as well as sea ice and other nonhost environments. To elucidate the relationship between these ecological niches and Psychrobacter’s evolutionary history, we performed tandem genomic analyses with phenotyping of 85 Psychrobacter accessions. Phylogenomic analysis of the family Moraxellaceae reveals that basal members of the Psychrobacter clade are Moraxella spp., a group of often-pathogenic organisms. Psychrobacter exhibited two broad growth patterns in our phenotypic screen: one group that we called the “flexible ecotype” (FE) had the ability to grow between 4 and 37°C, and the other, which we called the “restricted ecotype” (RE), could grow between 4 and 25°C. The FE group includes phylogenetically basal strains, and FE strains exhibit increased transposon copy numbers, smaller genomes, and a higher likelihood to be bile salt resistant. The RE group contains only phylogenetically derived strains and has increased proportions of lipid metabolism and biofilm formation genes, functions that are adaptive to cold stress. In a 16S rRNA gene survey of polar bear fecal samples, we detect both FE and RE strains, but in in vivo colonizations of gnotobiotic mice, only FE strains persist. Our results indicate the ability to grow at 37°C, seemingly necessary for mammalian gut colonization, is an ancestral trait for Psychrobacter, which likely evolved from a pathobiont. IMPORTANCE Host-associated microbes are generally assumed to have evolved from free-living ones. The evolutionary transition of microbes in the opposite direction, from host associated toward free living, has been predicted based on phylogenetic data but not studied in depth. Here, we provide evidence that the genus Psychrobacter, particularly well known for inhabiting low-temperature, high-salt environments such as sea ice, permafrost soils, and frozen foodstuffs, has evolved from a mammalian-associated ancestor. We show that some Psychrobacter strains retain seemingly ancestral genomic and phenotypic traits that correspond with host association while others have diverged to psychrotrophic or psychrophilic lifestyles.

Specialization to host association can also be seen at higher taxonomic levels, indicating that whole lineages may have diverged once animal hosts were first successfully colonized. For instance, within the Bacteroidetes, the taxa that are mammalian gut associated are derived from phylogenetically basal clades that include free-living and invertebrate-associated taxa (4). These patterns of distribution imply that specialization to the warm animal host habitat is mostly incompatible with fitness in other environments. The specific diet of the host, the chemical environment of the gut, competition with an extremely dense surrounding microbial community, and direct selection by the host immune system all likely contribute to the restricted taxonomy of gut-associated microbes (5).
One line of evidence for the adaptive evolution of gut microbes is the emergence of specialized functions, such as the ability to bind and degrade host-specific nutrient sources like mucin (6), or resistance to bile acids and their salt conjugates (7). Additionally, many gut-associated commensal taxa lack significant environmental reservoirs, due to the divergence in selective pressures between guts and nongut environments (8). In contrast, many gastrointestinal pathogens such as Vibrio cholerae (9) or Yersinia enterocolitica (10) have high fitness within the gut and outside the gut, typically in soil or water. One phylogenetic analysis has indicated that the evolutionary transition from a free-living lifestyle to a host-associated (including pathogenic) lifestyle is more common than the opposite (11).
Species of Psychrobacter have been recovered through culture-based and sequencebased methods from a range of animal microbiomes, including the respiratory blow of marine mammals (12); marine mammal skin (13) and guts (14)(15)(16); the throats and guts of birds (17,18) and fish (19); and many nonhost environments such as seawater (20), sea ice (21), marine sediment (22), glacial ice (23), and permafrost soil (24). Some Psychrobacter species are capable of causing disease in mammalian hosts (25,26); however, Psychrobacter infections are very rare, and the factors leading to infection are unclear. Psychrobacter's wide range of environmental sources and adaptation to cold temperatures raise the question whether Psychrobacter is truly associated with mammalian hosts, or if it is an allochthonous member of mammalian microbiomes. A comparative genomics analysis of 26 Psychrobacter spp. and metadata gleaned from public sources revealed differences in cold adaptation of protein coding sequences between warm-host-associated strains and derived marine and terrestrial strains (27). Furthermore, warm-adapted strains were basal in the rpoB gene phylogeny, suggesting that Psychrobacter evolved from a mesophilic ancestor. These observations make Psychrobacter an interesting candidate to assess how phenotype maps onto phylogeny and source of isolation and to probe into the evolutionary history of a genus with a wide habitat range.
Here, we investigated the evolutionary history of the genus Psychrobacter, a group of closely related bacteria with a broad environmental distribution. We generated 85 genomes of Psychrobacter accessions, which we phenotyped for growth under different temperature and salt conditions. We used a collection of wild polar bear feces collected from across the Canadian Arctic and assessed the presence of Psychrobacter ecotypes as a function of bear diet. Finally, we conducted tests with a subset of strains for colonization of the mammal gut using germfree mice. Our phylogenomic results confirm a mesophilic ancestry for Psychrobacter and a common ancestor with the genus Moraxella. Our phenotyping revealed that overall, Psychrobacter tolerates a wide range of salinity, but growth at 37°C divided the accessions into two ecotypes: those that retained the ability to grow at warm temperatures and can colonize mammalian hosts (flexible ecotype [FE]) and those that lost the ability to grow at warmer temperatures and are instead restricted to colder temperatures (restricted ecotype [RE]). A genomic analysis of the two ecotypes showed genome reduction in the FE strains with high transposon copy numbers, and increased biofilm formation capability in RE strains, as well as higher proportions of lipid metabolism genes. We showed that Psychrobacter strains that are basal are FE but that FE strains are also phylogenetically interspersed with RE strains, indicating either readaptation to the animal host or retention of the basal traits. Although both FE and RE strains were detected in the feces of wild polar bears, of the subset of Psychrobacter strains tested, only FE strains could colonize the germfree mouse gut. Together, our results indicate the genus Psychrobacter evolved from a pathobiont ancestor, having largely lost its ability to associate with animals in its adaptation to psychrophilic, nonhost environments.

RESULTS
Psychrobacter forms a clade whose basal members are of the Moraxella genus. To explore the evolutionary history of Psychrobacter, we built a phylogeny based on 400 conserved marker genes using publicly available whole genomes derived from cultured isolates of 51 species from the Moraxellaceae family. We included 18 species each of Acinetobacter and Moraxella obtained from NCBI and 15 Psychrobacter genomes that we generated in this study (see Table S1 in the supplemental material). For all genomes, we incorporated phenotypic data collected from previously published type-strain research. The 51 species formed 3 distinct clades, each with robust bootstrap support. The Acinetobacter clade consists uniquely of Acinetobacter species (labeled A in Fig. 1A). The Acinetobacter clade is a sister taxon to the Moraxella (M) clade, consisting entirely of Moraxella species, and to the Psychrobacter (P) clade, which contains all of the Psychrobacter species, as well as 4 Moraxella species-M. boevrei, M. atlantae, M. osloensis, and M. lincolnii-that are basal.
Consistent with the topology of the phylogeny, a principal-coordinate analysis (PCoA) of gene presence/absence data shows that the greatest variation within the family (i.e., principal coordinate [PC] 1) is the separation between the A clade and the M and P clades, which are grouped (Fig. 1). Genes annotated with very diverse functions, falling under almost every Cluster of Orthologous Groups (COG) category, strongly contribute to the separation between clades according to envfit analysis ( Fig. 1B; Table S2). The P-clade Moraxella spp. fall between the P-clade Psychrobacter and the M-clade Moraxella when visualizing PC1 and -2 ( Fig. 1B) as well as PC2 and -3 (Fig. 1C).
Despite their close phylogenetic relationship and high similarity in gene presence/ absence, Psychrobacter and Moraxella have different genomic properties. When examined by genus rather than clade, Psychrobacter species have an average genome size of 3.12 6 0.27 Mb, while the average of Moraxella is 2.41 6 0.28 Mb (pairwise Wilcoxon rank sum test [WRS], P value = 8e207). Moraxella species have an average coding density of 86.4% 6 1.33%, which is significantly higher than the Psychrobacter species average of 82.7% 6 1.33% WRS, P value = 1e25). Despite being phylogenetically more related to Psychrobacter species than to the other Moraxella species, the P-clade Moraxella strains are significantly different from Psychrobacter species for both genome size and coding density (pairwise WRS, P values = 0.004) while not significantly different from the M-clade Moraxella spp. (pairwise WRS, P values = 0.9).
Psychrobacter ranges of growth temperatures differ from those of Moraxella. To examine the phenotypic behaviors of the Moraxellaceae family, we applied continuous trait mapping of the ranges of temperatures at which species from the Moraxellaceae family can grow (16,17,20,21,26, onto the previously generated phylogeny (Fig. 2). The Psychrobacter spp. included here exhibit a broad range of growth temperatures (0 to 38°C), but several strains, such as P. frigidicola and P. glacincola, are psychrophilic (restricted to growth below 20°C), which is a phenotype that is not seen elsewhere in the family. Psychrobacter spp. have lower minimum growth temperatures than the included Moraxella spp. from either the P or M clades (pairwise WRS, P value = 5e206), which have a narrow range of temperatures at which they can grow (between 22°C and 40°C). In contrast to the minimum growth temperatures of the family, there is little variation in the maximum growth temperatures, except for the notable exceptions of several Psychrobacter spp. that are entirely restricted to low temperatures.
As with their genomic properties, P-clade Moraxella strains have more similar phenotypes compared to M-clade Moraxella strains than to Psychrobacter strains. Review of type strain descriptions revealed that Psychrobacter strains are consistently urease positive, nitrate reducing, salt tolerant, and nonfastidious, whereas P-clade Moraxella strains are inconsistent with their urease-and nitrate-reducing phenotypes, are sensitive to high salt concentrations, and are often nutritionally fastidious-they have complex growth requirements (in particular, often blood or bile for growth) (16,17,26,32,35,43,44,59).
Few Psychrobacter strains can grow at 37°C. To explore Psychrobacter phenotypic diversity, we established a strain collection of 85 Psychrobacter accessions isolated from diverse locations and ecological sources (Table S3). To characterize Psychrobacter phenotypes under a number of different conditions, we assessed the strain collection for resistance to bile salts as well as ability to grow under 24 different conditions: two different media combined with four different salt concentrations and with three different temperatures (Materials and Methods).
We calculated growth probabilities, or the fraction of growth-positive conditions out of total conditions tested, for every strain and given variable of the growth curve The Moraxellaceae family phylogeny was constructed with 51 diverse Moraxellaceae genomes using the software PhyloPhlAn, which constructs a phylogeny using FastTree with 1,000 bootstraps, refined by RAxML under the PROTCATG model. Amino acid sequences from 400 marker genes were used in the alignment. Branches with bootstrap support greater than 70% are represented by filled circles. The scale bar represents the average amino acid substitutions per site. Pseudomonas aeruginosa was used as an outgroup. Clades are highlighted in colored blocks, and branches are colored by genus. Isolation source is depicted in a color strip, along with a heatmap of scaled notable genome characteristics that differ between the genera, with 0 representing the smallest value present and 1 the largest value (P. aeruginosa not included). GS, genome size, ranging between 1.8 Mb and 4.5 Mb. PCoA, PC1 values from a PCoA based on gene presence/absence data. CD, genome coding density, ranging from 80% to 89%. Growth temperature range data were collected from type strain publications. (B) PC1 and PC2 of a PCoA of a binary matrix of gene presence/absence for 51 species of Moraxellaceae, explaining 32% and 14% of the variation, respectively. Each genome is represented by one point, colored by genus. The P-clade Moraxella spp. are represented by yellow points with red outlines. (C) PC2 and PC3, explaining 14% and 5% of the variation, respectively. screen and compared them along with the bile resistance data across the phylogeny and by isolation source (Fig. 3). We generated a robust genus-level phylogeny for Psychrobacter using 400 conserved marker genes, with M. lincolnii as an outgroup. In agreement with single marker gene trees generated using rpoB sequences (27) and 16S rRNA gene sequences (69) as well as the P-clade structure of the Moraxellaceae family tree generated in this study, we observed a phylogenetically basal group of strains mostly isolated from mammals, and a phylogenetically derived group isolated from mixed sources. Across the entire phylogeny, closely related strains have similar growth probabilities (Pagel's l ranging from 0.78 to 0.97, all corrected P values , 1e23). We observed that most Psychrobacter strains are tolerant of a wide variety of temperatures between 4 and 25°C and of salt concentrations between 0 and 5%; more than 90% of all strains can grow under these conditions. However, only 54% of the tested accessions can grow at 10% added salt, and only 31% can grow at 37°C.
Since we were interested in the possibility of Psychrobacter interacting with warmbodied hosts, we divided the strains into two ecotypes: the "flexible ecotype" (FE) corresponds to strains that could grow at 37°C, and the "restricted ecotype" (RE) corresponds to strains that could not grow at 37°C. FE strains are psychrotrophic (mesophilic organisms with a low minimum growth temperature but an optimal growth temperature above 15°C), and RE strains are either psychrotrophs or true psychrophiles (unable to grow at temperatures higher than 20°C).
Notably, the basal clade of the Psychrobacter-only tree (Fig. 3) consists solely of FE strains, while the rest of the phylogeny is made up both FE and RE strains. Furthermore, the basal FE strains have higher growth probabilities at 37°C than do other FE strains (Pagel's l = 0.89, P value = 2e25). Frequencies of RE and FE strains vary significantly across sources of isolation: the FE group is significantly enriched in strains derived from mammalian sources, and the RE group is significantly enriched in  The Psychrobacter phylogeny was constructed using FastTree with 1,000 bootstraps, refined by RAxML under the PROTCATG model. Amino acid sequences from 400 marker genes were used in the alignment. Branches with bootstrap support greater than 70% are represented by filled circles. The scale bar represents the average amino acid substitutions per site. M. lincolnii was used as an outgroup. Type strain isolate names are indicated in bold and italicized type. Strains indicated with * next to their name exhibited growth defects in liquid media and were tested on solid agar media instead. Strains indicated with ** exhibited growth defects on solid and liquid media and were tested on solid media supplemented with 0.1% Tween 80. Isolation source is depicted in column 1 as a color strip. Columns 2 to 10 represent the growth probabilities of each strain for each condition; medium complexity is represented in yellow, salt concentration is represented in blue, and temperature is represented in red. Type strain data support our temperature data except where indicated-colored triangles show conditions under which we expected growth but did not observe it, while white triangles represent conditions under which we observed growth that we did not expect. Bile salt resistance is shown in column 11, with black triangles indicating strains where we expected resistance but observed sensitivity, and green triangles representing the opposite. The "ecotype" is shown in column 12 in a color strip and in the color of the branches. strains derived from other hosts, food, and marine and terrestrial sources (x 2 test, 83% of P values adjusted for group size , 0.05). Nonetheless, both FE and RE contain strains from other environments, including mammalian-derived strains within the RE group.
FE strains show higher growth probabilities in complex media than in defined media, while RE strains showed little difference between the two (WRS, Wilcoxon test statistic [W] = 1.13e3, P value = 0.0005, phylogenetic P value = 0.002). FE strains also have higher growth probabilities at low to medium salt concentrations (WRS, all P values , 0.05, phylogenetic P values , 0.4), though there is no difference between FE and RE strains at 10% salt (WRS, W = 811.5, P value = 0.7, phylogenetic P value , 0.4). Finally, FE strains are more likely to be resistant to bile salts than RE strains (x 2 test, 93% of P values adjusted for group size , 0.05).
FE and RE Psychrobacter spp. have functional and structural genomic differences. When exploring gene presence/absence via PCoA, accessions from the basal FE-only subclade cluster closely together, indicating similar gene content. In agreement with the phylogeny, these basal FE accessions are separated from the other accessions on PC1, while the derived FE and RE accessions are more scattered, indicating more diverse gene content (Fig. 4A). According to an envfit analysis, the separation is most strongly driven by genes from the COG categories T, signal transduction (r 2 = 0.65, P value = 0.001); U, trafficking and secretion (r 2 = 0.63, P value = 0.001); P, inorganic ion transport and metabolism (r 2 = 0.56, P value = 0.001), and X, unassigned or no homologs in the COG database (r 2 = 0.51, P value = 0.001) ( Table S2).
The average Psychrobacter genome carries 2,763 genes, 1,598 of which are core (present in 90 to 100% of strains), 1,005 are shell (present in 2 or more strains but fewer than 90%), and 70 are cloud (present in only one strain). The average FE genome has significantly fewer core (WRS, W = 531.5, P value = 0.02, phylogenetic P value = 0.3) and shell (WRS, W = 462, P value = 0.003, phylogenetic P value = 0.4) genes but significantly more cloud genes (WRS, W = 1014, P value = 0.02, phylogenetic P value = 0.2) than the average RE genome (Fig. 4B).
Increased transposon activity can lead to interruption and decay of functional protein-encoding genes, leading to an increase in pseudogenes (70). Given the higher number of transposons in FE strains, we next examined the number of predicted pseudogenes between the ecotypes. FE genomes are predicted to have a higher proportion of pseudogenes than RE strains (WRS, W = 997, P value = 0.03, phylogenetic P value = 0.6) (Fig. 4C). Finally, we compared the average numbers of genes per genome between the ecotypes, as bacterial genomes are known to strongly select against the accumulation of pseudogenes resulting in gene deletions (71), and found that FE strains have significantly fewer genes than RE strains (WRS, W = 547, P value = 0.05, phylogenetic P value = 0.6).
We looked for specific gene clusters differentiating the FE and RE ecotypes while controlling for population structure by performing a microbial pangenome-wide association analysis with the R package treeWAS (72). This returned four gene clusters that significantly correlated with ecotype, each of which is enriched in RE strains compared to FE strains ( Table 1). The gene with the most significant enrichment is annotated as a transcriptional regulator in the TetR/AcrR family with the KEGG KOs of K16137 and K19335, which is predicted to be either nemR or bdcR. Other significant results include lpdA, dihydrolipoamine dehydrogenase related to amino acid metabolism (KEGG KO K00382/383), a hypothetical protein with the IG domain, and a hypothetical protein with no homologs in eggNOG database and no predicted annotation. Each of these genes has paralogs that are not significantly enriched in one ecotype over the other.
Polar bear feces collected from the Arctic ice have a high abundance of Psychrobacter. Psychrobacter spp. have been reported previously in the skin, respiratory, and gut microbiomes of several marine mammals, including whales, porpoises, seals, and sea lions; Psychrobacter presence may stem from the surrounding seawater. Polar bears, on the other hand, are marine mammals that do not spend as much time swimming. We surveyed the gut microbial diversity of 86 polar bear fecal samples, 76 wild and 10 captive, by 16S rRNA gene sequencing. Sequence variant clustering of the 16S rRNA gene sequences reveals that Psychrobacter is detectable in 88% of the samples (Fig. 5A). The large majority of Psychrobacter sequences were assigned to unclassified Psychrobacter spp., which are prevalent in 87% of samples with a mean abundance of 22%. We also detected RE strain P. immobilis in 49% of samples with a mean abundance of 3%, and FE strain P. pulmonis in 8% of samples with a mean abundance of 0.5%. Polar bear diet significantly impacts the abundance of Psychrobacter spp. (Kruskal-Wallis x 2 = 13.5, df = 3, P value = 0.004); we found that polar bears feeding on mammalian prey, including seals and reindeer, had significantly higher abundances of unclassified Psychrobacter spp. than polar bears feeding on avian prey or mixed diets (pairwise Wilcoxon rank sum test, adjusted P values , 0.05) (Fig. 5B). However, diet data are confounded with other metadata, as location (Kruskal-Wallis x 2 = 8.6, df = 4, P value = 0.0009) and year (Kruskal-Wallis x 2 = 16.6, df = 5, P value = 0.005) of sample collection also significantly impact unclassified Psychrobacter abundance. Captive status does not significantly impact abundance, but there is a trend of wild bear samples having higher unclassified Psychrobacter abundance than samples from captive bears (WRS, W = 222, P value = 0.08).
To elucidate the effect that host dietary nutrition may have on Psychrobacter growth, we tested the maximum change in absorbance for 190 different carbon sources by a subset of Psychrobacter accessions including 9 FE and 10 RE strains (Fig. 5C). All Psychrobacter spp. reached significantly higher optical densities at 600 nm (OD 600 ) growing on amino acid carbon sources compared to carbohydrates or sugar alcohols (pairwise WRS, adjusted P values , 0.05). Psychrobacter spp. reached the highest OD 600 growing on fatty acids, surfactants, and peptides, but there was no significant difference between FE and RE strains' changes in absorbance (WRS, W = 1.55e6, P value = 0.08), though there is a slight trend of RE strains having overall higher changes in absorbance.
Of the four FE strains tested, three were able to persist in the mice, while the FE strain P. faecalis PBFP-1 and none of the RE strains were detectable after 3 weeks. Phylogenetic relatedness does not correlate with ability to colonize, as FE strain P. Gene cluster refers to a group of protein-coding gene sequences with distant sequence homology which may be either orthologs or paralogs, while paralog refers to subsets within the gene cluster identified as exact gene matches. b ***, the paralogs identified as significantly associated with ecotype by the pangenome-wide association software treeWAS.
pacificensis was able to persist, while closely related RE strains P. namhaensis and P. okhotskensis were not. Phylogenetic placement does correlate with colonization density, however, as the two most basal strains tested, P. lutiphocae and P. ciconiae, colonize at significantly higher densities than the most derived strain that was successful, P. pacificensis (pairwise WRS, both adjusted P values = 0.0007).

DISCUSSION
The phylogenomic and phenotypic characterization of the Psychrobacter genus indicates a common ancestor with Moraxella, all of which are restricted to growth at higher temperatures. Furthermore, the most basal members of the Psychrobacter clade are Moraxella species and species of Psychrobacter that can grow at 37°C, unlike most of the derived Psychrobacter species. Our extensive phenotyping indicated that members of the Psychrobacter genus grow at a wide range of salinities and temperatures, but distinguishing strains by their ability to grow at 37°C results in two groups of strains with differences in response to rich versus defined media, bile salt resistance, and genomic characteristics. Our analysis of a collection of wild polar bear species  Fig. 3. Branches with bootstrap support higher than 70% are indicated with a filled circle. Means were compared using the Wilcoxon rank sum test. *, P value less than 0.05; **, P , 0.005; ***, P , 0.0005.
indicates both RE and FE strains are present; however, tests in germfree mice support the notion that only FE strains can colonize the mammal gut, whereas RE strains may be allochthonous members. Together with previous reports, this work indicates the genus Psychrobacter is a lineage descended from pathobionts, some of which have evolved to inhabit the colder environments of their warm-bodied hosts.
Our results corroborate those of Bakermans, who used the isolation source of Psychrobacter as a proxy for temperature adaptation to conclude the genus has a mesophilic ancestor (27). By assessing growth under the same 24 conditions for 85 strains, we remove any ambiguity that can stem from whether an isolate can indeed grow at the temperature of its source of isolation. This is particularly important given that several strains isolated from mammals proved to be RE and that many of the FE strains came from seawater or other relatively cold environments. We also expanded upon the results of Bakermans by including phylogenetically aware statistics. In any population, identifying traits that are shared due to ecological convergence versus those that are shared due to common ancestry is a well-recognized problem (73); this problem is particularly striking in the genus Psychrobacter, where not only is there phenotypic divergence between the more phylogenetically basal strains and the derived strains, but as evidenced by the gene presence/absence PCoA, there is a large difference in gene content based on phylogeny.
We found that a transcriptional repressor, either NemR or BdcR, is enriched in RE strains' genomes compared to FE strains. NemR plays a role in electrophile sensing (74), while BdcR plays a role in cyclic-di-GMP sensing (75). Both relate to increased biofilm formation when expressed. Similarly, we found a predicted Ig-domain protein, which is thought to be an outer membrane protein with a role in adhesion (76), enriched in RE strains. Adhesion and biofilm formation can increase during cold stress (77), indicating that it may be adaptive for psychrophiles or psychrotrophs to have increased biofilm formation capability.
In addition to specific genes being enriched in RE strains, we found broader functional differences between the ecotypes. FE strains had higher proportions of COG category "L" (replication) and "P" (inorganic transport and metabolism) genes. The higher proportion of "P" category genes could point to FE strains' association with hosts, as free iron availability is rather low within host bodies (78). The higher "L" proportion seems to be related to higher transposon copy numbers in FE strains, which could point to population bottlenecks relating to host association (79). RE strains, on the other hand, showed increased proportions of COG category "W" (extracellular structures), "S" (function unknown), and "I" (lipid metabolism) genes. The increase in the proportion of "W" genes is unlikely to have a large impact on phenotypic differences between the ecotypes, since there are very few "W" genes in genomes of either RE or FE strains. It is difficult to comment on the increased proportion of "S" genes, since their function is unknown or poorly understood. The increased proportion of "I" category genes, relating to lipid transport and metabolism, could point to increased flexibility in membrane lipids for RE strains, which is a well-characterized adaptation to cold temperatures (80).
The three of four FE strains tested were able to colonize germfree mice, whereas the RE strains could not, indicating that growth at 37°C may be necessary (although not sufficient) to colonize mammals. Opportunistic infections in mammals caused by Psychrobacter strains are limited to P. sanguinis, P. phenylpyruvicus, P. faecalis, and P. pulmonis (25), which are all FE strains but do not all belong to the basal clade. Our results suggest that the FE strains maintain an ancestral ability to grow at mammalian body temperatures and colonize mammalian host bodies, while RE strains have adapted toward psychrophilic or psychrotrophic lifestyles.
Psychrobacter's sister taxon Moraxella in particular is commonly isolated from host mucosal tissues and exhibits the reduced genome size and nutritional fastidiousness common to many host-dependent organisms. Moraxella contains species that are frequently associated with human respiratory infections, primarily M. catarrhalis (81), as well as livestock conjunctivitis, for example, M. bovis or M. equi (36). Since they are commonly found in healthy individuals and can cause disease in healthy individuals (82), Moraxella bacteria are best categorized as pathobionts and not dedicated or opportunistic pathogens. Several species of Moraxella appear basally in the P clade of the Moraxellaceae family-level phylogeny, suggesting that Psychrobacter evolved from a "Moraxella-like" ancestor. This is supported by the fact that both phylogenetically basal and derived Psychrobacter strains carry genes related to virulence functions and that many of the basal Psychrobacter strains exhibit growth defects in liquid culture, similar to the fastidiousness of Moraxella.
Despite clear phenotypic differentiation, Psychrobacter and Moraxella have similar genomic content, although Psychrobacter genomes are larger. A psychrophile emerging from an apparently mesophilic background through widespread horizontal gene transfer has been suggested before in the genus Psychroflexus (83), though the study was limited to comparing only two genomes. In fact, it has been suggested before that this is Psychrobacter's evolutionary trajectory (27), and although many Psychrobacter trees are constructed using a Moraxella outgroup, Psychrobacter's potential pathobiont origin has not been explored in detail. Horizontal gene transfer would explain Psychrobacter's larger genome size compared to Moraxella despite lower coding density, as many newly acquired horizontally transferred genes are expected to be inactivated and pruned by the recipient genome (84,85).
Our results indicate that the history of the genus Psychrobacter is that of an ancestral pathobiont or pathogen, some of the descendants of which broadened their ecological distribution, resulting in an attenuation of pathogenicity. The emergence of a psychrotroph-a remarkable generalist-from a background of a more specialized pathobiont or pathogen showcases the remarkable adaptability of bacteria, and particularly Proteobacteria, to their environments.
We analyzed the Moraxellaceae pangenome using the PanX pipeline (91) (v1.6.0, core gene cutoff = 0.9). The input genomes used by PanX were initially annotated using Prokka. After their assignment into orthologous clusters using MCL (v14.137), we reannotated gene clusters using eggNOG-Mapper (92) (v1.0.3). We explored genome content by calculating a distance matrix using the Jaccard metric through the R package ecodist (93) (v0.3.0) from a binary gene presence-absence table, followed by dimensional reduction of that distance matrix through principal-coordinate decomposition (PCoA) with the cmdscale function from the R package stats. We investigated variables contributing to the separation of the PCoA using the envfit function from the R package vegan (94) (v2.5-6); we tested whether genes significantly contributing to separation were "core" (genes present in 90% to 100% of strains), "shell" (genes present in greater than two strains, but in fewer than 90%), or "cloud" (genes present in only one strain) and if general gene function-summarized by Cluster of Orthologous Groups (COG) category (95)-contributed to the separation. We collected growth temperature range data from type strain publications (16,17,20,21,26,. We used the R package phytools (96) (v0.7-47) to map the temperature ranges onto the phylogeny.
Psychrobacter strains. We obtained 92 isolates of Psychrobacter from strain catalogues for phenotypic and genotypic characterization. These represent 38 validly published species of Psychrobacter as well as unclassified strains, all isolated from a wide variety of geographical locations and diverse environmental and host samples. All strains were purchased and maintained in compliance with the Nagoya Protocol on Access to Genetic Resources and the Fair and Equitable Sharing of Benefits Arising from their Utilization to the Convention on Biological Diversity. For a full list of accessions and their catalogue, isolation, and cultivation information, see Table S3. Unless mentioned, we grew accessions as recommended by the strain catalogue (medium and temperature) from which they were purchased.

Psychrotrophic Bacteria Descended from Pathobionts
March/April 2021 Volume 6 Issue 2 e00258-21 msystems.asm.org 13 P. fozii, P. fulvigenes, P. glacincola ACAM483, P. luti, P. lutiphocae, P. maritimus Pi2-25, P. okhotskensis MD17, P. pacificensis, P. piscatorii, and P. urativorans ACAM534) to evaluate for their ability to grow on 190 different substrates as their sole carbon source. We utilized Biolog plates PM1 and PM2 (Biolog Incorporated, Hayward, CA, USA) following a slightly modified protocol. Briefly, we streaked out strains on agar plates of their preferred medium incubated at their preferred temperature (Table S3). We scraped cells from these plates and resuspended them at a final optical density at 600 nm (OD 600 ) of 0.07 in the Biolog inoculation fluid (PM IF-0a GN/GP, 1.2Â) diluted to 1Â with sterile water. We inoculated 100 ml of cell suspension into each well of the PM1 and PM2 plates, mixing well to resuspend the carbon sources. All strains were grown under aerobic conditions at their preferred growth temperature (Table S3). Growth was monitored by measuring OD 600 for 14 days. We calculated total change in absorbance by subtracting the blank wells from the substrate wells and then taking the difference between the absorbances at t = 14 days and t = 0 (inoculation time point). We then averaged the total change in absorbance for all strains for every carbon source and assigned each carbon source to a "family" of compounds.
All strains reached a significantly higher max OD 600 in amino acid substrates than in other carbon sources. L-Glutamate was chosen as the carbon source for the defined medium described in our phenotypic screens below, as none of the strains tested failed to grow using it. Some of the other amino acids allowed strains to grow to a larger change in OD 600 but failed to allow all strains to grow.
Psychrobacter phenotypic screen. For the 85 Psychrobacter accessions that passed genome quality control (QC) (see Table S4), we tested their ability to grow under 24 different conditions, a growth condition being a combination of a medium (complex or defined), salt concentration (0, 2.5, 5, or 10% NaCl) and incubation temperature (4, 25, or 37°C).
We used lysogeny broth (LB) as a complex medium with high nutrient availability and a variation of M9 minimal medium (MM) as a defined medium with lower nutrient availability. We prepared LB following the manufacturer's instructions except by various concentrations of sodium chloride (NaCl) (see below) and autoclaving it at 121°C for 20 min to ensure sterility. We prepared MM by adding components in the following final concentrations, followed by filter sterilization: 33.7 mM Na 2 HPO 4 , 22.0 mM KH 2 PO 4 , 9.35 mM NH 4 Cl, 0.4% L-glutamic acid, 1 mM MgSO 4 , 0.3 mM CaCl 2 , 1Â ATCC trace vitamins solution, and 1Â ATCC trace minerals solution, pH 7 (adjusted with KOH). We added NaCl to each "base medium" to 1 of the 4 following concentrations of NaCl: low, with 0%; medium, with 2.5%; medium-high, with 5%; and high, with 10%. We grew cultures at 4°C, representing cold environments; 25°C, representing mesophilic environments; and 37°C, representing mammalian host body temperature.
We randomly assigned Psychrobacter accessions to blocks of 10 strains to be tested simultaneously (several accessions were included in multiple blocks; see Table S2). We first grew strains to saturation, washed them, diluted them to an optical density at 600 nm (OD 600 ) of 0.3 in sterile phosphate-buffered saline (PBS), and inoculated them in 100 ml of medium with a final ratio of 1:1,000. We used 96-well plates, with 10 inocula in 5 replicates and 10 uninoculated medium wells per plate; all periphery wells were filled with water to reduce edge effects. Plates were incubated at each temperature to reflect all 24 growth conditions. We measured the OD 600 (Spark plate reader; Tecan, Zürich, Switzerland) every 8 h during the first week of incubation and then every 24 h and let cultures grow until stationary phase (3 to 12 weeks).
We used the same culture dilutions from the growth assays as inocula for bile resistance assays. We spotted 10 ml of diluted culture onto both tryptic soy agar (TSA) and TSA amended with 0.5% (wt/vol) bile salt (Difco bile salts no. 3 from BD Biosciences, San Jose, CA, USA) and incubated the plates at the strains' preferred temperatures (Table S2). We incubated the plates for 2 weeks and then scored strains for growth on bile compared to the no-bile control as either "sensitive" if no growth or "resistant" for growth similar to the control. For several strains, no growth was observed on the control plate, and these were marked as "ND" or "no data." P. phenylpyruvicus and P. sanguinis exhibited the inability to grow in liquid culture under the conditions tested. Hence, we first streaked 5 strains of these 2 species on Columbian blood agar which we incubated for 3 days at 37°C, and then we washed and diluted them as for the liquid cultures. We spotted dilutions onto LB or MM medium with 3% agar, supplemented with the tested salt concentrations as above, and incubated them at 4°C, 25°C, and 37°C. P. sanguinis strain 1501 was unable to grow on any of the base media, so we repeated the same process with medium supplemented with 0.1% Tween 80. We scored growth for all plates after 2 weeks.
For 2 accessions, we could not confirm the purity of the cultures used in the phenotypic screen and subsequently removed them from analysis. They were removed from subsequent genomic analysis as well.
Growth probabilities. We scored each replicate as either "growth positive," meaning the accession grew, or else "growth negative," if the replicate never reached a maximum OD 600 of 0.15 over the course of the experiment. For each strain and condition (medium, salt, and temperature), we calculated a growth probability, which corresponds to the median value of growth positivity/negativity of all replicates for that condition.
For the type strains included in the phenotypic screen, we cross-checked growth probabilities at 4°C and 37°C with the published type strain descriptions where the original publications were clear about conditions tested. For the majority of strains, the type strain data and our data are in agreement; however, there are discrepancies between the type strain publications and our own data for the following strains: P. ciconiae, P. fozii, P. marincola, P. namhaensis, and P. submarinus. In some cases, we observed growth under temperatures where growth was previously unobserved; P. ciconiae and P. marincola have not before been reported to grow at 4°C, and P. fozii has not been reported to grow at 37°C. In other cases, we observed no growth in temperatures that have previously been reported to support growth; P. submarinus is expected to grow at 4°C and P. namhaensis is expected to grow at 37°C. The conditions tested in the type strain publications often differ from the conditions used in our phenotypic screen.
Genome sequencing, assembly, and annotation. We extracted genomic DNA from cultures grown in their preferred conditions using the Gentra Puregene tissue kit (Qiagen, Valencia, CA, USA). We initially sequenced samples using the MiSeq 2 Â 250-bp and the HiSeq 2 Â 150-bp paired-end read technology (Illumina, San Diego, CA, USA) as previously described (83). We constructed libraries using the Nextera DNA sample preparation kit (Illumina) with modifications: we sheared 1 ng of DNA with inhouse-generated Tn5 transposase and then amplified and barcoded it with custom primers for 7 to 14 cycles. We pooled samples and size selected using magnetic beads for MiSeq libraries or BluePippin (Sage Science, Beverly, MA, USA) for HiSeq libraries. After dilution to 4 nM for MiSeq and 2.5 nM for HiSeq, we stored libraries at 220°C until sequencing.
We followed the quality control cutoffs suggested by CheckM and removed two genomes for having contamination higher than 5% and one for having completion less than 90%. We removed two additional genomes as the taxonomic classification was outside the Psychrobacter genus. These accessions were removed from genomic and phenotypic analysis. All further analyses use the 85 accessions which passed all quality control (QC) measures. For a full description of sequencing for each accession, see Table S4.
We annotated genomes with Prokka and eggNOG-Mapper. A phylogeny of the accessions was generated using PhyloPhlAn, with Moraxella lincolnii as an outgroup. Again, the phylogeny was visualized and annotated using iTOL. We used PanX to analyze the Psychrobacter pangenome and R to explore gene presence-absence data as described above with the Moraxellaceae family. Pseudogenes were predicted using the DFAST core workflow (100). We used treeWAS (72) to perform a pangenome-wide association study relating data collected in the phenotypic screen to genomic data.
Microbiome diversity of polar bear feces. We collected 86 polar bear fecal samples from several regions in Canada, including samples from 10 captive bears, fed various diets or fasted, and 76 samples from an unknown number of wild bears, feeding on unknown diets. The feces from the captive bears were forwarded from institutions within Canada and did not require permitting for their passage to Queen's University. The captive samples comprised 5 fecal samples from a single bear sequentially fed various diets of Arctic char, harp seal, and a "zoo diet" at the Polar Bear Habitat in Cochrane during 2010; 2 samples from each of 2 bears held at the Metro Toronto Zoo, fed a consistent "zoo diet" during 2010; and a single sample from a bear held at the Churchill Polar Bear Holding Facility in Churchill during 2010, where the bears are given only water until release. We collected all wild bear feces from M'Clintock Channel and Hudson Strait in Nunavut in accordance with permits prior to shipping to Queen's University We confirmed that samples collected from wild bears originated from bears by sequencing a cytochrome b gene fragment using metabarcoding approaches based on Ion Torrent (Ion Torrent Systems Inc., Gilford, NH, USA) and 454 pyrosequencing (454 Life Sciences, Branford, CT, USA) next-generation sequencing technologies (101). We also used this method to analyze what prey animals the polar bears had been feeding on at the time of sample deposition. We visually inspected the samples for confirmation on the dietary analysis. We extracted DNA from the fecal samples using the DNeasy blood and tissue kit (Qiagen) and characterized the gut bacterial community by amplification and sequencing of the V4 region of the 16S rRNA gene as described previously (102). We used QIIME (103) (v2, DADA2 for quality control, F read trimmed to 200 bp, R read trimmed to 110 bp, samples rarefied to 20,000 sequences) for sequence processing and Silva (138 SSURef NR99 515F/806R) to assign taxonomy.
Isolation of Psychrobacter sp. from polar bear feces. Two wild polar bear fecal samples were combined and diluted to 1 mg/ml in PBS. We plated the solution onto LB agar plus 6% NaCl and incubated it at 14°C for 10 days. A single colony grew and was identified as Psychrobacter faecalis by colony PCR and Sanger sequencing of the full-length 16S rRNA gene (described below). This isolate is designated P. faecalis PBFP-1 and is included in the phenotypic screen described above and the genomic analysis.
Gnotobiotic mouse colonizations. We selected 8 accessions of Psychrobacter-P. cibarius JG-220, P. ciconiae, P. faecalis PBFP-1, P. immobilis S3, P. lutiphocae, P. okhotskensis MD17, P. namhaensis, and P. pacificensis-as inocula for gnotobiotic mouse studies based on their derivation sources and phylogenetic breadth. We grew each accession under its preferred conditions (Table S3) to saturation, spun it at 10°C at 2,500 rpm for 25 min, washed it with sterile PBS, resuspended it in 15% (vol/vol) glycerol in PBS, and flash froze it in liquid N 2 . Inoculum samples were stored at 280°C until the mouse experiments were performed by animal caretakers as follows. For P. ciconiae, P. faecalis, P. namhaensis, and P. pacificensis, experiments were performed by Taconic Biosciences (Rensselaer, NY, USA) staff, while for P. cibarius, P. immobilis, and P. okhotskensis, experiments were performed by Max Planck Institute for Developmental Biology (MPIDB) animal caretakers. P. lutiphocae was included in both Taconic and MPIDB experiments.
Five-to 6-week-old germfree male C57BL/6J mice were orally inoculated with approximately 10 7 CFU (n = 4 per Psychrobacter accession, n = 8 for P. lutiphocae). Mice were cohoused (Taconic) or singlehoused (MPIDB) in sterile cages (IsoCage P; Tecniplast) and provided autoclaved water and sterile chow (NIH31M) ad libitum. Three weeks postcolonization, mice were sacrificed via CO 2 asphyxiation (Taconic) or CO 2 asphyxiation followed by cervical dislocation (MPIDB), after which cecal contents were immediately collected, flash frozen, and stored at 280°C prior to use. The Taconic experiments were performed in compliance with Taconic's IACUC, and the MPIDB experiments were approved by and performed in accordance with the local animal welfare authority's legal requirements.
To determine the bacterial colonization density in the mouse gastrointestinal tract, we serially diluted 2 aliquots per mouse of cecal material (50 mg each) and incubated them on plates under their preferred conditions (Table S3) for 3 to 5 days. If no colonies were observed, we spread an inoculating loop of the undiluted aliquot onto brain heart infusion agar and incubated it at 37°C for 2 weeks. For samples where we again observed no colonies, we categorized these Psychrobacter accessions as "nonpersistent." For the samples that did show colony growth, we confirmed the identity of the colonies as Psychrobacter using Sanger sequencing of the full-length 16S rRNA gene as described below.
16S rRNA gene Sanger sequencing. We prepared "colony" PCR mixtures following the protocol for Phusion high-fidelity polymerase (New England Biolabs, Ipswich, MA, USA), swirling a colony of interest in the reaction mixture as a substitute for the DNA template, and using the 27F and 1391R universal full-length 16S rRNA gene primers (104) for amplification. The reaction mixtures were incubated on the Mastercycler Pro S thermocycler (Eppendorf, Hamburg, Germany) following the Phusion protocol, using a touchdown program for the annealing temperature (dropping the annealing temperature at a rate of 1°C per cycle from 70°C to 55°C, then annealing at 55°C for 15 cycles). We cleaned the products using the DNA Clean & Concentrator-25 kit (Zymo Research, Irvine, CA, USA) and checked the concentration using the DS 111 spectrophotometer (DeNovix, Wilmington, DE, USA). We used the cleaned DNA products as the templates for the Sanger sequencing reaction, following the ABI Prism BigDye Terminator cycle sequencing kit (ThermoFisher Scientific, Waltham, MA, USA) protocol with the 27F primer. The sequencing was performed using the 3730xl DNA analyzer (ThermoFisher). Upon sequencing completion, we performed a BLAST search (105) of the sample sequences against the NCBI nonredundant nucleotide database. We examined the top 10 hits to confirm the sample identity.
Statistical analysis. We performed all data processing and statistical analysis using R (106) (v3.6.2) or Python (v3.6.10). We compared means between groups using a Kruskal-Wallis test followed by a pairwise Wilcoxon rank sum when more than two groups were compared. We tested differences in frequencies between groups with more than 10 observations with a x 2 test, repeated 100 times with down-sampling in order to correct for sampling sizes between groups. We measured phylogenetic signal using a log-likelihood ratio test on Pagel's l (Pagel's l fitted using the phylosig function from the phytools R package [96] [v0.  and the null hypothesis being l = 0). When applicable, we tested for the confounding of phylogeny with our groups of interest using the aov.phylo function from the R package geiger (107) (v2.0.7). We adjusted all P values for multiple comparisons with the Benjamini-Hochberg (BH) correction method.
Data availability. Raw sequences for the Psychrobacter genome sequencing and polar bear feces 16S rRNA gene sequencing, as well as assembled Psychrobacter genomes, are available in the European Nucleotide Archive under the accession no. PRJEB40380. Annotated Psychrobacter genomes are available at ftp://ftp.tue.mpg.de/pub/ebio/dwelter. Raw data, R notebooks, and Python scripts for the analyses are available at https://github.com/dkwelter/Welter_et_al_2020.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.