Background

First described over a century ago, marine fungi have received increasing recognition for their roles in ocean biogeochemical cycles and microbial food webs [1,2,3,4]. While environmental DNA sequencing surveys have uncovered many previously unrecognized lineages of marine fungi [5, 6], understanding of the functional roles of marine fungi has been partially limited by the small number of genomic analyses. Efforts such as the 1000 Fungal Genomes Project have primarily focused on fungi isolated from terrestrial environments [7]. This bias in research efforts partially stems from the view that the relative low substrate availability in seawater, especially when compared to soil and other types of terrestrial systems, limits the diversity and abundance of marine fungi.

Nutrient scarcity and the intensity of competition in oligotrophic environments is proposed to impact microbial genome size and composition. This theory, known as the streamlining theory, states that microbes recovered from environments of low nutrient availability will tend to have reduced genomes with a low base-pair count, a low intergenic-to-coding DNA ratio, few paralogs or pseudogenes, and a highly conserved set of core metabolisms [8]. Selection in oligotrophic environments is thought to favor efficiency of transport and metabolism as well as a low cost of transcription and replication [8]. To minimize the cost of protein synthesis and increase efficiency, streamlined marine microorganisms are expected to primarily reduce genes for transcriptional regulations [9]. It is supposed that carbon, nitrogen, iron, and sulfur limitations all lead to a reduction in nitrogen-rich protein expression and transition to less costly amino acid synthesis, including a preferential reduction in N-containing side chains and ribosomal proteins [10, 11]. While the streamlining theory has been supported by numerous prokaryotic examples, it is unclear whether marine fungi have made similar adaptations as their prokaryotic counterparts.

We analyzed the genome of the marine yeast Rhodotorula sphaerocarpa ETNP2018 isolated from the water column of the eastern tropical North Pacific (ETNP) ocean. A recent metabarcoding survey showed the genus Rhodotorula is one of the most prevalent and abundant fungal taxa in the ETNP oxygen minimum zone (OMZ). The Rhodotorula strain we analyzed was isolated from the same research expedition as the metabarcoding survey [12]. Rhodotorula are basidiomycetous yeasts that are sometimes referred to as “red yeasts” due to their production of β-carotene, responsible for their red, pink, or orange pigmentation [13]. Carotenogenic yeasts are utilized by the biotechnological industry to produce β-carotene as well as other isoprenoid products (γ-carotene, torulene, and torularhodin) [14]. The genus is also considered oleaginous, capable of synthesizing lipids up to one third of their dry weight. They are known to inhabit a wide array of environments, both eutrophic and oligotrophic, including human and plant hosts, benthic marine sediments, Antarctic permafrost, eutrophic freshwater lakes, ocean and river waters, soil, wood pulp, and the International Space Station [15,16,17,18,19,20]. The species R. sphaerocarpa has been isolated from the marine environments such as the Marguerite Bay in the Antarctic Ocean and the coastal waters of Thailand [21, 22]. In this study, we used comparative genomics of the open ocean isolate R. sphaerocarpa ETNP2018 and fourteen other Rhodotorula strains from diverse environments to test the genome streamlining theory for marine fungi and further searched for evidence of adaptation of marine fungi to the oligotrophic ocean.

Methods

Sampling and isolate information

The R. sphaerocarpa ETNP2018 strain was isolated from seawater in the eastern tropical North Pacific (ETNP) oxygen minimum zone (OMZ) [12]. Seawater from 50 m depth at Station 1 (10°N 113°W) was collected onboard R/V Sally Ride using 30-L Niskin bottles and inoculated into aerobic artificial seawater [23] medium supplemented with cellobiose (2 g L−1) and antibiotics (200 mg L−1 penicillin and streptomycin). The initial batch of enrichment cultures that showed growth were inoculated on agar plates. Axenic cultures were obtained after five passages from the first batch of agar plates. Routine maintenance of the culture was performed in 10-ml of medium at 20 °C without shaking. The strain is accessible in the Agriculture Research Service Culture Collection (Northern Regional Research Laboratory, NRRL 64474).

Genome sequencing and assembly

Genomic DNA from 11-day cultures of R. sphaerocarpa ETNP2018 (OD = 0.5) was extracted using QIAGEN DNeasy Plant Mini Kit (Cat. No. 69104) following the manufacturer’s instructions. The DNA quality was checked by a NanoDrop 2000c spectrophotometer and by a Agilent TapeStation 4150 system. The DNA quantity was measured using Qubit dsDNA BR Assay Kit (Invitrogen). The DNA sample was stored at -20 °C. Sequencing libraries were constructed using the TruSeq DNA PCR Free kit (Illumina # 20015962). The draft genome of R. sphaerocarpa ETNP2018 was sequenced at the University of California Davis Genome Center on an Illumina HiSeq 4000 platform with 150 bp paired-end sequencing. Raw reads were filtered with the BBDuk tool in the BBMap software package (v38.73) [24] using the settings “ktrim = r ordered minlen = 51 minlenfraction = 0.33 mink = 11 tbo tpe rcomp = f k = 23 ftm = 5”. Adapters were trimmed from the BBDuk-filtered reads using the tool Trimmomatic v0.39 [25] with the settings “ILLUMINACLIP:$adapters:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:100”. De novo assembly of reads that passed both quality filtering and adapter trimming was performed for each individual sample using SPAdes v3.15.2 [26] with kmer lengths 21, 33, 55, 77, 99, and 127.

Phylogenomic reconstruction of Rhodotorula genomes

For phylogenomic analysis, all Rhodotorula genome assemblies from the National Center for Biotechnology Information’s (NCBI) Assembly Database were retrieved. The genome of Rhodotorula sp. FNED7-22 was excluded due to its exceptionally low quality: it contains 2,892 scaffolds with a scaffold N50 value of 5.9 kb. Five metagenome-assembled Rhodotorula genomes were excluded also due to similar quality issues (large number of scaffolds and small scaffold N50 values). This leaves 167 Rhodotorula genomes for phylogenomic analysis, as well as two genomes (Microbotrium intermedium GCA 900096595.1 and Leucosporidium creatinivorum GCA 002105055.1) selected as part of the outgroup. Benchmarking Universal Single-Copy Orthologs (BUSCO) v5.4.4 was used to retrieve single-copy orthologues from all 169 genomes [27, 28]. A total of 983 single-copy orthologues were present in all Rhodotorula genomes. They were aligned individually using MUSCLE v5.1 [29] and trimmed with trimAl v1.2 with the parameters “-gt 0.6 -w 3 -st 0.001” [30]. The dataset was concatenated with the perl script “catfasta2phyml.pl” which generated an accompanying partition file (https://github.com/nylander/). IQ-TREE v1.6.12 was used to find the best model for each partition and infer a maximum-likelihood tree with the parameters “-m MFP + MERGE -rcluster 10 -bb 1000 -alrt 1000” [31, 32]. All partitions shared the same set of branch lengths but are allowed to have its own evolution rate [33]. The tree was visualized using Interactive Tree Of Life v6 [34].

Structural and functional annotations

For genome annotations and metabolic reconstruction, 14 Rhodotorula genomes were selected so as to capture a broad range of native habitats, as well as different species (Table S1) [16, 35,36,37,38,39,40,41,42,43,44]. BBMap was used to determine each genome’s GC content, scaffold and contiguous sequence (contig) counts, scaffold and contig N50 values, and total length [24]. Direct statistical comparison between overall averages and individual species was performed using directional one-sample t-tests at 95% confidence. Statistical analysis of differences between source categories was performed using two sample t-tests assuming unequal variance at 95% confidence.

Both structural and functional annotations were facilitated by Funannotate v1.8.14, a program for prediction, annotation, and comparison of eukaryotic genomes [45]. Each genome assembly was first soft-masked using tantan [46] and contigs < 1500 bp in length were excluded from downstream annotations. Structural annotations from BRAKER2 were generated in EPmode with OrthoDB v10 and all existing protein sequences from Rhodotorula genomes retrieved from the Joint Genome Institute Mycocosm as protein evidence, using ProHint as the alignment tool [7, 28, 47,48,49]. Because BRAKER2 utilizes GeneMark-EP + , the structural annotations from BRAKER2 are fed into the “Funannotate predict” pipeline which also uses AUGUSTUS v3.3.3 (with at least 500 training models), GlimmerHMM, and SNAP [50,51,52]. The SWISS-PROT protein knowledgebase, as well as existing protein sequences from Rhodotorula genomes retrieved from the Joint Genome Institute Mycocosm, was supplied as protein evidence to the “Funannotate predict” pipeline [7, 53]. EVidenceModeler was used to generate a weighted consensus of structural annotations from all four tools [54]. The program tRNAscan-SE was used to detect tRNA-coding genes [55].

Predicted protein sequences from structural annotations were annotated by BlastKOALA against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database [56, 57], by InterProScan v5.51–85.0 [58], and by the eggnog-mapper v2.19 against the eggnog 5.0 database [59, 60]. Carbohydrate-active enzymes were identified using the dbCAN2 meta server against the Carbohydrate Enzyme (CAZy) database [61, 62]. The dbCAN2 pipeline uses three tools (DIAMOND, HMMER, and Hotpep) and we only kept annotations supported by at least two tools to be conservative and confident. To determine potential function of all well-annotated CAZymes, the CAZy family number of each reported CAZyme was compared against its entry on the online CAZypedia index [63]. Biosynthetic gene clusters (BGCs) were identified by the Antibiotics and Secondary Metabolite Analysis Shell (AntiSMASH) v6.0 using the fungal version (fungiSMASH) [64]. BGC sequences identified by AntiSMASH were submitted to the NCBI Conserved Domain Database for refined annotation of BGC genes with the location of conserved domain footprints and functional sites inferred from these footprints [65, 66].

Metabolic reconstruction

Functional annotations were manually inspected to determine key metabolic pathways conserved in the R. sphaerocarpa ETNP2018 genome. Metabolic pathways were reconstructed based on the presence of key enzymes using the KEGG mapper, the Saccharomyces Genome Database (SGD), and published literature on budding yeasts [67,68,69]. Cofactors were determined using the SGD, and localization was determined using the annotations given by BlastKOALA or from the literature. This information was then used to create a custom graphic depicting the intercellular transport and localized carbohydrate metabolic pathways contained within the genome of R. sphaerocarpa ETNP2018.

Results and discussion

Genomic characteristics

The R. sphaerocarpa ETNP2018 genome was assembled into 115 scaffolds, with a total size of 17.7 Mbp, and an N50 value of 377,844 (Table 1). With 6,451 gene models, the gene density is 364 genes/Mbp, on par with previously published genomes of marine fungi [70]. BUSCO estimated that the genome is 97.3% complete. Of the 1764 BUSCOs from the OrthoDB v10 database for Basidiomycota, 1713 are present as single copies, 3 as duplicated copies, and 13 as fragmented in the R. sphaerocarpa ETNP2018 genome. There are 5,324 eukaryotic cluster of orthologs (KOGs), 7,872 protein family (Pfam) domains, and 137 tRNAs in the assembly. A total of 3,210 (49.8%) genes were annotated by KEGG orthology, as well as 172 CAZymes. The average genome size of all 15 Rhodotorula strains is 20.3 ± 1.6 Mbp and the average number of proteins per genome is 3,275 ± 154 (Table 1). The 15 Rhodotorula genomes included an average of 5,610 ± 360 KOG assignments, 8,240 ± 418 Pfam domains, and 194 ± 19 CAZymes. Marine yeasts isolated from seawater including R. sphaerocarpa ETNP2018, R. sphaerocarpa GDMCC 60679, R. diobovata 08–225, and R. mucilaginosa CYJ03 have an average genome size of 19.0 ± 1.6 Mbp with 3,203 ± 99 annotated proteins (Table 1).

Table 1 General genomic characteristics of R. sphaerocarpa ETNP2018 and fourteen representative Rhodotorula strains. "I.S.S." represents International Space Station
Full size table

A one sample t-test found the difference in genomic size between R. sphaerocarpa ETNP2018 and the average of all strains to be significant (p = 1.059 × 10–5). R. sphaerocarpa ETNP2018’s genome is the smallest of the marine strains; however, a one sample t-test found this difference to be insignificant (p = 0.099). The difference was significant when compared with the mean genome size of freshwater, endophytic, and terrestrial strains (p = 0.04, 0.039, and 0.0008, respectively). In contrast, the number of genes with functional annotations was not different between R. sphaerocarpa ETNP2018 and the average of all strains, as well as marine, freshwater, endophytic, and terrestrial source category averages (p = 0.06, 0.44, 0.12, 0.57, and 0.24, respectively).

The genome size of R. sphaerocarpa ETNP2018 was 10.8% smaller than the average genome size of five representative terrestrial Rhodotorula strains (Table 2). This reduction of 10.8% was the largest amongst the marine strains of Rhodotorula, prompting the hypothesis that the environmental pressures imparted at the OMZ favor fungal strains with smaller genomes in comparison to other marine environments. This hypothesis can be tested in future studies once additional fungal genomes from the open ocean become available. A metabarcoding survey showed that Rhodotorula was one of the most abundant and prevalent genera of the fungal communities in the ETNP [12], indicating that Rhodotorula are active in the oligotrophic ocean despite their reduced genome size. The degree of genome reduction we observed in marine Rhodotorula strains was similar to that in the marine bacterium Pelagibacter ubique HTCC1062, a well-documented example of a streamlined microbe (Table 2) [71]. However, the genome reduction by the heterotrophic yeast and bacteria was not as large as the reduction (38%) found in the marine chemoautotrophic archaea Nitrosopumilus maritimus SCM1 when it was compared to five terrestrial strains of ammonia-oxidizing archaea (Table 2; Table S2).

Table 2 The percent reduction in the genomic size of Rhodotorula strains isolated from seawater, compared with the average of the representative terrestrial Rhodotorula. Pelagibacter ubique HTCC1062 and Nitrosopumilis matrimis SCM1 were chosen as well-known examples of streamlined marine microbes for comparison. P. ubique HTCC1062 was compared to five terrestrial strains of the alphaproteobacterial genus Rickettsia and N. matrimis SCM1 to five representative ammonia-oxidizing archaea, all retrieved from the NCBI assembly database (Table S2)
Full size table

The number of KOGs related to translation and ribosomal biogenesis, the transport and metabolism of amino acids, carbohydrates, lipids, secondary metabolites, and coenzymes were lower in the genome of R. sphaerocarpa ETNP2018 than the other 14 Rhodotorula strains (Tables S3 and S4) [12]. Nevertheless, R. sphaerocarpa ETNP2018 contained the core set of protein-coding genes despite nutrient scarcity and its small genome (Figure S1), which is consistent with streamlining in response to nutrient deprivation [8]. Previously conducted studies on streamlined microbes have found that the average genome size was the smallest for microorganisms isolated from oligotrophic seawater and the largest for those isolated from soil [8]. Microorganisms isolated from freshwater exhibit a broad spectrum of genome sizes [8]. We find Rhodotorula genomes from both ends of the size spectrum consistent with this theory. Among the 15 Rhodotorula strains, R. glutinis ZHK and R. kratochvilovae YM25235 were the largest in genome size, and they were isolated from eutrophic environments (the Pearl River and Chenghai Lake, respectively) [15, 18]. The genomes of two soil strains, R. frigidialcoholis JG-1b and R. sp. CCFEE5036, were smaller than the average genome size of all 15 strains. However, both of these strains were isolated from permafrost and hyper-arid soil in Antarctica’s McMurdo Dry Valley [37, 39]. Their reduced genome size could be related to the extreme conditions of their environment with low nutrient availability.

Phylogeny

Phylogenomic analysis revealed multiple monophyletic clades at the level of species, including R. sphaerocarpa, R. paludigena, R. kratochvilovae, and R. toruloides (Fig. 1 and Figure S2). R. sphaerocarpa ETNP2018 was closely related to the R. sphaerocarpa strain isolated from mariculture seawater in Maoming, Guangdong, China. Together, these two R. sphaerocarpa genomes form a clade distinct from all other Rhodotorula species, demonstrating the evolutionary pressure that likely contributed to the speciation of R. sphaerocarpa, which has been primarily isolated from seawater [22, 72,73,74]. Among the Rhodotorula, some strains of the same species have been isolated from drastically different environments (Fig. 1). This could be attributed to the ability of Rhodotorula yeasts to adapt to a diverse range of environmental conditions [14, 39]. R. mucilaginosa, for example, has been isolated from soil, both animal and plant microbiomes, industrial mineral deposits, the International Space Station, and the marine water column [19, 43]. Our phylogenomic reconstruction shows that the recently described R. frigidialcoholis JG-1b [20] is closely related to R. mucilaginosa (Fig. 1). The sister clade of the main R. mucilaginosa clade includes R. frigidialcoholis JG-1b, two genomes of undescribed species, and four R. mucilaginosa genomes, so it is possible that these four R. mucilaginosa genomes obtained from the international space station may have been misclassified and are actually R. frigidialcoholis.

Fig. 1
figure 1

A phylogenomic tree constructed using single-copy orthologues shared amongst 168 Rhodotorula strains (as well as two outgroup species, Microbotrium intermedium GCA 900096595.1 and Leucosporidium creatinivorum GCA 002105055.1). 983 single-copy orthologs were retrieved for comparison using BUSCO v5.4.4. The isolation source is denoted in the parentheses after the strain names (“ISS” = International Space Station). Bootstrap values (n = 1000) are displayed at each node. The collapsed clade includes 102 strains of Rhodotorula mucilaginosa primarily from the International Space Station. The two side lengths of the triangle representing the collapsed clade are proportional to the distances to the node’s closest and farthest child leaves. R. sphaerocarpa ETNP2018 is highlighted by bold font and an underline. The same tree that displays the 102 collapsed Rhodotorula mucilaginosa strains is included as Figure S2

Full size image

CAZymes

CAZymes identified in the R. sphaerocarpa ETNP2018 genome include 56 glycoside hydrolases (GH), 84 glycosyltransferases (GT), 17 related to auxiliary activities (AA), 3 carbohydrate binding modules (CBM), 8 carbohydrate esterases (CE), and 4 polysaccharide lyases (PL) (Table S5). Chitinase (GH18), xyloglucanase (GH16), β-hexosaminidase (GH20), both α- and β-glucosidase (GH3/31), invertase (GH32), α,α-trehalase (GH37), α-mannosidase (GH38/47), and cellulase (GH5) glycoside hydrolase CAZymes are all conserved across the Rhodotorula genus (Fig. 2). Therefore, Rhodotorula yeasts have the potential to digest chitin, xylan, hexoses, trehalose, some mannose, and cellulose [63]. Trehalase and glycogen debranching CAZymes (GH13) are also present in all 15 representative members of the Rhodotorula genus, suggesting the ability to use both storage polysaccharides to maintain energy production in response to potential carbon deprivation (Fig. 2). The most prevalent and abundant glycosyl hydrolase families include GH5, GH16, and GH18 (Figure S3).

Fig. 2
figure 2

Heatmap of major genes involved in carbohydrate, nitrogen, and sulfur metabolisms in the Rhodotorula genome. Filled circle (light blue) indicate the presence of a gene whereas empty circles indicate its absence. Abbreviations for the headers are as follows: LM, Lactate Metabolism; EM, Ethanol Metabolism; AM, Aldehyde Metabolism; CS, Carotenoid Synthesis. The described functions of glycoside hydrolases (GH), pectinases (PL) and esterases (CE) are in Table S6. Abbreviations for the gene names are as follows: HK, hexokinase; galK, galactokinase; MPI, mannose-6-phosphate isomerase; XYLB, xylulokinase; NRT, nitrate/nitrite transporter; NR, nitrate reductase (NAD(P)H); NIT-6, nitrite reductase (NAD(P)H); NIT4, nitrate assimilation; amiE, aliphatic amidase; glnA, glutamine synthetase; GLT1, glutamate synthase (NADH); GDH2, glutamate dehydrogenase; met3, sulfate adenylyltransferase; cysC, adenylylsulfate kinase; cysH, phosphoadenosine phosphosulfate reductase; cysJ, sulfite reductase (NADPH) flavoprotein alpha-component; cysI, sulfite reductase (NADPH) hemoprotein beta-component; lldD, L-lactate dehydrogenase; LDHD, D-lactate dehydrogenase; ACH1, acetyl-CoA hydrolase; ackA, acetate kinase; acs, acetyl-CoA synthetase; BCKDHA, 2-oxoisovalerate dehydrogenase alpha subunit; BCKDHB, 2-oxoisovalerate dehydrogenase beta subunit; DBT, dihydrolipoyl transacylase; ACOX1 acyl-CoA oxidase; ECHS1, enoyl-CoA hydratase; HIBCH, 3-hydroxyisobutyryl-CoA hydrolase; mmsA, acetyl-CoA carboxylase; ACACA, malonate-semialdehyde dehydrogenase; ABAT, 4-aminobutyrate aminotransferase; adh, alcohol dehydrogenase (NADP +); adhC, alcohol dehydrogenase; adhP, alcohol dehydrogenase, propanol-preferring; ALDH, aldehyde dehydrogenase (NAD +); ALDH3, aldehyde dehydrogenase (NAD(P) +); AL2, 15-cis-phytoene synthase / lycopene beta-cyclase; AL1, phytoene desaturase

Full size image

Chitin is the most abundant biopolymer found in the marine environments and thus an important source of carbon and nitrogen for marine microbes [75]. Chitin is produced throughout the water column by fungi, protists, and crustaceans, yet is utilized so rapidly that it is present only in trace concentrations in marine sediments. The importance of chitin as a source of nutrient for marine fungi is relatively understudied however, as most marine chitin degradation is attributed to bacteria [75]. Chitinase is conserved throughout the Rhodotorula genus. GH18 was the most prevalent in genomes of freshwater strains (7 ± 1.4) and the least in genomes of marine strains (4 ± 1.4). Chitin degradation via chitinase results in a wide array of oligomers including diacetylchitobiose. The endo-β-N-acetylglucosaminidase (GH85, EC: 3.2.1.96), which degrades diacetylchitobiose into monomeric residues of β-1,4-N-acetyl-D-glucosamine (GlcNAc) and subsequently soluble sugars and dissolved organic nitrogen, was present in all Rhodotorula genomes except for R. graminis WP1 and R. kratochvilovae YM25235 (Fig. 2) [75, 76].

GH2, a β-mannosidase, and GH76, an α-1,6-mannanase, are absent exclusively in the two genomes of R. sphaerocarpa strains isolated from seawater (Fig. 2). Polygalacturonase (GH28), β-glucoronyl hydrolase (GH88), chitonsanase (GH75), and β-mannanase (GH26) CAZymes are absent in R. sphaerocarpa ETNP2018 as well as several other strains (Fig. 2). Mannanase and mannosidase allow yeast to ferment mannose to ethanol [77]; reduction in mannose-hydrolyzing CAZymes in the R. sphaerocarpa ETNP2018 genome suggests that it is not a commonly utilized substrate for the strain. Mannans are typically found in plant vacuoles and the endosperm of seeds, as well as the cell walls of certain yeasts [77]. These sources place open ocean yeast such as R. sphaerocarpa ETNP2018 far from a consistent supply of mannans, suggesting that its reduction in mannanase and mannosidase CAZymes is a response to the low encounter frequency for the substrate [77, 78].

A recent study demonstrated a positive correlation between a fungal strain’s repertoire of CAZymes and its saprophytic tendencies [79]. This suggests that strains which encode few CAZymes, such as those isolated from the water column, have less saprophytic tendencies and encounter fewer carbohydrates than those with higher counts, such as freshwater or endophytic strains. Fungi are the dominant detritovores in eutrophic freshwater ecosystems, and endophytic fungi are reported to opportunistically utilize saprophytic feeding mechanisms after the death of their host plant [80, 81]. The availability of organic matter in these environments makes the synthesis of many different CAZymes more energetically favorable in comparison with the oligotrophic open ocean. Given the low availability of organic matter in the open ocean water column, R. sphaerocarpa ETNP2018 likely streamlined its genome to reduce unnecessary and biosynthetically expensive CAZymes.

Central carbon metabolisms

The Embden-Meyerhof-Parnas pathway, Tricarboxylic acid (TCA) cycle, glyoxylate cycle, and pentose phosphate pathway were present in their entirety in the R. sphaerocarpa ETNP2018 genome (Figure S1). Potential substrates for these pathways include glucose, the cell’s preferred substrate, as well as acetate, ethanol, D-lactate, L-glutamine, and oxaloacetate (Fig. 3). The glyoxylate cycle, a secondary shunt of the TCA cycle localized in the peroxisome, utilizes isocitrate lyase (EC: 4.1.3.1) to catalyze the conversion of isocitrate to glyoxylate as well as succinate to malate without requiring the energy intensive decarboxylation steps required to form S-succinyl-dihydrolipoamide-E from isocitrate during the TCA cycle (Fig. 2) [82]. Glyoxylate cycle genes in yeast have been shown to upregulate in macrophage-engulfed Candida yeasts, concurrent with a downregulation of transcriptional machinery and glycolytic enzymes, allowing the cell to acquire carbon through alternative sources to glucose and conserve energy [82]. This suggests preferential use of the glyoxylate cycle as a response to glucose deprivation which is typical in the oligotrophic ocean.

Fig. 3
figure 3

Reconstructed metabolic pathways for Rhodotorula sphaerocarpa ETNP2018. The outer bi-layered membrane represents plasma membrane surrounding the cytosol, the inner bi-layered membrane the mitochondrion, and the inner single-layered membrane the peroxisome; shapes embedded into membranes represent transport proteins. Arrow and transport protein color gradients represent requirement for a specific cofactor in reaction or transport, with the second color of the gradient indicating preferred direction: Black/White, no cofactor required; Light Blue/White, H2O; Purple/Pink, ATP; Purple/Blue, ADP; Yellow/White, NAD+; Teal/Yellow, NADH; Green/Yellow, NADP+; Green/White, NADPH; Red/Pink, FAD2+; Black/Red, O2; Purple/Red, H2O2. Abbreviations are as follows: EtOH, Ethanol; CoA, Coenzyme A; THF, tetrahydrofolate; Pi, Inorganic Phosphate; AICAR, 5-aminoimidazole-4-carboxamide-1-β-D-ribofuranoside; PRPP, phosphoribosyl diphosphate; AMP, adenosine monophosphate; ATP, adenosine triphosphate; UDP, uridine diphosphate; UTP, uridine triphosphate; UDPG, UDP-α -D-Glucose; sFA, Saturated Fatty Acid; FAD, Flavin Adenine Dinucleotide; FMN, Flavin Mononucleotide; CoQ6, Ubiquinone-6; CoQ6H2, Ubiquinol-6; CoQ10, Ubiquinone-10; Cyt. C, Cytochrome C

Full size image

In case physical processes transport R. sphaerocarpa ETNP2018 to the anoxic part of the water column, its genome shows the potential to ferment pyruvate via the enzyme pyruvate decarboxylase (EC: 4.1.1.1), creating acetaldehyde that can be further converted to acetate, ethanol, or carboxylic acids (Fig. 2). Acetate is synthesized from acetaldehyde via aldehyde dehydrogenase (EC: 1.2.3.1) to replenish acetyl-CoA using its acetyl group (Fig. 2). Ethanol is then synthesized via alcohol dehydrogenase (EC: 1.1.1.2) alongside the interconversion of NADH to NAD+ as a mechanism of replenishing the intracellular reducing agent (Figs. 2 and 3). It can then be excreted passively or converted to acetate in the peroxisome (Fig. 3). The small number of Pfam domains related to short chain dehydrogenase enzymes (PF00106), which are responsible for fermentative reactions on aldehydes and alcohols (Table 3), suggests the niche of R. sphaerocarpa ETNP2018 is not the anoxic portion of the water column. Reduced fermentative machinery may rather serve as an adaptation to their oligotrophic yet oxygenated environment, where biosynthetic resources are at a premium, and anaerobic metabolisms act as a backup for unfavorable changes in conditions.

Table 3 Pfam domains with major depletions in the genome of R. sphaerocarpa ETNP2018 as compared to the average of all Rhodotorula strains studied
Full size table

Transporters

Compared to other Rhodotorula strains, the genomes of both R. sphaerocarpa ETNP2018 and GDMCC 60679 are particularly low in the number of the Pfam domain encoding the major facilitator superfamily (MFS) of transporters (PF07690) (Table 3), which play a significant role in the cross-membrane transport of organic solutes. This suggests R. sphaerocarpa strains isolated from seawater have streamlined their genomes given the low substrate availability in the ocean. MFS transporters also symport H+ with siderophores, organometallic molecules formed by prokaryotes to sequester ferric iron [83, 84]. Although yeasts are considered to utilize siderophore assimilation as an opportunistic iron uptake mechanism [84], the low number of MSF transporters in the genomes of R. sphaerocarpa strains suggests a low competitiveness in siderophore acquisition, a tradeoff resulting from genome streamlining.

Nevertheless, the number of other metal transporters annotated by Pfam was not lower in the genomes of R. sphaerocarpa ETNP2018 and GDMCC 60679 in comparison to the other strains. High affinity iron permease (PF03239) was conserved in all 15 Rhodotorula genomes, suggesting ferrous iron intake provides them with much of the iron required for protein synthesis. One copy of PF10566, which contains natural resistance associated macrophage protein (Nramp) transporters, was conserved across all 15 Rhodotorula strains, aside from R. toruloides NBRC 0880, which contains two, and R. kratochvilovae YM25235, which contains zero. Nramp transporters, belonging to the Smf family of genes, are responsible for the cross-membrane transport of a variety of transition metals [85, 86]. Smf proteins demonstrate the highest affinity for Cu2+ and Mn2+ and are thought to be responsible for the high-affinity Mn2+ uptake system, but also show function in transporting ferrous iron, copper, nickel, cadmium, cobalt, zinc, and manganese [83, 85].

Nitrogen assimilation

Ten of the 15 representative Rhodotorula strains encode the genes for nitrate assimilation pathways (Fig. 2), through which nitrate and nitrite are transported into the cell by the nitrate/nitrite transporter narK and reduced to ammonium via the enzymes nitrate reductase (EC: 1.7.1.1) and nitrite reductase (EC: 1.7.1.4) (Fig. 3). Six of these ten genomes were isolated from aquatic sources: two from freshwater, R. glutinis ZHK and R. kratochvilovae YM25235, and four from the marine environment, R. sphaerocarpa ETNP2018, R. paludigena P4R5, R. sphaerocarpa GDMCC 60679, and R. diobovata 08–225 (Fig. 2). The only aquatic yeast lacking this genetic potential is R. mucilaginosa CYJ03, isolated from the Yellow Sea in Yunnan, China (Fig. 2) [43].

Nitrate assimilation, in particular the reduction of nitrate in the cytosol, is an energetically expensive process [87]. Strains lacking the genetic potential to assimilate nitrate were largely isolated from environments where competition for resources is less intense and alternative sources of nitrogen (e.g. ammonium, urea) are likely readily available. R. mucilaginosa CYJ03 was isolated from the northern Yellow Sea, which has been eutrophic for decades [88]. It can therefore be inferred that R. mucilaginosa CYJ03 encounters comparatively high ammonium concentrations in the water column and to synthesize nitrate reductase would constitute a waste of biosynthetic resources.

Yeasts in the genus Rhodotorula have previously displayed the ability to grow on acetonitrile as a sole nitrogen source [89]. Nitrile hydratase (NHase) proteins, together with amidases (EC: 3.5.1.4), mediate a two-step metabolism of nitrile compounds such as acetonitrile to amides and acids; a second nitrile-hydrolyzing enzyme found in yeast, nitrilase (EC: 3.5.5.1), can perform the same reaction in one step. One of either nitrilase or cyanoalanine nitrilase (EC: 3.5.5.4) was found in all Rhodotorula genomes except for the two R. sphaerocarpa strains, Rhodotorula sp. CCFEE5036, and R. taiwanensis MD1149 (Fig. 2). However, none of the representative strains contained genes for NHase synthesis. Acetonitriles are predominately released via terrestrial biomass burning and constitute only a trace gas in the global atmosphere, placing open ocean R. sphaerocarpa strains far from stable sources of incorporable nitriles [90]. The lack of NHase, CobW, and nitrilase genes exclusively in both R. sphaerocarpa ETNP2018 and R. sphaerocarpa GDMCC 60679 suggests that as the lineage was diverging, R. sphaerocarpa strains did not retain CobW or nitrilase genes potentially due to a lack of available nitrile compounds in the ocean.

Secondary metabolisms

All Rhodotorula genomes contained between four and six BGCs, primarily from the categories of non-ribosomal polyketide synthase (NRPS) and Terpene synthesis. The genome of R. sphaerocarpa ETNP2018 included NRPS-like clusters 1.1 and 6.1 as well as terpene clusters 8.1 and 9.1. The core biosynthetic gene of terpene cluster 9.1 functions in the formation of an isoprenoid biosynthetic complex. A BLASTp search identified both lycopene β-cyclase (EC: 5.5.1.19) and phytoene synthase (EC: 2.5.1.32) domains in the complex. Both lycopene and phytoene result from the digestion of cytosolic acetyl-CoA during the mevalonate (MVA) pathway, which converts acetyl-CoA into isopentyl diphosphate (Fig. 3) [91]. Phytoene is converted to lycopene via the enzymatic action of phytoene desaturase (EC: 1.3.99.30), where it can be further metabolized to form the carotenoid β-carotene via lycopene β-cyclase [91]. Yeast carotenoids are responsible for protection from over-exposure to ultraviolet light, in addition to proposed antimicrobial activity [92]. Rhodotorula have been shown to increase carotenogenesis as light intensity increases, indicating the molecules have a photoprotective role in the cell [92, 93]. R. sphaerocarpa ETNP2018 encodes five laccases (AA1), the most of all 15 Rhodotorula analyzed (Table S6). Fungal laccases can be ligninolytic enzymes, and are also known to function in plant pathogenicity, detoxification, and pigment modification [94]. Laccases degrade β-carotene and other carotenoids, so they may play a role in the breakdown of intracellular carotenoid pigments. Light is attenuated more rapidly in eutrophic lakes than in pelagic seawater due to the presence of particulate and dissolved organic matter, so lake yeasts would have less exposure to potentially harmful UV light and thus a reduced requirement for pigmented molecules [95].

Conclusions

Our analysis suggests that the marine yeast R. sphaerocarpa ETNP2018 has adapted to conditions in the oligotrophic marine environment through genomic and subsequent biosynthetic streamlining. Its genome is smaller than the typical Rhodotorula strain, allowing it to conserve limited nutrients during replication without a massive reduction in potential proteins. Reduction was expected for transcription-related genes but was primarily found in biosynthetic genes. In the genome of R. sphaerocarpa ETNP2018, the number of KOGs related to carbohydrate, lipid, and secondary metabolisms was lower than average, and depleted Pfam domains were mostly related to transport and non-essential biosynthetic pathways. The number of CAZymes was lower in the genome of R. sphaerocarpa ETNP2018 than the average of Rhodotorula genomes, and the reduction in CAZymes was primarily found among CAZymes involved in the metabolism of non-marine biopolymers. The conservation of core carbohydrate metabolisms in the genome of R. sphaerocarpa ETNP2018, including carotenoid production, suggests it has maintained an independent lifestyle despite streamlining pressures.