Rhodotorula toruloides: carbon-source fitness defects and annotation gaps

Rhodotorula toruloides: carbon-source fitness defects and annotation gaps

Back to Fungal growth and knockout phenotype resources

Bottom line: in genome-wide transposon fitness data for the basidiomycete
yeast R. toruloides, 370 genes are needed for growth on a specific carbon
source but not on glucose, and 137 of them strongly (fitness ≤ −3 in at most
six of 20 carbon conditions). These fall into recognisable catabolic modules:
aromatic compounds (96 genes), branched-chain amino acids (22), pentoses and
polyols (9), fatty acids and acetate (5), lactate (3), galactose (1) and
cellobiose (1). The method
recovers expected enzymes, such as galactokinase on galactose and
branched-chain keto acid dehydrogenase on leucine and valine. Of the 137
strong genes, 26 map to a UniProt entry with no GO annotation (5 more have no entry we
can confidently map), and 11 map to current
entries that are fused gene models, two or more neighbouring genes joined
into one protein and often named after the wrong part: a
3-methylcrotonyl-CoA carboxylase entry is called "Nup85" and an intradiol
dioxygenase entry "ADP-ribose pyrophosphatase". The fusions are analysed in
GENE_MODEL_ERRORS.
Joining the data to UniProt takes some work, because most accessions from the
source strain have been deleted from UniProtKB.

Data

Method

scripts/rhoto_specific_phenotypes.py calls a specific defect when a gene's
fitness in a carbon condition is ≤ −2 and its fitness in every glucose control
is > −1. 327 genes are excluded as sick on glucose. The thresholds are stricter
than the Fitness Browser's because there are no t statistics.
scripts/rhoto_candidates.py keeps the strong subset (strongest defect
≤ −3, defective in ≤ 6 carbon conditions) and assigns each gene to a module by
its strongest condition. A module records which growth condition needs the
gene, not what the gene does.

Mapping to current UniProt entries

The 2023 accessions point at the IFO0880 proteome (UP000239560), now a
non-reference proteome. Of the 370 genes with a specific defect, 209 have a
2023 accession that has since been deleted from UniProtKB; their
sequences survive only in UniParc, with no identical active entry.
The shared resolver
PROTEOME_REMOVAL/scripts/resolve_deleted_accessions.py
fetches each sequence from UniParc and searches it with phmmer against the
reference proteome (UP000199069, locus tags BN2166_*):

Call Genes Meaning
active 161 2023 accession still in UniProtKB
same_gene 146 ≥ 90% identity over ≥ 80% of the query
same_gene_partial_model 39 ≥ 95% identity over a shorter span of at least 100 residues (or half the query); the two assemblies' gene models differ in length
weak 10 lower identity, or a near-identical span too short to call; not used
no_hit 14 no significant match

So 346 of 370 genes can be reviewed against a current UniProt entry.

Modules

Validation: expected enzymes are recovered

Aromatic catabolism

Benzoate, p-coumarate, ferulate and phenylalanine together account for 96 of
the 137 strong genes. Many are probably general stress genes (respiratory
chain, ESCRT, actin regulators), since benzoate and p-coumarate are also weak
acids. The strongest hits, though, carry domains that fit a fungal
β-ketoadipate pathway. That is an inference from domain annotations, not an
established function:

Gene Strongest defects Domain evidence Current UniProt GO terms
RTO4_9789 benzoate −11.4, p-coumarate −5.0 HD domain; KEGG K06950 A0A0K3CMN6, "HD domain containing protein" 0
RTO4_12623 benzoate −7.5, p-coumarate −5.7 Intradiol ring-cleavage dioxygenase (PF00775) A0A0K3C7C1, "ADP-ribose pyrophosphatase" 2
RTO4_10304 benzoate −7.1 Cytochrome P450 A0A0K3CMI5 4
RTO4_11658 benzoate −8.1 Zn2Cys6 transcription factor A0A0K3CRQ3, "predicted protein" 4
RTO4_12620 benzoate −5.8, p-coumarate −3.9 Carboxymuconolactone decarboxylase (PF02627) plus α/β hydrolase (PF12697) A0A0K3CEH3, "alpha/beta hydrolase" 0
RTO4_12621 benzoate −5.7, p-coumarate −4.2 Dimeric α/β barrel A0A0K3CAC0, "Uncharacterized protein" 0
RTO4_13090 p-coumarate −6.6, benzoate −5.7, ferulate −4.1 Succinyl-CoA:3-ketoacid CoA-transferase A0A0K3CAS9 2
RTO4_15228, RTO4_9065 p-coumarate −5.5 and −5.9 3-oxoacyl-CoA thiolases A0A0K3CDS0, A0A0K3C8E1 4 each
RTO4_14934 p-coumarate −7.1, ferulate −3.5 α/β hydrolase A0A0K3CGB7 0

RTO4_12620, RTO4_12621 and RTO4_12623 sit at consecutive reference loci
(BN2166_0014570, _0014580, _0014600), so they probably form a gene cluster.
RTO4_9789, the strongest defect in the whole matrix, has no functional
annotation.

Annotation problems the fitness data exposes

Tiling each entry with the IFO0880 proteome and checking the fitness of
each part confirms the first two below as fusions, along with nine more
entries carrying strong candidates (column current_entry_fused in
data/rhoto_candidates.tsv; details in
GENE_MODEL_ERRORS).

Caveats

Next steps

  1. Review the aromatic cluster (RTO4_12620, RTO4_12621, RTO4_12623), RTO4_9789
    and MCC α (RTO4_12867) as gene reviews. These would be the first
    R. toruloides entries in this repository (genes/RHOTO/). Several sit in
    fused current entries, so the reviews need a way to say which part they
    cover; see the proposal in GENE_MODEL_ERRORS.
  2. Report the fused gene models to UniProt.
  3. Add t statistics if the authors' per-replicate data become available, and
    repeat with the Fitness Browser specificity criteria.

Files

Run from the repository root:

uv run python projects/FUNGAL_PHENOTYPES/scripts/rhoto_specific_phenotypes.py
uv run --with pyhmmer python projects/PROTEOME_REMOVAL/scripts/resolve_deleted_accessions.py \
    --from-tsv projects/FUNGAL_PHENOTYPES/data/rhoto_specific_defects.tsv \
    --column uniprot_2023 --proteome UP000199069 --cache tmp/fungal_phenotypes \
    --out projects/FUNGAL_PHENOTYPES/data/rhoto_accession_resolution.tsv
uv run python projects/FUNGAL_PHENOTYPES/scripts/rhoto_candidates.py

Downloads (supplements from Europe PMC and the UniProt reference proteome)
are cached in tmp/fungal_phenotypes/.