Mitochondrial Fatty Acid β-Oxidation (cross-species)

Overview

A cross-species curation of the mitochondrial fatty acid β-oxidation (FAO)
spiral — the four-step cycle that degrades saturated fatty acyl-CoA esters two
carbons at a time (acyl-CoA dehydrogenase → enoyl-CoA hydratase →
3-hydroxyacyl-CoA dehydrogenase → 3-ketoacyl-CoA thiolase). The project reviews
the full enzyme complement in human and Drosophila melanogaster, ties
them together in the cross-species module
MODULE:fatty_acid_beta_oxidation,
and uses the pathway as a testbed for several recurring curation problems:
chain-length substrate specificity, cross-paralog/cross-gene mis-annotation,
moonlighting functions, organelle (mitochondrion vs peroxisome) assignment, and
GO↔RHEA reaction mapping.

Scope: the core saturated spiral is 20 grounded gene products (10 human + 10
Drosophila; the module's derived QC reports 20/20 reviewed). The project has since
been extended with the fly step-3 enzyme (DROME/scu), the mouse flagship
mouse/Acadl (LCAD), and the unsaturated-FAO auxiliary enzymes (see the
cassette section below).

The gene set

Unsaturated fatty acid β-oxidation — the auxiliary-enzyme cassette

Saturated fatty acids are degraded by the four-step spiral above. Unsaturated
fatty acids need auxiliary enzymes because their double bonds end up in
positions the core enoyl-CoA hydratase cannot act on:

• Double bond at an odd-numbered carbon (e.g. oleate 18:1Δ9) → after chain
  shortening a (3Z)-enoyl-CoA forms, which a Δ³,Δ²-enoyl-CoA isomerase
  (EC 5.3.3.8) converts to the (2E)-enoyl-CoA that rejoins the spiral.
• Double bond at an even-numbered carbon (e.g. linoleate 18:2) → a
  (2E,4Z)-dienoyl-CoA forms, which 2,4-dienoyl-CoA reductase (NADPH,
  EC 1.3.1.34) reduces to a (3E)-enoyl-CoA, then the Δ³,Δ²-isomerase re-isomerizes
  it; a Δ³,⁵,Δ²,⁴-dienoyl-CoA isomerase (EC 5.3.3.21) handles the
  conjugated-diene variant.

This is the "cassette" Rossana's GO-CAM deck models on top of the saturated model.
The fly auxiliary enzymes are now curated:

Mitochondrial vs peroxisomal isomerase. For grounding the deck's mitochondrial
cassette, the Δ³,Δ²-isomerase step maps to the mitochondrial CG4592/CG4594/CG4598
cluster — not to Dci, which is the peroxisomal-type (PECI/ECI2) enzyme. This
resolves the mito-vs-peroxisome ambiguity noted on the Dci review.

Curation-model gap: no fly 2,4-dienoyl-CoA reductase (DECR1)

The reductase step of the even-position unsaturated cassette has no
confidently-assignable Drosophila ortholog. Four ortholog resources converge:
NCBI Datasets (841 DECR1 orthologs, zero in Drosophila), UniProt (no DROME
protein named 2,4-dienoyl-CoA reductase or EC 1.3.1.34), Ensembl (none), and
OrthoDB — whose only fly candidate, CG3603, is PANTHER-classified as a
(3R)-3-hydroxyacyl-CoA dehydrogenase / mitochondrial FAS-II 3-oxoacyl-ACP
reductase (fatty-acid biosynthesis), not DECR1, so it was not curated as
the reductase. A GO-CAM of the even-position PUFA cassette therefore cannot
currently ground the reductase node in the fly; the enzyme is likely divergent or
unannotated (a candidate for reciprocal-BLAST / experimental resolution).

Curation findings and patterns

Chain-length substrate specificity

Each step is run by a family of chain-length-specific isozymes. The curation
principle is use the chain-length-specific molecular-function term where one
exists, and fall back to the general term where it does not ("not all MF have
them"): VLCAD GO:0017099, MCAD GO:0070991, SCAD GO:0016937, long-chain
3-OH-acyl-CoA DH GO:0016509; general fall-backs for the hydratases (GO:0004300)
and the straight-chain thiolases (GO:0003988). See the flagship
Enzyme Specificity project for the worked table.

Cross-paralog and cross-gene mis-annotation

• ACADVL ← mouse LCAD (Acadl, P50544): 8 IEA/ISS annotations were
  cross-transferred from the paralog, including a self-contradictory "negative
  regulation of fatty acid oxidation" — removed.
• ACAT1 ← SOAT1/SOAT2 name collision: cholesterol O-acyltransferase +
  ER-localization annotations (from a SARS-CoV-2/CH25H paper about the ER
  acyl-CoA:cholesterol acyltransferase) were mis-attributed to the mitochondrial
  T2 thiolase — removed.
• HADHA: thiolase MF removed (that activity is the HADHB subunit's), plus a
  wrong-paper thioesterase term (a Them5/Acot15 paper).

Moonlighting functions

• HADH → GLUD1 (glutamate dehydrogenase) inhibitor / negative regulation of
  insulin secretion (definitive protein-protein interaction evidence,
  PMID:20670938).
• HADHA → monolysocardiolipin acyltransferase / cardiolipin remodeling
  (third core function).
• ACAT1 → tetramer protein-lysine acetyltransferase of PDH components
  (PMID:27867011, verified).

The mitochondrial trifunctional protein (MTP)

Steps ②–④ for long-chain substrates are carried by the membrane-bound MTP, an
α₂β₂ heterotetramer (α = HADHA / DROME/Mtpalpha; β = HADHB / DROME/Mtpbeta).
The reviews keep α- and β-subunit activities consistent across species.

GO↔RHEA reaction-chaining gap

The cross-species module's reaction-chaining check flags
step ①→② as a break: step ① makes (2E)-enoyl-CoA, but the enoyl-CoA hydratase
MF GO:0004300 maps only to RHEA:20724 (the (3E) variant), not the canonical
(2E) crotonase RHEA:16105. The chemistry is correct; the gap is in the
GO:0004300 → RHEA mapping. Recorded as a MAPPING_GAP override on the module
connection; see projects/RHEA.

Reactome cross-check

A pathway-level comparison against Reactome's mitochondrial β-oxidation reaction
set — see FAO module vs Reactome
— shows the module and Reactome agree on the skeleton and on steps ①–③
(the same chain-length-specific dehydrogenases; the same ECHS1/HADH-vs-MTP split
between short/medium and long chains; and Reactome models step ①→② with the
correct (2E)/crotonyl substrate, independently confirming the RHEA gap above is a
mapping artefact). The one substantive difference is step ④: Reactome routes
the thiolase across the whole spiral (C6→C16) through the MTP complex and never
invokes ACAA2 or ACAT1, whereas our reviews resolve it into ACAA2 (medium/long)
and ACAT1 (short-chain/ketone). We consumed Reactome at the annotation level
throughout (many Reactome:R-HSA-… TAS annotations adjudicated per gene); this
is the pathway-level cross-check.

Organelle specificity — mitochondrion vs peroxisome

β-oxidation runs in both mitochondria (this gene set) and peroxisomes
(ACOX1, HSD17B4, ACAA1, SCP2…). Several fly annotations carried peroxisome calls
of uncertain standing. Rather than run our own bioinformatics, these were phrased
as neutral function-assignment hypotheses and tested with the OpenScientist
workflow (independent AI scientist; our review conclusions held out for
comparison):

The dissociation is the point: two proteins annotated peroxisomal by the same
inventory paper (Faust et al. 2012) resolve oppositely under independent analysis —
Acat1's call is refuted (computational-only, non-canonical -EKL motif, contradicted
by LOPIT/subfamily mitochondrial evidence), while Mtpalpha's is supported (a
Drosophila-conserved SKL PTS1 and isoform-resolved dual targeting). Both verdicts
were produced by OpenScientist blinded to our review actions and agree with the
conclusions we reached independently.

Chain-length specificity — OpenScientist structural verdicts

Chain-length specificity in the acyl-CoA dehydrogenase (ACAD) family is set by a
computable feature: the depth and hydrophobic lining of the substrate-binding
cavity. That makes the chain-length calls testable structurally (AlphaFold +
sequence), not just from literature. We ran the two most decision-relevant
specificity questions — both testing our own MODIFY actions — through
OpenScientist, blinded to the prior review action (--as-function-hypothesis):

| DROME/Echs1 | Echs1's crotonase pocket has the same broad short/branched-chain enoyl-CoA range as human ECHS1 | Strongly supported — substrate range conserved, not narrowed. Both catalytic glutamates and the active-site/substrate-binding residues are 100% conserved with human ECHS1; in vivo, Echs1-loss flies accumulate the valine-pathway substrate methacrylyl-CoA. | Confirms the dual hydratase / valine-catabolism core functions |

| DROME/Mcad | Mcad substrate cavity matches medium-chain (C6-C12) like human MCAD | Supported. 20/21 active-site residues identical to human MCAD; the one substitution (F372 vs L376) is conservative; AlphaFold active-site geometry near-identical (r=1.0 CA-CA distances). No support for a different chain-length term. | Confirms our ACCEPT of medium-chain GO:0070991 |

The residue analyses are the value-add: OpenScientist supplies an independent
structural basis (ACAD9 Thr-139/Ala-143 channel geometry; CG4860 RIGIA+8
Leu→Thr; Echs1 fully-conserved crotonase pocket; Mcad 20/21 pocket identity) for
the chain-length / substrate-range calls our reviews made on enzymatic/genetic
grounds. All six blinded runs to date (Acat1, Mtpalpha, ACAD9, CG4860, Echs1,
Mcad) have agreed with the conclusions the reviews reached independently.

Resolving the step-3 fly ortholog gap (scully)

The module long marked the fly short/medium-chain 3-hydroxyacyl-CoA dehydrogenase
(human HADH step ③) as unresolved. It is unresolved for a real reason:
Drosophila lacks a clean 1:1 ortholog of the classical HADH1/SCHAD (type-I,
3HCDH fold, Pfam PF00725) — the FlyBase genes named Had1 (CG9914) and Had2
(CG10131) are actually CRYL1 / L-gulonate-3-dehydrogenase orthologs
(EC 1.1.1.45). The fly enzyme that covers this step is scully (scu, O18404),
the ortholog of human HSD17B10 (type-II 3-hydroxyacyl-CoA dehydrogenase, SDR
fold, Pfam PF00106; Torroja et al. 1998).

A blinded OpenScientist run (narrowed to a single Foldseek/catalytic-residue
analysis after a first attempt hit the 7200 s ceiling) returned SUPPORTED:
orthology confirmed by PANTHER/OrthoDB/eggNOG, 73.1% identity with 100%
conservation of all 11 catalytic/binding residues, near-identical AlphaFold
active-site geometry, and a direct enzymatic assay (PMID:12917011) confirming
(3S)-3-hydroxyacyl-CoA dehydrogenase activity in the fly protein. Caveats it
flags: scully is type-II (HSD17B10), not classical HADH (type-I); it is
multifunctional (also a steroid dehydrogenase and the MRPP2 subunit of
mitochondrial RNase P); and the long-chain step ③ is carried by MTP (Mtpalpha).
scully is now the grounded fly S/MC representative for step ③ in the module.

Validating datasets

Independent, in-vivo data that reports each enzyme's substrate specificity
(surveyed, not re-run):

• Acylcarnitine chain-length signatures (human newborn screening): MCAD
  deficiency → C8; VLCAD → C14:1/C14:2; LCHAD/HADHA → C16-OH — direct in-vivo
  readouts of which chain-length step is blocked.
• Fly CRISPR-KO acylcarnitine profiling (Geronazzo et al. 2025, PMID:40519079):
  the same clinical assay in Drosophila — Arc42 KO reproduces the ACADS/SCAD C4
  (butyrylcarnitine) signature, CG4860 KO does not, functionally disambiguating
  two paralogs indistinguishable by sequence/EC.
• MTBLS636 (MetaboLights): mitochondrial cardiolipin molecular-species atlas
  across ~20 human cell lines — the CL space HADHA's MLCL-acyltransferase remodels.
• Spatial proteomics for the organelle question: LOPIT-DC U-2 OS
  (PRIDE PXD011254) and Dynamic Organellar Maps (HeLa) resolve mitochondrion vs
  peroxisome per protein; MitoCoP (PMID:34800366) is the high-confidence
  human mitochondrial reference.

FlyBase curation gap (PMID:40519079)

The Geronazzo et al. 2025 CRISPR study is a rich functional resource, but FlyBase
has so far propagated only one GO annotation from it to GOA:

Arc42  involved_in  GO:0033539 (fatty acid beta-oxidation using acyl-CoA dehydrogenase)  IMP  PMID:40519079  (2026-06-12)

Not yet captured (candidate annotations):

Our reviews already encode the Arc42-vs-CG4860 disambiguation (Arc42 → core
short-chain MF; CG4860 → MODIFY to general acyl-CoA DH activity), so they are
ahead of FlyBase on this paper's implications.

Curation-model gap: representing experimentally-supported negatives

The CG4860 result is a clean negative: an experiment shows the paralog does
not carry the short-chain role that electronic annotation assigns it. The
schema supports proposing a novel NOT annotation (negated: true +
action: NEW) — but only for a GO term absent from the gene's GOA. The
NEW-exists check is term-level (it ignores the negated flag), so a NOT + NEW
on GO:0016937 is rejected because the positive GO:0016937 is already in GOA.

Consequently there is no first-class way to negate an existing positive
electronic annotation; the recourse today is MODIFY / MARK_AS_OVER_ANNOTATED
/ REMOVE. A schema/validation enhancement — keying the NEW-exists check on
(term, negated) so a curator-supported NOT can coexist with a positive
annotation — would let experimentally-supported negatives (like CG4860's) be
represented directly. Flagged here as a candidate improvement.

STATUS

Completed

• [x] 8 human FAO gene reviews (ACADVL, ACADM, ACADS, HADHA, ECHS1, HADH, ACAT1, ACAA2); ACAD9 + HADHB pre-existing
• [x] 10 Drosophila ortholog reviews
• [x] Cross-species module + reaction-chaining check (QC 20/20)
• [x] OpenScientist organelle hypothesis: DROME/Acat1 peroxisome (refuted)
• [x] OpenScientist organelle hypothesis: DROME/Mtpalpha peroxisome (supported — conserved SKL PTS1, isoform-dependent dual targeting)
• [x] OpenScientist chain-length/substrate specificity runs (ACAD9, CG4860, Mcad, Echs1) — all confirmed our reviews; see Chain-length specificity
• [x] Fly step-3 ortholog resolved + reviewed: DROME/scu (scully, HSD17B10 type-II); no classical HADH1 ortholog in fly
• [x] Pathway-level Reactome cross-check (sub-page)
• [x] Mouse arm started: mouse/Acadl (LCAD) — homes the ACADVL cross-paralog cleanup
• [x] Unsaturated-FAO auxiliary-enzyme cassette curated: DROME/Dci, DROME/Ech1, DROME/CG4592, DROME/CG4594, DROME/CG4598

In progress / open

• [ ] 2,4-dienoyl-CoA reductase (DECR1) fly ortholog — no clean ortholog found (see cassette gap above); needs reciprocal-BLAST / experimental resolution
• [ ] Remaining mouse FAO orthologs (Acadvl, Acad9, Acadm, Acads, Hadha, Hadhb, Echs1, Hadh, Acat1, Acaa2); then rat and worm arms
• [ ] FlyBase curation-gap candidates from PMID:40519079 (above)
• [ ] Negation-representation gap (NOT+NEW vs existing positive term)

Last updated: 2026-07-01

NOTES

2026-07-01

Created the project to capture the cross-species FAO curation, the validating
datasets (acylcarnitine + fly CRISPR + MTBLS636 + spatial proteomics), the
OpenScientist organelle work, the FlyBase under-curation of PMID:40519079, and
the NOT/NEW negation-representation gap surfaced by the CG4860 negative result.

Drosophila Echs1 accession/fetch-gap fix. The PR review flagged fly Echs1
as having only 2 (IEA) GOA annotations. Cause: fetch-gene resolved the gene to
the isoform accession Q0E987 (2 IEA terms) rather than the well-curated
primary Q7JR58 (12 annotations, including experimental IDA/HDA mitochondrion
and IMP L-valine catabolic process from PMID:40056416 / PMID:39727068). Re-fetched
with -u Q7JR58 --force (merged, preserving reviewed content), reviewed the 10
newly-surfaced annotations, and re-aligned the review id and the module
representative_member for the hydratase step to Q7JR58.