Repository navigation
Ubiquinone-linked reactions may have been incorrectly removed #1015
Description
Previously, many reactions catalysed by flavin-associated enzymes were present in both a FAD-linked and ubiquinone-linked forms. The ubiquinone-linked forms have since been removed for being duplicates (discussed in #599, #607; fixed in #613, #710).
The FAD associated reactions are maintained, because FADH2 electrons can move between them via ETF even though in reality almost all FAD is enzyme-bound (discussed in #608). However, not all of the duplicate reactions were connected to ETF. Their FADH2 is not connected to the rest of the matrix, except via ubiquinone. The consequence of removing the ubiquinone-linked form of these reactions is that their FADH2 can only enter the ETC via MAR06911, a lumped reaction with 'or' gene rules for both ETF and ETF-independent FAD-linked enzymes (eg. GPD2).
This creates problems in enzyme-constrained (eg. GECKO) models. MAR06911 contains a single EC (1.5.5.1, for ETFDH) and is therefore assigned a single Kcat. This Kcat is a shared for all FADH2 entry to the ETC, ie. GPD2-derived FADH2 is constrained by the Kcat of ETFDH. Effectively, the glycerol-phosphate shuttle is being constrained twice, as both MAR01169/MAR00449 and MAR06911 are independently assigned a Kcat. ETF-independent reactions (eg. GPD2) should be able to contribute electrons to the ETC without using the ETF reaction. Notably, at high ATP synthesis rates, the glycerol phosphate shuttle carries a much higher flux than ETF-linked beta-oxidation (ref); it does not make sense that they now share a constraint.
A second problem with the current formulation is that beta oxidation can contribute electrons to the ETC without involvement of ETF. Because of the 'or' gene rules, even when ETFDH/ETFA/ETFB are removed, FADH2 generated by beta-oxidation can still enter the ETC via GPD2. This does not make sense, as in reality GPD2 is on the outer-facing side of the inner membrane while beta oxidation occurs in the matrix. It is known that ETFDH is required for beta-oxidation electrons to enter the ETC (ref).
Essentially, I think these reactions should be split into two categories: matrix and intermembrane space. Those reactions that occur in the matrix should remain in the FAD[m]/FADH2[m] associated forms, as the FADH2 electrons are effectively shared via ETF. Reactions that occur in the intermembrane space should be replaced by their ubiquinone-linked forms, because their FADH2 does not contribute to the matrix FADH2 pool except via ubiquinone. Because MAR06911 is reversible, the electrons can still reach matrix FAD, however, they no longer require MAR06911 to enter the ETC, and beta oxidation electrons can no longer reach the ETC via the glycerol-phosphate shuttle.
Suggested changes:
Replace FAD-linked with ubiquinone-linked versions for ETF-independent reactions:
- restore
MAR00483, removeMAR01169,MAR00449(GPD2) - restore
MAR03838, removeMAR08611(PRODH, PRODH2)
Activity
[-]Q-linked reactions may have been incorrectly removed[/-][+]Ubiquinone-linked reactions may have been incorrectly removed[/+]edkerk commentedon Jun 1, 2026
Thank you for reporting this and contributing with your reasoning. We'll look into this in the coming days and come back to you.
edkerk commentedon Jul 11, 2026
Thanks, you're right, and it turns out to be slightly broader than the two enzymes you flagged.
Confirmed and implementing. GPD2 and PRODH reduce ubiquinone directly (EC 1.1.5.3 and 1.5.5.2), so we're restoring
MAR00483andMAR03838and removing the FAD-linkedMAR00449,MAR01169andMAR08611.MAR06911is specifically ETFDH (EC 1.5.5.1), but itsorrule had six direct-CoQ dehydrogenases that don't use ETF at all (CHDH, DHODH, GPD2, PRODH, PRODH2, SQOR), all of them listed in the review already cited on that reaction, Banerjee et al. 2022. We're reducing it to the ETF system only,ETFB and ETFA and ETFDH, which removes the shuttle double-constraint and the beta-oxidation bypass you described. DHODH needs no reaction change since it already has its own CoQ reaction,MAR20167.MAR00483sn-glycerol-3-phosphate[c] + ubiquinone[m] => DHAP[c] + ubiquinol[m](GPD2)MAR00449,MAR01169MAR00483)MAR03838proline[m] + ubiquinone[m] => 1-pyrroline-5-carboxylate[m] + H+[m] + ubiquinol[m](PRODH/PRODH2)MAR08611MAR03838)MAR06911ENSG00000105379 and ENSG00000140374 and ENSG00000171503(ETFB and ETFA and ETFDH)On the earlier work (#607, #608, #612). This partly reverses a deliberate decision. The direct-CoQ reactions were removed and these genes were added to
MAR06911in full knowledge that they reduce CoQ directly, on the premise that matrix FAD acts as one shared pool. That premise is fine for the acyl-CoA dehydrogenases, whose electrons really are pooled by ETF, so leave those as they are. It doesn't hold for these membrane quinone reductases, which was also the reason Complex II was kept as its own reaction (MAR04652) at the time. In plain FBA the two formulations are equivalent. They only diverge in GECKO models, so this extends the Complex II exception rather than correcting the earlier work.A few things we're less sure about:
MAR20186still writes electrons into the matrix FADH2 pool:FAD[m] + GSH[m] + H+[m] + H2S[m] => FADH2[m] + GSSH[m]so onceMAR06911is ETF-only, those electrons would still be drained through ETFDH. We could instead reduce ubiquinone directly:GSH[m] + H+[m] + H2S[m] + ubiquinone[m] => GSSH[m] + ubiquinol[m]MAR08441(choline[m] + FAD[m] => betaine aldehyde[m] + FADH2[m]), but unlike SQOR its acceptor is only proposed, not established (Salvi & Gadda 2013). Convert it tocholine[m] + ubiquinone[m] => betaine aldehyde[m] + ubiquinol[m], or leave it feeding the FADH2 pool for now?ENSG00000250799) is really a hydroxyproline dehydrogenase (Summitt et al. 2015), yet it is annotated only on the two proline reactions,MAR08610(FAD[c] + proline[c] => 1-pyrroline-5-carboxylate[c] + FADH2[c] + H+[c]) andMAR03838(proline[m] + ubiquinone[m] => 1-pyrroline-5-carboxylate[m] + H+[m] + ubiquinol[m]). The model has no hydroxyproline reaction. Drop PRODH2 from these, or keep it as an approximation?fix: couple non-ETF flavoenzymes to ubiquinone instead of the ETF pat…
tom-hobbs commentedon Aug 6, 2026
Thank you for looking into this and resolving the issue. To me the fix looks correct. To respond to each of the points:
Thanks for making the fix!
fix: couple non-ETF flavoenzymes directly to ubiquinone (#1028)
fix: route SQOR and CHDH through ubiquinone, drop PRODH2 from proline…
fix: route SQOR and CHDH through ubiquinone, drop PRODH2 from proline…
edkerk commentedon Sep 27, 2026
#1105 corrects the cross-references and retired-id bookkeeping of the ubiquinone reactions restored in #1028 and #1081. The reactions themselves (equations, GPRs) are unchanged.
MAR00483 (GPD2, sn-glycerol-3-phosphate + ubiquinone). When it was restored it carried the ids of the FAD-linked reaction it replaced. It now has its own ids, the same as in its old deprecated row:
MAR03838 (PRODH, proline + ubiquinone). KEGG R10507 -> R01253 (L-proline + quinone); the FAD-form ids BiGG/Recon3D PROD2m and MetaNetX MNXR103187 removed; Rhea RHEA:23785 with master RHEA:23784 added.
Retired-id bookkeeping. After #1028 some deprecated rows still listed the reactivated ids as retired into them, and MAR03838 and the deprecated MAR08611 pointed at each other:
MAR08441 (CHDH, choline + ubiquinone, after #1081). MetaNetX MNXR153198 and ModelSEED rxn10062 describe the FAD-coupled reaction and are removed; no quinone-coupled equivalent exists in either database (Rhea has only the generic RHEA:17433). The evidence for the quinone route is now cited on the reaction: choline dehydrogenase reduces ubiquinone analogues (PMID:1218095, rat liver mitochondria); patients with ETF deficiency (MADD) keep making betaine and its products, so choline oxidation does not depend on ETF (PMID:9630673); CHDH is listed among the CoQ-reducing dehydrogenases in PMID:34428349 (review). Direct evidence for human CHDH with ubiquinone is still lacking (UniProt Q8NE62 gives a generic acceptor). The Recon3D id CHOLD2m stays as provenance.
Implemented in #1105.
edkerk commentedon Oct 5, 2026
Fixed in Human 2.1.0.