All-trans-retinoic acid metabolism — RBP4/STRA6 retinol delivery, CRBP1 binding, RDH10/DHRS3 retinol-retinal interconversion, ALDH1A1-3 retinal oxidation, CRABP1/2 binding and CYP26A1/B1/C1 catabolism

All-trans-retinoic acid (atRA), the most potent vitamin A metabolite, is made on demand inside target cells from circulating retinol and is cleared by oxidative catabolism, so that its tissue level is set by the balance of local synthesis and local destruction. Retinol travels in plasma bound to retinol-binding protein 4 (RBP4), which circulates as a complex with the transthyretin (TTR) tetramer. The multi-pass membrane receptor STRA6 binds holo-RBP4 and moves retinol into the cell, where it is bound by cellular retinol-binding protein 1 (CRBP1, RBP1). The first and rate-limiting step of synthesis is the reversible, membrane-associated oxidation of all-trans-retinol to all-trans-retinal by the short-chain dehydrogenase/reductase RDH10. RDH10 works together with the retinaldehyde reductase DHRS3: the two proteins activate each other, and DHRS3 reduces retinal back to retinol, so the pair sets the flux of retinal toward acid. Cytosolic medium-chain alcohol dehydrogenases can also oxidise retinol. The second step, the oxidation of retinal to atRA, is irreversible and is catalysed by the NAD+-dependent retinaldehyde dehydrogenases ALDH1A1, ALDH1A2 and ALDH1A3, whose distinct tissue patterns define where atRA is produced in the embryo. Cellular retinoic acid-binding proteins bind atRA in the cytosol: CRABP2 delivers it to the nuclear retinoic acid receptors, and CRABP1 sequesters it and favours its catabolism. atRA is inactivated by the cytochrome P450 retinoic acid hydroxylases CYP26A1, CYP26B1 and CYP26C1, which form 4-hydroxy-, 4-oxo- and 18-hydroxy-atRA in the endoplasmic reticulum; CYP26A1 and DHRS3 are induced by atRA, giving negative feedback on its own level. Retinol not used for synthesis can be stored as retinyl esters, made by LRAT and hydrolysed back to retinol by retinyl ester hydrolases such as the carboxylesterase CES1. Loss of RDH10 or ALDH1A2 in mouse abolishes embryonic atRA synthesis, CYP26B1 deficiency causes craniosynostosis and skeletal anomalies, and STRA6 and RBP4 defects cause microphthalmia and coloboma.

Derived from the RGD Pathway Ontology record and diagram PW:0001004 (retinoic acid metabolic pathway) through the RGD_PATHWAY_ONTOLOGY project's diagram->module route; GO mapping GO:0042573 retinoic acid metabolic process (skos:closeMatch). The 18-gene RGD human set and the diagram give the boundary: circulating retinol in, atRA made, bound and catabolised, with an output to the retinoic acid signaling pathway (PW:0001014). It is modelled as six required parts and one optional part: (1) RBP4/TTR transport and STRA6 uptake; (2) CRBP1 binding; (3) the RDH10-DHRS3 retinol dehydrogenase/retinaldehyde reductase pair; (4) a variant set of the ALDH1A1, ALDH1A2 and ALDH1A3 retinaldehyde dehydrogenases; (5) a variant set of CRABP2 and CRABP1; (6) a variant set of the CYP26A1, CYP26B1 and CYP26C1 hydroxylases; and (7, optional) retinyl ester storage, with LRAT esterification by module_ref and CES1 hydrolysis. Second RGD pathway considered, PW:0000141 (retinol metabolic pathway, GO:0042572), is NOT listed as an evidence source and is not claimed as covered. Its record and diagram describe dietary vitamin A absorption in the intestine: carotenoid uptake by SCARB1 and CD36, retinyl ester hydrolysis in the gut lumen by PNLIP, PNLIPRP2 and PLB1, central cleavage of beta-carotene to retinal by BCO1, reduction of retinal by AKR1B10, binding by RBP2 (CRBP2), re-esterification by LRAT, chylomicron secretion and ABCA1 efflux, with the ISX feedback on SCARB1 and BCO1. Of its 11 RGD human genes only LRAT (here by reference) and ALDH1A2 fall inside this module. Its content is enterocyte absorption and chylomicron packaging, a distinct module upstream of the plasma RBP4 pool, so it should get its own module rather than be counted as covered here. Boundary decisions. (a) Downstream signalling is not modelled: RAR/RXR activation and coregulator exchange are in modules/retinoic_acid_receptor_signaling.yaml (RGD PW:0001014). That module has a compact retinoic acid production and transport part, with ALDH1A2 and CRABP2 grounded by PAINT nodes, which overlaps parts 4 and 5 here; it could be replaced there by a module_ref to this module's retinal_oxidation and retinoic_acid_binding nodes (not edited here). (b) The retinoid visual cycle (modules/retinoid_visual_cycle.yaml) shares LRAT and DHRS3. LRAT is included by module_ref to its retinol_esterification node, since the reaction and enzyme are the same. DHRS3 is modelled again here, with its RDH10 partner, because the visual-cycle node treats it as a stand-alone buffer and has no RDH10; the activity term (GO:0052650) is the same in both. That module does not contain RBP4 or STRA6, so the plasma delivery step is modelled here. (c) Retinyl ester storage is optional: it buffers the retinol supply rather than being a step from retinol to atRA. CES1 is the RGD-drawn hydrolase; its GO:0050253 rests on IBA, and the purified liver retinyl ester hydrolase is a non-specific carboxylesterase, so it is a representative, not the dedicated enzyme. The mouse GO-CAM gomodel:67086be200001721 uses Pnpla2 for the same step in stellate cells. (d) Context, not participants: cytosolic medium-chain alcohol dehydrogenases (ADH1 class, ADH4, ADH7), which oxidise retinol in vitro and in mouse GO-CAMs gomodel:66f5faaa00001204 and gomodel:67086be200002705 (Adh1 GO:0004745), but whose redundancy RDH10 genetics argues against as the main embryonic route (PMID:17473173); and the other SDR retinol dehydrogenases in the RGD gene list, RDH16 and DHRS9, which have experimental GO:0004745 rows in human GOA but whose contribution to atRA synthesis in vivo is less clear than that of RDH10 and which also act on 3-alpha-hydroxysteroids. They could be added as an optional variant of part 3 if a curator judges the in vivo evidence sufficient. (e) FABP5 is drawn by RGD as delivering atRA to PPARD. It is a broad fatty acid-binding protein (GO:0005504 IDA), and the atRA-PPARD route is contested, so it is not a participant. (f) Not drawn by RGD and not modelled: BCO1 cleavage of beta-carotene (an alternative retinal source, in PW:0000141), RETSAT, the 9-cis-RA routes, and DHRS3/CYP26A1 transcriptional induction by atRA, which is the feedback recorded in the description but which acts through the RAR module and is not drawn as a connection. Molecular functions come from each protein's experimental GO annotations in QuickGO (checked 2026-11-05): RBP4 retinol binding GO:0019841 (IDA, PMID:571335), with GO:0140104 molecular carrier activity used as the annoton function after the mouse GO-CAM; STRA6 GO:0034632 (IDA, PMID:21901792); RBP1 GO:1904768 (IDA, PMID:28057518); RDH10 GO:0004745 (IDA, PMID:12407145, PMID:19458327); DHRS3 GO:0052650 (EXP, PMID:9705317; core function of the completed DHRS3 review); ALDH1A2 GO:0001758 (IDA, PMID:29240402); ALDH1A3 GO:0001758 and GO:0002138 (IDA, PMID:27759097); CYP26A1 GO:0008401 (IDA) and GO:0062183 (IDA, PMID:22020119); CYP26B1 and CYP26C1 GO:0008401 and GO:0034653 (IDA). Exceptions, used without experimental GOA support: ALDH1A1 GO:0001758 (ISS/IEA; experimental rows are the parent GO:0004029); CRABP1 and CRABP2 GO:0001972 (IBA/IEA; CRABP2's only experimental MF row is cyclin binding); TTR GO:0140313 (TAS and mouse GO-CAM); CES1 GO:0050253 (IBA). Possible upstream issue: the RDH10 GO:0004745 (NAD+) IDA row from PMID:12407145 cites a paper whose abstract reports a preference for NADP as cofactor, so the NAD+ term may need checking against the full text. Comparator check for processes: GO:0002138 is placed on RDH10 and the ALDH1A enzymes, which do the chemistry, and not on the binding proteins or STRA6; GO:0034653 only on the CYP26 enzymes. GO-CAM: gocams/index.tsv (checked 2026-11-05) has no human production model for any member. The seed's best_gocam lead is empty for PW:0001004. Mouse models are linked at activity level: Retinoid Metabolic Process 2 (gomodel:67086be200001721; Rbp4, Ttr), Retinoid Metabolic Process 1 (gomodel:66f5faaa00001204; Rbp1) and 4 (gomodel:67086be200002705; Dhrs3), and gomodel:5fce9b7300001742 for Cyp26a1, whose title is about Strap and splicing, so only the Cyp26a1 activity is relevant. These mouse models describe absorption, storage and mobilisation and contain no Rdh10, Aldh1a2/3 or Crabp activity, which is the main GO-CAM gap for this pathway. Gene reviews exist for DHRS3 and LRAT (and RDH11, not a member); the other members have none. PANTHER family ids and PTNs are deliberately not asserted. Most cited publications are abstract-only in the cache; PMID:21901792, PMID:24733397 (PMC), PMID:17473173 (PMC), PMID:29240402, PMID:27759097, PMID:17175089 and PMID:22020119 have full text or a PMC record.