AFP

UniProt ID: P02771
Organism: Homo sapiens
Review Status: COMPLETE
📝 Provide Detailed Feedback

Gene Description

Alpha-fetoprotein is the major plasma protein of the mammalian fetus and the closest paralogue of serum albumin. It is a 609-residue, single-chain, three-domain member of the ALB/AFP/VDB family, made with a cleaved 18-residue signal peptide, N-glycosylated at Asn251, disulfide-rich, phosphorylated on six serines by FAM20C in the endoplasmic reticulum lumen, and constitutively secreted. Synthesis is restricted to the visceral and yolk-sac endoderm and the fetal liver; concentrations peak around the 12th-16th week of gestation, fall to trace levels after birth, and rise again when hepatocytes or germ-cell tumours re-acquire the fetal expression programme. Its molecular function is reversible binding of small hydrophobic and metal ligands in the albumin fold. A cryo-EM structure of the human protein resolves four fatty acids occupying sites in subdomains IIA, IIA/IIB, IIIA and IIIB, with palmitate and stearate identified by GC-MS as the species actually bound, and a metal ion coordinated by His22, His264, His268 and Asp280. Direct binding assays on protein purified from human cord serum and hepatoma ascites show one Cu(II) per molecule, dependent on an amino-terminal histidine, and one bilirubin per molecule with sub-micromolar affinity. The ligand repertoire overlaps serum albumin's but is not identical to it: human alpha-fetoprotein binds copper, nickel, fatty acids and, less well than albumin, bilirubin, whereas high-affinity estrogen binding, which is a defining property of rodent alpha-fetoprotein and the basis of the female-infertility phenotype of Afp-null mice, is essentially absent from the human protein. Beyond ligand carriage the protein has no established physiological role. Humans homozygous for null AFP alleles are asymptomatic with normal fetal development and normal male reproduction, so the protein is dispensable in humans. In hepatocellular carcinoma an intracellular pool has been reported to engage HuR, HSP90 and PTEN and to modulate apoptotic and PI3K/AKT signalling, and the secreted protein binds the cell-surface proteoglycan glypican-3 through its core protein; these are cancer-cell observations rather than properties of the fetal plasma protein. Clinically the protein is used as a concentration readout in prenatal screening and in hepatocellular and germ-cell tumour management, which is a diagnostic application and not a biological activity.

Proposed New Ontology Terms

bilirubin binding

Definition: Binding to bilirubin, a linear tetrapyrrole produced by the reduction of biliverdin during the catabolism of heme, and transported in blood plasma bound to carrier proteins.

Justification: GO can express bilirubin transport as a process, GO:0015723, and bilirubin transmembrane transporter activity, GO:0015127, but it has no molecular function term for simply binding bilirubin. This is a real gap rather than an oversight about an obscure ligand. Plasma carriage of bilirubin by albumin-family proteins is textbook physiology, it underpins the clinical management of neonatal jaundice, and it is directly measured for alpha-fetoprotein: two independent methods, difference spectroscopy and the Jacobsen peroxidase assay, give 1:1 stoichiometry and dissociation constants between 2.6 and 7.4 x 10^-7 M on protein purified from human cord serum and from hepatoma ascites. Because no molecular function term exists, that measurement cannot be captured in GO at all, and UniProt's statement that alpha-fetoprotein binds bilirubin less well than serum albumin does is likewise unrepresentable. A parent of GO:0046906 tetrapyrrole binding is appropriate because bilirubin is an open-chain tetrapyrrole and the definition of GO:0046906 covers compounds containing four pyrrole nuclei linked through alpha-position carbons. The term would immediately serve serum albumin as well as alpha-fetoprotein, and plausibly the bilirubin-binding glutathione S-transferases and the substrates already modelled under GO:0015723.

Parent term: tetrapyrrole binding

Supporting Evidence:

Existing Annotations Review

GO Term Evidence Action Reason
GO:0031667 response to nutrient levels
IBA
GO_REF:0000033
REMOVE
Summary: A family-node inference whose entire experimental foundation fails on reading. Not one alpha-fetoprotein, from any species, is a seed on this row - the donors are serum albumin from human, rat and chicken, and vitamin D-binding protein from rat. Querying each donor's own evidence under this branch leaves exactly three experimental papers, and all three measure something other than the protein participating in a response to nutrient levels: a rice sucrose-starvation promoter driving recombinant albumin production, exogenous albumin at 29-49 mg/ml protecting cells from drug-induced ER calcium release, and albumin concentration used as a nutritional-status biomarker in rats. Everything else in the donor set is an ISS or ISO copy of the same two human ALB annotations.
Reason: Two independent lines carry this, with a third recorded as an observation rather than an argument. The observation first, so that it is not mistaken for load-bearing: the seed list contains no member of the alpha-fetoprotein subfamily, so no ortholog-strength inference is available on this row. That is worth noting but it is not itself a defect, because an IBA is a judgement about a node rather than a pairwise transfer, and AFP descends from PTN000147344 whichever members happened to seed it. The removal rests on the next two points. First, the donors' own evidence does not support the term even for the donors. Human ALB holds this term by the same IBA; its only experimental rows in the branch are GO:0009267 by IDA from PMID:16245148 and GO:0072732 by IDA from PMID:16153637. PMID:16245148 expressed human serum albumin from the rice alphaAmy3 sucrose-starvation-inducible promoter in rice suspension culture, so the starvation response is the plant promoter's and albumin is the recombinant cargo. PMID:16153637 supplemented serum-free medium with 29-49 mg/ml albumin and challenged cells with dotarizine, flunarizine, thapsigargin and cyclopiazonic acid, which release endoplasmic reticulum calcium pharmacologically; that is not a nutrient level. Rat Alb adds one IEP, PMID:20227002, in which serum albumin concentration falls under fetal protein malnutrition and recovers afterwards, which is albumin as a readout of nutritional status and precisely the concentration-is-not-a-function error. Chicken ALB contributes only ISS copies of the two human ALB rows plus an unrelated response-to-vitamin-A IDA. So the apparent breadth of the donor set is one pair of experiments counted repeatedly. Second, the target's own biology gives no independent support: two unrelated human families homozygous for null AFP alleles are asymptomatic with normal fetal development, which is difficult to reconcile with participation in a nutrient-response process. This is a node-level defect rather than an AFP-level one, and the recommendation to PAINT is stated once in suggested_questions rather than repeated here.
Propagation Review
Root cause: SOURCE WEAK OR INFERRED
Failure modes: SOURCE EVIDENCE WEAK ROLE CONFLATION
Sources checked:
PANTHER:PTN000147344 · PANTHER internal tree node, family root of PTHR11385, not a protein SOURCE WEAK OR INFERRED
The root node of PTHR11385. Its reach is the whole serum-albumin-related family, so this term is asserted over ALB, AFP, GC and AFM alike. Compare the three other PAINT nodes on this family, which each place a specific ligand term on a single subfamily; this row is the only one that places a process term at the root.
RGD:2085 · Alb (Rattus norvegicus), rat serum albumin SOURCE WEAK OR INFERRED
Resolved through the RGD REST API. Its only experimental row in this branch is GO:0007584 by IEP from PMID:20227002, where serum albumin concentration is a readout of nutritional status; its GO:0009267 and GO:0072732 rows are ISS and ISO copies of human ALB's.
RGD:2667 · Gc (Rattus norvegicus), vitamin D-binding protein SOURCE WEAK OR INFERRED
Resolved through the RGD REST API as rat Gc. A vitamin D carrier rather than an alpha-fetoprotein, and no experimental response-to-nutrient-levels evidence was found for it. QuickGO rejects RGD identifiers in geneProductId with HTTP 400, so its evidence was approached through the rat UniProt accession; reported as a resolution route rather than left silent.
UniProtKB:P02768 · ALB (Homo sapiens), serum albumin, Swiss-Prot, 609 aa SOURCE WEAK OR INFERRED
The pivotal donor. It does not hold this term experimentally - its own GO:0031667 is the same IBA. Its two supporting IDAs are PMID:16245148, a rice-promoter recombinant expression study, and PMID:16153637, an exogenous-albumin cytoprotection study. Every other donor's ISS or ISO row copies these two.
UniProtKB:P19121 · ALB (Gallus gallus), serum albumin, Swiss-Prot, 615 aa CIRCULAR OR REDUNDANT
Contributes no independent evidence. Its GO:0009267 and GO:0072732 rows are both ISS with human ALB P02768 as the with/from, so counting it as a separate donor double-counts the same two human experiments. Its one IDA in the branch, GO:0033189 response to vitamin A, is a different stimulus.
Supporting Evidence:
PMID:16245148
Mature form of HSA was expressed under the control of the sucrose starvation-inducible rice alpha Amy3 promoter
PMID:16153637
In serum-free medium, albumin (29 or 49 mg/ml) fully prevented the apoptotic effects of dotarizine, flunarizine and cyclopiazonic acid.
PMID:20227002
there was a restoration of serum glucose, total protein, and albumin concentrations, which were reduced by fetal malnutrition
PMID:15280901
The affected individuals were asymptomatic and presented normal development.
GO:0036094 small molecule binding
IBA
GO_REF:0000033
ACCEPT
Summary: Correct, and general for the right reason. The instinct to call a broad term a granularity failure does not survive looking at the donor set, which is genuinely heterogeneous: serum albumin holds fatty acid, pyridoxal phosphate, oxygen, enterobactin and copper binding; vitamin D-binding protein holds vitamin D binding; afamin holds vitamin E binding. Small molecule binding is the least common ancestor of that list, not a curator declining to be specific. PAINT's own design confirms the reading, because it put the general term at the family root and the specific ligand terms on the three subfamily nodes.
Reason: Checked against the rule that a general term is only a granularity defect when the donors agree. They do not. Resolving all eight non-node seeds gives serum albumin from four species, human vitamin D-binding protein, human afamin, rat albumin and rat vitamin D-binding protein, and querying their own annotations gives ligands as different as palmitate, pyridoxal phosphate, molecular oxygen, enterobactin, copper, vitamin D and vitamin E. GO:0036094 is the correct common parent of those, and QuickGO ancestry confirms that both GO:0005504 fatty acid binding and GO:0005507 copper ion binding are descendants of it, so the term does cover AFP's own measured ligands rather than merely sitting alongside them. The term is also true of AFP on the target's own evidence, independently of the phylogeny: UniProt records copper, nickel, fatty acid and bilirubin binding, with an experimentally evidenced Cu(2+) site at His22. The right response is therefore additive rather than corrective - accept the family-level term and supply AFP's specific measured ligands as new annotations, which is what this review does. Note also what the donor set does not contain: no alpha-fetoprotein from any species is a seed, so the inference is entirely from the paralogous subfamilies.
Propagation Review
Root cause: NO FAILURE CORE
Sources checked:
PANTHER:PTN000147344 · PANTHER internal tree node, family root of PTHR11385, not a protein SUPPORTS TRANSFER
The family root. Placing the least common ancestor of the family's ligand repertoire here, while the specific ligands sit on the subfamily nodes PTN000147369 for vitamin E in afamin, PTN002604574 for vitamin D in the vitamin D-binding proteins and PTN008319942 for enterobactin in the albumins, is coherent node design. The gap is that the alpha-fetoprotein subfamily node has no assignment of its own.
RGD:2085 · Alb (Rattus norvegicus), rat serum albumin SUPPORTS TRANSFER
Carries its own fatty acid binding IDA from PMID:16753933 and zinc ion binding IDA from PMID:540055, so it is a genuinely characterised ligand-binding donor rather than a by-similarity label.
RGD:2667 · Gc (Rattus norvegicus), vitamin D-binding protein SUPPORTS TRANSFER
A vitamin D carrier. Supports the general term and is one of the reasons the general term is the correct level, because its ligand is not AFP's.
UniProtKB:P02768 · ALB (Homo sapiens), serum albumin, Swiss-Prot, 609 aa SUPPORTS TRANSFER
The closest human paralogue and the strongest donor. Holds fatty acid binding by two independent IDAs, plus pyridoxal phosphate, oxygen and enterobactin binding by IDA. Its breadth of ligands is itself the argument for the general term.
UniProtKB:P02769 · ALB (Bos taurus), serum albumin, Swiss-Prot, 607 aa SUPPORTS TRANSFER
Bovine serum albumin, the most extensively characterised ligand-binding protein in the family, and co-seed with human ALB on the enterobactin binding node.
UniProtKB:P02774 · GC (Homo sapiens), vitamin D-binding protein, Swiss-Prot, 474 aa SUPPORTS TRANSFER
Human vitamin D-binding protein, and notably shorter than the other members at 474 aa. Its ligands are vitamin D and G-actin, neither of which AFP binds, which again argues for the parent term rather than a transferred specific one.
UniProtKB:P08835 · ALB (Sus scrofa), serum albumin, Swiss-Prot, 607 aa SUPPORTS TRANSFER
Porcine serum albumin, with the same ligand profile as the other albumins.
UniProtKB:P19121 · ALB (Gallus gallus), serum albumin, Swiss-Prot, 615 aa SUPPORTS TRANSFER
Chicken serum albumin. Its presence is informative for this molecular function row because it shows the ligand-binding property predates the mammalian radiation, unlike the metal-site conservation pattern that PMID:38678117 reports as stronger in viviparous animals.
UniProtKB:P43652 · AFM (Homo sapiens), afamin, Swiss-Prot, 599 aa SUPPORTS TRANSFER
Human afamin, a vitamin E carrier and a Wnt solubilising factor. A third distinct ligand class in the donor set, and the seed for the vitamin E binding term on its own subfamily node.
Supporting Evidence:
file:human/AFP/AFP-uniprot.txt
Belongs to the ALB/AFP/VDB family.
PMID:38678117
We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
GO:0005576 extracellular region
IEA
GO_REF:0000120
ACCEPT
Summary: Correct and core. Alpha-fetoprotein is a constitutively secreted plasma protein: it has a cleaved 18-residue signal peptide, no transmembrane segment, and it was originally purified and sequenced from human fetal cord serum, amniotic fluid and hepatoma ascites. This is the compartment in which every measured ligand-binding property was determined, and it is also what makes the cytosolic and nuclear partners reported in high-throughput screens topologically implausible.
Reason: The evidence route is combinatorial rather than ARBA, drawing on four InterPro albumin-family signatures, the UniProt SubCell term SL-0243 and the mouse orthologue, and all of those converge on the same conclusion that direct measurement gives. The mature chain is defined as residues 19 to 609 after a signal peptide confirmed by direct protein sequencing of plasma-derived protein, and the clinical assays that made this protein famous all measure it in serum or amniotic fluid. Unlike the cytoplasm row, this localisation does not depend on the mouse projection at all: dropping UniProtKB:P02772 from the evidence set would leave the call unchanged. Accepted as a core location.
Supporting Evidence:
file:human/AFP/AFP-uniprot.txt
SUBCELLULAR LOCATION: Secreted.
GO:0005515 protein binding
IPI
PMID:26808496
Comparative Proteomics Reveals Important Viral-Host Interact...
MODIFY
Summary: A genuine interaction with a viral protein, recorded with a term that hides what is most informative about it. The partner accession resolves to Envelope glycoprotein E2, residues 384 to 750 of the hepatitis C virus genotype 2a JFH-1 polyprotein, which is a protein of a different species. GO has a term for exactly this situation, and the closest paralogue ALB already carries it by IDA.
Reason: GO:0140272 exogenous protein binding is defined as binding to a protein or protein complex from a different species, for example a pathogen molecule binding to a host protein, which is a literal description of this row. Replacing GO:0005515 with it loses nothing and records the cross-species character that is the only generalisable thing the experiment shows. The precedent is not hypothetical: human serum albumin, the closest paralogue, holds GO:0140272 by IDA, so this brings AFP into line with how the family is already curated rather than inventing a treatment. The interaction itself is adequately supported but not strong. It was found in an affinity purification of E2 complexes from infected hepatoma cells and then confirmed by co-immunoprecipitation of Flag-tagged AFP with E2 in co-transfected 293T cells, that is with both partners over-expressed, and the paper's functional follow-up went to UGT1 rather than to AFP. So this is a real binding event with no measured consequence, which is why it is not proposed as a core function. Note also that AFP is secreted and E2 is a viral envelope glycoprotein that transits the same secretory pathway, so unlike most of AFP's high-throughput partners the two are topologically capable of meeting.
Proposed replacements: exogenous protein binding
Supporting Evidence:
PMID:26808496
were further validated as novel E2 binding partners
PMID:26808496
To this end, we subcloned AFP, UDP-glucose:glycoprotein glucosyltransferase 1 (UGT1), and Cdc2 in a Flag-tagged expression plasmid and co-transfected with a HCV E2 expression plasmid into 293T cells.
GO:0005515 protein binding
IPI
PMID:39822733
A proteomic atlas of glypican-3 interacting partners: Identi...
MODIFY
Summary: Glypican-3, and the one AFP interaction that survives every check applied to it. IntAct holds three records that are not sub-methods of a single screen - a cell-based co-immunoprecipitation in HepG2, an in vitro co-immunoprecipitation, and a pull-down typed as a direct interaction with host recorded as In vitro - at MI-score 0.59, the highest of any of AFP's twenty-one interaction records. Glypican-3 is a heparan sulfate proteoglycan, so the informative term is available and bare protein binding understates what is known.
Reason: Corrected after review. This row was first kept at bare GO:0005515 on the argument that because a glypican-3 mutant lacking heparan sulfate chains still precipitates AFP, the binding is to the core protein and a proteoglycan term would misdescribe it. That reasoning confused two different objects. GO:0043394 is defined as binding to a proteoglycan, that is to the glycoprotein itself, and GO has separate terms for binding the glycosaminoglycan moiety - GO:0008201 heparin binding and GO:1904399 heparan sulfate binding. Checked against QuickGO rather than inferred from the label. So the heparan-sulfate-free mutant result rules out a glycosaminoglycan-binding term while leaving GO:0043394 untouched: it shows the sugar chains are not required for the interaction, not that the partner is not a proteoglycan. Glypican-3 is a GPI-anchored heparan sulfate proteoglycan, and the cell-based co-immunoprecipitation and the membrane colocalisation were both done on the native, glycanated protein. The interaction itself is well supported and orthogonally validated: purified Fc-tagged AFP pulls down GPC3-His while failing to pull down MUC16-His as a negative control, which distinguishes this row sharply from the eight yeast two-hybrid partners in IntAct, all of them cytosolic or nuclear and so topologically inaccessible to a constitutively secreted protein. It remains non-core in substance - no downstream signalling, proliferation or migration consequence is measured anywhere in the paper, so no activity-level term is available - but MODIFY is the right action because a strictly more informative term does exist and the earlier objection to it does not hold.
Proposed replacements: proteoglycan binding
Supporting Evidence:
PMID:39822733
The results show that AFP pulls down GPC3, but not MUC16, a control protein
PMID:39822733
suggesting that the GPC3 core protein might bind AFP
PMID:39822733
Our results indicate that the core protein of GPC3 is involved in binding with AFP.
Knowledge gap:
The AFP literature invokes a cell-surface AFP receptor, AFPR, repeatedly, but its molecular identity has never been established. OPEN BIOLOGY
GO:0005737 cytoplasm
IEA
GO_REF:0000107
KEEP AS NON CORE
Summary: The right term reached by a route that does not support it. GOA records this as an Ensembl Compara projection from mouse Afp, whose entire basis is a single MGI IDA from PMID:8607965, a paper whose own abstract reports that alpha-fetoprotein protein was not detected by immunostaining in the systems studied, with only serum albumin visible and AFP present as mRNA. The term nonetheless survives, on human evidence that GOA does not cite: confocal imaging shows AFP colocalising with HuR in the cytoplasm of HuH7 and HepG2 hepatoma cells.
Reason: Two separate judgements, kept apart deliberately. On provenance, this is a one-hop projection resting on one mouse annotation, and the cited paper's abstract states that AFP was not detected during the period examined while serum albumin was, with the intracellular-AFP claim appearing as a background citation to earlier work rather than as this study's result. Only the abstract is cached, so this is recorded as support that cannot be confirmed from the available record rather than as a curator error, and the mouse IDA itself is not second-guessed. On the term, an intracellular pool of AFP in AFP-producing hepatoma cells is independently supported in human cells by direct imaging, so removing the term would discard something true. It is non-core rather than core for two reasons: the physiological protein is a constitutively secreted plasma protein with a cleaved signal peptide, and the cytoplasmic pool is described specifically in hepatoma lines, which is a disease context rather than the fetal setting in which this gene does its work. Kept, downgraded, and with the recorded provenance flagged so that the reason it is kept is not confused with the reason GOA gives.
Propagation Review
Root cause: SOURCE WEAK OR INFERRED
Failure modes: SOURCE EVIDENCE WEAK CONTEXT OR TISSUE MISMATCH
Sources checked:
UniProtKB:P02772 · Afp (Mus musculus), alpha-fetoprotein, Swiss-Prot, 605 aa SOURCE WEAK OR INFERRED
The true orthologue, so the transfer is not a paralogue error. But its GO:0005737 rests on exactly one IDA, PMID:8607965, whose abstract reports that AFP protein was not detected by immunostaining in the embryos or teratocarcinomas examined, only serum albumin. This same donor also carries five IMP process rows from the Afp knockout, none of which has been propagated to human AFP, correctly, since they depend on estrogen binding that the human protein lacks.
ensembl:ENSMUSP00000041006 · Ensembl mouse protein for Afp, the Compara projection vehicle CIRCULAR OR REDUNDANT
The Ensembl protein identifier for the same mouse Afp entry as UniProtKB:P02772, listed alongside it. It is the mechanism of the projection rather than a second, independent source, so the two-token with/from field represents one donor, not two.
Supporting Evidence:
PMID:33009373
partial colocalization of AFP and HuR was observed in the cytoplasm of HuH7 and HepG2 cells
PMID:8607965
Only SA protein was detectable by immunostaining.
GO:0005788 endoplasmic reticulum lumen
TAS
Reactome:R-HSA-8952289
KEEP AS NON CORE
Summary: Correct but incidental. The Reactome reaction behind this row is FAM20C phosphorylates FAM20C substrates, which is annotated to the endoplasmic reticulum lumen compartment, and AFP is on the substrate list with real backing, because UniProt records six FAM20C phosphoserines on AFP at Ser-111, Ser-115, Ser-117, Ser-344, Ser-444 and Ser-445, each carrying ECO:0000269 to PMID:26091039. So the protein demonstrably is in the ER lumen. It is there because every secreted protein passes through it.
Reason: Retained because it is true and experimentally underpinned, and marked non-core because it is a biosynthetic transit compartment rather than a site of action. Alpha-fetoprotein has a cleaved signal peptide and is constitutively secreted; the ER lumen is where it is folded, disulfide-bonded, glycosylated at Asn251 and phosphorylated by FAM20C, not where it binds its ligands or encounters glypican-3. Distinguishing the two is the point of the non-core action here. No modification is proposed, because the term is already at the right granularity for what it describes, and a TAS to a Reactome reaction whose substrate assignment traces to an ECO:0000269-backed phosphosite set is a sounder provenance than most electronic routes on this gene.
GO:0005515 protein binding
IPI
PMID:26808496
Comparative Proteomics Reveals Important Viral-Host Interact...
MODIFY
Summary: The same AFP and HCV E2 interaction as the row above, from the same publication and the same partner accession, contributed independently by AgBase rather than IntAct. It is a second record of one experiment, not a second experiment, and it resolves the same way. It is restored here as its own entry because the fetch-gene stub had collapsed the two assigner rows into one, which would have hidden the redundancy rather than documenting it.
Reason: Same reasoning and same replacement as the IntAct row. The partner is Envelope glycoprotein E2 of hepatitis C virus, a protein of a different species, so GO:0140272 exogenous protein binding is the precise term and GO:0005515 understates what is known. The only thing this row adds is a reminder about how evidence gets counted. Two assigner records of a single co-immunoprecipitation look like two independent supports in any tally that counts GOA rows, which is the same shape as an interaction database logging one screen under three sub-methods. Recording that explicitly is the reason to keep the row separate rather than merge it away, and it is why the count of supporting experiments for this interaction is one, not two.
Proposed replacements: exogenous protein binding
Supporting Evidence:
PMID:26808496
were further validated as novel E2 binding partners
GO:0005504 fatty acid binding
IDA
PMID:38678117
Structural characteristics of alpha-fetoprotein, including N...
NEW
Summary: Proposed. The cryo-EM structure of human alpha-fetoprotein resolves four fatty acids bound in subdomains IIA, IIA/IIB, IIIA and IIIB, and the bound species were identified chemically by GC-MS as palmitate, at 57.42 per cent of total fatty acid, and stearate. PDB entry 8X1N deposits palmitic acid as a ligand of the human protein. UniProt independently states that the protein binds fatty acids as well as serum albumin does. GOA carries no fatty acid term for AFP, and PAINT places GO:0005504 at no node of PTHR11385 despite serum albumin holding it by two separate IDAs.
Reason: This is the best-evidenced molecular function of the protein and the clearest gap in its GO record. Three things make it an IDA rather than an inference from the albumin fold. First, the lipid was not modelled from density shape alone, because gas chromatography-mass spectrometry on fatty acids extracted from the purified protein identified which species were present, so the chemical identity is measured rather than assumed. Second, the fatty acids were carried by the protein through purification rather than soaked in, so the sites are occupied under ordinary conditions. Third, the sites are located in named subdomains of this protein's own structure rather than transferred from an albumin structure, and the authors superpose AFP on the HSA-palmitate complex separately. The honest caveat is that the protein was recombinant, expressed in HEK293F cells rather than purified from fetal serum, so the fatty acids are those available in that medium; this bears on which fatty acid predominates, not on whether the protein binds fatty acids. GO:0005504 is a descendant of the GO:0036094 already annotated by IBA, so this refines an existing row rather than contradicting it.
Supporting Evidence:
PMID:38678117
We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
PMID:38678117
it was identified that palmitic acid (C16:0) was the most abundant FA bound to AFP
PMID:38678117
These FA binding sites are located in AFP substructures IIA, IIA/IIB, IIIA, and IIIB, respectively
GO:0005507 copper ion binding
IDA
PMID:80265
Copper(II)-binding ability of human alpha-fetoprotein.
NEW
Summary: Proposed. Alpha-fetoprotein purified from human umbilical cord serum and from the ascites of a hepatoma patient binds one copper(II) ion per molecule above pH 6, measured by equilibrium dialysis and gel filtration. The site was localised by two independent chemical arguments: it is lost when histidyl residues are destroyed by methylene-blue photooxidation, and it is reconstituted by a synthetic undecapeptide corresponding to the protein's amino terminus. UniProt already encodes the conclusion as a Cu(2+) binding feature at residue 22 with ECO:0000269 evidence to this paper. GOA has no copper term.
Reason: This is a measurement on the human protein, from human tissue, with the binding site mapped, rather than a property inherited from serum albumin. It is also the clearest instance of the coverage problem on this gene, because UniProt has carried an experimentally evidenced copper-binding residue for decades and no GO annotation reflects it. The structural work corroborates it independently and from a different direction: PMID:38678117 resolves a metal site coordinated by His22, His264, His268 and Asp280, and His22 is the same residue UniProt annotates as the Cu(2+) ligand, so a 1978 solution measurement and a 2024 structure converge on one position. The pH dependence reported in PMID:80265, with binding falling to 0.5 mol per mol at pH 5.4, close to the pKa of the imidazole group, is itself consistent with histidine coordination. GO:0005507 is a descendant of the GO:0036094 already annotated by IBA, so this specifies an existing row. This row is deliberately the specific one of a parent/child pair, and the two are more nearly independent than first stated. PMID:80265 reconstitutes copper binding with a synthetic amino-terminal undecapeptide: eleven residues, which on a mature chain beginning at position 19 spans roughly 19-29. That peptide contains His22 but cannot contain His264, His268 or Asp280, so copper binding does not require the other three residues of the tetrahedral site resolved in the structure. PMID:38678117 points the same way from its own data, noting molar ratios above 1 and inferring metal binding at multiple locations. What is established is therefore separability and a shared His22, not that the two sites are disjoint in the full-length protein - sufficiency of a peptide does not establish that. On that reading the companion GO:0046872 row is not a generalisation of this one at all: it records a distinct, multi-metal site, which is a stronger reason to keep both than the different-experiments argument it replaces.
Supporting Evidence:
PMID:80265
Alpha-fetoprotein bound 1 mol of copper(II) ion per mol of protein above pH 6.0
PMID:80265
Photooxidation of alpha-fetoprotein in the presence of methylene blue resulted in the loss of the copper(II)-binding ability of the protein in parallel with the destruction of the histidyl residues.
PMID:80265
A synthetic amino-terminal undecapeptide of alpha-fetoprotein also bound copper(II) ion.
PMID:38678117
We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
PMID:38678117
Moreover, the molar ratio of some metal elements with AFP exceeds 1, it suggests the potential for AFP to bind to metal ions at multiple locations.
GO:0046872 metal ion binding
IDA
PMID:38678117
Structural characteristics of alpha-fetoprotein, including N...
NEW
Summary: Proposed, at the generality the measurement supports. The cryo-EM structure of human alpha-fetoprotein resolves a tetrahedral metal site in domain IA coordinated by His22, His264, His268 and Asp280, and the authors tested it directly: they mutated all four residues to alanine and measured metal content of wild-type and mutant protein by ICP-MS. The mutant lost magnesium, aluminium, nickel and zinc alike, and they conclude the four residues can bind different types of metal ion. The metal was carried from the culture medium rather than added, so the site is occupied under ordinary conditions.
Reason: This row was first proposed as GO:0008270 zinc ion binding and has been generalised after review, because the paper's own wet-lab metal experiment argues against zinc specificity rather than merely failing to establish it. Zinc entered the model as a modelling choice - the authors write that they selected Zn2+ to fit into the density map, having reasoned from a similar motif in Cu,Zn-superoxide dismutase - and the density is explicitly described as an average that may cover more than one metal species. Against that, the ICP-MS assay on the H22A/H264A/H268A/D280A mutant is a direct measurement on the human protein, and it shows the site losing four chemically dissimilar metals at once. Weighing a deposited ligand identity above the paper's own measurement would be the wrong way round. GO:0046872 metal ion binding is therefore the term the evidence supports, and it is still a descendant of the GO:0036094 already annotated by IBA, so the row remains a refinement rather than a restatement. Note that this does not weaken the companion copper row: GO:0005507 rests on a different and independent line of evidence, an equilibrium-dialysis stoichiometry on protein purified from human tissue with the site mapped chemically to histidine, which is exactly the kind of measurement the structure paper does not provide for zinc. Specific where a specific metal was measured, general where only a metal was. The ICP-MS result also supplies the first independent evidence on the human protein for UniProt's otherwise unsourced nickel claim. Two things a curator will reasonably ask about this pair. First, GO:0005507 copper ion binding is a descendant of GO:0046872, so why both: because they rest on different experiments making different claims, and neither subsumes the other in evidence. The copper row asserts a measured 1:1 stoichiometry with a chemically mapped site, from PMID:80265; this row asserts something the copper row cannot, namely that a four-residue site spanning His22, His264, His268 and Asp280 engages several chemically dissimilar metals, which is what the AFP-4mut ICP-MS actually shows. The two are not simply a general and a specific statement about one site: copper binding is reconstituted by an amino-terminal undecapeptide that cannot contain three of those four residues, and this paper infers metal binding at multiple locations from molar ratios above 1, so the sites are at least separable and share only His22. Dropping the general term would discard the promiscuity result, and dropping the specific one would discard the only quantitative metal measurement on this protein. Second, nickel is among the metals lost in the mutant, so why is GO:0016151 nickel cation binding not proposed alongside: because the ICP-MS gives a relative reduction in molar ratio in a supplementary panel, not a stoichiometry or an affinity, and that is a weaker basis than the copper work by some distance. GO:0046872 already covers nickel, so the evidence is represented without asserting a specificity the data do not carry. The same reasoning would extend to magnesium and aluminium, which is itself a reason to stop at the parent.
Supporting Evidence:
PMID:38678117
The findings indicated a reduction in the molar ratio of metal to protein in the mutant AFP-4mut, specifically for elements such as magnesium, aluminum, nickel and zinc
PMID:38678117
This implies that the identified four amino acids in AFP possess the capacity to bind to different types of metal ions
PMID:38678117
After we selected Zn2+ fit into the density map
PMID:38678117
This could be due to the presence of multiple metal ions at this binding site, and the metal ion density being obtained through an averaging algorithm
PMID:38678117
We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
PMID:38678117
the presence of metal ions was not intentionally introduced during the protein extraction process
PMID:38678117
Moreover, the molar ratio of some metal elements with AFP exceeds 1, it suggests the potential for AFP to bind to metal ions at multiple locations.
PMID:80265
A synthetic amino-terminal undecapeptide of alpha-fetoprotein also bound copper(II) ion.

Core Functions

Extracellular carriage of long-chain fatty acids in fetal plasma. Alpha-fetoprotein is the fetal counterpart of serum albumin and binds fatty acids at four sites distributed across subdomains IIA, IIA/IIB, IIIA and IIIB of the three-domain albumin fold. The identity of the bound species is measured rather than assumed, because gas chromatography-mass spectrometry of lipid extracted from the purified human protein identifies palmitate as the dominant species at 57.42 per cent of total fatty acid, with stearate the other major peak, and the deposited structure 8X1N contains palmitic acid. The ligands are retained through purification rather than soaked in, so the sites are occupied under ordinary conditions. Because the protein is the dominant plasma protein of the fetus at a time when serum albumin has not yet taken over, this binding activity is the best-supported molecular function the gene has, and it is the one the whole albumin family is built around.

Molecular Function:
fatty acid binding
Cellular Locations:
Supporting Evidence:
  • PMID:38678117
    We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
  • PMID:38678117
    it was identified that palmitic acid (C16:0) was the most abundant FA bound to AFP
  • file:human/AFP/AFP-uniprot.txt
    Belongs to the ALB/AFP/VDB family.

Extracellular binding of transition metal ions through an amino-terminal histidine site. Alpha-fetoprotein purified from human cord serum and from hepatoma ascites binds one copper(II) ion per molecule above pH 6, with binding falling to half a mole per mole near the pKa of the imidazole group, abolished by photooxidative destruction of histidyl residues, and reproduced by a synthetic undecapeptide from the protein's amino terminus. The structure of the human protein places a tetrahedral metal site in domain IA formed by His22, His264, His268 and Asp280, with His22 the same residue UniProt annotates as the copper ligand, so a solution measurement from 1978 and a 2024 cryo-EM model identify the same position. The site differs from serum albumin's, which substitutes a water molecule for one of the histidines, and this is the structural basis for the higher zinc affinity reported for alpha-fetoprotein. The identity of the metal is not settled and the paper's own measurement says it is promiscuous: mutating all four residues to alanine reduces bound magnesium, aluminium, nickel and zinc together by ICP-MS, and zinc entered the structural model only because the authors selected it to fit an averaged density. Copper is named as the molecular function here because copper is the one metal for which a stoichiometry and a site have actually been measured on the human protein; the accompanying GO:0046872 annotation carries the general capacity the structure supports. The copper site and the tetrahedral site are separable rather than demonstrably identical: the amino-terminal undecapeptide that reconstitutes copper binding cannot contain His264, His268 or Asp280, and the structure paper infers metal binding at multiple locations from molar ratios above 1. His22 is common to both.

Molecular Function:
copper ion binding
Cellular Locations:
Supporting Evidence:
  • PMID:80265
    Alpha-fetoprotein bound 1 mol of copper(II) ion per mol of protein above pH 6.0
  • PMID:80265
    A synthetic amino-terminal undecapeptide of alpha-fetoprotein also bound copper(II) ion.
  • PMID:38678117
    We observed and identified certain structural features of AFP, including N-glycosylation at Asn251, four natural fatty acids bound to distinct domains, and the coordination of metal ions by residues His22, His264, His268, and Asp280.
  • PMID:38678117
    The findings indicated a reduction in the molar ratio of metal to protein in the mutant AFP-4mut, specifically for elements such as magnesium, aluminum, nickel and zinc
  • PMID:38678117
    Previous studies have demonstrated that AFP exhibits a higher affinity for Zn2+ than HSA

References

Loading supporting content…

Download this section (compressed HTML)

Suggested Questions for Experts

Q: For PAINT, a node-placement question about PTHR11385 as a whole rather than about AFP alone. The family has seven IBD assignments. Three subfamily nodes each carry the specific ligand of their subfamily - vitamin E on PTN000147369 for afamin, vitamin D on PTN002604574 for the vitamin D-binding proteins, enterobactin on PTN008319942 for the albumins - while the family root PTN000147344 carries the general parent GO:0036094. That design is sound. Two things follow from it that are worth examining. First, PTHR11385:SF7, the alpha-fetoprotein subfamily, has no assignment at all, so AFP is the only one of the four human members that inherits no ligand term; a subfamily node exists and is empty. Second, GO:0005504 fatty acid binding sits at no node in the family, although serum albumin holds it by two independent IDAs, rat albumin by IDA, mouse Afp by UniProt function statement, and human AFP by the structural work cited here. Fatty acid binding looks like the property that most clearly transfers family-wide, and it is the one property that PAINT has not placed anywhere.

Q: Also for PAINT, and affecting every member of PTHR11385 rather than AFP specifically: is GO:0031667 response to nutrient levels sustainable at PTN000147344? Tracing the donors' own evidence leaves three experimental papers. One expressed recombinant human serum albumin from a rice sucrose-starvation-inducible promoter, where the starvation response belongs to the plant promoter. One added albumin to serum-free medium at 29-49 mg/ml and challenged cells with drugs that release endoplasmic reticulum calcium. One measured serum albumin concentration as a readout of nutritional status in rats. The remaining donor rows are ISS or ISO copies of the first two. This review removes the term from AFP, but the same argument applies to the node and therefore to ALB, GC and AFM as well.

Q: Should GOA import the human AFP ligand measurements at all? Two of them, copper binding from PMID:80265 and bilirubin binding from PMID:89900, are from 1978 and 1979 and have been sitting in UniProt's function line and feature table since. The copper site is encoded as a binding feature at residue 22 with ECO:0000269 evidence, which is a curated experimental assertion, yet there is no corresponding GO annotation. Is the absence a deliberate age or quality judgement, or simply a gap? If it is a judgement, the 2024 structure independently locating a metal site on the same His22 is new grounds to revisit it.

Q: What is the AFP receptor? The literature invokes a cell-surface AFPR to explain receptor-mediated uptake of AFP by embryonic, hepatoma and gastric tumour cells, and the AFP drug-delivery work depends on it, but no molecular identity has been established. Glypican-3 is now a validated direct binding partner of secreted AFP at the cell membrane, verified with purified proteins against a negative control. Is glypican-3 the receptor, one of several, or an abundant co-purifying surface partner with no uptake role?

Q: A UniProt question rather than a GO one, now partly answered. The function line states that the protein binds nickel, but the feature table records only the Cu(2+) site and no reference in the entry is flagged for nickel the way PubMed:80265 is flagged for copper and PubMed:89900 for bilirubin. PMID:38678117 supplies the first independent measurement bearing on it: its ICP-MS assay on the H22A/H264A/H268A/D280A mutant shows nickel among the metals lost along with magnesium, aluminium and zinc, so a metal site on the human protein does engage nickel. Which site is not resolved by that experiment: AFP-4mut removes His22 along with His264, His268 and Asp280, and His22 is shared between the amino-terminal site that a synthetic undecapeptide reconstitutes for Cu(II) and the tetrahedral site the structure resolves, so the result is silent on which of the two binds nickel. That does not tell us where UniProt's assertion came from, and it is worth asking whether the original basis was a measurement on alpha-fetoprotein or a transfer from serum albumin's amino-terminal metal site - but the claim itself now has support that UniProt could cite.

Q: How should GO represent a protein that is clinically important purely as a concentration? Nothing in the ontology distinguishes what alpha-fetoprotein does from what its level indicates, and the distinction is not merely philosophical here, because it has already produced a defective annotation. The rat albumin IEP that helps underpin this family's GO:0031667 is an experiment in which albumin concentration falls with malnutrition and recovers with refeeding, that is a biomarker measurement that has become a process annotation. The same confusion is waiting in AFP's own much larger biomarker literature.

Suggested Experiments

Experiment: Purify alpha-fetoprotein from human fetal cord serum and determine its bound fatty acid composition by GC-MS, alongside serum albumin from the same donors. The existing composition comes from recombinant protein grown in HEK293F medium, so it reports what that medium supplied. This would establish which fatty acids the protein actually carries in the fetal circulation, and in particular whether docosahexaenoic acid, the species most often invoked in the placental transfer argument, is present at a meaningful occupancy.

Hypothesis: The fatty acids bound by alpha-fetoprotein in vivo differ from those it acquires in HEK293F culture medium, and include docosahexaenoic acid.

Type: Lipid extraction and gas chromatography-mass spectrometry on protein purified from native human fetal serum

Experiment: Test the carrier half of the activity directly. Load purified human alpha-fetoprotein with a labelled fatty acid and measure transfer to acceptor membranes or to cells, with serum albumin as a positive control and a ligand-site mutant as a negative. Binding is well established and delivery is not, which is the single reason GO:0140104 molecular carrier activity is not proposed in this review. A positive result would justify that term and a transport process term; a negative result would be equally informative, because it would mean the fetal protein is a reservoir rather than a shuttle.

Hypothesis: Alpha-fetoprotein does not merely bind fatty acids but releases them to an acceptor, which would license GO:0140104 molecular carrier activity.

Type: Ligand transfer assay from purified loaded protein to acceptor membranes or cells, with serum albumin as positive control and a site mutant as negative

Experiment: Settle the number of metal sites and the preference of each by isothermal titration calorimetry on the purified human protein across Cu(II), Zn(II) and Ni(II), comparing wild-type against a His264A/His268A/Asp280A triple mutant that leaves His22 intact. The published AFP-4mut cannot do this job, because it also carries H22A and so disables part of the amino-terminal site; leaving His22 in place is what makes the two sites separable experimentally. The 1978 equilibrium dialysis work measured copper and localised it to an amino-terminal undecapeptide; the 2024 structure models zinc into an averaged density its authors say may cover several species, and its ICP-MS assay shows magnesium, aluminium, nickel and zinc all falling together; and UniProt's function line adds nickel without a flagged measurement. One titration series against the right mutant resolves all of that, and would also test whether the additional histidine relative to serum albumin really accounts for the higher zinc affinity attributed to alpha-fetoprotein.

Hypothesis: Alpha-fetoprotein has two separable metal sites sharing His22 - an amino-terminal site sufficient on its own for Cu(II), and a tetrahedral His22/His264/His268/Asp280 site that binds several metals - and each has its own preference order across Cu(II), Zn(II) and Ni(II).

Type: Isothermal titration calorimetry across three metals, wild-type versus a His264A/His268A/Asp280A triple mutant retaining His22

Experiment: Ask whether glypican-3 mediates uptake of alpha-fetoprotein. Compare internalisation of labelled AFP by hepatoma cells with and without GPC3 knockout, using the heparan-sulfate-free glypican-3 mutant to confirm that the core protein is sufficient, as the binding data indicate. This converts a validated physical interaction into a functional claim or refutes it, and it bears directly on whether glypican-3 is the long-postulated AFP receptor.

Hypothesis: Glypican-3 is a receptor for secreted alpha-fetoprotein, and its core protein is sufficient for uptake.

Type: Labelled-ligand internalisation in GPC3-knockout versus wild-type hepatoma cells, rescued with full-length or heparan-sulfate-free glypican-3

Experiment: Measure estrogen binding by human and mouse alpha-fetoprotein side by side under identical conditions, using protein purified from the respective fetal sera. The species difference is asserted throughout the literature and underlies the judgement that the mouse knockout phenotype must not be transferred to the human gene, but the quantitative comparison is old and method-heterogeneous. Since this asymmetry is what licenses excluding the mouse reproductive annotations from human AFP, it deserves a modern direct measurement.

Hypothesis: Human alpha-fetoprotein binds estrogens far more weakly than mouse alpha-fetoprotein, which is what excludes the mouse reproductive phenotype from the human gene.

Type: Side-by-side equilibrium binding measurement on protein purified from human and mouse fetal serum under identical conditions

Knowledge Gaps

What is not known — curated, literature-grounded statements of the open unknowns (the inverse of core functions).

Gap: This gene's GO problem is coverage rather than over-annotation, and the review should be read that way.

NARROWING CURATION MF_DARK

Significance: Human AFP's entire GO record is eight annotations over six distinct terms, with no experimental molecular function beyond three protein binding IPI rows and no experimental biological process row at all. Set against that: two solution-binding studies on protein purified from human tissue, a cryo-EM structure resolving four fatty acids and a metal site, six FAM20C phosphosites with ECO:0000269 evidence, and two human null-allele reports. None of those produced a GO annotation. Exactly one existing row, GO:0031667, is an over-annotation.

What would resolve it: Three new molecular function rows are proposed here from measurements that already exist, and a missing bilirubin binding term is proposed to GO.

Gap: What alpha-fetoprotein is for in humans is unknown, and the usual answer given is a rodent answer.

OPEN BIOLOGY BP_DARK

Significance: In mouse, Afp-null females are infertile and the phenotype is attributed to estrogen binding; human alpha-fetoprotein has essentially no estrogen-binding capacity, and humans homozygous for null alleles are asymptomatic with normal development and normal male reproduction. The human null reports say nothing about female reproduction, which is the one place the mouse phenotype would predict a defect, so the strongest human evidence has a gap precisely where it would be most informative.

Gap: Whether the intracellular alpha-fetoprotein described in hepatocellular carcinoma is a distinct functional pool or a by-product of very high biosynthetic flux is unresolved.

OPEN BIOLOGY CC_DARK

Significance: A secreted protein at extreme expression levels will be abundant in the secretory pathway and detectable in cytoplasmic fractions, so colocalisation alone does not establish a cytosolic pool with access to cytosolic partners such as HuR, HSP90 and PTEN. This matters because most of the mechanistic AFP literature depends on that pool being real and topologically cytosolic, and because it is the reason the GO:0005737 row here is kept as non-core rather than accepted.

Gap: The functional significance of alpha-fetoprotein glycosylation is unresolved even though it carries the clinical assay.

OPEN BIOLOGY

Significance: AFP-L3, the core-fucosylated glycoform, is a distinct approved tumour marker, and tumour-derived and cord-derived protein differ in glycan composition, but whether glycosylation alters ligand binding, receptor recognition or the reported immunosuppressive activity has not been established.

Deep Research

Affinage

(AFP-deep-research-affinage.md)

Loading supporting content…

Download this section (compressed HTML)

📚 Additional Documentation

Notes

(AFP-notes.md)

Loading supporting content…

Download this section (compressed HTML)

Bioinformatics Results

(RESULTS.md)

Loading supporting content…

Download this section (compressed HTML)

📄 View Raw YAML

Loading supporting content…

Download this section (compressed HTML)