AIGR TreeGrafter Function-Inference Stress Test — Final Report

Target gene: eryCIV (UniProt A4F7N3) · Saccharopolyspora erythraea NRRL 2338 (NCBITaxon:405948)
Seed hypothesis under test: eryCIV has transaminase activity (GO:0008483) — propagated by TreeGrafter/PANTHER, evidence type IEA, reference GO_REF:0000118
Source selector: existing_annotations[2].function_hypothesis in genes/SACEN/eryCIV/eryCIV-ai-review.yaml

Summary

The TreeGrafter/PANTHER graft assigned transaminase activity (GO:0008483) to eryCIV. My independent analysis finds this assignment partially supported but too general: eryCIV is a genuine pyridoxal-5′-phosphate (PLP)-dependent aminotransferase, but the propagated term stops at the broad family level when the protein in fact belongs to a well-defined subfamily carrying out a specific sugar C4-transamination reaction. Of the three characteristic phylogenetic-propagation failure modes the objective asks me to test, only Failure Mode #1 (granularity / family-vs-subfamily) applies. Failure Mode #2 (pseudo-enzyme / loss of catalysis) and Failure Mode #3 (within-superfamily mis-placement onto a functionally distinct neighbor) are both affirmatively ruled out.

The decisive evidence is active-site conservation. The DegT/DnrJ/EryC1 fold (Pfam PF01041) supports two divergent chemistries — C4 transamination versus C3,4-dehydration — within a single structural scaffold, making a wrong-subfamily graft a live risk that a tree graft cannot self-diagnose. I therefore directly tested the catalytic apparatus. EryCIV retains the catalytic PLP Schiff-base lysine (Lys200) in register with the catalytic lysines of the structurally characterized transaminases ArnB (Lys188) and WecE (Lys181), whereas the family members that lost transamination and became dehydratases (ColD, AscC) carry a histidine at that exact position. EryCIV has lysine, not histidine. It is 60.5% identical to DesI, an X-ray-characterized dTDP-sugar C4 aminotransferase, far above any other characterized member. The AlphaFold model (mean pLDDT 96.5) independently places Lys200 and the PLP-anchoring Asp171 in a compact, well-ordered pocket with geometry consistent with a bridging PLP cofactor.

The practical outcome for curation is retain-and-refine: keep the transaminase assignment but make it more specific to a dTDP-4-amino-4,6-dideoxy-D-glucose (C4) sugar aminotransferase in desosamine biosynthesis — the direct ortholog of DesI. The legacy "3,4-dehydratase" database name that still clings to eryCIV is a pre-structural mis-label, contradicted by the active-site analysis, and should not be propagated.

Executive Judgment

Verdict: PARTIALLY SUPPORTED / TOO GENERAL.

• Failure mode that applies: #1 (granularity / family-vs-subfamily). The propagated term is correct in kind but under-granular.
• Failure modes ruled out: #2 (pseudo-enzyme) — all catalytic and cofactor-anchoring residues are intact and well-ordered; #3 (within-superfamily mis-placement) — EryCIV is on the transaminase side of the DegT/DnrJ/EryC1 split, not a ColD-type PLP dehydratase.
• Single most decisive piece of evidence: EryCIV retains the catalytic PLP Schiff-base Lys200 (motif SFHA-T-K-AVTS), aligned residue-for-residue to the catalytic lysine of ArnB (Lys188) and WecE (Lys181). The non-transaminase member ColD carries His at this exact position — its hallmark loss-of-transamination substitution — and EryCIV does not. Combined with 60.5% identity to the structurally characterized C4 transaminase DesI, this fixes EryCIV as a genuine transaminase.

Independent Family / Function Assignment

Independent of the propagated PANTHER term, my own sequence and structural analysis assigns:

dTDP-4-keto-6-deoxy-D-glucose → dTDP-4-amino-4,6-dideoxy-D-glucose, a C4 sugar aminotransferase in dTDP-D-desosamine biosynthesis.

• Characterized homolog this is based on: DesI (Streptomyces venezuelae, UniProt Q9ZGH0), the X-ray-characterized PLP-dependent C4 aminotransferase of the methymycin/pikromycin desosamine pathway, solved in complex with its product dTDP-4-amino-4,6-dideoxyglucose (PMID: 17630700). EryCIV is 60.5% identical to DesI over ~400 aligned columns — by far its nearest characterized neighbor. Erythromycin and methymycin/pikromycin share identical D-desosamine machinery, so eryCIV is the erythromycin-pathway ortholog of desI.
• Domain architecture: Pfam PF01041 (DegT_DnrJ_EryC1), InterPro IPR030967 (PLP_DesI); single-domain fold-type I PLP enzyme, 401 aa (SACE_0716).
• GO term I would assign: a more specific child of GO:0008483 on the same branch — functionally a "dTDP-4-amino-4,6-dideoxy-D-glucose transaminase" / sugar aminotransferase (EC 2.6.1.–). Granularity relative to the seed term: more specific (same branch) — not a sibling and not a different branch. Also well-supported: GO:0030170 (pyridoxal phosphate binding), already present, and a desosamine/dTDP-sugar biosynthetic process term.

Key Findings

Finding 1 — The DegT/DnrJ/EryC1 transaminase active site is fully conserved; EryCIV is not a ColD-type dehydratase

The DegT/DnrJ/EryC1 family (PF01041) is a textbook case where one PLP-binding fold supports two divergent chemistries: C4 transamination (DesI, ArnB, WecE, EryCI/DesV type) and C3,4-dehydration (ColD, AscC). Because these diverged within a single fold, a graft could land eryCIV on the wrong side — the exact shape of Failure Mode #3 — and the tree alone cannot tell. I tested this directly with a Needleman–Wunsch global alignment (BLOSUM62) of EryCIV against characterized members, anchored on literature catalytic residues.

The two diagnostic PLP-handling residues are present and in register. The catalytic Schiff-base lysine, EryCIV Lys200 (motif SFHA-T-K-AVTS), aligns exactly to ArnB Lys188 (SFHA-I-K-NITC) and WecE Lys181 (SFHE-T-K-NYTA) — the lysine that forms the internal aldimine with PLP and is absolutely required for transamination. Crucially, in the dehydratase members that lost transaminase activity this position is occupied instead by histidine: cross-alignment confirms the ColD residue aligned to ArnB Lys188 is His. EryCIV has Lys, not His, placing it unambiguously on the transaminase side of the split. The PLP pyridine-N1 anchor is likewise conserved: EryCIV Asp170/171 in motif FFDAAHA aligns to ArnB EDAAHAV, the aspartate that hydrogen-bonds the protonated pyridine nitrogen of PLP.

Pairwise identities reinforce the placement by a roughly two-fold margin:

The ~2-fold identity gap between the nearest transaminase (DesI, 60.5%) and the dehydratases (≤23%), together with the Lys-vs-His discriminator, refutes Failure Mode #3.

Finding 2 — EryCIV's nearest characterized homolog is DesI, a C4 sugar aminotransferase: the seed term is correct but too general

EryCIV's closest characterized neighbor by a wide margin is DesI (60.5% identity over 400 columns). DesI is an X-ray-characterized PLP-dependent aminotransferase that converts dTDP-4-keto-6-deoxyglucose → dTDP-4-amino-4,6-dideoxyglucose (PMID: 17630700). Because erythromycin (in S. erythraea) and methymycin/pikromycin (in S. venezuelae) both deploy identical D-desosamine machinery, eryCIV is the desosamine-pathway ortholog of desI. This is precisely the granularity failure (Failure Mode #1): the propagated term names the broad family activity (any transaminase), while the protein carries out a specific C4 sugar-transamination step. UniProt annotation is concordant — EryCIV carries GO:0008483 (transaminase activity, IEA/TreeGrafter) and GO:0030170 (PLP binding), and the PLP-binding keyword is present.

This finding also surfaces the main competing / historical signal: EryCIV was originally deposited under a "3,4-dehydratase" name (it was said to "resemble the AscC 3,4-dehydratase"; PMID: 9353926). That label predates the structural era and is a pre-structural homology carry-over — the same vintage of error that for years left DesI itself mislabeled a "4-dehydrase" despite being a proven aminotransferase. It is directly contradicted by the active-site analysis in Finding 1 (EryCIV has the transaminase Lys, not the dehydratase His) and must not be propagated by a curator.

Finding 3 — AlphaFold confirms an intact, well-ordered PLP aminotransferase pocket (not a degenerate active site)

To test Failure Mode #2 (fold retained but catalysis lost), I examined AlphaFold DB model AF-A4F7N3-F1 (v6), overall mean pLDDT 96.5. The catalytic Schiff-base Lys200 and the pyridine-N1-anchoring Asp171 (motif DAAHA) co-localize in a single compact pocket, each modeled with very high local confidence (pLDDT 96–99). Measured geometry:

Active-site pocket residues average pLDDT 98.1, indicating a well-ordered, non-degenerate catalytic site. There is no evidence of active-site erosion, loop disorder, or catalytic-residue substitution that would mark a pseudo-enzyme. The geometry is consistent with a PLP cofactor bridging Lys200 (internal aldimine) and Asp171 — i.e., an intact, catalytically competent aminotransferase active site. Failure Mode #2 is refuted.

Mechanistic Model / Interpretation

EryCIV operates in the dTDP-D-desosamine branch of the erythromycin deoxysugar machinery. Desosamine is one of two sugars (with L-mycarose/L-cladinose) appended to the erythronolide macrolactone; eryC-class mutants accumulate 3-α-mycarosyl erythronolide B, the signature of a block in desosamine attachment (PMID: 9563840). Within that pathway, EryCIV performs the C4 transamination step, installing the amino group that is later elaborated into the dimethylamino function of desosamine:

 dTDP-4-keto-6-deoxy-D-glucose
    |
    |   EryCIV (= DesI ortholog)
    |   PLP-dependent C4 transaminase
    |   Lys200 forms internal aldimine with PLP;
    |   Asp171 anchors PLP pyridine N1;
    |   amino donor (L-Glu) -> PMP -> transfer to sugar C4
    v
 dTDP-4-amino-4,6-dideoxy-D-glucose
    |
    |   (downstream N,N-dimethylation, etc.)
    v
dTDP-D-desosamine  ---->  glycosyl transfer onto erythronolide

The family-level ambiguity that makes the graft under-specified — and its resolution:

      DegT/DnrJ/EryC1 fold (PF01041, fold-type I PLP)
                       |
+----------------------+------------------------+
   TRANSAMINASES                                    DEHYDRATASES
 (catalytic Lys at PLP site)                   (His at that position)
 DesI, ArnB, WecE, EryCI, DesV                     ColD, AscC
|                                               |
   EryCIV sits here  <-- Lys200 present            EryCIV is NOT here
   (60.5% id to DesI)                              (<=23% id; His discriminator absent)

The three findings converge on one coherent conclusion: EryCIV has the right fold, the right subfamily, the right nearest characterized neighbor, and an intact catalytic apparatus. The only defect in the propagated annotation is resolution — it stops at family-level "transaminase activity" rather than naming the specific sugar C4-aminotransferase reaction. This is a true-positive IEA requiring a granularity refinement, not a false positive.

Evidence Base

How the literature frames the conclusion. PMID: 17630700 provides the structural keystone: it establishes DesI — EryCIV's 60.5%-identical nearest neighbor — as a PLP-dependent aminotransferase and defines its exact reaction product, pinning EryCIV's specific function by orthology. PMID: 9563840 anchors the pathway role genetically, showing eryC-class disruptions produce the desosamine-deficient 3-α-mycarosyl erythronolide B phenotype. PMID: 9353926 documents the competing historical "3,4-dehydratase" label — the carry-over that motivated the active-site test and that this report recommends against propagating.

Active-Site / Placement Analysis (computed provenance)

Method: Needleman–Wunsch global alignment (BLOSUM62, gap −7) of EryCIV against characterized DegT/DnrJ/EryC1 members; catalytic positions anchored on literature residues (ArnB Lys188, WecE Lys181). AlphaFold geometry read from AF-A4F7N3-F1 v6.

Catalytic PLP Schiff-base lysine — INTACT:

          catalytic residue v
EryCIV   ...S F H A T K A V T S...   (Lys200)
ArnB     ...S F H A I K N I T C...   (Lys188, catalytic)
WecE     ...S F H E T K N Y T A...   (Lys181, catalytic)
ColD     ...aligned position = His (NOT Lys)  -> dehydratase hallmark

Second PLP anchor (pyridine-N1 Asp) — INTACT:
EryCIV FFDAAHA (Asp170) aligns to ArnB EDAAHAV.

Pairwise % identity to EryCIV (nearest-neighbor placement):

DesI  Q9ZGH0  60.5%   (characterized C4 transaminase)  <- nearest
ArnB  P77690  37.5%   (characterized transaminase)
EryCI P14290  36.1%   (characterized C3 transaminase)
DesV  Q9ZGH4  32.5%   (characterized C3 transaminase)
WecE  P27833  32.4%   (characterized transaminase)
AscC  Q57174  23.0%   (PLP dehydratase)
ColD  G4WJD4  19.8%   (PLP 3-dehydratase)             <- far

AlphaFold structural confirmation (AF-A4F7N3-F1, v6; mean pLDDT 96.5):

Lys200-NZ  <->  Asp171 carboxyl   6.98 Å   (spans a PLP ring; Asp anchors pyridine N1)
Lys200-NZ  <->  His174-NE2        3.15 Å
Lys200-NZ  <->  Ser195-OG         4.52 Å
pocket residues mean pLDDT = 98.1

Conclusion: Catalytic residues are intact and correctly spaced; EryCIV is unambiguously placed in the transaminase subfamily (DesI clade), not the dehydratase clade. The pocket is well-ordered, not degenerate — consistent with a catalytically competent aminotransferase.

GO Curation Implications

Recommended lead action (requires curator verification):

• GO:0008483 transaminase activity (IEA): RETAIN, but MAKE MORE SPECIFIC. Replace/augment with a child term reflecting a dTDP-sugar C4 aminotransferase (sugar aminotransferase; EC 2.6.1.–). The broad term is not wrong, merely under-granular for a protein with a 60.5%-identical, structurally characterized neighbor of known specific activity.
• GO:0030170 pyridoxal phosphate binding: RETAIN. The cofactor keyword and conserved PLP-binding residues confirm it.
• Biological process: flag any generic polysaccharide-biosynthesis propagation for the curator — the accurate BP is dTDP-sugar / desosamine (deoxysugar) biosynthetic process, not generic polysaccharide biosynthesis.
• Do NOT propagate the legacy "3,4-dehydratase" name — a pre-structural carry-over contradicted by the active-site analysis.
• This is a true-positive IEA with a granularity refinement. "Protein binding" is explicitly not an appropriate fallback.

Limitations and Knowledge Gaps

1. No direct enzymatic assay for EryCIV itself. The specific-function call rests on 60.5% orthology to DesI plus active-site conservation and genetics, not on a purified-enzyme kinetic assay of A4F7N3. Strong but inferential.
2. AlphaFold is a prediction. The 6.98 Å Lys200–Asp171 geometry and pocket ordering come from a predicted model (very high pLDDT, but not experimental coordinates); no PLP ligand is present in the model.
3. Regiochemistry inference. The exact C4-vs-C3 regiochemistry is inferred from DesI orthology; EryCI/DesV already account for C3 amination in related pathways, making C4 the parsimonious assignment for EryCIV, but this is not experimentally confirmed for A4F7N3.
4. Reference-set and method scope. Analysis is sequence/structure-homology based (programmatic UniProt + custom Needleman–Wunsch, linear gap, no curated MSA). The characterized-member panel is representative but not exhaustive; reference catalytic-residue numbering is from the literature while EryCIV numbering is from alignment.
5. Substrate promiscuity. PMID: 9563840 notes some ery deoxysugar enzymes "can accommodate alternative substrates," so the precise in vivo substrate channel is not fully pinned by sequence alone.

Proposed Follow-up Experiments / Actions

1. Curator action (immediate): Refine the EryCIV annotation from family-level GO:0008483 to the specific sugar C4-aminotransferase child term, citing orthology-to-DesI (PMID: 17630700) as ISS/ISO support; flag the "3,4-dehydratase" name as deprecated carry-over.
2. Direct biochemical confirmation: Express and purify EryCIV, reconstitute with PLP, and assay transamination of dTDP-4-keto-6-deoxy-D-glucose + L-glutamate, detecting dTDP-4-amino-4,6-dideoxyglucose by LC-MS; include the K200A mutant as a loss-of-function control. This directly closes the one inferential gap.
3. Cross-complementation genetics: Test whether eryCIV complements a desI mutant of S. venezuelae (and vice versa), confirming functional orthology in vivo.
4. Structural validation: Solve EryCIV with PLP (or PMP + sugar-nucleotide), or 3-D superpose the AlphaFold model onto the DesI structure, to confirm the predicted Lys200/Asp171 geometry and substrate pose.
5. Genome paralog check: Confirm no second DegT/DnrJ/EryC1 paralog in S. erythraea has been confused with eryCIV in the annotation pipeline, ruling out orthology-assignment error at the genome level.

Bottom line

EryCIV (A4F7N3) is a genuine PLP-dependent sugar aminotransferase — the desosamine-pathway ortholog of DesI. The TreeGrafter/PANTHER graft identified the correct kind of activity (transaminase) but stopped at the family level. The decidable, evidence-backed action is to retain the transaminase assignment and make it more specific (dTDP-4-amino-4,6-dideoxy-D-glucose C4 aminotransferase). The pseudo-enzyme and wrong-subfamily failure modes are affirmatively ruled out by the intact catalytic lysine, the Lys-vs-His discriminator, and the well-ordered AlphaFold active site.

Artifacts

• OpenScientist final report
• OpenScientist final report