AP3M2 (mu3B) bioinformatics results

All numbers below are computed at run time by analyze.py from live UniProt, PDBe/RCSB and Human Protein Atlas records. Nothing is hardcoded; re-running regenerates this file.

1. Human adaptor medium (mu) subunits

Pairwise global identity (%, BLOSUM62 global alignment)

The matrix is computed per ordered pair, and where several alignments are equally optimal the two directions can pick different ones, so it is very slightly asymmetric: the largest difference between a cell and its transpose is 0.6 percentage points. Both values in such a pair are correct for their own direction; the AP3M2 row, which is the one the review cites, is unaffected either way.

AP3M2's closest human paralogue is AP3M1 (84.2% identity). Identity to the AP-1 and AP-2 medium subunits is AP1M1 29.7%, AP1M2 31.6%, AP2M1 29.1%.

2. The AP-3 tyrosine-cargo site, observed and transferred

2a. mu3A residues that contact LAMP1 cargo in PDB 9C5B

9C5B is the human AP-3 holocomplex on a lipid nanodisc with the LAMP1 cytoplasmic tail engaged (chain Y, modelled sequence SHAGYQTI, which carries the GYQTI YxxPhi motif). The mu subunit in the structure is Q9Y2T2 (AP3M1/mu3A). Every mu3A residue with a heavy atom within 4.0 A of the cargo peptide is taken as the site: 9 residues qualify. Because AP3M1 and AP3M2 are 84.2% identical, transferring these positions to AP3M2 is a near-trivial alignment rather than a cross-family inference.

AP3M2 is identical to AP3M1 at 8 of the 9 cargo-contacting positions, with no deletions. The tyrosine-cargo site that the AP-3 cryo-EM structure resolves is therefore intact in the neuronal mu3B paralogue; there is no residue-level evidence that AP3M2 has lost cargo recognition.

2b. An independent AP-3 cargo complex: PDB 4IKN

4IKN is the Rattus norvegicus (Rat) mu3A C-terminal domain bound to the TGN38 cytoplasmic tail (chain B, modelled sequence DYQRL, carrying the DYQRL YxxPhi motif) - a different species and a different cargo from 9C5B, so it tests whether the site found there is structure-specific. 10 rat mu3A residues lie within 4.0 A of the peptide. Rat and human mu3A are 98.8% identical and rat mu3A and human mu3B are 84.2% identical.

Projected onto human AP3M1, the 10 contacts of 4IKN land on 10 positions, of which 9 are among the 9 that 9C5B shows contacting LAMP1 (shared: 180, 181, 389, 392, 402, 403, 404, 405, 406). Human AP3M2 carries the same residue as rat mu3A at 9 of the 10 positions. Two AP-3 structures, two different YxxPhi cargoes and two species therefore pick out the same site, and mu3B matches mu3A across it.

2c. Outgroup: the classical mu2 YxxPhi pocket (PDB 1BXX)

1BXX is the mu2 (AP50) C-terminal domain of Rattus norvegicus (Rat) bound to the TGN38 internalisation peptide (chain P, modelled sequence DYQRLN). 13 mu2 residues lie within 4.0 A of the peptide. Positions are in P84092 (AP2M1_RAT) numbering and are carried onto each human paralogue by pairwise alignment.

Positions where every human paralogue matches mu2: mu2 D176, mu2 V401 (2 of 13). Every other contact position has at least one paralogue that diverges.

AP3M2 aligns to 13/13 of the mu2 pocket positions, with no deletions, but matches mu2's residue at only 4 of them - fewer than AP1M1 (10), AP1M2 (9) or AP4M1 (11). The AP-3 medium subunits have diverged substantially from the AP-2 signal pocket even though, per section 2a, they bind tyrosine cargo through the structurally equivalent region.

2d. Do the alignment and structure routes agree?

Projecting the 13 mu2 pocket positions onto AP3M1 by alignment lands on 13 AP3M1 positions, of which 9 are among the 9 positions the 9C5B structure actually shows contacting LAMP1 cargo (overlap: 180, 181, 389, 392, 402, 403, 404, 405, 406). The alignment-only route and the structure-observed route therefore identify the same site, which is the check that the cross-family alignment in 2c is not drifting.

3. The mu-linker amphipathic helix

Begley et al. 2024 (PMID:39705307) report a membrane-inserting amphipathic helix in the mu3 linker of human AP-3. Below, the highest-hydrophobic-moment 18-residue window starting in the 45 residues that precede each protein's own UniProt MHD domain, using the Eisenberg consensus scale at 100 degrees per residue.

AP3M2 = 0.487 versus AP3M1 0.450; the AP-1/AP-2 subunits score AP1M1 0.433, AP1M2 0.373, AP2M1 0.227 and AP4M1 0.161. The two mu3 proteins hold the top two hydrophobic moments. Mean hydrophobicity is positive for AP3M2 (+0.10), AP3M1 (+0.12), AP4M1 (+0.44) - so a positive alone does not separate mu3 from the rest, and it is the moment that does.

4. AP3M2 isoform 2 (P53677-2)

• VAR_SEQ 268-273: NLVAIP -> KCCLGM (in isoform 2)
• VAR_SEQ 274-418: Missing (in isoform 2)

The MHD of AP3M2 spans 176-417 of 418 residues, so the isoform-2 variant removes the C-terminal portion of the very domain that carries the sorting-signal site analysed in section 2. Of the 9 cargo-contacting positions, isoform 2 deletes 7 (389, 392, 402, 403, 404, 405, 406) and retains 2 (180, 181), which lie N-terminal to the truncation.

5. Is human AP3M2 neuron-restricted?

Human Protein Atlas fields for ENSG00000070718:

• RNA tissue specificity: Low tissue specificity
• RNA tissue distribution: Detected in all
• RNA single cell type specificity: Cell type enhanced
• RNA single cell type specific nCPM: {'Differentiating spermatogonia': '82.0', 'Late spermatids': '110.1', 'Melanocytes': '103.5', 'T-cells': '88.5'}
• RNA single cell type group specific nCPM: {'Pigment cells': '103.5', 'Spermatogenic cell types': '110.1'}
• RNA single nuclei brain specificity: Low cell type specificity
• RNA brain regional specificity: Low region specificity
• Tissue expression cluster: Cluster 46: Brain & retina - Neuronal signaling