RFT1 (Q96AA3) — Family constraint: central cavity vs. lateral lipid portal

Hypothesis (slug family-constraint-cavity-vs-portal): In the RFT1 family, evolutionary
constraint is concentrated on positions lining the central substrate cavity of the MOP-superfamily
fold rather than on positions lining the lateral lipid portal — i.e., selection preserved substrate
capture/coordination rather than the lipid transit path.

Term in scope: GO:0140303 intramembrane lipid transporter activity (MF).

Executive Judgment

Verdict: SUPPORTED (robust, with one honest qualification).

A deep eukaryotic RFT1-family alignment mapped onto the AlphaFold model of Q96AA3 shows that
central-cavity-lining transmembrane (TM) positions are the single most evolutionarily constrained
structural class, well above the TM background, whereas lateral lipid-portal-lining positions are
statistically indistinguishable from background (slightly below it), and the outer lipid-facing
surface is the fastest-evolving class. The cavity-vs-portal contrast is large and highly significant
(rank-biserial −0.68, p = 3×10⁻⁵) and survives four independent controls (a second conservation
metric, amino-acid-composition/Pro-Gly exclusion, a 20,000-draw permutation null, and ±15–25°
perturbations of the estimated membrane axis).

Qualification (the alternative that is partly true): buried helix-packing ("neither") residues are
numerous (n≈180) and also modestly enriched above background (p = 1×10⁻³, effect −0.18). So a
meaningful share of total family constraint is fold maintenance. But this is a weaker effect than the
cavity (−0.43), and — critically — the cavity signal is not driven by helix-breakers: cavity
residues contain 0% Gly and 0% Pro; they are 28% charged and 47% polar, the chemistry of
substrate coordination. The hypothesis's own "Pro/Gly fold-maintenance" alternative is therefore
refuted for the cavity class specifically.

Most important caveats: (i) the family alignment is eukaryote-only — PF04506/RFT1 has no
alignable archaeal/bacterial members, so the requested prokaryotic MOP homologues (MATE/MurJ) could
not be folded into one MSA; (ii) the membrane axis and the cavity/portal geometry are derived from a
single static AlphaFold model with an operationally defined portal; (iii) evolutionary constraint
indicates importance, not mechanism per se.

What was actually computed (provenance)

All numbers below are direct outputs of code executed this run (not asserted). Pipeline:

1. Alignment. Pfam PF04506 ("Rft-1") full alignment via the InterPro API →
   3,826 sequences, 512 HMM match columns. Match columns identified by the Pfam convention
   (columns containing no . and no lowercase). Taxonomic spread: fungi (9PEZI, 9EURO, 9AGAR, 9HYPO,
   9PLEO, 9BASI, 9HELO…), metazoa (9BILA, 9PASS, 9TELE, 9AVES…), plants (9ROSI, 9POAL, 9ORYZ, 9LILI),
   stramenopiles (9STRA). 1,371 distinct clades; zero bacterial/archaeal mnemonics.
2. Per-position constraint. Henikoff position-based sequence weights → weighted per-column
   Kullback–Leibler relative entropy to background (primary metric) and normalized Shannon conservation
   (secondary). Columns with weighted gap fraction > 0.6 excluded as unreliable.
3. Human mapping. Profile Needleman–Wunsch (free end gaps) of Q96AA3 against the match-state
   profile → 497 human residues mapped (positions 10–528).
4. Structure & membrane frame. AlphaFold AF-Q96AA3-F1 (v6). Membrane normal estimated as the
   principal eigenvector of TM-helix direction vectors (helices detected from Cα i→i+3 / i→i+4 geometry).
   TM residues = |z − membrane-center| ≤ 15 Å (n = 312 mapped + reliable).
5. Geometric classification (explicit criteria). Occupancy grid (1.5 Å, 1.4 Å probe) in the
   membrane frame. For each TM residue, side-chain centroid → cylindrical (r, θ, z) about the central
   axis.
6. central-cavity-lining: side chain points inward (r_sidechain < r_Cα − 0.3 Å) and reaches the
   central channel wall (r ≤ lumen radius(θ) + 3 Å).
7. lateral-portal-lining: lipid-exposed (>60% of outward hemisphere sample points free of protein)
   and within 9 Å of a cavity-lining residue (i.e., lipid-facing residues bordering the cavity's
   lateral opening).
8. outer lipid-facing: lipid-exposed but not adjacent to the cavity.
9. buried helix-packing ("neither"): neither lumen-lining nor lipid-exposed.

Core result (KL relative entropy, bits)

Direct contrasts: cavity vs portal p = 2.9×10⁻⁵ (effect −0.68); cavity vs all lipid-facing
p = 1.0×10⁻¹²; all-lipid-facing vs background p = 9.3×10⁻¹² (below).

Robustness controls

• Second metric (normalized Shannon conservation): cavity 0.564 vs bg 0.438 (p = 5.7×10⁻⁴);
  portal 0.373 vs bg (p = 0.19, NS); cavity vs portal p = 2.3×10⁻⁴. Same ranking.
• Composition / not a Pro-Gly artifact: cavity = 0% Gly, 0% Pro, 28% charged, 47% polar, 16%
  aromatic; portal = 0% charged, mostly hydrophobic. Cavity enrichment unchanged after excluding
  Gly/Pro (none present), p = 6.2×10⁻⁵.
• Permutation null (20,000 random 32-residue TM draws): observed cavity mean 2.354 vs null mean
  1.676, empirical one-sided p = 1×10⁻⁴.
• Membrane-axis sensitivity: across ±15° to ±25° tilts of the estimated normal, cavity ≫ portal
  (p = 3.1×10⁻⁶ … 7.9×10⁻⁴) and cavity ≫ background persist.

Axis-independent corroboration (continuous, confounder-controlled)

To confirm the result does not depend on the cavity/portal class boundaries or on the precise membrane
axis, per-residue constraint was correlated with continuous geometric measures over all 312 TM positions:

Monotonic gradient by lipid-exposure quartile (least→most exposed): mean KL 2.31 → 1.95 → 1.48 → 0.98.
Restricting to high-confidence residues (pLDDT > 70; 308/312) leaves the correlations unchanged
(lipid ρ = −0.596; radial ρ = −0.675). Spatial clustering: the 15 most-constrained TM residues have a
mean pairwise distance of 19.2 Å vs a random-TM null of 30.1 Å (5,000 permutations, p < 0.0001) — they
form a single compact pocket, not dispersed fold positions. Together these confirm the cavity-over-portal
pattern without relying on the discrete geometric classification.

Candidate substrate-coordinating residues (most-constrained cavity positions, human numbering)

R290 (KL 4.05), N435 (3.81), Y378 (3.66), E64 (3.48), E298 (3.31), Q186 (3.28), N283 (3.10),
E156 (3.06), T257 (3.06), S286 (3.00), E258 (2.80), F415 (2.79), E260 (2.67), S412 (2.63), K152 (2.57).
These are overwhelmingly H-bond donors/acceptors and charged side chains — the chemistry expected to
coordinate the polar Man5GlcNAc2-pyrophosphate headgroup of the substrate. (Leads for mutagenesis;
require curator/experimental verification.)

Evidence Matrix

GO Curation Implications (leads — require curator verification)

• GO:0140303 (intramembrane lipid transporter activity, MF) → RETAIN. It is directly supported by
  reconstitution (PMID 38886340) and is reinforced by this analysis: the family's conserved core is
  precisely the polar substrate-handling cavity of a MOP flippase, not incidental surface.
• Consider a more specific MF child if one exists (curator to check the ontology): the physiological
  substrate is a lipid-linked oligosaccharide (Man5GlcNAc2-PP-dolichol), so a "glycolipid/oligo­saccharide-lipid
  flippase" style MF child would be more informative than the generic parent. Do not invent an ID;
  verify against current GO. Evidence code for the retained/refined MF should reference the direct assay
  (IDA, PMID 38886340) rather than IEA/ISS alone.
• This evolutionary analysis is CC/BP-neutral. It speaks to which residues matter (MF mechanism),
  not to localization (ER membrane, CC) or the glycosylation process (BP); those should be curated from
  their own primary evidence.
• "Protein binding" is not an appropriate summary here — a specific transporter MF is supported.

Mechanistic Scope

• Immediate molecular function tested: which wild-type residues selection has preserved across the
  RFT1 family, and where they sit relative to the substrate cavity vs. the lipid-facing/portal surface
  of the MOP fold. This is a statement about the flippase's substrate-coordination machinery.
• Direct gene-product activity: transbilayer flipping of Man5GlcNAc2-PP-dolichol (PMID 38886340).
• Separated (not part of this test): downstream N-linked glycosylation flux, CDG disease phenotypes,
  or any inference from loss-of-function alleles. Per instructions, no disease/pathogenicity data were
  used to select positions; the constrained set is defined purely by cross-species selection and
  geometry.

Conflicts and Alternatives

• Both-cavity-and-portal alternative: not supported — portal is at/below background while cavity is
  above.
• Fold-maintenance (Pro/Gly, helix-packing) alternative: partly true but not for the cavity. Buried
  packing residues carry a weaker-but-real constraint enrichment; the cavity class contains no Pro/Gly and
  is polar/charged. So "fold maintenance dominates everything" is refuted, but "fold maintenance also
  matters" is honestly acknowledged.
• Shallow/too-divergent alignment alternative: not the case — 3,826 sequences, 512 columns, human
  mapped over 497 residues; classes separate with large effect sizes and small p-values.
• Portal-definition sensitivity: the "portal" is operationally the lipid-facing rim of the cavity in a
  single, likely occluded AlphaFold state. A genuinely open lateral gate in another conformational state
  could be lined by different residues; my first (ray-tracing) portal detector found no wide-open
  lateral sector in this model, consistent with an occluded state.
• Cross-superfamily constraint (MATE/MurJ): deliberately requested but not computable by sequence
  alignment — those Pfam families are too divergent to co-align with RFT1; a structure-based comparison
  would be needed.
• Paralog/isoform confusion: none material — RFT1 is single-copy in human; PANTHER PTHR13117:SF5 and
  TCDB 2.A.66.3.2 are RFT1-specific.

Knowledge Gaps

1. Experimental structure of RFT1. Checked: only an AlphaFold model was available programmatically;
   the membrane axis/portal were inferred. Why it matters: the cavity/portal partition depends on the
   model. Resolver: a cryo-EM/OPM-oriented Rft1 structure (± bound LLO) to define the real cavity and any
   lateral gate.
2. Prokaryotic MOP constraint. Checked: PF04506 has no bacterial/archaeal members. Why it matters:
   the hypothesis invokes the broader MOP/MATE/MurJ architecture. Resolver: structure-based multiple
   alignment across the superfamily (e.g., Foldseek/DALI + 3D-anchored MSA).
3. Substrate-contact identity of the conserved cavity residues. Checked: only geometric proximity to
   the lumen. Why it matters: conservation ≠ contact. Resolver: substrate-bound structure or docking, and
   mutagenesis of R290/N435/Y378/E64/E298 etc.
4. Conformational state. Checked: one static model (apparently occluded). Why it matters: the portal
   may only exist transiently. Resolver: multi-state modeling / MD / alternate-state structures.

Discriminating Tests

• Substrate-bound Rft1 structure or MD-informed docking → do the constrained cavity residues
  (R290, N435, Y378, E64, E298, Q186, N283, E156, S/T cluster 257/286/412) actually contact the
  oligosaccharide/pyrophosphate? This is the single most decisive test.
• Alanine/charge-reversal mutagenesis of the top cavity residues vs. matched portal residues,
  scored in the reconstituted flipping assay (PMID 38886340): cavity mutants should impair transport far
  more than portal mutants if the hypothesis holds.
• Structure-based superfamily constraint (RFT1 vs MATE vs MurJ, aligned in 3D) to test whether the
  cavity-over-portal pattern generalizes across the MOP fold.
• Deep-mutational scanning / evolutionary-coupling (DCA) on the same alignment to check that
  co-evolving/coupled positions concentrate in the cavity.

Curation Leads (all require curator verification)

• Action: Retain GO:0140303 (MF) with IDA support (PMID 38886340); flag for possible refinement to a
  lipid-linked-oligosaccharide-flippase child term if the ontology provides one.
• Candidate reference + exact snippets to verify (PMID 38886340):
• "purified Rft1 catalyzes the translocation of M5GN2-PP-Dol across the lipid bilayer"
• "confirm the molecular identity of Rft1 as the M5GN2-PP-Dol ER flippase"
• Supporting computational provenance (this report): the constraint-by-class table, robustness controls,
  and the ranked cavity-residue list are curator-usable evidence that the conserved core of RFT1 is its
  substrate-handling cavity.
• Suggested curator questions: (1) Is there a GO MF child specific to oligosaccharide-lipid/glycolipid
  flipping? (2) Should the ER-membrane CC and N-glycosylation BP terms be (re)affirmed from their own
  primary evidence rather than from this MF-focused analysis?
• Suggested experiments: cavity-vs-portal mutagenesis in the reconstituted flipping assay; substrate-bound
  structure.

Bottom line

Within the deepest assemblable (eukaryotic) RFT1 family, evolutionary selection has preserved the polar
central-cavity residues of the MOP fold — the substrate-capture/coordination surface — far more than the
lateral lipid-facing/portal surface, which evolves at or below the transmembrane background. The seed
hypothesis is supported, with the honest caveat that buried fold-maintenance packing also carries a
weaker constraint signal, and that the analysis rests on a single static AlphaFold model of a
eukaryote-only family. The result reinforces retention of GO:0140303 and yields a concrete, testable
set of candidate substrate-coordinating residues.

Artifacts

• OpenScientist final report
• OpenScientist final report