ENDOG mtDNA-breakpoint motif-enrichment hypothesis — Deep Research Report

Gene: ENDOG (Q14249), Homo sapiens (NCBITaxon:9606)
Hypothesis slug: mtdna-breakpoint-motif-enrichment
Focus type: free_text (computational analysis on sequence + coordinate data)
Date: 2026-10-26

Seed hypothesis. Human ENDOG's cleavage-site preferences explain where mitochondrial
DNA breakpoints fall: breakpoints of nuclear mitochondrial insertions (NUMTs) and of
mtDNA deletions are enriched at G-quadruplex-forming and C-rich motifs in the mitochondrial
genome, relative to position- and composition-matched control sites.

This was executed as a computation on sequence and coordinate data, not a literature review.
All breakpoint coordinate sources named in the seed context were obtained programmatically.

Executive Judgment

Verdict: PARTIALLY SUPPORTED, and over-interpreted as stated — the NUMT arm is REFUTED and the
deletion arm is a direct-repeat confound rather than evidence of ENDOG causation.

• mtDNA deletions: breakpoints are statistically enriched at G4-forming / C-rich motifs,
  even after matching local GC composition (local max|G4Hunter| fold ≈ 1.17, z = +13, p < 1e-4;
  robust across G4Hunter windows and a stringent ±1 kb local null). However, a direct-repeat
  stratification shows the entire signal comes from repeat-mediated deletions (DR ≥ 5 bp:
  fold 1.24, z = +16) and disappears in non-repeat deletions (DR < 5 bp: fold 0.97, p = 0.91).
  mtDNA direct repeats are themselves G/C-rich, so the "motif" enrichment is collinear with the
  classical slipped-replication / direct-repeat mechanism and does not isolate ENDOG cleavage.
• NUMTs: No enrichment at G4/C-rich motifs over GC-matched controls, in three independent
  datasets — the exact 66,083-genome catalogue named in the seed (PMID:36198798 germline, z = +0.3,
  p = 0.39; somatic, p = 0.10) and the UCSC NumtS set (p = 0.59). The seed's NUMT claim is refuted.

Curation bottom line: This computational result should not be used to assign ENDOG a
mtDNA-breakpoint-determining function. It demonstrates co-localization of deletion breakpoints with
G/C-rich, repeat-prone sequence — a long-known property of mtDNA deletions — not a specific,
ENDOG-attributable cleavage signature. The enrichment is real but non-discriminating; the NUMT
prediction is wrong.

Computed Results (primary provenance)

GC-composition-matched permutation null (10 GC deciles, 10,000 permutations); statistic = local
max|G4Hunter| (window 25, ±5 bp slack) and local C-fraction (window 25). Reference: rCRS NC_012920.1
(16,569 bp; reference/light strand is C-rich: C = 31.3 %, G = 13.1 %, so negative G4Hunter =
C-rich on the reference = G4-forming potential on the complementary heavy strand).

Supporting robustness (deletions, all):
- Parameter sweep: fold 1.14–1.28 across G4Hunter windows 15/25/35 × ±0/5/10 bp slack, all p ≈ 2e-4.
- Stringent local null (±1 kb): breakpoints sit at the 57th percentile (median 60th) of G4
propensity within their own neighborhood (Wilcoxon p = 2.8e-23) → not purely a regional-composition artifact.
- Mann–Whitney vs whole genome: deletions p = 4.4e-43; NUMTs (germline) p = 0.015 (unmatched,
dissolves under GC-matching).
- Known hotspots sit at extreme motifs: 9 bp deletion hotspot 8271–8281 max|G4| = 2.0 (99.8th
genome percentile); CSB2 (99.9th); common 4977-del 5′ 8470–8482 (96.9th), 3′ 13447–13459 (90.3th).
These are all direct-repeat-flanked deletions, i.e., the hotspot G/C-richness is the repeat.
- Direct-repeat prevalence: 69.9 % of the 1,369 deletion pairs have a ≥5 bp perfect repeat at the
junction (median 5 bp), consistent with the known microhomology/direct-repeat mechanism.

Data provenance (all fetched programmatically this run):
- rCRS: NCBI efetch NC_012920.1.
- mtDNA deletions: MitoBreak human Del table (1,369 pairs → 1,963 unique breakpoints).
- NUMTs: Springer static-content supplementary of PMID:36198798 (MOESM4_ESM.xlsx, Suppl. Tables 1 & 4);
UCSC hg19 numtSMitochondrion track (766 features → 1,195 boundaries).

Evidence Matrix

GO Curation Implications (leads — require curator verification)

• Do NOT create/assign a BP term implying ENDOG determines mtDNA breakpoint/NUMT positions
  (e.g., nothing like "regulation of mitochondrial genome rearrangement / NUMT formation via
  sequence-specific cleavage"). The computational enrichment does not support an ENDOG-specific,
  motif-driven breakpoint-placement function; the deletion signal is a direct-repeat confound and
  the NUMT prediction is refuted.
• Molecular function: Evidence (cited, not re-derived) supports ENDOG's core MF as a
  nuclease acting on structured/damaged DNA — candidates such as endonuclease activity
  (GO:0004519) / deoxyribonuclease activity (GO:0004536), and possibly G-quadruplex DNA binding
  (GO:0051880) on the strength of PMID:36394256. These are MF claims about the enzyme, independent
  of the breakpoint-enrichment hypothesis, and should be curated from the primary assays, not from
  this coordinate analysis.
• This analysis is best logged as negative/qualifying computational context, not as an IEA/ISS
  basis for any term. Prefer: retain existing experimentally-supported nuclease MF terms; do not
  add a breakpoint-placement BP term on the basis of motif co-localization.
• Avoid "protein binding" as a recommendation; a more informative supported MF is the nuclease /
  G4-binding activity above (from primary literature, pending curator check of those PMIDs).

Mechanistic Scope

• Immediate molecular activity under test: sequence/structure-specific DNA cleavage by ENDOG
  (G4-forming / C-rich motif preference). The analysis probes whether where breakpoints sit in the
  mtDNA sequence matches ENDOG's reported preference.
• What the data actually show: a statistical association between mtDNA deletion breakpoints
  and G/C-rich, G4-prone, direct-repeat sequence. This is a property of the breakpoints' genomic
  neighborhood, not a demonstration that ENDOG cut there.
• Downstream / not tested here: actual deletion/NUMT formation, replication-slippage vs
  nuclease-cut causation, ENDOG loss-of-function effects on breakpoint spectra, disease phenotypes.
  The enrichment is agnostic to which nuclease (if any) acts; direct-repeat replication slippage
  explains the same localization without invoking ENDOG.

Conflicts and Alternatives

1. Direct-repeat mechanism (strongest competing explanation). 70 % of deletions are flanked by
   ≥5 bp repeats; the motif enrichment is entirely within this class. mtDNA deletion hotspots have
   long been attributed to direct/imperfect repeats and replication slippage — a mechanism that does
   not require ENDOG and fully accounts for the observed G/C-rich localization.
2. NUMT arm contradicted by its own source. The 66,083-genome catalogue shows no G4/C-rich
   enrichment after composition matching; its trinucleotide table is dominated by low-complexity
   microhomology (AT- and C-rich), consistent with non-homologous end joining at the nuclear insertion
   site rather than ENDOG motif cleavage in mtDNA.
3. Substrate determinant may be damage, not sequence (PMID:41854021). If ENDOG's site selection is
   driven by oxidative lesions / nicks / gaps, a static reference-sequence motif scan cannot capture it —
   the negative/confounded motif result is expected even if ENDOG is biologically involved.
4. Organism / substrate specificity. The C-rich cleavage evidence is mouse (PMID:39752519) and in an
   RNA/transcript context; the yeast transfer-limiting role (PMID:39227600) is species-specific. Neither
   licenses a human DNA-breakpoint-placement claim.
5. Composition collinearity. G4 propensity, C-richness, and GC content are mutually correlated in the
   GC-skewed major arc where deletions cluster; GC-decile matching mitigates but cannot fully disentangle
   these from repeat content.

Knowledge Gaps

• Causal attribution to ENDOG. Checked: motif co-localization (association only). Why it matters:
  association ≠ cleavage. Resolve with ENDOG-knockout vs WT mtDNA breakpoint spectra (ChIP/cleavage-seq,
  END-seq, or deletion mapping).
• Damage-conditioned motif preference. Checked: static rCRS scan only. Why it matters: ENDOG prefers
  oxidized/nicked DNA (PMID:41854021), invisible to reference-sequence scans. Resolve with in-vitro
  cleavage mapping on damaged mtDNA templates, then test breakpoint overlap.
• Repeat-independent deletions. Checked: DR<5 bp subset is null. Why it matters: this is the cleanest
  window to detect an ENDOG-specific motif effect, and it shows none. Resolve with a larger,
  uniformly-ascertained repeat-free deletion set.
• G4 strand/structure validation. Checked: G4Hunter propensity (sequence predictor). Why it matters:
  predicted ≠ folded G4. Resolve with experimental G4 maps (BG4 CUT&Tag, G4-seq) on mtDNA and overlap
  with breakpoints.

Discriminating Tests

1. ENDOG-null vs WT breakpoint spectra (cells/mouse): if ENDOG drives breakpoint placement, its loss
   should shift/deplete breakpoints at G4/C-rich sites — most informative single experiment.
2. In-vitro ENDOG cleavage-site mapping on mtDNA (undamaged vs oxidatively-damaged) → define the true
   human cleavage motif, then test breakpoint overlap (replaces the assumed motif with a measured one).
3. Repeat-controlled re-analysis at scale: restrict to repeat-free deletions and NUMTs and ask whether
   any residual ENDOG-specific motif survives (this run: none).
4. Experimental G4 maps (G4-seq / BG4) on mtDNA to replace G4Hunter predictions and re-test overlap.
5. Cross-nuclease comparison: test whether other mitochondrial nucleases' known motifs fit the
   breakpoints equally well — if so, the signal is non-specific.

Curation Leads (require curator verification)

• Action lead: Treat the breakpoint-motif hypothesis as non-core / not curation-actionable for a
  breakpoint-placement function. Flag the deletion enrichment as confounded by direct repeats and
  the NUMT enrichment as refuted in the review free-text.
• Candidate references to verify (exact snippets to pull from primary sources):
• PMID:36394256 — verify the specific claim that ENDOG binds G-quadruplex at the 9 bp hotspot and
  induces breaks (supports a G-quadruplex DNA binding MF, GO:0051880, if the assay is direct).
• PMID:41854021 — verify ENDOG's preference for oxidatively-damaged / nicked / gapped DNA
  (argues site selection is lesion-driven, tempering any pure-motif claim).
• PMID:36198798 — the authoritative NUMT coordinates used here; Suppl. Table 6 trinucleotide result is
  mixed, not G4/C-rich-specific (supports the refutation).
• Candidate GO terms (MF, from primary assays — not from this analysis): endonuclease activity
  (GO:0004519) / deoxyribonuclease activity (GO:0004536); conditionally G-quadruplex DNA binding
  (GO:0051880). Do not add a breakpoint-/NUMT-placement BP term.
• Suggested curator questions: (i) Is there any ENDOG-dependent (knockout-controlled) evidence that
  it determines human mtDNA breakpoint positions? (ii) Does the review already cite the direct-repeat
  mechanism as the explanation for deletion hotspots? (iii) Is the G4-binding claim (PMID:36394256) a
  direct binding/cleavage assay suitable for an MF annotation?
• Suggested experiment: ENDOG-null vs WT mtDNA breakpoint mapping (above) as the decisive test.

Limitations

• Reference-sequence motif scan cannot capture damage-conditioned or structure-conditioned cleavage.
• Direct-repeat detection used a longest-common-substring proxy (±15 bp); imperfect repeats and
  microhomology thresholds could shift the exact DR≥5 split, though the qualitative confound is robust.
• G4Hunter predicts, but does not prove, G4 folding.
• MitoBreak aggregates deletions of heterogeneous ascertainment; recurrent hotspots were collapsed to
  unique positions to avoid double-counting (results unchanged).
• No ENDOG perturbation data were available; causation cannot be tested from coordinates alone.

Conclusion

The seed hypothesis is partially supported for the descriptive claim (deletion breakpoints
co-localize with G4/C-rich motifs) but over-interpreted as an ENDOG function: the deletion signal is
a direct-repeat confound (absent in non-repeat deletions), and the NUMT arm is refuted in the
exact 66,083-genome catalogue and a second independent dataset. The analysis should inform the review as
a negative/qualifying computational result, not as support for a new ENDOG breakpoint-placement
annotation.

Artifacts

• OpenScientist final report
• OpenScientist final report