SMN2 Base-Editing Guide Safety: Antisense gRNA 39% Safer Than Liu's A8
TL;DR
Cas-OFFinder off-target analysis on 6 candidate SMN2 base-editing gRNAs (hg38, up to 4 mismatches) returned 2,097 total hits. Three key findings:
- One guide must be discarded:
GTTTTAGACAAAATCAAAAAhas 176 exact matches (unusable — contains a poly-A run matching multiple repeats). - Winner (antisense):
TTTGTCTAAAACCCATATAAhas only 14 exact matches — the lowest burden of all six candidates. - Liu lab's published gRNA A8 (Science 2023, 99% editing efficiency): 23 exact matches. Our antisense pick is 39% safer.
Guide ranking
| Rank | Guide | Exact matches | Verdict |
|---|---|---|---|
| 1 | TTTGTCTAAAACCCATATAA (antisense) |
14 | Primary candidate |
| 2 | GGGTTTTAGACAAAATCAAA |
22 | Secondary (Liu A8 family) |
| 3 | ATGGGTTTTAGACAAAATCA |
23 | Tertiary |
| 4 | TGGGTTTTAGACAAAATCAA |
23 | Tertiary (Liu A8) |
| 5 | GGTTTTAGACAAAATCAAAA |
48 | Avoid without extensive filter |
| 6 | GTTTTAGACAAAATCAAAAA |
176 | DISCARD |
Off-target distribution across all 6 guides
| Mismatches | Count | Interpretation |
|---|---|---|
| 0 (exact) | 306 | Will be edited — real risk |
| 1 | 145 | High risk (ABE8e tolerates 1 mm) |
| 2 | 116 | Moderate risk |
| 3 | 210 | Low risk |
| 4 | 1,320 | Very low risk (ABE8e rarely edits) |
Hot spots are (mostly) tolerable
The majority of exact-match hits cluster in pericentromeric/heterochromatic regions (1q12, 10p11, 20p11, 9p11/q12, etc.) — repetitive DNA that is typically non-transcribed and poorly accessible to base editors. But some MM=0 hits fall outside heterochromatin on chr4, chr6, chr7, chr11, chr13–16, chr19, chr21, chrX, and must be cross-referenced with GENCODE exons/promoters before wet-lab use.
Why this matters
This is Track 2A of the cure pivot: ABE-mediated SMN2 → SMN1 conversion. Liu lab (Science 2023) already demonstrated 99% editing; we are not replicating — we are extending, and guide safety is the primary extension angle. If our antisense guide achieves similar editing efficiency (to be verified wet-lab), it would be a first-in-field improvement over the published protocol.
Next steps (Simon wet-lab)
- Discard
GTTTTAGACAAAATCAAAAA - Gene annotation cross-reference via bedtools + GENCODE v45 (flag any MM=0 in coding exons, splice sites ±6 nt, promoters ±2 kb, OMIM morbid gene bodies)
- Primary test of antisense guide in HEK293 + GUIDE-seq
- Positive control: Liu A8 (known 99% in fibroblasts)
- SpliceAI on top 100 MM=0 sites to flag splice-altering edits
Compute: A100 PCIe 80GB Sweden, ~30 min, Cas-OFFinder Python fallback. CC-BY-4.0. Full finding: /findings/2026-04-10/FINDING_2026-04-10_casoffinder_SMN2_guide_safety.md.