Adenine Base Editor Guide Optimization for Sickle Cell Disease

Hematology

Adenine Base Editor Guide Optimization for Sickle Cell Disease

Clinical-stage cell therapy team needed ABE guide RNAs that achieve ≥60% HbF reactivation with minimal bystander edits before ex vivo manufacturing scale-up.

Company

Clinical-stage cell therapy biotech (Series B, 28-person team)

Timeline

April to May 2025

Engagement

Base Editing Guide Design & Edit Outcome Simulation

Base Editing Guide Optimization

12 days: Computational delivery
68%: Observed HbF reactivation (top guide)
<0.5%: Bystander edit rate (top 3)
4/5: Guides met edit threshold

The Challenge

A clinical-stage biotech was developing an ex vivo ABE8e base editing therapy to reactivate fetal hemoglobin (HbF) in sickle cell disease patient CD34+ cells. Their internal bioinformatics team had designed 12 guides manually over 6 weeks, but only 2 achieved acceptable on-target editing in primary cells, and 1 showed unacceptable bystander adenine edits at a nearby codon. They needed a ranked guide panel with predicted edit outcome distributions before locking a manufacturing process for their Phase 1/2 trial.

Business Constraints

Budget: $285K (remaining preclinical computational budget)
Timeline: Ranked base editor guides in 12 business days
Must achieve ≥55% on-target edit with <1% bystander rate at adjacent adenines

GeneForge Approach

Days 1 to 4: HBB Locus Modeling and Guide Enumeration

Input

HBB gene locus (GRCh38) with BCL11A erythroid enhancer coordinates
Patient haplotype data (HbS, HbC, and common beta-globin cluster variants)
Prior internal guide sequences and edit outcome data (12 guides, 2 validated)

Methods

320K ABE-compatible protospacer enumeration across HBB promoter and BCL11A enhancer
BE-Hive and DeepABE ensemble scoring for A→G edit efficiency
Bystander edit probability modeling across all editable adenines in editing window

Output

Top 1,800 guides ranked by on-target edit probability
Bystander risk heatmap across all enumerated protospacers

Days 5 to 9: Off-Target and Edit Outcome Simulation

Genome-wide off-target search (Cas-OFFinder + CHANGE-seq validated sites) eliminated 412 guides with high-risk off-targets in coding regions. Edit outcome simulation predicted indel and base edit distributions for top 200 guides in silico. Erythroid-specific chromatin accessibility (ENCODE CD34+ → erythroblast differentiation) weighted on-target predictions. Guides with predicted bystander edits affecting HBB coding sequence were removed (89 eliminated). Final pool: 156 high-confidence candidates.

Days 10 to 12: Ranking and Primary Cell Validation Protocol

Output

Top 15 base editor guides with on-target score, bystander risk, and off-target tier
Recommended primary CD34+ validation protocol (amplicon-seq + HbF flow cytometry)
gRNA + ABE8e mRNA co-delivery parameters for electroporation
Manufacturing-ready oligo sequences for top 5 guides

Final Ranked Base Editor Guides

Top 5 shown; full list of 15 delivered with edit outcome predictions.

Top ABE guides by on-target edit efficiency and bystander safety
Rank	Guide ID	On-Target Edit	Bystander Risk	Off-Target Sites	Status
1	GF-g101	0.91	0.02	0 critical	Priority A
2	GF-g102	0.89	0.03	0 critical	Priority A
3	GF-g103	0.87	0.04	1 low-risk	Priority A
4	GF-g104	0.85	0.05	0 critical	Priority A
5	GF-g105	0.83	0.06	0 critical	Priority B
6–15	GF-g106–g115	0.72–0.82	0.05–0.12	0–2 low-risk	Backup

Results and impact

Speed, validation, and business outcomes

Speed vs. Internal Manual Design

Metric	Internal Manual Design	GeneForge	Improvement
Timeline	6 weeks	12 business days	3× faster
Cost	$180K (FTE time)	$285K	Higher throughput + edit simulation
Guides Evaluated	12 manual designs	320K variants screened	26,000× larger search space
Guides Meeting Threshold	2/12 (17%)	4/5 top guides (80%)	4.7× higher success rate

Primary Cell Validation Outcomes (4 weeks post-delivery)

Amplicon sequencing and HbF flow cytometry on top 5 guides in CD34+ cells from 3 SCD donors.

Guide	Predicted Edit	Observed Edit (%)	HbF+ (%)	Bystander (%)	Notes
GF-g101	0.91	64%	68%	0.3%	Selected for manufacturing
GF-g102	0.89	59%	62%	0.4%	Backup lead
GF-g103	0.87	57%	58%	0.5%	Consistent across donors
GF-g104	0.85	55%	56%	0.6%	Met minimum threshold
GF-g105	0.83	48%	49%	0.8%	Below HbF target; deprioritized

80%: Top-5 guides met edit threshold (4/5)
68%: Max HbF reactivation (GF-g101)
0.3%: Lowest bystander rate (GF-g101)
4 wks: Timeline to manufacturing-ready guide

Immediate Wins

Manufacturing process locked: GF-g101 selected for Phase 1/2 ex vivo manufacturing within 4 weeks of delivery
Regulatory briefing supported: edit outcome distribution data included in pre-IND meeting package
Avoided trial delay: eliminated 6-week internal redesign cycle that would have pushed manufacturing start

Strategic Advantages

Bystander edit modeling prevented selection of guides that would have introduced silent coding mutations
Patient haplotype-aware scoring ensured guides effective across HbS and HbC allele backgrounds
Edit outcome simulation gave CMC team predicted edit distributions for release specification design

Follow-on engagement
Q3 2025: in vivo LNP delivery guide optimization for systemic HbF reactivation. Estimated cost: $310K. Target: ranked guide panel for NHP study within 2 weeks.

Model validation

Lessons and recommendations

What Worked

BE-Hive + DeepABE ensemble outperformed either model alone for primary cell edit prediction (R² = 0.88 vs. 0.79 single-model)
Erythroid chromatin weighting improved HbF reactivation correlation vs. generic on-target scores
Prior failed internal guides used as negative training signal improved ranking accuracy

Challenges and Mitigations

GF-g105 showed strong predicted edit efficiency but fell below HbF threshold in 2 of 3 donors. Root cause: donor-specific BCL11A enhancer accessibility variation.

Mitigation: Added donor haplotype and chromatin accessibility ensemble scoring; flagged guides sensitive to enhancer state.

One guide in the backup set (GF-g108) showed a low-frequency off-target edit at a pseudogene locus not captured by standard Cas-OFFinder parameters.

Mitigation: Expanded off-target search to include pseudogene and segmental duplication databases; added to standard pipeline.

When to use GeneForge for base editing programs

Ex vivo cell therapy programs requiring precise edit outcome control
Manual guide design producing inconsistent primary cell editing rates
Programs where bystander edits pose regulatory or safety risk
Timeline pressure before manufacturing lock or pre-IND meetings

ROI: approximately 5:1 (avoided trial delay + eliminated repeat design cycle + manufacturing slot value).

Next steps: extend base editor guide optimization to in vivo delivery formats; pair with LNP or AAV tissue targeting analysis.

About This Engagement

Client profile: Series B cell therapy biotech, 28 employees, limited bioinformatics capacity
Project duration: 12 business days (computational delivery) + 4 weeks (validation)
Total cost: $285K
Date: April to May 2025

This case study is anonymized at client request. Guide sequences, patient identifiers, and institutional affiliations have been redacted. Full protocols available under NDA.

Run a similar program with GeneForge.

Tell us about your target. We will scope a 2 to 4 week pilot.

Schedule a Consultation hello@bajpailabs.com