Design on What Was Actually Done

Recommended experimental pipeline (visual-first)

1. Design & in silico vetting: generate candidate sequences via interpretable DL-evolution (start random → greedy mutation steps), score with cell-type model (e.g., DeepFlyBrain / DeepMEL2) and importance maps (SHAP/DeepExplainer), and remove predicted repressor motifs while implanting activator motifs with preferred spacing (see motif-spacing graph) Deep-learning enhancer design (Nature 2023)
2. Reporter assay hierarchy (fast → definitive):
   • Luciferase in relevant cell-state lines (e.g., MM001 MEL vs MM047 MES) to get quick activity/specifity ratios;
   • Chromatin access assays (ATAC) after integration/transduction to check locus/context independence;
   • Stable genomic reporter (AAVS1 safe-harbor or phiC31 locus in Drosophila) for long-term expression and in vivo validation
   Use the AAVS1 safe-harbor integration approach for human iPS/lines to reduce position-effect variability when testing enhancers in differentiated lineages AAVS1-TALEN AAVS1 reporter protocol
3. In vivo Drosophila validation: clone candidate enhancers upstream of minimal-promoter GFP reporters and phiC31-integrate to a single landing site (21F) to compare patterns; quantify positive/negative lines and co-stain for cell markers (e.g., Dachshund for KCs). The primary study achieved 10/13 KC-positive synthetic enhancers in vivo using this pipeline In vivo validation rates (Nature 2023)
4. Chromatin & TF perturbation checks: mutate predicted repressor motifs (e.g., ZEB2 in IRF4 locus) and test effect on reporter/ATAC/ChIP; Enformer/ChromBPNet can simulate locus-level effects but require empirical validation (authors mutated ZEB2 and observed activity increases) ZEB2 repressor evidence (Nature 2023)
5. Functional cell-biology assays to tie enhancer → phenotype:
   • Migration/invasion assays (wound-healing & transwell) when enhancers control genes implicated in motility; use proliferation controls and Matrigel for invasion (protocol details and caveats below)
   • Secretion/trafficking assays (e.g., RUSH system) if enhancer targets secreted proteins (see IER3IP1 proinsulin trafficking study for assay examples)
   • Genotoxicity checks (micronucleus/comet) if constructs or delivery methods include potentially genotoxic elements
   For migration/invasion, use the JoVE wound/transwell protocols to avoid proliferation confounds and to standardize imaging/analysis JoVE wound/transwell protocols
6. Replication & falsification checklist (essential):
   • Biological replicates for all ATAC/ChIP (>=3) and luciferase (>=3 independent transfections); include negative random-sequence controls and R2oDNA-designed biologically neutral sequences as background controls
   • Randomize construct order, blind imaging/quantification, and deposit raw reads/construct sequences in GEO/Zenodo for reproducibility
   • Falsify by testing same motif content in shuffled-arrangement backgrounds and by moving the enhancer to a neutral genomic locus (AAVS1 / phiC31) to test context independence

Concise experimental checklist (materials & key parameters)

• Reporter vectors: minimal promoter luciferase; pENTR/GFP reporters for flies; AAVS1 donor with puromycin marker for human cells.
• Integration: phiC31 (Drosophila 21F); AAVS1-TALEN or CRISPR/Cas9 donor for human lines; include clonal PCR/Southern screening.
• Sequencing/epigenomics: ATAC-seq (50k cells min), ChIP-seq for predicted TFs (2–3 replicates desirable), ddPCR for copy number.
• Controls: neutral sequences (R2oDNA), random sequences, implanted-activator-only, added-repressor-only constructs.
• Statistics: predefine primary endpoint (e.g., % lines with cell-type-specific expression), power for luciferase effect size (~3 biological replicates for within-line comparisons), use sample-as-unit for chromatin assays.

• Model bias: deep models are trained on existing accessible regions and may not capture unobserved chromatin contexts or long-range regulation — empirical locus tests are required (authors note limited ATAC/ChIP replicates) Model/training-data limits (Nature 2023)
• Sequence context & position effects: motif function can depend on local GC, nucleosome occupancy, and 3D contacts; test multiple genomic insertions if possible.
• Off-target/immune responses: when moving to primary or in vivo mammalian systems, evaluate immune sensing of synthetic DNA and vector-related genotoxicity (micronucleus/comet as needed).

Practical 6-week plan (pilot)

1. Week 0–1: In silico design & rank top 30 candidates (evolution + SHAP maps + motif-implant filtering).
2. Week 2: Clone top 12 into luciferase reporters; transfect MEL and MES cell lines (3 biological replicates each) — measure relative activity and specificity.
3. Week 3: Integrate top 6 into stable lines (AAVS1 or lentiviral) and perform ATAC on high/low winners (2 replicates) to check accessibility gains.
4. Week 4–5: For Drosophila: clone 6 top KC candidates into pENTR-GFP, phiC31 integrate (21F), screen larvae/adult brains for cell-type expression; co-stain markers.
5. Week 6: Perturb top designs by mutating predicted repressor sites (e.g., ZEB2) and measure luciferase/ATAC/ChIP shifts to test causality.