Bio Crispr Screens In Vivo Screens

Version Compatibility

Reference examples tested with: MAGeCK 0.5.9+, MAGeCK-VISPR 0.5.6+, pandas 2.2+, numpy 1.26+.

Before using code patterns, verify installed versions match. If versions differ:

• CLI: mageck --version
• Reference focused libraries: Manguso 2017, Chen 2015, public Addgene aliquots

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

In Vivo CRISPR Screen Analysis

"Design or analyze an in vivo CRISPR screen" -> Account for the dramatic bottleneck during animal implantation and tumor growth; use focused libraries; recover DNA from tumor explants; analyze with bottleneck-adjusted hit calling.

• CLI: mageck count + mageck test for standard analysis
• Special handling: bottleneck-adjusted coverage thresholds; per-tissue per-animal replicate structure

The In Vivo Bottleneck Problem

Why in vivo screens differ from in vitro:

Math: A 70,000-sgRNA library at 500x coverage requires 35M cells in pool. Most syngeneic models can implant 1-5M cells. Result: real coverage is 70x at best; effective coverage at endpoint is even lower after bottleneck.

Solution: Use focused libraries (500-3,000 genes; ~3,000-15,000 sgRNAs) to maintain reasonable coverage despite the bottleneck.

Focused Library Design for In Vivo

Manguso et al 2017 Nature 547:413 established the canonical in vivo CRISPR screen methodology with a focused library:

• 2,398 genes covering kinases, phosphatases, cell-surface proteins, antigen presentation, immune regulation and chromatin remodeling; the 2,368 expressed in the melanoma line were the ones scored
• 4 sgRNAs per gene, delivered as four sub-pools of one sgRNA per gene plus 100 non-targeting controls each (9,992 sgRNAs total)
• Targeted at immune-evasion biology in syngeneic mouse melanoma
• Recovered the known immune-evasion genes Cd274 (PD-L1) and Cd47, and identified Ptpn2 loss as sensitizing tumors to immunotherapy through increased IFN-gamma signaling and antigen presentation

Standard focused-library principles:

1. Gene selection: Define the biology to be tested (e.g., immune evasion, metastasis); restrict library to genes plausibly involved (kinases, surface proteins, regulators)
2. Library size: 500-3,000 genes; 3-6 sgRNAs/gene; total 3,000-18,000 sgRNAs
3. Coverage achievable: With 5M cells implanted, 100-300x coverage is achievable

Public focused libraries:

• Manguso 2017 immune library (Addgene)
• DepMap focused panels for specific pathways
• Custom: order from Twist via CRISPick or CRISPOR

CRISPR-StAR (Stochastic Activation by Recombination; Uijttewaal 2025)

Uijttewaal et al 2025 Nat Biotechnol 43(11):1848 (published online Dec 2024) introduced CRISPR-StAR, which holds sgRNAs inactive until cells have engrafted and re-expanded, then activates each sgRNA in only half the progeny of a clone to generate matched active-vs-inactive internal controls.

How it works:

1. Library is delivered as sgRNAs held in an inactive state, alongside a tamoxifen-inducible CreERT2 recombinase
2. Cells are implanted in animal at MOI 0.3; they engraft and re-expand into single-cell-derived clones (no editing yet); library complexity preserved
3. Tamoxifen induces CreERT2 recombination, which stochastically activates the sgRNA in ~half the cells of each clone
4. Active and still-inactive (wild-type) cells of the same clone, tracked by UMI barcodes, form paired internal active-vs-control comparisons
5. Screen proceeds; tumor harvest, DNA extraction, sequencing
6. Per-clone active-vs-inactive contrast suppresses engraftment and clonal-drift noise

What it buys: CRISPR-StAR enables genome-scale in vivo screens (vs focused libraries) by generating intrinsic per-clone controls; outperforms conventional in vivo screens in therapy-resistant mouse melanoma models (Uijttewaal 2025).

Syngeneic vs Xenograft vs PDX

Decision rule: For immune-targeting drug screens, use syngeneic. For human-cancer cell-intrinsic biology, use xenograft. For patient-specific drug screens, use PDX. Each requires different cell numbers and bottleneck planning.

Tumor DNA Extraction and Sequencing

Goal: Recover sufficient sgRNA-containing DNA from tumor explants for sequencing.

Approach: Dissect tumor; lyse with proteinase K; extract genomic DNA; amplify the sgRNA locus by PCR; sequence on MiSeq / NextSeq / NovaSeq.

# Typical PCR + sequencing parameters for in vivo screens
# Per-tumor DNA: 0.5-5 mg yield from typical syngeneic tumor
# Per-sample sequencing depth: >=500 reads/sgRNA at endpoint (bottleneck-limited libraries need depth, not breadth)
# Multiple animals per condition (n=5-10) to account for clonal variation

# mageck count for in vivo
mageck count \
    --list-seq library.csv \
    --sample-label Plasmid,Animal1,Animal2,Animal3,Animal4,Animal5 \
    --fastq Plasmid.fq.gz A1.fq.gz A2.fq.gz A3.fq.gz A4.fq.gz A5.fq.gz \
    --norm-method median \
    --output-prefix in_vivo_screen

Hit Calling for In Vivo

Goal: Identify per-gene fitness effects despite high inter-animal variability.

Approach: Each animal is a "replicate" with high variance due to clonal dynamics. Use MAGeCK MLE with animal-as-batch covariate, or run MAGeCK RRA per animal and meta-analyze.

# Option A: MAGeCK MLE with batch covariate
cat > in_vivo_design.txt <<EOF
Samples         baseline    tumor   animal_2  animal_3  animal_4  animal_5
Plasmid         1           0       0         0         0         0
Animal1         1           1       0         0         0         0
Animal2         1           1       1         0         0         0
Animal3         1           1       0         1         0         0
Animal4         1           1       0         0         1         0
Animal5         1           1       0         0         0         1
EOF

mageck mle \
    --count-table in_vivo_screen.count.txt \
    --design-matrix in_vivo_design.txt \
    --output-prefix in_vivo_mle

Per-animal RRA + meta-analysis:

import pandas as pd

# Run mageck test on each animal vs plasmid
# Combine with Stouffer's Z method
from scipy.stats import norm

def meta_analyze_animals(per_animal_results):
    '''per_animal_results: list of MAGeCK gene_summary.txt per animal.'''
    merged = pd.concat([df.assign(animal=i) for i, df in enumerate(per_animal_results)])
    grouped = merged.groupby('id')
    meta = grouped.apply(lambda g: pd.Series({   # pandas 2.2+: pass include_groups=False
        'mean_neg_score': g['neg|score'].mean(),
        # clip to keep norm.ppf finite at p=0; negate so positive z = stronger depletion,
        # matching examples/per_animal_meta_analysis.py
        'stouffer_z': -norm.ppf(g['neg|p-value'].clip(1e-10, 1 - 1e-10)).sum() / (len(g) ** 0.5),
        'animals_significant': (g['neg|fdr'] < 0.05).sum(),
        'n_animals': len(g)
    }))
    return meta.sort_values('stouffer_z')

Failure Modes

Clonal dominance from low complexity

Trigger: Implanted cells lack sufficient library complexity; a few clones dominate the tumor. Mechanism: Inter-animal stochasticity in cell engraftment creates founder effects. Symptom: Per-animal hit lists vary dramatically; no genes appear across all animals. Fix: Use focused library to maintain coverage; increase animals per condition (n=10+); use CRISPR-StAR to delay bottleneck.