In Vivo CRISPR Screens - Usage Guide

Overview

Decision-grade design and analysis of in vivo CRISPR screens. Covers the bottleneck math (cell number limits force focused libraries); Manguso 2017 immune-evasion methodology; Chen 2015 tumor screens; CRISPR-StAR stochastic post-engraftment sgRNA activation (Uijttewaal 2025 Nat Biotechnol 43:1848) for escaping early bottlenecks; syngeneic vs xenograft vs PDX choice; per-animal clonal variability; tumor-explant DNA extraction; per-animal meta-analysis; in vivo CEGv2 calibration limitations.

Prerequisites

conda install -c bioconda mageck   # not on PyPI
pip install pandas numpy scipy
# Optional: focused library annotations from Manguso 2017
# Tumor genomic DNA extraction kit (proteinase K + column purification)

Required inputs: - Animal cohorts (n=10+ per condition for hit calling) - Focused library (3,000-15,000 sgRNAs covering 500-3,000 genes) - Cas9+ cell line (selected by FACS before infection) - Plasmid pool sequencing as baseline - Per-animal tumor DNA + sgRNA amplification primers

Quick Start

Tell the AI agent what to do: - "Design a focused in vivo CRISPR library targeting immune-evasion biology following Manguso 2017 methodology: 2,000 kinases + surface proteins + immune factors, 4 sgRNAs/gene" - "Compute bottleneck math: my B16 syngeneic model can take 2M cells; pick library size for 100x coverage maintainable through bottleneck" - "Apply CRISPR-StAR so a genome-scale library survives the in vivo bottleneck: activate sgRNAs after engraftment for paired internal controls" - "Run MAGeCK MLE on my in vivo screen with animal-as-batch covariate; output per-condition beta scores after batch adjustment" - "Diagnose: why is my in vivo CEGv2 PR-AUC only 0.45 despite passing all in vitro QC?"

Example Prompts

Library Design for In Vivo

"Design a focused library targeting immune-evasion biology in syngeneic B16-OVA mouse melanoma. Include: kinases, cell surface proteins, immune-regulatory genes (Manguso 2017 selection criteria). Total ~2,000 genes x 4 sgRNAs = 8,000 sgRNAs. Verify coverage with 2M cells implanted (= 250x cells/sgRNA before the engraftment bottleneck)."

"For a syngeneic colorectal cancer model with maximum 3M cells implantable: design library at 5,000 sgRNAs (= 600x effective coverage). Pick genes by pathway relevance (metabolism, immune, proliferation)."

CRISPR-StAR

"Set up CRISPR-StAR (Uijttewaal 2025) screen with tamoxifen-inducible CreERT2 sgRNA activation in a syngeneic model. Cells infected with library at MOI 0.3; implanted in 5 mice per condition; tamoxifen induction Day 5 post-implant activates the sgRNA in ~half of each clone for paired internal controls; tumor harvest Day 21. Compute expected coverage maintenance vs a conventional in vivo screen."

Per-Animal Analysis

"Run MAGeCK on each animal vs plasmid pool separately; meta-analyze with Stouffer's Z; identify genes consistent across animals. Compare to single combined-animal analysis."

"Run MAGeCK MLE with animal as batch covariate. Output per-condition beta after batch adjustment. Compare to per-animal RRA + meta-analysis."

Diagnostics

"In vivo CEGv2 PR-AUC is 0.45 despite in vitro on same line showing 0.85. Diagnose: context-specific essentialome, clonal dominance, or Cas9 selection failure?"

"My screen has 100x effective coverage at endpoint but only 60% of library detected. Investigate: bottleneck at engraftment or PCR amplification?"

Cross-Validation

"Validate top in vivo hits in vitro (matched cell line, no animal context). Identify hits that are in-vivo-specific (require tumor microenvironment) vs cell-intrinsic."

"Arrayed validation of top 10 in vivo hits with n=10 mice each. Confirm consistent effect across biological replicates."

What the Agent Will Do

Determine model: syngeneic vs xenograft vs PDX
Calculate bottleneck-adjusted library size: max cells implantable × target coverage = total sgRNAs
Design focused library following Manguso 2017 conventions
Plan animal cohort: n=10+ per condition for hit calling
Use CRISPR-StAR if genome-scale library needed (CreERT2 sgRNA activation in half of each clone + post-engraftment induction)
Verify Cas9+ cell selection before infection (FACS or selection marker)
Sequence plasmid pool as baseline
Implant cells; allow tumor growth 12-21 days
Harvest tumors; extract DNA via proteinase K + column purification
PCR amplify sgRNA cassette + sequence
Run MAGeCK count + MLE with animal as batch covariate (or RRA per animal + meta-analysis)
QC: in vivo-adjusted thresholds (CEGv2 PR-AUC >0.4 acceptable; lower than in vitro)
Per-condition hit calling; cross-validate with in vitro
Output: per-condition gene effects, per-animal consistency, arrayed validation plan

Tips

The single most common failure: trying to use a genome-wide library (70k+ sgRNAs) in vivo. Even at 10M cells implanted, this gives 140x coverage which collapses to 20-50x at endpoint after bottleneck. Use focused libraries unless implementing CRISPR-StAR.
For immune-targeting screens, syngeneic models are mandatory; xenografts have impaired immunity.
Per-animal clonal dynamics drive enormous inter-animal variability. Use n=10+ animals per condition; meta-analyze across animals rather than treating as single experiment.
Cas9 selection before implantation is non-negotiable. Cas9-negative cells persist with their sgRNA but no editing, diluting all signal.
In vivo CEGv2 calibration is poor because in vitro essentialome doesn't capture tumor-microenvironment biology. Use cell-type-and-context-specific reference essentialome (a matched in vitro screen in the same cell type).
CRISPR-StAR is the modern approach for genome-scale in vivo: hold sgRNAs inactive until after engraftment, then activate (tamoxifen/CreERT2) in half of each clone for paired internal controls. Uijttewaal 2025 (Nat Biotechnol 43:1848) reports intrinsic per-clone control and outperforms conventional screens in therapy-resistant melanoma models.
Multiple animals per condition is more important than depth per animal. n=10 mice at 100x coverage is better than n=3 at 500x.
Tumor heterogeneity arises during growth; sample multiple regions or pool whole-tumor DNA.
For metastasis screens, each metastatic site is a separate selection event; analyze per-site.

Decision Cheat Sheet

Model	Use case
Syngeneic	Tumor-immune interaction, checkpoint biology
Xenograft	Human cancer cell-intrinsic biology
PDX	Patient-specific drug testing
Humanized mouse	Tumor-immune in human context

Thresholds

Threshold	Value	Rationale
Cells per animal (syngeneic)	1-5M typical	Tumor model dependent
Library size for focused in vivo	3,000-15,000 sgRNAs	Maintainable coverage
sgRNAs per gene in library	4-6	Standard
Coverage at endpoint	>50x; ideally 100-200x	Bottleneck-adjusted
Animals per condition for hit calling	10+	Inter-animal variability
In vivo CEGv2 PR-AUC	>0.4 acceptable	Lower than in vitro
Days to harvest	12-21 days post-implant	Time for selection

Validation Checklist

[ ] Library size matches bottleneck math
[ ] Cas9+ cells selected before infection
[ ] Plasmid pool sequenced as baseline
[ ] n ≥10 animals per condition
[ ] Per-animal sequencing depth ≥100 reads/sgRNA
[ ] MAGeCK MLE with animal-as-batch covariate
[ ] Cross-validation with in vitro screen
[ ] Arrayed validation of top hits in matched cohort

crispr-screens/library-design - Focused library design
crispr-screens/mageck-analysis - MAGeCK MLE with batch covariate
crispr-screens/hit-calling - Per-animal meta-analysis
crispr-screens/screen-qc - In vivo-specific QC
crispr-screens/batch-correction - Animal cohort as batch
crispr-screens/combinatorial-screens - In vivo combinatorial
crispr-screens/copy-number-correction - Cancer-line in vivo
pathway-analysis/go-enrichment - Functional analysis