CRISPR screens
A lentiviral library of guides against every gene, several per gene, delivered once into Cas9-expressing cells; selection acts, and guide abundance before and after is compared by sequencing. The default format and the cheapest way to interrogate a genome.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Pooled knockout
A lentiviral library of guides against every gene, several per gene, delivered once into Cas9-expressing cells; selection acts, and guide abundance before and after is compared by sequencing. The default format and the cheapest way to interrogate a genome.
How it differs from the standard method
- Knockouts are complete but mosaic: in-frame indels leave some cells functional.
- Essential genes are hard to study because the cells carrying those guides die.
- Hit calling collapses several guides per gene to one value, which is the main defence against off-target effects.
1The library is the experiment
In an ordinary editing experiment you make one change and measure what happens to the cells. In a screen you make tens of thousands of different changes at once, each in a different cell, and measure nothing about any individual cell.
Cells per guide
Each guide needs hundreds of independent cells carrying it, maintained at every passage. Lose that and chance, not biology, decides which guides drop out.
One guide per cell
Infect at a multiplicity around 0.3, so most infected cells carry exactly one guide and a phenotype can be attributed to it.
A matched reference
Freeze a sample when selection starts. The result is a change in abundance, and no single time point contains a change.
2What the numbers look like
The raw data are counts: a few hundred reads for each of maybe 77,000 guides, at two time points. Counts are normalised to the total, logged, and expressed as a log2 fold change per guide; the guides belonging to one gene are then collapsed to one value per gene.
3How it fails
- Too few cells per guide: sampling noise at every passage swamps real effects, and the hit list does not reproduce.
- Too much virus: cells carry several guides, so phenotypes are smeared across unrelated genes and every effect is diluted.
- No selection for infected cells: uninfected cells make up most of the pool and dilute every fold change.
- A genotyping PCR from too little genomic DNA: the counts then measure the PCR tube rather than the cells.
- One guide per gene, or one infection: nothing internal tells you whether a hit is a hit.
4Cutting is not the only option
A catalytically dead Cas9 still binds DNA; fuse a repressor to it and the gene is turned down rather than broken, fuse an activator and it is turned up. Neither touches the sequence, so there are no frameshifts, no in-frame escapers and no indel lottery.
Key terms
Practise these as flashcards →Independent cells carrying each guide, kept up at every passage. 300–500× is a usual target; it decides the sampling noise.
Virions per cell. At MOI 0.3 about a quarter of cells are infected and ~86% of those carry a single guide.
The readout per guide: its normalised count at the end divided by its count at T0, logged.
A guide matching nothing in the genome. A thousand of them define the distribution a guide follows when the gene does nothing.
Growing the pool for a fortnight so that small fitness differences compound into measurable losses of abundance.
Killing nearly everything, so survivors are enormously enriched. Huge fold changes, but a brutal sampling bottleneck.
Dead Cas9 fused to a repressor or an activator: expression is changed, the DNA is not cut.
The noise floor times the hit threshold. Testing 19,000 genes pushes the threshold out to roughly 4.5 standard deviations.
When to use it
- Finding which genes a phenotype depends on when you have no candidate list, only a selectable or sortable readout.
- Mapping which genes a drug needs to work, by screening for resistance or sensitisation.
- Interrogating essential genes or non-coding regulatory elements, where CRISPRi's partial, cut-free knockdown is the only workable perturbation.
- Following up a pooled hit list gene by gene in an arrayed format, where the phenotype can be an image or a secreted protein rather than survival.
Compared with alternatives
Reach for single-gene Cas9 editing once a screen has given you candidates: a screen ranks genes, it does not characterise one.
Prefer base or prime editing when the question is a specific variant rather than which gene matters; screens of pegRNA or base-editor libraries exist but are far harder to run.
An RNAi screen knocks mRNA down transiently and tolerates essential genes, but suffers worse off-target noise; CRISPRi largely replaced it.
The readout of a pooled screen is a short-read run on a single amplicon, so library quantification and depth decisions belong there.
Sorting on a fluorescent reporter turns a pooled screen into one with an arbitrary readout, by comparing guide counts between sorted bins.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
A lentiviral library of guides against every gene, several per gene, delivered once into Cas9-expressing cells; selection acts, and guide abundance before and after is compared by sequencing. The default format and the cheapest way to interrogate a genome.
- Knockouts are complete but mosaic: in-frame indels leave some cells functional.
- Essential genes are hard to study because the cells carrying those guides die.
- Hit calling collapses several guides per gene to one value, which is the main defence against off-target effects.