Forward and Reverse Genetics: Screens, RNAi and CRISPR

A screen in flies that started with no gene names at all recovered 15 loci controlling the pattern of a larva’s segments. A CRISPR experiment starts from the opposite end: you name the gene first, and then find out whether the cell cares. Both are genetics. Which direction you run decides what you are able to find, and it decides which way you can be fooled.

Phenotype first: the forward screen

A forward screen begins with something you can see and works back toward its genetic cause. The order is fixed: mutagenize a population at random, score a defined phenotype, recover the mutant lines, map the lesion, then test whether the candidate gene really causes the phenotype. A locus is a position in the genome where a gene sits, and in a forward screen you often know the locus long before you know what its product does.

The classic example is the 1980 Drosophila screen by Nüsslein-Volhard and Wieschaus. They looked through embryonic-lethal mutants for changes in larval segmentation and reported 15 loci. The phenotypes sorted into classes acting at different scales, which indicated at least three levels of spatial organization in the segmentation process. Earlier, Brenner’s 1974 work in Caenorhabditis elegans characterized about 300 mutants from chemical mutagenesis and defined about 100 genes; in budding yeast, Hartwell and colleagues organized 148 temperature-sensitive mutants into 32 complementation groups. A complementation group is a set of mutations that fail to restore function in combination, which under the standard assumptions marks them as hits in one gene. That logic is the subject of complementation and pathways, and the mapping step is three-point mapping scaled up with sequencing.

The strength of this direction is that it is unbiased about mechanism. You do not have to guess which gene matters. The weakness is that a screen inherits the bias of its phenotype criterion: it finds only what the assay can see. The zebrafish screeners said so explicitly when they qualified their coverage estimate by what their screen was able to discover, and an Arabidopsis study noted that meiotic defects could fall below the detection threshold of a screen scored on fruit length. Mutagenized backgrounds also carry extra variants, so backcrossing and a causal test by complementation or rescue are part of the method, not optional extras. Mutation covers how the mutagens act.

Gene first: knockdown and knockout

Reverse genetics starts from a chosen gene and asks what changing its sequence or expression does. An early targeted mouse version was arithmetic as much as biology. Thomas and Capecchi disrupted endogenous mouse Hprt in embryonic stem cells with a construct that put neomycin resistance into an exon by homologous recombination; about one in a thousand drug-resistant colonies had also lost Hprt function. Germline transmission of a targeted allele came later, for c-abl.

The gentler route is RNA interference, which lowers how much protein a gene makes without touching the gene. In 1998 Fire, Mello and colleagues injected RNA into C. elegans and found that double-stranded RNA produced substantially stronger, specific interference than either single strand, and that the small effective amount suggested a catalytic or amplifying step. A companion study found evidence that the effect was post-transcriptional, targeting endogenous mRNA, with no evidence of any change in primary DNA sequence. By 2001, 21-nucleotide siRNA duplexes were suppressing expression in cultured human HEK293 and HeLa cells. The mechanism belongs to eukaryotic regulation; what matters here is the bookkeeping. Knockdown reduces output and leaves the gene intact; knockout alters the DNA.

That difference sets the controls. A knockdown needs a measurement of the perturbation itself, by immunoblotting or immunofluorescence for the target protein, two non-overlapping reagents against the same gene, and a rescue with a transgene made refractory to the reagent. Even a successful rescue does not formally exclude a phenotype that also requires an off-target effect. A knockout needs its own checks, because an indel is not a guarantee of lost function.

CRISPR came from a bacterial immune system

In 2005 Mojica and colleagues noticed that some CRISPR spacers, the variable sequences between the repeats, matched transmissible genetic elements, which implied a relationship to immunity against foreign DNA. In 2007 Barrangou and colleagues tested that in Streptococcus thermophilus: phage challenge added phage-derived spacers, and engineering the array by adding or removing particular spacers changed resistance, with specificity tracking spacer-phage sequence similarity. In 2011 the Charpentier laboratory found tracrRNA and showed that mature crRNA in the native Streptococcus pyogenes system is made with RNase III and the Cas protein Csn1. The natural guide is two RNAs.

In 2012 the Jinek study fused those two into one chimeric RNA and showed that purified Cas9 cut DNA wherever the RNA specified. A separate 2012 study from Siksnys and colleagues, with Barrangou and Horvath, showed Streptococcus thermophilus Cas9 cleaving DNA in vitro, and that the RuvC and HNH active sites nick opposite strands. Both were biochemistry with purified components. Editing inside cells came in 2013, when Cong and colleagues and Mali and colleagues independently directed Cas9 to endogenous loci in cultured human and mouse cells. The 2012 papers did not edit a genome.

Programming the cut, and what the cell does next

For Streptococcus pyogenes Cas9 the targeting rule is short. The guide’s target-binding segment is 20 nucleotides, the target must sit immediately next to an NGG motif called the PAM, and in the original plasmid assay cleavage fell three base pairs upstream of the PAM. The PAM is not a formality: Cas9 ignored DNA that matched the guide completely but had no nearby PAM. Mismatches are not a clean off switch either, since tolerance in human cells depended on the number, position and distribution of mismatches.

The cut is where your control ends. A double-strand break can enter non-homologous end joining, which needs no template, or homology-directed repair from a template you supply, which is more efficient in dividing cells. End joining is often described as error-prone, and it does produce insertions and deletions, but in a study of paired cuts at 71 sites in mouse and human cells about 50% of end-joining events had been joined accurately. Outcomes are also not random: a survey of more than 40,000 guides on synthetic targets found repair spectra that depended reproducibly on local sequence and on cell line. The repair pathways themselves are DNA repair and recombination.

So cut, mutated allele and functional knockout are three states, and only the third is a genetic result. Here is the arithmetic. You cut exon 2 of a gene in a diploid line with one guide, and sequencing of alleles in the pool gives: unedited 10%, minus 1 bp 35%, plus 1 bp 25%, minus 3 bp 15%, minus 6 bp 5%, minus 7 bp 10%.

frameshifted alleles = indels whose length is not a multiple of 3 = 35 + 25 + 10 = 70%

In-frame edits are 15 + 5 = 20%, and 10% is unedited, which accounts for all 100%. Now assume the two alleles in a cell are edited independently.

fraction of cells with both alleles frameshifted = 0.70 x 0.70 = 0.49

About half the cells still carry at least one in-frame or unedited allele. Frameshifts often trigger nonsense-mediated decay, and in-frame deletions can leave partial function, so a pool like this is a mosaic, not a null. You derive clones and genotype them, or use more than one guide, before you call the phenotype a knockout. Two more cautions sit underneath the table: amplicon sequencing reports only the window you amplified, and long-range assays have revealed large on-target deletions and rearrangements in mouse and human preparations. Editing also provoked a p53-associated damage response and arrest in immortalized human retinal pigment epithelial cells.

Design problem

An siRNA against gene X reduces cell migration. A CRISPR knockout clone of X migrates normally. Design experiments to decide which result reflects X’s function.

Somatic, germline, and who decides

A somatic edit changes cells in one person. A germline edit in an embryo can be passed to descendants, which is why the two sit in different places in every governance document. In November 2018, He Jiankui announced that twin girls had been born after editing aimed at CCR5 in human embryos, work carried out without the review such an application would require.

The published guidance draws the line at reproduction rather than at embryos. The 2021 ISSCR guidelines placed reproductive heritable genome editing in a prohibited category because the methods were not sufficiently safe or efficient, allowed non-reproductive human embryo editing research under robust review and oversight, and required meaningful public engagement before any clinical heritable application. The same year, WHO recommendations emphasized oversight, registries, and equitable access to the benefits of genome editing. Those are questions about consent, access and who speaks for future generations, and no efficiency measurement answers them.

Somatic editing has reached the clinic. In the first reported patients, a person’s own CD34-positive blood stem cells were edited at the BCL11A erythroid enhancer and returned after myeloablation, and fetal haemoglobin rose in both; that approach has since been approved for sickle cell disease and transfusion-dependent beta thalassaemia. It is a transplant procedure with an editing step inside it, and nothing here is clinical advice. For the sequencing and library methods these screens depend on, see recombinant DNA and PCR and sequencing and genomes.