Guide RNA design is a hypothesis

A spacer written beside an NGG is a proposal that Cas9 might cut there, in this genome, in this chromatin. The proposal can be ranked. It becomes evidence only when the alleles are read. This page is about building that proposal for a research experiment and about noticing when the ranking has outrun the cells. Cutting and repair are explained in how CRISPR-Cas9 editing works in research. Reading the result is covered in checking whether a genome edit worked.

Oligonucleotides, guide RNA and the polymerases used to check a locus are reagent classes in the molecular biology catalogue. A written specification belongs with the quote request.

What the design is deciding

The decision is where a nuclease is invited to look, not what the cell will write afterwards. For a loss-of-function experiment the useful sentence names the transcript, the exon and the assembly. For a base substitution the sentence is tighter: the cut, or the editing window, has to land on the bases you hope to change. A guide that is excellent for a frameshift in an early exon can be useless beside a codon in exon 7. Write that sentence before a design tool offers a ranked table.

The sequence you design against has a build identifier. Ensembl, GenBank and the UCSC Genome Browser are public places those assemblies live. A guide drawn on an old build, or on the wrong strain, is a hypothesis about a genome your flask may not contain. If the parental line carries a variant that breaks the PAM, the enzyme will not see the site, and no on-target score will confess the mismatch.

Spacer, PAM and the strand the file prints

Streptococcus pyogenes Cas9, the enzyme most tools assume unless you say otherwise, recognises a PAM commonly written NGG. The spacer is the stretch of about 20 nucleotides immediately upstream of that motif. Design pages usually print those 20 letters as they appear on the PAM-containing strand, with T where the RNA will carry U. The RNA pairs with the opposite strand. Reverse-complementing the export "so that it matches the guide" is a reliable way to order a molecule that cannot bind. Keep the tool's strand note in the same folder as the sequence.

The PAM is part of the site and not part of the spacer RNA. NAG can behave as a weak SpCas9 PAM in some contexts. A search that counts only NGG will miss those neighbours. A search that treats every NAG as equal to NGG will fill the table with sites the enzyme barely cuts. Record which PAM set was switched on.

Other proteins are different hypotheses. A Cas12a protospacer sits next to a T-rich PAM and uses a different scaffold. Staphylococcus aureus Cas9 uses a different PAM and a shorter practical cargo in some delivery formats. Moving an SpCas9 20-mer onto one of those proteins because the gene symbol matches does not move the activity with it.

Mismatches are not equal along the spacer. The PAM-proximal seed, often discussed as roughly the 8 to 12 nucleotides closest to the PAM, is where many guides lose cutting first. Mismatches farther from the PAM are more often tolerated. Specificity models use that gradient. Individual guides break it. A seed mismatch that looks minor on a logo can abolish cutting, and a distal mismatch can still be a real off-target site.

What a numerical score is allowed to do

On-target scores, including the family associated with Doench Rule Set 2 and the ranks shown by tools such as CRISPOR, estimate activity under the conditions the model saw. They do not know your chromatin, your delivery method or your cell cycle. Specificity scores, including MIT and CFD-style models, estimate how unusual the site looks against a mismatch search of a reference. They do not count cuts in a nucleus.

Use the numbers to discard the worst of a long list and to pick two or three guides worth testing. Do not use them to declare a guide free of off-target cuts. Activity at the locus in your cells is the observation the on-target score was guessing about. Off-target measurement is a separate design, set out in off-target checks and their limits.

Truncated spacers of about 17 or 18 nucleotides appear in the literature as one way to reduce some off-target cuts, sometimes with weaker on-target activity. That is a published variant with its own trade-off, not a default to apply because a slider is present. If you shorten the spacer, say so, and do not compare its score with a 20-nucleotide guide as if the models were interchangeable.

Where you place the cut also decides what a later protein claim can mean. An early constitutive exon is a common place to interrupt a coding sequence. An exon the relevant isoform skips is a wasted cut. A lesion in the final exon can leave a transcript that escapes nonsense-mediated decay. If the sentence you want is about protein loss, read indels, frameshifts and protein loss before the exon is frozen.

A path that stops when the reference is the wrong genome

Search only after the transcript is named. Confirm the PAM, the spacer length and the assembly identifier in the output, then look at the locus in a browser so exon edges and annotated variants are visible, not only the score column. If the favoured guide overlaps a variant your cells might carry, sequence that short interval in the parent or choose another site. A hypothesis about GRCh38 is not automatically a hypothesis about the flask.

Include a second on-target guide when the conclusion matters. A non-targeting guide, or a guide already known to cut in these cells, can separate "the reagents never arrived" from "this locus is quiet". The known-cutting guide is also extra nuclease in the culture. Keep it out of any clone you intend to keep.

If no acceptable PAM sits near the base you must change, stop and change enzyme or strategy. Hoping a distant SpCas9 cut will still support precise repair is how a knock-in plan becomes a collection of indels. That repair choice is its own experiment.

If the parental sequence matches and a well-delivered guide still leaves the locus untouched, treat chromatin and the guide identity as open questions. Check the scaffold matches the protein. Check you did not order the complement. Then change guide before you change the whole delivery platform on the strength of one failure.

When a favoured guide does nothing, or does too much

A high score with no scar at the locus has a short list of causes worth separating. The parent may not match the reference. The RNA may be the reverse complement, or the scaffold may belong to another protein. Delivery may have failed, which you can only know if some other readout shows the reagent arrived. The locus may sit in chromatin this cell type does not open, even though a model trained on other cells was optimistic. Less often, the enzyme cuts and the cell re-ligates the blunt ends so cleanly that a bulk indel assay looks empty. A second glance at the score does not distinguish those. Sequencing does.

The opposite failure is a phenotype too large for one locus. The guide may be cutting elsewhere, or the delivery may have stressed the culture. An unedited sister culture and a second guide aimed at the same gene are the comparison. If only one guide produces the phenotype, and a clone whose on-target allele is clean does not, the phenotype is not yet an on-target result.

Guides also fail at repair rather than at binding. Cutting is not a frameshift. Re-ligation, a three-base insertion or an in-frame deletion can leave a polypeptide. The design note should say which of those you will count as success, or a cheerful indel percentage will hide an allele you cannot use.

Institutional boundary

Drawing a guide is computational until RNA or plasmid enters a cell. From then, the work sits under the biosafety and ethics approval that covers the organism, the delivery and the nuclease. This page does not grant that approval, does not clear a diagnostic claim, and does not set out any procedure for editing human embryos or the human germline. A specificity score is not a biosafety argument. If the target sequence itself is hazardous, the design file remains under the rules that govern that sequence.

Tubes, humidity and a spacer you can reorder

A synthetic guide arrives as the specification you wrote: sequence, scaffold or chemical class, and the assembly behind the design. In a humid room, adhesive labels leave small tubes before the nucleic acid degrades. Write the identifier on the tube and in the book. Resuspend and store as the sheet for that material describes. A temperate-city phrase about room temperature is not the temperature of a hot afternoon bench. If a power cut warms a freezer, the sequence file is what lets you reorder the same hypothesis instead of a half-remembered 20-mer.

A shared list of favourite guides is a record only if it stores the genome build, the PAM convention and whether the parent was checked. Two benches can order the same gene symbol and two different spacers.

What a sourcing note should carry

Name the reagent class you have already chosen. That might be a synthetic guide with a stated scaffold, a plasmid that expresses it, or the Cas protein that matches that scaffold. Include the organism, the assembly, the spacer strings and whether you need a chemically stabilised RNA or an unmodified oligonucleotide. Say how the locus will be verified. The CRISPR validation sequencing enquiry reference is a prompt for discussing that sequencing. It is an independent method reference, not a statement that a validation line is being run for you. The wider experimental setting is the molecular biology pathway. Ask whether a quotation is possible against the specification. A gene symbol on its own is not one.