Bench philosophy: CRISPR design tools
Ask the Virtual CRISPR Experts
by Steven Buckingham, Labtimes 06/2014
Finding the optimal target sequence is crucial for successful gene editing experiments with CRISPR or TALENs. Online web-tools may facilitate the probe design.
The perfect CRISPR online tool should have a lot of functions and should be easy to use − just like John S. Holler’s early multi-tool with 100 functions, including a revolver.
Earlier this year, we at Lab Times told you about the exciting ways you can use the new CRISPR technology to edit your favourite genome (Lab Times 1, see page 44). Remember how excited we were? No wonder. This fast, reliable and versatile technology has made genome editing easier than ever before. Following its arrival, just a few years ago, CRISPR has now made itself comfortable and at home in the standard lab toolbox. You would expect a crop of design tools to have started growing out of this, as well, but there are surprisingly few that cater for the non-expert. There are, however, a couple of online tools that we have been keeping our eye on for you.
But first, let’s remind ourselves of how it works. Editing genomes with CRISPR works by exploiting the native CRISPR system, which is a bacterial defence against invading DNA. The CRISPR locus consists of a series of short, conserved repeat sequences separated by variable sequences called spacers. Once the bacterium suspects a piece of DNA may have gained unlawful entry − from a pathogenic virus, perhaps − it responds by chopping it up and incorporating it into the spacers. The spacers are then used as templates to make RNA strands (crRNAs), which effectively guide an attack on invading DNA, directing a DNA-cutting enzyme to the crRNA’s complimentary sequence.
Handy for the bacterium, and handy for the researcher, too − you can hijack it to knock down or even edit a gene in a target cell by injecting the crRNA and its machinery. But as if that wasn’t convenient enough, there is an extra bonus. You can stick all the elements of the machinery into a single guide RNA (sgRNA), drop that into the target cells and it still does its work. That will knock the gene out, or you can introduce arbitrary edits into the genome by combining CRISPR with template-based repair mechanisms, such as non-homologous end-joining or homologous recombination.
What gives this technique the real edge over TALENs (Transcription Activator-Like Effector endoNucleases), the other “Big Thing” in genome editing, is its modularity: to adapt it to a new gene, all you have to do is to build the sgRNA with a different complementary sequence.
But there is a down side (Ah, here we go...). Our readership, experienced as they are in bench-work, will not be surprised to hear that you can mess up, if you don’t know what you are doing!
Don’t get us wrong − these techniques are real “game-changers”, as some would say. It’s just that, like any other technique, there are limitations and pitfalls. For one thing, it is becoming clearer that CRISPR and TALENs aren’t always quite as specific as you might hope, and reports of off-target problems are piling up.
But that’s not all. For various reasons (such as variations in C-G content), some sgRNAs are better at binding their target than others. And even designing the probe in the first place is not always perfectly straightforward. The polymerase you decide to use for the sgRNA synthesis can impose, for example, its own constraints, such as the U6 (polymerase III) promoter forcing you to use GN (glycine followed by any amino acid) for the 5’ dinucleotides. Fussy!
So, to get the experiments to work, you’ll find yourself delving into some secret black arts, drawing on years of experience, navigating constraints, following guidelines. Just think how much could go wrong...
Or if you are lazy (“read sensible”) you can just get a computer to do all the hard work. After all, what computer programmes do is to encapsulate human expertise and make it widely available to the non-expert, as well as save time for the expert. So it’s unsurprising then that there has been a steady growth in computer programmes that help in designing the probe.
But which one to use? The various offerings differ in many ways: the kinds of input they accept, the algorithms they use for predicting the on-site and off-site binding, how fast they are and how easy they are to use, just to name a few.
Let’s start with the more basic online tools. These typically ask for little more information than the sequence you are targeting. Popular options include 'Jack Llin’s gRNA finder' (http://spot.colorado.edu/~slin/cas9.html), the tool by the University of Wisconsin (limited to Drosophila), available at http://tools.flycrispr.molbio.wisc.edu/targetFinder, and the tool from the Drosophila RNAi Screening Center available at www.flyrnai.org/crispr/.
Many commercial suppliers of CRISPR/TALEN-related products offer free design tools on their site, usually with a similarly basic functionality − presumably because they want you to move on to the checkout as quickly as possible. DNA2.0 (www.dna20.com) offer a design tool, which returns a result very quickly, along with ranking of results and some genomic context. However, we could find no information predicting off-target effects. Sigma-Aldrich boast an 'online design tool' (www.sigmaaldrich.com/catalog/product/sigma/crispr) but in reality it is a search engine for their predesigned gRNAs − in other words, let them worry about design, you just fill in the order form.
The CRISPR design tool available at the Broad Institute (www.broadinstitute.org/mpg/crispr_design) is of the “submit and we’ll email you later” variety. It does make off-target predictions but only works for the human genome. And when it comes to the algorithms used, it only confesses, somewhat cryptically, that it “leverages experimental data collected by the Zhang laboratory”.
But perhaps we are going the wrong way about this. Perhaps we should be asking what the perfect CRISPR/TALEN design tool would look like and then see who fits the bill. To make things easier, let’s stick with web-based tools. After all, who uses desktop applications these days? Let someone else take care of keeping the software up-to-date.
First off, a good online tool must be feature-rich, yet easy to use. A contender for first prize in this section would be the “CHOPCHOP’’ system announced by Eivind Valen, a post-doc in Alexander Schier’s lab in Harvard, this summer (Montague et al. 2014, Nucleic Acids Research 12, Web Server Issue, W401-W407). The tool (https://chopchop.rc.fas.harvard.edu) accepts a wide range of input types. You can tell it the gene name or sequence (most tools accept one or the other) and you can work with 12 species. I decided to give it a try with my favourite gene − CHRNA7, the alpha7 subunit of the human nicotinic acetylcholine receptor. The tool returned three alternatives and I chose one at random.
The search took about seven seconds, and I was then faced with a diagram showing the exons/introns annotated with the sgRNAs it thinks I ought to try, along with the results in chart form. Hovering over the chart caused the annotation on the diagram to be emphasised and vice versa. Each entry also included the number of predicted off-target hits. Clicking on one of the results took me to a primer design page, with the primers ranked along with the predicted off-target hits. I just went with the defaults but there are several choices of alternative settings, including the off-target prediction algorithm, whether to target the whole gene or just part of it (including specific splice sites) and which restriction sites. The intuitive interface, coupled with the speed of the search, made it easy to look at the effects of playing around with the parameters.
But CHOPCHOP isn’t able to hog things for itself − it has to head off its rival E-CRISP (www.e-crisp.org/E-CRISP/designcrispr.html), another easy-to-use and feature-rich tool. I found this tool to run slower than the other tools I tested but the results it returned were richer in information. You choose your species and fill in either a target sequence or a gene symbol then hit go. But watch out − you can easily miss the fact that the search result is dropped into an adjacent box, leaving you wondering if the programme is working.
But after that, everything is very straightforward. The user faces several options and is even asked whether the purpose of the experiment is for tagging or knock-down, although it was not clear to me what difference that made. The results page is impressive − you get a table with each hit given a 3-point “SAE’’ (specificity, annotation and efficiency) score and ranked accordingly. A visual diagram follows, and clicking the diagram takes you to the entry in the table (though not the other way around, sadly). The tool is the product of the Boutros laboratory at the German Cancer Research Center, a well-known name in the CRISPR field.
There are many other online tools that also offer some form of visualisation of their results. The CRISPR Optimal Target Finder (http://tools.flycrispr.molbio.wisc.edu/targetFinder) was easy to use but nothing like CHOPCHOP for features and options. However, when I tried it with a fly nicotinic receptor, it returned two results and invited me to evaluate them both with a click of a button (why that extra step, I wondered?). When I did so, I was presented with a simple graph showing the cleavage sites and the ranked results. The programme had predicted no off-targets for one of them and two for the other. It didn’t tell me how it worked it out, though. The tool works for several Drosophila species and a small number of other important invertebrates, such as Caenorhabditis elegans, Apis mellifera and Anopheles gambiae.
Both CHOPCHOP and the CRISPR Optimal Target Finder did their job pretty quickly and the software authors like to boast how fast their programmes work − but is speed really all that important? Perhaps not − even a tool in the crawler lane will hardly count much towards the total time taken by a project. Agreed, the better tools allow you to fine-tune your search by adjusting the input parameters, and waiting a couple of minutes for each iteration would be irritating. But again, this is a drop in the ocean compared to the total project length.
But surely what matters more is how good the tool is at predicting off-target hits. And this is no straightforward matter, given that most researchers agree that there is no single “right” way to do this. There are several algorithms out there; the better tools will provide you with the means to run several and compare their results. CHOPCHOP offers three methods − most others offer one, usually some version of an algorithm called 'bowtie'.
You may want to have an independent prediction of off-target effects using a specialised web tool but there are surprisingly few of them. The 'Cas-OFFinder' offered by Seoul National University (www.rgenome.net/cas-offinder) invites you to enter the type of endonuclease you are using along with the crRNA (CRISPR-RNA). If you are the type that likes to know what is going on “under the hood’’, you can read about the algorithm it uses in Bae et al. 2014, (Bioniformatics 30: 1473 -75). An alternative tool, but one more suitable for experts, is COD (Cas9 Online Designer at http://cas9.wicp.net), although it is not clear to us what algorithm is being used here.
CHOPCHOP and E-CRISP are pointing the way for online CRISPR design tools. CRISPR has become a mainstream method for a wide variety of applications and, for most laboratories, is just a means to an end. Most of us don’t care what algorithms are used for the search for off-targets or for gRNA design. We just want the experiment to work with no undesirable side-effects. We’re not interested in CRISPR, we are interested in answering biological questions. We need tools that do all the expert stuff for us − tools that are easy to use with sensible defaults and a readily understandable output.
Last Changed: 20.11.2014