Important links:
Protocol, probe design software, probe concentration calculator.
When we first were developing our variant of single molecule RNA FISH, we often got asked whether we could detect single base differences. And my answer was always no. There are two specificity problems: one is that it’s hard to make oligos that have enough specificity to reliably discriminate a single base difference, and the other is that you pretty much just get one oligo to detect a SNP, and each oligo will bind non-specifically, so how do you know when it’s actually bound to the target?
To address the first problem, my first student Marshall came up with the awesome idea (inspired by the work of David Zhang (Rice University)) of using a “mask probe” to leave only a “toehold” exposed, thereby greatly increasing the specificity of base-mismatch detection, while at the same time providing stability through the full oligo binding once the mask goes away. Great idea!
To the second problem, we used a “guide” probe, which is a conventional single molecule RNA FISH probe, to tell us where to look. Using this, we can determine which SNP FISH spots are legitimately binding vs. which are spurious binding events.
The purpose of this site is to keep a living document with our latest in probe design, FISH protocols and data analysis. Hope it is helpful.
Yes, it works! Well, we think it works around 50% of the time, meaning that if you are targeting a specific SNP, you have a 50% chance of getting really nice data. There’s a bit of a sliding scale here: you might get lower detection efficiency, but that might be fine for qualitative single cell calls of allele-specific expression. Depends on the question.
Well, it’s definitely more complicated than single molecule RNA FISH. Our primary recommendation is to make sure you are very comfortable with “conventional” single molecule RNA FISH first. Repeat after me: get comfortable with single molecule RNA FISH first. If you are having a hard time with that, it is essentially inconceivable that you will be able to get SNP FISH to work. Conversely, if you have pretty good single molecule RNA FISH signals, then we actually think you have a solid chance to get SNP FISH to work–mostly because it means your setup is probably pretty close to adequate. Just requires a bit more patience and fiddling.
Probe design involves designing the following components: SNV detection oligonucleotides (mutant and wild-type), mask oligonucleotides and “guide” probes. Guide probes are relatively straightforward: just use the regular single molecule FISH RNA probe designer and remove any oligos that may overlap with your SNP(s). For mutant and wild-type, we recommend getting a range of mask lengths and potentially multiple SNV detection oligonucleotides. Ultimately, while our computational tools will try and put you in the right vicinity, the final optimization is typically empirical.
Probe design software:
We have probe design software here that will take your sequences and design a series of SNV and mask oligonucleotides. Our probe design software currently exists in two versions. One requires sequence input in FASTA format and the other simply takes an arbitrary sequence input. As a note, you’ll have to be at least mildly comfortable with Python scripts in order to run the design code. The repository contains readme files that delve into the formatting specifics for actually doing the design.
Multiple SNPs:
For genes where you are lucky enough to have multiple heterozygous SNPs, you can make the designs for each of them and pool them together to amplify signal. As a clarification, this can only be done if you have phasing for these SNPs. This is simple in cases where you have crossed two mouse strains and can be significantly more complicated if you have done de-novo sequencing of a human sample. If you do not know the specific phasing for your SNPs, you can still design multiple independent SNP probes to improve the chances of having one of them work. We still do not have a great sense for predicting whether a single SNP probe will work well or not, so having more swings of the bat, if you will, increases the chances of getting a hit.
Dyes:
Unlike regular single molecule RNA FISH, there are some very important dye-specific effects that stem from the fact that you’re trying to detect individual dye molecules. The combinations that we use are Cy3 and Cy5 for the SNV oligonucleotides and Cal Fluor 610, Atto 700 and Atto 488 for guides. Our favorite for the guide is Cal Fluor 610, which you can only get as a “Stellaris” probe set from Biosearch. Atto 700 is okay, but can be finicky in terms of getting good signal, and Atto 488 works for some genes, but this wavelength has the highest background in biological samples and therefore can lead to suboptimal signal quality (and also it is hard to purify the probes very well). You may be asking “why not Alexa 594 or some other nice dye?”. Turns out that two of our other common dyes don’t work for SNP FISH. Alexa 594 has the problem that it always colocalizes with Cy5 instead of Cy3, and Atto 647N is generally attractive and will show spurious colocalization with SNV detection oligos (or it repulses them all, I can’t remember, either way=bad). Either way, the dye thing is definitely a limitation to be very aware of before beginning.
Ordering:
We order all the oligos from Biosearch with 3’ amine modifications for conjugation except for the Cal Fluor 610 labeled guides, which we get preconjugated as Stellaris probes. I think that getting Quasar 570 and 670 for preconjugated SNV detection oligos should work fine.
Have you done regular single molecule RNA FISH first? Are you comfortable with that? FIRST TRY THAT!
Here's the latest protocol that we have put together. Questions welcome, just send us an email and we'll update the doc. It's relatively straightforward, but there are some differences with regular RNA FISH, and it may require a bit more fiddling.
Here's some basic guidelines with deviations from regular single molecule RNA FISH
As a good first guess, we have found that a working stock concentration of 0.3 µM per SNP probe works pretty well. Here's our probe concentration calculator. For masks, we typically use 1.5x the SNP concentration, to allow for error in our nanodropped concentrations and make sure that the masks are in excess to fully bind with the SNP probes (SNP probes without a mask bound are more non-specific). So, for a standard SNP-FISH hybridization, your math would like something like:
WT SNP Working Stock (0.3 µM ) : 1 µL
MUT SNP Working Stock (0.3 µM) : 1 µL
Mask Probe Working Stock (0.9 µM) : 1 µL
Guide Probe Working Stock : 1 µL
Hybridization Buffer : 50 µL
To save time and not overly dilute the hybridization buffer, you can combine both SNP detection probes and masks into a single working stock, that way you only have to pipette once instead of 3 times.
If you have multiple SNP probes per allele
You can continue to scale up the concentration linearly (0.3 µM SNP WT-1 + 0.3 µM SNP WT-2 etc…) until you reach a total concentration of about 3 µM). Once you reach this “max” concentration, simply scale the proportion to each of SNPs. (ex: 3 µM/12 SNP probes = 0.25 µM per SNP).
This concentration guide serves as a good starting point, but it may be necessary to change the concentration depending on the specific binding properties of your SNP probe, so don’t overly fuss about being 100% precise here. In our experience, our results are relatively invariant as long as you are in this ballpark concentration range.
Optimization can definitely help you get better detection efficiency, but in our experience, optimization will rarely take you from a probe with very poor hybridization efficiency to very good efficiency. We don’t really know why some probes are good and some are bad, but we can sometimes turn okay signal into good signal. As we mentioned earlier, you’ll want to order a couple of different mask lengths. Different length masks mean different length free overhangs (what we term toeholds). We have performed SNP FISH on toeholds varying from 5 to 13 bases in length. The general principle is that a longer toehold (shorter mask) will lead to increased detection efficiency but potentially at the cost of specificity. As a good first guess, toehold lengths between 8 and 10 work best, but this may depend on your specific gene.
To make sure that dye-specific effects are not messing up your signal, you’ll want to always couple in pairs. For instance, if the wild-type SNP probe is in Cy3 and the mutant SNP probe is in Cy5, you’ll want to couple a reciprocal pair where the mutant probe is in Cy3 and the wild-type probe is in Cy5. Both experiments should give you the same result.
When possible, if you can test cell lines with different genotypes, that does wonders for estimating false binding rate! (ie: number of mutant binding events observed in a pure wild-type genetic background)
A critical control involves introducing a random pixel shift and then performing colocalization. SNP FISH’s specificity relies on the fact that a spurious spot single-oligo spot is unlikely to colocalize to a guide by random chance. To quantify this explicitly, we add about 1.25 μm to the x and y coordinates of the guide spots, and perform colocalization again. This random colocalization rate should ideally be below 5%, but can be higher if your gene of interest is very densely expressed within the cell (i.e. something like GAPDH). Keep that in mind when considering the statistical power for the scientific question at hand.
If you have multiple SNP probes, one of the best controls is to test two SNP probes independently and show that you obtain the same allelic bias in both cases. As a note, two SNPs may not necessarily agree when one has a very poor detection efficiency while another has a good detection efficiency.
First thing we would do is just do regular single molecule RNA FISH and increase the exposure time a lot (like 5-7 seconds) and see if you start to see single molecule spots showing up. You may have to adjust contrast because the intensity will definitely be lower than the regular single RNA molecules. Nevertheless, you should be able to see nice clean (though dim) spots. Then do SNV FISH!
Yes, these are very helpful for aligning images.
Autofluorescence can be a real issue because the method depends on colocalization, and autofluorescence often shows up in multiple channels. We recommend always taking images in the GFP channel as well (in case you’re not using Atto 488) because autofluoresce often shows up there. Look for colocalization of blobs across the GFP channel into other channels.
We use standard exposure times for the single molecule RNA FISH (2-3 seconds max), but then use 5-7 seconds for the SNV detection oligos. You may have to play with this to get spots.
We recommend imaging fresh samples under 24 hours post-FISH wash. Store in the dark at 4C in between imaging.
Remake your glucose oxidase solution. Cy5 is very prone to photobleaching, and glucose oxidase is sensitive to freeze-thaws, etc.
We have not found significant differences depending on the order of the channels we use to acquire data. One recommendation to keep in mind, however, is that Cy5 is more prone to photobleaching, so it may be better to take it earlier rather than later. The order we tend to use is: Cal Fluor 610 (guide), Cy5, Cy3, other channels, and finally DAPI.
Here’s our software package. It’s complicated, but will repay the effort–you can play with the included raw data to get a feel for it. Basically, you have to set a threshold for the guide probe, then thresholds for the SNV probes. For the guide probes, the threshold should hopefully be quite clean because it’s conventional single molecule RNA FISH. For SNV probes, the threshold will typically not be clean, especially if it’s a single dye for a single SNP. We recommend setting this threshold relatively low to let a lot of spots through. While this will raise the rate of random colocalization somewhat, it can really boost your detection efficiency.
Beyond just generating numbers, quantification is important to get a sense for how well the technique is working, for which there are a few metrics:
Detection efficiency
What percentage of guides colocalize with a SNV detection probe? We consider an experiment “good” if this is around 40-50%. If it’s down in the 10-20% range, that’s not great, although could be fine for qualitative things. We have seen as high as 70% or more, but that is probably not the norm.
Three color spots
Occasionally, you will get spots that colocalize with both SNV detection oligos (especially if you are doing multiple SNPs). This should be pretty low, especially with the masks, like well under 5%. Higher numbers indicates a lot of cross-reactivity. When you have multiple SNP probes, you can compare the relative intensities of the two SNP channels in the three color spot in order to potentially classify them as either wild-type or mutant. The intensity magnitude cutoffs will of course depend on the number of SNPs used, but the general idea is that if you have 10 SNP probes binding to a transcript, and 9 of them are MUT probes and 1 is wild-type, you are pretty safe in assuming that your transcript is from the mutant allele and that the one wild-type SNP is a false positive binding event
False positive rate
There is going to be some rate of false positive detection. This can come from two sources: one is the wrong SNV probe binding, and the other is spurious colocalization. For the former, the only real way to get at this is to use a genetically homogeneous sample and look for binding of the “wrong” probe. For the latter, we use a “pixel-shift” control. Basically, we shift the SNV oligo channel image over a few pixels and measure colocalization rate. Any colocalization here must be due to random colocalization, and so will give you a sense for a false positive rate. Again, these are typically around 5%.
Dye swap
As mentioned, just to eliminate any potential dye-specific effects in detection efficiency, we often run a dye swap where we swap the dyes for the two SNV detection oligos. It’s comforting, but honestly, we’ve never noticed a difference with this.
If you are getting good signal from your guide probes but not getting great signal from your SNP probes, here is a simple diagnostic algorithm for common problems. If you are not getting good guide probe signal, please look at the standard single molecule RNA FISH FAQs. You need to make sure regular single molecule FISH works robustly before attempting SNP FISH!
Cy5 is a dye that photobleaches very quickly if you are not using the proper anti-fade reagents. If your image stacks start out with decent Cy5 signal that disappears as you traverse up the stack, then you almost certainly have a photobleaching problem. Even if you don’t see that, you probably still have a photobleaching problem. Make sure you making anti-fade to exact specifications and keep in mind that the glucose oxidase is sensitive to freeze-thaw (it stops being able to do its job after about 20 freeze thaws), so make new anti-fade reagents and see if the problem persists!
In general, we have found that Cy5 single oligo spots are 25-50% brighter than Cy3 oligo spots, though these specifics may vary for your specific filter cubes.
Different cells and tissues have different levels of background and this can present a serious challenge to SNP-FISH, as detecting a single oligo can be challenging in highly autofluorescent samples. Sometimes, cellular stress leads to higher levels of background, so if possible, try culturing cells very carefully to see if that improves signal. More of than not, this will not help too much, and you will instead need to apply to some tissue clearing techniques to help improve your signal. We have found most success by washing with an 8%SDS solution in PBS before adding the hybridization solution to our sample, but this is definitely a tough challenge that we are working on!
Try increasing the mask length!