Experimental Design – Controls
Key point – Your experiment is only as good as your controls
Key point – Control for everything that could influence the interpretation of your data
“Controls” are the extra things that you build in to your experimental design to help you interpret and increase confidence in your data. The term “control” can mean different things to different people and is often used interchangeably for the following situations.
Positive and negative Experimental controls – to help with troubleshooting
Positive and negative Biological controls – to confirm your experiment “works” by checking that the results you obtain are really what you think they are; that a positive result is really positive and a negative result is really negative
Rule out alternative interpretation(s) – control for third or confounding variables
Calibrate the system – internal normalisation to control for biological variation (including loading controls)
In many experiments you will actually include more controls than test samples and usually you can only trust your results if your controls work. I’ll deal with the above points in turn.
Positive and negative Experimental controls
Most lab based experiments can (and will) go wrong in one way or another and you will want to know why. Therefore you should build in experimental controls along the way to help you trouble-shoot and adapt your protocols.
You should include:
A sample that will give you a positive result in the experiment
Usually this will be a sample that you have prepared prior to your main test experiment and which you know works. This could be a protein/DNA/RNA that you isolated on a separate occasion and have already tested. The point is that the more you know about this sample the better. You should know what to expect and therefore you will know there is a problem if you don’t get the result you expect.
Ideally you should have a control for each step where something can go wrong. Your positive biological control (next section) can serve as an experimental control but that might be a bit risky in multi-step experiments.
A sample that will give you a negative result at each stage of your experiment.
In practical terms the negative controls are usually easier as most often you can set them up at the same time as your main experiment. Again, controls at every stage are important. Often these are as simple as using a blank well in your plate assay or water in place of one key reagent (eg instead of the template in your PCR to control for contamination of any of your other buffer/enzyme solutions). An experimental negative control can occasionally be the same as your biological control but again, they are serving a different purpose: the stringency on biological controls should be highest.
The number and specifics of each control will depend on what you are actually doing. The point here is that you want to have some measure of success at every stage of your multistage process so that you can spot not only that there is a problem but also where any problem is.
Example; let’s say I want to do a qPCR experiment comparing the abundance of two mRNAs between cells after treatment with a drug or not treated. This experiment has multiple stages – cell growth, drug treatment, RNA isolation, RT-PCR and qPCR for both genes (don’t worry if you don’t know what these mean). If I ran this whole experiment and got no data back, I wouldn’t know where the problem was. So, in my experimental design I will think about what can go wrong and how will I know if it has happened. I will be able to look at my cells so I can tell if they are sick/dead or whatever – probably no extra control needed. The drug treatment bit is a bit different (will deal with that in 2. below). However, the RNA isolation could go wrong in a purely experimental reason – old reagents or inappropriate protocol for my cell type or whatever – so I would include a positive control of an RNA I already knew is OK (in addition to testing the RNA I had isolated for integrity, conc etc). Next the RT-PCR step could fail for some reason (eg PCR machine not working), so I could add a cDNA that I know is OK to my qPCR step. Now when I run the qPCR if my samples give no signal but my known cDNA and RNA samples are OK, I know the problem is early in the system but if the cDNA works and the RNA doesn’t then I know it’s the RT-PCR step that has failed.
2. Positive and negative Biological Controls
Your experiment is only as good as its controls (I know, I’ve said this already, but it’s important). Whereas you experimental controls are there to troubleshoot, biological controls are required to validate that your results are real. They are therefore critical in determining how good your experiment is!
key point – you can’t prove a negative without a positive control
The point is, that whenever you do an experiment you need something in there to show you that it would have been possible to see if a difference if a difference existed. I think this is easier to appreciate with examples.
Example 1 You are testing a new drug that you think will increase cell proliferation but you aren’t sure yet. Your experimental results might show that it doesn’t have any effect but how will you know that it was even possible to see an increase? Here you would include an already known compound that increases cell proliferation alongside your test sample. Your positive control, your known compound, will tell you whether an increase was possible or not. Technically you could use this sort of positive as a calibrator/comparator.
Example 2 Let’s say you want to prove that your cell line doesn’t express a particular protein on its cell surface. To do so, you could run a flow cytometry experiment and show that when you probed your test sample with antibodies against your protein you got the same results as when you probed the test sample with isotype matched non-specific antibodies. All well and good, but you can’t say for sure that something didn’t go wrong, that your antibody worked etc so you need to include a positive control – a cell line that you already know expresses your sample. In this case this control would be suitable as both a biological and experimental control.
key point – you can’t prove a positive without a negative control
The inverse of positive controls; you need to be able to prove that your result is really what you think it is. There are different levels of biological control and you should aim to be the most stringent you can. Consider everything that could give you a positive result and think of ways that that result could occur for reasons other than what you have set out to do. Then design in controls to account for those potential issues. Example time again…
Example 1 Let’s say you are planning a western blot to compare expression levels of a protein that you care about. You do your blot and get some bands, how do you know which of if any of the bands are the thing you are looking for? Your negative control helps. Options; a sample where you have genetically depleted (or, less good, overexpressed) your target protein, less good, a sample from a different species that you are confident that your antibody does not recognise. In fairness, for a western you have the advantage of knowing the size of your target protein so you might be able to make a reasonable inference. If you were doing flow cytometry, you wouldn’t even have that benefit.
Some of you might be thinking about isotype or secondary only controls. They are useful to show that the signal you receive are due to your primary antibody but they don’t tell you about the specificity of your primary for its real target.
Example 2 Treatment controls. Let’s say you are treating your sample (be it cells, mice, human whatever) with a compound diluted in some solvent (DMSO is quite common. The compound could have an effect and so could the solvent so you will need a control treatment with just the solvent. Often it is this solvent treated sample that will be your true comparison group. This sort of scenario goes for transfection reagents, viral particles etc etc.
3. Rule out alternative interpretation(s)
Remember how I commented about designing out third/confounding variables? Well this is it.
Back to the common sense/biology understanding thing again. Be really critical about the data that you might acquire to try and think of every other potential reasons why you might get the same outcome and think how you might control for it.
Example Studying if training at altitude has an impact on VO2 max I would go into the experiment thinking the age range, gender and BMI of my participants are clear confounders while level of exercise in general, exercise frequency and altitude to which they hill climb would all be relevant. I could assign my participants randomly to altitude training vs non-altitude training groups and then account for demographic differences in analysis or I could stratify my groups so that they proportionately reflect the study population. Both could work, what would be best would depend on study size and breakdown and likely the best option would be to stratify for some putative confounders and account for differences through analysis for the others.
4. Controls to calibrate the system
The fourth use of controls is to provide an internal measurement against which your findings can be compared. These can be the positive controls described above or could be experimental negative controls such as empty wells in an ELISA, or untreated cells in a flow cytometry experiment, from which background signal can be obtained and then subtracted from all your test samples.
You may also require internal normalisation. For example you might require reference transcripts (usually more than one) for qPCR/miRNA expression-type experiments or control proteins for western blotting (though total protein stains are generally preferred). What you use depends on the experimental set-up, likely you will need a pilot experiment to establish that your internal reference is not affected by the assay conditions. As with other parts of your experimental design, you ultimately will have to justify your decision if you attempt to make a broader interpretation. At very least, you will state your internal normalisation approach in your methods and figure legends.
The point to remember is that if you are going to need these things don’t forget to plan them in to your experimental set up!
One more time for luck; controls are critical to being able to interpret your experiments the way you want to. Check, check and triple check you have everything before you begin otherwise you will end up running the whole thing again.
Depending where you came from you might want to go
Or, if you came to this page directly you might want to start at the beginning of our experimental design series – here.