Your Western blots deceive you!

Ever since the technique to blot protein samples on nitrocellulose membrane was developed, Western blot has been a go-to technique for every molecular biologist. The method is simple enough and the results seem straightforward, but correct interpretation of your results is a crucial and often overlooked step.

In Figure 1, I present an example of a straightforward Western blot result. For this experiment, I differentiated the myeloid cell line PLB into functioning granulocytes by stimulating them with dimethylformamide (DMF) for 6 days . Afterwards, I took a sample of my stimulated and control cells as well as a sample of primary human granulocytes (PMN), dissolved these in sample buffer, ran the samples next to each other on an SDS-PAGE gels, blotted the gel to a membrane and probed that membrane with antibodies against the NADPH-oxidase components p67 and p47. Finally, I probed the blot with fluorescently labelled antibodies and scanned it on a Li-Cor Odyssey infrared scanner to detect the signal. As you can see, the neutrophils (PMN) express high levels of both p67 (in red) and p47 (in green) whereas the control PLB cells express nothing and the treated PLB cells express levels comparable to the neutrophils, demonstrating that the experiment was successful. I also probed the blot for β-actin to demonstrate equal loading of the samples. This is a perfectly straightforward result: the protein is either there or not there at all. No discussion about it! However, most blots won’t be so straightforward.

Figure 1. Western blot probed for p67phox (red) and p47phox (green) or actin (lower panel). Loading control on the far left lane with molecular weight bands as indicated in kDa. A sample of human neutrophils (PMN) is shown as a control as compared to non-differentiated PLB cells and PLB cells differentiated with 0.5% DMF.

In most cases, your protein of interest won’t be either there or not there. The level of expression will be reduced or increased by a certain margin. Western blots can be very deceptive in demonstrating such subtle effects, as I will outline below.

Controls

First and foremost in scientific practice come the controls. Your control and experiment samples should have received the exact same treatment to be valid and run side by side on a gel for proper comparison. I know it happens far too often that you go through the whole procedure only to discover a nasty spot in either your sample or control lane. Then you go and repeat the whole procedure, only to have the other lane smudged this time. If only you cropped both pictures and spliced them next to each other, you’d have a perfect result… Unfortunately, this is not allowed. Such practice reeks of scientific fraud and makes it impossible to judge whether or not your result was real, even if you mean well. Nope, my friend, either it’s back to the lab, or show the smudges! There’s nothing wrong with showing smudges, one should not be afraid of that. Smudges are much preferred over fraud!

Detection limits

The second issue to consider before stating whether or not a certain protein has disappeared or not (for example after siRNA treatment) is the detection limit of your protein of interest. The problem with all commercial antibodies is that nobody really knows just how strongly they interact with their target protein. Anybody who’s tried several different antibodies against the same target knows that some will give you a very strong signal and low background, while others will give you a weak signal and high background. Anything in between, in any possible combination, can also be encountered. It’s completely random! This indicates that the strength of the signal on your blot is not so much an indication of relative protein abundance as of antibody quality. Though, of course, protein abundance also helps.

Detection limit becomes especially relevant when your signal is very weak. This means you’re drawing conclusions close to the detection limit of your assay. Thus, when comparing two lanes on a blot, it may seem that in one lane the protein is there and in the other it’s absent, while in reality the difference is no more than 10% or so. What makes the blot deceptive is the detection limit of your protein. It could just happen that you loaded 10% less of your experimental sample and that this is enough to make your protein of interest undetectable, even though it’s still there. If you now compare the control and your experimental lane, it will look like your protein of interest has disappeared completely after treatment and since we usually use highly abundant proteins, such as actin, as our loading controls, this 10% difference in loading will go unnoticed. I see these kind of results quite often with siRNA controls; very weak signals that appear to show a black and white difference. Results collected on old fashioned X-ray film are particularly sensitive to this form of deception, since a pixel on the film is either black or not. The grey values on a film are entirely derived from pixel density. In contrast, results collected with a fluorescence imager are much less sensitive to deception, since every pixel can have a wide range of values. On the Li-Cor Odyssey, for example, a pixel can have a value between 1 and 40,000, providing a huge dynamic range.

Post-processing errors

Every digital image can be manipulated to show only what one desires others to see. Of course, manipulating only part of the image (say the control lane) and not the rest is fraud, but there’s a huge grey area of manipulation that is allowed, but not quite correct. I provide an example in Figure 2. For this experiment, I tried to knock down a gene with siRNA. The blot shows my control sample (Ctrl) and two different siRNA’s (1 and 2). Panel A is what the blot actually looks like after I acquired the data with the Odyssey infrared imager, panel B is manipulated to show what I want to see (protein levels decreased after siRNA treatment) and panel C is my loading control (HSP90). As you can see, the effect of siRNA treatment appears to be much greater in panel B than in panel A, even though both show the exact same blot. What I did to generate panel B was to adjust the image display curves, rather than merely the contrast. You should never, ever do that! Ever. Because by adjusting the curves you’re discarding data you don’t like. You’re telling the program that you don’t care about values above or below a certain threshold, thus you get rid of pixels with very low or very high values. Of course, this eliminates your background, but you also lose information that might be valuable, such as the weak bands visible in panel A below and above the main bands. In addition, you enhance and multiply a small difference to make it seem much greater. This might indeed help you get your work published in Nature, wherein most blots are suspiciously squeaky clean, but it really isn't the way to go.

Figure 2. Control sample (Ctrl) and samples treated with siRNA 1 or 2. A) unprocessed image, B) processed image, C) loading control (HSP90). 

Loading controls

Another obvious problem is with the loading control. As I mentioned before, we tend to choose a highly abundant protein as our loading control, such as Actin. However, if your control is much more abundant than your protein of interest, the result may be highly deceptive. I provide an example in Figure 3. For this experiment, I simply took a cell lysate and diluted this with sample buffer in 10%-steps (thus, lane 1 is 100%, lane 2 90%, 80% etc.). Then I probed the blot with an anti-actin antibody, followed by a fluorescently-labelled antibody and I scanned the blot with the Odyssey Infrared scanner. The nice thing about this technology is that I can actually quantify my signal and draw a graph, as shown in the figure. Now, you’ll notice that the difference between the first 5 lanes is very hard to spot with the naked eye, even though there’s an almost 2-fold difference in loading (100% vs 60%)! In addition, the computer can barely tell the difference between the first 3 lanes, even though I loaded only 80% of the sample in lane 3. Thus, also the intensity of your loading control can be very deceptive and you should be aware that the same phenomenon occurs for every protein you blot for. If you’re trying to draw conclusions outside the linear range of detection, whether by eye or computer, you’re going to be deceived.   

Figure 3. Decreasing amounts of sample were loaded on gel with steps of 10%. The blot was probed for human Actin and data acquired on the Li Cor Odyssey.

Finally, the molecular weight of your controls and experimental samples matter. Not every protein in your sample is going to transfer equally well and molecular weight is an important determinant for transfer efficiency. Large proteins tend to transfer slower whereas small proteins can actually be transferred straight through your blot if your transfer time is too long. Thus, ideally, your loading control and protein of interest should be of both similar size and similar abundance.

Solutions

It’s up to you to provide the right controls for your experiments and to make sure your samples and controls are on the same blot. You can not, and should not, compare samples from different blots as they may have been differently exposed and processed.

You should know the detection limit of your protein of interest and the affinity of your antibody. When you first start using a new antibody, run a control blot with a dilution range of your sample to make sure you’re measuring in the linear range of your protein.

Image manipulation is allowed, of course, you’re already doing it when you’re exposing your film for different lengths of time. However, only manipulate the whole blot and never discard data. Better to show some noise and background. Find some useful hints and tips on this website. Nowadays, I much prefer using fluorescence scanners, such as the Odyssey, the acquisition Western blot data because the scanned imaged can be quantified and I can easily make different exposures of my blot, simply by setting the exposure level. When using the Odyssey, always set the acquisition for the maximal possible exposure, without over-exposing the image (shown as white pixels in the colour view or blue pixels in the grey scale view). That way, you collect as much data as possible and you can always post-process your image to get a prettier picture. More important than a pretty picture, however, is the ability to graph and calculate your data.  

I hope these examples and guidelines give you some idea of how to interpret Western blot results, be it your own data or published work.