bioinformatics - Mass spectrometry versus western blotting for validation

I have mass spectrometry data (LC-MS/MS) from rat cortices under either drug or control treatments. The results were performed in triplicate (three pairs of rats, drug or control per pair). In addition to some of the bioinformatics analysis I am doing, I will have to validate some of the fold-changes from the mass spectrometry by Western blot in order to have my results published.

Some of the proteins I have chosen to to validate work very nicely by Western blot which captures the same trend as the mass spectrometry. However, other proteins that I have attempted to validate go in the exact opposite direction.

My questions:

1. In general, how accurate is mass spectrometry compared to Western blotting?



2. Is it unusual to have a lot of variability between the results of the two methods? I am worried that I might have to try multiple proteins before I see the results replicated and this seems problematic to me. That leads me to the third question.


3. Which result should I trust?

Other information - Some of the proteins I have chosen for validation come from lower in the list where not as many peptide fragments were detected. In some cases, fragments were detected in only 2 of the 3 biological replicates. I did not chose to validate proteins where the 2 or 3 replicates varied widely in normalized spectral count values so I would expect these readings to be accurate. Additionally, the mass spectrometry part of the experiment was performed by a reputable lab that has filtered erroneous data using FDR cutoffs. There is a high correlation overall between all three replicates for a particular treatment and low standard error between samples across the population so, statistically, I trust the mass spec results.

If you need more information please ask - I don't know what exactly will be useful to answer my questions.

EDIT :

This article discusses the justification for using western blotting to validate mass spec results. It suggests using selected reaction monitoring assays as an alternative. Some of the points made by commenters in this thread are reiterated in this article.

Answer

It is a common practice to prove a result using an orthogonal technique. Like RNAseq followed by qRT-PCR etc.

Western blotting is not a robust technique and cross comparisons are difficult because of difference in the avidities/affinities of different antibodies. So comparisons can be made only with one protein-control pair in different conditions.

LC-MS is more sensitive and less biased(IMO). So the general trick is- don't do westerns for all proteins just report the ones that behave well (say that you "chose" these because they are important). I know this is a wrong practice and I should not be advocating it. For your own scientific validation try it with another MS technique; if you used ESI-Quadrupole/ion-Trap etc then try with MALDI-TOF or iTRAQ. Some luddites will continue to cling to westerns.

I think it is better to not test the proteins that have low peptide counts. See what is known as MA plot. This is frequently used for microarrays. Here M denotes fold change and A denotes total expression in both samples. Don't pick proteins that are low in expression in both samples; they might not be meaningful. For example if you have 2 molecules of X and 100 molecules of Y in control condition and 5 molecules of X and 200 molecules of Y in test; then the fold change in X would seem important; however this change may not be relevant and can be a result of stochastic fluctuation/measurement error. If you have many samples you can see if something is stochastic or not but the limitation is the number of samples.

Note: I am not saying that 2→5 molecule increment should be meaningless. But to know if they have a meaning or not you would require more complex models; better avoid them at this moment.