You should also consider the possibility that perhaps one of your labels didn't work properly. Unfortunately, I see this far too often. It may be related to the sample (ie. wrong pH), or it may just be that particular label doesn't work properly for whatever reason. This does occur and it is common.

I always test the iTRAQ reagents by labelling BSA prior to my real samples. Even though I do this, I still have problems on occasion where one label doesn't work properly. Go through your MS/MS spectra and have a look at the reporter region. Do you see one label is always lower than the others? If so, this may be why. If this does occur, you would see down-regulation for that sample, not the up-regulation you are reporting, but it's still worth checking.

I'd also suggest you do a MASCOT search without any iTRAQ modifications specified. This will give you an idea of how successful the iTRAQ labelling was. ie. you should get very few matches without any iTRAQ labels specified.

Dear Kin,

Are you sure that these "housekeeping proteins" proteins are not up-regulated due to the conditions of your sample? (I do not know the system you are working on). There are lots of conditions (i.e. types of cancer) that cause the dys-regulation of GAPDH.

Instead if you think that the problem is in the dilution then yes: it should be appropriate to use these ratios to normalize. However, it is common if there is about 10% of error (the ratio is 1.1).

I wondered you could add an internal standard, i.e. you can use a pure protein you have in lab, split it in equal amounts (ratio 1:1 and/or 1:2 or both) label it with iTRAQ and mix it to your samples. You can use these peak areas as reference. Hope it can be useful.

good luck

C.

Dear Claudia,

Thanks for your feedback. To be honest, actually I am quite new to iTRAQ experiment so please bear with me.

I am also aware that GAPDH can be influenced by many conditions. In my case however, in one of my sample (115), not only GAPDH but also beta-tubulin and beta-actin were 4-fold higher than control (114). Still there might be a global change in the protein expression pattern in my sample, but seeing these up-regulation in those traditional "housekeeping proteins" worried me. Maybe because I am still naively seeing LC-MS proteomics in the way of western blot...

Your idea of using internal standard is very helpful. Next time I guess I will spike in some BSA as a quality control.

ok...4-fold increase is really too big, I understand your worry!

If you still have a little sample, you can do it!

Hi Kin,

I also add internal standard proteins in my experiments. I have noticed that if you have sample handling issues, for example, if one sample had significantly more sample than others, normalizing based on the internal standard won't really solve your problem. This is because the amount of your internal standard would be independent of actual amount of proteins in your samples

On the other hand, for other sample prep issues like digestion or labeling efficiency, normalization based on internal standard would help you.

I typically start with normalizing based upon total intensities of each of the iTRAQ channels. The reason for that is that I am able to normalize based upon total protein and at the same time I can also normalize for sample prep and handling.

But this type of normalization also is not really perfect. Therefore, I look at my data to check if it is skewed one way or the other. If it is skewed, I try other normalization approaches such as house keeping genes or internal standards.

You should also consider the possibility that perhaps one of your labels didn't work properly. Unfortunately, I see this far too often. It may be related to the sample (ie. wrong pH), or it may just be that particular label doesn't work properly for whatever reason. This does occur and it is common.

I always test the iTRAQ reagents by labelling BSA prior to my real samples. Even though I do this, I still have problems on occasion where one label doesn't work properly. Go through your MS/MS spectra and have a look at the reporter region. Do you see one label is always lower than the others? If so, this may be why. If this does occur, you would see down-regulation for that sample, not the up-regulation you are reporting, but it's still worth checking.

I'd also suggest you do a MASCOT search without any iTRAQ modifications specified. This will give you an idea of how successful the iTRAQ labelling was. ie. you should get very few matches without any iTRAQ labels specified.

Hi Yanli,

I don't see why you can't incorporate all the ideas, it seems sensible in many ways. If your results show issues with all three approaches, you may be far better off just repeating the experiment as it would suggest there are some fundamental problems.

As for how to normalise. That's the perennial question. Many proteomics papers say they normailse the data. Very few actually say how they normalise the data. Some seem to do very simple normalisation, others tend to baffle you with formulae. Both may be appropriate, depending on the data. Personally, I think this is an area that needs to be clarified in the field.

Hope that was helpful,

Peter

Hi Yanli,

I don't have the papers at hand right now. I've got a lot on at the moment, but I'll see if I can dig out some papers that actually describe their methods properly and post them here. Chase me up if I haven't done something in 7 days...

Peter

Hi Yanli,

I have come across many data analysis approaches for iTRAQ/TMT that use complex normalization procedures. I have pasted below the links to a few:

Addressing Accuracy and Precision Issues in iTRAQ Quantitation: http://www.mcponline.org/content/9/9/1885.short

A Statistical Model for iTRAQ Data Analysis: http://pubs.acs.org/doi/full/10.1021/pr070520u

Statistical Analysis of Relative Labeled Mass Spectrometry Data from Complex Samples Using ANOVA: http://pubs.acs.org/doi/full/10.1021/pr700734f

In addition, you may want to have a look at a seminal paper about normalization techniques for microarray data, “A comparison of normalization methods for high density oligonucleotide array data based on variance and bias.” http://bioinformatics.oxfordjournals.org/content/19/2/185.short

hope this is helpful.

Hi Yanli,

Parimal has a good list, but I'd also add the following. Note that many of these have some hard core (by my statistical knowledge standards anyway) calculations. My eyes do tend to glaze over when I see equations in papers, but they are still worth the read as there is some good discussion in there. You might need a tame stats expert to help with the processing of the data, not too mention someone who can work with R.

A. L. Oberg and O. Vitek. Statistical design of quantitative mass spectrometry-based proteomic experiments. J.Proteome Res. 8 (5):2144-2156, 2009.

Ann L. Oberg and Douglas W. Mahoney. Statistical methods for quantitative mass spectrometry proteomic experiments with labelling. BMC Bioinformatics 13 (16):1-18, 2012.

Douglas W. Mahoney, Terry M. Therneau, Carrie J. Heppelmann, LeeAnn Higgins, Linda M. Benson, Roman M. Zenka, Pratik Jagtap, Gary L. Nelsestuen, H. Robert Bergen, and Ann L. Oberg. Relative Quantification: Characterisation of Bias, Variability and Fold Changes in Mass Spectrometry Data from iTRAQ-Labelled Peptides. J.Proteome Res. 10 (9):4325-4333, 2011.

Ruiyan Luo, Christopher M. Colangelo, William C. Sessa, and Hongyu Zhao. Bayesian Analysis of iTRAQ Data with Nonrandom Missingness: Identification of Differentially Expressed Proteins. Stat Biosci 1 (2):228-245, 2009.

Shelley M. Herbrich, Robert N. Cole, Keith P. West, Kerry Schulze, James D. Yager, John D. Groopman, Parul Christian, Lee Wu, Robert N. OGÇÖMeally, Damon H. May, Martin W. McIntosh, and Ingo Ruczinski. Statistical Inference from Multiple iTRAQ Experiments without Using Common Reference Standards. J.Proteome Res., 2012.

Having listed all that, I still think many publications do a simple normalisation that uses basic, but still valid maths. Something along the lines of we have X proteins that we know are not affected by the treatment. We know they should have a ratio of 1. If they give a ratio of 0.9, they adjust all values accordingly. I see this as valid, but it depends on the sample complexity, how many proteins in their 'X list,' do they really know they don't change etc... The problem with normalisation is that the more you look into it, the harder it gets...

Good luck!!

Peter

I wasn't aware of that paper, but it looks interesting. The fact that they provide a spreadsheet means you can easily do that type of normalisation. Always a plus.

I can't comment on how well it works, but I'd certainly give it a try and see what the data looks like.

Peter