Are you doing any pre-fractionation? I found since the 4 and 8 plex labels are different sizes that I needed to optimize the pre-fractionation differently. We use Hp-Rp and so the larger 8 plex tags have a big effect on the hydrophobicity of the resulting labeled peptide digest. I really don't know of any other method specifically to do with the mass spec parameters that would drastically increase your spectra I'd's by more than 10 or 20% other than using a faster machine ie. a q-exractive over a velos ...

Our method is very simple post-labeling. Just ion exchange clean-up prior to LC-MS. Due to the expense of the iTRAQ labels, I've not been able to do a rigorous comparison myself, which is why I reference the study made by the Mechtler group. However, experiments that should have given similar results have been surprisingly different, in favor of the smaller labels.

I think there is no "rule" about which isobaric tagging technique is better than the other (provided the chemical reaction is equivalent).

In this paper from Potties (http://www.ncbi.nlm.nih.gov/pubmed/22594965) they reached the opposite conclusion (8-plex > 4-plex).

Did you tried to modify the collision energy parameter? Could it be that the 8 plex needs a slightly higher energy? After all, you don't really care about hog good your backbone fragmentation is, so far you get an acceptable ion score.

in this discussion they were talking about this 

https://www.researchgate.net/post/Peak_clusters_in_HCD_MS2_spectra_of_8plex_iTRAQ_labeled_samples

How much NCE are you using? I would guess something around 30/35 %?

If your using HCD for both identification and quantification when using a Velos MS you should be using around 40-45% NCE.  If your using an orbi XL which has the less than optimal HCD cell then you might want to be using around 40% NCE.  It depends on your acquisition parameters and data analysis pipeline.  Too little energy and you won't get good reporter ions as well as poor identification of the peptide (especially when using HCD only methods).  Too much HCD energy is okay if your using a parallel method with both HCD and CID because you can rely on the CID for your peptide identification.  However, I found that the parallel method really slows you down even on an orbirap xl and I found it much more advantageous to go with straight HCD no matter what machine your using.  Increasing the HCD collision energy for 8-plex may help a bit but i doubt it will increase the amount of identifications by more that 5% and may even have a negative affect if your using an HCD only method.

Ive been hearing good things about this TMT-10plex kit however I have not had a chance to take a look at it.  We may be testing it soon in the lab.  One caveat to this is that thermo is most likely using a Qexactive with an HCD only method for their studies.  In addition they are probably doing a fair bit of pre-fractionation.  In my experience the iTRAQ 8-plex kit has given us nothing but trouble and I regularly recommend steering clear of it unless you plan on spending about 6 months optimizing it.  If you have a 4-plex method thats working I would stick with that and daisy chain your experiments together to increase your biological or technical replicates with a pooled control.

Additionally, I would recommend switching from SCX to a high-pH pre fractionation.  We found fairly large gains in identifications because we were able to take advantage of the hydrophobicity shifts between high and low pH compared to SCX.  Take a peak at my recent paper PMID: 23956148  (you can take 24 fractions mix them down to 12 and still get almost as many proteins and peptides ID's as if you had ran all 24 fractions).  Cuts the run time in half and you don't have to desalt your fractions before running them on the MS reducing the amount of sample loss.

Fabrizio, thanks for pointing out the Pottiez et al. publication. I will read it carefully. The LTQ Orbitrap Velos instrument, which is used to analyze our iTRAQ and TMT samples, is located in a core facility, so I am not familiar with every operating parameter. However, HCD normalized collision energy is usually set to 33, right in the middle of the range you guessed. The discussion you linked was also interesting. I will check the raw data for evidence of partially fragmented label. I'm guessing increased collision energy would push the label fragmentation process toward completion and useful reporter ions. Nevertheless, although iTRAQ-8 samples are more prone to giving peptide-spectrum matches with missing reported ions, the total number of peptide-spectrum matches is typically lower than for iTRAQ-4. By the way, as Derek brings it up, I prefer to use HCD for both reporter ions and identifications. In comparisons between alternating between CID for identification and HCD for reporter ions and  HCD alone, HCD alone was clearly superior in terms of peptide-spectrum match counts.

EDIT: Normalized collision energy was actually 45% for the group of samples that allowed a rough comparison to be made between 4- and 8-plex iTRAQ.

Derek, thanks for your thoughts on HCD normalized collision energy. As mentioned above, the LTQ Orbitrap Velos used for this work is located in our mass spectrometry core facility, so I'm not familiar with all operating parameters. However, I was told yesterday that HCD normalized collision energy was usually set to 33. I'm sure I remember it being set to 40 in the past, so I will have to look into this. I agree with you that collecting both CID and HCD MS/MS spectra for each precursor ion slows things down too much for separate optimization of identification and reporter ion detection to be beneficial. It was also a real pain to deal with merging identifications from one series of spectra with reporter ions from another. I also agree with you that the TMT 10-plex route is very interesting as no separate optimization would be required if you already had a nicely optimized method using TMT 6-plex. Overall, my best guess is that the larger tags may be a problem simply by providing more fragmentation options that provide peaks useful for neither identification nor quantitation.

EDIT: Normalized collision energy was actually 45% for the group of samples that allowed a rough comparison to be made between 4- and 8-plex iTRAQ.

Derek, 

You have run much more isobaric labelled samples than me. Do you use a QC to check labelling performance and fragmentation conditions?

I was thinking about using a BSA digest (or other protein mix) and process it parallel to the original sample. Would this help you identify issues in the whole analytical experiment?