1) Biological samples are typically pretty heterogeneous in terms of cell types.

2) Samples change over time, particuarly PTMs (like phosphorylations) due to residual cellular enzymatic activity.

3) Relative Ionization efficiency changes in MS due to the presence of small amounts of contaminating ions in the sample that may change from sample to sample. You can partially control for this by referencing the abundances of every peak to an internal standard in all spectra before doing your analysis.

4) MZ Accuracy varies over time and if the "analysis window" isn't right peaks can shift apart and look like differnent peaks. Remember precision and accuracy are two different things.

5) At any mz, there will be overlapping chemical noise (or cross-comtamination of isotopic series) at every resolution, which can give false positives and negatives after centroiding.

Method Mass Spectral Data Processing

6) Peak detection and isotopic alignments in the time domain are very problematic by traditional chromatographic peak detection methods. This causes wide shifts in retention times for peaks and sometimes divides peaks in two or three.

Method Isotopic Vector Angle Analysis in Mass Spectrometry v2

7) The intrinsic variability of peak abundance can be almost an order of magnitude between samples, even by the most careful mass spectrometrist. This means up- or down- regulation of particular peaks has to be pretty large (almost yes/no) before you can claim any statistical confidence, particularly for low abundance species.

8) Furthermore, statistics in MS aren't parametric since noise is uni-directional in mass spectra (see above reference). Therefore, you shouldn't use traditional t-test, ANOVA, of PCA analyses on MS peak abundances and errors without considerable modification.

9) This is further compounded by the shear number of features that you are measuring in most LC/MS analyses. Basic chemimetric theory states that you need to analyze a minimum of 3 independent samples per feature being measured to have any statistical confidence in the results. When you are measuring 3000 features in an open peptidomics LC/MS spectra, that means nearly 10,000 independent samples are required before you can have any statistical confidence in any result. No one wants to hear that because they can't afford the cost. This is why Beyond Genomics failed as a company. They spent $20M running 20 disease model mice and 20 control mice through a complete genomic, proteomic, and metabolic profile, generating 300,000 features and tried to sell this approach to every Pharma company. All the Research directors said nice work, but how much did that cost? When they were told $20M their response was a universal, "Are you F-ing crazy?"

Hi Lorena,

When you say " only a very low percentage of proteins are repeated " do you mean the proteins identified or the PTM site occupancy?

Which software are you using to identify your proteins?

Is there a way you could show me the raw data of those replicates?

Hi Lorena, generally the reproducibility of technical replicates for the same sample on a QE orbitrap should be very high. Are you running your replicates in tandem with each other (that is, back to back)? Or are you freezing your sample in the same autosampler vial and then re-running the thawed sample at a later date?

How does the overall LC-MS trace look between the replicates, for instance if you pull up the baseline MS1 signal across the entire run? Do they look generally similar? For more detail, you could also pull up some XIC's for trypsin cleavage peptides (or whichever protease you use) and qualitatively look at the peak shape and retention time between replicates. If either the baseline MS1 or trypsin XICs deviate in retention time by more than 5% despite using the same LC gradient, or the overall signal/peak shapes look qualitatively different, this suggests a pump problem, a column problem, or an ionization problem which could account for difference in your data.

Is your PTM on a mis-cleaved peptide? Do you see an unmodified version of that peptide in the same run. If you don't detect an unmodified version and the modified peptide is also low abundant/intensity, it could have been a spurious MS/MS call in that one initial replicate (not necessarily a problem with the instrument but more of the peptide mapping software you're using). To help improve the signal quality, you could try to program your method to specifically target that modified peptide to acquire more MS/MS.

Thanks,

Barry

Thank you very much for your answers

did you find the problem?