Mass Spectrometry of Lipids SPPs

From LipidomicsWiki

Contents 1 Input 2 Output 3 Processing 4 Proposals and discussion points 5 Implementation 6 Processing Pipeline

Input

1) Raw MS files of lipid analysis (massspectrometry files).
2) Template file which describes each sample analyzed.
3) List of identified lipids and amount? (in case if lipids are identified by data producers)

Output

Data records in LipidomicNet database.
Raw MS files, sample information, identified lipids, lipid measurements (peak intensities/areas, relative abundance, absolute concentrations, mol.ratios) will be recorded into a database with a data retrieval interface connected to it.

Processing

General information about the experiment
Information from an MS experiment should be submitted to a LipidomicNet server for further recording into a database.

Prior to submission, format of the files, compliance of file entries with nomenclature adopted by LipidomicNet will be checked. Furthermore, quality of experiments should be assessed in order to assure that only high quality data reach the database.
Data recording starts with filling in or selecting already recoded information about the data producer.
In order to assure, that full information about analytical method is available before the data upload, method should be recorded on LipidomicNet WikiPages ().
After method is recorded its identification number (ID) can be used to identify all experiments performed with that method.

Information about samples

The following information about samples analyzed should be provided:
Sample Id, biological species, tissue/cell type, extraction method used etc.
During taskforce VIII workshop, it was decided to use controlled vocabularies (CV) and ontologies developed at EBI. When sample information is uploaded it will be checked for compliance with these ontologies. Data provider will have to edit all entries that don’t match the common format and resubmit the file.

Lipid information
Data providers that identify lipids from MS spectra can upload the list of lipids together with measures of lipid quantity. This file should contain sample id for with a lipid was identified (sample id should be in the list uploaded as described in the previous section). There should be a description of measurement units used and normalization steps performed (such as protein amount normalization, internal standard normalization etc).
Lipid list will be checked against a table of standard lipid names created by members of Task_Force_I

QC
In order to avoid recording and analysis of problematic dataset a quality control system will check results before uploading the data. QC will be evaluated based on reproducibility of lipid signal (if experiment contains replicas), ratios of internal standard and other measures depending on which MS method is used (Category:Lipid_mass_spectrometry).

Lipid identification from MS spectra files
Help to data providers who are interested in assistance with lipid identification will be provided upon request. Due to great variety of analytical MS platforms used by project members it is rather difficult and unpractical to develop a common framework for lipid identification. Instead, a custom solution will be provided based on analytical platform and method used to generate the data.

Proposals and discussion points

Write here any proposals and discussion points you may have.

Jos Brouwers:

1. Several MS data analysis packages are already available. Some of them in the public domain, such as MZmine, XCMS and OpenMS. Is it considered to adopt one of these packages for further development to suit our needs?
2. Sciex also has a software package available (at no cost): Lipidprofiler^TM, based on a MS Acces database for identification/annotation purposes. With very few modifications, the database structure may be usable as a template for identification in to-be-written software (-front-ends).
3. Most of the text I read on SPP pages only deal with MS data. I know that I'm not the only person that finds HPLC indispensable in many analyses.  Am I missing the LC/MS pages or do they still have to be written?
4. Is there agreement within the bioinformatics group about the data format that is most desirable from their point of view? mzXML? netCDF? RAW? Wiff? Or is it all the same (also with respect to multi-dimensional MS)? It might be a good idea to agree on a data format before data collection is started.

Gerhard Liebisch

1. In agreement with Jos, I think we need separate SPPs for different analytical approaches:
   • Untargeted LC-MS
   • Targeted LC-MS - to my opinion no issue because it is not necessary or included in the method protocol
   • Untargeted flow injection MS
   • Targeted flow injection MS - may depend on the MS software (ABI or Micromass?)
2. Untargeted approaches require SPPs different from targeted (we are no experts in untargeted analysis).
3. We mainly use targeted approaches with SPPs leading to quantitative data. These data may be used for further bioinformatic processing. Therefore, I think we may need two different data formats one for primary data from untargeted approaches and another for quantitative data.

Harald Koefeler:

1. I also agree that we need different SPPs for targeted and non targeted approaches
2. For the non targeted approach we will soon use SIEVE, which basically works the same way as MZmine and does data alignment and peak picking. In addition it also calculates up and down regulation ratios. Then we compare the lists of exact masses generated by SIEVE with our internal DB which at the moment covers the exact masses of about 1000 lipids. But all of this procedings are work in progress and we would like to implement MS/MS data. So I don't think this needs to be reflected in actual SPPs right now. In addition this system will be developed mainly for use of LTQ-FT and LTQ-Orbitrap systems.
3. For the targeted approach we could adopt the system used by Gerhard. I think a 4000QTrap with an Agilent 1100 system should cover the instrumental needs. This also fulfills the wishes of Prof. Schmitz who would like to see the same instrumental platforms running all across the LipidomicNet. This setup could be good for multi center clinical trials, as anticipated by Schmitz.
4. Research projects are handled by a TSQ couppled to UPLC (Accela) where we cover the usual triple quad analytics (precursor, neutral loss). With the data generated we go into Xcalibur QuanBrowser where mass and retention time are taken into account. Based on this information peak ID is established and peak areas are calculated. Data normalization to internal standards and sample weight are done in a simple Excel sheet. Lipid classes which don't give good fragmentation can also be run on the Accela LTQ-FT system. The data processing is done the same way as for the TSQ, except that we can utilize the exact mass instead of fragment spectra.
5. I don't see the big problem in data processing, at least for standard requests. I rather see a lot of questions coming up regarding data normalization and the output formate (relative vs absolute amounts). We will really need a good system of internal standards!!
6. As our systems so far are fully running on LC-MS I have no use for the Lipid Profiler, because not taking into account retention times would waste valuable information.
7. I know this is a little off topic here, but when it comes to multi center clinical trials I would rather prefer to distribute the analytes acquired on different labs than distributing the samples and acquiring everything everywhere!

Jos Brouwers:
First of all, let me thank everybody for their contributions so far. Second, let me clear up a misunderstanding that my previous post apparently caused. I do not (!) think we need different SPPs for targeted and non-targeted analysis, despite the obvious differences in experimental set-up. To clarify what I did mean, I made a sort of flow-diagram that shows a possible SPP.

In this diagram, the different steps in data-processing are modular, and depending on the question (or the experimental set-up if you like), some modules can be skipped.

Modules in this first draft are:

1. Peak recognition (m/z direction). This module might also contain a de-isotoping algorithm.
2. Peak recognition (Chromatographic direction). This module should detect peak start and -end in the time dimension. This module should not be necessary when direct infusion is used.
3. MSn data. This module adds information about product ions, or in the case of MRM/SRM, mass transitions.
4. Alignment. This module should align peaks across samples. Particularly important when chromatography of isobaric compounds is involved.
5. Integration. Convert the MS data to a peak list.
6. Compound database. This module should take care of several things: assigning a compound to a peak, and quantify it to absolute amount based on response factors and/or comparison with internal standards.
7. Statistics. I think everybody can fill this in for themselves.

I'm sure that everybody does these steps one way or the other. If I summarize earlier contributions by LipidomicNet members correctly, some never go through the whole process of the flow-diagram, others use commercial software. I also seem to understand that some people have made their own lipid database, although there are also excellent lipid databases in the public domain.

Let me be clear that I'm not saying that people shouldn't be using their own processing procedures. However, I do think that their should at least be a SPP that can fully process data from raw MS data (from whatever platform) to the point where the statistics come in. I have troubles to see any processing procedure that is not (MS-) platform independent, becoming a SPP. Moreover, programs like XCMS and mzMine have already rather advanced algorithms for many of these steps. Shouldn't we adopt these?

Gerhard Liebisch

Sorry Jos, for misunderstanding your suggestions. Although we are no experts in non-targeted LC-based approaches: I think it may not be possible to design one SPP for LC and non-LC approaches because for flow injection or infusion methods you usually do not need a peak recognition and alignment of chromatographic peaks. Therefor a SPP for non-LC approaches is much simpler compared to LC approaches. However, some modules may be very similar for both approaches.

1. Peak recognition (m/z direction) - similar for LC and non-LC
2. Peak recognition (Chromatographic direction) - only LC
3. MSn data - similar for LC and non-LC
4. Alignment - only LC
5. Integration. Convert the MS data to a peak list. - different for LC and non-LC
6. Compound database - different for LC and non-LC

Otherwise, for non-targeted approaches I like the concept of different SPP modules and the idea to use public domain programs like XCMS, mzMine or SIEVE (although we have no experience using these). For targeted approaches you optimize data processing to the steps necessary (which may overlap common SPP modules).

Implementation

Implementation of Mass Spectrometry of Lipids SPPs

Processing Pipeline

Processing Pipeline for Mass Spectrometry of Lipids SPPs

Back to Standard Processing Procedures