LipidMS workflow

LipidMS is an R-package aimed to confidently identify lipid species in untargeted LC-DIA-MS. It combines a set of fragmentation and intensity rules with a parent and fragment co-elution score PFCS), which is calculated in predefined retention time windows. Depending on the MS evidence reached by the identification function survey, LipidMS provides three levels of structural annotations: i) subclass level, e.g., PG(34:1); ii) fatty acyl level, e.g., PG(16:0_18:1); and iii) fatty acyl position level, e.g., PG(16:0/18:1).

LipidMS Overview

LipidMS annotation is based on coelution between parent and daughter ions, what means that peaks coming for the same molecule will have the same retention time (RT) and a good coelution score (based on Pearson correlation). As a general rule, parent ions will be found when no collision energy is applied, while fragment ions will be found when it is. Each lipid class has characteristic ionization and fragmentation properties that allow to filter informative fragments among all coeluting ions to reconstruct the parent’s structure. Next figure summarizes the basics of LipidMS.

To execute identification functions, LipidMS needs a named list of two data frames (peaklist and rawScans) for each collision energy applied. First data frame need to have 4 columns: m.z, RT (in seconds by default), int (peak intensity) and peakID, while the second one requires an extra column named Scan, which will inform about the scan (order) to which each observation belongs. The first data frame contains peak extracted information and the second, the raw data for each MS scan. These data are used to define parent and fragment peaks in order to calculate the PFCS for each lipid identification in a determined RT window.

We propose to convert vendor files to mzXML format, split them by collision energy (CE) and then, process these files separately using LipidMS (dataProcessing function) or any other package/software to obtain the required tables.

In case other software is employed for processing and raw scans data are not available or no coelution score wants to be applied, just the peak table data frame for each MS function have to be provided. The coelution score calculation may take long times of computation, but results improve substantially.

Files convertion

After MS acquisition using MSe/All ions/DIA mode, we have to get one peak table for each collision energy employed. As most peak picking tools handle only MS1 as input and do not allow to treat different CE separately, we propose to obtain as many files as different CE and process them individually as if they were MS1.

Vendor files can be converted to mzXML format using MSConvert (proteowizard), but their separation by CE is platform-dependent. When CE information is kept in the mzXML files (i.e. convertion from .d files from Agilent), they can be split using the sepByCE function.

library(LipidMS)
sepByCE(input = "mix_pos.mzXML",  output = "mix_pos_sep")

# to convert a batch of files you can use the following code:
files <- dir()[grepl(".mzXML", dir())]
outputs <- unlist(lapply(sapply(files, strsplit, ".mzXML"), "[[", 1))
outputs <- paste(outputs, "_sep", sep="")
mapply(sepByCE, files, outputs)

Otherwise, if CE information is not kept in the mzXML file, alternative procedures will have to been followed. Here we show an example for .raw files from a Waters Synapt G2-Si Q-TOF. In this case, no CE information is obtained when converting to mzXML format, and lockspray scans are mixed with MS1 and MS2 functions. We propose to 1) remove lockspray function files (three files in total), 2) convert to mzXML, where MS1 and MS2 scans will be alternative, and 3) separate mzXML files. The following code can be employed to save time.

## 1) remove lockspray function

# first, you will need to set your working directory where all you raw files are saved. Then run the following code:

folders <- dir()[grepl("raw", dir())]

for (f in folders){
  files <- dir(f)
  unlink(paste(f, files[grep("FUNC003", files)], sep = "/"))
}


## 2) convert to mzXML format using MSConvert

## 3) separate mzXML files

files <- dir()
files <- files[grep(".mzXML", files)]
outputs <- paste(unlist(sapply(files, strsplit, ".mzXML")), "_sep", sep="")

for (f in 1:length(files)){
  nCE <- 2  ## change if you have more than 2 remaining functions after removing lockspray
  lines <- readLines(files[f])
  scans <- grep("    <scan num", lines)
  length(scans)
  runs <- grep("</msRun>", lines)
  for (i in 1:nCE){
    pos <- seq(i,length(scans), nCE)
    lines2write <- c(1:(scans[1]-1))
    for (x in pos[1:length(pos)-1]){
      if (x == scans[length(scans)]){
        lines2write <- append(lines2write, c(scans[x]:c(runs-1)))
      } else {
        lines2write <- append(lines2write, c(scans[x]:(scans[x+1]-1)))
      }
    }
    lines2write <- append(lines2write, c(runs:length(lines)))
    write(lines[lines2write], file=paste(c(outputs[f], as.character(i), ".mzXML"),
                                         collapse=""))
  }
}

Data Processing

Once all files have been obtained, peak picking and deisotoping have to been performed. This can be done using dataProcessing function from LipidMS, which requires enviPick and CAMERA packages, or any other software/package. For further details see help(dataProcessing, package = "LipidMS").

ms1 <- dir()[grepl("fullMS.mzXML", dir())] #fullMS or any other nomenclature employed for MS1
ms2 <- dir()[grepl("sep40.mzXML"), dir()] #sep40 or any other nomenclature employed for MS2
data_ms1 <- sapply(ms1, dataProcessing, msLevel = 1, polarity = "positive")
data_ms2 <- sapply(ms2, dataProcessing, msLevel = 2, polarity = "positive")
head(data_ms1[[1]]$peaklist)
head(data_ms1[[1]]$rawScans)

This function will return a list with two data frames (peaklist and rawScans) for each file.

Lipid Annotation

LipidMS contains a total of 32 functions aimed to annotate lipid species: 30 class and polarity-specific functions (i.e. idPGneg) and two general functions (idPOS and idNEG) for ESI+ and ESI+, respectively. Class-specific functions allow to customize fragmentation rules, while general identification functions execute all functions for a given polarity sequentially using the predefined rules.

General annotation functions

If predefined fragmentation rules are convenient for your analysis, the easiest way to run the annotation step is to use idPOS or idNEG for ESI+ or ESI- data, respectively. This two functions will run all class-specific functions for the given polarity. The output will be a list with two data frames: the results table, which contains information for each annotated lipid, and the annotatedPeaklist table, which links the original MS1 data and the results table, and provides information for each feature.

pos_res <- idPOS(MS1 = data_ms1[[1]], MSMS1 = data_ms2[[1]], ppm_precursor = 10, 
                 ppm_products = 10, rttol = 10, coelCutoff = 0.8)

Then, you can use pos_res$results and pos_res$annotatedPeaklist to see the results.

Class-specific annotation functions

A more customizable option is to use the class-specific functions for lipid identification. These functions allow you to change fragmentation and intensity rules. In addition, they provide more detailed information about the fragments found for each identified lipid. For further information see the documentation page for each function.

MS1 <- data_ms1[[1]] # CE = 0
MSMS1 <- data_ms2[[1]] # CE > 0
ppm_precursor <- 10
ppm_products <- 10
rttol <- 10
dbs <- assignDB()

# example code for idPEpos function
pe <- idPEpos(MS1 = MS1, MSMS1 = MSMS1, 
          ppm_precursor = ppm_precursor, 
          ppm_products = ppm_products, rttol = 6, 
          chainfrags_sn1 = c("mg_M+H-H2O", "lysope_M+H-H2O"), 
          chainfrags_sn2 = c("fa_M+H-H2O", "mg_M+H-H2O"),
          intrules = c("mg_sn1/mg_sn2", "lysope_sn1/lysope_sn2"), 
          rates = c("3/1", "3/1"), intrequired = c(T),
          dbs = dbs, coelCutoff = 0.8)

# additional information about how to change rules is given in the documentation 
# of the following functions: chainFrags , checkClass, checkIntensityRules, 
# coelutingFrag, combineChains and organizeResults. These functions could be also
# empoyed to build customized identification functions.

The output of these functions is a list with sevaral data frames that contain: annotation results of lipids supported by fragments in MS2 (pe$results), all feasible candidates for the given lipid class based only on MS1 (pe$candidates), class-specific fragments found (pe$classfragments) and chain-specific fragments (pe$chainfragments).

To obtain similar tables than the ones obtained with idPOS, you can run the following code:

MS1 <- data_ms1[[1]] # CE = 0
MSMS1 <- data_ms2[[1]] # CE > 0
ppm_precursor <- 10
ppm_products <- 10
rttol <- 10

# example to customize several id functions
results <- vector()
results <- rbind(results, idLPCpos(MS1 = MS1, MSMS1 = MSMS1, 
          ppm_precursor = ppm_precursor, 
          ppm_products = ppm_products, rttol = rttol)$results)
results <- rbind(results, idLPEpos(MS1 = MS1, MSMS1 = MSMS1,
          ppm_precursor = ppm_precursor, 
          ppm_products = ppm_products, rttol = rttol)$results)
# in this case you should add the rest of functions for annotation in ESI+

# Once you have the complete results table, you can cross it with the MS1 original peak table.
annotatedPeaklist <- crossTables(MS1, results, ppm_precursor, rttol)


# To see results:
View(results)
View(annotatedPeaklist)

# if you want to see just the features annotated by LipidMS:
View(annotatedPeaklist[annotatedPeaklist$LipidMS_id != "",])

High customization of LipidMS

Data bases

By default, LipidMS data bases (for each lipid class), are based on the combination of the following chain building blocks: 30 fatty acyl chains and 4 sphingoid bases, which were selected based on their biological relevance. If you want to add or remove any of the building blocks, createLipidDB function can be employed to rebuild the data bases of interest.

fas <- c("8:0", "10:0", "12:0", "14:0", "14:1", "15:0", "16:0", "16:1",
"17:0", "18:0", "18:1", "18:2", "18:3", "18:4", "20:0", "20:1", "20:2",
"20:3", "20:4", "20:5", "22:0", "22:1", "22:2", "22:3", "22:4", "22:5",
"22:6", "24:0", "24:1", "26:0")
sph <- c("16:0", "16:1", "18:0", "18:1")
dbs <- createLipidDB(lipid = "all", chains = fas, chains2 = sph)

# to use for identification function two additional data frames need to be added
dbs$adductsTable <- LipidMS::adductsTable
dbs$nlsphdb <- LipidMS::nlsphdb

If just some DB need to be modified, you can use the following code:

fas <- c("8:0", "10:0", "12:0", "14:0", "14:1", "15:0", "16:0", "16:1",
         "17:0", "18:0", "18:1", "18:2", "18:3", "18:4", "19:0", "20:0", "20:1",
         "20:2", "20:3", "20:4", "20:5", "22:0", "22:1", "22:2", "22:3", "22:4",
         "22:5", "22:6", "24:0", "24:1", "26:0")
newfadb <- createLipidDB(lipid = "FA", chains = fas)
dbs <- assignDB() # This function loads all DBs required
dbs$fadb <- newfadb$fadb # Then, you can modify some of these DBs

Adducts

LipidMS uses specific adducts for each lipid class and polarity. All the adducts searched must be included in the adductsTable data frame that is within the package data. In case you want to use an adduct that is not included, you will need to add it:

adductsTable <- LipidMS::adductsTable
adductsTable <- data.frame(adduct = c(adductsTable$adduct, "M+X"), 
                          mdiff = c(adductsTable$mdiff, 52.65), 
                          charge = c(adductsTable$charge, 1), 
                          n = c(adductsTable$n, 1),
                          stringsAsFactors = F)

Once included, this adduct can be used when calling the identification function:

# The new adductsTable has to be also uploaded in the dbs list.
dbs <- assignDB()
dbs$adductsTable <- adductsTable

idPCpos(MS1 = LipidMS::mix_pos_fullMS, MSMS1 = LipidMS::mix_pos_Ce20,
MSMS2 = LipidMS::mix_pos_Ce40, adducts = c("M+H", "M+Na", "M+X"), 
dbs = dbs)

Fragmentation and intensity rules

For a higher customization of LipidMS rules, different arguments of the identification functions can be modified:

clfrags, clrequired and ftype are employed by checkClass function to confirm a given lipid class. This function allows three different types of fragments: fragments with a specific m/z as for example 227.0326 for PG in negative mode, which needs to be defined as clfrags = c(227.0326) and ftype = c(“F”); neutral losses such as the head group of some PL (i.e. NL of 74.0359 in PG in negative mode), which will be defined as clfrags = c(74.0359) and ftype = c(“NL”); or building blocks resulting from the loss of some groups, as for example, PA as M-H resulting from the loss of the head group (glycerol) in PG in ESI-, which will be defined as clfrags = c(“pa_M-H”) and ftype = c(“BB”). The last two options could define the same fragments. In this case just one of them would be necessary. When using the third type of fragment (“BB”), the building block will be specified in lower case (i.e. pa, dg, lysopa, mg, etc.) and the adduct will be given as it appears in the adductsTable, both separated by “_“. Names for the building blocks are the ones used for the LipidMS databases without the”db" at the end. In case the presence of a fragment indicates that the candidate does not belong to the lipid class (i.e. loss of CH3 in PE, which corresponds to a PC actually), this will be specified by using clrequisites = c(“excluding”).
chainfrags argument are employed by chainFrags function to look for specific fragments that inform about the chains of a given class. For example, in the case of PG subclass, the chain in sn1 position is identified by the LPG as M-H resulting from the loss of the FA chain of sn2; and the chain in sn2 position is identified as the free FA chain as M-H. These two fragments need to be searched in two different steps: in the fist step we will look for LPG coeluting with the precursor using chainfrags = c(“lysopg_M-H”); then, we will look for FA chains using chainfrags = c(“fa_M-H”). This information can be combined later using combineChains function. To indicate the fragments to be searched, the class of lipid is writen using the same names as the LipidMS databases without the “db” at the end (i.e. pa, dg, lysopa, mg, CE, etc.), and the adduct has to be indicated as it appears in the adductsTable, both parts separated by “_“. In case some chain needs to be searched based on a neutral loss, this can be defined using”NL-" prefix, followed by the database and the adduct. If this neutral loss is employed to find the remaining chain, “cbdiff-” prefix allows to calculate the difference in carbons and doubles bounds between the precursor and the building block found. For example, “cbdiff-dg_M+H-H2O” will look for DG as M+H-H2O and then, it will return the difference between their number of carbons and double bounds and the ones from the precursor. On the other hand, “NL-mg_M+H-H2O” will look for fragments coming from the loss of MGs. In case these fragments identified as losses from the precursors are going to be employed for the intensity rules, this same prefix has to be added. If a chain is calculated based on the difference of total number of carbons and double bounds between the precursor and a previously searched chain, chainfrags argument must be must be a character vector c(“”) and the candidates data frame and the chain fragments list must be provided.
intrules, rates and intrequired arguments are used by checkIntensityRules function to confirm chain structure and position. Taking PG subclass as an example, intensities of lysoPG fragments (informative for sn1) can be employed to confirm the chains structure (intrules = c(“lysopg_sn1/lysopg_sn2”)). In this case, the intensity of the lysoPG resulting from the loss of the FA chain in sn2 is at least 3 times greater (rates = c(“3/1”)) than the lysoPG resulting from the loss of the FA chain in sn1. For the intrules argument, “/” will be use to separate the fragments related to each chain (i.e. 16:0/18:1), and “_" will be use to indicate the list in which they (sn1 or sn2) will be searched. This will depend on the chain fragments rules defined previously. If you only search for LPG in chainfrags_sn1 but you want to compare in both positions, you need to use “_sn1" in both cases. For classes with more than one FA chain, if some intensity rule should be employed to identify their position but they are no defined yet, use “Unknown”. If it is not necessary because the fragmentation rules are informative enough to define the position (i.e. sphingolipid species), just leave an empty vector.

If you have any further questions, please do not hesitate to contact us at: maialba@alumni.uv.es or maribel_alcoriza@hotmail.com