BiocStyle::markdown()

Package: r Biocpkg("xcms")
Authors: Johannes Rainer, Michael Witting
Modified: r file.info("xcms-lcms-ms.Rmd")$mtime
Compiled: r date()

## Silently loading all packages
library(BiocStyle)
library(xcms)
library(pander)
register(SerialParam())

Introduction

Metabolite identification is an important step in non-targeted metabolomics and requires different steps. One involves the use of tandem mass spectrometry to generate fragmentation spectra of detected metabolites (LC-MS/MS), which are then compared to fragmentation spectra of known metabolites. Different approaches exist for the generation of these fragmentation spectra, whereas the most used is data dependent acquisition (DDA) also known as the top-n method. In this method the top N most intense m/z values from a MS1 scan are selected for fragmentation in the next N scans before the cycle starts again. This method allows to generate clean MS2 fragmentation spectra on the fly during acquisition without the need for further experiments, but suffers from poor coverage of the detected metabolites (since only a limited number of ions are fragmented).

Data independent approaches (DIA) like Bruker bbCID, Agilent AllIons or Waters MSe don't use such a preselection, but rather fragment all detected molecules at once. They are using alternating schemes with scan of low and high collision energy to collect MS1 and MS2 data. Using this approach, there is no problem in coverage, but the relation between the precursor and fragment masses is lost leading to chimeric spectra. Sequential Window Acquisition of all Theoretical Mass Spectra (or SWATH [@Ludwig:2018hv]) combines both approaches through a middle-way approach. There is no precursor selection and acquisition is independent of acquired data, but rather than isolating all precusors at once, defined windows (i.e. ranges of m/z values) are used and scanned. This reduces the overlap of fragment spectra while still keeping a high coverage.

This document showcases the analysis of two small LC-MS/MS data sets using r Biocpkg("xcms"). The data files used are reversed-phase LC-MS/MS runs from the Agilent Pesticide mix obtained from a Sciex 6600 Triple ToF operated in SWATH acquisition mode. For comparison a DDA file from the same sample is included.

Analysis of DDA data

Below we load the example DDA data set using the readMSData function from the r Biocpkg("MSnbase") package.

library(xcms)

dda_file <- system.file("TripleTOF-SWATH/PestMix1_DDA.mzML",
                        package = "msdata")
dda_data <- readMSData(dda_file, mode = "onDisk")
#' Silently sub-setting the object to speed-up analysis
dda_data <- filterRt(dda_data, rt = c(200, 600))

The variable dda_data contains now all MS1 and MS2 spectra from the specified mzML file. The number of spectra for each MS level is listed below.

table(msLevel(dda_data))

Next we perform the chromatographic peak detection on the MS level 1 data with the findChromPeaks method. Below we define the settings for a centWave-based peak detection and perform the analysis.

cwp <- CentWaveParam(snthresh = 5, noise = 100, ppm = 10,
                     peakwidth = c(3, 30))
dda_data <- findChromPeaks(dda_data, param = cwp)

In total r nrow(chromPeaks(dda_data)) peaks were identified in the present data set.

The advantage of LC-MS/MS data is that (MS1) ions are fragmented and the corresponding MS2 spectra measured. Thus, for some of the ions (identified as MS1 chromatographic peaks) MS2 spectra are available. These can facilitate the annotation of the respective MS1 chromatographic peaks (or MS1 features after a correspondence analysis). MS2 spectra for identified chromatographic peaks can be extracted with the chromPeakSpectra method which returns a Spectra object of MS2 spectra associated with a chromatographic peak (i.e. with their retention time within the retention time window of a chromatographic peak and their precursor m/z within the m/z range of the same chromatographic peak.

dda_spectra <- chromPeakSpectra(dda_data)
dda_spectra

The metadata column "peak_id" contains the identifiers of the chromatographic peaks the MS2 spectrum is associated with.

We next use the MS2 information to aid in the annotation of a chromatographic peak. As an example we use a chromatographic peak of an ion with an m/z of 304.1131 which we extract in the code block below.

ex_mz <- 304.1131
chromPeaks(dda_data, mz = ex_mz, ppm = 20)

A search of potential ions with a similar m/z in a reference database (e.g. Metlin) returned a large list of potential hits, most with a very small ppm. For two of the hits, Flumazenil (Metlin ID 2724) and Fenamiphos (Metlin ID 72445) experimental MS2 spectra are available. Thus, we could match the MS2 spectrum for the identified chromatographic peak against these to annotate our ion. Below we extract all MS2 spectra that were associated with the candidate chromatographic peak using the ID of the peak in the present data set.

ex_id <- rownames(chromPeaks(dda_data, mz = ex_mz, ppm = 20))
ex_spectra <- dda_spectra[mcols(dda_spectra)$peak_id == ex_id]
ex_spectra

There are 5 MS2 spectra representing fragmentation of the ion(s) measured in our candidate chromatographic peak. We next reduce this to a single MS2 spectrum using the combineSpectra method employing the consensusSpectrum function to determine which peaks to keep in the resulting spectrum.

ex_spectrum <- combineSpectra(ex_spectra, method = consensusSpectrum, mzd = 0,
                              ppm = 20, minProp = 0.8, weighted = FALSE,
                              intensityFun = median, mzFun = median)
ex_spectrum

Mass peaks from all input spectra with a difference in m/z smaller 20 ppm (parameter ppm) were combined into one peak and the median m/z and intensity is reported for these. Due to parameter minProp = 0.8, the resulting MS2 spectrum contains only peaks that were present in 80% of the input spectra.

A plot of this consensus spectrum is shown below.

plot(ex_spectrum[[1]])

We could now match the consensus spectrum against a database of MS2 spectra. In our example we simply load MS2 spectra for the two compounds with matching m/z exported from Metlin. For each of the compounds MS2 spectra created with collision energies of 0V, 10V, 20V and 40V are available. Below we import the respective data and plot our candidate spectrum against the MS2 spectra of Flumanezil and Fenamiphos (from a collision energy of 20V).

flumanezil <- spectra(readMgfData(
    system.file("mgf/metlin-2724.mgf", package = "xcms")))
fenamiphos <- spectra(readMgfData(
    system.file("mgf/metlin-72445.mgf", package = "xcms")))

par(mfrow = c(1, 2))
plot(ex_spectrum[[1]], flumanezil[[3]], main = "against Flumanezil",
     tolerance = 40e-6)
plot(ex_spectrum[[1]], fenamiphos[[3]], main = "against Fenamiphos",
     tolerance = 40e-6)

Our candidate spectrum matches Fenamiphos, thus, our example chromatographic peak represents signal measured for this compound. In addition to plotting the spectra, we can also calculate similarities between them with the compareSpectra method (e.g. using the dotproduct method as shown below).

compareSpectra(ex_spectrum[[1]], flumanezil[[3]], binSize = 0.02,
               fun = "dotproduct")
compareSpectra(ex_spectrum[[1]], fenamiphos[[3]], binSize = 0.02,
               fun = "dotproduct")

Clearly, the candidate spectrum does not match Flumanezil, while it has a high similarity to Fenamiphos. While we performed here the MS2-based annotation on a single chromatographic peak, this could be easily extended to the full list of MS2 spectra (returned by chromPeakSpectra) for all chromatographic peaks in an experiment. Respective code facilitating this will be implemented in future.

In the present example we used only a single data file and we did thus not need to perform a sample alignment and correspondence analysis. These tasks could however be performed similarly to plain LC-MS data, retention times of recorded MS2 spectra would however also be adjusted during alignment based on the MS1 data. After correspondence analysis (peak grouping) MS2 spectra for features can be extracted with the featureSpectra function which returns all MS2 spectra associated with any chromatographic peak of a feature.

Note also that this workflow can be included into the Feature-Based Molecular Networking FBMN to match MS2 spectra against GNPS. See here for more details and examples.

SWATH data analysis

In this section we analyze a small SWATH data set consisting of a single mzML file with data from the same sample analyzed in the previous section but recorded in SWATH mode. We again read the data with the readMSData function. The resulting object will contain all recorded MS1 and MS2 spectra in the specified file.

swath_file <- system.file("TripleTOF-SWATH",
                          "PestMix1_SWATH.mzML",
                          package = "msdata")

swath_data <- readMSData(swath_file, mode = "onDisk")
swath_data <- filterRt(swath_data, rt = c(200, 600))

Below we determine the number of MS level 1 and 2 spectra in the present data set.

table(msLevel(swath_data))

As described in the introduction, in SWATH mode all ions within pre-defined isolation windows are fragmented and MS2 spectra measured. The definition of these isolation windows (SWATH pockets) is imported from the mzML files and stored in the object's fData (which provides additional annotations for each individual spectrum). Below we inspect the respective information for the first few spectra. The upper and lower isolation window m/z can be extracted with the isolationWindowLowerMz and isolationWindowUpperMz.

head(fData(swath_data)[, c("isolationWindowTargetMZ",
                           "isolationWindowLowerOffset",
                           "isolationWindowUpperOffset",
                           "msLevel", "retentionTime")])

head(isolationWindowLowerMz(swath_data))
head(isolationWindowUpperMz(swath_data))

In the present data set we use the value of the isolation window target m/z to define the individual SWATH pockets. Below we list the number of spectra that are recorded in each pocket/isolation window.

table(isolationWindowTargetMz(swath_data))

We have thus 1,000 MS2 spectra measured in each isolation window.

Chromatographic peak detection in MS1 and MS2 data

Similar to a conventional LC-MS analysis, we perform first a chromatographic peak detection (on the MS level 1 data) with the findChromPeaks method. Below we define the settings for a centWave-based peak detection and perform the analysis.

cwp <- CentWaveParam(snthresh = 5, noise = 100, ppm = 10,
                     peakwidth = c(3, 30))
swath_data <- findChromPeaks(swath_data, param = cwp)

Next we perform a chromatographic peak detection in the MS level 2 data of each isolation window. We use the findChromPeaksIsolationWindow function employing the same peak detection algorithm reducing however the required signal-to-noise ratio. The isolationWindow parameter allows to specify which MS2 spectra belong to which isolation window and hence defines in which set of MS2 spectra chromatographic peak detection should be performed. While the default value for this parameter uses isolation windows provided by calling isolationWindowTargetMz on the object, it would also be possible to manually define the isolation windows, e.g. if the corresponding information is not available in the input mzML files.

cwp <- CentWaveParam(snthresh = 3, noise = 10, ppm = 10,
                     peakwidth = c(3, 30))
swath_data <- findChromPeaksIsolationWindow(swath_data, param = cwp)

The findChromPeaksIsolationWindow function added all peaks identified in the individual isolation windows to the chromPeaks matrix containing already the MS1 chromatographic peaks. These newly added peaks can be identified by the value of the "isolationWindow" column in the corresponding row in chromPeakData, which lists also the MS level in which the peak was identified.

chromPeakData(swath_data)

Below we count the number of chromatographic peaks identified within each isolation window (the number of chromatographic peaks identified in MS1 is r sum(chromPeakData(swath_data)$ms_level == 1)).

table(chromPeakData(swath_data)$isolationWindow)

We thus successfully identified chromatographic peaks in the different MS levels and isolation windows, but don't have any actual MS2 spectra yet. These have to be reconstructed from the available chromatographic peak data which we will done in the next section.

Reconstruction of MS2 spectra

Identifying the signal of the fragment ions for the precursor measured by each MS1 chromatographic peak is a non-trivial task. The MS2 spectrum of the fragment ion for each MS1 chromatographic peak has to be reconstructed from the available MS2 signal (i.e. the chromatographic peaks identified in MS level 2). For SWATH data, fragment ion signal should be present in the isolation window that contains the m/z of the precursor ion and the chromatographic peak shape of the MS2 chromatographic peaks of fragment ions of a specific precursor should have a similar retention time and peak shape than the precursor's MS1 chromatographic peak.

After detection of MS1 and MS2 chromatographic peaks has been performed, we can reconstruct the MS2 spectra using the reconstructChromPeakSpectra function. This function defines an MS2 spectrum for each MS1 chromatographic peak based on the following approach:

To illustrate this process we perform the individual steps on the example of Fenamiphos (exact mass 303.105800777 and m/z of [M+H]+ adduct 304.113077). As a first step we extract the chromatographic peak for this ion.

fenamiphos_mz <- 304.113077
fenamiphos_ms1_peak <- chromPeaks(swath_data, mz = fenamiphos_mz, ppm = 2)
fenamiphos_ms1_peak

Next we identify all MS2 chromatographic peaks that were identified in the isolation window containing the m/z of Fenamiphos. The information on the isolation window in which a chromatographic peak was identified is available in the chromPeakData (which contains arbitrary additional annotations to each individual chromatographic peak).

keep <- chromPeakData(swath_data)$isolationWindowLowerMz < fenamiphos_mz &
        chromPeakData(swath_data)$isolationWindowUpperMz > fenamiphos_mz

We also require the retention time of the MS2 chromatographic peaks to be similar to the retention time of the MS1 peak and extract the corresponding peak information.

keep <- keep &
    chromPeaks(swath_data)[, "rtmin"] < fenamiphos_ms1_peak[, "rt"] &
    chromPeaks(swath_data)[, "rtmax"] > fenamiphos_ms1_peak[, "rt"]

fenamiphos_ms2_peak <- chromPeaks(swath_data)[which(keep), ]

In total r sum(keep, na.rm = TRUE) MS2 chromatographic peaks match all the above condition. Next we extract their corresponding ion chromatograms, as well as the ion chromatogram of the MS1 peak. Note that we extend the retention time range by 1 seconds on both sides. In addition we have to filter the object first by isolation window, keeping only spectra that were measured in that specific window and to specify to extract the chromatographic data from MS2 spectra (with msLevel = 2L).

rtr <- fenamiphos_ms1_peak[, c("rtmin", "rtmax")]
rtr[1] <- rtr[1] - 1
rtr[2] <- rtr[2] + 1
mzr <- fenamiphos_ms1_peak[, c("mzmin", "mzmax")]
fenamiphos_ms1_chr <- chromatogram(swath_data, rt = rtr, mz = mzr)

rtr <- fenamiphos_ms2_peak[, c("rtmin", "rtmax")]
rtr[, 1] <- rtr[, 1] - 1
rtr[, 2] <- rtr[, 2] + 1
mzr <- fenamiphos_ms2_peak[, c("mzmin", "mzmax")]
fenamiphos_ms2_chr <- chromatogram(
    filterIsolationWindow(swath_data, mz = fenamiphos_mz),
    rt = rtr, mz = mzr, msLevel = 2L)

We can now plot the extracted ion chromatogram of the MS1 and the extracted MS2 data.

plot(rtime(fenamiphos_ms1_chr[1, 1]),
     intensity(fenamiphos_ms1_chr[1, 1]),
     xlab = "retention time [s]", ylab = "intensity", pch = 16,
     ylim = c(0, 5000), col = "blue", type = "b", lwd = 2)
#' Add data from all MS2 peaks
tmp <- lapply(fenamiphos_ms2_chr@.Data,
              function(z) points(rtime(z), intensity(z),
                                 col = "#00000080",
                                 type = "b", pch = 16))

Next we can calculate correlations between the peak shapes of each MS2 chromatogram with the MS1 peak. We perform the correlation below for one of the MS2 chromatographic peaks. Note that, because spectra are recorded consecutively, the retention times of the individual data points will differ for the MS2 and MS1 chromatographic data and data points have thus to be matched (aligned) before performing the correlation analysis. This is done automatically by the correlate function. See the help for the align method for more information on alignment options.

correlate(fenamiphos_ms2_chr[1, 1],
          fenamiphos_ms1_chr[1, 1], align = "approx")

After identifying the MS2 chromatographic peaks with shapes of enough high similarity to the MS1 chromatographic peaks, an MS2 spectrum could be reconstructed based on the m/z and intensities of the MS2 chromatographic peaks.

The reconstructChromPeakSpectra function performs the above analysis for each individual MS1 chromatographic peak in a SWATH data set. Below we reconstruct MS2 spectra for our example data requiring a peak shape correlation higher than 0.9 between the candidate MS2 chromatographic peak and the target MS1 chromatographic peak.

swath_spectra <- reconstructChromPeakSpectra(swath_data, minCor = 0.9)
swath_spectra

As a result we got a Spectra object of length equal to the number of MS1 peaks in our data. A peaksCount of 0 indicates that no MS2 spectrum could be defined based on the used settings. For reconstructed spectra additional annotations are available such as the IDs of the MS2 chromatographic peaks from which the spectrum was reconstructed ("ms2_peak_id") as well as the correlation coefficient of their chromatographic peak shape with the precursor's shape ("ms2_peak_cor"). Metadata column "peak_id" contains the ID of the MS1 chromatographic peak.

We next extract the MS2 spectrum for our example peak most likely representing [M+H]+ ions of Fenamiphos using its chromatographic peak ID:

fenamiphos_swath_spectrum <- swath_spectra[
    mcols(swath_spectra)$peak_id == rownames(fenamiphos_ms1_peak)]

We can now compare the reconstructed spectrum to the example consensus spectrum from the DDA experiment in the previous section (variable ex_spectrum) as well as to the MS2 spectrum for Fenamiphos from Metlin (with a collision energy of 10V).

par(mfrow = c(1, 2))
plot(fenamiphos_swath_spectrum[[1]], ex_spectrum[[1]],
     tolerance = 50e-6, main = "against DDA")
plot(fenamiphos_swath_spectrum[[1]], fenamiphos[[2]],
     tolerance = 50e-6, main = "against Metlin")

As a second example we analyze the signal from an [M+H]+ ion with an m/z of 376.0381 (which would match Prochloraz). We first identify the MS1 chromatographic peak for that m/z and retrieve the reconstructed MS2 spectrum for that peak.

prochloraz_mz <- 376.0381

prochloraz_ms1_peak <- chromPeaks(swath_data, msLevel = 1L,
                                  mz = prochloraz_mz, ppm = 5)
prochloraz_ms1_peak

prochloraz_swath_spectrum <- swath_spectra[
    mcols(swath_spectra)$peak_id == rownames(prochloraz_ms1_peak)]

In addition we identify the corresponding MS1 peak in the DDA data set, extract all measured MS2 chromatographic peaks and build the consensus spectrum from these.

prochloraz_dda_peak <- chromPeaks(dda_data, msLevel = 1L,
                                  mz = prochloraz_mz, ppm = 5)
prochloraz_dda_peak

The retention times for the chromatographic peaks from the DDA and SWATH data match almost perfectly. Next we get the MS2 spectra for this peak.

prochloraz_dda_spectra <- dda_spectra[
    mcols(dda_spectra)$peak_id == rownames(prochloraz_dda_peak)]
prochloraz_dda_spectra

In total 5 spectra were measured, some with a relatively high number of peaks. Next we combine them into a consensus spectrum.

prochloraz_dda_spectrum <- combineSpectra(
    prochloraz_dda_spectra, method = consensusSpectrum, msd = 0, ppm = 20,
    minProp = 0.8, weighted = FALSE, intensityFun = median, mzFun = median)

At last we load also the Prochloraz MS2 spectra (for different collision energies) from Metlin.

prochloraz <- spectra(readMgfData(
    system.file("mgf/metlin-68898.mgf", package = "xcms")))

To validate the reconstructed spectrum we plot it against the corresponding DDA spectrum and the MS2 spectrum for Prochloraz (for a collision energy of 10V) from Metlin.

par(mfrow = c(1, 2))
plot(prochloraz_swath_spectrum[[1]], prochloraz_dda_spectrum[[1]],
     tolerance = 50e-6, main = "against DDA")
plot(prochloraz_swath_spectrum[[1]], prochloraz[[2]],
     tolerance = 50e-6, main = "against Metlin")

The spectra fit relatively well. Interestingly, the peak representing the precursor (the right-most peak) seems to have a slightly shifted m/z value in the reconstructed spectrum.

Similar to the DDA data, the reconstructed MS2 spectra from SWATH data could be used in the annotation of the MS1 chromatographic peaks.

Outlook

Currently, spectra data representation, handling and processing is being re-implemented as part of the RforMassSpectrometry initiative aiming at increasing the performance of methods and simplifying their use. Thus, parts of the workflow described here will be changed (improved) in future.

Along with these developments, improved matching strategies for larger data sets will be implemented as well as functionality to compare Spectra directly to reference MS2 spectra from public annotation resources (e.g. Massbank or HMDB).

Regarding SWATH data analysis, future development will involve improved selection of the correct MS2 chromatographic peaks considering also correlation with intensity values across several samples.

Session information

sessionInfo()

References



xiaodfeng/DynamicXCMS documentation built on Aug. 6, 2023, 3:02 p.m.