Dear Skyline Users,
Our seventh Skyline Tutorial Webinar covered the complex topic of iRT retention time prediction -- so naturally, we had some great questions from our live audience. And when you have questions, its our job to provide the answers: so here they are:
Q: You just mentioned having accumulated several thousands of measured peptides. Do you have an iRT library ready to download?
Ans: I am currently working on a DIA experiment performed by Ludovic Gillet at the Aebersold lab involving around 90,000 peptide targets of Human, Yeast and E. coli origin. The Skyline document provided to me for that has iRT values for all peptides in it. I would expect that experiment to be published within the next year, and include those iRT values then in the supplementary materials. But other papers have been published already with extensive lists of iRT values, which can be very easily used in Skyline.
Q: What's the difference between Biognosys iRT and HRM Calibration kit for SWATH?
Ans: I have to admit I don’t completely know the answer to this one, since I have built Skyline to work with a variety of peptide standard mixes (and even analyte peptides) for iRT retention time normalization and prediction. It certainly works with the Biognosys iRT-kit, and that was the standard mix used in the tutorial webinar. My understanding is that the Biognosys HRM mix is a superset of the iRT-kit. That is, it contains all of the iRT peptides plus more. But, Skyline does not currently make any special use of these extra peptides. I believe you only need the HRM mix, if you plan on using Biognosys software for data analysis.
Q:Briefly, is it the same principle for DIA data? I want to use the IRT predictor with my internal standard to align the data of my DIA data (with spiked iRT)
Ans: Yes, it is the same principle, namely that you can use linear regression of standards to normalize peptides measured with the standards into a time-independent scale that can be stored in a library, and then mapped back into predicted retention times based on regression on measured times of the same standards in matrix with the peptides recorded in the library. In the case of DIA, rather than using the iRT values to predict RT before acquisition for the purpose of scheduling acquisition, the standard peptides can be detected post-acquisition and then used to aid in detection of the other peptides in the time dimension of already acquired DIA data. In DIA, these predicted times can help both to limit the length of chromatograms extracted from the data, like scheduling in SRM only post-acquistion, as well as to validate the correct chromatogram peak is integrated for quantification.
Q:iRT seems like a good solution for gradient time change as long as it stays linear, what about change from a linear gradient to a step/ramp gradient?
Ans: The iRT concept has certainly been most widely used so far on linear gradients, and it has proved reliable and very useful in that situation. In principal, the use of “landmark peptides” should be of use in any gradient where peptides elute somewhat consistently relative to each other. For non-linear gradients, the prediction might be improved by changing from linear regression to lowess regression. Although this is not implemented in Skyline yet, nor as rigorously proven to work as linear regression for a linear gradient.
Q:Can iRT retention time prediction be useful in appropriately scheduled SRM experiments of study samples spiked with heavy labeled standards. Namely, can it be used to account for sample-specific retention time shift which may not necessarily be linear across the whole gradient?
Ans: Potentially not. In a setting where retention time deviation is _on_average_ linear, iRT works quite well at predicting retention time. Note that this is on average and that your mileage may vary in specific cases. I have seen it work quite well in an experiment involving heavy labeled standards over around 70 runs, where most of the runs were performed a month before the final set of 15 or so. There is a marked shift in retention time, presumably caused by a column change, between the first set of runs and the second set, but with retention time alignment performed based on iRT the shift disappears almost entirely. Though, I cannot even promise that this is the case for every last peptide targeted in that experiment, but it certainly is on average. However, the use of heavy labeled standards pared with every targeted peptide would reduce the need for retention time normalization, and do a much better job of ensuring consistent measurement of the analytes of interest, because heavy labeled standards will always co-elute with the analyte targets. However, iRT may still be of interest in predicting retention times for scheduled acquisition in such settings.
Q:Is it possible to use iRTs from a linear gradient to make predictions about retention times for a non-linear gradient?
Ans: I wouldn’t recommend trying to take iRT values which are taken from empirical measurements on a specific chromatography across changes that you expect not to behave linearly, without also adapting the technique to account for those changes. For instance, if the peptides are expect to continue to elute in the same order but change in relationship to each other in a non-linear way, you might simply need to switch to lowess regression. However, as Mike MacCoss has pointed out before, certain changes in chromatography (e.g. stationary phase changes) can cause peptides to elute in different orders. iRT values are not transferable across this type of change, and you would need to re-derive the normalized ordering of your peptides for each such condition.
Q:What is the limit of deviation of chromatogram shift across runs which can be handled by iRT calibration?
Ans: As long as the shift is linearly related to prior measurements, then there should be not limit. Certainly it would not matter if all peptides suddenly eluted 30 minutes later but experienced not shifts relative to each other. In the tutorial presentation and the original iRT paper, we showed that iRT could handle very well a change from calibration on a 30 minute gradient to prediction and detection on a 90 minute gradient.
Q:What happens if the shift is being seen in half the chromatogram?
Ans: If the shift is non-linear, then your only hope would be to consider a different regression approach, and even then the prediction would likely be more error prone, especially at the ends of the gradient, as lowess regression is well known to overweight points at the extremes of the data set. I know of no foolproof way to predict retention time from one run to the next even under very controlled circumstances. iRT has proven extremely useful in many cases, though.
Q:What is the advantage of using external iRT standard peptides over internal (eg Apo AI) peptides in DIA experiments that utilize corresponding DDA libraries?
Ans: We don’t generally use iRT in any run where we also have DDA spectrum IDs either in the runs themselves or in closely associated runs on the same instrument to use for retention time prediction. That is, we do not use iRT for the DIA with DDA workflow presented in Webinar #2: Jump Start DIA Analysis with DDA Data in Skyline. However, iRT can be used very effectively in the “Prior Knowledge Workflow” also described briefly in that webinar, and we hope to make that a future webinar topic. In that case, we are finding that researchers doing largely multi-replicate studies, where they are likely to run hundreds to thousands of samples find it worth the method development time to work out endogenous standards, as we have done for Apo A1. It turns out that you could also work out endogenous standards after you have already made DIA measurements without adding a standard mix. All you would then need to do to gain access to all of the iRT values derived for all other peptides on the iRT-C18 scale (originally defined in the iRT paper for the Biognosys iRT-kit) would be to measure your sample at least once with another set of standards in that scale. Once you have calculated your standards in the iRT-C18 scale, then they can be used to predict retention times for all other peptides in that scale (assuming you are using compatible chromatography). The next public version of Skyline will contain iRT-C18 values for standard mixes from Biognosys, Sigma Aldrich and Pierce, as well as the Apo A1 peptides we use.