reverse external calibration curves

reverse external calibration curves chiva cristina  2018-04-26

I was wondering if Skyline supports reverse external calibration curves in which you inject different concentrations of heavy peptide and you use the light peptide (at fixed concentration) to normalize.


Jeff Whiteaker responded:  2018-04-26
I have used Skyline to determine linear ranges and approximate limits of quantification using reverse curves. In Skyline, set the internal standard to light, then proceed as normal.
For actual calibrated values using the reverse curve method it has to be an analyte free matrix – something often hard to come by in proteomics. If you’re performing an external calibration in this fashion, you probably would use the ‘surrogate analyte’ approach, incorporating two isotope labels (see for example: Leinenbach, Clinical Chemistry, 2014, 60:987-994).
Hope this help,
chiva cristina responded:  2018-05-02
Thank you Jeff for your reply,
In my external reverse calibration curve I use the light version to normalize. When I follow your suggestion Skyline can calculate the regression but then when I add my real samples in which I have used the heavy peptide to normalize the software is still using the light to normalize. Is there a way to tell Skyline to use different versions of the peptide to normalize depending if it is a standard (use light) or an unknown (use heavy)? I guess not....
Brendan MacLean responded:  2018-05-02
My understanding is that most people are not really using these "response curves" for "calibration" as we understand it, meaning actually using the curves to derive a calibrated quantity in an unknown sample. Instead, what people are typically doing is using the reverse response curves (varying heavy concentrations spiked into the endogenous background) to characterize "figures of merit", especially LOD and LOQ. Then they are typically using "internal calibration" (i.e. using the ratio of light to heavy times the spiked in heavy concentration) to estimate a calibrated quantity (e.g. if ratio light:heavy = 0.5 and spiked heavy = 10 fmol, then concentration is estimated to be 5 fmol), as long as the light peak area is above the LOQ estimated with the reverse curve.

Skyline is capable of doing all of this now, but you would never actually combine your reverse response curve runs with your unknown runs, as you would for external calibration (single point or multipoint), because you don't actually use the linear equation derived from the response curve to estimate the endogenous concentration. It is not valid to assume that inverse ratios of heavy:light can be applied to ratios of light:heavy to estimate the light concentration.

We are not big fans of internal calibration, but it is supported by Skyline as are reverse response curves. Someone can correct me if I am wrong and they see a valid way of using a reverse response curve of actually estimating an unknown concentration, but my understanding is that they are used for estimating the linear range and then single-point internal calibration is used with spiked in heavy to estimate the actual unknown concentration.

This is what I just heard again yesterday here at the NEU May Institute teaching with Andy Hoofnagle, Sue Abbatiello, and Lindsay Pino on calibrated quantification.

tobias.kockmann responded:  2019-06-06
Hi Brendan, Hi Jeff,

I have a related question: I am also performing PRM assay characterization using the reverse calibration curve approach and as Jeff pointed out by setting Modifications -> Internal standard type to light it becomes possible to plot calibration curves in skyline. Very nice! But since I did the reverse curve in matrix (incl. the endogenous peptides) I would like to see when the light signal is about 1:1 with respect to the heavy (because this would be the sweet spot for the following single point calibration as Brendan pointed out). Unfortunately, I can not see the light signals in the calibration curve plots and I can also not switch to a display of H:L or inverse ratios. Is there a way to get this done? The idea is actually shown here

Fig. 4

tobias.kockmann responded:  2019-06-06
My current work around is using the attached PeakArea and calibration curve plots side-by-side. But why do the peak areas differ by 10fold?