-
- Posts: 56
- Joined: Fri Feb 04, 2005 10:05 pm
Dear Colleagues,
Here's an interesting dilemma I'm sure we've all thought about before. Whenever someone does quantification by LCMS, they build a calibration curve of a standard and quantify identified unknowns in samples against this standard, using the linear portion of the curve. Calibration may be via an external or internal standard method, where a response factor is calculated in the form of a slope or ratio plot, respectively. For all practical purposes, it doesn't matter how the analyte concentration is calculated - e.g. nmol/mL, ppm, ng/mL, etc - as any unknown peak in a sample for which there is a corresponding standard can be quantified against that standard.
However, problems start to arise when semiquantification is performed. By semiquantification I mean you have, for example, 4 standards, one of which is perhaps an internal standard, but you are quantifying 10 compounds in the sample. Thus 3 compounds in the sample are correctly quantified against their respective calibration curves, but the remaining 7 are semiquantified. The most common approach for these 7 compounds would be to use either one the response factors (internal standard/standard) for the standard closest in retention time to the unknown peak to calculate the unknown's concentration, or perhaps an average of all the response factors for all the standards, assuming they are structurally similar.
This approach works well when calibration is by total mass (i.e. ng/mL, ng/sample, etc) using a mass selective detector, for example the FID in GC, where the detector repsonds more or less linearly to the total amount of carbon that passes through the detector. Thus, with GC-FID, 1 pg of a 20-carbon alkane will give a very similar detector response to 1 pg of a 10-carbon alkane. Therefore, for GC-FID methods, calibration curves are calculated on a mass basis - e.g. ng/mL, and unknowns can be semiquantified fairly reliably as response factors are very close to 1. The nature of the FID detector also means that calibration curves should never be made on a concentration basis i.e. nmol/mL. If this were done, in the example above, then 1 pg of a 10-carbon alkane would have double the molar concentration of 1 pg of a 20-carbon alkane. Thus, if only a 20-carbon alkane was available as a standard, then semiquantitative calculation of an unknown peak that happened to be a 10-carbon alkane would give an accurate result if the C20 ng/mL response factor was used, but a 50% underestimate of C10 would result if the C20 nmol/mL response factor was used. Thus, everyone uses ng/mL or similar calibrations for GC-FID analyses.
Now, if this same thinking is applied to LCMS semiquantification, then nmol/mL calibrations should be used. This is because the MS is a concentration selective dector; i.e. every molecule will form, for example an [M+H]+ ion that is then detected. Thus, 1 mol of a C20 alkane will form the same number of molecular ions as 1 mol of a C10 alkane (I'm ignoring the difficulties in ionizing alkanes by soft ionization techniques here, please bear with me...), whereas 1 ng of C10 will form approximately double the number of molecular ions compared to 1 ng of C20. Thus, it follows that any LCMS (or indeed any concentration selective detectors, which would include UV and fluorescent detectors too), should have concentration curves built on a molar concentration basis. Then any semiquantitative calculations fro unknowns will be closer to their real values.
My question is: why does this never seem to be done in practice? It's ususal to see calibrations for many forms of hyphenated MS in ng/mL or similar mass-specific units. This will then lead to substantial errors if semiquantification is done. I know semiquantification is not ideal, but if you are in the business of doing, for example metabolomics, you may have hundreds of unknowns and a handful of standards. It would then seem very dangerous to semiquantify a wide range of these unknowns by LC (or GC) MS using ng/mL calibration curves. I know there are a whole lot of other issues to consider, such as ionization efficiencies for different compounds, etc, but it seems to me to be important to at least consider the initial assumptions of how best to use response factors based on how the detection system works.
thanks
Tony
Here's an interesting dilemma I'm sure we've all thought about before. Whenever someone does quantification by LCMS, they build a calibration curve of a standard and quantify identified unknowns in samples against this standard, using the linear portion of the curve. Calibration may be via an external or internal standard method, where a response factor is calculated in the form of a slope or ratio plot, respectively. For all practical purposes, it doesn't matter how the analyte concentration is calculated - e.g. nmol/mL, ppm, ng/mL, etc - as any unknown peak in a sample for which there is a corresponding standard can be quantified against that standard.
However, problems start to arise when semiquantification is performed. By semiquantification I mean you have, for example, 4 standards, one of which is perhaps an internal standard, but you are quantifying 10 compounds in the sample. Thus 3 compounds in the sample are correctly quantified against their respective calibration curves, but the remaining 7 are semiquantified. The most common approach for these 7 compounds would be to use either one the response factors (internal standard/standard) for the standard closest in retention time to the unknown peak to calculate the unknown's concentration, or perhaps an average of all the response factors for all the standards, assuming they are structurally similar.
This approach works well when calibration is by total mass (i.e. ng/mL, ng/sample, etc) using a mass selective detector, for example the FID in GC, where the detector repsonds more or less linearly to the total amount of carbon that passes through the detector. Thus, with GC-FID, 1 pg of a 20-carbon alkane will give a very similar detector response to 1 pg of a 10-carbon alkane. Therefore, for GC-FID methods, calibration curves are calculated on a mass basis - e.g. ng/mL, and unknowns can be semiquantified fairly reliably as response factors are very close to 1. The nature of the FID detector also means that calibration curves should never be made on a concentration basis i.e. nmol/mL. If this were done, in the example above, then 1 pg of a 10-carbon alkane would have double the molar concentration of 1 pg of a 20-carbon alkane. Thus, if only a 20-carbon alkane was available as a standard, then semiquantitative calculation of an unknown peak that happened to be a 10-carbon alkane would give an accurate result if the C20 ng/mL response factor was used, but a 50% underestimate of C10 would result if the C20 nmol/mL response factor was used. Thus, everyone uses ng/mL or similar calibrations for GC-FID analyses.
Now, if this same thinking is applied to LCMS semiquantification, then nmol/mL calibrations should be used. This is because the MS is a concentration selective dector; i.e. every molecule will form, for example an [M+H]+ ion that is then detected. Thus, 1 mol of a C20 alkane will form the same number of molecular ions as 1 mol of a C10 alkane (I'm ignoring the difficulties in ionizing alkanes by soft ionization techniques here, please bear with me...), whereas 1 ng of C10 will form approximately double the number of molecular ions compared to 1 ng of C20. Thus, it follows that any LCMS (or indeed any concentration selective detectors, which would include UV and fluorescent detectors too), should have concentration curves built on a molar concentration basis. Then any semiquantitative calculations fro unknowns will be closer to their real values.
My question is: why does this never seem to be done in practice? It's ususal to see calibrations for many forms of hyphenated MS in ng/mL or similar mass-specific units. This will then lead to substantial errors if semiquantification is done. I know semiquantification is not ideal, but if you are in the business of doing, for example metabolomics, you may have hundreds of unknowns and a handful of standards. It would then seem very dangerous to semiquantify a wide range of these unknowns by LC (or GC) MS using ng/mL calibration curves. I know there are a whole lot of other issues to consider, such as ionization efficiencies for different compounds, etc, but it seems to me to be important to at least consider the initial assumptions of how best to use response factors based on how the detection system works.
thanks
Tony
