Method optimization is largely a colloquial term for LC/MS/MS users. The extent of optimization can range from cursory evaluations during core development experiments to substantive investigations of all variables associated with a method. Unlike design of experimentation which requires broad assumptions and large experimental parameter shifts, design of optimization relies on efficiently designed experimentation with the ability to produce valid conclusions. This paper will introduce a practical series of experimental frameworks to scientifically rationalize the parameters and processes involved in optimization of a clinical diagnostics assay. Examples will be shown from assays which analyze more than 500 specimen/day.
A clinical diagnostic assay can produce results for a number of years, analyzing thousands upon thousands of samples. In order to produce high quality data on a continuing basis, assays should be optimized for operational utility. Such optimization can result in cost savings as well as increased uptime of the assay, generally as a result of loading less material into the instrument to achieve the same degree of sensitivity. However, optimization is a largely personal task, relying on experience, unrefined development experiments, and often results in a highly subjective definition for “optimized”. This paper shall explain the modes and methods used to perform a scientific evaluation of optimization of an LC-MS/MS assay, such that the method can be accurately justified as “optimized”.
All experiments were performed on an API 5000 triple quadrupole mass spectrometer operating in electrospray mode (AB Sciex). Chromatographic separations were performed on an MPX-II system with LC-20AD pumps and autosampler (Shimadzu) using a Restek Biphenyl column (50x2.1 mm, 3 um). Example compounds include commonly analyzed drugs of abuse, purchased from Cerilliant; LC/MS grade mobile phases were purchased from MidSci, Acros, Fisher, VWR and EMD; formic acid was purchased from Fisher and MidSci. Modifications to parameters were assigned by determination of the distribution of response functions for each parameter change and variables which are coordinated. Signal to noise and peak area were calculated for each sample to assess increases or decreases in response relative to the provisionally developed method.
To establish the required n of experimentation, 96 samples (a single batch) were analyzed via a provisionally developed methodol. CV's of the internal standard response were calculated; using power = 0.95 and p = 0.05, a priori power analysis determined the sample size must be n = 11 for subsequent experiments. Variables were assessed for co-ordination in optimization, such that the changes in optimization of one variable would change the optimal value for another. For example, nebulizing gas and the heated desolvation gas are co-ordinated variables. Example molecules of morphine and oxymorphone demonstrate highest intensity at either relatively low nebulizing gas/low heated desolvation gas or at relatively high nebulizing gas/high heated desolvation gas rate, with lower sensitivities at combined values between, including high/low or low/high for the respective settings. As most experimentation regarding the optimization of these variables is undertaken as a univariate assessment, the co-ordination of these 2 gas settings is lost. This will be further shown in other MS parameters such as mobile phase modifiers/declustering potential and collisional dissociation energy/collisional gas volume. In order to reduce the overall number of experiments performed and the total volume of data produced, methods to determine the testing range and the step size between variables will be discussed. The analysis of all possible combinations of MS parameters is both infinitely cumbersome and exceptionally wasteful; the process of determining appropriate variables and how to assess them for optimization will be shown with example compounds.
The overall gain between provisionally optimized methods (settings determined in method development via infusion or coarse evaluations) and a fully optimized assay differs between approximately 240% (Carisoprodol) and >600% (Temazepam). This approach uses a simple statistical foundation and targeted experimentation to provide for the subjective, rather than objective, production of optimal methodologies.