A mysterious autosampler problem is solved.
We recently experienced some mysterious and sporadic liquid chromatography (LC) results using an ultrahigh-pressure LC (UHPLC)
system. This incident took us a week of intensive investigation involving many hypotheses and numerous blind alleys until
we found the root cause. We believe this issue has not been previously reported and could easily be missed by many LC users
(1–4). Here is our story.
Our problems began when we performed potency and impurity assays of two newly developed clinical drug products (capsules and
tablets). We followed a validated 42-min procedure that had been used successfully for two years and modified it into a faster
and equivalent 15-min UHPLC method (5). The sample preparation procedure in the modified method uses 20 mM ammonium formate,
pH 3.7, as the extraction solvent. We used an Agilent model 1290 UHPLC (Santa Clara, California, USA) with 3-μL injections
and performed many area-percent analyses without any issues. However, this time the potency assay failed system suitability
tests because the response factors for duplicate standard (calibrator) preparations differed by ~20%. Curiously, peak area
repeatability from the same vial was excellent (<0.3% RSD), indicating that the autosampler was not malfunctioning. Duplicate
standard preparations were then made by two analysts. Highly variable results and high responses for the calibrator were observed,
which led to artificially low potency values of the drug products. We decided to launch a more systematic investigation.
We started our investigation by interviewing the primary analyst and checking all the calculations for the method. We noted
higher than expected responses from some calibrator vials but not others — even though all calibrator solutions (~0.5 mg/mL)
should have provided nearly identical responses. It seemed quite inconceivable that we obtained responses from two duplicate
calibrator preparations that varied as much as 20–30%.
Peak areas in chromatograms of drug product extracts typically were quite consistent. Next, we scrutinized each analyst's
weighing techniques and the quantitative transfer procedure of the weighed powder to the volumetric flask. We verified the
analytical balance calibration and visually checked the homogeneity of the reference standard. At that point, all practices
appeared to be in order, and we found no clues for the disparate responses. Next, we turned our attention to the existing
sample preparation procedure and designed an experiment using stronger extraction solvents. For quicker problem diagnosis,
we switched to a 2-min fast LC, nonstability indicating method (Waters XBridge C18 column [Milford, Massachusetts], isocratic
mobile phase consisting of 25% acetonitrile in 0.1% trifluoroacetic acid, 1-mL/min flow rate, 2-μL injection) (6). The results
of the extraction study are summarized in Table 1, which shows encouraging results in study 1 with all three extraction solvents
(ammonium formate, 0.1 N hydrochloric acid and 20% acetonitrile in 0.1 N hydrochloric acid) yielding the expected potency
of ~100%. However, reinjecting the very same sample vials on a different model 1290 UHPLC system yielded data with 30% higher
peak areas for the standards and low potency results of ~74–78% (study 2, Table 1). A repeat of the same sample set in study
3 using the 15-min UHPLC method on the second system also yielded the low potency results, but at ~8–11% higher potency than
those in study 2 (fast LC). A review of historical response factors showed higher responses for some of the calibrator solutions
as the key issue. The high calibrator responses were consistent with an extra injection volume of ~0.6 μL; that is, 30% for
2 μL, 20% for 3 μL, and 6% for 10 μL. We were puzzled with these consistent observations that could be caused by an extra
0.6 μL that was injected at some times, but not at others.
Table 1: Drug product potency assay results with different extraction solvents.