Intelligent Peak Deconvolution

June 6, 2017
Lewis Botcherby
The Column

Volume 13, Issue 8

Page Number: 2–5

High quality, low cost is a standard mantra within the pharmaceutical industry, but with increasing structural complexity of drugs and drug candidates maintaining the core value of this mantra is becoming more difficult. Kanta Horie from Eisai Co., Ltd., recently spoke to The Column about the development of an intelligent peak deconvolution technique using multivariate curve resolution-alternating least squares (MCR-ALS) that allows accurate quantitation of multiple components with different absorbing spectra even if the peaks are not completely separated.

High quality, low cost is a standard mantra within the pharmaceutical industry, but with increasing structural complexity of drugs and drug candidates maintaining the core value of this mantra is becoming more difficult. Kanta Horie from Eisai Co., Ltd., recently spoke to The Column about the development of an intelligent peak deconvolution technique using multivariate curve resolution-alternating least squares (MCR-ALS) that allows accurate quantitation of multiple components with different absorbing spectra even if the peaks are not completely separated.- Interview by Lewis Botcherby

Q. What challenges have arisen from the introduction of high‑end ultrahigh‑pressure liquid chromatography (UHPLC) systems in pharmaceutical analysis?A: Ultrahigh-pressure liquid chromatography (UHPLC) systems using sub‑2-μm particle‑packed columns offer high efficiency (for example, peak capacity per unit of time), but not an absolutely high peak capacity that can be achieved without a time limitation (1).
The structural complexity of recent drugs and candidates (mid-size compounds, biologics, and engineered cells and tissues) is dramatically increasing because of the diversity of current molecular pharmacology, but this complexity brings an undesirable physicochemical property that often provides big challenges, especially for pharmaceutical development, despite using high-end UHPLC. In particular, impurity testing for complex pharmaceuticals needs higher chromatographic efficiency (for example, ultra-high theoretical plates > 100,000) to accomplish enough separation for accurate quantitation of each component. The challenge could be solved by special techniques, such as the use of meter-scaled long columns and long analytical times (2), however, time–benefit approaches are essential in pharmaceutical development because tremendous numbers of samples need to be analyzed, especially in downstream areas, including formulation research.
We knew a breakthrough technology was necessary, and intelligent peak deconvolution, using multivariate curve resolution-alternating least squares (MCR-ALS) with bidirectional exponentially modified Gaussian (BEMG) model function, was conceived. This involves a “change in thinking” and shows a hypothetical separation evidenced by reflected mathematical theory. Our algorithm enables the accurate quantitation of multiple components with different absorbing spectra, even though the peaks are not completely separated. If an incomplete separation is acceptable, a special technique that needs long analysis times is not necessary and our main goal can be accelerated.
Regarding “peak deconvolution”, everyone can imagine the application of liquid chromatography–mass spectrometry (LC–MS). It is without question that LC–MS-based peak deconvolution is an important technique; however, we believe our photodiode array (PDA)-based deconvolution approach offers the following advantages: high accuracy as a result of the “no ionization suppression” effect; cost-effective because a simple high performance liquid chromatography (HPLC) system equipped with conventional detectors, such as a PDA, is used; and it is also robust and easy to use-a trained specialist is not required for this approach.
The main requirement from the pharmaceutical industry is high quality with low cost. The quantitative accuracy in stacked peaks from LC–MS is vague because of the ionization characteristics, whereas the developed approach based on a PDA with a mathematical algorithm has the potential to ensure the quantitative accuracy, even though the peaks are stacked. No additional cost or special device is necessary because all that is required is an installation of a deconvolution algorithm (program) to process peak capacity. The pharmaceutical industry needs the expansion of production sites and that means that tech-transfer is frequently implemented in multiple sites. The peak deconvolution algorithm developed in this research (intelligent peak deconvolution using MCR-ALS with BEMG) offers faster analysis, streamlining of sample pretreatment, and the ability to obtain quantitative results based on the simple settings of time range and wavelength range. These features are particularly significant for applications such as drug research and development. Utilizing peak separation technology based on the subject method can be expected to further improve analytical efficiency and data reliability.

Q. What particular attributes does MCR‑ALS have that has led to an increase in its use?A: Many papers have been reported on the application of multivariate curve resolution techniques (3,4). MCR-ALS is conceptually simple and easy to implement. It is also fast to calculate while providing good separation results. We were motivated to integrate it into a commercially available chromatography data system. This will bring new insights to chromatographic analysis.

Q. Your recent paper details an algorithm that can be used to resolve overlapped peaks and obtain good separation results (5). How did you achieve this?A: MCR-ALS is a powerful approach to obtain chromatographic data and spectra without prior information. However, the mathematical decomposition of a single data matrix is subject to rotational ambiguity and intensity ambiguity. By selecting a model function appropriately as a constraint, the feasible range broadening is reduced. Restricting the degrees of freedom of a model function decreases the ambiguity. On the other hand, the fitting error is apt to increase. They are in a trade-off relationship. In the actual measured data, the chromatogram shape is not symmetrical, but it is leading or tailing at the edge of the peak. The BEMG function is a candidate to make the fitting error small at both the tailing edge and leading edge of the peak. The developed algorithm was applied to actual measured data and simulation data to confirm that using the BEMG model function for MCR-ALS as a chromatogram constraint can simplify specifying peak deconvolution parameter settings and enable accurate estimation of actual chromatograms and spectra.

 

Q. Could this approach be used in any other situation?A: This approach is definitely applicable to any field that has issues with stacked peaks and throughput within HPLC analysis. HPLC is an essential tool in many fields, including life sciences, chemistry, agriculture, soil and pesticides, food, and polymers as well as pharmaceutical science-so everyone that experiences reduced throughput because of incomplete peak separation, including users of UHPLC. We also expect this approach to be promising in chemistry fields because chemists frequently face the issue of byproduct contamination-including isomers in the synthesis process. Our approach may become a complementary tool to LC–MS to characterize the process. For example, for isomers that cannot be differentiated by mass spectrometry because they have the same mass-to-charge ratio (m/z), and the chromatographic separation is therefore essential for identification as well as quantification. To simplify this issue, our approach could be useful because the different absorbing spectra of stacked peaks have a potential to elucidate the quantitative deconvolution through MCR‑ALS with BEMG (that is, positional isomers and stereoisomers have basically different absorbing spectra.)
Our approach also expresses an original spectrum from each deconvoluted peak. That function can be utilized for the peak identification in multiple stacked peaks via verifying spectra library that is prospectively validated by standards or a different HPLC run. We think there are many possibilities with this approach.

Q. The peak deconvolution algorithm provided good separation results but could it be improved in anyway?A: Our paper (5) presents an analysis of a single dataset to demonstrate the chemometrics approach applied to pharmaceutical samples with a typical
HPLC–PDA system. By selecting the best model function for each HPLC system, each column and each component would improve the separation results.
The ambiguity of feasibility range broadening could be reduced when multiple datasets with different concentration ratios were used. A combination of MCR-ALS
and multidimensional chromatography (LC×LC–PDA) would be a candidate for
this purpose.

Q. What are you currently working on?A: A more accurate and high-throughput HPLC system than is currently available is necessary and we are assessing if our developed approach meets these requirements. The chemistry, manufacturing, and control (CMC) area of the pharmaceutical industry in particular seeks proven analytical methods that conform to regulations. The feasibility and validation study for the practical tasks is ongoing and the update will be planned accordingly. The instrumental update is also ongoing. The system reproducibility and the high dynamic range photodiode array detector can achieve the best qualitative and quantitative measurements with the help of a mathematical approach. We will continue to improve the instrumentation for the chromatographic challenges.

References

  1. K. Horie, Y. Sato, T. Kimura, T. Nakamura, Y.Ishihama, Y. Oda, T. Ikegami, and N. Tanaka, J. Chromatogr. A1228, 283 (2012).
  2. K. Horie, T. Kamakura, T. Ikegami, M. Wakabayashi, T. Kato, N. Tanaka, and Y. Ishihama, Anal. Chem. 86, 3817 (2014).
  3. M. Vosough, S. Ghafghazi, and M. Sabetkasaei, Talanta119, 17 (2014).
  4. R.L. Perez, G.M. Escandar, Anal. Chim. Acta835, 19 (2014).
  5. S. Arase, K. Horie, T. Kato, A. Noda, Y. Mito, M. Takahashi, and T. Yanagisawa, J. Chromatogr. A.1469, 35 (2016).

Kanta Horie is a scientist at the Translational Medicine Department of Neurology Business Group in Eisai Co., Ltd. in Japan. His current interests centre around mass spectrometry-based metabolomics and proteomics for biomarker research. He was previously in the section of chemistry, manufacturing, and control, and pharmaceutical analysis using (U)HPLC as well as other separation techniques.
The deconvolution algorithm presented here was developed based on results obtained from joint development work with Shimadzu Corporation.

E-mail: k2-horie@hhc.eisai.co.jpAddress: Eisai Co., Ltd., Translational Medicine Department, Medicine Creation, Neurology Business Group, Tokodai, Tsukuba, Ibaraki 300-2635, Japan