Evolving neural network optimization of cholesteryl ester separation by reversed-phase HPLC
- First Online:
- Cite this article as:
- Jansen, M.A., Kiwata, J., Arceo, J. et al. Anal Bioanal Chem (2010) 397: 2367. doi:10.1007/s00216-010-3778-5
- 376 Views
Cholesteryl esters have antimicrobial activity and likely contribute to the innate immunity system. Improved separation techniques are needed to characterize these compounds. In this study, optimization of the reversed-phase high-performance liquid chromatography separation of six analyte standards (four cholesteryl esters plus cholesterol and tri-palmitin) was accomplished by modeling with an artificial neural network–genetic algorithm (ANN-GA) approach. A fractional factorial design was employed to examine the significance of four experimental factors: organic component in the mobile phase (ethanol and methanol), column temperature, and flow rate. Three separation parameters were then merged into geometric means using Derringer’s desirability function and used as input sources for model training and testing. The use of genetic operators proved valuable for the determination of an effective neural network structure. Implementation of the optimized method resulted in complete separation of all six analytes, including the resolution of two previously co-eluting peaks. Model validation was performed with experimental responses in good agreement with model-predicted responses. Improved separation was also realized in a complex biological fluid, human milk. Thus, the first known use of ANN-GA modeling for improving the chromatographic separation of cholesteryl esters in biological fluids is presented and will likely prove valuable for future investigators involved in studying complex biological samples.
KeywordsLipidsCholesteryl linoleateInnate immunityBiological fluidsArtificial neural networksGenetic algorithms
All internal body surfaces are bathed with fluids that contain a variety of molecules, some of which are involved in innate immunity defense. Innate immunity forms the first line of defense against microbial invasion, and well-established mediators of this system include antimicrobial peptides . Lipids are also ubiquitously present in body fluids , and there is increasing evidence that selected lipids have antimicrobial activity [3, 4] and form a lipid-mediated arm of the innate immunity response. We have shown that cholesteryl esters have antimicrobial properties and contribute to the innate immunity of secretions of the airway mucosa [5, 6]. Developing an optimized separation method in order to characterize the role of individual lipids in innate immunity would be a significant advance to this field. Methods requiring minimal sample handling with high recovery for subsequent functional assays are essential. Current methods for cholesteryl ester analysis include HPTLC , TLC-GC , and ES-MS/MS [9, 10]. A high-performance liquid chromatography (HPLC) protocol for cholesteryl ester analysis has been described by Cullen et al. . However, this protocol is suboptimal because triglycerides are first removed, preventing complete lipid analysis and thus adding additional steps. To meet our interest in characterizing antimicrobial lipids, we developed a reversed-phase HPLC (rpHPLC) one-step protocol for total lipid extracts that is suitable for both analytical and preparative scale work, but in this method, cholesteryl esters were incompletely resolved . Because of the evidence of an antimicrobial role for cholesteryl esters [5, 6], optimization of the analytical protocol became of paramount importance.
The use of information processing techniques, in particular artificial neural networks (ANN), has proved valuable for a variety of separation methods [12–14]. When combined with experimental design techniques, ANN quickly optimizes the separation conditions and shortens analysis time, and does so without knowledge of the physical or chemical properties of the analytes . In terms of applications, a limited number of studies have utilized ANN for optimizing HPLC experimental conditions [15–17]. Others have used a hybrid artificial neural network–genetic algorithm (ANN-GA) approach to improve separation methods for isolated compounds  and for routine pesticide analysis . However, the incorporation of ANN-based methods for investigating the separation of complex human biological fluids is lagging, particularly with regards to lipids. Herein, we describe the first use of an ANN-GA approach for characterizing cholesteryl esters in human body secretions.
Chemicals and reagents
HPLC grade acetonitrile, reagent grade alcohol, water, and chloroform were obtained from Fisher Scientific (New Jersey, USA). HPLC grade dichloromethane was obtained from EMD Chemicals Inc. (Darmstadt, Germany). Cholesterol, cholesteryl arachidonate, cholesteryl linoleate, cholesteryl palmitate, tri-palmitin (a highly hydrophobic triglyceride with similar elution times as cholesteryl esters in rpHPLC), cholesteryl stearate, and heptadecanoic acid (internal standard for milk lipid extraction) were obtained from Sigma-Aldrich (Missouri, USA).
A mixture of cholesterol, cholesteryl arachidonate, cholesteryl linoleate, cholesteryl palmitate, tri-palmitin, and cholesteryl stearate, each at 1 mg/mL, was prepared in dichloromethane.
Lipid extraction from human milk
Human milk from three different donors was purchased from Mother’s Milk Bank, Denver, CO. Lipids were extracted according to Bligh and Dyer  using 10 µL aliquots for rpHPLC analysis and 20 µL aliquots for ESI/MS/MS.
Under the control of Chromeleon® software (version 6.60 SP2), solutions of standards and extracts re-dissolved in dichloromethane were manually injected (1–3 μL/injection, 20 μL injection loop) onto a reversed-phase column (Dionex Acclaim PolarAdvantage II, a silica-based column with a proprietary amide-embedded ligand, 150 mm × 2.1 mm ID, 3 μm particle size) preceded by a guard column (Dionex Acclaim PolarAdvantage II, 2.1 × 10 mm, 5-μm particle size) in a temperature-controlled compartment (Dionex model TCC-100 column oven). The column had been previously equilibrated in the selected eluant and was eluted (Dionex P680 low-pressure quaternary pump with degasser) at specified flow rates. The eluant was passed through an evaporative light-scattering detector (ELSD, Alltech model 800). For fraction collection, the ELSD was bypassed and collected fractions were dried in a stream of nitrogen and stored at −20 °C for further analysis.
Mass spectral analysis
Summary of factors and responses for screening runs
Column T (°C)
Flow rate (mL/min)
Means ± SD
Relative SD (%)
0.375 ± 0.154
0.840 ± 0.025
0.151 ± 0.004
0.217 ± 0.013
0.918 ± 0.009
0.605 ± 0.072
0.248 ± 0.011
0.120 ± 0.008
Derringer’s desirability function
For the transformation of the total number of peaks, the target value was set at 6 to match the number of analytes in the standard mixture. The minimum acceptable value was set at 1, while the maximum acceptable value was set at 8 to accommodate lipid oxidation or degradation that might result in additional peaks. The exponents s and t were both set at 3 to assign more weight to values closer to the target value of 6. For the transformation of the retention time of the last eluting peak, a target value of 25 min was chosen to accommodate future studies of biological fluids and high throughput analysis. The maximum value was set at 100 min, and s was set to 1 and t to 3 to match the preference for a shorter run.
Artificial neural network–genetic algorithm approach
Results and discussion
ANOVA calculations for the linear model
Sum of squares
Prob > F
ANN-GA prediction and validation
Summary of factors and responses for validation runs
Column T (°C)
Flow rate (mL/min)
Mean ± SD (n = 3)
Relative SD (%)
0.924 ± 0.025
0.889 ± 0.036
0.879 ± 0.012
0.912 ± 0.020
0.915 ± 0.030
0.896 ± 0.007
Considering that a satisfactory peak separation was achieved with a limited number of HPLC runs, this study confirms the suitability of the ANN-GA model in method optimization for rpHPLC. In contrast, manipulation of parameters via trial and error did not lead to a substantially improved separation of cholesteryl esters after more than 50 runs, and hence, this approach was abandoned (data not shown). Previously, ANN has been used successfully to separate structurally similar compounds, namely, indinavir and lactone, using HPLC , and complex mixtures of neuropeptides .
To assess whether the improved method (validation run 1) would be applicable to complex biological fluids, we subjected extracted human milk lipids to rpHPLC using the same experimental conditions dictated by the ANN-GA model. Choosing a lipid-rich fluid provided a robust test of the ANN-GA model. Since fatty acids and other lipids in human milk may interfere with cholesteryl ester separation , we included a water gradient (15% to 0% H2O, 16.5% to 19.4% reagent alcohol with acetonitrile in the first 5 min of the separation) before the isocratic gradient used for validation run 1. This water gradient had previously been successful in separating free fatty acids in our laboratory (unpublished data).
Elution profile of selected cholesteryl esters extracted from human milk in rpHPLC before and after ANN-GA modeling
Relative intensity (%)
Relative intensity (%)
This study provides guidance for the development and application of experimental design methodology and ANN-GA modeling tools for rapid optimization of the HPLC method used to separate cholesteryl esters. This approach optimally determined a set of conditions in which cholesteryl esters were fully separated in standard mixtures and better resolved in a complex biological fluid in a timely fashion. This method improvement will facilitate studies to unveil the biological function of cholesteryl esters in innate immunity. Moreover, as a general-purpose optimization approach, ANN-GA modeling will likely prove useful for a wide range of method optimizations concerning complex biological samples.
This research was funded by 1 P20 MD001824 (EP) from the National Institutes of Health. GH acknowledges support from the John Stauffer Charitable Trust and the Swenson Summer Research Internship Program. The authors also acknowledge Katina Landon and Christine Markowitz for their help with initial model development exercises.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.