ARTICLE pubs.acs.org/ac
Quantitative Evaluation of the Sensitivity of Library-Based Raman Spectral Correlation Methods Jason D. Rodriguez, Benjamin J. Westenberger, Lucinda F. Buhse, and John F. Kauffman* Division of Pharmaceutical Analysis, Center for Drug Evaluation and Research, U.S. Food and Drug Administration, 1114 Market St, Room 1002, Saint Louis, Missouri 63101, United States ABSTRACT: Library-based Raman spectral correlation methods are widely used in surveillance applications in multiple areas including the pharmaceutical industry, where Raman spectroscopy is commonly used in verification screening of incoming raw materials. While these spectral correlation methods are rapid and require little or no sample preparation, their sensitivity to the presence of contaminants has not been adequately evaluated. This is particularly important when dealing with pharmaceutical excipients, which are susceptible to economically motivated adulteration by substances having similar physical/chemical/spectroscopic properties. We report a novel approach to evaluating the sensitivity of library-based Raman spectral correlation methods to contaminants in binary systems using a hit-quality index model. We examine three excipient/contaminant systems, glycerin/diethylene glycol, propylene glycol/diethylene glycol, and lactose/melamine and find that the sensitivity to contaminant for each system is 18%, 32%, and 4%, respectively. These levels are well-correlated to the minimum contaminant composition that can be detected by both verification and identification methods. Our studies indicate that the most important factor that determines the sensitivity of a spectral correlation measurement to the presence of contaminant is the relative Raman scattering cross section of the contaminant.
’ INTRODUCTION Raman technology has evolved significantly over the past decade from being primarily a laboratory-based spectroscopic technique to one that is being increasingly deployed in the field. Driven primarily by technological advances and miniaturization, Raman devices are now being used in a wide variety of areas such as pharmaceutical screening,1 law enforcement,2,3 and explosives detection.4 Many field applications rely heavily on the statistical comparison of test spectra with spectra of known compounds stored in Raman libraries. These library-based spectral correlation methods have roots in UV/vis/IR spectroscopy57 and provide a rapid, nondestructive avenue for sample characterization. Currently, Raman spectral correlation methods are at the core of various commercial devices, which operate primarily as “answer box” systems8,9 in one of two modes: identification (ID) or verification. The ID mode compares a test spectrum to all of the spectra in a reference library and ranks the library spectra using a hit-quality index (HQI). The HQI is a numerical quantity that characterizes the correlation between two spectra, and in ID mode the library spectrum that gives the highest HQI identifies the most probable composition of the test material. The verification mode is more common in the pharmaceutical industry10 and is widely used in the testing of incoming pharmaceutical raw materials and excipients. Whereas the ID mode compares each sample spectrum acquired to all of the library spectra, the verification mode operates by comparing the spectrum of each sample to a predetermined library reference spectrum. Verification tests use a minimum HQI threshold to assign a “Pass” or “Fail” determination to the This article not subject to U.S. Copyright. Published 2011 by the American Chemical Society
sample under study. A typical minimum HQI threshold for a “Pass” is 0.95. In light of recent cases11,12 of economically motivated adulteration (EMA) and the widespread use of Raman spectrometers for material verification, it is important to determine the sensitivity of spectral correlation methods with regards to their ability to detect the presence of contaminants in pharmaceutical raw materials. In this paper we report our efforts to characterize the sensitivity of Raman spectral correlation methods to contaminants during verification of pharmaceutical raw materials. We present a technique to characterize the sensitivity of spectral correlation methods for any binary system and describe several key factors that influence the sensitivity. Throughout this work sensitivity will be characterized as the minimum contaminant composition (%) that can be detected by the spectral correlation method. We have selected three contaminant/excipient systems for this work: (1) diethylene glycol in glycerin (DEG/Gly), (2) diethylene glycol in propylene glycol (DEG/PG), and (3) melamine in lactose (Mel/Lac). The DEG/Gly system has been implicated in several adverse events over the past century, including the event that led to the creation of the Food and Drug Administration. Propylene glycol is a common pharmaceutical excipient that is also susceptible to adulteration with DEG. Both of these systems are composed of materials with similar physical, chemical and spectroscopic properties. The Mel/Lac system is a Received: January 6, 2011 Accepted: April 10, 2011 Published: May 06, 2011 4061
dx.doi.org/10.1021/ac200040b | Anal. Chem. 2011, 83, 4061–4067
Analytical Chemistry
ARTICLE
another, and this provides a contrast to the DEG/PG and DEG/ Gly systems. We have developed a straightforward approach to quantitatively evaluate the sensitivity of spectral correlation methods by modeling the HQI as a function of contaminant concentration. While the systems discussed in this paper focus on pharmaceutical materials, the methods described can be applied to any system where spectral correlation methods are applied to mixtures.
’ EXPERIMENTAL METHODS
Figure 1. Pure component excipient/contaminant and mixture spectra for (A) DEG/Gly, (B) DEG/PG, and (C) Mel/Lac. Even at high concentrations of contaminant, little change can be detected visually or with the HQI for the two DEG-contaminated systems.
surrogate for melamine in powdered protein substances. Lactose was selected because it is a pharmaceutical excipient that is extracted from whey, and it is therefore conceivable that melamine could appear as a contaminant in lactose, albeit at low levels. The spectra of melamine and lactose are very different from one
Sample Preparation. All samples in this study were stored and analyzed in sealed 8-mm clear glass vials. The pharmaceutical excipients used to construct the Raman library used in this work were taken from an in-house collection of excipients from various manufacturers without further purification. Pure glycerin, DEG, and PG were purchased from Fisher Scientific and analyzed for moisture content using Karl Fischer titration (Mettler Toledo DL 31). All three liquid reagents were found to contain small amounts of water (
