With the increasing proliferation of counterfeit drug products, there is an incentive to screen drugs for legitimacy. One
method is to examine the tablet itself, which is usually a destructive operation. Another method that has been explored is
to characterize the packaging, which enables nondestructive screening of the product. Raman microscopy has been found to be
a useful tool, and it is often the tool of choice for these measurements. Raman mapping of tablets enables not only the identification
of components, but also the mapping pattern can enable matching manufacturing sites of different tablets. The study of packaging
materials can provide an identification of coating and a depth-profile measurement of thickness. In addition, unlike Fourier
transform infrared spectroscopy, which can only measure surface materials, Raman microscopy can provide spectra of dyes and
pigments, even if they are printed below a surface film.
In this column installment, we present results of tablet matching measurements and characterization of packaging. The examples
show how these measurements aid in the detection of counterfeit drugs.
Raman Tablet Mapping
Two tablets of a well-known drug product were acquired, one from a pharmacy in the United States and one via the mail from
an internet pharmacy. Initially, maps of the entire tablets were produced using a DuoScan Raman imaging system (Horiba Scientific)
with a 10× objective and a 200-μm pixel. Although maps were acquired in reasonable amounts of time, there did not seem to
be any variability in the data. That is, on the 200-μm scale, the tablets were homogeneous, at least for this product. So,
smaller maps were collected to make chemical and textural comparisons. This work was performed on an Evolution Raman microscope
(Horiba Scientific), an 800-mm focal length system, using a 600-grooves/mm grating, and a 532-nm laser. Maps were acquired
in the fingerprint region (200–1800 cm-1) and the CH–OH region (2700–3700 cm-1). LabSpec multivariate analysis software (Horiba Scientific) was used to create the Raman maps, using either the classical
least squares (CLS) algorithm when identifiable species could be seen in the data set, or the multivariate curve resolution
(MCR) algorithm when the chemical factors were not apparent. As the results will show, operating on the two data sets independently
produced verification of the conclusions.
Figure 1: Left: Raman maps of tablet purchased from a pharmacy in the United States produced with fingerprint spectra (top)
and CH–OH spectra (bottom). Right: Spectral loadings used to create the maps on the left. The red species is presumably the
active drug, having bands near 1600 cm-1 and above 3000 cm-1 that indicate unsaturation. The reference spectrum shown in dark green is that of a cellulose paper filter.
In reality, it was Witkowski of the Food and Drug Administration (FDA) Forensic Chemistry Center in Cincinnati, Ohio, who
originally proposed the use of Raman mapping for identifying chemical species and morphological distribution as a means of
differentiating counterfeit from authentic tablets, and for differentiating counterfeits from various manufacturing sites
(1). Since that work was done, developments in mapping hardware have improved data collection times and multivariate software
provides better quality images resulting from the use of scores mapping rather than intensity between cursors, which is subject
to much more noise.
Figure 2: Left: Raman maps of tablet purchased over the internet produced with fingerprint spectra (top) and CH–OH spectra
(bottom). Right: Spectral loadings used to create the maps on the left. The red species is presumably the active drug, having
bands near 1600 cm-1 and above 3000 cm-1 that indicate unsaturation. The reference spectrum shown in green is that of lactose. Note that use of the MCR algorithm
in the CH region identified the presence of a third species (shown in grey) whose spectrum is similar to that of Mg stearate.
Before data collection, the tablets need to have the coating (usually opaque and pigmented) removed and the surface flattened.
The maps shown in Figures 1 and 2 were acquired with step sizes between 5- and 15-μm, using the DuoScan Raman imaging system
to distribute the laser intensity over each pixel in the map. The SWIFT imaging option (Horiba Scientific) was used to further
reduce the acquisition times by eliminating stop and start operations at each data point. After the acquisition of a map,
the entire hyperspectral cube was baseline-flattened to improve the quality of the multivariate results. Then we either surfed
the file to identify regions of homogeneous spectra that could be used with the CLS algorithm or the MCR algorithm was invoked
to "find" the spectral components in the file. Figures 1 and 2 show maps for each of the two tablets based on the fingerprint
spectra and then the CH–OH spectra. The loadings ("spectra") used to produce the maps are also shown with reference spectra
of the excipient presented for comparison. Because the identity of the active pharmaceutical ingredient (API) was not known,
its spectrum could not be shown for comparison. However, APIs tend to be aromatic with large intensities above 1500 cm-1; the loadings shown in red in Figures 1 and 2 are presumed to be API and one can see that the red factors show the same spectra
from the two tablets. That is, the counterfeit tablet did contain the active ingredient.
Figure 3: Left: Raman maps of tablet purchased over the internet produced with fingerprint spectra. Right: Spectral loadings
used to create the maps on the left. The red species is presumably the active drug, having bands near 1600 cm-1 and above 3000 cm-1 that indicate unsaturation. Both excipient spectra (grey = hydrocarbon and blue = lactose) are heavily contaminated with
the API probably because the materials do not segregate on this length scale. The reference spectra shown represent lactose
in green, and Mg stearate in black.
Note that for the internet sample, whose results are shown in Figure 2, the CH–OH map has more morphological detail than the
fingerprint map. Also, the use of the MCR algorithm revealed the presence of a hydrocarbon like magnesium (Mg) stearate. If
we go back to the fingerprint map and examine the spectrum in the region where Mg stearate appeared in the CH region, then
that component can be added to the factors, producing the map shown in Figure 3. Surprisingly, in this case the CH–OH region
provided more information than the fingerprint region, as originally examined. We believe that this is because the fingerprint
region will be dominated by the unsaturated API, whereas in the CH region, all species have similar intensities.