Raman Barcode for Counterfeit Drug Product Detection - Analytical

Subscriber access provided by UOW Library

Article

The Raman Barcode for Counterfeit Drug Product Detection Latevi Sini Lawson, and Jason D Rodriguez Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.5b04636 • Publication Date (Web): 04 Apr 2016 Downloaded from http://pubs.acs.org on April 12, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

The Raman Barcode for Counterfeit Drug Product Detection Latevi S. Lawson and Jason D. Rodriguez * Division of Pharmaceutical Analysis, Center for Drug Evaluation and Research, US Food and Drug Administration, 645 S. Newstead Ave Saint Louis, MO 63110

* EMAIL: [email protected]

CORRESPONDING AUTHOR INFORMATION: Tel (314) 539-3855 Fax (314) 539-2113

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Abstract Potential infiltration of counterfeit drug products — containing the wrong or no active pharmaceutical ingredient (API) — into the bona fide drug supply poses a significant threat to consumers worldwide. Raman spectroscopy offers a rapid, non-destructive avenue to screen a high throughput of samples. Traditional qualitative Raman identification is typically done with spectral correlation methods that compare the spectrum of a reference sample to an unknown. This is often effective for pure materials but is quite challenging when dealing with drug products that contain different formulations of active and inactive ingredients. Typically, reliable identification of drug products using common spectral correlation algorithms can only be made if the specific product under study is present in the library of reference spectra, thereby limiting the scope of products that can be screened. In this paper, we introduce the concept of the Raman barcode for identification of drug products by comparing the known peaks in the API reference spectrum to the peaks present in the finished drug product under study. This method requires the transformation of the Raman spectra of both API and finished drug products into a barcode representation by assigning zero intensity to every spectral frequency except the frequencies that correspond to Raman peaks. By comparing the percentage of nonzero overlap between the expected API barcode and finished drug product barcode, the identity of API present can be confirmed. In this study, 18 approved finished drug products and nine simulated counterfeits were successfully identified with 100% accuracy utilizing this method.


Page 2 of 25

Page 3 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Introduction The presence of counterfeit drugs in the global supply chain has the potential to reach consumers across the world. Counterfeit and substandard drugs, once thought to be a problem only in emerging countries,1,2 can reach consumers through Internet websites which may present themselves as legitimate pharmacies3 with lower costs. There is no clear agreed-upon definition1 of counterfeit medicines, but may include drugs that contain the wrong2 or no active pharmaceutical ingredient (API), contain more or less than the labeled API, or contain an undeclared API.4-6 While all types of counterfeit drugs adversely affect consumer health, we will focus exclusively on counterfeit drugs that contain the wrong or no API.

These type of

counterfeit drugs are especially dangerous for consumers who may have treatable conditions, such as bacterial or malarial infections, which could remain untreated if they are subjected to fake medicines.7 Counterfeit medicines with no or the wrong APIs made up an estimated 50% of the recorded counterfeits from South America, Africa and Asia between 2003-2015.8,9 Over the past decade, Raman spectroscopy has been used successfully for detection of counterfeit drugs10-15 and is well-suited for counterfeit detection due to its non-invasive15 and non-destructive nature towards the sample under study. This allows Raman to be used to analyze drugs through the original packaging and preserve samples, should further analysis be warranted. Most of the published work using Raman spectroscopy focuses on the comparison of samples under study—or unknowns—to authentic reference spectra of drug products. Two common algorithms for comparison are spectral correlation based library methods10,15-17 and multivariatebased12,13,18-21 approaches. The latter approach is capable of looking at finer differences between samples such as batch differences or variability between different manufacturers and can be used to quantify ingredients in drug formulations. The library-based methods are qualitative and



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 25

better suited to screen a wide variety of drugs since rigorous method development does not need to be made for each entry in the spectral library, which is a requirement for multivariate approaches. For spectral library methods, the same comparison is made for each entry by use of a metric such as the spectral correlation coefficient or hit quality index.16

However, a key

challenge of this approach is that it is impractical to include every possible drug product in the spectral library. This would amount to ~100,000 unique entries for U.S. FDA-Approved drug products alone.22

Comparison of an authentic finished product spectrum from a particular

manufacturer may not yield an accurate assessment when compared with other manufactures of the same drug product due to differences in the formulation because coatings and inactive ingredient profiles differ depending on the manufacturer. This is especially the case for generic drugs where there may exist many different formulations of the same off-patent drug. Herein, we introduce a novel algorithm to screen antibiotic and antiviral finished drug products (FDPs) using Raman spectroscopy. Along with antimalarials, these two classes represent some of the most commonly counterfeited drugs worldwide.

1,2,18,23

The algorithm described in this

paper introduces the Raman barcode which allows us to overcome the necessity to build a spectral library with every possible approved formulation. Previous studies have employed the use of barcode representations of Raman spectra24 combined with multivariate approaches to classify samples. While similar in the aim of use of barcodes to reduce the complexity of the Raman signatures, the novel method presented here employs a straight forward comparison without any external algorithm to classify samples. Our method requires the transformation of the Raman spectra into a barcode representation by assigning zero intensity to every spectral frequency except the frequencies that correspond to Raman peaks, which are assigned a value of one. Moreover, the main motivation of the Raman barcode method is to allow for screening


Page 5 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


FDPs without having to develop comprehensive libraries. This approach represents a paradigm shift of using only the API Raman signatures as the starting point for FDP library method development.25 All that is needed is the spectrum of the API present in the FDP be populated in the spectral library. The Raman barcode method was tested on US FDA-Approved solid oral dosage forms purchased commercially. These drugs contained API loadings in the 40-80% range. Extrapolation to lower or higher API loadings would require further study. The 40%80% range is representative of commercially-available antibiotics and antiviral drugs. In this work, a reference library containing Raman spectra for 98 unique APIs was transformed into a barcode library. The library was then compared to the barcode spectrum of the samples under study. Six types of antibiotic and antiviral drugs were used to test the barcode method. Three different manufactures were included for each type of drug. counterfeit samples to challenge the method.

We also prepared simulated

The simulated counterfeits were designed to

represent counterfeits having no or wrong API. Materials and Methods API and Finished Drug Product Spectral Collection The Raman spectra for 98 unique APIs were used to populate a reference library, as has been described in detail elsewhere26 and the list is included in Table 1. Raman spectra for the API reference samples were acquired on an EZ-Raman I portable Raman Spectrometer (Enwave Optronics Inc.) using the output of a 785 nm laser (340 mW). The TE-cooled CCD detector was operated at -50°C. Variable acquisition times ranging from 1 to 36 s were used for each API. Acquisition was determined by the software using an algorithm which first performs a quick ~ 1 second scan on each sample and determines the integration time needed to acquire a spectrum with approximately 50,000 counts. The FDPs tested for this study were obtained commercially



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

and used without further sample preparation. Except for simulated counterfeit drugs which were prepared for this study, all FDPs were analyzed intact and were of FDA-Approved drugs. Six different types of FDPs from eighteen different manufacturers were used to develop a test set. The six types of FDPs used were acyclovir, amoxicillin, cephalexin, ciprofloxacin HCl, doxycycline hyclate and levofloxacin. The FDP spectra used as the test set were acquired on Raman microscope spectrometer (Kaiser Optical Systems Inc., Ann Arbor, MI) using a 785 nm excitation with a spot size of 3 mm that produced roughly 260 mW of power at the sample. Each acquisition consisted of a 1 s integration time and 30 accumulations. For each manufacturer, a total of six tablets or capsules were analyzed twice, once for each tablet side. Two random orientations were used for capsule analysis. Thus, each FDP in the test set provided 12 unique tests for each test sample.


Page 6 of 25

Page 7 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table 1: The list of APIs in the reference Raman library. The highlighted entries correspond to the APIs that are expected to be present in the finished drug product evaluated in this study. Active Pharmaceutical Ingredient (API) in Library Code #

API

Code #

1

Acetaminophen

26

2

Acyclovir

27

API Dexamethasone Sodium Phosphate Dextromethorphan HBr

Code #

API

Code #

API

51

Hydroxyzine Pamoate

76

Prednisone

52

Ibuprofen

77

Promethazine HCl

3

Albuterol Sulfate

28

Diazepam

53

Incromega

78

Propranolol HCl

4

Amitriptyline HCl

29

Diclofenac Sodium

54

Lactulose

79

Pseudoephedrine HCl

5

Amlodipine Besylate

30

Diphenhydramine HCl

55

Levetiracetam

80

Pyridoxine HCl

6

Amoxicillin Trihydrate

31

Docusate Sodium

56

Levofloxacin

81

Quetiapine Fumarate

7

Aripiprazole

32

Donepezil HCl

57

Lidocaine

82

Quinine Sulfate

8

Ascorbic Acid

33

Doxycycline Hyclate

58

Lidocaine HCl

83

Ranitidine HCl

9

Atorvastatin Ca

34

Efavirenz

59

Lisinopril

84

Risperidone

10

Atropine Sulfate

35

Erythromycin

60

Losartan Potassium

85

Salicylic Acid

11

Azithromycin Monohydrate

36

Esomeprazole Mg Trihydrate

37

Ethinyl Estradiol

61

86

Simvastatin

87

Sodium Copper Chlorophyllin

12

Budesonide

13

Bupropion HCl

38

Famotidine

63

Metoprolol Tartrate

88

Sulfamethoxazole

14

Carisoprodol

39

Fenofibrate

64

Metronidazole

89

Sulfanilamide

15

Cefepime HCl

40

Fluconazole

65

Montelukast Sodium

90

Taurine

16

Centirizine HCl

41

Fluoxetine HCl

66

Naproxen

91

Thiamine HCl

67

Naproxen Sodium

92

Topiramate

68

Niacin

93

17

Cephalexin

42

Fluticasone Propionate

18

Chloroquine Phosphate

43

Furosemide

62

Methylprednisolone Acetate Metoprolol Succinate

Tramadol HCl Triamcinolone Acetonide Triamcinolone Diacetate

19

Ciprofloxacin

44

Gabapentin

69

Niacinamide

94

20

Ciprofloxacin HCl

45

Gentamicin Sulfate

70

Olanzapine

95

21

Clindamycin HCl

46

Guaifenesin

71

Oxcarbazepine

96

Valsartan

47

Hydrochlorothiazide

72

Pantethine

97

Venlafaxine HCl

48

Hydrocortisone

73

Pioglitazone HCl

98

Venlafaxine Sodium

22 23

Clindamycin Phosphate Clobetasol Propionate

24

Clopidogrel Bisulfate

49

Hydrocortisone Acetate

74

Potassium Iodide

25

Coenzyme Q10

50

Hydroxychloroquine

75

Prednisolone

Note: Highlighted Boxes correspond to API that match the label claim of products used to challenge the method.

Simulated Counterfeit Preparation Nine simulated counterfeit capsules were prepared by emptying the contents of authentic 500 mg amoxicillin capsules and refilling them with three different simulated counterfeit mixtures: excipient-only, acetaminophen/excipient mixture and naproxen sodium/excipient mixture. The



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

excipient mixture was prepared by mixing three common excipients: cornstarch, ethyl cellulose, and silicon dioxide. The excipients were placed in a screw-top amber jar and the contents in the amber jar were inverted and mixed with a SPEX 5100 mixer/mill for 2 minutes. The excipient mixture final composition (w/w%) was 53% cornstarch, 24% ethyl cellulose and 23% silicon dioxide. The excipient mixture was further combined with acetaminophen and naproxen sodium APIs. These API-containing mixtures were made by combining (w/w%) 73% API and 27% excipient mixture in an amber jar. Each of the API/excipient mixtures was mixed well for 2 minutes with a SPEX 5100 mixer/mill. The APIs used were taken from an in-house library and used without further sample preparation. Both of these APIs are listed in Table 1. Each of the three simulated counterfeit mixtures was used to fill the empty 500 mg amoxicillin capsules. Before refilling, the contents of each capsule were removed and the empty capsules were rinsed three times with cornstarch to remove trace amounts of the original contents. Three capsules were prepared for each of the three mixtures, producing a total of nine simulated counterfeits. The average mass of the original amoxicillin capsules used in this study was 685 ± 9 mg. The mass of the simulated counterfeits varied with the mixture used to fill the capsules since the mixtures vary in density. The final masses of the three capsules refilled with only excipients was roughly 457 ± 13 mg; the final mass of the capsules refilled with acetaminophen/excipient was roughly 652 ± 24 mg; and the final mass of the capsules refilled with naproxen sodium/excipient was roughly 621 ± 28 mg.

Raman Spectral Analysis and Model Development A flowchart detailing the steps in the Raman barcode method is shown in Figure 1. Analysis was done using custom code written in MATLAB (Version 8.4, R2014b, MathWorks Inc., Natick, MA).

As shown in Figure 1, data processing included the removal of fluorescence


Page 8 of 25

Page 9 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


background followed by smoothing of each Raman spectrum. Next, peak frequencies in each Raman spectrum were identified and a unique barcode representation for each Raman spectrum was generated. The barcode was then used for spectral comparison. A detailed summary of each section is given below. Background Subtraction The procedure used to generate the Raman barcode library using the API reference Raman library is described below.

First, each reference library spectrum underwent removal of

fluorescence background using the Lieber method, which utilizes an iterative polynomial fit to automate background subtraction.27

A total of 100 iterations with a 9th order polynomial was

used to subtract the background from each Raman spectrum. Smoothing and Peak Identification Following background subtraction, each API reference spectrum was smoothed (Savitzky-Golay 0th order, 14 point window) and a peak search algorithm was used to identify the peak frequencies using the “Peakfind” function from PLS Toolbox Version 7 (Eigenvector Research Inc., Manson, WA). The “Peakfind” function’s parameters were adjusted so that at least 95% of the peaks in each API spectrum were identified.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 1: Flow chart depicting the steps required to evaluate FDP for API using the Raman barcode method.

Generating API Raman Barcodes The peaks selected to represent the Raman barcode signature for an API were those having intensities greater than 20% of the average peak intensity, which were calculated based on the intensity of all peaks found in the Raman spectrum. This constraint limited those peaks used in the Raman barcode to those that were most prominent in the API Raman spectrum, therefore avoiding spurious peaks in the barcode. This algorithm is shown pictorially in Figure 2. The


Page 10 of 25

Page 11 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


peaks with intensity values greater than the 20% threshold are denoted by green circles in Figure 2A. These peaks are the only ones used for the Raman barcode, shown in Figure 2B. Figure 2B shows the typical Raman barcode signature which was generated by identifying the peak frequencies in the Raman spectrum above the 20% threshold in Figure 2A. The intensity value of one was assigned at the peaks found using the peak search algorithm and a value of zero is assigned at all other frequencies.

Figure 2: The figure depicts how to generate a Raman barcode. All peaks denoted with a circle are included in the barcode spectrum.

Generating the FDP Raman barcodes The entire process described above is used for the Raman spectra of the FDPs in the test set with a slight modification in that the 20% of the average peak intensity constraint was not used when generating a Raman barcode for FDPs. This is because some of the Raman signal



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 25

from the API may become obscured by the fluorescent background from excipients in the FDP. Thus, if a threshold were used to exclude the less prominent Raman peaks in the FDP spectrum from being identified, then some of the API peaks in the FDP spectrum would be erroneously eliminated from the barcode representation of the FDP.

Raman Barcode for Evaluation of API Identification In this study, Raman barcodes were used to confirm the identity and presence of the labeled API in FDPs. If an acceptable percentage of the API barcode signature overlaps with the barcode spectrum of a FDP, then the API is considered present in the FDP. The acceptable percent of an API barcode spectrum required for identification is called the % match criterion. For example, if the match criterion is set to 75%, then this means that at least 75% of the API barcode must overlap with FDP barcode. Comparison between the API reference Raman barcodes and FDPs Raman barcodes is a two-step process that is further explained in Figure 3. The first step is to pad the API’s bars. Padding assigns intensity values of one to frequencies adjacent to

Figure 3: The figure depicts how the Raman barcodes for the FDP and API are compared.

peak frequencies in both directions. Padding increases the width of the API bars in the API barcode spectrum to compensate for the resolution differences between different spectrometers.


Page 13 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


In this paper, we used 10 cm-1 padding for each bar in the barcode spectrum since this represented the middle resolution range for most commercial Raman spectrometers in our laboratory.17 Bars from the FDP barcode spectrum are removed if they do not overlap with bars in the padded API barcode spectrum. The frequency position of the remaining bars in the FDP spectrum were shifted to the center frequency of the matching bar in the padded API barcode spectrum. Padding in-depth An in-depth example of the padding process is shown in Figure 4. The image depicts an original reference API and a padded API barcode spectrum. The original binary spectrum has ones and zeros assigned to peak and non-peak frequencies respectively. The table at the bottom

Original Barcode

0

0

0

0

1

0

0

0

0

Padded Barcode

0

0

0

1

1

1

0

0

0

Raman Shift (cm-1)

771

772

773

774

775

776

777

778

779

Figure 4: The original API and a padded API Raman barcode.

of the image depicts a padding of 2 cm-1 for the Raman bar at 775 cm-1. To achieve this, values of one are added to the positive and negative adjacent frequency position of the Raman bar. To use a padding of four, values of one would extend from 773 cm-1 to 777 cm-1. Padding values must be even numbers to provide an equal distribution for the bar widths on either side of the



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 25

peaks. Padding accounts for spectral shifts by increasing the width of barcode bars and is depicted in Figure 3. This extends the utility of the Raman barcode method since small shifts resulting from differences in manufacturer calibration of the wavelength axis often contribute to difficulties when comparing Raman spectra acquired on different instruments. Relationship Between the % match and the Padding Size By varying the padding prior to generating results at various % match between the barcodes of the library spectrum and the sample under study, the parameters required for a robust algorithm without false positives and false negatives were determined. The parameters (pad size and % match) were optimized using the 216 spectra collected from 18 different manufacturers as a part of the FDP test set. The results are described in detail in the next section. Figure 5 shows the relationship between the pad size and the % match. It is also clear from Figure 5 that the pad size changed the optimal match criterion required.

Figure 5: The figure depicts how frequency padding effects the overlap between the API library barcodes and drug product barcodes.

A padding of 10 cm-1 requires the match criteria of at least 63% in order to avoid any false positive results and a maximum of 79% to avoid any false negative results. Therefore, match criteria should be set somewhere between 63% and 79% when using a padding of 10 cm-1. In general, higher match criteria are needed when the padding is increased; however, the false positive and false negative match criteria converge as the pad width is increased. A pad of 2 cm-


Page 15 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


1

, for example, requires match criteria between 32% and 57% to avoid false positives and false

negatives, respectively, while a pad of 16 cm-1 will always produce either a false positive or false negative, regardless of the match criteria. From Figure 5, it is obvious that regardless of the match criteria, a false positive or false negative was generated when no padding was applied in the evaluation. This indicated that the peak positions of the pure API and those of the API in formulation were offset slightly. This was likely due to differences in the spectrometers used for data collection since the API library and FDP test set were acquired on different Raman instruments. For the work presented in this paper, a 75% match was required for positive identification between an API reference barcode and FDP barcode.

Results and Discussion The API reference library listed in Table 1 was transformed into a collection of Raman barcode signatures shown in Figure 6. Each row in Figure 6 is a Raman barcode spectrum of an individual API. The row number can be used to identify the API utilizing the list in Table 1. The frequencies shown in white in Figure 6 correspond to peaks in the API Raman spectrum and those frequencies shown in black correspond to spectral regions where there were no peaks present in the API Raman spectrum or where peaks were less than 20% of the average peak intensity as discussed in the materials and method section of this paper.

Each entry in the

Raman barcode library shown in Figure 6 was compared against all other entries using varying match criteria to gain a better understanding on the specificity of the Raman barcode method and specifically to test the occurrence of false-positive identifications. The results of this survey are given in Figure 7, which shows a square



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 6: The barcode spectra for all the active pharmaceutical ingredients composing the library.

matrix with the index of APIs in the reference library listed on both the horizontal and vertical axes. These indices are the same as those listed in Table 1. Based on this type of comparison, a positive identification was made if the two elements being compared have a match percentage above the setting. All possible match percentages were tested from 25-100% and four different match criterias are shown in Figure 7: 25%, 50%, 75% and 100%. White off-diagonal elements indicate that two different APIs shared the given percent of barcode features. Elements on the diagonal should always have a positive identification made since these elements represent comparing the exact barcode against itself. Elements with positive identifications predicted were colored white in the square matrix. For example, a white-colored element can be seen in Figure 7B at the coordinate (10,59), indicating that no less than 50% of the bars of the barcode spectra for API 10 can also be identified in API 59. APIs 10 and 59 correspond to atropine and Lisinopril, respectively. A noticeable increase in specificity of the algorithm was observed when going from a 25% match requirement to a 50% match requirement. At 25%, all the API’s appear


Page 16 of 25

Page 17 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


to be similar as the overwhelming majority of the matrix is white-colored. A noticeable improvement was made when going from 50% to 75% match requirement and no noticeable improvement was apparent when requiring a 100% peak match, thus providing a firm basis for choosing 75% percent match for the Raman Barcode algorithm. Only two API have at least 75% of their barcode spectra overlap, venlafaxine HCl (97) and venlafaxine Sodium (98). This lone false positive identification pair can be avoided if the match criteria is set to 80 %, but it is clear from Figure 7 that a 75% match criteria would not yields any false positive results other than the venlafaxine pair.

Figure 7: The barcode spectrum of each of the 98 API in the library was compared to themselves and to all the other spectra in the API library.

A total of 18 FDPs (six different APIs—three different manufacturers each) were used to challenge the Raman barcode method for testing whether the declared API was present in the FDP. In Table 2, the declared API is indicated along with the three different FDPs from different manufacturers.

The columns indicate which API Raman barcode was used to screen the



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

samples. All 98 APIs were used to screen the 18 FDPs, but only the results for the six APIs declared on the labeling of the FDPs used in the test set are shown. Twelve spectra for each FDP tested were evaluated. If the Raman barcode method indicates that a FDP dose contains an API, then a green box is used as an indicator. Red boxes indicate that the FDP dosage did not contain the API for which it was screened. As seen in Table 2, each dose was identified as containing only one API (green). The results in Table 2 indicate that confirming the presence of the API on the label can be achieved by comparing the Raman barcodes for the pure API reference sample and finished products.


Page 18 of 25

Page 19 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Table 2: Results obtained from the Raman barcode method for six different types of drugs. The API was correctly identified in all products tested. -1

Results utilizing the Raman barcode method with a pad size of 10 cm and a match criteria of 75% Acyclovir (FDP)

Acyclovir (FDP)

Amoxicillin (FDP)

Cephalexin (FDP)

Ciprofloxacin (FDP)

Doxycycline Hyclate (FDP)

Levofloxacin (FDP)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

Amoxicillin (API)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

Cephalexin (API)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

Ciprofloxacin (API)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

Doxycycline HCL (API)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

Levofloxacin (API)

man 1

12 of 12

man 2

12 of 12

man 3

12 of 12

man = manufacturer; API = active pharmaceutical ingredient; FDP = finished Drug Product; green color indicates a match

Simulated counterfeit samples were also prepared to challenge the Raman barcode method. Prior to preparation of the simulated counterfeit samples, each of the 500 mg



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

amoxicillin capsules was analyzed using the barcode method to serve as an experimental control in order to show that the capsules would pass the method before being transformed into simulated counterfeits. As shown in Figure 8, all the nine capsules delivered a match for amoxicillin. All the simulated counterfeit capsules matched their known API contents with 100% accuracy.

Simulated counterfeit capsules (#1-3), which contained naproxen

sodium/excipient mixture, matched only for naproxen sodium. It should be noted that naproxen did not deliver a match and is a further proof that the 75% match criteria was sufficient to distinguish between different salt forms for the API library with the exception of venlafaxine which was discussed above. Simulated counterfeit capsules (#4-6), which contained acetaminophen/excipient mixture, matched only for acetaminophen. Simulated counterfeit capsules (#7-9) which contained only the excipient mixture did not exhibit any match. These results indicate that the barcode method can discriminate between closely related API.


Page 20 of 25

Page 21 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 8: The identification of simulated counterfeit capsules employing the barcode method at 75% match criteria.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 25

Conclusions In this paper, we introduced a binary barcode comparison method to identify APIs in FDPs. While this study included drugs with API loadings in the 40-80% range, the use of this method may be expanded with further testing on drugs with API loadings above or below this range. The range of API loadings tested is representative of FDA-Approved antibiotic and antiviral drugs. By utilizing only the frequency component of Raman spectra, we were able to generate unique Raman barcode representations of API and FDP spectra. By comparing the Raman barcode of FDP doses to that of APIs, we verified the presence of APIs in the FDPs. The method was shown to correctly predict the declared API present in 18 different commercial samples and nine simulated counterfeit samples. The barcode method parameters were optimized to eliminate both false positives and false negatives. Because of the padding process, the method was also able to achieve comparison across spectrometers without any extra correction or standardization required for comparison. The Raman barcode method streamlines spectral library development since padding accounts for any frequency shift that arise when spectral libraries are transferred to different instruments at different locations. Since the barcode method emphasizes frequency and not intensity, our method eliminates the need to correct the spectral intensities when they are transferred between different instruments. Although the intensity information is lost, the barcode method is effective because peak positions are strong indicators of molecular functional groups present in substances analyzed via Raman spectroscopy. Perhaps the biggest contribution of this work is that that the barcode method eliminates the need to have a comprehensive spectral library of FDPs in order to make qualitative authentication of API in FDPs. The Raman barcode has the potential to transform the way library-based screening is done by allowing for streamline library building and transfer across different spectrometers and users. Such distribution of


Page 23 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


spectral libraries could help in screening more medicines before reaching consumers and protect the pharmaceutical supply chain from counterfeits containing no API or the wrong API. ACKNOWLEDGMENT

This project was supported in part by the CDER Critical Path and Regulatory Science & Review Enhancement Programs. This project was supported in part by an appointment (L.S.L) to the Research Participation Program at the Center for Drug Evaluation and Research administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and the U.S. Food and Drug Administration. DISCLAIMER This article reflects the views of the authors and should not be construed to represent FDA’s views or policies. SUPPORTING INFORMATION Additional details on FDPs with no or wrong API and our peak finding algorithm are available free of charge via the Internet at http://pubs.acs.org. REFERENCES (1) Degardin, K.; Roggo, Y.; Margot, P. J. Pharm. Biomed. Anal. 2014, 87, 167. (2) Blackstone, E. A.; Fuhr, J. P., Jr.; Pociask, S. Am. Health Drug Benefits 2014, 7, 216. (3) (4) Buhse, L.; Gryniewicz-Ruzicka, C. M.; Dunn, J. D.; Arzhantsev, S.; Spencer, J. A.; Rodriguez, J.; Westenberger, B. J.; Kauffman, J. F. Abstracts of Papers of the American Chemical Society 2012, 244. (5) Dunn, J. D.; Gryniewicz-Ruzicka, C. M.; Kauffman, J. F.; Westenberger, B. J.; Buhse, L. F. J. Pharm. Biomed. Anal. 2011, 54, 469. (6) Dunn, J. D.; Gryniewicz-Ruzicka, C. M.; Mans, D. J.; Mecker-Pogue, L. C.; Kauffman, J. F.; Westenberger, B. J.; Buhse, L. F. J. Pharm. Biomed. Anal. 2012, 71, 18. (7) Liang, B. A. Am. J. Law Med. 2006, 32, 279.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 25

(8) Krech, L. A.; El-Hadri, L.; Evans, L.; Fouche, T.; Hajjou, M.; Lukulay, P.; Phanouvong, S.; Pribluda, V.; Roth, L. The Medicines Quality Database: a free public resource, World Health Organization, 2014. (9) "Promoting the Quality of Medicines in Developing Countries (PQM)" http://www.usp.org/global-health-programs/promoting-quality-medicines-pqmusaid. Accesed March 8, 2016. (10) Assi, S.; Watt, R.; Moffat, T. Spectrosc. 2011, 36. (11) de Veij, M.; Deneckere, A.; Vandenabeele, P.; de Kaste, D.; Moens, L. J. Pharm. Biomed. Anal. 2008, 46, 303. (12) Degardin, K.; Roggo, Y.; Been, F.; Margot, P. Anal. Chim. Acta 2011, 705, 334. (13) Fraser, S. J.; Oughton, J.; Batten, W. A.; Clark, A. S. S.; Schmierer, D. M.; Gordon, K. C.; Strachan, C. J. J Raman Spectrosc. 2013, 44, 1172. (14) Kwok, K.; Taylor, L. S. Vib. Spectrosc. 2012, 61, 176. (15) Willis, R. C. Anal. Chem. 2007, 79, 1773. (16) Rodriguez, J. D.; Westenberger, B. J.; Buhse, L. F.; Kauffman, J. F. Anal. Chem. 2011, 83, 4061. (17) Rodriguez, J. D.; Westenberger, B. J.; Buhse, L. F.; Kauffman, J. F. Analyst 2011, 136, 4232. (18) Been, F.; Roggo, Y.; Degardin, K.; Esseiva, P.; Margot, P. Forensic Sci Int. 2011, 211, 83. (19) de Peinder, P.; Vredenbregt, M. J.; Visser, T.; de Kaste, D. J. Pharm. Biomed. Anal.2008, 47, 688. (20) Kwok, K.; Taylor, L. S. J. Pharm. Biomed. Anal.2012, 66, 126. (21) Roggo, Y.; Degardin, K.; Margot, P. Talanta 2010, 81, 988. (22) U.S. Department of Health and Human Services, U.S. Food and Drug Administration http://www.fda.gov/Drugs/InformationOnDrugs/ucm142438.htm. Accessed March 24, 2016. (23) Mackey, T. K.; Liang, B. A.; York, P.; Kubic, T. Am. J. Trop. Med. Hyg. 2015, 92, 59. (24) Patel, I. S.; Premasiri, W. R.; Moir, D. T.; Ziegler, L. D. J Raman Spectrosc. 2008, 39, 1660. (25) Loethen, Y. L.; Rodriguez, J. D. Analyst 2015. (26) Rodriguez, J. D.; Skaggs, S. K.; Johny, M. M.; Srivastava, H. K.; Arzhantsev, S.; Loethen, Y. L.; Kauffman, J. F.; Buhse, L. F. Am. Pharm. Rev. 2014, 17, 10. (27) Lieber, C. A.; Mahadevan-Jansen, A. Appl. Spectrosc. 2003, 57, 1363.


Page 25 of 25

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Graphical abstract 82x53mm (300 x 300 DPI)


Raman Barcode for Counterfeit Drug Product Detection - Analytical

Recommend Documents