Double metal cyanide (DMC) catalyst is widely used for the alkoxylation reaction to produce polyether polyol from ethylene/propylene/butylene oxides. It is challenging to optimize the synthesis process, due to the lack of real-time understanding of the speciation of the reaction mixture. Powder X-ray diffraction (XRD), transmission electron microscopy (TEM), and actual alkoxylation reaction performance are effective ways for the evaluation of success of each synthesis, but it is difficult to guide process optimization. An in situ Raman method is developed in this study to monitor the DMC catalyst synthesis in real time to accelerate the process optimization. The synthesis involves the reaction of ZnCl

 (ZHCC). ZHCC is then converted to DMC in the presence of excess of t-butanol and ZnCl

 in the second step. Characteristic KHCC Raman peaks were observed at 2138 and 2153 cm

, respectively. This enables realtime tracking of both steps’ conversion. Both t-butanol and ZnCl

 concentrations were found to substantially influence the kinetics of DMC formation, but not the Raman spectra of the final DMC products. The reaction time could be adjusted from hours to minutes through the control of reactant concentrations.

Double metal cyanide (DMC) and potassium hydroxide serve as catalysts for the manufacture of polyether polyols from ethylene oxide (EO), propylene oxide (PO), and butylene oxide (BO). The heterogeneous catalyst, DMC, is replacing the incumbent potassium hydroxide (KOH) technology which requires an added neutralization step (1). Another drawback with KOH catalyst is the competitive side polymerization initiated by allylic alkoxides affecting the polyol end properties, but this can be resolved by using DMC. The DMC technology has a reduced “footprint” compared to KOH, as it has lower unreacted oxide concentrations (hence safer design), lower capital and maintenance costs, and reduced energy consumption.

DMC catalyst technology was developed in the 1960s; however, there is still debate over the actual structure of the Lewis acid and the active site for alkoxylation. The quality of DMC is highly sensitive to the method of preparation and is dictated by a combination of factors: order and rate of addition of reactants, type of agitation, concentration of reactants and rinse solutions, and mixing time. A potential risk with making DMC is the final material can be inactive. Synthesizing DMC in a reproducible manner is therefore the first step in de-bottlenecking the scale-up process. In the invention by Kuyper and associates, it is noted that alkali metal ions, in particular sodium and potassium ions, act as catalyst poisons in the polymerization reaction of epoxides using DMC catalysts (2). Another observation by Le-Khac is the necessity to wash away potassium chloride (KCl), as it will render the catalyst inactive if not removed (3). Based on prior literature, catalyst characterization is carried out after the entire work flow (multiple washing and drying steps) which is time- and effortintensive. These characterization tools typically include X-ray powder diffraction, elemental analysis, and scanning electron microscopy (4).

Huang and associates synthesized DMC with various water soluble Zn salts such as zinc chloride, zinc sulfate, and zinc acetate with complexing agents such as t-butanol, dioxane, tetrahydrofuran, and dimethoxyethane to observe that the DMC made using zinc chloride and t-butanol (t-BuOH) showed highest specific rate of polyether polyol formation (5). Catalytic activity is strongly dependent on the composition of catalysts and the morphology of DMC complexes (6). Several earlier studies demonstrated that Raman spectroscopy can potentially be a sensitive technique to investigate metal cyanides (7–9). In this work, the use of Raman spectroscopy for real-time monitoring of DMC formation is reported, in particular the DMC containing Zn and Co. The following two steps are monitored.

Reaction between potassium hexacyanocobaltate (KHCC) and zinc chloride (ZnCl

) resulting in the formation of zinc hexacyanocobaltate (ZHCC): 3ZnCl

Complexation of ZHCC with t-butanol (t-BuOH) and water in the presence of excess ZnCl

Note that from here on, DMC specifically refers to the active DMC catalyst, but not the KHCC starting material, or the ZHCC intermediate.

In industrial research, it is important to find optimal processes to produce product. Often, many competing fac tors are involved. Therefore, it is challenging to understand how these variables may impact the process. Raman spectroscopy is found to be a powerful tool that yields real-time process information enabling researchers to track reaction conversion without having to carrying out tedious ex situ analysis, and thus, accelerates overall project progress. To demonstrate the utility of in situ Raman spectroscopy for such process development and optimization, two example synthesis routes are included, with DMC-A route being the initial recipe prior to optimization, and DMC-B route being the improved process.

Potassium hexacyanocobaltate (III) (≥ 97.0%), zinc chloride (≥ 98.0%), t-butanol (≥ 99.0%), potassium chloride (≥ 99.0%) were used as received from MilliporeSigma. Zinc hexacyanocobaltate was purchased from Alfa Chemistry.

A design of experiment (DOE) approach was used to explore the vast process variable space. Results from two such experiments are presented to demonstrate the utility of in situ Raman monitoring. These two synthesis protocols will be referred to as DMC-A and DMC-B, with the major difference between them being the relative molar ratio of ZnCl

 to KHCC, the sequence of reagent addition, and also several other process differences as described below.

In Step 1 of DMC-A, aqueous zinc chloride solution (97.5 g, 38.5 wt% ZnCl2) is added at 4 mL/min using a peristaltic pump to a solution containing KHCC and water (222 g, 5.4 wt% KHCC). The process is carried out in a 1 L jacketed glass reactor under atmospheric pressure at 300 in the aqueous solution, which was almost identical to those observed in the solid form. As ZnCl

 solution was added, both peaks diminished in a proportional manner, and concurrently a pair of new peaks at 2185 and 2204 cm

started to gain intensity. These new peaks are believed to be from ZHCC, since their positions closely matched the two lower frequency peaks observed in solid ZHCC at 2192 and 2205 cm

 It can then be inferred based on such results that the 2224 cm

 peak observed in the earlier sample (see Figure 1) is likely from a partially dehydrated material.

The almost identical peak positions of ZHCC to that reported for DMC-I (12) provided strong evidence that the DMC-I in the earlier paper was actually ZHCC hydrate, instead of DMC. A close examination of the reference showed that the DMC-I solid cake was washed with DI water, which is believed to convert DMC back to ZHCC. Such sample preparation induced changes further highlight the importance of in situ analysis for such unstable samples.

The sapphire window gave rise to several sharp features at 378, 418, 431, 450, 577, and 750 cm

. Several other weak features could also be observed between 200 and 300 cm

 solution itself showed a weak feature around 300 cm

 feature was consistent with that of ZHCC catalyst (see red spectrum in Figure 2 for comparison). It should be noted that, although the end ZHCC signal appeared to be substantially weaker than that of starting KHCC, this was mostly due to the decrease in the sampling volume as the system changed from a clear KHCC solution to a highly turbid ZHCC suspension. Such a turbidity change also led to the suppression of the water signal above 3000 cm

Figure 3 shows representative spectra from DMC-A during the ZHCC to DMC conversion. The number in the figure refers to the spectrum number, with 115 being the first spectrum collected during this stage. All spectra are normalized to the t-BuOH 750 cm

 peak for ease of comparison. A monotonic decrease of the ZHCC peaks at 2185 and 2206 cm

 was observed concurrent with a monotonic increase of the DMC peaks at 2203 and 2225 cm

, spectral changes corresponding to the ZHCC to DMC transformation could also be observed, though they were of weaker intensity and sitting on a large slope of a strong feature (unknown origin).

The close proximity of the DMC and ZHCC CN stretch peaks would lead to significant inter ference based on simple peak area analysis. As a result, classical least squares (CLS) analysis was used for spectral deconvolution (13,14). The two pure component spectra used were the starting ZHCC spectrum and the end spectrum of the final DMC product. The 2160 to 2240 cm

spectral region of each reac tion spectrum, after a linear baseline correction, is assumed to be a linear combination of these two pure component spectra. The weight, or CLS response, of both components can then be used to track their relative concentration more effectively than univariate spectral analysis such as peak height or area. Figure 4 shows the CLS response of the ZHCC and DMC components.

The rate of ZHCC to DMC conversion appeared to accelerate in the first 40 min of the reaction, presumably due to the continuous feed of ZnCl

 during this period, which led to a faster reaction rate. Upon completion of ZnCl

 feed, the conversion continued but at a slower rate, though the rate appeared to increase about 5 hours into the reaction, and full conversion appeared to be reached after about 15 hours.

DMC-A is time-intensive and thus costly to scale up as an industrial process. Based on reaction kinetics insights from both the in situ Raman results and other complementary tools, a new DMC-B procedure is identified. Instead of adding tBuOH in a separate step, it is included initially in the solution. A higher ZnCl2 concentration was also used. These process changes led to a greatly accelerated process. Figure 5 shows the representative reaction spectra from synthesis of DMC-B. The overall spectral changes are consistent with those observed in DMC-A, with KHCC bands first giving way to transient ZHCC bands (similar to that observed in DMC-A Step 1) which eventually all converged to DMC bands (similar to that observed in DMC-A Step 2).

The rate of reaction for DMC-B was an order of magnitude faster than that observed for DMC-A, as clearly illustrated by comparing Figure 4 to Figure 6. The ZHCC signal appeared to be completely consumed after 0.5 hr, which coincided with the DMC signal reaching its maximum.

In situ Raman was used to monitor the DMC synthesis for the first time. Characteristic CN stretch bands were identified from KHCC, ZHCC, and DMC, enabling real-time tracking of the reactions throughout the KHCC to ZHCC and finally to DMC stages. While the KHCC to ZHCC appears to be robust and straightforward, the ZHCC to DMC reaction is demonstrated to be more sensitive to process conditions. The use of the in situ Raman method was especially useful in this study because it eliminated the need of multiple sample withdrawal for off-line analysis. Real-time reaction progress information from in situ Raman results enabled researchers to accurately determine the reaction end point. Even though conventional off-line analytical tools such as XRD, TEM, and actual alkoxylation reaction performance test are needed to fully characterize the final DMC catalyst product, in situ Raman results accelerated the completion of the DOE study, which drastically reduced reaction time from more than 10 hours to about 0.5 hours. This study clearly demonstrates the power of in situ Raman monitoring of industrial processes.

(1) Herzberger, J.; Niederer, K.; Pohlit, H.; Seiwert, J.; Worm, M.; Wurm, F. R.; Frey, H. Polymerization of Ethylene Oxide, Propylene Oxide, and Other Alkylene Oxides: Synthesis, Novel Polymer Architectures, and Bioconjugation. 

 2170–2243 (2016). DOI: 10.1021/acs.chemrev.5b00441

(2) Kuyper, J.; Schaik-Struykenkamp, P. v. Process for the Polymerization of Epoxides. U.S. Patent US4472560A, 1984.

(3) Le-Khac, B. Highly Active Double Metal Cyanide Catalysts. U.S. Patent US5482908A, 1996.

(4) O’Connor, J. M.; McAdon, M. H.; Laycock, D. E. 

New DMC Catalysts for Manufacturing Polyols. In API Polyurethanes Expo 2001.

(5) Huang, Y. J.; Qi, G. R.; Chen, L. S. Effects of Morphology and Composition on Catalytic Performance of Double Metal Cyanide Complex Catalyst. 

 263–271. DOI: 10.1016/S0926-860X(02)00452-0

(6) Kim, I.; Ahn, J.-T.; Ha, C. S.; Yang, C. S.; Park, I. Polymerization of Propylene Oxide by Using Double Metal Cyanide Catalysts and the Application to Polyurethane Elastomer.

 3417–3428. DOI: 10.1016/S0032-3861(03)00226-X

(7) Lukey, G.; Van Deventer, J.; Huntington, S.; Chowdhury, R.; Shallcross, D. Raman Study on the Speciation of Copper Cyanide Complexes in Highly Saline Solutions. 

 233–244. DOI: 10.1016/S0304-386X(99)00047-X

(8) Cho, K.; Jang, Y. S.; Gong, M.-S.; Kim, K.; Joo, S.-W. Determination of Cyanide Species in Silver and Gold Plating Solutions by Raman Spectroscopy. 

 1147–1151. DOI: 10.1366/000370202760295377

Infrared and Raman Spectra of Inorganic and Coordination Compounds.

(10) McAllister, W. A. Raman Spectrum of Potassium Hexacyanocobaltate (III). 

(11) Swanson, B. I. Aspects of the Structure and Bonding in Prussian Blues. Single-Crystal Raman Study of Trimanganese Hexacyanocobaltate Hydrate (Mn

O) and Tricadmium Bis(hexacyanocobaltate) Hydrate (Cd

(12) Sebastian, J.; Darbha, S. Structure-Induced Catalytic Activity of Co–Zn Double-Metal Cyanide Complexes for Terpolymerization of Propylene Oxide, Cyclohexene Oxide and CO

(13) Haaland, D. M.; Easterling, R. G. Application of New Least-Squares Methods for the Quantitative Infrared Analysis of Multicomponent Samples. 

 665–673. DOI: 10.1366/0003702824639114

(14) Haaland, D. M.; Easterling, R. G.; Vopicka, D. A. Multivariate Least-Squares Methods Applied to the Quantitative Spectral Analysis of Multicomponent Samples. 

 is with Analytical Science at Dow Chemical, in Midland, Michigan. 

Mrunmayi Kumbhalkar, Jason Fisk, and Brian Murdoch

 are with Engineering & Process Sciences, at Dow Chemical, in Midland, Michigan. Direct correspondence to: 

Articles

Peer-Reviewed Article

Double metal cyanide (DMC) can be monitored in real time using 
an in situ Raman spectroscopy method. In this study, real-time reaction 
progress information from in situ Raman results enabled researchers 
to accurately determine the reaction end point of DMC.

In situ Monitoring of Double  Metal Cyanide (DMC) Catalyst  Synthesis by Raman Spectroscopy

In the field of document identification, it is always challenging to determine the printing sequences of crossed writings and seal stamps, which can serve as important evidence in litigation. Some common methods to determine the intersection sequence are expensive, destructive, cumbersome, and time-consuming. In this study, we provide several new potential ideas to help solve those problems by using the Raman spectral area scanning method to identify the intersection sequence (printing order) quickly and nondestructively. The results show that for red or blue seal inks, Raman spectroscopy can be used for good verification and as a supplementary method in determining the intersection sequence of writings and seal stamps. It is also effective for pigments analysis and can powerfully complement other analytical methods.

The sequence of crossed writings and seal stamps of documents must be determined in many economic and civil litigation cases (1–4). Some specific or important documents are formed in the order of writing or printing first, and then applying seal stamps after verification. Some criminal suspects often use the reverse order of the document formation process by applying seals first and then the writing or printing of the document to forge such documents. The intersection sequence (printing order) identification of writings and seal stamps confirms the authenticity of the document, and it provides a basis for the court to judge the facts stated in the documents for both economic and civil cases (5–8).

The main methods commonly used to identify the intersection sequence of writings and seal stamps include methods such as stripping, decolorization (9), sticking, fluorescence testing (10), and microscopy observation (11–13). Inspection methods such as microscopic inspection, fluorescence inspection, spectral inspection, and experimental inspection have been specified in the national standard “Specification for forensic examination of document formation time (GB/T 37233-2018).” The microscopic inspection method and experimental verification have also been mentioned in the ministerial standard “The intersection sequence identification of writings and seal stamps,” issued by the Ministry of Justice in 2010. The above commonly used methods are mostly expensive, destructive, cumbersome, and time-consuming (14). Testing and judging these sequences mainly rely on personal experience; therefore, strong subjectivity may lead to unexpected errors during examination. In 2010, the passing rate of proficiency testing organized by the Ministry of Justice in the intersection sequence identification of writings and seal stamps was only 71% in some provinces of China. Therefore, an objective and effective detection method to improve the reliability and accuracy of document identification is urgently needed. Some effective detection methods, including spectral inspection and mass spectrometry (MS) (15), are currently of high importance. Infrared (IR) spectroscopy is one of the most widely used spectral inspection methods in forensic sciences. It can effectively identify blue ballpoint pen inks, including dyes, solvents, film-forming agents, dispersants, and so forth. The literature on such studies has been extensive (16–19). However, the IR method does not adequately eliminate the paper signal interference. For example, good information can be obtained by detecting the ink alone, but if the ink is written on paper, only the signal of the paper can be obtained while the entire ink signal is obscured. The inspection materials and samples obtained from actual cases are usually with ink written on paper; therefore, using IR spectroscopy to identify writing ink is not applicable when handling actual forensic cases. The problem of a missing ink signal on paper can be solved by using Raman spectroscopy (20–21). The Raman spectrum of ink measured on glass slides is identical to those signals obtained on paper, indicating that the paper had no influence on the detection of inks using Raman spectroscopy. Therefore, Raman spectroscopy is different from IR spectroscopy, and it is more suitable for practical use cases (22–23). As a result, we have attempted to provide a nondestructive detection method using Raman spectral area scanning in this study.

Three groups of seal samples were randomly collected from different manufacturers and markets. A Renishaw Invia confocal Raman microscope system with three laser sources emitting at 532, 633, and 785 nm with a charge-coupled device (CCD) detector was employed for collecting the Raman spectra. Laser power within 10–100% was optimized to achieve strong signals and protect the samples from burning. Before scanning the sample, silicon with a fixed peak position at 520 cm

 was used for calibration. The spectra were collected in extensive mode over a Raman shift range of 200–2000 cm

. No characteristic peaks appeared below 200 cm

. Every spectrum was obtained in using three replicate measurements to enhance the signal-to-noise (S/N) ratio. The area scanning function was employed for the overall analysis and differentiation. In this study, the morphology of seal ink near the intersection of seal ink and the writing stroke can be observed using Raman spectrum surface scanning. The laser wavelength used was 785 nm, the laser intensity was 10%, the magnification was 5–50 times, and the scanning step length was 8 μm in both the horizontal and vertical directions using horizontal progressive scanning mode.

Figure 1a shows the Raman spectra of the carbon powder with strong peaks at 993 cm

. Figure 1b shows the Raman spectra of the intersection where the document was written first and then sealed. It can be seen that the Raman spectrum peaks of carbon powder approximately at 660 cm

 are covered by the seals. The Raman spectra of writings above the seals is shown in Figure 1c, where the strong absorption of carbon powder at 660 cm

 can be seen. This result indicated that the writings were not thick and broad enough to fully cover the seals. Ideal results can be achieved using a Raman spectrometer when analyzing a single point at the intersection of writings and seals. However, because of the unevenness of writings and seals, the writings may be on the upper layer at the edge portion. If only a single point is used for Raman analysis, a wrong conclusion is possible.

Figure 2 shows the experimental results of two test samples with red ink. Figure 2a and 2b display the photographs of the scanned intersection area. Figure 2c shows the Raman spectra of the sample sealed first and then written, and Figure 2d shows the Raman spectra of the sample written first and then sealed. Both were obtained by scanning from top to bottom and from left to right. Figure 3 further shows the Raman shifts in color depths. The main Raman shift peaks of the seal ink include 1597 cm

. In the intersection area of Figure 2a, especially in the red circled area (data sets 50–60), the Raman shift peak intensity is much lower than that of data sets 10–20 (seal only), suggesting that the seal ink is covered by the writing strokes. This is consistent with the conclusion that the samples were sealed first and then written after. In the intersection area of Figure 2b, especially in the red circled area (data sets 50–60), the Raman shift peak intensity is at similar level to that of data sets 10–20 (seal only). This indicates that the seal ink was not covered by the writing strokes, which is consistent with the conclusion that the sample was written first and then sealed.

Figure 4 shows the experimental results of other two test samples with different red inks. The Raman shift peaks of this type of red ink mainly appear at 1528 cm

. Figures 4a and 4b display the photographs of the scanned intersection area. Figure 4c shows the Raman spectra of the sample sealed first and then written, and Figure 4d shows the Raman spectra of the sample written first and then sealed, both obtained by scanning from top to bottom and from left to right. Figure 5 further shows the Raman shifts in color depths. The results of “seal first” series were consistent with those in Figure 2, where the seal ink was covered by the writing strokes, and coincided with the experiment results (seal first). In the red circled area (data sets 55–60) in Figure 4b and 4d, there were still several spectra in which the Raman shift peak intensities were at high levels (seal only). These spectra were obviously identified in the intersection area; therefore, the seal inks were not covered by the writing strokes and coincided with the experiment results (writing first).

Figure 6 showed the experimental results of two test samples with blue ink for better discrimination. The main Raman shift peaks of the blue ink include 1523 cm

. For the write-before-seal samples, the seal ink spectra were clearly identified in all data sets in Figure 4c. For the seal-then-write samples, no seal ink spectra was observed in the intersection area, especially in the red circled area, indicating that the seal ink were completely covered by the writing strokes. The Raman scanning spectrum results are also consistent with the sample design.

Compared to the method of single-point Raman spectroscopy, the method of Raman area scanning greatly improves document inspection and analysis, particularly when it comes to reliability, automation, and accuracy. It also reduces the possibility of false identification caused by the unevenness of margins at the surface top layer. Raman spectroscopy can be used as a good verification and supplementary method for the intersection printing sequence identification of writing and seal stamps with samples using red or blue seal inks. Reducing fluorescence interference, improving detection signal and stability, and avoiding sample burning still need to be explored in future studies.

(2) Chai, T.; Li, B. The Factors Influencing the Identification of the Chronological Intersecting Sequence Between Ink and Toner. 

 (6), 2469–2477. DOI: 10.1111/1556-4029.14830

(3) Luo, Z. G.; Wang, W. X.; Pang, X. C.; Zhang, J.; Sun, C. L.; Zhou, X.; He, J.; Zhang, R.; Li, X.; Lian, Z.; Abliz, Z. Writing Sequence Identification of Seals and Signatures in Documents Using Ambient Mass Spectrometry Imaging with Chemometric Methods. 

, 122804–122804. DOI: 10.1016/j.talanta.2021.122804

(4) Wang, J. Research on Several Inspection Methods to Determine the Sequence of Photocopying and Stamping. 

(2), 93–98. DOI: 10.4103/2349-5014.170601

(5) Farid, S.; Kasem, M. A.; Zedan, A. F.; Mohamed, G. G.; El-Hussein, A. Exploring ATR Fourier Transform IR Spectroscopy with Chemometric Analysis and Laser Scanning Microscopy in the Investigation of Forensic Documents Fraud. 

, 106704 (2021). DOI: 10.1016/j.optlastec.2020.106704

(6) Spagnolo, G. S. Potentiality of 3D Laser Profilometry to Determine the Sequence of Homogenous Crossing Lines on Questioned Documents. 

 (2-3), 102–109. DOI: 10.1016/j.forsciint.2005.12.004

(7) Aneja, P.; Mathur, S.; Dahiya, M. S. Determining Sequence of Strokes of Green Pen Ink and Blue Seal. 

 (1), 555628. DOI: 10.19080/JFSCI.2017.03.555628

(8) Lee, K. Y.; Lee, J.; Kong, S. G.; Kim, B. Sequence Discrimination of Heterogeneous Crossing of Seal Impression and Ink-Printed Text Using Adhesive Tapes. 

, 120–125. DOI: 10.1016/j.forsciint.2013.11.003

(9) Stewart, R.; Li, L. Selective Decolorization of Pigments Within Two and Three Pigmented Inks for Colored Marking by a Q-Switched Laser. 

(10) Voorde, L. V.; Vekemans, B.; Verhaeven, E.; Tack, P.; Wolf, R. D.; Garrevoet, J.; Vandenabeele, P.; Vincze, L. Analytical Characterization of a New Mobile X-ray Fluorescence and X-ray Diffraction Instrument Combined with a Pigment Identification Case Study. 

, 14–19. DOI: 10.1016/j.sab.2015.05.002

(11) Black, D.A. The Microscope in Document Examination. 

(12) Dreger, A. A. A Comparison Microscope for Document Examination. 

 (4), 175–179. DOI: 10.1080/00085030.1973.10757311

(13) Baier, P. E. Technical Improvements of Scanning Electron Microscope Methods in Document Examination. 

 (2-3), 265–278. DOI: 10.1016/0379-0738(83)90023-3

(14) Heudt, L.; Debois, D.; Zimmerman, T. A.; Köhler, L.; Bano, F.; Partouche, F.; Duwez, A. S.; Gilbert, B.; Pauw, E. D. Raman Spectroscopy and Laser Desorption Mass Spectrometry for Minimal Destructive Forensic Analysis of Black and Color Inkjet Printed Documents. 

 (1-3), 64–75. DOI: 10.1016/j.forsciint.2011.12.001

(15) Morais, D. R.; Campelo, J. M.; Razzo, D.; Eberlin, M. N.; Costa, J. L; Santos, J. M. Forensic Determination of Crossing Lines Involving Stamp and Pen Inks by Mass Spectrometry Imaging. 

(16) Zhao, Y. X.; Zeng, L. Y. Z.; Li, K. K. Identification of Different Brands Erasable Pens by Infrared Spectroscopy Combined with Chemometrics Methods. 

 (8), 2420–2426. DOI: 10.3964/j.issn.1000-0593(2021)08-2420-07

(17) Ameh, P.; Ozovehe, M.S. Forensic Examination of Inks Extracted from Printed Documents using Fourier Transform Infrared Spectroscopy. 

 (1), 10–17. DOI: 10.33805/2576.8484.107

(18) Kumar, R.; Kumar, V.; Sharma, V. Fourier Transform Infrared Spectroscopy and Chemometrics for the Characterization and Discrimination of Writing/Photocopier Paper Types: Application in Forensic Document Examinations. 

Spectrochim. Acta A Mol. Biomol. Spectrosc.

, 19–28. DOI: 10.1016/j.saa.2016.06.042.

(19) Udriştioiu, E. G.; Bunaciu, A. A.; Aboul-Enein, H. Y.; Tănase, I. Gh. Infrared Spectrometry in Discriminant Analysis of Laser Printer and Photocopy Toner on Questioned Documents. 

 (2), 230–240. DOI: 10.1080/10739140902735688

(20) Fikiet, M. A.; Khandasammy, S.R.; Mistek, E.; Ahmed, Y.; Halamkova, L.; Bueno, J.; and Lednev, I. K. Forensics: Evidence Examination via Raman Spectroscopy. 

 (2), 20170049. DOI: 10.1515/psr-2017-0049

(21) Johnson, C. E.; Martin, P.; Roberts, K. A.; Trejos, T.; Corzo, R.; Almirall, J. R.; Safer, A. M. J. The Capability of Raman Microspectroscopy to Differentiate Printing Inks. 

(1), 66–79. DOI: 10.1111/1556-4029.13508

(22) Liu, S.; Feng, J. M.; Lv, J. G. Discrimination of Blue Ballpoint Pen Inks in Chinese Market with Confocal Raman Microscope. 

 (1), 45–51. DOI: 10.1108/PRT-12-2012-0089

(23) Chen, R.; Lv, J. G.; Feng, J.; Liu, Y. M. Discrimination of Seal Inks Used for Seals by Confocal Raman Microscopy. 

 (6), 389–393. DOI: 10.1108/PRT-10-2013-0096

 are with the Procuratoral Technology and Information Research Center, at the Supreme People’s Procuratorate, in Beijing, China. Direct correspondence to: 

Determining the printing sequences of crossed writings and seal stamps 
is often difficult because the most common methods used are expensive, 
time-consuming, and cumbersome. A new method using Raman spectral 
area scanning offers a better alternative while conducting pigment analysis and determining intersection sequences of writings and seal stamps. We explain why.

A Raman Spectral Area Scanning Method to Identify the Sequences  of Crossed Writings and Seal Stamps

This study aims to discriminate different types of illicit drugs (MDMA and THC) in blood samples using surface-enhanced Raman spectroscopy (SERS) combined with chemometric techniques including principal components analysis (PCA) and partial least squares discriminant analysis (PLS-DA). A PLS-DA classification model was built using a training data set containing Raman spectra from control and experimental groups (drug-detected blood). PLS-DA was performed for discrimination and classification among blood samples. The scores obtained in the PLS-DA model were used to evaluate the performance of the created model. The leave one out cross-validation (LOOCV) method was used for calibration and validation of the PLS-DA model. In the study, it was observed that the SERS method and chemometric techniques together could be used in drug analysis, even at low concentrations in complex body fluids such as blood. As a result, Raman spectroscopy with PCA and PLS-DA methods of data analysis could be used extensively to build similar or different classification models.

Illicit drug use or addiction has become an important problem all over the world, as it causes many serious problems (1). The World Drug Report in 2020 stated that the estimated number of past-year users of any drug utilization globally increased from 210 million to 269 million from 2009–2018. Around 20.5 million people globally were estimated to have used “ecstasy” in the past year, corresponding to 0.4% of the global population aged 15 and older in 2018 (2).

Since 1971, amphetamine, meth- amphetamine, methcathinone and “ecstasy”-group substances (3,4-methylenedioxymethamphetamine [MDMA] and its analogues) have been called amphetamine-type substances (3). Ecstasy and cannabis (tetrahydrocannabinol [THC]) are two of the most frequently encountered drugs in blood analyses for the detection of drug use. THC is the most important psychoactive ingredient in cannabis (4).

Biological samples such as blood, urine, hair, nails, and saliva are used to determine whether a person uses drugs or not (5). Today, different analysis methods are used for body fluid identification. These analysis methods can be classified as chemical tests, immunological tests, protein catalytic activity tests, or spectroscopic and microscopic methods. The methods are applied for the harmless examination of obtained evidences, such as blood, urine, and saliva, as they continue to develop. Although the results of both screening and validation tests are reliable, the methods take time and may delay treatment (6,7).

Raman spectroscopy is one of the most recommended methods, since it is a method that does not damage the sample and provides rapid results, with even a small amount of sample being sufficient for examination (7–11). Raman spectroscopy has been increasingly used in all scientific areas in recent years in different analyses, and has enabled the determination of illicit drugs without using additional chemicals (8). However, although Raman spectroscopy has many advantages, it is sometimes not sufficient by itself to analyze a small amount of a component in blood. This is because the components in biological samples may generate very weak Raman peaks. For this reason, the use of the surface-enhanced Raman spectroscopy (SERS) technique is more advantageous in Raman examinations. The SERS method significantly increases the applicability of Raman spectroscopy (12). The SERS method is a spectroscopic technique used to induce vibrational transmission of adsorbed molecules on a rough metallic surface. Thanks to their electromagnetic and charge transfer mechanisms, metallic nanoparticles (Ag, Au, and others) are widely used in SERS applications. In addition, being able to control properties such as the size and the shape of nanoparticles facilitated the widespread application of the SERS technique in biomedical analytics and clinical diagnostics (13–16).

It is difficult to obtain an exact result, due to many different reasons (such as the classification of the spectra obtained by Raman spectroscopy examination of biological samples, as well as the limited number of samples). For this reason, chemometric methods are needed to classify many variables (17) and used for both qualitative and quantitative analysis of the data while being categorized as supervised or unsupervised methods. Principal component analysis (PCA) is an unsupervised method, and is convenient for the analysis of unknown spectra. A small amount of unknown data can be obtained, such as principal components (PC) and loadings (eigenvectors). Another preferred reason for the application of PCA to the data obtained from the spectra in the studies is that only some components in a mixture are known (18). Partial least square (PLS) regression is a supervised method, and generally applied in cases where there is a limited number of samples. Calibration and validation are a two-stage process for the supervised methods (19–21). Conditions such as the fact that the PCA method is an unsupervised method do not necessarily reveal the relevant patterns, and the performance of PLS largely depends on the number of components to be used in discriminant analysis (DA) (17). PLS-DA is a linear classification method based on PLS regression, and it is the most popular chemometric tool used for qualitative and quantitative modeling of multidimensional data. The large data in human blood samples can give rise to problems in the analysis. Thus, the study aims to discriminate different type of illicit drugs (MDMA and THC) in blood samples using the SERS method combined with multivariate analyses. For this purpose, SERS spectra of the blood samples were analyzed. PLS-DA and PCA were performed for discrimination and classification among blood samples.

This study was approved by the Health Sciences Central Ethical Committee (14.10.2021/01). Consent forms were signed and a total of 22 blood samples were taken from two groups: an experimental group (drug users) who were brought to the health service for different reasons, and the control group (non-drug users) in whose blood drugs were not contained. Blood samples were taken into ethylenediaminetetraacetic acid (EDTA) tubes and stored at −20 

C in a freezer until the examination was made. In all examinations, slides that were compatible with Raman spectroscopy and did not change the peak of the sample were used. Silver nanoparticles (AgNPs) were synthesized by reduction of AgNO

 using trisodium citrate according to the Lee and Meisel method (22). Chemicals were supplied to synthesized AgNPs from Sigma Aldrich. An ultraviolet-visible (UV-Vis) spectrophotometer and Scanning Electron Microscopy (SEM-EDS) analysis unit (JEOL 5500-OXFORD Inca-X) were used to characterize AgNPs. Before SERS measurements, 15 μl of AgNPs and 15 μl of blood samples were separately dropped on the slides and allowed to dry at room temperature and in the dark. After drying, Raman spectra of blood samples were obtained by the SERS method.

All samples were measured with Renishaw’s Via Raman spectroscopy instrument equipped with a charge-coupled device (CCD) and the same parameters. The parameters were taken as 5 mW laser power, 50x objective, 785 nm diode laser for excitation, and 40 s exposure time. Raman bandwidth was chosen as 100 to 3200 cm

Spectral collection and data pre-processing (such as smoothing, fourth polynomial baseline correction, and vector normalization) from Raman spectroscopy were acquired using the WiRE 3.2 software (Renishaw plc). In the study, each sample was analyzed five times, and then averaged to form a spectrum representing one sample. A total of 889 spectral data were created in the region between 500–1600 cm

, which forms the fingerprint region and has different peaks, from the Raman spectra of 22 (11 positive, 11 negative) blood samples. The data were transferred to the MATLAB (Matlab 7.13, The Mathworks) program and analyzed using PLS Toolbox (Eigenvector Research) for Matlab users, accounting for both the PCA and PLS-DA methods.

The PCA method was used for the analysis of 11 blood samples (2 MDMA, 1 THC + MDMA, and 8 THC), which were determined to be different drugs. Second, the PLS-DA method was used to optimize the separation between drug-containing and drug-free blood.

In the study, a SERS spectrum of the blood sample was obtained, and Raman bands were analyzed (Figure 1). The Raman band at 495 cm

 involved vibrational modes of amino acids (23). The peaks at 1003, 1341, 1434, 1598 and 1663 cm

 were assigned to hemoglobin and its derivatives (10). Raman bands at 1341 and 1434 cm

 corresponded to deformations of the carbon rings and the characteristic bending motions of the CH, CH

 groups amino acid deformation modes. The bands at 1003 cm

 involved stretching motions of the phenylalanine, C-C skeletal and amide, C=C groups in unsaturated fatty acids correspondingly (24,25). The bands at the 1176 cm

 correspond to the vibrations of C-C and C=C stretching.

The SERS spectra obtained in blood samples containing THC and MDMA are shown in Figure 2 and Figure 3, respectively. It was observed that the strong SERS bands obtained in blood samples containing THC were the same in the fingerprint range of 500–1600 cm

 (Figure 2). However, it was observed that there were changes at the 615, 696, 1341, and 1434 cm

 bands of blood samples containing different amounts of THC. The weak Raman bands at 830–860 cm

 were assigned to stretching of the middle tetrahydropyran ring, the weak Raman bands at 1003 and 1088 cm

 corresponded to phenyl ring peak intensities, and the weak Raman bands at 1341 cm

 were assigned to CH deformation modes (26). The characteristic Raman bands of blood samples containing different amounts of THC were detected at 712, 1006, 1310, 1390, 1450, and 1590 cm

. These bands corresponded to vibrations of, respectively, (C-H) deformation, (C-C) stretching, (C-H) deformation, (=C-H) deformation, (C-H) deformation, and C=C stretching (27).

In a study by Barnett and Rathmell, Raman peaks of THC were examined by changing the THC concentrations and it was observed that these peaks were observed at 710, 999, 1130, 1232, 1352 and 1390 cm

. It was also shown that stronger peaks could be obtained at increasing concentrations (28).

The MDMA peaks differed in a blood sample taken from a person using both MDMA and THC (Figure 4). The most noticeable Raman peaks that distinguished MDMA from THC were found at 750, 1231, 1392, 1570, and 1624 cm

. These bands can be used to confirm the presence of MDMA. The Raman band at 1624 cm

 corresponded to vibrations of NH deformation.

Sagmuller and associates stated that there are peaks of C-H and C-C vibrations belonging to methyl and ethyl groups in the region between 900 and 1650 cm

 for MDMA analysis (1). Westa and coauthors (9) stated that the most prominent Raman peaks for MDMA were at 716 cm

. Ryder and associates applied different chemometric analyses by dividing into regions between 450 and 1100 cm

 by micro-Raman spectroscopy. Four different Raman peaks were observed between 700 cm

 as MDMA peaks in solid mixtures (29). It was observed that the results obtained from the studies were compatible with our study.

Inscore and coauthors performed a drug substance analysis in saliva with the SERS method in which it was shown that the obtained Raman peaks according to the type of drug used and the varying amount of the same substance were also different in their study (30). In the study, SERS of MDMA was found to be at 530–535, 715–720, 1250, 1365–1370, 1430–1435, 1470–1480, and 1620 cm

, largely due to the dioxole ring alone or coupled with the phenyl ring (26).

PCA, which is used as the most preferred method for qualitative classification, has proven very useful in different studies (21,31,32), as well as in our study, to visualize spectral data and examine possible sample groupings according to spectral characteristics.

When the SERS spectra of the blood samples containing THC and MDMA were examined, distinctive peaks were obtained at 500–1600 cm

. The PCA method was performed for pattern identification of different types of objects in cluster form among illicit drugs of blood samples. In the PCA model, three different principal components (PCs) with different percentages were obtained. It was observed that the percentages of these variables were 55.77% and 19.68% respectively, and the model was classified depending on two PCs, which would make up 75.45% of the total variance. When PCA was applied to the data sets obtained for the analysis of blood samples containing different drugs, it was observed that the data were clustered differently. The PC1 and PC2 scores interpreted the highest variance, and the plot and loadings of two factors are shown in Figure 5a and Figure 5b. The figure of PC2 versus PC1 was illustrated in Figure 5b. PC1 was the most discriminatory, and illustrated in Figure 5a. The PCs indicated the importance of drugs in a given blood sample. In Figure 5a, PC1, which explains most of the variance, was located on the top as positively and strongly correlated with MDMA.

Using a small number of PCs is sufficient, as it uniquely captures the effect of a certain combination of relevant identifiers. For this reason, two PCs were sufficient for the visualization of the investigated blood samples (33). Another preferred reason for the application of PCA to the obtained data from the spectra in the study was that only some components in a mixture were known (18).

In this study, since the variability of the data obtained by Raman spectroscopy was high, PLS-DA analysis was applied to determine the variable to be examined. Blood taken from the experimental group (drug users) and the control group (non-drug users) were divided into two groups for PLS-DA analysis. SERS spectra obtained from these two separated groups were transferred into the PLS toolbox as a training and test set. A PLS-DA classification model was built using a training data set containing SERS spectra from the control and experimental groups. The scores obtained in the PLS-DA model were used to evaluate the performance of the model. The leave one out cross-validation (LOOCV) method was used for calibration and validation of the PLS-DA model. Sensitivity, specificity, class error, the root mean square error of calibration (RMSEC), the root mean square error of cross-validation (RM-SECV), and 

 values were calculated to measure the difference. These values were used to describe the performance of the classification model.

RMSEC values were used to differentiate between the reference and the test set. CV values were used for the low number of samples for validation of a classification model, and RMSEC values were also used to evaluate the compatibility of the calibration model with the calibration set (31,34).

The PLS-DA models gave sensitivity and specificity values on the test set in the range of 88%–100%. The sensitivity values were found to be on the test set in 100%, and specificity values were found to be on the test set in the range of 81.8%-90.9% for the PLS-DA models. RMSEC values were found to be 2% and 3%. RMSECV values were found to be 0.37% and 0.58%. Class error values were found to be between 0 and 9%. The obtained values were found separately as calculated and cross-validated. Obtained scores using the PLS-DA model with the control and experimental groups were loaded as predictions (Figure 5c, and the model was built with three latent variables (LVs). When the loading plots were examined, it was observed that there were significant differences in blood samples, blood samples with THC, and blood samples with MDMA in the region between 500 and 1600 cm

 (Figure 6a). When the wavelengths of LVs were examined, it was observed that they resembled the spectra obtained from blood samples, blood samples with THC, and blood samples with MDMA (LV1, LV2, and LV3, respectively) and were compatible with the stated respective Raman bands. Scores on LV2 were used to create three-dimensional plots of blood using latent variables (Figure 6b). The LV values obtained in this method allowed for the visualization of different variables. Over-fitting in PLS models reduced the probability of estimation of the results. Therefore, the selection of latent variables was important. Too many latent variables can cause high bias and low variance (35).

The combination of Raman spectroscopy and chemometric techniques allows for the analysis of even the smallest analyte that can be found in biological samples. We have demonstrated an analysis of Raman spectroscopy that is a viable tool for the rapid, non-invasive detection of different illicit drugs in blood samples. It has been observed that the SERS method and chemometric techniques together can be used in drug analysis, even at low concentrations in complex body fluids. RS with PCA and PLS-DA methods of data analysis can be used extensively to build similar or different classification models.

(1) Sagmüller, B.; Schwarze, B.; Brehm, G.; Schneider, S. Application of SERS Spectroscopy to the Identification of (3,4-methylenedioxy)amphetamine in Forensic Samples Utilizing Matrix Stabilized Silver Halides. 

(2) World Drug Report 2020: Drug Use and Health Consequences

. (United Nations Publication, Sales No. E.17.XI.6) 

https://wdr.unodc.org/wdr2020/field/WDR20_Booklet_2.pdf

 (United Nations publication, Sales No. E.14.XI.7) 

https://www.unodc.org/documents/wdr2014/World_Drug_Report_2014_web.pdf

(4) Khiabani, H. Z.; Bramness, J. G.; Bjorneboe, A.; Morland, J. Relationship Between THC Concentration in Blood and Impairment in Apprehended Drivers. 

https://doi.org/10.1080/15389580600550172

(5) Yayatürk, A. E.; Ertas, H.; Akgür, S. A.; Ertas, F. N. 

, Akgür, S. A.; Coşkunol, H., Eds. Ege University Press, 2014, pp. 211–236.

(6) Farquharson, S.; Shende, C.; Sengupta, A.; Huang, H.; Inscore, F. Rapid Detection and Identification of Overdose Drugs in Saliva by Surface-Enhanced Raman Scattering Using Fused Gold Colloids. 

https://doi.org/10.3390/pharmaceutics3030425

(7) An, J. H.; Shin, K.-J.; Yang, W. I.; Lee, H. Y. Body Fluid Identification in Forensics. 

https://doi.org/10.5483/bmbrep.2012.45.10.206

(8) de Oliveira Penido, C. A. F.; Pacheco, M. T. T.; Novotny, E. H.; Lednev, I. K.; Silveira, L. Jr. Quantification of Cocaine in Ternary Mixtures Using Partial Least Squares Regression Applied to Raman and Fourier Transform Infrared Spectroscopy. 

(9) West, M.J.; Went, M.J. Detection of Drugs of Abuse by Raman Spectroscopy. 

(10) Sikirzhytski, V.; Virkler, K.; Lednev, I. K. Discriminant Analysis of Raman Spectra for Body Fluid Identification for Forensic Purposes. 

(11) Zapata, F.; Gregório, I.; García-Ruiz, C. Body Fluids and Spectroscopic Techniques in Forensics: A Perfect Match? 

https://doi.org/10.4172/2472-1026.1000101

(12) Premasiri, W. R.; Lee, J. C.; Ziegler, L. D. Surface-Enhanced Raman Scattering of Whole Human Blood, Blood Plasma, and Red Blood Cells: Cellular Processes and Bioanalytical Sensing. 

(13) Laserna, J. J. Combining Fingerprinting Capability with Trace Analytical Detection: Surface-Enhanced Raman Spectrometry. 

https://doi.org/10.1016/0003-2670(93)85274-N

Detecting illicit drugs in blood samples requires a rapid, non-invasive 
technique. The combination of surface-enhanced Raman spectroscopy 
(SERS) and chemometric techniques, such as principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), can aid in this endeavor. 

Raman Technology for Today's Spectroscopists

In situ Monitoring of Double Metal Cyanide (DMC) Catalyst Synthesis by Raman Spectroscopy

A Raman Spectral Area Scanning Method to Identify the Sequences of Crossed Writings and Seal Stamps

Illicit Drug Analysis in Blood Samples with Multivariate Analysis Using Surface-Enhanced Raman Spectroscopy

Portable Raman Spectrometers: How Small Can They Get?