The origin and specifications of all the analyzed samples are given in Table 1. The real samples, which were collected with standard ethical procedure, were provided as a part of the seizures performed by the Police of the Czech Republic, while the standards (purity of all the used standards were higher than 98%) were acquired from different sources (Table 1). Samples 5F-ADB I. and II. were provided as a part of seizures performed by the Police of the Czech Republic, while the samples 5F-ADB IV., V., and VI. were obtained from a commercial vendor in the framework of a darknet study. Sample 5F-ADB III. was prepared in house.

The sample crystals were crushed with a microscope slide on a silicon pod (see Supplementary Figure 1) and, thus, a narrow surface was created. For the remeasurement, JWH-073 samples were ground extensively in the agate mortar to show the differences in relative intensities.

The XRPD data were collected with a Bruker 2nd generation D2 Phaser powder diffractometer (Bruker AXS, Germany) at room temperature with parafocusing Bragg-Brentano geometry using CuKα radiation (λ = 1.5418 Å, U = 30 kV, I = 10 mA). The data were scanned with an ultrafast LYNXEYE XE detector over the angular range of 5 to 60 °2θ with a step size of 0.019 °2θ and a counting time of 1 s per step. The software HIGHSCORE PLUS 3.0e (PANalytical, Almelo, Netherlands) was employed to fit the background using a polynomial method, to smooth the data and to eliminate the Kα2 component. The top of the smoothed peaks was used to determine the peak positions and intensities. Determination of peak positions was made by an in-build algorithm in the HIGHSCORE PLUS 3.0e. Thus, processed diffraction patterns were subjected to the pattern searching procedure within PDF4+ database via the HighScore software.

To compare similarity of XRPD diffractograms quantitatively, a cross-correlation score was used. To accentuate peaks and attenuate background noise and minor pollutant effects, we pre-prepare the normalized patterns by squaring them (Equations 1, 2) and only then calculating their cross-correlation (Equations 3, 4). With f'(θ) as the normalized XRPD diffractogram of a known standard and g'(θ) as the normalized XRPD diffractogram of a measured sample, we calculate a cross-correlation score CCS_fg as defined below(Equation 5):

A discrete calculation method for CCS_fg, directly using ASC files, was implemented as a Python script (available at https://github.com/dehaenw/cross-correlation).

Crystal shapes were visualized by confocal microscope Olympus Lext OLS 3100 without any additional image processing.

The IR spectra of the standards and real samples were measured on a FT-IR Nicolet iS50 spectrometer (Thermo Scientific, USA) with a Tungsten-halogen MIR radiation source, KBr beam splitter and DLaTGS detector. All the samples were in the form of a powder and they were analyzed by the ATR technique with a diamond crystal. The spectra were recorded in a spectral region of 4,000–400 cm⁻¹ with a resolution of 4 cm⁻¹ and they are presented as an average of 256 scans. The spectral background was collected before every sample measurement.

The Raman spectra were acquired on a DXR SmartRaman spectrometer (Thermo Scientific, USA) equipped with two lasers (excitation wavelengths 532 and 780 nm). In the case of the 532 nm laser, a diffraction grid comprised of 900 lines per mm, a laser power of 5 mW and 10 accumulations each of 10 s exposure time were used. A diffraction grid with 400 lines per mm, a laser power of 65 mW and 10 accumulations each of 10 s exposure time were used for the measurements with the 780 nm laser. The spectra were recorded in a spectral region of 400–3,000 cm⁻¹ with a resolution of 2.4–4.4 cm⁻¹ and all the samples were analyzed in glass vials. The spectra were processed with the correction of fluorescence (6th order polynomial).

We have demonstrated in our previous work (Jurásek et al., 2019) that XRPD can effectively distinguish between particular chemical entities even with similar chemical structures, which is essential in the forensic analysis of NPSs. However, most of the tested standards were prepared in our laboratory and the results were not extensively compared with authentic samples that might occur on the drug market (Jurásek et al., 2019). Therefore, in the current study, we analyzed 7 samples of psychoactive substances (cocaine, heroin, and 5 NPSs) that were seized on the black market and the acquired results were compared with the diffraction patterns of the respective standards (Figure 1).

PDF4+ is a commercially available database that contains more than 410 000 diffraction patterns. Although this database does not contain most NPSs, this database contains diffraction patterns of heroin and cocaine. Such a big amount of patterns in this database made us wonder if it would be possible to use it for identification of street samples of heroin and cocaine. The results were quite impressive, as we were able to identify both cocaine and heroin in street mixtures (see non-modified search data in the Supplementary Data) using this commercial database. However, we were unable to assign the cutting agents, as this database mostly does not contain their respective patterns. Because there is no suitable database of illicit substances yet, street samples of NPSs were just compared with the diffraction patterns of their respective standards.

JWH-073 was identified by XRPD (Figure 1) in one of the seized materials despite the observation that relative intensities in the diffraction pattern differed considerably. The most intensive peak of the standard sample was 15.1 °2θ whereas in the seized sample it was 21.5 °2θ. However, these differences in the relative intensities might be caused by different crystal shapes. To confirm that the standard and seized sample had different crystal shapes, they were subjected to a visual analysis by the optical confocal microscope (Figure 2). The crystal proportions of the seized sample were approximately the same in all three dimensions, but the standard formed needle-like shapes and so one dimension was significantly larger than the other two. This was presumably caused either by the type of crystallization or the synthetic process of the respective samples (Morris et al., 2000). Therefore, to reduce the differences in relative intensities both JWH-073 sample and the respective standard were extensively ground in an agate mortar and remeasured with the same setting of the goniometer (see Figure 3). The differences in the signal intensities did not have any effect regarding the identification of the compound in the seized sample as the peak positions did not change. The sample was successfully identified according to the peak positions and no other peaks were observed suggesting a high purity of the JWH-073 in the seized material.

Mephedrone, ephylone, naphyrone, and butylone were successfully identified in the seized material by comparison of the seized samples and respective standard diffraction patterns. Although the relative intensities of some peaks differed slightly in both samples, which was presumably caused again by different crystal shapes, there were no other peaks at different positions suggesting that the seized materials were of high purity.

Although cocaine and heroin do not belong to the NPSs, the prevalence of these compounds on the drug market is high and therefore we have included these “classic” street drugs in the examination test of XRPD in a similar manner as in the case of the NPSs. Since cocaine and heroin have been already measured and their diffractograms were included in the database of PDFs (powder diffraction file), database cards were used for the identification instead of using the respective standards.

This approach was chosen mainly to prove that the samples could be identified without the need of a standard only by using a suitable database. The seized sample of heroin was successfully identified as diacetylmorphine with card PDF 00-033-1635 when most of the peaks belonged to the drug. The relatively intensive peaks 11.5 and 11.8 °2θ might be attributed to possible diluting agents (e.g., caffeine, PDF 00-049-2058). However, the aim of this study was to prove that XRPD can be used for drug identification and therefore, these impurities were not further investigated. Cocaine was identified by XRPD in the last seized sample. All of the major peaks were attributed to cocaine hydrochloride (PDF 00-030-1629) with the exception of the less intensive peak 10.7 °2θ and a few minor peaks, which might be attributed to specific adulterants in the future.

To compare the efficiency of XRPD with other non-destructive methods that are often used in forensic practice, all of the samples and standards were measured by the IR and Raman spectroscopies. The differences of the results provided by these methods have been highlighted.

The Raman spectroscopy suffered from the high fluorescence level with the use of the 532 nm excitation wavelength, where only measurements of the real sample of JWH-073 provided an interpretable spectrum. After the application of the 780 nm excitation wavelength, the fluorescence level decreased in most cases and the active substances were identified by a simple comparison with the spectra of the corresponding standards (Figure 4 and Supplementary Data). However, the high level of fluorescence made the analysis of the naphyrone street sample impossible (Supplementary Data) and the high amount of the adulterants in the heroin sample did not allow a reliable identification of the active substance (Figure 4).

The IR spectroscopy performed slightly better, as it allowed the reliable identification of 6 of the 7 analyzed samples (Figure 5 and Supplementary Data), but the identification of the active substance in the heroin street sample was not possible due to the presence of many interfering bands. Both IR and Raman spectroscopies can offer spectra within several minutes, whereas XRPD instrumentation is usually more time demanding (about 15–20 min). All the measured data can be found in the Supplementary Data.

Vibrational spectroscopy proved to be a powerful tool in the analysis of illicit drug samples as was expected. However, in the case of the heroin sample, both methods struggled in the identification of the active substance. On the other hand, heroin was easily identified by XRPD, we thus believe that its potential in the forensic practice is promising.

Although differences in relative intensities in the XRPD patterns may seem to complicate the identification of unknown substances, on the contrary, in some cases, it might further provide valuable data about the analyte (e.g., differences between the production procedures). To demonstrate this ability, six different samples of the 5F-ADB, which were obtained during our NPS survey in the Czech Republic, were analyzed and compared with the standard. The same peak positions in the diffraction patterns enabled the identification of the 5F-ADB in the samples. Two of the samples offered diffraction patterns with not only the same peak position but the relative intensities corresponded as well. The other samples exhibited differences in relative intensities (see Figure 6).

This difference in relative intensities makes qualitative estimation of similarity by visual comparing and simple pattern subtracting suggestive but tricky and non-trivial. As a quantitative similarity approach, we proposed a cross-correlation score. This showed that the XRPD patterns of all seven different samples of 5F-ADB have a much higher score than the diffractograms of other, unrelated compounds (see Figure 7).

This suggests that peak positions are essential for the substance identification, whereas an exact match of the relative intensities is not needed, as is a well-known property of XRPD patterns (Pecharsky and Zavalij, 2009). A match of the relative intensities occur when the shape of the crystals and partly their size correspond. Interestingly, we observed a similar relative intensity pattern for sample V. and VI., which could be possibly attributed to their shared origin. If so, such information may be useful for investigators as samples with the same relative intensities could either be from the same source or possibly even from the same batch. Yet, drawing such conclusions based on the agreement of relative intensities may be unreliable, therefore the use of LC-MS remains the only reliable and generally powerful method for this purpose. In case a reliable correlation could be found generally, XRPD may be useful as a pre-screening method for this purpose. Notably, significant grinding of the samples undoubtedly affects the relative intensity pattern of the measured samples (see Figure 3), which unfortunately further complicates the option to compare “similarity of the samples” in a straightforward manner.

Nevertheless, it is essential to note that if it is not possible to assign all the signals in the diffraction pattern, then further analyses may be required. However, after creating a robust database of diffraction patterns of NPSs and cutting agents, such database would enable the identification of not only the main compound but also help with assigning all of the other signals in the pattern to other compounds.