Introduction

A key tenet of malaria control lies in antimalarial treatments being accessible, affordable, and effective1. Poor quality antimalarials are thus a cause for serious concern. Falsified medical products are defined by the World Health Organization (WHO) as those “that deliberately/fraudulently misrepresent their identity, composition or source,” in contrast to substandard or “out of specification” medicines that “are authorized medical products that fail to meet either their quality standards or their specifications, or both”2.

Falsified medicines may contain no or incorrect active pharmaceutical ingredients (APIs), or incorrect API amounts, and impurities or poor bioavailability. Incorrect unstated excipients may also be damaging. Falsified medicines harm patients, weaken health systems, damage economies and, for antimicrobials, endanger antimicrobial resistance. WHO estimated that globally ~ 10% of medicines in low- and middle-income countries are substandard or falsified2. There have been numerous examples of falsified antimalarials, particularly in Southeast Asia and Sub-Saharan Africa, which must have had a major negative impact on malaria morbidity and mortality2,3,4,5. Their sources remain poorly understood and to provide actionable evidence innovative methods are needed to identify their origins and trade routes6.

There has been minimal research published on innovation of forensic techniques to trace where falsified medicines were manufactured, or their ingredients sourced. Pollen and calcium carbonate analyses were important in providing evidence that falsified oral antimalarial artesunate—in a large epidemic that afflicted mainland Southeast Asia in the late 1990s and early 2000s—was from southern China 6. More recently, environmental DNA has been shown to offer promise for providing signatures, specific for time and place, for the ingredients and manufacturing sites of falsified medicines7. There has been much more research and significant recent innovations in the forensic investigation of the illegal wildlife trade to provide evidence, for instance, on the habitat of elephant victims of the ivory trade. This uses stable isotope ratio mass spectrometry (IRMS) to measure small differences in amounts of isotopes in materials that are characteristic of geographic origin8,9,10,11.

Most elements in the periodic table have multiple isotopic forms, distinguished by different numbers of neutrons. Isotopes can be considered Nature’s recorders, useful for reconstructing biological, chemical, and ecological processes8. For example, carbon and nitrogen isotope delta (δ) values can be used to trace the fixation and movement of carbohydrates and protein, while oxygen isotope δ values can provide geolocation information as they are related to the systematic global variations in environmental water. Natural and artificial transfers lead to changes in relative amounts of stable isotopes within materials, in a phenomenon called isotopic fractionation12,13,14,15,16,17,18,19,20,21,22,23.

Differences in the carbon isotopic composition of biological material can be related to differences in the photosynthetic pathways of plants. The C3 pathway, used by temperate grasses, trees, and crops such as rice and wheat, typically results in relatively low δ13C values for plant tissues (e.g., − 35 to − 20‰) while the C4 pathway, used by tropical grasses and crops such as maize and sugar cane, typically results in higher values (e.g., − 14 to − 10‰)14. Due to geochemical processes, fossil fuels generally have δ13C values lower than modern C3 plants, ranging between − 60 to − 20‰, while most marine carbonates (e.g., limestone and dolomite) have δ13C values higher than C4 plants, ranging between approximately − 5 and + 2‰15.

Differences in δ15N values within the environment are influenced by chemical changes within the nitrogen cycle. Common inputs to the terrestrial part of the cycle include atmospheric nitrogen deposition (soils) and nitrogen fixation (plants) while outputs include gaseous losses and hydrologic leaching. The isotopic fractionation factors associated with these processes are often dependent upon nitrogen quantity and the conditions (e.g., temperature, aridity, enzyme properties, etc.) under which nitrogen is cycled16,17,18.

The oxygen isotopic composition of terrestrial surface water varies with geographic location in a predictable manner19,20,21. Precipitation causes isotopic fractionation as water molecules move from oceans onto land surfaces22. The amount of fractionation varies with distance from the ocean, elevation, and temperature. Surface water in warmer climates typically has higher δ18O values than water found in colder, higher latitude locations. The local water signals resulting from this predictable isotopic fractionation are transferred to plants and animals and recorded in their tissues e.g., plant carbohydrate through photosynthesis. Once recovered, this signal can provide geolocation information23. Plants and animals with lower δ18O values more likely originated from colder climates, higher elevations, and/or regions more inland than plants or animals with higher δ18O values.

IRMS techniques have been used for the characterization of both illicit drugs and medications. For illicit drugs, isotopic profiling has been useful for collecting source intelligence, elucidating production processes, and making sample-to-sample comparisons24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39. For medications, isotopic profiling can aid in the authentication of genuine pharmaceuticals and, conversely, the detection of falsified medicines40,41,42,43,44,45,46,47,48,49. We investigated the δ13C and δ18O values of calcite in genuine and falsified artesunate antimalarials, suggesting that the calcite was of high temperature intrusive origin, probably in southern China6.

We therefore conducted a further pilot study to assess whether stable isotope analysis of falsified antimalarial medicines could provide evidence of their source. We used both bulk and component-specific approaches for isotopic profiling. In an initial sample set, 18 antimalarial samples, stated to contain artesunate, artemether-lumefantrine, dihydroartemisinin-piperaquine and sulphamethopyrazine-pyrimethamine, were analyzed without any prior purification (i.e., bulk) for carbon (C), nitrogen (N), and oxygen (O) element concentrations and stable isotope ratios. In a follow-on set, 26 antimalarial samples, stated as including artesunate or artemether-lumefantrine, were first analyzed as bulk material before an aliquot of each was extracted using a series of different solvents to collect the insoluble fraction (“starch”); that fraction was also analyzed for element concentrations and stable isotope ratios. Starches extracted from samples of known origin maize (Zea mays), a common excipient, were used to produce a comparison dataset to determine whether predictions could be made on the potential source of starch found in falsified antimalarials.

Methods

Samples

The medicine samples were collected as part of studies in the Lao PDR, Cambodia, Angola, Cameroon, China, and Myanmar5,6,50,51,52. All genuine samples were stated as containing starch and all but two falsified samples (1/29 and 2/12056) were recorded as containing starch (Supplementary Material-Tables 1 and 2). Information provided with samples included unique identification code, brand/stated manufacturer, and quality classification (genuine vs. falsified, based on previous HPLC, LC–MS, and packaging analyses6).

Table 1 Samples (n = 18) included in the initial study of antimalarials, with measured element concentrations and stable isotope ratios analyzed in bulk.
Full size table
Table 2 Samples (n = 26) included in the follow-on study of artesunate/artemether antimalarials with bulk C, N, and O element concentrations and stable isotope ratios.
Full size table

Initial set of antimalarials (n = 18)

At least one tablet of each sample was available for analysis; in some cases, two tablets were available. The active pharmaceutical ingredients (APIs) included were artesunate (4 genuine and 5 falsified), artemether-lumefantrine (2 genuine and 1 falsified), dihydroartemisinin-piperaquine (1 genuine and 2 falsified), and sulphamethopyrazine-pyrimethamine (2 genuine and 1 falsified) (Table 1). The Supplementary Material includes further details on excipient content as determined by ATR-FTIR.

Tablets were ground to a fine powder using a ceramic mortar and pestle. In cases where two tablets were provided, only one tablet was ground. In cases where one tablet was provided, the single tablet was cut in half using a clean razor blade and only half the tablet was ground. One sample was largely powdered upon receipt and all material was subsequently ground. Ground material was stored in capped 1-dram glass vials at room temperature. Samples were analyzed in duplicate and means presented.

Follow-on set of artesunate/artemether antimalarials (n = 26)

Between one and five tablets of each sample were available for analysis (Table 2). All tablets available per sample were ground to a fine powder using a ceramic mortar and pestle. Ground material was stored in capped 1-dram glass vials at room temperature. To isolate the “starch” fraction, a 250 mg aliquot of each powdered sample was extracted using 2 mL deionized water, then 2 mL 95% ethanol, 2 mL acetone, and finally 2 mL hexanes (Fisher Scientific). In each extraction, powder and solvent were mixed well and then centrifuged (1600 g for 3 min), the supernatant carefully decanted, and the insoluble fraction was extracted once more with the same solvent for a total of two extractions per solvent. The insoluble fraction remaining after all solvent extractions was air dried at room temperature. The dried starch fractions were stored in capped 1-dram glass vials at room temperature.

Survey of plant starches (n = 21)

Nine fresh ears of maize were collected in the continental USA and 12 fresh ears of maize were collected from Asia and Africa, from retail outlets (Supplementary Material-Table 3). Maize collection complied with available institutional, national, and international guidelines and legislation. To extract starch, kernels were steeped in warm water for 24 h and then grated using a kitchen grater and the grated kernels covered with deionized water and soaked and agitated for five minutes. The liquid was left undisturbed until starch precipitated, and the supernatant was then carefully decanted. Solids were filtered using a paper coffee filter from the liquid and discarded. The starch was resuspended in 1–2 ml of deionized water and then centrifuged at 1600 g for 3 min to pellet starch. The supernatant was removed without disturbing the pellet. The starch remaining after decanting was extracted with the same series of solvents used to extract starch from antimalarials.

Table 3 Location of collected maize samples and estimated mean annual precipitation (MAP) δ18O values and average growing season precipitation (GSP) δ18O value for each collection location.
Full size table

Isotope analysis

The relative amounts of two stable (i.e., non-radioactive) isotopes in the elements carbon (C), nitrogen (N), or oxygen (O) are presented as the ratio (R) of the heavy to light isotope—i.e., R = 13C/12C, 15N/14N, 18O/16O. Since these ratios are small, it is typical to express a sample ratio (Rsamp) in “delta notation” (δ) as parts per thousand (‰) difference relative to an internationally accepted standard zero-point (RRM), where δ = (Rsamp/RRM − 1)12. The standard used for expressing a δ value varies, with C referenced to Vienna-Pee Dee Belemnite (VPDB), N referenced to the average N2 in the atmosphere (Air), and O referenced to Vienna Standard Mean Ocean Water (VSMOW)13.

Dry material was weighed into capsules that were crimped closed and then stored in covered 96-well PCR plates. Approximately 500 μg (± 10%) of sample material was sealed into tin capsules for C and N isotope ratio analysis. Approximately 100 μg (± 10%) of sample material was sealed into silver capsules for O isotope ratio analysis. Capsules prepared for O isotope ratio analysis were stored under vacuum until analyzed except benzoic acid reference materials as they would sublime. Laboratory reference materials for normalization of measured isotope ratios to the isotope δ scales and for quality control purposes were weighed at the same time as samples and included in each analytical sequence.

Carbon and nitrogen

Measurements of δ13C values, δ15N values, wt% C, and wt% N were performed using a Thermo Scientific MAT 253 isotope ratio mass spectrometer with an attached Costech elemental analyzer (ECS4010). Two laboratory reference materials of known δ13CVPDB and δ15NAir values, glutamic acids UU-CN-1 and UU-CN-2, were included at defined positions within the analytical sequence for correction of drift (time) and linearity (peak area), as needed, and for data normalization. UU-CN-1 had a calibrated δ13C value of + 23.328 ± 0.088‰ and δ15N value of + 49.28 ± 0.27‰. UU-CN-2 had a calibrated δ13C value of − 28.254 ± 0.039‰ and δ15N value of − 4.58 ± 0.01‰. A third laboratory reference material with long-term measured δ13C and δ15N values, glutamic acid UU-CN-3, was analyzed to provide a quality control assessment; this reference material has a long-term δ13C mean of − 12.629‰, within-sequence repeatability of 0.087‰, and between-sequence uncertainty of 0.039‰. It has a long-term δ15N mean of + 9.18‰, with within-sequence repeatability of 0.28‰, and between-sequence uncertainty of 0.12‰. Additionally, two commercially available starches (C3str1, a C3 plant starch, Aldrich Chemical Corp) and C4str3, a C4 plant starch, Sigma Chemical Co.) were analyzed alongside starch fractions for comparison of element concentrations.

Oxygen

Measurements of δ18O values were performed using a Thermo Scientific MAT 253 with an attached high temperature conversion elemental analyzer (TCEA). Laboratory reference materials of known δ18O values, benzoic acids UU-OH-5 and UU-OH-7, were included at defined positions within the analytical sequence for correction of drift (time) and linearity (area), as needed, and for data normalization. UU-OH-5 had a calibrated δ18O value of + 36.35 ± 0.18‰. UU-OH-7 had a calibrated δ18O value of − 2.78 ± 0.38‰. An additional laboratory reference material with long-term oxygen measurements, benzoic acid UU-OH-6, was analyzed to provide a quality control assessment. It has a long-term δ18O mean of + 26.06‰, within-sequence repeatability of 0.57‰, and between-sequence uncertainty of 0.17‰.

Results

Initial set of antimalarials

As an initial study of antimalarials, measured element concentrations and stable isotope ratios of 18 samples analyzed in bulk are presented in Table 1, grouped by API. Only 5 of the 18 samples contained nitrogen above the limit of quantitation of 1%. Note that not all APIs in the study contain detectable nitrogen (e.g., artesunate) so nitrogen cannot be a factor for predicting authenticity. Measured element concentrations ranged from 12 to 55% for C, 1.2–15.8% for N, and 21–50% for O. Measured stable isotope ratios ranged from − 27.8 to − 10.4‰ for δ13C values, + 2.3 to + 3.4‰ for δ15N values, and + 10.3 to + 30.9‰ for δ18O values.

Comparisons of genuine (n = 9) and falsified (n = 9) antimalarials, using Mann–Whitney tests, found that both wt% C and δ13C values were significantly different between the two quality classifications (wt% C: p = 0.0003; δ values: p < 0.0001). As compared to the genuine antimalarials, the falsified antimalarials had a significantly lower median wt% C (30 vs. 43%) and a higher median δ13C value (− 11.8 vs. − 25.5‰). At p < 0.05, there were no significant differences in either wt% O or δ18O values observed between the genuine and falsified samples. Nitrogen data were not statistically tested as the majority (13 of 18, 72%) of samples contained no measurable N; however, it should be noted that the only samples containing measurable N (n = 5) were genuine antimalarials.

The lower C concentrations observed for the falsified antimalarials suggest the addition of inorganic materials, such as minerals that have lower wt% C (0–12%) than most plant and animal products (40–55%). The higher δ13C values of the falsified antimalarials also suggest the addition of either mineral carbonates (e.g., chalk) or a C4-based organic material. Further examination of differences in element concentrations and stable isotope ratios between genuine and falsified antimalarials required a component-specific approach in which excipients are separated for analysis. A follow-on survey of 26 antimalarial samples that contained or claimed to contain artesunate or artemether-lumefantrine was used for this component-specific investigation.

Follow-on set of artesunate/artemether antimalarials

The bulk C, N, and O element concentrations and stable isotope ratios of 26 additional samples containing (or claiming to contain) artesunate or co-formulated artemether-lumefantrine are presented in Table 2, grouped as genuine (n = 17) or falsified (n = 9). Like the observations made in the initial survey, only a few samples contained measurable N (n = 9). However, in contrast to the initial survey, three falsified antimalarials in the follow-on survey contained measurable N.

Measured element concentrations ranged from 29 to 49% for C, < 1.0–15.5% for N, and 20–49% for O. Measured stable isotope ratios ranged from − 27.7 to − 10.7‰ for δ13C values, − 3.1 to + 3.8‰ for δ15N values, and + 19.1 to + 28.0‰ for δ18O values.

Comparisons of the genuine and falsified antimalarials using Mann–Whitney tests found that both C concentrations and δ13C values were significantly different (%C: p < 0.0001; δ values: p < 0.0001). As compared to the genuine antimalarials, the median wt% C of the falsified antimalarials was significantly lower (37 vs. 43%) while the median δ13C value of the falsified antimalarials was significantly higher (− 12.5 vs. − 25.5‰). These results mirrored those from the initial survey, as did that there was no significant difference in either wt% O or δ18O values observed between the genuine and falsified antimalarials in the follow-on survey. Nitrogen data were not statistically tested as 17 of 26 (65%) of samples in the follow-on survey contained no measurable N; however, it should be noted that none of the falsified pills had greater than 1.1% N (w/w). As nitrogen content is not a useful metric for predicting authenticity for this general class of antimalarials, neither can the nitrogen isotope ratio be used in this manner.

Both studies combined

A diversity of APIs and brands were included in both genuine and falsified categories in the initial and follow-on surveys (Supplementary Material-Tables 1 and 2). Between the two, 7 genuine Guilin Pharmaceuticals Co. artesunate samples and 9 falsified versions of this product were tested. Comparisons of the genuine versus falsified Guilin samples found that both δ13C and δ18O values were significantly different between the two quality classifications (Mann–Whitney; p = 0.0002 and 0.0288, respectively). For carbon, the median δ value of the falsified antimalarials was higher than the genuine antimalarials (− 10.8 vs. − 19.5‰). The mean δ13C value for genuine antimalarials was − 23.6‰ with a standard deviation (SD) of 3‰; for falsified antimalarials, the mean δ13C value was − 11.8‰ with an SD of 1‰. For oxygen, the opposite was found; the median δ value of the falsified antimalarials was lower than their genuine counterparts (+ 23.8 vs. + 27.6‰). The mean δ18O value for genuine was + 24.5‰ with SD = 3‰; for falsified, the mean δ18O value was + 24.3‰ with SD = 0.5‰.

As noted above, the higher δ13C values of the falsified antimalarials analyzed as bulk suggests the addition of C4 plant-based organic material—e.g., starch derived from maize40. Another possibility is the addition of inorganic carbonates, especially where the wt% C is low. To investigate the potential sources of the carbon-containing material present in the falsified antimalarials, the “starch” fraction was extracted from each sample and analyzed. We define the starch fraction as the material that was insoluble in a series of solvent washes (water, ethanol, acetone, and hexanes; see Methods).

The measured element concentrations of the starch fractions ranged from 30 to 41% for C and 32–51% for O (see Table 2). Considering just the falsified antimalarials in the follow-on survey (n = 9), all had starch fractions with δ13C values indicative of C4 plants (see Table 2). To identify the potential growth locations of the C4 starch used in these falsified antimalarials, we collected maize from continental US, Africa and Asia and examined the correlation between the isotopic composition of growth water and maize starch.

Starch source predictions

The coordinates of known growth locations of collected maize (see Methods) were used to estimate mean annual precipitation (MAP) δ18O values through the Online Isotopes in Precipitation Calculator (OIPC)53 (Table 3). Monthly data from the OIPC were also used to calculate an average growing season precipitation (GSP) δ18O value for each collection location (Supplementary Material-Table 3). Growing seasons in the USA were identified using an online planting date calculator54 and those in Asia and Africa using online crop calendars55.

There was no significant correlation between the δ18O values of MAP and maize starch or between the δ18O values of GSP and maize starch (Pearson correlation coefficient, p > 0.05 for both). For the maize grown in Utah, USA, the relationship between the estimated δ18O values of MAP and GSP and the measured δ18O value of the starch was particularly counterintuitive. These water δ18O values were the lowest estimated from the OPIC, while the starch δ18O value was among the highest measured. Removing this sample from consideration, the correlation between the δ18O values of MAP and corn starch was still not significant; however, there was a significant correlation between the δ18O values of GSP and maize starch (r = 0.505, p = 0.023). The GSP average was not year-specific and may explain some of the unexplained residuals in the model.

We performed a linear regression with the dependent variable as the maize starch δ18O value and the independent variable as the GSP δ18O value. The slope was 1.273 ± 0.5123‰ (standard error) and the intercept was 30.86 ± 2.254‰ (Fig. 1). Uncertainties in the GSP as well as the measured δ18O values of the starch were considered insignificant relative to the residuals in the model; therefore, they were ignored. Using the inverse model, we calculated the δ18O values of growth water available to maize used as an excipient in the falsified antimalarials analyzed in the follow-on survey (see Table 4).

Figure 1
figure 1

Plot of the sampled maize starch δ18O value versus the GSP δ18O value with linear regression line and equation.

Full size image
Table 4 Predicted GSP δ18O (‰) for the follow-on set (Table 2) using the inverse model for predicting maize starch from GSP.
Full size table

Notably, the stable isotope examination of the starch from falsified samples of artesunate labeled as “Guilin Pharmaceutical Co., Ltd” and “Mekophar Chemical Pharmaceutical Joint-Stock Co., Vietnam” (4 and 5 samples, respectively) both were derived from C4 plants, but the δ18O values in extracted starch differed sufficiently (23.6 ± 0.6 and 28.7 ± 0.1‰, respectively) to indicate that the starch was derived from different sources for the two falsified sample sets. This finding suggests different localities and/or different manufacturers of the two falsified sample sets with different labels.

Discussion

The variation observed in the element concentrations and stable isotope ratios of antimalarials analyzed in bulk or as the extracted starch fraction suggests that IRMS may be a useful tool for profiling falsified antimalarials. Here, we were able to identify excipients as C3 or C4 plant-based materials based on measured δ13C values and found that C4 ingredients were exclusively used in falsified antimalarials as opposed to genuine antimalarials. By extracting and analyzing starch from maize grown in known locations, we may be able to predict potential growth water δ18O values for the starch present in falsified antimalarials. To improve predictions, future work would need to refine starch extraction methods, add more known-origin plants to the comparison database, collect more accurate data on growth water δ18O values, and generate specific prediction models for different starches commonly used as excipients in antimalarials.

Limitations of this pilot work include the relatively small sample size that precluded more detailed analysis and the lack of comprehensive information on the origins of the starch in both genuine and falsified samples to support the precipitation water isotope model. Further work is needed using the creation of simulated medicines with different API and excipient combinations, linked to isoscapes with a wider geographical spread of maize IRMS data, and the potential effect of technological processing on manufacturing food grade and pharmaceutical grade starch, to understand the potential accuracy of this technique in tracking the origin of falsified pharmaceuticals. A large overlap of regions with similar GSP δ18O values coupled with similar maize growing seasons limits the use of a simple oxygen isotope-based model to predict the location of maize-based excipients in falsified malarial drugs.

Established research has shown that stable isotope analysis can be used as a tool to elucidate the origin of many pharmaceutical products, both commercially and clandestinely made. More research is needed on a diversity of approaches for estimating the origins of pharmaceutical excipients and APIs, such as for water residues in tablets and liquids in falsified vaccines, using larger numbers of samples of known origin and greater diversities of pharmaceuticals (https://www.cghr.ox.ac.uk/research/medicine-quality-research-group/mqrg-projects/foresfa). If it is demonstrated to be helpful in providing actionable evidence for estimating falsified medicine origin, an international infrastructure for consensus protocols and appropriate data sharing will be needed.