Decoding synthetic pathways of chemical warfare precursors: advanced impurity profiling of methylphosphonothioic dichloride using GC × GC-TOFMS-chemometrics hybrid platforms

Abstract

The chemical identification of precursor synthesis pathways is crucial for enforcing the Chemical Weapons Convention (CWC) by facilitating the forensic tracking of organophosphorus nerve agents. This study introduces the initial systematic impurity-profiling platform for methylphosphonothioic dichloride, a critical precursor of V-series CWC-controlled substances. Our analysis identified 58 unique compounds, offering valuable insights using comprehensive two-dimensional gas chromatography/time-of-flight mass spectrometry in conjunction with advanced chemometric workflows. We devised a hierarchical analytical approach: (1) unsupervised pattern recognition (HCA/PCA) revealed the inherent clustering of two primary synthetic pathways, (2) oPLS-DA modeling achieved 100% classification accuracy (R² = 0.990) with 15 VIP-discriminating features, and (3) rigorous validation through permutation tests (n = 2000) and external samples (n = 12) demonstrated 100% prediction accuracy. Notably, traceability was established at impurity levels as low as 0.5%, exceeding the OPCW verification standards. The established impurity database, in combination with the dual-mode chemometric approach, provides a robust framework for identifying chemical warfare-related precursors.

---

1. Introduction

Chemical attribution involves identifying and characterizing relevant chemical samples from a target substance through forensic chemistry principles.^1,2 The research strategy utilizes various analytical techniques to examine a chemical substance or its components,^3,4 aiming to identify markers called chemical attribution signatures (CASs) that offer insights into its origin.^2,5 Impurity profiling's crucial in chemical forensics and attribution,⁶ as impurities can provide valuable information for tracing hazardous substances, including their sources and synthesis pathways.⁷ Chemometrics is integrated to assist in data analysis and improve classification accuracy through modeling.^8,9

Although chemical warfare agents have been prohibited, ongoing threats persist.^10,11 Characteristic impurities are crucial for tracing toxins, particularly in identifying the synthetic pathways of chemical warfare agents,^12,13 which is highly significant in chemical safety. The investigation of precursors is essential for attributing synthetic pathways to compounds, as key precursor compounds can offer conclusive evidence for tracking toxic agents.¹⁴ In 2010 Fraga et al.¹⁵ integrated the bioinformatics tool XCMS with chemometrics methods to analyze liquid chromatography-mass spectrometry (LC-MS) data from methylphosphonyl dichloride (DC) samples for source traceability. Subsequently, in 2018, Fraga et al.¹⁶ extended their analysis of DC synthesis and identified that commercially synthesized DC, methylphosphonic difluoride (DF), or methylphosphonic acid (MPA) was derived from DC. In 2021, Lu et al.¹⁴ conducted a comprehensive study of the potential classification of the crucial tabun precursor, N,N-dimethylphosphoramidic dichloride (DMPADC), using GC-MS for 27 samples from three synthetic pathways.

Methylphosphonothioic dichloride (MPTDC, CAS# 676-98-2) is commonly used as a primary raw material for phosphorylation reagents in organic synthesis. It serves various purposes such as fungicides, flame retardants, and surfactants.^17,18 This compound is a vital precursor for organophosphorus nerve agents, particularly the V-series.^19,20 Owing to its classification as a recognized precursor for chemical weapons and its limited non-weapon applications, it was categorized as a Schedule 2 compound by the Organisation for the Prohibition of Chemical Weapons (OPCW).²¹ Despite its significant role, information on the chemical properties of MPTDC is scarce. Therefore, conducting chemical attribution research on MPTDC is essential for forensic tracing of toxic agents and for supporting efforts to prevent the illicit use of chemical warfare agents.

In this study, two synthetic routes were explored. Gas chromatography/time-of-flight mass spectrometry (GC × GC-TOFMS) was used for monitoring and analysis. Chemometrics was integrated to apply both supervised and unsupervised models. Impurities suitable for identification were classified using the obtained chemical information, achieving the chemical identification of MPTDC (Fig. 1).

Fig. 1 Illustration of the strategy used in the present study.

2. Materials and methods

2.1. Safety

MPTDC and its precursors, DC and methyldichlorophosphine, are toxic and corrosive organophosphorus compounds that react strongly with water. All reaction experiments were carried out in a laboratory fume hood while wearing appropriate laboratory attire, including goggles and nitrile gloves.

2.2. Procedures for the synthesis of MPTDC

Two classical synthetic routes were chosen for MPTDC,²² as shown in Fig. 2. The precursor organophosphorus compounds used in the synthesis of MPTDC, DC, and methylphosphonium dichloride were synthesized and purified by distillation from commercially available feedstocks to compounds of over 99% purity. The initial precursor compound, dimethyl methylphosphonate (DMMP), was procured from Rhawn, phosphorus trichloride was procured from Alladin, P₂S₅ and toluene were procured from Macklin, and AlCl₃ was procured from Alfa Aesar, all with purities exceeding 99%.

Fig. 2 Synthetic routes of MPTDC.

Three parallel samples were synthesized using the designated methods, achieving purities ranging from 80% to 92%. Route a entailed synthesizing DC from DMMP and chlorosulfoxide, followed by the production of MPTDC using DC and and P₂S₅.²² This solvent-free route was utilized. Route b initiated with the production of methyl dichlorophosphine from phosphorus trichloride and iodomethane under anhydrous and oxygen-free conditions. Subsequently, methyl phosphorothioformyl dichloride was created from methyl dichlorophosphine and sublimated sulfur, catalyzed by AlCl₃ in a nitrogen atmosphere using toluene as the solvent.²²

2.3. Instrumentation

The samples synthesized using MPTDC were analyzed with a two-dimensional GC × GC-TOFMS (Guangzhou Hexin Instrument Co., Ltd, GGT 0620). Sample injection utilized a fully automatic multifunctional injection platform (CTC PAL RTC). The front inlet temperature was set at 290 °C with a split ratio of 10 : 1. The primary column was a DB-5MS (30 m × 0.25 mm × 0.25 μm, Agilent Technologies), and the secondary column was a DB-17 (1.1 m × 0.18 mm × 0.18 μm, Agilent Technologies). Helium served as the carrier gas at a flow rate of 1 mL min⁻¹. The GC column oven was initially held at 50 °C for 2 min and then ramped up to 280 °C at a rate of 5 °C min⁻¹, with a total runtime of 48 min. The modulation column was a high-voltage (HV) modulation column (1.2 m × 0.25 mm) with a modulation period of 5 s. The interface and ion source temperatures were maintained at 280 °C and 250 °C, respectively. The MS detector voltage was −1750 V, ionization energy was 70 eV, MS acquisition mass range was 40–500 amu, data acquisition rate was 100 spectra per s, and solvent delay was 3 min. ³¹P-NMR spectra were obtained in chloroform-d at room temperature on a Bruker 121.5 MHz instrument.

The synthesized samples were analyzed in the electron ionization (EI) mode using an Agilent 7693 A-8890 GC/5977B mass spectrometer. The detector was equipped with a DB-5MS chromatographic column (30 m × 0.25 mm × 0.25 μm, Agilent Technologies). A 1 μL sample was injected in splitless mode at 250 °C. Helium served as the carrier gas with a flow rate of 1 mL min⁻¹. The GC column oven started at 50 °C and then heated at a rate of 15 °C min⁻¹ to 280 °C, where it was held constant for 2 min. The total analysis time for the sample was 15.33 min, with a solvent delay of 2.50 min. The mass spectrometer scanned the range of 50–500 m/z at a speed of 3.21 scans per s. The transmission line, ion source, and quadrupole temperatures were maintained at 280 °C, 230 °C, and 150 °C, respectively.

2.4. Identification of CASs

The samples, solvent blanks, and method blanks were analyzed using GC × GC-TOFMS. A two-dimensional total ion chromatogram was produced using the Canvas data processing software, with a signal-to-noise ratio threshold of 20 for peak recognition. Specific compounds, identified by their retention time pairs, were automatically detected as peaks. The X-axis represents the first-dimensional retention time (min), and the Y-axis represents the second-dimensional retention time (s).

Compounds were identified by comparing the mass spectra with the NIST 17 (v2.3) and OCAD (v.20_2018) libraries, supplemented by relative retention indices. When reliable matches were lacking, molecular structures were inferred through mass spectrometry fragment analysis. Peak area normalization was employed to determine the relative compound proportions. Potential CAS numbers were assessed to eliminate interferences such as solvent peaks and column bleeds. For qualitative assessment, peak centers were utilized for low-concentration compounds, while peak edges were employed for high-concentration compounds. Validated CAS numbers specific to synthetic pathways and not present in control samples were organized into a peak table, with samples as rows and impurities as columns.

The system automatically screened compounds detected in the peak list, excluding solvents and starting materials, based on their presence in the blank samples. Compounds appearing more than twice in the samples and with a match score exceeding 800 from the NIST library matches were prioritized.

3. Results and discussion

3.1. Chemical analysis of chemical attribution signatures

Following the screening criteria outlined in Section 2.4, we identified 58 characterized compounds, mainly organophosphorus compounds and aromatic derivatives with benzene rings. A detailed analysis showed that Route a produced 47 CASs, while Route b yielded 26, with 17 common factors between the two routes. Trace amounts of methyl phosphate compounds were detected in Route a samples, as shown in Fig. 3. Considering the experimental procedures of Route a, which utilized DMMP as the initial material, it is likely that the methyl phosphate compounds originated from this precursor. The reaction conditions and chemical changes in Route a seem to form and preserve these methyl phosphate derivatives, acting as distinct markers for this synthetic pathway.^16,23 Methylphosphonic dichloride reacts with P₂S₅ in the absence of a solvent, potentially leading to increased impurities due to the high reactivity of P₂S₅ and the lack of a solvent, causing uncontrolled conditions and side reactions. In contrast, methyl dichlorophosphine reacts with sublimated sulfur in the presence of the AlCl₃ catalyst under a nitrogen atmosphere, with toluene as the solvent in Route b. The use of AlCl₃ as a catalyst and toluene as a solvent in Route b can improve the reaction's selectivity,²⁴ thereby reducing impurity formation.

Fig. 3 Organophosphorus impurities in MPTDC samples.

3.2. Route classification

Parallel samples synthesized via each route displayed broadly consistent results. ³¹P-NMR analysis indicated specificity between the two pathways. The ³¹P-NMR chemical shift of the pure product of MPTDC was 79.46 ppm.²⁵ The ³¹P-NMR spectrum of the MPTDC samples synthesized via Route a revealed the presence of phosphorus-containing impurities alongside the unreacted raw material DC (44.83 ppm ²⁶). In contrast, the sample synthesized via Route b did not exhibit additional phosphorus-containing impurities in the ³¹P-NMR spectrum (Fig. 4). Further characterization of these impurities necessitates integration with GC × GC-TOFMS and subsequent data analysis for definitive conclusions.

Fig. 4 ³¹P-NMR spectra of MPTDC and synthetic samples from the two routes.

The results of GC × GC-TOFMS facilitated enhanced understanding of these compounds. Fig. 5 illustrates notable distinctions in the detected compounds between the two pathways. The GC × GC-TOFMS spectra utilized a third dimension (comprising color shades or heights) to depict signal intensity, with the X and Y coordinates denoting the retention times of the two columns, Rt₁ and Rt₂ (retention time, min), respectively.

Fig. 5 GC × GC-TOFMS chromatograms of MPTDC crude samples.

Table 1 presents 58 CASs screened to identify potential candidates with high matches for MPTDC attribution. Certain impurities were found to be crucial in distinguishing between the two synthesis pathways. Specifically, methyl phosphate compounds were uniquely associated with Route a, likely due to the initial material, DMMP. The presence of these methyl phosphate compounds indicates that the sample originated from Route a. Moreover, impurities common to both routes include specific sulfur compounds (e.g., compounds 12 and 26) and organophosphorus compounds (e.g., compounds 15 and 53). These compounds are generated during synthesis when sulfur elements are introduced, particularly through the use of elemental sulfur and P₂S₅. While these shared impurities may not directly indicate the route, they confirm the sample's association with this chemical system and offer insights into the general synthetic approach.

Table 1 Potential CASs in samples synthesized using two synthetic routes

Entry	Name	MW^a	Rt1^a	CAS number^b
a MW: molecular weight; Rt1: retention time 1. b CAS number: “—” for compounds with no CAS number.
1	2-Methyl-3-buten-2-ol	86.07	38.88	115-18-4
2	3-Methyl-2-buten-1-ol	86.07	42.12	556-82-1
3	2-Butene-1,4-diol	88.05	35.07	110-64-5
4	2-Propanone, 1-cyclopropyl-	98.07	37.57	4160-75-2
5	1,3,5-Trioxepane	104.04	37.73	5981-06-6
6	Aminomethanesulfonic acid	110.10	28.92	13881-91-9
7	1,2-Pentadiene, 4-methoxy-4-methyl-	112.08	36.07	49833-91-2
8	Methylmalonic acid	118.02	40.57	516-05-2
9	Thiopropionic acid, S-ethyl ester	118.04	35.23	2432-42-0
10	Dimethyl methylphosphonate	124.03	5.32	756-79-6
11	N-Nitroso-N-methyl-3-aminopropionic acid	132.05	39.37	10478-42-9
12	S-Methyl-L-cysteine	135.03	47.29	1187-84-4
13	Phosphoric acid, trimethyl ester	140.02	6.55	512-56-1
14	Methyl dichlorophosphate	147.92	5.47	677-24-7
15	O,O,O-Trimethyl thiophosphate	156.00	7.80	152-18-1
16	Phosphorothioic acid, O,O,S-trimethyl ester	156.00	11.22	152-20-5
17	Phosphorochloridothioic acid, O,O-dimethyl ester	159.95	7.38	2524-03-0
18	3,3-Dimethylglutaric acid	160.07	40.23	4839-46-7
19	DL-Ethionine	163.06	40.57	67-21-0
20	Phosphorodichloridothioic acid, O-methyl ester	163.90	6.55	2523-94-6
21	1,4′-Bipiperidine	168.16	6.48	4897-50-1
22	Phosphorodithioic acid, O,O,S-trimethyl ester	171.98	13.30	2953-29-9
23	Phosphorodithioic acid, O,S,S-trimethyl ester	171.98	16.47	22608-53-3
24	2-Ethylhexanal ethylene glycol acetal	172.14	46.51	—
25	Diethyl succinate	174.09	33.48	123-25-1
26	Methylsulfonyl-methanesulfonyl chloride	191.93	11.13	22317-89-1
27	Tri(propylene glycol) methyl ether	206.15	39.63	25498-49-1
28	2,4-Di-tert-butylphenol	206.17	22.07	96-76-4
29	3,5-Dimethoxycinnamic acid	208.07	45.37	16909-11-8
30	4-Morpholinopropanesulfonic acid	209.07	26.29	1132-58-2
31	9-Ethyl-9H-carbazol-3-ylamine	210.11	32.82	132-32-1
32	Benzoxazol-5-amine, 2-(2-pyridyl)-	211.07	31.01	58431-37-6
33	2,5-Diethoxy-3-methyl-3-vinylhexane	214.19	33.32	—
34	Diphenylmethylphosphine oxide	216.07	31.07	2129-89-7
35	1-(3,4-Dichlorophenyl)-3-methylurea	219.07	8.13	3567-62-2
36	2-Hydroxy-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5-one	224.08	34.32	17910-73-5
37	2-Methoxyacridin-9-ol	225.08	20.73	857574-55-1
38	4-Methoxyacridin-9-ol	225.08	23.82	73663-88-4
39	2,4-Dimethyl-6-(1-phenylethyl)phenol	226.31	42.48	92673-75-1
40	3,4′-Isopropylidenediphenol	228.11	19.87	46765-25-7
41	Clomazon	239.07	36.23	81777-89-1
42	Trichlorfon	255.92	8.29	52-68-6
43	6-Ethoxy-4-methyl-3-phenylcoumarin	280.11	24.82	263365-04-4
44	Mefexamide	280.18	27.38	1227-58-8
45	2,6′-Dimethoxy-2′-hydroxychalcone	284.10	26.03	1435451-87-8
46	3,4′-Dimethoxy-2-hydroxychalcone	284.10	15.88	18778-37-5
47	2,5-Dimethoxy-2′-hydroxychalcone	284.10	15.07	5452-99-3
48	6,2′-Dimethoxy-3-hydroxyflavone	298.08	28.17	1203801-43-7
49	6-Isopropoxy-9-oxoxanthene-2-carboxylic acid	298.08	23.73	33458-93-4
50	1,4-Piperazinediethanesulfonic acid	302.37	19.38	5625-37-6
51	5-Benzoyl-4-hydroxy-2-methoxybenzenesulfonic acid	308.03	24.57	4065-45-6
52	2-Propenethioamide, 3-[3,5-bis(1,1-dimethylethyl)-4-hydroxyphenyl]-2-cyano-, (2E)-	316.16	7.13	148741-30-4
53	Bithionol	353.88	18.73	97-18-7
54	3,4,2′,4′,6′-Pentamethoxychalcone	358.14	32.29	76650-20-9
55	2′,3,4,5,6′-Pentamethoxychalcone	358.39	21.87	944447-14-7
56	Thiencarbazone-methyl	390.03	23.88	317815-83-1
57	4′,5,7-Trihydroxy 3,3′,6,8-tetramethoxyflavone	390.09	18.62	58130-91-9
58	14-O-Acetyldaunomycinone	456.10	7.87	29984-41-6

---

To simplify guidelines, we suggest that analysts initially screen for route-specific impurities. For instance, identifying methyl phosphate compounds specific to Route a can promptly indicate the route. If these compounds are absent, subsequent analysis may concentrate on common impurities to validate the sample's connection with either route. This sequential method streamlined the analysis, offering dependable data to differentiate between the two routes.

Among the compounds detected, we identified derivatives of aromatic compounds introduced into the solvents. These include compounds 45, 46, 48, and 54, which are derivatives of the chalcone found in toluene.^27–29 Additionally, compounds 43 and 49 exhibit a benzoxy heterocyclic structure. Route a utilized no solvents, while Route b utilized toluene as the solvent. Consequently, impurity derivatives found in toluene, such as oxides and other alkoxy-substituted compounds, can serve as characteristic compounds for subsequent data and model development. This highlights the significance of investigating factors such as solvents.

During our thorough analysis of impurities in the MPTDC synthesis process, we identified compounds such as sildenafil, bromosporine, and tecloftalam whose presence could not be explained by the synthetic routes used. Despite employing this method and solvent blanks in our analytical protocol, the appearance of these compounds remains puzzling and requires further investigation. We suspect that these unexpected impurities may stem from various sources, such as unexpected by-products of synthesis reactions and environmental contamination. Environmental factors, including storage and transfer methods, need to be considered during the experimental process.^5,30,31 Impurities were introduced at each experimental step. The potential of these impurities to act as CASs is a subject of debate and requires further examination for more precise attribution. Long-chain olefins and alkanenitrile compounds such as octadecanitrile were found in most samples but were not identified as CASs in this study. Their presence is likely due to their use as plasticizers and surfactants,³² possibly introduced during sample handling. Interestingly, these compounds were absent in samples transferred using glass pipettes, indicating that their introduction was linked to specific handling procedures rather than the inherent chemical composition of the samples.^33,34

3.3. Multivariate statistical models

A 6 × 58 data matrix was created by utilizing the peak areas of the labeled compounds as distinctive information. A multivariate statistical classification model was then established for the MPTDC samples synthesized through two different routes using Metaboanalyst. Data preprocessing included sum normalization of the matrix and autoscaling (centered using mean and scaled using standard deviation) to enhance the data.

Hierarchical cluster analysis (HCA) is a clustering algorithm that constructs nested trees by evaluating similarity between data points, commonly using Euclidean distance. After applying different pretreatment methods to all the samples, normalization yielded the most satisfactory classification. Within the hierarchical framework of the analysis, each sample type initially formed its own cluster, which then progressively merged with other clusters. Cluster analysis results were visualized as dendrograms at varying levels of granularity. These dendrograms reveal that samples from a particular class consistently have nearest neighbors from the same class.

The HCA dendrogram (Fig. 6) from the CAS data of each MPTDC sample demonstrates that the samples can be precisely grouped using the CAS associated with their synthetic pathways. The vertical axis indicates sample similarity or distance, with lower values denoting higher similarity or closer proximity, while the horizontal axis represents the MPTDC samples and illustrates how they merge into clusters based on similarities. Notably, all samples were correctly classified, achieving 100% accuracy. Route a samples (a-1, a-2, and a-3) clustered together, and Route b samples (b-1, b-2, and b-3) formed a separate cluster, clearly distinguishing between the two routes. This highlights the effective discriminatory capability of CASs between the two synthetic pathways.

Fig. 6 HCA dendrogram of the MPTDC samples.

Subsequently, we utilized principal component analysis (PCA) on the dataset in an unsupervised manner. PCA is a robust multivariate analysis method that reduces numerous correlated variables into a few principal components (PCs), capturing most of the dataset's variation. In the two-dimensional PCA plot for the MPTDC samples (Fig. 7), the X-axis (PC₁) represents the first principal component, explaining 25.1% of the total variance, while the Y-axis (PC₂) represents the second principal component, explaining 40.7% of the total variance. The percentages indicate the proportion of total variance explained by each principal component. Each data point in the plot corresponds to an individual MPTDC sample, with Route a samples shown in pink and Route b samples in green. The ellipses show 95% confidence intervals for each route. As shown in Fig. 7, the PCA model effectively distinguished the samples from the two routes, with a clear separation between the pink and green clusters. This indicates that the PCA model successfully captured the differences in CASs between the two synthetic routes.

Fig. 7 PCA results of the MPTDC samples.

The variable importance plot of 58 CASs of MPTDC samples is shown in Fig. 8, ranking features based on their contributions to classification accuracy (mean decrease accuracy, MDA). The selected characteristic compounds for the two synthetic pathways added substantial value to the model, clearly differentiating Route a from Route b. Notably, 13 compounds had MDA values below 0.01, signifying their limited impact on the model's classification accuracy.

Fig. 8 Variable importance plot of the MPTDC samples.

Based on the constructed data matrix, we also utilize a classification model supervised by orthogonal partial least squares discriminant analysis (oPLS-DA). This method integrates PLS-DA with orthogonal signal filtering to enhance the separation of information unrelated to predefined categories from the original matrix. As a result, oPLS-DA optimizes the distinction between sample groups by reducing within-group differences and emphasizing between-group variances, making it highly effective in differentiating two sample groups. Based on these findings, we constructed the oPLS-DA model using the data. The results demonstrated that the oPLS-DA score plot (Fig. 9) enables a clear classification of the two sample classes, validating the robustness of our feature compound selection and endorsing the feasibility of route attribution.

Fig. 9 oPLS-DA scoring chart of the MPTDC samples.

Supervised classification models are prone to overfitting; that is, the model performs well on training data but poorly on unknown samples. Therefore, it is crucial to thoroughly validate the reliability of supervised classification models. Permutation testing is a widely used method for model evaluation, involving the random disruption of group labels for each sample, followed by modeling and prediction. R²X and R²Y represent the explanatory power of the model in the X and Y matrices, respectively, while Q² indicates the predictive ability of the model. Ideally, the model's performance is better when R² and Q² values are closer to 1. The validation of the oPLS-DA model, depicted in Fig. 10, stabilized Q² = 0.785 and R² = 0.990 after multiple training sessions. These results indicate that the model is both reliable and robust, with strong predictive capabilities.

Fig. 10 Permutation test plots for the oPLS-DA trained with 20, 100, 1000, and 2000 iterations.

In this study, an external test set was used to validate the performance of the oPLS-DA model. The test set comprised six unknown samples of MPTDC. As shown in Fig. 11, these samples were classified with 100% accuracy. The predicted values for both the training set and the test set confirmed that the validation samples were correctly assigned to their respective categories, demonstrating the high accuracy and robustness of the oPLS-DA model. The practical detection capability was confirmed using external samples with impurities as low as 0.5% abundance,³⁵ All markers remained distinguishable with >95% peak intensity reproducibility (RSD < 15%, n = 6). The empirical detection limit was established at 0.5% abundance, based on consistent identification in external validation samples.

Fig. 11 oPLS-DA score plot of test samples.

4 Conclusion

This study is the first comprehensive investigation into the classification of synthetic routes to MPTDC. It evaluates the feasibility of chemical attribution analysis using characteristic data derived from MPTDC synthesis. By combining GC × GC-TOFMS and ³¹P-NMR techniques, we identified 58 CASs through two distinct synthetic routes. HCA, PCA, and oPLS-DA models were developed using GC × GC-TOFMS spectra of the identified CASs. These models effectively distinguished samples from the two synthetic routes. Test sample validation demonstrated that the oPLS-DA model achieved outstanding classification accuracy, correctly categorizing 100% of the six averaged test samples from both routes. These findings underscore the potential of chemical attribution analysis in identifying precursor compounds in harmful substances.

Future research will focus on investigating the stability of CASs in the synthesized MPTDC samples, which is crucial for understanding the long-term applicability of chemical attribution methods. Extending these findings, we aim to enhance our understanding of the traceability of other related compounds. This will further enhance the scope of chemical attribution analyses, providing new insights and tools for identifying and tracking toxic substances and their precursors.

Data availability

Additional data are made available in the ESI† of this manuscript.

Conflicts of interest

The authors declare that there is no conflict of interest.

Acknowledgements

The authors wish to thank the temporary working group (TWG) on chemical forensics of the Scientific Advisory Board (SAB) of the OPCW.

References

1. S. Bell, Forensic Chemistry, Annu. Rev. Anal. Chem., 2009, 2(1), 297–319,  DOI:10.1146/annurev-anchem-060908-155251.
2. C. G. Fraga, Chemical forensics, Talanta, 2018, 186, 585,  DOI:10.1016/j.talanta.2018.04.057.
3. X. Lu, X. Zhu, R. Gao, H. Tang, C. Pei and H. Wang, et al., Chemometrics-assisted analysis of chemical impurity profiles of tabun nerve agent using comprehensive two-dimensional gas chromatography-time-of-flight mass spectrometry, J. Chromatogr. A, 2022, 1685, 463643,  DOI:10.1016/j.chroma.2022.463643.
4. J. Wang, X. Lu, Z. Zhang, R. Gao, C. Pei and H. Wang, Application of chemical attribution in matching OPNAs-exposed biological samples with exposure sources- based on the impurity profiles via GC × GC-TOFMS analysis, J. Chromatogr. A, 2024, 1718, 464718,  DOI:10.1016/j.chroma.2024.464718.
5. S. P. B. Ovenden, R. L. Webster, E. Micich, L. J. McDowall, N. W. McGill and J. Williams, et al., The identification of chemical attribution signatures of stored VX nerve agents using NMR, GC-MS, and LC-HRMS, Talanta, 2020, 211, 120753,  DOI:10.1016/j.talanta.2020.120753.
6. Y. Cho, S. W. Song, J. Sung, Y.-S. Jeong, C. R. Park and H. M. Kim, Hyperspectral depth-profiling with deep Raman spectroscopy for detecting chemicals in building materials, Analyst, 2017, 142(19), 3613–3619,  10.1039/C7AN00894E.
7. S. Säde, L. Mörén, K. Höjer Holmgren, H. Hakulinen, A. Larsson and M. Engqvist, et al., Chemical impurity profiling: Linkage of starting materials and an intermediate synthesis product of a carbamate chemical warfare agent, Forensic Chem., 2024, 39, 100581,  DOI:10.1016/j.forc.2024.100581.
8. R. Kumar and V. Sharma, Chemometrics in forensic science, TrAC, Trends Anal. Chem., 2018, 105, 191–201,  DOI:10.1016/j.trac.2018.05.010.
9. G. Sauzier, W. van Bronswijk and S. W. Lewis, Chemometrics in forensic science: approaches and applications, Analyst, 2021, 146(8), 2415–2448,  10.1039/D1AN00082A.
10. M. Kloske and Z. Witkiewicz, Novichoks–The A group of organophosphorus chemical warfare agents, Chemosphere, 2019, 221, 672–682,  DOI:10.1016/j.chemosphere.2019.01.054.
11. P. R. Chai, E. W. Boyer, H. Al-Nahhas and T. B. Erickson, Toxic chemical weapons of assassination and warfare: nerve agents VX and sarin, Toxicol. Commun., 2017, 1(1), 21–23,  DOI:10.1080/24734306.2017.1373503.
12. X. Lu, Z. Zhang, R. Gao, H. Wang and J. Xiao, Recent progress in the chemical attribution of chemical warfare agents and highly toxic organophosphorus pesticides, Forensic Chem., 2021, 39(2), 334–349,  DOI:10.1007/s11419-021-00578-7.
13. Z. Zhang, J. Wang, X. Lu and H. Wang, Recent progress in the application of chemical attribution in forensic profiles, Forensic Chem., 2024, 37, 100556,  DOI:10.1016/j.forc.2024.100556.
14. X. Lu, Z. Zhang, H. Liu, H. Tang, R. Gao and C. Pei, et al., Forensic signatures of a chemical weapon precursor DMPADC for determination of a synthetic route, Talanta, 2021, 232, 122476,  DOI:10.1016/j.talanta.2021.122476.
15. C. G. Fraga, B. H. Clowers, R. J. Moore and E. M. Zink, Signature-Discovery Approach for Sample Matching of a Nerve-Agent Precursor Using Liquid Chromatography–Mass Spectrometry, XCMS, and Chemometrics, Anal. Chem., 2010, 82(10), 4165–4173,  DOI:10.1021/ac1003568.
16. J. J. Moran, C. G. Fraga and M. K. Nims, Stable-carbon isotope ratios for sourcing the nerve-agent precursor methylphosphonic dichloride and its products, Talanta, 2018, 186, 678–683,  DOI:10.1016/j.talanta.2018.04.021.
17. H. Saeidian, M. Babri, D. Ashrafi, M. Sarabadani and M. T. Naseri, Fragmentation pathways of O-alkyl methylphosphonothionocyanidates in the gas phase: Toward unambiguous structural characterization of chemicals in the Chemical Weapons Convention framework, Anal. Bioanal. Chem., 2013, 405(21), 6749–6759,  DOI:10.1007/s00216-013-7094-8.
18. J. Jiang, B. Lu, B. Zhu, X. Li, G. Rauhut and X. Zeng, Hydrogen-Bonded π Complexes between Phosphaethyne and Hydrogen Chloride, J. Phys. Chem. Lett., 2023, 14(8), 4327–4333,  DOI:10.1021/acs.jpclett.3c00695..
19. K. H. Holmgren, C. A. Valdez, R. Magnusson, A. K. Vu, S. Lindberg and A. M. Williams, et al., Part 1: Tracing Russian VX to its synthetic routes by multivariate statistics of chemical attribution signatures, Talanta, 2018, 186, 586–596,  DOI:10.1016/j.talanta.2018.02.104.
20. D. Jansson, S. W. Lindström, R. Norlin, S. Hok, C. A. Valdez and A. M. Williams, et al., Part 2: Forensic attribution profiling of Russian VX in food using liquid chromatography-mass spectrometry, Talanta, 2018, 186, 597–606,  DOI:10.1016/j.talanta.2018.02.103.
21. Organisation for the Prohibition of Chemical Weapons (OPCW), Article IV–Chemical Weapons, OPCW, https://www.opcw.org/chemical-weapons-convention/articles/article-iv-chemical-weapons.
22. F. W. Hoffmann, D. H. Wadsworth and H. D. Weiss, Organic Phosphorus Compounds. III.l O,O′-Dialkyl Alkylphosphonothioates and O-Alkyl Alkylphosphonochloridothioate, J. Am. Chem. Soc., 1958,(80), 3945–3948,  DOI:10.1021/ja01548a037.
23. L. Maier, Organic Phosphorus Compounds 90. A Convenient, One-Step Synthesis of Alkyl- and Arylphosphonyl Dichlorides, Phosphorus, Sulfur, Silicon Relat. Elem., 1990, 47(3–4), 465–470,  DOI:10.1080/10426509008038002.
24. M.-Y. Huang, J.-C. Wu, F.-S. Shieu and J.-J. Lin, Preparation of high energy fuel JP-10 by acidity-adjustable chloroaluminate ionic liquid catalyst, Fuel, 2011, 90(3), 1012–1017,  DOI:10.1016/j.fuel.2010.11.041.
25. L. Maier, Organic Phosphorus Compounds 59. A new method for the synthesis of phosphonic dichlorides, Helv. Chim. Acta, 1973, 56(1), 492–494,  DOI:10.1002/hlca.19730560143.
26. H. Fakhraian, M. Zarinehzad and H. Ghadiri, Reexamination of the Friedel - Crafts Reaction of Methyl Chloride with Phosphorus Trichloride for the Preparation of Methyl Phosphonic Dichloride, Org. Prep. Proced. Int., 2005, 37(4), 377–385,  DOI:10.1080/00304940509354970.
27. K. M. Kadish, X. Mu and J. E. Anderson, Recommended methods for the purification of solvents and tests for impurities: Benzene and toluene, Pure&App.Chern., 1989, 61(10), 1823–1828.
28. N. Li and J. Zou, Determination of impurities in industrial caprolactam produced from toluene by SPME and GC–MS, J. Appl. Polym. Sci., 2006, 100(4), 3141–3144,  DOI:10.1002/app.23075.
29. M. Mahmoudi and J. Behin, Detrimental effect of industrial toluene organic impurities on the density of rigid polyurethane foam and their removal, Korean J. Chem. Eng., 2021, 38(1), 204–214,  DOI:10.1007/s11814-020-0679-y.
30. S. Lociciro, P. Hayoz, P. Esseiva, L. Dujourdy, F. Besacier and P. Margot, Cocaine profiling for strategic intelligence purposes, a cross-border project between France and Switzerland. Part I. Optimisation and harmonisation of the profiling method, Forensic Sci. Int., 2007, 167(2–3), 220–228,  DOI:10.1016/j.forsciint.2006.06.052.
31. C. G. Fraga, G. A. Pérez Acosta, M. D. Crenshaw, K. Wallace, G. M. Mong and H. A. Colburn, Impurity Profiling to Match a Nerve Agent to Its Precursor Source for Chemical Forensics Applications, Anal. Chem., 2011, 83(24), 9564–9572,  DOI:10.1021/ac202340u.
32. H. Wang, S. Zhang, N. Weng, B. Zhang, G. Zhu and L. Liu, Discovery and identification of a series of alkyl decalin isomers in petroleum geological samples, The Analyst, 2015, 140(13), 4694–4701,  10.1039/C4AN00815D.
33. V. Krishnan, N. Mahalingam and H. V. K. Udupa, Electroreduction of lauronitrile and stearonitrile, J. Chem. Technol. Biotechnol., 1979, 29(11), 703–706,  DOI:10.1002/jctb.503291106.
34. R. I. S. Romão and A. M. Gonçalves Da Silva, Phase behaviour and morphology of binary mixtures of DPPC with stearonitrile, stearic acid, and octadecanol at the air–water interface, Chem. Phys. Lipids, 2004, 131(1), 27–39,  DOI:10.1016/j.chemphyslip.2004.03.009.
35. D. Ballabio and V. Consonni, Classification tools in chemistry. Part 1: linear models. PLS-DA, Anal. Methods, 2013, 5(16), 3790–3798,  10.1039/c3ay40582f.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5ay00870k

---