Introduction

The world population is predicted to continue to rise sharply to 9.7 billion by 2050, according to the United Nations1. As of December 2024, the global population was approximately 8.1 billion and continues to grow steadily2. This dramatic population growth significantly increases food demand3. However, concerns about food sustainability and security arise due to limited natural resources and the effects of climate change, which impact food production, water quality, and nutritional security4. These issues affect manufacturing capacity and disrupt food supply networks3, while they also contribute to the rise of waterborne, foodborne, plant, and vector-borne diseases caused by bacteria and viruses5,6.

Antimicrobial resistance (AMR) is one of the most important global health emergencies7. Drugs, including antibacterial, antifungal, antiparasitic, and antivirals, are used to prevent and treat infectious diseases in humans, livestock, and the agricultural industry, with their misuse and overuse, leading to the adaptation and mutation of pathogenic organisms8,9,10. The effects of AMR on humans include medical treatment failures with previously effective drugs, increased healthcare costs due to the use of second-line drugs, prolonged hospital stays, and a very worrying increase in death rates due to the lack of viable treatments7,11. According to one published report, AMR causes 700,000 deaths globally each year, but without interventions, this number could rise to 10 million deaths annually by 205012. Additionally, the treatment and personal use of antibiotics during the COVID-19 pandemic since 2019 has led to high drug contamination in wastewater treatment plants and urban communities13,14. Quarantine and self-treatment may also encourage behavior where individuals purchase their own medicine rather than receiving a prescription from a medical practitioner or hospital. The lack of effective removal from sewage treatment plants has resulted in high levels of drug residues in the effluent, which enter the environment. Various classes of antibiotic contamination have been reported in rivers and groundwater15,16, and also in agricultural soils caused by groundwater infiltration, irrigation, surface runoff, and livestock manure17,18. Oxytetracycline is the most widely reported drug found in swine manure, with concentrations up to 1260 µg.g−1 dry weight19 in the manure, 406.7 ng.g−1 in the soil20, and 8400 ng.g−1 in vegetable farming where manure was used21. These high levels of contamination may have resulted from both legal and illegal use in crops22. Unless alternative ways of preventing and treating disease are found, due to the increasing demand for food production, the use of drugs in the food and feed industry will likely further increase. Data on the use and detection of antimicrobials in global crop production across many countries is limited, particularly in the Developing World. Therefore, widespread monitoring of antimicrobial residues in food crops, including both primary and processed products, is essential to ensure food safety and public health.

Legumes are a key source of plant-based alternative proteins23. Common types of legumes include soybeans, peanuts, dried beans, broad beans, lupins, lentils, cowpeas, peas, and chickpeas24. Consuming legumes can help address various aspects of malnutrition by providing essential nutrients and benefit weight control. Moreover, legume crops promote sustainable agriculture due to their low CO2 emissions and their ability to be cultivated in many regions25. In Thailand, legumes like soybeans, mung beans, and black beans are commonly planted after rice harvests because they improve soil fertility and increase crop productivity by adding organic matter and fixing nitrogen, which enriches the soil for subsequent crops26. The Food and Agriculture Organization of the United Nations (FAO) encourages greater public awareness of legumes and pulses, advocating for increased consumption to help combat poverty, hunger, and malnutrition, while promoting the sustainability of agricultural and food systems27. However, plant diseases caused by bacteria are widespread and devastating28, often requiring the use of antibiotics, with oxytetracycline being the most commonly used29. This raises concerns about the contamination of crops with antibiotics or other antimicrobial agents, which could transfer to humans and negatively impact both human health and the environment. The schematic of contamination of drugs in crops is illustrated in Fig. 1. Moreover, many countries have established regulations for the maximum residue limits (MRLs) of antimicrobials in beans. In Thailand, the MRLs for carbendazim is set at 0.1 µg.g–1 for peanuts and 0.5 µg.g–1 for dry soybeans30. In Japan, the MRLs for difenoconazole are set at 0.01 µg.g–1 for dry peanuts and 0.2 µg.g–1 for dry soybeans and peas, while tetraconazole limits are set at 0.2 µg.g–1 for soybeans and 0.09 µg.g–1 for peas31. However, regulations for only a few antimicrobial drugs exist for beans and nuts. Given these data, effective monitoring of antimicrobial contamination in legume products is crucial to ensure food safety and regulatory compliance.

Fig. 1: Sources of pharmaceutical contamination in agriculture systems.
Fig. 1: Sources of pharmaceutical contamination in agriculture systems.
Full size image

Antimicrobial drugs can spread from point sources, such as run off, untreated sewage, and municipal effluents from wastewater treatment plants (WWTP). These contaminants disperse into water bodies, soil, and agriculture, potentially entering the food chain and affecting human health.

The most widely accepted tool for the analysis of antibiotic and veterinary drug residues is an liquid chromatography-tandem mass spectrometry (LC-MS/MS), due to its high sensitivity and selectivity for the simultaneous determination of multiple classes with high accuracy and precision32. Various samples, such as animal tissues33, manures19, animal feeds34, milks35,human plasmas and urines36 have been analyzed for antibiotic residues using LC-MS/MS. However, no reports have yet been found regarding the detection of antimicrobial residues in alternative protein food samples, either individually or in multiple classes, such as antibacterial, antifungal, and antiparasitic drugs. Moreover, the extraction and cleanup of drug residues from legume samples is challenging due to the matrix effects, which contribute significantly to the uncertainty of the sensitivity of LC-MS/MS37. The Dilute-and-Shoot method (DnS) has been reported for sample preparation of animal feed and milk using LC-MS/MS analysis32,37,38. demonstrating good sensitivity and acceptable intermediate precision, with no need for extensive cleanup39. The use of DnS reduces sample preparation time and is high throughput40, while also negating the use of expensive clean-up steps sometimes used, such as solid-phase extraction (SPE) or immunoaffinity columns (IAC), while simultaneously reducing solvent waste compared to other techniques.

In this study, the aim was to develop a multi-class method for the simultaneous detection of antimicrobial residues using LC-MS/MS to monitor contaminants in plant-based alternative proteins, such as legume products. Antibiotic, antifungal, and antiparasitic drugs were those considered for inclusion in the method, with DnS selected for the extraction and preparation. Additionally, the validation studies including repeatability, intermediate precision, matrix effects, and recovery were reported. This research could enable the analysis of drug residues in legumes, both qualitatively and quantitatively, and further apply to monitoring and preventing the antimicrobial contamination in a wide range of foods and environmental samples.

Results and discussion

Optimization of LC-MS/MS conditions

The antimicrobial ionization process was investigated as part of the method’s development. A comprehensive scan mode was configured for 27 distinct antimicrobial substances, including 19 antibiotics, 4 antifungals, and 4 antiparasitics (Supplementary Material 1). Standard reference solutions (>1000 μg.mL−1) were individually prepared at a final concentration of 50 ng.mL−1 per analyte in 1 mL of water:ACN (50:50 v/v). The MS/MS parameters obtained from Q1 and Q3 are presented in Table 1. Chromatographic scanning was performed with a runtime of 1 min per substance, providing two fragmentation reactions (product ions) per precursor ion. Investigations showed that there were 54 product ions in both +ESI and –ESI modes for the 27 antimicrobial standards. Figure 2 shows the single extracted ion chromatograms of all analytes. Three analytes, i.e., CRP, FA, and TPN, were fragmented in negative polarity, while the other analytes were fragmented in positive polarity. The structures of CRP and TPN contain nitro group (-NO2)41, which has high electron affinity, and FA deprotonates to form a carboxylate anion (-COO⁻)42, resulting in stable negative ion formation. In contrast, the other analytes contain basic nitrogen atoms or nitrogen-containing groups capable of accepting a proton (H+) during ionization43. The MS/MS system successfully detected most drug substances as singly charged ions ([M + H]+), except for BCT and VCM, which were detected as doubly charged ions ([M + 2H]2+) due to their large molecular structures. However, the instrument specification is suitable for target analytes with precursor ions of less than 1400 m/z.

Fig. 2: The single extracted ion chromatograms of 27 antimicrobial compounds.
Fig. 2: The single extracted ion chromatograms of 27 antimicrobial compounds.
Full size image

The antimicrobial analytes were dissolved in solvent (water:ACN, 50:50 v/v) at the concentration of 150 ng.mL–1. The chromatograms were acquired under optimized LC–MS/MS conditions using multiple reaction monitoring (MRM) mode.

Table 1 MS/MS parameters for the detection of 27 antimicrobial substances in positive (+ESI) and negative (–ESI) modes
Full size table

The final standard solution was prepared in water:ACN (50:50 v/v) at a concentration of 200 ng.mL–1. Flow rates of 0.3, 0.5, and 0.7 mL.min–1 were evaluated. The total ion chromatography (TIC) results for these flow rates are presented in Supplementary Material 2. The results showed that a flow rate of 0.3 mL.min–1 extended the total runtime to over 20 min, with noticeable peak broadening. This prolonged the separation process and increased the consumption of eluent solutions. In contrast, flow rates of 0.5 and 0.7 mL.min–1 provided shorter runtimes of 16 and 14 min, respectively, with 0.5 mL.min–1 yielding a higher response intensity compared to 0.7 mL.min–1. However, increasing the flow rate resulted in an increase in pressure, which could pose issues if the pressure exceeds the system’s capacity or affects column integrity44. Based on these findings, a flow rate of 0.5 mL.min–1 was selected, as it offered the highest intensity with well-shaped peaks.

As previously mentioned, the detection protocol was developed for the quantification of veterinary drugs in complex feed and milk matrices38. However, this study utilized a different instrument model and column. Therefore, only the eluent solutions (mobile phases A and B) were transferred from the protocol. In contrast, MRM transitions, chromatographic separation, and MS/MS parameters were investigated to suit the 27 antimicrobial substances on our LC-MS/MS system, including adjustments of sample preparation for legumes using the DnS method. The results of MS/MS studies are shown in the TIC comparison with MRM mode (Supplementary Material 3). The optimized LC-MS/MS method employed a flow rate of 0.5 mL.min–1, an injection volume of 5 µL, and a column temperature of 25 °C. The MS/MS system was configured with the following parameters: a sheath gas temperature of 350 °C, sheath gas flow of 12 L.min–1, nebulizer gas pressure of 40 psi, capillary voltage of 3500 V for both +ESI and –ESI, nozzle voltage of 0 V, gas temperature of 320⁰C, and gas flow of 6 L.min–1. These fragmentation parameters provided the highest signal intensity for all antimicrobial analytes. This outcome includes a broader range of target drugs not covered in the previous report38, such as CBD, DFN, FCN, FTD, FZD, FA, MG, MCN, RFX, STM, TCN, and VGM.

Optimization of legume sample preparation

Legumes were chosen for the analysis of antimicrobial residues due to the indiscriminate use of veterinary drugs and other antimicrobials in aquaculture and livestock45. The water bodies can be contaminated by discharge and manure46 that may also release to soil. Many beans are widely grown and valued for both food and green manure. They need little water and have a short growing cycle. This method effectively monitors drug residues in soil that may transfer to crops, including legumes. Complex legumes exhibit a wide range of chemical and physical properties47, making the development of the analytical method for detecting veterinary drug residues in a bean matrix a challenging task. An acid-free solvent was selected for legume extraction based on a previous report38, as acidic conditions were unsuitable for maintaining the stability of veterinary drugs in the feed matrix. The effect of the legume matrix, specifically soybean (blank matrix in this study), was investigated to assess its impact on the stability of the 27 analytes and their ionization efficiency during detection. The DnS method was modified for sample preparation, as described above. However, different volumes of the extraction solvent (water: ACN, 20:80, v/v) were evaluated for the extraction of 0.5 g of soybean powder. The final standard solution was prepared at a concentration of 100 ng.g–1 in the soybean-extracted solution. The effect of the solvent volume on extraction efficiency is illustrated in Fig. 3, only 26 analytes were investigated. The results showed that increasing the volume of the extraction solvent slightly increased the matrix effect (%SSE) for all analytes, except for CBD (data on %SSE in this study is presented in Supplementary Material 4). This could be due to the higher solubility of CBD in the high-polarity solvent48, which in this case was a 50:50 v/v mixture of water and acetonitrile used in the final preparation before injection. However, the matrix effects observed across five solvent volumes for 0.5 g of soybean powder were not significantly different from each other (ANOVA, P-value = 0.97, Fcal < Fcrit). Additionally, a comparison between 2 mL and 3 mL of extraction solvent was performed in a preliminary test using recovery percentages after spiking at 100 ng.g–1. The results showed %recovery values of 0.03 ± 0.42% to 99.99 ± 17.36% for 2 mL and –0.03 ± 0.21% to 102.57 ± 21.40% for 3 mL (Supplementary Material 5). No significant difference was observed between the two volumes (P-value = 0.85, two-tailed t-test). Finally, a volume of 2 mL of water:ACN (20:80, v/v) was selected for legume sample preparation due to its efficiency in using less solution, reducing waste, and saving costs, while matching the ratio used in the previous study38. A volume of 1.5 mL was not selected, as it was insufficient during the sample preparation step. However, the extraction efficiency might be improved by adding 1% formic acid to the extraction solvent in the DnS method (would be ACN/water/formic acid, 79:20:1 v/v/v), following the literature that reported higher recovery in chicken feed (High level was spiked at 41.7 ng.g–1)32.

Fig. 3: Effect of extraction solvent volume on legume matrix effects during dilute-and-shoot sample preparation.
Fig. 3: Effect of extraction solvent volume on legume matrix effects during dilute-and-shoot sample preparation.
Full size image

The matrix effect of 26 antimicrobial compounds (except spectinomycin) were evaluated using different extraction solvent volumes (1.5 to 3.5 mL) of water:ACN, 20:80, v/v during dilute-and-shoot sample preparation of blank legume sample.

The study of the extraction ratio for soybean was performed using the DnS method with centrifugation at 3500 rpm before dilution of the supernatant into the LC vial. However, high levels of components in the soybean matrix affected the sensitivity of AJS-MS/MS, leading to sensitivity drop and clogging from precipitate inside the LC column and/or the MS/MS capillary. Figure 4a shows the split and broadened peaks in the TIC after only one centrifugation process. To protect the instrument and C18 column, a second centrifugation step at 12,000 rpm (Fig. 4b) was introduced after dilution of the sample extract). This additional step offered advantages, such as preventing a drop in LC-MS/MS sensitivity, reducing maintenance costs, and extending instrument uptime. After centrifugation of the final extract, the supernatant was filtered using a 0.22 µm PTFE filter before injection.

Fig. 4: Effect of centrifugation conditions on chromatographic performance.
Fig. 4: Effect of centrifugation conditions on chromatographic performance.
Full size image

The comparison of total ion chromatogram between different centrifuges during sample preparation; a single centrifugation at 3,500 rpm and b sequential centrifugation at 3,500 rpm and then 12,000 rpm. Analyses were performed using the same C18 column (4.6 × 150 mm, 5 µm) and LC-MS/MS.

Method Validation

The coefficient of linearity was determined by calculating the calibration curves over a five-point linear range, 0.5–500 ng.mL−1 in the solvent and 0.5–200 ng.g−1 in the blank soybean matrix. The correlation coefficients (R²) indicated linearity for each antimicrobial with values ≥ 0.990 in the solvent; however, the results from blank soybean extracted solution presented R² ≥ 0.970 due to the matrix effect on the extraction and detection (Table 2). This result demonstrated excellent linearity for each analyte, underscoring the reliability of the developed analytical method in both solvent and soybean matrix. Even though the linear range in the blank matrix was narrower than in the solvent, they still covered the MRLs for beans. However, STM was not detected in the bean matrix due to a high impact of the sample components other than the analyte (Table 2), as no peak could be accepted observing in the chromatogram (Supplementary Material 6). The retention times (RT) of the analytes in blank soybean samples are shown in Fig. 5, with all analytes eluting within the range of 3.4–10.0 min. In addition, the LOD and LOQ were also investigated in both solvent and soybean matrix conditions. The results of the detection limit and quantification limit ranged from 0.05 to 17.81 ng.g−1 and 0.12 to 53.96 ng.g−1 for solvent and 0.04 to 15.64 ng.g−1 and 0.10 to 47.40 ng.g−1 for sample solution, respectively. When considering only the MRLs for beans, the limits of detection and quantification from this developed method were lower than the guideline levels, mentioned for CBD, DFN, and TCN residues30,31.

Fig. 5: The multiple reaction monitoring (MRM) chromatogram and retention time of 26 antimicrobial residues in a spiked bean matrix.
Fig. 5: The multiple reaction monitoring (MRM) chromatogram and retention time of 26 antimicrobial residues in a spiked bean matrix.
Full size image

Chromatograms and retention times were obtained for 26 antimicrobial compounds at the concentration of 150 ng.g–1. Analyses were conducted using a 150 mm column and a flow rate of 0.5 mL.min–1 under optimized LC–MS/MS conditions.

Table 2 The retention time, linearity, limit of detection, and limit of quantification of 27 antimicrobial substances in solvent (water:acetonitrile, 50:50, v/v) and soybean extracted solution
Full size table

The precision study was conducted using spiked antimicrobial standard solutions at a concentration of 100 ng.mL−1 in both solvent and matrix solutions (Table 3). The results demonstrated RSD values ≤ 20% for both single day analysis (intra-day precision) and three different days analysis (inter-day precision).

Table 3 The precision in a day and three different days of the developed method for antimicrobials analysis in solvent (water: acetonitrile, 50:50 v/v) and soybean matrix solutions
Full size table

Additionally, the recovery of soybean extraction was evaluated using the modified DnS method. The extraction efficiency was compared between spiking the mixed antimicrobial agents before and after the extraction process. All determinations utilized the same soybean product, identical sample preparation steps, and the same detection method on the LC-MS/MS system at the IJC-FOODSEC laboratory. Recovery results are presented in Table 4. The developed method achieved good recovery, ranging from 72.08 to 108.11% for low spiking levels and from 74.79 to 108.60% for high spiking levels, with acceptable recovery for 69% of all analytes. However, the other 31% of analytes showed extraction efficiencies below 65.40%RE for low spiking and 62.34%RE for high spiking. VCM exhibited the lowest recovery which might related to the significant matrix effects in soybean samples, potentially due to binding interactions with soybean components, which impaired ionization efficiency and led to a substantial reduction in peak areas during the before-spiking extraction. The low and high spiking levels did not differ significantly, with a two-tailed t-test result of 0.99 at a 95% confidence level.

Table 4 The recovery and matrix effect of the developed method for antimicrobials analysis in soybean matrix solution
Full size table

The method’s reproducibility efficiency was also assessed through a matrix effect study. Guo et al. (2016)49 summarized potential sources of matrix effects, including co-eluting compounds and the surface activity during AJS droplet formation. Table 4 shows that over 73% of analytes experienced ion suppression, while 19% showed ion enhancement, including CPR, DXC, FTD, MCN, and OTC. Among the six analytes exhibiting ion enhancement, three substances (DXC, MCN, and OTC) demonstrated reduced soybean extraction efficiency. LCM displayed suppression with minimal matrix effect, achieving an %SSE value of 91.54%. However, %recovery values for LCM at both low and high levels remained below acceptable thresholds.

Legumes application

In this study, 97 legume samples were analyzed to monitor the risks of antimicrobial residue contamination using the developed method described above, including black beans, kidney beans, soybeans, mung beans, peanuts, and chickpeas. The samples included 10 products imported into Thaila, 80 products cultivated and sold in Thailand, and seven processing products, such as mung bean vermicellis and peanut snacks, sourced through supermarkets and online shops. Most brands have factories and warehouses located in central region of Thailand, which collected primary seeds from various farms and distributed products across the country.

The analysis of these samples revealed no drug residues were detected, or they were below the LOD of the method, in both raw and processed products. Quantification was verified based on the following criteria: (i) s/n must be >10 to indicate contamination, and (ii) the response ratio between two product ions of each antimicrobial analyte must fall within the acceptable range of 80–120%. Both organic and non-organic products yielded similar results. All samples demonstrated low or negligible response peaks, with s/n < 10 and product ion ratios <80% or exceeded 120%.

Barros et al. (2023)50 reported various studies focusing on the determination of antibiotic residues in food, utilizing numerous sample extraction and cleanup methods. Each analytical approach was developed to detect at least one analyte, with some targeting up to 175 compounds in a single run. However, none of these studies included bean samples. Instead, the literature predominantly investigated food matrices, such as milk, animal tissues (including poultry, pork, fish, and meat), butter, eggs, honey, and animal urine. Moreover, Chan et al. (2022)51 detected residues in swine feces and animal drinking water, while Krishna et al. (2012)52 analyzed human plasma samples. The widespread use of antibiotics, projected to exceed 70 billion doses globally by 203053, underscores the importance of monitoring contamination in various food sources. This study presents a novel approach for detecting different groups of antimicrobials in a single run, specifically targeting legume samples, which a trend in alternative protein food nowadays and were successfully analyzed. Future efforts should focus on applying this method to more complex bean matrices and increasing target analytes for more detection of all three major groups of antimicrobials to enhance monitoring and prevention efforts across a broader range of food products.

The novel LC-MS/MS method with a modified sample preparation technique, the DnS method was successfully developed for the simultaneous detection of 27 antimicrobial residues (including antibacterials, antifungals, and antiparasitics) in a single run (14 min). However, due to matrix effects, the method was able to detect 26 analytes in six types of legumes, except for spectinomycin, which was less responsive in the matrices. The results demonstrated good linearity and repeatability under both solvent and matrix conditions. Over 18 analytes achieved acceptable recovery rates, meeting the required standards. In application, all 97 tested samples showed quantification results below the detection limit. To further improve the method, the inclusion of additional analytes is recommended, along with applying the technique to other alternative protein-based samples or other sample types potentially contaminated by drug residues. The study shows that bean-based products produced and imported into the country do not pose AMR risks and this adds further weight to claims about their value as part of a healthy diet.

Methods

Chemicals

A total of 27 antimicrobial reference standards belonging to three different groups—antibiotic, antifungal, and antiparasitic drugs—were used for this study. Most target analytes were selected according to the critical guidelines from the World Health Organization (WHO)54 and European Commission (EC)55. Powdered standards of bacitracin (BCT), carbendazim (CBD), cefazolin sodium (CFZ), chloramphenicol (CRP), dicloxacillin sodium monohydrate (DCX), difenoconazole (DFN), doxycycline hyclate (DXC), enrofloxacin (ERF), fluconazole (FCN), furaltadone hydrochloride (FTD), furazolidone (FZD), lincomycin hydrochloride monohydrate (LCM), malachite green oxalate (MG), oxytetracycline hydrochloride (OTC), penicillin G potassium (PenG), piperacillin (PRC), rifaximin (RFX), spectinomycin sulfate (STM), sulfadiazine (SFD), tetraconazole (TCN), thiamphenicol (TPN), tiamulin (TML), vancomycin hydrochloride (VCM), and virginiamycin (VGM), were purchased from Dr. Ehrenstorfer (Augsburg, Germany). They were synthesized as a reference material for residue analysis under ISO17034 and EURACHEM/CITAC guidelines with a 95% confidence level. Erythromycin (ETM) and mecillinam (MCN) standards were purchased from Merck (Darmstadt, Germany), with fusidic acid (FA) purchased from the European Pharmacopoeia (Strasbourg, France). Methanol (MeOH) LC-MS grade was obtained from RCl Labscan (Bangkok, Thailand), with acetonitrile (ACN) LC-MS grade and dimethyl sulfoxide (DMSO) ordered from Fisher Chemical (Geel, Belgium). Additionally, acetic acid analytical grade and ammonium acetate were purchased from Merck (Darmstadt, Germany). Ultrapure water (18.2 MΩ) was prepared from an in-house Milli-Q water purification system (Merck, Darmstadt, Germany).

Standard preparation

Stock solutions of individual antimicrobial agents were weighed and dissolved in four different solvents based on their solubility, including i) water, ii) MeOH, iii) water:MeOH (50:50 v/v), and iv) DMSO:MeOH in different ratios, namely 10:90 and 50:50 ratio, v/v (Supplementary Material 1). All stock solutions could be in the concentration of ≥1000 µg.mL–1 and kept in a freezer at –20 °C until further experiment.

LC-MS/MS method optimization

The protocol of the LC-MS/MS method and sample preparation were transferred and modified from Kenjeric et al. (2024)38. In this work, a different model of instrument and column was used to develop, implement, and validate a rapid multi-group method for the determination of antimicrobial residues in legume samples. The laboratory instrument used was a 1260 infinity II prime liquid chromatography (LC) system coupled to an ultivo triple quadrupole mass spectrometry (MS/MS) equipped with a jet stream electrospray ionization source (AJS-MS/MS, Agilent Technologies, Singapore). LC was performed using a ZORBAX Eclipse Plus C18 column (4.6 × 150 mm i.d., 5 µm particle size, Agilent Technologies, Singapore), no guard cartridge was used. Both mobile phases contained water, MeOH, acetic acid, and 5 mM ammonium acetate, with 89:10:1 v/v/v for eluent A and 2:97:1 v/v/v for eluent B.

The results showed high peak responses and clear separation of a number of compounds with varying polarity, which required chromatographic optimization of both the LC and MS systems. Several factors were optimized, including the type of mobile phase, column conditions, injection volume, flow rate, column temperature, and gradient elution. In this study, the LC-MSMS parameters were optimized step-by-step as follows; The flow rate studied was from 0.3 to 0.7 mL.min–1 due to the limitation of the instrument. The injection volume was set at 5 µL and the column temperature was maintained at 25 °C. As preliminary, each run was completed in 20.5 min. Chromatographic elusion started from 100% mobile phase A to 2.0 min, then increasing mobile phase B to 50% hold time of 1.0 min. Mobile phase B was then further increased to 100% within 9.0 min, then holding time of 6.0 min at 100% mobile phase B. Finally, the system was re-equilibrated at 100% mobile phase A for further 2.5 min. The ESI-MS/MS parameters were conducted in multiple reaction monitoring (MRM) mode both in positive and negative polarity as follows: nebulizer gas of 35 psi, capillary voltage of +4000 and –4000 V, nozzle voltage of 0 V for both, gas temperature of 300 °C, gas flow of 8 L.min–1, sheath gas temperature of 350 °C, and sheath gas flow of 10 L.min–1. The MS/MS method systematically varied in six parameters. The sheath gas temperature was controlled at 350 °C. The nebulizer gas was studied from 20, 25, 30, 35, 40, and 45 psi. The capillary voltage was varied between 3000, 3500, 4000, 4500, 5000, 5500, and 6000 V in both (+ESI) and (–ESI) polarities. The nozzle voltage was studied at 0, 500, 1000, and 1500 V. The gas temperature was set from 240 °C, 260 °C, 280 °C, 300 °C, and 320 °C, the gas flow was started from 6, 8, 10, and 12 L.min–1. Additionally, the sheath gas flow was studied from 8, 10, and 12 L.min–1. The MRM mode was used throughout the experiment and the flow rate was set at 0.5 mL.min–1. The chromatographic separation process was developed to simultaneously analyze 27 antimicrobial analytes. The flow rate, column temperature, and injection volume were set as previously described. The gradient elution was optimized by reducing the run time between mobile phases A and B while still providing the good peak for all analyte responses.

Legume sample preparation

The sample preparation in this study modified the DnS method according to Kenjeric et al. (2024)38. In this work, the process was studied as follows; The soybean was homogenized by grinding to a fine powder using a Retsch mixer mill MM400 (Haan, Deutschland) at a frequency of 30 Hz for 15 s. The ground homogenized powder was kept at –20 °C before further use. This powder (blank sample) was weighed at 0.5 g into the 15 mL centrifuge tube and then extracted with different volumes of water:ACN (20:80, v/v), using 1.5, 2, 2.5, 3, and 3.5 mL of extraction solvent. The mixture (soybean and water:ACN) was vigorously shaken by a vortex (OHRUS, NJ, United States) in a vertical position for 90 min at a speed of 2,500 rpm. This experiment was studied into two processes i) The mixture was centrifuged at 3500 rpm for 10 min at 4 °C, after which, 500 µL of supernatant was diluted 1:1 (v/v) with water:ACN (80:20, v/v). The final diluted extract was filtered through a 0.22 µm hydrophobic polytetrafluoroethylene (PTFE) membrane and transferred into an LC vial for injection by LC-MS/MS. ii) the same process was followed, but after reconstitution, the mixture was centrifuged a second time at 12,000 rpm for 10 min at 4 °C to rid of the remaining matrix precipitates. The supernatant was filtered and transferred to an LC vial before injection.

Method Validation

The analytical validation of the method was performed according to the guidelines of the Directorate-General for Health and Food Safety (SANTE /11312/2021)56. For determination, the developed method was changed to dynamic multiple reaction monitoring (dMRM) mode with a cycle time of 500 ms and the detection window width at 30 s. Following the validation guidelines, two MRM transitions per analyte were acquired to ensure accurate confirmation. Soybean powder was selected to validate the developed method. For selectivity study, the final standard solutions were freshly prepared at the concentration of 50 ng.mL–1 in water:ACN (50:50, v/v). This was obtained by diluting the stock solution ( > 1000 μg.mL–1) of 27 reference standards to 10 μg.mL–1 in 100%MeOH, which was able to store at –20 °C for a month. The standard solution was further diluted to 1 μg.mL–1 in water:ACN (50:50 v/v) before reaching the final concentration. In addition, each standard solution was individually prepared and scanned using LC-MS/MS to identify the precursor ion and two product ions. Of these, the instrument’s fragmentor voltage and collision energy (CE) were optimized during this process to provide the highest response peak. Then the optimal parameters of all analytes were confirmed by the retention time (RT), signal-to-noise ratio (s/n) (more than three for detection and ten for quantification), and response ratio between two product ions (80-120% was acceptable). Moreover, the calibration curves of the solvent (water:ACN, 50:50 v/v) and matrix from soybean extraction were established for sensitivity study. The curve was prepared over five points between 0.5 to 500 ng.mL–1. To verify the blank matrix, the soybean was analyzed in six replicates to ensure that the blank matrix was free of antimicrobial residues. Moreover, the linearity, the limit of detection (LOD), and the limit of quantification (LOQ) were investigated following ICH Q2(R1) guidelines57. The analytical process can reliably differentiate from background levels (s/n ≥ 3) and the lowest concentration of substances can be quantified (s/n ≥ 10) were defined as LOD (Eq. (1)) and LOQ (Eq. (2)), respectively.

$${\rm{LOD}}=3\frac{\sigma }{s}$$
(1)
$${\rm{LOQ}}=10\frac{\sigma }{s}$$
(2)

Where σ is the standard deviation of the lowest level (five injections) s is the slope of the calibration curve

The precision of the repeatability analytical approach was reported in the percentage of relative standard deviation (%RSD). The matrix solution was spiked at the concentration of 100 ng.g–1 of the final stock solution. The process was studied within a day (Intraday-precision) and different three days (Interday-precision). The precision was accepted at RSD < 20%. This work studied both precision in solvent (water:ACN, 50:50 v/v) and soybean matrix. The extraction efficiency was evaluated and presented in percentage of recovery (%RE). Response areas before and after spiking the sample with the extracted solution were used to calculate, as shown in Eq. (3). The condition was prepared by spiking two concentration levels at 10 ng.g–1, excepted BCT, FZD, FA, OTC, PenG, and VCM were investigated at 50 ng.g–1, and 100 ng.g–1 for all analytes. Before extraction, the soybean powder (0.5 g) was combined with the final standards solution and the mixture was left for 1 h at 4 °C to equilibrate between analytes and matrix. Six replicates per condition were used. The recovery was accepted in the range of 70–120% following SANTE/11312/202156. Moreover, the matrix effect was tested as a signal suppression/enhancement (SSE) by calculating the ratio of a spiked extract against a solvent standard at the same concentration (100 ng.g–1), as shown in Eq. (4). If the values of %SSE were in the range of 0 to < 85%, it indicated ionization suppression, whereas values exceeding 115% represented ionization enhancement58. Moreover, the values higher than 90% would be assumed to have no matrix effects or have less effects49.

$$\% {\rm{RE}}=\frac{{\rm{Area}}({\rm{spiking\; before\; extraction}})}{{\rm{Area}}({\rm{spiking\; after\; extraction}})}x\,100$$
(3)
$$\% {\rm{SSE}}=\frac{{\rm{Area}}({\rm{spiking\; after\; extraction}})}{{\rm{Area}}({\rm{spiking\; in\; solvent}})\,}x\,100$$
(4)

Application of the developed method

The optimal method was used to monitor the antimicrobial residues in legume samples. A total of 97 samples consisted of 87 domestic legume samples (80 packages of primary bean seeds and 7 packages of processed bean products, such as peanut snacks and vermicelli made from mung beans) and ten imported legume samples (primary products). The warehouse of products is shown in Supplementary Material 7. The legume samples were ordered randomly from local shops, online shops, and grocery stores in Thailand. The conditions for the selection of samples included i) six types of legumes that are commonly consumed in Thailand and are rich in protein, such as soybeans, peanuts, mung beans, black beans, kidney beans, and chickpeas. ii) organic and non-organic products and iii) different brands that represent various growing areas in Thailand. Each sample package was transferred for analysis at the International Joint Research Center on Food Security (IJC-FOODSEC) Laboratory (Biotec, Pathum Thani, Thailand). All samples were homogenized and ground. The sample powder was stored in the 50 mL centrifuge tubes and kept in the freezer at –20 °C until further use. Some samples were soaked in liquid (canned food), the seeds and water were poured into the mortar and then ground before freezing at –80 °C for a night. The frozen sample was crushed to a fine powder with liquid nitrogen and transferred to the sample preparation process as described above.