Multiple mycotoxin exposure assessment through human biomonitoring in an esophageal cancer case-control study in the Arsi-Bale districts of Oromia region of Ethiopia Girma Mulisa a,b, Roger Pero-Gascon c,*, Valerie McCormack d, Jordan E. Bisanz e, Fazlur Rahman Talukdar d,f,g, Tamrat Abebe a, Marthe De Boevre c, Sarah De Saeger c,h,** a Department of Microbiology, Immunology and Parasitology, Addis Ababa University, Ethiopia b Department of Biomedical Sciences, Adama Hospital Medical College, Adama, Ethiopia c Centre of Excellence in Mycotoxicology and Public Health, Faculty of Pharmaceutical Sciences, Ghent University, Belgium d International Agency for Research on Cancer, Lyon, France e Department of Biochemistry and Molecular Biology, The Pennsylvania State University, USA f Cancer Research UK Cambridge Institute, Li Ka Shing Centre, University of Cambridge, Cambridge, UK g Cancer Research UK Cambridge Centre, Li Ka Shing Centre, University of Cambridge, Cambridge, UK h Department of Biotechnology and Food Technology, Faculty of Sciences, University of Johannesburg, South Africa A R T I C L E I N F O Keywords: Human biomonitoring Mycotoxins Risk factors Esophageal cancer Case-control study A B S T R A C T Background: Esophageal cancer (EC) is a malignancy with a poor prognosis and a five-year survival rate of less than 20%. It is the ninth most frequent cancer globally and the sixth leading cause of cancer-related deaths. The incidence of EC has been found to vary significantly by geography, indicating the importance of environmental and lifestyle factors along with genetic factors in the onset of the disease. In this work, we investigated mycotoxin exposure in a case-control study from the Arsi-Bale districts of Oromia regional state in Ethiopia, where there is a high incidence of EC while alcohol and tobacco use – two established risk factors for EC – are very rare. Methods: Internal exposure to 39 mycotoxins and metabolites was assessed by liquid chromatography-tandem mass spectrometry in plasma samples of EC cases (n ˆ 166) and location-matched healthy controls (n ˆ 166) who shared similar dietary sources. Demographic and lifestyle data were collected using structured question- naires. Principal Component Analysis and machine learning models were used to identify the most relevant demographic, lifestyle, and mycotoxin (co-)exposure variables associated with EC. Multivariate binary logistic regression analysis was used to assess EC risk. Result: Evidence of mycotoxin exposure was observed in all plasma samples, with 10 different mycotoxins being detected in samples from EC cases, while only 6 different mycotoxins were detected in samples from healthy controls. Ochratoxin A was detected in plasma from all cases and controls, while tenuazonic acid was detected in plasma of 145 (87.3%) cases and 71 (42.8%) controls. Using multivariable logistic regression analysis, exposure to tenuazonic acid (AOR ˆ 1.88 [95% CI: 1.68–2.11]) and to multiple mycotoxins (AOR ˆ 2.54 [95% CI: 2.10–3.07]) were positively associated with EC. Conclusion: All cases and controls were exposed to at least one mycotoxin. Cases were exposed to a statistically significantly higher number of mycotoxins than controls. Exposure to tenuazonic acid and to multiple myco- toxins were associated with increased risk of EC in the study population. Although aflatoxin B1-lysine and the ratio of sphinganine to sphingosine (as a biomarker of effect to fumonisin exposure) were not assessed in this study, our result emphasizes the need to characterize the effect of mycotoxin co-exposure as part of the exposome and include it in risk assessment, since the current mycotoxin safety levels do not consider the additive or synergistic effects of mycotoxin co-exposure. Moreover, a prospective study design with regular sampling should * Corresponding author. Centre of Excellence in Mycotoxicology and Public Health, Faculty of Pharmaceutical Sciences, Ghent University, Ottergemsesteenweg 460, 9000 Ghent, Belgium. ** Corresponding author. Centre of Excellence in Mycotoxicology and Public Health, Faculty of Pharmaceutical Sciences, Ghent University, Ottergemsesteenweg 460, 9000 Ghent, Belgium. E-mail addresses: roger.perogascon@ugent.be (R. Pero-Gascon), sarah.desaeger@ugent.be (S. De Saeger). Contents lists available at ScienceDirect International Journal of Hygiene and Environmental Health journal homepage: www.elsevier.com/locate/ijheh https://doi.org/10.1016/j.ijheh.2024.114466 Received 29 May 2024; Received in revised form 16 September 2024; Accepted 18 September 2024 International Journal of Hygiene and Environmental Health 263 (2025) 114466 Available online 21 September 2024 1438-4639/© 2024 Published by Elsevier GmbH. This is an open access article under the CC BY-NC-ND IGO license ( http://creativecommons.org/licenses/by-nc- nd/3.0/igo/ ). be considered in this high incidence area of EC in Ethiopia to obtain conclusive results on the role of mycotoxin exposure in the onset and development of the disease. 1. Introduction Cancer is a disease of uncontrolled proliferation of transformed cells which undertake genetic and epigenetics changes in the microenviron- ment following the interaction with host and other factors (Brown et al., 2023). Cancer is increasingly a global health issue. The International Agency for Research on Cancer (IARC) estimated that there were almost 20 million new cases of cancer and nearly 10 million cancer deaths worldwide in 2022. Projections considering population growth and aging predict over 35 million new cases of cancer in 2050, an increase of 77% compared to 2022 (Bray et al., 2024). Esophageal cancer (EC) is the ninth most common cancer worldwide and the sixth leading cause of cancer-related deaths, characterized by a poor prognosis and a five-year survival rate of less than 20% (Sheikh et al., 2023). The two common histological subtypes are Esophageal Squamous Cell Carcinoma (ESCC) and Esophageal Adenocarcinoma (EAC) of which ESCC accounts for more than 85% of its histological types (Abnet et al., 2018; Morgan et al., 2022). EC is a major public health concern worldwide, although its inci- dence and mortality has been found to vary significantly by region (Zhou et al., 2023). The highest incidence of EC and associated mortality globally is found in the areas referred as the African and Asian EC belts, which extend from eastern to southern Africa (Ethiopia to South Africa) and from western to eastern Asia (eastern Turkey to northern and central China), respectively (Abnet et al., 2018). Marked histology variation was also noticed based on geography. ESCC is the major histological type in economically less developed countries in the African and Asian EC belts while EAC is most common in more developed countries (Abnet et al., 2018; Liu et al., 2023; Sheikh et al., 2023). Different risk factors were established for both histologies: alcohol consumption is strongly associated with increased risk of ESCC, gastro esophageal reflux disease and obesity are risks factors for EAC, while smoking is a risk factor for both ESCC and EAC (Dong and Thrift, 2017; Rustgi and El-Serag, 2014). In most high-risk areas in Asia and Africa, the lack of awareness and screening programs of EC contribute to late diagnosis, leading to poor outcomes (Codipilly et al., 2018; Liang et al., 2017). Furthermore, treatment options are often limited due to the high cost of treatment and the limited availability of cancer care facilities in many African coun- tries. In Ethiopia, EC is among the top ten cancer type (Timotewos et al., 2018) with a significant increasing trend in its incidence (Wondimagegnehu et al., 2020), which is particularly high in Arsi-Bale districts of Oromia regional state (Deybasso et al., 2021a; Bulcha et al., 2018; Shewaye and Seme, 2016). Like other African countries, there is no screening program of EC in Ethiopia, leading to late diagnosis and poor survival (Middleton et al., 2021; Mwachiro et al., 2021). The treatment options available to this cancer is limited. This warrants the need to prioritize the implementation of primary preventive strategies based on the identification of specific etiology as well as associated risk factors of EC in Ethiopia. In addition to alcohol consumption and smoking, several risk factors have been linked to ESCC in Africa. These include poor nutrition and exposure to environmental contaminants, such as high levels of nitro- samines present in traditional food preservation methods (Alaouna et al., 2019; Ohashi et al., 2015). Exposure to aflatoxins from contami- nated food, has also been associated with the development of ESCC in Africa (Brown et al., 2020). ESCC has also been associated with the consumption of local alcoholic beverages, particularly kachasu in Malawi and Zambia, chang’aa in Kenya, and gongo in Tanzania (McCormack et al., 2017). In Ethiopia, where a high incidence of EC has been reported, very few etiological studies have been conducted. In recent years, already established important risk factors for ESCC, i.e. alcohol consumption and tobacco use, were observed to be rare in the affected Ethiopian population (Deybasso et al., 2021a; Shewaye and Seme, 2016). However, in African and Asian population where alcohol and tobacco are not common, exposure to mycotoxins, including fumonisins from contaminated cereal crops and maize, was identified as a significant risk factor for EC (Shephard et al., 2000, 2007; Sun et al., 2007). Based on these findings, we hypothesized that exposure to my- cotoxins, among other environmental exposures, could play a role in areas of high EC incidence in Ethiopia. Mycotoxins, i.e. secondary metabolites produced by fungi such as Aspergillus, Penicillium, Fusarium, and Alternaria toxigenic species that contaminate agricultural crops and commodities, have been associated with increased risk of various cancer types (Claeys et al., 2020; Marchese et al., 2018) with proven mechanism of carcinogenesis (Mace, 1997). IARC listed aflatoxins (including aflatoxin B1, B2, G1, G2 and M1) as carcinogenic to humans (Group 1) (Centre international de recherche sur le cancer, 2012) and fumonisins (including fumonisin B1 and B2) (International Agency for Research on Cancer, 2002) and ochratoxin A (International Agency for Research on Cancer, 1993) as possibly carci- nogenic to humans (Group 2B), while other some mycotoxins were also evaluated and classified as not classifiable as to its carcinogenicity to human (Group 3) based available evidence (Ostry et al., 2017). In the African and Asian EC belt, common cereal crops used for food have been reported to be contaminated with mycotoxins (Alizadeh et al., 2012; Chu and Li, 1994; Ghasemi-Kebria et al., 2013; Lipenga et al., 2021; Sun et al., 2007). This geographical overlap of mycotoxin occurrence in food and the epidemiology of EC may suggest exposure to mycotoxins may be a potential risk factor for EC. For example, the level of FB1 in corn and rice samples collected from areas of high risk of esophageal cancer in Iran and China were associated with increased risk of EC (Alizadeh et al., 2012; Chu and Li, 1994; Sun et al., 2007). Mycotoxin contamination has been reported to be a health concern in Ethiopia, as major agricultural food commodities were contaminated by carcinogenic mycotoxins (Ayelign and De Saeger, 2020). Maize, wheat, barley and raw milk, which are known sources of human expo- sure to mycotoxins, are used as common foodstuff in areas of high EC incidence in Ethiopia (Deybasso et al., 2021b). Coffee, which is also reported to be contaminated with mycotoxins (Soliman, 2005), is consumed at least three times per day in that local community. Most studies in the literature reported the use of food occurrence data combined with population data on food consumption to assess mycotoxin exposure. However, this approach is known for its intrinsic limitations due to the non-uniform distributions of mycotoxins in food, individual variations in toxicokinetics and bioavailability, and inaccu- rate estimations of food consumption (Heyndrickx et al., 2015). Assessment of human internal exposure to mycotoxins is best achieved using a human biomonitoring approach by analyzing biomarker com- pounds in biological fluids and tissues. The evidence generated by epidemiological (Lipenga et al., 2021; Sun et al., 2007), biomonitoring (Xue et al., 2019) and mechanistic experimental (Yu et al., 2021) studies suggests the important role of mycotoxins in the high incidence of EC in areas where food safety policy on mycotoxins is not established or not enforced. In this work we determined the concentrations of multiple mycotoxins in plasma sam- ples of EC patients and healthy controls to investigate associations be- tween mycotoxin exposure and the onset of EC in the high incidence area of Arsi-Bale districts of Oromia regional state of Ethiopia. G. Mulisa et al. International Journal of Hygiene and Environmental Health 263 (2025) 114466 2 2. Materials and methods 2.1. Study design and setting A health care facility-based case-control study design was used. Cases were pathologically confirmed newly diagnosed and treatment- naive esophageal cancer patients. Controls were residence matched endemic healthy relatives of the patients. Because controls were recruited at the health care facility from accompanying relatives of the case, other potential confounders, such as age, sex, and body mass index, were not matched. Female participants were more frequent in the con- trol group because female relatives were more likely to accompany the patients to the hospital. Cases and controls were recruited from purpo- sively selected hospitals located in the catchment area of high EC inci- dence namely Adama Hospital Medical College (AHMC), Adama General Hospital and Medical College (AGHMC), Muse General Hospital, Asella Rehoboth Hospital and Meda Wolabu Hospital. After signed written informed consent, socio-demographic data, dietary patterns, and mycotoxin awareness data were collected from 166 cases and 166 controls using interviewer administered semi-structured questionnaires. A whole blood sample of 5 mL was collected using an EDTA coated tube from each participant. EDTA plasma was immediately separated by centrifugation at 5000 rpm for 5 min, transferred to a sterile cryotube using a sterile pipet and stored at 80 C until processing. The study protocol was approved by Institutional Review Boards of Addis Ababa University, College of Health Sciences with protocol number of 024/21/ DMIP and by the Federal Ministry of Education National Ethics Com- mittee with protocol number of 03/246/221/22. 2.2. Chemicals and reagents ULC–MS grade glacial acetic acid and LC–MS grade absolute meth- anol (MeOH) from Biosolve (Deuze, France), analysis grade ammonium acetate from Merck (Darmstadt, Germany), and ultrapure water (18 mΩ cm) from an Arium® Pro water purification system from Sartorius (Goettingen, Germany) were used to prepare the chromatographic mo- bile phases and the injection solvent. LC–MS grade acetonitrile from Biosolve was used for plasma samples preparation. Analytical standards of 3-acetyldeoxynivalenol (3-ADON), aflatoxin B1 (AFB1), aflatoxin G1 (AFG1), aflatoxin G2 (AFG2), beauvericin (BEA), cyclopiazonic acid (CPA), enniatin A (ENN A), enniatin A1 (ENN A1), enniatin B (ENN B), fumonisin B1 (FB1), fumonisin B2 (FB2), fumonisin B3 (FB3), hydrolysed fumonisin B1 (HFB1), neosolaniol (NEO), ochratoxin alpha (OTα), roquefortine-C (ROQ-C), ster- igmatocystin (STC), T-2 tetraol, and zearalanone (ZAN) were purchased from Fermentek (Jerusalem, Israel); aflatoxin B2 (AFB2), alternariol methylether (AME), alternariol (AOH), citrinin (CIT), diacetoxy- scirpenol (DAS), deoxynivalenol (DON), enniatin B1 (ENN B1) HT-2 toxin (HT2), nivalenol (NIV), tenuazonic acid (TA), alpha-zearalanol (α-ZAL), beta-zearalanol (β-ZAL), zearalenone (ZEN), and alpha- zearalenol (α-ZEL) were purchased from Sigma-Aldrich (Overijse, Belgium); aflatoxin M1 (AFM1), deepoxy-deoxynivalenol (DOM), fusarenone-X (FUS-X), ochratoxin A (OTA), T-2-toxin (T2), and beta- zearalenol (β-ZEL) were purchased from Food Risk Management B.V. (Oostvoorne, Netherlands). Two multi-mycotoxin standard working solutions were prepared: (i) containing 60 μg/L of the 5 AFs and OTA standards and 600 μg/L of the other 36 mycotoxin standards in meth- anol; and (ii) containing 6 μg/L of the 5 AFs and OTA standards and 60 μg/L of the other 36 mycotoxin standards in methanol. These solutions were stored at 20 C when not in use. Isotopically labeled internal standards 13C-AFB1, 13C-CIT, 13C-DON, 13C-FB1, 13C-HT2, 13C-T2 and 13C-TA were purchased from Food Risk Management B.V. An internal standards working solution was prepared containing 19 μg/L of 13C-AFB1 and 300 μg/L of the other 6 13C-labeled internal standards in methanol. This solution was stored at 20 C when not in use. 2.3. Plasma samples for matrix-matched calibration curves and quality controls EDTA plasma purchased from the Red Cross East Flanders (Ghent, Belgium) was used to prepare matrix-matched calibration curves and quality controls. Upon receipt, the EDTA plasma was pooled and analyzed for the presence of mycotoxins by UHPLC-MS/MS. The result was negative for all mycotoxins, except OTA. The concentration of OTA was 0.13 μg/L, quantified using the standard addition method spiking with an OTA standard in the range of 0.03–0.50 μg/L. No additional testing was performed to characterize the EDTA plasma purchased from the Red Cross East Flanders and compare it with the EDTA plasma collected in Ethiopia. 2.4. Plasma samples pretreatment EDTA plasma samples were transferred from 80 C to 20 C and stored overnight. Then, they were taken to room temperature, thawed and homogenized by vortex for a few seconds. Three hundred microliter (300 μL) of each sample was transferred to Eppendorf tubes of 2 mL. Twenty μL of the internal standards working solution was added to all samples and appropriate concentrations of mycotoxin standards were added to the EDTA plasma samples from the Red Cross East Flanders to prepare matrix-matched calibration curves for mycotoxin quantification and quality controls. Then, 300 μL of acetonitrile was added to all samples for protein precipitation. Samples were vortexed for 2 min and centrifuged at 3300 g for 15 min at 4 C using a Multifuge 3 S-R centrifuge from Heraeus (Hanau, Germany). The supernatants were transferred to Eppendorf glass tubes in a Turbo Vap LV evaporator from Biotage (Dusseldorf, Germany) and evaporated at 40 C under gentle nitrogen gas flow until completely dry. The residues were re-dissolved in 150 μL of injection solvent (60:40 v/v mobile phase A/mobile phase B) by vortexing for 2 min, then centrifuged at 2100 g for 1 min. The dis- solved residue was transferred to a tube with a PVDF centrifuge filter of 0.22 μm from Millipore (Cork, Ireland) and centrifuged at 9000g for 5 min at 4 C. The filtrate was transferred into a HPLC vial with insert and closed with a cap. Air bubbles on the bottom of the insert were removed by gently tapping the vial. 2.5. UHPLC-MS/MS analysis of multiple mycotoxins, identification criteria and method validation A Waters Acquity UPLC I-class system coupled to a triple quadrupole XEVO TQ-XS mass spectrometer (Waters, Manchester, UK) equipped with an electrospray ionization (ESI) source was used for targeted analysis. Analytes were separated chromatographically using an Acquity UPLC HSS T3 analytical column (1.8 μm particle size, 2.1 mm id  100.0 mm) from Waters with an Acquity UPLC HSS T3 VanGuard pre- column (1.8 μm, 2.1 mm id x 5 mm). The chromatographic separation has already been described by Martins et al. (Martins et al., 2019, 2020). The optimized UHPLC-MS/MS parameters for the analysis of mycotoxins are shown in Table S-1. Matrix-matched calibration curves were obtained by analyzing EDTA plasma samples from the Red Cross East Flanders spiked with the mycotoxin standards at 7 concentration levels, while the concentration of internal standards was constant. Response was calculated as the in- tegrated area of the chromatographic peak of each mycotoxin divided by the area of the corresponding internal standard. To obtain accurate calibration models, a weighting factor of 1/x2 was applied to the regression modeling of the calibration curves to account for the fact that the variability (standard deviation) of the response increased propor- tionally with the concentration of the analytes over the entire concen- tration range, as is typical in bioanalytical LC-MS/MS assays (Gu et al., 2014). The midpoint of the calibration curve (at the 4th concentration level) was reinjected throughout the analytical run every 11th analysis as a quality control. G. Mulisa et al. International Journal of Hygiene and Environmental Health 263 (2025) 114466 3 Stringent criteria were used for the identification of mycotoxins by UHPLC-MS/MS based on SANTE/12089/2016 guidelines (European Commission, 2016): the retention time of the analyte in the sample extract should correspond to that of the average of the calibration standards measured in the same sequence with a tolerance of 0.2 min, chromatographic peaks with a similar peak shape should be observed in the extracted ion chromatograms of the two product ions and the ion ratio should be within 30% (relative) to that obtained from the average of the calibration standards from the same sequence. The chromatographic peaks should have a signal-to-noise ratio of at least 3. Validation assays were performed constructing three calibration curves during four different working days. The lower limit of quantifi- cation (LLOQ) was calculated using the formula LLOQ ˆ k⋅SD, where k ˆ 5 to ensure an absolute coefficient of variation within 20% for acceptable precision and SD represents the standard deviation of the replicate analyses (of the lowest point of the calibration curve), while the accuracy was within the range 80–120% (Gonzalez and Alonso, 2020). The linearity was interpreted graphically using a scatter plot. The limit of detection (LOD) was calculated as three times the standard error of the intercept, divided by the slope of the calibration curve (Kruve et al., 2015) (except for BEA, ENN A, ENN A1 and ENN B1 for which the linearity in the calibration range was poor and the LOD was determined experimentally analyzing samples spiked at low concentration and considering the identification criteria of SANTE/12089/2016 guide- lines). Accuracy was expressed as: Accuracy (%) ˆ 100 * measured concentration/spiked concentration (Gonzalez and Alonso, 2020); the measured concentration was obtained by averaging the results of three analyses conducted over four days (12 replicates). Precision was esti- mated by the coefficient of variation using the following formula: CV ˆ SD/measured concentration (Gonzalez and Alonso, 2020); the SD and the measured concentration were obtained by considering the results of three analyses conducted over four days (12 replicates). Selectivity was evaluated by analyzing (non-spiked) EDTA plasma from the Red Cross East Flanders. Matrix effect was not evaluated during method validation as it was expected to have a negligible impact on quantification due to the use of matrix-matched calibration curves, and no plasma samples indicated hemolysis or special conditions (icterus, lipemia). The absence of carry-over was confirmed by analyzing (non-spiked) EDTA plasma from the Red Cross East Flanders after a calibrant spiked at the highest (7th) concentration level. The results of the method validation are shown in Table S-2. 2.6. Statistical analysis For statistical analysis, binary categories were replaced with integers (0, 1) for the following demographic and lifestyle variables: group (case or control), gender, soup drinking, coffee drinking, porridge eating, alcohol drinking, use of separate dwelling house, use of separate kitchen and smoking of utensils. Left-censored data were substituted following the guidelines of the European Food Safety Authority (EFSA) (European Food Safety Authority et al., 2018) considering the middle-bound sce- nario: mycotoxin biomonitoring measurements with values between the limit of detection (LOD) and the lower limit of quantification (LLOQ) were assigned a concentration of LLOQ/2, and non-detects (i.e. below the LOD) were replaced by LOD/2. Data was analyzed using IBM SPSS software version 29 (p-val- ue<0.05 was considered as statistically significant). Normality of continuous data was tested using the Shapiro-Wilk test. Differences between the case and control groups in age, binary categorical variables for demography and lifestyle, and mycotoxin (co)-exposure were tested using independent samples t-test, chi-square test or Mann–Whitney U test, respectively. Multivariate binary logistic regression analysis was used to identify demographic, lifestyle and mycotoxin (co-)exposure variables that may potentially be risk factors for the development of EC. JupyterLab (version 3.6.3) software based on Python programming language (version 3.9.7) was used to create violin plots, perform Principal Component Analysis (PCA), and compute classification and regression models based on 13 machine learning algorithms (Logistic Regression, K Neighbors Classifier, Naïve Bayes, Decision Tree Classi- fier, SVM – Linear Kernel, Ridge Classifier, Random Forest Classifier, Quadratic Discriminant Analysis, Ada Boost Classifier, Gradient Boost- ing Classifier, Linear Discriminant Analysis, Extra Trees Classifier, Light Gradient Boosting Machine) included in the PyCaret (Classification and Regression Training) library (https://pycaret.gitbook.io/docs). When considering only participants under age of 50, Synthetic Minority Over- Sampling Technique (SMOTE) was used to address the imbalance be- tween the number of cases and controls in the data set (Chawla et al., 2002). 3. Results 3.1. Demographic and lifestyle variables Table 1 shows the descriptive statistics of demographic and lifestyle variables for the cases and location-matched controls in Ethiopia. The mean age of the case and control groups was statistically significantly different (independent samples t-test, α ˆ 0.05), with the case group being older. The minimum and maximum age of the cases was 18 years and 105 years, respectively. Females were more frequent in the control group (chi-square test, α ˆ 0.05). All participants were residents of the rural district of Oromia regional state with agricultural occupations. Based on food frequency questionnaires, all participants were reported to use a similar dietary source for lunch and dinner, predominantly wheat, barley and teff (Eragrostis teff). However, many cases were Table 1 Descriptive statistics of demographic and lifestyle variables for esophageal cancer cases and location-matched controls in Ethiopia. Variables Cases (n ˆ 166) Controls (n ˆ 166) Statistical test t df b p-value Continuous Mean ± standard deviation ​ ​ ​ Age (years)a 52  14 39  7 11.4 246 <0.001 Categorical Frequency (n (%)) χ2 df b p- value Gendera Female 96 (57.8) 126 (75.9) 12.2 1 <0.001 Male 70 (42.2) 40 (24.1) Established risk factors Alcohol drinkinga Yes 8 (4.8) 26 (15.7) 10.6 1 0.001 No 158 (95.2) 140 (84.3) Smoking Yes 3 (1.8) 0 (0) 3.0 1 0.082 No 163 (98.2) 166 (100) Potential risk factors Thermal injury Soup drinking Yes 142 (85.5) 130 (78.3) 2.9 1 0.087 No 24 (14.5) 36 (21.7) Coffee drinkinga Yes 166 (100) 149 (89.8) 17.9 1 <0.001 No 0 (0) 17 (10.2) Porridge eatinga Yes 164 (98.8) 145 (87.3) 15.9 1 <0.001 No 2 (1.2) 20 (12.0) Exposure to indoor air pollution Use of separate dwelling housea Yes 62 (37.3) 121 (72.9) 42.4 1 <0.001 No 104 (62.7) 45 (27.1) Use of separate kitchena Yes 75 (45.2) 102 (61.4) 8.8 1 0.003 No 91 (54.8) 64 (38.6) Smoking of utensilsa Yes 143 (86.1) 80 (48.2) 54.2 1 <0.001 No 23 (13.9) 86 (51.8) a Variables with a statistically significant difference between the case and control groups. b Degrees of freedom. G. Mulisa et al. International Journal of Hygiene and Environmental Health 263 (2025) 114466 4 unable to eat solid food at the time of sample collection due to severe dysphagia. The majority of study participants were unaware of myco- toxins (90 (54%) cases and 88 (53%) healthy controls). Most study participants reported no history of exposure to alcohol and tobacco. Of the two histological types identified, ESCC accounted for 86%. There was a relationship (chi-square test, α ˆ 0.05) between disease status and the variables alcohol drinking, coffee drinking, porridge eating, use of separate dwelling house, use of separate kitchen and smoking of utensils. 3.2. Mycotoxin exposure Evidence of exposure to 10 of the different mycotoxins investigated was observed in plasma samples of the participants as shown in Fig. 1. Ochratoxin A (OTA) was detected from all participants both in the cases and control groups, while tenuazonic acid (TA) was detected in plasma of 145 (87.3%) cases and 71 (42.8%) controls. Whereas the mycotoxins citrinin (CIT), cyclopiazonic acid (CPA), deoxynivalenol (DON), and zearalanone (ZAN) were detected both from cases and controls. Afla- toxin B2 (AFB2), enniatin B (ENNB), nivalenol (NIV), and α-zearalenol (α-ZEL) were detected only in the plasma of cases. Representative chromatograms of all mycotoxins detected in plasma samples and a comparison with a plasma sample spiked with the mycotoxin standard at a similar concentration are shown in Figure S-1. Table 2 shows the LOD, LLOQ, median and the highest concentration for the mycotoxins detected in the plasma samples. Violin plots were prepared for the mycotoxins with a frequency of detection >50% in cases and/or controls, i.e. OTA and TA (Fig. 2). The violin plots pointed out some differences in exposure for case and control groups. The con- centration of OTA in case and control groups was compared using the Mann-Whitney U test and there was a statistically significant difference (p-value<0.001), with OTA plasma concentrations being lower in cases than in controls. Regarding exposure to TA, there was also a statistically significant difference between the groups (p-value<0.001), with TA plasma concentrations being higher in cases than in controls. 3.3. Identification of potential risk factors for esophageal cancer Principal component analysis (PCA) was used to investigate the re- lationships among demographic and lifestyle variables, and mycotoxin concentrations to identify patterns and potential clusters in the data (Figure S-2). The Scores plots of the PCA show that PC2 was useful in differentiating cases and controls. Variables positively correlated with the case group were age (which is a risk factor for the development of most cancers) (White et al., 2014), smoking of utensils, coffee drinking, soup drinking, porridge eating, and the exposure to several mycotoxins, e.g. AFB2, CIT, and NIV, consistently with the higher number of positive samples in cases compared to controls. Concentration of OTA and use of separate dwelling house were positively correlated with the control group. To further investigate the datasets, machine learning algorithms included in PyCaret library were used to compute classification and regression models. The model with the highest accuracy was based on the Gradient Boosting Classifier algorithm. The area under the Receiver Operating Characteristic curve (AUC) of the computed model was 0.97 for both case and control groups and the five most important features for classification of cases and controls were age, concentration of OTA, concentration of TA, smoking of utensils and use of separate dwelling house (Fig. 3). In view of the importance of mycotoxin exposure in the classifier model, the machine learning models were recomputed to discriminate EC patients from controls based solely on mycotoxin con- centrations. The model with the highest accuracy was again based on the Gradient Boosting Classifier algorithm, the AUC was 0.92 for both cases and controls and the most relevant mycotoxins for classification of cases and controls were OTA and TA (Figure S-3). Considering the promising results of the machine learning models to classify EC patients and controls, a multivariate binary logistic regres- sion model was computed for the onset of EC based on demographic and lifestyle variables and mycotoxin exposure quartiles (Q1-low exposure, Q2-medium-low, Q3-medium-high, Q4-high exposure) (Table S-3). Table 3 shows the results of the final multivariate binary logistic regression analysis considering the five most important predictor vari- ables from the Gradient Boosting Classifier model (Fig. 3-B). Variables statistically significantly associated with an increased probability of developing EC were age, smoking of utensils and mycotoxin exposure quartile for TA. Use of a separate dwelling house and mycotoxin expo- sure quartile for OTA were positively correlated with the control group. Sensitivity analysis was performed to evaluate the influence of participant age on mycotoxin exposure. To balance the age of partici- pants in the case and controls groups, only participants younger than 50 years were considered for further analysis (n ˆ 61 and 157 for case and control groups, respectively). However, the imbalance in the number of participants in the two groups negatively affected the sensitivity of the multivariate binary logistic regression models. SMOTE was used to address the imbalance between the number of cases and controls in the data set by over-sampling of cases. Table 4 shows the results of multi- variate binary logistic regression analysis considering age and myco- toxin exposure for participants under age of 50. TA exposure was statistically significantly associated with an increased probability of developing EC, independently of age, while OTA was statistically Fig. 1. Frequency of detection of mycotoxins in plasma from esophageal cancer cases and location-matched controls in Ethiopia. *Variables with a statistically significant difference between the case and control groups. Results of the chi-square test (χ2, p-value): AFB2- aflatoxin B2 (3.0, 0.08), CIT-citrinin (11.8, <0.001), CPA-cyclopiazonic acid (0.3, 0.59), DON- deoxynivalenol (0.3, 0.57), ENNB- enniatin B (2.0, 0.16), NIV- nivalenol (6.1, 0.01), OTA- ochratoxin A (-,-), TA-tenuazonic acid (73, <0.001), ZAN- zearalanone (4.6, 0.03),α-ZEL- α-zearalenol (2.0, 0.16). G. Mulisa et al. International Journal of Hygiene and Environmental Health 263 (2025) 114466 5 significantly associated with the control group. Fig. 4 shows the histogram of the number of mycotoxin exposures per participant in the case and control groups. All participants were positive at least for one type of mycotoxin (i.e. OTA). There was a statistically significant difference in the number of mycotoxin exposures between case and control groups (Mann-Whitney U test; p-value<0.001), indi- cating that cases were exposed to more types of mycotoxins than con- trols. Multivariate binary logistic regression analysis based on number of mycotoxin exposures, age and gender for all participants of the case- control study revealed the number of mycotoxin exposures as a pre- dictor of EC independently of gender (Table S-4). Table 5 shows the results of multivariate binary logistic regression analysis considering age and number of mycotoxin exposures for participants under age of 50. The number of mycotoxin exposures was statistically significantly associated with the probability of developing EC, independently of age. 4. Discussion In Ethiopia, where there is a high esophageal cancer incidence and already established important risk factors for EC, i.e. alcohol con- sumption and tobacco use, are rare (Deybasso et al., 2021a; Shewaye and Seme, 2016), we assessed the exposure to 39 mycotoxins and me- tabolites through human biomonitoring and their association with occurrence of EC. The risk associated with individual as well as multiple mycotoxin exposure and their levels was evaluated using classification and regression models. All cases and controls were found to be exposed to at least one mycotoxin, suggesting evidence of (multi-)mycotoxin exposure as a potential risk factor for the onset EC for the first time in Ethiopia. In EC patients, a wider variety of mycotoxins and a statistically significant greater number of co-exposures were observed, including higher frequency and concentration of tenuazonic acid (TA). Multivar- iate binary logistic regression analysis indicated that TA exposure and the number of mycotoxin exposures were positively associated with EC, independent of age. These findings warrant the importance of consid- ering mycotoxins in the study of the etiology of EC in Ethiopia. We also identified other potential risk factors for EC, i.e. coffee drinking and porridge eating as proxy indicator of thermal injury and use of separate dwelling and smoking utensil as indicators of indoor air pollution levels. Environmental exposures have been identified as potential risk fac- tors of EC in Africa including exposure to heavy metals (Ahmad et al., 2011; Pritchett et al., 2017), polycyclic aromatic hydrocarbons metab- olites from in-door air pollution (Mwachiro et al., 2021), N-nitrosamines from traditional brews (Isaacson et al., 2015), smoking and daily use of spicy chilies and salted foods (Mmbaga et al., 2021). Dietary and various environmental exposure determinants of EC were reported from Ethiopia (Deybasso et al., 2021b). Mycotoxins are a major cause of food intoxication in Sub-Saharan Africa, where the hot and humid tropical climate favors the growth of fungi (Bankole et al., 2006). Recently, dietary exposure to mycotoxins in Nigerian mothers and infants was investigated by analyzing breast milk, complementary food and urine (Braun et al., 2022; Ezekiel, 2022). Multiple mycotoxin exposure was higher in urine samples from non-exclusively breastfed compared to exclusively breastfed infants, demonstrating dietary exposure to myco- toxins through complementary foods. Using an epidemiological approach the significance of mycotoxin exposure as a risk factor of EC was reported in high incidence areas in Africa (Kigen et al., 2017; Mlombe et al., 2015). However, these epidemiological studies are known for their limitations due to the non-uniform distributions of mycotoxins in food and inaccurate esti- mations of food consumption and recall bias associated food frequency questionnaire. To bridge this gap we assessed the human internal exposure to multiple mycotoxins using a human biomonitoring approach which better estimates human exposure to mycotoxins (Habschied et al., 2021; Heyndrickx et al., 2015). A notable gender imbalance in the study participants was observed, in particular a higher representation of female controls. This is because the majority of the patients’ relatives who accompanied them were fe- male. Globally, the incidence of EC is higher in males (70%) than in females (Kamangar et al., 2020; Sung et al., 2021). Gender-based vari- ations in the incidence of EC have also been reported in Africa EC high-risk region (Middleton et al., 2018). In South Africa, EC is more common among males (Ferndale et al., 2020) whereas, a study in Sudan by Gasmelseed et al., in 2015 found a higher incidence of EC among females (Gasmelseed et al., 2015). Similarly, other studies conducted in Ethiopia (Hassen et al., 2021; Shewaye and Seme, 2016) found a slightly higher prevalence among female cases. This variation may be due to difference in risk factors based on geographical and gender-associated cultural activity variation across countries. In this study, OTA was detected in all cases and controls, with OTA concentrations being higher in controls than in cases. A possible explanation for the lower OTA concentrations in cases may be related to the disease status. From the same district, a large proportion (87.6%) of cases presented with dysphagia, which is associated with weight loss due to reduced food intake (Deybasso et al., 2021a), probably causing a decrease in the level of exposure to OTA. The median OTA concentration in plasma in the present study was 0.28 and 0.73 μg/L for cases and controls, respectively (Table 2), which is higher compared to the results of mycotoxin biomonitoring studies from China (median OTA concen- tration was 0.16 μg/L) (Fan et al., 2020) and Italy (median OTA con- centration was 0.17 μg/L) (Di Giuseppe et al., 2012), and similar to a study in Bangladesh (median OTA concentration was 0.57 μg/L) (Ali et al., 2014). In contrast, a study in Northern Spain reported a median OTA concentration in plasma of 2.60 μg/L (Arce-Lopez et al., 2020a), which is around 10 times and 2 times higher compared to the results in the present study for cases and controls, respectively. The high frequency of OTA detection is consistent with previous Table 2 Limit of detection (LOD), lower limit of quantification (LLOQ), median, 95th percentile (P95) and highest concentration for the mycotoxins detected among esophageal cancer cases and location-matched controls in Ethiopia. Mycotoxins LOD (μg/L) LLOQ (μg/L) Median (μg/L) P95 (μg/L) Highest concentration (μg/L) Cases Controls Cases Controls Cases Controls AFB2 0.018 0.024 LOD and LOD and LOD and LOD and LOD and LOD and LOD and LOD and LOD and