The study protocol was approved by the Institutional Review Boards at the Kenya Medical Research Institute (KEMRI), Centers for Disease Control and Prevention (CDC), and Texas A&M University, bearing protocol IDs 2603, 6535.0 and 2013-0311F, respectively. The study's clinicaltrial.gov identifier is NCT02188953. The study started in July 2014 and ended in August 2014. A double-blinded, crossover clinical trial was conducted in which 50 participants were randomly assigned to group I or group II (Figure 1). Group I participants consumed 1 g of calcium carbonate placebo three times per day for 7 days. They did not consume any test material for 5 days (washout period), and completed the study with 1 g of ACCS100 three times per day for the last 7 days. Group II participants followed the same schedule, except they began with ACCS100 and finished with placebo. Dose ranged from 3 packet/40 kg to 3 packet/94.5 kg (Table 1). The dose in the current study has been shown to be safe and effective in several human trials. Since the ultimate goal of this mitigation strategy is for emergency use during outbreak situations, a single minimal effective dose in adults and children, respectively, would be the most field-practical.

Texas EnteroSorbents, Inc. provided ACCS100 and placebo as 1-g foil sachets packaged for individual use. Participants were instructed to consume one sachet with each of their three main meals by diluting the sachet in water provided by the study; thus, participants would consume 3 g of ACCS100 or placebo per day during the study arms. The sachets were coloured pink or green to designate the content (ACCS100 or placebo respectively). The study investigators and participants were both blinded regarding treatment type. Each day during the study each participant was provided a sterile urine cup, three treatment sachets and three 500-ml bottles of clean water.

The study was implemented in Kalimani and Kamboo villages of Kamboo sub-location in Makueni county, Eastern Kenya. The region is predominately agricultural and composed of subsistence farming, and residents eat primarily a maize-based diet. These villages were chosen because they had experienced multiple aflatoxicosis outbreaks in the prior decade.

Field teams employed convenience sampling and went door to door to recruit participants, enrolling one adult per household. Inclusion criteria consisted of the following: (1) age ≥ 18 years; and (2) consumed maize and/or groundnuts at least four times per week. Once a participant was identified, written informed consent was obtained. Consent forms were supplied in the local language (Akamba) as well as the national language (Kiswahili). Participants could decline to participate in any part of the study and were free to withdraw at any point. Upon completion of the study, participants were reimbursed with cooking items worth approximately 400 Kenya shillings (US$4).

All study participants were provided with detailed study objectives and procedures. After consent was provided, each person's medical history, height and weight were recorded and their current health status was assessed through a physical example and onsite urine testing. Urine specimens were analysed for protein and glucose levels using Chemistrip^®2 GP test strips from Roche Diagnostics (Indianapolis, IN, USA). These tests helped to rule out many health conditions involving the metabolic system, e.g., diseases of the liver (e.g., hepatitis, cirrhosis, etc.), kidney (e.g., nephrotic syndrome, renal failure, etc.) and thyroid gland (e.g., hypothyroidism, etc.). Participants who reported a history of thyroid disease, heart disease, lung disease, kidney disease, gastrointestinal disease or diabetes, or who had protein or glucose levels outside the normal range were informed of their result and excluded from the study. Female participants aged 18–49 were assessed for pregnancy status onsite using Sure-Vue® urine human chorionic gonadotropin strips (Fisher Healthcare, Pittsburgh, PA, USA). Participants who were pregnant were excluded because of unknown effects of ACCS100 on human pregnancy (although studies on pregnant rats have reported no serious adverse effects). Participants not excluded were enrolled in the study and randomly assigned to group I or II.

An enrolment survey was administered that included questions regarding the duration of residency in the village, source of maize and the shelf-life of their maize stock.

Participants provided a first morning void urine sample at baseline, and again for each of the 7 days of each treatment arm. Given the short half-life of aflatoxin in the urine (1–2 days) and the transit time of ACCS100 in the gastrointestinal tract (1–3 days), aflatoxin M1 should have completely rebounded during the wash-out period. Urine samples were collected, aliquoted and frozen each morning, then kept frozen until laboratory analysis. Two blood samples were also collected from each participant at baseline (day 0) and completion (day 20) of the trial for AFB₁-lys analysis.

Following each treatment arm, a questionnaire was administered that asked participants to rate the taste, aftertaste, smell, texture, appearance and colour of the sachet contents they had been consuming using a five-point Likert type scale (1 = really bad; 5 = really good).

Study monitors completed a worksheet each day during the study to record participant adherence to protocol (i.e., daily use of ACCS100 or placebo), the occurrence of side effects and diet. For adherence, daily use of ACCS100 or placebo was obtained by asking participants to self-report their consumption, and by collecting empty treatment sachets. For side effects, additional information was collected regarding side-effect severity (i.e., mild, moderate or severe), time of day the side effect occurred (i.e., a.m., noon or p.m.), and whether or not the participant sought treatment.

Acceptance was also assessed by administering a questionnaire at the end of the study to collect the participant's perceptions of ACCS100 and whether they would be willing to consume ACCS100 in future, or whether they would be willing to have their children consume ACCS100.

The exposure of AFM₁ in humans is by ingestion of food contaminated with AFB₁ which is then metabolised to AFM₁, or consumption of contaminated milk and dairy products. Urine samples were analysed at Texas A&M University. Analysis of urinary AFM₁ levels followed methods reported by Groopman et al. (1992) with the modifications of Sarr et al. (1995) and Wang et al. (1999). Urine samples were centrifuged at 2300 rpm, and 5.0 ml of supernatant were collected, acidified with 0.5 ml of 1.0 M ammonium formate (pH 4.5) and diluted with water to a total volume of 10.0 ml. Samples were then loaded onto a 3-ml preparative Aflatest^® WB immunoaffinity column (VICAM, Watertown, MA, USA) at a flow rate of 1 ml min⁻¹. Following washing of the column, the aflatoxin fraction was eluted from the column with 2 ml of 80% methanol, dried under N₂ and resuspended in 200 μl of a 1:1 solution of methanol:20 mM ammonium formate. Samples were analysed using a Waters HPLC system (Waters Corporation, Milford, MA, USA) with fluorescence detection capabilities. A 250 × 4.6 mm Luna C-18 column with pore size 100 Å and particle size 5 μm (Phenomenex, Torrance, CA, USA) was used to resolve AF metabolites. The mobile phase consisted of 22% ethanol buffered with 20 mM ammonium formate (pH 3.0) in water. Isocratic elution of the mobile phase for 20 min at a rate of 1 ml min⁻¹ allowed for proper chromatographic separation. External AFM₁ standards were prepared weekly and injected following every five injections of samples. The LOD for this method was 12 pg ml⁻¹ of urine for AFM₁. Random samples were aliquoted for additional verification using a Waters Acquity H-Class UPLC-MS/MS. Separation was achieved using a 2.1 × 50 mm Acquity UPLC BEH C18 column with a particle size of 1.7 μm. Isocratic separation was achieved with 70% water buffered with 1% formic acid and 30% ACN buffered with 1% formic acid. Samples (10 μl) were injected onto the column and the elution rate was 0.325 ml min⁻¹. The column effluent was directly coupled to the MS, which was operated in the positive electrospray ionisation mode. MS/MS conditions were optimised for AFM₁ and based on Warth et al. (2012). The precursor ion was set to 329.00 Da and the two product ions were 273.00 Da (quantifying ion) and 259.1 Da (qualifying ion). Urinary AFM₁ concentrations were expressed as pg mg⁻¹ creatinine to correct for variations in urine dilution among samples. Creatinine concentrations were measured at Baylor Scott & White Hospital (Temple, TX, USA).

Previous measurements of aflatoxin exposure in Kenya have been based on the serum aflatoxin B₁-lysine adduct from serum albumin (AFB₁-lys), a biomarker for long-term aflatoxin exposure. This allowed us to compare aflatoxin exposure with levels seen during previous aflatoxicosis outbreaks. The CDC's National Center for Environmental Health Division of Laboratory Sciences analysed serum specimens for AFB₁-lys adduct, which consisted of two measurements: (1) analysis AFB₁-lys by LC-MS/MS (McCoy et al. 2005); and (2) albumin measurement. To allow the release of AFB₁-lys from albumin, protein in serum specimens was digested in the presence of stable-iso-topically labelled internal standard (²H₄-AFB₁-lys) for at least 15 h at 37°C by use of a commercially available mixture of proteinases (Pronase™). AFB₁-lys and ²H₄-AFB₁-lys were then extracted by use of mixed-mode anion exchange reversed-phase SPE. Each SPE eluate was evaporated, reconstituted in mobile phase and injected onto a reversed-phase C₁₈ column. AFB₁-lys was chromatographically separated from other compounds using gradient mobile phase. Both AFB₁-lys and ²H₄-AFB₁-lys were detected with positive electrospray ionisation (ESI) in SRM mode using tandem quadrupole mass spectrometry (McCoy et al. 2005). Quantitation was based on peak area ratios interpolated against a seven-point aqueous linear calibration curve with 1/x weighting. The calibration range for serum AFB₁-lys was 0.025–10 ng ml⁻¹. There are no established critical call values for serum AFB₁-lys concentrations, i.e., there are no defined concentration thresholds distinguishing a normal or acceptable serum AFB₁-lys concentration from one that would be considered abnormal or life threatening. The LOD for AFB₁-lys was 0.02 ng ml⁻¹. Serum albumin was analysed on the Hitachi Modular P clinical analyser using the Roche^® colorimetric assay. The LOD for albumin was 0.2 g dl⁻¹. Human serum albumin – and subsequently albumin-corrected serum AFB₁-lys – has a half-life of approximately 20 days. Thus, detection of AFB₁-lys in this assay suggests a likelihood of exposure to aflatoxin within the previous 1–2 months.

We used Epi Info™ 7 (CDC, Atlanta, GA, USA) for data entry and SAS Enterprise Guide version 4.3 (SAS Institute, Cary, NC, USA) for data analysis. We performed all statistical analyses blinded, and we considered p < 0.05 to be statistically significant.

We compared demographic variables by group using paired t- and chi-square tests. We compared the palatability by treatment using Wilcoxon rank-sum test, and compared serum AFB₁-lys levels between days 0 and 20 using the Wilcoxon signed-rank test. We assessed intra-individual correlation in urinary AFM₁ levels using a Spearman correlation coefficient.

To assess efficacy, we first created an outcome variable for each participant. The outcome variable was the average of a participant's urinary AFM₁ level during the 7 days of follow-up (i.e., days 2–8) for each treatment. Urinary aflatoxin levels below the LOD were substituted with the LOD divided by the square root of 2 before being averaged. These averages were then log-transformed. Thus, each participant had two summary outcome variables: log-transformed average urinary AFM₁ during treatment; and log-transformed average urinary AFM₁ during placebo. We then fitted a general linear model with fixed effects for subject, treatment and period. We performed a modified intent-to-treat analysis, meaning that we performed an intent-to-treat analysis on all individuals who completed the study.

In order to enrol the study population of 50 participants, a total of 68 potential participants were consented and assessed. Eighteen participants were not enrolled because they did not meet the inclusion criteria (Figure 1). Study retention for randomised participants was 98%. One (2%) male participant dropped out on day 3 after being randomised and completing two treatment sessions. Thus, we included 49 participants in the statistical analyses.

The majority of participants were female (n = 36, 72%) (Table 1). Participants ranged in age from 21 to 75 years (mean = 39 years). Participant sex, age, weight, height and amount of time living in the village did not differ by group. Participants consumed an average of 1.9 maize-containing meals per day; this did not differ by treatment (placebo, 1.8 maize-containing meals per day; ACCS100, 1.9 maize-containing meals per day).

The majority of participants consumed all 21 sachets during both the ACCS100 (n = 45) and placebo (n = 44) treatment. The majority (n = 46) of participants always ingested the sachet with water; four individuals reported consuming a sachet without water (but with food) once during the study. There were 23 participants who consumed a sachet without food (but with water) between one and three times during the study.

Forty-nine participants contributed data to both arms of the study and thus were included in the efficacy analyses. Participants provided 784 (98%) of the potential 800 urine samples. Overall, 48% of samples contained detectable levels of urinary AFM₁ (range = < LOD–1986 pg mg⁻¹ creatinine).

Baseline urinary AFM₁ levels also did not vary statistically by treatment. There was no statistical correlation in urinary AFM₁ aflatoxin levels when comparing baseline of arm 1 with the baseline of arm 2, or when comparing day 1 of arm 1 with day 2 of arm 1.

Table 2 depicts geometric mean average urinary AFM₁ levels during follow-up (days 2–8)by arm and group. Our general linear model found that geometric mean urinary AFM₁ was lower during ACCS100 compared with placebo (β = –0.7093, 95% confidence limit = –1.14 to –0.28; p < 0.01), while controlling for subject and period. Once exponentiated to account for the log transformation, the results indicate that urinary AFM₁ was approximately half during treatment compared with placebo (0.49, 95% confidence limit = 0.32–0.75).

Figure 2 illustrates daily urinary AFM₁ levels by treatment. As indicated by the error bars, there was not a statistically significant difference between ACCS100 and placebo when looking at any individual day.

The average level of consumption of calcium montmorillonite clay was 3 g day⁻¹. Previous research has demonstrated that 3 g day⁻¹ of clay was the minimal effective dose that significantly decreased the AFM₁ bio-marker (Wang et al. 2008). This dose of clay (3 g day⁻¹) is equivalent to 0.25% w/w in the adult Ghanaian diet, based on an intake of approximately 1200 g day⁻¹.

Thirty-nine participants provided serum data at both time points and thus were included in this analysis (Figure 3). Serum aflatoxin median levels exhibited a statistically significant decrease from day 0 (n = 39; median = 9.3 pg mg⁻¹ albumin) to day 20 (n = 39; median = 6.4 pg mg⁻¹ albumin; p < 0.01). The day 0–20 interval includes both the placebo and ACCS100 treatments, as well as the wash out.

There was a statistically significant difference in smell by treatment, with participants on average rating the placebo as 0.25 Likert points better than ACCS100 (p < 0.05). While the placebo and treatment had slight difference in colour, there were no statistically significant differences in ratings of taste, aftertaste, appearance, colour or texture by treatment (Table 3).

The majority of participants rated aftertaste (98%), texture (97%), appearance (99%), smell (99%), taste (99%) and colour (99%) as either okay, good or really good for both the placebo and ACCS100. No one reported appearance, colour and taste as bad for the placebo. However, ACCS100 received a ‘bad’ rating from one participant (2.2%) for appearance, one (2.2%) for colour, one (2.2%) for taste and one (2.2%) for texture. Two (4.2%) people rated the placebo texture as bad.

Forty-seven participants completed the end-of-study questionnaire. Most participants (96%) had heard of aflatoxin prior to the study, and most (91%) worried about becoming sick as a result of exposure to afla-toxin. Two-thirds of participants (67%) knew of someone who had become sick from aflatoxin exposure in the past, and 9% believed they themselves had become sick from aflatoxin exposure at some point.

Most participants (98%) did not have any concerns about ACCS100 and would be willing to take ACCS100 (98%) or give it to their children (98%) if they knew it would protect them from aflatoxicosis. The majority of participants (72%) reported that they would prefer to take the clay in water as they had done during the study; the other commonly mentioned option was taking the clay plain (i.e., licking it; 13%). Participants would be willing to take the clay for at least 2 weeks (40%) or as long as recommended (38%). Similarly, participants would be willing to let their children take the clay for at least 2 weeks (36%) or as long as recommended (47%).

Approximately one-quarter of participants reported at least one adverse event when taking placebo (n = 12, 26%) or ACCS100 (n = 14, 28%). The most commonly reported adverse events reported while on placebo were nausea (14%) and abdominal discomfort (8%); for ACCS100, they were abdominal discomfort (14%) and increased appetite (8%) (Table 4). Other side effects reported at least once were diarrhoea (reported once, while on ACCS100), and dizziness with headache (reported once, while on ACCS100). These percentages were calculated by dividing the number of times an adverse event was reported by the total potential number of times when it could have been reported. There were no statistically significant differences in the reporting of adverse events by treatment.

No adverse events were graded as severe. Approximately one-third of side effects on ACCS100 (n = 6, 35%) and placebo (n = 7, 39%) were rated as moderate.