2.3.1. Method Accuracy and Precision

Extraction efficiency and variability associated with various aflatoxin extraction procedures for recovery studies are shown in Table 3 and raw data in Supplementary Materials (Dataset S1. Aflatoxin recovery data). Wet milling method (Group II) had mean aflatoxin recovery of 121%, which at *p* = 0.05 is not significantly different from mean recovery of 80% of conventional dry milling procedure (Group I). For estimation of spiked aflatoxin, variability associated with dry milling was significantly higher compared to wet milling. Coefficient of variation (CV) also referred to as relative standard deviation (RSD) of dry milling procedure (Group I) was 2.3-fold compared to slurry, wet milling method (Group II). CV effect associated with Group II (wet milling procedure) was less (CV rank 2) compared to CV rank 4 of Group I (dry procedure) (Table 3). Multiple comparisons employing Welch ANOVA-associated Games–Howell post-hoc test showed that mean aflatoxin recovery associated with Group III procedure was significantly (*p* < 0.05) different from those of both Groups IV and V (Table 3). Of the three wet milling methods, variability (Observed CV) associated with Group V (CV rank 1) was remarkably low compared to Groups III (1.5-fold) and IV (3-fold) procedures. Additionally, Group V had the least bias and repeatability precision (HorRat value < 1). Wet milling procedure associated with Groups II and V had acceptable precision level (expressed as CV or RSD) as prescribed by modified Horwitz equation. Analyzed together, the extraction procedure associated with Groups II and V had the lowest percent bias of −4.95, acceptable variability and recovery (Table 3: HorRat value < 2; recovery = 95%).

Precision (variability) was evaluated using the Horwitz equation where the measure of variation, predicted relative standard deviation (RSDp) or CV is a function of the analyte concentration [50–54]. This is given by modified and unmodified Horwitz equations:

For modified Horwitz equation, RSD<sup>p</sup> < 2 (1−0.5logC) <sup>×</sup> 0.67 (2)

$$\text{For unmodified Horwitz equation, } \text{RSD}\_{\text{P}} < 2^{\left(1 - 0.5 \log \text{C}\right)} \tag{3}$$

where C = AFB1 concentration.

The modified Horwitz equation was used to predict RSD under repeatability and routine inter-assay conditions while the unmodified form was used for within-laboratory reproducibility conditions (intermediate precision). The observed relative standard deviation (RSDO) was compared with the RSD<sup>p</sup> to give a Horwitz Ratio (HorRat) value, thus:

$$\text{HorRat value} = \text{RSD}\_{\text{O}} / \text{RSD}\_{\text{P}} \tag{4}$$

During method validation, the Commission of the European Communities (CEC) precision requirement for both repeatability and within-laboratory reproducibility conditions is a HorRat value < 2 [55]. However, for routine work (Sections 2.3.5 and 4.2.8), we adopted inter-assay precision level of HorRat ≤ 1 [56].

Variability associated with estimation of natural AFB1 in chicken feed employing two sample splitting techniques at two sampling stages are shown in Table 4 and raw data in supplementary material (Dataset S2. Precision data). HorRat values for repeatability (HorRat**r**) and within-laboratory reproducibility (HorRat**R**) ranged 0.3–5.8 and 0.9–1.3, respectively, with all groups but one having HorRat values below the maximum allowable

limit of 2 set by European legislation [55]. The lowest intra-laboratory variation was observed in secondary sampling for splitting 25 g dry aliquots from the laboratory sample (Group VI; RSD**r**= 6.5%, HorRat**<sup>r</sup>** value = 0.3; rank H1), and highest for preparing 1.3 g dry aliquots at tertiary sampling employing modified coning and quartering procedure (Group VIII; RSD**r**= 91.5%, HorRat**<sup>r</sup>** value = 5.8; rank H6). For preparation of 1.3 g analytical samples, intra-laboratory variation associated with coning and quartering procedure (Group VIII) was 4.4-fold that for water slurry (wet milling) method (Group IX, RSD**r**= 20.6%, rank H3). At tertiary sampling stage, intermediate precision (within-laboratory reproducibility) associated with Group X (RSD**R**= 30.3%, HorRat**<sup>R</sup>** value = 1.3, rank H4) and Group XI (RSD**R**= 26.9%, HorRat**<sup>R</sup>** value = 0.9, rank H2) of water slurry procedure were almost the same (HorRat value > 2), the latter having slightly lower variability. One-way ANOVA showed no significant effect (*p* > 0.05) of the analyst or the day of analysis on means of AFB1 levels for each condition of the two water slurry procedures.

The second most important source of variability after sample collection is the sample preparation segment of the aflatoxin test procedure [41,48]. Variability associated with splitting 25 g test portions from the comminuted laboratory sample employing modified coning and quartering procedure was much below the threshold level prescribed by European Union [52] and therefore suitable mass reduction method for this purpose. Density effects and matrix particle size influence performance of sample splitting methods [57]. We minimized this by efficient dry comminution of the aggregate sample prior to mass reduction [58]. Preparation of smaller size aliquots did not yield desirable precision under intra-laboratory conditions. Indeed, the FAO sampling tool [59] will not accept mass reduction in granular products beyond paired 25 g test portions for aflatoxin analysis because this will compromise representativeness. To enhance sampling precision, aggregate sample collected should not be less than 2 kg. Variability at this level can be reduced by increasing aggregate sample size before comminution through collection of 200 g incremental samples from all potential sampling locations. Removal of test portions larger than 25 g from a 2 kg (or larger) comminuted laboratory sample will reduce variability and can still be water slurried and a small (1.3 g) slurry aliquot selected for extraction. There are also commercial laboratory mills that incorporate a sample-splitting mechanism [4,41]. However, these are expensive and not readily available. Our novel aflatoxin test procedure is designed especially for laboratories that do not have automated sample splitting facilities.

Another critical innovation described here is inclusion of wet milling (water slurring), an additional comminuting step in the extraction procedure followed by processing of a smaller slurry aliquot. This allowed analysis of an adequately large test portion (25 g and larger), minimizing huge sampling uncertainty associated with aflatoxin estimation in animal feed and with reduced extraction cost. As reported in the literature, aflatoxin contamination is characterized by heterogeneous spatial distribution and nugget effect [37,60]. Wet milling is more efficient than dry milling in producing a more homogenous sample [35,39,43,61]. Indeed, water slurring was recently incorporated as a sample homogenizing procedure for aflatoxin analysis in maize [42] and dried figs [35]. This is the first report of using wet milling sample preparation method for aflatoxin analysis in animal feed. Through optimization of sample selection and mass reduction procedures and minimizing spatial heterogeneity of aflatoxin distribution in the test portion by wet milling, we were able to reasonably reduce measurement uncertainty and extraction cost. However, our modification does not completely eradicate inherent variability associated with sample selection, sample preparation and analytical steps of the aflatoxin test procedure, but minimizes this variability at each step, as well as reducing bias at both test portion selection and analytical segment. The aflatoxin diagnosis kit used in this study was not validated specifically for chicken feed. It is designed for various food matrices and animal feed grouped as one matrix. Because animal feed is an amorphous matrix with diverse physicochemical properties which can be a source of variation due to within-class matrix effects [62], we generated validation data associated with chicken feed. Single-laboratory validation of the modified aflatoxin test procedure was carried out through collection of

data on spiking and recovery, replication, LOD and LOQ, robustness and ruggedness as the CEC prescribes [50–52,54,63].

Accuracy is a trade-off between bias and recovery data variability. We observed least bias (−4.95) and good aflatoxin recovery (95.1%) in wet milling procedure (Group. V). Recovery, precision and efficiency were compliant with CEC requirements of 75–125% for recovery and HorRat < 2 for precision [52,54,64]. AOAC and other authorities also recognize HorRat < 2 as a reliable precision criterion [64–66]. For replication studies using native aflatoxin, precision data collected under repeatability and reproducibility conditions met the EU guidelines. In absence of collaborative trial data, we estimated inter-laboratory precision using modified Horwitz equation against our within-laboratory reproducibility data, an internationally accepted practice [62]. Repeatability was both within the EU requirements and the same range for surrogate and native aflatoxin contents. Since native aflatoxin contamination is characterized by heterogeneity [4,35], we attribute the observed reduced variability to effective sample homogenization through the wet milling innovative procedure in our novel method described herein. By CEC requirements, the wet milling method described here has good repeatability and reproducibility. Reducing intrinsic variability associated with aflatoxin heterogeneity is cost effective in terms of time and resources [67].

**Table 3.** Extraction efficiency (percent recovery), bias and measurement precision of surrogate AFB1 content in chicken feed associated with dry and wet sample homogenization procedures at various extraction conditions (dilution factors) and ranked using HorRat\* value effect.


\*\* Observed residue standard deviation/predicted residue standard deviation; R is n = number of replicates; **\*** acetonitrile:water (80:20, *v*/*v*), <sup>a</sup> figures marked with this superscript were statistically compared by independent *t*-test; <sup>b</sup> phosphate buffered saline tween 20; DF, Dilution factor; StDev, Standard deviation; CV, Coefficient of variation (relative standard deviation); b,c figures marked with the same superscript were statistically compared by Welch's ANOVA (Games-Howel post-hoc test); R1−<sup>5</sup> method ranking based on CV effect with descending order of preference; \* 75–125% mean recovery is the accepted recovery range for trace AFB1 levels [52].



\* Observed residue standard deviation/predicted residue standard deviation; R is n = number of replicates; EC (Extraction conditions) 1 enumerated as i–iv: (i) dry milling (ii) matrix/ 80% acetonitrile (25:130, *w*/*v*); (iii) 80% acetonitrile extract/ phosphate buffered saline tween 20 (100:900, *v*/*v*); (iv) final dilution factor = 52: EC 2 enumerated as i–iv: (i) wet milling with matrix/water (25:37.5, *w*/*w*); (ii) slurry/80% acetonitrile (1.3:130, *w*/*v*); (iii) 80% acetonitrile extract/ phosphate buffered saline tween 20 (100:900, *v*/*v*); (iv) final dilution factor = 2500: EC 3 enumerated as i-iv: (i) dry milling; (ii) matrix/ 80% acetonitrile (1.3:130, *w*/*v*); (iii) 80% acetonitrile extract/ phosphate buffered saline tween 20 (100:900, *v*/*v*); (iv) final dilution factor = 1000: EC 4 enumerated as i–iv: (i) wet milling with matrix/water (25:37.5, *w*/*w*); (ii) slurry/80% acetonitrile (1.3:86.5, *w*/*v*); (iii) 80% acetonitrile extract/ phosphate buffered saline tween 20 (100:650, *v*/*v*); (iv) final dilution factor = 1247: Sd, Standard deviation; RSD**<sup>r</sup>** , within-laboratory repeatability relative standard deviation; RSD**R**, within-laboratory reproducibility relative standard deviation; H1–6 method ranking based on HorRat value effect with descending order of preference.
