Faecal sample collection and faeces-blood preparation
In order to obtain blood-free field-collected sheep faeces from a range of different environments and sheep backgrounds, faecal samples were collected during winter-spring of 2019 (August–October) from multiple sheep properties in SA known to be free from H. contortus infection (based on previous worm egg counts and larval differentiations). Faecal samples were collected both as individual samples from single sheep, and pooled samples where ten individual samples of faeces were combined and later subsampled for vis-NIR analysis. Faecal consistency was described as very dry pellets (VDP), dry pellets (DP), dry aggregates (DA) moist pellets (MP) and very moist pellets (VMP). Fresh samples were collected off the ground from ewes, rams, lambs (< 1 year old) and hoggets (1 – 2 years old) into separate sample bags shortly after defecation. Faecal samples were transported on ice to Brisbane, QLD and stored at –20°C until further analysis. Faecal samples were also collected from Armidale, NSW as described in Kho et al. (2020). These samples were collected from a single uninfected sheep kept in an animal house, fed a diet consisting of ground wheat grain and ground lucerne hay. Samples from this animal are defined as uninfected control samples for this study. All animal procedures were approved by the FD McMaster Animal Ethics Committee, CSIRO Agriculture and Food (Animal Ethics Approval Number AEC 18/09).
The frozen samples were thawed over a two-hour period, and the weights of faeces from each sample were recorded. Faeces from each collection bag was thoroughly homogenised, and 10 g subsamples were mixed with various concentrations of diluted defibrinated sheep blood. Faeces-blood samples and controls were sandwiched between two sheets of polyethylene film (Woolworths Australia) and thoroughly dispersed by rolling and folding each sample multiple times to form a ‘slab’ (10 cm x 20 cm) before measurement at randomly selected points with a vis-NIR spectrometer .
Defibrinated sheep blood used throughout the study was purchased from a commercial supplier (Serum Australis, Australia) and serially diluted with distilled water at the ratio of 1:2. Each slab consisted of 10 g of faeces mixed with 5 mL of diluted blood to provide final concentrations of 0, 2, 2.3, 4, 4.25, 4.5, 8, 8.5 and 9 µg Hb/mg faeces. Control samples were prepared by adding 5 mL water to 10 g faeces.
Faeces were also collected from The University of Queensland (UQ) Gatton campus, QLD, where H. contortus is endemic, on two days in February and March 2020, respectively. A total of 221 mm of rainfall had been recorded over the preceding two months, providing favourable conditions for H. contortus transmission [44, 45]. These samples were included to represent samples collected from a different region known to be prone to H. contortus infections. Sheep faeces were collected off the ground shortly after defecation and were placed into individual sample bags. Sheep ID was recorded on all collection bags. The FAMACHA© scores and history of anthelmintic treatments for the corresponding sheep ID were obtained and recorded on the same day of faeces collection by trained and experienced veterinarians at UQ Gatton, QLD. Faecal samples from each bag were homogenised and subsamples were processed within 24 h of collection using the McMaster method to determine FWEC and Hemastix® to estimate the amount of blood present in the faeces. The remainder of each sample was stored at –20°C until further analysis with vis-NIR spectroscopy. Prior to spectroscopic measurement, the faecal samples were thawed for two hours and later scanned as a slab (10 cm x 20 cm) sandwiched between two sheets of polyethylene film. All animal procedures here were approved by the UQ Animal ethics committee (Anima Ethics Approval Number AEC: SVS/452/17).
Calibration and validation dataset:
As the diagnosis of H. contortus infection using FWEC and Hemastix® is typically performed by pooling faeces collected from multiple individual animals [18, 46], we determined the effects of using pooled or individual samples for building calibration models. Two calibration models were built using faeces collected from different locations in SA and NSW. Model 1 consisted of faeces from individual sheep, while Model 2 included faeces from both individual and pooled samples. Table 1 shows the description of samples included in each model.
Four groups of samples were prepared and measured to form the validation datasets. The Hb concentrations of the faecal samples used in the validation datasets were 0, 2, 2.3, 4, 4.25, 4.5, 8, 8.5 and 9 µg Hb/mg faeces. For samples collected as ‘individual’ sample type, each bag of faeces was subsampled, spiked with blood and scanned as individual samples for each Hb concentrations. For samples designated as ‘pooled’, replicates of each faeces were subsampled and spiked with blood before scanning using vis-NIR spectrometer. To include variations that may be present in future validations, a randomly selected portion of faecal samples included in Models 1 and 2 were mixed with various concentrations of Hb and scanned as a slab on separate days using the vis-NIR spectrometer (Val1 and Val2).
Faecal samples were collected from the Lower North and Yorke Peninsula in SA to represent locations not included in the models as validation dataset 3 (Val3). Faeces were also collected from UQ Gatton, QLD as validation dataset 4 (Val4) to represent samples from different regions with naturally occurring H. contortus infections. Table 2 shows detailed descriptions of the validation datasets used in this study.
Spectral acquisition with vis-NIR spectroscopy:
A Felix F-750 portable vis-NIR spectrometer (Felix instruments, Camas, WA, USA) was used to collect spectra throughout this study. The Felix F-750 is equipped with a xenon tungsten lamp as the light source. The vis-NIR spectra of blood in sheep faeces were obtained with the Felix F-750 within the wavelength range of 300–1200 nm at the spectral resolution between 8–13 nm and data resolution of 3 nm. Each spectrum was obtained using 32 scans. Faeces-blood mixture slabs were placed on top of the scanning window of the Felix F-750, and a white Teflon disc was placed above the slab to augment the signal-to-noise ratio . Spectra were acquired randomly between prepared faeces slabs and across ten random scanning points on each faeces-blood mixture slab using a randomised list generated from Microsoft Excel (2016). The averages of ten spectra from each slab were calculated and used for the development of the vis-NIR calibration models.
Faecal worm egg count, Hemastix® and FAMACHA©
Faecal worm egg counts were performed using the modified McMaster method . Briefly, faeces from each collection bag were homogenised, and a subsample of 2 g faeces was mixed with 16 mL of concentrated salt solution (magnesium sulphate, MgSO4, specific gravity = 1.18SG). The faecal slurry was strained through a tea strainer to remove large particulate matter, and a subsample of the egg suspension was examined by microscopy using the 0.3 mL McMaster counting chamber, giving a lower detection level of 30 epg.
As the faecal samples collected from QLD were naturally infected with H. contortus, the blood content in the faeces was unknown. Furthermore, there was no available method to accurately measure the blood concentrations in the faeces. Therefore, the amount of faecal occult blood in the faeces from QLD samples was estimated using Hemastix® (Bayer Australia, Pymble, NSW, Australia) based on a modified protocol described by Colditz and Le Jambre . Briefly, two subsamples from each sample bag were diluted in water to yield a final dilution of 1:500. Samples were boiled for 20 min and cooled for 3–5 mins in a container with tap water before testing. Hemastix® scores were obtained by dipping the reagent strips into the cooled diluted faecal mixture, and the colour change of the strip was assessed against a reference colour chart (provided on the reagent bottle) after 60 s to provide a score between 1 (no color change) and 5 (dark green).
The FAMACHA© scores of individual sheep were obtained by examining the colour of the lower eyelid mucous membranes and comparing it to a chart of colour standards, with a score of 1–2 representing 'not anaemic', and scores of 3–5 indicating levels of anaemia requiring treatment . Of the 22 samples collected from Gatton, eight were collected at random with no prior knowledge of infection history or FAMACHA© score; thus, these samples were removed from the analysis. The vis-NIR predicted Hb for the faecal samples were then compared with the results from the FWEC, Hemastix® and FAMACHA© assessments.
Chemometric analysis and statistics
All spectral data were exported using F750 DataViewer (v 1.2) and analysed as raw spectra using The Unscrambler X (v. 10.5.1; CAMO A/S, Oslo, Norway). RStudio (v. 1.0.153; https://rstudio.com/products/rstudio/) was used to establish the plots presented here. The spectral data were pre-treated using the Savitzky-Golay filter with second derivative order smoothing, second polynomial order, and seven smoothing points to remove scattering effects from the spectra prior to further analysis. Spectral measurements were initially investigated using principal component analysis (PCA) to identify spectrally similar samples, and outliers were identified using the Hotelling T2 statistics. Spectra that had extreme spectral signals, leverage and residual means relative to other samples were considered outliers. The impact of sample origin (location), class of sheep, faecal consistencies and sample type on the Savitzky-Golay transformed spectra was analysed using PCA within the wavelength range of 387–609 nm.
The calibration models for the detection of Hb in sheep faeces were developed using PLS regression analysis with randomised blocks of cross-validation. Calibration models built in this study were confined within the wavelength range of 387–609 nm, which contained the relevant Hb absorption bands . The coefficient of determination for correlation in calibration (R2cal), root-mean-square error of calibration (RMSEC), coefficient of determination for correlation in cross-validation (r2cv) and the root-mean-square error of cross-validation (RMSECV) were used to assess the model performance. The number of latent variables (LVs) was determined by the leave-one-out cross-validation method [47, 48, 49]. For analysis, FWEC was transformed using log10(FWEC + 10). A transformed epg of 2.82 (650 epg) was used as the threshold for anthelmintic treatment . Prediction statistics of the validation dataset was evaluated using the coefficient of determination for correlation of prediction (r2p), and root mean squared error of prediction (RMSEP). Predictions are considered good if they have r2p > 0.80 with the smallest RMSEP. A level of 3 µg Hb/mg faeces was used as a threshold for anthelmintic treatment to calculate the prediction accuracies for the calibration models built in this study [42, 50]. The sensitivity (%SN) of the model refers to the percentage of samples correctly identified as indicating the need for treatment (more than 3 µg Hb/mg faeces), while the specificity (%SP) of the model referred to the percentage of the samples correctly predicted as 'healthy' samples where treatment was not needed (less than 3 µg Hb/mg faeces). The formulas used for the calculation of sensitivity and specificity of the predicted results are as follow:
True positive indicates samples containing more than 3 µg Hb/mg faeces that were correctly predicted, whereas true negative indicates samples containing less than 3 µg Hb/mg faeces that were correctly predicted.
PCA was also applied to the QLD samples for analysis of the pre-processed spectra (Savitzky-Golay 1st derivative, 2nd polynomial order and 5 smoothing points) within the wavelength region of 387 – 609 nm for analysis of the loading and score plots.