Animals, cages, and sample collection

All pigs were housed in individual metabolic cages equipped with feeders, nipple water dispensers, and feces collection trays, all feed was in the form of meal and is described in the references cited; pigs were fed at 2.5 times their maintenance DE requirements¹⁸. The facility has a temperature control system which was maintained at 19 ± 2 °C. Ileal digesta samples were freeze-dried, and feces samples were dried in a forced-air oven at 55 °C for 48 h. Subsequently, freeze-dried digesta samples were ground through a 0.5 mm mesh, and dried stool samples, were through a 1 mm mesh in a laboratory mill (Arthur H. Thomas Co. Philadelphia, PA).

Laboratory analysis

Dry matter and crude protein analyses were performed according to AOAC official methods 934.01 and 976.05¹⁹. The digestibility markers used in the experiments were determined for chromium oxide according to Fenton²⁰ and titanium oxide according to Myers²¹. The AA were determined by AOAC¹⁹ method No. 994.12, which consists of hydrolyzing the samples at 110 °C for 24 h in 6 mol/L HCl. Amino acid analysis was performed according to Henderson²² or Csapó²³. Energy analyses were carried out in an adiabatic calorimetric pump (1281, Parr, Moline, IL).

Procedure for obtaining samples in NIRS equipment

The samples were placed in a quartz cup and scanned with a Nicolet 6700 FT-IR spectrophotometer (Thermo Fisher Scientific, Inc); the spectra were generated with the OMNIC™ software. The spectra were obtained in the near infrared (NIR) region (1000-2500 nm); the data absorbance was expressed as the logarithm of the reciprocal of R "(log 1/R)", where R= reflectance. The spectra thus generated were fed with the values obtained in the laboratory. The multivariate statistical model used was partial least squares, the first or second derivative, and the Savitzky-Golay filters with the TQ Analyst v8™ software. The variables entered in TQ Analyst v8 were: dry matter, protein, energy, and amino acid content. Calibration was performed with 2/3 of the samples, and validation, with 1/3, either for stool sample or ileal content. The selection of the best predictive model was based on the minimization of the standard error of prediction (SEP) and of the standard error of cross-calibration (SECV), and the maximization of the coefficient of determination (R²). In order to evaluate The predictive power of the calibration model was assessed considering the index known as RPD, i.e., the predictive deviation ratio. The predictive power of the calibration is considered as follows: if the RPD is less than 1.5, the calibration is not useful; if the RPD is between 1.5 and 2.0, the calibration may distinguish between low and high values; if the RPD is between 2.0 and 2.5, the calibration has possibilities to make quantitative predictions; if the RPD is between 2.5 to 3.0, the calibration prediction is good, and if RPD is greater than 3.0, the calibration leads to excellent predictions (Williams, 2003) quoted by Saeys²⁴. Samples with differences of over 2.5 standard deviations were considered "outliers" and eliminated from the study.

Scanning of samples

Ileal digesta: A total of 119 ileal digesta samples were used. The mathematical analysis utilized was partial least square (PMS). The mathematical treatment was 1,4,13,6 (the numbers represent the degree of the derivative, the width of the interval over which the derivative is calculated, the smoothness level, and the order level of the polynomial).

Feces: A total of 222 samples were utilized. The statistical model used was the partial least square (PMS), and the mathematical treatment utilized was 1,4,11,5 (see explanation above).

Graphs. The graphs were made using the "ggplot2"(25) and "patchwork"(26) packages of R software²⁷.

Spectra

Representative spectra of ileal digesta and feces samples are shown in Figure 1.

Figure 1 Representative spectra of ileal digesta and feces samples 

Calibration for ileal digesta analysis

A total of 119 samples were used, of which 9 were eliminated because they were considered to be "outliers" according to the aforementioned criteria. Thus, the models were developed over 110 samples.

Protein. The mean and standard deviation (SD) for protein were 15.57 ± 2.15, ranging from 11.1 to 20.8 (Table 1). The coefficient of determination R² was 0.98, the calibration standard error (SEC), cross-validation standard error (SECV) and prediction standard error (SEP) were 0.330, 1.020 and 0.640, respectively. Its RPD was 2.11 and the SD/SEP ratio was 3.34 (Table 2, Figure 2A).

Table 2 reports the values of the equations

Figure 2 Relationship between the predicted values by NIRS and those determined in the laboratory for ileal digesta 

Leucine. The mean and the standard deviation were 0.85 ± 0.21, with a range of 0.33 to 1.31 (Table 1). The coefficient of determination R² was 0.95. The values for SEC were 0.040; for SECV, 0.123, and for SEP, 0.102. The results obtained for RPD were 1.70, and for the SD/SEP ratio, 2.07 (Table 2, Figure 2B).

Lysine. The mean and the standard deviation were 0.58 ± 0.29, with a range of 0.17 to 1.82. The coefficient of determination R² was 0.93. The values were 0.077 for SEC, 0.217 for SECV, 0.143 for SEP, and 1.31 for RPD, and the SD/SEP ratio was 2.00.

In order to improve the lysine prediction model, a second calibration was performed for lysine only. The SD/SEP and RPD statistics improved, and there was a minimal decrease (one percentage point) in R². The mean standard deviation, and the minimum and maximum values were the same as described in the previous paragraph; therefore, the coefficient of determination (R²) was 0.92. The results were 0.070 for SEC, 0.137 for SECV, and 0.115 for SEP. The result obtained for RPD was 2.08, and 2.48 for SD/SEP. (Table 2, Figure 2C).

Threonine. The mean and standard deviation for threonine were 0.71 ± 0.36, with a range of 0.32 to 2.13 (Table 1). The coefficient of determination R² was 0.67. The values of SEC, SECV and SEP were 0.209, 0.230 and 0.187, respectively. Its RPD was 1.57, and the SD/SEP ratio, 1.94 (Table 2, Figure 2D).

Calibration for fecal analysis

Of the 222 samples, 20 considered "outliers" were eliminated. Table 3 shows summary statistics generated by NIRS.

Protein. Its mean and standard deviation were 20.05 ± 6.65, respectively, with a range of 4.5 to 29.2 (Table 3). Its coefficient of determination R² was 0.98. The value of SEC was 0.950; that of SECV was 1.600, and that of SEP was 1.190. The statistics for RPD and SD/SEP were above 4, i.e., of 4.16 and 5.59, respectively (Table 4, Figure 3A).

Table 4 reports the values of the equations.

Figure 3 Relationship between the predicted values by the NIRS and those determined in the Laboratory for the feces 

Energy. The mean and standard deviation were 4,444 ± 269, ranging from 3,684 to 4,966 (Table 3). The coefficient of determination R² was 0.94. The results for errors were 60.8 for SEC, 106.3 for SECV, and 95.3 for SEP. The results obtained for RPD were 2.53, and 2.82 for SD/SEP (Table 4, Figure 3B).

Dry matter. The mean and standard deviation were 93.43 ± 2.48, with a range of 85.83 to 97.50 (Table 3). The coefficient of determination R² was 0.87; the values of SEC, SECV, and SEP were 0.830, 1.390, and 1.150, respectively. Its RPD was 1.78, and the SD/SEP ratio was 2.16 (Table 4, Figure 3C).

Protein and aminoacids

This is the first work in which the NIRS technology was used to predict protein and amino acid content in ileal digesta in pigs. The results obtained show that, based on the R² statistic = 0.97 and the RPD value of 2.11, the calibration for crude protein is acceptable for predicting ileal digesta composition. These values are in agreement with those indicated by Saeys²⁴, who report that the calibration to estimate pig excreta N had a R²= 0.89 and an RPD above 3.0, suggesting that NIRS can predict crude protein content. According to the RPD statistic (1.70, 1.31, and 1.57), the lysine and leucine concentration prediction is unreliable, but the calibration obtained can be used to estimate their concentration in the ileal digesta. However, in the case of threonine, the results were not favorable, as the calibration and prediction values obtained and the RPD (R²=0.67, SEP=1.87, and RPD=1.57) suggest that this is a poorly performing calibration and therefore it is not recommended. The low reliability of threonine prediction may be due to its richness in endogenous ileal protein losses, given that mucin is rich in threonine and mucin secretion is modulated by several dietary factors such as fiber and antinutritional factors^31,32. Thus, the concentration of threonine is more variable than that of the amino acids lysine and leucine; this argument is supported by the ability of NIRS to predict amino acid content in wheat (213 samples), barley (185 samples), and corn (258 samples)³³, since the variability in amino acid content in these samples is lower than in ileal digesta. The prediction was improved in the (second) calibration, in which only lysine was included; this was attributed to the fact that the correlations obtained were only for that amino acid ⸻a similar situation to that reported by Owens et al³⁴, who predicted more than 20 wheat traits and obtained good predictions for only a few of them, including protein. The results of the present work for predicting stool protein content were superior to or equal to those obtained in previous studies^35-38.

The range of crude protein values in pig feces (45 to 292 g/kg) is higher than that reported in previous works^35,36,38, due to the diets used in them; for example, in the study by Bastianelli³⁶, the samples came from a single experimental diet, whereas in the present study they came from several experiments and, consequently, from different diets. However, the greater variation in the data of the present work makes the resulting calibration more robust; this can be seen in the R², which is higher than those reported in those other works (0.98 vs 0.84, 0.88, and 0.89)^35,36,38) and indicates that the calibration was excellent. According to Saeys²⁴ calibration can be used to predict the crude protein content of the stools. This is reflected in the results of the present study, where the error for prediction was 11.9 g/kg, a lower value than those reported in other studies. In addition, the calibration is considered to have an RPD statistic value of 4.15, which is higher than the value of 3 recommended by Saeys²⁴, which indicates that the prediction is accurate and reliable.

Gross energy

The energy prediction is achieved thanks to the organic C-H bond absorption bands present in the scanned samples and corresponds to those bonds present in the carbohydrates, lipids, and protein of the food. Carbohydrates present in the feed constitute the main fraction of the pigs’ diet, and are classified into sugars (mono and disaccharides), oligosaccharides, and polysaccharides (starch and non-starch polysaccharides (NSP)). Most of the disaccharides and starch are digested and absorbed in the small intestine as shown by the work of Reis et al¹⁰; while the pig's digestive enzymes do not hydrolyze NSP, and therefore their degradation is carried out by the microflora of the cecum and colon through fermentation³⁹.

The dietary protein source varies in its amino acid content according to the pig's stage of production. For example, piglets consume mainly animal sources such as whey proteins, fish, animal plasma, etc., which are highly digestible. As the pig’s digestive system matures, vegetable proteins (mainly from soybean meal and other oilseeds) increase until only vegetable proteins are used, their digestibility being more variable than that of carbohydrates or fats⁴⁰. In ileal digesta samples, these bonds correspond mostly to the undigested bonds of the consumed feed and to the endogenous losses contributed by the animal's organism⁴¹. Substrates not absorbed in the small intestine reach the large intestine and are fermented by the microbiota or excreted in the feces⁴². In the case of amino acids, they can be used for microbial protein synthesis or fermented and used as a source of energy⁴³. Therefore, the C-H bonds in the feces correspond mainly to the bonds in the microbiota and the fermentation products produced by it.

The results obtained in the calibration to predict energy were excellent; the R² was 0.94, which indicates an excellent calibration that can be used in the prediction of the energy content of feces²⁴. The results of the present study are consistent with other studies^37,38. In addition, the robustness of the energy prediction (RPD of 2.52) means that quantitative predictions are possible.

The results regarding the predictive ability of stool protein and energy content are similar to those of Cruz-Conesa⁴⁴, whose objective was to develop predictive models to estimate the chemical composition of the feces of three different poultry species: broilers, layers, and turkeys. These authors⁴⁴ reported that their calibrations accounted for over 93 % of the variation, and the RPDs were greater than 3.7, concluding that NIRS is capable of estimating feces' protein and energy content.

Dry matter

The chemical composition of the samples is affected by the collection site (ileal or fecal), since, although most of the water is reabsorbed in the small intestine, the large intestine is where the digesta is dehydrated⁴⁵; therefore, the feces had a higher dry matter content.

The moisture content of the samples has a direct impact on the generation of the spectrum, as water increases the absorption of light producing a higher absorption in the whole spectrum, on the other hand, the solid particles cause the light to be reflected and thus to reduce the absorption and generate higher peaks. For example, the region between 1,450 nm and 1,900 nm corresponds to the O-H chemical bonds present in water; consequently, there are more pronounced peaks in this region.

The coefficient of determination of the calibration for dry matter R² was 0.87, which indicates that the calibration is accurate and reliable for determining the dry matter content in ileal samples according to Saeys²⁴, with a calibration error of 8.3 g/kg DM. On the other hand, the prediction error was 11.5 g/kg, and the RPD was 1.78; this value indicates that a quantitative approximation of the dry matter in the feces samples is possible²⁴. The relatively low prediction values obtained for dry matter compared to protein and energy values result from the fact that the scanned samples were from experiments that had previously been analyzed for digestibility, and, therefore, the samples had been stored. Although the samples were stored correctly, they may have become rehydrated with ambient moisture. This argument is supported by the work of Garnsworthy et al⁴⁶, who evaluated 156 wheat grain samples and compared the dry matter values obtained by NIRS on the previously dehydrated samples versus the dry matter values obtained on the samples that were dehydrated at the time of scanning, lower values were observed in the previously dehydrated samples. Another factor is that the variation of the laboratory analysis of the dry matter was narrow (2.48 %); this small variation means that, when the samples are hydrated, they easily fall out of the calibrated range, affecting the results.