A model that predicts resistant starch in a dog kibble using a small-scale twin-screw extruder

The objective of this work was to modify extrusion parameters to yield greater resistant starch (RS) in a kibble and create a model to predict its concentration. A dog food was extruded through a small-scale twin-screw extruder as a central composite design with 6 central points (replicates) and 14 single replicates. There were three factors tested at three levels: corn particle size, extruder shaft speed, and in-barrel moisture (IBM). The remaining processing inputs were kept constant. Chemical and physical starch analyses were performed. A model to predict RS was created using the REG procedure from SAS. Pearson correlations between extrusion parameters and starch analyses were conducted. A model to predict RS was created (R2adj= 0.834; P < .0001). Both SME and extrudate temperature had a high negative correlation with RS and RVA raw starch. Results suggest that low mechanical energy and high IBM increase kibble RS.

and one bar of air pressure for dispersion of the sample. 100 Particle size distribution of the dry mix and corn samples were determined by rotating 101 tapping sieve analyses (Ro-tap) (11). This procedure involved 100 g sample and rotating-tapping   Fig 1). This system utilized a 25 L dual cylinder preconditioner with 127 Corn was ground through a hammermill at three screen sizes: 0.793 mm, 1.19 mm and 128 1.586 mm in order to produce a fine, medium and coarse grind size. In sequence, ground corn 129 was mixed with the other ingredients of the "dry mix" (Table 1)  Extruder data for input and output variables were recorded every 5 minutes (twice) 135 during each sample production and then averaged ( Table 3). The dry mix was delivered to the 136 PC at a constant feed rate of 30 kg/h with a PC shaft speed set to 100 rpm during the whole 137 experiment. The PC steam flow rate was set to 4.5 kg/h, while PC water was varied from 1.5 to

Processing details and data collection
Where: ρd = extrudate density inside the die; Md= moisture content of the extrudate in 185 the die; ρe= apparent density of the wet kibble; and Me= moisture content of the wet kibble.

186
Moisture content inside the die (Md) equaled IBM, while moisture content of the extrudate after 187 exiting the die (Me) was calculated as IBM minus steam loss. Steam loss was estimated 188 according to (13). Density of the kibble inside the die (ρd) was calculated using a model (14).

189
Lastly, longitudinal expansion index (LEI) was calculated as a function of VEI divided by SEI.   The median circle equivalent (CE) diameter distribution measured by the Morphologi G3 233 for the three ground corn samples were between 3.6 and 5.5 times lower than the sieve sizes used 234 to grind each corn (Table 2), and high sensitivity (HS) circularity, aspect ratio and elongation 235 were numerically similar within ground corn and dry mixes. Particle size distributions were 236 bimodal for corn ground at the three levels (Fig 1). Corn ground using a 0.793, 1. Dry mix 1, 2 and 3 were the same diet recipe mixed with corn 1, 2 and 3, respectively. 243 1 CE= circle equivalent; (v, 0.5) = median of the volume distribution. The volume is calculated by converting average 3D image of the particles to a 244 2D circle of equivalent area. 245 2 HS= high sensitivity circularity; (n, 0.5) = median of the number distribution. Calculated as 4πA/P 2 ; where A is the particle area, and P its 246 perimeter. Measurements ranged from 0 to 1 and described the deviation from a perfect circle (the closer to 1, the more it approximates to a perfect 247 circle). 248 3 Aspect-Ratio-Number Distribution; (n, 0.5) = median of the average width/length of the particles. It ranges from 0 to1. A low aspect-ratio would 249 assume a rod shape. 250 4 Elongation= average particle elongation (the closer to 1, the longer it is). Calculated as 1 -aspect-ratio (ranges from 0 to 1).    Table 4).

303
Resistant starch was less than 50% negatively correlated (P < 0.05) with starch cook and 304 RVA cooked AUC, and also less than 50% positively correlated (P < 0.05) with RVA 305 raw:cooked AUC and setback viscosity ( in-barrel moisture; SME= specific mechanical energy; PS= particle size. Resistant starch had a strong negative correlation with dough temperature at the end of the barrel, shaft speed and SME (r= -0.829, -0.724 and -0.862, respectively), and a less strong positive correlation with IBM (r= 0.48; Table 6). The regression equations of interest follow below: Starch cook had a low correlation with all parameters tested but was significant (P < 0.05) and positively correlated with dough temperature and SME and negatively correlated with IBM. The RVA AUC of cooked and raw starches were mostly affected by dough temperature and SME, but not IBM. Lastly, setback viscosity had a strong negative correlation with dough temperature and SME. Due to difficulties visualizing a 3D surface response plot using 3 independent variables (PS, SS, and IBM), two variables were plotted against RS for each (Figs 3-5).

Surface response plots and model building
The final model to predict viscosity (RVA) with endpoints cooked peak and raw:cooked ratio was not significant (P > 0.05). The model to predict RVA raw peak AUC was significant (P =

Discussion
To date, the present study is the first to determine the effect of pet food processing parameters on RS yield and to create a model that would predict RS content in a dry extruded kibble. A secondary goal was to explore methods of starch gelatinization, correlate these among themselves, as well as correlate extrusion processing parameters with starch transformation measures and kibble parameters.
The RS targeted in the present study was likely a combination of types II and III (16,17).
Recently, a broader classification was suggested which identifies this type of RS as starches not digestible by enzymes due to their tightly packed crystalline structure (18) which is present in raw starches (19).
The ideal concentration of RS that benefits dog colonic health without decreasing stool quality still hasn't been established for all breeds. Although it is known that large dogs may produce loose stools when fed a diet with RS as low as 2.5% (8). In the present study the raw dry mix with the most RS (dry mix 3) had only 2.2% RS. Thus, the maximum RS in a treatment could not have been over 2.2%. Ideally, we would extrude a dog food with the minimal energy necessary to destroy antinutritional factors and pathogens, while minimizing starch cook. This is challenging since extruded pet foods need hydration as well as thermal and mechanical energies to form an expanded kibble.
Starch granules of cereals like corn, wheat or rice have an X-ray diffraction type A and also contain pores and channels that facilitate alpha-amylase adhesion and digestion of the substrate (18,20,21). Conversely, tubers have large smooth starch granules with less enzyme adhesion sites, and legume starches are trapped within cotyledon cells, which are little disrupted after thermal processing (22). Although tubers and legumes tend to be higher in naturally occurring RS than cereals (23), corn was chosen as the sole starch ingredient for two reasons: 1.) it is one of the most common cereal grains used in pet food due to its high apparent total tract digestibility coefficients, palatability and low cost. 2.) by choosing one starch source and not adding any fiber ingredients the effect of recipe would be minimized and the focus would be on the effects of processing on RS yield. Moreover, whole ground corn was selected instead of corn flour because it has lower in vitro starch digestibility when compared to fine milled flours (24).
It is well known that raw cereals, legumes or tubers possess a greater amount of RS than their cooked forms (25,26), but thermal processing is usually required to destroy antinutritional factors (27), increase acceptability (25), and microbiological safety of products. Thermal and mechanical processing with some moisture gelatinizes the starch making the α-glucan chains less ordered and more available for enzymatic adsorption and digestion (28,29). The RS of kibbles in the present study had a high and inverse correlation with dough temperature at the end of the extruder barrel confirming that the lowest thermomechanical energy from the process resulted in the most RS. There are many processing inputs that contribute to changes in extrudate temperature such as feed rate, steam and water additions, extruder shaft speed, and die open area.
When corn is extensively ground, starch gelatinization and digestion are improved due to a large surface area to mass of the small particles. This happens because the starch ground to smaller particles becomes more exposed to hydration, which increases gelatinization.
Conversely, coarsely ground corn has a lower surface area to mass which slows water penetration and starch gelatinization, thereby decreasing its digestion (4,5). Grain milling also destroys cell walls and disrupts the protein matrix making starch more available (30). For these reasons, an important factor selected to construct the surface response model for RS in dry extruded kibbles was the corn grind size. In the present study, the largest mean geometric diameter was less than initially targeted. This likely happened because the hammermill used was more effective than expected, and the corn (and the ration it was contained within) was ground through the hammermill with a 1.59 mm screen size twice. A better model may have been created if wider differences among mean geometric diameter had resulted. In future work using a larger screen size would be recommended. (5) and (4) reported that extruded dog food produced with raw corn ground to geometric mean diameter of 224 (low) and 312 μm (high) and at high and low SME inputs yielded 0.21-0.22 and 1.46-1.54% RS, respectively. In their work the lowest geometric mean was greater than the grain-mix coarsely ground in the present study. Moreover, in their studies the SME was controlled by differing the die open area rather than altering screw speed like in the present work, and both methods were effective in controlling SME. (31) reported that fine, medium and coarse maize (360, 452 and 619 μm mean geometric diameter, respectively) produced foods with a starch gelatinization of 79.9, 73.8 and 63.2%, respectively.
The starch cook values they obtained were wider than ours due to the greater differences in particle size. While RS was not measured in that work (31), it would likely be above what was determined in the present study. Nevertheless, particle size still had a significant effect on RS concentration when plotted against shaft speed and in-barrel moisture; wherein, the RS prediction was the greatest at the largest mean geometric diameter.
Water, which is required for starch gelatinization, has a quadratic effect on starch cook: too little water does not provide enough hydration of the starch granule and subsequently the kibble will have small cell structure and little expansion (3). On the other extreme, too much water decreases temperature in the extruder and therefore starch cook, acting as a plasticizer (32). When water is below what is required to hydrate the dry mix, the mechanical shear inside the extruder barrel can cause starch damage which will create a premature RVA peak in cold water, meaning that starch has been mechanically damaged. This phenomenon was present in the RVA profile of sample 16 (Fig 2). (3) demonstrated that 22% and 37% IBM were the extremes because neither resulted in kibble expansion or to have a good cell structure according to scanning electron micrographs. (33) reported that extruded wheat flour with the most RS was produced with excess water. At the other end of the spectrum, (34) reported that corn starch extruded at both high and low SS (600 and 300 rpm, respectively) at lower moisture content (17% water) yielded the most RS. In their study, the low water content was likely not enough to cook the starch, and the treatment with more water (at 22%) caused greater starch gelatinization.
The treatment in the present study with high IBM, low SS and high PS yielded the greatest predicted RS content. The IBM at 36% likely helped to dissipate energy and lower starch cook.
The process with higher moisture could also have created an environment to develop retrograded starch (RS type III) (33).
Extruder shaft speed (SS) is a controllable input that affects specific mechanical energy directly. Simultaneous to a higher SS and increased mechanical energy in the process the residence time decreases. This exposes the starch to a shorter cook time in the extruder barrel.
Therefore, altering shaft speed can have mixed results. (35) reported that RS content of corn and mango were higher when extruded at a SS of 30 rpm as compared to 65 rpm. (33) did not find a difference in RS of wheat flour extruded at the same moisture contents and different shaft speeds. This is likely because they increased SS at 50 rpm increments, which was not a significant change to affect RS yield. In the present study, increasing the shaft speed in 400 rpm increments led to significant cross-product effects with particle size on RS yield.
Among methods used to determine the degree of starch gelatinization starch cook was the least consistent. This happened because even raw corn possesses pores and channels that facilitate enzyme adhesion and digestion (18). Thus, the starch cook method in corn does not account for only the portion that was gelatinized. A better method to estimate starch gelatinization in this study was RVA as it correlated highly with dough temperature and SME.
The RVA provides a broader characterization of starch transformations during extrusion. With cold water raw starch does not swell and presents a low viscosity, while high molecular weight starch derivatives produced from chain scission during extrusion easily swell and increase viscosity (36). This was well illustrated in the RVA plots (Fig 2) where the treatment with the highest thermomechanical energy had a significant initial cold swelling and the less cooked sample had no cold swelling. As the RVA temperature increased to the gelatinization range of corn starch (mid 60ºC) (37) the sample that preserved some raw starch had an increase in relative viscosity, while the treatment that suffered more thermomechanical energy had little to no raw starch left to promote a hot viscosity peak. Samples 15 and 16 that were used as examples of RVA plots also had the highest and lowest IBM, respectively. The first sample had a much higher setback viscosity than the latter. According to (38), extruded puffs produced with high and low water contents had similar final viscosity as samples from the present study, which meant that less water led to more mechanical shear that dextrinized the starch, reducing final viscosity. Extrusion causes mechanical disruption of molecular bonds within the starch granule, resulting in loss of crystallinity and gelatinization (39,40). The RVA method to measure cooked and raw starch should be employed more frequently in pet foods, but it is important that the same protocol is used so that results can be comparable across studies.
Extrudate temperature is a result of thermomechanical energy inputs during the process.
Mechanical energy is largely affected by viscosity, which is affected by water content. The driving force for bubble expansion in the kibble can be represented as a function of water vapor pressure inside the vapor bubble and viscosity (specific volume of extrudate = P vs /η, where P vs is the water vapor pressure and η the molten viscosity) (41). In the present study both VEI and LEI had a high negative correlation with moisture content. This meant that the dough at 30% IBM likely had a lower viscosity compared to treatments produced at 35% IBM, which allowed for bubble expansion according to (41). Extrudates with high IBM won't absorb as much mechanical energy and therefore won't be as hot, so vapor pressure decreases (42), decreasing overall expansion. Moreover, a wetter extrudate takes a longer time to drop below glass transition temperature (Tg) and the shrinkage effect is greater (42).

Conclusion
This study was successful at describing the effect processing has on starch gelatinization and RS content in the kibble. Results suggested that a higher IBM, lower extruder shaft speed and larger particle size should contribute to the survival or development of RS during thermomechanical processing. Higher water content and lower extruder shaft speed lower the dough viscosity which directly affects SME. Resistant starch had large negative correlations with dough temperature and SME, which means that extrusion should target a low thermomechanical to increase RS yield. This strengthens the RS prediction model built in this study since SME is a function of multiple factors such as viscosity, and treatment inputs extruder shaft speed, water content and particle size. The physical method RVA to characterize starch gelatinization was preferred over the enzymatic starch cook as it had a strong correlation with thermomechanical parameters. The model created to predict RS can be used as a platform for future studies. Future work should focus on improving this model by using a wider range of corn particle sizes.

Declaration of Interest
analysis. CW did the experimental design and oversaw the extrusion processing. CGA and Lewis C. Keller (LCK) edited and provided intellectual contribution to the work.