Physicochemical property characterization, amino acid profiling and sensory evaluation of plant-based ice cream incorporated with soy, pea and milk proteins

This study examined the effects of incorporating milk protein concentrate (MPC), pea or soy proteins isolates (PPI and SPI) on the physicochemical, sensorial properties, and amino acid composition of ice creams containing 7% protein, in comparison to dairy ice cream as a reference. As protein ingredients, PPI exhibited higher water and oil holding capacity but lower surface hydrophobicity than SPI and MPC. Viscosity of the mixes were proportional to the firmness of ice cream, and both were highest with use of PPI. MPC ice cream had most similar physical and sensory properties to reference. PPI and SPI ice cream mixes showed higher extent of fat coalescence than MPC and reference. PPI and SPI conferred structural stability to ice cream with lower melting rate and better shape retention, and ability to delay ice recrystallization during temperature flocculation as compared with SMP and MPC. Confocal laser scanning microscope images indicated that higher extent of protein aggregation and more air cells were found in PPI ice cream. Sensory and amino acid profile results revealed that PPI and SPI ice creams were inferior in taste, texture, and essential amino acids like methionine. This study offers insights for the development of high protein frozen desserts.


INTRODUCTION
Ice cream is a popular dairy-based frozen dessert widely consumed, with global market share of US $82 billion in 2022.Driven by its popularity and consumers' demand for healthier products, this led to the trend of using ice cream as a vehicle to deliver health-related com-pounds or nutrition (Soukoulis et al., 2014;Spence et al., 2019;Pimental et al., 2021).Beyond that, high-protein ice cream is already present in the market, with positive consumer outlook (Teixeira et al., 2023).
Proteins in ice cream perform an important role in not only forming and stabilizing the foam and lipid emulsion, but also its subsequent limited desorption from the lipid interface for partial coalescence of fat globules (Goff, 2006;Dickinson, 2003).While increasing protein content could boost its nutritional value, this affects the formation of partially coalesced fat during freezing, and hence the structure and texture of ice cream (Daw & Hartel, 2015).Additionally, in conventional ice cream, the principal source of dairy protein is from skim milk powder (SMP).However, there will be presence of off-flavors and increased lactose contents in ice cream when higher amount of SMP applied (Roy et al., 2022).Alternatives such as milk and whey protein concentrates are commonly used to boost protein content, in which different sources of proteins were applied in ice cream formulation to either replace or were added on top of SMP (Patel et al., 2006;Daw & Hartel, 2015;Roy et al. 2022).All of which would have different effects on the emulsion stability and propensity for partial coalescence, resulting in different ice cream textural properties (Dickinson, 2003).
Plant-based frozen desserts are gaining popularity in recent years with consumers (Shoup, 2018).Beyond health reasons, plant-based alternatives address consumers' concerns about animal mistreatment and have a lower carbon footprint (McCarthy et al., 2017;Mintel, 2020).Soy and pea proteins are most common plant-based proteins in the market with excellent properties of water holding, gelling, fat absorbing and emulsifying capacities (Ismail et al., 2020).However, concerns are raised for their lower protein quality and anabolic capacity due to unbalanced amino acid composition, undesirable tastes, off-flavors, and reduced protein digestibility due to antinutritional factors (Gorissen et al., 2018;Nasrabadi et Physicochemical property characterization, amino acid profiling and sensory evaluation of plant-based ice cream incorporated with soy, pea and milk proteins Cheryl Kwoek Zhen Ng, 1 Wei Qi Leng, 1 Churn Hian Lim, 1 and Juan Du123* al., 2021).On the other hand, animal proteins (meat, dairy, egg) not only contain the full range of 9 essential amino acids vital for body needs but also have a high digestibility (>95%) (Qin et al., 2022).As a result, the usage of plant proteins at higher levels might not provide the same amounts of essential amount acids as ice cream made using dairy proteins.In addition, plant proteins differ greatly in structure and functionality from dairy proteins, causing challenges in development of plant-based ice creams with outstanding taste and texture comparable to dairy ice creams (Akesowan, 2009;Pereira, Resende, Abreu, Oliveira Giarola, & Perrone, 2011).
There are limited studies on high-protein ice cream formulated using plant proteins.Hence, in this study, PPI, SPI and MPC 80 were selected due to their availability and functional properties, to investigate the effects of enriching protein from 4% to 7%, namely soy, pea, and milk on the functionality, physicochemical, sensorial properties, and amino acid profile of ice cream, in comparison to dairy ice cream formulated with SMP and milk fat at 4% protein as a benchmark.
Preparation of ice cream.Four ice cream mixes and frozen samples were produced based on formulation indicated in Table S1.Reference represents a typical commercial dairy ice cream, which is used for comparison.Protein enriched ice creams formulated with MPC 80, PPI and SPI with crude protein content of 7.30 ± 0.26% determined by Kjeldahl method (AOAC, 2005), were labeled as MPC, PPI and SPI respectively in this study.
The ice cream processing method was modified from Lipsch & Van Schie (1998).Dry blend of sucrose, dextrose, and emulsifier-stabilizer blend were dissolved in water at 80°C and mixed for 10 min using a Silverson mixer (L5M-A, Silverson Machines Ltd., Chesham, UK) at 2000 rpm.SMP, MPC, PPI and SPI from each formulation were hydrated for 30 min at 50°C, with a gradual increase in speed to 6000 rpm.The melted fat was added and mixed for another 10 min.The mix was pre-heated to 65°C, upstream homogenized at 115/15 bar (Panda Plus Homogenizer, GEA Niro, Italy), followed by continuous high-temperature short time (HTST) pasteurization (HT220 series Lab HTST/ UHT Machine, OMVE, Netherlands) at 80°C with a holding time of 15 s.Subsequently, the mix was cooled to 7°C and aged for 24 h at 4°C.Flavors were added into the ice cream mix and churned using a batch ice cream freezer (Taylor, Model 104, Rockton, Illinois).The churning time was controlled between 8 to 10 min to achieve target draw temperature of -5.5 ± 0.5°C and target overrun of 45 -50%.Ice cream was filled into 3.5 Oz ice cream pints and plastic cups and immediately stored in blast freezer (Friulinox, Model BF121AG, UK) at -40°C for at least 2 h for hardening followed by storing in the freezer at -18°C for 24 h before performing physicochemical and sensory analysis.The total solid content of all samples controlled at 36.1 ± 1.55%, and the final pH of ice cream mixes were 6.5 ± 0.3.

Water holding capacity (WHC) and oil holding capacity (OHC)
The WHC and OHC determination method of the protein powders was performed according to Zhao, Chen, Bi, & Du, (2023) with modifications and were calculated using Equation (1) below:

( ) ( )
For WHC analysis, 1 g of protein powder was added into a pre-weighed 50 mL centrifuge tube, followed by addition of 10 mL distilled water.The samples were then vortexed (Vortex-Genie 2; Scientific Industries, New York, USA) for 1 min and left aside to stand for 30 min.The samples were centrifugated at 4000 x g at 25°C for 25 min.After centrifugation, the supernatant for all the samples were decanted, and the tubes were left to drain for 15 min before the residues were weighted.Similar procedures were performed for OHC analysis, where 10 mL of canola oil was added to 1 g of protein powder instead.
Surface hydrophobicity Surface hydrophobicity (H 0 ) of PPI, SPI and MPC were determined using method according to Asen & Aluko (2022).The stock mixture for each protein was prepared using 10 mg/mL in 10 mM phosphate buffer, pH 7.0.Fluorescence intensity (FI) measurement at excitation and emission wavelengths of 390 and 470 nm respectively (Biotek Synergy HTX Microplate Reader, Agilent Technologies, USA).The surface hydrophobicity was calculated as the slope of a plot of the FI against protein concentration.
Particle size distribution Particle size of the protein powders (PPI, SPI & MPC) in 2% wt/wt distilled water were determined with slight method modification from Binte Abdul Halim et al., (2023).The refractive index was set at 1.46 for PPI, and 1.45 for both MPC and SPI.Measurement was conducted within range of 5 to 12% obscuration values, at stirrer speed of 1500 rpm.Size measurements of d [4,3] and d [3,2] , D 10 , D 90 and D 50 were recorded.

Characterization of ice cream mix pH, total solid content and apparent viscosity
The pH and total solid content of ice cream mix were analyzed after aging at 4°C for 24 h, whereas apparent viscosity of ice cream mixes before and after aging with reference to Hashim & Shamsi (2016).The pH values were measured using a digital pH meter (FiveGo pH meter F2-Field-Kit, Mettler Toledo, Shanghai, China).Total solid content was measured using a pocket digital refractometer (Pocket Refractometer PAL-3, Atago®, Japan).The ice cream mix with 250 mL sample size was stirred for 50 rounds within 30 s before viscosity measurement at 10°C using a viscometer (DV2T Viscometer, AMETEK® Brookfield, Middleboro, MA, USA) with spindle LV64, measured for 1 min at 30 rpm.
Fat droplet size distribution Ice cream mixes, before aging, were analyzed for fat particle size distribution with slight method modification from Binte Abdul Halim et al., (2023).The refractive index for dispersed phase (fat) and dispersant (DI water) was set to 1.46 and 1.33 respectively.The ice cream mix was diluted 10 times with deionized water and the measurement was conducted with the range of 5 to 12% obscuration values, at stirrer speed of 1500 rpm.

Characterization of ice cream
Overrun Overrun was measured by weighing ice cream mix and the same volume of ice cream using the same container to determine the amount of air incorporated into the ice cream upon churning.For each batch, the overrun was taken for all samples throughout the ice cream production, with target overrun of 45 to 50%.
Texture analysis (TA) Firmness (peak force during penetration) of ice cream was measured according to method modified from Rolon, Bakke, Coupland, Hayes & Roberts (2017), using a texture analyzer (Texture Analyzer TA.XT plus, Stable Micro Systems, UK), equipped with a 30 kg load cell and a 3 mm cylindrical probe (P/3).A 5 kg standard weight load and height of 70 mm was calibrated prior measurement.The pre-test, test and post-test speed was set to 2 mm/s, 1 mm/s and 1 mm/s respectively and sample was compressed at distance of 10 mm with a trigger force of 5 g.Samples stored at -20°C, were taken out from a portable freezer placed beside the equipment sequentially and quickly placed on the center of the metal base of equipment, held at room temperature.Analysis was completed within 30 s to minimize variability due to sample warming.Six measurements were taken per sample to ensure precision and accuracy.
Meltdown test Melting rates were determined with method modified from Yeon et al. (2017).Ice cream sample stored at -18°C was taken out and placed on a stainless-steel mesh (2 mm x 2 mm square hole, 0.1 mm wire diameter) at ambient temperature (25 ± 1°C) over 100 min.The mesh was placed above a tray and the dripped weight of the ice cream was recorded over an analytical balance (Sartorius Electronic Balance Practum, Germany) once every 10 min for the first 60 min, followed by once every 20 min for the next 40 min.The melting behavior is plotted in a graph, expressed as the amount of melted ice cream as a function of time.
Ice crystal size distribution Ice crystal size distribution of samples obtained from section 2.2 was determined according to the method described by Lomolino et al. (2020).ECLIPSE Ci-L upright light microscope (Nikon, Tokyo, Japan), equipped with a PE120 Peltier Heating and Freezing Stage (Linkam Scientific Instruments, UK) and DS-Ri2 camera (Nikon, Tokyo, Japan) was used.All materials and solvents were kept at −20°C to avoid thermal shocks.Ten mg of ice cream samples (stored at −19 ± 2°C) were obtained 2 cm from the surface, and was spread on the slide surface.A drop of n-butanol was added at −20°C to the sample and covered with a coverslip to avoid condensation.Images were captured at about 40 x magnification for up to 2 min.The sizes of the initial ice crystals were then measured was defined as sample of t 0 .Ice cream samples were subjected to temperature fluctuation treatment for 4 h, inclusive of 2 cycles in total, with method modified from Lomolino et al. (2020).A sample of ice cream was removed from the freezer (−19 ± 2°C) then subject to temperature cycle of -6°C for 1 h and -18°C for 1 h, with a heating rate of 0.2°C/min.Images were captured at about 40 x magnification for up to 2 min after each cycle.The sizes of the ice crystals were manually measured.Ice crystal size measured after 1 and 2 cycles of heat shock were defined as t 1 and t 2 respectively.
Shrinkage test after heat shock treatment Heat shock treatment was conducted with 3 cycles for each sample in triplicates.The ice cream was removed from the -18°C freezer and tempered at ambient temperature (25 ± 1°C) for 1 h followed by re-freezing back into the freezer.The steps were repeated for 3 times to obtain 3 heat shock cycles.Visual appearance was observed by taking image of the ice cream (top surface view) before and after heat shock.Shrinkage test was carried out by volumetric measurement.A 40% brine (sodium chloride) solution was prepared and stored at -18°C for at least one day to ensure equilibrium temperature as the ice cream samples and prevent sample from melting.The cold brine solution is poured onto the surface of the heat-shocked samples, filled to top of the brim.The results are expressed in % shrinkage, calculated as follows in Equation ( 2 (2)

Confocal laser scanning microscopy (CLSM)
This method was modified from Voronin, Ning, Coupland, Roberts & Hartel (2021).The ice cream samples were left to melt in 4°C refrigerator for 24 h.The distribution of proteins and fats in the melted ice cream samples was assessed under a confocal laser scanning microscope (A1R+si DUVB, Nikon, Tokyo, Japan) in fluorescence mode.Fast Green FCF (30 µL, 0.1 mg/mL in ethanol) and Nile Red (30 µL, 0.1 mg/mL in ethanol) were mixed into an aliquot (2 mL) of each sample.Samples were observed at excitation wavelengths of 488 nm and 633 nm respectively.All the images were recorded at 20 x magnification.

Amino acid analysis
Ice cream samples were tested for its amino acid profile for the 9 essential amino acids (EAA) and non-essential amino acids using method modified from AOAC 994.12 (2005).The results of amino acid profile were expressed in % EAA.

Sensory evaluation of ice cream
The ice cream samples were evaluated by 25 semitrained panelists (aged 18 to 55 years old, 20 women and 5 men) using acceptance test.Recruited participants were required to have some experience or familiarity with ice cream, and they are mostly research professionals who work with dairy ingredients on a daily basis or students in related fields.The 3-digit random coded samples were stored in −18°C freezer and the serving order for panelists was reference, followed by MPC, PPI and lastly SPI samples.The acceptance degree for several attributes; firmness, creaminess, iciness, sweetness, grittiness, mouthcoating and aftertaste were rated by the panelists using a 9-point hedonic scale shown in Table S2.A rating for the overall acceptability of ice cream samples made using MPC, PPI and SPI were also given.

Statistical analysis
All tests were performed in duplicates or triplicates and reported as mean ± standard deviation (SD).Statistical analysis was performed on Minitab software (Minit-ab®19, MiniTab Inc., State College, PA, USA).One-way ANOVA followed by Tukey pairwise comparisons at 95% confidence level (P < 0.05) was used to compare the mean values of the physicochemical parameters and hedonic scores between different ice cream formulations.

Water and oil holding capacities of protein powders
As seen in Figure 1, PPI had significantly higher WHC as compared with SPI and MPC at neutral pH of 7. The tendency of PPI and SPI to have higher WHC could be attributed to presence of polar and charged amino acid groups exposed that enables binding of water molecules (Ma, Grossmann, Nolden, McClements, & Kinchla, 2022;Schnepf, 1992).As for MPC, the low WHC could be attributed to high surface hydrophobicity as observed in Figure 1, thus limiting protein-water interactions.The WHC values obtained for MPC in our study was similar to other publication (Banach, 2012), however the WHC values obtained for PPI and SPI were lower than other studies in which WHC of PPI was about 5 g/g and for SPI, it falls in the range of 4.5 to 7.5 g/g (Liang, et al., 2022;Ma, et al., 2022).This observation might be because the protein powders were obtained from different commercial sources, and different processing conditions were used during protein extraction (Stone, Karalash, Tyler, Warkentin, & Nickerson, 2015).The OHC of all the samples were lower than the WHC.This is expected as the protein tested contains hydrophilic regions exposed on the surfaces that contain polar amino acid residues.PPI had the highest OHC, followed by SPI and MPC.OHC is generally caused by binding of non-polar side groups of proteins (Ma, Grossmann, Nolden, McClements, & Kinchla, 2022).The results obtained were not in line whereby PPI having significantly higher OHC despite having the least percentage of non-polar amino acid groups and lowest surface hydrophobicity.However, OHC is also affected by other factors such as matrix structure of the protein, type, distribution and stability of lipids present (Yang, et al., 2023).The protein extraction method used could have induced protein unfolding and caused structural changes, which influences the functionality of the proteins (Yang , Zamani, Liang, & Chen, 2021).

Surface hydrophobicity of protein powders
With reference to Figure 1, MPC had significantly higher H 0 (173 ± 0.59) than SPI (109 ± 5.42) followed by PPI (49.6 ± 3.09) at pH 7.0.The high H 0 of MPC  suggests that more hydrophobic regions were exposed to the surface.The results obtained for H 0 were higher than other studies for both SPI (Tang , Roos, & Miao, 2023) and MPC (Banach, Lin, & Lamsal, 2013) but opposite was observed for PPI as compared with a study by Tang, Roos, & Miao (2023) where H 0 value was about 100.A potential reasoning for the above variation could be attributed to different protein extraction procedure used as the proteins were provided by different suppliers, which denatured the protein more and therefore exposed the non-polar region to the surface of the proteins (Mune & Sogi, 2015).

Particle size of protein powder suspension
Particle size and distribution profiles of protein suspensions are shown in Table 1and Figure S1A.The d [4,3] values of dry MPC, PPI and SPI powders were 71.2 ± 5.9 µm, 156.0 ± 7.8 µm and 53.4 ± 1.6 µm respectively.Table 1showed that the sizes of the particles in PPI, SPI and MPC suspensions ranged between 1 to 2000 µm, with all suspensions displaying monomodal distribution.The mean particle diameter and particle size values at different cut-off all indicated that of particles suspended in PPI solution were significantly bigger than SPI and MPC.Overall, these results were agreeable to those published by other authors (Peng, Kersten, Kyriakopoulou, & Goot, 2020;McSweeney, Maidannyk, Montgomery, O'Mahony, & McCarthy, 2020).Comparing the particle size of proteins before and after water suspension, there was observable increase in size which could be accounted for by powder particle swelling caused by water uptake and hydration.The results could also indirectly show that PPI displayed better solubility properties in water as compared with SPI and MPC (Tang , Roos, & Miao, 2023).

Viscosity, overrun, solid content and firmness of ice cream
As shown in Table 2, viscosity of protein enriched ice cream mixes formulated with PPI and SPC were significantly higher than reference and MPC.The increase in viscosity for SPI and PPI ice cream could be attributed to the ability of the proteins to form a more stable gel matrix due to their water-binding properties (Boye, Zare, & Pletch, 2009).A correlation between WHC and viscosity could be observed whereby ice cream formulated with PPI had the highest WHC and highest viscosity among the  4 ice cream samples.Moreover, proteins in ice cream are usually adsorbed at the fat and air interfaces to provide stability.This suggests that proteins that are not adsorbed to the air interface becomes concentrated in the serum phase, which in turn contributes to higher viscosity (Goff & Hartel, 2012).However, viscosity of ice cream mix generally falls in the range of 0.1 to 0.8 Pa.s after aging.The high viscosity of PPI and SPI samples might cause processibility issues, especially in a manufacturing setting.
Overrun of ice cream samples ranged from 43.1% to 47.9%.No significant differences were observed across the 4 ice cream formulations indicating that the air content in the final product are similar.Ice cream enriched with PPI had the highest firmness, followed by SPI, and lastly, MPC and reference.The high firmness obtained for ice cream formulated with PPI and SPI were also likely attributed to their high WHC and viscosity, which formed more rigid and tight structures, thus giving a harder texture (Mutlu, Avkan, & Akin, 2021).

Fat droplet size distribution
The development of fat coalescence is commonly determined by d [4,3] and particle size distribution (Chen, et al., 2019;Méndez-Velasco & Goff, 2012).In Figure S1B, reference exhibited bimodal particle distribution while other samples had polydisperse distribution with various peaks.The results showed that sample with MPC, PPI and SPI behaved more similarly due to the same type of fat blend used.However, it can be assumed that peaks in the range of 50 to 1000 µm were attributed to the protein particles, and not the fat globules when referred to section 3.3.At protein concentration of 7.34 ± 0.28%, fat droplet size followed the order of PPI > SPI > MPC (Figure S1C & D).The results from this study also show that d [4,3] values were much higher than d [3,2] values, which indicates that the PPI and SPI mix contained a wide range of droplets that vary in sizes.Other researcher had reported similar observation that plant proteins results in broad particle size distribution in ice cream mix (Ma, Grossmann, Nolden, McClements, & Kinchla, 2022).
Previous studies have stated that partial coalescence is characterized as bimodal, representing both individual fat droplets and fat aggregates (Sung & Goff, 2010).As more coalescence dominates, peaks appear at progressively larger sizes.Generally, MPC forms a layer onto the fat globules via steric forces (Alvarez, Wolters, Vodovotz, & Ji, 2005;Daw & Hartel, 2015) while PPI and SPI improves stability of emulsions by formation of physical barrier at oil or water interfaces to prevent flocculation and coalescence of fat droplets (Li, et al., 2016).However, in this study, there were higher amounts of larger particles (above 10 µm) observed for samples with PPI, followed by SPI, then MPC, indicating large and unstable fat droplets formed after homogenization process in protein enriched samples.

Melting rate and shape retention
The melt rate of the reference ice cream was the highest at 0.15 ± 0.01 g/min, followed by MPC (0.09 ± 0.04 g/min), SPI (0.01 ± 0.00 g/min) and lastly PPI (0.00 ± 0.00 g/min).Ice cream sample enriched with SPI had the best shape retention, followed by PPI, reference and MPC.As observed from Figure 2, the liquid dripped off from the reference and MPC samples were slightly translucent, whereas for SPI samples the liquid was clear.On the other hand, no dripped liquid was recorded for PPI samples which could be attributed to the good WHC and OHC, and high viscosity properties of PPI observed in section 3.1 and 3.4, thus forming stronger gel network structure (Guler-Akin, Avkan, & Akin, 2021) which entrapped the melted lipid and water from dripping out from the system.Moreover, a negative correlation were observed between melt rate and viscosity in our study which agrees with findings by Sivasankari, et al. (2019) who found an increase of viscosity corresponding to the flow resistance in pulse protein concentrate enriched ice cream mix, that caused immobilisation of water molecules to move freely among other molecules in the ice cream matrix, thus reducing the meltdown rate.

Ice crystal size distribution
Referring to Figure 3, the mean ice crystal sizes were all less than 50 µm across all formulations even after thermal stress treatment, indicating ice crystal growth of all samples were under control as ice crystal size larger than 50 µm could contribute to rough texture and icy mouthfeel of ice cream (Drewett & Hartel, 2007).At T 0 , it can be observed that there were no significant differences across the formulations.After thermal stress treatment, the ice crystals in PPI and SPI ice creams had minimal increase and were similar to reference, but not for MPC.This indicated that pea and soy proteins but not milk proteins at 7% were able to delay ice recrystallization during temperature flocculation.With reference to Figure 7. Mean ratings in a radar plot using 9-point hedonic scale to compare sensorial attributes for the 4 ice cream formulations.Lomolino, Zannoni, Zabara, Da Lio, & Iseppi, (2020), the lower growth rate of ice crystals could be due to the high viscosity and strong network or gel-like structure of ice cream that reduced the mobility of water around the ice crystals.A correlation can be observed between viscosity and ice crystal size for the protein-enriched ice cream samples whereby ice cream mixes with higher viscosity had lower ice crystal growth when exposed to thermal fluctuations.

Shrinkage
As depicted in Figure 4, the shrinkage of all the samples were below 5% after heat shock treatment, indicating acceptable stability.Comparing the effect of protein type on shrinkage, all protein enriched ice creams had lower extent of shrinkage as compared with reference when stored under the same condition, indicating the air stabilizing effect of proteins at 7% regardless of its type.During freezing, viscosity of the matrix surrounding the air cells increased, thus reducing the mobility of molecules and lowering the possibility of disproportionation which causes shrinkage and drainage of ice cream (VanWees, Rankin, & Hartel, 2021).Furthermore, fat destabilization also plays a part whereby the extent of fat coalescence lead to reduced interfacial tension, which enhanced the collapse of proteinaceous air cell wall and allow air to escape, hence resulting in greater shrinkage (Dubey & White, 1998).

Confocal laser scanning microscopy (CLSM)
As shown in Figure 5, clumps of fat globules were found to be present in the serum phase for all the ice cream samples, indicated by the reddish-pink regions, which could represent the extent of fat destabilization.Proteins were indicated by green regions in the micrographs, whereby higher extent of aggregation can be observed for reference and PPI samples.For all the ice cream samples, there were proteins attached to the surface of the air cells.There was greater extent of fat displacement with PPI and SPI observed as compared with MPC at the air-serum interface.The fat droplets also seemed to be smaller and more dispersed in PPI and SPI samples, suggesting a more stabilized emulsion.Droplet size in emulsions typically depend on the type of protein used, varying in size, surface chemistry and aggregation state (Ma, Grossmann, Nolden, McClements, & Kinchla, 2022).The observed results could be due to the larger protein particle size in suspension and surface chemistry of PPI and SPI.There could be faster adsorption of smaller proteins to the interface which facilitates initial droplet formation, whereas the larger proteins get adsorbed slower but induce more steric repulsion that aids in stabilization.In addition, there were more obvious protein aggregation in PPI samples followed by SPI, MPC and reference.This is in line with the particle size distribution results shown in Figure S1B.In addition, there seemed to be presence of more air cells for samples with PPI and SPI.This observation can be attributed to the high viscosity which requires greater shear to break down the air cells (Muse & Hartel, 2004).

Profile of amino acids in ice cream
To visualize the effect of different proteins on the profile of amino acids, z-scores were used to plot the heatmap in Figure 6.In the heat map, lower and higher relative concentrations were illustrated with blue and orange color respectively.From the heat map graph, it can be observed that all samples contained higher amounts of polar acidic amino acids as compared with other amino acids while tryptophan and glycine were the lowest.In addition, z-scores of most polar no charge amino acids were negative indicating their lower abundance when compared with other categories of amino acids.Low presence of majority of the essential amino acids listed above agreed well with literature (Roquette;Zeng, et al., 2022).In general, literature stated that plant proteins are rich in glutamic acid, leucine, lysine, arginine, and deficient in methionine, cystine and tryptophan.As for the branched amino acids (BCAAs), results showed that MPC sample contained more BCAAs than SPI and PPI samples which meant that there could be better synthesis of proteins (Neinast, Murashige, & Arany, 2018).Lower amounts of valine were found in ice cream enriched with dairy proteins and higher amounts of leucine and isoleucine were found in SPI ice cream.Essential amino acids were found to be lacking even for dairy ice cream (SMP formulated).These results indicate that amino acid enrichment is needed for protein enriched ice creams to improve the amino acid profile.

Sensory analysis
Figure 7 expresses the sensory attributes results for all ice cream samples.With regards to creaminess, iciness and grittiness, there were no significant difference from reference (P > 0.05) across all samples.For firmness and aftertaste, significant difference can be observed between PPI and SPI as compared with reference but not for MPC.Ice cream formulated with PPI had the highest firmness, which supports the analytical firmness results obtained in Table 2.In terms of aftertaste, the ice creams formulated with PPI and SPI, strong pea and beany notes were perceived by panelists although masking flavor was added into the formulations.The perceived sweetness for all protein enriched samples were all lower than the ref- erence, in which the sweetness could have been masked by proteins, which agrees with a previous report (Yuceer & Drake, 2003).For mouth-coating attribute, PPI had no significant difference when compared with reference.The overall acceptability for ice cream enriched with MPC (6.56 ± 1.85) was significantly more accepted than that formulated with PPI (3.92 ± 2.02) and SPI (3.76 ± 1.74), where all samples were rated using the 9-point hedonic scale.Aside from acceptance test, other sensory method such as just-about-right (JAR) could also be utilized for future studies to obtain more objective measurement of sensory attributes (Peh et al., 2024, Ribeiro et al., 2024).

CONCLUSION
The pea, soy, and milk protein concentrate or isolate powders used in this study had significantly different functional properties.The water-holding capacities were higher than the oil holding capacities for all the proteins samples, with PPI being the highest.The surface hydrophobicity was the highest for MPC, followed by SPI and PPI.In general, this study had shown that elevated protein level at 7% reduces ice cream melting rate and shrinkage.Certain correlations among functional properties were observed, whereby protein with higher WHC results in higher ice cream mix viscosity, firmer ice cream and lower melting rate.In this case, PPI has the highest WHC, that contributed to high viscosity in mix and firmness in ice cream compared with SPI and MPC.However, high viscosity might lead to processibility issues in manufacturing setting.Reduction of fat droplet size was observed in ice cream samples containing PPI and SPI.Interestingly, it was notable that PPI and SPI had better physicochemical properties especially shape retention and ice recrystallization inhibition but were inferior in terms of taste and acceptability.The study reveals that protein concentration and protein variation still needs to be explored to improve the sensory acceptability of plant-based protein ice cream at elevated protein level.
Figure 1.WHC, OHC (A) and H 0 (B) of protein powders (Milk Protein Concentrate 80, Soy Protein Isolate and Pea Protein Isolate).Data are presented as mean values ± standard deviation.WHC: Water holding capacity; OHC: Oil holding capacity; H 0 : Surface Hydrophobicity.Data are statistically compared within each column.

Figure 2 .
Figure 2. Melting curves (A) and shape retention (B) of the 4 ice cream formulations at 0, 60 and 100 min.Data are presented as mean values ± standard deviation.

Figure 3 .
Figure3.Ice crystal diameter determination on the 4 ice cream formulations, before and after subjecting to 1 and 2 h of thermal stress, respectively.Data are presented as mean values ± standard deviation.Values with same big superscript letter are not significantly different from each other between formulations for each cycle at α < 0.05.Values with same small superscript letter are not significantly different between cycles for each formulation α < 0.05.

Figure 4 .
Figure 4. volumetric shrinkage comparison of the 4 ice cream formulations (A) and visual observation for shrinkage before and after 3 cycles of heat shock treatment (B).Values are expressed as mean ± standard deviation; mean values with different superscripts differ significantly (P < 0.05) with each other.

Figure 5 .
Figure 5. Confocal laser scanning micrographs observed in the ice cream emulsions containing (A) anhydrous milk fat, (B) milk protein concentrate, (C) pea protein isolate and (D) soy protein isolate.Samples were dyed with FCF Fast Green and Nile Red.Red indicates fat; green indicated protein (scale bar 250nm).

Figure 6 .
Figure 6.Heat map displaying z-scores of amino acids profile of protein powder.Essential amino acids are indicated with '#' and purple, red, green and blue color represents amino acids that are polar acidic, polar with no charge, polar basic and non-polar respectively.