The effect of process variables on the physical properties and microstructure of HOPO nanoemulsion flakes obtained by refractance window

Refractance window (RW) drying is considered an emerging technique in the food field due to its scalability, energy efficiency, cost and end-product quality. It can be used for obtaining flakes from high-oleic palm oil (HOPO) nanoemulsions containing a high concentration of temperature-sensitive active compounds. This work was thus aimed at studying the effect of temperature, thickness of the film drying, nanoemulsion process conditions, and emulsion formulation on the flakes’ physical properties and microstructure. The results showed that HOPO flakes had good physical characteristics: 1.4% to 5.6% moisture content and 0.26 to 0.58 aw. Regarding microstructure, lower fractal dimension (FDt) was obtained when RW drying temperature increased, which is related to more regular surfaces. The results indicated that flakes with optimal physical properties can be obtained by RW drying of HOPO nanoemulsions.

Preparation of coarse emulsions. The coarse emulsions were homogenized in an mixer (Imusa, Bogota, Colombia), first the aqueous phase of the emulsion was prepared incorporating whey powder (formulation A: 29.76% (w/w); formulation B 29.43% (w/w)), determined according to preliminary results 28 , followed by the sequential addition of native corn starch (0.24% w/w, formulation A) or gum Arabic (0.57% w/w, formulation B). Then it was followed by the addition of HOPO (14%, w/w) to the aqueous phase and mixed over 2 min. The soy lecithin concentration was held constant at 10% w/w with respect to the HOPO concentration (1.4%, w/w) and it was added to the oil phase in the preparation of the coarse emulsion. Subsequently, emulsions were processed to obtain the nanoemulsions. Nanoemulsion preparation. The nanoemulsions were obtained following the methodology of Quintanilla-Carvajal et al. 29 with some modifications. They were homogenized in an LM10 microfluidizer (Microfluidics, England) variating the operation pressure between 0 and 20,000 psi following a Box-Behnken optimization design obtained from the software Design Expert Version 10.1.0 (Stat-Ease Inc., MN, USA), in which the following two numerical factors and one categorical were also varied accordingly to their impact on the process: drying temperature (60-80 °C), sample thickness (1-2 mm), and the formulation (A and B), respectively. Table 1 shows the conditions provided by the Box-Behnken optimization design for the preparation of HOPO nanoemulsions. Figure 1 shows a graphical description of the procedure used to obtain the HOPO nanoemulsions and flakes.

Refractance window drying (RW).
A pilot scale Refractance Window (RW) dryer was used for drying nanoemulsions of HOPO. The dryer has an effective surface drying area of 0.43 m 2 and length of 0.92 m. The main components of the dryer included a belt made of "Mylar" (polyethylene terephthalate) plastic, a water pump, a hot water tank, a heating unit, two water flumes, a suction blower, a spreader, and a scraper. The drying was accomplished by spreading HOPO nanoemulsions on the plastic belt. Molds of 14 × 24 × 0.3 cm were used to control the size of the samples. The thickness of the nanoemulsion on the belt was 1-2 mm and was controlled using a spreader bar. The thermal energy from the circulating hot water (transferred to the nanoemulsion through the belt) was used to remove moisture from the product 30 . During drying operation, the temperature of circulating hot water was maintained at 60 °C, 70 °C and 80 °C according to the experimental design. The temperature of the circulating hot water was continuously monitored at the flume inlet and outlet section using pre-calibrated Type T thermocouple sensors. Water vapor removal from the samples was facilitated by using a suction blower. The residence time to dry the HOPO nanoemulsions into flakes was set at 1 h, this time was determined according to preliminary studies (data not shown). www.nature.com/scientificreports/ with reference to illuminant C and a visual angle of 10°3 2 . The colorimeter was calibrated with a black and white standard patterns. Hue (h ab ) was determined using Eq. (1).
In this case, b* is the yellow/blue coordinate, and a* the red/green coordinate. Total color difference (ΔE*) regarding fresh nanoemulsions was calculated as follows 33 .
The ΔL*, Δa* and Δb* variables are the difference values between a reference color and the color of the sample; ΔL* represents the difference of lightness, Δa* represents the difference of red/green coordinate, and Δb* represents the difference of yellow/blue coordinate. The values used to determine whether the total color difference was appreciable by the human eye were the following 34 : ΔE* < 1 color differences are not obvious for the human eye. 1 < ΔE* < 3 color differences are not appreciated by the human eye. ΔE* > 3 color differences are obvious for the human eye.
Measurements were performed in triplicate in three different flakes obtained by RW drying.
Scanning electron microscopy (SEM). The analysis of microstructure of flakes was performed using a scanning electron microscope (Phenom Pro, Cecoltec Ltda, Bogotá, Colombia) at an acceleration voltage of 5 kV. The samples of 2 mm wide were placed on the scanning electron microscope slides with the aid of colloidal silver and a magnification of 4000x was used for the analysis.
Image analysis. Images of the flakes of 2048 × 2176 pixels were captured using an electronic microscopy and stored as bitmaps in a gray scale with brightness values between 0 and 255 for each pixel constituting the image. A generalization of the box counting method was used to evaluate the fractal dimension of the images (FDt). In this work, the shifting differential box-counting method (SDBC) 35 was used to evaluate the fractal dimension of texture of SEM images using the ImageJ 1.34 software. Four different images at the same magnification (4000×) were evaluated for each flake HOPO nanoemulsion. The size of the crops was 67.81 μm × 67.81 μm. The texture parameters, contrast, correlation, inverse difference moment (IDM), and entropy of SEM images were evaluated using the GLCM and surface plot tools of ImageJ.  www.nature.com/scientificreports/ The textural feature contrast is a measure of the intensity contrast between a pixel and its neighbour over the whole image. It measures the local variation in the GLCM. Contrast is 0 for a constant image 34 . The textural feature correlation is a measure of how correlated a pixel is to its neighbour over the whole image. Its range lies between − 1 and + 1. Also, the correlation is 1 or − 1 for a perfectly positively or negatively correlated image. Correlation measures the joint probability of occurrence of pixel pairs of GLCM 36 . The textural feature inverse difference moment (IDM) measures the texture uniformity or orderliness of an image but normalized for distance. As defined by Yang and collaborators, higher inverse difference moment values indicate a variation in image contrast 37 , that is, greater heterogeneity of the flakes texture. The textural feature entropy is a statistical measure of randomness that can be used to characterize the texture of the input image 38 .
Statistical analysis. The statistical analysis was performed using the Box-Behnken optimization experimental design methodology in the Design Expert Software Version 10.1.0 (Stat-Ease Inc., MN, USA). Quadratic models were used to express the response variables as a function of the independent factors, where A, B, C, and D, are the coded values of the drying temperature, the sample thickness, pressure, and formulation, respectively. A statistical significance test was used in the total error criteria with a confidence level of 95%. The significant terms in the model were found through analysis of variance (ANOVA). The fit of the model was evaluated by the R 2 value.
The graphic and numerical optimization of the Design Expert software was used for response optimization. Two optimisation systems (Opt A and Opt B) were formulated; one for a nanoemulsion made from formulation A (Opt A) and the other for a nanoemulsion made with formulation B (Opt B), both having minimum moisture and water activity (a w ) after RW drying.

Results and discussion
Moisture. A sample's moisture during drying is the ratio of a sample's mass of water after drying to a sample's total mass. Table 2 gives analysis of variance (ANOVA) results for moisture response. The results were fitted to a quadratic model, giving 0.81 R 2 . Table 2 shows that the model was significant (p < 0.05) and had 0.0584 lack of fit (p > 0.005); such result indicated that the model was suitable and could thus predict the moisture content for flakes from RW dried-nanoemulsions, prepared from a determined concentration of buttermilk, starch or gum Arabic and oil. The flakes' moisture content ranged from 1.8 to 5.65% (Table 1).
Drying temperature was the variable which significantly affected (p < 0.05) the resulting moisture pattern; the higher the drying temperature, the lower the moisture content in the resulting flake. Even though the pattern was the same for both formulations, formulation B had lower moisture content at the same temperature than that for formulation A. The above could be due to the difference in wall material in both formulation (formulation A was prepared with corn starch and formulation B with gum Arabic), the molecular weight of wall materials impact its capacity of absorb water, affecting directly the moisture content of the sample. In the supplementary file Table S1 are shown the equations describing the pattern of moisture content for both formulations. www.nature.com/scientificreports/  www.nature.com/scientificreports/ The samples' moisture content was only affected by water temperature during drying; this was the variable directly influencing RW drying's three most important aspects: heat emission by the drying source (hot water), propagation of heat by the medium (Mylar polyester film) and the heat absorbed by the product (emulsion). Zotarelli et al. 39 have reported how using Mylar film led to more than 40% transmittance regarding maximum heat emission (60 °C to 100 °C) by water radiation (10-100 W/m 2 ); such values were more than sufficient for good heat transfer by radiation between the heat source and the water in the emulsion 40 . This data suggested the effectiveness of heat transfer during RW drying.
The forgoing highlighted the fact that moisture content varied from 1.8 to 5.65% for the proposed temperature interval and gave favourable values for RW dried products, as values below 15% prevent antimicrobial growth, increase a sample's structural stability and retard deterioration reactions (i.e. sugar crystallisation and non-enzymatic browning) 41 .
Water activity. Water activity (a w ) is an important parameter which can influence food shelf-life since it reflects the amount of water available for chemical reactions and microorganism growth (i.e. bacteria, fungi and yeasts) 42 . Table 2 gives ANOVA results for water activity (a w ) response. The results were fitted to a quadratic model (0.79 R 2 ). Table 2 shows that the model was significant (p < 0.05) and had a 0.28 lack of fit, indicating that the proposed model's a w provided a suitable fit and could thus fit and predict an evaluable water activity pattern for flakes from RW-dried nanoemulsions.
Only temperature had a significant impact on water activity pattern (p < 0.05); at higher temperatures, less water activity. The temperature squared (A 2 ) value also significantly affected a w (p < 0.05); Table S1 of Supplementary file shows the equations describing water activity pattern for each formulation.
This study emphasised the relationship between drying temperature and a w , as increasing the drying temperature led to a reduction in the flakes' a w ; this could have been related to the fact that more water would have been evaporated at higher temperatures 43 . However, no mention was found in the pertinent literature of variation concerning drying temperature and its effect on a w , as most studies compare RW with other drying methods at a single temperature or vary drying times or sample thickness.
Drying is one of the most efficient methods for conserving food, as reducing a w reduces microbial growth and decreases the speed of degradation reactions, the latter being of great interest for drying products having antioxidant activity. Around 0.3 a w values were obtained for the flakes (Table 1) which authors like Pavan et al. 24 (whose group has dried açai pulp) have considered a standard value for pulp dried by any method, since water having higher a w is sufficient to behave like a solvent, thereby increasing the mobility of products used in/available for degradation reactions 13 . Other authors have focused more on the relationship of a w with moisture and storage temperature which are very important aspects when considering food conservation 12,13,39 . Contact angle. Contact angle is a measurement quantifying a substance's hydrophilic or hydrophobic behaviour as an important parameter when analysing surfaces' interaction with water; a smaller angle measurement indicates hydrophilic behaviour whilst a greater angle indicates a hydrophobic pattern 44 .
ANOVA was used for ascertaining quadratic fit ( Table 2) and square root transformation (Supplementary file  Table S1) of data regarding contact angle (0.78 R 2 ). Table 2 shows that the model was significant (p < 0.05) and had 0.72 lack of fit, suggesting the model's great reliability or describing its pattern and possibly acting as a tool for analysing the contact angle of RW dried nanoemulsion flakes in similar conditions. Like a w , temperature (A) and its quadratic relation (A 2 ) were the variables which significantly affected the contact angle. Contact angle varied from 15.72° to 31.99° (Table 1); a sample's lower drying temperature meant a smaller contact angle whilst formulation A had a slightly more hydrophobic pattern (CA = 24°) than formulation B (B = 22°), given the same temperature and thickness values. This pattern could have been due to the chemical nature of gum Arabic (Formulation B) since its structure (consisting of a branched carbohydrate chain (d-glucuronic acid, l-rhamnose, d-galactose and l-arabinose) has glycoproteins bound by covalent linkages, making it a highly hydrophilic molecule 45 . Amylose, which makes up 50% of corn starch, consists of glycosidic bonds producing a simple helix whose interior only consists of hydrogen atoms, thereby giving a lipophilic pattern 46 . In the supplementary file Table S1 are shown the equations describing contact angle pattern for both formulations.
Studying contact angles is closely linked to the science of materials and polymers, so studying it in the agrifood industry has been very limited and (to the best of our knowledge) no research has yet been published on contact angles on surfaces obtained from RW-dried matrices.
Authors who have researched food area-related contact angles have based their work on studying the properties of materials' surfaces in response to contact with liquid foodstuffs (i.e. interfacial interactions). Their usefulness as a mechanism for predicting interactions between liquids and solid surfaces (i.e. biofilms, membranes, heat interchange or packed food surfaces) has been studied by authors like Güleç et al. 47 . Such authors have studied the hydrophilicity, hydrophobicity and surface free energy of three different materials used in the food packaging industry (glass, polyethylene and stainless steel); they studied contact angle variation regarding surface and interfacial tension to ascertain the effects of packaging surface contaminants on food and microorganism growth. Other studies have focused more on predicting food packaging surface structure by using mathematical relationships to modify surfaces to control contact angles and surface energy 48,49 . Color. www.nature.com/scientificreports/ Variable h ab was significantly affected (p < 0.05, Table 2) by flake thickness, microfluidization pressure and formulation. Increasing flake thickness significantly reduced (p < 0.05) h ab , this could be due to the exposure of the sample film to the drying temperature; lower thickness values, less layers to receive the heat transfer, which results in a yellowing of the sample. On the other hand, increasing microfluidization pressure significantly increased (p < 0.05) h ab wich could be due to the organization of the droplets of the sample, that at higher microfluidization pressure, lower the droplet size of the emulsion [50][51][52][53][54] . The same happened for formulation B, h ab values were higher than those for formulation A; in this case, the main difference between formulation A and B was the wall material (corn starch and gum Arabic, respectively), letting conclude that gum Arabic preserve better the color of the original food. The flakes h ab values ranged from 73.24° to 78.80° (Table 1), related to high purity regarding yellow tonality since such values were close to 90°5 5 . Table 3 gives the fresh nanoemulsions' color parameters (L*, a*, b*, h ab ). Color difference (ΔE*) values regarding flakes and nanoemulsions before RW drying varied from 3.81 to 13.29 (Table 1), i.e. an obvious color difference between flakes and fresh nanoemulsion for the human eye (ΔE * > 3) 34 . Such results showed that the temperature of RW drying of a nanoemulsion does produce important changes in parameters (L*, a*, b*), also the pressure of microfluidization which led to encapsulation of the oil, thereby producing high color differences obtained at lower pressure values and higher temperatures. Such difference may have been due to carotene degradation in palm oil (i.e. natural antioxidants controlling color 56 during drying as it has been widely reported that they are temperature-sensitive (i.e. the higher the temperature, the greater carotene degradation, especially when 60 °C has been exceeded) [57][58][59] . Figures 2 and 3 show SEM microstructures of flakes from formulations A and B for every drying temperature and microfluidization pressure used in the study; the porous structure of flakes obtained by drying HOPO nanoemulsions from both formulations was revealed by microstructural study. Nanoemulsions which were not microfluidized (0 psi) had a less homogeneous structure, having larger pores which joined together forming large holes; pore size became reduced when increasing microfluidization pressure. It would thus seem that using higher microfluidization pressure gives more homogeneous microstructure nanoemulsions, as RW drying results in flakes having a more uniform structure, with smaller sized pores, since the continuous phase is more homogeneous in the emulsion. Increased nanoemulsion homogeneity related to decreased particle size when increasing microfluidization pressure has already been reported 19,20 . Image analysis. Table 2 gives ANOVA results for FDt, contrast, correlation, IDM and entropy response. The results for all texture parameters analysed here fit a quadratic model; texture parameters R 2 were 0.85 for FDt and contrast, 0.8 for correlation and 0.78 for IDM and entropy. Table 2 shows that the model was significant (p < 0.05) and had > 0.05 lack of fit, thereby indicating its suitability for predicting the texture parameters which would be obtained when a nanoemulsion prepared at a specific microfluidisation pressure was dried at a determined temperature and lamina thickness.

Scanning electron microscopy (SEM).
Texture FDt analysis regarding drying temperature and drying temperature squared (A 2 ) was statistically significant (p < 0.05) ( Table 2). Increased temperature produced a significant reduction (p < 0.05) in the flakes' FDt for both formulations. Lower FDt values were related to more homogenous and regular surfaces. FDt was also affected by the microfluidization pressure squared (C 2 ) used in preparing the nanoemulsion.
The values obtained came within the range reported to date for food surface images [60][61][62][63] . The pertinent literature states that low FDt values at high drying temperatures could be associated with images having a smooth fractal texture, while high FDt values could be related to images having rough fractal texture 62 . Hernández-Carrión et al. 63 found that Lamuyo red pepper FDt was lower when less structural damage was caused by high hydrostatic pressure. Such results would explain why lower FDt values were obtained at higher microfluidization pressures, related to a more homogeneous and regular structure, as can be seen in the microstructure study. Authors like Aragón-Rojas et al. 64 have established a relationship between freeze-dried powders' FDt and moisture. Such authors have stated that freeze-dried powders have greater surface area exposed to the environment when their surface is more rugged (higher FDt values) and their moisture will thus be higher. This would explain the results obtained when analysing the flakes' moisture, as flakes with higher values of FDt also had the higher values of moisture content.
Texture contrast, correlation and IDM parameters were significantly affected (p < 0.05) by RW drying temperature, as well as such variable squared (A 2 ) ( Table 2). Increased drying temperature led to a significant reduction (p < 0.05) in image contrast, whilst producing a significant increase (p < 0.05) in correlation and IDM for both formulations. Such results indicated that increasing nanoemulsion drying temperature led to a more homogenous and regular texture of flakes after drying. www.nature.com/scientificreports/ A high contrast value indicates a high degree of local variation 65,66 , this being typical of rougher and more heterogeneous surfaces. This would explain the higher contrast values for rougher surfaces observed at low drying temperatures and the lower values for a more homogenous surface at high drying temperatures. It is known that IDM values indicate the degree of image contrast variation and high IDM values can be associated with homogeneous images 65 , such as images obtained at high drying temperatures. An increase in temperature tended to increase the IDM which was related to the more homogenous structures observed at high drying temperatures. Similar results were obtained by Barrera et al. 67 when evaluating mechanical damage to wheat starch granules. They concluded that damaged granule surface had lower IDM values than those for native starch granules, suggesting that the mechanical process decreased IDM. Hernández-Carrión et al. 63 evaluated structural damage to Lamuyo red pepper caused by high hydrostatic pressure treatment and pasteurisation; they found that treatment causing more structural damage to red pepper tissue had lower IDM values and higher contrast values than those causing lower structural damage, thereby suggesting that structural damage decrease IDM values and increase contrast values. The above suggest that drying nanoemulsions at low temperatures could a more heterogeneous flake structure as the IDM values obtained for these temperatures decreased.
Entropy was significantly affected (p < 0.05) by drying temperature and the microfluidization pressure of nanoemulsions (Table 2). Increased drying temperature and microfluidization pressure led to a significant reduction (p < 0.05) of the flakes' entropy for both formulations. The higher entropy values obtained at low drying temperature and microfluidization pressure could have been related to their structure's greater heterogeneity 37 since more complex images are associated with higher entropy values 65,67 . www.nature.com/scientificreports/ The lower entropy values found at high drying temperatures and microfluidization pressures could have been related to their structures' greater homogeneity 37,65 . Hernández-Carrión et al. 63 when evaluating structural damage to Lamuyo red pepper subjected to high hydrostatic pressure treatment and pasteurization stated that treatments causing more structural damage to red pepper tissue had higher entropy values than those causing lower structural damage, suggesting that structural damage increased entropy.
Increased nanoemulsion drying temperature was thus mainly responsible for improving the parameters regarding image texture as a more homogeneous and regular texture was thereby obtained at higher temperatures.  www.nature.com/scientificreports/ Optimisation. Table 4 and Fig. 4 show that high drying temperature (72 °C), intermediate thickness (1.44 mm) and high microfluidization pressure (20,000 psi) must be used for obtaining flakes from nanoemulsions made from formulation A having low moisture content of 2.69% and a w of 0.2847. Higher drying temperature (80 °C), greater thicknesses (2 mm) and lower microfluidization pressure (16,237 psi) would have to be used with formulation B for ensuring low moisture (2.02%) and a w (0.2916). High drying temperatures and microfluidization pressures were thus mainly responsible for obtaining flakes having low moisture and a w ; flakes having lower moisture and aw were obtained from nanoemulsions made from formulation B than formulation A.
It should be stressed that optimal solution desirability was > 92% and moisture and a w experimental values were very close to those estimated by the model's prediction equations (Table 4). Maximum error was 5.76% for optimal nanoemulsion A moisture whilst a 2.70% minimum error was recorded for optimal nanoemulsion B moisture. www.nature.com/scientificreports/

Conclusions
Using response surface methodology for studying the effect of emulsion formulation drying conditions resulted in mathematical models for predicting the pattern of the variables analysed in this work. These results suggested that RW drying, and microfluidization as a nanoencapsulation technique of high-oleic palm oil, could lead to producing dried products in the form of flakes, however, it causes an effect in the color of the emulsion which could be perceived by the human eye. In this case, they were formulated with biopolymers such as corn starch and gum arabic, having high oil content, allowing optimum physical properties represented in lower moisture content and lower water activity, mostly affected by the pressure of microfluidization and drying temperature. For the best of our knowledge, this is the first article that analyses the microstructure of flakes containing oil using RW, that was affected only by the drying temperature. These results provide evidence of new ways of dried structures with high oil contents different from powders. This paper shows the importance of develop technologies and processes that enhance products and allow the addition of compounds at cheaper prices than those that can be offered by technologies such as freeze-drying.