Impact of a minimally processing route for the production of infant formulas on their sensory properties

Infant formulas (IFs), the sole adequate substitute to human milk, undergo several thermal treatments during production that can damage milk proteins and promote the formation of Maillard reaction products, modifying nutritional and sensory properties. The aim of this study was to determine the impact of a minimally processing route based on membrane filtration associated with different levels of heat treatment, on the odor, taste, texture and color attributes of IFs, then to compare with those of commercial milks. Three experimental IFs (produced with membrane filtration associated with low – T - , medium – T + , or high thermal treatments – T +++ ) were evaluated. Triangular tests conducted with a panel of 50 adults highlighted clear disparities between all the IFs. The same panel applied the Check-All-That-Apply method to evaluate the IFs: the range of variability between T - and T +++ was similar to that between the 2 commercial IFs, and the sensory characteristics of the experimental IFs were not far from the commercial brands for flavor and texture attributes. Analysis performed on the citation frequencies for each descriptor differentiated T - /T + from T +++ , but all the experimental IFs were described with positive sensory characteristics, unlike one commercial IF. Vola-tile organic compounds (VOCs) content of IFs with low and high thermal treatments were analyzed. Forty VOCs were identified by gas chromatography-mass spectrometry. T - contained a higher quantity of VOCs than T +++ , except for benzaldehyde (Maillard reaction product), and aldehydes (oxidation-related products) were the most represented compounds. In conclusion, the processing was associated with sensory differences among IFs, but no marked difference in flavors was found according to CATA and physicochemical analysis. Additionally, no unpleasant sensory descriptors were noted. This shows that the minimally processed route leads to IFs that could fit well within the market from a sensory point of view.


INTRODUCTION
The European Union is a leading international producer in the infant nutrition field, with 1.2 million tons of infant formulas (IFs) produced in 2019 (Marguet, 2020).Within Europe, France is the top producer and one of the main worldwide exporters.Despite the international recommendation of breastfeeding infants for at least 6 mo, many newborns are fed with IFs, and this feeding practice starts early in life, with the proportion of infants receiving IFs increasing as they grow older.In France, by the age of 1 mo, 57% of children have already received some form of IF (Wagner et al., 2015;de Lauzon-Guillain et al., 2018).Approximately 40% of parents have declared that they pay attention to the quality of industrial IFs when making a purchase, considering factors such as the origin of components and preservation methods, without necessarily taking the price into consideration (Nielsen, 2015).
Infant formulas are primarily manufactured from a blend of bovine skim milk, whey proteins, lactose, vitamins, and minerals.These ingredients undergo various unit operations, including thermal treatments, during IF production.These thermal treatments ensure microbiological safety and extend shelf life but also impact the structure of milk protein components by inducing protein denaturation/aggregation, and promoting the Maillard reaction.During IF processing, thermal treatment can also induce lipid oxidation, resulting in the formation of molecules responsible for off-notes (Thomsen et al., 2005).The aggregation of whey proteins can increase viscosity and reduce emulsion stability, which can have technological implications (Buggy et al., 2017;Joyce et al., 2017).Protein denaturation and aggregation can also lead to nutritional implications, as previous studies have shown their impact on the in vitro digestion rates of both caseins and β-lactoglobulin (Halabi et al., 2020;Peram et al., 2013).Furthermore, heat-induced protein denaturation has been reported to modulate protein allergenicity (Golkar et al., 2019).The Maillard reaction results in the development of typical odors associated with heated or cooked food.It can also reduce the nutritional value, primarily by decreasing the bioavailability of essential amino acids such as lysine, and may lead to the formation of carcinogenic compounds (Damjanovic Desic and Birlouez-Aragon, 2011).Some molecules resulting from the Maillard reaction, such as furans, have been described as potentially carcinogenic in humans (Knutsen et al., 2017).Consequently, all these alterations can impact the sensory and nutritional properties of IFs.
Product health safety is a critical concern, especially for infants, who constitute a very sensitive population.Microfiltration, conducted with a 0.8-µm pore size ceramic membrane, is an operation that can replace a part of the thermal treatments for bovine milk debacterization, as previously reported (Yu et al., 2021).Furthermore, the 0.1-µm pore size membrane microfiltration allows for the separation of whey proteins, enabling them to remain in their native form.Microfiltration, combined with other low-heat unit operations, is currently an underutilized processing route to reduce the use of thermal treatments during IF production.
The flavor of milk is influenced by a complex combination of various small, volatile, inorganic, or organic molecules, which collectively serve as quality indicators (Muelbert et al., 2021).Significant variability is observed in the composition of volatile organic compounds (VOCs), whether among milks from different species or within the same species.In studies on bovine milk, the content of ketones and aldehydes may be more prominent in some cases (Hougaard et al., 2010), while others report higher levels of esters and alcohols (Moio et al., 1993).In human milk, carboxylic acids are identified as the most significant component according to He et al. (2022), whereas aldehydes and ketones are more prominent in the study by Hausner et al. (2009), and esters are more present according to Sun et al. (2023).Therefore, even for a given milk origin, the composition of VOCs in milks varies widely from one study to another.Infant formulas are relatively simpler in terms of flavor content than natural milks and are well-documented in the literature.Previous studies focusing on the chemical characterization of VOCs in IFs have highlighted the predominance of aldehydes, alcohols and esters (Wang et al., 2019).However, there is also significant heterogeneity among different brands of IFs, similar to the variances observed between natural milks (Hausner et al., 2009), attributable to various processing methods and ingredients.
The sensory properties of IFs could significantly impact their acceptability by infants, who are capable of perceiving olfactory and gustatory information even at low intensities (Lipchock et al., 2011).Therefore, it is crucial to assess the sensory properties of IFs produced using a minimal processing route.Innovation in this field goes beyond a simple increase in market share.With growing consumer interest in natural and organic food products, IFs lack crucial biomimicry due to the numerous treatments during their production, distancing the composition of IFs from that of breast milk.
The objective of this study was to understand the influence of the IF processing route combining microfiltration and minimal heat treatment on the flavor properties of IFs.The tested hypotheses aimed to evaluate this impact on sensory perception as well as the composition of VOCs.Initially, the sensory properties of several experimental IFs (microfiltered IFs produced with low, medium, or high thermal treatments) were compared.Subsequently, we attempted to position the experimental formulas in relation to 2 well-known commercial brands in the French market.Finally, the VOCs profiles of IFs with low and high thermal treatment were analyzed; indeed, this is a crucial indicator of milk integrity and quality, with these VOCs contributing to the overall palatability of the food.

Tested products
The 3 experimental IFs were produced as previously described (Yu et al., 2021).In summary, the bacterial purification process involved a 0.8-µm pore size membrane filtration of fresh bovine skim milk, combined with liquid whey proteins obtained through a 0.1-µm membrane microfiltration, along with lactose and minerals.This was followed by vacuum evaporation, homogenization after the addition of lipids and vitamins, and a final spraydrying of the mixture.Based on these processing route, 3 different types of IFs were manufactured: the T -formula with no additional heat treatment; the T + formula, with a moderate heat treatment (pasteurization at 75°C for 2 min) applied before spray, whereas the T +++ formula received successive heat treatments (72°C for 30 s on the milk; 90°C for 2-3 s before evaporation; 85°C for 2 min before spray-drying), thus mimicking those received by industrially-processed powdered IF.The resulting protein denaturation level in T +++ (58%) was similar to that observed in commercial IFs, in contrast to T -(6%) and T + (10%).These 3 experimental IFs were subsequently packaged in metal tin cans under a modified atmosphere, Martin et al.: Flavors of minimally processed infant formulas similar to standard IFs.The composition of the experimental IFs was developed to comply with EU regulations regarding nutritional quality and microbiological safety (Commission Regulation N°2016/127, 2016).The 3 experimental IFs were compared with 2 commercial IFs from different brands: Guigoz 'first age Optipro' (Nestlé, Vevey, Suisse), and Blédilait 'first age' (Bledina, Limonest, France).These 2 IFs, hereafter referred to as G and B, were selected due to their leading position in the French market and their respective rates of protein denaturation, close to T +++ (72.3 ± 0.4% and 76.7 ± 0.0% respectively, as reported by Yu et al., 2021).

Samples preparation
All 5 IFs were prepared to a final volume of 500 mL before each session.For T -, T + , and T +++ , 62.5 g of powder was weighed and blended with 437.5 g of preheated Evian® water at 37°C.The commercial IFs were prepared according to the brand's specifications: 62.7 g of G powder was added to 437.5 g of preheated Evian®, while 67.1 g of B powder was mixed with 432.9 g of preheated Evian®.All the IF mixtures were stirred for 3 min then kept refrigerated until service.For sensory testing, 10 mL of IF was served in 80 mL plastic cups, at room temperature (20-23°C).Each IF was assigned a 3-digit anonymization code.

Regulatory approval and panel recruitment
Conducting sensory evaluations with adults is more practical for assessing sensory differences between various food products (Schwartz et al., 2017).For this reason, this study was conducted with a panel of adults as an initial step toward their sensory characterization.
The study took place in the sensory evaluation laboratory of the Centre des Sciences du Goût et de l'Alimentation (CSGA, Dijon, France).Fifty adults, aged 18 to 40, were recruited from the PanelSens database of the ChemoSens Platform at CSGA, declared to the competent commission (CNIL, Commission Nationale Informatique et Libertés, No. 1148039).Eligible panelists had to be over 18 years old and consume cow's milk and a dairy product at least once a month and once a week, respectively.Exclusion criteria included allergies to cow milk proteins and lactose intolerance.The recruited panelists had no prior experience with triangular test or Check-all-that-apply (CATA) questionnaires.
All panelists participated in 2 sessions and received compensation for their participation in the form of a gift voucher.The study was approved by the Ethics Evaluation Committee of INSERM (reference No. 20-648, granted on 2020/01/21).Panelists were informed of the study's purpose and provided their consent before participating.

Triangular tests
A series of sensory discrimination triangular tests were conducted with the participation of 50 panelists to compare perceptible differences between IFs, based on overall sensory characteristics.In each triangular test, 3 samples were presented to each panelist in different orders.Among these samples, 2 were similar, and the third one was different.Assessors were required to identify the sample that differed from the others.The 5 IFs were compared against each other, resulting in a total of 10 triangular tests.These 10 tests were conducted during the same session.After completing 2 triangular tests, panelists were instructed to rinse their mouths with water and consume a thin slice of apple to remove any fatty film that may have formed in their mouths after swallowing the IF.Additionally, a 5-min break was provided to the panelists following the fourth and seventh triangular tests.The presentation of the 10 triangular tests was organized following a Williams Latin square to mitigate order effects in the evaluations.For each triangular test, the presentation order among the panelists was also arranged in accordance with a Williams Latin square.Data were collected using FIZZ software (Biosystèmes, Couternon -France).

CATA method
CATA questions (Varela and Ares, 2012) were used to obtain a sensory characterization of the products.An internal panel of 5 adults generated the list of descriptors.They relied on literature data and their expertise regarding these products to choose the most appropriate terms for describing the IFs (Garrido et al., 2015;Hausner et al., 2009;Moio et al., 1993;He et al., 2022;Houggard et al., 2010).They also conducted tastings.Twenty attributes were then presented to the 50 panelists who participated in the triangular tests, encompassing categories such as color (white, yellow), visual texture (liquid, thick), odor (fresh cream, butter, vanilla, caramel, cut grass, dairy, fish, rancid oil, plastic, cardboard), taste (sweet, acid, metallic), and texture in mouth (fat in-mouth, dry-in-mouth, velvety).The panelists independently tasted the 5 IFs and selected descriptors that they felt were appropriate to characterize each IF after tasting.The CATA session took place on a separate day following the triangular tests.The FIZZ software was used to create the experimental design, ensuring that the orders of the 5 IFs and the descriptors were balanced among panelists using a Williams Latin square.

Aroma extraction
The VOCs in T -and T +++ were extracted using a modification of the solvent-assisted flavor evaporation technique (Engel et al., 1999).Initially, 120 mL of ultrapure water was added to 350 µL of a 110 ng/µL solution of deuterated hexanal (used as an internal standard with purity >99%, Sigma-Aldrich, Saint Quentin-Fallavier, France) in a glass flask.This internal standard was used for relative quantification.The mixture was homogenized for 30 min, after which 20 g of the IF powder was added and stirred for 5 min.The flask was placed in a 35°C bath, and vacuum distillation at 2.10 −2 mbar was carried out for 90 min after the sample boiled.The distillate containing the flavor compounds was collected in another glass flask cooled with liquid nitrogen.This mixture was gathered at the end of the distillation and then extracted 3 times with 15 mL of dichloromethane (99.5% purity, Carlo Erba, Val-de-Reuil, France).Following liquid-liquid separation, the organic phase was collected and concentrated using the Kuderna Danish apparatus.The extracted volume was adjusted to 260 µL with dichloromethane.Three replicate extracts were obtained from each IF.All of these extracts were stored at −20°C for subsequent analysis by gas chromatographymass spectrometry.

Analyses of extracts by gas chromatography-mass spectrometry
T -and T +++ replicates were analyzed by gas chromatography-mass spectrometry.For each sample, 1 µL was injected into a split/splitless injector set at 240°C.VOCs separation was performed using an Agilent 7890A gas chromatography apparatus (Palo Alto, CA, USA), equipped with a polar capillary DB-WAX column (30 m long × 0.25 mm ID × 0.50 µm film thickness; Agilent J&W, Santa Clara, CA, USA).Helium was employed as the carrier gas at a flow rate of 40 cm/s.Chromatographic separation conditions were set from 40°C to 240°C, with a ramp rate of 4°C/min, and held for 10 min.VOCs detection was carried out using an Agilent 5975C mass spectrometer (Palo Alto, CA, USA), operating in electron impact mode at 70 eV, covering a mass-to-charge ratio (m/z) range from 29 to 250.All VOCs were identified by comparing their experimental linear retention indexes and obtained spectra with several databases, including the US National Institute of Standards and Technology (NIST 2.0), the Wiley Flavors and Fragrances (Wiley 138), INRA Mass (internal VOCs database, Dijon, France), and Volatile Compounds in Food (VCF version 15.2).Data processing was performed using the MSD ChemStation software.

Statistical analyses
The Thurstone model was used to analyze the triangle test data.This model assumes that the sensory profile of each product follows a normal distribution, and the hypothesis that these distributions are similar is tested using an index called d-prime (or Thurstone delta).A d-prime not significantly different from 0 (null hypothesis) indicates that the subjects did not differentiate the 2 samples compared.The p-value is calculated using the Binomial distribution.For the CATA analysis, Cochran's Q test, calculated for each descriptor, was used to determine whether the products differed from each other in terms of citation frequencies.A low p-value beyond a significance threshold (0.1) indicates that at least one product was different from the others.When the p-value was significant, we examined multiple pairwise comparisons using the Critical Difference (Sheskin) procedure.A correspondence analysis (CA), using the Chi-squared distance, was carried out to graphically summarize the main differences between the products.Non-discriminating descriptors (Cochran's Q test, P > 0.1) were not selected for this analysis.

Impact of heat treatment on the sensory properties of experimental IFs
Since experimental IFs differ only in the intensity of heat treatment applied, conclusions drawn from comparisons between these milks can be entirely attributed to this factor.The results of the triangular tests (Table 1) made it possible to highlight significant sensory differences between the IF that underwent the most intense heat treatment (T +++ ) and that obtained with the minimal heat treatment (T -).On the other hand, the IF obtained with the moderate treatment (T + ) was not judged to be different from the 2 other experimental IFs (T +++ and T -).These results indicate that heat treatment has an impact on sensory properties, but also that the differences are significant only when comparing the most different heat treatments (T -and T +++ ).Moderate heat treatment (T + ) makes it possible to obtain an intermediate IF in terms of sensory properties, not different from T -and T +++ .
The triangular tests also revealed that T +++ and T -milks could be discriminated by the panel.The 2 corresponding heat treatments therefore resulted in perceptible sensory differences, which we attempted to clarify using CATA analysis.However, the results presented in Table 2 show that none of the 20 preselected descriptors made it possible to discriminate the 3 IFs.The citation frequencies of these terms do not vary significantly depending on the Several explanations can be considered.The first concerns the list of descriptors selected for the CATA.It is possible that, despite all the attention paid to this exercise, the internal panel at the origin of this list missed certain decisive descriptors for discriminating between the 3 experimental IFs.A second explanation relies on the CATA method itself.CATA is a rapid method with advantages and disadvantages.Among these disadvantages, we can wonder whether it is suitable for the analysis of very fine differences requiring a certain expertise of the panel.In our case, it is possible that the selected panel, composed of 50 consumers without prior experience with the CATA test or descriptive analysis in general, was not able to specify the nature of the perceived differences between T -and T +++ by selecting descriptors from a list, while being able to detect global differences between these 2 products during the triangular test.The task carried out during a discriminative test such as the triangular test is, in fact, much simpler than that of the CATA.These 2 explanations constitute one of the limitations of this study.However, it seems reasonable to consider that if obvious differences had existed between T -and T +++ , for one or more of the 20 descriptors on the list, the CATA analysis would have shown it.

Comparison of experimental and commercial IFs
The results obtained from the triangular tests (Table 1) indicate that, based on their overall sensory properties, all experimental IFs (T -, T + , and T +++ ) were judged to be different from B and G, the 2 commercial IFs (P < 0.0001, d-prime values between 1.976 and 2.860).The CATA analysis helped determining the nature of the differences separating experimental IFs from commercial IFs.Table 2 summarizes the results of Cochran's Q tests carried out for each descriptor, as well as the multiple comparisons making it possible to determine whether there are significant differences between the citation frequencies of the different IFs. Figure 1 presents the results of the correspondence analysis carried out on the CATA data and summarizes, using a map, the main sensory properties of the products studied.Only discriminating descriptors were taken into account for this analysis.In the section below, general comments will be based on the map (Figure 1), and more specific comparisons between products will be based on Table 2 (multiple pairwise comparisons).The map resulting from the correspondence analysis (Figure 1) shows a clear separation between the IF G and the other IFs, according to axis 1. Axis 1 is mainly based on an opposition between the descriptors acid, fish, oil rancid, and metallic on the one hand and, on the other hand, the descriptors fresh cream, velvety, thick, sweet, fat-in-mouth, dairy, and white.It should be noted that the second set of descriptors, pointing toward the left part of the graph, is rather positively connoted for milks.In a study of 14 powdered infant formulas, Xi et al. (2023) showed that the descriptors fatty flavor, fishy flavor and sourness contributed negatively to consumer preferences, and conversely, sweet and milk flavor was appreciated.
Overall, the subjects primarily used the first set of descriptors (acid, fish, rancid oil, and metallic) to describe IF G and less frequently used the descriptors from the second set (fresh cream, velvety, thick, sweet, fat-inmouth, dairy, and white).Conversely, IFs B, T -, T + , and T +++ were not often described with the descriptors acid, fish, rancid oil, and metallic, and more frequently associated with the descriptors fresh cream, velvety, thick, sweet, fat-in-mouth, dairy, and white.
To further explore the comparison between experimental and commercial IFs, it is necessary to analyze the results of multiple pairwise comparisons (Table 2, Citation frequencies for each IFs).These results reveal that the 3 experimental IFs are very close to IF B. In fact, T -differs from B only by a less pronounced dairy note.T + differs from B only by a more pronounced cardboard note.Finally, T +++ and B obtained non-significantly different citation frequencies for all the descriptors studied.Among all the experimental IFs, T +++ is therefore the closest to B. The differences between experimental IFs and commercial IF G are more numerous.Compared with G, T - was significantly more often described as white, sweet, and velvety.Conversely, T -was significantly less often described with the descriptors fish, rancid oil, acid, and metallic than G. Compared with G, T +++ was significantly more often described with the descriptors fresh cream and velvety and significantly less often with the descriptors rancid oil, fish, and acid.Finally, T + differs from G only because it was significantly less often described with a fishy note.
Overall, these results demonstrate that the experimental IFs obtained using the microfiltration process fit well into the product space defined by the 2 commercial milks chosen for this study.A comparison with a larger number of commercial milks would make it possible to refine the position of experimental IFs in a broader product space.The experimental IFs are closely aligned with IF B, a market-leading product described with positively connoted descriptors.

Chemical characterization
Triangular tests revealed no significant difference between T + and T -on one hand, and T + and T +++ on the other hand, indicating that T + represented an intermediate IF treatment.Therefore, T + was not chosen for the physicochemical characterization of the VOCs.Forty compounds were identified in T -and T +++ : 19 aldehydes, 10 ketones, 8 alcohols, and 3 acids.Table 3 presents their molecular names, odor descriptions (according to the Volatile Compounds in Food 15.2 database), and their concentrations in these experimental IFs.Among the identified compounds, isomers of 3 species could not be determined using the databases: 2,4-heptadienal, 2,4-decadienal, and 3,5-octadien-2-one.
The solvent-assisted flavor evaporation method was selected because it facilitates the characterization of compounds found within the matrix, and not only in the vapor phase, in a fairly exhaustive manner.In the T -condition, we observed that some compounds had a coefficient of variation (CV) greater than 20% (hexanal, heptanal, octanal, nonanal, 2-undecenal, (E)-, 1-octen-3-one, 2,2'-oxybis-ethanol, and hexanoic acid).Despite solvent-assisted flavor evaporation being the least reproducible extraction method (Liu et al., 2017), this outcome is undesirable.The CV remains high even when the most extreme replicate is eliminated, indicating significant differences among the 3 concentrations measured from the same sample.This observation applies to all chemical families, making it challenging to explain, especially considering that this repeatability issue is not observed in 32 out of 40 compounds.We are aware of the approximation of relative quantification, as the ionization response in mass spectrometry is very dependent on the physicochemical properties of the molecules (Le Quéré and Lucchi, 2022).However, these results allowed us to determine an order of magnitude for each of the identified molecules.
All the chemical classes identified in the present study have been previously reported in the literature to characterize commercial IFs (Jia et al., 2019).Within each class, there are numerous similar species, such as hexanal, heptanal, octanal, and 1-pentanol, for instance.Aldehydes are the most commonly found VOCs in commercial IFs (Fenaille et al., 2003), as well as in human milk (Garrido et al., 2015;Hausner et al., 2009) and bovine milk (Hougaard et al., 2010).Aldehydes, ketones, and alcohols could result from lipid oxidation, a phenomenon described in other studies (Romeu-Nadal et al., 2004;van Ruth et al., 2006).The unsaturated fatty acids may have been peroxidised, leading to the formation of hydroperoxides.The decomposition of these species results in a complex mixture of secondary products, including aldehydes and ketones.Ketones can also be produced through the β-oxidation of unsaturated fatty acids or amino acid degradation (Bertolino et al., 2011).The odor descriptions of these carbonyl compounds play a significant role in the flavor development of human milk.For example, among the compounds we identified, aldehydes such as octanal, nonanal, 2-Nonenal-(E), and 2-Decenal-(E) are the most important contributors to dairy-fat odor, one of the significant descriptors mentioned by our panelists (Yu et.al, 2024a).In the latter study, the fat content exhibited the strongest correlation with the aldehyde content (Yu et.al, 2024b).
Since the volatiles were only partially quantified, making definitive conclusions about their contribution to the flavor of the milk is not feasible.Quantities of hexanal and pentanal (2 of the most oxidative markers) were previously compared in several commercial IFs (Chávez-Servín et al., 2008).According to this study, when the boxes were opened, hexanal ranged from 26 to 6,230 µg/kg, indicating significant variability in the quantities of this aldehyde among the 20 IFs tested.In our study, hexanal quantification was 9,513 µg/kg for T -and 3,660 µg/kg for T +++ .T +++ is in the range measured previously, while T -is more consistent with the range observed after 15 d of opening (mean ± SD ranging from 88 to 13,200 µg/kg in Chávez-Servín et al., 2008), despite the fact that T -and T +++ boxes were opened the same day.Hexanal is naturally present in grass, milk, and vegetable oils, rich in unsaturated fatty acids, and added to the present IFs (Yu et al., 2023).Moreover, there may have been some hexanal degradation during heat treatment in T +++ , such as previously reported (Grebenteuch et al., 2021).However, it is crucial to note that this IF does not exhibit any unpleasant characteristics or off-flavors, as determined by the panelists during sensory analysis.They cited a 'fat in-mouth' taste rather than 'rancid oil,' for instance.In the same study of Chávez-Servín et al., pentanal was quantified at levels ranging from 167 to 1,720 µg/kg when the boxes were opened, which aligns with our relative quantification of this compound (1,228 for T -and 575 for T +++ ).The majority of the compounds showed a difference of a factor of 2 between T -and T +++ .The highest ratio is observed for an isomer of the 3,5-Octadien-2-one (7.9).Maybe it is not sufficient to induce a difference in sensory perception.Indeed, the perceptual quality of odors is generally robust to concentration variability (Cleland et al., 2007).Diaz, in 2004, investigated the effect of the perception of 3 esters in a single solution (without mixing) on the intensity of their perception through retro-and ortho-nasal ways.Overall, a concentration difference of a factor of 2 does not distinctly alter the perceived aroma intensity, even though some minor differences exist depending on the concentration and type of molecule.In complex mixtures, an aroma compound can moreover have multiple odors depending on its concentration and the interactions it may form within the matrix and with the other molecules.Even more crucially, the concentration ratio of odorants in a mixture is clearly a key factor that can determine the perception of food (Thomas-Danguin et al., 2014).Many components would contribute more strongly than others to the overall perceptual quality of the odor blend, even at levels below the threshold (Hummel et al., 2013).Synergistic effects are also significant as described by Tian et al. (2020): only 3 aroma compounds are enough to exhibit synergistic effects in a yogurt matrix, and provide a consequence in sensory response.Therefore, it is impossible to predict the perception of a matrix solely based on its composition.Finally, in our study, the VOC composition appears sufficiently similar to explain the minimal difference observed between the 2 formulas.Physicochemical analysis of commercial brands could have helped us understand the perception differences observed in sensory analysis.The absence of certain chemical classes after SAFE extraction (such as lactones and terpenes) and the lack of reproducibility suggest, however, that we may have overlooked something.Despite the fact that the VOCs analysis revealed some off-flavor characteristics (according to the Volatile Compounds in Food database), such as 'fish' ((E)-2-octenal, 2,4-heptadienal, (E)-2-decenal) and 'plastic' ((Z)-2-penten-1-ol) or 'acrid' (1-hydroxy-2-propanone, hexanoic acid), many panelists did not perceive these unpleasant odors.The fact that T -exhibits a slightly higher level of VOCs compared with the most Martin et al.: Flavors of minimally processed infant formulas heated formula, combined with a lower protein denaturation rate (Yu et al., 2021) and a tendency toward better protein digestibility (Calvez et al., 2023) demonstrates the promising potential of minimal processing route, based on microfiltration, for producing IF.

CONCLUSIONS
The impact of a minimal processing route on IF sensory properties was evaluated through the chemical characterization of VOCs and sensory evaluation conducted by a panel of tasters.The experimental IFs, with low and high intensity of the thermal treatment, were globally perceived by panelists as different, but this was not significantly highlighted by the CATA test and VOC characterization.Experimental IFs were close to the commercially available IF B, a market-leading product, but further away from the IF G, which exhibits less appealing sensory characteristics.The nature and quantity of VOCs identified from the experimental IFs were minimally affected by the manufacturing process, except for a few specific aldehydes tending to differ between T -and T +++ .In conclusion, microfiltration associated with low heating for IF production was associated with no specific sensory and chemical effects on IF flavors, and with no unpleasant sensory descriptors and a sensory proximity to major IF brands.According to the results of the physicochemical analysis, minimal heat treatment could lead to a greater quantity of VOCs, thus preserving more flavor compounds in the IFs.Coupled with much lower protein denaturation rate and the trends toward better protein digestibility, this demonstrates the interest of IF production with microfiltration and low heating without any sensory cost.
Martin et al.: Flavors of minimally processed infant formulas different experimental IFs, therefore depending on the intensity of the heat treatment of these milks.

Table 1 .
Martin et al.: Flavors of minimally processed infant formulas Triangular tests: number of correct answers (cor.ans.) and p-values.A minimum of 23 correct answers is necessary to conclude a difference in perception between infant formulas (IFs) at P < 0.05

Table 2 .
Check-all-that-apply analysis for each infant formula (IF): + (experimental IF with medium temperature treatment), T +++ (experimental IF with high temperature treatment).*Indicates discriminating descriptors (P < 0.1).a-b Multiple pairwise comparisons are represented by the letters within the cells (for each descriptor, two citation frequencies with no common letters are significantly different at the 5% threshold).

Table 3 .
Yaylayan, 2008)c compounds identified in T -and T +++ formulas and their concentration in µg/kg of powder infant formula (IF)Yaylayan, 2008)and reducing sugar (such as lactose present in the milk) and promoted by heat treatment.It is in agreement with the higher heat treatment received by this IF and its tendency to contain more Maillard reaction products, such as Nε-carboxymethyllysine, than T -(Yu