Effect of Lactic Acid Fermentation on Volatile Compounds and Sensory Characteristics of Mango (Mangifera indica) Juices

Fermentation is a sustainable bio-preservation technique that can improve the organoleptic quality of fruit juices. Mango juices were fermented by monoculture strains of Lactiplantibacillus plantarum subsp. plantarum (MLP), Lacticaseibacillus rhamnosus (MLR), Lacticaseibacillus casei (MLC), Levilactobacillus brevis (MLB), and Pediococcus pentosaceus (MPP). Volatile compounds were sorbed using headspace solid phase microextraction, separated, and identified with gas chromatography-mass spectrometry. Forty-four (44) volatile compounds were identified. The control, MPP, and MLB had higher amounts of ethyl acetate, ethyl butyrate, 2-hexenal, 2,6-nonadienal, 2,2-dimethylpropanal, β-selinene, γ-gurjunene, α-copaene, and δ-cadinene, while MLC, MLP, and MLR had higher amounts of 2,3-butanedione and a cyclic hydrocarbon derivate. Consumers (n = 80) assessed their overall liking and characterized sensory attributes (appearance, color, aroma, flavor, consistency, acidity, and sweetness) using check-all-that-apply, and penalty analysis (just-about-right). Overall liking was associated with ‘mango color’, ‘pulp’, ‘mango aroma’, ‘sweet’, ‘natural taste’, and ‘mango flavor’ that described the control, MLB, MLC and MPP. Juices MLR and MLP were described as ‘bitter’, ‘sour’, ‘aftertaste’, and ‘off-flavor’. Multivariate analysis revealed relationships between the volatile compounds, mango juices fermented by different lactic acid bacteria, and sensory characteristics. Thus, the type of lactic acid bacteria strains determined the volatile and sensory profile of mango juices.


Introduction
Mango (Mangifera indica L.) is globally an important commercial fruit with high demand in the international market. It is among the top 10 major fruits cultivated in sub-Saharan Africa with a production of over 8 million tonnes/year [1]. Approximately 35% of produced fruits are lost post-harvest every year given it is a seasonal climacteric fruit with a few harvesting seasons, and its fresh fruit has a very short shelf-life. Thus, there is a need to transform this perishable fruit into products such as fruit juices with a long shelf-life and diversify its products through fermentation.
Fermentation using lactic acid bacteria increases food shelf-life by lowering pH and producing antagonistic metabolites such as organic acids and bacteriocins that are lethal to pathogens [2]. Furthermore, fermentation improves the nutritional and organoleptic quality fruits were selected based on their maturity, uniform color, no visible infection, and no mechanical damage. They were washed using distilled water, peeled, chopped, and mixed using a domestic blender (Joseph, MI, USA) to obtain mango juice without the addition of water. The obtained mango juice was homogenized using an Ultra-Turrax (IKA T18, Staufen, Germany) at 1422× g (10,000 rpm) for 15 min and pasteurized according to Shaheer et al. [16]. Briefly, 50 mL of mango juice dispensed in a sterile 100 mL flask was pasteurized at 80 • C (internal temperature) for 5 min in a water bath (Memmert WNB 45, Schwabach, Germany) with an external temperature of 100 • C under continuous shaking. The pasteurized juice was rapidly cooled to room temperature using an ice-water bath (0 • C).

Fermentation of Mango Juice
Pasteurized mango juices were inoculated with monoculture washed bacterial cells (1% v/v) and incubated at optimal growth temperatures (30 or 37 • C) for 24 h to obtain mango juice fermented by Levilactobacillus brevis (MLB), Lacticaseibacillus casei (MLC), Lacticaseibacillus rhamnosus (MLR), Lactiplantibacillus plantarum subsp. plantarum (MLP), and Pediococcus. pentosaceus (MPP). The control was pasteurized mango juice without the addition of lactic acid bacteria and incubated under the same conditions (30 • C, 24 h). The juices were then kept at 4 • C to stop fermentation and analyzed within 12 h. Three independent fermentation experiments were carried out for each bacterial strain.

Extraction of Volatiles
Volatile compounds in the samples were analysed using HS-SPME GC-MS according to a method described by Hinneh et al. [19] with some modifications. Briefly, 2 g of juice sample was added to each HS-SPME vial (20 mL) and thoroughly mixed with 2 mL of saturated sodium chloride (NaCl) (Merck, Belgium) previously brought to pH 3.0 (with 0.8 M acetic acid solution, Merck, Belgium) and 3 µL of the internal standard, 2-octanol (Sigma-Aldrich, Belgium) at a concentration of 213.6 mg/L methanol (Merck, Belgium). The vial was hermetically sealed and then incubated (Gerstel, Müllheim an der Rur, Germany) at 40 • C for 20 min in a thermostatic agitator to extract the volatiles. The released volatiles in the headspace were subsequently sorbed onto the divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) fiber (75 µm, Sigma-Aldrich, Belgium) for 20 min at 40 • C. Three independent experiments for each lactic acid fermentation were carried out and between each GC-MS analysis, the fiber was conditioned for 7 min at 270 • C.

GC-MS Analysis
The gas chromatograph (GC) (Agilent 6890, Agilent Technologies, Santa Clara, CA, USA) was connected to a mass spectrometer fitted with a ZB-Wax plus column (30 m × 0.25 mm i.d. × 0.25 µm film thickness, Zebron, Phenomenex, Macclesfield, UK). Helium gas was used as a carrier gas with a constant flow rate of 1 mL/min. The DVB/CAR/PDMS fiber was inserted and desorbed for 180 s into the splitless injection port (250 • C) of the GC oven. The following time-temperature program was applied: 40 • C for 5 min, then increased at 5 • C/min to 80 • C, at 3 • C/min to 134 • C, and at 8 • C/min to 230 • C, where it was held for 2 min. Mass spectrometry was performed at a 230 • C ion source temperature with a mass range from m/z 40 to 300 (full scan mode) and 70 eV ion current using no solvent delay and a threshold of 50 [20]. The extracted volatile compounds detected were identified.

Identification of Volatile Compounds
Volatile compounds were identified by comparing retention indices on the ZB-Waxcolumn with literature data, matching the MS-spectrum of each peak to those of the Wiley275 library (quality match > 85%), and their retention index (RI) values were calculated using a series of n-alkanes (C 9 -C 16 ) as standards according to Vandendool & Kratz [21]. An internal standard method was used to quantify the identified volatiles [22]. Therefore, data have been expressed as nanograms of the internal standard (2-octanol) equivalents per mL of sample and were calculated as: (1) where M s is the identified volatile concentration, expressed as ng/mL; M i is the weight of the internal standard, expressed as ng; M o is the weight of mango juice used, expressed as mL; A s is the peak area of identified volatiles; and A i is the peak area of the internal standard. Percentage differences of the volatiles in each fermented mango sample versus the control were calculated to evaluate any differences between the samples.
2.6. Consumer Sensory Acceptability 2.6.1. Participants Eighty (80) participants were randomly recruited from students, staff, and visitors of the Nelson Mandela African Institution of Science and Technology. Eligibility for participation followed the criteria of Meilgaard et al. [23]: no food allergies (oral allergy syndrome) or dietary intolerances, consumption of fruits, willingness, and availability. All participants' demographics are described in Table S2. The majority were aged between 18 and 49 years with 55% males and 45% females. No prior information regarding the aim of the study or content of the products was given, and no reimbursements were made for their participation.
This study was approved by the Health Research Ethics Committee of Kibong'oto Infectious Diseases Hospital, the Nelson Mandela African Institution of Science and Technology, and the Centre for Educational Development in Health Arusha under the protocol number KNCHREC0008, and each participant gave informed consent for inclusion before they participated in the study.

Sensory Data Collection
Juice samples (20 mL) were served cold (4 ± 1 • C) in styrofoam cups identifiable by a random three-digit code. Each consumer received six samples (5 fermented and control juices) one at a time and between each different sample, two unsalted crackers and bottled water were provided to rinse their mouths, and a 2 min break was taken. The samples were served in a completely randomized order using William's Latin square design [24] to balance bias caused by first-order and carry-over effects. This experiment took place in a room with a classroom arrangement, adequate lighting, noise-free uninterrupted environment, and participants did not face each other.
For each sample, consumers first rated their overall liking using a 9-point hedonic scale [25] with 1 = 'dislike it extremely', and 9 = 'like it extremely'. This hedonic scale also assessed the appearance, aroma, sweetness, flavor, consistency, acidity, and color attributes of the juices.
Secondly, consumers used the check-all-that-apply (CATA) method [26] to characterize the samples. This is a multi-choice question that comprises a list of terms from which the consumers select. The terms used were based on prior work [27] and included: 'mango aroma', 'mango color', 'mango flavor', 'thick', 'pulp', 'sweet', 'sour', 'off-flavor', 'natural taste', 'intense flavor', 'light color', 'bitter', and 'aftertaste'. Consumers were requested to check all applicable terms. For each sample, these terms were randomized in a monadic sequence following a balanced order by using William's Latin square design [24].
Finally, consumers stated their intent on whether they would likely purchase the product in the market using a 5-point scale ranging from 'certainly would not buy' to 'certainly would buy' [29]. Consumers were also asked questions regarding their age, gender, frequency of fruit consumption per month ('more than once a week', 'once a week', 'more than once a month but less than every week' or 'less than once a month') [30], and whether they paid attention to their diet.

Statistical Analysis
Statistical analyses were performed using XL-STAT, (version 2020.1, Addinsoft, Paris, France), IBM SPSS for macOS (Version 23, IBM Corporation, Armonk, New York, NY, USA), and GraphPad Prism (Version 8.0.0 for macOS, San Diego, CA, USA). All the microbiology and volatile assays were performed in triplicates in three independent experiments, and results were expressed as the assay's average. Data of volatile compounds were analyzed using one-way analysis of variance (ANOVA) followed by a post-hoc Tukey t-test to determine any significant differences (p < 0.05) between samples. Principal component analysis (PCA) was used to study relationships between samples in terms of volatile profiles.
For the sensory data, repeated measures one-factor ANOVA and Bonferroni post-hoc test was used to check for differences in the overall liking and sensory attributes between the different samples. A frequency analysis assessed attributes on the JAR scale and thereafter penalty analysis [31] examined if any of the attributes influenced a mean drop in the overall acceptability for each sample. Based on Pareto's principle, significant (p < 0.05) results were considered when a proportion of >20% consumers criticized an attribute either as too 'low' (−) or too 'high' (+) and caused a mean drop of >1 point on overall liking [28]. CATA data were analyzed using Cochran's Q test [32], which analyses a two-way randomized block design (data matrix) to check if the samples as treatments have similar effects (McNemar post-hoc) when the consumer response is binary (checked/not checked) [8].
A multiple factorial analysis (MFA) [33] was used to determine relationships between the samples based on the overall liking, liking of key sensory attributes, CATA characteristics, and volatiles data.

Growth of Lactic Acid Bacteria in Mango Juice during Fermentation
The mango juices had an initial (T 0 ) lactic acid bacteria concentration of 7-8 log CFU/mL, but after 24 h fermentation (T 24 ), the viable counts increased to a maximum of 9.16 log CFU/mL in MLB (Table 1). This increment was significant (p < 0.05) in MLB, MLP, and MLR. Counts in the control were below the detectable limit of <1 log CFU/mL. The microbial analysis also showed that the control had a total plate count (<3 log CFU/mL) and yeast and molds (<2 log CFU/mL) below permitted levels (<4 and <3 log CFU/mL, respectively) according to the Codex Alimentarius Commission of the Food and Agricultural Organization [34]. Similarly, besides the lactic acid bacteria, no other microorganisms were observed using total plate count (<3 log CFU/mL) and yeast and mold were below detectable limits (<2 log CFU/mL) in the samples. Total coliforms in the samples were below detectable limits (<1 log CFU/mL), as well as pathogenic microorganisms Escherichia coli and Salmonella spp. (<1 log CFU/mL).
In the control sample, mainly monoterpenes were detected: δ-3-carene, α-pinene, β-myrcene, α-terpinene, limonene, and β-phellandrene ( Figure S1). After fermentation (24 h), some variations in volatile concentrations were observed, for instance, the levels of 2,6-nonadienal (cucumber notes) and 2-hexenal (apple and green notes) fell sharply in all samples while a cyclic hydrocarbon derivate, originally not in the control, was detected in all fermented juices ( Table 2). The percentage change (%) of volatile compounds in fermented mango juices compared to the control was therefore calculated (Figure 1). Representation of volatile concentrations −100% mean complete degradation and +100% mean production after fermentation.
The total level of the monoterpenes (Figure 1a) did not significantly change (<15%) after fermentation. However, there was a significant increase in β-ocimene in MLP (p = 0.001), limonene in MLC (p = 0.010), and β-myrcene in MLP and MLC (p = 0.004) while the p-cymene decreased (p = 0.033) in MLC and MLP. Most of the sesquiterpenes decreased ( Figure 1b) in MLB, MLP, and MLR except for an unknown sesquiterpene which increased in MLP and MLR by over 40% (p < 0.05). In MPP and MLC, a slight increase (<10%) in the total level of sesquiterpenes was observed. Especially for the level of β-caryophyllene, a significant (p < 0.05) decrease was recorded in MLB and MLR. In "other volatiles" (Figure 1d), a cyclic hydrocarbon derivate was unique to fermented juices with a production of >+100% after fermentation and was not detectable in the control. The chemical structures of the terpene compounds are shown in Table S3.   Alcohol concentrations (Figure 1c) significantly decreased in MLR by 63% in 1-hexanol (p ≤ 0.001) and 26.1% in 3-hexen-1-ol (p = 0.027). Unsaturated aldehydes 2,6-nonadienal and 2-hexenal were degraded by more than 80% and 42% after fermentation in all the fermented juices, but the concentration of an unknown aldehyde significantly increased (p = 0.006) in MPP.
Four (4) esters were found in the samples. After fermentation, in all the fermented juices, the levels of ethyl butyrate and ethyl acetate (except MPP) decreased while concentrations of linalyl propanoate significantly increased (35-158%). Among other volatiles (Figure 1d), the amount of 2,3-butanedione increased (p < 0.05) tremendously in MLC and MLR juices by 282% and 419%, respectively, as opposed to MPP and MLB where it significantly decreased (p < 0.05) after fermentation.
The relationships among the samples based on their volatile data were illustrated using a principal component analysis (PCA) plot ( Figure 2). Considering the 44 compounds, the first two PCA dimensions accounted for 30.8% (PC1) and 22.3% (PC2) of the variance. PC1 separated MLC from MLR, but the results showed PC2 was the main axis for the separation of the control from MLC, MLR, and MLP. The control, MPP, and MLB were localized on the PC2 positive semi-axis due to higher levels of esters (ethyl acetate, ethyl butyrate), aldehydes (2-hexenal, 2,6-nonadienal, and an unknown aldehyde), and sesquiterpenes (β-selinene, γ-gurjunene, α-copaene, and δ-cadinene). MLC, MLP, and MLR were localized on the PC2 negative semi-axis and had higher amounts of 2,3-butanedione, an unknown sesquiterpene, and a cyclic hydrocarbon derivate.

Consumer Sensory Acceptability
All the samples were liked moderately ranging from 7.71 in the control to 6.71 in MLR ( Table 3). The fermented mango juices did not differ significantly (p > 0.05) from the control except for MLR, which was rated significantly lower. Although aroma and flavor were most liked in MLB (7.63 and 7.66, respectively), MLB was similar to the control, MLC, and MPP in these attributes. However, MLP and MLR juices received the lowest scores (6.28-6.96) and differed (p < 0.05) from the control in terms of aroma, flavor, consistency, and sweetness. The CATA question obtained binary responses of terms that consumers perceived to describe the samples. The terms 'mango flavor', 'mango color', 'mango aroma', 'sweet', 'thick', and 'natural taste' were most frequently (>60% of consumers) used to describe the samples (Figure 3). Fermented juices did not differ from the control in most of the sensory terms but 4 out of 14 terms were significantly different, i.e., 'natural taste' (p = 0.003), 'sour' (p = 0.001), 'sweet' (p = 0.011), and 'watery' (p = 0.008). MLB had the highest mention of 'natural taste' at 64% followed by MPP (59%) and the control (50%), whereas MLR had the highest mention of 'sour' (49%) and least mention of 'sweet' (41%).
A sensory map from multi-factorial analysis (MFA) evaluated whether the allocation of these terms contributed to overall liking and showed any relationships between the product categories. The first two MFA dimensions (Figure 4) explained 73.9% of the total variability. There was a good correlation between different samples, sensory terms, and overall liking. The positive F1 semi-axis represented the control, MLB, MPP, and MLC. These juices were closely associated with overall liking and characterized with 'mango color', 'pulp', 'mango aroma', 'sweet', 'natural taste', and 'mango flavor' terms. Conversely, MLP and MLR juices were in the negative F1 semi-axis (separate level) and characterized with 'sour', 'bitter', 'aftertaste', and 'off-flavor' terms.
Penalty analysis obtained information on the intensity level of modifiable sensory attributes of each sample. Overall, consumers who found the samples to deviate from just-about-right was less than 50% ( Figure 5). Moreover, attributes that fell in the upper right corner were considered most concerning as they have the highest skews and had the greatest mean drop, while those in the lower-left corner are those with minimal concern. As observed, the control, MLB, MPP, and MLC had the least penalized attributes in the upper right corner compared to MLR and MLP. No sensory attribute exceeded the 20% threshold for MLB. The control and MPP had only two out of six attributes above the threshold, MLC registered three attributes, MLP had four attributes, while for MLR, all six attributes were criticized by > 20% of consumers.
Sweetness and aroma were the most penalized attributes and were considered too low in the control, MPP, MLC, MLR, and MLP, causing a mean drop for overall liking scores ranging from 1.10 in MLC to 1.69 in MLP juice. Acidity was penalized for being too high in MLC, MLR (mean drop 1.43), and MLP (mean drop, 1.35). Consumers had conflicting opinions on the flavor of MLP as 21.3% found it to be too high (1.73 mean drop) and 30% too low (1.51 mean drop). Consistency and color were only criticized in MLR as too low and too high, respectively.
Regarding consumers' purchase willingness, the control had the highest percentage of 'certainly would buy' at 46% followed by MLB and MPP at 40% ( Figure S2). The highest scores for MLR and MLP juices were recorded at 'might buy' at 32.5% and 33.75%, respectively.

Discussion
Lactic acid bacteria grew and survived in mango juice, and this may be attributed to nutrients (carbohydrates, organic acids, vitamins, and minerals) in mango that are a source of energy for metabolism. Other studies have also demonstrated mango as a suitable medium for lactic acid bacteria growth [4][5][6][7]. After fermentation, the acidity of fermented mango juices increased, and this may be antagonistic to pathogenic microorganisms increasing juice shelf-life.
During fermentation, lactic acid bacteria produce various products that may directly or indirectly be involved in the decrease or increase in volatile compounds. Although the total level of the monoterpenes did not significantly change after fermentation, a significant increase was observed in β-ocimene, limonene, and β-myrcene. Lactic acid bacteria produce acids such as lactic acid and acetic acid during fermentation that may damage the fruit cells leading to the release of these compounds [38]. In addition, it is generally recognized that monoterpenes originate from the plastids of pyruvate and glyceraldehyde-3-phosphate via the 2-C-methyl-D-erythritol 4-phosphate (MEP) pathway [39]. Lactic acid bacteria possess an extensive array of enzymes including terpene synthases, which can be produced by Levilactobacillus brevis and Pediococcus pentosaceus and are involved in their biosynthesis and biochemical reactions [40]. On the other hand, most sesquiterpenes decreased (Figure 1b) after fermentation, which is in agreement with Park et al. [41] who also found that lactic acid bacteria significantly decreased terpenes in a mixed berry juice. This reduction could be due to their oxidation to secondary products, hydroxylation, acylation, or isomerization [39,42].
2,3-Butanedione was the only ketone detected. Lactic acid bacteria have plasmidencoded citrate transporter genes and together with enzyme citrate lyase, they can degrade citrate present in mango to 2,3-butanedione [43]. Strains such as Lacticaseibacillus casei, Lacticaseibacillus rhamnosus, and Lactiplantibacillus plantarum subsp. plantarum can convert citric acid during citric acid metabolism to acetate and oxaloacetate under the catalysis of citric acid lyase. The oxaloacetate is decarboxylated by oxaloacetate decarboxylase to produce pyruvate [44]. The pyruvate is then condensed by α-acetolactate synthase to α-acetolactate, which is chemically unstable and can be converted to diacetyl (2,3-butanedione) in a non-enzymatic oxidative decarboxylation reaction or by α-acetolactate decarboxylase [45]. After fermentation, 2,3-butanedione tremendously increased by >800% in MLC and MLR (Figure 1d). Lacticaseibacillus rhamnosus produces high amounts of 2,3-butanedione (64 mg/g glucose) [46] which have a profound effect on the flavor and aroma of fermented products as it is characterized by a strong buttery odor that may probably not be organoleptically acceptable. Hence, 2,3-butanedione may be an index for product quality control. Another compound, i.e., 2-pentylfuran, which has odor notes of beany, oxidized, and green could also give an undesirable buttery flavor to the sample of MLC (Figure 2).
Strong fruity aromas such as apple-like (ethyl butyrate) and pineapple-like notes (ethyl acetate) were also found in the samples. After fermentation, their levels decreased (Table 2) unlike in other fruit juices such as apple juice, where a slight increase after Lactiplantibacillus plantarum subsp. plantarum, Lacticaseibacillus rhamnosus, and Lacticaseibacillus casei fermentation was reported [47]. Linalyl propanoate, which gives citrus-like notes, significantly increased (35-158%) in all fermented juices ( Table 2). This aliphatic (straight-chain) ester may be formed from the metabolism of fatty acids through β-oxidation.
Aldehydes (2-hexenal and 2,6-nonadienal) could also influence sample sensory attributes as they give fatty-grassy and cucumber notes, respectively. Liu et al. [48] also identified 2,6-nonadienal in fresh Tianong mango pulp. After fermentation, 2-hexenal and 2,6-nonadienal were degraded. This result is in agreement with Jin et al. [4] who reported a decrease in aldehydes in mango slurries fermented by Lactiplantibacillus plantarum subsp. plantarum. During fermentation, aldehydes may be reduced to their corresponding alcohols or oxidized to acids [49]. A high level of aldehydes may cause off-flavors whichnegatively impact the sensory characteristics of fermented food [50].
Sweetness and consistency were most liked in the control, MLB, MLC, and MPP, while MLP and MLR were the least scored. This correlated with volatile analysis which showed that δ-3-carene, with sweet and limonene-reminiscent odor responsible for ripe mango flavor [51], were highly concentrated in MLC while β-ocimene, responsible for the warm, herbaceous, and floral odor characteristic of raw (unripe) mango flavor was mainly present in MLP, MLR, and MLC ( Figure 6). MLR had a significant decrease in the concentration of alcohols (Table 2); 1-hexanol (fruity and aromatic flavor) and 3-hexen-1-ol (intense green grassy odor) that give desirable sweet flavor notes.
The CATA data showed that 'mango flavor', 'mango color', 'mango aroma', 'sweet', 'thick', and 'natural taste' (Figure 3) were the main drivers of consumer liking as they were the most frequently used terms. Thus, consumers like fermented mango juices that still maintain the natural taste and flavors of mango juice. Consumers also detected differences between the samples, as there were significant differences in the frequency of mention using Cochran's Q test (p < 0.05) similar to other studies that characterized orange juices [52] and chocolate milk deserts [26]. The MFA (Figure 4) showed relationships between the samples based on their CATA characteristics, overall liking, liking of key sensory attributes, and volatiles. Overall liking was strongly associated with the control, MLB, and MPP which were characterized by 'mango aroma', 'natural taste', 'sweet', and 'mango flavor' terms, sesquiterpenes, monoterpenes, alcohols, and aldehydes. MLP, MLR, and MLC were associated with 'off flavor', 'sour', 'aftertaste', and 'intense flavor' terms probably due to high levels of 2,3-butanedione.
Regarding penalty analysis, different attributes had a differential effect on the overall acceptability of each product. MLB was highly accepted as none of its attributes led to a mean drop in overall acceptability ( Figure 5). Its attributes were perceived as optimal requiring no adjustments, hence this product could be scaled up by food producers but potential changes in flavor during storage should be further investigated. MLR and MLP were the most penalized juices, especially the sweetness, aroma, acidity for MLR; and sweetness, aroma, acidity consistency, flavor, and color for MLP. Hence, these attributes should be modified during reformulation. Consumers disagreed on the ideal intensity of MLP flavor, and this polarity may be attributed to the quality of this attribute rather than its quantity [53]. Furthermore, the use of just-about-right scales has been found to make respondents more aware and critical of imperfections in samples [53].
Consumers rely on sensory attributes to purchase foods and make a re-purchase on products that they like. This study showed that MLB and MPP had a higher purchasing intent than MLC, MLP, and MLR ( Figure S2). However, it should be noted that all the samples scored above 30% on 'would buy' showing that they would compete favorably on the market. In addition to volatiles, other non-volatile compounds such as sugars and organic acids may be accumulated or depleted by lactic acid bacteria during fermentation, affecting the sensory characteristics of fermented juices.

Conclusions
Following the lactic acid bacteria fermentation, the content of sesquiterpenes, aldehydes, alcohols, and esters decreased while ketones and furans increased in mango juice. The control, mango juice fermented by Pediococcus. pentosaceus and Levilactobacillus brevis had higher amounts of ethyl acetate, ethyl butyrate, 2-hexenal, 2,6-nonadienal, 2,2dimethylpropanal, β-selinene, γ-gurjunene, α-copaene, and δ-cadinene, while juice fermented with Lacticaseibacillus casei, Lactiplantibacillus plantarum subsp. plantarum, and Lacticaseibacillus rhamnosus had higher amounts of 2,3-butanedione and a cyclic hydrocarbon derivate. There was an association between the volatile compounds of the fermented mango juices and their sensory acceptability. Fermentation of mango juice with Levilactobacillus brevis, Lacticaseibacillus casei, Lactiplantibacillus plantarum subsp. plantarum, and Pediococcus. pentosaceus except Lacticaseibacillus rhamnosus did not affect the overall liking. Overall liking was related to 'mango aroma', 'natural taste', 'sweet', and 'mango flavor'. Mango juices fermented by Levilactobacillus brevis were most accepted and are a potential product for scaling up. However, juices fermented by Lactiplantibacillus plantarum subsp. plantarum and Lacticaseibacillus rhamnosus were most criticized and require modifications/reformulation. A follow-up study is recommended to confirm if the changes made are effective in improving these mango products. Moreover, the flavor of mango juice fermented by Lactiplantibacillus plantarum subsp. plantarum received conflicting consumer critics, hence an appropriate consumer group should be targeted during product development and marketing. Further research may be carried out to investigate the effect of non-volatile compounds on the sensory acceptability of fermented mango juices.