Two mix of yogurt lactic acid bacteria coming from the lab collection were used in this study, N1 and N2. Both included one strain of Streptococcus thermophilus (STN1, STN2) and one strain of Lactobacillus delbrueckii subsp bulgaricus (LBN1, LBN2). The four strains were stored at −80˚C in a medium containing 0.4 mL of an overnight culture (42˚C), 0.4 mL of sterilized broth and 0.4 mL of glycerol (30%, SIGMA). The broths used culture the bacteria were either M17 (BIOKAR) for streptococci or MRS (BIOKAR) for lactobacilli.

During the course of this experiment, we inoculated milks with different mix of bacteria. The cow milks used systematically originated from the same batch and from two different origins. The first type of milk was purchased from OXOID. Before use, this milk powder was reconstituted (10%) and sterilized (110˚C, 10 min). The second type of milk was half-skimmed and microfiltered; it was bought in a supermarket (Marguerite). Each mix included one strain of ST and one strain of LB. N1 and N2 bacteria were never mixed together. For each bacterial association (N1 or N2), three ratios were tested: 50:50, 99:1 and 1:99 (ST:LB). The standardization of each ratio was made according to the methodology already described by Demarigny et al. (1994) [6]. Each bacterium was cultured at 42˚C in milk. The pH and the bacterial levels were followed from time to time and the corresponding curves were drawn. To adjust the initial ratio, the two bacterial strains were separately cultured in milk (42˚C) during one night. The measure of the pH of this preculture allowed us to calculate the volume to be added to reach the appropriate concentration in the experimental milk culture (1 L, prepared as indicated above). In this study, all the cultures were standardized to start the fermentation steps at approximately 10⁶ cfu/mL (LB + ST, Table 1).

After inoculation with the proper ratio, milks were incubated at 42˚C. The growth of the two bacterial strains and the pH of the medium were monitored over time. At initial time and after 7 h, two samples were also taken for gas chromatography and mass spectrometry analysis (GC/MS) and NeOse Pro analysis.

Unless specified, all the experiments were made in triplicate.

pH was determined with a pH-meter (HI 2211, pH/ORP Meter, Hanna instruments) on 10 mL of milk. The total titratable acidity (TTA) value was defined as the amount of a 0.1 N NaOH solution required to reach a pH-value of 8.5. The results were expressed in milliliters [7]. Bacterial levels were determined by colony enumeration on MRS (LB) or M17 plates (ST). Samples were first diluted in trypton salt (BIOKAR) before being spread on petri dishes.

Volatile compounds were analysed using the Solid-Phase Micro Extraction Gas-Chromatography technique (SPME/GC, 6890 Series GC System, Hewlett Packard), coupled with Mass-Spectrometry (MS, 5973 Mass Selective Detector, Hewlett Packard). Ten millilitres of supernatant were placed in a vial and heated at 42˚C for 10 min. SPME-fiber was conditioned at 270˚C for 30 min. Compounds were absorbed onto this fiber (SPME Fiber Assembly DVB/CAR/PDMS, Supelco, SIGMA) at 42˚C for 30 min. They were then desorbed and separated with the temperature gradient proposed by Aggelopoulos et al. (2014) [8]. Helium was used as the vector gas with a flow rate of 1 mL/min. Molecules were separated through a BPX 5 capillary column, 30 m × 0.25 mm i.d. × 0.25 μm, 5% diphenyl and 95% dimethylpolysiloxane (Phenomenex, Le Pecq, France). They were identified using Chemstation software by spectra comparison with NIST 05 and NIST 98 libraries.

The NeOse Pro apparatus was developed to mimic the sensory cells of the nose. Sixty-four sensors are fixed on a gold micro array. A fan sucks up the volatile molecules originating from the product analysed. These molecules interact with 64 peptides arrayed on the gold layer of an SPR imaging device optical sensor. The resulting interactions are monitored versus time as a change of reflectivity and are ploted as sensorgrams [9] [10].

The preparation of the NeOse Pro apparatus was made according to the preconisation of Aryballe.

Three replicates were made for each analysis. A blank was made with a flask identical to those used for the samples and containing 10 mL of water. The blank measure was performed before the first analysis and after any new test. The methodology was standardized as best as possible. For instance, the temperature of the samples was carefully controlled throughout the experiment.

NeOse Pro measurements were made at different steps during the yoghurt fermentation (T0, T1, T2, T3, T4, T5, T6, T7, T8, T24 h). Experiments were carried out in reconstituted milk (OXOID, 10% w/v), sterilized at 110˚C/10min, or in half skimmed microfiltered milk bought in supermarket.

The extraction of total DNA was based on the method of Masco et al. (2003) [11]. An overnight broth culture (MRS for LB and M17 for ST) was centrifuged for 3 min at 13,000 rpm at 4˚C. The supernatant was removed and the pellet re-suspended in 170 µL of T1 buffer (Genomic DNA from tissue, MACHEREY- NAGEL) and 20 µL of a lysozyme solution (0.1 mg/µL) was added in an Eppendorf tube. The mixture was incubated for 30 min at 37˚C. 20 µL of RNase (1 mg/mL), 30 µL of proteinase K (1 mg/mL) and 20 µL of SDS (Sodium Dodecyl Sulfate) 10% were then added in each tube. Except when specified, all the chemicals used were from SIGMA. The mixture was left to incubate overnight at 37˚C in a water bath. The day after, Eppendorf tubes were incubated for 45 min at 64˚C; then, 200 µL of buffer B3 (Genomic DNA from tissue, MACHEREY- NAGEL) were added. The tubes were then placed at 70˚C for 10 min. 600 µL of CTAB (cetyltrimethylammonium bromide) were added. The tubes were incubated for 30 min at 64˚C and 800 µL of chloroform-isoamyl 24:1 ethanol were added. The tubes were then centrifuged during 5 min at 13,000 rpm and the upper phase was recovered in a clean collecting tube with an elution column. 210 µL of absolute ethanol were added; the tubes were centrifuged 1 min at 13,000 rpm. The collector tube was emptied and 500 µL of BW buffer (Genomic DNA from tissue, MACHEREY-NAGEL) were added prior to another centrifugation (1 min at 13,000 rpm). The collector tube was emptied and 600 µL of B5 buffer (Genomic DNA from tissue, MACHEREY-NAGEL) were added. The tubes were successively centrifuged 1 min and 3 min at 13,000 rpm, the collecting tube being emptied at the end of each centrifugation. The column was then placed in a 1.5 mL Eppendorf tube; 70 µL of BE buffer (Genomic DNA from tissue, MACHEREY-NAGEL) preheated at 70˚C were added. After a rest 5 min at room temperature, the tube was finally centrifuged twice for 1 min at 13,000 rpm.

The protocol followed for DNA amplification and migration was identical to the procedure already published by Masco et al. (2003) [11].

Principle component analysis (PCA) and hierarchical cluster analysis (HCA) were performed using the XLStat software (Microsoft, 2016).

In all the experiments, the ST on LB ratio was carefully standardized to be as closed as possible to the theoretical values aimed at (Table 1). The measured values obtained at the beginning of the fermentation steps are shown in Table 2. Whatever the ratios, the differences between theoretical and analytical ratios were small.

During the acidification step, the evolution of the pH was nearly identical irrespective of the mix of ST and LB used (N1 or N2) or the ratio (1/99, 50/50, 99/1). In particular, if slight differences were occasionally observed at the beginning of the culture (less than 0.1 pH unit), they disappeared rapidly. At the end, the pH was equivalent in all the cultures (±0.02).

This experiment was performed on half skimmed microfiltered milk. Two NeOse Pro measures were made, one just after the inoculation of the milk with one of the two mix of bacteria (STN1 + LBN1 or STN2 + LBN2) and the other at the end of the fermentation step (T = 7 h). Figure 1 shows an example of radar charts obtained after NeOse Pro analysis with the N1 bacterial mix. The radar chart sums up the normalized intensity values for the 64 sensors involved in the volatile molecule detection. Each graph is specific of the aromatic balance of each sample; as such, it is an image or a sign of the sample. As one can notice, the fermentation step led to a change in the aromatic profile of the milk as a consequence of the bacterial development. The same observations could be drawn from the study of the second mix (N2).

A principal component analysis was performed with the data obtained to build the radar charts shown on Figure 1 (Figure 2). Experiments were made in triplicate. The first axe of the PCA totalized 90.78% of the information inertia. It means that the discrepancy between non fermented and fermented milk (“yoghurt” on Figure 2) follows the first axe, confirming the difference of aroma profiles between the two states. As indicated above the same results were obtained with the second yogurt mix (N2, data not shown).

The sterilization of the milk led to the generation of empyreumatic compounds in high concentrations. These chemical products created a background signal which hided the volatile compounds released by the microbes, whatever the fermentation step. The initial NeOse Pro profile (t = 0) was so subtracted from the profiles obtained thereafter, the addition of the microorganisms at initial time being still ineffective. On this base, the profiles obtained after 1 and 2 h turned out to be quite similar. Thereafter, the intensity of the volatile compounds perceived by certain sensors increased as the fermentation progressed. This observation resulted from the influence of the bacterial development. On Figure 3, are displayed, as an example, the evolutions of the response of 9 sensors. The fermentation was carried out with the N1 starter mix, each bacterium being brought to the milk at equal levels (50/50). Two behaviours could be observed: sensors 3, 4 and 5 increased from t = 2 h to t = 24 h—as a consequence of the bacterial growth—whereas the other sensors remained stable or slightly decreased. The scattering of the intensity responses into two groups could be generalized to the 64 sensors. It is noteworthy that the evolution of S3, S4 and S5 was strongly correlated with the increasing of the bacterial level and the diminution of the pH (r > 0.99; p < 0.05).

From these results, two consequences were deduced for the other experiments:

It is preferable to work with microfiltered milk, which avoid to subtract the profile of the heated milk

It is necessary to wait at least 3 h before observing a detectable influence of the bacteria. We decided to work with the results obtained after a 7 h delay.

Two samples of half skimmed microfiltered milk were fermented either with St thermophilus or with Lb delbrueckii subsp bulgaricus and analysed with the NeOse Pro apparatus. The radar charts of the two milks after a 7 h fermentation step appear on Figure 4. The results presented here correspond with the strains STN2 and LBN2, knowing that the study was also made with STN1 and LBN1. To the eyes, the shapes of the radar charts seem clearly different on the basis of the answer of the 64 sensors.

A principal component analysis was built (Figure 5). The first two axes of the PCA totalized 97.10% of the information inertia (82.95% for the first). The discrepancy was less evident than the difference observed between non-fermented and fermented milks. Overall, the scattering of the repetitions was more important. Nevertheless, the two groups appeared distinct. As waited, the same results were observed with STN1 and LBN1.

Three cultures were made with the two yogurts mix N1 and N2 with the three theoretical bacterial ratios (ST/LB, 1/99, 50/50, 99/1). Experiments were made with microfiltered half skimmed milk. NeOse Pro data were obtained after a 7 h culture step. They were successively analysed by PCA (results not shown) and HCA (Figures 6(a)-(c)). The data were analysed separately since the discrepancy between the different ratios was not evident, as indicated later in the article. The two yogurt mix appeared clearly separated whatever the ratio considered. A REP-PCR was performed on the four bacterial strains. Figure 7 shows the agarose gel after staining. The four strains appeared clearly distinct, LB vs ST and STN1 vs STN2 or LBN1 vs LBN2. They were then supposed to lead to different results after fermentations steps.

In this experiment, we compared the three different ratios ST/LB of the two mix; as previously, analyses were made at the end of the fermentation step (T7 h). A PCA followed by a HCA was performed on the NeOse Pro data, mix by mix. After 7 h, the pH values and the TTA of the six cultures were respectively equal to 4.16 ± 0.23 and 69.6 ± 6.77 mL. The levels of the two microbial populations were also close, 1.70.10⁸ ± 2.33.10⁸ cfu/mL for LB and 2.50.10⁸ ± 2.44.10⁸ cfu/mL, whatever the initial ratio or the mix. The NeOse Pro profiles were also similar (Figure 8). It was nearly impossible to separate the three ratios from the same mix.

A GC/MS analysis was made after 7 h of fermentation on the fermented milk (two mix and three ratios) to characterize the chemical nature of the volatile compounds and to associate them with the results of the NeOse Pro analysis. 28 molecules were found at significant levels, among them the most important were: acetaldehyde, 2,3-butanedione, ethyl acetate, 3-hydroxy 2-butanone.

NeOse Pro data were crossed with the GC/MS results and analysed by PCA and CHA, the former being considered as the main parameters whereas the latter were just additive data (not shown).

Out of the 28 main molecules detected by GC/MC analysis, only one—ace- taldehyde—was clearly linked with the same set of sensors, whatever the bacterial mix considered (Table 3). Acetaldehyde was detected by sensors C19, C13, C44 and C45. However, following the mix, some other sensors could be excited as well—C4, C14, C51, C6, C46 for N1; C39 for N2. For the other molecules, no clear relation was noticeable. For example, ethyl acetate, was linked to C1 to C12, C14 to C18, C20, C21, C24 to C31, C33 to C38, C40, C41, C43, C46 to C56 sensors, for N1, and C58 to C62 sensors for N2.