Biopurification using non-growing microorganisms to improve plant protein ingredients

Securing a sustainable global food supply for a growing population requires a shift toward a more plant-based diet. The application of plant-based proteins is therefore increasing, but unpleasant off-flavors complicate their use. Here, we screened 97 microorganisms for their potential to remove off-flavors in a process with limiting amounts of fermentable sugar. This allowed the production of a more neutral-tasting, purified food ingredient while limiting microbial growth and the production of typical fermentation end products. We demonstrate that various lactic acid bacteria (LAB) and yeasts remove “green” aldehydes and ketones. This conversion can be carried out in less than one hour in almond, pea, potato, and oat proteins. Heterofermentative LAB was best at aldehyde and ketone neutralization with minimum de novo formation of microbial volatiles such as ethylacetate (sweet, fruity) or alpha-diketones (butter- and cheese-like). While sensory properties were improved, changes in protein solubility, emulsification, foaming, and in vitro digestibility were limited.


Data normalization of GC-MS data (sterile unfermented plant proteins)
In the absence of internal standards, the obtained GC-MS data is subject to matrix effects which influences the response factor of a volatile compound.This complicates comparison between samples.Within our experiment, this matrix effect is caused by differences between plant proteins, dry matter content (5% for potato protein vs 10% for others), initial pH, and analysis batch.This can be seen through the range and distribution of volatile peak areas (Supplementary Figure 1).Despite lower dry matter content, potato protein showed an overall shift to a higher peak area resulting in an impression of high relative abundance of most volatiles.
Among detected volatiles, we observed that ethylacetate and 2-pentylfuran levels are consistently among the lowest and the highest, respectively.In the case of almond, one of the four replicates (Almond 3, removed from further analysis) seemed to be an outlier resulting in the larger standard deviations in the boxplots (Supplementary Figure 1).However, the presence of the outlier sample did not influence ethylacetate measurements and its standard deviation remained small despite being at the lower range of the data with generally a lower signal to noise ratio.For 2-pentylfuran, it is commonly reported as the product of lipid oxidation.While its presence is expected, its unusually high level among other furans (100-to 1000-fold higher) is likely to indicate high response factors in all matrices.Altogether, the presence of a consistent volatile species at minimum and maximum as well as overall shift in data distribution justifies scaling and centering of data to compare between samples.Scaling and centering for this blank (no incubation with microorganism) dataset was done within samples.For heatmap visualization, scaling and centering within volatiles was done (using the generic "scale" function in R).Averages of peak areas of detected volatiles are summarized in Table S1.

Common plant proteins shared the majority of detected volatiles
To understand similarities and differences between commercial protein isolates and concentrates, we performed an analysis on the volatiles detected in pea, almond, oat, and potato protein, representing legumes, stone fruit, cereals, and tubers, respectively.Based on normalized data (Supplementary Figure 2A), the relative composition of pea and oat proteins seems to be dominated by similar volatiles, while potato and almond proteins show distinctive volatile signatures.This is potentially attributed to differences in raw material and their processing (Supplementary Information S2).From a total of 42 volatiles measured across proteins, 27 of them were detected in all proteins and 7 of the remaining ones were detected in at least 3 proteins (Supplementary Figure 2B).These volatiles can be grouped based on their compound classes (Supplementary Figure 2C).Overall, the reported aldehydes, alcohols, ketones, and furans can be associated with lipid oxidation and/or Maillard reaction, and they are largely shared, albeit at different ratios.
Supplementary Figure 1: Distribution of volatile peak areas (raw data) for each sterilized plant protein (n=3 for potato, n=4 for pea, almond and oat) shown as boxplots.Ethylacetate (consistently small peak area) and 2-pentylfuran (consistently large peak area) are depicted in red.
A commercial potato protein isolate used in this study shows overall low abundance of most detected volatiles (Supplementary Figure 2) with only a few volatiles including a few ketones, heptanal and its key compound dimethyl trisulfide (DMTS) ("cooked potato" 1,2 ) found in higher abundance relative to other proteins.In contrast, almond protein concentrate displayed a high relative abundance of key compound benzaldehyde 3 and "green" aliphatic aldehydes (C5-C7), of which the latter are expected from an oleaginous protein concentrate.Despite purity differences in pea protein isolate and oat protein concentrate, they share 25 abundant volatiles.Additionally, 2,4-decadienal which is considered a less volatile compound, could only be detected in oat protein.While other proteins may contain 2,4-decadienal, their concentrations are likely too low to allow detection by headspace analysis.Furans seem to be abundant in all samples, which may be indicative of thermal degradation 4 and likely formed during heat treatments and/or sterilization of proteins.Aside of furans, aldehydes and ketones are consistently abundant in all four proteins and, consequently, are the target priority of off-flavor removal.

Supplementary information S2: Differences in raw material and processing
Commercial potato protein used in this study shows overall low abundance of most detected volatiles (Figure 1A) with only 6 volatiles (2-heptanone, DMTS, heptanal, benzaldehyde, 2-octanone and 2ethylfuran) found in higher abundance relative to other proteins.Among these volatiles, DMTS ("sulfurous", "onion") was particularly found only in potato protein and it contributed considerably to the total volatiles (Supplementary Figure 2C).The prominence of DMTS agrees with other studies where the compound was reported to be one of the main contributors to cooked potato aroma 5,6 .
Aside from DMTS, the ketones of 2-heptanone ("cheese", "fruity") and 2-octanone ("cheese", "earthy") are other main contributors which further distinguish the volatile composition of potato proteins.Overall, potato protein is still a relatively clean matrix, and this can be associated with its high purity isolation using EBA (expanded bed adsorption) chromatography 7 .Also, a low-fat content might contribute to undetectable off-flavor precursors of unsaturated fatty acids (0.2% fat, Table S2).
In almond protein concentrate, higher amounts of benzaldehyde were observed.This is expected since benzaldehyde is a signature volatile that gives characteristics to almond aroma.Additionally, we observed a high contribution of both alcohols and aldehydes with "green" characteristics such as pentanal, pentanol, hexanal, hexanol, and heptanol.Additionally, all Strecker aldehydes (3methylbutanal, 2-methylbutanal, and 2-methylpropanal) are higher in almond protein.This is in line with other reports 8 and expected from the higher abundance of branched chain amino acids in almond.The latter suggests Maillard reaction that occurs during processing.This protein concentrate is produced by partially defatting almond through mechanical press and subsequent thermal and pneumatical treatment.Around 10-11% unsaturated fat remains (product specification sheet), which is the highest among the other proteins and corresponds to the abundance of various "green" products of lipid oxidation.
In pea protein isolate and oat protein concentrate, 25 volatiles are shared as abundant volatiles despite purity differences.Furan volatiles are particularly more abundant in these proteins, accounting for more than 50% of total scaled volatiles.During processing, moisture was removed from oat and pea proteins using roller-drying and spray-drying, respectively.High formation of furans suggests that temperatures above 110°C were used during drying 9 , allowing Maillard reactions to occur.It remains to be explained how furan formation occurs in pea protein isolate where sugars are barely present (Table S3).However, it was observed earlier that furan abundance increases at 30C storage of pea protein isolate, while its diversity increases with exposure to light 10 .In terms of lipid oxidation products, oat protein concentrate contains roughly 14% unsaturated fat, while pea protein isolate contains approximately 3.4%.This difference is reflected in the formation of green aldehydes, particularly heptanal and nonanal, which are found to be higher in oat protein concentrate.
As described above, plant protein concentrates and isolates typically contain a wide range of molecules.While we have no detailed data on this, it is commonly accepted that e.g.due to seasonal differences variations of the composition occur even between different batches of the same product which is a challenge for product development.For the optimization of biopurification, strains that are already industrially used in food fermentations are attractive to apply.However, such strains might be homofermentative and therefore be low in the expression of alcohol dehydrogenase, resulting in limited removal of aliphatic aldehydes.Slow growth of facultative homofermentative organisms is associated with a shift towards heterofermentative metabolism, which is expected to come with higher ADH expression levels 14,15 .
Such growth rates can be manipulated for instance by changing the type of sugar a strain is cultured on.To understand the effect of different sugars during culture preparation on the biopurification process, pea protein was inoculated with cells of the typically homofermentative strain L. cremoris MG1363 16 at roughly 1E8 cells/mL.Biopurification was performed either with or without the addition of sugar and after a 2-hour incubation period at 30 °C, volatiles were measured.The cells used were precultured with different sugars for a minimum of 15 generations, which allowed metabolic adaptation to different growth rates.
In the absence of added sugar in pea biopurification, volatile reduction was similar across sugar precultures.In the presence of added sugar, volatile reduction became more pronounced, particularly by cells cultured with sugar that led to low growth rates (< 0.4 /h).Up to 20-fold higher reduction of aldehydes in sugar-added pea was observed.However, the formation of pyruvate-derived compounds such as ethylacetate, 2,3-butanedione, and 2,3-pentanedione concomitantly increased with the addition of sugar (Supplementary Figure 4).
In summary a growth rate reduction caused by using different carbon sources allowed to significantly increase the reduction of aldehydes during biopurification with a homofermentative organism. Supplementary

Supplementary information S5: Selection of biopurifying strains
An analysis of the heatmaps in Figure 2 (main text) (Supplementary Figure 5  showing limited reduction of aldehydes.Both seem to be associated with either lower or nonfunctional ADH activity as a result of niche-specific evolution 17,18 . Based on our screening, strain rankings were calculated.For each volatile, log10(fold-change) was weighted by the volatile's sensory score (0,1,2,3,4 -Table S4; Table S5) and multiplied by -1.Higher scores indicate better biopurification.Fold change values were capped at 3 or -3 as the upper and lower limits.Values higher than 3 or lower than -3 were converted to 3 and -3, respectively.For each strain in each substrate, the weighted scores were averaged by chemical groups (e.g., aldehydes, ketones, etc).Subsequently, the average weighted scores per chemical group were added, which resulted in the strain's score in that substrate (Table S6).Total scores in all or individual substrates, combined with pH information were the bases for choices of individual experiments on e.g.technofunctional properties or biopurification optimizations.
The sensory scores were based on the ranking of qualitative descriptions (more or less offensive) as outlined in Table S4.Relevant thresholds or levels to which the concentration of volatiles eventually should be reduced depend on their ratios and the end application.For this reason, sensory evaluation of biopurification was tested in applications mimicking aqueous beverages and a semi-skim milk.
Table S4: Odour active compounds present in the sample which were perceived during GC-O and identified by GC-MS.The score in the first column indicates the potential impact of the volatile compound for off-flavor removal.A higher score indicates that a compound has a higher contribution to the off-flavor and therefore removal is more important.However, aroma perception will be strongly matrix dependent.Scores were estimated by a sensory expert based on known flavor defects as listed in the column description.exhibited phase separation (creaming).Altogether, foaming properties may be slightly improved and other properties remain unaltered by our approach of off-flavour biopurification.

Semi dynamic in vitro digestion
INFOGEST static in vitro digestion was performed as described before 23 .Prior to digestion, pea protein solutions were subjected to biopurification by LAB61, NCDO712, YEAST 20, and LAB 56, as described.Pea protein solutions at 10% (w/w) were incubated in electrolyte simulated salivary fluid 26 in the presence of salivary amylase (Sigma, cat.no.A1031) for 2 min at 37C in a shaking water bath.Following this step, the gastric phase of digestion was mimicked by incubation for 30 min in electrolyte simulated gastric fluid and pepsin (Sigma, cat.no.P7000) while gradually lowering the pH to 2.0 by titration with 10 M HCl.Finally, small intestinal digestion was addressed by adding electrolyte simulated intestinal fluid containing bile extract (Sigma, cat.no.B3883), and pancreatin (Sigma, cat.no.P7545) for 120 min while increasing the pH to 7.0.Experiments were performed as independent triplicates and enzyme-containing samples in the absence of protein and non-biopurified pea protein solutions were taken as controls.

Mono and di-saccharide analysis
Neutral mono and di-saccharides were determined by high performance ligand exchange chromatography (HPLEC) with refractive index (RI) detection.Proteins and fat in the sample were precipitated by Grimbely Biggs reagents and removed after centrifugation.25 µL supernatant was injected on a mixed bed guard in series with two Aminex HPX-87P, 300 x 7.8 mm analytical columns (Bio-Rad).The neutral mono and di-saccharides were eluted isocratic with Milli-Q water with a flow rate of 0.400 mL/min.The column oven was held at a temperature of 80°C.Data analysis was done with Chromeleon software version 7.2.9 (ThermoFisher Scientific).Quantitative analyses were performed by using standards of the neutral mono and di-saccharides (Sigma-Aldrich).

B
72h -yeastsSupplementary Figure 3: Fold-change (Log10) of volatiles in biopurified pea samples compared to sterile unfermented samples.Protein solution was inoculated with 100-fold diluted early stationary cells and incubated for 24 hours or 72 hours at 30°C for LAB (panel A) or yeast (panel B), respectively (each bar represents a different strain; n=1).When indicated, Glucose (1%) was added (bars from left to right correspond to each other for samples with and without glucose).

Figure 4 :
Fold-change (Log10) of volatiles in biopurified pea samples in comparison to sterile unfermented samples.Biopurification was performed with roughly 1E8 cells/mL for 2 hours at 30°C.Cells were precultured for a minimum of 15 generations in sugars indicated on the left.Subsequently, cells were harvested at stationary phase, washed, and added to pea protein solution (10% dry matter) with or without the corresponding sugar added during preculture.Yaxis shows growth rate (/h) and 3-letter initial of sugar added during preculture: glucose (Glu), mannose (Man), fructose (Fru), sucrose (Suc), maltose (Mal), galactose (Gal), trehalose (Tre), and mannitol (Mnt).Error bars show standard deviation of the mean (n=3).The bottom panel is identical to Fig.4and shown here for completeness.

Table S1 :
GC-MS peak areas [log10 transformed] (unprocessed) of detected volatiles in sterile,unfermented plant proteins.The mean and standard deviation of 3 biological replicates are shown.

Table S2 :
General chemical analysis of plant proteins used in this study

Supplementary information S3: Analysis of neutral mono, di-saccharide in plant proteinTable S3 :
High Performance Ligand Exchange Chromatography (HPLEC) detection of sugars in plant protein powders.Duplicate measurements are reported individually.

Table S5 :
GC-Olfactometry (GC-O) results reflecting the aroma active compounds after biopurification.Compounds are listed in the order of eluting from the GC.

Table S6 :
Top 10LAB and yeast strains, weighted score of volatile neutralization, and final pH in plant proteins.The pH of sterile, unfermented solutions are: 6.4 (almond and oat), 7.2 (pea), and 6.0 (potato).