Metabolomics and sensory evaluation of white asparagus ingredients in instant soups unveil important (off-)flavours

Highlights

• Upcycling asparagus waste streams into flavour-rich ingredients is feasible.
• Spray-dried powder had similar flavour to flavour-supplemented oven-dried powder.
• Dimethyl sulphide is confirmed as key asparagus odorant.
• 3-Methyl-1-butanol was highly correlated with asparagus odour.
• Asparagus fibre negatively influenced the taste and mouthfeel of soup formulations.

Abstract

Split-stream processing of asparagus waste stream is a novel approach to produce spray-dried powder and fibre. Asparagus ingredients processed by this method and a commercial asparagus powder were compared by evaluating their flavour profile in a soup formulation. Professional sensory panel and untargeted metabolomics approaches using GC–MS and LC–MS were carried out. Unsupervised and supervised statistical analyses were performed to highlight discriminatory metabolites and correlate these to sensory attributes. The spray-dried powder scored higher on asparagus flavour compared to the commercial powder. The fibre negatively impacted the taste and mouthfeel of the soups. GC-O-MS confirmed the role of dimethyl sulphide, 2-methoxy-3-isopropyl pyrazine and 2-methoxy-3-isobutyl pyrazine in asparagus odour. Seven new volatile compounds are also proposed to contribute to asparagus flavour notes, most of which were more abundant in the spray-dried powder. This research demonstrates the feasibility of upcycling asparagus waste streams into flavour-rich ingredients with good sensorial properties.

2. Materials & methods

2.4. Sensory evaluation

The sensory evaluation was carried out at the facilities of Essensor BV (Wageningen, The Netherlands). The sensory panel was selected from the top 10 % of the population after screening on sensory abilities and sensitivities following the ISO 8586 criteria. Ten selected professional panellists first became fully acquainted with the six soup prototypes during four training sessions, to be acquainted with the samples without knowing the composition of ingredients, before the final descriptive evaluation. The sessions were carried out on separate days and each started at 10:00 AM. A set of 24 attributes was determined for the soups during the four training sessions, which covered odour, taste, mouthfeel and aftertaste attributes (Table S2).

During the final descriptive sensory evaluation, the six soup prototypes were served to the panellists twice in randomized order. Each sample was labelled with a unique three-digit code. The serving temperature was 65–70 °C. The panellists first evaluated the soups on the odour attributes. Subsequently, each sample was tasted at 60 °C to evaluate the taste, mouthfeel and after-feel attributes. Each panel member used thermometers to confirm and monitor the temperature. Evaluation scores were within the range of 0–100.

2.5. Metabolomics

To profile the volatile and non-volatile chemical composition of the six soup prototypes, 15 mL was taken from each soup on each training sensory session day as well as the final descriptive evaluation and stored at −20 °C. After the final descriptive evaluation was complete, all samples were transported to the lab on ice. Once fully defrosted, these were placed in a water-bath at 70 °C to mimic the temperature as served to the panellists. Afterwards, they were respectively aliquoted in glass vials and Eppendorf tubes for further analysis using SPME GC–MS and LC–MS as described below. All samples were analysed in a single sequence per metabolomics platform, in randomized order.

2.5.1. Analysis of volatile compounds

To profile the volatile compounds, 1 mL per soup replicate was pipetted in a 10 mL ND18 headspace screw glass vial (BGB®, Germany) and vials were closed with ND18 magnetic screw caps (8 mm hole) with Silicone/PTFE septa (BGB®, Germany). Before extraction, each sample was preconditioned at 65 °C for 10 min agitating at 350 rpm, to release the volatiles to the headspace mimicking how the panellists smelled the samples. Volatiles in the headspace were extracted at 65 °C for 10 min without agitation and absorbed onto a Polydimethylsiloxane/Divinylbenzene/Carboxen 50/30 μm diameter, 1 cm length fibre (Supelco, PA, USA). After extraction, SPME fibres were desorbed onto the GC–MS by heating the fibre at 250 °C for 2 min. The GC–MS analysis settings were as previously described (Siccama, Pegiou, Eijkelboom, et al., 2021). A mixture of all samples was used for each quality control (QC) sample. QCs were analysed in the same way as all biological samples and were distributed along the analysis series. A range of n-alkanes (C₆ – C₂₁), prepared from a set of stock solutions of the individual alkanes, was analysed in the same way to calculate retention indices.

2.5.2. Analysis of non-volatile compounds

To profile the non-volatile compounds, ultra-performance LC–MS was performed. The semi-polar compounds were extracted by mixing 0.3 mL of each soup replicate with 0.9 mL 32.04 M methanol and 0.035 M formic acid followed by sonication and centrifugation, as described previously by De Vos et al. (2007). The LC–MS calibration and analysis settings were as previously described (Pegiou et al., 2021) using both negative and positive ionization modes of the Q Exactive™ Plus Hybrid Quadrupole-Orbitrap™ Mass Spectrometer (Thermos Fisher Scientific™, Germany). A 0.3 mL from the same mixture of all samples as prepared and mentioned in 2.5.1, was used for each QC sample and analysed in the same way as the biological samples.

2.5.3. Untargeted metabolomics data processing workflow

All GC–MS and LC–MS data were processed following the untargeted metabolomics workflow centred around the software packages MetAlign and MSClust as described before (Pegiou et al., 2021). The obtained relative abundances of the reconstructed metabolites in the processed data were log-transformed and a correction for signal drift was carried out, based on the QC samples (Wehrens et al., 2016).

2.5.4. Metabolite identification

Volatile metabolites were identified based on matching the reconstructed mass spectra and calculated RIs with authentic reference standards and those present in the NIST17 Mass Spectral Library and in-house databases. Non-volatile compounds were putatively identified based on matching their molecular ion mass and associated in-source fragments with the detected LC-MS asparagus compounds described by (Pegiou et al., 2021) and the online databases KnaPSAck (https://www.knapsackfamily.com/) and mzCloud (https://www.mzcloud.org/). The given level of identification (LOI) follows the guidelines of the Metabolomics Standards Initiative (Sumner et al., 2007).

2.6. Multivariate statistical analysis

The log-transformed metabolomics data were mean-centred, Pareto-scaled and variation between samples was initially explored by applying PCA. PCA analyses were performed using the R package ropls (Thevenot, Roux, Xu, Ezan, & Zunot, 2015). Random Forest was applied for the supervised analysis and variable selection, considering the composition of the data matrices obtained. The bootstrapping method, which is applied during random forest analyses, converges to the leave-one-out cross-validation method which is suitable for datasets with a limited number of observations, enabling a higher prediction performance. For the Random Forest analyses, log-transformed data were used and analyses were performed using the R package randomForest (Liaw & Wiener, 2002). The number of trees (ntree) and the number of variables (mtry) for each decision rule were optimized for the minimum prediction error for each model. The Random Forest classification approach was followed to determine those GC–MS and LC–MS compounds indicating the ingredient composition of a specific soup prototype. Two classification analyses were performed per dataset; one based on the primary asparagus component (3 classes: commercial, concentrate, spray-dried) and one based on the presence of asparagus fibre (2 classes: yes, no) (Table 1). The commercial powder was classified as fibre-containing as it is produced from whole asparagus pieces, thus, including the fibres. The performance of each model was assessed by the out-of-bag (OOB) error rate which corresponds to the prediction error after leave-one-out cross-validation. Variable importance was estimated by the mean decreased accuracy which expresses how much accuracy the model loses by excluding each variable. Compounds with Mean Decrease Accuracy > 1 were considered relevant. Hierarchical clustering (HCA) of the soups was subsequently performed focusing on the selected variables after log-transforming and autoscaling their abundances, and this was visualised in a heatmap. The HCA analysis and visualisation were performed using the R packages pheatmap and ggplot2 in RStudio with R version 4.1.1 (2021-08-10). The Random Forest regression approach was followed to determine which individual compounds have a predicting relevance to sensory attributes having a QI > 0.65 and p-adjusted < 0.05 (Table S2). The performance of each model was assessed by the mean square error (MSE). Variable importance was estimated by the increase in mean square error (%IncMSE) which expresses how much the percentage of prediction error increases by excluding each variable. Compounds with %IncMSE > 1.50 were considered relevant. Microsoft Excel and PowerPoint (version 2104, 2021) were used for additional data analysis and visualisation.

2.7. GC-Olfactometry-MS

The regression analyses highlighted 18 volatile compounds as being correlated to specific sensory attributes of which eight could be unambiguously identified (level 1). A mixture of the reference standards of these eight compounds was analysed with GC-O-MS, to determine whether these volatiles are odour-active within the concentration range detected in the soups without the extra addition of the flavour mix. A GC column splitter outlet was used to split 1:1 towards the MS and the olfactory detection port (ODP2, Gerstel, The Netherlands). The reference standards were dissolved in methanol and were used to prepare a solution in water comprising dimethyl sulphide (5 µg/mL), pentanal (5 µg/mL), 1-hexanol (5 µg/mL), 1,3-dimethylbenzene (0.2 µg/mL), 1-octen-3-ol (0.5 µg/mL), octanal (0.05 µg/mL), (E)-2-heptenal (0.2 µg/mL) and 2-methoxy-3-isopropyl pyrazine (0.2 µg/mL). The same SPME GC–MS conditions as described in section 2.5.1 were used except that the GC oven temperature program was adjusted to shorten total run time to 24 min by increasing the ramp to 20 °C/min after 15 min. For sniffing and assigning the individual peaks to a specific aroma attribute (if any), three assessors were recruited within the laboratory. Each assessor smelled the GC-O profile of the prepared standard solution and noted down the perceived aroma per compound at the given retention times without knowing which compound corresponds to which peak.

2.8. Physical characteristics

Selective physical properties of the dry powders and soup prototypes were analysed. The particle size distributions of the commercial asparagus powder, spray-dried powder and asparagus fibre were measured using a Mastersizer 3000 analyser (Malvern Inc, Malvern, UK) with the dry powder disperser Aero S. The size distributions of the commercial asparagus powder and asparagus fibre were analysed in the non-spherical analysis mode using the refractive index (RI) of cellulose, i.e. 1.468, since cellulose was considered the most abundant compound. The spray-dried powders were analysed with the spherical analysis mode using the RI of maltodextrin, i.e. 1.670. Furthermore, the particle size distributions of the soup prototypes were analysed in the Hydro MV module and the Mastersizer 3000 with the non-spherical analysis mode. The RI of cellulose (1.468) was used for the dispersed phase and 1.330 for the continuous phase (water). In addition, the particle size distribution of pure starch dissolved in hot water was measured using the spherical analysis mode and the RI of starch (1.450).

The viscosity of the soups was determined with the Anton Paar rheometer (MCR301, Anton Paar GmbH, Graz, Austria) with a concentric cylinder geometry (CC-27). A shear rate sweep with a logarithmic increasing shear rate from 1 to 1000 s⁻¹ was performed. The rheology measurements were performed at 40 °C, which is assumed to be the relevant temperature inside the mouth before swallowing the soup (Deblais et al., 2021). The viscosities of the soups at 50 s⁻¹ are reported (Table 1). This shear rate has been adopted as the oral shear rate standard by the National Dysphagia Diet task force and is considered a reasonable order of magnitude for swallowing liquids (Ong et al., 2018, Popa et al., 2013).

Microscopy images of the soups were taken using a light microscope (Carl Zeiss AxioScope, Jena, Germany) after the soup samples were vortexed to ensure homogeneity. A drop of the sample was placed on a microscopic slide under a coverslip. The images were captured (AxioCam MRc 5 camera) at a 20x magnification.

5. Conclusion

This sensory descriptive analysis showed that soup prototypes prepared with spray-dried asparagus powder made from concentrated asparagus juice had similar flavour notes to those prepared from a flavour-supplemented commercial powder. Adding asparagus fibre negatively affected the flavour and mouthfeel of the soup prototypes. Using advanced metabolomics tools, a chemical characterization of the soups was performed and the datasets of the two analyses (sensory and metabolomics) were fused and examined following Random Forest approaches. Not only key known asparagus odorants were highlighted and confirmed, but we also suggest new compounds with potential relevance for the sensory profile of the asparagus ingredients. In conclusion, this study has revealed that spray drying of asparagus concentrate is a promising processing method to produce flavour-rich asparagus powder, as compared to the conventional oven-drying process. A similar process may be tested to upcycle other vegetable waste streams to produce flavour-rich food ingredients, in turn contributing to the sustainability of food systems. Ultimately, this could reduce the usage of supplemental flavourings in food products to make them more natural.