Extraction strategies integrated with digitalization processes for the characterization of proteins from broad bean and red beetroot waste leaves

Highlights

• •Ultrasound-assisted extraction (UAE) was integrated with Internet of Things (IoT)
• •Underutilized plant waste leaves are affordable sources of high-value compounds
• •Leaves of broad bean and red beetroot represent exemplary starting material
• •Food-grade suspensions enriched in amino acids, peptides and proteins were produced
• •Advanced UAE-IoT represent accomplished setup for forthcoming green extractions

Abstract

A real-time monitored Internet of Things (IoT)-integrated ultrasound-assisted extraction laboratory-scale prototype was developed and evaluated as a green and scalable processing platform for the recovery of high-value compounds from food-grade plant by-products. The key novelty of this study lies in the integration of IoT-based process monitoring and control with ultrasound-assisted extraction, enabling reproducible and energy-efficient recovery of proteins and bioactive compounds from plant matrices. Leaves of Vicia faba L. (broad bean) and Beta vulgaris L. (red beetroot), characterized by a protein content exceeding 30% of dry matter, were selected as representative raw materials. Ultrasound-assisted aqueous extraction yielded protein-rich extracts with mildly acidic pH values (5.94–6.81) and low electrical conductivity (0.043–4.355 mS/cm), indicating minimal matrix degradation and favourable processing conditions.

The extracts contained water-soluble flavonoids, free amino acids, and proteins at concentrations suitable for further food and bioproduct applications. Proteomic profiling by nanoLC-MS/MS revealed the presence of ribulose-1,5-bisphosphate carboxylase/oxygenase as the dominant chloroplast protein, alongside hundreds of additional water-soluble proteins originating from multiple plant cell compartments, confirming the broad extraction capability of the ultrasound-assisted system. Continuous IoT-based monitoring ensured stable operating conditions and process reproducibility throughout extraction. The results demonstrate that the proposed IoTenabled ultrasound-assisted extraction prototype represents an effective green technology for the valorisation of plant-based by-products and shows strong potential for process optimization and scale-up toward industrial applications.

1. Introduction

Over the past few decades, the growing need for digitalization has been reinforced by, of course, the economic indicators of flexible businesses [1], specifically in the integrated food systems [2], but also by scientific goals of the research, development and innovative society including academic community [3], [4]. It is reasonable to suppose that information technology and biotechnology/food technology are the two of the most innovative industries today, and through close collaboration between experts from these two key technologies, numerous time-consuming complexities across the (bio)processes development could be carried out in a faster and simplified manner. Moreover, mature innovative (bio)processes integration in the biorefinery concept [5], based on utilization of biomass materials as a feedstock, can be greatly improved by the digitalization and, consequently, achieve desirable technical, socio-economic and environmental effects. Mass agricultural production and food processing offer a quite number of potential plant-based feedstocks acceptable to the end-consumer, especially in the case of much needed but insufficiently optimized manufacturing of protein concentrates to be used in food and pharmaceutical industries [6], [7], [8], [9], [10]. Up to now, advanced green protein extraction methods particularly from plant biomass and its by-products by ultrasound (US)-assisted methods, and ionic liquids and deep eutectic solvents with high protein extraction yield and recovery rate are operable as separated non-automated processes at the laboratory level [11], [12], [13], and more prosperous digitalized scaled up extraction structure are in demand [14], [15]. There are several legitimate reasons for the lack of such an integrated digitalized system for green protein extract production, and here are some of those reasons that illustrate the complexity of the system: the availability of plant by-products is determined by geographical and climatic conditions, often is seasonal and it requires smart management; the heterogeneity of these potential raw materials, especially their physical and chemical characteristics, e.g., their consistency and viscosity when ground and suspended in a particular solvent, and concentration of desirable but also undesirable compounds, largely rules upstream processing and more broadly − the entire (semi-)industrial production line; then design and concept of central (bio)process, its monitoring strategy as well as quantity and diversity of collected data has nonlinear character and affects greatly final products safety, quality, and stability. To fulfil such manifold tasks, a specific infrastructure is required − the Internet of Things (IoT), which can be defined as a unique system of interconnected computing devices, mechanical and digital objects, and operators that are assigned unique identifiers and the ability to transfer data over a network [14]. Real-time process monitoring and automated modifications that can improve new, smart, sustainable, reproducible and rational protein extractions and plant-based products with optimized costs are just some of the outstanding criteria that make the IoT inevitable. Therefore, this research addresses the complexity of such a system with the main goal of integrating the central process − the green protein extraction from waste leaves with the IoT for characterization of resulting extracts, and, to our knowledge, this type of integration has been carried out for the first time. To achieve this, we have set several goals, as follows below.

One of the goals of this research was to design, develop and test a digitalized process line using leading green extraction – ultrasound (US)-assisted extraction as non-thermal (NT) processing (P) technology, connected to the IoT, and prepare an accomplished ready-to-use laboratory scale IoT-US system. Here effect of the high-power US (the NT-P part of the NT-IoT-P system) on the IoT part i.e., sensors of the system have to be taken into account because US-generated physical stress can damage the sensors, and therefore a smart two-chamber IoT-US system with separated extraction unit and sensor unit was designed. Then, real-time sensors for pH (−), temperature (°C), oxidation–reduction potential (mV), electrical conductivity (mS/cm), and energy consumption (kWh) were connected via local network (LAN) to a central controller that sends data to the cloud (DEEP cloud) every second. The operator can monitor and save the data via a web application, which allows for reproducibility and subsequent optimization of the protein extraction and experimental design, and create and manage mandatory analyses.

The second goal of our work was to assess the applicability of water as a green solvent (without the addition of any other chemicals) and the significance of the US alone in the extraction of proteins from waste leaves, without the assistance of elevated temperature (≤ 35 °C). Several types of affordable leaves, e.g., sugar beet (Beta vulgaris L.) leaves, broad bean (Vicia faba L.) leaves, and red beetroot (Beta vulgaris L.) leaves, were collected, dried, grounded and sieved to obtain powder fractions with different particle sizes (fraction 1, particle size <500 µm, fraction 2, particle size 500–1000 µm, and fraction 3, particle size >1000 µm). All leaves powder suspensions were prepared in an identical manner in water at room temperature, and different ratios of solids to liquid were tested. The aim of this preliminary experiments was preparing a suspension with as high a solids fraction as possible and extract as much protein as possible at the same time making the suspensions suitable for undisturbed pumping through the two-chamber IoT-US system. Firstly, particles of all fractions 1 tend to remain on the surface of the aqueous suspensions, and significant additional effort needs to be made when mixing these suspensions in water to make them homogeneous. Then, so to say a more favourable viscosity and easier flow of the obtained aqueous suspensions with a ratio of solid to liquid equal to or less than 1:16.7 (w/v) (3.0–10 g of leaf powder in 50 mL of water), through the laboratory IoT-US system, were achieved with the fraction 2 of the broad bean leaves and the red beetroot leaves than with the other fractions of all tested types of leaves, and especially with all three fractions of the sugar beet leaves, whose characterization is not the subject of this research. For that reason, the IoT-US extraction of proteins from the fraction 2 of two food-grade vegetable by-products − broad bean leaves and red beetroot leaves were selected for development of such user-friendly IoT-US web application. With both fractions 2 water was used at a solid to liquid ratios from 1:5 to 1:16.7 (w/v) to create operational suspensions, which were intended to have smooth flow through the IoT-US system, and the US is applied briefly (up to 9 min, with amplitude up to 100%) while the temperature of water-based extraction suspension should remain below 35 °C, thus, the extraction is physically non-thermal process.

Results of well-established thermal protein extractions and protein separations by organic and other solvents were usually employed as indicators of efficacy of novel green extraction procedures. The potency of the IoT-US extraction was compared by three additional control types of protein extractions performed: (i) IoT-thermal (IoT-H) extraction from the fraction 2 of both leaves, (ii) Osborne-based extraction [16] from the fraction 2 of the reed beetroot leaves assisted by the US and H, and (iii) cold acetone-assisted extraction from the fraction 1 of both plant by-products, carried out at 4˚C without support of the US or the H (Fig. 1). The IoT-H was performed by applying heat (H) instead of the US, with temperature of the water-based extraction suspensions ≤40 °C, and for period up to 9 min (duration of the IoT-H extractions and the IoT-US extractions were the same). The Osborne-based extraction was selected as a classical sequential protein extraction with water, NaCl, ethanol and NaOH, here shortened from 16 h, as described previously [16], to 9 min in each step. Four steps of the Osborne-based extraction were upgraded by application of the IoT-US or the IoT-H under (almost) the same conditions because the temperature of the extraction suspensions was ≤35 °C and ≤40 °C, respectively, and duration of the extraction procedures was rather short (9 min). Our preliminary results showed that proteins are more difficult to extract with water from the red beetroot leaves than from the broad bean leaves by assistance of both treatments − the IoT-US and the IoT-H, so our hypothesis was that the sequential salt/alcohol/alkali-mediated (the Osborne-based) extraction of proteins from the red beetroot leaves only with the assistance of the US or the H would be more successful and thus highlight the (potentially negligible) effect of the US. The cold acetone-assisted extraction represents an understated organic precipitation of proteins, and the fraction 1 of both leaves was used to test whether particle size reduction would result in increased yield of extracted proteins, which puts additional pressure on the efficiency of the IoT-US extraction system. So, this design (Fig. 1) includes relevant control extractions: (1) the extraction of proteins in water and treatment of the water-based suspension with the US vs the extraction in the water and the treatment of the water-based suspension with the H, then (2) standard extraction of proteins with water/salt/alcohol/alkali supported with the US or the H, which allows a critical evaluation of the contribution of water as effective green solvent, and (3) the organic protein precipitation without boost by the US, to assay augmentation of the extraction by the US.

Fig. 1.

Approaching strategies employed for the extraction of proteins from the waste leaves. Internet of Things, IoT; H, heat (H)-assisted extraction; US, ultrasound-assisted extraction. Symbols and refer to treatments of suspended solids using extraction parameters highlighted in the IoT-H and the IoT-US extractions.

Third goal of this investigation was to characterize resulting extracts and determine if the prepared extracts are suitable for use in the food and the pharmaceutical industry. Thus, content of metals and non-metals, proteins and amino acids as well as other highly valuable extracted compounds were estimated. As an additional criterion, identity of the extracted proteins with focus on ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), was chosen.

This investigation contributes to straightforward routes for advanced digitalized green extraction of proteins in the green solvent − water from economical plant by-products and other agro-industrial sources. Food-grade aqueous extracts with a high yield of expensive compounds can be formed by such scalable and sustainable systems, and function as a starting raw material to be used further, presumably in upstream and also in downstream processing in all bio-based industries.

2. Materials and methods

2.1. Materials

Broad bean (Vicia faba L.) leaves, collected in June 2024, were donated by the company NutriS Ltd. (Zagreb, Croatia). Red beetroot (Beta vulgaris L.) leaves, harvested in September 2024, were provided by the company Naturala Ltd. (Fortenova group, Zagreb, Croatia).

2.2. Experimental methods

Triplicate experiments and analyses were performed unless otherwise stated. Determined content of metals and non-metals in the bblp and the rblp (Table 1), concentration of protein (Supplementary Table S2), and content of chlorophyll a, chlorophyll b, and carotenoid (Supplementary Table S8) were expressed as the mean values ± standard deviation (SD). Concentration of amino acids in the six selected bblp and rblp extracts by HPLC method (Table 2) and RuBisCO large subunit by LC-MS/MS was assayed in duplicate with maximal deviations of <3% for both analyses.

Table 1.

Content of metals and non-metals in the bblp and the rblp.

Metals and non-metals	bblp	rblp
	content (mg/kg dm)
Zinc (Zn)	82.506 ± 2.370	16.185 ± 0,540
Arsenic (As, total)	0.106 ± 0.010	0.127 ± 0.020
Lead (Pb)	0.164 ± 0.010	0.207 ± 0.020
Cadmium (Cd)	0.038 ± 0.010	0.927 ± 0.010
Mercury (Hg, total)	0.015 ± 0.000	0.040 ± 0.010
Tin (Sn)	<0.010	<0.010
Cooper (Cu)	13.795 ± 0.270	9.352 ± 0.180
Magnesium (Mg)	5181.703 ± 139.910	28856.494 ± 577.130
Phosphorus (P)	4531.963 ± 90.640	2830.480 ± 56.600
Calcium (Ca)	10814.923 ± 27.035	28082.701 ± 70.207
Potassium (K)	22452.446 ± 4.491	24075.084 ± 4.815
Sodium (Na)	1293.560 ± 23.274	8758.346 ± 157.651
Iron (Fe)	169.732 ± 1.867	194.854 ± 2.534
Selenium (Se)	0.060 ± 0.001	0.172 ± 0.002
Aluminium (Al)	412.343 ± 19.793	246.486 ± 11.832
Antimony (Sb)	<0.010	<0.010
Barium (Ba)	6.418 ± 0.116	60.301 ± 0.965
Beryllium (Be)	<0.010	<0.010
Boron (B)	9.415 ± 0.190	10.873 ± 0.218
Cobalt (Co)	0.336 ± 0.004	0.116 ± 0.003
Chrom (Cr, total)	0.553 ± 0.017	0.602 ± 0.018
Lithium (Li)	0.232 ± 0.004	5.707 ± 0.229
Manganese (Mn)	53.002 ± 1.590	107.123 ± 3.214
Molybdenum (Mo)	<0.010	1.565 ± 0.024
Nickel (Ni)	2.459 ± 0.038	0.446 ± 0.007
Silicon (Si)	717.145 ± 7.172	261.880 ± 2.619
Silver (Ag)	<0.010	0.035 ± 0.011
Thallium (Tl)	<0.010	0.095 ± 0.008
Vanadium (V)	0.567 ± 0.021	0.398 ± 0.014

Table 2.

Concentration of amino acids in the six selected bblp and rblp extracts.

Compound	bblp/US/9/100	bblp/H/9	rblp/US/9/100	rblp/H/9	bblp/C	rblp/C
	(µg/mL)
Asxa	189.87	241.85	93.63	56.17	99.88	64.16
Ser	48.56	58.27	30.37	15.34	36.95	29.08
Glxa	66.18	67.27	196.09	151.18	78.48	69.64
Gly	23.41	22.96	33.14	17.27	36.10	28.64
His	16.44	18.29	17.36	10.72	11.30	9.39
Arg	28.18	29.72	44.34	28.54	28.34	25.52
Thr	34.83	38.42	30.58	15.55	32.25	31.20
Ala	42.37	47.15	42.19	21.66	39.80	31.20
Pro	188.70	223.50	34.38	19.39	38.79	25.27
Tyr	36.39	40.82	21.82	11.83	25.38	14.66
Val	46.46	52.60	39.16	21.77	34.20	28.74
Metb	3.10	3.37	6.81	0.00	4.98	2.14
Lys	24.61	22.26	34.17	16.04	46.05	36.12
Ile	28.81	31.41	23.81	12.19	25.85	19.89
Leu	32.41	32.05	38.67	17.23	40.02	29.07
Phe	37.43	42.28	37.44	23.46	20.64	17.15

^aaAspartic acid and asparagine, as well as glutamic acid and glutamine, were not individually quantified and their concentrations are reported as Asx (Asp and Asn) and Glx (Glu and Gln), respectively.

^bMethionine may undergo oxidation during sample preparation, and reported values may deviate from the actual concentrations.

2.2.1. Plant material preparation

The broad bean leaves (bbl) and the red beetroot leaves (rbl) were dried at room temperature and atmospheric pressure through June and July i.e., September and October 2024, respectively, with occasional mixing until a constant mass was achieved. The dried leaves were grinded to powder (bblp and rblp) by using a Retsch GM 200 mill (Verder Scientific Austria GmbH & Co. KG, Verder Group, Golling, Austria) at 3000 rpm for 5 s with cut and non-interval options, and then fractionated by using two sieves (Retsch, diameter 200 mm/203 mm, mesh size of 500 μm and 1000 μm; Verder Scientific Austria GmbH & Co. KG, Verder Group, Golling, Austria). So, three powder fractions of leaves were collected: fraction 1 (particle size <500 μm), fraction 2 (particle size 500–1000 μm), and fraction 3, (particle size >1000 μm), and stored separately in plastic closed containers and kept in cool and dry conditions until further use. Two out of the three powder fractions − fraction 1 and 2 were used for characterization (Chapter 2.2.2.) and in extraction (Chapter 2.2.3.).

2.2.2. Characterization of leaves

2.2.2.1. Determination of moisture content

The moisture content of the bblp and the rblp was determined gravimetrically. The standard method involved drying of 3.000 ± 0.001 g sample in a ventilated oven at 103 ± 1 °C (Inko d.o.o., Zagreb, Croatia) for 4 h (Croatian Standards Institute, 2001).

2.2.2.2. Determination of nitrogen content and crude protein estimation

A sample of the bblp and the rblp (1.000 ± 0.001 g) was digested using a block digestion system (FOSS, Hillerød, Denmark) with sulfuric acid, potassium sulphate, and a copper catalyst. After digestion, ammonia was released by adding sodium hydroxide, distilled (Foss Kjeltec 8200, FOSS, Hillerød, Denmark), and collected in boric acid. The distillate was titrated with standardized hydrochloric acid, and nitrogen content was calculated using the standard method (HRN EN ISO 5983–2, 2008). Crude protein was obtained by multiplying the determined nitrogen content by a conversion factor of 6.25.

2.2.2.3. Determination of metals and non-metals by inductively coupled plasma mass spectrometry

Samples of the bblp and the rblp (approximately 1.000 ± 0.300 g) were transferred into TFM vessels with deionized water and concentrated nitric acid, and digested using a microwave digestion system (Milestone Ethos, Milestone Srl., Italy) by following the manufacturer’s protocol. After cooling, digested solutions were diluted to 20 mL and homogenized. Aliquots were analysed on inductively coupled plasma mass spectrometry Agilent 7900 ICP-MS (Agilent Technologies, Santa Clara, USA) (ICP-MS, Supplementary Table S1).

2.2.3. Extractions

Four different extraction set ups were carried out: Internet of things (IoT) ultrasound-assisted (US) extraction (IoT-US) and IoT-thermal (or heat, H, IoT-H) extraction (Chapter 2.2.3.1.), Osborne-based extraction (Chapter 2.2.3.2.), and cold acetone-assisted (C) extraction (Chapter 2.2.3.3.). The IoT-US extraction and the IoT-H extraction were performed by using the fraction 2 of the bblp and the rblp, the Osborne-based extraction was conducted by exploiting the fraction 2 of the rblp, while the cold acetone-assisted extraction was managed by utilizing the fraction 1 of the bblp and the rblp. These selected leaves powder fractions − fractions 2 and/or 1 of the bblp and/or the rblp, were chosen to optimize the IoT-US protein extractions in water, as well as to perform rigorous control extractions, in order to evaluate the efficiency of the optimized IoT-US laboratory-scale system, as detailed in the introductory section. Additionally, characteristic viscosity and flow characteristics of the selected bblp and the rblp fractions suspensions in water at relatively high solids to water ratios (from 1:16.7 to 1:5 (w/v)) through the system, also directed described selection (see Introduction). All extractions started by suspending the leaves powder (3.0–10.0 g) in the in-house prepared deionized water or 50 mM K-phosphate buffer, pH 6.0 (both 50 mL), as indicated hereinbelow.

2.2.3.1. IoT-ultrasound-assisted extraction (IoT-US) and IoT-thermal extraction (IoT-H)

2.2.3.1.1. Internet of things

The collected sensor data are first read by a central controller, which forwards the collected data to DEEP cloud storage and processing system. With fully real-time process monitoring, test data were globally available for further analysis and adjustment of test parameters. Once an experimental setup with IoT functionality has been prepared, the next step was to collect data through such a system. Depending on the partners' local communication resources, some types of long-range communication (LAN or WLAN) are used to establish a connection between the IoT system and the central experimental data storage system. The designed IoT-US system concept, which is applicable to all experimental setups of this investigation, uses a completely autonomous and portable measurement system consisting of a control unit, a communication and control interface, and industrial sensors for measuring pH, oxidation reduction potential (ORP), temperature, conductivity, electricity consumption and pressure. Since the operating conditions of US devices do not allow direct placement of measurement sensors in these extraction unit (due to relatively strong ultrasound powers, which can damage the sensors), a system of pumps, tanks and liquid overflows was designed, with which the liquid under measurement is continuously sampled during the protein extraction processing experiments from leaves, without any danger to the measurement system. This measurement system must also have a secured connection to the internet for its operation, through which it establishes a connection with the system for storing and displaying measurement data located in Zagreb (called DEEP cloud). Software support has been developed for the control unit, which ensures simple operation with the measurement system without unnecessary user intervention. Selected sensors were: (1) for pH measurement − JUMO tecLine HD pH combination electrode (pH 0 to 14); (2) for ORP measurement − JUMO tecLine HD Rd redox combination electrode (−1500 to + 1500 mV); (3) for conductivity measurement − JUMO BlackLine CR 4P four-electrode conductivity sensor; (4) for temperature measurement − 1CH DV41 with optical temperature sensor (fiber optic) accuracy +/- 0.2 K; and (5) electricity consumption measurement. The electricity consumption data are used to assess life cycle assessment (LCA), which will be elaborated on in an upcoming publication. All the sensors were connected to a central processing module. Block diagram of the unit is presented in Fig. 2.A. Communication with all sensors was carried out via the local RS485 serial bus and the Modbus RTU protocol.

Fig. 2.

(A) Block diagram of the DEEP cloud central system for receiving and storing data from the remote NT-IoT system (B) test NT-IoT setup, (C) real NT-IoT setup, and (D) block diagram of the NT-IoT solution.

A peristaltic pump was selected for the circulation (suction and return) of the liquid so that the transported medium is only in contact with the hose, for easier cleaning, but also for increased electrical insulation and protection of the entire system. For the purposes of monitoring the experiments, it was concluded that it is sufficient to perform sampling every second, so that all sensors perform measurements once per second. All measurements are stored in the control unit and sent to the DEEP cloud. In the event that the control unit has no connection to the Internet or the DEEP cloud system is unavailable for any other reason, all measurement data is stored locally in the control unit (in the MMC memory) until the first connection to the DEEP cloud is established, after which it is sent without delay. This ensures that the measurement data is protected from accidental loss. Detailed descriptions of the individual parts of the system are given in Supplementary Material.

2.2.3.1.2. Ultrasound-assisted extraction (US) and thermal extraction (H)

A defined mass (3.0–10.0 g) of the fraction 2 of the bblp and the rblp was suspended in deionized water (50 mL, in a glass beaker) at room temperature, then briefly and thoroughly mixed with a glass rod to prepare a homogeneous suspension, and the resulting suspensions were transferred to the extraction unit of the two-chamber IoT-US system. All suspensions prepared in this way were treated by ultrasound (US, Q700 Sonicator; Qsonica, Newtown, CT, USA) over defined period (3.0, 6.0 or 9.0 min) with amplitude of 50, 75 and 100%, power between 60 and 100 W and with temperature of the water suspension ≤35 °C, or by heat (H, maximal temperature of the water suspension ≤40 °C) over given period (3.0, 6.0 or 9.0 min, at 200 rpm; magnetic stirrer, DLAB MS-H-S, Dlab, Beijing, PRC), to create comparable IoT-US and IoT-H extraction procedures, as indicated in Supplementary Table S2. Extracts in bold letters in Supplementary Table S2, together with samples obtained by the cold acetone-assisted extraction − bblp/C and rblp/C (Chapter 2.2.3.3.), were used for further analyses (Chapter 2.2.4.). After defined period of the IoT-US and the IoT-H treatments, resulting suspensions were immediately centrifuged (15 min/ 4 °C / 4200 rpm; Eppendorf 5804 R; Eppendorf, Leipzig, Germany), and obtained supernatants and precipitates removed and stored separately in marked containers. Acquisition of absorption spectra of the obtained supernatants aliquots (4–5 mL, Chapter 2.2.4.1.) was completed immediately after centrifugation, while the remaining aliquots of the supernatants and the precipitates were stored at −20 °C for further analyses: determination of amino acids concentration (Chapter 2.2.4.2.), and protein concentration (Chapter 2.2.4.3.) as well as for separation of extracted proteins (native PAGE and SDS-PAGE, Chapter 2.2.4.4.) and their identification by nanoLC-MS/MS method (Chapter 2.2.4.5.). Quantification of RuBisCO large subunit was performed as described in Chapter 2.2.4.6.

2.2.3.2. Osborne-based extraction

Osborne-based extraction was a combination of a modified Osborne fractionation [16] and the IoT-US and the IoT-H extraction. The suspension of the fraction 2 of the rblp in the deionized water (6% w/v) was treated successively with 5% (w/v) NaCl, 70% (v/v) ethanol and 0.1 M NaOH over 9 min. Each step for the four solvents was performed by treating the suspension by the US (rblp/US/9/100; Q700 Sonicator; Qsonica, Newtown, CT, USA) or the heat (rblp/H/9; 200 rpm, magnetic stirrer; DLAB MS-H-S, Dlab, Beijing, PRC) and followed by centrifugation (45 min/ 4 °C / 4200 rpm; Eppendorf 5804 R; Eppendorf, Leipzig, Germany). Absorption spectra of the eight resulting supernatants were collected (see Chapter 2.2.4.1.) to briefly review extracted compounds.

2.2.3.3. Cold acetone-assisted extraction

Aqueous extracts of the bblp and the rblp fraction 1 were prepared by using 3.0–10.0 g of the powder in 50 mL of 50 mM K-phosphate buffer, pH 6.0 (in a glass beaker), shortly and intensely mixed with a glass rod, and then the glass beaker was left over 30 min embedded in ice (4 °C), and container with ice was placed on a magnetic stirrer (200 rpm, magnetic stirrer DLAB MS-H-S, Dlab, Beijing, PRC). Ensuing suspensions were filtered through filter paper using a vacuum pump, and then obtained filtrates were immediately centrifuged (15 min/4 °C/4200 rpm; Eppendorf 5804 R; Eppendorf, Leipzig, Germany) to remove remaining powder particles from both filters derived from the bblp suspension and the rblp suspension. Aliquots of resulting supernatants were used to determine pH and conductivity, and to record absorption spectra (Chapter 2.2.4.1), while one of the remaining aliquots of the supernatants was transferred to another clean glass beaker and used for protein precipitation by addition of acetone (the aqueous extract: acetone = 1.0:0.67, v/v), and the mixtures were left overnight at 4 °C (the glass beaker was embedded in ice, and container with ice was left overnight in the refrigerator). After centrifugation (45 min/ 4 °C / 4200 rpm; Eppendorf 5804 R; Eppendorf, Leipzig, Germany) two-thirds of the supernatant was removed, while the pellets with the remaining third of the supernatant (bblp/C and rblp/C) were used to further characterization, as described below (Chapter 2.2.4.).

2.2.4. Characterization of extracts

2.2.4.1. Absorption spectra and assessment of chlorophyll and carotenoid content, and total flavonoid content

Absorbance spectra (λ = 230–900 nm) of the extracts were acquired by using 20x diluted extracts in 50 mM K-phosphate buffer, pH 6.0. UV–Vis 96 well microplates were used for the spectra acquisition (spectrophotometer Synergy™ Mx Monochromator-Based Multi-Mode Microplate Reader, Bio Tek Instruments, Inc.). Specific values at defined wavelengths were used for estimation of chlorophyll and carotenoid content. Thus, method described by Zhang et al. [17] was slightly modified and the rblp/US/9/100 and the rblp/H/9 extracts after addition of 70% (v/v) of ethanol over only 9 min were used in the acquisition of absorbance at 665 nm, 649 nm, and 470 nm (UV–Visible spectrophotometer UV-2600, Shimadzu Corporation, Kyoto, Japan). Then, concentration of chlorophyll a (Ca), chlorophyll b (Cb), and carotenoids (Cxc) (mg/mL) were calculated [17]. Total flavonoid content was determined as described previously [18].

2.2.4.2. Amino acids profile

2.2.4.2.1. Sample preparation for quantitative analysis of total amino acids

Selected extract supernatants (Chapters 2.2.3.1. and 2.2.3.3., Supplementary Table S2) were centrifuged for 10 min at 10,000 g, and 8 °C (Centric 200R, Tehtnica, Železniki, Slovenia), and diluted 1:10 with ultrapure water (Synergy® Water Purification System, Merck Millipore, Burlington, USA) containing α-aminobutyric acid (Sigma-Aldrich, St. Louis, USA) as internal standard at a final concentration of 0.05 mM. Derivatization of the dried hydrolysates was carried out using the AccQ•Fluor Reagent Kit (Waters Corporation, 2008) according to the manufacturer’s instructions.

2.2.4.2.2. Quantitative analysis of total amino acids

HPLC analysis was conducted on an Agilent 1260 Infinity system equipped with a G1311B binary pump and a fluorescence detector (Agilent Technologies, Santa Clara, USA). Data acquisition and processing were carried out using OpenLab CDS ChemStation software (version 3.3.42, Agilent Technologies, Santa Clara, USA). Chromatographic conditions are listed in Supplementary Table S3. Amino acids were identified and quantified using an external calibration mixture of derivatized amino acids (100 pmol/µL; Waters), and normalized to the internal standard α-aminobutyric acid.

2.2.4.3. Protein concentration

Concentration of protein was assessed by PierceTM BCA Protein Assay Kit (Thermo Scientific, USA; spectrophotometer Synergy™ Mx Monochromator-Based Multi-Mode Microplate Reader, Bio Tek Instruments, Inc.) and by Bradford method (Bradford, 1976) (spectrophotometer Shimadzu UV-2600i, Shimadzu Corporation, Kyoto, Japan). Standard curves for the protein determination were prepared by using Albumin Standard (BSA) (data not shown).

2.2.4.4. Native PAGE and SDS-PAGE

Native PAGE was performed using a polyacrylamide gradient gel (4–20%, Bio-Rad), and a Tris-glycine running buffer (25 mM Tris, 192 mM glycine). It was carried out at a constant voltage of 180 V and 380 mA. For SDS-PAGE, protein samples were denatured by incubation in Laemmli buffer (final concentration 32.5 mM Tris-HCl pH 6.8, 10% glycerol, 2% SDS, 5% β-mercaptoethanol, 0.005% bromophenol blue), followed by heating at 100 °C for 5 min. SDS-PAGE was accomplished by using Novex™ Tris-Glycine Mini Protein Gels (4–12%, 1.0 mm, WedgeWell™ format, 15-well; Thermo Fisher Scientific, USA). Protein bands were visualized by Coomassie Brilliant Blue staining, following the Wong et al. (2000) protocol [4].

2.2.4.5. Identification of extracted proteins and peptides by nanoLC-MS/MS

2.2.4.5.1. Sample preparation

Proteins were isolated from extract supernatants (Chapters 2.2.3.1. and 2.2.3.3.) following a modified protocol described by Chevallet et al. (2007) [19]. For rblp/US/9/100 and rblp/H/9, the starting material was doubled to compensate for the notably lower protein concentrations. After addition of sodium deoxycholate in the final concentration of 0.5% and brief mixing, proteins were precipitated with trichloroacetic acid (final concentration 8.5%) on ice for 2 h. Pellets were collected by centrifugation (15,000 g, 15 min, 4 °C; Centric 200R, Tehtnica, Železniki, Slovenia), washed twice with ice-cold tetrahydrofuran, and reconstituted in 50 µL of 25 mM ammonium bicarbonate (ABC; pH 7.8). Proteins were reduced with tris(2-carboxyethyl) phosphine hydrochloride (5 mM, 30 min, RT) and digested with trypsin (final concentration 0.02 mg/mL) for 16 h at 37 °C, 600 rpm. Digestion was quenched with 1 µL of 10% formic acid (FA), followed by centrifugal filtration through a 0.2 µm pore size filter (5,000 g, 5 min) to remove residual impurities. Tryptic peptides were dried in a vacuum centrifuge (Concentrator plus, Eppendorf) and stored at −80 °C until analysis.

2.2.4.5.2. Nanolc-MS/MS analysis

Dried peptides were reconstituted in 20 µL of 0.1% FA and separated on the UltiMate™ 3000 RSLCnano (Thermo Fisher Scientific) system coupled to a TripleTOF 6600+ (SCIEX, Framingham, MA, USA) mass spectrometer (MS), equipped with a nano-electrospray ionization (nano-ESI) source. Summary of LC-MS/MS Parameters, including chromatographic conditions and ion source settings are listed in Supplementary Table S4. The MS/MS spectra were acquired using Data‑Dependent Acquisition (DDA) mode, with the top 50 precursor ions with the highest intensities in each MS scans selected for MS/MS fragmentation. Rolling collision energy was used for fragmentation.

2.2.4.5.3. Protein identification

Obtained MS/MS spectra were processed with Analyst software (version 1.8.1, SCIEX) and analyzed using ProteinPilotTM Software (version 5.0.2, SCIEX). Experimental spectra were matched to theoretical spectra derived from the Vicia faba reference proteome for samples bblp/US/9/100, bblp/H/9, and bblp/C, or the Beta vulgaris subsp. vulgaris reference proteome (accessed on 15. 05. 2025.) for samples rblp/US/9/100, rblp/H/9, and rblp/C. False discovery rate (FDR) analysis was performed to ensure identification reliability.

2.2.4.6. Quantification of RuBisCO large subunit by LC-MS/MS

2.2.4.6.1. Sample preparation

Quantification of RuBisCO large subunit was performed according to Dukić et al. [6] with several adjustments, as follows. Protein extracts (Chapters 2.2.3.1. and 2.2.3.3.) were centrifuged (10,000 g, 10 min, 8 °C) to remove residual cell debris, and supernatants were diluted 5-fold with ultrapure water to a total of 1 mL. Samples were prepared in two sets: one for the RuBisCO quantification, and the other for calibration curve matrix simulation. Protein precipitation, reduction and digestion were performed as described in the Chapter 2.2.4.5.1. Filtered tryptic peptide solutions were diluted 1:1 with ultrapure water. Matrix simulation samples followed the same workflow but were digested with Glu-C (final concentration 0.02 mg/mL) at 37 °C, 600 rpm, for 16 h. Synthetic peptide AQAETGEIK (purity >98%, Thermo Scientific) was used for calibration curve preparation (Supplementary Fig. S1.) and the RuBisCO quantification was based on the AQAETGEIK.

2.2.4.6.2. nanoLC-MS/MS analysis

Parameters for Multiple Reaction Monitoring (MRM) were adjusted according to [6]. Sample analyses were conducted on a 6460 Triple Quad LC-MS system (Agilent technologies, Santa Clara, CA, USA) equipped with an ESI source. Chromatographic conditions and ion source settings are listed in Supplementary Table S5. The MRM transition list is given in Supplementary Table S6. The spectra were analyzed using Agilent MassHunter Work station software (Agilent technologies, Santa Clara, CA, USA).

3. Results and discussion

3.1. Characterization of leaves

In this research plant by-products − broad bean (Vicia faba L.) leaves (bbl) and red beetroot (Beta vulgaris L.) leaves (rbl) were used as a promising starting material in several regards [20], [21], but primarily as affordable sources of plant-based proteins. The leaves were characterized including the moisture content as well as the nitrogen content and, based on the experimental data, the dry weight was estimated to be 9.23 ± 0.05% for the bbl and 9.36 ± 0.06% for the rbl, while portion of proteins in its dry matter assessed via the nitrogen content was 35.0 ± 0.04% and 31.0 ± 0.02%, respectively. These levels are comparable with those of commonly consumed greens such as spinach (2.9% of proteins in a fresh matter) and amaranth (4.0% of proteins in a fresh matter) [22]. The protein content of leafy vegetables varies widely across species due to physiological and environmental factors, and can be efficiently withdrawn, as it was shown for cabbage and cassava leaves and bamboo shoots, where up to 53% of the leaves proteins was extracted [23], [24]. Highly valuable leaves protein concentrates have already been obtained from spinach, sugar beet and moringa [9]. In this context, Jerusalem artichoke leaves [25] and tobacco leaves clearly were of less importance with up to 21% of proteins by weight [26]. Suitability of the bblp and the rblp for manufacturing of innovative food-grade commodities was confirmed by the ICP-MS analysis of metals and non-metals (Table 1), as defined by the Commission Regulation (EU) 2023/915.

Table 1.

Hence, it was of prime importance to measure content of arsenic, lead, cadmium and mercury, and collected data precisely shows its load in the both samples within acceptable food safety limits. Distinct metal and non-metal profiles in the bblp and the rblp reflects species-specific physiology and environmental influences. In the bblp and the rblp the highest content of calcium, magnesium and potassium were observed, then phosphorus, iron, manganese and zinc. These findings highlight the nutritional potential of both leaves and underscore the importance of monitoring elemental accumulation in edible plant materials [27], [28].

3.2. Characterization of extracts

The Internet of Things (IoT) was integrated with the ultrasound (US) extraction unit employed in the green extraction of high-value compounds from the bblp and the rblp, as described in the Supplementary material. In short, the DEEP cloud system successfully collected and stored measurement data from remote instruments in a structured format, ensuring data integrity. During experiments, measurement parameters were continuously transmitted to the DEEP cloud in one-second intervals, where they were automatically stored with time stamps. A dedicated web application was developed to extend the functionality of the DEEP cloud by enabling users to create and manage analyses. Through its interface, users defined the analysis time range and automatically retrieve the relevant data from the cloud. The application provided tabular overviews of the measured parameters, including pH value (−), temperature (°C), redox potential (mV), and electrical conductivity (mS/cm). Data export to Excel format was implemented to facilitate further statistical processing (Supplementary Table S7). The IoT-US extraction was highly efficient in isolating proteins and other compounds of interest (see below) was performed and compared to the non-green IoT-thermal (heat, H) extraction (the IoT-H; Chapter 2.2.3.1.) in terms of concentration of extracted compounds in resulting aqueous extracts. In addition to the two IoT-integrated extraction procedures, the Osborne-based fractionation of proteins assisted by the IoT-US and the IoT-H (Chapter 2.2.3.2.) was performed to rationalize use of water as an extractant, and also the cold acetone-assisted extraction of proteins (Chapter 2.2.3.3.), to emphasize role of the US i.e., the H in the extraction.

It is important to highlight range of the IoT-collected values for the pH, the temperature, and the conductivity of all water-based bblp and rblp extracts. In all cases the temperature of the extracts was below 35.0 °C, pH values were in mild acidic range: 6.00–6.76 units for the bblp/US/9/100, 5.94–6.10 units for the bblp/H/9, then 6.41–6.81 for the rblp/US/9/100 and 6.23–6.58 for the rblp/H/9. The conductivity of the two bblp extracts was in the range 0.043–1.215 mS/cm, while the two rblp extracts had higher conductivity with values between 0.935 and 4.355 mS/cm. There is no comparable data for similar aqueous leaves extracts in available literature.

3.2.1. Absorption spectra and assessment of chlorophyll and carotenoid content, and total flavonoid content

Absorption spectra of all water-based extracts with two main peaks at around 300–380 nm and 240–295 nm, especially in the bblp/US/3/100 and the bblp/US/9/100 extracts (Fig. 3A1), suggest presence of water-soluble flavonoids, and can be overlapped with the absorption spectra of the bblp/H/3 and the bblp/H/9 extracts (Fig. 3A2). So, in this case efficacy of the IoT-US extraction is comparable to the IoT-H extraction.

Fig. 3.

Absorption spectra of: (A) the bblp extracts obtained by the IoT-US extraction (US, A1), the IoT-H (H, primary y-axis, A2) and the cold extraction (C, secondary y-axis, A2); and (B) the rblp extracts accomplished by the IoT-US extraction (US, B1 and B2), the IoT-H extraction (H, B2), and the cold extraction (C, B3).

According to the absorption spectra, the rblp extracts (Fig. 3B) in general contain less water-soluble flavonoids than the bblp extracts (Fig. 3A), and the most intensive bands were obtained for the rblp/US/3/50 and the rblp/H/9 (Fig. 3B1 and B2). In addition, total flavonoids concentration was determined [18] in the four extracts and expressed as gallic acid equivalents per mL of the extracts (GAE; μg/mL). The total flavonoid concentration of 76.98 ± 0.25 and 56.38 ± 0.18 μg GAE/mL in the bblp/US/3/100 and the bblp/US/9/100 extract, respectively, and of 76.17 and 39.12 μg GAE/mL in the rblp/US/9/100 and the rblp/H/9 extracts, respectively, highlight the bblp and the IoT-US protocol as more promising choice when water-soluble flavonoids are in focus. However, the absorption spectra of the bblp/C and rblp/C extracts do not have characteristic absorption patterns, and it is not possible to draw a clear conclusion about extracted acetone-soluble absorbing compounds. Nevertheless, in all these absorbance spectra contribution of extracted phenolic compounds and proteins at around 280 nm was kept in mind, too (see Supplementary Table S2). Then, the Osborne-based fractionation of proteins assisted by the ultrasound (US) and the heat (H) was performed to rationalize use of water as an extractant in the IoT-US and IoT-H extractions (see Chapter 3.2.3.). For this extraction the rblp (rblp/US/9/100 and rblp/H/9) was chosen as a raw material less susceptible to extraction in terms of proteins and flavonoids, as discussed above. But first, in this context, extraction of ethanol-soluble pigments − chlorophylls and carotenoid from the rblp will be commented. In comparison with the Osborne fractionation [16] considerably shorter time (9 min vs 16 h) of extraction with 70% (v/v) of ethanol was used in the case of the rblp extracts. As expected, rather low content of chlorophyll a, chlorophyll b, and carotenoid (0.21–0.54 mg/mL for all three pigments) was estimated (Supplementary Table S8). Absorption spectra of the rblp/US/9/100 and the rblp/H/9 Osborne-based extracts in water (data not shown), then in 5% NaCl, 70% (v/v) ethanol and 0.1 M NaOH (Supplementary Fig. S2) with less than 0.4 mg/mL of proteins, do not give hint about possible extracted compound(s).

3.2.2. Amino acids profile

All in all, 14 amino acids (Ser, Gly, His, Arg, Thr, Ala, Pro, Tyr, Val, Met, Lys, Ile, Leu, Phe), and also aspartic acid and asparagine (Asp and Asn → Asx) and glutamic acid and glutamine (Glu and Gln → Glx), were quantified in the six selected water extracts (Table 2).

Tryptophan and cysteine were not identified in any sample. The highest concentration of Asx and Pro, then Glx, Ser, and Val, and also Ala, Phe and Tyr were detected in the bblp/H/9 extract. Analogous concentration of these eight compounds in the same order was found in the bblp/US/9/100 extract. Obtained data seems to reflect the amino acids profile in the fava bean flour and deep eutectic solvent (DES)-based extracted fava bean protein isolates [12]. Except for the Glx, with its highest concentration in the rblp/US/9/100 and rblp/H/9 extract, in all rblp extracts concentration of amino acids was noticeably lower than in the bblp extracts. Although rather short (30 min) and carried out at low temperature (4 °C), the cold extraction resulted in respectable concentration of extracted amino acids. Actually, this type of extraction pinpoints importance of implementation of the US i.e., the H procedures, as is evident from the results in Table 2.

3.2.3. Quantification and identification of extracted proteins

As previously reported for the fava bean [16], concentration of extracted proteins from the rblp in the water (2.431 mg/mL in the rblp/US/9/100, and 10.540 mg/mL in the rblp/H/9; Supplementary Table S2) was higher than in the Osborne-based extracts − the NaCl fraction, the ethanol and the NaOH fraction of the rblp (data not shown). Therefore, the aqueous extracts were used in further evaluation of the IoT-US in the protein extraction from the leaves. The highest protein concentration was approximated in the bblp/H/3 extract (40.69 mg/mL). Then, between 35.69 and 31.23 mg of protein per mL of extract was estimated in the bblp/US/9/50, the bblp/US/3/100, the bblp/US/9/100 and the bblp/H/9 extracts. Less efficient were the bblp/US/3/50 and rblp/C extractions, and the most ineffective the rest of the IoT-US and the IoT-H rblp extractions (protein concentration in range from 2.43 to10.54 mg/mL; Supplementary Table S2). Previously has been shown that the rblp extracts principally contain less highly valuable compounds − water-soluble flavonoids and amino acids than the bblp extracts. Thus, it seems that the mildly acidic (pH in range between 5.94 and 6.81) bblp extracts, especially the green bblp/US/9/100 extract, is certainly promising candidate for preparation of the plant-based protein-rich food-grade commodity.

The native PAGE and SDS-PAGE gels (Supplementary Fig. S3A and B, respectively) gave valuable insight in status of variety of the bblp and the rblp extracted proteins. Maybe except for the bblp/C extract, in the rest five extracts majority of the protein fail to migrate in the native PAGE gel, and remained at >250 kDa. In the bblp extracts with protein concentration estimated to be >30 mg/mL, weak band at approximately 17 kDa was visualized, as well in the rblp extracts with much lower protein concentration −2.43 and 10.54 mg/mL (Supplementary Table S2). Unfortunately, in several attempts we were not able to achieve better separation of the extracted proteins under the native conditions, and there might be reasonable explanation for that. It can be assumed how exactly the US and the H extraction procedures affected the bbl and the rbl water-extracted proteins, peculiarly ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), and if (re-)aggregation of extracted proteins occurred under used conditions [7], [29]. In the mild acidic conditions (pH in the range 5.94–6.81, Chapter 3.2.) also precipitation of e.g., RuBisCO might be foreseen [30], but not experimentally proven. However, in all sample preparation procedures (Chapter 2.2.4.) the centrifugation was required to remove the precipitate (or protein pellets) and their analysis by the SDS-PAGE was also not successful (data not shown). The two bblp extracts of major interest − the bblp/US/9/100 and the bblp/H/9 (line 13 and line 14, respectively, Supplementary Fig. S3B) were analyzed by the SDS-PAGE and compared to all rblp extracts. Again, in the SDS-PAGE gel in all 14 lines a fair portion of extracted proteins did not migrate and remained at >180 kDa position. However, in all lines more than several prominent protein bands were visualized, notably at approximately 130 kDa, 100 kDa, 70 kDa, then between 55 kDa and 25 kDa, similar to the (DES)-based extracted fava bean protein isolates [12]. Besides numerous extracted proteins, here is quite understandable to presume the contribution of eventually separated RuBisCO subunits − the large subunit (50–56 kDa) and the small subunit (12–18 kDa), primarily in the bblp/US/9/100 extracts. Majority of the extracted proteins in the bblp and the rblp extracts were identified by the nanoLC-MS/MS and those with the sequence coverage (SC) higher than 40% were listed in Table 3.

Table 3.

Proteins identified in the six selected bblp and rblp extracts with sequence coverage > 40%^*.

No	extract	species	total identified proteins	protein name	UniProt Accession	SC (%)
1	bblp/US/9/100	Vicia faba L.	177	Plastocyanin	P00288	96.00
				Nodulin-related protein 1	A0AAV1B4H0	71.70
				23 kDa subunit of oxygen evolving system of photosystem II	A0AAV0ZF78	64.90
				Ribulose bisphosphate carboxylase large chain	A0A023I2J6	59.20
				Bet v I/Major latex protein domain-containing protein	A0AAV0YZK5 A0AAV0Z5U4 A0AAV0YXH1 A0AAV0YVE3	57.60 42.00 42.00 40.50
				Ferredoxin	A0AAV0YG12 A0AAV0Z3H9	56.10 54.40
				Triose-phosphate isomerase	A0AAV0Z3J5	54.60
				Stromal 70 kDa heat shock-related protein, chloroplastic	A0AAV0Z017	52.60
				Superoxide dismutase [Cu-Zn]	H9BPH7	49.00
				20 kDa chaperonin, chloroplastic	A0AAV0YZU0	48.80
				Fructose-bisphosphate aldolase	A0AAV0YNR1 A0AAV1B2V7 A0AAV1B5N1 A0AAV0YMX2	48.50 46.60 44.40 40.80
				Glycine cleavage system H protein	A0AAV0YUX3	44.20
				Transketolase	A0AAV0ZUE2	43.40
				NAD-dependent epimerase/dehydratase domain-containing protein	A0AAV1AYW6	42.30
				thioredoxin-dependent peroxiredoxin	A0AAV1B031	41.30
				Enoyl reductase (ER) domain-containing protein	A0AAV1AE39	41.20
				Ribulose bisphosphate carboxylase/oxygenase activase, chloroplastic	A0AAV0ZGE5	36.20
				RuBisCO large subunit-binding protein subunit beta, chloroplastic	A0AAV0ZA35	30.30
				Ribulose bisphosphate carboxylase small subunit, chloroplastic	A0AAV0Z4X9 A0AAV0Z5U4 A0AAV0ZB70	27.80 27.80 27.20
				RuBisCO large subunit-binding protein subunit alpha	A0AAV1B6J5	25.10
2	bblp/H/9	Vicia faba L.	107	Plastocyanin	P00288	75.80
				Superoxide dismutase [Cu-Zn]	H9BPH7 A0AAV1AZN6	65.80 45.40
				Ribulose bisphosphate carboxylase large chain	A0A023I2J6	65.70
				Nodulin-related protein 1	A0AAV1B4H0	56.50
				Bet v I/Major latex protein domain-containing protein	A0AAV0YZK5 A0AAV0YXH1 A0AAV0YVE3	55.10 49.00 46.80
				Ferredoxin	A0AAV0Z3H9 A0AAV0YG12	54.40 49.30
				Cytosolic copper/zinc superoxide dismutase (Fragment)	B5LRC1	52.50
				Dehydrin	A0AAV0ZC41	49.10
				Glycine cleavage system H protein	A0AAV0YUX3	46.10
				23 kDa subunit of oxygen evolving system of photosystem II	A0AAV0ZF78	42.90
				Thioredoxin-dependent peroxiredoxin	A0AAV1B031	40.90
				Ribulose bisphosphate carboxylase small subunit, chloroplastic	A0AAV0Z5U4 A0AAV0Z4X9 A0AAV0ZB70	31.70 31.70 26.10
				Ribulose bisphosphate carboxylase/oxygenase activase, chloroplastic	A0AAV0ZGE5	11.70
				RuBisCO large subunit-binding protein subunit beta, chloroplastic	A0AAV1A8N7	6.40
				RuBisCO large subunit-binding protein subunit alpha	A0AAV1B6J5	6.10
3	rblp/US/9/100	Beta vulgaris subsp. vulgaris L.	24	Ribulose bisphosphate carboxylase large chain	A0A023ZPS4	51.00
				Induced stolen tip protein TUB8-like	A0A0J8BW28	41.10
				DUF6598 domain-containing protein (Fragment)	A0A0J8BFA2	41.20
4	rblp/H/9	Beta vulgaris subsp. vulgaris L.	16	Ribulose bisphosphate carboxylase large chain	A0A023ZPS4	50.70
5	bblp/C	Vicia faba L.	107	Superoxide dismutase [Cu-Zn]	H9BPH7 A0AAV1B208	65.80 64.30
				Cytosolic copper/zinc superoxide dismutase (Fragment)	B5LRC1	61.00
				peptidylprolyl isomerase	A0AAV1AI87	58.90
				Plastocyanin	A0AAV0ZRF4	56.60
				Malate dehydrogenase	A0AAV1A8Z1	46.70
				CP12 domain-containing protein	A0AAV1A5M5	43.90
				SCP domain-containing protein	A0AAV0YF40	43.80
				Small ubiquitin-related modifier	A0AAV0ZIS4	43.00
				Stromal 70 kDa heat shock-related protein, chloroplastic	A0AAV0Z017	41.40
				RuBisCO large subunit-binding protein subunit beta, chloroplastic	A0AAV0ZA35	4.70
6	rblp/C	Beta vulgaris subsp. vulgaris L.	7	Superoxide dismutase [Cu-Zn]	A0A0J8BQL4	48.30

Sequence coverage, SC.

^*all identified RuBisCO proteins are listed regardless of corresponding SC.

The largest number of extracted proteins was identified in the bblp/US/9/100 extract − in total 177 identified proteins, and then in the bblp/H/9 extract and the bblp/C extract (both with 107 identified proteins). About five times less (up to 24) proteins were identified in the rblp/US/9/100, the rblp/H/9 and the rblp/C, and again these data definitely show that the rblp is far less affected by all three extraction procedures − the IoT-US, the IoT-H and the cold aceton-assisted procedure, than the bblp. In the bblp/US/9/100 and the bblp/H/9 extracts Ribulose bisphosphate carboxylase large chain (the SC of 59.20% and 65.70%, respectively) and Ribulose bisphosphate carboxylase small subunit, chloroplastic (the SC of 27.80% and 31.70%) were identified. In addition, the large chain with the SC of 51.00% and 50.70% in the two analysed rblp extracts were identified. Unfortunately, the large chain and also the small subunit were not detected in the bblp/C and the rblp/C extracts. Obviously, the cold acetone-assisted extraction of proteins from the leaves is not method of choice when the leaves proteins are to be analysed, characterized and used further. Moreover, sequences of the RuBisCO large chain and/or the RuBisCO small subunit, chloroplastic, with the SC > 99% corroborated extraction of this protein by the bblp/US/9/100, the bblp/H/9, the rblp/US/9/100 and the rblp/H/9 procedures, as detailed in Table 4.

Table 4.

Sequences of RuBisCO large chain (A0A023I2J6) and RuBisCO small subunit, chloroplastic (A0AAV0YZP7), identified with the sequence coverage (SC) of 99% in the four selected bblp and rblp extracts.

No	sample/extraction	sequence
1	bblp/US/9/100	RuBisCO large chain (A0A023I2J6)
		VTPQPGVPAEEAGAAVAAESSTGTWTTVWTDGLTSLDRYK ELGVPIVMHDYLTGGFTANTTLSHYCR KVGFQAGVKDYKLTYYTPEYQTK LTYYTPEYQTKDTDILAAFR EITLGFVDLLRDDYIEKDR GGLDFTKDDENVNSQPFMR MSGGDHIHAGTVVGKLEGER DKLNKYGRPLLGCTIKPK WSPELAAACEVWK TFQGPPHGIQVER EIKFEFPAMDTL LEDLRIPNAYVK DNGLLLHIHR AQAETGEIK AKVGFQAGVKDYK GHYLNATAGTCEEMLKR SQAETGEIKGHYLNATAGTCEEMLKR
2	bblp/H/9	RuBisCO large chain (A0A023I2J6)
		ALRLEDLRIPNAYVK GGLDFTKDDENVNSQPFMR DNGLLLHIHR DTDILAAFR EIKFEFPAMDTL EITLGFVDLLRDDYIEKDR ELGVPIVMHDYLTGGFTANTTLSHYCR GHYLNATAGTCEEMLKR LTYYTPEYQTKDTDILAAFR TFQGPPHGIQVERDKLNK VTPQPGVPAEEAGAAVAAESSTGTWTTVWTDGLTSLDRYKGR WSPELAAACEVWK YGRPLLGCTIKPK HMPALTEIFGDDSVLQFGGGTLGHPWGNAPGAVANR TSIVGNVFGFK DLINILEDAIR AQAETGEIK DTTTIVGDGSTQEAVNKR EVELEDPVENIGAK VVAAGANPVLITR AAMQAGIDKLADAVGLTLGPR
		RuBisCO small subunit, chloroplastic (A0AAV0YZP7)
		ELDEVIAAYPEAFVR FETLSYLPPLTR LPMFGTTEASQVLK
3	rblp/US/9/100	RuBisCO large chain (A0A023I2J6)
		WSPELAAACEVWK GGLDFTKDDENVNSQPFMR TFQGPPHGIQVER LEDLRIPVAYVK DKLNKYGRPLLGCTIKPK DTDILAAFR AQAETGEIK
4	rblp/H/9	RuBisCO large chain (A0A023I2J6)
		GGLDFTKDDENVNSQPFMR LEDLRIPVAYVK TFQGPPHGIQVER AQAETGEIK

Also, based on the RuBisCO large chain peptide AQAETGEIK concentration, concentration of the RuBisCO was assessed by the nanloLC-MS/MS method and estimated to be around 1408.64 μg per mL of the bblp/US/9/100 extract and 83.66 μg per mL of the rblp/US/9/100 extract.

Apart from the RuBisCO, and e.g., plastocyanin, two proteins located within the chloroplast, it is worth to mention that other identified proteins are successfully extracted by the IoT-US from various bbl cell compartments, such as (i) nodulin-related protein 1, a membrane protein (aquaporin), (ii) glycine cleavage system H protein, located at the mitochondrial inner membrane, and, as expected, (iii) soluble cytosol proteins, e.g., glycolytic enzymes fructose-bisphosphate aldolase and triose-phosphate isomerase. In comparison with two other green procedures − the DES-assisted extraction of fava bean proteins [12] and enzyme-assisted extraction of sugar beet leaves proteins [9], the IoT-US extraction of the broad bean leaves proteins in water is far less demanding if consumption of chemicals/enzymes is taken into account as well as the content of extracted proteins in resulting extracts.

4. Conclusion

Presented results clearly demonstrate potential of a smart integration of the non-thermal (NT) extraction unit with the Internet of Things (IoT), connected to the cloud system for collection and storage of measured data in the structured format, and extended with the web application to enable operator to create and manage analyses, in sustainable and safe extraction of highly valuable compounds. The accomplished IoT-US prototype tested in the laboratory scale showed high applicability in the green − ultrasound (US)-assisted extraction, which was compared to three non-green procedures − the IoT-heath (IoT-H) extraction, then the Osborne-based extraction assisted by the US or the H to justify use of water as a green solvent in the withdrawal routine, and the cold acetone-assisted extraction without assistance of the US or the H, to pinpoint the role of the US i.e., the H in the procedures. The most promising extraction of proteins, amino acids and flavonoids from affordable but more importantly suitable plant by-products with particularly low level of toxic metals, and >30.0% of protein of its dry matter − Vicia faba L. (broad bean) leaves (bblp) and Beta vulgaris L. (red beetroot) leaves (rblp) in water was achieved by the IoT-US. The IoT-US and the IoT-H generated aqueous bblp extracts (the bblp/US/9/100 extract and the bblp/H/9 extract) with mild acidic pH values (from 5.94 to 6.76 units) and moderate conductivity (0.043–4.355 mS/cm) contained the highest concentration of water-soluble flavonoids (76.98 ± 0.25 μg GAE/mL and 56.38 ± 0.18 μg GAE/mL, respectively), amino acids (Asx and Pro, then Glx, Ser, and Val, and also Ala, Phe and Tyr, all amino acids in the range 241.85–36.39 μg/mL), and proteins (31.685 ± 0.301 mg/mL and 31.231 ± 0.150 mg/mL, respectively). Although precipitation of proteins in the aqueous extracts under the conditions used was observed, for fair portion of water-soluble proteins from both bblp extracts molecular weight was estimated by the SDS-PAGE, and more than 100 proteins from various plant cell compartments, including membrane proteins and also RuBisCO from the chloroplast, were identified by nanoLC-MS/MS analysis. In addition, comprehensive peptide sequences with sequence coverage of 99% for the RuBisCO large chain and the RuBisCO small subunit, chloroplastic, were identified. Based on the RuBisCO large chain peptide AQAETGEIK concentration, the RuBisCO concentration of 1408.64 μg per mL of the bblp/US/9/100 extract and 83.66 μg per mL of the rblp/US/9/100 extract was estimated. In general, the rblp seem less susceptible to the selected extractions than the bblp.

According to all acquired data, the advanced IoT-US laboratory scale prototype can be used for testing raw material sensitivity to different extraction protocols in the small scale but also it can be scaled up for forthcoming green extractions from revealed plant-based by-products i.e., the broad bean leaves as renewable feedstock with its unenforced capacity to be used in the food industry and beyond.

This research represents a very exciting step forward from previous hypotheses and apparently optimized extraction processes [14], [18], [31], [32], [33], [34]. For the first time, a unit for the green extraction of primarily proteins, but also other desirable components (water-soluble flavonoids and amino acids) from waste leaves was integrated to set IoT. For each individual experiment performed in the designed IoT-US system, a huge amount of data was experimentally generated, collected, and stored in the DEEP cloud, and combined in user-friendly application (sensor-controlled extraction procedures set up approach). For comparison, the conventional IoT-H extractions (under similar conditions as much as possible to perform similarly) of proteins from the waste leaves were tested and equipotent results were assessed with these non-green procedures. Water (without the addition of other compounds) was used as an extractant in this kind of extraction protocols. Furthermore, although the IoT-US extractions caused minimal degradation of the plant matrix, proteins from different plant cell compartments were extracted in a predominantly soluble form under rather mild conditions (e.g., ≤ 35 °C). Moreover, content of the IoT-US derived protein-water solution is far more attractive and relevant than well-established salt/alcohol/alkali protein solutions gained under completely comparable non-green conditions in the IoT system − the control Osborne-based extraction. In addition, acetone, widely used and cost-effective organic solvent for the precipitation and concentration of proteins, particularly for removing interfering contaminants such as detergents, lipids, and salts, was also employed, but was efficient only in the non-US-assisted protein extraction from the red beetroot leaves. It remains to be seen if acetone precipitation of proteins from certain plant by-products (and waste) can be effectively combined with the IoT-US. In general, acetone is generally considered a “greener” organic solvent compared to many alternatives (like benzene or dichloromethane) due to its low toxicity, biodegradability, and high volatility, which allows for easy recovery. However, in a direct comparison with water, water is considered the superior green solvent.

The main idea of the entire IoT setup is to get closer to sustainable plant-based by-products and waste processing, in terms of low temperature extraction, exclusive utilization of green-solvent i.e., water, as shorter as possible treatment time, and, consequently, as lower as possible energy consumption. The optimized IoT-US system with all its components resemble such an encouraging structure.

Funding source

This work was supported by the Croatian Natural Science Foundation under the project „Digitalisation of nonthermal extraction of proteins from plant by-products and electroforming as output product“ (HRZZ IP-2022-10-2207).

CRediT authorship contribution statement

Anita Slavica: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Josipa Dukić: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Data curation. Tea Martinić Cezar: Writing – original draft, Methodology, Formal analysis, Data curation. Karla Košpić: Writing – original draft, Methodology, Formal analysis, Data curation. Janko Diminić: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation. Filip Jukić: Writing – original draft, Software, Methodology, Investigation, Formal analysis, Data curation. Marko Jurčević: Writing – original draft, Methodology, Investigation, Formal analysis, Data curation. Mojca Čakić Semenčić: Writing – original draft, Supervision, Methodology, Investigation, Formal analysis, Data curation. Sanja Ostojić: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Data curation. Iva Sabljak: Writing – original draft, Validation, Methodology, Investigation, Formal analysis, Data curation. Anet Režek Jambrak: Writing – review & editing, Supervision, Resources, Project administration, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: