Plant protein: extracting it from crops

In one paragraph. This chapter looks at extracting protein from plants. The pattern across every candidate is the same: the protein science is good and the Vietnamese supply is not. Mung bean makes the best protein in the set and Vietnam barely grows it. Peanut, coconut, sesame and cashew all work in the laboratory, and Vietnam crushes too little of the first two, does not report the third, and does not publish the tonnage for the fourth. Rice bran is the exception on supply, because Vietnam produces it at genuine scale, and the exception in the other direction on process, because no published extraction route gets both good yield and good purity. Then there is duckweed, which is not a crop at all, grows on farm wastewater, and produces more protein per hectare than anything else in this study, as feed.

11.1 Rice protein: broken rice, not bran#

Rice is the obvious place to start in Vietnam, and the obvious assumption is wrong. Rice bran is the side stream everyone points to, and it is the harder of the two routes. Commercial rice protein is made mainly from broken rice and milled endosperm, the fraction left when grains fracture during milling. [general]

Both rice proteins share the virtues that make rice interesting here: gluten-free, low allergenicity, a balanced amino acid pattern and good digestibility. Both share the weakness: poor solubility limiting what they can do in a food product, improvable by gentle structural modification or partial hydrolysis, sometimes at the cost of oil binding.

The position. Rice protein ranks second in chapter 5, and the entry point is broken rice, which Vietnam produces in quantity as a milling fraction and already trades. The bran route, assessed in section 11.3, remains the more interesting research problem and the weaker commercial proposition. Vietnamese broken rice volumes and prices were not established in this study: open question NP-G7. An external reviewer puts the broken rice stream at 1.5 to 2.5 million tonnes a year at 7.5 to 8.5 percent protein, which if confirmed would close that question; it is unverified and not quoted as a finding.

11.2 Mung bean protein#

Performance. Bench-scale isolates reach 81 to 92 percent protein on a dry basis at 40.9 to 66.5 percent yield, depending on the extraction route; concentrates reach 50 to 63 percent protein at about 31.9 percent yield. Commercial mung bean ingredients are typically 80 to 85 percent protein. Bench isolates show a nitrogen solubility index above 85 percent, against below 60 percent for commercial pea and soy isolates, with minimum solubility at pH 4.0 to 5.0 and maximum at pH 6.2 to 7.2. Water holding is 3.0 millilitres per gram and oil holding 1.7 millilitres per gram. Foaming stability is above 0.8 millilitres per millilitre after 90 minutes. Mung bean protein concentrate forms a gel at the lowest concentration in its comparison set, at 10 percent, while isolates generally need 16 to 18 percent. Flavour is reported as milder and cleaner than soy. [general]

Protein quality. DIAAS reported from 57.7 to 93, with cooked mung bean at 68, 76, 86 and 93 across different studies and age groups. PDCAAS 0.58 to 0.76. Digestibility from about 52 percent measured in the laboratory to 79 percent measured in pigs and 89.4 percent crude protein digestibility in one comparison of pulses. The limiting amino acid is usually methionine and cysteine, and is sometimes reported as lysine, leucine, threonine or valine depending on the scoring basis. [general]

The Vietnamese problem. Mung bean does not appear as a separately reported crop in USDA oilseeds coverage of Vietnam, which means it falls below the reporting threshold. No Vietnamese production figure was established in this study. See open question NP-G1.

Verdict. The best plant protein in this set on what it can do in a food and on flavour, with a protein quality that is respectable but much less certain than usually claimed. It fails the local-availability filter as things stand, and would need a deliberate agricultural push to pass it. An external reviewer argues this is settled enough that measuring national acreage is not worth doing, since mung bean is a smallholder rotational crop with no mechanised harvest or extraction infrastructure in Vietnam. That is a fair reading; the query is cheap enough to make anyway.

11.3 Rice bran protein#

Why it is the structurally interesting candidate. Vietnam mills 26.95 million tonnes of rice from 43.12 million tonnes of paddy. Rice bran is the one protein-bearing side stream the country produces at genuine scale, it is concentrated at mills rather than scattered across fields, and rice bran protein is gluten-free and causes few allergies, which matters because every other candidate in this chapter carries an allergen problem. [VN-direct] on scale, [general] on the protein

Note the correction in section 11.1: for a commercial product, broken rice is the better entry point. This section remains the assessment of the bran route, which is the better research problem.

Why it is hard. Rice bran protein sits in dense, poorly soluble protein bodies, cross-linked by sulfur bonds and tightly associated with phytate and fibre. Heat stabilisation of the bran, which is necessary to stop it going rancid, makes extraction harder still. [general]

Extraction, one line per published route. [general]

The pattern to notice. Yield and purity never occur together. The routes that recover most of the protein do not report a purity, and the route that reports the highest purity recovers only 15 percent. This is the central obstacle. The high-recovery methods tend to co-extract soluble arabinoxylans, phytates and silica, producing dilute hydrolysates that are hard to filter, which is a plausible reason the purities go unreported.

What the protein can do in a food. Concentrates show only 8 to 26 percent soluble protein at pH 4 to 7. Isolate nitrogen solubility is 52.3, 8.4, 63.5, 79.1, 83.7 and 81.5 percent at pH 2, 4, 6, 8, 10 and 12 respectively. Emulsifying capacity is 424 ± 14 millilitres per gram at pH 4 and 530 ± 21 millilitres per gram at pH 7.0, described as comparable to bovine serum albumin. Gelation is not reported anywhere in the source set. High hydrostatic pressure at 200 megapascals significantly increases solubility and emulsifying activity and stability. [general]

Scale and cost. The literature shows limited movement beyond laboratory scale: one pilot drying study and one small-scale concentrate process with cost estimates, plus an explicit statement that large-scale commercial production is not established. The only cost figures found come from a weaning pig feeding trial and are expressed as relative cost reductions against skim milk, not as a cost per kilogram of protein. They are not usable for a business case. See open question NP-G4. [general]

Verdict. The right raw material in the right country with the wrong process economics. This is where a Vietnamese research programme would have the most defensible claim to be doing something the field needs rather than importing a solved problem.

11.4 Peanut, sesame, coconut and cashew press cakes#

The protein science is good. Peanut concentrates and isolates from defatted flour or oil cake reach about 80 to 90 percent protein at laboratory scale. Sesame defatted meal starts at 35 to 50 percent protein and sometimes above 50 percent, and alkaline extraction produces isolates of around 90 to 94 percent protein with useful emulsifying, foaming and gelling behaviour; lysine is the first limiting amino acid. Coconut protein products range from moderate-protein dried extracts to high-protein concentrates, with performance depending heavily on whether the source is coconut milk cake, oil cake or copra meal. Cashew work is on defatted kernel flour and broken kernels rather than on the cashew apple, with isolates reaching very high protein content in laboratory studies. All follow the same route: remove the oil, extract with alkali, precipitate at the isoelectric point. [general]

The Vietnamese volumes rule out three of the four. [VN-direct]

Allergenicity is the other filter. Peanut, cashew and sesame are all major allergens, and the literature names allergenicity as a principal reason these proteins have not become larger food commodities despite performing well. For an export framing this is a labelling and regulatory burden; for a domestic framing it is a genuine safety consideration. [general]

Verdict. Cashew is the only one worth a second look, and only because Vietnam's processing industry is the largest in the world. The question there is not the protein, which is well characterised, but whether the broken-kernel stream is large enough and whether allergen management is tractable.

11.5 Protein quality and complementation#

How these scores work. DIAAS and PDCAAS both compare a protein's amino acid content, adjusted for how much of it the body actually absorbs, against human requirements. A score of 100 on DIAAS, or 1.0 on PDCAAS, means the protein supplies everything needed. Animal proteins sit at or above that line. Plant proteins sit below it, usually because they are short of one amino acid.

The comparison everyone quotes. A first-pass report gave DIAAS as soy 91, mung bean 86, pea 70, with animal references, meaning milk, whey, casein, egg and many meats, at or above 100 by DIAAS and at or near 1.0 by PDCAAS. Among the plant options, soy and mycoprotein come closest to animal reference proteins, pea is intermediate, and mung bean, rice, rice bran, peanut and sesame score lower. The usual limiting amino acids are the sulfur amino acids for legumes and soy, and lysine for cereals, rice bran, peanut and sesame. [general]

The comparison that survives scrutiny. When the same tool was asked for values backed by named sources, mung bean alone returned a DIAAS range of 57.7 to 93. The honest reading is that protein quality scores for plant proteins are wide ranges determined as much by processing, product form, test model and the age-reference pattern used as by the crop itself. See contradiction NP-C1. Publish ranges with their basis, never a single DIAAS number.

Complementation. Pairing a cereal protein short of lysine with a legume protein short of sulfur amino acids should raise the quality of the blend. That question was carried into a dedicated later search and is answered in section 13.2. In short: the improvement is real and well supported directionally, and a measured blend score at or above 100 does not exist in the evidence set. This closes open question NP-G3 with a qualified negative result.

11.6 Carbon and nitrogen co-feeding, which partly closes open question CF-G5#

Documented co-feeding cases pair carbon-rich residues with nitrogen-rich organic side streams, including fish and meat hydrolysates, digestate from biogas plants, potato processing water, soybean meal and hydrolysed chicken material, against carbon sources such as starch, glycerol, hydrolysates or mixed streams. These reach about 14.8 to over 30 grams per litre of biomass at protein contents around 40 to 46 percent, which sits inside the 40 to 60 percent range commonly reported for fungal systems on side streams generally.

Direct head-to-head tests against urea or ammonium sulfate are limited. Where they exist, organic nitrogen side streams can match or exceed mineral nitrogen for biomass in some systems, but mineral nitrogen often remains competitive and in several comparisons gives equal or better biomass. Waste-based fungal systems generally need some nitrogen added to reach high biomass. [general]

What this does to the cassava and seafood pairing. It stays interesting as a circularity and waste-disposal argument, and as a food-security argument about not importing nitrogen. It does not stand up as a yield or cost argument on present evidence. An external reviewer adds an operational objection: raw seafood processing waste carries 0.5 to 3.0 percent salt and live spoilage bacteria, both awkward in a continuous sterile fermentation, so a first plant would probably buy urea or ammonia for nitrogen and pH control. If the project pursues the seafood pairing, the honest framing is a Vietnamese waste-valorisation case, not a superior process.

11.7 Plant-based scoring table#

Dimensions: D1 local availability at mass scale, D2 protein content and extraction yield, D3 protein quality, D4 what the protein can do in a food, D5 process maturity and commercial precedent, D6 allergenicity and safety, D7 evidence quality.

11.8 The duckweed candidate#

Duckweed is a small floating water plant. It is not a crop in the usual sense: it is grown on the surface of ponds, doubles in days, and is harvested by skimming. It was not named in the original brief and entered the study in the first expansion round.

Read the limit first. Moved to the front in this version. There are two duckweed propositions and they are not variants of each other. Duckweed grown on farm effluent is a feed material and cannot enter the human food chain. Duckweed grown as food requires closed cultivation on clean water, which is horticultural infrastructure rather than a waste pond, and a different capital profile, a different site and a different business. Everything below should be read with that split in mind. The first version of this report stated the distinction in this section and then summarised duckweed in Part 2 without it, which oversold it.

What the general literature shows [general]#

• Protein content. Duckweeds are typically 20 to 35 percent protein of dry matter, reaching 40 to 50 percent in selected strains or optimised cultivation. Amino acid quality is strong for lysine, threonine and the branched-chain amino acids; the sulfur amino acids are the limiting point.
• Productivity. Around 6 to 18 grams of dry matter per square metre per day in the cited studies, which converts to roughly 22 to 65 tonnes of dry matter per hectare per year, with protein yields exceeding 10 tonnes per hectare per year in well-run wastewater systems. Soybean yields well under 1 tonne of protein per hectare per year. This is the highest protein yield per hectare encountered anywhere in this study.
• Human food evidence. Strongest for Wolffia globosa, marketed as Mankai, where essential amino acids were shown to be available to humans at levels similar to peas and, for several amino acids, broadly similar to soft cheese.
• Antinutrients. Oxalate in some Lemna species is the main compositional concern. Wolffia looks safer for direct human consumption. Tannins, phytates and nitrates need control species by species and batch by batch.
• Safety when grown on wastewater. Lemnaceae are hyperaccumulators of dissolved minerals and metals. Material grown on pangasius pond effluent or swine wastewater takes up cadmium, arsenic and lead, and enteric pathogens including Salmonella and E. coli. Wastewater-grown biomass is suitable for nutrient recovery and for feed or technical protein, subject to meeting the feed contaminant limits in QCVN 01-190:2020. Human food use requires controlled water, food-grade handling and species-specific contaminant and microbial checks.
• Extraction. Leaf protein concentrate and RuBisCO extraction both work, but yield and purity trade off against each other in the same way as with rice bran protein. Using fresh biomass and ultrasound-assisted extraction improves recovery; removing colour and purifying raises purity and lowers yield.

The regulatory precedent, and what it actually demonstrates [general]#

The EU novel food dossier for Wolffia globosa Mankai powder was submitted by Hinoman of Rishon LeZion, Israel. The product is grown in cultivation basins under closed, controlled greenhouse conditions with continuous growth and harvesting, certified to FSSC 22000 and ISO 9001. Composition: protein 40 to 48 percent, carbohydrate 24 to 40 percent of which 75 percent is dietary fibre, fat 6 to 12 percent, minerals below 10 percent, PDCAAS 0.89, with vitamins A, E, B6, B9 and B12 plus iron, magnesium and zinc. Safety data showed no genotoxicity across the AMES test and two chromosome studies, with microbiological and heavy metal analyses meeting standards and oxalic acid at or below the levels in common vegetables such as spinach. The European Food Safety Authority panel did raise questions about manganese levels in Wolffia powders.

A PDCAAS of 0.89 puts this above every plant protein assessed in this chapter. But note what the dossier is a precedent for: closed greenhouse cultivation on controlled water by one company. It is not a precedent for effluent-grown duckweed as food, and it should not be cited as one.

What Vietnam already has [VN-direct]#

• A research base going back three decades. A Can Tho study in the Mekong Delta grew Lemna spp. on ponds enriched with biodigester effluent and household wastewater, harvested twice daily, at 38.6 percent crude protein on a dry matter basis, and used it to replace roasted soya beans in a broken rice diet for Muscovy ducks. Growth was lower on the duckweed diets, at 27.6 to 28.3 grams a day for males against 36.1 for the control, and feed conversion was worse, at 4.12 to 4.23 against 3.24 to 3.76, but feed cost per kilogram of weight gain fell substantially when farmers grew their own duckweed, with the largest saving when soya bean was replaced completely.
• A pangasius-linked pilot has already been attempted. FiBL, led by Timo Stadtlander, ran a 2019 project with Binca Seafoods cultivating duckweed on nitrogen-rich wastewater from an organic pangasius farm in Vietnam, explicitly to reduce soybean meal imports, citing up to 70 tonnes of dry matter per hectare annually at 25 to 40 percent protein. It was funded by the Coop Sustainability Fund. No outcome data was published in the project record. Open question NP-G8.
• Wastewater treatment work continues. A study of nutrient removal by duckweed from anaerobically treated pig farm wastewater in laboratory-scale ponds in Vietnam was published in 2020.

Why this matters for the core question#

Every other category in this study is held up by the same thing: Vietnam has no submerged fermentation capacity, and building it is the expensive step. Duckweed does not need it. It needs ponds, a nutrient stream, harvesting and drying. Vietnam has the nutrient stream in industrial quantity, because pangasius and shrimp farming generate exactly the nitrogen-rich water duckweed grows on, and the aquafeed market that would buy the output imports its protein. Note that the buyer here is the whole aquafeed market rather than the narrower fishmeal-displacing pool of chapter 9, because duckweed at 25 to 40 percent protein competes with plant meals rather than with fishmeal.

The honest limits. Duckweed as feed is near-term, low-capital and already demonstrated in Vietnam at farm scale, and the binding question is contaminants rather than biology. Duckweed as a food-grade protein isolate is a different and much harder proposition: it needs closed cultivation on clean water, it runs into the same yield-versus-purity trap as rice bran protein, and it has one novel-food precedent held by one Israeli company. Do not conflate the two, and do not use the Mankai PDCAAS to describe an effluent-grown feed product. Whether Vietnamese duckweed grown on pangasius effluent would meet feed safety limits for heavy metals and pathogens is open question NP-G9, and it is the first thing to test.