Industries / Food, farming and forestry / Food and beverage processing / Plant-based and alternative protein
Texture
Texture work turns plant protein into pieces that bite, chew and tear like meat, mostly by lining the protein up into fibres.
7
Milestones achieved, from 1954 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1954
Protein spun into fibres
Robert Boyer patented a way to make meat substitutes by forcing a protein solution, such as soy protein, through the fine holes of a spinneret into a bath where it sets into thin threads, much as textile fibres are made. The threads were stretched to line up the protein molecules, coated with a binder and fat, and bundled into pieces that could be cut and cooked like meat. The patent was filed in 1952 and granted on 29 June 1954.
Credit
Robert A. Boyer
1962
Frozen, not canned
Worthington Foods in Ohio bought rights to Boyer's patent in 1956, and Ralston Purina built a plant making about 10,000 pounds (4,500 kg) a day of food-grade spun soy fibres for it. Canning at high temperature spoiled the flavour and texture of the new fibre products, so Worthington chose to sell them frozen, and launched frozen chicken-style products in 1962.
Lessona new texture can fail at the last step; the canning heat that suited older analogues destroyed the spun fibres.
Credit
Worthington Foods; Ralston Purina; Robert Boyer
1965
Extruded soy chunks
William Atkinson of Archer Daniels Midland (ADM) patented a product made by mixing defatted soy meal with 20 to 60% water, kneading it above 250 °F (121 °C) under pressure, and forcing it through a small opening. Steam puffs the mass into an aligned, open structure that takes on the texture of cooked meat when soaked in boiling water. The patent was first filed in 1964 and granted in 1970. The Soyinfo Center says ADM introduced it as TVP in 1965, that it was widely used by 1968, and that 12 or 13 US companies made similar products by 1980.
Credit
William T. Atkinson; Archer Daniels Midland
2012
Wet extrusion makes chicken
In high-moisture extrusion, a wet protein mix is cooked in a heated screw barrel and then cooled in a long die, so it sets into aligned fibres instead of puffing up. A University of Missouri team spent more than a decade on a soy product made this way that felt like chicken breast. Beyond Meat licensed it and began selling it in a few Whole Foods stores in northern California in 2012.
Credit
Fu-hung Hsieh and Harold Huff (University of Missouri); Beyond Meat
2016
Shear cell makes thick cuts
Researchers at Wageningen University and TU Delft scaled up a shear cell, two cylinders that turn against each other while heating a protein paste, to a 7-litre device. At 120 °C, about 30 minutes and about 25 revolutions per minute it made fibrous soy pieces 30 mm thick, the size of whole cuts, and the study found no barrier to further scale-up. The Wageningen spin-off Rival Foods, founded in 2019, was supplying shear-cell chicken fillets to restaurants in the Netherlands and Germany by April 2024, AgFunderNews reported.
Credit
Atze Jan van der Goot (Wageningen University); Georgios Krintiras, Andrzej Stankiewicz and Georgios Stefanidis (TU Delft); Birgit Dekkers and Ernst Breel (Rival Foods)
2021
Printed steaks in restaurants
Redefine Meat, an Israeli company, makes plant-based whole cuts by 3D printing, building a cut layer by layer, which FoodNavigator said also deals with how fat is spread through the meat. In late 2021 it launched them in high-end restaurants in the United Kingdom, Germany, the Netherlands and Israel, FoodNavigator reported.
Credit
Redefine Meat; chief executive Eshchar Ben-Shitrit
2025
Spun fibres in supermarkets
Project Eaden, a Berlin company, makes threads from pea, wheat and faba bean protein, each about as thick as a muscle fibre, twists them into bundles and presses the bundles into blocks for ham-style cold cuts and sausages. The Berliner Zeitung reported the products on the shelves of the Rewe supermarket chain in May 2025, 71 years after Boyer's spinning patent. The company says its brand, LEKKA, is now in more than 4,500 supermarkets in Europe.
Credit
Project Eaden
Watch for
Marbled cuts for US shops
Steakholder Foods, an Israeli maker of food 3D printers, says it will launch Perfecta, a line of whole-cut steak, chicken and seafood substitutes, in the United States in the second half of 2026, using a proprietary process that gives the products marbling-like fat. No retailers, prices or capacity have been given.
Achieved whenthe products are on sale in US shops and an outside party reports on them or measures their texture.
Claimed by
Steakholder Foods, May 2026
Fibre bundles without printers
Offbeast, a Baltimore company, says its "cross fiber assembly" shreds and bundles soy and wheat fibres to copy the bundles of animal muscle, replacing the 3D printing it started with, which its founder calls too costly to scale. It says its former chicken plant can make up to 250 tons a year and makes about 55 tons now.
Achieved whenan outside party measures the texture against beef or reports the plant's output.
Claimed by
Offbeast, January 2026
Open
01
A shelf-bought texture benchmark
A public, repeated set of measurements of how meat substitutes on sale compare with the meat they copy: tension along and across the grain, compression and shear, a photo-based fibre score for each piece, and a short tasting panel. Publish the raw data and repeat each year as products change.
The briefCredit · unclaimed02
A shared extrusion run log
One open table, in one format, of high-moisture extrusion runs from many labs: recipe, water content, barrel and die temperatures, screw speed and throughput, and the result (tensile strength along and across the fibres, fibre score, photos). Then train a model on all labs but one and test it on the one left out.
The briefCredit · unclaimed
Open 01 · the brief
A shelf-bought texture benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, repeated set of measurements of how meat substitutes on sale compare with the meat they copy: tension along and across the grain, compression and shear, a photo-based fibre score for each piece, and a short tasting panel. Publish the raw data and repeat each year as products change.
Why it is still open
The largest open studies test a handful of products. A Stanford team tested eight (five plant-based, three animal) in 157 mechanical tests and published its data; a 2025 study of deli slices tested four of each. A 2025 Wageningen paper says meat analogue texture "remains poorly defined". No set covers the whole cuts launched since 2021.
Built with what exists
A texture analyser or universal testing machine such as the Stable Micro Systems TA.XTplus or an Instron 5900; Fiberlyzer, Wageningen's published image method for scoring fibres from photos, or Meta's Segment Anything 2 for the same job; the Stanford lab's open analysis code (github.com/LivingMatterLab/CANN).
First step
Buy 10 chicken-style products and one chicken breast, cut strips along and across the grain, run tension tests, and publish the numbers.
Done looks like
At least 50 products from at least three countries, including 10 whole cuts, plus 10 animal references, each tested at least five times in each mode, with photos and raw data under an open licence, and a second round a year later that retests at least 30 of them.
Open 02 · the brief
A shared extrusion run log
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One open table, in one format, of high-moisture extrusion runs from many labs: recipe, water content, barrel and die temperatures, screw speed and throughput, and the result (tensile strength along and across the fibres, fibre score, photos). Then train a model on all labs but one and test it on the one left out.
Why it is still open
Published machine learning studies of meat analogue texture use small, private data sets: one 2025 study compared methods on 15 extrusion trials, and a 2023 model predicted hardness with a 22.9% average error. Labs do publish single-study extrusion data on Zenodo, each in its own format.
Built with what exists
The extrusion data sets already on Zenodo; a laboratory twin-screw extruder with a cooling die, for example a Thermo Fisher Process 16, for new runs; XGBoost and Meta's Ax or BoTorch for models and for choosing the next runs.
First step
Convert the high-moisture extrusion data sets already on Zenodo into one table, and publish the schema with an invitation to other labs.
Done looks like
At least 500 runs from at least five labs in one schema under an open licence, and a model trained on four labs that predicts the fifth lab's tensile strength across the fibres within a stated error, compared against a simple baseline.