Which crop biodiversity is used by the food industry throughout the world? A first evidence for legume species.

doi:10.21203/rs.3.rs-3452458/v1

Food biodiversity is a challenging issue for sustainable agri-food systems, particularly in the European one-health context. Most often studied at the levels of agricultural systems and food diets, few works examined plant species diversity used in food products. As the market is a main driver for farmers’ crop choices, analyzing how crop biodiversity is supported by the food industry will inform actions to promote it. By text-mining the ingredient lists of nearly 350,000 packaged food products extracted from the MINTEL database, we first provide a market assessment of legume biodiversity in processed food over the last decade, on a global scale. Soy appears as the major global legume crop for food, and there is still a clear lack of pulse diversity reflected in food products; some progress however was observed over the last years - particularly in Europe. Results suggest that lock-in continues to hamper stronger crop biodiversity.

Social science/Science, technology and society

Scientific community and society/Agriculture

Business and commerce/Economics

plant species biodiversity

food product launches

packaged food

soybean

legumes

pulses

ingredient list

The environmental impacts of conventional agri-food systems and markets are widely acknowledged^1–3, and particularly in terms of biodiversity loss⁴. Global diets and crop systems rely on a poor biodiversity of species³. According to the most recent FAO assessment⁵, out of 6,000 different plant species cultivated in 2014, just 9 accounted for 66% of the total crop production, and only 3 of them - wheat, rice and maize - represented over 50% of plant-based human food. This low biodiversity of cropping systems makes crops much more sensitive to pest infestation^6,7, less resilient to climate change^8–11, and more dependent on synthetic inputs such as fertilizers and pesticides^12,13. This may also have detrimental consequences on human diet^3,14. Agroecological transition (towards more biodiversity) and food transition (toward more plant-based diets) are both required^11,15,16, but drivers are obviously complex and multidimensional, concerning technological, social, institutional, educational, political, and economic levers, all at once^17,18.

Concerning processed food markets, more data on food product composition is required to uncover the relationship between biodiversity and agri-food systems^2,15,19. However, this is hampered, on the one hand, by a lack of information in scientific literature studying the types of ingredients used in food products²⁰, and on the other hand, by a lack of consensus on how to categorize ingredients, notably according to their levels of processing²¹. Our research tackles these challenging questions by using text-mining methods applied to the ingredient lists of processed foods, whatever the degree of processing.

Our goal is to question the premises of a shift toward more crop biodiversity from a market point of view: on which crop biodiversity does the food industry currently rely? To tackle this question, we focus on food product launches which reflect the innovation dynamics of food markets. According to innovation theories²² and transition studies^23,24, we assume that a broader transition could result from firms investing in new species for their products. In short, we consider new food product launches to be a proxy of the capacity of the food industry to develop species diversity in its offer.

This diversity is assessed at the ingredient list level from three complementary angles: (a) the analysis of the variety of species used, (b) the market’s degree of concentration on certain dominant species, and (c) a preliminary approach to species’ contribution to the overall product formulation. These three approaches to species biodiversity in product launches are based on three assumptions. Firstly, the term biodiversity suggests a wide variety of species. If markets feature only a few species, it would be speculative to speak of biodiversity. Secondly, the predominance of a particular species can be a disabling factor for more biodiversity. To assess these effects, we need to look at how agri-food markets concentrate on certain species. Thirdly, we make the hypothesis that the ways in which a species is used by the food industry, can either promote or limit species biodiversity in the market. When a species is used as a main component for a given food product, this does not mean the same thing in terms of biodiversity as when used as a side-component designed to carry functionalities to the food product.

Therefore, this paper constitutes an original response to the need for further scientific knowledge on the composition of the food products²⁵ that shape diets and agroecological systems, thanks to the development of text-mining methods.

We focused on the diversity of pulse species as pulses are increasingly promoted as a main lever for both agroecological and food transition^26,27. Pulses benefit from growing public support since the 2016 International Year of the United Nations, which created a great momentum for these plant species²⁸. While pulses are increasingly praised for their contribution to healthy and sustainable agri-food systems²⁹, they face a lock-in situation compared to the huge development of major crops such as wheat and soy^30–32. That is why we chose to compare the diversity of pulse species with soy - pulses being advanced in the literature as the major legume species used in food^24,26,32.

Data were retrieved from the Mintel Global-New-Products-Database (GNPD) which tracks food product innovations (i.e., product launches) in over 80 countries. Mintel-GNPD features about 5 million registered products, and is currently the only database that provides such a global coverage, with detailed information at product level³³.

We identified about 350,000 products launched in 2010-2021 worldwide, with at least one pulse species in their ingredients among the products from the Mintel-GNPD platform. The search criterion matched all products recorded containing at least one pulse or soy ingredient. However, Mintel-GNPD does not distinguish between the use of fresh and dried legumes the term “pulses” is reserved for pulses harvested once their grains have dried. To avoid any ambiguity, because it is sometimes impossible to deduce from food product compositions if the legumes were used fresh or dried, we labelled the pulse species identified in our corpus as “non-soy legumes” (NSL).

Data were retrieved as a tabulated file gathering all metadata available for each product retrieved by the search query, such as the ingredient lists, production and marketing places, types of product launches (new product, range extension, reformulation, repackaging or relaunch), targeted markets, claims mentioned. The results we present stem from text and data-mining applied to this dataset, at product and ingredient levels.

Firstly, we parsed the ingredient lists of products, originally stored as unstructured, free-form text, in Mintel-GNPD. On the packaging of any product, ingredients appear in decreasing order of importance: those with the highest concentration in the product formulation are listed, provided they represent a significant amount of the product (the threshold used varies according to national regulations). We developed a method³³ to decompose the ingredient lists at ingredient level. It extracts the rank and depth of appearance of each ingredient of an ingredient list. The position of an ingredient is determined both by: (a) its rank, i.e. its index within an ingredient list without considering if the ingredient is included in a nested ingredient list; and (b) its depth of appearance, i.e. at which level of a nested ingredient list it belongs. From this position we infer the importance of a given ingredient. Two main positions of legume-based ingredients were deemed of interest: situated among the top five ingredients of level 1 (that we called the ‘In Top List’) or among the remaining ingredients (that we called the ‘In Remaining List’). This distinction was adopted considering both the average distribution of soy-based ingredients and the ingredients related to other legume species within ingredient lists, which have variable lengths. We completed these observations by checking if the legume species were also mentioned in the product description provided on the product packaging (and registered in Mintel-GNPD data). Such additional mentions of species reveal a greater interest in the legume species used for the identity of the product.

Secondly, in order to disambiguate the NSL species mentioned (in the ingredient list or in the product description) we designed an automatic tagging process under the supervision of six food scientists. Plant species are most often designated by their vernacular names or by part of their vernacular names, the latter sometimes being misspelled. Matching an ingredient with the species from which it is wholly or partly derived is challenging. In order to associate ingredients rigorously to the scientific name of the NSL species they refer to, we first retrieved all ingredients containing a term that could refer to a pulse species (“pea”, “bean”, “lens” or “lentil”, “gram”, ...) and then examined the citation context of these terms (i.e. the n-gram in which the term used is appearing). The words that precede and follow these terms are very important to correctly identify the species that any given ingredient refers to. For example, in the following ingredients, “black eyed peas” and “grass pea”, it would be erroneous to deduct the presence of Pisum sativum L.. Here, each of the terms preceding "pea" refers to a different pulse species - Vigna unguiculata L. for “black eyed pea” and Lathyrus sativus L. for “grass pea”. This fine-grained analysis, conducted with pulse food science experts, resulted in a dictionary of expressions linked to the scientific name of each plant species identified. This dictionary was then translated into a computer automaton to label all the ingredients in our corpus that refer to a legume species. As mentioned in the paragraph above, this labelling process was also applied to product descriptions to identify legume species mentioned on the food packaging that we considered as an indicator of species promotion.

After discarding ambiguous mentions of species, we obtained a dataset of 343,444 products, not considering “specific cases” (n=4,734) - ingredients that refer ambiguously to a legume species (as in the expression “pulse bean”), false friends (as in the expression “coffee bean” or “coco pea”), or legume species that do not enter into the sub-family of pulses as Arachis hypogaea L..

The dataset is made of three sub-corpora as illustrated in Figure 1: i) products whose lists of ingredients contain soy-based ingredients and no NSL-based ingredients (n=249,425), ii) products whose lists of ingredients contain NSL-based ingredients and no soy-based ingredients (n=65,653), and iii) products containing both soy-based and NSL-based ingredients (n=28,366). Descriptive statistics presented in the paper were computed considering these three sub-corpora.

Species variety in an unbalanced market. The tagging of legume-based ingredients from the dataset led to the identification of 32 different species including soy(presented in Table S1). This variety is a rather unexpected observation, and at first glance could be considered as an encouraging result in terms of biodiversity. Yet the Figure 1 shows a highly asymmetrical distribution of identified species. Products for which only soy was identified within the ingredient lists, account for 72% of all products. Conversely, products for which only one or more NSL ingredients were identified, represent 19% of all products. The 9% of other products concern those containing both soy and NSL-based ingredients. There are nearly 4 times more product launches containing soy-based ingredients than products containing NSL-based ingredients.

The analysis of the frequencies of the tagged species (see Table S1 in appendices ) according to their position, confirms this highly unbalanced market in favor of few species - primarily soy, and pea among pulses. More precisely, a quartet of species, Pisum sativum L., Phaseolus vulgaris L., Cicer arietinum L. and Lens culinaris Medik L.., account for almost 80% of NSL ingredients, while around twenty NSL-species have a frequency of appearance inferior to 1% among products containing NSL-species. However, NSL ingredient products are more frequently associated with a mention of the NSL species in the product description (on the packaging) compared with soy, for which this frequency is only 4%. For instance, 70% of products containing Lens culinaris Medik. present a mention of the Lens species within the product description.

This unbalanced situation between soy and NSL-based ingredients could be linked to economic factors, such as the availability of each species for agri-food companies. According to the FAO[1], soy remains the most cultivated legume in the world, with an annual production of more than 300 million tons over the last decade. This is 3 times higher than the total production of the most frequent NSL species mentioned above. These observations first confirm the existence of a strong “technological lock-in” around one legume species - soya - which is used for food worldwide, the dominance of this species could be analyzed as a structural trend in agri-food markets^29,32.

This dominance of soy is observed across all market segments (Figure 2), with the exception of segments “Spreads” and “Fruits & Vegetables”, where the balance between products containing soy-based ingredients and products containing NSL-based ingredients is almost negligible or even inversed respectively. However, growth rates for products containing soy-based and NSL-based ingredients lead us to several observations. While soy dominates in terms of volume, the cumulative growth rate of products containing NSL-based ingredients is much higher than those containing soy, whatever the market segment considered (see also Table S1). Some products containing NSL-based ingredients experienced a very high cumulative growth rate, particularly in the “Dairy” segment - almost 12 times higher than that of soy-based products. The “Desserts” segment was almost 9 times higher, and the “Breakfast” segment - 7 times higher. More generally, these observations point to a growing interest of agri-food companies for NSL-based products³⁰. Such growing interest, if confirmed over time, could play in favor of greater biodiversity of legume species used in the food supply.

Figure 2 - Shares of products containing soy-based ingredients or NSL-based ingredients in each market segment (%). The percentage sum may exceed 100% because some products have both soy and NSL-based ingredients. The color intensity reflects the cumulative-growth of product launches in the market segment for soy and NSL over the decade. The market segment categories are those established in Mintel-GNPD and detailed in another work³².

[Here comes Table 1]

Europe presents the less unbalanced market between soy and NSL ingredients. Differences between products containing soy-based ingredients and those containing NSL-based ingredients are also more significant when we observe the share of these two categories in each main geographical area of our corpus (Figure 3).

Figure 3 - Soy and NSL-based products in main geographic areas covered by the corpus.

Soy is dominant in every geographical area. Let us now consider three groups of geographic areas. A first one encompasses areas for which the share of products containing soy-based ingredients and NSL-based ingredients are almost balanced: this is the case for Europe and Southern Asia, where around 40% of product launches contain NSL-based ingredients. A second one gathers areas where the share of products containing soy-based ingredients is extremely dominant (around 80%): this group concerns mainly Asian markets (excepting the Southern Asian market) and the North and South American markets. A third group includes areas where the share of products containing soy-based ingredients is less dominant than in the second group, but still concerns the majority (over 60% and under 77 %). This last group concerns smaller markets compared to the other areas, for which MINTEL-GNPD does not have full coverage, as it is the case for Africa.

The interpretation of such differences between geographical areas is probably multi-factorial. The structuring of the different markets may reflect differences in terms of food cultures. For instance, soy products dominate Asian markets, with the exception of the Southern Asian market which includes India, a country where pulse (and particularly lentils) production and consumption are among the highest in the world^35,36. Meanwhile, such differences can also be interpreted as the consequences of the different national or international public support schemes for pulse consumption, as is the case for Europe³⁷. Still regarding the European area, this quasi-balance between soy and NSL-based food products could indicate a shift in the technological lock-in that European countries have encountered until now^30–32, which would be beneficial in a near future to the biodiversification of processed food supply.

More generally, the overall structure of the corpus, whether in terms of market segments or geographical areas, reveals that a small number of species account for the bulk of legume-based food innovation production. This concentration stands in the way of greater biodiversity in the food markets. The more we use a small number of species to produce a larger and growing variety of foods (soy is present in all market segments), the less room there is for the development of other species. This situation, partly resulting from historical and economic factors that led to a lock-in situation, could undergo contemporary changes. But to confirm an actual possible shift that would favor the use of NSL species in food markets, we also need to look at the ways these species are used in product formulations.

Product-context use of legumes: entering the importance of the ingredient. To go further in assessing the biodiversity of legume species used in product launches, we looked at the ‘product-context of use’ of these species. In terms of biodiversity, we assumed that it could be misleading to consider on the same level the food products that use these species for different reasons that we do not know about. Notably, from the point of view of product formulation, we may consider that the functional properties (technological, organoleptic, nutritional, etc.) derived from the parts of the species used, account for more than the species itself. This way, we propose to approach what we call the ‘product-context of use’ by analyzing jointly the different positions of appearance of species in ingredient lists.

A good starting point is to examine where the identified species appears in ingredient lists. Regulations require ingredients to be listed in descending order of importance, with the first one weighing the most and the last one the least. Hence, we assume that a species that is only used for few of its functionalities (i.e. which received a treatment process aimed at extracting one or other of its parts such as peptides, starches, a gelling compound, etc.) is more likely to be found among the least important ingredients of an ingredient list (i.e. those weighing the least). This approach can be further refined by assessing if the species identified in the food products are part or not of the marketing pitch. We assume that the mention of the species on the product packaging (in addition to its appearance in the ingredient list) gives higher specificity to the species used, as it is positively associated with the identity of the product. From this point of view, differences between soy-based ingredients and NSL-based ingredients are quite striking.

[Here comes Table 2]

The Table 2 reports the mean position of the soy and NSL-based ingredients according to the ingredient list length, grouped in deciles. We observe that half of all products (52%) containing NSL-based ingredients are concentrated within the first four deciles; the first decile gathering almost 20% of the products containing NSL-based ingredients. For products containing soy-based ingredients, this threshold is reached from the 6^th decile upwards. On the other hand, unlike products containing NSL-based ingredients, the decile distribution of products containing soy-based ingredients seems relatively balanced, which would support the idea of flex uses of this species.

More generally, soy-based ingredients tend to appear, more frequently in food products with complex formulations (i.e., longer ingredient lists), and almost systematically at a higher rank (column ‘Soy ingr. mean position’ in Table 2) than NSL-based ingredients. In all the deciles except the last three, the mean position of NSL-based ingredients is always lower than soy-based ingredients. This means that NSL-based ingredients tend to appear earlier in ingredient lists, suggesting their amount used in the product formulations is probably larger than for soy-based ingredients. This result could be explained by the fact that soy has been much more widely studied than pulses in the field of Food Sciences & Technology, particularly during the last decade³⁸. Research and development in this domain led to a broader knowledge base for various uses and functionalities of soy, compared to all other pulses/NSL. In view of this, our results could confirm that soy use is associated to a larger array of functional ingredients than NSL-based ingredients. The likelihood of finding soy-based ingredients used as additives in product formulation is then probably higher than for NSL-based ingredients, whose position is most often among the top ingredients of ingredient lists. Thus, we assume that NSL-based ingredients are probably more frequently used as specific components for the product formulation.

We then considered only the top five ingredients (“InTop” in Figure 4) of depth 1 in the ingredient lists. This allowed us to eliminate most ingredients that are part of the composition of higher-depth ingredient lists (i.e., ingredients written between brackets). This selection criterion does not completely change the overall structure of the corpus. We observe here the same general trend. Soy-based products still account for almost twice (n=93,359) the number of NSL-based products (n=53,161). The latter still tend to appear more among the top 5 ingredients (69% to 78%), while soy-based ingredients appear more in the lower part of ingredient lists.

Based on this new criterion, we refined our analysis by classifying species according to their frequency of appearance among the top five ingredients, and in product packaging. The Figure 4 presents the result of a k-means clustering performed on these two dimensions, for species present in at least one hundred product launches. The 5 resulting groups are identified by categorical colors, and species are displayed in a three-dimensional space showing their frequency of appearance (in percent) among the top 5 ingredients, the remaining ingredients, and on the product packaging. In addition, to help interpret the results of this clustering, we also examined the most frequent ingredient expressions associated with the species in each group.

Figure 4 - 3D Scatter plot of most frequent legume species. Each species is plotted in the 3D graph according to its frequency of appearance in the top five ingredients (InTop%), in the remaining ingredients (InRemList %), and in the product description (InDesc %). Each color represents a cluster resulting from the k-means clustering (5 clusters were requested, according to results provided by the silhouette coefficients method³⁹).

A central axis structures the cluster distribution in Figure 4. This axis distinguishes the species mostly found among the top 5 ingredients and frequently in product descriptions (the green cluster), from those found more frequently among the remaining ingredients and rarely cited in product descriptions (the blue cluster). More precisely, at one end of this axis we find a set of 6 species highlighted in green which are Phaseolus coccineus L., Cajanus cajan L., Lens culinaris Medik., Phaseolus vulgaris L., Cicer arietinum L., Vicia faba L., characterized by a high frequency of appearance among first ingredients and a high rate of mentions in product descriptions. These features lead us to suggest that product identity is more closely associated with those NSL species whatever their functional use. In that sense, species from this group could have a more positive impact on market biodiversity as they are of key interest for the food industry, in comparison to species only used for a functional interest, and therefore could be substituted by another species. Most frequent ingredients associated with species from this group do not seem to indicate a fractionated use of them. For example, in the case of Lens culinaris Medik., the most common ingredients mentioning this species quote it directly by its vernacular name, without mentioning specific parts ("lentils", n=2522; "red lentils", n=1250; “green lentils”, n=818). When this ingredient is associated to a processing term, the most frequent one is milling (“lentil flour”, n=1195). The same is true for Cicer arietinum L. (“chickpeas”, n=7467; “chickpea flour”, n=3295). This cluster gathers ingredients that seldom undergo processing.

At the opposite end of this central axis, plotted in blue, we find a group of 4 NSL-species (Canavalia gladiata Jacq., Pachyrhizus erosus L., Dolichos lablab L., Glycine max L.) and soy. They present a reversed profile: a low frequency among first ingredients and in product descriptions. This cluster could also include Ceratonia siliqua L. (plotted in purple), which has been identified as a cluster in its own right due to its extreme behavior: it is hardly ever mentioned - neither in product descriptions, nor among first ingredients. In this group, for the two most frequent species, Glycine max L. and Ceratonia siliqua L., the ingredients most frequently associated correspond to fractionated uses (“soybean oil”, n=58584; “soy lecithin”, n=50490; “soy protein”, n=18569; “locust bean gum”, n=4903; ”carob bean gum”, n=2242).

This axis, that contrasts species according to their frequency of appearance (within the top 5 ingredients and in the product description) may also tend to oppose different product-context of use of species. This brings us back to our initial hypothesis: the more frequently a species is used in a fractionated way, the more likely it is to be found among ingredients of lesser importance (in terms of volume and therefore rank and level) in ingredient lists, and the less prominence it is given in product packaging.

Hence, the case of the median cluster (plotted in red on Figure 4,) is very interesting. Here we find species characterized by a balanced score between their frequency of appearance among the top and remaining ingredients, but not systematically mentioned in product descriptions (Pisum sativum L., Vigna unguiculata L., Vigna angularis L., Vigna radiata L., Lupinus angustifolius L.). According to our main hypothesis, this median position between species could reveal various strategies of the food industry for those species that could become more “identity” or “generic” species, according to the future uses those species will encounter. In other words, those species can be used as effective key-components for product formulation, and sometimes not. The analysis of the most frequent ingredients quoting the most major species of this group seems to substantiate this observation. For Pisum sativum L., the two most common ingredients are “peas” (n=10,132) and “pea protein” (n=6,423), and for Lupinus angustifolius L., the most common ingredient is lupin flour (n=1,095).

Finally, a group of three species (plotted in orange on Figure 4), made up of Vigna aconitifolia Jacq., Vigna mungo L. and Phaseolus acutifolius L., seems to be opposed to the first group described (plotted in green) due to the lower propensity of these species to be cited in product descriptions. These “discrete” species know very low frequency of appearance, but the analysis of the most cited ingredients related to them brings them closer to the first group. For example, the most common ingredients referring to Vigna mungo L. mention the species by its vernacular name (“black gram lentils”, n=1412; “black lentils”, n=114), and when a process is mentioned, in most cases it concerns flours, the resulting product from grinding possibly coupled with sieving, (“black lentil flour”, n=56). We observe the same thing for Phaseolus acutifolius L. (“tepary bean flour”, n=128; “tepary beans”, n=25).

While there is abundant and growing scientific literature on the benefits of biodiversity for sustainable agri-food systems, this paper constitutes the first attempt to assess food market biodiversity through processed food product launches over the main regions in the world. We developed original text-mining analysis methods for tagging species and interpreting species biodiversity through the food ingredients used. We based this approach on the different positions of the species under study within ingredient lists, and from additional descriptions on packaging.

Focusing here on legume species, a botanical family at the heart of sustainability issues in contemporary agri-food systems, this work makes it possible to compare the development of soy on processed food markets with the development of pulses. By analyzing ingredient compositions of about 350,000 food product launches worldwide, we identified over 30 different non-soy legume species used in product formulation. This could suggest consistent biodiversity in the use of these species. Nevertheless, we mitigated this result by taking an in-depth look at frequency of appearance and ways in which those species are used in product formulation and then promoted on packaging. Thanks to this first analysis we can assess to which extent the food market is concentrating on certain dominant species, which could hamper the development of biodiversity. Indeed, one of our main hypotheses was that the more a market locks-in on certain species, the more difficult it becomes for others to emerge, and this lock-in situation becomes a major hindrance for more biodiversity. We thereby confirm at the food market level what other studies stated at the crop field level: there is still a strong lock-in situation around soy-based products, even if there is a perceptible shift in favor of other pulses. In this regard, our results particularly highlight the specificity of the European market, which presents a more balanced use between soy-based and non-soy-legumes-based ingredients.

Furthermore, we assume that the position in ingredient lists of ingredients related to soy or other non-soy legumes is a proxy of the ways in which these species are used. Considering this position (in the ingredient list), is a way to distinguish species which are more valued in their entirety as opposed to more “flex” species used for their decomposition into various functional ingredients. Crossing this information with the ways species get promoted or not on packaging, we suggest that the notion of species, and therefore that of biodiversity, tends to be replaced by that of functionality. The food industry’s lack of interest in highlighting the species simply as it is, before any processing, reflects a system where agri-food production is more and more valued by a compositional/processed paradigm (i.e. where food is first considered as a technological process of assembling elements fulfilling a desired goal such as increasing protein level, improving texture, avoiding specific flavors, etc.). If this work constitutes a first step towards a deeper analysis of what we called “the product-context use” of species in food products, further research should undertake a deeper assessment of the processing profile of those ingredients.

Finally, this work also demonstrates the interest and feasibility of analyzing the agri-food market supply at the ingredient scale. While some other works studying composition of food products tackled food safety²⁰ and climate issues⁴¹, here we focused on biodiversity issues. Being able to identify how processed food impacts agri-food systems remains a key challenge for informing public policymakers and consumers. We argue that this type of analyses, at the ingredient level, could serve as a tool for public policy to steer agri-food markets towards more sustainable goals. However, this depends on the availability of food databases and accessible controlled vocabularies and ontologies. Except for the USDA branded food database and the crowdsourced database Open Food Facts, the most extensive food databases required to conduct such an analysis are privately owned. Despite the intrinsic quality of these databases, whether for research purposes or to support public policy, this does not guarantee the control and transparency of information from collection to processing data. In this context, benefiting from controlled vocabularies validated by the academic community on the ingredients linked to processed foods would not only ensure this transparency but would also constitute a powerful tool for public policy⁴². We also advocate the need for enriched public databases on the processed food supply.

Acknowledgments

The research was supported by the PPR SPECIFICS project (ANR-20-PCPA-0008) funded by the “Growing and Protecting crops Differently” French Priority Research Program (PPR-CPA), part of the national investment plan operated by the French National Research Agency (ANR) and the Occitanie Region (KING project). Thanks to Alice Thomson-Thibault for her English editing checking of the manuscript.

Rockström, J., Edenhofer, O., Gaertner, J. & DeClerck, F. Planet-proofing the global food system. Nat Food 1, 3–5 (2020).
Campbell, B. M. et al. Agriculture production as a major driver of the Earth system exceeding planetary boundaries. E&S 22, art8 (2017).
Tilman, D. & Clark, M. Global diets link environmental sustainability and human health. Nature 515, 518–522 (2014).
Benton, T. G., Bieg, C., Harwatt, H., Pudasaini, R. & Wellesley, L. Food system impacts on biodiversity loss. (2021).
Pilling, D. & Bélanger, J. The state of the world’s biodiversity for food and agriculture. (2019).
Thomine, E., Mumford, J., Rusch, A. & Desneux, N. Using crop diversity to lower pesticide use: Socio-ecological approaches. Science of The Total Environment 804, 150156 (2022).
Bedoussac, L. et al. Ecological principles underlying the increase of productivity achieved by cereal-grain legume intercrops in organic farming. A review. Agron. Sustain. Dev. 35, 911–935 (2015).
Zimmerer, K. S. & De Haan, S. Agrobiodiversity and a sustainable food future. Nature Plants 3, 17047 (2017).
Gil, J. Forgotten crops confer resilience under climate change. Nat Food 4, 275–275 (2023).
Labeyrie, V. et al. The role of crop diversity in climate change adaptation: insights from local observations to inform decision making in agriculture. Current Opinion in Environmental Sustainability 51, 15–23 (2021).
Mustafa, M. A., Mabhaudhi, T. & Massawe, F. Building a resilient and sustainable food system in a changing world – A case for climate-smart and nutrient dense crops. Global Food Security 28, 100477 (2021).
Altieri, M. & Nicholls, C. Biodiversity and Pest Management in Agroecosystems, Second Edition. (CRC Press, 2004).
Beillouin, D., Ben‐Ari, T., Malézieux, E., Seufert, V. & Makowski, D. Positive but variable effects of crop diversification on biodiversity and ecosystem services. Glob Change Biol 27, 4697–4710 (2021).
Webb, P. et al. The urgency of food system transformation is now irrefutable. Nat Food 1, 584–585 (2020).
Leite, F. H. M. et al. Ultra-processed foods should be central to global food systems dialogue and action on biodiversity. BMJ Glob Health 7, e008269 (2022).
Burlingame, B. & Dernini, S. Sustainable diets and biodiversity: directions and solutions for policy, research and action: proceedings of the intenational scientific symposium Biodiversity and sustainable diets united against hunger, 3-5 november 2010, FAO headquarters, Rome. (FAO, 2012).
Tilman, D. Extinction, climate change and the ecology of Homo sapiens. Journal of Ecology 110, 744–750 (2022).
Köhler, J. et al. An agenda for sustainability transitions research: State of the art and future directions. Environmental Innovation and Societal Transitions 31, 1–32 (2019).
Campbell, B. M., Thornton, P. K. & Nelson, G. C. Upping our ambition for food system adaptation. Nat Food 3, 970–971 (2022).
Ahuja, J. K. C. et al. IngID: A framework for parsing and systematic reporting of ingredients used in commercially packaged foods. Journal of Food Composition and Analysis 100, 103920 (2021).
Sadler, C. R. et al. Processed food classification: Conceptualisation and challenges. Trends in Food Science & Technology 112, 149–162 (2021).
Rogers, E. M. Diffusion of innovations. (Free Press ; Collier Macmillan, 1983).
Geels, F. W. Socio-technical transitions to sustainability: a review of criticisms and elaborations of the Multi-Level Perspective. Current Opinion in Environmental Sustainability 39, 187–201 (2019).
Raven, R., Bosch, S. V. den & Weterings, R. Transitions and strategic niche management: towards a competence kit for practitioners. IJTM 51, 57 (2010).
Toledo, Á. & Burlingame, B. Biodiversity and nutrition: A common path toward global food security and sustainable development. Journal of Food Composition and Analysis 19, 477–483 (2006).
Semba, R. D., Ramsing, R., Rahman, N., Kraemer, K. & Bloem, M. W. Legumes as a sustainable source of protein in human diets. Global Food Security 28, 100520 (2021).
Ditzler, L. et al. Current research on the ecosystem service potential of legume inclusive cropping systems in Europe. A review. Agron. Sustain. Dev. 41, 26 (2021).
Foyer, C. H. et al. Neglecting legumes has compromised human health and sustainable food production. Nature Plants 2, 16112 (2016).
Cusworth, G., Garnett, T. & Lorimer, J. Legume dreams: The contested futures of sustainable plant-based food systems in Europe. Global Environmental Change 69, 102321 (2021).
Cusworth, G., Garnett, T. & Lorimer, J. Agroecological break out: Legumes, crop diversification and the regenerative futures of UK agriculture. Journal of Rural Studies 88, 126–137 (2021).
Magrini, M.-B. et al. Why are grain-legumes rarely present in cropping systems despite their environmental and nutritional benefits? Analyzing lock-in in the French agrifood system. Ecological Economics 126, 152–162 (2016).
Magrini, M.-B. et al. Pulses for Sustainability: Breaking Agriculture and Food Sectors Out of Lock-In. Front. Sustain. Food Syst. 2, 64 (2018).
Salord, T., Magrini, M.-B. & Cabanac, G. Packaged foods with pulse ingredients in Europe: A dataset of text-mined product formulations. Data in Brief 42, 108173 (2022).
Magrini, M.-B., Salord, T. & Cabanac, G. The unbalanced development among legume species regarding sustainable and healthy agrifood systems in North-America and Europe: focus on food product innovations. Food Sec. (2022) doi:10.1007/s12571-022-01294-9.
Joshi, P. K. & Rao, P. P. Global pulses scenario: status and outlook: Global pulse scenario. Ann. N.Y. Acad. Sci. 1392, 6–17 (2017).
Gurusamy, S., Vidhya, C. S., Khasherao, B. Y. & Shanmugam, A. Pulses for health and their varied ways of processing and consumption in India - A review. Applied Food Research 2, 100171 (2022).
Szczebyło, A., Halicka, E., Jackowska, M. & Rejman, K. Analysis of the Global Pulses Market and Programs Encouraging Consumption of This Food. PRS 19, 85–96 (2019).
Magrini, M.-B. et al. Peer-Reviewed Literature on Grain Legume Species in the WoS (1980–2018): A Comparative Analysis of Soybean and Pulses. Sustainability 11, 6833 (2019).
Rousseeuw, P. J. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20, 53–65 (1987).
Clark, M. et al. Estimating the environmental impacts of 57,000 food products. Proc. Natl. Acad. Sci. U.S.A. 119, e2120584119 (2022).

Data were retrieved from the “FAOStat” website (https://www.fao.org/faostat/en/#data) on July 25, 2023. These are production data extracted worldwide over the decade corresponding to the period covered by our data.

Legumes Type	Product Category	Number of Launches	Share %	Cumulative Growth	Category Cumulative Growth
Soy	Baby Food	4894	76.769	109.091	112.121
NSL Spec.	Baby Food	1716	26.918	117.742	112.121
Soy	Bakery	47765	88.664	43.969	62.789
NSL Spec.	Bakery	9912	18.399	249.324	62.789
Soy	Breakfast	3997	84.45	87.64	137.705
NSL Spec.	Breakfast	957	20.22	637.5	137.705
Soy	Dairy	10892	90.525	33.907	61.258
NSL Spec.	Dairy	1853	15.401	396.364	61.258
Soy	Desserts	17707	89.778	21.648	39.184
NSL Spec.	Desserts	5099	25.853	193.373	39.184
Soy	Drinks & Soups	13928	69.713	60.117	94.595
NSL Spec.	Drinks & Soups	8127	40.678	180.936	94.595
Soy	Fruits & Vegetables	1818	12.924	35.165	143.228
NSL Spec.	Fruits & Vegetables	12759	90.702	154.812	143.228
Soy	Meals & Side Dishes	49620	77.519	75.744	91.727
NSL Spec.	Meals & Side Dishes	22627	35.349	132.373	91.727
Soy	Processed fish,meat & egg	37600	88.914	146.253	174.55
NSL Spec.	Processed fish,meat & egg	7129	16.858	493.464	174.55
Soy	Sauces & Seasonings	32736	92.043	70.196	69.053
NSL Spec.	Sauces & Seasonings	4283	12.042	61.026	69.053
Soy	Snacks	51551	83.233	77.956	103.332
NSL Spec.	Snacks	14879	24.023	324.128	103.332
Soy	Spreads (sweetened & salted)	5139	58.061	88.115	171.028
NSL Spec.	Spreads (sweetened & salted)	4540	51.294	324.561	171.028

Table 1 - Soy and NSL Species market segment features (number of launches, share of the market segment in the corpus, cumulative growth over the period, category cumulative growth). Market shares, expressed as percentages, are calculated for each legume type in relation to all food products in the market segment concerned. For example, 76,77% of Baby food products in our corpus contain soy-based ingredients. As a product can contain soy-based and NSL-based ingredients, aggregate market shares may exceed 100%.

ListSize_Deciles	Nb. Launches	NSL Launches	NSL Class Share %	NSL ingr. mean position	Soy Launches	Soy Class Share %	Soy ingr. mean position	Launches w. NSL and Soy
(1,7]	39188	18135	19.32	1.57	21888	7.88	2.54	835
(8,11]	33769	10841	11.55	3.04	24440	8.8	4.51	1512
(12,14]	30859	8561	9.12	4.25	23957	8.63	5.61	1659
(15,18]	43771	11170	11.9	5.4	35385	12.74	6.81	2784
(19,21]	31697	7558	8.05	6.5	26432	9.52	8.21	2293
(22,24]	27773	6278	6.69	7.56	23705	8.54	9.34	2210
(25,29]	37431	8246	8.78	9.46	32678	11.77	10.64	3493
(30,35]	31052	7117	7.58	11.79	27343	9.85	12.13	3408
(36,46]	35386	7772	8.28	15.13	32057	11.55	13.45	4443
(47,278]	32383	8206	8.74	25.6	29771	10.72	17.74	5594

Table 2 Soy-based and NSL-based ingredients mean position according to ingredient list length. Here, ‘Deciles’ column corresponds to the discretization into deciles of the ingredient list lengths of the products in the corpus. For each one of this class, we look at the number of products for which the length of the ingredient list corresponds to the decile (‘Nb. Launches’), the average position of the soy or NSL ingredient in the list (‘NSL/Soy ingr. mean position’), and the share these products represent of all products containing soy or NSL (‘NSL/Soy Class Share’). The results are displayed for products containing both NSL-based and soy-based ingredients.

There is NO Competing Interest.

TableS1SpeciesFrequencies.xlsx
Supplementary Table 1 - Species Frequencies

Which crop biodiversity is used by the food industry throughout the world? A first evidence for legume species.

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Results and Discussion

Conclusion

Declarations

Acknowledgments

References

Footnotes

Tables

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1