The compatibility of animal-sourced food and circularity in healthy European diets

Several dietary guidelines are developed that propose limiting the intake of animal protein to stay within planetary boundaries and improve human health. Simultaneously, circular food systems are receiving signicant attention in the European Union as an option to improve the current food system. In a circular system, animals are solely fed with low-opportunity-cost-biomass (LCB), resulting in substantially fewer animals and reduced supply of animal-sourced nutrients to humans. We assessed whether this circularity principle within the EU-28 is compatible with the recommended animal-source food consumption in healthy and environmentally friendly dietary guidelines such as the EAT-LANCET dietary guidelines. Our results show that the overall quantity of animal-sourced protein in EAT-LANCET dietary guidelines can be met, but that the precise levels of inclusion of different animal-sourced foods in such a diet cannot be achieved. The EAT-LANCET guidelines recommend larger quantities of chicken meat over beef and pork while a circular food system produces mainly milk, dairy-beef, and pork. All three circularity diets outperform the EAT-LANCET diet in nutritional value while reducing GHG emissions (up-to 31%) and arable land use (up-to 42%). Careful consideration of the permissible substitutability between animal-sourced foods is urgently needed to dene the role of animal products in circular human diets. In this way the consumption of animal products - based on the circularity principle of only feeding animals with LCB benets both human health and the environment.

, while others show that farm animals reared under a circular paradigm can play a crucial role in feeding humanity [11][12][13][14][15] . Circular food systems aim to optimally utilise resources by prioritising arable land to produce plant biomass for human consumption, thus avoiding feed-food competition 16,17 . Currently about 40% of our global arable land is used to produce high-quality feed for farm animals, which to a large extent is human-edible 18 . From a resource-e ciency point of view, farm animals, instead, could be fed low-opportunity-cost-biomass (LCB), which includes co-products from the food industry (e.g. wheat middling's or slaughter waste from farm animals), food waste, and grassland resources 13 . In this case, the resource use e ciency of the farm animals is increased which has potential to reduce the environmental impacts 15,19 .
Our aim was to assess whether adhering to the circularity principle of feeding LCB to farm animals within the EU-28 is compatible with the recommended animal-sourced food consumption in healthy and environmentally friendly dietary guidelines. We took a reference diet derived from the EAT-LANCET dietary guidelines as an example of a future healthy diet of which the environmental impacts of the food system were kept within the safe operating space of the planetary boundaries' framework 4 . To adhere to circularity principles applied to the EU-28, animals were fed co-products and food waste resulting from the plant-sourced fraction of EAT-LANCET diet (Fig. 1). In addition, grassland resources and slaughter byproducts from farm animals could be used as animal feed. A resources allocation model was used to distribute the LCB among animal production systems (dairy, beef, pigs, broilers, layers, Atlantic salmon and Nile tilapia) to maximize protein production while respecting recommended animal-sourced food intake levels of the EAT-LANCET dietary guidelines as well as land use and GHG emission boundaries. Crop and animal production systems in the EU-28 were based on current management and yields (i.e., kg per hectare or kg per animal). Nutrient adequacy of the EAT-LANCET dietary guidelines was assessed against the European Food Safety Agency (EFSA) human nutrient intake requirements 20 . Four scenarios were investigated (Table 1), rstly an EAT-LANCET reference scenario which represented the EAT-LANCET dietary guidelines in their current form (EL Reference). Secondly, a healthier whole-grain diet with a xed composition of animal-sourced food (EL Circular Wholegrain Fixed). Thirdly, a re ned grain diet with a xed composition of animal-sourced food (current grain consumption is dominated by re ned grains and consuming wholegrains results in less by-products from cereal processing and can therefore strongly affect the role of animals when adopting circularity principles, van Hal et al., 2019a) (EL Circular Re nedgrain Fixed). Lastly, a wholegrain diet with an unrestrained quantity of animal-sourced food to demonstrate the production potential of animals fed LCB (EL Circular Wholegrain Potential). Scenarios two to four provide insight into the debate about which and how many animals to keep in a circular food system and the trade-offs and synergies with health recommendations.

Results
Animal-sourced Food Supply From Circular Food Systems Figure 1: Framework to assess the supply of animal-sourced food from animals fed LCB. Example shown represents the EL Circular Wholegrain Fixed scenario. All ows are in fresh matter except grass which is in dry matter.
Our analysis revealed that animals exclusively fed LCB were unable to provide the combination of meat, milk, eggs and sh recommended in the EAT-LANCET dietary guidelines, largely due to an insu cient quantity of high quality LCB. In total, the reference diet derived from the EAT-LANCET dietary guidelines contained 71 grams of meat and sh, 250 grams of milk and 13 grams of eggs per capita per day. It was, nevertheless, possible to ful l these recommendations by adjusting the share of meat and sh while respecting the healthy range. The reference value for pork, for example, is 7 grams, while the healthy range is 0-14 grams of pork per capita per day.
In the EL Circular Wholegrain Fixed scenario, recommended quantities of milk and sh could be met while meat and eggs were 5% and 92% short of meeting the recommended intake in the EAT-LANCET dietary guidelines. In the EL Circular Re ned-grain Fixed scenario, the recommended quantities of meat, milk, eggs and sh could be met due to the additional LCB available from the re ning of grains (e.g., wheat bran). However, adjusting the shares of meat and sh was still required. Compared to the EL Circular Wholegrain Fixed, the EL Circular Re ned-grain Fixed scenario could produce more poultry meat (4 vs. 2 grams of poultry meat) and meet the recommended intake of eggs in the EAT-LANCET dietary guidelines (13 eggs per capita per day). From a health externalities perspective, the consumption of poultry meat is preferred over the consumption of beef and pork 4 . Broilers and laying hens, however, were limited in their ability to upcycle all types of LCB and mainly required the co-products from re ned grains. This creates a trade-off between consuming healthy whole grains or producing healthy white poultry meat and eggs.
The EL Circular Wholegrain Potential scenario showed the optimal allocation of LCB (in terms of maximising protein production) to different animals (Fig. 2). This scenario resulted in an increase in pork production (to 40 grams per capita per day) due to a pig's ability to convert low quality co-products and food waste into animal-sourced food. Milk production also increased (to 563 grams per capita per day) as dairy cattle are e cient converters of LCB (especially grassland) to protein. Increased milk production increased the supply of cull cows which produced additional beef. The increase of pork and milk was at the expense of poultry and sh production, thus showing a trade-off between optimally utilising LCB and producing the preferred white meat.

Human Nutrient Supply From Circular Food Systems
Our results showed that the EL Circular Wholegrain Fixed, EL Circular Re ned-grain Fixed scenarios and the EAT-LANCET Reference did not meet zinc, calcium, vitamin B12 average nutrient requirements of the human population set out by EFSA ( Fig. 3; Supplementary Material). Notably, the EAT-LANCET reference also fails to meet EPA/DHA average nutrient requirements (Fig. 3). The EL Circular Wholegrain Potential did meet the calcium and vitamin B12 requirements but not zinc however, largely due to an increase in milk production (250 grams vs 563 grams). For all nutrients except EPA/DHA (due to less sh), nutrient supply was greatest in the EL Circular Wholegrain Potential scenario. Besides calcium, all three circularity diets outperformed the EAT-LANCET diet on available nutrients.

Greenhouse gas emissions and land use impacts of circular food systems
Overall, GHG emissions were 31% and 28% lower in the EL Circular Wholegrain Fixed and EL Circular Re ned-grain Fixed compared to the EAT-LANCET reference scenario. The reduction in emissions was due to the avoided emissions related to the production of animal feed (e.g., nitrous oxide (N 2 O) from nitrogen fertilisation) and the EL Circular Wholegrain Fixed scenario requiring less grain production (i.e., more grain was destined for human consumption, due to no re ning). . In all EL circular scenarios, the upper limit to the range of uncertainty was beyond the safe operating space.
Overall, cropland use was lower in all EL circular scenarios compared to the EAT-LANCET reference diet. However, it was important to note that the EAT-LANCET reference scenario was a global land use average, while the EL circular scenarios were based on EU land use. Further, utilising cropland to produce animal feed also led to an increase in land use in the EAT-LANCET reference scenario. Cropland use was lowest in the EL Circular Wholegrain Fixed and EL Circular Wholegrain Potential scenarios due to the use of wholegrains requiring less land (i.e., less co-products from wheat results in less land required), though differences with using re ned grain were marginal. Grassland use of the EL Circular Wholegrain Fixed and EL Circular Re ned-grain Fixed were similar while the EL Circular Wholegrain Potential scenario resulted in a higher grassland use, as the use of grassland resources was increased for milk production (Fig. 3).
The milk and beef production in circular food systems was highly dependent on the availability of grassland. Variation exists in the data of quantity and quality of current grassland in the EU-28 depending on the study and de nition of grassland (i.e. between managed and natural grassland) and available data sources. We compared the animal-sourced food output (e.g., milk) of the EL Circular Wholegrain Potential scenario with different areas of managed grassland resulting from three different studies/models [21][22][23] . Milk production and beef (from dairy cattle) ranged, respectively, from 326 to 780 and 11 to 37 grams per capita per day (Supplementary Material). Including natural grasslands could further increase the output of animal-sourced food.

Discussion
Our results show that the overall quantity of animal-sourced protein in EAT-LANCET dietary guidelines can be met, but that the precise levels of inclusion of different animal-sourced foods in such a diet cannot be achieved by only feeding LCB to animals. The extent to which the recommended quantities of animalsourced food could be met largely depended on the availability of the LCB. The EL Circular Wholegrain Fixed scenario versus the EL Circular Re ned-grain Fixed scenario revealed that the role animals can play in circular food systems will be narrowed as we move towards healthier consumption of plant-sourced foods. With today's food consumption patterns, several food groups are consumed in highly processed forms, resulting in additional by-products on the one hand, but increasing the risk for non-communicable diseases on the other hand 24 . The example employed here, wheat, results in by-products such as wheat bran and wheat germ if wheat is consumed in a re ned manner. If it is however consumed as whole grain, as recommended in the EAT LANCET dietary guidelines, no by-products occur. Potatoes, vegetables, and fruits would be other examples; if potatoes for example are industrially processed, potato peels can be collected and thereafter used as animal feed.
Although dietary guidelines could not be met, it was possible to meet the nutrient recommendations when the EAT-LANCET diet restrictions were removed (i.e., the EL Circular Wholegrain Potential scenario).
Comparing the two EL circular xed scenarios with the EL Circular Wholegrain Potential scenario showed that some animal species were more e cient at upcycling LCB (e.g., dairy cattle and pigs) than others (e.g., poultry). Grass resources for example were utilised most e ciently by dairy cattle as ruminants are well adapted to value this feed. Wet or brous food leftovers are used most e ciently by pigs that are known to have a high feed intake capacity. Milk, furthermore, includes relatively high amounts of calcium and beef and pork are high-quality sources of bioavailable vitamin B12 and zinc 25 . In other words, each animal has its own unique capacity to convert LCB into speci c nutrients. The nutrients provided by animals are of high bioavailability and some, such as vitamin B12 and the omega-3 fatty acids EPA and DHA are predominantly provided animal-sourced foods and are almost absent in plant-source-foods 25 .
The circular scenarios and in particular the EL Circular Wholegrain Potential scenario showed that animals raised in a circular food system can play an essential role in providing nutrients. This is in line with earlier ndings of for example Röös  None of those studies, however, assessed the importance of the dietary recommendations. Our results made clear that although the EL Circular Wholegrain Potential scenario met all nutrient recommendations except zinc, it exceeded the intake of beef and pork. Findings from cohort studies suggest that occurrence of several noncommunicable diseases, such as cardiovascular disease, was associated with a relatively high intake of red meat, i.e., beef and pork (e.g., Etemadi et al., 2017). The higher recommended amounts of poultry as compared to beef in the EAT-Lancet reference diet was justi ed by the fact that poultry meat does not show associations with increased mortality, and poultry fat moreover disposes over a higher content of essential poly-unsaturated fatty acids (21% vs. 4%) 4 . The above illustrates that although it is important to optimize essential nutrients from animal-source food, following upper limits of dietary guidelines is essential to avoid dietary related diseases.
Our results furthermore showed that circularity principles were adopted, GHG emissions and land-use were reduced compared to the EAT-LANCET diet, as feed-food competition was avoided. In Willett et al.
(2019), broilers perform better than e.g., cattle meat from an environmental perspective, due to their favourable feed conversion ratio. However, their assessment is based on impact intensities (e.g., GHG emissions per kg of food product) calculated for the current system. Thereby, it was not considered whether the feed for broilers would also have been suitable as food for humans, or whether the area it was grown upon would have been suitable to grow food for humans. Our analysis clearly shows that as soon as we move towards a circular food system and hence restrict the role for animals to converting LCB, broilers cannot compete with e.g., cattle anymore. This does not mean that broiler cannot play a role in circular food systems, but it demonstrates that the broilers of today are less suited to convert LCB. This stresses the importance of adapting future breeding goals and feeding strategies towards their ability of utilizing LCB. This is essential as our results showed that animals can reduce their environmental impact (and the impact of the entire food system) if they increase their e ciency in converting LCB into healthy food.
Our study and model focused on feeding LCB, including food waste to animals as a principle of circularity. The nutrient content of food waste was a weighted average based on the amount of food consumed in the human diet and the proportion wasted. By combining these products into one mix the feed value in terms of energy and protein of higher quality waste products (e.g., grains) is diluted by lower quality waste products (e.g., vegetables). Expanding the model employed to separate some streams of food waste may increase the amount of animal-sourced food produced due to a greater availability of high quality LCB. Further expanding the optimisation model to include plant-sourced food production within the EAT-LANCET diet could offer further opportunities to reduce GHG emissions and land-use. In addition, more circularity principles could be captured, including returning nutrients in manure and crop residues to the soil. Applying alternative objective functions (e.g., minimising GHG emissions while meeting the nutrient requirements of the human diet) could also in uence the animal production systems selected.
We would like to end by stressing the importance of future technologies 28

Conclusion
We demonstrated that feeding low-opportunity-cost-biomass to animals has the potential to reduce GHG emissions and land-use. Our results showed that the quantity of animal-sourced protein in EAT-LANCET dietary guidelines could be met, but that the precise animal-sourced food composition of the EAT-LANCET dietary guidelines could not be met by only feeding LCB to animals. Dietary guidelines recommend chicken meat over beef and pork while in a circular food system mainly milk, dairy-beef, and pork are produced. Careful consideration of the permissible substitutability between animal-sourced foods is urgently needed to de ne the role of animal products in the human diet. In this way the consumption of animal products -based on the circularity principle of only feeding animals with LCB -bene ts both human health and the environment.

Methods
In this study we extended the resource allocation model developed by van Hal, (2020) to include GHG emissions and land-use. The model of van Hal, (2020) allocates co-products and food-waste resources from the EAT-LANCET example diet derived from the EAT-LANCET dietary guidelines 4 , and grassland resources. We compared environmental impacts of the EAT-LANCET diet with three EL circular scenarios.
Each scenario varied based on the type of grain (wholegrain or re ned grain) and the animal-sourced food composition. (Table 1).

Quantifying leftovers from the EAT-LANCET diet
We took the example diet developed by the EAT-LANCET Commission (derived from the EAT-LANCET dietary guidelines) as a starting point for this study. To better re ect the EU diet, some adjustments were made to grain consumption (i.e. more wheat and less rice, total quantity of grain remained unchanged) based on FAOSTAT 4,34 .
To calculate the amount (i.e., tonnes) and area of crop required, and co-products available, reverse calculations were made using food consumption as a starting point. Quantities of co-products (e.g. wheat bran) from crops (e.g. wheat) were calculated using so-called technical conversion factors 35,36 . In some scenario's grains were re ned to increase the availability of co-products as animal feed and to better re ect current dietary habits (e.g., wheat bran, Table 1). Quantities of food waste were calculated using food waste fractions developed by Gustavsson et al. (2011). This process was performed for each of the EU-28 countries.
Resource allocation model

The model of van Hal, (2020) is a resource allocation model of the EU-28 developed in General Algebraic
Modelling System (GAMS) version 30.3. The objective of the model is to maximise animal protein output from a given availability of animal feeds while meeting the nutritional requirements of the animals.
Animal systems include livestock (dairy, beef, pigs, broilers, and layers) and farmed sh (Atlantic Salmon and Nile Tilapia). The two sh systems are a proxy for a range of species with similar characteristics (e.g. rainbow trout for Atlantic salmon). Livestock systems include three productivity levels (high, medium and low) while farmed sh only include a high productivity level. The model included the parent stocks (e.g. sow in pig system) and reproduction stocks (e.g. heifer in a dairy system) to account for the entire lifecycle of the animal. The nutritional requirements of livestock and farmed sh can be found in Supplement Material of van Hal, (2020).
Livestock and farmed sh were exclusively fed co-products, food waste, grassland resources and animal by-products, referred to as LCB. In our model, co-products and animal by-products could be traded between EU-28 countries while food waste and grassland must be used in the country it is produced. The availability of co-products and food waste was set by the EL reference.
Thirty-ve percent of the available food waste could be fed to animals as a wet feed, which is considered achievable if the feeding of food waste to animals were to be legalised 38 . Food waste could only be consumed by monogastric animals and sh due to food safety risks 38 . The availability of European grassland was based on the Miterra-Europe model 23 , and it was assumed grassland could only be consumed by ruminants. Our analysis only included managed grassland due to the uncertainty in quantity and quality of natural grasslands in Europe.
The availability of animal by-products was a fraction of the predicted live weight output of each livestock system (Supplement Material of van Hal, (2020)). Cannibalism was prevented in livestock systems, in farmed sh systems cannibalism was allowed due to the species being a proxy of a range of species. This enabled intraspecies recycling of by-products from farmed sh, meaning farmed sh can consume by-products of the same species. The nutritional value of LCB for livestock was obtained from the Dutch animal feed board; known as the CVB system 39 . While the nutritional value of LCB for farmed sh was obtained from the IAFFD 40 .
In addition to aquaculture, the model includes capture sheries. Capture sheries produced sh for human consumption and sh by-products (e.g., sh meal) which could be fed in the animal systems. Dutch GHG inventory methodologies 41 . It was assumed that fossil energy was replaced by renewable energy sources by 2050, causing no CO 2 emissions to keep our assumptions in-line with the EAT-LANCET study 4 . The only contribution to land-use from livestock was the grassland used as feed for ruminants.
Other livestock systems were considered landless or had a very small, negligible land-use as livestock are fed exclusively LCB.
GHG emission calculations were performed using a food systems approach. A food systems approach assesses emissions from the total diet as opposed to emissions per individual products in a life cycle assessment. Emissions were limited to on-farm, including manure management, enteric fermentation and grassland production. No other animal feed emissions were considered due to the food-based allocation method 42 .
GHG emissions from terrestrial animals (dairy, beef, pig, broiler and layer) included CH 4 and N 2 O from manure management. Methane emissions from manure management were calculated by multiplying volatile solid excretion by the methane conversion factor (i.e. the conversion factor for each manure management system), B 0 (i.e. the maximum methane producing capacity for manure) and 0.67 (i.e. the conversion of methane from m 3 to kg CH 4 ) 43 . Volatile solid excretion was calculated using digestibility of protein and organic matter of feed consumed by the animal species 44 . Nitrous oxide emissions from manure management included direct and in-direct emissions (the latter resulting from the volatilisation of ammonia and nitrogen (di)oxide) from nitrogen excretion in housing systems 43 . Nitrogen excretion was calculated by subtracting nitrogen retained in meat/milk/eggs from nitrogen intake 41 .
In addition, ruminant systems included CH 4 from enteric fermentation and N 2 O from grassland fertilisation. Methane emissions from enteric fermentation was calculated by multiplying gross energy intake by Y m (i.e. percentage of gross energy in feed converted to CH 4 ) and dividing by 55.65 (i.e. the gross energy content of methane) 43 . Nitrous oxide emissions from grassland included direct and indirect emissions (the latter resulting from the volatilisation of ammonia and nitrogen (di)oxide and the leaching of nitrate) from nitrogen fertilisation and manure excretion while grazing 45 . Grassland fertilisation rates were estimated by the Miterra-Europe model based on the assumption that all organic fertiliser produced by grazing animals is applied to fodder crops (e.g., grassland, fodder maize) in the same region and based on FAOSTAT data on N mineral fertilizer 23 .
GHG emissions from aquatic animals (e.g., high-tropic and low-tropic aquaculture) included N 2 O emissions from the aquaculture system. Nitrogen in un-consumed feed and excreta (nitrogen intake minus nitrogen retained in body tissue) was multiplied by 1.8% and converted from nitrogen to N 2 O 46 .
GHG emissions were summed into carbon dioxide equivalents (CO 2 e; 100-year time horizon, 28 for biogenic CH 4 and 265 for N 2 O; 47 ), and summed with plant-sourced food emissions (see next section) to calculate total GHG emissions. Results were given in GHG emissions per diet per capita per year.
GHG emissions and land-use from plant-sourced food For plant-sourced food, average national crop yields and nitrogen inputs per hectare were estimated using the Miterra-Europe model 23 with 2017 as a reference year. From national crop yields and nitrogen inputs, direct and indirect N 2 O emissions were calculated using an IPCC tier 2 approach 45 . A food-based allocation method was applied where all GHG emissions were allocated to the main food product (e.g. wheat our) 42 . To calculate GHG emission intensities and crop yields at an EU level, a weighted average was applied based on harvested area in each country. Not all plant-sourced food was included in the Miterra-Europe model (Lentils, groundnuts, tree nuts, and bananas). The GHG emission intensities and crop yields per hectare were then estimated using global data, reference year circa 2000 29 . Processing of crop into edible food products and food waste along the supply chain (see section Quantifying leftovers from the eat-lancet diet) were then considered to calculate GHG emission intensities and land-use per kg of plant-sourced food consumed. Figure 1 Framework to assess the supply of animal-sourced food from animals fed LCB. Example shown represents the EL Circular Wholegrain Fixed scenario. All ows are in fresh matter except grass which is in  Animal-sourced food production from the three EL Circular scenarios. 100% is equal to the recommended intake in EAT-LANCET dietary guidelines, i.e., 28 grams of sh, 7 grams of pork, 13 grams of eggs, 29 grams of poultry meat, 7 grams of beef and 250 grams of milk