Quantifying the utilisation of blue, green and brown resources by riparian predators: A combined use of amino acid isotopes and fatty acids

1. Global change drives multiple facets of biodiversity including interaction diver - sity, which is fundamental for ecosystem functioning. However, studying trophic


| INTRODUC TI ON
Global change has far-reaching impacts on various facets of biodiversity (Scherer-Lorenzen, 2014).One key facet that has been little studied to date is interaction diversity, which is of fundamental importance as trophic interactions form food webs and maintain ecosystem functions and services (Pugh & Field, 2022).This interaction diversity links a variety of ecosystems together.For example, there are notable trophic interactions between aquatic (blue), terrestrial (green), and decomposing (brown) compartments of ecosystems.These interactions have been conceptualised within the meta-ecosystem framework (Loreau et al., 2003), which describes ecosystems that are connected by spatial flows of energy, materials, and organisms across ecosystem boundaries.Characterising these flows is crucial in ecology for understanding trophic dynamics and the risks they face in a changing world.
Blue-green links have been increasingly recognised in sustaining vital ecosystem processes.For example, terrestrial organic carbon can be an important diet component for lake plankton (Cole et al., 2011), many aquatic invertebrates feed on terrestrial input in the form of leaf litter (Little & Altermatt, 2019), and fish can rely on terrestrial prey (Sato et al., 2011).In addition, emerging aquatic insects are an important food source for a wide variety of terrestrial predators, including birds (Twining et al., 2018), bats (Fukui et al., 2006), beetles and spiders (Paetzold et al., 2005).Blue-green links are however particularly threatened by global change drivers, including climate change (Shipley et al., 2022), habitat modification (Wesner et al., 2020) and eutrophication (Greig et al., 2012).
While much attention has recently been paid to the exchange between blue and green ecosystems, an often overlooked facet within food web ecology is the connection between green and brown food webs (Potapov, Guerra, et al., 2023;Potapov, Lindo, et al., 2023).The dependence of brown food webs on green sources is evident, since most of the organic matter that is decomposed in brown food webs comes from photosynthetic sources (Elfstrand et al., 2008).However, the reciprocal link is understudied even though detritivorous invertebrates may be an important prey to green predators (Birkhofer et al., 2008), with cascading effect on green ecosystem functioning (Halaj & Wise, 2002).The scarcity of studies in this area can be attributed to the difficulty to sample brown sources, and to the complex interpretation of traditional dietary markers (Traugott et al., 2013).
When more than two cross-ecosystem linkages are considered, it is harder to track dietary sources and energy flows (Bowes & Thorp, 2015).
The study of cross-ecosystem linkages has been significantly enhanced by the development of compound-specific stable isotopes of amino acids (AAs).In particular, essential amino acids (EAAs), which can only be synthesised by primary producers, are of particular interest as they retain their distinctive isotopic signatures up the food chains (Whiteman et al., 2019).Each EAA can be characterised by its δ 13 C, that is the ratio of the heavy but rare carbon-13 stable isotope compared to the lighter but common carbon-12.Another notable advancement lies in the capacity to differentiate primary producers from various phyla based on the relative difference between EAA δ 13 C, an approach commonly referred to as fingerprinting (Larsen et al., 2009).Based on the fingerprinting, studies have estimated the relative importance of producers as food source of several consumers (Arsenault et al., 2022;Liew et al., 2019;Manlick et al., 2024;Manlick & Newsome, 2022;Saboret et al., 2023).
Sampling and isolating primary producers in the wild remains challenging, especially for bacteria, whose fingerprinting is typically derived from lab cultures.An alternative approach involves sampling primary consumers, which integrate basal fingerprinting over time and provide a site-specific baseline (Skinner et al., 2021;Vane et al., 2023).However, primary consumers themselves may exhibit overlaps, particularly for brown and blue consumers that derive carbon from bacteria, fungi, and plant, and bacteria, algae, and plant, respectively.Consequently, it remains unclear whether primary consumers of blue, green, and brown sources exhibit distinct fingerprinting patterns that could be used to track source reliance in natural environments.
This pursuit holds substantial significance owing to the pivotal role that PUFAs play in shaping the carbon transfer efficiency in ecosystems (Müller-Navarra et al., 2000).Notably, the sources of PUFAs (blue, green, or brown) exhibit distinct abundance profiles.For instance, blue sources are known for their rich abundance in long-chain (20 carbon atoms and more) PUFAs, such as eicosapentaenoic acid (EPA) and docosahexaenoic acids (DHA), predominantly produced by algae (Hixson et al., 2015).Even a small amount of subsidised PUFAs can sustain vital ecological processes.For instance, swallow's chicks show an increased performance when fed with PUFA-rich food (Twining et al., 2016).Researchers have used this uneven abundance of PUFAs to elucidate dietary reliance.However, the composition of PUFAs might vary markedly between these preys and the consumers that depend on them (Ebm et al., 2021).Consumers possess the 4. We show for the first time the effectiveness of combining AA isotopes and PUFA abundances, particularly relevant for complex trophic interactions in a metaecosystem context.Our study illustrates the trophic uncoupling of proteins and PUFAs, highlighting the necessity in combining both approaches.capacity to accumulate and modify PUFAs (but rarely synthesise them de novo, but see Kabeya et al., 2018), further complicating the understanding of PUFAs transfer (Galloway & Budge, 2020).
A promising alternative avenue could be the use of AA isotopes, which disentangle dietary preferences, in combination with the relative abundances of PUFAs, which inform on the transfer and utilisation of high-quality resources in food webs.In addition, combining the two approaches could provide novel insight into the coupling of protein and PUFA transfers in food webs, which is particularly important if the sources might differ in their nutritional quality.
However, it is unknown if those two approaches can be combined in mixing models for inferring dietary preferences.A good study system, where green, brown and blue systems collide, are riparian areas where spiders have access to all three types of sources.Here, we use spiders in a temperate riparian meta-ecosystem as a study case to test the applicability of combining the two analytical approaches to infer dietary preferences in a natural system where a previous study revealed a wide range of resource utilisation (Kowarik et al., 2021).In this system, we focus on ground-dwelling spiders and web-spiders that mainly feed on brown and green sources, respectively, while both groups might benefit from blue sources close to water bodies (Nyffeler, 1999).
In this study, our specific aim is to examine how a combination of AA isotopes and PUFA abundances can enhance our understanding of carbon utilisation in linked ecosystems.For the first time, we report PUFA mass fractions and AA isotopes of the same samples and address three key methodological questions: 1. Considering that blue, green and brown energy channels share some composition of primary producers, we explore whether the δ 13 C of EAAs in primary consumers provides consistent and distinct fingerprints in consumers.
2. We further ask if PUFA relative abundance and δ 13 C of EAAs can be combined in a mixing model, and whether this combined model yields different estimates and/or higher precision than the single approach models.
3. Finally, we ask how the combined approach of PUFAs and EAA δ 13 C can disentangle the coupling of PUFAs and protein transfers in a natural environment, enabling to link carbon fluxes (reflected by proteins) to food quality (reflected by PUFA abundance).To address this, we first test whether spider PUFA profiles alone are linked to dietary preferences, and second, evaluate how PUFA concentration factors from dietary sources to predators can be calculated.

| Sample collection and identification of specimen
We conducted our study at the Necker, a sixth order stream in northeast Switzerland (47.39281 N,9.09015 E) in June 2020 and May 2023.In 2020, we collected spiders as an important predator group in riparian systems at different distances from the shoreline (0-2 m, 5-10 m, 20-30 m, 40-50 m, 80-100 m), thereby distance to stream categories later refer to those categories.The spiders represented two types of hunting strategy, ground-dwelling and webbuilding spiders, thereby designated as ground and web spiders.We focused on two important ground and web spider families, Lycosidae and Tetragnathidae, respectively.Previous analysis based on bulk isotopes and PUFAs suggested a broad spectrum of source utilisation by spiders, depending on the hunting strategy and distance to the stream (Kowarik et al., 2021(Kowarik et al., , 2023)).This made spiders good candidates for investigating the transfer of subsidies in a natural system.We also collected stream invertebrates (kick samples), including mayfly and caddisfly larvae (blue sources), and terrestrial invertebrates (sweep net samples), including major herbivorous groups such as plant bugs and leaf beetles (green sources) in 2020.In addition, we collected by hand detritivorous invertebrates, comprising the most important taxa of this functional group (e.g.ground beetles, springtails, mites), in 2023 from the same site to supplement the brown sources (for details see Table S1).It is worth noting that our study uses this natural variation in source use to ask methodological questions, rather than asking the environmental and biological factors driving resource use.After transport to the laboratory, all samples were identified to the lowest practicable taxonomic level and stored at −20°C until further analysis.We categorised the potential prey in three groups, based on their habitat: blue (aquatic), brown (leaf litter) and green (grass and shrubs).No specific permit was required for our sampling.All fieldwork was conducted in accordance with Swiss guidelines and regulations.

| Sample analysis
We analysed fatty acids (FA), and AA isotopes and mass fractions, from the identical 44 samples.For the combined FA and AA analyses, we required a dry weight above 2 mg, aiming for 10 mg.Therefore, where necessary, we pooled prey individuals of the same taxonomic group, and spiders of the same distance to stream category and family.After pooling, the number of samples was nine for ground spiders (Lycosidae), 10 for web spiders (Tetragnathidae), five for aquatic (blue) invertebrates, seven for terrestrial (green) invertebrates and 13 for leaf litter (brown) invertebrates (in total 19 spider samples and 25 prey samples).We assigned each sample to coarse trophic levels: detritivores, herbivores, omnivores and predators.A list of all samples can be found in Table S1.
To analyse FAs, we followed a previous protocol (Kowarik et al., 2021) which was slightly modified, more information can be found in Data S1 (Protocol A).
We extracted AAs from identical samples.Addressing potential concerns, existing research has indicated that preceding FA extraction do not impact AA recovery in organisms (Yun et al., 2020).Crucially, the isotopic composition of AAs remains unaffected.After extracting the supernatant (lipids, see above), we retained the pellets containing the non-lipidic portion that had settled at the bottom of the tubes.We hydrolysed the pellets and subsequently derivatized the free AAs to N-acetyl methyl ester (NACME-AA) derivatives following previous protocols (Corr et al., 2007;Larsen et al., 2013).A full description of the protocol can be found in Saboret et al. (2023).We measured AA mass fractions (see Protocol B, Data S1), and isotopes of carbon (see Protocol C, Data S1).AA isotopes were run in duplicates, with a mean propagated standard deviation of 0.7‰.

| Statistical analysis
Statistical analysis was performed in R (R Core Team, 2020).

| Testing source differences in EAA δ 13 C and PUFA abundance
We first tested for mean differences in EAA δ 13 C and PUFA abundance (as % of total FA), between blue, green and brown sources, and the two spider families.Because the data were not normally distributed, we used pairwise Kruskal-Wallis tests with post-hoc comparisons using the Fisher's least significant difference, in the package agricolae (De Mendiburu & Simon, 2015).

| Using EAA δ 13 C to track blue, green and brown sources
We tested if EAA δ 13 C could disentangle blue, green and brown sources of diet.The EAA δ 13 C fingerprinting, hereby "fingerprinting", has emerged as a strong method to characterise different producer phyla (Larsen et al., 2013).The method is based on the fact that different primary producers leave consistent δ 13 C differences between EAAs (Scott et al., 2006), largely independent of growth rates, sources of carbon and δ 13 C baseline (Elliott Smith et al., 2022;Larsen et al., 2013;Stahl et al., 2023).As EAAs show little fractionation in consumers (~0-0.5‰), the fingerprinting is conserved up the food chain (Manlick & Newsome, 2022).Here, because we collected diet sources on sites that also showed a contrast in the mean EAA δ 13 C (Figure S1), we used a mixed approach based on EAAs' not-normalised values, similar to Skinner et al. (2021).First, we applied a linear discriminant analysis (LDA) using not-normalised δ 13 C of four EAAs (Isoleucine, Leucine, Phenylalanine and Valine).We considered only the primary consumers of each source (i.e.detritivore or herbivore), as they specialise on one dietary source only (e.g.only brown or only green source reliance).LDA is a powerful tool to maximise the contrast between the source's isotopic signatures, but posterior probabilities of classification do not necessarily represent mixture reliance (Elliott Smith et al., 2021).Rather, we used LD coordinates in mixing models, similar to Manlick et al. (2023).This method showed conclusive results in a control feeding experiment that distinguished between plants, fungi and bacteria diets (Manlick & Newsome, 2022).The underlying idea is to apply LDA to maximise fingerprint difference, and use the MixSIAR framework to model consumer use (see below).

| Mixing models
We implemented three mixing models based on EAA δ 13 C (model 1), PUFA profiles (model 2) and their combination (model 3) using MixSIAR, following classical guidelines (Stock et al., 2018).To this regard, we followed the general approach of using all PUFAs that fulfil model structure requirement (Galloway et al., 2015).For all models, we considered the same sources, used the same MixSIAR parameters, but used different tracers.To combine the two tracer groups in model 3, we min-max normalised LD coordinates and PUFA concentrations within tracer groups, so that the relative weight of each tracer was equal and data structure was conserved.A summary can be found in Table 1.
For the sources, we only considered samples composed of herbivores or detritivores, as those were representative of one source only (i.e.predators, like ground beetles, were excluded because they might have fed on prey of mixed sources).Some of the sources had concentration of PUFAs below detection limits (e.g.we did not TA B L E 1 Mixing models' summary.PUFAs: C18:2 n-6, C18:3 n-6, C18:3n-3, C18:4 n-3, C20:2 n-6, C20:3 n-6, C20:5 n-3 and C22:6 n-3 detect any C22:6 n-3 in brown sources).We then assigned a mean of 0, and a standard deviation equal to the minimum standard deviation observed (0.02), allowing the model to consider some natural variation we could not detect.To explore whether the three sources were significantly distinct for the three sets of markers, we compared the dissimilarity of the three sources for the different tracer sets based on ANOSIM test, using euclidean distances and 10,000 permutations.
A key assumption using isotopic mixing models, which is often overlooked when using EAA δ 13 C, is that the sources have similar relative abundance of EAAs.This holds particularly true for differentiating terrestrial and aquatic sources, which may have distinct EAA abundances (Anderson et al., 2004).To ask whether our three sources had different EAA abundances, we used a Kruskal-Wallis test.We found that blue and green sources had, relatively, lower content of Ile and Phe, respectively (Figure S2).However, the mean differences were rather marginal (maximum average difference was a 1.14 ratio), so we did not further consider those differences.
Another key assumption in mixing models is that consumers' tracers are in the range of the source (Phillips et al., 2014).To check this assumption, we first excluded every tracer for which at least one of the consumers was not in the source range.This was the case for C20:4 n-6 (a long-chain ω6-PUFA) because some ground spiders had higher values than all the sources (Figure S3).Secondly, to ensure that consumers were within the source polygon (i.e.there is a possible solution to the mixing), we performed a non-metric multidimensional scaling (NMDS) based on the Bray-Curtis distance in vegan (Oksanen, 2018) (Figure S4).NMDS is particularly suitable to check this assumption as NMDS conserves rank orders between individuals, and classifies samples based on dissimilarity distance regarding all biomarkers, rather than maximising variability.The NMDS was not necessary for the EAA δ 13 C as the source polygon could be visualised in two dimensions.
We considered no fractionation in the PUFAs mixing models (i.e.mean = 0), assuming that the relative PUFA abundance of predators largely reflects those of their prey.This approach is supported by  (Manlick & Newsome, 2022).We considered a null fractionation with small variations around it, setting fractionation factor to 0 and SD to 0.1.To make sure that setting the factor could not drive the model inferences, we performed a sensitivity analysis on the mean or SD of fractionation factor.We found that changes (mean ± 0.1 and/ or SD + 1) had little effects on the mean estimates; see Appendix A1.
For MIXSIAR parameters, we treated individuals as random factors, as individuals could randomly variate from the mean.
We set the model using "long" run, meaning 300,000 iterations (burn-in = 200,000) on three parallel Monte Carlo Markov chains with a thinning interval of 300 using non-informative priors and considering process error only (as individuals were treated as factors).
2.3.4 | Comparing mixing models based on EAA δ 13 C, PUFAs profiles, or their combination Increasing the number of biomarkers should in theory increase the discriminating power of mixing models (Brett et al., 2016).However, combining different sets of tracers in one mixing model can prove challenging (Pickett et al., 2024), as for instance AA and PUFA may show different turnover rates.Here, we asked if mixing models based on EAA δ 13 C, PUFAs profiles, or their combination provide similar estimates of dietary sources, in terms of mean estimates and precision (i.e.confidence intervals around the mean).As ground and web spiders have distinct dietary preferences and metabolism, we asked this question independently for the two families.Because of non-normality in data distribution, we employed a pairwise Wilcoxon rank-sum test, using Bonferroni corrections, for each of the three sources to examine differences between the three models in terms of mean estimates and associated standard deviation.

| Determining the link between PUFA abundance and dietary source of proteins
First, we asked which PUFA relative abundances were the most linked to dietary sources of proteins.To do so, we performed a redundancy analysis (RDA, van den Wollenberg, 1977) using the package vegan (Oksanen, 2018).RDA allows regression of multiple response variables on parameters, and is a powerful tool to fit community variance (here the relative source of protein, as inferred by EAA δ 13 C mixing model) to explanatory variables (here PUFAs).We used concentrations of nine PUFAs relative to the total FA (six ω6-and three ω3-PUFAs, see Figure 2).We used a permutation test (10,000 permutations) to measure the marginal effect of each predictor and to calculate statistical significance.
Second, we directly estimated concentration factors for PUFA mass fractions between dietary sources and spiders.Details on the calculation can be found in Appendix A2.In brief, the concentration factor represents the ratio between the mass fraction in individual spiders and the available mass fraction derived from the three dietary sources, as inferred from the EAA δ 13 C model.Therefore, we normalised PUFA abundance to protein mass fraction, as the dietary model estimated protein fluxes.Some spiders had PUFA mass fraction under detection limit, therefore the concentration factor was null and was excluded from mean estimates.We examined the influence of PUFA dietary availability on concentration factors using linear regressions.
Blue, green and brown sources showed distinct EAA δ 13 C patterns.On average, blue sources had the lowest δ 13 C values for all EAAs (Figure S1, Kruskal-Wallis, p < 0.05).The difference was the strongest for phenylalanine (5-6‰), which distinguished the most aquatic sources from green and brown sources in the LDA (Figure 1a).
Brown sources were further distinguished from green ones by relatively 13 C-enriched leucine and isoleucine (Figure 1a; Figure S1).

| Using EAA δ 13 C to track blue, green and brown sources of proteins
We used EAA δ 13 C of sources to determine consumers' source of proteins.Spiders overlapped with all sources on the LDA, supporting the reliance on the three sources (Figure 1a).The mixing model supported a wide range of sources and source mixtures used by spiders (Figure 1b), but we observed different trends depending on hunting strategy: while ground spiders mainly relied on brown sources (average 69%), web spiders were more dependent on green sources (average 42%).Among ground spiders, only the two samples with individuals collected at the shore (0-2 m from the stream) showed a strong blue reliance (Figure 1b).Among web spiders, samples collected at the shore showed higher reliance on blue sources, but the almost exclusive reliance above 80% was found in two samples collected at 5-10 and 80-100 m from the stream (Figure 1b).Details on reliance for other samples are detailed in Figure S5.

| Comparing mixing models based on EAA δ 13 C, PUFAs profiles, or their combination
We compared the output (mean estimates and precision) of mixing models based on three diet tracer groups: EAA δ 13 C, PUFAs profiles, or their combination (see summary in Table 1).Details on reliance for all samples are shown in Figure S5.
There were no overall significant differences between the three models in the two spider families (Wilcoxon rank-sum test, all p > 0.1).The three models agreed on specialisation on brown resources for several ground spiders (mean 68%-72% reliance), and a more mixed diet for web spiders on green (44%-47%), blue (24%-32%) and brown (21%-27%, Figure 2).The combined model did not show average estimates between EAA δ 13 C and
We determined concentration factors between sources and spiders, relative to protein flow.We found that concentration factors ranged between 0.2 and 0.9, except DHA that showed a high concentration factor of 5 (Figure 3b).Spiders showed high variability in concentration factors, but PUFA availability in diet was only a predictor of concentration factor for three PUFAs (Figure S6).

| General findings
Understanding trophic interactions in meta-ecosystems, such as in riparian areas, is challenging.Our study shows that carbon isotopes of EAAs can track the source of proteins in riparian areas, whether they are of blue, green and brown origins.We demonstrate that PUFA abundance and EAA δ 13 C can be combined in mixing models, which can result in higher precision.We found that the combination of the two approaches unveil different trophic transfers for the two macromolecules.Although PUFA profiles scale with protein sources, some long-chain PUFA abundance did not correlate with flow of proteins, suggesting transfer uncoupling.We showcase how to estimate concentration factors in spiders, which were variable among individuals but consistent between PUFAs, supporting their use as relative abundance in mixing models.

| Primary consumers of blue, green and brown integrate distinct EAA fingerprinting
Different phyla of primary producers evolve different anabolism pathways, resulting in distinct δ 13 C differences between EAAs, referred to as fingerprinting (Larsen et al., 2009).A challenge when applying this approach to the wild is to sample the different producers (e.g.difficulties to collect and isolate fungi and bacteria), and to assign them to a source (e.g.plants are a source of carbon for green, brown and blue consumers).Here, we found that primary consumers of the three sources had distinctive EAA δ 13 C.
Notably, blue sources were characterised by lighter Phenylalanine values, consistent with algae fingerprinting (Saboret et al., 2023), while brown sources exhibited heavier Isoleucine and Leucine values, likely influenced by bacterial de novo synthesis that results in relative 13 C-enrichment of those EAAs (Larsen et al., 2013).This supported that primary consumers of blue, green and brown integrated fingerprinting originating from producers of distinct phylogenetic origins.
Higher trophic level consumers, such as spiders, showed intermediate EAA δ 13 C, showing the reliance on different sources.Ground spiders displayed valine and isoleucine values approximately 2‰ higher than their sources, indicating potential isotopic discrimination between the prey and the spiders.Such small trophic discrimination has been reported for EAAs before (Whiteman et al., 2019), also in spiders (Pollierer et al., 2019).A combined LDA and Bayesian mixing model (Manlick & Newsome, 2022) could alleviate trophic discrimination uncertainties by relying on the dimensions that maximise differences between sources.Our EAA δ 13 C showed ecologically sound results; ground spiders near the shore (<2 m) and beetles of the genus Bembidion (that are known to rely on blue sources, Paetzold et al., 2005) showed blue reliance.In contrast, we mostly found that ground spiders and web spiders, further away from the shore, mainly relied on brown and green sources, respectively, consistent with their hunting strategies, although there was one exception in web spider, illustrating spatial heterogeneity in feeding habits.

| Combining PUFA abundance and EAA δ 13 C in mixing models
Determining the optimal for tracking resource use presents a fundamental question to ecologists, with the choice contingent upon the specific inquiry and context at hand.Increasing the number of biomarkers could increase discriminating power between sources (Brett et al., 2016), or provide additional insights.For instance, Hambäck et al. (2016) used a combination of bulk isotopes and genetic analysis to give fine estimates of spider diets.While mixing models using PUFA abundance (Galloway et al., 2015), and EAA δ 13 C (Manlick & Newsome, 2022), show great perspectives in foodweb studies, their combination has never been tested.Combining biomarkers in a mixing model can prove challenging due to different transfer pathways and mixing model assumptions (Pickett et al., 2024).
Here, we found that three mixing models, based on EAA δ 13 C, PUFAs profiles or their combination, can be easily implemented and showed similar mean estimates.We found that models based on EAA δ 13 C and PUFA yielded similar precision.This result seems conflicting with the expectation that a higher number of biomarkers would increase confidence in predictions (Galloway et al., 2015).
Instead, it appears that the substantial dissimilarity in source EAA δ 13 C values and/or the good preservation of source isotopic signal in trophic transfer, counterbalanced the effect of a higher number of biomarkers (in the PUFA-based model), of which only a few are strongly associated to dietary source of proteins (see below).
However, we found that the combined model showed on average higher precision (i.e.narrower estimate intervals).This was particularly relevant for some individuals that had large uncertainties for the unique approaches (e.g.uncertainty on green use for some ground spiders), while the combined model alleviated uncertainties.
Some individuals exhibited marked discrepancies between the two models based on EAA δ 13 C and PUFAs, such as one web spider displaying "blue EAAs" and "green PUFAs", which resulted in This might not hold for other tissues, as relative abundance of macromolecules and turnover rates depend on organs (Ebm et al., 2021;Robinson et al., 2024;Villamarín et al., 2016).Here, we used a wholebody approach, but the organ should be considered with caution.

| Quantifying protein and PUFA coupling in the wild
Polyunsaturated fatty acids are not only biomarkers, but they play a pivotal role in ecosystems, influencing trophic transfer efficiency (Müller-Navarra et al., 2000) and individual fitness (Twining et al., 2018).In particular, long-chain PUFAs (LC-PUFAs, with 20 carbon atoms and more), such as EPA and DHA, are rarely produced de novo and production from bioconversion is energy demanding (Parrish, 2009).Because they are much more abundant in blue sources (Twining et al., 2019), many studies have focused on the transfer of EPA and DHA from blue sources to terrestrial ecosystems (Fritz et al., 2019;Gladyshev et al., 2013;Moyo et al., 2017).
Our study is in line with those expectations, as DHA abundance was quasi null in non-blue sources, and given that blue sources had an order of magnitude higher EPA content than green sources.
However, we found that brown sources harboured equal abundance of EPA, and even higher abundance of several long-chain ω6-PUFAs, compared to blue sources.This finding is in line with previous studies, as some species of springtails, which are a substantial part of brown food webs, are able to biosynthesized LC-PUFAs (Chamberlain & Black, 2005;Parmar et al., 2022), and even able to make de novo synthesis of linoleic acid (Malcicka et al., 2017).There is increasing evidence that LC-PUFAs other than EPA and DHA, such as ARA, can be equally important for consumer nutrition (Ilić et al., 2019).The high abundance of ARA and other LC-PUFAs in brown sources suggest another source for terrestrial consumers to fulfil their metabolic requirements.
Such observations question the origin and transfer of PUFAs in food webs.Compound specific isotopes of FA show promising perspectives (Pilecky et al., 2021;Twining et al., 2020), although their interpretation can be difficult (Burian et al., 2020).Instead, we show here how protein-based isotope approach and PUFA abundance data can unveil their coupling.
First, we uncovered a lack of association between LC-PUFAs' abundance in spiders and their source of proteins, as inferred by EAA δ 13 C.We found that three short-chain PUFAs (SC-PUFAs, with Second, our estimates of concentration factors relative to protein fluxes showed that PUFAs were actually "diluted" from the sources to consumers, especially for SC-PUFAs.We found that concentration factors were variable between individuals, and were weakly correlated to bulk dietary preferences, in terms of assimilated proteins.This suggests that in our case other metabolic factors influenced PUFA transfers, highlighting their physiological role (Danger et al., 2022;Keva et al., 2021).
Last, it is worth noting that findings can be dependent on analytical accuracy.In our study, we could report γ-linolenic acid with confidence, despite its low concentration (<1% of total FA), and highlight its coupling to dietary proteins.In addition, we found higher concentrations of EPA in ground spiders than previously reported from the same sample collection, but this did not affect the conclusion of the study (Figure S7).These discrepancies underscore the demands placed on analytical methods and emphasise the need for more consistent practices in the literature.

| Ecological relevance and future directions
We show by using PUFAs and AA isotopes, that riparian spiders rely on a diversity of protein and PUFA sources, including blue, green and brown sources.The brown ecosystems usually receive less attention in food-web studies, despite their importance in terms of biomass and ecological process (Potapov, Guerra, et al., 2023;Potapov, Lindo, et al., 2023).However, our results supported the transfer of PUFAs from brown sources, especially springtails, highlighting the importance of brown sources as high-quality food sources.
Here, we propose a synergistic use of AA isotopes and PUFAs to describe the consumer's niche: the EAA fingerprinting aligns with resource use and processes at the food web's foundation, while PUFAs illuminate the metabolically realised niche, offering a nuanced view of ecological dynamics.

2041210x, 0 ,
Downloaded from https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.14371by Paul Scherrer Institut PSI, Wiley Online Library on [03/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License acids (Ile, Leu, Val, Phe) δ 13 C Pickett et al. (2024) that found that mixing models are robust to variations in PUFA fractionation factors.We accounted for slight metabolic effects (Chiapella et al., 2021), by setting the standard deviation of trophic discrimination factor in MixSIAR to 10% of the source mean values.For the combined mixing model, we set this standard deviation to 5% to all of the tracers, an intermediate value between EAA δ 13 C and PUFA.EAAs can show little fractionation in consumers (McMahon & Newsome, 2019), but because the LD axes are linear combinations of EAA δ 13 C, they are conservative along the food chains

F I G U R E 1
Carbon isotopes of essential amino acids (EAAs) elucidate blue, green and brown dietary sources of proteins.(a) Linear discriminant analysis of blue, green and brown sources based on four EAAs δ 13 C of primary consumers of the three groups.EAAs (Ile, Leu, Phe and Val) positions show their relative contribution to the linear discriminant axis.Ellipses show 50% confidence intervals of the sources.(b) Ternary diagram showing the relative sources of amino acids in spiders, as estimated by the mixing model (based on δ 13 C of EAAs).Numbers show the category of distance to the stream at capture.PUFA model: rather, it slightly increased the reliance on blue sources (1%-8%) for both spider groups compared to both single approaches.The combined model showed reliance that were not significantly different (when considering precision around estimates) from the other two models, for all individuals, except for two web spiders (5 and 40 m distance from the shore).Those two individuals had very distinct reliance between the EAA δ 13 C and the PUFA model, which predicted reliance on blue and brown (>80% reliance, EAA δ 13 C model), and green and green (PUFA model), respectively.The combined model confirmed blue reliance of the web spider collected at 5 m distance (as in the EAA δ 13 C model), and green reliance of the web spider collected at 40 m distance (as in the PUFA model), but did not show intermediate results (Figure2A).Some spiders' reliance differed drastically between models (FigureS6).For instance, two web spiders (40 m distance) had a main predicted reliance on green and brown sources, as derived from the EAA δ 13 C and the PUFA model, respectively.However, credible intervals overlapped (as they were particularly large for the EAA δ 13 C model), and in that case the combined model showed an intermediate reliance between brown and green, with however narrower credible intervals.We found that the combined model had on average a higher precision than the EAA δ 13 C model, which was significant regarding brown estimates in web spiders (p < 0.05).F I G U R E 2 Comparison of methods to infer dietary preferences of spiders.Mixing models' outputs (source reliance) in spiders for the three sources, from top to bottom: blue, green and brown.(A, B) Show the source reliance for web and ground spiders, respectively.Distributions are visualised using boxplots, where the central line represents the median.The box encompasses the middle 50% of the data (from the 25th to the 75th percentile), while the whiskers extend to show the full range of the data, excluding outliers.Numbers on top show the means of mean reliance (top row) and standard deviation (bottom row) for each model.Letter groupings in the bottom left panel show significant differences between pairs based on Wilcoxon rank-sum test for p < 0.05.EAA, essential amino acid; PUFA, polyunsaturated fatty acid.

F
Identification of PUFAs that are linked to dietary sources of carbon.(a) Redundancy analysis on the source of proteins (as predicted by EAA δ 13 C), constrained by PUFAs proportions, relative to total fatty acid concentration.Triangle colours and shape show diet source (see bottom right triangle) and spider hunting strategy (see bottom left legend), respectively.Arrows show the relative contribution of PUFAs to the axis.Arrow thickness shows the proportion of variance explained in the spiders' diet (see upright legend).Stars show individual significance of ANOVA permutation test (10,000 permutations): **p < 0.01, *p < 0.05.(b) Concentration factors for the different PUFAs in spiders.Dotted line shows the one line (i.e.same relative abundance between sources and spiders).Triangles show individuals [see legend in (a)], and red dots and above bars show mean and standard deviation for all spiders, excluding spiders with null concentration.On top, red numbers show the mean, and stars denote positive relationship (linear regression) between PUFA abundance in diet and concentration factor: **p < 0.01, *p < 0.05.EAA, essential amino acid; PUFA, polyunsaturated fatty acid.2041210x, 0, Downloaded from https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.14371by Paul Scherrer Institut PSI, Wiley Online Library on [03/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License larger uncertainty in the combined model.This divergence could be attributed to differences in turnover rates between the two macromolecule families, with the possibility that these individuals recently shifted their diet, resulting in bad model combining.While EAAs' turnover can vary considerably in animals [the only reported turnover rates ranged from days/weeks in fish(Matley et al., 2016), to years in sharks(Whiteman et al., 2018)], limited information is available regarding PUFAs' turnover.While to the best of our knowledge no information exists for arthropods, the consistency of most outcomes indicates a similar turnover rate between EAAs and PUFAs.

2041210x, 0 ,
Downloaded from https://besjournals.onlinelibrary.wiley.com/doi/10.1111/2041-210X.14371by Paul Scherrer Institut PSI, Wiley Online Library on [03/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 18 carbon atoms or less) were strongly correlated to spiders' protein source.This observation could be explained by the potential ability of spiders to bioconvert SC-PUFAs into LC-PUFAs or preferably accumulate LC-PUFAs in their tissue (Mathieu-Resuge et al., 2022), making it more difficult to track the source of these PUFAs.