Contents lists available at ScienceDirect Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc Diet selection of white-tailed deer supports the nutrient balance hypothesis Jacob L. Dykes a, * ,1 , Bronson K. Strickland a , Stephen Demarais a , Daniel B. Reynolds b , Marcus A. Lashley c a Department of Wildlife, Fisheries, and Aquaculture, Mississippi State University, Mississippi State, MS, 39762, USA b Department of Plant and Soil Sciences, Mississippi State University, Mississippi State, MS 39762, USA c Department of Wildlife Ecology and Conservation, University of Florida, 110 Newins Ziegler Hall, Gainesville, FL, 32611, USA A R T I C L E I N F O Keywords: herbivore nutrition forage quality plant nutrients nutrient balance hypothesis A B S T R A C T Herbivores must navigate a heterogeneous matrix of nutrients in plant communities to meet physiological re- quirements. Given that the only difference between an essential nutrient and a toxin is the concentration in the herbivores diet, heterogeneity of nutrient concentrations in plant communities likely force wild herbivores to balance intake of abundant nutrients that may reach toxic levels with the need to meet nutritional demands of rare nutrients (i.e., nutrient balance hypothesis). While this hypothesis has been demonstrated in controlled studies with captive herbivores, experiments testing the nutrient balance hypothesis with wild herbivores are rare. We designed a cafeteria-style experiment to measure use of forages with differing nutritional compositions by wild white-tailed deer ( Odocoileus virginianus ) to test the nutrient balance hypothesis. We predicted deer diet selection would be explained by attraction to some nutrients and avoidance of others. Deer selected forages with low sulfur concentrations, a nutrient that commonly reaches toxic levels in herbivores. However, deer secon- darily selected forages with greater digestibility and crude protein. Thus, our data indicate that the nutrient balance hypothesis may explain diet selection in wild herbivores where they avoid reaching toxicity of abundant nutrients while secondarily maximizing intake of limiting nutrients. 1. Introduction Optimal foraging theory explains animal foraging behavior with a key prediction that animals strive to optimize a tradeoff between en- ergy intake and expenditure (MacArthur and Pianka, 1966). Support for this theory has demonstrated that balancing time with energy intake is essential to animal fitness (Gillette et al., 2000; Sinervo, 2013). Large herbivores alter bite size, cropping rates, and patch choice to optimize intake rate of digestible plants during grazing events (Distel et al., 1995; Kenney and Black, 1984). Large herbivores may also prioritize diet quality over quantity to minimize digestive constraints when available daily grazing time is restricted (Ginane and Petit, 2005; Westoby, 1974). When forage quality is held constant, handling time becomes the key factor in selection as demonstrated by clam size selectivity of common crow ( Corvus caurinus , Richardson and Verbeek, 1987) and mussel size selectivity of oystercatchers ( Haemotpus astralegus , Meire and Ervynck, 1986). When handling time is held constant but forage quality varies, selection for limiting nutrients becomes the key factor, as demonstrated with white-tailed deer ( Odocoileus virginianus ) selection of energy over protein in mixed rations with similar handling times (Berteaux et al., 1998). Because these factors influencing intake are rarely standardized or stable in the environment and they are overlain with the need to meet fluctuating nutritional demands, herbivores have developed a complex physiology allowing diet selection to quickly ad- just to post-ingestive feedbacks and social learning (Forbes and Provenza, 2000). Nutrient availability fluctuates frequently in natural environments for many reasons (e.g., seasonal changes, plant senescence, and rainfall, Bormann and Likens 1967; Himelblau and Amasino, 2001; Lashley and Harper, 2012). As such, natural selection should favor highly selective and flexible foraging behaviors in herbivores to navigate the hetero- geneous matrix of nutrients in plant communities and allow animals to seek out and exploit limiting nutrients (Hättenschwiler et al., 2008). Evidence from a variety of study systems have provided support for this selectivity and flexibility in diet selection (Forbes and Provenza, 2000). For example, Ceacero et al. (2009) observed captive Iberian red deer ( Cervus elaphus hispanicus ) selected forages high in Na and Co when made deficient in feeding trials. Similarly, wild ungulates have the https://doi.org/10.1016/j.beproc.2020.104196 Received 2 March 2020; Received in revised form 6 July 2020; Accepted 8 July 2020 ⁎ Corresponding author. E-mail address: jacobd4092@gmail.com (J.L. Dykes). 1 Present Address: Caesar Kleberg Wildlife Research Institute, Texas A&M University - Kingsville, Kingsville, TX 78363, USA. Behavioural Processes 179 (2020) 104196 Available online 23 July 2020 0376-6357/ © 2020 Elsevier B.V. All rights reserved. T ability to select plants high in phosphorus (Lashley et al., 2015; Dykes et al. 2018), protein (Dostaler et al., 2011), or energy (Berteaux et al., 1998) when limiting, and modify plant selection to accommodate fluctuations in resource availability, such as drought induced changes in the plant community (Lashley and Harper, 2012; Cain et al., 2017). Likewise, when many nutrients are limiting, diet selection should maximize an array of limiting nutrients. For example, McNaughton (1990) reported seasonal movements of African migratory grazers were linked to a wide range of forage nutrients (i.e., Ca, Cu, N, Na, Zn, Mg, P, and Ca/P balance). Contrastingly, when no nutrients are limited, forage nutritional composition may poorly explain diet selection. For example, Vangilder et al. (1982) analyzed white-tailed deer rumen samples and were unable to detect any nutrient that explained diet selection. Simi- larly, Dumont (1997) reported ruminants chose a mixed diet when of- fered different forages despite one forage adequately meeting the ani- mal’s nutritional requirements. Importantly, evidence from many herbivore systems reflect a highly specialized foraging trait in herbi- vores that allows them to adjust diet selection rapidly based on context (Forbes and Provenza, 2000). Although the majority of studies test predictions related to max- imizing intake of limiting nutrients (i.e., nutrient maximization hy- pothesis), recent experiments provide evidence that avoidance of toxi- city is a more important consideration in diet selection (i.e., nutrient avoidance hypothesis, Ceacero et al. 2015). For example, Ceacero et al. (2015) determined that Iberian red deer avoided toxic levels of sulfur even when other essential nutrients were limiting. They posed that avoiding toxicity may supersede selection for limiting nutrients because toxicity has a more acute negative effect on fitness than deficiency. Thus, animal foraging behavior seems to be flexible enough to ac- commodate the fluctuations in nutrient availability and physiological state of the animal and thus may vary widely even within an herbivore species. In fact, herbivorous insects have been documented to select diets in a way that not only prevents a surplus or deficiency of a nu- trient but also maintains a specific balance between specific nutrients (Chambers et al., 1995; Raubenheimer and Simpson, 1997; Rau- benheimer and Simpson, 2003). Given the flexibility of foraging behavior, herbivores may actually optimize a tradeoff between acquiring limiting nutrients while avoiding nutrient toxicity, supporting Westoby’s (1997) nutrient balance hy- pothesis. This is a fundamental shift in thought from previous hy- potheses because it predicts animals will maximize intake of limiting nutrients until reaching toxic amounts of another nutrient (Forbes and Provenza, 2000). Thus, abundant nutrients are more likely to regulate the consumption of limiting nutrients rather than vice versa as was the conventional line of thought. However, studies often consider too few nutrients to adequately test the nutrient balance hypothesis of diet se- lection (e.g., in white-tailed deer studies; 4 in Nelms, 1996; 2 in Berteaux et al., 1998; 2 in Dostaler, 2011). Moreover, even when many nutrients are considered, quantifying diet selection in wild populations is notoriously difficult and often researchers utilize indirect measures of selection such as bite counts (Lashley et al., 2011; Lashley and Harper, 2012), microhistological surveys (Alipayo et al., 1992; Marrero and Nogales, 2005; Lashley et al., 2015; Jung et al., 2015; Lashley et al., 2016), estimates of plant biomass removal (Lunceford, 1986), or cap- tive animal observations (Nelms, 1996; Ceacero, 2009; Ceacero, 2015). However, each of those measurements have associated biases making accurate estimates difficult (Lashley et al., 2016). Captive populations allow more controlled experimental manipulations of nutrient avail- ability and precise measures of animal selection, but may be limited in inferential power to predict how environmental stressors and resource Fig. 1. Hypothetical representation of the nu- trient maximization hypothesis (a), the nu- trient avoidance hypothesis (b), and the nu- trient balance hypothesis (c) when two nutrients (i.e., nutrients X and Y) vary in their rate of procurement when foraging (diagonal lines), the amount required by the animal (horizontal dashed lines), and the level at which the respective nutrient becomes toxic (horizontal solid lines), given the animal cannot increase intake once reaching a phy- siological threshold in handling time (i.e., in- take constraint) or a threshold in nutrient toxicity (i.e., toxicity constraint). The shaded area represents the compensation accom- plished by selective foraging to meet each re- quirement and constraint under the given hy- pothesis. The nutrient maximization hypothesis (panel a), assumes that nutrient toxicity does not limit intake because it is not prevalent enough to exceed the toxicity con- straint before reaching the physiological intake constraint and thus, selective foraging pri- marily accomplishes increased intake of the limiting nutrient Y. Contrastingly, the nutrient avoidance hypothesis (panel b) assumes that some nutrients can easily become toxic before an animal reaches its physiological intake constraint but other nutrients are not limiting if the animal maximizes its physiological in- take constraint and thus, selective foraging primarily accomplishes the reduction of intake of the toxic nutrient X, allowing the animal to increase intake to meet the requirements of limiting nutrient Y. The nutrient balance hy- pothesis (panel c) assumes that given the physiological intake constraint, nutrient X is abundant enough to reach toxicity and nutrient Y is limiting even when the animal maximizes the physiological intake constraint and thus, selective foraging reduces the rate of intake of nutrient X but also secondarily increase the rate of intake of nutrient Y, allowing the animal to avoid toxicity of the abundant nutrient while also meeting the requirement of the limiting nutrient. J.L. Dykes, et al. Behavioural Processes 179 (2020) 104196 2 heterogeneity affect diet selection in wild populations (Spalinger et al., 1997; Forbes and Provenza, 2000). Thus, a robust experimental design that allows manipulation of a wide variety of nutrients, and a more unbiased direct quantification of diet selection in wild herbivore po- pulations is needed to test diet selection. We designed a cafeteria-style field experiment where we manipu- lated the availability of nutrients using 15 plant species that varied in nutritional composition to test the nutrient balance hypothesis in a wild population of white-tailed deer (hereafter deer). To improve previous limitations in quantification of diet selection in wild populations, we directly monitored deer use of plants remotely with camera traps rather than relying on indirect measures of selection or potential biases as- sociated with observer presence. Conceptually, diet selection driven by maximizing limiting nutrients assumes that toxicity is not an important constraint given the possible intake of an animal (Fig. 1a). That is, the animal can consume as much as physiologically possible to meet the limiting nutrient constraint. Contrastingly, diet selection driven by avoidance assumes that limited nutrients are only limited because the toxicity constraint limits intake (Fig. 1b). That is, diet selection is simply avoidance of plants with high levels of toxic nutrients which allows animals to increase forage intake and as an artifact of increased intake, meet the limiting nutrient constraints. However, in nature, avoiding nutrient toxicity and acquiring limiting nutrients are neces- sary if both simultaneously exist as constraints (Fig. 1c), and in many cases, actively avoiding toxicity while secondarily maximizing intake of limiting nutrients may be necessary through diet selection (Forbes and Provenza, 2000). That is, only maximizing limiting nutrients may result in reaching toxicity in abundant nutrients while only avoiding abun- dant nutrients may result in deficiencies in limiting nutrients. To test the nutrient balance hypothesis, we tested a key prediction that deer diet selection should be partially explained by attraction to some nu- trients and partially by avoidance of others if animals indeed selectively forage to meet both constraints. 2. Materials and Methods 2.1. Study Area We conducted this study at the Andrews Forestry and Wildlife Laboratory, a 220 ha research property, located in the Interior Flatwoods soil region of Oktibbeha County, Mississippi, USA. The property is comprised of about 10 ha of agronomic fields with the re- maining 210 ha being loblolly pine ( Pinus taeda ) forest. 2.2. Experimental Design We used a cafeteria-style experiment in a randomized block design by planting monoculture plots of 15 cool-season plants varying in nu- tritional composition in each of 8 blocks. In addition, we included 4 blocks that were part of a secondary experiment (Dykes et al. 2018). These blocks consisted of only 2 of the 15 cool-season plants each planted in 4 monoculture plots (block differences addressed in analysis, see Data Analysis section). Each block contained a fallow plot which was primarily bare soil with little plant cover. We randomized place- ment of plots within each block to ensure estimates were not biased by microsite conditions (e.g., moisture, shade, slope gradients, animal trails). Each plot was approximately 0.02 ha in size (Fig. 2A) and protected from herbivory by electric fencing. The 15 plants were common cool-season agronomic crops: ladino white clover ( Trifolium repens ), durana white clover ( T. repens ), crimson clover ( Trifolium in- carnatum ), arrowleaf clover ( Trifolium vesivulosum ), balansa clover ( Trifolium michelianum ; used in additional blocks), berseem clover ( Trifolium alexandrium ), red clover ( Trifolium hirtum ), winter wheat ( Triticum aestivum ; used in additional blocks), bob oats ( Avena sativa ), rye ( Secale cereal ), rape ( Brassicas spp. ), turnips ( Brassicas spp. ) chicory ( Cichorium intybus ), ryegrass ( Lolium multiflorum ), and Austrian winter peas ( Pisum sativum subsp. arvense ). To avoid potential biases associated with a novel forage being introduced onto the landscape, each plant was cultivated and left unprotected on the property for two consecutive years prior to the study. Before planting for the study, we cultivated (e.g., herbicide application, mowed, disked, tilled) fields and amended soils to maximize productivity of the respective plant species according to soil analyses performed at Mississippi State University’s Soil Lab (http://extension.msstate.edu/content/soil-testing). All planting was completed in fall 2016 and no herbicide was applied after planting. The electric fencing was designed to exclude deer from the plots to allow each plant species to establish before beginning trials. This is an im- portant strength of the design because the plants varied widely in their maturity succession and thus, without allowing all plants to become established, diet selection could have reflected the successional pattern of the plants if herbivores had access before the later succession species were established. To test the nutrient balance hypothesis, we collected plant samples and removed protective fences from 2 spatially separated blocks for each trial (n = 6). Trial 1 began on February 1, 2017 (Trial 2 on March 12, 2017, Trial 3 on April 8, 2017; additional blocks, Trial 4 on April 26, 2017; additional blocks, Trial 5 on May 10, 2017, and Trial 6 on June 15, 2017) at which point all plant species had established and were considered available. Each trial ran for a period of 14 days be- cause that time period was short enough duration that no plots were depleted by herbivory but long enough to ensure appropriate sample sizes were obtained to make accurate and precise estimates of behavior (i.e., minimum of 100 detections and 2 weeks of sampling re- commended for other behavioral inferences, Rowcliffe et al., 2014; Lashley et al., 2018, Kays et al. 2020). We temporally replicated the experiment beginning 2–4 weeks after the previous trial concluded until completing all trials on June 29, 2017 (completion of Trial 1 on February 15, 2017, Trial 2 on March 26, 2017, Trial 3 on April 22, 2017, Trial 4 on May 9, 2017, Trial 5 on May 25, 2017, and Trial 6 on June 29, 2017). Plants were maturing at different rates during our trials and because forage maturity effects the nutritional composition of the plant (Lashley et al., 2014a; Fig. 2B), temporal replication effectively allowed us to reorganize relative nutrient concentrations of the same plants over time. Thus, our unique design allowed us to detangle if herbivores were selecting specific species or if selection was better explained by the change in relative nutritional composition over time. All forage plots and the fallow plot within each block were monitored with motion-triggered camera traps set to a 1 minute delay between pictures. 2.3. Plant Sampling To measure forage availability and nutritional composition, we sampled all forage biomass ≥ 4 cm in height within a randomly placed 0.80 m 2 sampling quadrat. We followed forage handling protocol pre- sented by Lashley et al. (2014a) where samples were dried in a con- vection oven at a temperature of 47 °C to constant mass. Once dry, samples were weighed to the nearest gram and shipped to Dairy One Forage Lab (http://dairyone.com) for nutrient analysis, which is certi- fied by the National Forage Testing Association. We measured 14 nu- tritional parameters potentially influencing diet selection: crude protein (CP), protein solubility (Prot. Solub), neutral detergent fiber (NDF), calcium (Ca), phosphorus (P), magnesium (Mg), potassium (K), sodium (Na), iron (Fe), zinc (Zn), copper (Cu), manganese (Mn), molybdenum (Mo), and sulfur (S); (McDowell, 1992; Robbins, 1993; Table 1). 2.4. Animal Sampling We set camera traps monitoring the entirety of each plot, 0.5 m off the ground on a stake 1 m from the plot edge, and set to a 1 minute delay between photographs. Because camera sensitivity could vary among units, we took a subsample of the cameras and tested them J.L. Dykes, et al. Behavioural Processes 179 (2020) 104196 3 within a captive deer facility and within a captive feral pig facility. None of the units were obvious outliers in sensitivity. Thus, we ran- domly assigned cameras to each forage plot in each trial and assumed similar sensitivity. We tallied deer detections and posture of each deer. Posture was recorded as alert, actively feeding, or searching. Similar to previous experiments using camera traps to monitor foraging behavior, we used the actively foraging posture as an indication of forage selec- tion (Lashley et al., 2014b; Cherry et al., 2015; Biggerstaff et al., 2017; Schuttler et al., 2017). Although animals could be handling plants during other behaviors, we considered our metric of use conservative because non-feeding behaviors such as antipredator vigilance are likely obscured by peripheral vegetation in the head down posture (Lashley et al., 2014b). We considered each photograph a single detection if at least one individual in the picture was in the feeding posture. Because the amount of time an individual chose to forage in a particular forage has relevance to diet selection, we chose to use all detections in our analysis as opposed to setting an arbitrary time frame threshold for independent observations which is more common in camera trap stu- dies based on occupancy metrics. Chance et al. (2020) validated this approach by using camera traps to estimate deer habitat selection. 2.5. Data Analysis We used JMP 13 (SAS Institute Inc., Cary, NC) for all analyses. To test the nutrient balance hypothesis, we used a multistep model selec- tion process with Standard Least Squares (SLS) Regression. To ensure results were not biased by failed plots (i.e., too little biomass to be attractive), we plotted the distribution of biomass for each and removed the 25% quartile for each forage type considering them failed plots. As a result, chicory and Austrian winter pea plots were removed because of drought-induced crop failure. Trial 5 was also excluded from analysis due to missing biomass data. We then quantified the severity of mul- ticollinearity between plant nutritional qualities and removed qualities with Variance Inflation Factor ≥ 5 (Belsley et al., 2005; Craney and Surles, 2002). We adjusted deer use data for each planted plot by subtracting fallow plot use (i.e., number of individuals photographed in feeding position) to account for random visitation not associated with selection. We used deer detections in the feeding position in fallow plots, which largely were absent of vegetation, as an index of how often we detected deer in the feeding position but not foraging. This was an important aspect of the design because some observations may be spurious because deer are in a feeding position but not eating or within the plot because of travelling to an adjacent plot. Because the fallow Fig. 2. (A) Experimental design quantifying white-tailed deer ( Odocoileus virginianus ) se- lection across plants of varying nutritional composition. Fifteen plants were randomly as- signed and planted in equal-sized adjacent plots (i.e., Block), protected by electric fencing until establishment, and then monitored using camera traps at Andrew’s Forestry and Wildlife Laboratory, Oktibbeha County, MS, USA, Spring 2017. (B) Example of changing forage nutrient concentration over time. Crude pro- tein (CP) concentration in (a) ladino clover ( Trifolium vesivulosum) , (b) bob oats ( Avena sativa ), and (c) turnips ( Brassicas spp .) changes across trials as plants mature at Andrew’s Forestry and Wildlife Laboratory, Oktibbeha County, MS, USA, Spring 2017. J.L. Dykes, et al. Behavioural Processes 179 (2020) 104196 4 plot had little forage and was randomly assigned within each replicate, we used them as an indicator of those spurious observations. We were unable to confidently discern males from females due to spring antler casting (Jacobson and Griffin, 1982). Thus, we did not distinguish be- tween males and females in our analysis even though there is evidence that diet selection among males and females could differ (Kie and Bowyer, 1999). We log-transformed deer use data to address assump- tions of normality. We used trial as a random effect to control for varying deer physiological requirements over time and block differ- ences (Kammermeyer and Miller, 2006). We fit the final model with deer use as the response variable and plant nutrient concentrations as predictor variables and alpha of 0.05 as the threshold to determine significance. 3. Results We collected 13,758 photographs containing 18,527 deer with an average of 1.6 individuals per photograph. Of those, 54% of photo- graphs had an individual in the feeding posture and was classified as a use event. We obtained a sample size > 100 in all plots. Due to high multicollinearity between plant nutrients, we removed Ca, Mg, and Fe from our model. After removing those variables, the remaining 11 plant nutrients explained over two-thirds of the variation in deer use (R 2 = 0.68; Table 2). Deer use was explained by attraction to CP (p = 0.0285) and the avoidance of S (p = 0.0276) and NDF (p = 0.0004; Table 1). 4. Discussion Diet selection was positively influenced by a potentially limited nutrient (i.e., crude protein) and negatively influenced by a potentially toxic nutrient (i.e., sulfur) supporting the nutrient balance hypothesis. Had we only evaluated a few nutrients, as is common in the literature, we may have misinterpreted our results. For example, if we had only considered CP and fiber content such as done in Donstaler et al. (2011), we too would have concluded that nutrient maximization was driving selection in full support of the nutrient maximization hypothesis. Similarly, if we had only considered mineral nutrients, we would have concluded that nutrient avoidance was driving diet selection in full support of the nutrient avoidance hypothesis (Ceacero et al., 2015). However, as suggested by Ceacero et al. (2015), considering a wide range of plant nutrients allowed us to determine both nutrient max- imization and avoidance were simultaneously influencing diet selection in support of the nutrient balance hypothesis. Moreover, it is important to note that the specific nutrient being selected or avoided in this study is likely context dependent and should not be interpreted as an absolute Table 1 Mean (highest observation-lowest observation) nutrient concentrations for each agronomic forage collected at Andrew’s Forestry and Wildlife Laboratory, Oktibbeha County, MS, USA, Spring 2017. Forage CP Prot.Solub NDF P K Na PPM. Zn PPM.Cu PPM.Mn PPM.Mo S Arrowleaf 18 (10-26) 52 (38-67) 41 (23-57) 0.32 (0.22-0.45) 2.13 (1.46-2.96) 0.08 (0.02-0.18) 32 (15-55) 8 (6-9) 224 (97-740) 2.01 (0.4-6.2) 0.21 (0.15-0.29) Balansa 19 (12-26) 55 (31-70) 40 (25-64) 0.29 (0.14-0.5) 2.57 (1.32-5.04) 0.16 (0.02-0.4) 33 (14-79) 10 (7-16) 177 (62-483) 1.05 (0-3.8) 0.24 (0.15-0.39) Berseem 16 (12-18) 43 (39-46) 52 (45-61) 0.27 (0.2-0.32) 1.74 (1.05-2.9) 0.18 (0.04-0.47) 26 (22-30) 9 (7-12) 525 (106-1050) 1.6 (0.4-3.1) 0.19 (0.13-0.28) Chicory 17 (9-21) 45 (32-52) 37 (26-45) 0.34 (0.24-0.49) 3.55 (2.58-5.22) 0.53 (0.1-1.54) 40 (24-56) 9 (5-14) 393 (113-860) 1.15 (0.4-2.1) 0.38 (0.29-0.6) Crimson 14 (6-21) 38 (33-52) 50 (24-74) 0.24 (0.11-0.35) 1.55 (0.75-3.32) 0.06 (0.03-0.1) 43 (21-98) 10 (6-25) 883 (68-2650) 1.31 (0.4-2.8) 0.18 (0.1-0.26) Durana 20 (19-21) 50 (42-57) 42 (36-46) 0.33 (0.25-0.38) 2.43 (1.91-3) 0.15 (0.02-0.36) 23 (20-29) 10 (8-13) 220 (107-371) 1 (0.5-1.8) 0.23 (0.18-0.37) Ladino 19 (11-23) 52 (37-65) 42 (32-46) 0.34 (0.21-0.5) 2.57 (1.82-3.22) 0.23 (0.03-0.4) 27 (23-36) 12 (9-23) 582 (78-2070) 1.18 (0.5-2.7) 0.19 (0.14-0.24) Oats 12 (4-23) 49 (34-60) 50 (18-73) 0.27 (0.08-0.43) 1.85 (0.54-2.86) 0.23 (0.02-0.48) 24 (12-33) 6 (3-8) 314 (84-973) 0.69 (0.3-1.6) 0.16 (0.05-0.27) Peas 18 (13-26) 52 (39-70) 49 (32-74) 0.31 (0.23-0.43) 2.14 (1.54-3.21) 0.02 (0.01-0.06) 36 (25-51) 9 (6-16) 416 (56-1180) 2.11 (0.6-3.5) 0.24 (0.15-0.27) Rape 14 (8-17) 59 (46-73) 37 (18-46) 0.35 (0.24-0.45) 2.25 (1.64-3.01) 0.05 (0.01-0.13) 35 (21-100) 5 (3-9) 272 (76-762) 1.5 (0.6-3.2) 0.53 (0.35-0.64) Red 19 (12-23) 49 (43-56) 42 (33-56) 0.28 (0.22-0.34) 2.8 (2.44-3.32) 0.02 (0-0.06) 24 (17-33) 10 (7-18) 208 (62-439) 1.53 (0.8-2) 0.19 (0.11-0.28) Rye 13 (7-17) 51 (36-59) 56 (41-76) 0.24 (0.2-0.32) 1.55 (0.82-2.23) 0.01 (0.01-0.02) 21 (15-27) 5 (4-7) 230 (70-720) 0.32 (0.2-0.6) 0.15 (0.11-0.2) Ryegrass 11 (5-21) 55 (45-67) 56 (34-80) 0.25 (0.14-0.53) 2.24 (1.16-3.63) 0.07 (0.02-0.21) 32 (15-93) 6 (3-10) 263 (91-636) 1.41 (0.8-2.5) 0.18 (0.1-0.3) Turnips 15 (5-24) 59 (54-66) 44 (28-73) 0.39 (0.24-0.5) 2.67 (2.15-3.33) 0.08 (0.02-0.19) 29 (10-39) 5 (2-8) 581 (30-1660) 1.29 (0.5-2.5) 0.49 (0.25-0.66) Wheat 11 (6-18) 59 (36-69) 53 (26-70) 0.23 (0.16-0.38) 1.94 (1.04-2.6) 0.01 (0-0.04) 22 (10-93) 6 (2-8) 130 (40-624) 0.41 (0-1.3) 0.16 (0.1-0.23) All nutrient concentrations are presented as percentages except where specified PPM (Parts Per Million). Nutrients included are crude protein (CP), protein solubility (Prot. Solub), neutral detergent fiber (NDF), phosphorus (P), potassium (K), sodium (Na), zinc (Zn), copper (Cu), manganese (Mn), molybdenum (Mo), and sulfur (S). Table 2 Model output for standard least squares regression predicting white-tailed deer ( Odocoileus virginianus ) plant selection as a function of twelve plant nutrient concentrations at Andrew’s Forestry and Wildlife Laboratory, Oktibbeha County, MS, USA, Spring 2017. Term Estimate Standard Error DFDen t Ratio Prob > |t| Intercept 5.4039 2.6026 53.9000 2.0800 0.0426 Biomass 0.0026 0.0054 57.5600 0.4900 0.6241 Crude Protein 0.1754 0.0780 56.2100 2.2500 0.0285 Protein Solubility 0.0085 0.0404 53.5800 0.2100 0.8338 Neutral Detergent Fiber −0.0929 0.0248 58.7600 −3.7500 0.0004 Phosphorus −5.8525 4.0412 57.4600 −1.4500 0.1530 Potassium 0.6874 0.5050 56.8600 1.3600 0.1789 Sodium 5.6961 3.0806 55.7300 1.8500 0.0698 Zinc −0.0042 0.0158 58.3900 −0.2600 0.7932 Copper −0.1388 0.1259 58.7600 −1.1000 0.2749 Manganese 0.0002 0.0007 58.8100 0.2800 0.7797 PPM Molybdenum 0.2448 0.2409 58.8600 1.0200 0.3138 Sulfur −5.9447 2.6290 56.7100 −2.2600 0.0276 J.L. Dykes, et al. Behavioural Processes 179 (2020) 104196 5 mechanism of diet selection in this species. For example, concurrent studies in the same population reported that common native plants in the study area contain relatively high leaf S concentrations (i.e., Soli- dago spp. and Ambrosia spp.; Chance et al. 2019, 2020). It is plausible that this population is exposed to a high background S abundance in the native plant community and that manifested in S avoidance in our trial. A population with lower background S may have resulted in a change in nutritional constraints driving diet selection. Likewise, if no nutrients are abundant or none are limiting, either of those constraints on plant selection could be relaxed. None-the-less, our study provides evidence that diet selection is driven by nutrient balance and supports the idea that avoiding nutrient toxicity is considered first in the hierarchy of decision making (Ceacero et al., 2015). Deer selected plants low in sulfur, and although the level of S that results in toxicity has not been reported in wild ungulates, 2000 mg/kg is toxic to cattle ( Bos taurus ) and domestic ruminants (Drewnoski et al., 2014; Zinn et al. 1997). Moreover, Spears et al. (2011) reported a linear decrease in average daily gains and dry matter intake when dietary S increased from 0.12 to 0.46%. Other studies involving domestic rumi- nants have reported S concentrations above 0.2% resulted in defi- ciencies in Cu and Se, possibly leading to severe diseases (Spears et al., 2011; Cammack et al., 2010; Ivancic and Weiss, 2001; McBride, 2007). We expect the threshold of S toxicity and subsequent effects in deer to be comparable to those of domestic ruminants; thus, deer may have avoided consuming large concentrations of S to avoid toxicity. Our data support this notion as some of our plants consistently exceeded 0.2% S (e.g., avg. 0.526% in rape and 0.489% in turnips) and were subse- quently avoided in our experiment. Diet selection for acquiring limiting nutrients must compensate for this avoidance secondarily, as animals will avoid toxicity even at the cost of deficiency (Hill et al., 2009; Ceacero et al., 2015). Again, this species may not avoid S when it is not abundant in the available forages. Deer in our study likely increased selection of plants as neutral detergent fiber decreased because NDF is a measure of forage bulkiness which is negatively correlated with intake (Mertens, 1987). Other stu- dies reported similar decreased voluntary intake as forage NDF in- creased in domestic ruminants, attributable to the overall decrease in digestibility (Mertens, 1987; Harper and McNeill, 2015). And, Don- staler et al., (2011) reported attraction to highly digestible forages in captive white-tailed deer. Thus, deer in our study likely avoided high NDF to relax their physiological intake constraint. The positive selection for crude protein corresponds to other diet selection studies in ruminants (Danell et al., 1994; Tixier et al., 1997; Deguchi et al., 2001; Dostaler et al., 2011). Selection of CP would support increased protein demands to meet the physiological needs of deer during the spring season (Robbins, 1993) and would have in- creased in demand over the course of our experiment because protein requirements are heightened in the third trimester of gestation (Parker et al., 2005). Thus, using the trial period as a random effect was an important consideration in our design given the expected changes in physiological requirements of the herbivore. After controlling for that effect, CP still emerged as an attractive limiting nutrient which has been suggested as limiting before energy in southern portion of the range of other ungulates (Cain et al., 2017). Like the toxicity constraint, if a different nutrient had been more limiting in the environment than crude protein, we expect diet select would have reflected a different constraint. Our data provide support for the nutrient balance hypothesis and as predicted, deer diet selection was explained simultaneously by attrac- tion and avoidance. Future research using a similar design could ad- vance our knowledge of diet selection behavior further by incorporating more populations with varying nutritional constraints to determine how these processes change with fluctuating nutritional demands and differences in relative abundance of available nutrients. Understanding how diet selection of a species changes with differing nutritional con- straints could aid in conservation of those species by informing management decisions to target the relaxation of nutritional con- straints. Author Contributions JLD, BKS, and MAL originally formulated the idea, JLD, BKS, SD, DBR, and MAL developed methodology, JLD and DBR conducted field work, JLD, BKS, and MAL performed statistical analyses, and JLD, BKS, SD, DBR, and MAL wrote the manuscript. Declaration of Competing Interest None Acknowledgements We thank the Mississippi State University Extension Service for funding and support. We are grateful to Bill Hamrick for his extensive involvement and assistance with this research. We thank the multiple graduate students and technicians for their help collecting data. We also thank Grassland Oregon, Inc. and Mossy Oak, Inc. for their donations to the project. Appendix A. Supplementary data Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.beproc.2020.104196. References Alipayo, D., Valdez, R., Holechek, J.L., Cardenas, M., 1992. Evaluation of micro- histological analysis for determining ruminant diet botanical composition. Journal of Range Management 45, 148–152. Belsley, D.A., Kuh, E., Welsch, R.E., 2005. Regression diagnostics: Identifying influential data and sources of collinearity Vol. 571 John Wiley & Sons. Berteaux, D., Crête, M., Huot, J., Maltais, J., Ouellet, J.P., 1998. Food choice by white- tailed deer in relation to protein and energy content of the diet: a field experiment. Oecologia 115, 84–92. Biggerstaff, M.T., Lashley, M.A., Chitwood, M.C., Moorman, C.E., DePerno, C.S., 2017. Sexual segregation of forage patch use: Support for the social-factors and predation hypotheses. Behavioural processes 136, 36–42. Bormann, F.H., Likens, G.E., 1967. Nutrient cycling. Science 155 (3761), 424–429. Cain, J.W., Gedir, J.V., Marshal, J.P., Krausman, P.R., Allen, J.D., Duff, G.C., Jansen, B.D., Morgart, J.R., 2017. Extreme precipitation variability, forage quality and large her- bivore diet selection in arid environments. Oikos 126 (10), 1459–1471. Cammack, K.M., Wright, C.L., Austin, K.J., Johnson, P.S., Cockrum, R.R., Kessler, K.L., Olson, K.C., 2010. Effects of high-sulphur water and clinoptilolite on health and growth performance of steers fed forage-based diets. Journal of Animal Science 88, 1777–1785. Ceacero, F., Landete-Castillejos, T., García, A.J., Estévez, J.A., Martinez, A., Calatayud, A., Gaspar-López, E., Gallego, L., 2009. Free-choice mineral consumption in Iberian red deer ( Cervus elaphus hispanicus ) response to diet deficiencies. Livestock Science 122, 345–348. Ceacero, F., Landete-Castillejos, T., Olguín, A., Miranda, M., García, A., Martínez, A., Cassinello, J., Miguel, V., Gallego, L., 2015. Avoiding toxic levels of essential mi- nerals: a forgotten factor in deer diet preferences. PloS one 10, e0115814. Chance, D.P., McCollum, J.R., Street, G.M., Strickland, B.K., Lashley, M.A., 2019. Native species abundance buffers non-native plant invasibility following intermediate forest management disturbances. Forest Science 65, 336–343. Chance, D.P., McCollum, J.R., Street, G.M., Strickland, B.K., Lashley, M.A., 2020. Vegetation characteristics influence fine-scale intensity of habitat use by wild turkey and white-tailed deer in a loblolly pine plantation. Basic and Applied Ecology 43, 42–51. Cherry, M.J., Conner, L.M., Warren, R.J., 2015. Effects of predation risk and group dy- namics on white-tailed deer foraging behavior in a longleaf pine savanna. Behavioral Ecology 26, 1091–1099. Craney, T.A., Surles, J.G., 2002. Model-dependent variance inflation factor cutoff values. Quality Engineering 14, 391–403. Danell, K., Mikael Utsi, P., Thomas Palo, R., Eriksson, O., 1994. Food plant selection by reindeer during winter in relation to plant quality. Ecography 17, 153–158. Deguchi, Y., Sato, S., Sugawara, K., 2001. Relationship between some chemical compo- nents of herbage, dietary preference and fresh herbage intake rate by the Japanese serow. Applied Animal Behaviour Science 73, 69–79. Distel, R.A., Laca, E.A., Griggs, T.C., Demment, M.W., 1995. Patch selection by cattle: maximization of intake rate in horizontally heterogeneous pastures. Applied Animal Behaviour Science 45, 11–21. J.L. Dykes, et al. Behavioural Processes 179 (2020) 104196 6 Dostaler, S., Ouellet, J.P., Therrien, J.F., Côté, S.D., 2011. Are feeding preferences of white-tailed deer related to plant constituents? Journal of Wildlife Management 75, 913–918. Drewnoski, M.E., Pogge, D.J., Hansen, S.L., 2014. High-sulfur in beef cattle diets: a re- view. Journal of animal science 92, 3763–3780. Dumont, B., 1997. Diet preferences of herbivores at pasture. Annales de zootechnie. INRA/EDP Sciences 46, 105–116. Dykes, J.L., Strickland, B.K., Demarais, S., Reynolds, D.B., Lashley, M.A., 2018. Soil nu- trients indirectly influence intraspecific plant selection in white-tailed deer. Basic and Applied Ecology 32, 103–109. Forbes, J.M., Provenza, F.D., 2000. Integration of learning and metabolic signals into a theory of dietary choice and food intake. In: Cronje, P.B. (Ed.), Ruminant physiology: digestion, metabolism, growth and reproduction’, pp. 3–19. Gillette, R., Huang, R.C., Hatcher, N., Moroz, L.L., 2000. Cost-benefit analysis potential in feeding behavior of a predatory snail by integration of hunger, taste, and pain. Proceedings of the National Academy of Sciences 97, 3585–3590. Ginane, C., Petit, M., 2005. Constraining the time available to graze reinforces heifers’ preference for sward of high quality despite low availability. Applied Animal Behaviour Science 94, 1–14. Harper, K.J., McNeill, D.M., 2015. The rol