Release from sheep-grazing appears to put some heart back into upland vegetation: A comparison of nutritional properties of plant species in long-term grazing experiments

Rewilding or wilding is a popularised means for enhancing the conservation value of marginal land. In the British uplands, it will involve a reduction, or complete removal, of livestock grazing (sheep), based on the belief that grazing has reduced plant species diversity, the ‘ Wet Desert ’ hypothesis. The hope is that if livestock is removed, diversity will recover. If true, we hypothesise that the species extirpated/reduced by grazing and then recover on its removal would more nutritious compared to those that persisted. We test this hypothesis at Moor House National Nature Reserve (North-Pennines), where seven sets of paired plots were established between 1953 and 1967 to compare ungrazed/sheep-grazed vegetation. Within these plot-pairs, we compared leaf properties of seven focal species that occurred only, or were present in much greater abundance, in the absence of grazing to those of 10 common species that were common in both grazed and ungrazed vegetation. Each sample was analysed for macro-nutrients, micro-nutrients, digestibility, palatability and decomposability. We ranked the species with respect to 22 variables based on effect size derived from Generalised Linear Modelling (GLM) and compared species using a Principal Components Analysis. We also assessed changes in abundance of the focal species through time using GLMs. Our results support the ‘ Wet Desert ’ hypothesis, that is, that long-term sheep grazing has selectively removed/reduced species like our focal ones and on recovery, they were more nutritious (macro-nutrients, some micro-nutrients) palatable, digestible and decomposable than common species. Measured changes in abundance of the focal species suggest that their recovery will take 10 – 20 years in blanket bog and 60 years in high-altitude grasslands. Collectively, these results suggest that sheep grazing has brought about biotic homogenization, and its removal in (re)wilding schemes will reverse this process eventually! The ‘ white woolly maggots ’ have eaten at least part of the heart out of the highlands/uplands, and it will take some time for recovery.

these results suggest that sheep grazing has brought about biotic homogenization, and its removal in (re)wilding schemes will reverse this process eventually! The 'white woolly maggots' have eaten at least part of the heart out of the highlands/uplands, and it will take some time for recovery.
conservation, digestibility, ecological restoration, land abandonment, long-term experiments, palatability, plant nutrients, sheep grazing 1 | INTRODUCTION Rewilding or wilding are terms that describe a range of management approaches, ranging from the introduction of wide-ranging large animals, especially top carnivores (Soulé & Noss, 1998), through to the abandonment of land, and a reduction in stock-grazing pressure (Corlett, 2016;Merckx & Pereira, 2015). The effects of stock grazing pressure were first identified in the United Kingdom by Frank Fraser Darling, who coined the term 'wet desert' to describe the species-poor vegetation of the Scottish Highlands, which he ascribed to a high, longterm, sheep-grazing pressure (Crumley, 2000;Darling, 1955;Stewart, 2010). Monbiot (2013) continues this debate, arguing that the British uplands are species-poor wastelands, 'sheep-wrecked', because of the high sheep-grazing pressure. In this situation, sheep are often referred to as 'white woolly maggots' or 'hoofed locusts' that 'have eaten the heart out of the Highlands/uplands' (Toogood, 1995;Monbiot, 2013; Baroness Young of Old Scone, pers. comm.).
If it is true that high sheep grazing has reduced species diversity, this can be translated into two hypotheses. First, that the historic high grazing pressure has removed certain species selectively (biotic trait homogenization; Smart et al., 2005Smart et al., , 2006, and second that removal of that grazing pressure will allow those species to return. If this were to be the case, we would predict that: 1 Species that have survived grazing will tend to have similar traits with respect to nutritional value, digestibility and palatability; 2 The species extirpated or reduced by grazing will have greater nutritional value, be more digestible and palatable and because of the higher nutritional status, they will decompose much faster. We can test these hypotheses by comparing the response of vegetation where sheep grazing can be compared with comparable ungrazed areas, usually within fenced exclosures. A good example of a series of such exclosure studies are those set up on the Moor House National Nature Reserve (NNR) in the North of England between 1953 and 1967 (Marrs, Rawes, Robinson, & Poppitt, 1986;Milligan, Rose, & Marrs, 2016). These experimental plots are distributed across the reserve; each compares sheep-grazed and ungrazed comparator plots, thus allowing an assessment of the effects of grazing removal on a range of plant community types encompassing a large proportion of British upland plant communities (Averis et al., 2004;Rodwell, 1991Rodwell, , 1992. The plant communities included vegetation dominated by dwarf-shrubs, grasses and sedges, growing on soils ranging from deep blanket peat through to brown-earth soils, and subject to very different, and indeed changing, sheep grazing pressures, which were related to forage quality (Eddy, Welch, & Rawes, 1968;Rawes & Welch, 1969). These vegetation types, in common with elsewhere in upland Britain, are described as degraded by sheep overgrazing (Darling, 1955;McGovern, Evans, Dennis, Walmsley, & McDonald, 2011).
A first assessment  of both soils and the quality of the total herbage (macro-nutrients and digestibility) in eight of these experiments showed almost no difference in macro-nutrient concentrations or digestibility. The only significant result was for one of the digestibility measures (acid detergent fibre concentration, ADF), which was lower where sheep were removed, indicating the vegetation had become more digestible . However, although there were few differences at the vegetation scale (total herbage), it was obvious that some species had either colonised or increased markedly in abundance within the ungrazed plots compared to those sheep-grazed (Milligan et al., 2016). Where sheep continued to graze, there was a reduction in species diversity and in the abundance of vascular plants, grasses, lichens, liverworts, and mosses; but an increase in herbs, sedges, and shrubs. Removal of sheep grazing reduced the abundance of grasses and liverworts compared to their grazed counterparts but herbs, mosses, sedges, and shrubs all increased (Milligan et al., 2016). The species that have increased after grazing removal are presumably those that have been reduced or extirpated by sheep grazing, and have recovered as a result of the zero sheep grazing pressure.
Here, we capitalise on these long-term experiments by comparing the traits of seven species (here termed focal species), which have either colonised the ungrazed plots since grazing ceased, or have become much more abundant than under grazed conditions, with 10 common species that occur widely in both grazed and ungrazed plots. For each species, we measured the concentrations of macro-nutrients, micro-nutrients, dry matter, fibre, lignin, protein and surrogate measures of metabolizable energy, digestibility (Si) and decomposability (C: N ratio). If our two hypotheses were correct, we would expect the seven focal species to be more nutritious, palatable, digestible and be capable of decomposition faster than the common species that have survived sheep-grazing. At the same time, we quantified the time taken for focal species to become abundant after grazing stopped.

| METHODS
This study used seven sheep-exclosure experiments located across the major moorland vegetation types found across Moor House National Nature Reserve in the northern Pennines of England ( Figure S1). These experiments were set up between 1953 and 1967 and designed to assess the impact of stopping sheep grazing (ungrazed exclosure) relative to free-range, sheep grazing. The vegetation types covered were representative of many upland ecosystems found in much of upland Britain with six National Vegetation Classification (NVC) plant community types included (Table S1). These communities cover ca. 80% of this reserve where they are grazed at a range of sheep densities (Table S1).
It is important, however, to realise that whilst long-term effects of sheep grazing versus no grazing are visible at these sites, the background grazing pressure has not been static. The reserve is a Common under English law, which means that designated farm-holdings from outside the moorland have the 'right' to graze their sheep on the land.
In the late 1960s, detailed studies by Rawes and Welch (1969)  House under the Commons Registration Act (1965), grazing density reduced >50% to 7,000 sheep or 2 sheep ha −1 . In the early 2000s, buy-out of some of the common rights-of-grazing led to further reductions in sheep numbers to c. 3,500 sheep or one sheep ha −1 (Milligan et al., 2016;Milligan, Rose, O'Reilly, & Marrs, 2018). The conservation objective for these reductions was the hope that it would lead to an improved vegetation quality. Rawes and Welch (1969) also showed that sheep grazing pressure in the plant communities available to the sheep was not random, with 11.6-23.2% greater densities on the most-grazed grassland communities compared to the least-grazed Blanket bogs (Table S1). Changing pollutant loads (SO 2 and NO x ) have also varied during this time and may also have affected species responses Rose et al., 2016). Hence, our experiments reflect an assessment of the effects of no sheep grazing relative to a dynamic 'business-as-usual' grazed scenario where the grazing pressure has reduced (Milligan et al., 2016(Milligan et al., , 2018.

| Vegetation sampling
At the end of July 2016, the seven experiments were visited, and individual species sampled in two groups based on visual inspection of the plots. Group 1 denoted here, as common species were present in reasonable abundance in both grazed and ungrazed plots. Some species (Group 1a) were found in only one experiment and comprised: Carex bigelowii, Nardus stricta, and Vaccinium myrtillus. Juncus squarrosus was also sampled in the ungrazed plot on one experiment.
Group 2 denoted here as focal species, were either present or abundant in the ungrazed plot of one experiment but were absent from, or present in very low abundance in, the grazed plots. This group comprised Dryopteris dilatata, which was present at low densities, and six species that were present in abundance (Chamaenerion angustifolium, Geum rivale, Narthecium ossifragum, Potentilla erecta, Rumex acetosa and Rubus chamaemorus) in at least one ungrazed plot.
For all species, three patches were selected randomly and plant parts harvested; shrubs = new annual shoots, graminoids = green leaves, dicotyledons = new shoots, fern = whole frond. In the laboratory, the samples were oven-dried at 80 C for 48 hr and milled to pass a 1 mm mesh.

| Chemical analyses
Total N and C determinations were made using a Thermo Scientific Flash 2000 Organic Elemental Analyser. For P and cations (K, Na, Ca and Mg), plant samples were analysed using the dry-ashing method (Allen, 1989). P was analysed by colorimetry (P) using a Seal Analytical AA3 HR AutoAnalyser and cations by absorption (Ca and Mg) and emission spectrophotometry (K and Na) on a Thermo Electron Corporation Solaar S4 AAS. The C:N ratio was used as a surrogate measure for decomposability.

| Micro-nutrients
Micro-nutrient element concentrations (Cl, Co, Cr, Cu, Fe, Mn, Mo, Ni, S, Si, Zn) were determined on the plant samples after using an Energy Dispersive X-ray Fluorescence Analyser (ED-XRF). Dried samples were pressed (1.5 t) in 20 mm pots and measured under a He atmosphere using a Spectro XEPOS 3 ED-XRF that emits a combined binary Pd and Co excitation radiation and uses a high resolution, low spectral interference silicon drift detector. The XRF analyser undergoes a daily standardisation procedure, with accuracy verified using 18 certified reference materials (Boyle, Chiverrell, & Schillereff, 2015). Si concentration was used as a surrogate measure of palatability (Massey, Massey, Ennos, & Hartley, 2009

| Statistical analysis
All statistical analyses were performed in the R statistical environment (R Core Team, 2017); the "vegan" package was used for the multivariate analyses (Oksanen et al., 2019).
Initially, the common species were tested for differences in leaf properties between grazed and ungrazed plots. Where a common species was collected at only one site (Group 1a), differences in leaf properties between samples collected in the grazed and ungrazed plots were tested using a t-test ("t.test" function). Where a common species was collected at more than one site (Group 1b), differences in leaf properties were tested using analysis of variance ("aov" function) using sites and grazing treatment as factors. Of the 168 grazed versus ungrazed contrasts (both t-tests and aov) only three produced a significant difference between grazed and ungrazed treatments (p < .01), with A. flexuosa having greater concentrations of Ca, Fe and Mn in grazed plots compared to ungrazed ones (Table S2). Accordingly, A. flexuosa, was treated as two species (A. flexuosa-G and A. flexuosa-U) for the analyses for these three elements and the multivariate analysis. Otherwise, as there were no other significant differences between grazed/ungrazed plots, all data for common species were pooled in all other analyses.
Generalised linear models (GLM) were used to investigate the relative differences in leaf nutritional properties between species. The GLMs were, therefore run with species as a fixed factor and as the variables were all continuous ones, a Gaussian error structure was used with transformed data (elements = log e (x) and percentages (digestibility variables = asin(sqrt[x/100]). For five variables where a high value represented low nutritional quality (C, ADF, NDF, Si concentrations and C:N ratio), the model intercepts were set to the species with the largest mean value. All other species were then ranked in graphs by effect size (estimate) away from this intercept (an example of this analysis is presented in Figure 1a for ADF). For all other variables, a similar approach was used except the intercepts were set at the species with the lowest mean value, that is, the species with the least nutritional value (an example of this analysis is presented in Figure 1b for Mg concentration). The approach allowed the spectrum of response to be ranked in terms of nutritive value from the worst to the best species. Assuming the hypothesis is accepted, that is, that the focal species were more nutritious/digestible/palatable/ decomposable than the common species then the focal species should be ranked 1-7 out of the 17 species assessed (18 when Df-G and Df-U were separated) and be plotted at the positive end of the graph. The graphs for all variables are presented as Figure S2 and all statistical outputs are presented in Tables S3-S6. For brevity, the discussion centres around a summary table of ranks, derived from these individual analyses for each focal species, assessed against their respective intercept.
In addition, the difference between the focal and common species groups were also analysed for each variable using GLM using the same analytical methodology except that species group was the fixed factor (i.e., common versus focal). These statistical outputs are presented in Table S7.
In addition, to summarise all the measured variables in one analysis, the combined dataset (macro-, micro-nutrients, digestibility, palatability and decomposability) were analysed together using principal components analysis (PCA), an unconstrained ordination technique using the rda function after standardization (mean = 0, s 2 = 1) using the 'decostand' function. The relative positions for each species on the biplots were visualized using standard-deviational ellipses with 95% confidence intervals, fitted using the 'ordiellipse' function.  Tables S4-S6). The rank of the seven focal species (Table 1)  Database for each experiment. These data were collected using pin quadrats within random quadrats (Marrs et al., 1986). The species abundance in each grazed and ungrazed plot were summed at quadrat level for each year and modelled against time. GLM modelling was described as above but as the data were counts a Poisson error structure with a log-link function was used (Crawley, 2013). The statistical outputs are presented in Table S8.  (Table 1). Importantly, each species had a different combination of elements that were greater in the top ranks (Table 1). The focal species group had significantly greater concentrations than the common species group for all macro-nutrients elements except C, which was significantly lower (Table 1).
Compared to macro-nutrients the pattern for the micro-nutrients was less clear with some species for some elements showing no significant difference from the intercept species (Table 1, Figure S2,   Table 1).

| Digestibility
Here, there are two things to consider (a) the fibre concentrations (ADF, NDF) and (b) the quality of the plant material, assessed through DOMD (a surrogate energy measure) and protein concentration.
For fibre, five species (C. angustifolium, G. rivale, N. ossifragum, P. erecta, R. acetosa [ADF only] and R. chamaemorus) were ranked in the top seven, that is, they had lowest concentrations of ADF, NDF or both (Table 1, Figure S2, Table S5), indicating that they are more digestible than the common species. All were significantly less than the intercept species (p < .001). D. dilatata was ranked twelfth and eighth for ADF and NDF respectively and R. acetosa was ranked fifteenth for ADF.   Figure S2, Table S5).
The focal species group had significantly lower concentrations of ADF and NDF and greater concentrations of DOMD and protein than the common species (Table 1).

| Decomposability and palatability
All focal species were significantly different from the intercept (p < .001) for both Si concentration (p < .001) and C:N ratio (p < .001; Table 1, Figure S6, Figure 2). For palatability (Si concentration), five of the seven focal species (C. angustifolium, D. dilatata, G. rivale, N. ossifragum and R. acetosa) were ranked in the top seven, and for decomposability (C:N ratio), but six (C. angustifolium, D. dilatata, G. rivale, N. ossifragum, P. erecta and R. acetosa) were ranked in the top seven (Table 1). The focal species group had significantly lower Si concentrations and C:N ratios than the common species indicating greater palatability and decomposability (Table 1).

| Multivariate analysis
The PCA produced eigenvalues of 7.477 and 3.174 for the first two axes explaining 51% of the total variation in the dataset. The Summary of the ranks of the seven focal species found in ungrazed plots at Moor House NNR with respect to a range of macroand micro-nutrient concentrations along with measures of digestibility, palatability and decomposability Decomposability C:N ratio 2 4 1 3 5 7 8 F < C, p < .0001 Note: The ranks have been derived from glm modelling and subsequent interpretation of boxplots (Figure 1, Figures S1-S4). The shaded data illustrates species that are in the top 7 with respect to nutritional value (n = 17, except * where n = 18) and values in bold type indicate a significant difference (p < .05) from the intercept (Tables S3-S6). Summary results of a glm to test for differences between focal (F) and common (C) species groups are also presented (full results in Table S7); ns, not significant (p > .10). located in the lower right quadrant, indicating correlations with high energy, protein, N, P, Ca, Mg, Na, K and micro-nutrients (Figure 2b).

| Species change through time
Two focal species showed no significant change through time. D. dilatata was patchily distributed and was below the detection limits of the sampling methodology and C. angustifolium showed an increase in abundance after 60 years in the enclosed plot but no significant temporal relationship. Of the remaining five focal species, only N.
ossifragum showed a slight increase in the sheep grazed plots, no relationship could be fitted for the other four species when sheep grazed ( Figure 3). The responses of the three species found predominantly in grasslands (G. rivale, P. erecta and R. acetosa) were relatively small when grazing was removed, with increases being detectable after 60 years ( Figure S5). The two species found in predominantly Calluna-Eriophorum-dominated vegetation (N. ossifragum and R. chamaemorus) showed greater and faster increases after sheep grazing was removed, that is, over a 10-to 20-year period (Figure 3). suggests that they will decompose more slowly and act as a negative feedback on primary production. Second, when sheep grazing was removed, the focal species, that is, those that either colonised or F I G U R E 3 Change in the abundance of the seven focal species in both grazed and ungrazed plots at Moor House National Nature Reserve in northern England; data are derived from pin quadrat touches. Plotted lines are fitted significant relationships from a Generalised Linear Model; if no line is present a significant relationship could not be fitted; statistical properties of the fitted relationships are presented in Table S8 increased markedly in the exclosures, were more nutritious in terms of at least some macro-nutrients, digestibility and palatability; a few also had greater concentrations of some micro-nutrients. Moreover, their lower C:N ratio suggests that where present they would decompose faster and produce a positive feedback on production. These local increases in abundance could contribute to reversing biotic/trait homogenization (Smart et al., 2005(Smart et al., , 2006. The larger Ca and/or Mg concentrations in six of the seven focal species, and their improved digestibility, relative to the common ones is consistent with Mládková, Mládek, Hejduk, Hejcman, and Pakeman (2018), who suggested that high values of these elements increase digestibility. This is partly associated with evolutionary development and the differential concentrations of these elements in cell walls (White, Broadley, El-Serehy, George, & Neugebauer, 2018). High concentrations of Ca and Mg have also been used as indices of litter decomposition rate (Cornelissen & Thompson, 1997).
Taken together, the results support Frank Fraser Darling's (1955) 'wet desert' hypothesis and Monbiot's (2013) view of upland degradation, in that nutritious plants are selectively removed/reduced by sheep. It also suggests that relaxation of the grazing pressure will allow species that have extirpated or reduced to recover, but if the rates from these experiments at Moor House are typical then this will take some time; 20-40 years on blanket bog and 60 years for high-altitude grasslands. It should also be noted that increasing N.
ossifragum may prove problematic if there were any stock grazing around as it is highly poisonous (Angell & Ross, 2011) as a result of a toxic component identified as 3-methoxy-2(5H)-furanone (Langseth et al., 1999). The faster recovery of blanket bog species might be because the stock grazing pressure on this vegetation type is lower than grasslands (Table S1; Rawes & Welch, 1969). Hence, the degradation of vegetation in terms of species loss might be slower, it might not reach as low a base, and hence recovery is faster; this hypothesis remains to be tested. At least some of the heart has been eaten out of the Highlands/uplands, and it will take some time to replace it.

| Limitations on the use of exclosures
There are several drawbacks to the approach of merely comparing the change in vegetation in sheep-free exclosures versus the business-asusual sheep grazing pressure. First, the use of exclosures means that the sheep grazing pressure is set at zero and hence no information can be derived on impacts of very low grazing pressures that might be associated with (re)wilding schemes. Second, we know that the business-as-usual stock grazing pressure has been reduced at the landscape scale during the experimental period, and possibly from a higher historic base in the 19th century (Darling, 1955 These reductions will almost certainly have intensified differences in grazing pressures between the different vegetation types brought about by sheep selection (Table S1), for example by maintaining a high pressure on the more-productive grasslands but reducing it on the least-productive blanket bogs (Eddy et al., 1968;Rawes & Welch, 1969

| Implications of these results
These results have important implications in terms of ecological theory and conservation management practice.
From a theoretical perspective, the focal species have larger quantities of at least some nutrients in their tissues. There were no significant differences in soil chemical properties between any of the grazed and exclosure plots when sampled in 2015 .
As a consequence, the focal species must either be able to extract nutrients in a more efficient way than the common species from the same soil resource and/or they invest resources into more palatable/ digestible tissues making them more susceptible to grazing. That species have differing elemental compositions is well known (Lambers & Poorter, 1992;Thompson, Parkinson, Band, & Spencer, 1997), and reflects evolutionary status, and differences in root characteristics, relative growth rates, root: shoot ratios, rhizosphere interactions, improved mycorrhizal associations, absorption processes, foraging behaviour or differential resource allocation within the shoots (Chapin, 1980;Hutchings & de Kroon, 1994;Thompson et al., 1997).
It remains unclear whether internal plant physiological functions, mycorrhizal or rhizosphere interactions brought about these increased elemental concentrations. The relative relationship between fungal and bacterial communities at the micro-level may also be affected by the colonisation by these focal plant species; low fertility soils tend to be fungal-dominated, higher fertility soils bacterial-dominated (de Vries et al., 2012;Smith et al., 2003). Reductions in airfall acid deposition in these upland communities  may also have affected the species responses. Further studies are needed to elucidate the exact mechanisms involved.
From a conservation management perspective, the removal of grazing livestock in (re)wilding schemes should eventually increase plant species diversity, but it may as seen here take a considerable time. There are at least two potential constraints. The first is one of seed limitation. We have very little information on the potential for species dispersal into these large areas, or on the existing seedbanks, which may or may not contain a legacy of potential colonists. If management wishes to accelerate the colonisation of new species, then it may be necessary to add seed (Hester, Gimingham, & Miles, 1991;Miles, 1974;Mitchell, Rose, & Palmer, 2008). The second constraint is the lack of 'safe-sites' (Harper, 1977); bare gaps necessary for the seed to germinate and establish. The herbage mass in these experiments range from 850 to 2,900 g m −2 , and almost certainly this vegetation will prevent seeds getting into the soil pool (Ghorbani, Le Duc, McAllister, Pakeman, & Marrs, 2006) and subsequently germinating (Lowday & Marrs, 1992;Miles, 1974). A range of approaches could be used to create these gaps by disturbing the extant vegetation including the use of herbicides, fire or physical damage by cutting, rotavating or screefing (Humphrey & Coombs, 1997;Lee et al., 2013;Miles, 1974;Milligan et al., 2004).
Experimental research to develop integrated approaches combining disturbance with seed addition will be needed if the intended objective is to accelerate (re)wilding as part of a conservation management strategy.